Jet Reconstruction in Athena

1
Jet Reconstruction in Athena

• Atlas Calorimeter Energy Calibration Workshop
• Ringberg Castle
• 23/07/02
• Ambreesh Gupta, University of Chicago

Outline Introduction Athena The Atlas
Software Framework Jet Reconstruction Tau
Reconstruction Energy Flow Summary
2
People Involved
The group of people involved in jet
reconstruction in the atlas software framework -
Martine, Monika, Peter, Ed, Srini, Tom, Hong,
Jim, Jon, Frank, Frank
People interested can join the JetRec bi-weekly
telephone meeting at 5 pm CERN time on
Wednesdays.
3
q
Initial Parton in Hard Scattering
p
p
q
ISR/FSR Fragmentation Underlying Event
Event Properties
LHC Design ?s 14 TeV L 10-34 cm-2 sec-1
Collision Every 25ns 23 Interaction/Collision
e/? Dead material Magnetic field Electronic
Noise Pile Up
Detector Effects
Reconstructed Jet in Calorimeter
The challenge is to identify reconstructed jet
with the initial parton Nontrivial
4
Goal of Hadron Calorimetry . Resolution
50/?E ? 3. . Jet Scale 1.
5
The Steps of Jet Measurement
Jet Measurement can be broadly divided in three
steps
Jet Reconstruction
Energy deposit in the calorimeter is clustered by
Jet reconstruction algorithms, e.g Cone and Kt.
Energy Calibration
Energy of a Jet is calibrated for
non-compensation, dead material, magnetic field,
etc.
Flavor Identification
b-tag, ??-tag, etc.
6
Athena The Atlas Software Framework
Framework Provides Services Execute, Monitor
and Output From Algorithms.
Framework Philosophy Data Classes are
Stable, Algorithms Change. User Can Plug in New
Algorithms.
Transient Event Data Store
Transient Detector and Calibration Data Store
Algorithm A
Algorithm B
Algorithm C
7
Jet Reconstruction Package
Many Kind of Input Data
Many Kind of Algorithms
Abstract The Inputs For Algs
Cone
CalCell
CaloTower
Kt Cluster
CaloCluster
ProtoJet
..
Tracks
ProtoJet Class . Can be created for different
sub system. . Homogenous.
Energy Flow
MC Truth
8
Jet Reconstruction Package
Jet Reconstruction Algorithms
Energy Correction Algorithms
Sampling Based Correction
Cone
Jet
Jet
Kt Cluster
H1 Style Correction
..
Working with data abstractions is great but the
information lost is required by correction
Algorithms..
Jet Class . Jets are composed of ProtoJet .
Provide Mechanism to deal with Overlapped
Jets. . Provide mechanism to deal with
Recombination Schemes.
9
Navigating Composite Objects
Problem Statement Given a Jet made of ProtoJet
How do I know what kind of CaloCell (LAr,Tile
,etc.) are they made of ?
Record structural pattern in Tokens. Access them
through special processor classes.
Jet
JetToken
PJet
PJetToken
PJet
PJet
Tower
TowerToken
Tower
Cell
Cell
Cell
CellToken
10
Available Tools and Packages

• There are many Packages that help in Jet
  Reconstruction. The
• three listed below are the typical names
  associated with, jet, tau
• and energy flow reconstruction -
• . JetRec
• . tauRec
• . eflowRec

11
JetRec
12
Jet Algorithms
Jet algorithms are employed to map final states,
both in QCD pert. theory and in the data, onto
jets. The motivating idea is that these jets are
surrogates for the underlying energetic parton.
Clustering - hadrons, Calorimeter Cells,Towers
etc., for nearness
Nearness in angle gt Cone Algorithm.
Nearness in relative transverse momentum gt Kt
algorithm.
Recombination Scheme - The momentum addition
rule of particles in a jet.
Historically hadron collider use cone algorithms
easier calibration
13
Cone Algorithm
Cluster particles within a radius R ???2
???2. Cone iterated until a stable Et weighted
cone is achieved.
Possible to produce overlapped cones Needs a
Split-Merge step.
Various version of cone algorithm dealing with
issues of speed and theoretical uncertainty.
Implementation of a seedless cone and split-merge
algorithm in JetRec.
Configurable through jobOption file.
?/E
Chrono services provided by Athena time profile
algorithms. Average reconstruction time for 1
GeV jet 0.7 sec.
1/?E GeV-2
14
Kt Algorithm
Cluster particles in order of increasing
relative transverse momentum. Requires a method
to terminate clustering.
No overlapping Jets. Theoretically well behaved
by design.
The algorithm is O(n3) gt Pre-Cluster particles.
Implementation of one Pre-Cluster and kT
algorithm in JetRec.
Reco time for 1000 GeV jet . Apply an Et cut
of 100 MeV gt 200 input ProtoJet. Average
time 1 sec. . Not Et cut gt 800 input
ProtoJet. Average time 1 min.
?/E
1/?E GeV-2
15
Energy Correction Algorithms
Various method of energy correction used by Atlas
in studies Parametric minimization, in-situ,
e/h.

• Setup of example algorithm to do energy
  correction
• - weights taken from earlier studies (Lefevre
  Santoni, Martine)
• iterative procedure to approximate the true
  energy for parameters
• use Navigation packages.

The correction improves both linearity and
resolution, but not in exact match with earlier
studies.
?/E
New studies with H1 style
calibration see later
1/?E GeV-2
16
tauRec
17
Tau Reconstruction
Taus have unique importance in SUSY searches.


Reconstruction Steps
A Sliding Window algorithm with window 0.5x0.5
Identify Cluster
Three highest pT tracks with pT gt 2 GeV And
within ?R lt 0.4
Associate Tracks to Cluster
1) H1 style calibration 2) Hadronic calibration
Calibrate Cluster
Variables to distinguish hadronic tau decays
from QCD jets. 20 fold reduction in bkg.
with 50 acceptance for SUSY signal.
Tau ID cuts
18
Hadronic Calibration (Used in Physics TDR)
Weights derived from single pion
Corrects the overall energy scale but does not
give optimal resolution.
19
H1 Type Calibration
Makes use of the fact that hadronic showers are
more localized than EM showers.
Derive weights in bins of Et for each layer -
EM3, Tile and HEC No weights for EM1, EM2.
Minimize function
With constraint
Samples of single pion with pT 10, 20, 40, 80,
160 GeV were used.
20
Calibrated ET/PT (p) vs h (H1 weights)
Pion resolution with H1 weights much better than
fixed weights. ?/E 38.56/?E ? 3.56
Applying the H1 weights to the reconstructed
tau events gave significant improvement in
both the average response and the resolution.

21
Extending H1 style Tau Calibration to Jets.
Jet35 sample was chosen for this study. 35 GeV
dijet sample with
. pT ( hard.scat. ) gt 35 GeV. . Electron rich
trigger.
Calibration weights re-derived for this sample.
Since sample is dominated by low Et jets, f(ET)
chosen to be f(ET) ET2
To compare to MC, same jet reconstruction
algorithm applied to ProtoJets generated from
Truth in ATLFAST.
22
Comparing Et/Et,MC
kt Jet Mean close to 1. Increases with Jet Et
cut. Resolution curves show tails in
gaussian fit. Fits fairly well with double
gaussian.
cone Jet Mean close to 1. Similar to above,
increases with jet Et cut. Tails,
substantially reduced compared to above.
23
ET/ET(MC) after calibration ( kT jet )
24
Fitted H1 weights for Jets
ET Bins 1/32, 1/16, 1/8, 16 GeV
Bin W(EM2) W(EM3) W(Tile) W(HEC)
01 1.203 2.000 3.999 3.123
2 1.645 1.393 2.230 1.163
3 1.423 1.500 1.894 0.950
4 1.137 1.165 1.355 0.922
5 1.043 1.079 1.223 1.000
6 0.984 1.451 1.135 1.078
7 0.964 1.128 1.104 1.202
89 1.041 1.206 1.052 1.241
Cryo term 0.471 ?(EEM3 ETile) Gap weight
0.915
25
Investigating the Tails
Possible effect of miss measurement of jet
energy due to magnetic field studied by taking in
to account of tracks that sweep in(out) of the
jet.
No noticeable change.
Cleaner Jet sample ( without electron rich
trigger ) for several ET bins being requested for
further studies on the tails.
26
eflowRec
27
Energy Flow Concept
Basic Idea The well measured particle momentum
substitutes random fluctuation of energy in the
calorimeter gt better resolution.
Introduced first by LEP experiments led to
significant improvement in jet energy
resolution.
The idea is simple but challenge to realize
requires building the particle ID associated
with the track. This starts running in to
difficulties in high track multiplicity
environment and coarse calorimeter granularity.

28
Energy Flow Package
eflowRec is a first attempt to combine
calorimeter, tracking and PID information to
improve energy resolution for jet and ETMiss
Algorithm flow
ECAL
HCAL
Loop over tracks
Tracks and cluster matched in ? and ? (
neutral particle ) and using helix parameter of
tracks ( charged particles ) to form
topologically connected eflow objects.
neutral
neutral
charged
Subtract expected energy deposit in EM and Had
cluster.
Estimated based on the particle ID hypothesis.
Loop over the remaining EM clusters and subtract
expected energy in HCAL clusters
29
Plans eflowRec
Limitation at present due the availability of
full PID packages.
Combined muon algorithm as input.
Brem recovery and conversion finding.
scalar sum pT
missing pT
PID internal to eflowRec.
3D cluster reconstruction.
Various choices to be made Need MC studies to
constraint.
y-comp. of missing pT
x-comp. missing pT
30
Summary

• The jet reconstruction setup in Athena is fairly
  mature.
• The important things that should be on the
  priority are documentation and a setup to test
  and validate released code in a simple way.
• New studies in calibration studies driving the
  software setup
• to produce the right tools.
• Progress in tau reconstruction and H1 style
  calibration.
• First implementations of an Energy Flow package
  available in Athena.
• Please send comment/suggestions to the
  jets-combined mailing list
• or join the bi-weekly phone meetings for
  further details.