1. Kevin Black Meenakshi Narain Boston University
Measurement of the Top Quark Mass The Low Bias Template Method using Lepton + jets events
4 high PT Jets
2 from the b-quarks
2 from hadronically decaying W
1 Isolated high PT lepton
Muon or electron
Tau not used in this analysis
Neutrino
Kevin Black
Meenakshi Narain
Boston University

2. Backgrounds W+Jets Multijet Events Real high PT lepton Neutrino
Multiple Jets
 ~12 pb ± 25%
Multijet Events
One Jet fakes the signature of an isolated lepton
Mismeasured transverse momentum

3. Event Selection At least 4 jets with PT > 20 GeV , || <2.5
From Cross-Section Measurement
At least 4 jets with
PT > 20 GeV , || <2.5
Isolated Muon or Electron with PT > 20 GeV
Muon || <2.0
Electron || <1.1
Missing ET> 20 GeV
e+jets

4. Topological Variables
Form log likelihood of signal over background for each variable and combine
Low Bias (“LB”) discriminant is built from four topological variables to give extra discrimination between top and background E T A 3 2
Smallest eigenvalue of P K m i n ( R j ) _ W P a p ~ L t b = 4 f D B 1 + : 5 e H k

5. Topological Variables

6. Topological Likelihood
Using each topological variable separately there is some separation between signal and background
however combining
them in a likelihood
gives optimal separation
Discriminant
LB disc > 0.4 (optimal cut)
Remove ~50% background
Retain ~ 85% Top

7. Identifying b-quark Jets
Each top event has two b-quarks
travel a few mm and decay
Search for jets which point back to secondary vertex
Largest backgrounds for top do not have heavy flavor
Same event selection, except
Jet Pt > 15 GeV
Require one or more tagged jets

8. Kinematic Fitting Kinematic Fit:
1 unknown: longitudinal momentum of the neutrino
3 constraints
Top quark and antiTop quark have the same mass
2 jets have the invariant mass of the W
lepton+neutrino have invariant mass of the W
2C fit for top mass
Experimentally see Jets not partons ! try out every permutation

9. Kinematic Fitting Complications
Final stage gluon radiation
- extra jet, splits the energy
from one parton->2 jets
Neutrino:
- 2 possible solutions
- different starting points for
fit g g
Initial state gluon radiation
- extra jet, not from the
top-antitop decay

10. Performance of the Fit In parton matched correct solution,
RMS ~ 18 GeV,
Lowest 2 all events RMS ~ 30 GeV
In parton matched events,
the lowest 2 solution is correct
~37% of the time

11. Effect of b-tagging
Requiring one or more b-tagged jets increases the expected S/B dramatically
without tag ~ 1/2.6
with b-tagged ~ 3/1 (also lower jet pT requirement)
Reduction of incorrect permutations:
12: untagged, 6 single tag, 2 double tag
Higher Probability of getting correct solution !
better resolution

12. Method to measure the top quark mass
Make Templates for signal events
9 different mass points between GeV
Background Templates from W+jets MC
Use simple Poisson likelihood for the number of signal and background compared to data
constrain the number of background events to the expected.
Signal 175 GeV
Background

13. Multijet Distributions

14. Event Yield - Topological Analysis
e+jets
mu+jets
Preselection 87 80
+fit2 <10 78 68
+DLB >0.4 49 45 b  p
E T
jet
e+jets
mu+jets
top
27.5 ± 2.2
20.4 ± 0.9
W+Jets
9.5 ± 2.2
22.0 ± 2.6
Multijet
12.0 ± 0.55
2.6 ± 0.5

15. Event Yield – b-tagged Analysis
e+jets single tag
e+jets double tag
mu+jets single tag
mu+jets double tag
Preselection 38 9 24 5
Fit Convergence 36 6 22
e+jets
mu+jets
top
30.5 ± 2.4
22.0 ± 1.75
W+Jets
7.0 ± 0.6
4.3 ± 0.3
Multijet
4.5 ± 0.4
0.7 ± 0.05

16. Ensemble Tests Calibration Consistent Pull: With Slope 1  ~ 1
And 0 offset
Pull:
 ~ 1
Mean ~0
Ensemble Tests
Pull
Mass Calibration
Topological:
Expected Error
~6.1 GeV
Tagged:
Expected Error
~4.2 GeV
Statistical Error (GeV)
Statistical Error (GeV)

17. Topological Analysis Results:
Red: background only
Blue: background + top
Fit 44.2§ 6.6 ttbar events, expect 47.9  8.8 events

18. Tagged Analysis Results
Red: background only
Blue: background + top
Fit 49.2§ 6.3 ttbar events, expect 52.4  4.2 events

19. Comparison with Expectation
Tagged Selection
Topological Selection

20. Topological Discriminant in Data
Tagged Selection
Topological Selection

21. Systematic Errors: Jet Energy Scale
Photon + Jet Events
- electromagnetic scale more
precisely known
- Compare the transverse energy
of the jet and photon and
compute  S
- Do the same with Monte Carlo
events and compare the
difference between data
and MC
ETJet- ET
S =
ET

22. Variation of Distributions with the Jet Energy Scale
JES up and down for top
JES up and down for Wjets
Perform Ensemble Tests with +1, -1
compared to nominal

23. Systematic Errors: Gluon Radiation
To obtain bounds:
- consider the variation
seen in events with
and without gluon
radiation g g
Conservative approach
but not much is known
about gluon radiation in
top decays!

24. Gluon radiation: Kinematics of Jets
Matched
Matched –Tagged
Gluon Jet
Gluon Jet –Tagged

25. Systematic Uncertainty
Source
Topological (GeV/c2)
b-tagged (GeV/c2)
Statistical
±5.8
±4.2
Jet Energy Scale
Jet Resolution
±0.9
Gluon Radiation
±2.6
±2.4
Signal Model
+2.3
Background Model
+0.7
±0.8
b-tagging
±0.7
Calibration (fitting bias)
±0.5
Trigger
MC Statistics
Total Systematic
±6.0

26. Cross-Checks Vary the Discriminant Cut Drop Constraint on background
170.5 ± 5.5 GeV
D > 0.3
170.6 ± 5.5 GeV
D > 0.4
169.9 ± 5.8 GeV
D > 0.45
170.3 ± 6.5 GeV
D > 0.5
169.2 ± 5.6 GeV
D > 0.55
167.0 ± 6.5 GeV
Vary the Discriminant Cut
Drop Constraint
on background
Constrained (Poisson)
Unconstrained
Topological
169.9 ± 5.8 GeV
170.7 ± 6.5 GeV
Tagged
170.6 ± 4.2 GeV
171.8 ± 4.8 GeV

27. Summary and Outlook
First measurement of the top quark mass at D using b-tagging
best for Run II so far
Result
Topological analysis
b-tagged analysis

28. Backup Slides

29. Event Selection: e+jets
4 jets PT > 20 GeV, || <2.5
1 electron PT > 20 GeV, || <1.1
EM Fraction > 0.9
Isolation < 0.15
Hmatrix < 50 (shower shape)
EM Likelihood > 0.85 (multivariate)
Track Match
ET > 20 GeV, lepton , ET  triangle cuts
Second high PT isolated lepton veto
Primary Vertex in SMT fiducial range, with at least 3 tracks
Primary Vertex within 1 cm of z position of electron track

30. Muon +Jets 4 jets PT > 20 GeV, || <2.5
1 muon PT > 20 GeV, || <2.0
Isolated from jets
ET > 20 GeV, lepton , ET  triangle cuts
Second high PT isolated lepton veto
Primary Vertex in SMT fiducial range, with at least 3 tracks
Primary Vertex within 1 cm in z of muon track

31. Parton Matched Jets from W
Jet Corrections
Parton Matched Jets from Top
Parton Matched Jets from W
Parton Matched Jets Z ! b

32. Likelihood Function L(mt, ns, nb) = P(N; N(nb)) P(Nj, Nj(ns, nb))
Nj= nspsj(mt)+ nbpjb
The product is over the number of bins
The ps and pb are the fraction of events from that template that fall into bin j
For each template the likelihood function finds the signal fraction which minimizes the negative log likelihood and returns the value of the likelihood, and the fit number of top events.
Repeated through each top mass template and then a parabola is fit through the resulting likelihoods

33. Discriminant Cut

34. Can we combine the two analyses
Simplest way – remove the tagged events from the topological analysis
Find 95 events
Anti-tag efficiency
Top ~40%
W+jets and QCD ~90%
Expect 80% are background after removal of the tagged events.
Redo ensemble tests, expected statistical error ~11.0 GeV
Combined expected error would be ~3.9 (rather then 4.2)
However, systematic dominated so only a few percent overall improvement