Noise in jets study

1
Noise (in) jets study
Bernard Andrieu (LPNHE,Paris) JetMET meeting
02/27/2003
2
Goal Separate good from fake jets
Definition fake made out of noise (
possible minimum bias energy)
good not fake (still it can include some
noise!)
Data Run 169183, 169222, 169262 (????K events),
p13.05.00 No trigger requirement. Processed with
top_analyse. Cone 0.5 jets, standard ID cuts
(without CHF cut) ? 0.05 lt EMF lt 0.95 ? n90 gt
1 ? HotF lt 10
Monte-Carlo ctf_p1308_qcd_pt40_sig2.5_tmb
3
How to select good jets? (I)
Trigger validation L1set a Could bias
selection (CH not included in L1 readout)
Other selection variables
Jet-track matching dR(jet,track)
Jet-Jet matching in the transverse plane
DeltaPhi (0 for back-to-back jets)
a Same remark as for track-jet matching
Missing Et Ptmiss/sqrt(ET)
a Could also suppress noisy (but still) good
jets
4
How to select good jets? (II)
(without introducing selection biases)
Solution (?)
- Select events with only two jets (even before
quality cuts)
f90
- Cross-checks between variables
l1set/PT
(see effects on one variable when cutting on
another independent one)
- Compare with Monte-Carlo
5
How to select good jets? (III)
6
How to select good jets? (IV)
dR(cal-track)
DeltaPhi
7
Final good jets selection
Exactly two jets (no 3rd jet, even not passing
the quality cuts)
Leading jet validated by L1 energy (L1set/PTgt0.3)
Minimal requirement on L1 energy for second jet
(L1set/PTgt0.)
40 lt S PTjet lt 60
60 lt S PTjet lt 80
80 lt S PTjet lt 100
DeltaPhi
100 lt S PTjet lt 140
140 lt S PTjet lt 200
200 lt S PTjet lt 300
Reasonable agreement DATA/MC
8
f90 for good jets (data vs MC)
Data
MC
f90 depends mainly on the number of merges
otherwise data and MC similar
9
emf chf for good jets (data vs MC)
emf
chf
0 merge
1 merge
gt 1 merge
Discrepancy DATA/MC for high emf (0 merge) and
high chf

10
f90 correlations for good jets (data)
f90 vs log10(E)
f90 vs chf
mergegt1
merge0
merge1
f90 depends on E for mergelt1, doesnt depend on
chf (at a fixed merge)
11
f90 correlations for good jets (MC)
f90 vs log10(E)
f90 vs chf
merge0
merge1
mergegt1
Lack of statistics at low E, similar correlations
as in data
12
How to select fake jets?
13
Distributions for fake jets
Leading jet
Second jet
merge
eta
f90
More merges than in data, but similar f90 at
fixed merge
14
f90 correlations for fake jets
f90 vs log10(E)
f90 vs chf
Leading jet
Second jet
Lack of statistics, similar correlations as in
data (for mergegt 1)
15
Summary

• ? Confirmation of effect seen by Vishnu on
  Monte-Carlo
• good jets may have high values of f90
• ? Pure fake jets dont seem to be a problem
• How to reduce the noise for good jets?
• ? Several (complementary?) approaches
• 1) get rid of cells with high occupancy (run by
  run) Robert
• 2) get rid of individual isolated noise cells
  (event by event)
• T42 algorithm (Jean-Roch)
• 3) get rid of noisy preclusters in CH to avoid
  merges in excess
• modify seeding algorithm to raise
  threshold in CH (Emmanuel)
• 4) subtract mean noise contribution to jets,
  depending on eta, number of merges,
• energy in CH?
• ? To do estimate efficiency and purity of jets
  with present quality cuts, compare with MC at
  lower PT, improve quality cuts, re-run jet
  algorithms with modified seeding on physics
  sample (e.g. Wjet(s)), subtract noise, etc