Pile-Up in Jets in ATLAS - Max Planck Societymenke/pdf/boost2013.pdfPile-Up in Jets in ATLAS BOOST...

Pile-Up in Jets in ATLASBOOST 2013 Sven Menke, MPP Munchen 12-16. Aug 2013, Flagstaff, AZ

on behalf of the ATLAS Collaboration

� Introduction• ATLAS and the LHC

• Pile-Up

• Jets

� Jet Area Method• transverse momentum density ρ

• area based pile-up correction

� Jet Vertex Fraction• track based pile-up suppression

� Jet Shapes� Conclusions

�

http://www.mppmu.mpg.de/~menke

http://www.mppmu.mpg.de/

http://atlas.ch

http://www.mppmu.mpg.de

ATLAS and the LHC

� LHC: pp collisions at√

s = 7TeV 2010 - 2011 and at√

s = 8TeV sinceMarch 2012

� several upgrade phases (energy and luminosity) are foreseen until 2023:

� end of 2012: ∼ 25 fb−1 pp collisions at 7-8TeV� Phase 0 (until 2019): ∼ 100 fb−1 pp collisions at 13/14TeV� Phase 1 (until 2022): ∼ 300 fb−1 pp collisions at 14TeV� Phase 2 (beyond 2023): ∼ 3000 fb−1 pp collisions at 14TeV

Year2010 2015 2020 2025 2030

]1

Inte

gra

ted L

um

inosity [fb

110

1

10

210

310

410

Ph

ase

0 S

hu

tdo

wn

Ph

ase

1 S

hu

tdo

wn

Ph

ase

2 S

hu

tdo

wn1

78

Te

V,

~2

5 f

b

11

4 T

eV

, ~

10

0 f

b

11

4 T

eV

, ~

30

0 f

b

11

4 T

eV

, ~

30

00

fb

S. Menke, MPP Munchen � Pile-Up in Jets in ATLAS � BOOST, 12-16. Aug 2013, Flagstaff, AZ 2


The ATLAS Detector

A Torodial LHC ApparatuS: 25m high; 44m long; 7000 t heavy



The ATLAS Detector � Inner Detector

� Precision Tracking up to |η| < 2.5 in Silicon Detectors (Pixel- andStrip-detectors)

� Immersed in 2TB-field

� 80.4× 106 pixelsensors

� 6.3× 106 siliconstrips

� 351× 103 strawtubes add trackpoints andmeasuretransitionradiation up to|η| < 2.0

� Tracking performance: σpT /pT = 0.038%pT (GeV)⊕ 1.5%



The ATLAS Detector � Calorimetry

� Calorimeter coverage up to |η| < 4.9

� EM: LAr/Pb Accordion |η| < 1.475/1.375 < |η| < 3.2

� Had Barrel:steel/scintillating tiles|η| . 1.7

� Had Endcap: LAr/Cuparallel plates1.7 . |η| . 3.2

� Forward: LAr/Cu(W)tubular electrodes3.2 . |η| . 4.9

� 3 to 7 samplings

� 188× 103 cells in total

� e/γ: σE/E ' 10%/√

E � jets: σE/E ' 50%/√

E ⊕ 3%



Pile-Up � important parameters• µ: average number of interactions per

bunch• NPV: number of rec. primary vertices∼ 44%µ + 3 at high µ

• ∆t : bunch distance50 ns in 2012

• signal integration time∼ 500 ns for the LAr

Mean Number of Interactions per Crossing

0 5 10 15 20 25 30 35 40 45

/0.1

]1

Record

ed L

um

inosity [pb

0

20

40

60

80

100

120

140

160

180 Online LuminosityATLAS

> = 20.7µ, <1Ldt = 21.7 fb∫ = 8 TeV, s

> = 9.1µ, <1Ldt = 5.2 fb∫ = 7 TeV, s

26.7 km ≡ 3560 bunch places each ∆t = 25 ns

Bunch ID

350 400 450 500 550 600 650 700 750 8000

0.2

0.4

0.6

0.8

1

Bunch ID

380 400 420 440 460 480 5000

0.2

0.4

0.6

0.8

1

Bunch ID

500 1000 1500 2000 2500 3000 35000

0.2

0.4

0.6

0.8

1

� typical LHC bunch structure in 2012• ∼ 1400 colliding bunch pairs• up to 11 trains ∼ 1000 ns appart• 2-4 sub-trains each ∼ 250 ns• 36 filled bunches

� Pile-up impact depends on bunch crossingNumber (BCID)� up to 10 colliding bunch pairs contribute tosignal



Pile-Up � Z0 → µµ candidate event with NPV = 25 from 2012



Jets� jets in ATLAS are made out of “topological clusters”• 3d energy blobs of neighboring calorimeter cells around seed cell

with |E | > 4σ• direct seed neighbors with |E | > 2σ become seeds too• re-clustering of this reduced cell set around local maxima|E| > 2σnoise |E| > 4σnoise 4/2/0 topological clusters

φ cos ×| θ|tan -0.05 0 0.05

φ s

in

×| θ|t

an

-0.05

0

0.05

210

310

410

510

FCal1C

φ cos ×| θ|tan -0.05 0 0.05

φ s

in

×| θ|t

an

-0.05

0

0.05

210

310

410

510

FCal1C

φ cos ×| θ|tan -0.05 0 0.05

φ s

in

×| θ|t

an

-0.05

0

0.05

210

310

410

510

FCal1C

� 2σ cut is removing cells from the signal region� 4σ cut shows seeds for the cluster maker� after clustering all cells in the signal regions are kept� cluster splitter finds hot spots



Jets � Noise Thresholds� Noise is σelec ⊕ σpile−up

• σelec relevant for no pile-up• σpile−up grows with

√µ

� a 20% increase in noise means 20× moreclusters for fixed thresholds!

� thresholds and filter weights are adjusted toexpected maximum 〈µ〉• modified weights slightly increase σelec and decreaseσpile−up from

√µ scaling

� average cell-level bias• f(BCID) . O(0.3σpile−up) corrected in 2012 t (s)

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

610×

Pile

Up S

ignal fo

r one b

unch tra

in

0.04

0.02

0

0.02

0.04 = 25nsBC t∆

= 50nsBC t∆

t [s]

0.2 0.3 0.4 0.5 0.6 0.7 0.8

610×

LA

r S

ignal [a

.u.]

0.2

0

0.2

0.4

0.6

0.8

1

LAr Ioniziation Current

LAr Readout Signal

bunch train

2010 µ = 0 2012 µ = 30 2022 µ = 200?

|η|

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Tota

l N

ois

e [

MeV

]

10

210

310

410FCal1

FCal2

FCal3

HEC1

HEC2

HEC3

HEC4

PS

EM1

EM2

EM3

Tile1

Tile2

Tile3

Gap

ATLAS

Simulation

= 0µ = 7 TeV, s

|η|

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Tota

l N

ois

e [

MeV

]

10

210

310

410

FCal1

FCal2

FCal3

HEC1

HEC2

HEC3

HEC4

PS

EM1

EM2

EM3

Tile1

Tile2

Tile3

Gap

ATLAS

Simulation

Preliminary

50 ns bunch spacing

= 30µ = 8 TeV, s

|η|

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

Tota

l N

ois

e [

MeV

]10

210

310

410

FCal1

FCal2

FCal3

HEC1

HEC2

HEC3

HEC4

PS

EM1

EM2

EM3

Tile1

Tile2

Tile3

Gap

ATLAS Preliminary

Simulation

50 ns bunch spacing

= 200µ = 14 TeV, s



Jet Area Method ATLAS-CONF-2013-083

� jet area Ajet (arXiv:0802.1188) measures jetsusceptibility to soft particles

2RπJet area/

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6

Nor

mal

ised

ent

ries

0

0.1

0.2

0.3

0.4

0.5 R = 0.4tanti-k

R = 0.6tanti-k

Cam/Aach R = 1.2

< 80 GeVjet

T p≤40

LCW Topo-clusterPythia Dijet 2012

ATLAS Simulation Preliminary

[GeV]ρ0 5 10 15 20 25 30

Nor

mal

ised

ent

ries

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

= 6PVN = 10PVN

= 14PVN = 18PVN


< 21⟩µ⟨ ≤20

= 8 TeVsPythia Dijet LCW TopoClusters

η-5 -4 -3 -2 -1 0 1 2 3 4 5

[GeV

]⟩ρ⟨

0

5

10

15

20

25

30 = 1⟩µ⟨ = 8⟩µ⟨ = 16⟩µ⟨ = 26⟩µ⟨

| < 4.9, R = 0.4η|ATLAS Simulation Preliminary

= 8 TeVsPythia Dijet 2012, LCW Topo-clusters

• by adding many soft particles(ghosts) during the jet forming

• Ajet = N jetg /νg

N jetg number of ghosts in jet

νg number density of ghosts in y − φ

� momentum density: ρjet = pjet⊥ /Ajet

� ρ = median{ρjet,i} measures event pile-upactivity (arXiv:0707.1378)• median suppresses bias from hard scatter• event-by-event follows pile-up fluctuations• kT jets with R = 0.4 for |ηjet| < 2.0

� less pile-up jets in forward region

� any jet in the event can then be corrected:pjet,corr⊥ = pjet

⊥ − Ajet × ρ• jet algorithm/size can differ from the one used for ρ


https://atlas.web.cern.ch/Atlas/GROUPS/PHYSICS/CONFNOTES/ATLAS-CONF-2013-083


Jet Area Method � Performance

� dependency of pjet⊥ on in-time pile-up (NPV)

• before and after jet area pile-up correction• in simulated di-jet events for 160 < p⊥ < 320GeV jets vs. |η|• for AntiKt jets with R = 0.4, R = 0.6 and R = 1.0

|η|0 0.5 1 1.5 2 2.5 3 3.5

[GeV

]P

VN∂

Tp∂

-1

0

1

2

3

4

5 < 320 GeVtrueT

p≤160

< 24PV7 < N LCW R = 0.4tAnti-k

= 8 TeVsPythia Dijet, = 21⟩µ⟨


Uncorrected

Corrected

|η|0 0.5 1 1.5 2 2.5 3 3.5

[GeV

]P

VN∂

Tp∂

-1

0

1

2

3

4

5 < 320 GeVtrueT

p≤160




Uncorrected

Corrected

|η|0 0.5 1 1.5 2 2.5 3 3.5

[GeV

]P

VN∂

Tp∂

-1

0

1

2

3

4

5 < 320 GeVtrueT

p≤160




Uncorrected

Corrected

� 0.5− 2GeV/vtx before correction with large dependency on R

� < 0.2GeV/vtx after jet-area correction� and no dependency on R



Jet Area Method � Residual Corrections� jet area method reduces mostly pile-up

from additional calorimeter clusters notbelonging to the hard scatter

� inside the jets belonging to the hardscatter the noise suppression is reduced –the pile-up overlaps with the signal

� this can be addressed by a residualcorrection after the jet area basedcorrection• linear fits to preco⊥ − ptrue⊥ vs. NPV (〈µ〉)• most relevant in forward direction

� dependency of pjet⊥ on in-time pile-up: NPV

for fixed 〈µ〉 and out-of-time pile-up: 〈µ〉for fixed NPV

• before, after jet area and after residual correction

• for AntiKt jets with R = 0.4

• small improvement for in-time pile-up

• large improvement in the forward region for out-of-timepile-up

|η|

0 0.5 1 1.5 2 2.5 3 3.5 4

[G

eV

]P

VN∂/

Tp∂

0.4

0.2

0

0.2

0.4

0.6

0.8

1ATLAS Simulation Preliminary

Pythia Dijets 2012

LCW R=0.4t

antik

Before any correctionA subtraction×ρAfter

After residual correction

|η|

0 0.5 1 1.5 2 2.5 3 3.5 4

[G

eV

]⟩

µ⟨∂/T

p∂

0.4

0.2

0

0.2

0.4

0.6

0.8


Pythia Dijets 2012

LCW R=0.4t

antik

Before any correctionA subtraction×ρAfter

After residual correction



Jet Area Method � p⊥ Resolution

� p⊥ resolution for jets in simulated di-jetevents• uncorrected

• offset corrected

• jet area corrected

� as a function of in-time pile-up, NPV

� as a function of out-of-time pile-up, 〈µ〉• for 20 GeV < ptrue⊥ < 30 GeV

• |η| < 2.4

� jet area method improves resolutionw.r.t. previous offset method

� still local pile-up fluctuations cause RMSto remain rising with NPV and 〈µ〉

PVN

5 10 15 20 25 30

) [G

eV

]tr

ue

T

pre

co

TR

MS

(p

5

6

7

8

9

10

11

12


=8 TeVsPythia Dijet

LCW R=0.6tantik

< 30 GeVtrue

T p≤20

| < 2.4η|

uncorrected

) correctionPV

, N⟩µ⟨f(

A correction×ρ

⟩µ⟨

5 10 15 20 25 30 35 40

) [G

eV

]tr

ue

T

pre

co

TR

MS

(p

5

6

7

8

9

10

11

12


=8 TeVsPythia Dijet

LCW R=0.6tantik

< 30 GeVtrue

T p≤20

| < 2.4η|

uncorrected

) correctionPV

, N⟩µ⟨f(

A correction×ρ



Jet Area � Jet Multiplicity

� pile-up subtraction with jet area method reduces amount of pile-upinside jets

� but also suppresses jets with pjet⊥ < ρ× Ajet

⟩µ⟨

0 5 10 15 20 25 30 35 40

> 2

0 G

eVT

, p⟩je

tN⟨

2

2.5

3

3.5

4

4.5

5

5.5

6

⟩µ⟨0 5 10 15 20 25 30 35 40

Dat

a/M

C

0.90.95

11.05

1.1

MC, No Correction

Data, No Correction

MC, Area Correction

Data, Area Correction

PreliminaryATLAS

+ jets µµ →Z

LCW R = 0.4tanti-k

2.1≤| η |≤0.0 � jet multiplicity as function of 〈µ〉in Z→ µµ+ jets events beforeand after pile-up correction• pile-up correction improves

data/MC agreement• reduced µ dependency

� still some jets remain due tolocalized pile-up fluctuations



Jet Vertex Fraction

� track-based observables can help to remove those jets� the Jet Vertex Fraction (JVF)

JVF(jet,PVj) =

∑k

p⊥(trackjetk ,PVj)∑n

∑l

p⊥(trackjetl ,PVn)

• counts fraction of p⊥-sum of alltracks associated to the jet and agiven vertex over thetrack-p⊥-sum for all vertices forthat jet

• properties of JVF assuming onehard vertex w.r.t. this vertex:� 0 for pile-up� 1 for the hard scatter signal� −1 for jets without associatedtracks



Jet Vertex Fraction � Performance

� Z→ ee + jets events to identify onehard PV with leptons

� check in MC the JVF distributionfor jets from the hard scatter andfrom pile-up

Jets JVF

1 0.5 0 0.5 1

Je

ts/0

.05

210

310

410

510

Pileup jets

Hardscatter jets


ee + jets→Z

, R=0.4, LC+JEStAntik

2.4≤|η<50 GeV, |T

p≤20

Jets JVF

-1 -0.5 0 0.5 1

Jets

/0.0

5

310

410 <15µMC <15µData

<25µ ≤MC 15<25µ ≤Data 15

25≥µMC 25≥µData

ATLAS Preliminary ee + jets→Z

, R=0.4, LCW+JEStAnti-k>20 GeV

T|<2.5, pη|

Jets JVF

-1 -0.5 0 0.5 1

Dat

a/M

C

0.60.8

11.21.4

� compare JVF in data and MC fordifferent µ• for larger µ the peak at JVF = 0

rises• MC reproduces data distributions

quite well



Jet Vertex Fraction � Performance

⟩µ⟨

0 5 10 15 20 25 30 35 40

> 2

0 G

eVT

, p⟩je

tN⟨

2

2.2

2.4

2.6

2.8

3

3.2

⟩µ⟨0 5 10 15 20 25 30 35 40

Dat

a/M

C

0.960.98

11.021.04

⟩µ⟨0 5 10 15 20 25 30 35 40

Dat

a/M

C

0.960.98

11.021.04

MC, No JVF Cut

Data, No JVF Cut

MC, |JVF| > 0.25

Data, |JVF| > 0.25

MC, |JVF| > 0.50

Data, |JVF| > 0.50

JVF Uncertainty

PreliminaryATLAS

+ jets µµ →Z

LCW+JES R = 0.4tanti-k

2.1≤| η |≤0.0

� use JVF to suppress pile-up:

� jet multiplicity for two |JVF | cuts inZ→ µµ+ jets data and MCcompared to no cut vs. 〈µ〉• |JVF| > 0.25

� reduces pile-up jet alreadysignificantly

• |JVF| > 0.5� the dependency on µ vanishes

• both lead to good data/MCagreement



Jet Shapes ATLAS-CONF-2013-85

� pile-up subtraction for jet shapes (arXiv:1211.2811) is an extension ofthe jet area correction approach beyond the p⊥ of the jet• first ingredient is the ghost’s 4-vectors

gµ = g⊥[cosφ, sinφ, sinhy , coshy ] with area Ag = 1/νg

• their addition during the jet forming change jet shapes V(ρ, g⊥)slightly

• by varying the ghost’s transverse momentum scale g⊥ the effect onany shape can be evaluated

• and eventually subtracted

� instead of subtracting the average ρ× Ajet from the jet p⊥ thecorresponding subtraction is done for the ghosts• measured shape: V(ρ = ρ0, g⊥ = 0)

• desired shape: V(ρ = 0, g⊥ = 0)

• correction principle: V(ρ+ δ, g⊥) = V(ρ, g⊥ + δ × Ag)

• evaluate: V(ρ = 0, g⊥ = 0) = V(ρ = ρ0, g⊥ = −ρ0 × Ag)

� calculated with first 3 terms of Taylor expansion


https://atlas.web.cern.ch/Atlas/GROUPS/PHYSICS/CONFNOTES/ATLAS-CONF-2013-085


Jet Shapes � Performance� several shapes tested in 2012 data and MC

• splitting scale√

dij = min(p⊥,i , p⊥,j )∆Rij , with the distance

of two subjets ∆Riji = 1, j = 2 for the last two sub-jets in jet forming

• N-subjettinessτN =

∑k p⊥,kmin(∆R1,k , ...,∆RN,k )/

∑k p⊥,k R, close

to 0 if the jet can be described by N or less sub-jets• ratios of τN , τij = τi/τj• energy-energy correlations (EEC) of the jet constituents,

C(β)1 =

(∑i<j p⊥,i p⊥,j (∆Rij )

β)/(∑

k p⊥,k)2

� examples for di-jet events� corrected shapes closer to truth

0 20 40 60 80 100 120 140

Jets

/ 3

.2 G

eV

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

610×

MC: Uncorrected

MC: Corrected

Data: Uncorrected

Data: Corrected

Truth-particle jets

ATLAS Preliminary -1 L dt = 20.3 fb∫=8 TeV, s

dijets (Herwig++) LCW jets with R=1.0

tanti-k

< 300 GeVjet

T p≤200

[GeV]12

d0 20 40 60 80 100 120 140

Data

/ M

C

0.60.8

11.21.4

0 0.1 0.2 0.3 0.4 0.5 0.6

Jets

/ 0

.014

0

10

20

30

40

50

60

70

80

310×

MC: Uncorrected

MC: Corrected

Data: Uncorrected

Data: Corrected

Truth-particle jets



tanti-k

< 300 GeVjet

T p≤200

1τ

0 0.1 0.2 0.3 0.4 0.5 0.6

Data

/ M

C0.60.8

11.21.40 0.2 0.4 0.6 0.8 1 1.2 1.4

Jets

/ 0

.032

0

10

20

30

40

50

60

70

310×

MC: Uncorrected

MC: Corrected

Data: Uncorrected

Data: Corrected

Truth-particle jets



tanti-k

< 300 GeVjet

T p≤200>0.01

1τcorr

21τ

0 0.2 0.4 0.6 0.8 1 1.2 1.4

Data

/ M

C

0.60.8

11.21.4

0 0.05 0.1 0.15 0.2 0.25

Je

ts /

0.0

06

0

0.02

0.04

0.06

0.08

0.1

0.12

0.146

10×

MC: Uncorrected

MC: Corrected

Data: Uncorrected

Data: Corrected

Truth-particle jets



tanti-k

< 300 GeVjet

T p≤200

(3)

1C

0 0.05 0.1 0.15 0.2 0.25

Data

/ M

C

0.60.8

11.21.4



Jet Shapes � µ Dependence� shapes before and after

correction for different µ

� for di-jet events with500GeV < p⊥ < 600GeV

� correction reduces µdependence and movescorrected data close to MCtruth

[GeV]12d

0 20 40 60 80 100 120 140

Arb

itra

ry N

orm

aliz

atio

n

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

Truth-particle jets

Uncorrected Data

9≤⟩µ⟨≤4

17≤⟩µ⟨≤12

31≤⟩µ⟨≤24



tanti-k

< 600 GeVjet

T p≤500

[GeV]12d

0 20 40 60 80 100 120 140

Arb

itra

ry N

orm

aliz

atio

n

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

Truth-particle jets

Corrected Data

9≤⟩µ⟨≤4

17≤⟩µ⟨≤12

31≤⟩µ⟨≤24



tanti-k

< 600 GeVjet

T p≤500

√

32τ0 0.2 0.4 0.6 0.8 1 1.2 1.4

Arb

itra

ry N

orm

aliz

atio

n

0

0.05

0.1

0.15

0.2

Truth-particle jets

Uncorrected Data

9≤⟩µ⟨≤4

17≤⟩µ⟨≤12

31≤⟩µ⟨≤24



tanti-k

< 600 GeVjet

T p≤500

leading jets > 30 GeV

jetcorr m

> 0.121

τcorr

32τ0 0.2 0.4 0.6 0.8 1 1.2 1.4

Arb

itra

ry N

orm

aliz

atio

n

0

0.05

0.1

0.15

0.2

Truth-particle jets

Corrected Data

9≤⟩µ⟨≤4

17≤⟩µ⟨≤12

31≤⟩µ⟨≤24



tanti-k

< 600 GeVjet

T p≤500

leading jets > 30 GeV

jetcorr m

> 0.121

τcorr

10 15 20 25 30

[G

eV

]⟩

12

d⟨

40

45

50

55

60

65

70MC: Uncorrected

MC: Corrected

Data: Uncorrected

Data: Corrected

Truth-particle jets



tanti-k

< 600 GeVjet

T p≤500

10 15 20 25 30

Data

/ M

C

0.90.95

11.05

1.1

⟩µ⟨10 15 20 25 30

0.80.9

11.11.2 MC Corrected / Truth-particle jets

Data Corrected / Truth-particle jets

√

� average shapes for same events vs. µ

� splitting scale√

d12 as typical example� also for the average shapes thecorrection removes µ dependence andimproves agreement with non pile-up truth



Conclusions

� Pile-up reached in 2012 unprecedented levels in ATLAS• even surpassed design levels at the beginning of some fills

� It will become even larger after the long shutdown� Methods to subtract and to suppress pile-up have been improved for the

2012 data� Corrections:• corrections are mainly based on jet areas and the median transverse

momentum density in the event• works as p⊥ subtraction for jets• and also with ghost subtraction approach for jet shapes

� Suppression:• direct suppression by adapting the calorimeters noise definition• track based suppression with the jet vertex fraction to identify jets

with mostly pile-up� similar methods also used for missing transvere momentum

measurements

S. Menke, MPP Munchen � Pile-Up in Jets in ATLAS BOOST, 12-16. Aug 2013, Flagstaff, AZ 21


Pile-Up in Jets in ATLAS - Max Planck Societymenke/pdf/boost2013.pdfPile-Up in Jets in ATLAS BOOST...

Documents

Transcript of Pile-Up in Jets in ATLAS - Max Planck Societymenke/pdf/boost2013.pdfPile-Up in Jets in ATLAS BOOST...