Large R jets and boosted. object tagging in ATLAS. Freiburg, 15/06/2016. #BoostAndNeverLookBack. Physikalisches Institut Universität Heidelberg

Large R jets and boosted object tagging in ATLAS Christoph Anders Physikalisches Institut Universität Heidelberg #BoostAndNeverLookBack Freiburg, 15/06/2016

??? Cambridge-Aachen 2 arxiv:1506.00962 Higgs tagger trimming n-subjetiness Jet substructure large R jets R2D2 BOOST anti kt HEPTopTagger subjets fully hadronic, di-jet mass grooming End of last year! there was a bump

Have: Standard Model Want: New physics! Extended gauge sector: Z > tt - W > WZ Extra dimensions: GRS >tt/ww/zz - Susy: stop decays In this context new heavy resonances decaying to pairs of SM bosons (W/Z/H) or top quarks would be nice (we need one to better calibrate our jets!) Hadronic final states? BR(W > hadrons) ~ 2/3 -

Detecting particles in ATLAS center of mass energy LHC Run 1: 7 & 8TeV LHC Run 2: 13TeV How about jets? 4

Jets Jets: collimated bunches of stable particles originating from partons after hadronization Jet finding: an approximate attempt to reverse engineer this QM process " More than one way to do this! 5 Sequential jet clustering algorithms

Sequential jet clustering algorithms Use clusters of calorimeter energy as input particles (also tracks or truth particles can be used) Distance: dij=min(pti 2p,pTj 2p ) ΔRij 2 /R 2 algo: find pair with smallest dij if dij > dib = pti 2p > i is already a jet, remove it else > merge i,j repeat until all particles are clustered into a jet parameters: R: geometrical separation, radius parameter, not a radius! p: energy vs geometry, 2=kt ; 0=C/A ; -2=anti-kt 6

kt Cambridge-Aachen Take home message: kt = soft first C/A = closest first anti kt = hard first anti kt Towards Jetography, G. Salam, Eur.Phys.J. C67 (2010)

What about BOOST? 8

What about BOOST? 18-22 July 2016 Early bird registration until tomorrow!!! 8

9 subjets

9 subjets large R jet

Angular separation Rule of thumb: standard jet in ATLAS: R= 0.4 JHEP09 (2013) 076 pt>200gev to contain W in R=1.0 jets top in R=1.5 jets 10

Two problems Larger jets pick up more junk! https://twiki.cern.ch/twiki/bin/view/atlaspublic/luminositypublicresults How to reject jets from light quarks and gluons? 11

recluster into small subjets JHEP 02 (2010) 084 Trimming JHEP09 (2013) 076 ATLAS typical values (for R=1.0 jets): Run 1: Rsub=0.3, fcut = 0.05 Run 2: Rsub=0.2, fcut = 0.05 12 remove low pt subjets We ll see this at work in a bit! One easy example:

- Side note: lep+jets tt hadronic side leptonic side 13 Graph by D0 experiment

- Side note: lep+jets tt study large R jets in data for top decays W decays require BOOST: pt>200gev identify events with 14 Graph by D0 experiment

Jet mass Top vs QCD Trimming signal and background JHEP09 (2013) 076 15 Normalized entries 2 where i = all clusters 0.25 ATLAS Simulation Preliminary anti-k t LCW jets with R=1.0 No jet grooming, no pileup correction 0.2 0.15 s = 14 TeV, 25 ns bunch spacing jet < 1000 GeV jet η <1.2, 500 < p T Pythia8 Z tt (m =2 TeV) Z µ=0 µ=80 µ=140 µ=200 µ=300 0.1 Un-trimmed 0.05 0 0 50 100 150 200 250 300 350 400 450 500 ATLAS-CONF-2015-036 Leading jet mass [GeV]

Jet mass Top vs QCD Trimming signal and background JHEP09 (2013) 076 15 Normalized Normalized entries entries 2 where i = all clusters 0.25 ATLAS Simulation Preliminary LCW jets with R=1.0 0.25 ATLAS Simulation Preliminary 0.2 0.2 0.15 0.1 0.1 anti-k t t LCW jets with R=1.0 No Trimmed, jet grooming, pileup no corrected pileup correction s s = = 14 14 TeV, 25 25 ns ns bunch spacing jet jet η jet <1.2, 500 500 < < p p < < 1000 GeV T T Pythia8 Z Z tt tt (m (m jet η =2 =2 TeV) Z Z µ=0 µ=0 µ=80 µ=140 µ=200 µ=300 Un-trimmed Trimmed 0.05 0 0 0 50 50 100 150 200 250 300 350 400 450 500 ATLAS-CONF-2015-036 Leading jet jet mass [GeV]

JHEP09 (2013) 076 Trimming signal 0 0 and background 15 ATLAS-CONF-2015-036 Leading jet jet mass [GeV] 0 0 50 50 100 150 200 250 300 350 400 450 500 arxiv:1603.03127 0.05 0.1 0.1 Un-trimmed Trimmed Normalized Normalized entries entries 0.15 jet η 0.2 0.2 anti-k t t LCW jets with R=1.0 No Trimmed, jet grooming, pileup no corrected pileup correction s s = = 14 14 TeV, 25 25 ns ns bunch spacing jet jet η jet <1.2, 500 500 < < p p < < 1000 GeV T T Pythia8 Z Z tt tt (m (m =2 =2 TeV) Z Z!=300!=200!=140!=80 0.25 ATLAS Simulation Preliminary LCW jets with R=1.0!=0!=0 0.25 ATLAS Simulation Preliminary Top vs QCD Works in data too! where i = all clusters 2 Jet mass

kt splitting scales go back one step in the kt clustering Z >qq vs QCD Example: Z >qq: symmetric two-body decay: both subjets apart and of similar pt > large d12 For top decay: go back one more step: d23 JHEP09 (2013) 076 16

N=2 n-subjetiness Observables related to Nsubjet reclusters jet constituents with kt into N subjets subjets define axes within the jet constituent k distance of k to subjet i 17 Signal JHEP 03 (2011) 015 ratio: background arxiv:1510.05821 1 subjet like 2 subjet like

JHEP 12 (2014) 009! Energy correlation functions m of particle in 2-body decay for beta=1 18 Signal no subjets! background arxiv:1510.05821

Variables arxiv:1510.05821 Are the same, apart from the mass cut! W vs Z Compare rejection @ fixed signal efficiency Jet collections bkg rej @ eff = 50% Pick the following: Jet type and radius parameter, e.g. anti kt R=1.0 Groomer and its parameters, e.g. trimming, Rsub=0.3, fcut=5% Choose substructure variables (of course there are many more!) Building a simple W tagger

Run 1 boson tagging in data Large R jet mass arxiv:1510.05821 20 Select D2, before mass cut

Efficiency Signal Di-jets (background) 21 arxiv:1510.05821

q q Large R jet W/Z Heavy Resonance W/Z q q Large R jet

Back to the start! arxiv:1506.00962 ish Full disclosure: C/A R = 1.2 jets, modified mass-drop filtering (actually no mass drop), Rsub = 0.3 ntracks < 30 Also WW,ZZ. 23

Limits on high mass resonance and other final states jet jet No bump in leptonic channels! lep, neutrino + jet arxiv:1506.00962 2 lep + jet Run 2 will have to tell! Eur. Phys. J. C (2015) 75:209 Eur. Phys. J. C (2015) 75:69 24

q q Large R jet On to Run 2 Di-boson searches: W/Z Heavy Resonance W/Z 25 lepton pair or 2. large R jet

Multi jet background == fake bosons Looking good! https://atlas.web.cern.ch/atlas/groups/physics/plots/jetm-2015-004/ Basics in Run 2 data

Simple Run 2 tagger W/Z boson tagger based on: anti kt, R=1.0, Rsub=0.2, fcut=5% Mass and D2 Constant signal efficiency (25% and 50%) ATL-PHYS-PUB-2015-033 27

Run 2 di-boson (W/Z) searches Channel VV > JJ VV > J ll VV > Jlν VV > Jνν Trigger Large R jet Boson Tag Large R jet, 360GeV 2 with pt>400/200 GeV 50% W/Z + Ntrack<30 Leptons/MET no Backgrounds mutlijet, shape fit in data electron/muon triggers 1 with pt>200 GeV electron/muon triggers 1 with pt>200 GeV MET trigger, 80GeV 1 with pt>200 GeV 50% W/Z 50% W/Z 50% W/Z e + e - / μ + μ - compatible with Z decay Z+jets, from mj sideband MET>100GeV 1 e or mu W+jet mj sideband top b-tagged CR MET>200 GeV Zmumu and btagged ATLAS-CONF 2015-073 2015-071 2015-75 2015-068 28

Fully hadronic search == Tagging in action 2 large R jets, W or Z tagged, =50%, rej>90% Ntrack<30, exploiting bigger track multiplicity in background, ~30% improvement in sensitivity efficiency checked in data ATLAS-CONF-2015-073 Leading jet, D2 cut 9<Ntrack < 20 Simulation to match data Fit fraction of W/Z in data In data lower by 5%+- 6% > 6% systematic uncertainty 29

Evolution in data Full W/Z w/o mass w/o D2 w/o Ntrack No boson tagging WZ 30 ATLAS-CONF-2015-073 ZZ WW

ATLAS-CONF-2015-073 Signal eff. & background fit Reco d signal mass Fit! to MC Stable vs mass Low/high mj control region: Fit! to data CR

Putting it all together ATLAS-CONF-2015-073 + = WZ ZZ WW 32

ATLAS-CONF-2015-073 Limits Run 1 excess not excluded Need more data 33

The other channels J ll J l# J ## Same conclusion > Need more data! ATLAS-CONF-2015-071 ATLAS-CONF-2015-075 ATLAS-CONF-2015-068 34

Boosted Higgs bosons b RSG >hh >4b b large R jets h b jets inside large R jets X match small (R=0.2) b-tagged track jets to large R jets h b b 35 larger jets ATL-PHYS- Boost smaller jets

Higgs boson tagging b-tagging + Jet substructure jet mass ATL-PHYS-PUB-2015-035 = D2 anti kt R=1.0 Rsub=0.2 fcut=5% Most discrimination from b-tagging, but JSS can help. 36 New in Run 2

Back to the original fat jets subjets large R

Top tagging Simple taggers possible as for boson tagging Top decays have a few more handles There are more advanced taggers on the market Large R jet Simple tagger example: Select http://arxiv.org/abs/1603.03127

Adv. Tagger Example: JHEP 1010 (2010) 078 HEPTopTagger Identify top to hadron decays with pt top >200GeV Use Cambridge/Aachen R=1.5 jets and their substructure Filter against pile-up b Identify top quarks via mass ratios C/A R=1.5 top candidate top g g q q Pictures G. Kasieczka 39

mw/mtop http://arxiv.org/abs/1603.03127 HEPTopTagger in lep+ jet data = m(subjet 2 and 3) m(subjet 1, 2 and 3) top candidate mass 40

http://arxiv.org/abs/1603.03127 Efficiency measurement ~40% ~80% Simple Tagger HEPTopTagger ~25-30% ~2-3% 41

Fully hadronic tt - resonance search Lots of non top background (QCD) JHEP 1301 (2013) 116 42

Fully hadronic tt resonance - search (Run 1) It works! JHEP 1301 (2013) 116 43

ATL-PHYS-PUB-2015-053 Run 2 top tagging ~ constant efficiency Pick two variables 44 https://atlas.web.cern.ch/atlas/groups/physics/plots/jetm-2015-004/

- tt resonance search in lepton+jet events (Run2) Lepton Large R jet, top tagged Di-top mass ATLAS-CONF-2016-014 45 main background contains top quarks high efficiency wanted! (comes with low rejection)

- tt resonance search in lepton+jet events (Run2) Lepton Large R jet, top tagged Di-top mass ATLAS-CONF-2016-014 45 main background contains top quarks high! low efficiency wanted (comes with rejection)

Jet Reclustering or doing it all backwards Large R jets are great, but they require extra work (calibrations, uncertainties etc) Reclustering: Use standard small R(=0.4) jets as inputs to the large R jet finding! by B. Nachman Advantages: inherit calibrations and uncertainties from well understood small R jets, easy to correlate with MET, faster, Disadvantages: less information used, need to understand close-by-jet effects, it wasn t my idea 46 http://arxiv.org/abs/1407.2922

A use case Has been pioneered in ATLAS SUSY analyses How about VLQ: } Many heavy objects to tag! Standard top, W or H taggers not ideal Require a lepton and MET and a few b-tags Reclustered large R jet(s) with mj>100 GeV (i.e. keep H and top) ATLAS-CONF-2016-013 47

A use case Has been pioneered in ATLAS SUSY analyses } How about VLQ: Many heavy objects to tag! Standard top, W or H taggers not ideal Require a lepton and MET and a few b-tags Reclustered large R jet(s) with mj>100 GeV (i.e. keep H and top) ATLAS-CONF-2016-013 47

A use case Has been pioneered in ATLAS SUSY analyses How about VLQ: } Many heavy }objects to tag! 1 of many Standard SRs top, W or H taggers not ideal Require a lepton and MET and a few b-tags Reclustered large R jet(s) with mj>100 GeV (i.e. keep H and top) ATLAS-CONF-2016-013 47

Of course there is more! There will be/ is more W > tb SM W/Z xs Eur. Phys. J. C (2015) 75:165 arxiv:1407.0800 stop search http://arxiv.org/abs/1606.03903 arxiv:1510.03818 SM boosted ttbar xs 48 at 13TeV in Run 2!

Of course there is more! Eur. Phys. J. C (2015) 75:165 JHEP 09 (2014 ) 015 W > tb stop search 49 SM W/Z xs arxiv:1407.0800 arxiv:1510.03818 SM boosted ttbar xs W/Z discrimination area subtraction template overlap method There will be more at 13TeV in Run 2! quark gluon tagging q jets mass calibration shower deconstruction in situ JES from top mass subjet calibration variable R jets uncertainty estimation

Summary Dramatic increase in understanding of hadronic final states in the last few years Boosted/jet substructure techniques have been shown to work in Run 1 of LHC + been employed in analysis They are even more important at the higher center of mass energy we are running at now First Run 2 results are in > Tagging still works well! We are already thinking about the even more boosted regime G. Salam (not touched on today) 50

Closing remark