Observation of top quark pairs produced in association with a vector boson in pp collisions at sqrt(s) = 8 TeV

Measurements of the cross sections for top quark pairs produced in association with a W or Z boson are presented, using 8 TeV pp collision data corresponding to an integrated luminosity of 19.5 inverse femtobarns, collected by the CMS experiment at the LHC. Final states are selected in which the associated W boson decays to a charged lepton and a neutrino or the Z boson decays to two charged leptons. Signal events are identified by matching reconstructed objects in the detector to specific final state particles from ttW or ttZ decays. The ttW cross section is measured to be 382 +117 -102 fb with a significance of 4.8 standard deviations from the background-only hypothesis. The ttZ cross section is measured to be 242 +65 -55 fb with a significance of 6.4 standard deviations from the background-only hypothesis. These measurements are used to set bounds on five anomalous dimension-six operators that would affect the ttW and ttZ cross sections.


Introduction
Since the LHC at CERN achieved proton-proton collisions at center-of-mass energies of 7 and 8 TeV, it has become possible to study signatures at significantly higher mass scales than ever before.The two heaviest sets of particles produced in standard model (SM) processes that could be observed using the data already collected are top quark pairs produced in association with a W or Z boson (ttW and ttZ), which have expected cross sections of σ(ttW) = 203 +20 −22 fb and σ(ttZ) = 206 +19 −24 fb in the SM in 8 TeV collisions [1].The dominant production mechanisms for ttW and ttZ in pp collisions are shown in Fig. 1.The ttZ production cross section provides the most accessible direct measurement of the top quark coupling to the Z boson.Both σ(ttW) and σ(ttZ) would be altered in a variety of new physics models that can be parameterized by dimension-six operators added to the SM Lagrangian.
The ttZ cross section was first measured by the CMS experiment in 7 TeV collisions, with a precision of about 50% [2].Measurements in events containing three or four leptons in 8 TeV collisions at CMS [3] have constrained σ(ttZ) to within 45% of its SM value, and yielded evidence of ttZ production at 3.1 standard deviations from the background-only hypothesis.The CMS collaboration also used same-sign dilepton events to constrain σ(ttW) to within 70% of the SM prediction, with a significance of 1.6 standard deviations from the background-only hypothesis.Most recently, the ATLAS experiment used events containing two to four leptons to measure σ(ttW) = 369 +100 −91 fb at 5.0 standard deviations from the background-only hypothesis, and σ(ttZ) = 176 +58 −52 fb with a significance of 4.2 standard deviations from the background-only hypothesis [4].
We present the first observation of ttZ production and measurements of the ttW and ttZ cross sections using a full reconstruction of the top quarks and the W or Z boson from their decay products.We target events in which the associated W boson decays to a charged lepton and a neutrino (W → ν) or the Z boson decays to two charged leptons (Z → ).In this paper, "lepton" ( ) refers to an electron, a muon, or a τ lepton decaying into other leptons.The top quark pair may decay into final states with hadronic jets (tt → bqq bqq), a lepton plus jets (tt → b ν bqq), or two leptons (tt → b ν b ν).The ttZ process is measured in channels with two, three, or four leptons, with exactly one pair of same-flavor opposite-sign leptons with an invariant mass close to the Z boson mass [5].The ttW process is measured in channels with two same-sign leptons or three leptons, where no lepton pair is consistent with coming from a Z boson decay.Additional b-tagged jets and light flavor jets are required to enable full or partial reconstruction of the top quark and W boson decays.The charge conjugate process of ttW + produces ttW − .Channels defined by lepton charge and multiplicity are further subdivided by lepton flavor and the number of jets, in order to provide an initial separation between signal and background (Section 5).Background processes with leptons from W and Z boson decays are estimated using Monte Carlo simulations that are validated in separate control regions (Section 6. 1).Processes with leptons from other sources are estimated directly from the data, using events in which one or more leptons fail to satisfy a strict set of selection criteria (Sections 6.2 and 6.3).In each channel, we attempt a full or partial reconstruction of the ttW or ttZ system with a linear discriminant that matches leptons and jets to their parent particles using mass, charge, and b tagging information (Section 7).Additional kinematic variables from leptons and jets are combined with output from the linear discriminant in a multivariate analysis that is used to make the final measurement of the ttW and ttZ cross sections (Sections 8 and 10).Finally, the measured cross sections are used to constrain the coupling of the top quark to the Z boson, and to set bounds on five anomalous dimension-six operators (Section 11).

The CMS detector
The central feature of the CMS apparatus is a superconducting solenoid of 6 m internal diameter, providing a magnetic field of 3.8 T. Within the solenoid volume are a silicon pixel and strip tracker, a lead tungstate crystal electromagnetic calorimeter (ECAL), and a brass and scintillator hadron calorimeter (HCAL), each composed of a barrel and two endcap sections.Extensive forward calorimetry complements the coverage provided by the barrel and endcap detectors.Muons are detected in gas-ionization muon chambers embedded in the steel flux-return yoke outside the solenoid.
A global event description is obtained using the CMS particle-flow (PF) algorithm [6,7], which combines information from all CMS sub-detectors to reconstruct and identify individual particles in collision events.The particles are placed into mutually exclusive classes: charged hadrons, neutral hadrons, photons, muons, and electrons.The primary collision vertex is identified as the reconstructed vertex with the highest value of ∑ p 2 T , where p T is the momentum component transverse to the beams, and the sum is over all the charged particles used to reconstruct the vertex.The energy of photons is directly obtained from the ECAL measurement, corrected for zero-suppression effects.The energy of electrons is determined from a combination of the electron momentum at the primary interaction vertex as determined by the tracker, the energy of the corresponding ECAL cluster, and the energy sum of all bremsstrahlung photons spatially compatible with originating from the electron track.The energy of muons is obtained from the curvature of the corresponding track and hits in the muon chambers.The energy of charged hadrons is determined from a combination of their momentum measured in the tracker and the matching ECAL and HCAL energy deposits, corrected for zero-suppression effects and for the response function of the calorimeters to hadronic showers.Finally, the energy of neutral hadrons is obtained from the corresponding corrected ECAL and HCAL energy.
A more detailed description of the CMS detector, together with a definition of the coordinate system used and the relevant kinematic variables, can be found in Ref. [8].

Data and simulated samples
This search is performed with an integrated luminosity of 19.5 ± 0.5 fb −1 of proton-proton collisions at √ s = 8 TeV, collected in 2012 [9].Dilepton triggers were used to collect data for all channels.The dilepton triggers require any combination of electrons and muons, where one lepton has p T > 17 GeV and another has p T > 8 GeV.A trielectron trigger with minimum p T thresholds of 15, 8, and 5 GeV was also used for channels with three or more leptons.These triggers approach their maximum efficiency for leptons with p T values at least 2 GeV higher than the thresholds.
Expected signal events and some of the background processes are modeled with simulation.The signal processes ttW and ttZ, as well as background processes producing a single Z boson, WZ, ZZ, W ± W ± , WWW, WWZ, tt, ttγ, ttγ * , ttWW, and the associated production of a Z boson with a single top quark (tbZ), are all generated with the MADGRAPH 5.1.3[10] tree-level matrix element generator, combined with PYTHIA 6.4 [11] for the parton shower and hadronization.The associated production of a Higgs boson with a top quark pair (ttH) is modeled using the PYTHIA generator assuming a Higgs boson mass of 125 GeV.Samples that include top quark production are generated with a top quark mass of 172.5 GeV.The CTEQ6L1 parton distribution function (PDF) set [12] is used for all samples.
The CMS detector response is simulated using GEANT4 software [13].Both data and simulated events are required to pass the same trigger requirements and are reconstructed with identical algorithms.Effects from additional proton-proton collisions in the same bunch crossing (pileup) in the simulation are modeled by adding simulated inclusive proton-proton interactions (generated with PYTHIA) to the generated hard collision, with the pileup interaction multiplicity in simulation reflecting the profile inferred from data.Correction factors are applied to individual objects and events to bring object properties and efficiencies in simulation into better agreement with data, as described in Section 4.

Object reconstruction and identification
Certain types of particles reconstructed with the PF algorithm are particularly useful in identifying and reconstructing ttW and ttZ events.These objects are electrons, muons, charged and neutral hadrons clustered into jets, and the imbalance in p T arising from neutrinos in the event.
Electrons with p T > 10 GeV are reconstructed over the full pseudorapidity range of the tracker, |η| < 2.5.The reconstruction combines information from clusters of energy deposits in the ECAL and the electron trajectory reconstructed in the inner tracker [14,15].A multivariate analysis technique combines observables sensitive to the amount of bremsstrahlung, spatial and momentum matching between the track and associated ECAL clusters, and shower shape observables, to distinguish genuine electrons from charged hadrons [14].
Muons with |η| < 2.4 and p T > 10 GeV are reconstructed using information from both the silicon tracker and the muon spectrometer [16].Track candidates must have a minimum number of tracker hits, be compatible with hits in the muon chambers, and match the associated energy deposits in the calorimeters, to be selected as PF muons [17].
The τ leptons decay before reaching the ECAL, and are not identified in this analysis.Their decay products are instead identified as hadrons, which may be clustered into jets, or as electrons or muons, depending on whether the τ lepton decays to hadrons or leptons.
Prompt leptons (electrons or muons from a W, Z, or Higgs boson, or the decay of a τ lepton) are distinguished from non-prompt leptons (misidentified jets or leptons from hadron decays) in part by assessing their isolation from surrounding hadronic activity.Lepton isolation is calculated by summing the p T of other particles in a cone of radius ∆R = √ (∆η) 2 + (∆φ) 2 = 0.4 around the lepton direction, where ∆η and ∆φ are the pseudorapidity and azimuthal angle difference (in radians) from the lepton direction.Contributions from charged particles not originating from the primary collision vertex are subtracted from the isolation sum, multiplied by a factor of 1.5 to account for the neutral pileup contribution [18].The relative isolation of the lepton is defined as the ratio of the corrected isolation sum to the lepton p T .
Prompt leptons are also identified by having low impact parameter (IP) and impact parameter significance (S IP ) values, where the impact parameter is the minimum three-dimensional distance between the lepton trajectory and the primary vertex, and its significance is the ratio of the IP value to its uncertainty.(These values tend to be higher for electrons and muons from the decay of τ leptons, which have a nonnegligible lifetime.)Furthermore, the properties of the nearest jet enclosing the lepton (within ∆R < 0.5) can be used to identify non-prompt leptons.The ratio of the lepton p T to the p T of this enclosing jet tends to be lower for non-prompt leptons.Also, an enclosing jet identified as coming from a bottom quark indicates that the lepton is likely non-prompt and originates from a b-hadron decay.
Three levels of lepton selection are defined: preselected, loose, and tight.The preselection includes leptons in data sidebands used to compute non-prompt backgrounds, the loose criteria select signal leptons in channels dominated by prompt lepton events, and the tight selection is used when the largest backgrounds contain non-prompt leptons.Loose leptons form a subset of the preselected leptons, and tight leptons form a subset of the loose leptons.The selection requirements are described below and summarized in Table 1.
The preselection removes leptons with an enclosing jet identified as a bottom jet, as described below, and imposes very loose requirements on the distance from the lepton trajectory to the primary vertex in the z direction and in the x-y plane, and on the S IP value.Preselected leptons must also have a relative isolation less than 0.4.The preselection has ≈100% efficiency for prompt leptons, and accepts a substantial number of non-prompt leptons.Loose leptons must lie below certain thresholds on the relative isolation calculated using only charged particles (0.15 for electrons and 0.20 for muons), and loose muons pass a tighter requirement on S IP .The loose selection retains 93-99% of prompt muons and 89-96% of prompt electrons, depending on p T and η, and rejects ≈50% of non-prompt leptons that pass the preselection.Tight leptons must pass several selection criteria: the charged relative isolation must be less than 0.05 for electrons and 0.10 for muons; the ratio of lepton to enclosing jet p T must be more than 0.6; and for electrons, the IP must be less than 0.15 mm.The tight selection efficiency is ≈90% for prompt muons and ≈80% for prompt electrons, with efficiency ranges of 68-98% for muons and 49-93% for electrons, depending on p T and η.The tight selection rejects ≈80% of nonprompt muons and ≈85% of non-prompt electrons that pass the preselection.
In order to reject leptons with misreconstructed charge, the preselected, loose, and tight leptons in some channels must pass additional charge identification (ID) requirements.Electrons must pass a veto on electrons from photon conversions and have no missing hits in the inner tracker, and muons must have more than five inner tracker hits.Electrons must also have the same charge assignment from the tracker and from the relative location of ECAL energy deposits from the electron itself and from its bremsstrahlung radiation.This charge ID selection efficiency ranges from 85 to 100% for tight electrons with correctly identified charge, depending on p T and η, while more than 97% of electrons with misreconstructed charge are rejected.The charge ID selection has 99% efficiency for tight muons with correctly identified charge and rejects ≈100% of muons with misreconstructed charge.Lepton selection efficiencies are measured using same-flavor (SF) lepton pairs with an invariant mass near the Z boson mass.The charge ID selection requirements are summarized in Table 1.
Charged and neutral PF particles are clustered into jets using the anti-k T algorithm with a distance parameter of 0.5 [19,20].Selected jets must be separated by ∆R > 0.5 from the selected leptons, and have p T > 25 GeV and |η| < 2.4.Charged PF particles not associated with the primary event vertex are removed from jet clustering, and additional requirements remove jets arising entirely from pileup vertices [21].A neutral component is removed by applying a residual energy correction following the area-based procedure described in Refs.[22,23], to account for pileup activity.Fake jets from instrumental effects are rejected by requiring each jet to have at least two PF constituents and at least 1% of its energy from ECAL and HCAL deposits.
The combined secondary vertex (CSV) algorithm [24,25] is used to identify (or "tag") jets originating from a bottom quark.The CSV algorithm utilizes information about the impact parameter of tracks and reconstructed secondary vertices within the jets to assign each jet a discriminator, with higher values indicating a likely b-quark origin.For a selection with the medium working point of the CSV discriminator, the b tagging efficiency is around 70% (20%) for jets originating from a bottom quark (charm quark), and the chance of mistagging jets from light quarks or gluons is about 1%.For the loose working point, the efficiency to tag jets from b quarks (c quarks) is approximately 85% (40%), and the probability to tag jets from light quarks or gluons is about 10%.These efficiencies and mistag probabilities vary with the p T and η of the jets.
The missing transverse momentum vector, arising from the presence of undetected neutrinos in the event, is calculated as the negative vector sum of the p T of all PF candidates in the event.This vector is denoted as p miss T , and its magnitude as p miss T .Since pileup interactions can cause missing transverse momentum not associated with the primary interaction, the magnitude of the negative vector sum of the p T of only selected jets and leptons (H miss T ) is also used.The H miss T variable has worse resolution than p miss T , but it is more robust as it does not rely on lowp T objects in the event.
The simulation is corrected with data-to-simulation scale factors in order to match the performance of reconstructed objects in data.Simulated events with leptons are corrected for trigger efficiency, as well as for lepton identification and isolation efficiency.Scale and resolution corrections accounting for residual differences between data and simulation are applied to the muon and electron momenta.All lepton corrections are derived from samples with a Z boson or J/ψ decaying into two leptons.Jet energy corrections based on simulation and on γ+jets, Z+jets, and dijet data are applied as a function of the jet p T and η [26].Separate scale factors ranging from 0.6 to 2.0 are applied to light and heavy flavor jets to correct the distribution of CSV values [27].

Event selection
Events for this analysis are divided into five mutually exclusive channels, targeting different decay modes of the ttW and ttZ systems.For all channels, at least one lepton is required to have p T > 20 GeV, and the remaining leptons must have p T > 10 GeV, to satisfy the dilepton trigger requirements.In addition, to reject leptons from Υ, J/ψ, and off-shell photon decays, no pair of leptons can have an invariant mass less than 12 GeV.The selection requirements for each channel are described below and summarized in Table 2.The opposite-sign (OS) dilepton channel targets ttZ events where the Z boson decays into an OS pair of electrons or muons, and the tt system decays hadronically.We select events with loose OS leptons forming an invariant mass within 10 GeV of the Z boson mass and at least five jets, where one or more jets pass the medium CSV working point.The channel is split into categories with SF lepton pairs (targeting events with a Z boson) and different-flavor pairs (to calibrate the tt background).It is further subdivided into events with exactly five jets and those with six or more jets, which have a higher signal-to-background ratio.This categorization provides an initial separation of the ttZ signal from the dominant Z boson and tt backgrounds, which are estimated from simulation.
The same-sign (SS) dilepton channel selects ttW events in which the associated W boson, and the W boson of the same charge from the tt system, each decay to a lepton and a neutrino, and the remaining W boson decays to quarks.Events are selected with two SS tight leptons which pass the charge ID criteria, plus three or more jets, of which at least two pass the loose CSV threshold or at least one passes the medium CSV threshold.In addition, in dielectron events, the ee invariant mass must be at least 10 GeV away from the Z boson mass, to reject Z boson decays in which the charge of one electron is misidentified.This channel is divided by lepton flavor (ee, eµ, and µµ), and further into categories with exactly three jets and four or more jets.The dominant background is tt with one non-prompt lepton, which is estimated from data by computing a misidentification rate.Diboson WZ events (modeled with simulation) are selected if one lepton from the Z boson decay does not pass the preselection, or if the Z boson decays to a pair of τ leptons, of which only one produces a muon or electron.For the ee and eµ categories, dileptonic Z boson and tt events with a charge-misidentified electron also appear in the final selection.
The three-lepton (3 ) ttW channel targets events in which both the associated W and the pair of W bosons from the tt pair decay leptonically.Events are selected in which the lepton charges add up to ±1, and the two leptons of the same charge pass the tight identification and the charge ID criteria.Furthermore, no SF OS pair of leptons can have a mass within 10 GeV of the Z boson mass.Events must have at least one medium b-tagged jet, or at least two loose b-tagged jets, and are divided into categories with exactly one jet, or with two or more jets.
The main backgrounds are tt decays with a non-prompt lepton, estimated from data, and WZ events, estimated using simulation.
The 3 ttZ channel selects events in which the Z boson decays to a pair of electrons or muons, and one W boson from the tt system decays to a charged lepton and a neutrino, with the remaining W boson decaying to quarks.The selection is identical to the one used for the 3 ttW channel, except that at least one SF OS pair of leptons must have an invariant mass within 10 GeV of the Z boson mass, and the categories have exactly three jets, or four or more jets.The dominant backgrounds are Z boson and tt events with a non-prompt lepton, and WZ events with prompt leptons, estimated in the same manner as in the 3 ttW channel.
Events with four leptons (4 ) come from ttZ decays in which the Z boson and both W bosons decay leptonically.This channel requires four leptons that pass the loose identification and the charge ID, and whose charges add up to zero.At least one SF OS dilepton pair must have a mass within 10 GeV of the Z boson mass, and at least one loose b-tagged jet must be present.In addition, H miss T must exceed 30 GeV.These criteria, and the categorization of events into those with exactly one lepton pair consistent with a Z boson decay, and those with two or more, help separate ttZ events from the dominant ZZ background, which is estimated using simulation.Small backgrounds from tt, WZ, and Z boson events with one or two non-prompt leptons are also estimated using simulation.

Signal and background modeling
Events in the signal channels fall into three broad categories.Signal and "prompt" background events have enough leptons from W or Z boson decays, with the correct charges, to satisfy the lepton selection of the channel."Non-prompt" backgrounds have at least one lepton which is a jet misidentified as an electron, or which comes from the in-flight decay of a hadron, or from photon conversion.The "charge misidentified" background has an electron whose charge was misidentified.The expected yields for these processes after the final selection are shown in Tables 3-5 in Section 6.4.

Signal and prompt backgrounds
The signal and prompt backgrounds are estimated using simulation, normalized to their predicted inclusive cross sections.We use next-to-next-to-leading-order (NNLO) cross sections for tt [28] and single Z boson [29] production; next-to-leading-order (NLO) cross sections for ttW and ttZ [1], ttH [30], ttγ [10,31], WZ and ZZ [32], and WWW, WWZ, and tbZ [10] production; and leading-order cross sections for W ± W ± , ttγ * , and ttWW [10] production.Additional corrections are derived from data for Z boson, WZ, and ZZ processes with multiple extra jets.
Rare processes such as SS diboson (W ± W ± ) and triboson production (WWW, WWZ), associated production of a Z boson with a single top quark (tbZ), and tt with an on-shell or off-shell photon (ttγ/ttγ * ) or two W bosons (ttWW) are subdominant backgrounds.The associated production of a Higgs boson with a top quark pair is included as a background, with uncertainties derived from theoretical predictions.All of these are minor backgrounds, with fewer expected events than the signal in each channel.
The main prompt backgrounds are tt and Z boson production (in the OS dilepton channel), WZ events (in the SS and 3 channels), and ZZ events (in the 3 and 4 channels).Because the Z boson, WZ, and ZZ simulation samples are produced with fewer extra partons from QCD radiation than there are jets in the final selection, their estimated contributions to the signal channels are approximations with large uncertainties.To get a more accurate estimate of these yields, scale factors are derived from events with SF OS leptons consistent with a Z boson decay and no medium b-tagged jets.Using about 5000 data events, of which 97% are expected to come from Z → events, we correct the predicted yield from the Z boson simulation as a function of the number of jets for events with five or more jets.To validate this technique, we derive a scale factor from four jet events with no medium b tags and apply it to events with at least one medium b tag, and find that it yields good agreement between data and the Z boson simulation.These scale factors range from 1.35 to 1.7, and each has an uncertainty of 30%, based on the level of data-to-simulation agreement in Z boson events with four jets.Additional uncertainties in the η distribution of jets in Z boson and tt events, and on the p miss T distribution in Z boson events with extra jets, are assessed due to possible data-to-simulation discrepancies in OS dilepton events with four or more jets (excluding the OS ttZ signal region).Scale factors for simulated WZ and ZZ events with three or more jets are derived from 80 three-lepton data events (70% from WZ) with no medium b-tagged jets and at most one loose b-tagged jet.The scale factors of 1.4 for three-jet events, and 1.6 for events with four or more jets, have 40% and 60% uncertainties, respectively, based on the limited number of 3 data events used to derive the scale factors.
In addition, there is significant uncertainty associated with the simulation of events with extra heavy flavor partons.Simulated tt, WZ, and ZZ events with one or two extra c jets, an extra b jet, or two extra b jets are separated from their inclusive samples and assigned extra rate uncertainties of 50%.The single Z boson simulation is divided similarly.However, by comparing the expected and observed numbers of b-tagged jets in SF OS events with low p miss T and exactly four jets, we are able to constrain the uncertainty in each of the Z boson plus heavy flavor jet processes to 30%.
The top quark p T spectrum in tt simulation (from MADGRAPH) is corrected to agree with the distribution predicted by higher-order calculations [33] and observed in tt differential cross section measurements in √ s = 8 TeV data, using the techniques described in Ref. [34].

Non-prompt backgrounds
Backgrounds with at least one non-prompt lepton are expected to have larger yields than the signal in the SS and 3 ttW channels, about the same yields in the 3 ttZ channel, and very low yields in the 4 channel.Non-prompt backgrounds in the SS and 3 channels are estimated from data.A sideband region dominated by non-prompt processes is defined by events which pass the same selection as the signal channels, but in which one or both of the preselected SS leptons fail the tight lepton criteria.Extrapolation to the signal region is performed by weighting the sideband events by the probability for non-prompt leptons to pass the tight lepton selection (the misidentification rate, ).Events in which one of the SS leptons fails the tight lepton requirement enter the signal region estimate with weight /(1 − ).Events where both SS leptons fail the tight lepton selection get a negative weight this accounts for events with two non-prompt leptons contaminating the sideband sample of events with a single non-prompt lepton.
The misidentification rate is measured with SS and 3 data events, separately for electrons and muons, and as a function of the lepton p T .Same-sign dilepton events with two or more jets (excluding the ttW signal region) are dominated by tt decays with a non-prompt lepton.Threelepton events with two or fewer jets, a lepton pair consistent with a Z boson decay, and low p miss T come mostly from Z boson production with an extra non-prompt lepton.These events usually have exactly one prompt and one non-prompt SS lepton, so we use a modified tag-and-probe approach in which the prompt lepton is tagged with the tight lepton selection, and the fraction of preselected probe leptons passing the tight selection measures .Because both leptons in the numerator of this ratio are tight, there is a ≈50% chance that the tag lepton was actually non-prompt, and the probe lepton was prompt.We estimate the size of this contamination by weighting events where the tag lepton fails the tight selection by /(1 − ), and subtract those with a tight probe lepton from the numerator, and those with a preselected probe from the denominator.
Since this correction term depends on itself, we cannot solve for explicitly.Instead, we find the set of p T -and flavor-dependent values that minimizes the difference between the data and predicted yields in the SS and 3 derivation regions, binned by lepton p T and flavor.Events in which both SS leptons are non-prompt naturally cancel to zero with the correction term, while those with two prompt SS leptons are estimated from simulation and subtracted explicitly.The misidentification rate in all the p T bins is computed to be ≈20% for muons and ≈15% for electrons, except for the muon bin with p T > 30 GeV, whose rate is 36%.This rate is uncorrelated with variables that do not depend on the lepton flavor or p T , including most of those used to separate signal from background events (Section 8).The relative uncertainty in is assessed at 40% for electrons and 60% for muons, equal to the maximum observed discrepancy between predicted and observed yields in any of 20 background-dominated selection regions with two SS leptons and two or more jets, or three leptons and two or fewer jets.There is an additional statistical uncertainty of 50% for leptons with medium p T and 100% for leptons with high p T , due to low event yields in the derivation regions.
In the 4 channel, there are too few events passing the kinematic requirements to use a data sideband to model the non-prompt background.Instead we use simulated WZ, Z boson, and tt samples to estimate non-prompt yields after the final selection, which are expected to be much smaller than the signal yields.We derive a scale factor for the simulation estimate of nonprompt leptons passing the loose selection using simulated Z boson and tt events with exactly three loose leptons and one or two jets, where at least one passes a medium b tag.Events with a SF OS lepton pair close to the Z boson mass are dominated by Z boson plus non-prompt lepton events; those without such a pair are dominated by tt plus non-prompt lepton events.The derived scale factor of 2.0 per non-prompt lepton is then applied to the simulation in the 4 category, with 100% rate uncertainties.

Charge-misidentified backgrounds
The misidentified charge background in SS dilepton events is estimated from OS dilepton events in data that pass all the other signal channel selections, weighted by the probability for an electron passing the charge ID requirement to have misidentified charge.This probability is derived from data as a function of electron η from the ratio of SS dielectron events with an invariant mass within 10 GeV of the Z boson mass and zero or more jets, to OS events with the same selection.The probability ranges from 0.003% for central electrons to 0.1% for endcap electrons.The absence of a Z boson mass peak in SS dimuon events indicates that the probability is negligible for muons.Opposite-sign eµ events enter the SS prediction region with a weight equal to the probability for the electron to have its charge misidentified; ee events enter with the sum of the probabilities for each electron.The charge misidentification probability has a 30% rate uncertainty, based on the agreement between predicted and observed SS dielectron events with multiple jets and with the ee invariant mass close to the Z boson mass.We expect to see fewer events with charge misidentified electrons than ttW signal events in all the SS dilepton channels.

Expected yields
Expected yields for the signal and background processes after the final fit described in Section 10, along with the observed data yields, are shown in Tables 3-5.
Table 3: Expected yields after the final fit described in Section 10, compared to the observed data for OS ttZ final states.Here "hf" and "lf" stand for heavy and light flavors, respectively. OS

Full event reconstruction
Even after the selection requirements have been applied, the final signal categories are dominated by background events.To help identify the ttW and ttZ signals, and the tt background, we attempt a full reconstruction of the events, by matching leptons, jets, and p miss T to the decaying W and Z bosons, and to the top quark and antiquark.
In all channels targeting the ttZ signal, the SF OS pair of leptons with an invariant mass closest to the Z boson mass is assumed to be from the Z boson decay.In selected ttW events, there Table 5: Expected yields after the final fit described in Section 10, compared to the observed data for 3 ttW and three and 4 ttZ final states.The 4 "Z-veto" channel has exactly one lepton pair consistent with a Z boson decay; the "Z" channel has two.The multiboson process includes WWW and WWZ; tt+X includes ttγ, ttγ * , and ttWW.are at least two leptons and two undetected neutrinos, so the associated W boson cannot be reconstructed.Thus, for both ttW and ttZ events, as well as tt events, it is the tt system which remains to be reconstructed.In selected OS ttZ events, both W bosons from the tt pair decay into quarks; we refer to this as a fully hadronic tt decay.In SS ttW and 3 ttZ events, the tt pair decays semileptonically.The 3 ttW and 4 ttZ channels target leptonic tt decays.While background tt events have genuine top quarks to reconstruct, they decay in a different mode than the signal does, e.g. in OS tt events both W bosons decay leptonically, and in SS and 3 tt events one lepton usually comes from a b-hadron decay.
The leptons, jets, and p miss T from tt decays preserve information about their parent particles.Pairs of jets from hadronic W boson decays have an invariant mass close to the W boson mass; adding the b jet from the same top quark decay gives three jets with an invariant mass close to the top quark mass.In semileptonic tt decays, the transverse component of the lepton momentum vector and p miss T give a Jacobian mass distribution which peaks around 60 GeV and quickly drops as it approaches the W boson mass.Additionally, the lepton and b jet coming from the same top quark decay will have an invariant mass smaller than the top quark mass.Jets from b quarks tend to have higher CSV values, while light flavor jets have lower values, and c jets have an intermediate distribution.The jet charges of b jets from top quarks and quark jets from W boson decays are also used.Finally, the ratio of the invariant mass using only the transverse component of momentum vectors (M T ) to the full invariant mass tends to be higher for the set of jets coming from top quark and W boson decays than for sets with jets from extra radiated partons.These variables are all used in the event reconstruction described below, and are listed in Appendix A, Table 10.
To optimally match jets and leptons to their top quark and W boson parents in tt decays, we construct a linear discriminant, similar to a likelihood ratio, which evaluates different permutations of object-parent pairings.The discriminant is created using millions of simulated tt events, so the true parentage of each object is known, and the variable distribution shapes have high precision.For each input variable to the discriminant, we take the ratio of the distribution using correctly matched objects (e.g. the invariant mass of two jets coming from the same W boson decay) to the distribution using any set of objects (e.g. the invariant mass of any two jets in the event), and rescaled the ratio histogram to have a mean value of one for correctly matched objects, as shown in Fig. 2. Variables with more discriminating power, such as the reconstructed W boson mass, have ratio histograms with some bin values very close to zero, and others above one; less discriminating variables such as b jet charge have values well above zero in all bins.Ratio mass q q q bq ν bl → t t We use these ratio histograms to match objects to tt decays in selected events in the ttW and ttZ signal regions, where the object parentage is not known.In ttZ channels the leptons matched to the Z boson decay are excluded from the tt reconstruction, and in ttW channels the lepton with the worst match to a tt decay is assumed to come from the associated W boson.For each permutation of leptons and jets matched to parent particles, we find the value of every variable (mass, CSV, charge, etc.) associated with an object-parent pairing.The matching discriminant is then computed as the product of the corresponding bin values from all the ratio histograms.The permutation with the highest discriminant value is considered to be the best reconstruction of the tt system.To more easily display the full range of values, we take the log of the discriminant value of the best reconstruction to calculate the match score.Events that contain all of the jets and leptons from the tt decay have match scores around zero, while events without all the decay products typically get negative scores.For semileptonic tt decays in events with exactly four jets, all from the tt system, the highest scored permutation is the correct assignment 75% of the time.For events with five or more jets, of which four are from the tt system, the exact correct match is achieved in 40% of cases, as there are five times as many permutations to choose from.Since one or two jets from the tt decay often fail to be reconstructed, we also attempt to match partial ttW and ttZ systems, with one or two jets missing.Output match scores in the OS ttZ, SS ttW, and 3 ttZ channels are shown in Fig. 3, along with the 68% confidence level (CL) uncertainty in the signal plus background prediction.
Since the background processes do not have the same parent particles as the signal in each channel, their best reconstructed match scores are typically lower.Thus, the match scores for full or partial reconstructions of the tt system in ttW, ttZ, and tt decays, along with the values of input variables to the chosen match (e.g.dijet mass of the hadronically decaying W boson in a semileptonic tt decay), provide good discrimination between signal and background processes, especially those without any genuine top quarks.Figure 3: Distributions for match scores with signal and background yields from the final fit described in Section 10.Plot (a) shows the match score for partially reconstructed hadronic tt systems in SF OS dilepton events with six or more jets.Plots (b) and (c) show scores for fully reconstructed semileptonic tt systems in events with at least four jets, and a SS eµ pair, or three leptons, respectively.The 68% CL uncertainty in the signal plus background prediction is represented by hash marks in the stack histogram, and a green shaded region in the data-toprediction ratio plot.The orange line in plot (a) shows the shape of the ttZ signal, suitably normalized."Ch.misID" indicates the charge-misidentified background."Other" backgrounds include ttγ, ttγ * , ttWW, tbZ, WWW, WWZ, and W ± W ± .

Signal extraction
The match scores and other event reconstruction variables are combined with kinematic quantities (e.g.lepton p T and jet CSV values) in boosted decision trees (BDTs) [35] to distinguish signal events from background processes.The linear discriminant for event reconstruction combines a large number of variables into maximally distinctive observables, achieving better separation than a BDT alone would, since fewer variables can be used in a BDT when the number of simulated events for training is limited.A separate BDT is trained for each jet category in each analysis channel, for a total of 10 BDTs.The input variables to these BDTs are described below, and listed in Appendix B, Tables 11-15.
An initial BDT in the OS channel is trained with ttZ events against the tt background, using the Z boson mass and p T , and ∆R separation between leptons as inputs, as well as H miss T , the number of jets with p T > 40 GeV, and the ratio of the M T to the mass of a four-momentum vector composed of all the jets in the event.Event reconstruction variables include match scores to leptonic and fully hadronic tt decays, and the CSV values of jets matched to b quarks from the leptonic tt decay.The final BDT is then trained against Z boson and tt events, using the ttZ vs. tt BDT as an input, along with the two highest jet CSV values, the fifth-highest jet p T , the number of jets with p T > 40 GeV, and the ratio of the M T to the mass of all the selected leptons and jets.Match scores to the partial five-jet and full six-jet hadronic tt system are also included, along with the minimum χ 2 value of a fit to the full hadronic tt system that uses only the W boson and top quark masses as inputs.
The SS channel BDT is trained with ttW events against tt simulation, using the lepton p T values, p miss T , the second-highest jet CSV value, and the M T of the system formed by the leptons, jets, and p miss T .Event reconstruction variables include match scores to three-and four-jet ttW decays and three-jet tt decays, the matched top quark candidate mass from two jets from the W boson and the non-prompt lepton from the b-hadron decay, and the other top quark candidate M T from the prompt lepton, p miss T , and the b jet in tt decays.
The BDT for the 3 ttW channel is trained against tt simulation, using the p T of the SS leptons, the highest jet p T , the second-highest jet CSV value, p miss T , and the M T of the leptons, jets, and p miss T .Match scores for the two-jet ttW system and one-jet tt systems, along with the invariant mass of the prompt and non-prompt leptons matched to the same top quark in a tt decay, are also used.
The 3 ttZ BDT is trained against simulated WZ and tt events, which contribute equally to the background in this channel.The input kinematic variables are the reconstructed Z boson mass (which discriminates against tt), the M T of the p miss T , leptons and jets, and the number of medium b-tagged jets.In the three-jet category, match scores for ttZ reconstructions with one or two jets missing from the semileptonic tt decay are used as inputs; in the four jet category, match scores for three-jet systems and the full four-jet system are used.
The 4 channel has too few signal and background events to train a BDT; here the number of medium b-tagged jets is used as a discriminant instead.This variable effectively separates ttZ events from the dominant ZZ background, and from subdominant non-prompt WZ, Z boson, and tt backgrounds.
The expected and observed distributions of the BDT output for each channel and category are shown in Figs.5-6.The expected signal and background distributions represent the best fit to the data of the SM predicted backgrounds and signal, where the signal cross section is allowed to float freely.The 68% CL uncertainty in the fitted signal plus background is represented by hash marks in the stack histogram, and a green shaded region in the data-to-prediction ratio plot.The 95% CL band from the fit is shown in yellow.Events in the 3 ttZ channel with high BDT values (>0.3 for three jet events, > − 0.2 for events with four or more jets) should provide a high-purity sample of ttZ events.Data distributions of the reconstructed Z boson and top quark properties are consistent with the SM ttZ signal, as shown in Fig. 7. -

Figure 5:
The final discriminant for SS ttW channel events with 3 jets (top) and ≥4 jets (bottom), after the final fit described in Section 10.The lepton flavors are ee (a, d), eµ (b, e), and µµ (c, f).The 68% CL uncertainty in the fitted signal plus background is represented by hash marks in the stack histogram, and a green shaded region in the data-to-prediction ratio plot.The 95% CL band from the fit is shown in yellow."Ch.misID" indicates the charge-misidentified background."Other" backgrounds include ttγ, ttγ * , ttWW, tbZ, WWW, WWZ, and W ± W ± .

Systematic uncertainties
There are several systematic uncertainties that affect the expected rates for signal and background processes, the shape of input variables to the BDTs, or both.The most important uncertainties are on the b tagging efficiency, signal modeling, and the rates of non-prompt backgrounds and prompt processes with extra jets.Some uncertainties affect the simulation in all of the channels, and are correlated across the entire analysis.The integrated luminosity has an uncertainty of 2.6% [9].The total inelastic proton-proton cross section is varied up and down by 5%, which affects the number of pileup vertices, and is propagated to the output distributions [36].
The properties and reconstruction efficiencies of different objects have their own uncertainties.The uncertainty in the jet energy scale [26] is accounted for by shifting the energy scale up and down by one standard deviation for all simulated processes, and evaluating the output distributions with the shifted energy scale.The shape of the CSV distribution for light flavor or gluon jets, c jets, and b jets has uncertainties associated with the method used to match the CSV shapes in data and simulation, as detailed in Ref. [27].Calibration regions for light flavor jets have some contamination from heavy flavor jets, and vice versa.The associated uncertainty  decay, after the final fit described in Section 10.The 68% CL uncertainty in the fitted signal plus background is represented by hash marks in the stack histogram, and a green shaded region in the data-to-prediction ratio plot.The 95% CL band from the fit is shown in yellow.The orange line shows the shape of the ttZ signal, suitably normalized.The tt+X background includes ttW, ttH, and ttWW; "Other" backgrounds include ttγ, ttγ * , ttWW, tbZ, WWW, and WWZ. in the final light or heavy flavor CSV shape is accounted for by varying the expected yields of contaminating jets up and down by one standard deviation, and propagating the result to the final CSV distribution.The weights for these alternate shapes are applied to produce alternate final discriminant histograms in each channel.Likewise there are uncertainties from the limited number of events in the calibration regions; these are assessed using the maximum linear and quadratic deformations of the CSV shape within an envelope whose size is determined by the magnitude of the statistical uncertainty.Because there is no calibration region to determine the CSV shape of c jets in data, they receive no correction factors, but have all the b jet uncertainties applied to them, multiplied by a factor of two so that they include the scale factor values for b jets.
Prompt electron and muon efficiency uncertainties are computed using high-purity dilepton samples in data from Z boson decays.These include rate uncertainties associated with the trigger efficiency, reconstruction efficiency, and the fraction of prompt leptons passing the tight, loose, and charge ID selection criteria.
The rate of non-prompt leptons passing the tight lepton selection receives a 40% uncertainty  , and the mass of the best fit dijet pair from a W boson decay (c) and trijet system from a top quark decay (f).The plots show signallike events from the 3 ttZ channel (3 jets with BDT > 0.3 and ≥4 jets with BDT > −0.2) before the final fit described in Section 10 is performed.The green band in the data-to-prediction ratio plot denotes the 68% CL rate and shape uncertainties in the signal plus background prediction."Other" backgrounds include ttγ, ttγ * , ttWW, tbZ, WWW, and WWZ.
for electrons and a 60% uncertainty for muons, based on the agreement between expected and observed yields in control regions in data, as described in Section 6.Additional uncertainties of 50% and 100% are assessed on the rates of non-prompt leptons with medium and high p T , respectively, because of the limited number of events in the sample used to find the misidentification rates.These uncertainties are applied separately for electrons and muons, and are uncorrelated between the SS and 3 channels, to account for possible differences in the sources of non-prompt leptons.While the uncertainties on event yields with non-prompt electrons and muons are initially large, the final fit constrains them to 10-15% using bins in the final discriminants which contain mostly non-prompt backgrounds.
The rates of charge misidentified electrons in the SS channels receive a 30% rate uncertainty, based on the agreement between predicted and observed SS dielectron events consistent with a Z boson decay.
Theoretical uncertainties from the PDFs of different simulated processes, as well as the choice of renormalization and factorization scales, are accounted for with rate uncertainties in all signal and backgrounds processes.The rate uncertainties for ttW and ttZ are 10% and 11%, respectively, from the choice of scales [1], and 7.2% and 8.2% from the PDFs [37,38].In addition, shape uncertainties derived from simulation generated with different PDF sets and PYTHIA tunes are applied to the ttW, ttZ, and ttH processes using linear and quadratic deformations of 10-11% on the final discriminant shape.The ttW and ttZ rate uncertainties are not included in the ttW and ttZ cross section measurements, respectively, and neither is used in the simultaneous measurement of the ttW and ttZ cross sections.
The systematic uncertainty in top quark p T reweighting in simulated tt events is assessed by applying no top quark p T weight for the lower systematic uncertainty, and twice the weight for the upper systematic uncertainty.Since neither higher-order theoretical calculations [39] nor independent control region studies currently constrain the normalization of the t t+cc, t t+b, or t t+bb processes to better than 50% accuracy, an extra 50% uncorrelated rate uncertainty is assigned to each process.An additional shape uncertainty is applied to the ratio of the M T to the invariant mass of the system of jets in tt events with five, or six or more jets.
Because the Z boson, WZ, and ZZ simulations are used to model events with more jets than there are extra partons in the generated event, rate uncertainties are assigned to these processes.Events with a Z boson plus five jets and six or more jets receive uncorrelated 30% rate uncertainties, based on the extrapolation from Z boson events with four jets and no medium b-tagged jets to those with at least one medium b tag.Diboson WZ and ZZ events with three jets and four or more jets have uncorrelated 40% and 60% uncertainties, respectively, due to the limited number of events in the light flavor sideband used to calibrate jet multiplicity.Diboson events with extra heavy flavor jets receive uncertainties identical to the tt plus heavy flavor simulation.The good data-to-simulation agreement in dileptonic Z boson events with four jets and one or two medium b tags constrains the Z + cc, Z + b, and Z + bb uncertainties to 30% each.Simulated Z boson events have extra shape uncertainties in H miss T and the M T -to-mass ratio of jets, uncorrelated between events with five and six or more jets, and between the different jet flavor subsamples.These account for possible data-to-simulation differences seen in Z boson events with four or more jets (excluding the ttZ signal region).Although these uncertainties are large, the Z boson and diboson backgrounds are well separated from the ttZ signal using the final discriminants, so they have a small effect on the final measurement.
Rare processes with low expected yields such as triboson production (WWW, WWZ), associated production of a Z boson with a single top quark (tbZ), and tt with an on-shell or off-shell photon (ttγ/ttγ * ) or two W bosons (ttWW) get 50% rate uncertainties, because they are either calculated at leading order or require extra jets or b jets to enter the signal region.
The expected impact of different sources of systematic uncertainty is estimated by removing groups of uncertainties one at a time and gauging the improvement in the signal strength precision, as measured using pseudo-data from simulation.(The measurement technique is described in the next section.)If we expect to measure a signal strength of 1 ± δ i with all the systematic uncertainties included, and expect to measure 1 ± δ i =j with fewer uncertainties, a large reduction in uncertainty j = δ i − δ i =j indicates that the removed uncertainties have a significant impact on the measurement.Uncertainties in b tagging efficiency, signal modeling, and rates of prompt processes with extra jets are found to have the greatest effect on the ttZ signal precision, while the ttW measurement is most impacted by uncertainties in the non-prompt backgrounds, b tagging efficiency, and signal modeling.The full set of systematic uncertainties and their expected effects are shown in Table 6.Because we are measuring j and not δ j , we do not expect the quantities in Table 6 to add in quadrature.10 Cross section measurement The statistical procedure used to compute the ttW and ttZ cross sections and their corresponding significances is the same as the one used for the LHC Higgs boson analyses, and is described in detail in Refs.[40,41].A binned likelihood function L(µ, θ) is constructed, which is the product of Poisson probabilities for all bins in the final discriminants of every channel.The signal strength parameter µ characterizes the amount of signal, with µ = 1 corresponding to the SM signal hypothesis, and µ = 0 corresponding to the background-only hypothesis.Systematic uncertainties in the signal and background predictions are represented by a set of nuisance parameters, denoted θ.Each nuisance parameter represents a different source of uncertainty.When multiple channels have the same source of uncertainty, the nuisance parameter is correlated across the channels, allowing certain initially large systematic uncertainties (such as the rate of non-prompt leptons passing the tight selection) to be constrained in bins with a large number of data events but few expected signal events.
To test how consistent the data are with a hypothesized value of µ, we consider the profile likelihood ratio test statistic q(µ) = −2 ln L(µ, θµ )/L( μ, θ), where θµ denotes the set of values of the nuisance parameters θ that maximizes the likelihood L for the given µ.The denominator is the likelihood maximized over all µ and θ.This test statistic is integrated using asymptotic formulae [42] to obtain the p-value, i.e. the probability under the signal-plus-background hypothesis of finding data of equal or greater incompatibility with the background-only hypothesis.Results are reported both in terms of the best fit cross section and µ values and their associated uncertainties, and in terms of the significance of observation of the two signal processes.
We perform separate one-dimensional fits for the ttW and ttZ cross sections using the relevant channels for each process.The fit for each cross section is performed with the other cross section set to the SM value with the uncertainty coming from theory calculations.The resulting measurements and significances are reported in Tables 7 and 8.The ttZ cross section is measured with a precision of 25%, and agrees well with the SM prediction.The observed ttW cross section is higher than expected, driven by an excess of signal-like SS dimuon events in the data.Most of the signal-like dimuon events with four or more jets also contributed to a similar excess seen in the CMS ttH search [27].In both analyses, a close examination yielded no evidence of mismodeling or underestimated backgrounds, and the excess events appear consistent with a ttW or ttH signal.The best fit values for the ttW and ttZ cross sections are compatible with the SM expectation at the 13% and 60% CL, respectively.Taking into account significant differences in event selection, this result is also consistent with the previous CMS measurement [3], which it supersedes.
We also perform a simultaneous fit of both processes using all the channels.Figure 8 shows the two-dimensional likelihood scan over σ(ttW) and σ(ttZ).The respective best fit values are found to be 350 +150 −123 fb and 245 +104 −80 fb, compatible at the 15% CL with the SM expectation [1,43].

Extended interpretation
Direct measurement of the ttZ and ttW cross sections can be applied to searches for new physics (NP) within the framework of an effective field theory.The effects of new particles or interactions can be captured in a model-independent way by supplementing the SM Lagrangian with higher-dimensional operators involving SM fields.The effective Lagrangian can be written [44] as an expansion in the inverse of the cutoff energy scale, 1/Λ: where L SM is the SM Lagrangian density of dimension four, L 1 is of dimension five, etc.The Wilson coefficients c i and c j are numerical constants that parameterize the strength of the nonstandard interactions, and O i and O j are operators corresponding to combinations of SM fields.Hermitian conjugate terms are denoted h.c.Good agreement between data and SM expectations suggests that deviations due to NP are small and it is reasonable to work in the first order of c i and c j [45]; we limit ourselves to this domain.
It is not possible to construct a dimension-five operator that conserves lepton number [44], so only dimension-six operators are considered in this work.Assuming baryon number con- servation, there are 59 independent dimension-six operators [46].We follow the notation and operator naming scheme introduced in Ref. [47].We study the effect of these operators on the tZ coupling constants and the ttW and ttZ cross sections, and compare them to the measured values.

Constraints on the axial and vector components of the tZ coupling
Indirect measurements of the top quark to Z boson coupling include, for example, precision studies of the Z → bb branching fraction at LEP and the SLC [48][49][50][51][52][53].The ttZ process provides the first experimentally accessible direct probe of the tZ coupling.
The SM ttZ interaction Lagrangian can be written in terms of the vector and axial couplings C SM V and C SM A , which can be precisely calculated.In the effective field theory approach, the modified couplings C 1,V and C 1,A are considered, which can be written in terms of the SM contribution plus deviations due to the Wilson coefficients c j of dimension-six operators [54], scaled by Λ, the Higgs field vacuum expectation value v, and the weak mixing angle θ w : A method for calculating σ(ttZ) in terms of C 1,V and C 1,A has been presented in Ref. [54].The cross section depends on a constant term, linear and quadratic terms in C 1,V and C 1,A , and a mixed term.Each of these six terms is scaled by a factor which was evaluated in Ref. [54] by calculating the 7 TeV cross section at six points and solving the system of equations.To extrapolate to 8 TeV, we scale σ(ttZ)(C 1,V , C 1,A ) linearly by the ratio of the theoretical ttZ cross sections at 7 and 8 TeV.From this we define the signal strength parameter µ ttZ in terms of C 1,V and C 1,A , and a profile likelihood ratio test statistic, as described in Section 10.We perform a two-dimensional scan of the (C 1,V , C 1,A ) phase space to extract the best fit values, which are found to satisfy the constraint: The difference between the profile likelihood and the best fit profile likelihood is plotted as a function of the relative vector and axial components ∆C

Constraints on dimension-six operators
Both indirect and direct constraints on dimension-six operators are documented in Refs.[45,[54][55][56][57][58][59].To study the effects of NP on the ttW and ttZ processes, we use the FEYNRULES [60] implementation from Ref. [47].This implementation is used with MADGRAPH 5 [10] to compute cross sections as a function of (v 2 /Λ 2 ) c j , henceforward simply denoted by c j .Cross sections were computed for the production of tt, a Higgs boson, ttZ, and ttW, sampling 20 points for each c j .For each sampled point, all c k =j were fixed at zero.From this survey, we select five operators as of particular interest because they have a small effect on inclusive Higgs boson and tt production, and a large effect on ttZ, ttW, or both: cuB , c HQ , cHQ , cHu , and c3W .An alternative way to display the effect of each c j is shown in Fig. 10, where sampled values are plotted in the (σ(ttW), σ(ttZ)) plane.From these it is clear that cuB , cHu , and cHQ affect only ttZ, whereas c3W only affects ttW, and c HQ affects both processes.For each of the five operators, we perform a finer scan of 200 cross section points and use a spline fit to obtain an expression for the cross section in terms of c j , σ(ttZ) SM+NP (c j ).We define the signal strength µ ttZ (c j ) to be the ratio of the ttZ production cross section to the combined expectations from SM and NP σ(ttZ) SM+NP (c j ), and likewise for ttW.From this we can define a profile likelihood ratio in terms of c j , similarly to what is described in Section 10.Best fit values, along with 1σ and 2σ CL ranges are summarized in Table 9. Operators that affect either the ttW or the ttZ cross section, but not both, have symmetric likelihood distributions and thus have two best fit values.Bounds on c HQ , cHQ , and cHu are stricter than those derived in Ref. [54] from CMS and ATLAS searches for ttZ using LHC data at 7 TeV.Constraints on cuB are tighter than those derived in Ref. [59].

Summary
An observation of top quark pairs produced in association with a Z boson and measurements of the ttW and ttZ cross sections have been made, using 19.5 fb −1 of 8 TeV pp collision data collected by the CMS detector at the LHC.Signatures from different decay modes of the top quark pair resulting in final states with two, three, and four leptons have been analyzed.Signal events have been identified by uniquely matching reconstructed leptons and jets to final state particles from ttW and ttZ decays.Results from two independent ttW channels and three ttZ channels have been presented, along with combined measurements.The combined ttW cross section measurement in same-sign dilepton and three-lepton events is 382 +117 −102 fb, 4.8 standard deviations from the background-only hypothesis, where a significance of 3.5 standard deviations is expected in the standard model.Combining opposite-sign dilepton, three-lepton, and four-lepton channels, the ttZ cross section is measured to be 242 +65 −55 fb, an observation with a significance of 6.4 standard deviations from the background-only hypothesis, and in agreement with the standard model expectation.Using the measured cross sections, limits have been placed on the vector and axial couplings of the Z boson to the top quark, and on the Wilson coefficients of five dimension-six operators parameterizing new physics: cuB , c HQ , cHQ , cHu , and c3W .These measurements are compatible with the standard model predictions, and are the most sensitive reported to date to these high mass scale processes.
[4] ATLAS Collaboration, "Measurement of the t tW and t tZ production cross sections in pp collisions at √ s = 8 TeV with the ATLAS detector", (2015).arXiv:1509.05276.Submitted to JHEP.

Figure 1 :
Figure 1: Dominant leading order Feynman diagrams for ttW + and ttZ production at the LHC.The charge conjugate process of ttW + produces ttW − .

Figure 2 :
Figure 2: Distributions from simulated tt → b ν bqq events with exactly four jets.Shown are the invariant mass of two jets matched to a hadronic W boson decay (a), the invariant mass of any two jets (b), and the rescaled ratio of the two (c).

Figure 4 :
Figure4: The final discriminant for 3 ttW channel events with 1 jet (a) and ≥2 jets (b) after the final fit described in Section 10.The 68% CL uncertainty in the fitted signal plus background is represented by hash marks in the stack histogram, and a green shaded region in the data-toprediction ratio plot.The 95% CL band from the fit is shown in yellow."Other" backgrounds include ttγ, ttγ * , ttWW, btZ, WWW, and WWZ.

Figure 6 :
Figure 6: The final discriminant for ttZ channel events with two OS leptons and 5 jets (a) or ≥6 jets (d), three leptons and 3 jets (b) or ≥4 jets (e), or four leptons and two lepton pairs (c) or exactly one lepton pair (f) consistent with a Z →decay, after the final fit described in Section 10.The 68% CL uncertainty in the fitted signal plus background is represented by hash marks in the stack histogram, and a green shaded region in the data-to-prediction ratio plot.The 95% CL band from the fit is shown in yellow.The orange line shows the shape of the ttZ signal, suitably normalized.The tt+X background includes ttW, ttH, and ttWW; "Other" backgrounds include ttγ, ttγ * , ttWW, tbZ, WWW, and WWZ.

0 1 2Figure 7 :
Figure 7: Distributions of the mass (a) and p T (d) of the lepton pair identified with the Z boson decay, the number of jets (b) and medium b-tagged jets (e), and the mass of the best fit dijet pair from a W boson decay (c) and trijet system from a top quark decay (f).The plots show signallike events from the 3 ttZ channel (3 jets with BDT > 0.3 and ≥4 jets with BDT > −0.2) before the final fit described in Section 10 is performed.The green band in the data-to-prediction ratio plot denotes the 68% CL rate and shape uncertainties in the signal plus background prediction."Other" backgrounds include ttγ, ttγ * , ttWW, tbZ, WWW, and WWZ.

Figure 9 :
Figure 9: Difference between the profile likelihood and the best fit profile likelihood functions for the relative vector and axial components of the tZ coupling.Contours corresponding to the best fit and the 1, 2, and 3 standard deviation (σ) CLs are shown in lines.

Figure 10 :
Figure 10: Sampled coefficient values for cuB (a), c3W (b), c HQ (c), cHu (d), and cHQ (e), plotted in the (σ(ttW), σ(ttZ)) plane.There are typically two best fit values, one greater and one less than zero, which lie on top of one another in the plane.
BDT inputs: SS ttW vs. tt 3 jet ≥4 jets M T of p miss T and p T of leptons and jets 1 quark M T from b ν 7 7 Match score for ttW → b ν bq 9 8 Match score for ttW → b ν bqq -9 tt matched top quark mass from b qq 3 -

T and p T of leptons and jets 1 2 13 : 1
Match score for ttW → ν b ν b ν Input variables to BDT that distinguishes 3 ttZ from WZ and tt, ranked by signalbackground separation.BDT inputs: 3 ttZ vs. WZ and tt 3 jet ≥4 jets Match score for ttZ → b ν bq 1 Match score for ttZ → b ν bqq -ttZ → b ν b 2 -Match score for ttZ → ν bq 5 -Match score for ttZ → b ν q 6 -

Table 1 :
Summary of preselected, loose, tight, and charge ID lepton selection requirements.The charge ID requirements are applied in addition to the preselected, loose, or tight lepton criteria.

Table 2 :
Summary of selection requirements for each channel.

Table 4 :
Expected yields after the final fit described in Section 10, compared to the observed data for SS ttW final states.The multiboson process includes WWW, WWZ, and W ± W ± ; tt+X includes ttγ, ttγ * , and ttWW.

Table 6 :
Reduction in the expected signal strength uncertainties produced by removing sets of systematic uncertainties.The quantities in each column are not expected to add in quadrature.

Table 7 :
Expected and observed measurements of the cross section and signal strength with 68% CL ranges and significances for ttW, in SS dilepton and 3 channels.

Table 8 :
Expected and observed measurements of the cross section and signal strength with 68% CL ranges and significances for ttZ, in OS dilepton, 3 , and 4 channels.

Table 9 :
Constraints from this ttZ and ttW measurement on selected dimension-six operators.

Table 11 :
Input variables to the BDT that distinguishes SS ttW from tt, ranked by signalbackground separation.

Table 12 :
Input variables to BDT that distinguishes 3 ttW from tt, ranked by signal-background separation.

Table 14 :
Input variables to BDT that distinguishes OS ttZ from tt (used as input to the final discriminant), ranked by signal-background separation.

Table 15 :
Input variables to BDT that distinguishes OS ttZ from Z boson and tt (the final discriminant), ranked by signal-background separation.