Document 7402222

Download Report

Transcript Document 7402222

ATLAS muon system performance
ATLAS Trigger and Physics Workshop
Muon Trigger and Combined Performance Session
David Adams
Brookhaven National Laboratory
May 29, 2006
ATLAS
Updated 19:00 May 29, 2006
Contents
Goals
AOD muons
Selecting muons
Performance evaluation
• Monte Carlo muons
• Matching
• Performance parameters
Example results
Summary results
• Type contributions to
efficiency and fakes
• Misreconstructed fraction
• Type contributions to good
efficiency and misrecos
• Resolutions
• PTsum tail
Summary comments
Issues
Conclusions
• Efficiency
• Fakes/event
• Fakes/reals
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
2
Goals
Assess the performance of the muon reconstruction software
• Release 11.0.X
• CSC production samples
Examine the view presented for physics analysis
• Two AOD collections
– MuidMuonCollection
– StacoMuonCollection
• Learn how to select a sensible set of muons from these collections
– Kinematic and quality cuts
Establish metrics for evaluating performance
• Efficiency, fake rate, resolutions, …
• For different selections
• As functions of kinematic variables
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
3
AOD muons
AOD has two muon containers
• StacoMuonCollection
• MuidMuonCollection
Each muon has one or more of four track particles
• inDetTrackParticle
– Inner detector track, e.g. from low-pT algorithm
• muonSpectrometerTrackParticle
– From standalone muon finding
– Parameters at entrance to the muon system
• muonExtrapolatedTrackParticle
– Previous extrapolated to the beamline DCA
• combinedMuonTrackParticle
– Combination of inner and standalone muon tracks
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
4
AOD muons
I divide these into four categories
• Combined (0100)
– Have a combined inner detector and muon spectrometer track,
– are “best match” and
– have good fit and match chi-squares
• Split (1000)
– Combined with bad fit or match chi-square
– Inner detector track is used
• Standalone (0010)
– Tracks which have an muon spectrometer track propagated to the
beam line
– Excluding combined or split if enabled
• Inner (0001)
– Muons which have an inner detector track but do not have a muon
spectrometer track
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
5
AOD muons (cont )
Quality parameters
• Needed to distinguish real from fake muons
• Fit and match chi-squares are present but don’t provide much
discrimination
• Isolation helps but we don’t want to lose all non-isolated muons
• There are methods to return the number of hits for each detector
type but they return incorrect values
• Associated digit counts are available only for Muid low-pT
– I use a crude cut to reduce the number of Muid low-pT fakes
• It would be very helpful to have information about misses, i.e.
active detectors that were crossed with no hit or segment found
– Inner detector calls these “holes” and, for selected subdetectors,
records the number of such layers
– Complicated muon geometry would probably be better handled with
some sort of likelihood estimate for each miss and/or the whole track
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
6
Selecting muons
I use the following job options to select muons:
• Name of the muon container (Staco… or Muid…)
• Flags indicating which categories of muons to accept
– I consider three cases for each of the two containers:
> 0010 = standalone only
> 0100 = combined only
> 1111 = all four (combined, split, standalone and inner)
• Kinematic cuts
– pT above some threshold (3 GeV/c in the following)
– abs(z0) (z at the DCA) below some value (not used)
• Fit chi-square limit for all track types
– Fit chi-square/DOF (below 100.0)
– Loose to allow for kinks (brem); tighter cuts follow
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
7
Selecting muons
• Chi-square limits to define good combined muons
– Fit chi-square/DOF below 5.0
– Match chi-square below 20.0
– Those above these values are “split”
• Isolation
– Muon is said to be isolated if it has an isolation energy less than 10
GeV (0.45. cone)
– Split, standalone and inner muons are required to be isolated
– Non-isolated, combined muons are subject to tight chi-square cuts
> Fit chi-square below 2.5
> Match chi-square below 10.0
• Associated digit counts
– Muid low-pT (inner) muons are required to have more than 10 digits
(precision or trigger)
– To reduce the large number of fakes
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
8
Selecting muons
Overlaps
• Muons are said to have spatial overlap if they are within dR of 0.50
• The following are removed:
– Split, standalone and inner muons overlapping a combined muon
– Split and standalone muons overlapping an inner muon
– Inner muons overlapping an inner muon with smaller fit chi-square
Shared tracks
• Muons may removed if they share tracks with another muon
• The following would be removed
– Combined muons sharing any track with a combined muon with larger match
chi-square
– Split, standalone and inner sharing any track with a combined muon
– Split muons sharing an inner track with an inner muon.
• However, shared track cuts were not applied because some muons in the
Muid container have invalid track assignments
– Expect that shared track filter is redundant with overlap filter
– Future studies will likely add shared filter and reduce the filtering in the
overlap filter
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
9
Selecting muons
Cut summary
•
•
•
•
•
Container: Muid or Staco
Categories: standalone (0100), combined (0010) or all (1111)
Kinematics: pT > 3 GeV/c, no cut on z0
Chi-square/DOF (all muons): below 100.0
Combined/split discrimination (split if either fails)
– Fit chi-square/DOF below 5.0
– Match chi-square below 20.0
• Isolation: 10 GeV/c in 0.45 cone
– Split, standalone and inner muons required to be isolated
• Chi-squares for non isolated combined muons:
– Fit chi-square/DOF below 2.5
– Match chi-square below 10.0
• Muid inner muons required to have more than 10 digits
• Overlaps removed if dR below 0.5
• Shared track filter disabled
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
10
Performance evaluation
For Monte Carlo samples, we can evaluate the performance
of the muon reconstruction using truth information
• Select a collection of reconstructed muons (evaluation sample)
• Select a collection of Monte Carlo muons (reference sample)
• Match the two (one reco to one MC)
– Efficiency is (# matched MC)/(# MC)
– Fake rate is the (# unmatched reco)/event
– Compare matched reco and MC parameters to calculate resolutions
and fraction of misreconstructed tracks
Following sections describe
•
•
•
•
ATLAS
Selection of MC muons
Matching algorithm
Precise definitions of efficiency, fake rate, …
Results for selected CSC samples
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
11
Monte Carlo muons
Monte Carlo muons are taken from the truth particle container
• Includes prompt muons and those produced in the tracking volume
• Assume all such muons (above our pT threshold) are recorded
Monte Carlo selections:
• Loose selection to construct the reference sample for matching
– Use final state particles (ATLAS algorithm)
– pT above job option threshold (2 GeV/c in the following)
– PDG ID must correspond to muon (+/-13)
• Tight selection to evaluate efficiency (after matching)
–
–
–
–
ATLAS
Increase pT threshold (4 Gev/c)
Maximum value for eta (2.7)
Maximum value for muon production radius (2 cm)
All can be set with job options
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
12
Matching
Each reconstructed track is matched to the nearest MC track
• Using reco-to-MC distance DR = sqrt[(TR – TMC) WR (TR – TMC)]
– TR = reconstructed track vector
– WR = inverse of error matrix for track vector
– TMC = Monte Carlo value for the track vector
• If no MC track is found within some maximum distance DRmax,
then the reconstructed track is left unmatched
– In the following, DRmax is set to 1000
A MC-to-reco distance is calculated for each match: DMC =
sqrt[((jR-jMC)/Dj)2 + ((hR-hMC)/Dh)2 + ((pTR-pTMC)/pTMC/DpTrat)2 ]
• Matches with DMC > DMCmax are dropped
• DMCmax, Dj , Dh and DpTrat are all job options
– For this study: 100, 0.005, 0.005 and 0.03 were used
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
13
Matching (cont)
Only one reco track is allowed to match each MC track
• Matches with larger DMC are dropped
• Not DR because it favors tracks with large errors
Tracks classified according to their match status
• Matched tracks are found
• Unmatched MC tracks are lost
• Unmatched reco tracks are fake
Found tracks are well-reconstructed (good) if
• DR < DRgood and DMC < DMCgood
– Limits are job options
– Both set to 10 for this study
• Good tracks are used to evaluate resolutions
• Tracks found but not good are called bad or misreconstructed
Efficiency and resolutions are calculated using tight MC cuts
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
14
Matching (cont)
Reco
MC
Found
Fake
Lost
Matching example
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
15
Matching (cont)
Reco-to-MC distance
whbb, Muid combined
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
16
Matching (cont)
MC-to-reco distance
whbb, Muid combined
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
17
Matching (cont)
MC parameter summary
• Efficiency is presented for all MC muons with
– pT > 4 GeV/c
– eta < 2.7
– r0 < 2.0 cm
• Found muons are labeled as “good” if
– DMC < 10.0
– DR < 10.0
 Dj = Dh = 0.005
> DpTrat = 0.03
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
18
Performance parameters
The following are used to evaluate the performance
•
•
•
•
•
•
•
Efficiency = (# found) / (# MC)
Good fraction = (# good) / (# found)
Misreco fraction = 1 – (good fraction)
Good efficiency = (# good) / (# MC)
Fakes/event = (# fake) / (# events)
Fakes/reals = (# fake) / (# found)
Resolution in X = RMS of (XR – XMC)
– X may be h, j, pT or any of the track parameters
– Evaluated for good (well-reconstructed) tracks
In addition, pTsum tails are calculated
• Each event, vector sum over pT is calculated for both MC and reco muons
• The magnitude of the vector difference between these sums is calculated
• Fraction of events with this difference exceeding some threshold is the
pTsum tail
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
19
Example results
Following slides show some example plots
• Efficiency
• Fakes
• pT sum
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
20
Found
Lost
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
21
Fakes
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
22
MC
Found
Residual
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
23
Summary results
Summary results are presented on the following pages
• Six sets of muons:
– Staco and Muid for three cases
> Standalone
> Combined
> All (combined, split, standalone and inner)
• Eleven datasets
–
–
–
–
–
–
–
–
–
–
–
ATLAS
s050 - csc11.007231.singlepart_mu50.recon.AOD.v11004103
s100 - csc11.007233.singlepart_mu100.recon.AOD.v11004103
zmm - csc11.005151.McAtNloZmumu.recon.AOD.v11004103
wmu - csc11.005101.JimmyWmunu.recon.AOD.v11004103
whbb - csc11.005850.WH120bb_pythia.recon.AOD.v1100410
h1304l - csc11.005300.PythiaH130zz4l.recon.AOD.v11004103_bnl
h170wwll - csc11.005320.PythiaH170wwll.recon.AOD.v11004107
G500mm - csc11.005621.Gmm_500_pythia_photos.recon.AOD.v11004103
j1 - csc11.005010.J1_pythia_jetjet.recon.AOD.v11004103
j3 - csc11.005012.J3_pythia_jetjet.recon.AOD.v11004103
j6 - csc11.005015.J6_pythia_jetjet.recon.AOD.v11004103
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
24
Efficiency
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
25
Fakes/event
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
26
Fakes/reals
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
27
Type contributions to efficiency and fakes
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
28
Misreconstructed fraction
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
29
Type contributions to good efficiency and misrecos
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
30
Resolutions
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
31
Resolutions (cont)
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
32
Resolutions (cont)
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
33
Resolutions (cont)
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
34
Resolutions (cont)
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
35
Resolutions (cont)
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
36
Resolutions (cont)
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
37
Resolutions (cont)
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
38
pT sum tail
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
39
pT sum tail (cont)
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
40
pT sum tail (cont)
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
41
Summary comments
Standalone
• Both Muid and Staco do a good job of finding isolated muons
– Efficiency: 95% in single muon, Zmm and Wm samples
– 80% in busier Higgs events
– Falls dramatically in high-pT jets
> Due to isolation cut required to reduce fakes
> Need some work here to improve crude isolation cut
 Probably best to calculate a likelihood during reconstruction
that can be used to discriminate between reals and fakes
• Fakes are 2-5X worse in Muid
– In j6, Muid is 0.02/event, Staco is 0.007/event
• Resolutions and pT sum residuals are similar
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
42
Summary comments (cont)
Combined
• Staco efficiency is better
– 3-5% in single muons and high-pT samples
– 10-15% in jets
• Efficiencies flat around 88% for Staco
– Much of the decrease from standalone is due to the eta coverage of
inner tracker
• Fakes, resolutions and pt sum residuals are similar for Staco and
Muid
– Fakes up to 0.05/event in j6
• Muid has 3-8X fewer misreconstructed muons in high-pT samples
– But Staco and Muid are the same in the dijet events
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
43
Summary comments (cont)
All (combined, split, standalone and inner)
• Staco and Muid efficiencies similar in high pT events: 95%
– In jets, Staco performs better
> For j6, Staco is 86%, Muid is 73%
• Staco fakes range from comparable to 2X less than Muid
– Muid getting efficiency boost and fakes from inner muons
• Misreconstructed fractions and resolutions are similar
– Staco has slightly better pT resolution
– Muid is slightly better with eta and phi resolutions
• pT sum residual
– In both cases, combined does better than all
> Less fakes and misrecos more important than extra efficiency
(mostly at high eta)
– Staco tails are generally 2-3X lower than those for Muid
> But Staco and Mui jet results for combined muons are very
similar
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
44
Issues
Hit counts are missing or wrong in the Muid AOD
• Could use this as a measure of track quality
There are incorrect track assignments in the Muid AOD
• Cannot filter muons with shared tracks
Poor fit quality at high eta
• Presumably due to CSC problems
Need better AOD handles to discriminate against fakes for
inner (low-pT) tracks
• E.g. likelihood that track would produce the observed signals (and
absence of signals) in the muon detectors
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
45
Issues (cont)
Like to have track positions at muon spectrometer “entrance”
• To filter standalone overlaps
• Some common surface (cylinder for central, z-plane for endcaps)
• Or can (and should) we use the muon spectrometer track particle?
Like to have truth muons produced outside tracking volume
• To identify “fakes” due to pi/K decays in calorimeter
Like to make use of calorimeter information in finding
• Find muons where muon spectrometer is blind
• E.g. Wisconsin topo cluster method
• YATA (Yet Another Tracking Algorithm)
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
46
Conclusions
It is now (release 11.0.X) possible to analyze AOD muons from both
Staco and Muid
• Very similar analysis algorithms
• Allows direct comparison
Good performance for combined muons for either collection
• Efficiency 80-90%
• Fakes around 0.01/event approaching 0.1/event in high-pT jets
Addition of split, standalone and inner muons improve performance
• Efficiency increases to 95% for high-pT muon
– Biggest gain at high eta
• Fakes kept at similar levels with isolation cuts
– Require all split, standalone and inner muons to be isolated
– Apply tighter cuts for non-isolated combined muons
Need better discriminants (and more work) to simultaneously improve
efficiency and fake rejection
ATLAS
D. Adams
Muon performance
ATLAS Trigger and Physics
May 29, 2006
47