Update on Cluster Splitting: TTbar sample at 8 TeV

Introduction MC efficiency and fakerate PU 0 MC efficiency and fakerate PU 0
Update on Cluster Splitting:
TTbar sample at 8 TeV
Silvia Taroni
Vincenzo Chiochia, Hella Snoek, Morris Swartz, Petar Maksimowic,
Kevin Nash, Marc Antoine Osherson
12th April 2013
Silvia Taroni Vincenzo Chiochia, Hella Snoek, Morris Update
Swartz, on
Petar
Cluster
Maksimowic,Kevin
Splitting: TTbar
Nash,
sample
Marc
at Antoine
8 TeV Osherson
12th April 2013
1 / 13
Introduction MC efficiency and fakerate PU 0 MC efficiency and fakerate PU 0
Outline
Introduction
MC efficiency and fakerate PU 0
MC efficiency and fakerate PU 0
Silvia Taroni Vincenzo Chiochia, Hella Snoek, Morris Update
Swartz, on
Petar
Cluster
Maksimowic,Kevin
Splitting: TTbar
Nash,
sample
Marc
at Antoine
8 TeV Osherson
12th April 2013
2 / 13
Introduction MC efficiency and fakerate PU 0 MC efficiency and fakerate PU 0
Remind: the problem
I
Tracks close together can have their clusters merged
I
The reconstructed hit has typically a small error (due to
being large) and wrong location
I
Tracking performance (efficiency and quality) seriously
affected
I
It happens in dense tracking environments in the pixel
detector and the first layers of the strip tracker
Silvia Taroni Vincenzo Chiochia, Hella Snoek, Morris Update
Swartz, on
Petar
Cluster
Maksimowic,Kevin
Splitting: TTbar
Nash,
sample
Marc
at Antoine
8 TeV Osherson
12th April 2013
3 / 13
Introduction MC efficiency and fakerate PU 0 MC efficiency and fakerate PU 0
Remind: when/what/how to split
Splitting merged cluster is not straightforward.
I
the tracking performance is lost at the seeding step.
I
to recover lost tracks, the splitting has to be done before seeding.
I
identification of merged clusters possible but relies on tracking (which relies
on seeding)
Identification of merged cluster is based on cluster charge corrected for the track
angle.
Splitting algorithm designed that floats cluster templates to the identified
merged cluster.
Track hypothesis to split: straight line from the first PV (since we are interested
in boosted object)
NB: we split clusters that should not be split, they are coming from a different
PV, or from low pt tracks
Silvia Taroni Vincenzo Chiochia, Hella Snoek, Morris Update
Swartz, on
Petar
Cluster
Maksimowic,Kevin
Splitting: TTbar
Nash,
sample
Marc
at Antoine
8 TeV Osherson
12th April 2013
4 / 13
Introduction MC efficiency and fakerate PU 0 MC efficiency and fakerate PU 0
samples and analysis
We compared the t t¯ samples (with and without PU) before and after the use of
the Cluster splitting algorithm.
Release CMSSW 5 3 5.
Selections:
I
“Standard” top muon selection, with pt > 45 GeV and PF isolation < 0.2
I
≥ 2 PFJets with pt > 30 GeV
I
Leading PFJet with pt > 200 GeV
I
≥1 PFJets have b-tag with SSVHE discriminant > 2.74
I
≥ 2 PFJets with deltaR(muon, jet) > 2.0 (labeled “hadronic-side” jets)
I
≥1 “hadronic-side” jet must have jet mass > 60 & jet mass < 100, and
explicitly fail the b-tag requirement
I
The other “hadronic-side” b-jet is the probe jet. It is over 85% pure b.
Silvia Taroni Vincenzo Chiochia, Hella Snoek, Morris Update
Swartz, on
Petar
Cluster
Maksimowic,Kevin
Splitting: TTbar
Nash,
sample
Marc
at Antoine
8 TeV Osherson
12th April 2013
5 / 13
Introduction MC efficiency and fakerate PU 0 MC efficiency and fakerate PU 0
efficiency using t t¯ MC sample (PU = 0)
track reconstruction efficiency
track reconstruction efficiency
Efficiency improvement at high pt and at low dR.
1
0.8
0.6
No splitting
Cluster splitter
Truth based splitter
0.4
0.2
0
10-1
1
0.8
0.6
No splitting
Cluster splitter
Truth based splitter
0.4
0.2
1
10
102
103
track pT [GeV]
0
0
0.005
0.01
0.015
0.02
0.025
track opening angle [mRad]
obtained using TrackAssociatorByChi2.
Silvia Taroni Vincenzo Chiochia, Hella Snoek, Morris Update
Swartz, on
Petar
Cluster
Maksimowic,Kevin
Splitting: TTbar
Nash,
sample
Marc
at Antoine
8 TeV Osherson
12th April 2013
6 / 13
Introduction MC efficiency and fakerate PU 0 MC efficiency and fakerate PU 0
MC fakerate PU 0
0.5
0.45
track reconstruction fakerate
track reconstruction fakerate
Only a very small increase of fakerate after splitting.
No splitting
0.4
Cluster splitter
0.35
Truth based splitter
0.3
0.25
0.2
0.15
0.5
0.45
0.4
0.35
0.3
0.25
0.15
0.1
0.1
0.05
0.05
0
10-1
1
10
102
103
track pT [GeV]
No splitting
Cluster splitter
Truth based splitter
0.2
0
0
0.005
0.01
0.015
0.02
0.025
track opening angle [mRad]
obtained using TrackAssociatorByChi2.
Silvia Taroni Vincenzo Chiochia, Hella Snoek, Morris Update
Swartz, on
Petar
Cluster
Maksimowic,Kevin
Splitting: TTbar
Nash,
sample
Marc
at Antoine
8 TeV Osherson
12th April 2013
7 / 13
Introduction MC efficiency and fakerate PU 0 MC efficiency and fakerate PU 0
on/off track cluster statistics in 0 PU MC
Clusters in the different categories for the different samples:
cluster type
on track
off track
No Split
323647 (56%)
250923 (44%)
Sim Split
327029 (54%)
275586 (46%)
Temp Split
328194(51%)
313240 (49%)
After splitting:
Number of tracks: + 1.6%
Number of PV: + 0.6%
The splitting is more ‘active’ on the off-track cluster.
The template splitting splits more than the sim split (on both off and on track
cluster). Why?
Silvia Taroni Vincenzo Chiochia, Hella Snoek, Morris Update
Swartz, on
Petar
Cluster
Maksimowic,Kevin
Splitting: TTbar
Nash,
sample
Marc
at Antoine
8 TeV Osherson
12th April 2013
8 / 13
Introduction MC efficiency and fakerate PU 0 MC efficiency and fakerate PU 0
average number of tracks
Looking to all the events, on average after the splitting we have 2 tracks more:
1 true track and 1 fake track if comparing with the sim split.
Silvia Taroni Vincenzo Chiochia, Hella Snoek, Morris Update
Swartz, on
Petar
Cluster
Maksimowic,Kevin
Splitting: TTbar
Nash,
sample
Marc
at Antoine
8 TeV Osherson
12th April 2013
9 / 13
Introduction MC efficiency and fakerate PU 0 MC efficiency and fakerate PU 0
splitted cluster
Looking at the clusters that have been splitted and performing a match with the
no splitted sample:
cluster type
on track
off track
No Split
27529
58951
Temp Split
30849
122441
The 15% of clusters has been splitted.
if the original cluster was on/off track, the splitted cluster are associated to 0, 1
or 2 track as in table:
original cl
on track
off track
0
1085
56390
1
24777
2362
2
1657
198
the 4% of on-track clusters becomes off-track but more off-tracks clusters
become on-track
Silvia Taroni Vincenzo Chiochia, Hella Snoek, Morris Update
Swartz, on
Petar
Cluster
Maksimowic,Kevin
Splitting: TTbar
Nash,
sample
Marc
at Antoine
8 TeV Osherson
12th April 2013
10 / 13
Introduction MC efficiency and fakerate PU 0 MC efficiency and fakerate PU 0
BACKUP
Silvia Taroni Vincenzo Chiochia, Hella Snoek, Morris Update
Swartz, on
Petar
Cluster
Maksimowic,Kevin
Splitting: TTbar
Nash,
sample
Marc
at Antoine
8 TeV Osherson
12th April 2013
11 / 13
Introduction MC efficiency and fakerate PU 0 MC efficiency and fakerate PU 0
efficiency using t t¯ MC sample (PU = 0)
track reconstruction efficiency
track reconstruction efficiency
Efficiency improvement at high pt and at low dR.
1
0.8
0.6
No splitting
Cluster splitter
Truth based splitter
0.4
0.2
0
10-1
1
0.8
0.6
No splitting
Cluster splitter
Truth based splitter
0.4
0.2
1
10
102
103
Track pT [GeV]
0
0
0.005
0.01
0.015
0.02
0.025
track opening angle [mRad]
obtained using quickTrackAssociatorByHits.
Silvia Taroni Vincenzo Chiochia, Hella Snoek, Morris Update
Swartz, on
Petar
Cluster
Maksimowic,Kevin
Splitting: TTbar
Nash,
sample
Marc
at Antoine
8 TeV Osherson
12th April 2013
12 / 13
Introduction MC efficiency and fakerate PU 0 MC efficiency and fakerate PU 0
MC fakerate PU 0
0.5
track reconstruction fakerate
track reconstruction fakerate
Only a very small increase of fakerate after splitting.
0.45
0.4
0.35
0.3
0.25
0.2
No splitting
Cluster splitter
Truth based splitter
0.15
0.5
0.45
0.4
0.35
0.3
0.25
0.15
0.1
0.1
0.05
0.05
0
10-1
1
10
102
103
track pT [GeV]
No splitting
Cluster splitter
Truth based splitter
0.2
0
0
0.005
0.01
0.015
0.02
0.025
track opening angle [mRad]
Silvia Taroni Vincenzo Chiochia, Hella Snoek, Morris Update
Swartz, on
Petar
Cluster
Maksimowic,Kevin
Splitting: TTbar
Nash,
sample
Marc
at Antoine
8 TeV Osherson
12th April 2013
13 / 13