Top Mass Template measurement in the Lepton Jets Channel PREBLESSING 1'7 fb1

1
Top Mass Template measurement in the LeptonJets
ChannelPREBLESSING (1.7 fb-1)
2
Cast of characters

• Jahred Adelman (UC)
• Erik Brubaker (UC)
• Sebastian Carron (Toronto)
• Wojciech Fedorko (UC)
• Young-Kee Kim (UC)
• Hyun Su Lee (UC)
• Young-Jang Lee (Seoul)
• Mel Shochet (UC)
• Pekka Sinervo (Toronto)
• George Velev (FNAL)

Plus a (not completely all-volunteer) army of CAF
warriors
3
History and documentation

• Published result Mtop 173.5 /- 3.9 GeV 355
  pb-1
• PRL 96 022004 (2006)
• PRD 73 032003 (2006)
• Most recent preliminary result
• Mtop 173.4 /- 2.8 GeV 680 pb-1
• Recent CDF notes
• CDF 8909 Lepton-Photon Mtop in ljets TMT2D
• CDF 8824 Gen6 Pythia top-specific corrections
• CDF 8761 Using KDE in Gen6 TMT
• CDF 8190 KDE and its application to TMT
• Soon well have a dilepton measurement too!
• See Wojteks preblessing tomorrow
• Web page
• QA on out TWiki
• http//www-cdf.fnal.gov/htbin/twiki/bin/view/TopMa
  ssTemplate/AnalysisLJ07

4
Reconstructing the top quark mass

• Use standard LeptonJets ?2 fitter to reconstruct
  a single top quark mass (one number) per event
• Correlated with true top quark mass, but not the
  same thing
• Use overconstrained kinematics of ttbar system
• New Pythia Gen6 TS corrections!

5
In-situ calibration using dijet mass

• Kinematic fitter works well, but distributions
  are highly correlated not only to top quark mass,
  but also to JES
• Introduce the dijet mass from the hadronically
  decaying W
• Use as in situ calibration of JES
• Strong JES correlation, also some weak top mass
  correlation
• We use the dimass closest to the well-known W
  mass from any pair of untagged jets among leading
  4 jets (studied other choices, this was best)
• New Throw away any events failing ?2 cut, just
  like for mtreco

6
Our likelihood
OLD

• Likelihood simplified w.r.t. published analysis
• Removed unwanted cross-terms
• Event counting simpler, more correctx
• Now mjj is a single number per event, so no
  correlations hidden there.
• We want to account for correlations between
  mtreco and mjj, too!

NEW
7
Shape probability?

• How do we get the probability to observe an event
  with mtreco and mjj?
• Previously, assume the two observables are
  uncorrelated, and parameterize as a function of
  mtop and dJES
• Near-impossible to account for correlation
  between observables
• Parameterizations difficult, mathematically bad

New Use a Kernel Density Estimate-based approach
to form PDFs that are two-dimensional in
observables
8
A 1d KDE pictorial tutorial
Probability
Mtopreco
9
Types of KDE
Our nominal (2d) measurement
A cross-check (1dx1d) measurement
1Dx1D
Mtop RMS
2D
10
2d PDFs
11
Adaptive KDE

• Want smoothing to depend on statistics within
  sample.
• First pass Pilot density estimatesfixed kernel
  width.
• Second pass Adaptive density estimates.
• Narrower kernel in high-stats region ? sharper
  peak.
• Wider kernel in low-stats region ? smoother tail.

12
How do we use this?

• Have a grid of 2002 points in mass-JES space
  (mass 140-200, JES -3.0 to 3.0)
• Minimize ns and nb for each subsample at every
  point in the grid
• Fit a 2d parabola (including correlation
  cross-term) to the points

Example PE
13
Sample division

• Old 1-tag(L) and 0-tag categories do not improve
  the expected total uncertainty at current
  luminosity
• Increase systematics more than they reduce
  statistical error
• After study, we require exactly four jets in
  1-tag category, but allow 3.5, 4, or more jets in
  2-tag

14
Boundary cuts

• KDE doesnt know about hard cutoffs in the
  observables
• Probability leaks into unpopulated regions
• Easiest fix is to explicitly set boundaries and
  force kernels to stay inside.
• Amounts to extra selection cut on mtreco, mjj
• Efficiency high for signal events passing chi2 cut

15
Backgrounds (technical details)

• Use MII4U numbers wonderfully provided by Tom
• Use HFOR removal (jet-based)
• Weight all the background samples by the
  individual pre-chi2 boundary efficiencies
  (50-80)
• Use Alpgen weights to stitch together samples
• Assume specific flavor backgrounds (ie Wbb) come
  from matrix element (ie btop?w samples)
• Assume mistags come from Wlf (ie ptop?w samples)
• Assume two-tag mistags are both mistags (not real
  tags)
• Use event-by-event mistag probabilities
• Assign QCD shape to mistags
• Studied in previous incarnations, more
  cross-checks to come
• Wbb/cc0-2P, Wc1-3P, W2-4P, s/t-stop, WW,
  WZ, ZZ
• Events in PEs drawn according to their correct
  weight (including individual event mistag
  probabilities)

16
1dx1d Background PDFs
17
2d Background PDFs
18
Bias checks
JES gt 0 JES lt 0

• No significant bias bias!
• Slight JES bias (no matter)?
• Pull width lt 1.0?!?!

19
On the pull width

• Removing strange point at 168, pull width goes to
  98.1 and chi2/ndof becomes 5.3/6.0
• Examined points at 168/log files/density
  estimates/PEs
• Cannot find anything wrong with the point!
• In any case, not reducing our reported error

Recall
20
Residual JES systematics
Systematics studied using large 175 GeV
sample Sum in quad of L1-L8 0.55 GeV/c2 Bkgd
JES uncertainty added linearly to uncertainty due
to first-order JES ( 0.38 GeV/c2 when converted
to a number in quadrature)
21
Other systematic samples
PEs 1.2 fb-1 run range vs PEs using 1.2 fb-1
ERR (in correct proportions). Take a systematic?
Or not?
22
Summary of systematics

• Take MC uncertainty on sample for FSR (instead
  of 50 MeV)
• PEs for b-jets and underway, should be done in
  next 1 day
• No prescription yet for bkgd shape, taken from
  Gen5 Q2 variation on 1-tag and 2-tag samples
  only
• Smaller now that we dont have 0-tag events!

23
Systematic vs Winter06 measurement
24
Nevt
25
The fit!
The fit!
26
Full fit
27
Cross-check 1dx1d fit
28
Are we lucky?
19.8
29
More cross-checks
n evt taken from minimization at single point
with best -ln(L), so not a full fit for number of
events (but a good approximation)
30
Cross-checks
31
Division by data period
No background constraint used (but JES constraint
used throughout)
32
2tag-1tag study
3.2 of PEs with observed num events have larger
2tag-1tag differences (8.0 for 2d sided
differences)
33
Fitted nbkgd
Paying no attention to strange spikes/dips
(stupid ROOT plotting issue), we prefer 0 2-tag
background in a wide area of the mass-JES plane
when bkgd constraint is removed
34
Old-style 1dx1d plots
35
Conclusions

• Made a first measurement of the top quark mass in
  the LeptonJets Channel using gt 1 fb-1 of data
• So far, single best top quark mass measurement
  from CDF
• Machinery also being used to make a dilepton
  measurement (see Wojteks talk tomorrow!)
• Measurements to be combined in future/for
  publication
• Compare with previous TMT result using 680 pb-1
  Mtop 173.4 ? 2.3 GeV/c2

Mtop 171.6 ? 2.3 GeV/c2
36
Wait for it
Backups
37
Study of 2tag-only nobkgd corr PEs
38
Background composition
39
Anti-ele vs Wlf
KS prob (top group normalized PE version 77.5)
40
More adaptive KDE (signal)
41
More adaptive KDE (signal)