jet reconstruction and jet physics in hi collisions with cmsheinz/jet2013/lecturenotes/roland... ·...

Gunther Roland JET summer school

Jet reconstruction and jet physics in HI collisions with CMS

JET collaboration summer schoolOSU June 2013

Gunther Roland


Overview

• I will discuss selected aspects of jet finding and jet physics in HI collisions using CMS

• Emphasis on tools, algorithms, analysis methods, not on physics - look behind the curtain

• What are the main experimental challenges?• Caveats for comparison to theory• What can one expect in the future?

2

Gunther Roland JET summer school4

Inner tracker:charged particlesvertex, isolation

solenoid

EM and Hadron calorimetersphotons, isolation

MuonHCALECALTracker |η|< 2.5

|η|< 3.0

|η|< 5.2

|η|< 2.4

Pb

Pb

Calojet

Particle Flow Jet (track pT> 0.9GeV/c)

CMS detector


Trigger and Event Selectionl MinBias Trigger:

l Coincidence of BSC or HF signall Trigger efficiency: 97% ± 3%

• Di-Muon Trigger:l Two tracks in the muon detector

• Photon Trigger:l Uncorrected photon ET> 15 GeV

• Jet Trigger: l Uncorrected jet ET> 35, 50 GeV

• High pT track trigger:l Charged particles pT> 12 GeV/c

l Centrality determination:l Forward calorimeter (HF) energy

Dec. 2011: ≈ 150 µb–1

Dec. 2010: ≈ 7 µb–1

4


CMS Jet Jargon

5

• Some CMS jargon– “gen level”: final state particle truth info of event generators

(e.g. PYTHIA) before decay of long-live particles (cτ > 1cm)

– “reco level”: after simulation of all decays, full simulation (GEANT 4) and reconstruction of tracks, photons, jets etc

– “embedding”: combining a single generator hard scattering event (e.g. a PYTHIA pp dijet event) with the heavy-ion underlying event (from data or from HYDJET). Called “overlay” by ATLAS

– “closure test”: comparison of an observable (e.g. dijet Aj) for e.g. PYTHIA gen level truth vs reco level result after embedding for the same set of PYTHIA events


Backgrounds and jet finding performance


Jets and underlying event

2

?

partons+Underlying Event(UE)

hadrons:parton-associatedUE-associatedJETS: well defined by the clustering algorithm, FastJet anti-kT, R = 0.3

Energy-corrected to particle-level jets (PYTHIA) w/o constituent pT threshold


?

Challenges:• Under/over subtraction of UE • Some parton associated particles are lost because of reconstruction inefficiency• Some parton associated particles are lost because of bkg subtraction• Calorimeter energy deposit of the final particles may fluctuate• Particle composition may be different from assumptions used for energy scale calibration

Jets and underlying event


HCAL clusters•Neutral hadrons•Capture charged hadrons that tracking missed•Event-by-event shower fluctuations:

• Non-linearity• Wide resolution

•Acceptance limited due to B-field•Low granularity

4T

5

ECAL clusters•Photons, electrons

Tracks•Best pT resolution•Blind to neutral energy•Not 100% efficient•Limited acceptance (Details: CMS-PAS-HIN-11-004)

CMS Physics objects


Jet Measurements

(Tracking for only the primary vertex)

UE subtraction and

jet clustering

ParticleFlow algorithm

Towers

6

Δη x Δϕ0.076 x 0.076

in barrel

Calorimeter clusters and tracks are matched(Details: CMS-PAS-HIN-11-004)

Candidates are merged into pseudo-towers for UE subtraction(Energy calibration is done for UE-subtracted found jets)

Gunther Roland JET summer school 7

UE-subtraction algorithm

Tunable parameters:• Coefficient of RMS

Estimate background for each tower ring of constant η estimated background = <pT> + σ(pT)•Captures dN/dη of background•Misses ϕ modulation – to be improved


Subtract background from all towersRun the clustering algorithm (anti-kT)



Start over, knowing where the jets roughly are



Start over, knowing where the jets roughly areExclude a certain area around the jets Re-estimate the background for all towers

Tunable parameters:•Coefficient of RMS•Raw jet threshold•Radius of exclusion(not necessarily = R)



Start over, knowing where the jets roughly areExclude a certain area around the jets Re-estimate the background for all towersSubtract final backgroundCluster jets

Tunable parameters:•Coefficient of RMS•Raw jet threshold•Radius of exclusion(not necessarily = R)



Limits on low pT jet finding

16



17

Jets that fluctuated below threshold in first step of iteration(got added to UE estimate, leading to oversubtraction)


Tracking performance

12

Validate efficiency by:

•Analysis of hadron spectra in pp•Track multiplicity distributions•CaloTower-track matching•CaloJet-PFJet matching

Important to understand in jet-track correlation analyses

•Tracking effiency vs distance to jet axis•Autocorrelations

CMS-PAS-HIN-12-013


• Hydjet 1.8 default tune successfully reproduces:• Charged hadron multiplicity• Charged hadron pT spectrum• Azimuthal asymmetry of low-pT particles (Elliptic Flow)

• Pythia dijets embedded into Hydjet and fully simulated

http://lokhtin.web.cern.ch/lokhtin/hydro/plots

-- Hydjet•ALICE PRL106(2011)032301 0-5% Central

-- Hydjet•ALICE PLB 696 (2011) 30 0-5% Central

-- Hydjet•ALICE PRL 105 (2010) 252302 10-20% Central

15

Effects in the reconstruction:•Tracks : Efficiency, fakes•Jets : Energy scale, resolution, fake-jets

Understand jet performance in HI


UE data vs MC comparison


UE data vs MC comparison

17

Estimated by the PU algorithm

(GeV/c)TJetp

100 150 200 250 300

> (G

eV/c

)T

< Ba

ckgr

ound

p

0

20

40

60

80

100PbPb data

0-5% 5-10% 10-30% 30-50% 50-70% 70-90% PYTHIA + HYDJET

Particle Flow Jets R = 0.3TAnti-k| < 2

Jetη|

CMS Preliminary(a)

-1bµ dt = 129 PbPb L∫

(GeV/c)TJetp

100 150 200 250 300

) (G

eV/c

)T

RM

S ( B

ackg

roun

d p

0

5

10

15

20 PbPb data0-5% 5-10% 10-30% 30-50% 50-70% 70-90% PYTHIA + HYDJET

Particle Flow Jets R = 0.3TAnti-k| < 2

Jetη|

CMS Preliminary(b)

-1bµ dt = 129 PbPb L∫


Jet background in PbPb

• Mean and width for background pT subtracted from jets– before jet energy correction

• Again, data and MC agree

22

PbPb

PbPb


Jet energy scale and resolution

14

See: CMS-PAS-HIN-12-004

• Corrections derived from PYTHIA (pp) simulations• Resolution increases (N term – see later slide) by ~ 5 GeV with centrality• Energy scale shifted by ~2-1% due to subtraction of low-pT

• Dependence on other properties (parton-type, fragmentation) are examined to evaluate systematic uncertainties


Jet energy smearing

Unfolding vs smearing for data/MC comparison

21

For 30 GeV jets : C component ~ 0.7 GeV, S ~ 6.5 GeV, N ~ 5 GeV

For 120 GeV jets : C component ~ 3 GeV, S ~ 13 GeV, N ~ 5 GeV

From PLB 718 (2013) 773 : Photon events

Not exactly inclusive jet resolution


Modified PYTHIA quark fraction

Standard PYTHIA

Some systematic checks

20

Fully simulated events are studied for determination of systematics

MODIFIED PYTHIA: Parton fractions are selected/reweighted to bias the sample for a specific type of hard fragmentation

PYQUEN: Different tunes with Radiational/Collisional energy loss


Quark vs gluon jet responseCMS-PAS-HIN-11-004

29

JEC is done assuming PYTHIA quark/gluon fractionChange in this ratio (e.g. “gluon filtering”) will bias JEC


Dijet asymmetries


Dijet balance centrality evolution from 2011 data set

9

DIjet asymmetries


Fraction of matched jets

29


Dijet imbalance vs pT

11

• Dijets in reference (pp, PYTHIA) more balanced with increasing pT

• <pT,2/pT,1> in PbPb consistent with constant offset from reference MC


Caveats

• The order of cuts is important for defining measurement– e.g. rapidity cuts and leading jet selection don’t commute

• When smearing MC jets for data comparison, order of smearing, cuts, selection is critical– smearing may flip order of leading and subleading jet!!!

• Careful reading of papers, but probably also direct communication is often useful...

31


Impact of jet energy smearing

0-20%

Jet energy smearing

Swapped leading and subleading jet

Subleading jet out of acceptance

Generator level leading and subleading jets matches reco level


Gen-level folding vs full simulation

32

peripheral central


l Need realistic MC generator (for both jet and UE)l Iterative feedback cycle is very important (like PYTHIA v.s pp data

in high energy community)l CMS is willing to use and check if you offer a jet event generator

l To compare with CMS data: Need to include reconstruction effect properly

l Genjet à energy loss à apply smearing

Inputs to the MC discussion

Generator Experiment

Used to derive correction and compare with data

Feedback and improve the generator

Jet MC and data comparison


Photon-jet correlations


• Azimuthal decorrelation:|ΔφJγ|, and its parametrized width σ(|ΔφJγ|)

• Transverse momentum ratio:xJγ = pT

Jet/pT γ, and its mean ⟨xJγ⟩

• Fraction of photons with associated jets: RJγ

• pT γ > 60 GeV/c (to have sufficient xJγ phase space)

• pTJet > 30 GeV/c (constrained by efficiency)

pTγ

|ΔφJγ|

pTJet

Photon-Jet Observables

36


Photon Candidates

Anti-kT 0.3 pile-up subtracted

Jet candidates |η|<1.6

Z to eeCross check the

energy correction

Photon Candidates- AllJet

Super Clustersspike rejection

Passing loose photon IDH/E<0.1

Photon isolationEnergy correction

Shower shapeSignal likeσηη<0.01

Background clusters(Fakes from jet)

Shower shapeBackground like0.011<σηη<0.017


Selected Jet candidates

Selected Jet

Exclude photon candidate |ΔR| > 0.3 w/r to leading photon

Electron rejectionRemaining contribution

1-3%

Leading Cluster

Background template

Photon Candidates- All JetSignal region

Photon-Jet correlations in MIXED events

Event Mixing|Δφ| > 7π/8

PurityDetermination

Dijet Background subtractionPhoton- All JetSignal region Photon-Jet

Final result

PurityDetermination

Photon-Jet Analysis Flow

37


• SumIso = uncorrected Track + ECAL + HCAL ET in R < 0.4• GenIso = generator level particle energy in R < 0.4• Isolated prompt (non-decay) photons with SumIso < 1 GeV• Comparison to MC definition GenIso < 5 GeV• SumIso ≠ GenIso due to PbPb underlying event fluctuation

γ

Isolated Isolated Non-isolated

Leading order Higher orders

γ

γ

Final state not

differentiableISOLATED ISOLATED NON-ISOLATED

Isolated Photons

38


Background energysubtraction

Photon Isolation in PbPb

39


• Shower shapeσηη = ∑i

5×5 wi (ηi − η5×5)²/∑i5×5 wi

wi = max(0, c + ln Ei/E5×5)

• Signal photons selected by cutting on σηη < 0.01

• Decay photons contribution determined by fit of (signal + background templates in σηη

• Background σηη template found using data from photons failing the SumIso cuts

Rejection of Decay Photons

40


Jet Finding in PbPb

41

• Anti-kT particle flow jets, R = 0.3

• UE estimation/subtraction using φ-rings in η, excluding jet candidates (two iterations)

• Reconstruction > 90% efficient for pTJet > 30 GeV/c in

PbPb



42

Jets that fluctuated below threshold in first step of iteration(got added to UE estimate, leading to oversubtraction)


Photon Candidates




energy correction










Selected Jet



1-3%

Leading Cluster

Background template




PurityDetermination


Final result

PurityDetermination


43


Δφ

Photon-Jet

Photon–UE Combinatorics

Background Photon–UE Combinatorics

Background Photon–Jet

Signal region

Estimated from event mixing using min-bias data

Data, cf. Slide 13

Background Subtraction, Part I

44


Photon Candidates




energy correction










Selected Jet



1-3%

Leading Cluster

Background template




PurityDetermination


Final result

PurityDetermination


45


Statistical Subtraction

Δφ

Background Photon–Jet

Signal region

Estimated from shower shape sideband & purity

Photon-Jet

Background Subtraction, Part II

46

Double-subtraction yields photon-associated jet correlation


• Distribution is consistent with pp & PYTHIA tune Z2 + Hydjet

• To quantify the centrality dependence, peak region is fit with an empirical formula

47

Photon-Jet Angular Correlation

γJφΔ

0.0001 1.0473 2.0944 3.1416

Pair

Frac

tion

-210

-110

1PbPb DataPYTHIA + HYDJETpp Data

50% - 100%

(a)

0 π31 π3

2

γJφΔ γJ

φΔ0.0001 1.0473 2.0944 3.1416

Pair

Frac

tion

-210

-110

1=2.76TeV NNs

-1bµ L dt = 150 ∫30% - 50%

(b)

0 π31 π3

2

γJφΔ γJ

φΔ0.0001 1.0473 2.0944 3.1416

Pair

Frac

tion

-210

-110

1| < 1.44γη > 60GeV/c |

Tγp

| < 1.6Jetη > 30GeV/c |TJetp

10% - 30%

(c)

0 π31 π3

2

γJφΔ γJ

φΔ0.0001 1.0473 2.0944 3.1416

Pair

Frac

tion

-210

-110

1CMS Preliminary

0% - 10%

(d)

0 π31 π3

2 π

γJφΔ


Angular width σ(|ΔφJγ|) is consistent, both PbPb to pp and PbPb to PYTHIA tune Z2 + HYDJET

48

Photon-Jet Angular Correlation

partN0 100 200 300 400

) γJφΔ(

σ

0

0.050.1

0.15

0.2

0.250.3

0.35

0.4

0.450.5

(a)

CMS Preliminary

PbPb DataPYTHIA + HYDJETpp DataPYTHIA

| < 1.44γη > 60GeV/c |Tγp

| < 1.6Jetη > 30GeV/c |TJetp


• Momentum ratio shifts/decreases with centrality• Unitary normalized distribution, points anticorrelated• Red/blue boxes try to indicate possible, anticorrelated

systematic variation

49

Photon-Jet Momentum Balance



photon-jet momentum balance shifts in central events(relative to PYTHIA reference, calibrated in 7GeV pp)



Reasonable agreement between ATLAS and CMSNOTE: CMS correlates photon w/ associated jet, ATLAS w/ leading jet

partN0 100 200 300 400

γJR

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1CMS(b)

π87 >

γJφΔ

NOTE: ATLAS data scaled such that references match


• Azimuthal decorrelation:|ΔφJγ|, and its parametrized width σ(|ΔφJγ|)

• Transverse momentum ratio:xJγ = pT

Jet/pT γ, and its mean ⟨xJγ⟩

• Fraction of photons with associated jets: RJγ

• pT γ > 60 GeV/c (to have sufficient xJγ phase space)

• pTJet > 30 GeV/c (constrained by efficiency)

pTγ

|ΔφJγ|

pTJet

Example: photon-jet closure test

52


Example: photon-jet closure test

53

For this analysis, the effect of background fluctuations, UE background subtraction, jet

finding, photon isolation, background jets etc etc etc is fully accounted for by simple Gaussian

smearing of jet energy


Closure test for modified jets (PYQUEN)

54

For this analysis, the effect of background fluctuations, UE background subtraction, jet finding, photon

isolation, background jets etc etc etc is fully accounted for by simple Gaussian smearing of jet energy

True also for models that show a strong quenching effect (e.g. PYQUEN)


“Lost” energy


Missing-pT||

arXiv:1102.1957 [nucl-ex]

0-30% Central PbPb

balanced jets unbalanced jets

Missing pT||:

excess away from leading jet

excess towards leading jet

Calculate projection of pT on leading jet axis and average over selected tracks with pT > 0.5 GeV/c, |η| < 2.4

56


Missing-pT||


0-30% Central PbPb


Missing pT||:

excess away from leading jet

excess towards leading jet

Integrating over the whole event final state,momentum balance is restored


57


Missing-pT||


0-30% Central PbPb

in-cone

out-of-cone


In-ConeΔR<0.8

Out-of-ConeΔR<0.8


58


Missing-pT||


in-cone

out-of-cone

In-ConeΔR<0.8

Out-of-ConeΔR<0.8

Momentum balance in the event is carried by low momentum particles at large angles to jets


59


More differential measurement?

• 2011 data should allow more differential measurement – vs asymmetry, dR, jet pT

• Method requires uncorrelated UE to cancel to 0.1%– 10000 tracks x 0.5GeV pT each– desired resolution is ~5GeV in missing pT

• Tracking efficiency depends on– particle pT, mass, eta, phi– vertex location, event multiplicity, local multiplicity density– vicinity to jet, jet pT,...

• More differential measurement requires– improvement in tracking efficiency from ~70% to ~ 85-90%– improved jet UE subtraction to reduce jet-UE correlation

60


Summary


Jet analysis and systematic uncertainties

13


Fragmentation functions


Fragmentation functions• Distribution of associated track pT in cone, relative to

measured jet pT

– plot Z = pT, track/pT,jet cos(ΔR), ξ = ln(1/z) and track pT

– No direct connection to parton pT at scattering

•

64



• Distribution of associated track pT in cone, relative to measured jet pT

– plot Z = pT, track/pT,jet cos(ΔR), ξ = ln(1/z) and track pT

– No direct connection to parton pT at scattering

• “Associated tracks” are defined relative to track population in same-event displaced cone– ATLAS: “cone grid”; CMS: “eta-reflected cone”– No distinction between in-cone “associated tracks”

from medium response or from hard-parton fragmentation

65



pp smeared and reweighted to match PbPb jet pT spectrum


z

19

Jet fragmentation functions

z • The hard part of the fragmentation is slightly modified

• May affect calorimeter-related resolution• Effects can be estimated by

• Modified Pythia parton content• Various Pyquen tunes

• There appears to be an enhanced soft component

- May interfere with PU subtraction to affect energy scale- Bias towards upwards fluctuations in UE

• Effects can be estimated by• Embedding tracks into jets• Various Pyquen tunes

CMS-PAS-HIN-12-013


Fragmentation Function in Data

Pb Pb PbPb

• Particle Flow Jet Reconstruction– Anti kT, R=0.3– Same background subtraction as

calorimeter based jets– Fully efficient for pT > 40GeV/c– Good control of jet pT scale– Applied in pp and PbPb

• Dijet selection– pT

Jet1>100GeV/c– pT

Jet2>40GeV/c– Δφ12>2π/3

• Compare Leading and Subleading Jet– Select Tracks in ΔR=0.3 cone– pT> 4 GeV/c

68


b-tagging


Selection of b-jet Results from CMSIdentification of b-quark jets with the CMS experiment CMS Physics Analysis Summary BTV-11-004

Inclusive b-jet production in pp collisions at √s = 7 TeV JHEP 1204 (2012) 084, arXiv:1202.4617Measurement of BB Angular Correlations based on Secondary Vertex Reconstruction at √s = 7 TeV JHEP 1103 (2011) 136, arXiv:1102.3194Measurement of the b-jet to inclusive jet ratio in PbPb and pp collisions at √sNN = 2.76 TeV with the CMS detector CMS Physics Analysis Summary HIN-12-003

33

http://cdsweb.cern.ch/record/1427247/files/BTV-11-004-pas.pdf






http://arxiv.org/abs/arXiv:1202.4617




http://arxiv.org/abs/1102.3194





https://cdsweb.cern.ch/record/1472721/files/HIN-12-003-pas.pdf





Secondary Vertex Mass Fits

17

• After enriching sample in b-jets with the SSVHE tagger, we fit the SV mass distribution

• Shapes of b and non-b templates taken from MC, normalizations allowed to float

• The shapes of the non-b templates are cross-checked with a data-driven method

• The stability of the fits and the shapes of the templates are the dominant sources of systematics uncertainty )2Secondary vertex mass (GeV/c

0 1 2 3 4 5 6

Num

ber o

f jet

s10

210

310

410

510

)2Secondary vertex mass (GeV/c0 1 2 3 4 5 6

Num

ber o

f jet

s10

210

310

410

510PbPb datab-jet templatec-jet templateusdg-jet template

< 100 GeV/cT

80 < p

/NDF = 10.1 / 112χ

CMS Preliminary = 2.76 TeVNNs

b-tagged sample

(GeV/c)T

Jet p80 100 120 140 160 180 200

b-je

t fra

ctio

n0

0.01

0.02

0.03

0.04

0.05

0.06CMS Preliminary = 2.76 TeVs

-1 L dt = 231 nb∫pp DataPythiaSyst. uncertainty

(GeV/c)T

Jet p80 100 120 140 160 180 200

b-je

t fra

ctio

n0

0.01

0.02

0.03

0.04

0.05

0.06

(GeV/c)T

Jet p80 100 120 140 160 180 200

b-je

t fra

ctio

n

0

0.01

0.02

0.03

0.04

0.05

0.06CMS preliminary = 2.76 TeVNNs

Centrality 0-100%

-1bµ L dt = 150 ∫PbPb DataPythia+HydjetSyst. uncertainty

(GeV/c)T

Jet p80 100 120 140 160 180 200

b-je

t fra

ctio

n

0

0.01

0.02

0.03

0.04

0.05

0.06


b-Tagging Purity and Efficiency

18

Purity: b-jet fraction in SV tagged sample extracted from SV mass fit

Efficiency: Fraction of b-jets which are tagged by their SV

• Efficiency is extracted from simulation and with a data-driven method using the JP tagger, i.e., w/o requiring a SV

• For both efficiency and purity, MC is fairly close to data “out of the box”

PbPb pp

partN0 50 100 150 200 250 300 350 400

b-je

t fra

ctio

n

0

0.01

0.02

0.03

0.04

0.05

0.06CMS preliminary = 2.76 TeVNNs

< 100 GeV/cT

80 < Jet p

-1bµ L dt = 150 ∫PbPb DataPythia+HydjetSyst. uncertainty

50-100% 20-50% 0-20%

partN0 50 100 150 200 250 300 350 400

b-je

t fra

ctio

n

0

0.01

0.02

0.03

0.04

0.05

0.06


b-jet Fraction vs. Centrality

20

b-jet fraction does not show a strong centrality dependence


Summary

• Fantastic set of new results on jets from LHC and RHIC• In general, good consistency between experiments• Interface to theory needs to be refined:

– unfolding?– folding?

74


Underlying Event Subtraction (CMS)

η

φ

1. <ET> calculated in strips of η. Subtract <ET> + σ

η

φ

2. Run anti-kT algorithm on background-subtracted towers

η

φ

3. Exclude reconstructed jetsand re-estimate background

η

φ

4. Re-run anti-kT algorithm to get final jets

For details see:•CMS, arXiv:1102.1957•Kodolova et al., EPJC 50 (2007) 117



http://www.springerlink.com/content/jr1k6t53n02k4268/?MUD=MP

http://www.springerlink.com/content/jr1k6t53n02k4268/?MUD=MP


Jet finding performance: ATLAS vs CMS

76

jet reconstruction and jet physics in hi collisions with cmsheinz/jet2013/lecturenotes/roland... ·...

Documents