Calorimetry: Validation and Performance of PandoraPFA€¦ · −1 −0.8 −0.6 −0.4 −0.2 0...
Transcript of Calorimetry: Validation and Performance of PandoraPFA€¦ · −1 −0.8 −0.6 −0.4 −0.2 0...
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
1
Calorimetry: Validation and Performance of PandoraPFA
Matthias Weber (CERN)
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
PandoraPFA algorithm
Pandora LC Reconstruction
Pandora Reclustering Example
16
3. Track momentum much greater than cluster energy – bring in nearby clusters and reconfigure.
2. Cluster energy much greater than track momentum – split cluster.
1. Multiple tracks associated to single cluster – split cluster.
4. If, and only if, no E/p match emerges, can force track-cluster
consistency ⇒ energy flow.
If identify significant discrepancy between cluster energy and associated track momentum, choose to recluster. Alter clustering parameters until cluster splits to obtain track-cluster consistency.
• Exploit calorimeter granularity to gradually build up picture of events
• More than 70 algorithms à address different topologies, à correct identification à avoid accidental splitting and merging of particles
2
62 Photon Reconstruction in PandoraPFA
u−axis
05
1015
2025
3035
40
v−axis
05
1015
2025
3035
40
0
20
40
60
80
100
Ener
gy /
GeV
Figure 5.3: Two 500 GeV photons (yellow and blue), just resolved in a transverse planeorthogonal to the direction of the flight by projecting their energy deposition inthe electromagnetic calorimeter. U and V are orthogonal axes. Z axis is the sumof the calorimeter hit energy in GeV.
dimensional peak finding is a collection of Shower Peak objects. Each Shower Peakobject corresponds to a photon candidate and contains associated calorimeter hits.
5.3.3 Photon ID test
Shower Peak objects are tested for photon tagging. A set of discriminating variablesthat exploit features of electromagnetic showers are calculated. A multidimensionallikelihood classifier is implemented, which needs to be trained before applying. Theresponse from the classifier determines if a Shower Peak object is a photon. If it isa photon, it would be tagged and separated from the event. If it is not a photon, thecalorimeter hits of the Shower Peak object will be passed on to the next stage of thereconstruction (see figure 5.2). Because the classifier is complicated, it is discussed in aseparate section 5.5.
5.3.4 Photon Fragment removal
The optional photon fragment removal algorithm aims to merge small photon fragmentto identified photons. Since this step shares the same logic as the fragment removalalgorithm in section 5.6, this step is described in details in section 5.6.
This step marks the end of the photon reconstruction algorithm. The output are acollection of reconstructed photons, separated from non-photon calorimeter hits. Figure5.2 summarises key steps in the Photon Reconstruction algorithm.
Energy of two nearby photons in transverse plane to the direction of the flight
NIM A 700 (2013) 153
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
ALICE Investigator
7 Timepix3 telescope planes
Cracow SOI SPS beam
Single Particle Identification
3
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
trueθ cos1− 0.5− 0 0.5 1
0.85
0.9
0.95
1pion identification efficiency, E=10 GeV
ILCSoft-17-08-23ILCSoft-17-10-14
CLICdp work in progress
Pion Identification Efficiency
4
Studied on isolated single pion events Example of recent developments in PandoraPFA algorithms: Modified track cluster matching algorithm in barrel-endcap calorimeter transition
Now achieve also in barrel endcap transition efficiencies beyond 93 %
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
trueθ 0 2 4 6 8 10 12 14 16 18 20 220
0.2
0.4
0.6
0.8
1pion identification efficiency, E=1 GeV
trueθ 0 2 4 6 8 10 12 14 16 18 20 220
0.2
0.4
0.6
0.8
1
pion identification efficiency, E=10 GeV
Forward Pion Identification Efficiency
5
trueθ 0 2 4 6 8 10 12 14 16 18 20 220
0.2
0.4
0.6
0.8
1pion identification efficiency, E=10 GeV
trueθ 0 2 4 6 8 10 12 14 16 18 20 220
0.2
0.4
0.6
0.8
1
pion identification efficiency, E=20 GeV
10 GeV, conformal tracking 10 GeV, truth tracking
20 GeV, conformal tracking 1 GeV, conformal tracking
Reconstruct particles, starting at 8 degrees, reach more than 60 % at 10 degrees
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
trueθ cos1− 0.5− 0 0.5 1
0.7
0.75
0.8
0.85
0.9
0.95
1
muon identification efficiency, E=10 GeV
Muon Identification Efficiency
6
Studied on isolated single muon events Around 98 % for almost the whole range
Investigation ongoing for point exactly at 0, where efficiency drops to 87 %, also for higher energies (checked up to 500 GeV), the inefficiency at 90 degrees remains, effect a lot larger than in the pion case à muon hits are there, track is also there, PFA expert comment: maybe something wrong in pseudo-layer calculation
CLICdp work in progress
trueθ cos1− 0.5− 0 0.5 1
0.75
0.8
0.85
0.9
0.95
1
muon identification efficiency, E=50 GeV
CLICdp work in progress
10 GeV 50 GeV
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
)trueθcos(1− 0.8− 0.6− 0.4− 0.2− 0 0.2 0.4 0.6 0.8 1
Effic
ienc
y
0.7
0.75
0.8
0.85
0.9
0.95
1
1.05
1.1all photonsunconverted photons
photon identification efficiency, E=100 GeV
transition region
CLICdp work in progress
Photon Identification Efficiency
7
Studied on isolated single photon events for several energies: Require angular matching with true photon as well as energy matching
For unconverted photons achieve identification beyond 99 %
100 GeV photons
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
)trueθcos(1− 0.8− 0.6− 0.4− 0.2− 0 0.2 0.4 0.6 0.8 1
Effic
ienc
y
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
all photonsunconverted photonsconverted photonsphoton merging
photon identification efficiency, E=100 GeV
Photon Identification Efficiency: Conversions
8
Photon conversions need to be treated separately, typically conversion rate below 15 %, happens late in the tracking system, thus default track cluster matching requirements in pandoraPFA not met by default àtwo very close resolved photons, which fails each the energy matching requirement
Recover largely performance for converted photons by merging both resolved reconstructed photons (spatial requirement less hard than for unconverted photons) à over 90 % of merged pairs survive energy cut
CLICdp work in progress
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
total event E [GeV]8 8.5 9 9.5 10 10.5 11 11.5 120
5000
10000
15000
20000
25000
30000
35000h_energyTot_pionInEvent_copy
Entries 154126Mean 10.24Std Dev 0.5478
, E=10 GeV-eelectrons
Electrons and Bremsstrahlung
9
Electron energy resolution impacted by photon bremsstrahlung à Track fit also distorted à adding all photons up leads to a partial double counting as well à At 10 GeV Identification efficiency around 90 % in barrel, in endcap down to 75-80
%, for higher energies identification efficiency above 95 % both in barrel and endcap
Work ongoing for photon bremsstrahlung recovery algorithm to recover electron response tail and avoid double counting of energy from bremsstrahlung photons
[GeV]recoE8 8.5 9 9.5 10 10.5 11 11.5 120
5000
10000
15000
20000
25000
30000
35000
40000
45000
h_energy_pions_copy
Entries 154143Mean 9.846Std Dev 0.326
, E=10 GeV-eelectrons
Bremsstrahlung tail Double counting
of energy
CLICdp work in progress CLICdp work in progress
reconstructed electron energy total reconstructed
energy in event
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
ALICE Investigator
7 Timepix3 telescope planes
Cracow SOI SPS beam
Software Compensation (SWC)
10
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
Neutral Hadrons and Software Compensation
11
Response of electromagnetic component in hadron showers on average larger than for hadronic component àtreating all hits in HCAL the same way leads to a less accurate energy measurement Follow description of Software Compensation paper EPJC 77 (2017) 698 Electromagnetic component of shower typically denser: Software compensation at CLIC reweights hits in HCAL depending on the hit energy density, assuming these originate from electromagnetic component à assume monotonic falling weight with energy density à Weight includes an energy dependence à in total 9 different parameters are used in software compensation à Software compensation reweighting integrated into PandoraPFA software CALICE investigates alternative parametrisations of software compensation weights, including hits in ECAL à See talk by Yasmine Default weights tuned for ILD up to 100 GeV à SWC used for Ehadrons < 100 GeV at CLIC expect to reach higher hadron energies, at 3 TeV sometimes beyond 500 GeV àretune parameters for CLIC, use SWC for all hadron energies
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
Software Compensation at work at CLIC
12
Energy [GeV]100 150 200 250 300 350 400 450 500
arbi
trary
uni
ts
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
0.04
100 GeV
150 GeV
200 GeV
250 GeV
400 GeV
ClusterEnergy after (solid) andbefore (dashed) SoftwareCompensation
CLICdp work in progress
Software compensation weights derived from MC using neutron and K0L events
àMean and resolution after software compensation largely improved àSoftware compensation corrects for nonlinear response of hadrons on the fly
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
MCtrue/Eneut
recoE0.6 0.8 1 1.2 1.4
1
10
210
310
Fit relative response with a gaussian àfor 75 GeV 6.4 %, mean of fitted response 1.014
K0L Energy Resolution
13
Fit neutral hadron response with a gaussian, several single particle gun energy pointsàapply CLIC software compensation weights three regions: barrel, endcap, transition, non gaussian tails to lower responses
MCtrue/Eneut
recoE0.6 0.8 1 1.2 1.40
500
1000
1500
Barrel region, K0L, E=75 GeV Barrel region, K0L, E=75 GeV
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
[GeV]K0LtrueE
0 100 200 300 400
) [%
]K0
Ltru
e/E
reco
PFO
(Eσ
5
10
15
K0L
|<0.65trueK0Lθ|cos
|<0.80trueK0Lθ0.65<|cos
|<0.94trueK0Lθ0.80<|cos
Neutral Hadrons Energy Resolution Summary
14
Neutral Hadron energy resolution determined in three regions, barrel, transition region and endcap
Fit relative response with a gaussian àaround 16.2 % at 10 GeV, around 150 GeV we reach a values around 5.5 %
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
[GeV]reco
-πE480 500 520 540
0
1000
2000
3000
4000default weights
CLIC weights
[GeV]recotot
total event energy E500 550 600 650
0
500
1000
1500
2000
2500 default weights
CLIC weights
Software compensation and charged particle
15
Software compensation applied to all HCAL hits, also impact on charged pions
With new CLIC weights significant reduction in reconstructed original cluster energy àless “excessive” energy to split and create additional neutrons (when compared to track energy) à Software compensation reduced confusion due to reduced splitting of clusters originating from very high energetic tracks (energies that high relevant for decays of high energetic hadronic taus, pion energies in jets rarely beyond 100 GeV)
500 GeV pions
500 GeV pions reconstructed pion energy
total reconstructed energy
CLICdp work in progress CLICdp work in progress
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
Software compensation at CLIC in dijet events
16
CLIC specific software compensation weights compared to previous default
Significant improved jet energy with new CLIC specific weights for large energetic jets, comparable performance at low jet energies à applying software compensation always improves jet energy resolution
)|θ|cos(0 0.2 0.4 0.6 0.8 1
) [%
]j
(E90
) / M
ean
j(E
90R
MS
3
4
5
6750 GeV Jets
no software compensation
default weights
CLIC weights
CLICdp work in progress
)|θ|cos(0 0.2 0.4 0.6 0.8 1
) [%
]j
(E90
) / M
ean
j(E
90R
MS
3
4
5
6100 GeV Jets
no software compensation
default weights
CLIC weights
CLICdp work in progress
100 GeV jets 750 GeV jets
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
Jet energy resolution at CLIC
17
CLIC specific software compensation weights à achieve jet energy resolution between 3.1 % and 4.5 %
Compared to previous default software compensation weights comparable performance for jets up to 190 GeV, for larger jet energies improvement of resolutions by around 10%
)|θ|cos(0 0.2 0.4 0.6 0.8 1
) [%
]j
(E90
) / M
ean
j(E
90R
MS
3
4
5
6
7 50 GeV Jets
190 GeV Jets
250 GeV Jets
750 GeV Jets
1500 GeV Jets
CLICdp work in progress
CLICdp work in progress
CLIC weights
)|θ|cos(0 0.2 0.4 0.6 0.8 1
) [%
]j
(E90
) / M
ean
j(E
90R
MS
3
4
5
6
7 50 GeV Jets
190 GeV Jets
250 GeV Jets
750 GeV Jets
1500 GeV Jets
CLICdp work in progress
CLICdp work in progress
default weights
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
Jet energy resolution at CLIC: zoomed out
18
in forward region jet energy values up to 11.5 %, uncertainties of RMS values larger due to less statistics
)|θ|cos(0 0.2 0.4 0.6 0.8 1
) [%
]j
(E90
) / M
ean
j(E
90R
MS
4
6
8
10
12 50 GeV Jets
190 GeV Jets
250 GeV Jets
750 GeV Jets
1500 GeV Jets
CLICdp work in progress
CLIC weights
)|θ|cos(0 0.2 0.4 0.6 0.8 1
) [%
]j
(E90
) / M
ean
j(E
90R
MS
4
6
8
10
12 50 GeV Jets
190 GeV Jets
250 GeV Jets
750 GeV Jets
1500 GeV Jets
CLICdp work in progress
default weights
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
Jet energy resolution at CLIC: forward jets
19
Endcap resolutions a bit worse than resolutions in outer barrel bins, large increase starting from bin at |cos θ| >0.925 (18.2-22 °), a lot worse for jets with |cos θ| >0.975 (12 °) à
)|θ|cos(0.6 0.7 0.8 0.9 1
) [%
]j
(E90
) / M
ean
j(E
90R
MS
4
6
8
10
12 50 GeV Jets
190 GeV Jets
250 GeV Jets
750 GeV Jets
1500 GeV Jets
CLICdp work in progress
CLIC weights
)|θ|cos(0.6 0.7 0.8 0.9 1
) [%
]j
(E90
) / M
ean
j(E
90R
MS
4
6
8
10
12 50 GeV Jets
190 GeV Jets
250 GeV Jets
750 GeV Jets
1500 GeV Jets
CLICdp work in progress
default weights
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
true/ErecoE0.6 0.8 1 1.2 1.4
0
50
100
150
200
250|<1.00θ uds, 500 GeV, 0.975<|cos→Z
true/ErecoE0.6 0.8 1 1.2 1.4
0
100
200
300
|<0.975θ uds, 500 GeV, 0.950<|cos→Z
true/ErecoE0.6 0.8 1 1.2 1.4
0
100
200
300|<0.950θ uds, 500 GeV, 0.925<|cos→Z
Energy Resolution: Zàuds at 500 GeV
20
Check energy resolution for 500 GeV dataset in different cos θ bins
true/ErecoE0.6 0.8 1 1.2 1.4
0
500
1000
1500
|<0.20θ uds, 500 GeV, 0.10<|cos→Z
0.10-0.20
true/ErecoE0.6 0.8 1 1.2 1.4
0
500
1000
1500
2000
2500
|<0.90θ uds, 500 GeV, 0.80<|cos→Z
true/ErecoE0.6 0.8 1 1.2 1.4
0
200
400
600
|<0.925θ uds, 500 GeV, 0.900<|cos→Z
0.80-0.90 0.900-0.925
0.925-0.950
0.950-0.975
0.975-1.000
Resolution core not drastically different in forward region (compared to endcap), but long tails
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
ALICE Investigator
7 Timepix3 telescope planes
Cracow SOI SPS beam
Effects of beam background
21
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
Beamstrahlung γγàhadrons background reduction
22
Backup: Forward WW
Mark Thomson LCWS 2011, Granada 42
No background
Forward WW, no background
Backup: Forward WW
Mark Thomson LCWS 2011, Granada 42
No background
with γγàhadrons background in reconstruction time window ~1.2 TeV
Backup: Forward WW
Mark Thomson LCWS 2011, Granada 42
No background
after timing and pT cuts: ~ 100 GeV remain
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
[GeV]xmissing p200− 100− 0 100 200
0
200
400
600 uds, 3000 GeV→Zall PFOs, no Overlayall PFOs, OverlaylooseSel PFOs, Overlaysel PFOs, OverlaytightSel PFOs, Overlay
Background in Zà light dijets at 3 TeV
23
At 3 TeV expect 3.2 hadronic event per bunch crossing Check events of Zàuds quarks, dijet events without genuine missing energy
Out of originally almost 2 TeV additional energy after timing and pT cuts: ~ 100 GeV remain (tight selection à targets to deal with 3 TeV conditions)
[GeV]totPFOE
3000 4000 5000 60000
200
400
600
800
1000
1200 uds, 3000 GeV→Zall PFOs, no Overlayall PFOs, OverlaylooseSel PFOs, Overlaysel PFOs, OverlaytightSel PFOs, Overlay
CLICdp work in progress CLICdp work in progress
No genuine missing momentum component in these events à all selection cuts recover performance before background overlay
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
Background in ZZàqqνν at 3 TeV
24
At 3 TeV expect 3.2 hadronic event per bunch crossing Check change in missing transverse momentum
Comparable performance of CLICdet with CDR CLIC_ILD model in presense of background
[GeV]T
missing p∆100− 50− 0 50 100
0
2000
4000
6000 uds, 3000 GeV→Zall PFOs, no Overlayall PFOs, OverlaylooseSel PFOs, Overlaysel PFOs, OverlaytightSel PFOs, Overlay
CLICdp work in progress
Here tightSelected PFO cuts applied on red curve
New model CLICdet
CDR model CLIC_ILD
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
Background and mass reconstruction
25
Study timing cuts in 3 TeV WWàqqqq events, boosted topology à cluster event in two jets
Using tight selected PFOs as input for jet clustering almost recovers mass peak at 80 GeV completely
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
Summary
ALICE Investigator
7 Timepix3 telescope planes
Cracow SOI SPS beam
Performance of PandoraPFA has been studied with new detector model CLICdet for several particle species, jets and missing transverse momentum distributions • CLIC specific software compensation parameters have been determined
- Comparable jet energy resolution for low energetic jets (<250 GeV) - Jet resolutions improved by around 10 % for high energetic jets (>250 GeV)
when using previous default parameters, in barrel values of 3-4.5 % are achieved, for very forward jets resolution around 7-11 %
à Confusion term reduced by more accurate cluster energy estimation
• Study impact of pT and timing cuts on missing energy resolution assuming background levels of a 3 TeV machine - Use timing cuts determined for previous detector model CLIC_ILD for new
CLICdet model - Good performance of tinming cuts show by examples of jet masses in hadronic
WW events and missing pT resolutions in dijet and ZZàqqνν events
26
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
ALICE Investigator
7 Timepix3 telescope planes
Cracow SOI SPS beam
BACKUP
27
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
Definition of PandoraPFA timing cuts
ALICE Investigator
7 Timepix3 telescope planes
Cracow SOI SPS beam
28
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
trueθ cos1− 0.5− 0 0.5 1
0.65
0.7
0.75
0.8
0.85
0.9
0.95
1pion identification efficiency, E=50 GeV
Pion Identification Efficiency
29
Studied on isolated single pion events, around 95 % for almost the whole range, for very high energetic pions in barrel slightly higher inefficiency than for endcap
Point at exactly 90 degrees a bit lower than neighbouring region
CLICdp work in progress
trueθ cos1− 0.5− 0 0.5 1
0.6
0.65
0.7
0.75
0.8
0.85
0.9
0.95
1
pion identification efficiency, E=100 GeV
CLICdp work in progress
50 GeV 100 GeV
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
trueθ cos1− 0.5− 0 0.5 1
0.7
0.75
0.8
0.85
0.9
0.95
1
muon identification efficiency, E=10 GeV
Muon Identification Efficiency
30
Studied on isolated single muon events Around 98 % for almost the whole range
Investigation ongoing for point exactly at 0, where efficiency drops to 87 %, also for higher energies (checked up to 500 GeV), the inefficiency at 90 degrees remains, effect a lot larger than in the pion case
CLICdp work in progress
trueθ cos1− 0.5− 0 0.5 1
0.75
0.8
0.85
0.9
0.95
1
muon identification efficiency, E=50 GeV
CLICdp work in progress
10 GeV 50 GeV
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
trueθ cos1− 0.5− 0 0.5 1
0.6
0.65
0.7
0.75
0.8
0.85
0.9
electron identification efficiency, E=10 GeV
Electron Identification Efficiency
31
PandoraPFA not designed to perform perfect isolated electron ID, but to find electrons within jets à ID might need further development for low energies
Close to 90 % in inner barrel, drop to 75-80 % in endcap
CLICdp work in progress
10 GeV
trueθ cos1− 0.5− 0 0.5 1
0.65
0.7
0.75
0.8
0.85
0.9
0.95
1
electron identification efficiency, E=250 GeV
CLICdp work in progress
250 GeV
Above 95 % almost everywhere
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
trueθ cos1− 0.5− 0 0.5 1
0.6
0.65
0.7
0.75
0.8
0.85
0.9
electron identification efficiency, E=10 GeV
Electron Identification Efficiency
32
PandoraPFA not designed to perform perfect isolated electron ID, but to find electrons within jets à ID might need further development for low energies
Close to 90 % in inner barrel, drop to 75-80 % in endcap
CLICdp work in progress
10 GeV
trueθ cos1− 0.5− 0 0.5 1
0.65
0.7
0.75
0.8
0.85
0.9
0.95
1
electron identification efficiency, E=250 GeV
CLICdp work in progress
250 GeV
Above 95 % almost everywhere
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
trueφ
3− 2− 1− 0 1 2 3
true
θ
1.4
1.45
1.5
1.55
1.6
1.65
1.7
0.45
0.5
0.55
0.6
0.65
, E=200 GeV+τ-τ 3prong Efficiency -τ
trueφ
3− 2− 1− 0 1 2 3
true
θ
1.4
1.45
1.5
1.55
1.6
1.65
1.7
0.74
0.76
0.78
0.8
0.82
0.84
0.86, E=200 GeV+τ-τ 1prong Efficiency -τ
Hadronic Tau Identification Efficiency
33
Checked on tau tau events in central part of detector à true tau jet spatially matched to reconstructed tau jet, check for correct charge identification and correct tau decay mode
Tau-energy 200 GeV, 1 prong Around 74-82 %
CLICdp work in progress
CLICdp work in progress
Tau-energy 200 GeV, 3 prong Around 48-60 %
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
trueφ
3− 2− 1− 0 1 2 3
true
θ
1.4
1.45
1.5
1.55
1.6
1.65
1.7
0.160.180.20.220.24
0.260.280.30.320.34
, E=500 GeV+τ-τ 3prong Efficiency -τ
trueφ
3− 2− 1− 0 1 2 3
true
θ
1.4
1.45
1.5
1.55
1.6
1.65
1.7
0.48
0.5
0.52
0.54
0.56
0.58
0.6
0.62, E=500 GeV+τ-τ 1prong Efficiency -τ
Hadronic Tau Identification Efficiency
34
Large drop when going to higher energies à not understood so far Reduced efficiency at polar angles of 90 degrees à not understood so far
Tau-energy 500 GeV, 1 prong Around 50-60 %
CLICdp work in progress
CLICdp work in progress
Tau-energy 500 GeV, 3 prong Around 18-30 %
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
Photon Performance: unconverted photons
35
[GeV]γtrueE
0 500 1000 1500
[%]
γtru
e)/E
PFO
REC
O(Eσ
0
1
2
3
4|<0.25θPhoton |Cos
|<0.50θPhoton 0.25<|Cos
|<0.75θPhoton 0.50<|Cos
|<0.87θPhoton 0.75<|Cos
|<0.98θPhoton 0.87<|Cos
Relative Photon Energy Resolution in different detector regions
CLICdp
[GeV]γtrueE
0 500 1000 1500
) [ra
d]PF
OR
ECO
(phi
σ
0
0.0002
0.0004
0.0006
0.0008
0.001|<0.25θPhoton |Cos
|<0.50θPhoton 0.25<|Cos
|<0.75θPhoton 0.50<|Cos
|<0.87θPhoton 0.75<|Cos
|<0.98θPhoton 0.87<|Cos
Photon phi position resolution in different detector regions as function of true photon energy
CLICdp
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
MC truth Hadron Energy spectrum for CLIC
36
E [GeV]0 500 1000 1500
1
10
210
310
410 uds→Z
neutrons, 500 GeVK0L, 500 GeVneutrons, 3000 GeVK0L, 3000 GeV
E [GeV]0 50 100 150 200 250
1
10
210
310
410 uds→Z
neutrons, 500 GeVK0L, 500 GeVneutrons, 3000 GeVK0L, 3000 GeV
For 500 GeV dataset neutral hadron energies beyond 90 GeV are 1.9 %, for 3000 dataset 13.7 % à if we want same coverage of neutral hadron energy spectrum need to calculate weights for samples up to far higher energies (1.7 % of hadrons with energies beyond 400 GeV for 3000 GeV sample)
Zàdijets at 500 vs 3000 GeV
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
] Samples and Software used
37
Produce single particle gun samples of neutrons and K0L’s separately, for each point simulate and reconstruct 70000 events Use the PandoraSettingsSoftwareCompensationTraining script for reconstruction Cleaning of clusters in the Pandora training script identical to cleaning for default reconstruction àThen run PandoraPFACalibrate_SoftwareCompensation script in PandoraAnalysis/calibration Merge Kaons and neutrons in one sample (relative weight 1:1) and use energy points of 2,5,10,30,75,150,200,400,1000 for software compensation training Density binning: 0 2 5 7.5 9.5 13 16 20 23.5 28 33 40 50 75 100, overflow 110
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
] Check on hit energy densities for very high K0L
38
So far did not yet change the binning of weights, maybe should extend weights to densities of 100/150 GeV/dm3
]3 [GeV/1000 cmρHit energy density 0 5 10 15 20 25 30
hits
N
210
310
410 neutrons, 30 GeV
K0L, 30 GeV
]3 [GeV/1000cmρHit energy density 0 50 100 150 200
[HC
AL]
hits
N
310
410
510
610 neutrons, 1000 GeV
K0L, 1000 GeV
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
] Software Compensation
Follow the procedure from software compensation paper draft https://arxiv.org/pdf/1705.10363.pdf
39
The three parameters depend on the energy of the cluster àweight shape changes quite a bit for different input hadron energies à Use automated procedure from PandoraAnalysis/calibration
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
Particle Flow Analysis: PandoraPFA
40
Particle identification at CLIC is based on the Pandora Particle flow identification algorithm • Pandora aims at reconstructing and identifying each particle reaching the
detector - The particle types are charged pions, muons, electrons, photons and neutrons
• The main objective of Pandora algorithm is to achieve very excellent jet energy resolution, needed to achieve the desired precision involving hadronic final states
Documentation: PandoraPFA algorithm: Nucl.Instrum.Meth A 611 (2009) 25 PandoraPFA at CLIC: Nucl.Instrum.Meth A 700 (2013) 153 Documentation of PandoraPFA Software Toolkit: EPJC 75 (2009) 439
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
PandoraPFA: Confusion term
41
Confusion term: effect of all pattern recognition on reconstruction, which includes errors in clustering of calorimeter hits and associating of tracks to those cluster à Important contribution to jet energy resolution, particularly at large jet energies
• Example1: energy of pion fluctuates low, then energy of a nearby “true” photon might be assigned to pion energy cluster as the sum matches the charged track momentum better, thus we measure less energy than we should have
• Example2: a lot of bremsstrahlung in the case of electrons: might give rise to additional photons, charged track from electron correctly measured, thus too large energy measured
Nonlinear non compensating natures of hadron calorimeters: • idea to correct with software for (on average) larger response of hadron
showers with large electromagnetic component, improves energy measurement of cluster energies, thus improves the confusion term
• Software compensation technique developed by CALICE à implemented in PandoraPFA now
Matthias Weber CERN
Calorimeter Session, Jan 25 Workshop 2018
]
Example: Beamstrahlung background reduction
42
Backup: Forward WW
Mark Thomson LCWS 2011, Granada 42
No background
Forward WW, no background
Backup: Forward WW
Mark Thomson LCWS 2011, Granada 42
No background
with background in recronstuction time window ~1.2 TeV
Backup: Forward WW
Mark Thomson LCWS 2011, Granada 42
No background
after timing and pT cuts: ~ 100 GeV remain: