Discovery potential for T ′ →tZ in the trilepton channel at the LHC
aa r X i v : . [ h e p - ph ] F e b Preprint typeset in JHEP style - HYPER VERSION
Discovery potential for T ′ → tZ in the trileptonchannel at the LHC Lorenzo Basso a ∗ and Jeremy Andrea b Universit`e de Strasbourg, IPHC, 23 rue du Loess 67037 Strasbourg, FranceInstitut Pluridisciplinaire Hubert Curien/D´epartement Recherches Subatomiques,CNRS-IN2P3, 23 Rue du Loess, F-67037 Strasbourg, France a Email: [email protected] b Email: [email protected] ∗ Corresponding author
Abstract:
The LHC discovery potential of heavy top partners decaying into a top quarkand a Z boson is studied in the trilepton channel at √ s = 13 TeV in the single productionmode. The clean multilepton final state allows to strongly reduce the background contam-inations and to reconstruct the T ′ mass. We show that a simple cut-and-count analysisprobes the parameter space of a simplified model as efficiently as a dedicated multivariateanalysis. The trilepton signature finally turns out to be as sensitive in the low T ′ massregion as the complementary channel with a fully hadronic top quark, and more sensitivein the large mass domain. The reinterpretation in terms of the top- Z -quark anomalouscoupling is shown. Keywords:
Top partner, trilepton, discovery potential, simplified model, MVA. ontents
1. Introduction 12. Simplified model 2
3. Analysis strategy 4
4. Conclusions 13A. BDT variables 15
1. Introduction
In 2012 the Run-I of the LHC at the center of mass energies of 7 and 8 TeV has finallydiscovered the long sought-after Higgs boson [1, 2]. In Spring 2015 the LHC will startagain to produce pp collisions in the so-called Run-II, at the increased energy of 13 TeV. Itis expected to accumulate 100 −
150 inverse femtobarn (fb − ) of data in the first two yearsand up to 300 fb − in the following ones. The primary scope of the Run-II is to furtherunderstand the newly discovered Higgs boson and to eventually make new discoveries. Inall generality, it is very common in beyond the standard model theories that new heavyfermions arise to stabilise the Higgs boson mass and to protect it from dangerous quadraticdivergences. In many cases, these new fermions are heavy partners of the third generationquarks with vector-like couplings. They are commonly predicted by many new physicsscenarios, including Extra Dimensions, Little Higgs Models, and Composite Higgs Mod-els [3, 4, 5, 6, 7, 8, 9, 10]. The observation of new heavy quarks thus plays an importantrole in the investigation of the Higgs sector. The common feature of these heavy quarks isto decay into a standard model quark and a W ± boson, a Z bosons, or a Higgs boson. The relative branching ratios are purely determined by the weak quantum numbers of themultiplet the new quark belongs to [11]. Here we will focus on the case of a singlet heavyquark: the top partner or T ′ . Experimental collaborations are now considering the inter-play between the various decay channels in their searches. Recent limits from ATLAS [12]and CMS [13] lie within 690-780 GeV, depending on the considered final state. Other possibilities like 3-body decays are also possible, but are not covered here. – 1 –hese searches however do not typically consider intergenerational mixing. Althoughvector-like quarks are usually assumed to only mix with the third generation followinghierarchy or naturalness arguments [14, 15, 16, 17, 18], the top partners can mix in asizable way with lighter quarks while remaining compatible with the current experimentalconstraints [19, 20]. This possibility has been recently pursued and must be consideredwith attention. The top partners interactions with the electroweak and Higgs bosons aregenerically allowed through arbitrary Yukawa couplings, implying that the branching ratiosinto light quarks can be possibly competitive with the top-quark one. Here we consideronly the case of mixing to the first generation of quarks.Beside opening up the decay channel into a standard model boson plus a light quark,the mixing with the first generation also enhances the single production, especially dueto the presence of valence quarks in the initial state. Even without mixing, the singleproduction cross sections at the upcoming LHC energies become competitive with the pairproduction ones. Furthermore, the T ′ → tZ final state is only recently being experimentallyinvestigated, both in the dilepton and in the trilepton channels by ATLAS [12]. Followingthe investigation first pursued at the 2013 Les Houches workshop [21], in this paper westudy the LHC discovery potential of the T ′ → tZ channel in the trilepton decay mode insingle production at √ s = 13 TeV, for a singlet T ′ quark mixing with the first generation.To capture all the essential features of the new heavy top quark while remaining as modelindependent as possible, the study here presented is performed in the framework of sim-plified models. We will employ the dedicated model for the heavy singlet top partner aspresented in ref. [20], comprising only 3 independent couplings.The paper is structured as follows. In section 2 the simplified model under considera-tion is described. In addition, we propose a convenient way to explore its parameter space.In section 3 we describe our analysis, comparing the discovery reach obtained in a simplecut-and-count approach to that obtained in a dedicated multivariate analysis. Results for L = 100 fb − are collected in section 3.3. The main result of this paper is that the simplecut-and-count analysis probes the T ′ parameter space as efficiently as the more sophisti-cated multivariate analysis. Also, a comparison to the complementary dilepton channel,recently presented in [22] for a T ′ coupling only to the third generation, is pursued. Thereinterpretation of our results to the case of the top-quark flavour-changing neutral cou-pling to a light quark and a Z boson, which shares the same final state as the T ′ discussedhere, is overviewed in section 3.4. Finally, we conclude in section 4.
2. Simplified model
A simple Lagrangian that parametrises the T ′ couplings to quarks and electroweak bosonis (showing only the couplings relevant for our analysis) [20] L T ′ = g ∗ (r R L R L g √ T ′ L/R W + µ γ µ d L/R ] + r
11 + R L g √ T ′ L/R W + µ γ µ b L/R ]+ (2.1)+ r R L R L g θ W [ T ′ L/R Z µ γ µ u L/R ] + r
11 + R L g θ W [ T ′ L/R Z µ γ µ t L/R ] ) + h.c. , – 2 – q/q ′ u/d Z/W ± T ′ Zt (a) A q q ′ b W ± T ′ Zt (b) A Figure 1: Feynman diagrams for the process pp → T ′ j → tZj via couplings of the T ′ to(a) first generation quarks and (b) third generation quarks.where the subscripts L and R label the chiralities of the fermions. Only 3 parameters aresufficient to fully describe the interactions that are relevant for our investigation. Besides M T ′ , the vector-like mass of the top partner, there are the 2 couplings appearing in eq. (2.1)- g ∗ , the coupling strength to SM quarks in units of standard couplings, which is onlyrelevant in single production. The cross sections for the latter scale with the couplingsquared;- R L , the generation mixing coupling, which describes the rate of decays to first gen-eration quarks with respect to the third generation, so that R L = 0 corresponds tocoupling to top and bottom quarks only, while the limit R L = ∞ represents couplingto first generation of quarks only.For some possible reinterpretation of this effective Lagrangian in terms of completemodels, see refs. [10, 22]. In this paper we study the LHC discovery potential of the T ′ in single production mode,in association with a light jet. Then, the T ′ decays into a top and a Z boson. The overallsignature reads: pp → T ′ j → tZj . This process is given by the set of Feynman diagramsdisplayed in figure1.The are two sets of diagrams, i.e. where the T ′ is produced due to the interaction withlight quarks ( A ) or due to the interaction with the b quark ( A ). From the Lagrangian ineq. (2.1), these 2 sets of diagrams give production cross sections that scale differently with R L , the mixing coupling. Further, the decay into a top quark and a Z boson scales with R L too. We parametrise the production cross section and branching ratio BR( T ′ → tZ ) asfollows: σ pp → T ′ ( M T ′ , R L ) = A ( M T ′ ) R L R L + A ( M T ′ ) 11 + R L , (2.2) BR T ′ → tZ ( M T ′ , R L ) = B ( M T ′ ) 11 + R L , (2.3)– 3 –here A i ( M T ′ ) ( i = 1 ,
3) and B ( M T ′ ) are numerical coefficients. In details, A i representsthe production cross section for the heavy quark due to interactions to partons belongingto the i th generation, while B is the T ′ branching ratio for its decay into a top quark and aZ boson. They have been evaluated at the LHC at √ s = 13 TeV for the CTEQ6L1
PDF [23]and collected in table 1, for the first and third generations only. M T ′ (GeV) A ( M T ′ ) (pb) A ( M T ′ ) (pb) B ( M T ′ ) (%)800 1.2614 0.07242 22.41000 0.7752 0.03518 23.51200 0.5001 0.01826 24.01400 0.3331 0.00994 24.21600 0.2265 0.00561 24.4Table 1: Numerical coefficients for 1 st and 3 rd production cross sections at the LHC at √ s = 13 TeV and T ′ branching ratios to tZ .These formulas will allow us to draw the LHC discovery power curves as a functionof R L in section 3.3 in a simple way. The product of eq. (2.2) and eq. (2.3) gives thecross section for pp → T ′ j → tZj as a function of R L in the narrow-width approximation.However, this also depends on the choice of the PDF, on the top mass, the EW couplings,etc. By considering the ratio of the cross sections evaluated at different values of R L (andall other input fixed) one can factorise the impact of the former. Such ratio can then beused to rescale a given cross section, evaluated at the user’s choice of the input parameters,as a function of R L . In our case, we computed the cross section σ ( M T ′ , .
5) of the variousbenchmark points for R L = 0 .
5, hence their cross section for an arbitrary value of R L isgiven by σ ( M T ′ , R L ) = σ ( M T ′ , . σ pp → T ′ ( M T ′ , R L ) BR T ′ → tZ ( M T ′ , R L ) σ pp → T ′ ( M T ′ , . BR T ′ → tZ ( M T ′ , . . (2.4)Furthermore, this rescaling can be applied anywhere in the cutflow. It is sufficient toreplace σ ( M T ′ , .
5) with the cross section after any cut, to obtain the corresponding crosssection at a different value of R L after the same cut.
3. Analysis strategy
All samples employed in this study have been generated in
MadGraph5 aMC@NLO v2.1.2 [24]with the
CTEQ6L1
PDF [23]. Events have subsequently been hadronised/parton showeredin
PYTHIA 6 [25] with tune Z2 [26]. Detector simulation is performed with a customisedversion of
Delphes 3 [27] to emulate the CMS detector. Jets have been reconstructed with
FastJet [28] employing the anti- k t algorithm [29] with parameter R = 0 . FeynRules [30,31]. We generate 5 benchmark points varying the T ′ mass in steps of 200 GeV in the range M T ′ ∈ [800; 1600] GeV, (3.1) Eq. (2.3) holds when all the T ′ decay products are much lighter than its mass, which is the case underconsideration. – 4 –ith g ∗ = 0 . R L = 0 .
5. We do not apply here any k -factor to the signal. Contraryto the backgrounds, tau leptons have not been included in the generation of the signalsamples.Backgrounds (B) that can give 3 leptons in the final state which are considered in thisanalysis are: tt and Z/W + jets with non-prompt leptons, and ttW , ttZ , tZj and V Z ( V = W, Z/γ ) with only genuinely prompt leptons. We generated leading order sampleswith up to 2 merged jets normalised to the (N)NLO cross section where available, takenfrom [24, 32]. For both signal and background, a suitable number of unweighted event isgenerated. We checked that the statistical uncertainties at any point of our analyses arebelow 1%, and for this reason they will not be quoted.We do not simulate multijet backgrounds, which can be reliably estimated only fromdata. Further, we do not consider jets faking electrons, since this is a feature which is notsupported in
Delphes . It is however seen to be negligible in multileptons analyses (see forinstance, ref. [33]).The analysis is carried out in
MadAnalysis 5 [34, 35]. Leptons ( ℓ = e, µ ) and jets areidentified if passing the following criteria: p T ( ℓ ) >
20 GeV, | η ( e/µ ) | < . / . , (3.2) p T ( j ) >
40 GeV, | η ( j ) | < , (3.3)∆ R ( ℓ, j ) > . b -tagging and for lepton isolation have been implemented. Re-garding the former, here we adopted the medium working point [36], which has an average b -tagging rate of 70% and a light mistag rate of 1%. To apply the b -tagging, we consid-ered jets within the tracker only, i.e. with | η ( j ) | < .
4. This means that the b -taggingprobability for jets with larger pseudorapidities is vanishing. For the latter, the combinedtracker-calorimetric isolation is used to identify isolated leptons. The relative isolation I rel is defined as the sum of the p T and calorimetric deposits of all tracks within a cone ofradius ∆ R = 0 .
3, divided by the p T of the lepton. The latter is isolated if I rel ≤ .
10. Thischoice has been taken as a compromise to strongly reduce backgrounds with non-promptleptons without suppressing the signal, where the two leptons coming from the Z bosonget closer and closer as the top partner mass increases.After the object reconstruction and selection, we apply some general preselections asfollows: we require at least 1 jet and no more than 3, of which exactly one is b-tagged, andexactly 3 leptons (electrons or muons). The requirement of less than 3 jets removes the T ′ pair production isolating the single production channel.Efficiencies and event yields are evaluated for L = 100 fb − and are collected in table 2.The requirement of 3 isolated leptons strongly reduces the tt + X backgrounds, especiallythe tt + jets one. The diboson component is instead suppressed by the b -tagging. Regardingthe signal, the requirement of 3 isolated leptons is less efficient as the T ′ mass increases.This is because the 2 leptons stemming from the Z boson gets closer to each other as the T ′ gets heavier, due to the larger boost of the Z boson in the T ′ → t Z decay.– 5 –ackground no cuts 1 ≤ n j ≤ n ℓ ≡ n b ≡ tt (+ X ) 7 . (100%) 6 . (81 . . . tZj . . . W Z . (100%) 5 . (41 . . . . (100%) 6 . (80 . . . M T ′ (GeV) no cuts 1 ≤ n j ≤ n ℓ ≡ n b ≡ . . . . . . . . . . . . . . . g ∗ = 0 . R L = 0 . Z boson massis chosen, and a cut around their invariant mass distribution is performed, | M ( ℓ + ℓ − ) − M Z | <
15 GeV . (3.5)The lepton from the top decay is therefore identified as the remaining one in our trileptonchannel and labelled ℓ W .We describe in the following the 2 analyses we performed, that differentiate from thispoint on. The first one is a traditional cut-and-count strategy, where subsequent cuts areapplied to the most important kinematic variables to maximise the signal-over-backgroundratio. The second one is a multivariate analysis (MVA), where several discriminatingobservables are used at once to distinguish the signal from the background, cutting at theend only on its output. The first strategy to study the LHC discovery potential illustrated here is the cut-and-countone (C&C). As for the previous reconstruction of the Z boson from its decay products,one could also reconstruct both the top quark and the W boson stemming from its decay.The presence of only one source of missing energy (one neutrino) allows to reconstruct thefour momentum of the latter by imposing suitable kinematical constraints. Hence one canreconstruct the W boson and the top quark as resonances in the invariant mass as well asin the transverse mass distributions of the decay products. We chose to use the latter, inthe following formulation [37]: m T = (cid:18)q M ( vis ) + P T ( vis ) + | /P T | (cid:19) − (cid:16) ~P T ( vis ) + /~P T (cid:17) , (3.6)because of the sharper peaks as compared to those in the invariant mass distributions.Instead to draw suitable window mass cuts, we decided to only apply loose selections,– 6 –hich maximise their efficiencies while retaining most of the signal. The cuts we appliedare as follows: 10 < M T ( ℓ W ) / GeV < , (3.7)0 < M T ( ℓ W b ) / GeV < . (3.8)In particular, the lower cut in eq. (3.7) is inspired by experimental analyses to suppressthe multijet background, which we did not simulate.Surviving events and relative efficiencies are collected in table 3. The numerical valuesof the cuts appearing in eqs. (3.5)–(3.8) have been chosen to maximise the signal-over-background ratio while keeping at least 90% of the signal.Background n b ≡ tt (+ X ) 243.8 (47 . . . . tZj . . . . W Z . . . . . . . . M T ′ (GeV) n b ≡ . . . . . . . . . . . . . . . . . . . . g ∗ = 0 . R L = 0 . T ′ decay products (the 3 charged leptons and the b -jet), as can been seen infigure 2 for the various signal benchmark points.We select a window around the peak of each benchmark point to get the best signal-over-background significance and we collect the final numbers in table 5. In section 3.1 we showed that suitable cuts on the most straightforward distributions weresufficient to isolate the signal from the background. We made use of the signal topology,which has a Z boson decaying leptonically, and a top-quark decaying into a b -jet and intoa leptonic W -boson. The presence of the intermediate T ′ was then seen as a peak in thetransverse mass of all its visible decay products. One could wonder if this was the beststrategy, i.e. cutting on those variables with the values we chose. There are in fact manyadditional variables that one could analyse to distinguish the signal from the background.However, cutting on any of these variables will unavoidably reduce also the signal. To– 7 – b3l) (GeV) T M200 400 600 800 1000 1200 1400 1600 1800 ) - E ve n t s ( L = f b -2 -1 WZTT (+X)TZj = 0.8 TeV T’ M = 1.0 TeV T’ M = 1.2 TeV T’ M = 1.4 TeV T’ M = 1.6 TeV T’ M = 13 TeVs Figure 2: Transverse mass distribution for the T ′ decay products: the 3 charged leptonsand the b -jet.overcome this, several variables can be combined using a multivariate analysis (MVA) toobtain the best signal/background discrimination [38]. We identified some discriminatingvariables in table 4, ranked according to their discriminating power when a boosted decisiontree (BDT) is employed. Here, ∆ ϕ is the difference of the azimuthal angles between 2objects, ∆ η is the difference of their pseudorapidities, and ∆ R = p (∆ ϕ ) + (∆ η ) .Variable Importance Variable Importance M T ( b ℓ ) 2 .
60 10 − ∆ R ( b, ℓ W ) 9 .
77 10 − p T ( Z ) /M T ( b ℓ ) 9 .
41 10 − ∆ ϕ ( t, Z ) 8 .
17 10 − η max( j ) 6 .
02 10 − ∆ ϕ ( ℓℓ | Z ) 5 .
89 10 − ∆ ϕ ( Z, /p T ) 5 .
37 10 − p T ( j ) /M T ( b ℓ ) 5 .
08 10 − ∆ η ( ℓℓ | Z ) 5 .
05 10 − ∆ η ( b, ℓ W ) 5 .
03 10 − η ( t ) 4 .
99 10 − ∆ ϕ ( Z, ℓ W ) 4 .
63 10 − η ( Z ) 4 .
61 10 − Table 4: Ranking training variables for M T ′ = 1 . ℓℓ | Z identifies the 2 leptons that reconstruct the Z boson.To define some of the angular variables, the whole four-momentum of the neutrino– 8 –temming from the semileptonic top-quark decay has been reconstructed as describedabove. Furthermore, we did not include the top mass (neither as invariant mass noras transverse mass of its decay products) in the MVA because it did not show a strongdiscriminating power. This is understandable because signal and the largest sources ofbackground both have a top quark in their intermediate states. Finally, the presence ofa forward jet is a prominent feature of the signal. To account for this, we use the largestpseudorapidity of all jets η max( j ) in the event.It is interesting to notice that there are few variables which behaviour is directlyproportional to the T ′ mass, like the p T of the leading jet or the p T of the 2 leptonsreconstructing the Z boson. These correlations are efficiently removed if one considerratios of the p T ’s over M T ( b ℓ ), which in fact decorrelate them. We checked that thehighest sensitivity is reached when the ratio of said variables to the T ′ reconstructed mass( M T ( b ℓ )) is considered instead of the actual observables. All other variables are almostuncorrelated, with a degree of correlation of ±
30% at most.The variables in table 4 are used to train the BDT to recognise the signal against thebackground. They are selected after the Z mass reconstruction, i.e. after applying eq. (3.5).A pictorial representation of these variables is in the Appendix, for the M T ′ = 1 . R L = 0 . g ∗ = 0 . R L = 0 . g ∗ – R L plane, without the need to retrain the BDT any further. However, by controlling theshare of the T ′ coupling between first and third generation quarks, R L changes the ratio ofincoming sea/valence quarks, which ultimately alters the process kinematics. We directlychecked that the loss in performance, as compared to the results obtained with the suitabletraining, is at most of O (10%) when R L goes to zero, thereby validating our extrapolatingprocedure. The very same check is carried out for the cut-and-count analysis, with asimilar performance behaviour.Finally, the reader could wonder if the ratio of variables as described above might letthe training procedure be less M T ′ -dependent. It is in fact the case, but still it is notpossible to use a universal MVA training and apply it to all the various T ′ samples. Thisis because, as clear from table 4, still M T ( b ℓ ) is by far the best discriminating variable. Ifwe remove it from the training, this becomes less M T ′ -dependent for large T ′ masses, butthe overall performance of the MVA is even lower than previously. This let us conclude Notice that the typical precision that one can aim at with a fast simulation is O (30%). – 9 –hat it is not possible to create an efficient M T ′ -independent training scheme. We collect here the final results for the discovery power at the LHC. In the case of the cut-and-count analysis of section 3.1, we need to select a window around the signal peaks in the M T ( b ℓ ) distribution. For the MVA of section 3.2, we need to perform a cut on the BDToutput that maximises the significance. The maximum significance for the benchmarkpoints, evaluated as σ = S/ √ S + B after selecting a window around the mass peak orcutting on the BDT output, are collected in table 5. Analysis M T ′ = 0 . M T ′ = 1 . M T ′ = 1 . M T ′ = 1 . M T ′ = 1 . M T ( b ℓ ) cut (GeV) [800 − − − − − σ σ Table 5: Signal and background events and maximum significance for the benchmark pointsfor L = 100 fb − , after selecting a mass window (for the C&C), or after cutting on theBDT output (MVA).One of the most important result in this paper is that the dedicated BDT analysisdoes not significantly improve on the cut-and-count strategy, as clear from table 5. Thelatter analysis is certainly sufficient and easier. The cuts as displayed in eqs. (3.7)–(3.8)are already best optimised, as is the signal peak selection. No further variable/cut need tobe considered/applied.The significances in table 5 are for the benchmark points, evaluated for g ∗ = 0 . R L = 0 .
5. We can now extrapolate them to the full g ∗ – R L parameter space using eq. (2.4).The extrapolation is done by rescaling the cross section and every time reoptimising thecuts to get the highest significance. The 3 and 5 sigma discovery lines are drawn as afunction of g ∗ and the T ′ mass for some fixed values of R L in figure 3(left), and as afunction of g ∗ and R L for the benchmark T ′ masses in figure 3(right). figure 3 shows thatwith 100 fb − of data, T ′ masses up to 2 TeV can be observed, depending on the valuesof the couplings. The cross section for the trilepton decay channel of the T ′ (and hencethe LHC reach) increases considerably when R L is non-vanishing, getting to a maximumfor R L ≃
1, corresponding to 50%–50% mixing. This effect is simply due to the increasedadmixture of valence quarks in production, mitigated by a reduced T ′ -to- tZ branchingratio, as R L increases.The reach in g ∗ is here roughly twice than for the no mixing case ( R L = 0). Then,for larger values of R L , g ∗ needs to slightly increase to compensate for the decrease incross section due to the larger mixing with the first generation quarks, that suppress the T ′ → tZ branching ratio.The discovery power of the trilepton channel can be compared to the one of the dileptonchannel as studied in [22]. The R L = 0 line in figure 3 is the one considered therein.– 10 – (GeV) T’ M800 1000 1200 1400 1600 1800 2000 2200 g * T’ g* vs M =0 (0%) L R =1.0 (50%) L R =3.0 (75%) L R =9.0 (90%) L R T’ g* vs M L R0 1 2 3 4 5 6 7 8 9 g * L g* vs R =1.6 TeV T’ M =1.4 TeV T’ M =1.2 TeV T’ M =1.0 TeV T’ M =0.8 TeV T’ M L g* vs R Figure 3: Significance σ = 3 (solid lines) and σ = 5 (dashed lines) for L = 100 fb − .However, to be able to draw a meaningful comparison, we shall set ourselves in the sameconditions, which correspond to the end of the LHC run-II. The plot for this setup isin figure 4. The curve to be compared is the R L = 0 one on the left-hand side plot. Atlow T ′ masses, the dileptonic channel of ref. [22] performs slightly better, meaning that amarginally lower value of g ∗ can be probed. At larger T ′ masses though the trileptonicchannel is more sensitive, extending the reach by 200 −
300 GeV. In these conditions, ouranalysis is sensitive to g ∗ couplings down to 0 .
05 and T ′ masses up to 2 . (GeV) T’ M800 1000 1200 1400 1600 1800 2000 2200 g * T’ g* vs M =0 (0%) L R =1.0 (50%) L R =3.0 (75%) L R =9.0 (10%) L R T’ g* vs M L R0 1 2 3 4 5 6 7 8 9 g * L g* vs R =1.6 TeV T’ M =1.4 TeV T’ M =1.2 TeV T’ M =1.0 TeV T’ M =0.8 TeV T’ M L g* vs R Figure 4: Significance σ = 3 (solid lines) and σ = 5 (dashed lines) for L = 300 fb − and k f = 1 .
14 as in ref. [22]. ref. [22] used an integrated luminosity of 300 fb − and rescaled the signal by a mean k -factor of 1 . – 11 –ut κ tZu κ tZc no cuts 2263(100%) 5360(100%)1 ≤ n j ≤ n ℓ ≡ n b ≡ M T ( b ℓ ) >
400 GeV >
200 GeVS 68.0 304.5B 102.9 241.7 σ κ tZu / Λ (at current limit) and κ tZc / Λ (forBR( t → Zc ) = 1%), and signal/background events that maximise the significance. We conclude this paper by presenting a reinterpretation of our investigation in terms ofthe top-quark FCNC coupling to a light quark and a Z boson. In this scenario, the topquark interacts with a Z boson and a up- or charm-quark via the κ tZq FCNC coupling [39] L = X q = u,c g √ c W κ tZq Λ tσ µν (cid:0) f LZq P L + f RZq P R (cid:1) q Z µν , (3.9)where Λ is the scale of new physics. The Lagrangian in eq. (3.9) gives a similar final stateas the one subject of this paper, pp → tZ , with a top-quark and a Z boson produced back-to-back. The only difference with the T ′ -induced topology is the absence of the forwardjet at leading order. The analyses of the T ′ -mediated signature subject of this paper couldtherefore be as well sensitive to the one induced by the top effective coupling. We tested itby producing at leading order a pp → tZ sample when turning on one FCNC coupling atthe time. We label them κ tZu and κ tZc , respectively. The samples have been produced asdescribed in section 3. Finally, they have been analysed at detector level by running themon the cut-and-count analysis of section 3.1. Efficiencies and event yields for 100 fb − arecollected in table 6.The M T ( b ℓ ) distribution after the application of the cuts of eqs. (3.5)–(3.8) is shownin figure 5. The significance for the κ tZu sample is maximised by selecting M T ( b ℓ ) >
400 GeV, reaching the value of 5 . κ tZu / Λ = 0 . − (or BR( t → Zu ) = 0 . κ tZc sample, we chose a coupling yielding BR( t → Zc ) = 1%to compare the results. For this value, the highest significance of 13 . σ is obtained byselecting M T ( b ℓ ) >
200 GeV. Notice that the only available limit to this coupling is suchthat BR( t → Zc ) ≤ .
4% [41]. For comparison, we applied the MVA trained on each T ′ signal to the FCNC case. Also in this case however it did not improve the sensitivity.– 12 – b3l) [GeV] T M200 400 600 800 1000 1200 1400 1600 1800 ) - E ve n t s ( L = f b -2 -1 WZTT (+X)TZj -BR=1% zct
K -lim zut K = 13 TeVs Figure 5: M T ( b ℓ ) distribution with the present best limit on the top-Z-up FCNC couplingand for BR( t → Zc ) = 1% for the top-Z-c one.Finally, the cut-and-count discovery power curves at the LHC as a function of the topFCNC couplings for 100 fb − are shown in figure 6. Contours for a significance σ = 2 , , κ tZu / Λ( κ tZc / Λ) couplings leading to BRs above 0 . . − . In case of non observation, 95% C.L. limits can be put for BRs down to 0 . .
1% for the two couplings, respectively.
4. Conclusions
In this work we described the LHC run-II discovery potential of the trilepton channel fora singlet top partner in the single production mode and its subsequent decay into a topquark and a Z boson. A simple cut-and-count analysis has been designed, by selectingand cutting the most straightforward distributions. A suitable multivariate analysis didnot improve significantly on the cut-and-count results. The comparison was performed onseveral signal benchmark points. Further, we proposed a simple way to extend our resultsto the whole parameter space of a simplified model.Overall, a search at the LHC in the trilepton channel can be sensitive to top partnersdecaying into tZ for masses up to 2 . .
1) TeV and couplings down to 0 . .
05) with– 13 –
R(t->Zu) (%)0.01 0.02 0.03 0.04 0.05 0.06 0.07 B R ( t - > Z c ) ( % ) BRs
BRs
Figure 6: Significance as a function of the top FCNC couplings for the cut-and-countanalysis, and σ = 2(white), σ = 3(red), and σ = 5(black) contour lines.100(300) fb − of data. We compared to the reach in the dilepton channel of ref. [22],concluding that the trilepton mode can extend the former by 200 −
300 GeV in T ′ massesfor suitable values of the couplings. Finally, we reinterpreted our analyses in the context ofa top FCNC coupling to a Z boson and a light quark, which provides a similar final state.We showed that this channel can discover at 5 σ values of the couplings at the present bestexclusion limit (for 100 fb − ), probe at 3 σ FCNC branching ratios down to 0 . . κ tZu / Λ( κ tZc / Λ), or eventually extend the exclusion limits down to 0 . . Acknowledgements
We would like to sincerely thank our colleagues A. Alloul, C. Collard, E. Conte and G. Ham-mad for the help in generating the background samples used in this work and for usefulcomments. The work of LB is supported by the Theorie-LHC France initiative of theCNRS/IN2P3 and by the French ANR 12 JS05 002 01 BATS@LHC.– 14 – . BDT variables
We present here the training variables for the M T ′ = 1 . R L = 0 . (b3l) (GeV) T M500 1000 1500 2000 2500 / N d N Background = 1.0 TeV T’ M ) W R(b l ∆ / N d N T (Z)/M T p0.2 0.4 0.6 0.8 1 1.2 1.4 / N d N φ ∆ -6 -4 -2 0 2 4 6 / N d N max η -4 -2 0 2 4 / N d N Z (ll φ ∆ / N d N Figure 7: BDT training variables (1). (Z MET) φ ∆ / N d N T )/M (j T p0.2 0.4 0.6 0.8 1 1.2 1.4 / N d N Background = 1.0 TeV T’ M ) Z (ll η ∆ -3 -2 -1 0 1 2 3 / N d N W (b l η ∆ -4 -3 -2 -1 0 1 2 3 4 / N d N η -6 -4 -2 0 2 4 6 8 / N d N W (Z l φ ∆ / N d N η -4 -2 0 2 4 6 / N d N Background = 1.0 TeV T’ M Figure 8: BDT training variables (2).– 15 –
DT response -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 d x / ( / N ) d N Signal (test sample)Background (test sample) Signal (training sample)Background (training sample)
Kolmogorov-Smirnov test: signal (background) probability = 0.921 (0.336) U / O -f l o w ( S , B ): ( . , . ) % / ( . , . ) % TMVA overtraining check for classifier: BDT
Figure 9: BDT overtraining.
References [1] CMS Collaboration, S. Chatrchyan et al., Phys.Lett.
B716 , 30 (2012), arXiv:1207.7235.[2] ATLAS Collaboration, G. Aad et al., Phys.Lett.
B716 , 1 (2012), arXiv:1207.7214.[3] R. Contino, L. Da Rold, and A. Pomarol, Phys.Rev.
D75 , 055014 (2007),arXiv:hep-ph/0612048.[4] R. Contino, T. Kramer, M. Son, and R. Sundrum, JHEP , 074 (2007),arXiv:hep-ph/0612180.[5] R. Contino and G. Servant, JHEP , 026 (2008), arXiv:0801.1679.[6] J. Cabarcas, D. Gomez Dumm, and R. Martinez, Eur.Phys.J.
C58 , 569 (2008),arXiv:0809.0821.[7] C. Anastasiou, E. Furlan, and J. Santiago, Phys.Rev.
D79 , 075003 (2009), arXiv:0901.2117.[8] N. Vignaroli, Phys.Rev.
D86 , 075017 (2012), arXiv:1207.0830.[9] O. Matsedonskyi, G. Panico, and A. Wulzer, JHEP , 164 (2013), arXiv:1204.6333.[10] S. A. Ellis, R. M. Godbole, S. Gopalakrishna, and J. D. Wells, JHEP , 130 (2014),arXiv:1404.4398.[11] F. del Aguila, M. Perez-Victoria, and J. Santiago, JHEP , 011 (2000),arXiv:hep-ph/0007316.[12] ATLAS Collaboration, G. Aad et al., (2014), arXiv:1409.5500.[13] CMS Collaboration, S. Chatrchyan et al., Phys.Lett.
B729 , 149 (2014), arXiv:1311.7667. – 16 –
14] J. Berger, J. Hubisz, and M. Perelstein, JHEP , 016 (2012), arXiv:1205.0013.[15] A. De Simone, O. Matsedonskyi, R. Rattazzi, and A. Wulzer, JHEP , 004 (2013),arXiv:1211.5663.[16] J. Aguilar-Saavedra, EPJ Web Conf. , 16012 (2013), arXiv:1306.4432.[17] J. Aguilar-Saavedra, R. Benbrik, S. Heinemeyer, and M. Perez-Victoria, Phys.Rev. D88 ,094010 (2013), arXiv:1306.0572.[18] C. Han, A. Kobakhidze, N. Liu, L. Wu, and B. Yang, (2014), arXiv:1405.1498.[19] G. Cacciapaglia et al., JHEP , 070 (2012), arXiv:1108.6329.[20] M. Buchkremer, G. Cacciapaglia, A. Deandrea, and L. Panizzi, Nucl.Phys.
B876 , 376(2013), arXiv:1305.4172.[21] G. Brooijmans et al., (2014), arXiv:1405.1617.[22] J. Reuter and M. Tonini, JHEP , 088 (2015), arXiv:1409.6962.[23] J. Pumplin et al., JHEP , 012 (2002), arXiv:hep-ph/0201195.[24] J. Alwall et al., JHEP , 079 (2014), arXiv:1405.0301.[25] T. Sjostrand, S. Mrenna, and P. Z. Skands, JHEP , 026 (2006), arXiv:hep-ph/0603175.[26] R. Field, Acta Phys.Polon.
B42 , 2631 (2011), arXiv:1110.5530.[27] DELPHES 3, J. de Favereau et al., JHEP , 057 (2014), arXiv:1307.6346.[28] M. Cacciari, G. P. Salam, and G. Soyez, Eur.Phys.J.
C72 , 1896 (2012), arXiv:1111.6097.[29] M. Cacciari, G. P. Salam, and G. Soyez, JHEP , 063 (2008), arXiv:0802.1189.[30] A. Alloul, N. D. Christensen, C. Degrande, C. Duhr, and B. Fuks, Comput.Phys.Commun. , 2250 (2014), arXiv:1310.1921.[31] http://feynrules.irmp.ucl.ac.be/wiki/VLQ tsingletvl.[32] M. Czakon, P. Fiedler, and A. Mitov, Phys.Rev.Lett. , 252004 (2013), arXiv:1303.6254.[33] CMS Collaboration, V. Khachatryan et al., Eur.Phys.J.
C74 , 3060 (2014), arXiv:1406.7830.[34] E. Conte, B. Fuks, and G. Serret, Comput.Phys.Commun. , 222 (2013), arXiv:1206.1599.[35] E. Conte, B. Dumont, B. Fuks, and C. Wymant, Eur.Phys.J.
C74 , 3103 (2014),arXiv:1405.3982.[36] CMS Collaboration, S. Chatrchyan et al., JINST , P04013 (2013), arXiv:1211.4462.[37] V. D. Barger, T. Han, and R. Phillips, Phys.Rev. D36 , 295 (1987).[38] A. Hoecker et al., PoS
ACAT , 040 (2007), arXiv:physics/0703039.[39] J. Aguilar-Saavedra, Nucl.Phys.
B812 , 181 (2009), arXiv:0811.3842.[40] CMS Collaboration, S. Chatrchyan et al., Phys.Rev.Lett. , 171802 (2014),arXiv:1312.4194.[41] CMS Collaboration, CERN Report No. CMS-PAS-TOP-12-021, 2013 (unpublished)., 171802 (2014),arXiv:1312.4194.[41] CMS Collaboration, CERN Report No. CMS-PAS-TOP-12-021, 2013 (unpublished).