Quark vs Gluon Jet Tagging Francesco Rubbo APS DPF 2017 1 q/g - - PowerPoint PPT Presentation

▶
quark vs gluon jet tagging
SMART_READER_LITE
LIVE PREVIEW

Quark vs Gluon Jet Tagging Francesco Rubbo APS DPF 2017 1 q/g - - PowerPoint PPT Presentation

Mar 19, 2024 35 likes •359 views

Quark vs Gluon Jet Tagging Francesco Rubbo APS DPF 2017 1 q/g discrimination hep-ph/1106.3076 For background (gluon) suppression - e.g. VBF, SUSY, For final state reconstruction - e.g hadronic top or W Requires detailed

slide-1
SLIDE 1

Quark vs Gluon Jet Tagging

1

Francesco Rubbo APS DPF 2017

slide-2
SLIDE 2

2

q/g discrimination

hep-ph/1106.3076

  • For background (gluon)

suppression - e.g. VBF, SUSY, …

  • For final state reconstruction -

e.g hadronic top or W

  • Requires detailed

understanding of jet fragmentation properties

  • Probes generator PS/tunes

Many observables exploiting different fragmentation properties are useful for q/g discrimination

slide-3
SLIDE 3

3

q/g tagging with jet images

hep-ph/1612.01551 red = transverse momenta of charged particles green = transverse momenta of neutral particles blue = charged particle multiplicity RGB jet images: represent the jet with multiple images, exploiting tracking information, in addition to calorimeter images.

slide-4
SLIDE 4

4

Outline

  • Performance and uncertainties for quark/gluon tagging

with charged particle multiplicity

  • First look quark/gluon tagging with ATLAS jet images

ATL-PHYS-PUB-2017-009 ATL-PHYS-PUB-2017-017

slide-5
SLIDE 5

5

ntrack

  • A simple observable: number of tracks ghost-matched to the jet (ntrack)
  • Proxy for charged multiplicity (ncharged) in quark/gluon fragmentation
  • Particle multiplicity scales

with the color charge (CF/CA)

  • —> powerful q/g discriminant

ATL-PHYS-PUB-2017-009

slide-6
SLIDE 6

6

ntrack

  • A simple observable: number of tracks ghost-matched to the jet (ntrack)
  • Proxy for charged multiplicity (ncharged) in quark/gluon fragmentation
  • Particle multiplicity scales

with the color charge (CF/CA)

  • —> powerful q/g discriminant

Improved gluon rejection at higher pT ATL-PHYS-PUB-2017-009

slide-7
SLIDE 7

7

ncharged modeling

  • The challenge is assessing the

modeling of the discrimination performance in MC

  • Run 1 ncharged measurement

shows large data/MC discrepancies depending on generator and tuning choices

  • Eur. Phys. J. C76(6), 1-23 (2016)
slide-8
SLIDE 8

8

Recasting Run 1 measurement for Run 2 tagger

  • Fragmentation modeling uncertainties from Run 1

charged multiplicity measurement.

  • Experimental uncertainties from track reconstruction

uncertainties at 13 TeV (Run 2).

slide-9
SLIDE 9

9

ncharged for quark and gluon jets

NNPDF Collaboration

  • Extract ncharged separately for quarks and

gluons by exploiting rapidity difference in dijet events (mostly qg—>qg)

“valence” quark at high x “sea” gluon at low x

p p

slide-10
SLIDE 10

10

ncharged for quark and gluon jets

Solving for <nq>, <ng>

more forward jet more central jet

  • Eur. Phys. J. C76(6), 1-23 (2016)
slide-11
SLIDE 11

11

ncharged for quark and gluon jets

  • Eur. Phys. J. C76(6), 1-23 (2016)

ATL-PHYS-PUB-2017-009 Excellent closure in wide kinematic range Contamination from gg and qq events at low and high pT, respectively

slide-12
SLIDE 12

12

ncharged for quark and gluon jets

  • Eur. Phys. J. C76(6), 1-23 (2016)
  • Unfolded measurement of

⟨ncharged⟩ separately for quark and gluon jets as a function of jet pT

  • Uncertainties from
  • ME and PDF used to estimate

quark/gluon fractions from MC

  • detector/physics objects

(through unfolding)

slide-13
SLIDE 13

13

From Run 1 measurement to Run 2 tagger

ATL-PHYS-PUB-2017-009 Use Run 1 measurement to assess modeling uncertainty Add Run 2 experimental tracking uncertainties for use at 13 TeV

Uncertainties

  • n quark and

gluon efficiency @ ~fixed 60% quark efficiency

slide-14
SLIDE 14

14

Overall uncertainties

  • ~5% overall uncertainties across a wide range of pT
  • Precision limited by statistics and gg(qq) contamination at low and high pT
  • Caveat: small(*) additional uncertainties for topologies other than dijet

(*) to be assessed case-by-case with MC comparisons

ATL-PHYS-PUB-2017-009

slide-15
SLIDE 15

First look quark/gluon tagging with ATLAS jet images

slide-16
SLIDE 16

16

Jet images for q/g tagging

  • Boost and rotate jets so that η,ɸ = 0,0
  • Different types of constituents:
  • truth-particles
  • charged particle tracks
  • topological calorimeter clusters
  • calorimeter towers
  • Build 16x16 pixel image so that the intensity (I)
  • f a pixel is the sum of the pT of the

constituents within the pixel

  • Normalize each image so that Σpixels I = 1

ATL-PHYS-PUB-2017-017

slide-17
SLIDE 17

17

The CNN tagger

  • Convolutional neural network (CNN) learns non-linear representations of

the input image with the goal of discriminating between quark and gluon jet images.

  • Similar network architecture as in Komiske et al., hep-ph/1612.01551

Pixel Intensity 0.05 0.1 0.15 0.2 φ Translated Azimuthal Angle 0.4 − 0.2 − 0.2 0.4 η Translated Pseudorapidity 0.4 − 0.2 − 0.2 0.4

/GeV < 500

T

, R = 0.4, 400 < p

t

anti-k

convolutional filters Max-pooling dense layer quark jet gluon jet 3x

5 10 15

Translated Azimuthal Angle ϕ

5 10 15

Translated Pseudorapidity η

ATLAS Simulation Preliminary

Anti-k R=0.4, 150 GeV < pT < 200 GeV Tower Constituents

−0.15 −0.10 −0.05 0.00 0.05 0.10 0.15 0.20

Pearson Correlation Coefficient

ATL-PHYS-PUB-2017-017

slide-18
SLIDE 18

0.4 0.5 0.6 0.7 0.8 0.9 1.0

Quark Jet Efficiency

1.0 10.0

Gluon Jet Rejection

ATLAS Simulation Preliminary

s = 13 TeV Anti-k EM+JES R=0.4 |η| < 2.1, 400 GeV < pT < 500 GeV

CNN Truth Particle CNN Topo Clusters CNN EM Towers CNN Topo Clusters + Tracks CNN EM Towers + Tracks

0.4 0.5 0.6 0.7 0.8 0.9 1.0

Quark Jet Efficiency

1.0 10.0

Gluon Jet Rejection

ATLAS Simulation Preliminary

s = 13 TeV Anti-k EM+JES R=0.4 |η| < 2.1, 150 GeV < pT < 200 GeV

CNN Truth Particles CNN Topo Clusters CNN EM Towers CNN Topo Clusters + Tracks CNN EM Towers + Tracks

18

Input for jet images

  • Comparison of CNN tagger performance with different input images

(network retrained for each input type) Low pT:

  • Calo+tracks tagger performs

as well as truth-particle one. High pT:

  • Gap wrt truth-particle image.
  • Some differences from treatment of

calo information (towers vs clusters). ATL-PHYS-PUB-2017-017

slide-19
SLIDE 19

0.4 0.5 0.6 0.7 0.8 0.9 1.0

Quark Jet Efficiency

1.0 10.0

Gluon Jet Rejection

ATLAS Simulation Preliminary

s = 13 TeV Anti-k EM+JES R=0.4 |η| < 2.1, 400 GeV < pT < 500 GeV

CNN EM Towers + Tracks LLH (# Charged Particles + Jet Width) # Charged Particles Jet Width

0.4 0.5 0.6 0.7 0.8 0.9 1.0

Quark Jet Efficiency

1.0 10.0

Gluon Jet Rejection

ATLAS Simulation Preliminary

s = 13 TeV Anti-k EM+JES R=0.4 |η| < 2.1, 150 GeV < pT < 200 GeV

CNN EM Towers + Tracks LLH (# Charged Particles + Jet Width) # Charged Particles Jet Width

19

CNN tagger vs high-level observables

  • Comparison of CNN tagger performance wrt single observables (ntrack, jet

width) and their combination (likelihood ratio). CNN tagger performs at least as well as ntrack+width At high pT, CNN tagger performs better/worse than ntrack+width depending on quark efficiency working point ATL-PHYS-PUB-2017-017

slide-20
SLIDE 20

0.4 0.5 0.6 0.7 0.8 0.9 1.0

Quark Jet Efficiency

1.0 10.0

Gluon Jet Rejection

ATLAS Simulation Preliminary

s = 13 TeV Anti-k EM+JES R=0.4 |η| < 2.1, 150 GeV < pT < 200 GeV

CNN Truth, Trained on Pythia, Test on Pythia CNN Truth, Trained on Herwig, Test on Herwig CNN Truth, Trained on Pythia, Test on Herwig CNN Truth, Trained on Herwig, Test on Pythia

20

MC generators

  • Comparison of CNN tagger performance vs fragmentation modeling
  • Similar performance when

comparing Sherpa and Pythia (for either training or testing)

  • Moderate difference in

performance between taggers trained on Pythia vs Herwig and tested on Pythia

  • Large difference in performance

when testing on Pythia vs Herwig (irrespective of training)

  • Difference in fragmentation modeling plays a role, but features learned by

network only partially sensitive to it. ATL-PHYS-PUB-2017-017

slide-21
SLIDE 21

21

Summary and outlook

  • Established strategy for calibrating and deriving

uncertainties for q/g tagging observables using dijet events

  • ~5% uncertainties for tagger based on charged particle

multiplicity

  • Begun exploration of q/g tagging using jet images and

convolutional neural networks

  • Next steps: understand differences between input objects

and what is learnt (e.g. compared with high-level

  • bservables).
slide-22
SLIDE 22

22

References

  • Measurement of the charged-particle multiplicity inside jets

from sqrt(s) = 8 TeV pp collisions with the ATLAS detector

  • Eur. Phys. J. C76(6), 1-23 (2016)
  • Quark versus Gluon Jet Tagging Using Charged Particle

Multiplicity with the ATLAS Detector ATL-PHYS-PUB-2017-009

  • Quark versus Gluon Jet Tagging Using Jet Images with the

ATLAS Detector ATL-PHYS-PUB-2017-017

slide-23
SLIDE 23

Additional slides

slide-24
SLIDE 24

24

ntrack - Run 2 tunes

ATL-PHYS-PUB-2017-009

slide-25
SLIDE 25

25

ntrack - uncertainties

ATL-PHYS-PUB-2017-009

slide-26
SLIDE 26

26

ntrack - uncertainties

ATL-PHYS-PUB-2017-009

slide-27
SLIDE 27

27

CNN - track-truth and η range

ATL-PHYS-PUB-2017-017

slide-28
SLIDE 28

28

CNN - pileup

ATL-PHYS-PUB-2017-017

slide-29
SLIDE 29

29

CNN - fragmentation modeling

ATL-PHYS-PUB-2017-017

slide-30
SLIDE 30

0.4 0.5 0.6 0.7 0.8 0.9 1.0

Quark Jet Efficiency

1.0 10.0

Gluon Jet Rejection

ATLAS Simulation Preliminary

s = 13 TeV Anti-k EM+JES R=0.4 |η| < 2.1, 150 GeV < pT < 200 GeV

CNN Truth, Trained on Pythia, Test on Pythia CNN Truth, Trained on Herwig, Test on Herwig CNN Truth, Trained on Pythia, Test on Herwig CNN Truth, Trained on Herwig, Test on Pythia

30

CNN - fragmentation modeling

ATL-PHYS-PUB-2017-017

Komiske, Metodiev, Schwartz - JHEP 01 (2017) 110

Pythia with Monash tune Pythia with A14 tune

slide-31
SLIDE 31

31

CNN - average convolution differences

ATL-PHYS-PUB-2017-017

slide-32
SLIDE 32

32

CNN - correlation

5 10 15

Translated Azimuthal Angle ϕ

5 10 15

Translated Pseudorapidity η

ATLAS Simulation Preliminary

Anti-k R=0.4, 150 GeV < pT < 200 GeV Tower Constituents

−0.15 −0.10 −0.05 0.00 0.05 0.10 0.15 0.20

Pearson Correlation Coefficient

ATL-PHYS-PUB-2017-017