Top Mass Template Status Report

1
Top Mass Template Status Report

• Wojciech Fedorko
• for Top Mass Template Group

2
Outline

• Motivation for 700pb-1 analysis
• Current state of the analysis
• Code validation
• Data Validation
• Parameters of the analysis
• Event selection
• JetProb tagging
• Analysis extensions for 1fb-1
• Schedule

3
Template Analysis Overview
?2 mass fitterFinds best top mass and
jet-parton assignmentOne number per
eventAdditional selection cut on resulting ? 2
Massfitter
Templates
Likelihoodfit
Result
Likelihood fitBest signal bkgd templates to
fit dataCompare to paramizn, not
directlyConstraint on background normalization
4
Motivation for 700pb-1 analysis

• We have potential to make best top mass
  measurement
• Excellent way to prepare for 1fb-1 round
• Transfer of expertise

• Strategy is to keep everything as close to
  318pb-1 as possible. Some changes are necessary.
• Gen6 / Gen5
• New data

5
Data Validation work in progress

• At this moment highest priority (but also shared
  responsibility) to be completed ASAP
• Target date is Dec 6
• Un-Ki (and many others) is working on
  understanding event yield jumps in 5.3.3 and
  6.1.2 datasets.
• Guram, Wojtek working on understanding basic jet
  quantities between two datasets
• We also need to understand 5.3.3 MC VS 6.1.2 data

6
Data Validation

• Plots below are from inclusive Leptonjets
  sample. (No met cut)
• Data samples shown are exclusive
• 5.3.3 Feb 2002 - Aug 2004
• 6.1.2 Dec 2004 Sep 2005
• Muon sample includes CMX

7
Data Validation
Loose jets 8GeVltETlt15GeV
Excess in events with high jet counts probably
due to increase in inst. lumi.
8
Data Validation
Tight jets ETgt15GeV
No excess in high jet count events
9
Data Validation
Excess in high jet count events for electrons in
6.1.2 dataset
6.1.2
6.1.2
10
Data Validation
Excess seen also in 5.3.3 data set- this is not a
new issue
5.3.3
5.3.3
Probable explanation sample composition (bg
fraction, type)
11
Data Validation
Lxy behavior unchanged between the datasets
despite changes in tracking
12
Data Validation
Turn-off different possible explanation is
movement of beam at the end of 5.3.3 data period,
causing different effect of radius cut on
Lxy distribution
13
Code Validation -why do we need to?

• New people running 2D analysis (mainly Jahred,
  Tomo and me)
• Some work done on the templating script
  (bugfixes)
• We overhauled parametrization and
  pseudo-experiments machinery
• Should not change function
• Please take a look at hep.uchicago.edu/jahred/tmt
  2d for excellent documentation

14
Code Validation -discrepancies withJean-Francois
(Gen 5 publication)

• Mean and RMS values shifted (up to 1GeV) with
  respect to JFs templates.
• Traced to accidental use of old Jet correction
  code

15
Code Validation

• Issue is most likely resolved
• We will have template and parametrization code
  fully validated by end of this week.
• Pseudo-experinent code by early next week

16
Code Validation pseudo-experiment machinery

• Preliminary check
• We run 3000 pseudo-experiments over a range of
  (mass,JES) points
• Our tools work for running on CAF or on UChicago
  cluster
• Signal/Background expectations used were s6.7pb
  LBL estimates (used by JF) for 318pb-1 dataset

17
Code Validation JES pull width
JES
18
Data Validation Mass pull width
JES
19
Code Validation bias on Mass?
JES
20
Code Validation further steps

• Eliminate or fully understand template
  differences between our templates and 5.3.3
• If no further issues appear -by the end of this
  week
• Validate pseudo-experiment code against JFs
  results.
• Investigate possible biases

21
Analysis Parameters

• Where no strong reason for change, stay close to
  gen5 analysis.
• Event selection L4 vs L5
• Change would entail new templates, top-specific
  corrections, backgrounds. ( intangibles)
• Decided to stay with L4.
• Jet probability (5 JPB cut) for second tag
• Gen5 had 1DJPB analysis, not 2D.
• Improvement using 2DJPB under study now.
• Final decision lt 2 wks. (Winter? 1 fb-1?)

22
Background Estimates

• Recall we constrain nb using estimates from stt
  analysis.
• Quite insensitive to value and error of
  constraint.
• What constraint to use for 680 pb-1?

23
Changes in new data

• Example New data has increased CMX acceptance
  from miniskirt. Gen5 MC doesnt have it.
• If NCMX/NCEMCMUP changes with miniskirt, and
  template shapes differ CMX vs CEM CMUP, we
  introduce bias.
• No apparent biasplots.
• Other differences in gen6 data?
• B-tagging/tracking
• Inst. Lum.

24
On to 1 fb-1

• Weve got lots of ideas, though many of them
  probably wont work. Will spend the time to study
  them all closely to find which will help the
  analysis
• Work on 1 fb-1 in conjunction with the winter
  analysis
• Many of the ideas have been around for awhile
• Its probable that no single one will shrink
  error by 25
• One simple thing we will do is revisit event
  selection/division
• Divide by Njet
• Divide by low and high ?2
• Do we still want to keep 1-tagL (3.5-jet bins)
  and 0-tag events?

Jets matched to partons -gt better stat error,
reduced systs
fraction
Also, separating like this makes combination with
ME methods much easier
fraction
25
More Ideas

• B-tagging
• Want to take advantage of best tagging possible.
  JetProb? SLT? Loose tagger? Tight tagger? NN
  tagger? Anti-charm taggers?
• Top-specific corrections
• Machinery in place for new corrections if we
  decide to use them
• Tried improving corrs with EMFraction no added
  benefit. Will look at tracking information for
  possible improvement
• JES measurement with subdivision by ?
• Mjj in different ? regions have different
  shapes, different SB
• Had some discussion of this, but its not clear
  how this would be used a single JES? Integrate
  out multiple JES?
• Helps with method systematic

Mjj 2tag Central-Central
Mjj 2tag Central-Central
JES -2s
JES 2s
Mjj 2tag Plug-Plug
Mjj 2tag Plug-Plug
JES -2s
JES 2s
26
Still more ideas
Naïve estimate from Erik Assuming SM stt(Mt),
counting expt measures Mt to /- 6.5 GeV."

• Simultaneously measure of mtop and ?tt
• Increased acceptance (plug electrons, other
  ideas?)
• Add in dileptons
• An obvious way to utilize JES knowledge/measureme
  nt into dilepton method
• Simply combine likelihoods for both channels
  when we minimize for the top mass
• Already have 3 template dilepton methods blessed
  by the top group
• Using more than 1 combination in mtop reco
  template? Using ?2 information to weight events?
  Using error returned by Minuit fit?
• Not trivial but worth revisiting

27
Winter Analysis Schedule
Aggressive schedule, builds in 2 wks cushion