Upload
others
View
15
Download
0
Embed Size (px)
Citation preview
Gluon Tagging and Quark & Gluon SamplesHow well can we do at the 7TeV LHC?
Jason Gallicchio
Harvard
12 April 2011
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 1 / 70
Short Outline
Biggest Motivation: Reject Gluey LHC Backgrounds
Part 1: The Gluon Tagger
Part 2: Finding Pure Samples of Quark and Gluon Jets
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 2 / 70
Gluon Tagging Motivation
Other than “Wouldn’t it be nice to know?”
Most new physics gives quark rather than gluon jets:
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 3 / 70
Gluon Tagging Motivation
Other than “Wouldn’t it be nice to know?”
Most new physics gives quark rather than gluon jets:
8-jet Gluino event: pp → g̃g̃ and each g̃ → qq̄χ01 + qq̄χ0
2
Tagging is especially important without W , Z, γ, ℓ±, B-Tags, or /ET
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 3 / 70
Gluon Tagging Motivation
Other than “Wouldn’t it be nice to know?”
Most new physics gives quark rather than gluon jets:
8-jet Gluino event: pp → g̃g̃ and each g̃ → qq̄χ01 + qq̄χ0
2
Tagging is especially important without W , Z, γ, ℓ±, B-Tags, or /ET
Assignment: Your favorite example of gluon or non-b quark signals.
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 3 / 70
Motivation
Interesting standard model physics also tends to be quark-heavy
Tops (tt̄ → 4 or 6 quarks)
W ’s decaying hadronically (there’s no b-tag): W+ → ud̄ or cs̄
WW → 4 light quarks
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 4 / 70
Motivation
Interesting standard model physics also tends to be quark-heavy
Tops (tt̄ → 4 or 6 quarks)
W ’s decaying hadronically (there’s no b-tag): W+ → ud̄ or cs̄
WW → 4 light quarks
Vector Boson Fusion
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 4 / 70
Motivation
Interesting standard model physics also tends to be quark-heavy
Tops (tt̄ → 4 or 6 quarks)
W ’s decaying hadronically (there’s no b-tag): W+ → ud̄ or cs̄
WW → 4 light quarks
Vector Boson Fusion
The generic catchall: “Understand QCD”
Q & G Jets as backgrounds to boosted top, W , H, etc.
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 4 / 70
Motivation
Interesting standard model physics also tends to be quark-heavy
Tops (tt̄ → 4 or 6 quarks)
W ’s decaying hadronically (there’s no b-tag): W+ → ud̄ or cs̄
WW → 4 light quarks
Vector Boson Fusion
The generic catchall: “Understand QCD”
Q & G Jets as backgrounds to boosted top, W , H, etc.
NOTE!LEP found b-jets look more like gluon jets than light quark jets(in terms of size and particle count)
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 4 / 70
Motivation
Interesting standard model physics also tends to be quark-heavy
Tops (tt̄ → 4 or 6 quarks)
W ’s decaying hadronically (there’s no b-tag): W+ → ud̄ or cs̄
WW → 4 light quarks
Vector Boson Fusion
The generic catchall: “Understand QCD”
Q & G Jets as backgrounds to boosted top, W , H, etc.
NOTE!LEP found b-jets look more like gluon jets than light quark jets(in terms of size and particle count)
Eventually combine Gluon-Tagging with B-Tagging and τ -Tagging
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 4 / 70
But There’s a Lot of Glue to Get Stuck In
50 100 200 400 800 1600
Chance EACH Jet is Quark
0%
100%
80%
60%
40%
20%
2j3j
4j
pT Cut on All Jets (GeV)
So chance that all 4 jets & 100GeV are quark: (30%)4 = 0.8%
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 5 / 70
Outline
Biggest Motivation: Reject Gluey LHC Backgrounds
Part 1: The Tagger
Example Observables: Jet Mass and Charged Track CountEvaluating the power of observables: Background RejectionFamiliar presentation of ‘Jet Shape’ and its problemsMeasuring ‘size’ of jetsCombining 2 or more observables
Part 2: Finding Pure Samples of Quark and Gluon Jets
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 6 / 70
Disclaimer
There will be NO single way to unambiguously sort jets.
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 7 / 70
Disclaimer
There will be NO single way to unambiguously sort jets.
What’s different now than 20 years ago?
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 7 / 70
Disclaimer
There will be NO single way to unambiguously sort jets.
What’s different now than 20 years ago?
LHC calorimeter spatial resolution. (Sometimes 10x CDF or D0)
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 7 / 70
Disclaimer
There will be NO single way to unambiguously sort jets.
What’s different now than 20 years ago?
LHC calorimeter spatial resolution. (Sometimes 10x CDF or D0)
LHC is very jetty and pp is more gluey.
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 7 / 70
Disclaimer
There will be NO single way to unambiguously sort jets.
What’s different now than 20 years ago?
LHC calorimeter spatial resolution. (Sometimes 10x CDF or D0)
LHC is very jetty and pp is more gluey.
The most difficult signals are buried under multi-jet events.
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 7 / 70
Goal
Which observables are useful:
Individually?
When combined?
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 8 / 70
Goal
Which observables are useful:
Individually?
When combined?
Interesting variables can be:
Studied theoretically
Verified experimentally
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 8 / 70
Differences in Quarks vs Gluons
Color Charge: CF vs CA → jet mass, size, track count
Color Connections: 1 vs 2 → eccentricity and pull
Electrical Charge → charge-weighted track pT
Spin: 1/2 vs 1 → not explicitly used
...
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 9 / 70
Brief QCD Showering Theory
Gluon has a greater effective color charge (squared) than quark:
Gluon adjoint’s CA vs Quark fundamental’s CF
CA
CF=
9
4
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 10 / 70
Brief QCD Showering Theory
Gluon has a greater effective color charge (squared) than quark:
Gluon adjoint’s CA vs Quark fundamental’s CF
CA
CF=
9
4
Jet Mass in the small angle limit:
⟨
M2⟩
= Cαs
πp2T R2
where C ∼ CA for gluon jets, and ∼ CF for quark jets.
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 10 / 70
Jet Mass in Detail
Normalizing by pT (200GeV in this sample) generalizes better.
0 0.05 0.1 0.15 0.2 0.25 0.3 0.350
1
2
3
4
5
6
mass/Pt
QG
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 11 / 70
Evaluating the Observable: Sliding Cut
0 0.05 0.1 0.15 0.2 0.25 0.3 0.350
1
2
3
4
5
6
mass/Pt
QG
Sliding Cut
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 12 / 70
ROC Curve
0 0.05 0.1 0.15 0.2 0.25 0.3 0.350
1
2
3
4
5
6
mass/Pt
QG
Sliding Cut
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
ROC Curve for mass/Pt
Quark Signal Efficiency
Glu
on B
ackgro
und R
ejec
tion
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 13 / 70
ROC Curve
0 0.05 0.1 0.15 0.2 0.25 0.3 0.350
1
2
3
4
5
6
mass/Pt
QG
Sliding Cut
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
ROC Curve for mass/Pt
Quark Signal Efficiency
Glu
on B
ackgro
und R
ejec
tion
50/50
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 14 / 70
Other Jet Sizes and pT s
0 0.05 0.1 0.15 0.2 0.25 0.3 0.350
1
2
3
4
5
6
mass/Pt
QG
Sliding Cut0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
ROC Curve for mass/Pt
Quark Signal Efficiency
Glu
on B
ackgro
und R
ejec
tion
50/50
0.35
0.40
0.45
0.50
0.4 0.6 0.8 1.0
Bac
kgro
und R
ejec
tion
Jet Size 50 GeV
1600 GeV800 GeV400 GeV
100 GeV200 GeV
0.20
0.2
mass/Pt @ 80% Signal Efficiency
Rather than showing 10 ∗ 6 = 60 ROC Curves, pick 80% point on eachJason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 15 / 70
Brief Theory II
Gluon adjoint’s CA vs Quark fundamental’s CF
CA
CF=
9
4
Lore: quark jet first emits a gluon, and then it dominates the cascade.
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 16 / 70
Brief Theory II
Gluon adjoint’s CA vs Quark fundamental’s CF
CA
CF=
9
4
Lore: quark jet first emits a gluon, and then it dominates the cascade.
Multiplicity of any particle in a gluon jet should be CA/CF = 9/4times greater (confirmed at LEP).
〈Ng〉〈Nq〉
=CA
CF
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 16 / 70
Brief Theory II
Gluon adjoint’s CA vs Quark fundamental’s CF
CA
CF=
9
4
Lore: quark jet first emits a gluon, and then it dominates the cascade.
Multiplicity of any particle in a gluon jet should be CA/CF = 9/4times greater (confirmed at LEP).
〈Ng〉〈Nq〉
=CA
CF
σ2g
σ2q
=CA
CF.
(Calculated to N3LO by Capella, et al. hep-ph/9910226)
No detector simulation, but require charged particles pT > 500MeV.
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 16 / 70
Charged Particles Count
...this old favorite does quite well at high pT :
0 5 10 15 20 25 30 35 400
1
2
3
4
5
6
7
components_jet0_ak05_charged_countcomponents_jet0_ak05_charged_count
0.4 0.6 0.8 1.0
0.45
0.50
0.55
0.60
0.65
0.70
50 GeV
1600 GeV
800 GeV
400 GeV
100 GeV
200 GeV
Bac
kgro
und R
ejec
tion
Jet Size
Charged Track Count @ 80% Signal Efficiency
Higher pT means more charged tracks and more ‘time’ to establishCA/CF .
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 17 / 70
Visual Differences
Accumulate 3 million back-to-back dijet events
Quark Jets Gluon Jets
(Same total amount of pT , which is hidden by logarithmic color bands.)
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 18 / 70
Jet Shape
r
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 19 / 70
Jet Shape
r
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 20 / 70
Jet Shape Distribution vs Average
0 0.2 0.4 0.6 0.8 10
1
2
3
4
5
Integrated Jet Shape out to r = 0.1 for 100 GeVIntegrated Jet Shape out to r = 0.1 for 100 GeV
Distribution is not narrow gaussian around averageCorrelations between bins is also useful
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 21 / 70
Radial Moment – a measure of the “girth” of the jet
Weight pT deposits by distance from jet center
Radial Moment, or Girth : g =∑
i∈jet
piTpjetT
|ri|
0 0.1 0.2 0.3 0.4 0.5 0.60
1
2
3
4
5
6
7
radial momentradial moment
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 22 / 70
Angularities
Jet Angularities : Aa =∑
i∈jet
Ei fa(θ)
fa(θ) = sinaθ (1− cos θ)1−a with θ =π|ri|2R
or just θ
a=+1.9
a=-3a=0
Π
4
Π
2
Π Θ
2 R
1
1.5
fa
Normalization? none, mass, jet energy, jet pT .Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 23 / 70
Angularities
-3 -2 -1 1 2a
0.50
0.55
0.60
0.65
0.70
0.75
Angularity vs a for different Jet Sizes
Small R=0.2 Jets (red) perform better than Large R=1.0 Jets (pink)
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 24 / 70
Radial Moments and Their Kernels
Linear Moment Quadratic Moment
Tapered Moment Tapered Radial Kernel
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 25 / 70
Optimal Kernel
0.1 0.2 0.3 0.4 0.5 0.6∆R
-0.2
0.2
0.4
0.6
0.8
1.0
Optimal Kernel for 200 GeV R=0.7
0 0.1 0.2 0.3 0.4 0.5 0.60
1
2
3
4
5
Optimal Kernel DistributionOptimal Kernel Distribution
Positive kernel weights mean gluon-like.
Overall vertical shift or scaling leads to same distribution.
Quarks have most of their pT near the center.
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 26 / 70
Optimal Kernel for different pT ’s and R = 1.0
1600GeV → 50GeV
0.2 0.4 0.6 0.8 1.0
-0.5
0.5
1.0
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 27 / 70
Types of Variables
The menu, including varying jet size
Distinguishable particles/tracks/subjets
multiplicity, 〈pT 〉, σpT, 〈kT 〉,
charge-weighted pT sum
Moments
mass, girth, broadeningangularitiesoptimal kernel2D: pull, planar flow
Subjet properties
Multiplicity for different algorithms and Rsub
First subjet’s pT , 2nd’s pT , etc.Each subjet’s massSplitting kT scale
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 28 / 70
Best Variables in Each Category
Signal eff0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
LHC 200 : Background RejectionLHC 200 : Background Rejection
charged ct R=0.5
subjet ct Rsub=0.1
mass/Pt R=0.3
cos moment R=0.5
ang a=1 R=0.5
girth R=0.5
1st subjet pT
decluster kT
optimal R=1.0
pull η R=0.3
—pull— R=0.3
planar flow R=0.3
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 29 / 70
Best Variables in Each Category
Signal eff0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 11
10
210
310
LHC 200 : Background RejectionLHC 200 : Background Rejection charged ct R=0.5
subjet ct Rsub=0.1
mass/Pt R=0.3
cos moment R=0.5
ang a=1 R=0.5
girth R=0.5
1st subjet pT
decluster kT
optimal R=1.0
pull η R=0.3
—pull— R=0.3
planar flow R=0.3
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 30 / 70
Story So Far...
Keep only 40% of gluons while keeping 80% quarks
For multiple jets, these fractions exponentiate
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 31 / 70
Story So Far...
Keep only 40% of gluons while keeping 80% quarks
For multiple jets, these fractions exponentiate
Where to put the Cut? 80%?How do we rank the variables?
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 31 / 70
Story So Far...
Keep only 40% of gluons while keeping 80% quarks
For multiple jets, these fractions exponentiate
Where to put the Cut? 80%?How do we rank the variables?
Always demand 80% quark, see how much glue you can reject?
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 31 / 70
Story So Far...
Keep only 40% of gluons while keeping 80% quarks
For multiple jets, these fractions exponentiate
Where to put the Cut? 80%?How do we rank the variables?
Always demand 80% quark, see how much glue you can reject?
Optimize S/B?
Optimize S/√B?
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 31 / 70
Cutting and S/√B
Cutting gives some signal efficiency and some background efficiency.
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 32 / 70
Cutting and S/√B
Cutting gives some signal efficiency and some background efficiency.
If you start with S signal events and B background events,
S
B→ Sǫs
Bǫb
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 32 / 70
Cutting and S/√B
Cutting gives some signal efficiency and some background efficiency.
If you start with S signal events and B background events,
S
B→ Sǫs
Bǫb
Improvement in statistical significance scales differently
σ =S√B
→ Sǫs√Bǫb
= σǫs√ǫb
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 32 / 70
Cutting and S/√B
Cutting gives some signal efficiency and some background efficiency.
If you start with S signal events and B background events,
S
B→ Sǫs
Bǫb
Improvement in statistical significance scales differently
σ =S√B
→ Sǫs√Bǫb
= σǫs√ǫb
Cutting can improve significance only to a point...
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 32 / 70
Significance Improvements for Best Single Variables
Signal eff0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
2.2
LHC 200 : SignificanceLHC 200 : Significance
charged ct R=0.5
subjet ct Rsub=0.1
mass/Pt R=0.3
cos moment R=0.5
ang a=1 R=0.5
girth R=0.5
1st subjet pT
decluster kT
optimal R=1.0
pull η R=0.3
—pull— R=0.3
planar flow R=0.3
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 33 / 70
Combining Variables: Ex. Girth vs Charged Count
0
50
100
150
200
250
300
0 2 4 6 8 10 12 14 16 18 20 22 240
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
QuarkQuark
Charged Count
Girth
0
10
20
30
40
50
60
70
80
90
0 2 4 6 8 10 12 14 16 18 20 22 240
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
GluonGluon
Charged Count
Girth
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 34 / 70
Combining Variables: Ex. Girth vs Charged Count
0
50
100
150
200
250
300
0 2 4 6 8 10 12 14 16 18 20 22 240
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
QuarkQuark
Charged Count
Girth
0
10
20
30
40
50
60
70
80
90
0 2 4 6 8 10 12 14 16 18 20 22 240
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
GluonGluon
Charged Count
Girth
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 2 4 6 8 10 12 14 16 18 20 22 240
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
Likelihood: q/(q + g)Likelihood: q/(q + g)
Charged Count
RadialM
om
ent
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 34 / 70
ROC Curves for Best Combinations
Signal eff0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
LHC 0100 : Background RejectionLHC 0100 : Background Rejection
1: optimal kernel R=0.31: angularity a=+11: optimal kernel R=0.42: optimal and girth2: a=+1 and <kT >2: optimal and charged count
3:3:3:
444
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 35 / 70
Final Results: Best Pairs
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 11
10
210
310
20%eff
50 GeV, QT
p
80%eff
50 GeV, QT
p
20%eff
100 GeV, QT
p
80%eff
100 GeV, QT
p
20%eff
200 GeV, QT
p
80%eff
200 GeV, QT
p
20%eff
400 GeV, QT
p
80%eff
400 GeV, QT
p
20%eff
800 GeV, QT
p
80%eff
800 GeV, QT
p
20%eff
1600 GeV, QT
p
80%eff
1600 GeV, QT
p
Gluon Background RejectionGluon Background Rejection
Quark Signal EfficiencyJason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 36 / 70
Final Results: Best Pairs
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.5
1
1.5
2
2.5
3
3.5
4
20%eff
50 GeV, QT
p
80%eff
50 GeV, QT
p
20%eff
100 GeV, QT
p
80%eff
100 GeV, QT
p
20%eff
200 GeV, QT
p
80%eff
200 GeV, QT
p
20%eff
400 GeV, QT
p
80%eff
400 GeV, QT
p
20%eff
800 GeV, QT
p
80%eff
800 GeV, QT
p
20%eff
1600 GeV, QT
p
80%eff
1600 GeV, QT
p
Significance ImprovementSignificance Improvement
Quark Signal EfficiencyJason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 37 / 70
Summary of Gluon Tagging
Can reject 80% of gluons while keeping 80% quarksCan reject 95% of gluons while keeping 50% quarks
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 38 / 70
Summary of Gluon Tagging
Can reject 80% of gluons while keeping 80% quarksCan reject 95% of gluons while keeping 50% quarks
We can improve S/√B on quark-only signals by a factor of 2x-3x.
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 38 / 70
Summary of Gluon Tagging
Can reject 80% of gluons while keeping 80% quarksCan reject 95% of gluons while keeping 50% quarks
We can improve S/√B on quark-only signals by a factor of 2x-3x.
For 4-jets, all of these get raised to the 4th for glue-heavy background.Differences in these variables are worth verifying & calculating.
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 38 / 70
Summary of Gluon Tagging
Can reject 80% of gluons while keeping 80% quarksCan reject 95% of gluons while keeping 50% quarks
We can improve S/√B on quark-only signals by a factor of 2x-3x.
For 4-jets, all of these get raised to the 4th for glue-heavy background.Differences in these variables are worth verifying & calculating.
TODO:
Real Signals and Backgrounds (e.g. SUSY, WW → 4jet)
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 38 / 70
Summary of Gluon Tagging
Can reject 80% of gluons while keeping 80% quarksCan reject 95% of gluons while keeping 50% quarks
We can improve S/√B on quark-only signals by a factor of 2x-3x.
For 4-jets, all of these get raised to the 4th for glue-heavy background.Differences in these variables are worth verifying & calculating.
TODO:
Real Signals and Backgrounds (e.g. SUSY, WW → 4jet)
Use tagger to measure Q/G fraction (e.g. W+jets)
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 38 / 70
Summary of Gluon Tagging
Can reject 80% of gluons while keeping 80% quarksCan reject 95% of gluons while keeping 50% quarks
We can improve S/√B on quark-only signals by a factor of 2x-3x.
For 4-jets, all of these get raised to the 4th for glue-heavy background.Differences in these variables are worth verifying & calculating.
TODO:
Real Signals and Backgrounds (e.g. SUSY, WW → 4jet)
Use tagger to measure Q/G fraction (e.g. W+jets)
Encourage Atlas and CMS to squeeze out the most spatialresolution that they can.
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 38 / 70
Summary of Gluon Tagging
Can reject 80% of gluons while keeping 80% quarksCan reject 95% of gluons while keeping 50% quarks
We can improve S/√B on quark-only signals by a factor of 2x-3x.
For 4-jets, all of these get raised to the 4th for glue-heavy background.Differences in these variables are worth verifying & calculating.
TODO:
Real Signals and Backgrounds (e.g. SUSY, WW → 4jet)
Use tagger to measure Q/G fraction (e.g. W+jets)
Encourage Atlas and CMS to squeeze out the most spatialresolution that they can.
Integrate these variables into FastJet or SpartyJet.
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 38 / 70
Outline
Biggest Motivation: Reject Gluey LHC Backgrounds
Part 2: The Gluon Tagger
Part 2: Finding Pure Samples of Quark and Gluon Jets
Goal: High Purity with a high Cross SectionUse madgraph tree-level samplesFind kinematic variables to cut on2D combinations and/or Multivariate Techniques99% Quark purity from γ+2jets95% Gluon purity from 3jetsarXiv:1104.1175 [hep-ph] with Matt Schwartz
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 39 / 70
Detailed Multi-Jet Composition
50 100 200 400 800 1600
2 Jets
0%
100%
80%
60%
40%
20%
GG
QG
pT Cut on All Jets (GeV)50 100 200 400 800 1600
0%
3 Jets
100%
80%
60%
40%
20%
GGG
QGG
QQG
QQQ
pT Cut on All Jets (GeV)
Result of Parton Distribution Functions
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 40 / 70
X+2jet Composition
50 100 200 400 800 1600
QG
GG
0%
100%
80%
60%
40%
20%
γ+2jets
pT Cut on All Jets (GeV)50 100 200 400 800 1600
0%
100%
80%
60%
40%
20%
b+2jets
QG
GG
pT Cut on All Jets (GeV)
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 41 / 70
Starting Samples
50 100 200 400 800 1600
Chance EACH Jet is Quark
0%
100%
80%
60%
40%
20%
2j
Z/W+2j
Z/W+1j
γ+1j
3j
4j b+2j
bb+j
γ+2j
b+j
pT Cut on All Jets (GeV)
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 42 / 70
Cro
ss S
ection
(p
b)
10-6
10-3
1
103
106
50 100 200 400 800 1600
2j
2j1jZ+
2j1jW+
2j1jγ+
3j
4j
b+j
b+2j
bb+j
2j
Z+2jZ+j
W+2j
W+jγ+j
γ+2j
b+j
b+2j
bb+j
3j
4j
pT Cut on All Jets (GeV)
tt-
tt-
semi-leptonic
100%10-3
Cro
ssSec
tion
inpb
200 GeV Quark Purity
Quark Purity (zoom)
40% 50% 60% 70% 80% 90% 100
1
103
2jet
3jet hardest
harder softer
softerZ+2jet
γ+1jet
harder
softerW+2jet
harder
Z+1jet
W+1jet
γ+2jet
Quark Purification in γ+1jet
-2 -1 0 1 20
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
2.2
Quark
Gluon
ηγ
200 250 300 350 4000
1
2
3
4
5
6
Quark
Gluon
pjT
Kinematics are too similar to do much:
gq → γq: Quark signal
qq̄ → γg: Gluon background
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 45 / 70
Quark Purification in γ+2jet: Look at Softer Jet
2 2.5 3 3.5 4 4.5 50
0.5
1
1.5
2
2.5
3
3.5
4
Quark
Gluon
∆Rγj1 (to harder jet)
0.5 1 1.5 2 2.5 3 3.5 4 4.50
1
2
3
4
5
6
Quark
Gluon
∆Rγj2 (to softer jet)
When the softer jet is quark, The photon is often radiated off of it,rather then the harder jet.
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 46 / 70
Quark Purification in γ+2jet: Look at Softer Jet
-2 -1 0 1 20
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
Quark
Gluon
ηγ
-2 -1 0 1 20
0.5
1
1.5
2
2.5Quark
Gluon
ηj1
Other useful kinematics (?)
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 47 / 70
Quark Purification in γ+2jet: Look at Softer Jet
2D version of the same
0
50
100
150
200
250
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
Quark
η j1
ηγ0
10
20
30
40
50
60
70
80
90
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
Gluon
η j1
ηγ
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 48 / 70
Quark Purification in γ+2jet: Look at Softer Jet
2D version of the same
0
50
100
150
200
250
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
Quark
η j1
ηγ0
10
20
30
40
50
60
70
80
90
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
Gluon
η j1
ηγ0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
Likelihood
η j1
ηγ
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 48 / 70
Quark Purification in γ+2jet: Look at Softer Jet
Approximating the Likelihood Contours with f(x, y) = xy
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
Likelihood
η j1
ηγ -2 -1 0 1 2
-2
-1
0
1
2
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 49 / 70
Quark Purification in γ+2jet: Look at Softer Jet
Approximating the Likelihood Contours with f(x, y) = xy
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
Likelihood
η j1
ηγ -2 -1 0 1 2
-2
-1
0
1
2
-4 -2 0 2 4 60
2
4
6
8
10Quark
Gluon
ηγ ηj1
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 49 / 70
Quark Purification in γ+2jet: Look at Softer Jet
Do it again to find an even better combination
-4 -2 0 2 4 6 8 100
1
2
3
4
5
6
7
Quark
Gluon
ηγ ηj1 +∆Rγj2
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 50 / 70
Automating the process with Boosted Decision Trees
Some totally crazy illustrative example from my Higgs+Z work:
0
1000
2000
3000
4000
5000
6000
-3 -2 -1 0 1 2 3
-4
-2
0
2
4
6
SignalSignal
∆θb1,ℓ1
∆ηb2,ℓ1
0
200
400
600
800
1000
1200
1400
-3 -2 -1 0 1 2 3
-4
-2
0
2
4
6
BackgroundBackground
∆θb1,ℓ1
∆ηb2,ℓ1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
-3 -2 -1 0 1 2 3
-4
-2
0
2
4
6
LikelihoodLikelihood
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 51 / 70
Automating the process with Boosted Decision Trees
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 52 / 70
Do Not Fear Boosted Decision Trees
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
-3 -2 -1 0 1 2 3
-4
-2
0
2
4
6
BDT 2BDT 2
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
-3 -2 -1 0 1 2 3
-4
-2
0
2
4
6
BDT 8BDT 8
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
-3 -2 -1 0 1 2 3
-4
-2
0
2
4
6
BDT 32BDT 32
-0.6
-0.4
-0.2
0
0.2
0.4
-3 -2 -1 0 1 2 3
-4
-2
0
2
4
6
BDT 64BDT 64
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
-3 -2 -1 0 1 2 3
-4
-2
0
2
4
6
BDT 256BDT 256
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
-3 -2 -1 0 1 2 3
-4
-2
0
2
4
6
LikelihoodLikelihood
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 53 / 70
100%10
-3
Cro
ssSecti
onin
pb
200 GeV Quark Purity
Quark Purity
40% 50% 60% 70% 80% 90%
1
103
2jet
3jet hardest
γ+2jet harderγ+2jet softer
softerZ+2jet
γ+1jetharder
Quark Purity
70% 80% 90% 100%
1
103
10-3
10-6
Cro
ssSec
tion
inpb
Quark Purity for Different pT
50 γjj
100 γjj200 γjj
400 γjj
800 γjj
1600 γjj
50 γj
100 γj
200 γj
400 γj
800 γj
1600 γj
10-1
Cross
Sectionin
pb
200 GeV Gluon Purity
Gluon Purity
2jet
b+jet
3jet softest
3jet hardest
b+2jet hardest b+2jet softest
3jet mid
50% 60% 70% 80% 90% 100%
1
101
102
103
104
0% 20% 40% 60% 80% 100%
1
103
106
10-3
10-6
Cro
ssSec
tion
inpb
Best Samples for Gluon Purity
Gluon Purity
50 jj
100 jj
200 jj
400 jj
800 jj
1600 jj
50 jjj100 jjj
200 jjj
400 jjj
800 jjj
50 bjj
100 bjj
200 bjj
400 bjj
800 bjj
Summary of Finding Samples
Quark samples at 99% purity for γ+jet
Gluon samples at 90%-95% purity for 3jets
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 58 / 70
Summary of Finding Samples
Quark samples at 99% purity for γ+jet
Gluon samples at 90%-95% purity for 3jets
Gluon samples at 95%-99% purity for b+2jets with perfectB-Tagging and B-Anti-Tagging
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 58 / 70
Summary of Finding Samples
Quark samples at 99% purity for γ+jet
Gluon samples at 90%-95% purity for 3jets
Gluon samples at 95%-99% purity for b+2jets with perfectB-Tagging and B-Anti-Tagging
Now go forth and use these tools for good.... Thanks!
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 58 / 70
Backup
In case waving my hands proves insufficient ...
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 59 / 70
History
Number of charged tracks
Jet Shape (shown)
LEP e+e− → Zg → bb̄g
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 60 / 70
Event Generation
180 185 190 195 200 205 210 215 2200
0.5
1
1.5
2
2.5
3
3.5
ak05_Pt : ak05_Pt > 180 && ak05_Pt < 220
Entries 167796
Mean 196.8
RMS 11.29
ak05_Pt : ak05_Pt > 180 && ak05_Pt < 220
100 150 200 250 3000
2
4
6
8
10
hard_Pt : ak05_Pt > 180 && ak05_Pt < 220
Entries 167796
Mean 188.9
RMS 28.65
hard_Pt : ak05_Pt > 180 && ak05_Pt < 220
Jet pT : anti-kT R=0.5 Hard Parton pT
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 61 / 70
Moment of Inertia / Covariance / Eccentricity
Covariance Tensor: C =∑
i∈jet
piTpjetT
(
∆ηi∆ηi ∆ηi∆φi
∆φi∆ηi ∆φi∆φi
)
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 62 / 70
Moment of Inertia / Covariance / Eccentricity
Covariance Tensor: C =∑
i∈jet
piTpjetT
(
∆ηi∆ηi ∆ηi∆φi
∆φi∆ηi ∆φi∆φi
)
Combination of Eigenvalues
Girth: g =√a2 + b2
Determinant: det = a · bRatio: b/a
Eccentricity: ǫ =√a2 − b2
Planar Flow: pf = 4ab(a+b)2
Orientation: θ
Not useful for Q vs G: first emission sets this shape, and has similar2-body kinematics.
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 62 / 70
Subjets – Smaller is Better
Subjet Algorithm: anti-kT , CA, kTSubjet Size: Darkest is Rsub = 0.1, lightest Rsub = Rjet
R jet
60
70
80
90
100
Grejection
components_jet0_ak_sub__count
(Background Rejection at 50% Quark vs Initial Jet Size)Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 63 / 70
Explosion of Variables
Explosion of Variables
Different Jet sizes (R = 0.1, 0.2, 0.4, 0.7, 1.0, 1.4, ...)
Different Jet definitions (anti-kt, kt, CA, SisCone)
Different Generators: Pythia vs Herwig
Different Samples: Dijet vs γ+jet vs 8-Jets
Different Subjet sizes and types
Different Powers in the various moments
Charged Tracks or Calorimeter deposits?
And different variables are better for different Jet pT ranges.
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 64 / 70
Jet Broadening
Jet Broadening similar to linear moment for small-angles: kT ≈ pT r
Bjet =
∑
i |~pi × n̂jet|∑
i |~pi|=
∑
i |~kT i|∑
i |~pi|
0 0.05 0.1 0.15 0.2 0.250
2
4
6
8
10
12
14
16
components_jet0_ak05_particles_broadening Q 0050
G 0050
Q 0100
G 0100
Q 0200
G 0200
Q 0400
G 0400
Q 0800
G 0800
Q 1600
G 1600
components_jet0_ak05_particles_broadening
-12 -10 -8 -6 -4 -2 00
1
2
3
4
5
6
7
8
9
components_jet0_ak05_particles_broadening_beta Q 0050
G 0050
Q 0100
G 0100
Q 0200
G 0200
Q 0400
G 0400
Q 0800
G 0800
Q 1600
G 1600
components_jet0_ak05_particles_broadening_beta
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 65 / 70
Charge Weighted by pT
-100 -50 0 50 1000
0.5
1
1.5
2
2.5
3
3.5
4
4.5
50GeV50GeV
-200 -150 -100 -50 0 50 100 150 2000
0.5
1
1.5
2
2.5
200GeV200GeV
-300 -200 -100 0 100 200 3000
0.5
1
1.5
2
2.5
400GeV400GeV
-1000 -500 0 500 10000
0.5
1
1.5
2
2.5
1600GeV1600GeV
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 66 / 70
Compare to ATLAS B-Taggnig
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 67 / 70
0% 20% 40% 60% 80% 100%
1
103
106
10-3
10-6
CrossSectioninpb
Softest Trijet Gluon Purity
Gluon Purity
50 GeV
100 GeV
200 GeV
400 GeV
800 GeV
Gluon Purification: Lesson about Harsh Cuts
-2 -1 0 1 20
0.5
1
1.5
2
2.5
Quark
Gluon
ηj3
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 69 / 70
Gluon Purification: Lesson about Harsh Cuts
-2 -1 0 1 20
0.5
1
1.5
2
2.5
Quark
Gluon
ηj3
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.5 1 1.5 2 2.50
0.5
1
1.5
2
2.5
3
3.5
4
4.5
Likelihood
|ηj3 |
|ηj1−
ηj2|
-4 -3 -2 -1 0 1 20
0.5
1
1.5
2
2.5
3
3.5
4
Quark
Gluon
|ηj3 | − |ηj1 − ηj2 |
Jason Gallicchio (Harvard) Gluon Tagging and Quark & Gluon Samples 12 April 2011 69 / 70
0% 20% 40% 60% 80% 100%
1
103
106
10-3
10-6
Cro
ssSec
tion
inpb
Trijet Sample with Different Kinematic Cuts
Gluon Purity
BDT
800 jjj
400 jjj
200 jjj
100 jjj
50 jjj
η j3|η
3
η1 −η
2
j
j
j |
|−|