PandoraPFA and LCFIVertex with GLD data

1
PandoraPFA and LCFIVertexwith GLD data

• S. Uozumi (Kobe)
• Apr-23th ILD detector optimization WG meeting

2
PandoraPFA Performancewith the GLD
3
With LDC00 Z-pole full tracking PandoraPFA,
Marks result JER 23.5 (pandora v2-01) My
result 28.5 (pandora
v2-00) Where does 5 difference come from ?

• At TILC08 Sendai, Mark told us
• We are more-or-less doing a right thing,
• there seems to be nothing apparently wrong.
• Mark generates events with a pythia setting tuned
  for LEP experiment, which gives 20 less neutral
  particles in a jet. It will give 1.5 effect on
  JER.
• Then Mark gave us his z-pole data (both lcio file
  
• after detector simulation and stdhep files)
  and
• steering file.

4
JER with various configurations (Z-pole)

• There are still some difference, but JER values
  are OK (lt30) for
• ILD optimization study.
• We decided to leave more detailed study to be
  done sometime
• in future, and move ahead on the optimization
  studies anyway.

5
Flavour Tagging Performancewith the GLD
6
Z-pole data Process for FT study

• Jupiter data (GLD) 10k events
• Jupiter data converted to lcio
• Z -gt anything, but leptonic decays (30) are
  rejected
• No ISR
• Mokka (LDC01_05Sc) 10k events
• Told by Sonja, copied from DESY GRID
• Z -gt qq (qu,d,s,c,b) events
• Flavour tagging by FullTracking PandoraPFA
  LCFIVertex with LDC-tuned neural net parameters.

Breakdown with true jet flavours




7
c-tagging performance
Jupiter GLD
uds
b
c
NN output for c-tag
Mokka LDC01_05Sc w/ conv
uds
b
c

• Result with LDC01fulltracking doesnt perfectly
  reproduce
• Sonjas result. But with cheated result,
  agreement with Sonja becomes better.
• Result with GLD data looks consistent with
  Sonjas LDC01_05Sc result anyway.

8
b-tagging performance
Jupiter GLD
uds
c
b
NN output for b-tag
Mokka LDC01_05Sc w/ conv
uds
c
b
NN output for b-tag
Results with different configuration are slightly
different, but b-tag peformance is almost
acceptable with the GLD data.
9
Summary

• We still can not reproduce the Marks JER result
• (3 worse), but further investigation is kept
  for future.
• Also JER with GLD data is still worse than Marks
  result,
• but OK for starting analyses of benchmark
  processes.
• FT performance we get with LDC01_05Sc is still
  slightly different with Sonja. Maybe issue of
  full tracking ?
• FT Performance with GLD data is consistent with
  Sonjas result with LDC01_05Sc.
• Now we are comparing JER and FT performance among
  GLD, GLDprime and J4LDC geometries.
• Also starting analyses of benchmark processes
  which use Pandora LCFIVertex.

10
Backups
11
Input variables for flavour tagging with
Neural-Net

• NumVertices
• PTCorrectedMass
• RawMomentum
• SecondaryVertexProbability
• Z0Significance1
• Z0Significance2
• D0Significance1 (zoomed)
• D0Significance2 (zoomed)
• Z0Significance1 (zoomed)
• Z0Significance2 (zoomed)

• D0Significance1
• D0Significance2
• DecayLength
• DecayLength(SeedToIP)
• DecayLengthSignificance
• JointProbRPhi
• JointProbZ
• Momentum1
• Momentum2
• NumTracksInVertices

12
b-jet events (any number of vertices)
Black GLD Red LDC01_05Sc
DecayLength Significance
D0Significance1
D0Significance2
DecayLength
DecayLength (SeedtoIP)
JointProbZ
JointProbRPhi
Momentum1
Momentum2
NumTracks InVertices
NumVertices
SecVertexProb
Z0Significance1
PTCorrectedMass
RawMomentum
D0Significance1 (zoom)
D0Significance2 (zoom)
Z0Significance1 (zoom)
Z0Significance2 (zoom)
Z0Significance2
13
c-jet events (any number of vertices)
Black GLD Red LDC01_05Sc
DecayLength Significance
D0Significance1
D0Significance2
DecayLength
DecayLength (SeedtoIP)
JointProbZ
JointProbRPhi
Momentum1
Momentum2
NumTracks InVertices
NumVertices
SecVertexProb
Z0Significance1
PTCorrectedMass
RawMomentum
D0Significance1 (zoom)
D0Significance2 (zoom)
Z0Significance1 (zoom)
Z0Significance2 (zoom)
Z0Significance2
14
uds-jet events (any num. of vertices)
Black GLD Red LDC01_05Sc
DecayLength Significance
D0Significance1
D0Significance2
DecayLength
DecayLength (SeedtoIP)
JointProbZ
JointProbRPhi
Momentum1
Momentum2
NumTracks InVertices
NumVertices
SecVertexProb
Z0Significance1
PTCorrectedMass
RawMomentum
D0Significance1 (zoom)
D0Significance2 (zoom)
Z0Significance1 (zoom)
Z0Significance2 (zoom)
Z0Significance2