Preblessing : Top quark mass measurement using TMT framework 3.0 fb1

1
Preblessing Top quark mass measurement using
TMT framework (3.0 fb-1)
Hyunsu Lee, Jian Tang, Young-Kee Kim University
of Chicago
2
Supporting Notes

• 9661 Main note
• 9549 2.7 fb-1 LJ and DIL combo
• 9153 1.9 fb-1 LJ and DIL combo
• 8919 Studies of DIL fakes
• 8653 Studies on combining DIL and LJets
  measurements
• 8989 1.7 fb-1 LJ measurement
• 8917 1.9 fb-1 DIL measurement
• 9081 Studies of the bootstrap method
• 9110 LPS bandwidth studies
• 8824 Gen6 Top Specific Corrections
• 8761 Studies of KDE and LJ subsample division
• 8190 Application of Kernel Density Estimation to
  TMT analyses

2
3
Brief method outline

• 2d templates
• mtreco, wjj in LeptonJets
• mtreco, mT2 in Dilepton
• signal pdf's depend on Mtop, ?JES
• background pdf's depend on ?JES

• MC
• tt
• backgrounds

DATA
Likelihood Fit
3
4
Updating Since last blessing

• We are currently using jetCorr17
• Include p18 data set (300 pb-1 good silicon)
• Good Run List v24
• Slightly different CMIO category for Dilepton
  selection
• Introducing mT2 as a second observable in the
  Dilepton channels
• Some of run in the p18 use loose CMUP trigger
• Background estimation was done using p17 data and
  scaled up
• We added one more measurement which we used only
  mT2 for Dilepton channel
• It can be first test of mT2 in the real data

4
5
CMIO category

• We had one less events compared with X-section
  groups
• We used same event selection with X-section group
• We traced and found that muon fiducial cut was
  slightly different
• Two groups requested same muon fiducial cut which
  is not matching with any of muon detector but, we
  additionally requested muontype0

NWA mass
mT2
DIL 0.18 GeV Combo 0.07 GeV DIL (mT2 only)
0.14 GeV
PE difference between old and new CMIO categories
5
6
What is mT2 (in the two missing particle system)

• We can define similar transverse mass in the two
  missing particle case which is mT2
• We dont know transverse momentum of each missing
  particle. We only know sum of that (missing
  transverse momentum)
• Miminization can determine transverse momentum

6
7
Why mT2 is quite interesting especially in the
top dilepton channel

• Most of exotic particle can be produced by pair
  production in the LHC
• They will have two missing particle(ex, SUSY-two
  neutralino)
• If we find new particles in the LHC, we dont
  know which theory is corresponding with this new
  particle need to know its characteristics
• For the mass measurement in the two missing
  particle system, mT2 is one of best observable
• There are bunch of communications to use mT2 in
  the LHC by phenomenologist but they didnt test
  with real data
• Gluino Stransverse Mass Phys.Rev.Lett.100171801
  ,2008
• Measuring superparticle masses at hadron
  collider using the transverse mass kink
  JHEP0802035,2008
• Inclusive transverse mass analysis for squark
  and gluino mass determination JHEP0806035,2008
• Using subsystem mT2 for complete mass
  determination in Decay chains with missing energy
  at hardron colliders arXiv0810.5576
• .
• Top dilepton channel can be a standard candle of
  using mT2
• Measuring the top quark mass with m_T2 at the
  LHC Phys.Rev.D78034019,2008
• Measuring mass with real data could be very
  interest
• We added one more measurement for Dilepton
  channel which only use mT2 observable

7
8
Top mass measurement in the dilepton channel
using mT2 in the LHC
PRD 78.034019
Template method 170.3-0.3 GeV
8
9
Template using mT2

• mT2 have mass dependence but resolution is poorer
  than NWA mass (better than Ht)
• JES dependence is relatively smaller than Ht and
  NWA mass

9
10
Mass Error estimation (1D input) for different
observable
10
11
mT2 improve real mass resolution in the dilepton
channel

• We had been used 2D observable which are
  reconstructed mass with neutrino weighting and
  Ht.
• The mass resolution of Ht is quite poor.
• We can replace Ht with some others.
• Introducing mT2 as second observable give 10
  better mass resolution.

11
12
LJ selection/signal background estimation

• We used standard Leptonjet/MII4U selection
• Require at least one tight secvtx b-tag
• Require exactly 4 tight jets for 1-tag events
• Allow 3.5 jet events and events with more than 4
  tight jets for 2-tag events
• We requested chi2
• We had boundary cut for reconstructed top mass
  and dijet mass

Background estimation was done using MII with p17
data and scaled up
12
13
2D template for LJ channel
13
14
DIL selection/ signal background estimation

• Standard DIL selection
• Separate sample into 0tag and tagged
• Boundary cut applied
• 100GeV
• 20GeV

• We used number from DIL X-section measurement
  with p17 data and scaled up

14
15
2D template for DIL channel
15
16
Bias check (mass bias)
16
17
Bias check (mass pull width)
17
18
Systematics
18
19
Residual JES systematics
19
20
Generator/IFSR
20
21
PDF
LJ
DIL
0.17 (GeV/c2) Diff. Group 0.10 QCD scale
0.06 20 eigenvector 0.13
0.48 (GeV/c2) Diff. Group 0.17 QCD scale
0.36 20 eigenvector 0.27
Combo
DIL mT2
0.19 (GeV/c2) Diff. Group 0.12 QCD scale
0.10 20 eigenvector 0.11
0.47 (GeV/c2) Diff. Group 0.21 QCD scale
0.36 20 eigenvector 0.22
21
22
LJ background systematics
Qfactor varying very tiny effect
Take largest shift from nominal as systematic
0.25 for LJ and 0.18 for combined
22
23
DIL background shape

• Shift /- sigma of each background type (fake,
  DY, diboson)
• DY reweighting with Met distro from data
• Fake reweighting to have maximum change of shape
  within the error of fake rate

23
24
B-JES
24
25
Pile-up

• nzV data2.075, MC1.504

Known 0.09 GeV Unkonwn0.24 GeV
Known 0.08 GeV Unkonwn0.10 GeV
Correction 0.16 GeV Known 0.17
GeV Unkonwn0.12 GeV
Correction 0.33 GeV Known 0.14
GeV Unkonwn0.33 GeV
25
26
gg fraction/lepton energy

• gg fraction
• Lepton energy scale

26
27
Data
DIL
LJ
27
28
Validation (LJ)
28
29
Validation (DIL)
29
30
The fit!
30
The fit!
30
31
Combined fit
171.8 1.5 (statJES) 1.1 GeV/c2(syst)
1.8
171.8 GeV/c2
-1.9
31
32
LeptonJet channel
172.5 1.6 (statJES) 1.1 GeV/c2(syst)
172.5 1.9 GeV/c2
32
33
Dilepton channel
2.6
169.0 (stat) 3.2 GeV/c2(syst)
-2.7
4.1
169.0 GeV/c2
-4.2
33
34
Dilepton (only using mT2)
4.8
167.9 (stat) 2.8 GeV/c2(syst)
-4.1
5.6
167.9 GeV/c2
-5.0
34
35
Cross check
35
36
Cross check
37
Data distribution LeptonJet channel
37
38
Data distribution Dilepton channel
38
39
Conclusion

• We had top mass measurement with up to p18 data
  using template method
• Hope to bless in 2weeks.
• Would like to publish top mass measurement in the
  dilepton channel only using mT2 observable

1.8
171.8 GeV/c2
-1.9
5.6
167.9 GeV/c2
-5.0
39
40

• Backup

41
LeptonJets Reconstructed Mass
41

• 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!

41
42
LeptonJet Dijet mass
42

• Use single dijet mass per event to help constrain
  the JES
• Easy choice for 2-tag events (one dijet among
  leading 4 jets)
• For 1-tag events, choose single dijet mass
  closest to W mass
• New since last summer correct jets fully using
  TS corrections, not just to L5

42
43
Dilepton NWA mass
43

• Use Neutrino Weighting Algorithm to integrate
  over under etas of neutrinos, sum over masses and
  give one number per event
• Correlated with true top quark mass, but not the
  same thing

43
44
Dilepton 2nd observable
44

• Can use same KDE that gives 2d PDFs in
  LeptonJets channel to form 2 PDFs in Dil events
• Look for variable giving SB separation and mass
  information with as little correlation to NWA
  mass as possible
• We used mT2 as a second observable

44
45
Likelihood
45

• Use Kernel Density Estimation to form PDFs that
  are 2d in observables (mtreco, wjj), (mtnwa, Ht)
  and 2d in quantities of interest mt, dJES
• Make measurement in single likelihood
• Keep two additional measurements standard 2d
  LeptonJets-only measurement and standard 1d DIL
  measurement (1d in quantity of interest, but 2d
  in observables)
• Use local polynomial smoothing algorithm to
  smooth out the likelihood in the grid of (mt,
  dJES) space

45
46
Boundary cuts
46

• 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
• Efficiency high for signal events

46
47
Tagged fake template
Tagged fake template
mT2
mNWA
mT2
mNWA
47