Search for invisible Higgs boson decays in vector boson fusion at √

(1)

Search for invisible Higgs boson decays in vector boson fusion at √

s = 13 TeV with the ATLAS detector

.TheATLASCollaboration

a r t i c l e i n f o a b s t ra c t

Articlehistory:

Received18September2018

Receivedinrevisedform12February2019 Accepted9April2019

Availableonline15April2019 Editor:M.Doser

We report a search for Higgs bosons that are produced via vector boson fusion and subsequently decayintoinvisible particles.The experimentalsignature is anenergetic jetpairwith invariantmass ofO(1)TeV andO(100)GeV missingtransversemomentum.Theanalysisuses36.1 fb⁻¹ofppcollision dataat√

s=¹³TeV recordedbytheATLASdetectorattheLHC.Inthesignalregionthe2252 observed eventsareconsistentwiththebackgroundestimation.Assuminga125GeV scalarparticlewithStandard Modelcrosssections,theupper limitonthebranchingfractionoftheHiggsbosondecayintoinvisible particlesis0.37 at95% conﬁdencelevelwhere0.28 wasexpected.ThislimitisinterpretedinHiggsportal modelstosetboundsonthewimp–nucleonscatteringcrosssection.Wealsoconsiderinvisibledecaysof additionalscalarbosonswithmassesupto3TeV forwhichtheupperlimitsonthecrosssectiontimes branchingfractionareintherangeof0.3–1.7pb.

©²⁰¹⁹^TheÂuthor(s).^Published^byÊlsevier^B.V.^Thisîsânôpenâccessârticleûnder^the^CC^BY^license (http://creativecommons.org/licenses/by/4.0/).FundedbySCOAP³.

1. Introduction

We presenta search forthe decays of the Higgs boson [1,2], produced via the vector boson fusion (VBF) process [3,4], into invisible particles (χχ_¯) with an anomalous and sizable O(10)% branching fraction. The hypothesis underconsideration [5–16] is thattheHiggsbosonmightdecayintoapairofweaklyinteracting massiveparticles(wimp)[17,18],whichmayexplainthenatureof darkmatter (seeRef. [19] andthereferencestherein).The search carriedout forthe125GeV Higgsbosonisrepeatedforhypothet- ical scalars with masses up to 3TeV. The search is independent onthedecayofthemediatorbecausethefinal stateparticles are invisibletothedetector,whileit isdependentonits E^miss_T distri- bution (defined below) because that quantity is reflective of the mediator’s p_Tdistribution.

The data sample corresponds to an integrated luminosity of 36.1fb⁻¹ ofproton-proton(pp)collisionsat√

s=¹³TeV recorded by the ATLAS detector at the LHC in 2015 and 2016. The ex- perimental signature of the VBF production process is a pair of energetic quark jetswith a wide gap in pseudorapidity (η) cor- responding to the O(1)TeV value of the invariant mass (mj j) of thehighest-pTjetsintheevent.¹ Thesignatureforthedecaypro-

E-mailaddress:atlas.publications@cern.ch.

1 ATLASusesaright-handedcoordinatesystemwithitsoriginatthe nominal interactionpointinthecenterofthedetectorandthez-axisalongthebeamdi- rection.Thex-axispointsfromtheinteractionpointtothecenteroftheLHCring;

they-axispointsupward.Cylindricalcoordinates(r,φ)areusedinthetransverse plane,whereφistheazimuthalanglearoundthez-axis.Thepseudorapidityisde- ﬁnedasη= −^ln(tan(θ/2)),whereθisthepolarangle.

cessistheO(100)GeV valueofthemissingtransversemomentum E^miss_T

thatcorrespondstotheHiggsbosonp_T.TheVBFtopology offersapowerfulrejectionofthestronglyproduced² backgrounds due to single vector boson plus two jets, and the multijetback- groundproduced fromQCD processes. In thisanalysis, the Higgs productionviathegluonfusionmechanismissubdominanttoVBF andisconsideredaspartofthesignal.

DirectsearchesforinvisibleHiggsdecayslookforan excessof eventsoverStandardModelexpectations.Theabsenceofanexcess isinterpretedasanupperlimitonthebranchingfractionofinvis- ible decays(Binv) assuming theStandard Modelproduction cross section [20] of the125GeV Higgs boson.Other published results have targeted a variety of production mechanisms—gluon fusion, VBF,W or Z associatedproduction[21–25]—tosetupperlimitson Binv.Thebestlimitsarefromthestatisticalcombinationofsearch results for which ATLAS reports an observed (expected) limit of 0.26 (0.17) [26] and CMS reports 0.26 (0.20) [27] at 95% conﬁ- dencelevel(CL).Forthesecombinationsthesingleinputwiththe highestexpectedsensitivityisVBF, thechannelpursued here.For theVBFchannelusingRun-1 data,ATLAS reports0.28 (0.31)[28]

andCMSreports0.43 (0.31)[29]. Inamorerecentupdateofthe VBFchannelusingRun-2 data,ATLASreports0.37 (0.28)[thispa- per]CMSreports0.33 (0.25)[27].

Global ﬁts to the measurements of visible decay channels of the Higgsbosonplace indirectconstraints onthe beyond-the-SM

2 FortheW and Zbackgroundprocessesinthispaper,electroweak(EW)refers todiagramsthatareofO(αew⁴ )orgreater,whilestrongreferstodiagramsthatare ofO(α²s)orgreateraccompaniedbyO(αew² ).

https://doi.org/10.1016/j.physletb.2019.04.024

0370-2693/©²⁰¹⁹^TheÂuthor(s).^Published^byÊlsevier^B.V.^Thisîsânôpenâccessârticleûnder^the^CC^BY^license(http://creativecommons.org/licenses/by/4.0/).Fundedby SCOAP³.

(2)

decaybranching fractionB^bsm^. ^The B^bsm îs^the ^sum ôf Binv that representsinvisibledecaysandBundet thatrepresentsthechannels thatareundetected,i.e.,thosethatarenotincludedinthefollow- ing combination. ForB^bsm using Run-1 data, ATLAS reports 0.49 (0.48)[30] andCMSreports0.57 (0.52)[31] withsimilar butnot identical assumptions. A combination of ATLAS and CMS results usingRun-1 datagives0.34 (0.39) [32]. Ina morerecentupdate usingRun-2data,CMSreportsanobservedlimitonBundet of0.38 [33].As notedinRef. [28],thereiscomplementarity betweenthe directsearchforinvisibleHiggsdecaysandtheindirectconstraints fromtheglobalfits.

In this analysis, several changes and improvements are made withrespectto the previous ATLASpaper onthistopic[28].The trigger and hadronic objects are deﬁned considering the simul- taneous pp collisions in the same and nearby bunch crossings (pileup)(Section2).Theleading backgroundsare simulatedusing state-of-the-art QCD predictions (Section 3). The eventselections arechangedtoretainagoodsensitivitydespitethehigherpileup.

Theanalysisextractsthesignal yieldusingabinnedlikelihoodﬁt to the mj j spectrum in 3 bins to increase the signal sensitivity (Section4). Theestimation ofthe importantanddominantback- groundfor the Z→νν process (denoted Zνν) relies only onthe Z_eeandZμμcontrolsamples,andisnotaffectedbytheoreticalun- certaintiesoftheW-to-Z extrapolation(Section5).Thesystematic uncertaintiesareevaluatedseparatelyforeachmj j bin(Section6).

Thesearch isrepeatedforother scalars withmassesup to 3TeV, whichcaneasilybereinterpretedformodelsnotconsideredinthis Letter(Section7).Severalaspectsoftheanalysishavenotchanged compared tothe ATLAS Run-1 analysis—e.g., subdetectordescrip- tions, transferfactormethod, Higgsportal models—and detailsof thesemaybefoundinRef. [28].

2. Detector,trigger,andanalysisobjects

ATLAS is a multipurpose particle physics detector with a forward–backwardsymmetriccylindricalgeometryconsistingofa trackingsystem,electromagneticandhadroniccalorimeters,anda muonsystem[34].

Thetrigger torecord eventsinthesample containing theVBF signalcandidatesusedatwo-levelE^miss_T algorithmwiththresholds adjusted throughoutthe data-taking periodto cope withvarying levelsofpileup[35,36].Thelevel-₁systemusedcoarse-granularity analog sums ofthe energy depositsin the calorimetertowers to require E^miss_T >50GeV. The second-level highleveltrigger sys- tem[37] usedjetsthatare reconstructedfromcalibratedclusters of cell energies [38] and requires E^miss_T >70–110GeV depending ontheluminosityandthepileuplevel.Thetriggereﬃciency[39]

forsignal events is98% for E^miss_T >180GeV when comparingthe triggerselectionwiththeoﬄine E^miss_T deﬁnitionthatcontainsad- ditionalcorrections.

Thetriggerstorecordthecontrolsamplesforbackgroundstud- iesusedleptonandjetalgorithms[40].Thesampleswithleptonic W and Z decayswere collected witha single-electronor-muon triggerwithpT>24GeV (26GeV)andan isolationrequirementin 2015(towardstheendof2016).Thesampleofmultijeteventswas collectedusingasetoflow-thresholdsingle-jettriggerswithlarge prescalevaluestokeeptheeventraterelativelylow.

For each event, a vertex is reconstructed from two or more associated tracks (t) with pT>400MeV. If multiple vertices are present, we consider the one with the largest

t(pT,t)² as the primaryvertexofourcandidates.

Leptons ( =ê,μ) are identified to help characterize events withleptonic finalstatesfromdecaysof vectorbosons. Sincethe signal process contains no leptons, such events are used forthe backgroundestimation,whichisdescribed inSection 5.Electrons

(muons)musthavepT>7GeV,|η|<2.47 (2.5),andsatisfyaniso- lationrequirement. Electronsare reconstructedby matchingclus- teredenergydepositsintheelectromagneticcalorimetertotracks fromtheinnerdetector[41,42] andmuonsbymatchinginnerde- tector and muonspectrometer tracks [43]. Forelectrons (muons) withap_T valueofatleast30GeV (20-100GeV),thereconstruction eﬃciency 80% (96%) witha rejectionfactorof around 500 (600).

Allleptonsmustoriginatefromtheprimaryvertex.

Jetsarereconstructedfromtopologicalclustersinthecalorime- ters using the anti-kt algorithm [44] with a radius parameter R=⁰.4. Jets must have pT>20GeV and |η_|<4.5. The subset of jetswithpT<60GeV and|η_|<2.4 arejetvertextagged(jvt) [45]

tosuppresspileupeffects,usingtrackingandvertexing.Thejvtis 92% eﬃcientforthejetsinthesignalprocessfromtheprimaryin- teractionwitharejectionfactorofaround100 forpileupjetswith pT valueintherangeof20-50GeV [45].

Cleaningrequirementshelpsuppressnon-collisionbackgrounds [46].Fakejetsduetonoisycells areremovedbyrequiringagood ﬁt to the expected pulse shape for each constituent calorimeter cell.Fakejetsinducedbybeam-halointeractionswiththeLHCcol- limatorsareremovedbyrequirementsontheirenergydistribution andthefractionoftheirconstituenttracksthatoriginatefromthe primaryvertex.

Ineventswithidentifiedleptons,anoverlapremovalprocedure is applied toresolve the ambiguitiesin caseswhere ajet is also identified in the same η-φ area, which could occur in situations such ashavingaheavy-flavorhadrondecaywithinajet[47].The lepton–jet overlapinR distance³ isresolved sequentiallyasfol- lows.IfanelectronisnearajetwithR<0.2,thejetisremoved to avoidthedoublecountingof electronenergydeposits.Ifa remaining jet is nearan electron with0.2≤R<0.4,the electron is removed. If a muon is near a jet with R<0.4 and the jet is associated with at least(less than) three charged tracks with pT>500MeV,themuon(jet)isremoved.

The E_T^missvariableisthemagnitudeofthenegativevectorsum of the transverse momenta, −

i^pT,i, where i represents both the “hard objects” andthe “soft term.” The hard objects consist of leptonsandjets, whichare individually reconstructedandcal- ibrated; thelistexcludespileupjets, whichare removedbya jvt requirement. The soft termis formed from inner detector tracks not associatedwiththehard objects,butmatchedtotheprimary vertex.Inthesearchregion,theE^miss_T producedbytheHiggsdecay isbalancedinthetransverseplanebythedijetsystem.

The jvt procedure is intended to remove pileup jets, but can cause large fake E^miss_T if it removesa high-p_T jet from the hard scatter, e.g., a jet from a pT-balanced three-jet event. In order to reduce this, a correlated quantity H_T^miss—deﬁned as |

jpT,j|^, where j represents all jets without the jvt requirement—is re- quired to be H^miss_T >150GeV. In the three-jet example, H^miss_T wouldbenearzero.

The E^miss_T significance (Smet) is used only in events with one identified electron in the final state and is defined as E^miss_T /√_p

T,j1+^pT,j2+^pT,e,wherethe pT quantitiesare forleading jet (j1),subleading jet(j2), andelectron,respectively.The useof thisquantity to reducethe contamination fromjetsmisidentiﬁed aselectronsisdiscussedinSection5.

3. Eventsimulation

Monte Carlo simulation (MC) consists of an event generation followedbydetectorsimulation[48] usinggeant₄[49].Simulated eventswere correctedforthesmalldifferencesbetweendataand

3 ThedistancevariableisdeﬁnedasR=

(η)²+(φ)².

(3)

troweak.The showering simulation followedthe same procedure asfortheVBFsample. Forboth theVBFandgluonfusionevents, the H→^{Z Z}^∗→⁴ν process isincludedin thesample asinvisible decaysof the Higgs boson.Additional scalars withmassesup to 3TeV were simulated as described above for VBF signal process, assumingafullwidthof4MeV.

TheW andZ eventsweregeneratedusingsherpa_2.2.1[60] with comix [61] and openloops [62] matrix-element generators, and mergedwithsherpapartonshower[63] usingtheme₊ps_@nlopre- scription[64]. Thennpdf_3.0 NNLO PDF setwas used.In termsof theorderofthevariousprocesses,thestrongproductionwas calculated atNLO forup to twojetsand leading order(LO) forthe third andfourthjets. The electroweakproduction was calculated atLOforthesecond andthirdjets.The levelsoftheinterference between electroweak and strong processes were computed with madgraph₅_amc_@nlo[65]. Theinterferenceon thetotalexpected backgroundisonly0.1% andthusneglected.

Other potential background processes involve top quarks, dibosons, and multijets. Top quarks and dibosons were generated withpowheginterfacedwithpythiaandevtgen[66],whichsim- ulate the heavy-ﬂavor decays. The diboson backgrounds include electroweak-mediated processes. The multijet estimate does not directlyusetheMC.

Toeachhard-scatterMCevent,pileupcollisions(30 onaverage) wereaddedtomimictheenvironmentoftheLHC.Thepileupcol- lisions,simulatedwithpythia₈[52] usingmstw₂₀₀₈ PDF[67] and thea₂setoftunedparameters[68],weresubsequentlyreweighted toreproducethepileupdistributionindata.

The sizes of the MC samples vary depending on the process.

TheeffectiveluminosityrangesfortheMCsamplesvariesdepend- ing on the process and on the selections, which are deﬁned in Section4.FortheW process,theMCsampleisapproximatelyhalf ofthatofthedataselectedfortheW controlregionandalsohalf forthesignalregion.Forthe Z process,theMCsampleforthe Z subprocessisapproximatelytwicethatofdatainthe Z controlre- gion;theMCsampleforZννsubprocessisapproximatelythesame asthatofdatainthesignalregion.

4. Eventselection

Alleventsmusthaveaprimary vertex.Theselectionlistedbe- low divides thedata sample into a signal-enrichedsearch region (SR)andbackground-enrichedcontrolregions(CR).Thecontrolre- gionsandthestatisticalﬁtarediscussedindetailinSection5.The restofthissectionfocusesontheSRandthepreﬁteventyields.⁴

FortheSR,aneventisrequiredtohave

•^noîsolatedêlectronôr^muon,

•^a^leading^jet^with^pT>80GeV,

•^a^subleading^jet^with^pT>50GeV,

4 “Prefit”indicatesthattheeventyieldsarenotadjustedaccordingtothestatis- ticaltreatmentofthebackgroundpredictions,whichisdescribedinthesecondhalf ofSection5.“Postfit”labelsthe quantitiesthatcomeoutofthefitprocedure.

Z→νν 1111 [18%] – –

Z→ee,μμ 12 [9%] 38 [9%] 181 [23%]

Z→τ τ 10 [16%] 11 [16%] –

W→eν,μν 540 [16%] 1400 [30%] –

W→τ ν 533 [20%] 130 [34%] –

Other 36 67 2

S, signal 1070 – –

VBF 930 – –

Gluon fusion 140 – –

• ^no^additional^jets^with^pT>25GeV,

• ^E^missT >180GeV,

• ^HT^miss>150GeV.

Thetwojetsarerequiredtohavethefollowingproperties:

• ^not^be^aligned^withE^miss_T ,|φj-met|>1,

• ^not^beback-to-back,|φj j|<1.8,

• ^be^well^separatedⁱⁿ η,|ηj j|>4.8,

• ^beⁱⁿ^opposite ηhemispheres, ηj1·ηj2<0,

• ^mj j>1TeV.

TheSRincludesbackgroundeventscontainingaW or Z plustwo jets, wherethe W decaysintoeν, μν,and τ ν,andthe Z decays into twoneutrinos. Here theleptons fromthe W decays arenot reconstructed since they would otherwise be rejectedby the se- lection.

Table1 givestheprefitSRyields inthe firstcolumn. TheVBF productionprocessgivesthebiggestcontribution(87%)tothesig- nalsample(fixedasBinv=^1).^Thecontributionfromgluonfusion accompaniedbypartonradiationissmall(13%)andotherproduc- tionmodescontributenegligibly.ThefractionofVBFsignalevents thatpassthesignalregioneventselections,definedasacceptance times reconstruction efficiency, is 0.7%. As is discussed in Sec- tion7,thesignalsignificanceisimprovedbyconsideringthreebins ofm_{j j} definedasfollows:1<m_{j j}≤¹.5TeV, 1.5<m_{j j}≤²^TeV,ând mj j>2TeV.Theprefit S/B ratio(forBinv=¹⁾ⁱⁿ^these^binsîsâp- proximately0.3,0.4,0.8,respectively.

For thebackgrounds, both the strong production andthe EW productioncontribute in theSR. The strongproduction processes contributes more than 70% of the backgrounds in all of the mj j bins. There is variation in the EW fractions for the background processes due to a combination of the following factors: known differencesintheproductiondiagramsbetween W and Z,differ- ences in kinematic acceptance for the particular W or Z decay, anddifferencesintheMCsamplesizeforeachEWprocess.

5. Controlsamplesandstatisticaltreatment

The main backgrounds in the SR, comprising of 98% of the background,arethe W and Z processes.The minorbackgrounds, comprising the remaining 2%, are the diboson, tt,¯ and multijet processes. Accurate estimation of the W and Z processes is the biggest challengeoftheanalysis. Themain backgroundyieldsare extractedusingdedicatedcontrolsamplesindata.