september 20, 2002 all dzero meeting
DESCRIPTION
TARC: Report from the Mini-Workshop. September 20, 2002 All DZero Meeting (Jianming Qian), Valentine Kouznetsov, Avto Karchilava, Rick Van Kooten, HTD. T racking A lgorithm R ecommendation C ommittee Charge. Collect information on performance of various tracking algorithms about - PowerPoint PPT PresentationTRANSCRIPT
September 20, 2002
All DZero Meeting
(Jianming Qian), Valentine Kouznetsov, Avto
Karchilava, Rick Van Kooten, HTD
TARC:Report from the
Mini-Workshop
Tracking Algorithm Recommendation Committee Charge
Collect information on performance of various tracking algorithms about Efficiencies, fake rates, and misreco
rates using standard procedures developed by the global tracking group and standard (both beam and MC generated) datasets
Reconstruction logistics such as CPU time per event, memory, and luminosity dependence
Input from physics/id/algorithm groups to be solicited
Make recommendation on how we should run tracking in p13 on the taking into account the farm resources available in October assumed to be 25 Hz events 58 seconds on 500 Mhz machine
where 29 seconds is available for tracking
Note p13 is frozen October 1st, implying the TARC should move as quickly as is possible following this meeting
Tracking Algorithm Recommendation Committee Charge
Acknowledgements
A great deal of help and cooperation was available.
Special notes of thanks to: The people who make SAM work
including Lee Lueking The D0 Data Farm Reco Group and
Heidi Schellman and Mike Diesburg Tracking Group esp. V. Kouznetsov Mark Sosebee and the UTA farms Mike Strauss The physics, ID, and algorithm
groups including the speakers from Wednesday’s workshop
The Tracking Algorithm Developers
Contents of this Talk
Data Samples & Procedures Definitions Presentations from the Mini-
workshop Tracking Algorithms and performance
report (S. Khanov) Lifetime B-tagging (B. Wijngaarden) Tau Reconstruction (S. Duensing) B Hadron Reconstruction (V. Jain) EM ID Issues (R. Zitoun) Dimuon Studies (R. Hooper) Higgs Gp. Report (L. Feligioni) Secondary Vertex B-Tagging in Top
Samples (A. Schwartzman) Top Gp. Report (E. Chabalina)
Common Elements in their reports Summary
Data and Monte Carlo Samples
Data files in SAM (picked events have been merged) Run 155554 – test run of 10,000
events Run 157708 - 90,000 events TV7.31
w/ inst. lum ~ 5e30 – SMT grade C, CFT in the good run range with
full stereo readout, prior to calor zero suppression change.
38,000 dimuon events picked by B-physics Group for J/psi post full stereo readout
16,000 mu+jet events picked by BID group
5,400 picked high pT dimuons prior to full stereo readout
6,800 picked high pT diem events post full stereo readout
SAM Definitions Data
Run 155554: %reco_all_0000155554%tk-p11.11-%.root
Run 157708: %reco_all_0000157708%tk-p11.11-%.root
Mujets: %merge_mujet%tk-p11.11-%.root J/Psi to dimuons: %dimuon_third_merged
%tk-p11.11-%.root Z to ee: %pick_diem%tk-p11.11-%.root Z to mumu (not isolated): %pick_dimuon
%tk-p11.11-%.root
The third %’s above are gtr, htf, gtrela, htfela, gtrhtf, aa, aa_vtx, or trkall.
Data and Monte Carlo Samples
Monte Carlo files in SAM 5,000 Z to ee B MC includes
8,000 Bs to Ds eX
8,000 B to J/psi(muons) Ks
5,600 Bs to Ds pi 2*10,000 Top Group lepton + jets with
average of 0.5 and 2.5 additional minbias 10,000 bbH to bbbb Higgs events 10,000 hadronic tau events A light quark sample?
Not all same simulation used in generation
SAM Definitions Monte Carlo
5,000 Z to ee %z-ee%tk-p11.11-%.root B MC %bbbarQQ%tk-p11.11-%.root 2*10,000 Top Group lepton + jets with
average of 0.5 and 2.5 additional minbias %ttbar-wjj+wlnu%tk-p11.11-%.root
10,000 bbH to bbbb Higgs events %bbh-bbbb%tk-p11.11-%.root
10,000 hadronic tau MC %tau_tauhcw%tk-p11.11-%.root
A light quark sample?
The third %’s above are gtr, htf, gtrela, htfela, gtrhtf, aa, aa_vtx, or trkall.
Links to Workshop Talks
Track Finders
smt cft Ptcut GeV/c
gtr cft-smt XOR
smt-cft
2d 1d(2d) 0.4
htf cft-smt OR
smt-cft
2d 2d(1d) 0.5
ela global 2d 2d -
aa smt-cft 1d 1d 0.18
•Experts may be willing to describe their algorithms in more detail.•We asked the algorithm developers to set their own parameters.
Stuff on this and next few pages from S. Khanov’s talk at the mini-workshop.
Reconstruction Procedure
All samples were reconstructed with p11.11
Individual and Combinationsgtr: (no H-disks in data, yes MC)
htf: with grt refit (all but aa did that)
gtrela: gtr (no overlap) + elastic on leftover hits
htfela: htf + elastic on leftovers
gtrhtf: OR of gtr (no overlap) + htf
aa: aa (some samples had no vertex info, look for aa_vtx, also gtr-refit now available also aa tracks have wrong chi^2 and d0hitmask)
Trkall (all 6) for cross checks 15% failed to finish the two steps Main problems are not thought to be
the fault of the tracking algorithms
Analysis Tools
“gtr_analyze” Filled roottuples with reco tracks and
their parameters Fills info needed for comparison
with MC Some MC samples didn’t retain the
MC hits so a hit-by-hit comparison could not be made.
“gtr_examine” (M.S.) Root macros which calculate track
efficiency and fake rates … Tons of plots Primarily for MC samples
Definitions
Track Quality is described by a 2. Good Tracks have matching 2 < 25 Misreco Tracks 25 < 2 < 500 Fake Tracks 2 > 500
Tracking Algorithm CPU vs memory
Algorithms take similar amounts of time in tracking except AA, which is ~3 times faster. D0Reco time spec. ~ met. Improves possible.
No hard info on occupancy dependence. Top MC took longer than spec. for some combos.
aa
Results from Tracking Algo. Gp. (S. Khanov)
A lot of material was presented. It was clear from the first that no
algorithm or combination solved our problems.
Studies were shown of data and MC efficiencies and fake rates. Number and distributions of tracks
in eta and phi for all algorithms and combinations.
Z to dimuons, J/psi to dimuons, psi’, phi to KK bumps shown and fitted with background estimates.
Detailed comparison of Z to ee including a diagram comparing which Z’s were identified between three algorithms.
Results from Tracking Algo. Gp. (S. Khanov)
3) Z to eeoverlaps
1) Bumps
2) Split J/psi mass in Eta regions
BID Group Results (Bram W.)
Studied B-tagging in Jets in Run 157708 and the mu-jet data Number of tracks per jet and number
of good tracks (positive DCA, pT>1.5 GeV/c) in jets with an 0.5 cone
“Efficiency” is fraction of b-jets (defined using pTrel mu-jet) tagged
“Mistag Rate” is fraction of jets in Run 157708 that are tagged.
Fleura studied top Monte Carlo Found the tagging probabilities and
mistag rates for 1- and 2- tagged jets for each algorithm
Presented a clear table.
BID Group Results (Bram W.)
Example plot (one of several)
Eff’y
Mistag Rate.
Tau ID group (Silke and Yuri)
Studied the hadronic tau MC (signal) and the 4b MC (backgd) Noted tau ID is highly sensitive to
the efficiency Counted the number of 1 & 3-prongs Looked at (pT) of additional tracks
in cones around the tau Mass distributions of matched tracks
(seemed independent of algorithm) Studied tracking efficiency in 1-
prong vs 3-prong vs pT Mapped lost track eta-phi
distribution
Tau ID group (Silke and Yuri)
Left Plot shows the # of tracks in 1 & 3 prong tau events Right plot shows the eff’y vs pT of the prongs for 3 prong taus.
B Hadron Reconstruction (Vivek)
Analyzed the dimuon data sample to search for J/psi, Ks, and . Interested in low pT to reconstruct
the pion in the lambda decay. Showed the mass resolution, #signal
and #background for the 6 cases.
Analyzed the three B MC samples: Bs to Ds*-+, B0 to D*-
+, and Bs to Ds*-e+X. Showed direct comparisons of the
eff’y, widths, misreco rates in the samples.
EM ID (Robert Z.)
Studied Z to ee data and MC and applied the track-matching used in obtaining the W and Z cross sections recently shown at ICHEP.
2
/
22
2 1/
pEz
pEz
z=7.6 mm =4.6mrad
E/p = 0.18 P(2)
Plots for p11.09 gtrEff’y depends on 2 cut. Select p(2)>1%.
EM ID
Studied Z to ee data and MC and applied the track-matching used in obtaining the W and Z cross sections recently shown at ICHEP in the region |eta|< 0.8
Studied eta dependence of the efficiencies in the Monte Carlo
)(2
2
210
21
NNN
NN
Efficiencies are per track
EM ID%
%
%%
TARC Z events run with p11.11
various algorithms ||<0.8
New Phenomena and Muon ID (Ryan H.)
Concentrated on the large dimuon data sample
Compared the 6 cases against J/Psi, upsilon, and Z to dimuons Number identified Mass and width
New Phenomena and Muon ID (Ryan H.)
Higgs Group Report (Lorenzo F.)
Studied the bh to bbbb MC sample with all tracking algorithms and combinations Tracking eff’y vs pT and eta Misreconstruction and Fake rates vs
pT and eta DCA resolution for various pT min. Eff’y and fake rate for Track
reconstruction in jets B-tagging eff’y and mis-tagging rate
A lot of information!
Higgs Group Report (Lorenzo F.)
Higgs Group Report (Lorenzo F.)
Higgs Group Report (Lorenzo F.)
Higgs Group Report (Lorenzo F.)
Secondary Vertex B-Tagging (Ariel)
Studied b-tagging in ttbar MC events for all 6 cases. B-quark vs light quark tagging eff’y
vs jet pT, eta, jet-track multiplicity, and jet multiplicity
Secondary Vertex B-Tagging (Ariel)
Same plot, knee of curve
Tracking in Top Samples (E. Chabalina)
Studied Tracking Eff’y, mis-reco rate, fake rate, and purity in top MC events using gtr_examine using all cases 0.5 and 2.5 additional min-bias
events overlaid pT dependence, eta dependence, pT
jet dependence …
Studied the Z to ee data Numerically rated the 6 cases in
tables of criteria!
gtr htf gtrela htfela gtrhtf aa
eff 0.769 0.856 0.834 0.898 0.878 0.776
purity 0.959 0.974 0.948 0.960 0.967 0.984
goodeff 0.617 0.742 0.658 0.698 0.777 0.602
misreco 0.069 0.073 0.094 0.088 0.072 0.118
fake 0.041 0.026 0.052 0.040 0.033 0.016
gtr htf gtrela htfela gtrhtf aa
eff 0.753 0.854 0.819 0.898 0.873 0.776
purity 0.954 0.961 0.937 0.948 0.947 0.988
goodeff 0.584 0.690 0.615 0.675 0.691 0.595
misreco 0.064 0.069 0.089 0.085 0.069 0.137
fake 0.046 0.039 0.061 0.052 0.053 0.012
0.5 mb (~500 events)
2.5 mb (> 1000 events)
Tracking in Top Samples (E. Chabalina)
gtr htf gtrela htfela gtrhtf aa
Reconstruction efficiency of isolated leptons in data
Z(ee) 4 3 1 1 2 4
Z(μμ) 4 5 2 1 3 5
sum 8 8 3 2 5 9
Reconstruction/b-tagging efficiency in MC
All tracks 5 2 4 3 1 6
4 2 5 4 3 1
tracks in jets
4 2 3 1 1 5
4 2 5 4 3 1
IP b-tag 3 1 4 5 2 3
SV b-tag 3 1 4 2 1 5
Sum MC 23 10 25 19 11 21
Total 31 18 28 21 16 30
Tracking in Top Samples (E. Chabalina)
Summary
I described the data samples, procedure and summarized the mini-workshop
Common Elements in the presentations No magic algorithm For high pT physics the 3
combinations outperform any single one and aren’t strikingly different
For low-pT physics there was consensus that combinations including htf had best eff’y vs mistag fraction
This is a start but isn’t as good as we’d like to be!