L2 B Tagging

1
L2 B Tagging

• Bad news Pre-STT parameters
• Generator level
• L2 trigsim
• Good news STT parameters
• Purity vs. Efficiency
• Correlation
• Good news SVM

2
MC Data Generator Level

• Pythia
• b b-bar incl. / QCD incl.
• 2000 events each
• 0 min bias
• p08.10

3
MC Data L2 trigsim

• Pythia, Pt gt 20
• b b-bar incl. / QCD incl.
• 3260 events / 5000 events
• 0 min bias / 0.5 min bias
• mixture
• p10.06.01
• Data dumped by filter as l2 analyze packages
  broken

4
Pre-STT parameters

• Generator level

5
Pre-STT parameters

• L2 trigsim data

6
Pre-STT Parameters

• Not feasible.
• Main problems
• small number of tracks / jet
• binning of tracks phi (160 segments)
• Jets eta binned into 0.05 eta segments, only Jet
  Et known, large error on Jet E for high eta jets
• Only way out STT
• DCA parameters have much higher suppression power
  (about 1 OoM)

7
STT Parameters

• Decay length
• Axial tracks (within 1.6 cone) projected onto jet
  theta
• Sum over all distances primary vertex - track /
  jet intersection

Purity
Efficiency
8
STT Parameters

• Ratio of tracks with high decay length
• Number of tracks within 1.6 cone and LD gt 10mu
  (OoM of STT resolution) divided by number of
  tracks within 1.6 cone

9
STT Parameters

• Decay Length Area
• Areas of the triangle spanned by
• Primary vertex
• PCA
• Intersection of track and jet
• Summed over all tracks with positive DL and
  relative angle lt 1.6

Jet
DLA
Track
Primary Vtx
10
STT Parameters

• Decay Length Area
• Track with spatial error small relative angle
  can result in huge DL, but small DLA
• Track with spacial error high relative angle
  results in precise DL, big DLA
• Relatively independent of jet axis errors
• Phi no error if as many tracks left of jet as
  right
• Eta Projected
• Tracks DCA and phi resolution well known, and so
  is DLAs.

11
STT Parameters

• Decay Length Area

12
STT Parameters

• Correlations

x DLA y DLA
x DL y DLR
x DLA y DLR
13
SVM

• SVM output for MC Generator pre-STT parameters
• 3400 jets in training sample (approx 50 QCD)
• 1000 jets in test sample (50 QCD)
• Non-standard implementation of SVM Improvement
  by using gaussian kernel with parameter dependant
  width

14
SVM

• SVM output for MC Generator pre-STT parameters