AATL-PHYS-PROC-2014-247October 12, 2018
Searches with Boosted Objects
Katharina Behr
On behalf of the ATLAS and CMS CollaborationsSub-department of Particle Physics, University of Oxford,Keble Road, Oxford, OX1 3RH, United KingdomBoosted objects - particles whose transverse momentum is greater thantwice their mass - are becoming increasingly important as the LHC con-tinues to explore energies in the TeV range. The sensitivity of searchesfor new phenomena beyond the Standard Model depends critically on theefficient reconstruction and identification (“tagging”) of their unique de-tector signatures. This contribution provides a review of searches for newphysics carried out by the ATLAS and CMS experiments that rely on thereconstruction and identification of boosted top quarks as well as boosted W , Z and Higgs bosons. A particular emphasis is placed on the differentsubstructure techniques and tagging algorithms for top quarks and bosonsemployed by the two experiments.PRESENTED AT XXXIV Physics in Collision SymposiumBloomington, Indiana, September 16–20, 2014 a r X i v : . [ h e p - e x ] N ov Introduction
Despite its tremendous success - once again impressively demonstrated by the discov-ery of the long-predicted Higgs Boson in 2012 - the Standard Model (SM) of particlephysics is widely considered an incomplete theory. For one, it cannot explain the factthat the mass of the Higgs boson is light (hierarchy problem) nor does it offer a can-didate for dark matter or satisfactorily explain the matter-antimatter asymmetry inthe observed universe. Hence a number of extensions to the SM have been proposedwhose predictions are currently under scrutiny by the LHC experiments. These ex-tensions include theories with warped extra-dimensions (Randall-Sundrum models),new strong interactions (Technicolour and others), an additional quark generation orvector-like quarks as well as supersymmetry. (For details of the specific models seethe references in Sections 3, 4, 5 and references therein.) Many of these modelspredict the existence of new heavy particles with large branching fractions into topquarks, heavy gauge bosons or the Higgs boson. If these new states are sufficientlyheavy their decay products are likely to have transverse momenta exceeding twicetheir rest masses. These decay products are called boosted objects .The sensitivity of searches for new phenomena at high energies depends criticallyon the efficient reconstruction and identification of boosted object decays. Boostedtechniques first were applied in searches at the Tevatron (see [1] for a recent review)and developed into a fast growing field of research during Run I (2010-2012) of theLHC, as its higher center-of-mass energies of 7 TeV (2011) and 8 TeV (2012) allow forabundant production of boosted objects across many final states. This enabled theexperiments to push the exclusion limits for many new particles into the TeV regime.This document provides a review of the most recent searches with boosted objectscarried out by the ATLAS [2] and CMS [3] experiments and presents some of themost commonly used reconstruction and identification techniques. The rapid growthof the field makes it impossible to cover every single technique within the scope of thisdocument. More details can be found in the proceedings of the BOOST workshop [1].
The defining property of boosted object decays is the fact that their decay productsappear collimated in the momentum direction of the boosted mother particle in therest frame of the detector. Their angular separation ∆ R is inversely proportionalto the transverse momentum p T of the mother particle with mass m according to asimple rule of thumb: ∗ ∆ R ≈ m/p T . Figure 1(a) illustrates this for the boosted ∗ ∆ R = (cid:112) ∆ η + ∆ φ with pseudorapidity η = -ln tan( θ /2). θ ( φ ) is the polar (azimuthal) angleof the ATLAS/CMS standard coordinate system, a right-handed orthogonal system with the z-axistangential to the beam pipe and the nominal interaction point in the detector centre as its origin. a) (b) Figure 1: (a) Angular separation between the two b quarks from the decay of aHiggs boson as a function of p HT [4]. (b) Mass distribution of the leading- p jetT jet forungroomed and trimmed jets on a Z (cid:48) → t ¯ t signal and a dijet background sample [5].decay H → b ¯ b . Consequently, at high p T , the decay products of a hadronicallydecaying object merge into a single large- R jet with a characteristic substructurethat allows one to distinguish these jets from those initiated by a single parton.While large- R jets can be reconstructed using any of the three common sequentialrecombination algorithms — anti- k T [6], Cambridge-Aachen (C-A) [7, 8] or k T [9, 10]— only the last two are suited for a substructure analysis: they start by clusteringclose-by (C-A) or close-by and soft particles ( k T ), effectively reversing the orderingof the parton shower. By undoing the last clustering step(s) and analysing propertiesof the subjets such as their relative p T fraction or angular separation, the presence ofhard splittings in the jet can be probed. ATLAS and CMS use a number of differentsubstructure variables to tag boosted object jets, the most prominent among whichare the k T splitting scales (see for example [11] and [12]) and n-subjettiness [13] whichmeasures how compatible the jet structure is with the “n subjets” hypothesis.At high-luminosity hadron colliders, a major obstacle for analyses relying on large- R jets is the presence of pile-up and the Underlying Event, both of which lead to soft,wide-angle contaminations that dilute the jet substructure. Various grooming tech-niques that remove these contaminations have been developed, the most widely usedof which are trimming [14], pruning [15] and filtering [16]. Figure 1(b) illustrates theeffect of trimming on the jet mass, a variable widely used for boosted top identifica-tion. The trimmed distributions exhibit a significantly improved separation betweensignal and background compared to the ungroomed case.The efficient identification of jets originating from b -quarks is another crucialaspect for new physics searches. Dedicated performance studies targeted particularlyat boosted topologies have been published [17].2 Searches with Boosted Top Quarks
Both ATLAS and CMS have conducted a wide range of searches in final states withboosted top quarks which rely on different combinations of substructure variables andgrooming techniques to efficiently identify hadronic decays of boosted top quarks.Many such dedicated “top taggers” have been developed and optimised for differentfinal states and kinematic regimes. A review of these techniques can be found in theperformance notes by ATLAS [11] and CMS [12]. t ¯ t Resonance Searches
Searches for heavy resonances decaying into t ¯ t pairs have traditionally been the flag-ship applications for boosted techniques, both at the Tevatron and the LHC. AT-LAS and CMS have published results in the single-lepton+jets (1 (cid:96) +jets) and theall-hadronic decay channels which rely on the reconstruction of hadronic decays ofboosted top quarks. Two benchmark models, a leptophobic top-colour Z (cid:48) boson(narrow resonance) and a Kaluza-Klein gluon g KK arising in Randall-Sundrum (RS)models (wide resonance), have been considered. For the sake of brevity, only resultsfor the Z (cid:48) model will be discussed here.CMS has published results from the combination of the semileptonic and theall-hadronic decay channels using 19.7 fb − of √ s =8 TeV pp collision data [18]. Inthe semileptonic channel both a traditional resolved selection which relies on thereconstruction of the individual decay products of the top quarks and a selectionrelying on boosted techniques are used: while the resolved selection requires exactlyone isolated electron or muon as well as ≥ k T jets with R = 0 .
5, the boostedselection requires ≥ k T jets with R = 0 . b -jetis small on average. The boosted channel dominates the sensitivity of the expectedupper limit for m t ¯ t > R = 0 . CMSTopTagger [12] are required. This top tagger analyses the subjets and usesvarious kinematic criteria such as a W and top mass window requirement to identifythe three-pronged substructure compatible with the decay t → W b → q ¯ q (cid:48) b . Thecombination of all channels yields a 95% CL lower limit on m Z (cid:48) of 2.1 TeV.ATLAS has published two separate searches in the semileptonic (resolved andboosted) and all-hadronic (boosted only) decay channels. The semileptonic search [19]uses 14 fb − of √ s =8 TeV data. In the boosted channel the hadronically decaying topquark is reconstructed as a trimmed anti- k T k T splitting scale √ d . In addition, this analysis3 a) (b) Figure 2: (a) Upper cross-section limits on the production of Z (cid:48) using the combinationof the semileptonic and all-hadronic channels. Above 1 TeV the sensitivity is drivenby the boosted channel [18]. (b) Performance of mini-isolation (blue curves) for highlyenergetic leptons compared to traditional fixed-cone isolation concepts [19].uses a novel lepton isolation concept, mini-isolation , which replaces the traditionalfixed-R isolation cone by one that shrinks inversely with increasing lepton transversemomentum p lepT , thus taking into account the collimation of the decay products ofboosted particles. Mini-isolation provides consistently high signal efficiency over thewhole p lepT range and outperforms traditional isolation concepts for higher p lepT asshown in Figure 2(b). The 95% CL lower limit on m Z (cid:48) resulting from the combinationof the boosted and resolved channels is 1.7 TeV.The ATLAS all-hadronic search [20] is based on the full 2011 dataset which com-prises 4.7 fb − of pp collision data at √ s =7 TeV. Two top taggers optimised fordifferent kinematic regimes are used: The HEPTopTagger [21, 22] is applied to C-A R = 1 . p jetT >
200 GeV and uses a combination of mass-drop tagging andfiltering to identify top jets. The
TopTemplateTagger [23] uses overlap functions thatcompare the energy flow in an anti- k T R = 1 . t ¯ t decays in order to quantify the resemblance ofa jet with boosted top jets and jets initiated by a single parton. The p T threshold forthe leading (subleading) jet is 500 (450) GeV. Z (cid:48) bosons are excluded at 95% CL inthe mass ranges 0.70-1.00 TeV as well as 1.28-1.32 TeV.Figure 3(a) shows a comparison of the performance of different top taggers interms of signal efficiency and rejection of multijet background. The HEPTopTagger (blue markers) provides a high background rejection rate at the cost of a low signalefficiency, suited for all-hadronic searches. The lower background in the 1 (cid:96) +jets searchallows for a tagger with a higher signal efficiency. The red cross marks the top taggerused in [19]. 4 a) (b)
Figure 3: (a) Comparison of top taggers in terms of signal efficiency and rejec-tion of multijet background [11]. Different final states require different top tag-gers, see text. (b) Contribution of the boosted and resolved channels to the over-all acceptance × efficiency as a function of the “bulk” RS graviton mass M G ∗ in thesemileptonic diboson resonance search [34]. W (cid:48) → tb Both ATLAS and CMS have published searches for a heavy W partner, W (cid:48) , decayingvia W (cid:48) → tb with a hadronically decaying top quark using 20 fb − of pp collision dataat √ s =8 TeV. The ATLAS search [24] is optimised for W (cid:48) masses above 1.5 TeV wherethe boosted top quark is identified using a dedicated W (cid:48) Top Tagger . This taggeruses trimmed anti- k T R = 1 . p jetT >
250 GeV and cuts on the k T splittingscale √ d and two n-subjettiness variables to identify the 3-pronged substructure ofa boosted top decay. CMS [25] relies on the CMSTopTagger with n-subjettiness asan additional identification criterion. Only jets with p jetT >
450 GeV are consideredas top jet candidates.ATLAS sets upper limits at 95% CL on the W (cid:48) → tb cross-section times branch-ing ratio ranging between 0.16 pb and 0.33 pb for W (cid:48) bosons with purely lef-handedcouplings, and between 0.10 pb and 0.21 pb for W (cid:48) bosons with purely right-handedcouplings. The sensitivity of the CMS search is further enhanced by combining theresults with those from the semileptonic channel where the top quark decays lepton-ically. This leads to an exclusion of right-handed W (cid:48) with masses below 2.15 TeV at95% CL. Both experiments also set upper limits at 95% CL on the couplings to tb asa function of m W (cid:48) thus allowing for a model independent interpretation of the results. The application of boosted techniques in searches for supersymmetry is a relativelyyoung field that has gained importance as the lower limits on the masses of super-5ymmetric particles have been pushed towards the TeV-scale. Searches for stop quarkpair production where boosted top quarks and W bosons arise from the decay ˜ t → tχ have been published by both ATLAS (0- and 1-lepton channel [26, 27]) and CMS (0-lepton channel [28]). For sake of brevity only the ATLAS 0-lepton search [26] will bediscussed here because it employs a novel substructure technique: jet reclustering [29].Here the large- R jets are not built from calorimeter clusters or inner detector tracksbut from small- R jets. This particular search uses anti- k t jets with R = 0 . p T >
25 GeV as input to anti- k t clusterings with R = 0 . p T cut on thesmall- R jets acts like trimming. The event selection requires ≥ R = 1 . p jet, . T requirements. These are considered top jetcandidates. Another mass requirement is placed on the reclustered R = 0 . p jet, . T in order to suppress backgrounds without hadronic W candidates.For a branching fraction of 100% into tχ stop quark masses in the range 270-645 GeVare excluded at 95% CL for χ masses below 30 GeV. Boosted techniques are becoming increasingly common in searches with hadronicallydecaying W , Z and Higgs bosons in the final state. Searches in all-hadronic finalstates especially benefit from substructure techniques as these allow for an effectivecontrol of the large multi-jet background. Dedicated tagging techniques for boostedbosons have been studied by both ATLAS [30] and CMS [31].CMS has published an inclusive search for resonances decaying to qW , qZ , W W , W Z or ZZ with fully hadronic boson decays [32]. In the boosted regime, the eventsexhibit a simple dijet topology. Signal events with boosted bosons are tagged byrequiring one or two pruned C-A jets with m jet between 70 and 100 GeV and two-pronged substructure identified using the n-subjettiness variable. The lower mass lim-its on excited quarks, RS1 gravitons and W (cid:48) bosons, set using 19.7 fb − of √ s =8 TeVdata, are all above 1 TeV hence accounting for the presence of boosted bosons.Resonance searches in the semileptonic final state have been conducted by bothATLAS [34] and CMS [33]. The ATLAS search focuses on decays to W Z and ZZ with at least one leptonic decay Z → (cid:96) + (cid:96) − . Both resolved and boosted scenarios areconsidered. In the boosted case, modified lepton isolation criteria are used to maintaina high selection efficiency for events where the two leptons from the boosted Z decayget into each others isolation cones. Boosted hadronic boson decays are identified byapplying a slightly modified version of the mass-drop-filtering technique from Ref [16]to C-A R = 1 . B (cid:48) quarks which are predicted by many modelsinvolving top partners and decay via B (cid:48) → Hb . The search has been optimised forboosted H → b ¯ b decays, the dominant Higgs decay mode with 56% branching ratio.In this difficult final state, substructure techniques are a major asset in reducingthe multi-jet background. Higgs jets are reconstructed as pruned C-A R = 0 . B (cid:48) quarks for masses below 846 GeV at 95% CL based on 19.7 fb − of √ s =8 TeV data. Top quark partners play an important role in many extensions of the SM since theyallow for cancellation of the quadratically divergent quantum-loop corrections to theHiggs boson mass introduced (predominantly) by the top quark. Their decays usuallyinvolve both top quarks and heavy gauge or Higgs bosons in the final state which areboosted if the hypothetical top partner is sufficiently heavy.A search for pair production of a vector-like top partner with charge ± e/ T / ,has been conducted by CMS [36] using 19.5 fb − of √ s =8 TeV pp collision data. Thetop partner is expected to decay via T / → tW , and the search focuses on the same-sign dilepton final state where the top quark and W boson from at least one T / both decay leptonically. Boosted top quarks are identified via the CMSTopTagger while boosted hadronic W bosons are reconstructed as pruned C-A R = 0 . m W . T / masses below 800 GeV areexcluded at 95% CL as illustrated in Figure 4(a).Several searches have been conducted for pair production of vector-like quarks, T ,with the same charge as the top quark. A T quark can decay into three final states: W b , Zt and Ht where the branching ratios are model dependent. CMS has conductedthe first inclusive search [37] for all three decay modes in final states with at least oneisolated lepton using the 19.7 fb − of √ s =8 TeV data. Boosted hadronic top quarksare tagged using the CMSTopTagger , and boosted W bosons are reconstructed aspruned C-A R = 0 . m W . Mass limits are set as a functionof the branching ratios as illustrated in Figure 4(b). A search in the same final statebut assuming a 100% branching ratio into W b has been conducted by ATLAS [38].Boosted hadronic W decays are reconstructed as anti- k T R = 0 . m W and mini-isolation is required on muons. Searches optimised for the Ht decay mode with H → b ¯ b have been published by both experiments. Only the CMSsearch [39] uses boosted techniques and will be discussed here. Both boosted topquarks and boosted Higgs bosons are reconstructed from filtered C-A R = 1 . p T >
150 GeV. The top jet is identified using the
HEPTopTagger whereas the7 a) (b)
Figure 4: (a) Upper cross-section limit as a function of m T / [36]. (b) Branching-fraction triangle with observed 95% CL limits on the mass of a 2 e/ T quark [37].Higgs jet is required to have at least two b -tagged R = 0 . Ht , the observedlimit on m T at 95% CL is 747 GeV based on 19.7 fb − of √ s =8 TeV data. Boosted objects are key elements in searches for new physics at the high energyand mass scales accessible at the LHC because they provide sensitivity in kinematicregimes where traditional reconstruction techniques fail. No deviation from the SMhas been observed during Run-I and upper limits on many benchmark models havebeen pushed into the TeV regime. This together with the planned increase of thecenter-of-mass energy of the LHC to 13 (later 14) TeV from 2015 onwards will fur-ther boost the number of searches (as well as measurements) relying on these noveltechniques. The era of boosted objects has only just begun.
ACKNOWLEDGEMENTS
For fruitful discussions and comments on both the presentation and these proceedings,I would like to thank N. Gutierrez, C. Heidemann, M. Martinez, A. Schmidt, F.Spano, J. Tseng, S. Willoq and J. Zhong. I gratefully acknowledge support from theRhodes Trust. I would further like to thank the C R Barber Trust Fund as well asthe conference organisers for funding this travel.8 eferences [1] A. Altheimer et al. , Eur. Phys. J. C , 2792 (2014) [arXiv:1311.2708].[2] ATLAS Collaboration, JINST , S08003 (2008).[3] CMS Collaboration, JINST , S08004 (2008).[4] CMS Collaboration, CMS-PAS-B2G-14-001, http://cds.cern.ch/record/1752557.[5] ATLAS Collaboration, JHEP , 076 (2013) [arXiv:1306.4945].[6] M. Cacciari, G. P. Salam and G. Soyez, JHEP , 063 (2008)[arXiv:0802.1189].[7] Y. L. Dokshitzer, G. D. Leder, S. Moretti and B. R. Webber, JHEP , 001(1997) [hep-ph/9707323].[8] M. Wobisch and T. Wengler, In *Hamburg 1998/1999, Monte Carlo generatorsfor HERA physics* 270-279 [hep-ph/9907280].[9] S. D. Ellis and D. E. Soper, Phys. Rev. D , 3160 (1993) [hep-ph/9305266].[10] S. Catani, Y. L. Dokshitzer, M. H. Seymour and B. R. Webber, Nucl. Phys. B , 187 (1993).[11] ATLAS Collaboration, ATLAS-CONF-2013-084,https://cds.cern.ch/record/1571040.[12] CMS Collaboration, CMS-PAS-JME-13-007, https://cds.cern.ch/record/1647419/.[13] J. Thaler and K. Van Tilburg, JHEP , 015 (2011) [arXiv:1011.2268].[14] D. Krohn, J. Thaler and L. T. Wang, JHEP , 084 (2010) [arXiv:0912.1342].[15] S. D. Ellis, C. K. Vermilion and J. R. Walsh, Phys. Rev. D , 051501 (2009)[arXiv:0903.5081].[16] J. M. Butterworth, A. R. Davison, M. Rubin and G. P. Salam, Phys. Rev. Lett. , 242001 (2008) [arXiv:0802.2470].[17] CMS Collaboration, CMS-PAS-BTV-13-001, https://cds.cern.ch/record/1581306/.[18] CMS Collaboration, Phys. Rev. Lett. , no. 21, 211804 (2013),[arXiv:1309.2030]. 919] ATLAS Collaboration, ATLAS-CONF-2012-136,https://cds.cern.ch/record/1478974.[20] ATLAS Collaboration, JHEP , 116 (2013) [arXiv:1211.2202].[21] T. Plehn, G. P. Salam and M. Spannowsky, Phys. Rev. Lett. , 111801 (2010)[arXiv:0910.5472].[22] T. Plehn, M. Spannowsky, M. Takeuchi and D. Zerwas, JHEP , 078 (2010)[arXiv:1006.2833].[23] L. G. Almeida, O. Erdogan, J. Juknevich, S. J. Lee, G. Perez and G. Sterman,Phys. Rev. D , 114046 (2012) [arXiv:1112.1957].[24] ATLAS Collaboration, submitted to EPJ C, [arXiv:1408.0886].[25] CMS Collaboration, CMS-PAS-B2G-12-009, https://cds.cern.ch/record/1751504.[26] ATLAS Collaboration, JHEP , 015 (2014) [arXiv:1406.1122].[27] ATLAS Collaboration, submitted to JHEP, [arXiv:1407.0583].[28] CMS Collaboration, CMS-PAS-SUS-13-015, https://cds.cern.ch/record/1635353.[29] B. Nachman, P. Nef, A. Schwartzman and M. Swiatlowski, [arXiv:1407.2922].[30] ATLAS Collaboration, ATL-PHYS-PUB-2014-004,https://cds.cern.ch/record/1690048/.[31] CMS Collaboration, CMS-PAS-JME-13-006, https://cds.cern.ch/record/1577417/.[32] CMS Collaboration, JHEP , 173 (2014) [arXiv:1405.1994].[33] CMS Collaboration, JHEP , 174 (2014) [arXiv:1405.3447].[34] ATLAS Collaboration, ATLAS-CONF-2014-039,https://cds.cern.ch/record/1735253.[35] ATLAS Collaboration, Phys. Rev. Lett. , no. 4, 041802 (2014),[arXiv:1309.4017].[36] CMS Collaboration, Phys. Rev. Lett. , 171801 (2014) [arXiv:1312.2391].[37] CMS Collaboration, Phys. Lett. B , 149 (2014) [arXiv:1311.7667].[38] ATLAS Collaboration, Phys. Lett. B718