Partially reconstructed electrons for ZZ4l search

1
Partially reconstructed electrons for ZZ?4l search

• Azeddine Kasmi
• Bob Kehoe
• Southern Methodist University
• Thanks to H. Ma, M. Aharouche, P. Renkel

2
Motivations

• 3 leptons has higher acceptance than 4 leptons
  reconstructed.
• Get the full acceptance for
• ZZ/Higgs? 4 leptons searches by considering
  the Higgs/ZZ ? 3l X
• However, background is higher
• In this talk, the emphasis is on cluster based
  approach
• Use Mohameds forward electron as soon as MC data
  is available (14.4.0)

Medium electron N3e 2 N4e
Number of medium electrons/event
3
The Monte Carlo Samples

• Data reconstructed in 14.2.21
• Z ? ee mc08.106050.PythiaZee_1Lepton.recon.e347_
  s462_r522
• Reconstructed on release 13
• ZZ(5931) ?4l MC_at_NLO
• WZ(6359) ?4l Jimmy
• Zbb(5176) ?3l AcerMC
• Zb(6540) ?3l AcerMC
• Ttbar(5211) ?4l MC_at_NLO

4
CaloTopoClusters performances (Z ? ee)
Eta converge gt 2.5
Eta of topocluster that matches the truth electron
DR between a topocluster candidate and the
unfound truth electron in Z
e 96
Matched Topocluster candidate
Truth unfound Z electron
All the cluster have to satisfy a PT gt 7 GeV
5
RMS 0.21
RMS 0.27
PT gt 20 GeV
PT in the range 7GeV to 20 GeV
Very similar to those of the egamma algorithm
6
Summary table of performances for ZZ Sample
Similar
N.B. We are looking for a cluster without regard
to a track. i.e. a photon like object.
7
Particle ID of the topoclusters The normalized
second longitudinal moment

• l is the distance of the cell from the shower
  center along the shower axis
• long2 ltl2gt , with l 0 for the
• two most energetic cells
• longmax ltl2gt , with l 10 cm
• for the two most energetic cells and l 0
  for all other cells
• 0 means shower more compact in 3D
• 1 means shower more spread

ZZ Zb Zbb
8
Particle ID of the topoclusters The normalized
second lateral moment

• r is the distance of the cell from the shower
  axis
• lat2 ltr2gt , with r 0 for
• the two most energetic cells
• latmax ltr2gt , with r 4 cm
• for the two most energetic cells and r 0
  for all other cells
• 1 means big showers,
• and 0 means small showers

ZZ Zb Zbb
9
Isolation Max energy fraction
ZZ Zb Zbb

• Energy fraction of the most energetic cell

The layer energy weighted fraction of
non-clustered neighbor cells on the outer
perimeter of the cluster
10
Longitudinal vs. Isolation
Background
Signal
11
The likelihood method
These distributions can be used to assign a
probability for a given topocluster to be signal
or background
and
Multiplication of these variables gives the
overall probability for the event.
The likelihood discriminant is defined as
following
12
Getting the Ps and Pb

• Get the Ps from Z ? ee by requiring one medium
  electron in the event and the second one will be
  a cluster from unreconstructed electron
• Get the Pb from tt sample
• Normalize to 1 to get the pdfs

13
The probabilities for signal and background (pdf)
S B
S B
Normalized second lateral moment
Normalized second longitudinal moment
14
The probabilities for signal and background (pdf)
S B
S B
Energy fraction in the most energetic cell
The layer energy weighted fraction of
non-clustered neighbor cells on the outer
perimeter of the cluster
15
Likelihood Signal is ZZ and Background is Zbb
S B

• e of signal events that passes a given
  likelihood/total of signal event
• Fake rate (f) of background events that
  passes a given likelihood/total of background
  event

16
Shapes of the backgrounds likelihoods
For L gt 0.3 e 82 fZbb 19 fZb 25 fWZ
23 fttbar 15 fb 12
WZ Zbb Zb b
The shapes are similar
17
Likelihood dependence on h

• The pdfs are strongly h dependent
• The barrel is divided into three regions
• Barrel-1 hlt 0.7
• Barrel-2 h gt 0.7 and hlt1
• Barrel-3 hgt1 and hlt1.375
• End Cap
• EC-1 hgt1.375 and hlt1.9
• EC-2 hgt1.9 and hlt2.5
• EC-3 hgt2.5 and hlt3.2
• FCal
• hgt3.2

18
Signal efficiency and fake rates for L gt 0.5
19
Efficiency vs. PT
Fake vs. PT
ZZ diboson
Zbb sample
Low PT lowers the efficiency
20
Conclusion

• In the case of ZZ?3l X analysis, The use of
  CaloTopocluster seems to be promising.
• The use of the likelihood on particle ID improves
  the signal efficiency and lowers the fake rate.
• The Likelihood dependence on PT and h should be
  studied more carefully to optimize the use of the
  likelihood.
• I will use Mohameds implementation of the
  forward electrons in 14.4.0 as soon as MC data
  becomes available

21

• Back up slides

22
The Z mass using a medium electron and a matching
cluster to the truth unfound electron
s 3.15
s 3.60
Z in the event where 2 Medium isolated electrons
where reconstructed. With opposite charge
Medium electron and a topo cluster matching the
unfound truth electron
23
Lateral vs. Isolation
Background
Signal
24
Lateral vs. Max_Efraction
Background
Signal
25
Longitudinal vs. Lateral
Signal
Background
26
Correlations Matrices
Background
Signal