Validation of the new menu

1
Validation of the new menu study of the PF
Clusters in HLT(addition to Friday's slides)

• M.Bachtis
• University of Wisconsin

2
The new Tau menu(I)
L1DoubleTau(ZTT)
L1SingleTau(W-gttaunu)
L1DoubleTau(ZTT)
L2Isolation(sliding)
L25Leading Track
Goes from 95-gt 100 We could do it 90-gt100 to
reduce rate
Not bad for MC only Matching! However the
alternative photon seeding will make it better
3
The new Tau menu (II)Double tau comparison
CMSSW_2_1_8
New menu
Efficiency Ref To MC L1 35 L2Threshold
25 L2Isolation 24 L25LeadTrack 17
Efficiency Ref To MC L1 18 L2ThresholdIsolati
on 16 L25Pixel Isolation 7
MC Match
ZTT
The Double Tau Efficiency for L2 threshold
corresponds to two taus. The difference makes
sense since we allow lower pt jets to pass
L1(right?).
QCD
4
PF Clusters in HLT

• The L2 code had to be modified to import the PF
  Clusters in the L2 sequence
• New module L2 TauCaloHitsSelector
• Selects the ECAL hits/towers around the L2 jet
  and feeds them to the PF Cluster Sequence
• New module L2TauModularIsolationProducer
• Calculates ECAL Isolation Cluster Shapes using
  any combination of Simple Clusters, RecHits and
  PF Clusters input
• Modules to be imported to the Trigger cfg from PF
• ParticleFlowRecHitECAL(HCAL)
• ParticleFlowClusterECAL(HCAL)
• Just 3 modules!

5
L2 Sequence with PF Clusters
L2JetsMerger
pfRecHitECAL
pfClustersECAL
jets
L2TauCaloHits
pfClustersHCAL
pfRecHitHCAL
L2TauModularIsolationProducer
L2TauRelaxingIsolationSelector
6
L2 Isolation with PF Clusters
Cone Isolation(ECAL/HCAL)
ZTT
QCD
ECAL Cluster shape
HCAL Cluster shape
7
L2 Isolation with simple Clusters
Cone Isolation(ECAL/HCAL)
ZTT
QCD
ECAL Cluster shape
HCAL Cluster shape
8
L2 Isolation with RecHits
Cone Isolation(ECAL/HCAL)
ZTT
QCD
ECAL Cluster shape
HCAL Cluster shape
9
Note

• In the PF ECAL cluster shape plots zeroth bin is
  not shown
• A lot of QCD jets with 0 clusters since the seed
  thresholds are tuned for particle ID
• However higher threshold make the tail in the
  cluster shape vanish so we can reject some more
  background using PF Clusters without sacrificing
  significant signal.

10
Timing Studies QCD50-80 particle flow
AMD Athlon 64 X2 Dual Core 4800/ Time
Summary(Double Tau)
11
Timing Studies QCD50-80standard trigger
AMD Athlon 64 X2 Dual Core 4800/ Time
Summary(Double Tau)
12
Timing Studies QCD50-80 particle flow (scaled
Time)
AMD Athlon 64 X2 Dual Core 4800/ Scaled Time
Summary(Double Tau)
13
Timing Studies QCD50-80standard trigger (scaled
time)
AMD Athlon 64 X2 Dual Core 4800/ Scaled Time
Summary(Double Tau)
14
Summary

• There seems not to be a problem with PF Clusters
  in the timing
• Assuming that we will gain a lot later we could
  add them in the paths
• We can improve even more the separation
  performance in Cluster shape variables by
  reducing the cluster seed threshold a bit
• However maybe we need to stay with the default
  which is tuned for particle identification so
  that the clusters will be more useful in creating
  the PF Candidates in L2.5?