? 1 / 22 Jet and E miss Reconstruction and Calibration T - - PowerPoint PPT Presentation

SLIDE 1

Jet and E miss

T

Reconstruction and Calibration

Christopher Young, CERN 16th July 2018

?

1 / 22

SLIDE 2

Jet and E miss

T

Reconstruction and Calibration

Christopher Young, CERN

Introduction

◮ Jets are very important to almost all analyses at the LHC. ◮ While this workshop clearly focusses on boosted object reconstruction

here I will cover the reconstruction and calibration of the R = 0.4 Anti-kt jets that are used as standard by ATLAS analyses as well as the reconstruction of missing transverse momentum.

◮ Jets are used from 20 GeV to 3.8 TeV and up to |η| < 4.5 by analyses

with their uses varying from jet vetoes, signal enhancement through their presence to unfolding their kinematic distributions.

◮ Missing transverse momentum, E miss

T

, is used to infer the existence of weakly interacting neutral particles that pass through the detector undetected, for example, neutrinos or other more exotic particles.

◮ The reconstruction of this requires the accurate measurement of all

• bjects in the event to check if they balance in the transverse plane.

2 / 22

SLIDE 3

Jet and E miss

T

Reconstruction and Calibration

Christopher Young, CERN

The ATLAS Detector

◮ The ATLAS detector - multi-purpose detector: inner tracker, EM + HAD

calorimeters, muon spectrometer.

◮ Magnetic fields provided by thin solonoid (inside calorimeters) and outer

toroid for muon measurements.

◮ The calorimeters are particularly important for jet measurements.

◮ >∼9 interaction lengths gives good jet containment. ◮ High granularity: 2nd EM layer 0.025×0.025, HAD barrel 0.1×0.1. 3 / 22

SLIDE 4

Jet and E miss

T

Reconstruction and Calibration

Christopher Young, CERN

Topocluster Reconstruction arXiv:1603.02934

◮ Calorimeter object reconstruction starts with topological clustering of

calorimeter cells.

◮ Cells 4σ above the noise (inc. pile-up) seed clusters. ◮ Neighboring cells 2σ above the noise are added iteratively. ◮ Finally a surrounding layer of cells is added. ◮ A splitting algorithm is then run to split local minima. ◮ For large-R jets these are then calibrated to account for EM and HAD

differences, dead material and out-of-cluster deposits.

φ cos × | θ |tan

• 0.05

0.05 φ sin × | θ |tan

• 0.05

0.05

2

10

3

10

4

10

5

10

E [MeV] ATLAS simulation 2010

Pythia 6.425 dijet event φ cos × | θ |tan

• 0.05

0.05 φ sin × | θ |tan

• 0.05

0.05

2

10

3

10

4

10

5

10

E [MeV] ATLAS simulation 201

Pythia 6.425 dijet event φ cos × | θ |tan

• 0.05

0.05 φ sin × | θ |tan

• 0.05

0.05

2

10

3

10

4

10

5

10

E [MeV] ATLAS simulation 2010

Pythia 6.425 dijet event

4 / 22

SLIDE 5

Jet and E miss

T

Reconstruction and Calibration

Christopher Young, CERN

Pile-Up in the Calorimeter arXiv:1703.10485

◮ Pile-up is the resulting signals from other interactions - both from the

same crossing and residual signals from close-by crossings.

◮ While the tracker can distinguish pile-up, the additional energy in the

calorimeter pollutes jet measurements and also results in the reconstruction of additional jets.

5 / 22

SLIDE 6

Jet and E miss

T

Reconstruction and Calibration

Christopher Young, CERN

Particle Flow Reconstruction arXiv:1703.10485

◮ Particle Flow reconstruction starts from tracks and topological clusters. ◮ Tracks where the tracker is expected to be much better than calo are

selected;

◮ Low pT- better tracker resolution ◮ Not in very dense areas of calorimeter - easier to do the subtraction

◮ The energy deposited by tracks is subtracted cell-by-cell. ◮ Objects built from remaining clusters and hard-scatter tracks.

π+ π0 π+ π0

6 / 22

SLIDE 7

Jet and E miss

T

Reconstruction and Calibration

Christopher Young, CERN

Jet Reconstruction and Calibration Sequence arXiv:1703.09665

◮ Jets are reconstructed using the Anti-kt algorithm with radius parameter

R = 0.4 - although we are also looking at other radii.

◮ The inputs are either topological clusters at the electromagnetic scale (we

• nly use the calibrated clusters for sub-structure) or particle flow objects -

tracks from the hard-scatter and remaining calorimeter clusters.

◮ Below is the full calibration sequence - I will go through each step in turn.

7 / 22

SLIDE 8

Jet and E miss

T

Reconstruction and Calibration

Christopher Young, CERN

Pile-Up Correction arXiv:1703.09665

◮ To correct for pile-up falling within the jet cone first a ρ × A subtraction

is performed.

◮ ρ is the average pile-up density per unit area determined in the region

|η| < 2.0

◮ An additional correction is then applied based on the number of vertices

and µ to account for residual pile-up dependence.

8 / 22

SLIDE 9

Jet and E miss

T

Reconstruction and Calibration

Christopher Young, CERN

MC-based Calibration and GSC arXiv:1703.09665,ATL-PHYS-PUB-2018-013

◮ A Monte Carlo based calibration corrects the jet energy to the truth jet

scale - particle level jets formed from stable hadrons.

◮ Following this the η of jets is corrected to account for biases due to

cracks in the calorimeter.

◮ The next stage of the calibration is to improve the resolution and reduce

quark/gluon differences by removing the dependence on fraction of energy in different calorimeter layers, number of tracks, track width and muon spectrometer hits (which accounts for punch-through).

◮ Now looking at using Machine Learning for this - see A. Cukierman’s

poster!

9 / 22

SLIDE 10

Jet and E miss

T

Reconstruction and Calibration

Christopher Young, CERN

η-Intercalibration JETM-2017-008

◮ Different detector technologies are utilized as a function of |η|. ◮ To ensure that the data-to-MC ratio is uniform as a function of η di-jet

events are used as they are expected to balance in the transverse plane.

◮ Events are selected with no 3rd jets and large ∆φ but still some truth

imbalance remains.

◮ The modeling of this imbalance forms one of the major systematics for

the forward JES.

◮ The size of the corrections required is ∼ 5% in the most forward regions.

10 / 22

SLIDE 11

Jet and E miss

T

Reconstruction and Calibration

Christopher Young, CERN

In situ V+jet Calibration arXiv:1703.09665

◮ The energy scale of electrons, muons and photons is very well known. ◮ A boson (Z → ll or γ) recoiling off a jet should balance in pT. ◮ We look at both the direct balance between the jet and boson and also

the Missing ET Projection Fraction (MPF) method where we look at the full hadronic recoil against the boson.

◮ The methods are found to be compatible and one is chosen as they are

not statistically independent.

11 / 22

SLIDE 12

Jet and E miss

T

Reconstruction and Calibration

Christopher Young, CERN

Multi-Jet Balance + Combination arXiv:1703.09665,JETM-2017-003

◮ The V+jet balance techniques run out of statistics around 1 TeV so a

different technique is required beyond this.

◮ The balance of a single leading jet against a multi-jet system is used to

extend the data driven techniques to higher pT.

◮ The methods are then all combined to form the final JES in situ

correction and its uncertainty.

◮ The methods are found to agree well in the regions of overlap and the

independence of their uncertainties reduce the overall level of uncertainty.

12 / 22

SLIDE 13

Jet and E miss

T

Reconstruction and Calibration

Christopher Young, CERN

Jet Energy Scale Uncertainties JETM-2017-003

◮ The full JES uncertainties contain the previously described in situ

uncertainties as well as additional uncertainties for the modeling of pile-up, the flavour composition and response differences between generators, and finally single particle response at the highest pT.

◮ At low pT the pile-up uncertainties dominate, then the flavour response

• f gluon jets which are not directly probed by the in situ measurements,

then the photon energy scale and finally single particle uncertainties.

◮ At high |η| we are dominated by modeling issues of the balance between

forward and central jets.

13 / 22

SLIDE 14

Jet and E miss

T

Reconstruction and Calibration

Christopher Young, CERN

Jet Energy Resolution Measurement (Run I) ATLAS-CONF-2015-037

◮ These same balance distributions (γ-jet, Z-jet and di-jet) can be used to

extract the Jet Energy Resolution.

◮ The truth level imbalance of the systems is corrected for by subtracting it

in quadrature.

◮ The results from the 3 systems are combined with a measurement of the

noise from pile-up taken from the fluctuations seen in random cones in unbiased data.

[GeV]

jet T

p 20 30 40

2

10

2

10 × 2

3

10

T

) / p

T

(p σ 0.1 0.2 0.3 0.4 0.5 0.6 R=0.4, EM+JES

t

anti-k | < 0.8 η | ATLAS Preliminary = 8 TeV s

∫

• 1

L dt = 20 fb

• jet

γ Z-jet Dijets Total uncertainty Statistical component [GeV]

jet T

p 20 30 40

2

10

2

10 × 2

3

10

T

) / p

T

(p σ 0.1 0.2 0.3 0.4 0.5 0.6 14 / 22

SLIDE 15

Jet and E miss

T

Reconstruction and Calibration

Christopher Young, CERN

Pile-Up Effect on Jet Resolution arXiv:1703.10485,ATLAS-CONF-2017-065

◮ Pile-up falling within a jet cone affects the scale and also the resolution. ◮ While the pile-up corrections correct for the former effect they cannot

eliminate the latter such that the resolution grows with increasing µ - particularly at low pT.

◮ Particle flow mitigates this by subtracting pile-up track-by-track. ◮ Additional constituent based subtraction techniques are being

investigated as well.

15 / 22

SLIDE 16

Jet and E miss

T

Reconstruction and Calibration

Christopher Young, CERN

Pile-Up Jets and Rejection JETM-2017-009,JETM-2017-006

◮ Pile-up creates additional reconstructed jets - affects analyses and E miss

T

◮ Jet Vertex Tagger is a likelihood based on the tracks pointing at the jet,

designed to reject pile-up jets while keeping hard-scatter jets. ATLAS-CONF-2014-018

◮ It uses the fraction of track pT from the HS as a fraction of the total,

and the ratio of track pT from the HS and the calorimeter pT.

◮ Particle Flow also helps the rejection of pile-up jets (but maintains high

efficiency).

Average Number of Interactions per Bunch Crossing

10 15 20 25 30 35 40 45 50 55

〉 Number of Jets 〈

0.5 1 1.5 2 2.5

< 101 GeV

µ µ

81 < M =0.4 EM+JES R

t

k Anti- | < 2.4 η > 20 GeV, |

jet T

p

ATLAS Preliminary

• 1

= 13 TeV, 20.8 fb s 2017, All jets JVT>0.59 jet

η

• 4
• 3
• 2
• 1

1 2 3 4

〉 Fake Jets / 0.2 / Event 〈

• 3

10

• 2

10

• 1

10 1 ~ 22 〉 µ 〈 = 13 TeV s > 20 GeV

jet T

p ATLAS Simulation Preliminary

EM+JES Jets EM+JES Jets JVT>0.59 Particle Flow Jets Particle Flow Jets JVT>0.2

16 / 22

SLIDE 17

Jet and E miss

T

Reconstruction and Calibration

Christopher Young, CERN

Forward Pile-Up Jets and Rejection arXiv:1705.02211

◮ Pile-up jets can either be from a combination of other vertices

(stochastic) or a single vertex (QCD).

◮ For forward jets we try to identify central pile-up jets that balance

forward jets as these are mainly QCD.

◮ Also looking at calo timing and jet shapes to reject stochastic pile-up.

| η | 0.5 1 1.5 2 2.5 3 3.5 4 4.5 QCD Pile-up Jet Fraction 0.2 0.4 0.6 0.8 1 1.2 1.4

<50 GeV

T

30<p <30 GeV

T

20<p

ATLAS Simulation

Pythia8 dijets =22 〉 µ 〈 = 13 TeV, s EM+JES R=0.4

t

k Anti-

fJVT 0.5 1 1.5 2 Fraction of Jets / 0.05 0.1 0.2 0.3

Hard-scatter jets Inclusive pile-up jets

ATLAS Simulation

µ µ → Powheg+Pythia8 Z =13.5 〉 µ 〈 = 13 TeV, s EM+JES R=0.4

t

k Anti- <50 GeV

T

|<4.5, 40<p η 2.5<|

Hard-scatter Jet Efficiency 0.7 0.75 0.8 0.85 0.9 0.95 1 Pile-up Jet Efficiency 0.2 0.4 0.6 0.8 1 1.2 1.4

<30 GeV

T

20<p <40 GeV

T

30<p <50 GeV

T

40<p

ATLAS Simulation

µ µ → Powheg+Pythia8 Z =13.5 〉 µ 〈 = 13 TeV, s EM+JES R=0.4

t

k Anti- |<4.5 η 2.5<|

17 / 22

SLIDE 18

Jet and E miss

T

Reconstruction and Calibration

Christopher Young, CERN

E miss

T

• NEW CONF ATLAS-CONF-2018-023

◮ ATLAS uses a object-based definition of the E miss

T

◮ We take all the hard objects in the event; muons, electrons, photons, taus

and jets, which are above threshold.

◮ We resolve the overlap between these at the cluster/track level - this has

been further optimized in this CONF.

◮ For soft energy flow we take tracks from the primary vertex (soft neutral

particles are not included as pile-up contaminates that calorimeter).

Events / GeV

2 −

10

1 −

10 1 10

2

10

3

10

4

10

5

10

6

10 Data Top Z+jets Diboson Stat+Syst Preliminary ATLAS

• 1

= 13 TeV 36 fb s

miss T

Tight E ee → Z [GeV]

miss T

Tight E 100 200 300 400 500 600 700 800 900 1000 Pred Data 0.5 1 1.5 Events / GeV

2 −

10

1 −

10 1 10

2

10

3

10

4

10

5

10

6

10 Data Z+jets Top Diboson Stat+Syst Preliminary ATLAS

• 1

= 13 TeV 36 fb s

miss T

Tight E ee → Z (Soft) [GeV]

miss T

PFlow E 20 40 60 80 100 120 140 Pred Data 0.5 1 1.5 − − Events / 20 GeV

4 −

10

3 −

10

2 −

10

1 −

10 1 10

2

10

3

10

4

10

5

10

6

10

7

10

8

10

9

10 Old OR New OR Simulation Preliminary ATLAS

• 1

= 13 TeV 36 fb s ee → Z [GeV]

miss T

• true E

miss T

E 200 − 100 − 100 200 300 400 500 600 New/Old 1 2

18 / 22

SLIDE 19

Jet and E miss

T

Reconstruction and Calibration

Christopher Young, CERN

E miss

T

Working Points ATLAS-CONF-2018-023

◮ Several different working points are provided for analyses. ◮ In particular as pile-up increases the forward pile-up jets contribute to the

E miss

T

resolution a lot.

◮ Therefore better performance is achieved either using the forward pile-up

rejection, or increasing the threshold for forward jets to be included in the E miss

T

.

◮ Using particle flow jets improves the resolution, particularly in events

without forward jets.

PV

Number of primary vertices N 5 10 15 20 25 30 35 40 Resolution [GeV]

miss y

, E

miss x

E 10 15 20 25 30 35

TST > 30 GeV

forward jet T

TST and p TST and fJVT

ATLAS Simulation Preliminary

µ µ → Powheg+Pythia8 Z = 13 TeV s Events / GeV

1 −

10 1 10

2

10

3

10

4

10

5

10

6

10 EMTopo jets PFlow jets Simulation Preliminary ATLAS

• 1

= 13 TeV 36 fb s | > 2.4 η , 0 jets with | µ µ → Z [GeV]

miss T

E 50 100 150 200 250 300 350 400 450 EMTopo PFlow 0.5 1 1.5 5 10 15 20 25 30 35 40 〉 µ 〈 Average number of interactions 12 14 16 18 20 22 24 26 28 30 RMS Resolution [GeV]

y miss

,E

x miss

E Preliminary ATLAS

• 1

= 13 TeV 36 fb s

miss T

Tight E µ µ → Z

EMTopo MC EMTopo MC Syst Err EMTopo Data PFlow Data

19 / 22

SLIDE 20

Jet and E miss

T

Reconstruction and Calibration

Christopher Young, CERN

E miss

T

Significance - NEW CONF ATLAS-CONF-2018-038

◮ As the E miss

T

consists of a series of well defined hard objects and a soft term we can determine an object-based significance of the E miss

T

.

◮ The resolutions of all the hard objects are propagated as well as a

gaussian to account for the missing neutral particles in the soft term.

◮ This exploits both the scale of the E miss

T

as well as its direction.

◮ Good data-to-MC agreement is seen both for the E miss

T

significance.

20 / 22

SLIDE 21

Jet and E miss

T

Reconstruction and Calibration

Christopher Young, CERN

E miss

T

Significance Performance ATLAS-CONF-2018-038

◮ To test the performance of this significance we create ROC curves for a

signal of ZZ → llνν against a background of Z → ll.

◮ For an inclusive selection little gain is found but when combined with a

soft E miss

T

cut or a more realistic analysis selection significant improvements in the performance are observed.

◮ Also see D. Portillo’s poster at the conference on this topic!

21 / 22

SLIDE 22

Jet and E miss

T

Reconstruction and Calibration

Christopher Young, CERN

Conclusions

◮ Jet reconstruction is an important part of the ATLAS physics program. ◮ The Jet Energy Scale is derived from data and the uncertainty is < 1%

for a 0.1 < pT < 1 TeV in the central region.

◮ Despite this the JES uncertainty remains one of the leading experimental

uncertainties in many analyses.

◮ Therefore work continues to measure this more precisely using the larger

full Run II dataset.

◮ E miss

T

reconstruction also important for many searches and measurements.

◮ Careful reconstruction of this to avoid fake E miss

T

tails continues.

◮ The newly developed E miss

T

significance shows promise for extracting signals more effectively from background.

◮ We still find that the E miss

T

is dependent on pile-up and work continues to reduce this dependence.

22 / 22