Muon Identification in CMS

1
Muon Identification in CMS

• Eric James (Fermilab), Yurii Maravin (Kansas
  State),
• Norbert Neumeister (Purdue)
• December 6th, 2005
• CMS Physics Meeting

2
Muon Identification Algorithm

• Muon Reconstruction (Outside-In) Start by
  reconstructing a stand-alone track in muon
  detectors and attempt to match with a
  reconstructed silicon track.
• Muon Identification (Inside-Out) Attempt to
  quantify the muon compatibility of any
  reconstructed silicon track.

3
Muons in CMS
4
Basic Idea

• Propagate silicon track outwards into calorimeter
  and muon systems.
• Search in cone around extrapolated track for
  associated energy deposits in ECAL, HCAL, and HO
  (or hits and segments in the muon detectors).
• Define probabilities corresponding to the
  compatibility of track with muon hypothesis based
  on these quantities.

5
CMS Calorimeter
6
Calorimeter Algorithm

• Extrapolate track into each calorimeter volume
  taking into account energy loss in material and
  bending in B field.
• Search in a ?R cone around extrapolated position
  for towers above a minimum threshold and sum the
  observed energies.
• Cone sizes and tower thresholds are optimized
  separately for muon barrel, overlap, and endcap
  regions.

7
Calorimeter Probability

• Calorimeter Probability is obtained from a
  three-dimensional likelihood function.

PS(x) PS(y) PS(z)
PS(x) PS(y) PS(z) PB(x) PB(y) PB(z)

• PS and PB are signal and background probabilities
  as functions of the observed energies in ECAL(x),
  HCAL(y), and HO(z).

8
ECAL Distributions
pT 10 GeV/c
Barrel Region
PB(x)
PS(x)
9
HCAL Distributions
pT 10 GeV/c
Barrel Region
PB(y)
PS(y)
10
HO Distributions
pT 10 GeV/c
Barrel Region
PS(z)
PB(z)
11
Calorimeter Probability
Single pT 10 GeV/c muons and pions
Barrel
Endcap
12
Calorimeter Probability
Single pT 10 GeV/c muons
13
CMS Muon Detectors
14
Muon Algorithm

• Extrapolate track into each muon detector layer
  and define a search road for associated
  hits/segments.
• For each candidate, calculate a ?2 measuring the
  compatibility of the position and direction (for
  segments) with those of the extrapolated track.
  Define hits and segments as associated if
  corresponding ?2 is below a programmable
  threshold (default 100).

15
Muon Probability Calculation

• Probability is based on the list of observed
  hits/segments found to be associated with track.
• Each associated object is weighted based on its
  dimensional content (e.g. segments are more
  valuable than hits).
• Objects are also weighted by layer (outer layer
  hits are considered more valuable than inner
  layer hits).

16
Muon Probability Contributions

• DT/CSC Stations (1?4) .10, .15, .20, .25
• Barrel RPC Layers (1?6) .02, .04, .04, .06,
  .06, .08
• Endcap RPC Layers (1?4) .05, .05, .08, .12
• 4-dim Segments 1.00 (1 max)
• 2-dim Segments 0.50 (1 max)
• CSC Hits 0.0833 (6 max)
• DT Hits 0.0417 (12 max)
• RPC Hits 1.00

17
Example Calculation
DT S1 ? 0.10 CSC S2 ? 0.15 CSC S3 ? 0.20 RPC B1 ?
0.02 RPC B2 ? 0.04 RPC E1 ? 0.05 RPC E2 ? 0.05
RPC E3 ? 0.08 Max. Score 0.69
18
Muon Probability
Single pT 10 GeV/c muons and pions
Barrel
Endcap
19
Algorithm Performance

• Have attempted to evaluate the performance of the
  current algorithm on three simulated samples.
• pT 5 GeV/c single muons
• H ? WW ? ????
• B-quark jets (soft lepton tagging)
• For the purposes of these studies, a muon is
  considered identified if the calorimeter and muon
  detector based probabilities are above 0.8 and
  0.4.

20
Low pT Single Muons
Reconstruction Efficiency 68.7
Reconstruction plus Identification Efficiency
78.6
21
Non-Reconstructed Muons
Cut Here
22
Non-Muon Fake Rate
Single muons generated on top of minimim-bias
events (Lum. 2x1033).
Measured muon ID fake rate for all non- muon
tracks in these events (zero values are due to
insufficient statistics).
23
H ? WW ?????
Muon ID efficiency for lower pT muon in these
events (20 are below 10 GeV/c).
Muon Identification improves overall event
selection efficiency by roughly 5.
24
Non-Muon Fake Rate
H?WW????? events also generated on top of
minimim-bias events (Lum. 2x1033).
Measured muon ID fake rate for all non- muon
tracks in these events (zero values are due to
insufficient statistics).
25
Soft Lepton Tagging
Reconstruction Efficiency 71
Reconstruction plus Identification Efficiency
84
26
Soft Lepton Tagging Fake Rates (??)
Reconstruction Efficiency for Fakes 0.28
Reconstruction plus Identification Efficiency for
Fakes 0.31
27
Soft Lepton Tagging Fake Rates (K?)
Reconstruction Efficiency for Fakes 0.55
Reconstruction plus Identification Efficiency for
Fakes 0.62
28
Conclusions

• Muon Identification algorithm provides an
  additional tool for muon selection (complimentary
  to the standard muon reconstruction).
• The algorithm is potentially useful for
  recovering selection inefficiencies and could
  play an important role in detector commissioning.
  
• The algorithm can still be significantly improved
  with additional tuning.

29
Backups
30
Endcap Energy Distributions
31
Muons with Zero HCAL Deposit
32
Muons with Zero HO Deposit
33
Muon Probability
Single pT 10 GeV/c muons
34
Muons with Zero Probability
Based solely on muon detector information.
35
H?WW????? Trailing Muons
Higgs Mass 200 GeV/c2
36
Muons in Bottom Quark Jets
b-jet pT 50 to 80 GeV/c