measurment of the top quark - smu physics · 12/07/2012 . standard model of elementary particles 2...
TRANSCRIPT
1
Precision measurement of the top quark
mass in events with two leptons
Yuriy Ilchenko
(Southern Methodist University)
Thesis Defense for Ph.D.
in Experimental High Energy Physics
12/07/2012
Standard Model of elementary particles
2
Interaction particles – gauge bosons (integer spin):
mediators of the following interactions
Weak
Electromagnetic
Strong - SU(3)
Standard Model (SM) is the theoretical
framework to describe elementary particles
Matter particles – fermions (half-integer
spin): leptons ( e, μ, τ, νe , νμ , ντ ) and
quarks ( u, d, c, s, t, b )
Elementary Particles in the
Standard Model
Electroweak - SU(2)L x U(1)Y
Fundamental Interactions
Fermions acquire mass by interacting with Higgs field through
their Yukawa coupling
Excitation of the Higgs field is the Higgs boson
Standard Model of elementary particles
3
To generate masses for W, and Z bosons
Electroweak gauge symmetry is broken by introducing Higgs field
Higgs mechanism eliminates Goldstone Boson and produces mass
SU(2)L x U(1)Y → U(1)Q
Leads to massive bosons
Lagrangian with Higgs doublet
The top quark
4
Top quark properties:
Fermion, spin sz=1/2
Heaviest known elementary particle –
about 175 GeV
short lifetime – τ t ≈ (3.3+1.3-0.9) x 10-25 s
τ t < τ had (5.23 x 10-25 s), decays
before hadronizing
the only quark whose properties
can be studied in isolation
Top quark history:
Predicted by the theory (SM) – Γb(Z→bb) consistent with b weak isospin
partner
Observed in 1995 at by D0 and CDF
Completed the existence of 3rd
generation of quarks
Fermions in the Standard Model (SM)
Top quark is the heaviest
elementary particle known
5
The top quark in the Standard Model
Mass of the top quark mt is free
parameter in the SM
Top quark role in the SM :
o Yukawa coupling to Higgs is
close to 1 (0.994 ± 0.007) –
indicate a potential fundamental
role of top in EWSB
o sets an indirect constraint on the
Higgs mass and other particles
through radiative loop
corrections:
Feynman diagrams for radiative corrections
=> Important to measure mt precisely!
Constraints on mt and mW @ 68% CL
(2009). Shaded region – range of Higgs
mass.
6
Top quark production
At hadron collider two production ways:
In pairs (tt) via strong interaction
Single quark via electroweak interaction
Single top leading order diagrams Double top leading order diagrams
tt pair Single t
Single top:
• σ ≈ 3.4 pb @ 2 TeV
• large background
Double top
• σ ≈ 7.0 pb @ 2 TV
• low background
qq annihilation gluon fusion
(quark-antiquark annihilation -85%,
gluon fusion 15%)
=> The analysis is performed on top pair events
First evidence by D0 in 2006 and CDF in 2009
7
Top quark decay
In the Standard Model top quark decays
through electroweak interaction:
In 99.8% cases: to W and b-quark
(t→W+b )
Other cases < 0.02%: W and s or
even d quarks
Top pair decay products
b-quark hadronizes into a jet
W that decays further
two light quarks that hadronize into
two jets, or lepton + neutrino
Top quark decay mode
Top pair final states are dictated by possible decay modes of W boson:
Dilepton (WW → llvv)
Lepton + jets (WW → lvqq)
All jets (WW → qqqq)
8
Top pair final states
Dilepton channel (WW → llvv):
• Distinct signal signature – two
leptons, two jets, two neutrinos
• Low background
• Low statistics – 9% of ttbar
events
• Lepton + jets channel (WW → lvqq)
• Modest background
• Large statistics – 45% of ttbar
events
• All jets channel (WW → qqqq)
• Large background
• Large statistics– 46% of ttbar
events
e,μ e,μ
b-jet
ν ν
b-jet b-jet b-jet
ν e,μ
jet
jet
Dilepton mode Lepton +jets mode
ttbar branching ratios
Tevatron
9
Proton-antiproton collider
2 km in diameter
Located at Fermilab near
Chicago
Center-of-momentum energy
is 1.96 TeV
Has two experiments: CDF
and D0
Delivered about 12 fb-1 of data
Each experiment
recorded about 10 fb-1
It was shut down in Sep 2011
Downtown Chicago Fermilab
CDF D0 Tevatron
10
The D0 detector
Multi-purpose detector at one of
the ppbar collision points:
Central Tracking (Silicon
Microstrips and Fiber tracker)
Determine particle
momentum
Vertex, charge, particle ID
Ur/Liquid Argon Calorimeter
Measure energy of
photons, electrons, jets
Muon Spectrometer (Drift
Tubes, Scintillators, and Toroid)
Identify muons
Complement the tracker in
muon momenta
measurement
Schematic view of the D0 detector
Central Tracking System
Spherical (r, θ, φ) or cylindrical (ρ, θ, φ)
coordinate systems are used,
where θ → η
(pseudorapidity)
11
Data Quality Monitoring
Data Quality Monitoring (DQM) is an important and integral part of the data taking
process.
It is performed
online throughout the data acquisition
offline processing at computer tiers
during production of simulated data
D0 and ATLAS many similarities in DQM
Online monitoring
Incoming data comes in histograms
DQM Framework executes algorithms
to check quality of the data
DQM Display is a platform to monitor results, debug problems, etc.
I have worked on Online DQM at ATLAS detector – in particular, DQM Display
Online Monitoring Data Flow at ATLAS
12
DQM Display (DQMD) DQM Display serves as visualization tool of the automatic data quality assessment
Main features
Shifters can validate the quality of the incoming data
Be warned about the problems related to data quality
Can see relevant information about the origins of problems
Provides graphical representation of the detector with navigation functionality
Main Panel of DQMD Detailed Panel of DQMD
Y. Ilchenko et al` `Data quality monitoring display for ATLAS experiment at the LHC,'‘ J.Phys.Conf.Ser. 219, 022035 (2010)
13
Mass measurement in dilepton channel
Statistical Uncertainty is expected to be low
Emphasis on improving Systematical Uncertainty
Standard Jet Energy Calibration is the main contribution to
Systematical Uncertainty
Improve Jet Energy Calibration
Neutrino Weighting (NuWT) method is used to measure mt in dilepton channel –
Based on MC templates
Use Likelihood approach – gives central value and stat. uncertainty
Choose best
discriminating
variables xi
Form signal and
background MC
templates of xi
Perform likelihood
fit between data
and templates
Method Calibration
(mt,σstat)
14
Justifying templates (Bayes’ theorem)
Weight produced by NuWT indicates how probable the event to be from ttbar
event of a certain mass mt
P({o}|mt) – probability to observe final state {o} given top quark mass mt
Bayes’ theorem relates P({o}|mt) and P(mt|{o})
P(mt|{o}) – probability measure top quark mass mt given ttbar decayed to
observed final state {o}
Assuming no prior knowledge about P(mt):
or
It allows
to form templates W({o}|mt) from MC for different mt hypotheses
use Likelihood to extract mt and stat. uncertainty when analyzing data
15
Kinematic reconstruction and weight
Kinematic parameters of leptons and b-jets
measured directly, neutrinos leave
undetected
Assuming neutrinos η and mt, event can be
fully kinematically reconstructed
Assign weight − W(η1,η2|mt) − how
consistent an event with top-quark mass
equal to assumed mt and particular choice
of neutrino rapididities η1,η2
Dilepton (eμ) event
Each neutrino - 2 solutions
from quadratic equations.
Additionally, two-fold ambiguity
in lepton-jet pairing.
Results in Nsoln <=8
σx,,σy – missing energy resolution
16
Neutrino rapidity distribution
MC neutrino rapidity distribution
The total event weight with assumed mt
obtained by
• integrating over η1,η2
• ρ(η1) and ρ(η2) – probability
distribution functions from MC
simulation
ρ(η)
where {o} is observed final state
Width of rapidity distribution vs
assumed top quark mass
Neutrino rapidity distribution
simulated for different masses of the
top quark, fit to Gaussian
dependence of the Gaussian width on
mt is fit to a straight line
17
NuWT: neutrino rapidity distribution
Schematical example of an event weight
distribution
μ
σ
Scanning over different mt values
produce weight distribution for each
event
𝑊(𝑚𝑡)
𝑚𝑡
Average event weight distribution
has broader peak due to resolution
effects and jet permutations
ISR/FSR shift mean value and add
high-mass tail
Every weight distribution and so every
event is characterized by moments - mean
value μ and - width σ
Average event weight distribution
18
Maximum Likelihood: templates
Construct probability density histograms –
“templates” - from moments
signal template (hs): moments, input top
quark mass
background template (hb): moments
For every MC event (either signal or
background) – produce weight distribution
Signal: put (μ, σ) in 3D histo at mt
Background: put (μ, σ) in 2D histo
Normalize templates
hs
hb
Slice of the signal template at
at mt=170 GeV
Background template
eμ
eμ
19
Maximum Likelihood: likelihood function
Likelihood function is given as
where
xi=(μ, σ) – weight distribution moments
ns – expected number of signal events
nb – expected number of background
events
Negative log likelihood with fit
(from previous analysis iteration)
t̂
tm̂
Fit –log(L(mt )) with parabola
Based on the assumption of the Gaussian shape of L(mt )
Minimum of parabola defines top quark mass
estimate mt
Half width where –log(L) rises to 0.5 unit above its
minimum defines statistical uncertainty
20
The NuWT Method calibration
Our method is based on number of assumptions
Likelihood function L(mt ) is Gaussian
Neutrino rapidities are distributed according to SM expectations
Additionally background shifts mt from its true value
NuWT calibration is needed
verify method performance and precision
accounts for possible residual effects
Gives more accurate central value and stat. uncertainty
NuWT is tested with pseudoexperiment technique (ensemble testing)
Based on events from MC
Each pseudoexperiment is similar to data events within
fluctuations
21
Ensemble testing
Ensemble testing - running many pseudo-
experiments
For each mt hypothesis, measurement is
performed in 1000 pseudo-experiments
for pseudo-experiment choose events randomly
from the signal (S) and background (B) MC
samples
• Number of S and B events are obtained from
Poisson distribution
• Mean of the Poisson is random from
Gaussian
• Mean and Width of Gaussian – expected
yield and uncertainty
• the average event numbers match the
expected yield
• Select ones where total number of events
equals those in data
Background MC Sample
Signal MC Sample
…
Pse
ud
oe
xp
mn
ts
22
Ensemble testing: calibration curve
The relationship between fitted top quark mass and the actual
input top quark mass is fitted to a line.
Linear fit is used as a calibration tool on the results from data.
Calibration points and fit:
ee eμ μμ
Calibrated cent. value and stat.
uncertainty:
α β
High χ2 of calibration curve
Pseudo-experiment (PE) studies to
calibrate the measurement show
calibration curve high χ2
high χ2 main sources
signal template statistics
background oversampling
non-unitary ALPGEN weights
Dominating contribution is found to be
due to template statistics
23
Calibration curve for eμ channel
High χ2
Template statistics effect on
calibration curve
To estimate the template statistics effect
varied templates within uncertainties for
each pseudoexperiment
produce fitted mass distribution for
each pseudoexperiment
plot RMS of those distributions
average RMS is 0.3 GeV
24
Template statistics contributes additional uncertainty of 0.3 GeV added in
quadrature with statistical uncertainty at every input mass point
Calibration curve for eμ channel with template statistics uncertainty
Good χ2
25
ee eµ µµ combined
without template stat. 170.2 ± 6.5 174.2 ± 3.2 183.5 ± 17.5 173.7 ± 2.8
with template stat. 170.3 ± 6.4 174.2 ± 3.2 183.8 ± 18.0 173.7 ± 2.8
Calibrated results
=> Template statistics has no or negligible effect on the
measurement
Template statistics effect on
measurement in data
26
Ensemble testing: stat. uncertainty
To test calibrated statistical uncertainty, ensemble testing technique is used
Pull variable evaluates correctness of calibrated statistical uncertainty
Pull points and fit:
ee eμ μμ
Pull corrected calibrated stat.
uncertainty:
λ
27
Event Yields and Data Sample
Apply kinematic and topological cuts – best signal to background ratio
Two isolated leptons with opposite charge
leptons Pt > 15 GeV
Two jets inclusive.
Jets Pt > 20 GeV
Topological cuts
ee: METsig>5.0;
eµ: Ht>120 GeV;
µµ: METsig>5.0, MET>40 Gev
Ht cut is effective against Z->tau tau and diboson that are largest in eμ
METSig cut rejects events with MET from resolution fluctuation
MET cut is powerful against for Z->ll background
σ is variance of prob. distr.
p(MET). σ and p(MET) are
computed from object momentum
resolutions.
28
32.95
8.1
2.21 0
ee
138.7
10.6 5.6
8.95 eμ
44.6 24.5
3.2
4.5 μμ
-Z->ll -Diboson -ttbar -Instrum.
Event Yields and Data Sample Integrated luminosity is 4.3 fb-1 of Run II
MC and data samples for modeling
Expected yields in MC and number of events in data after kinematic reconstruction
Data: 49 Data: 190 Data: 80
29
Calibrated result:
Data Measurement
mt(combined)= 173.7 ± 2.8 (stat.) GeV
Negative log likelihood (before calibration)
Combined likelihood is
obtained multiplying
likelihoods of all channels
ee eμ
μμ combined
mt(ee)= 170.3 ± 6.4 (stat.) GeV mt(eμ)= 174.2 ± 3.2 (stat.) GeV
mt(μμ)= 183.8 ± 18.0 (stat.) GeV
30
Expected statistical uncertainties
Distributions of pull-corrected calibrated statistical uncertainties
ee eμ
μμ combined
Good agreement for ee, eμ, and combined. μμ-channel is consistent with
expectation at probability level of 7%
31
Systematic uncertainties
Statistical uncertainty is calculated with Likelihood
Systematic uncertainty is due to systematic effects - due to inaccurate
equipment, assumptions, imperfect calibration
cannot be eliminated by using bigger statistics
some are of comparable scale to the statistical uncertainty
Systematic uncertainties are divided in four categories
Jet Energy Calibration
effects of overall energy scale
residual biases from jet Pt and η dependences
data to MC jet response differences
QCD interactions modeling
ISR/FSR, b-fragmentation etc.
Object reconstruction and identification
Resolutions of jets and leptons, ID efficiencies etc.
Systematics of the method
uncertainty from calibration, limited statistics for templates etc.
Replaces the standard
Jet Energy Calibration
32
Standard Jet Energy Scale
Standard JES systematic is the dominating uncertainty in previous iteration of
the analysis
Standard JES
reconstruct energy back to particle jet
includes number of corrections
Parton evolution in calorimeter
Eoffset – energy in calorimeter from
underlying event
Fη – correct for calorimeter non-uniformity
R – absolute jet response
S – correct for losses due to showering
33
Jet Energy Scale: response calculation
JES correction is dominated by error on absolute Response => needed to be
measured accurately with robust and reliable method
Missing Energy Projection factor (MPF) – method to measure absolute Response
Use back-to-back photon + jet events in CC
Employs momentum conservation
Missing Energy Projection factor
Independent of Eoffset, S effects
Does not depend on jet reconstruction algorithm
34
Jet Energy Scale: sample purity
Jet may have leading π0→γγ, thus misidentified as a photon
Photon + jet events suffer from contamination of events with two jets
(dijet)
needed to be taken into account for R calculation
Purity variable is introduced
S – number of photon + jet events
B – number of dijet events
To estimate purity
Simple event counting: MC
Hollow Cone Method (HC07): Data Example of MC based and
HC07 purity
35
Jet Energy Scale: HC07 method
Hollow Cone Method
HC07 variable discriminates a photon against a jet
HC07 – scalar Pt of track in hollow cone 0.05 < ΔR < 0.7
Fit normalized MC templates (photon+jet, dijet) to data
Purity for tight photon definition (HC07 <1.0 GeV) estimated from medium
HC07 distribution Purity in central region
36
Standard JES
Derived from γ+jet events, systematic
limited (2% averaged over top jets)
l + jets channel
extract additional JES (kJES ) using
constraint of dijet mass from W→qq’ decay
simultaneously measure kJES and mt
dilepton channel
no W→qq’
Replace standard JES with l+jets result
• 2.5x smaller uncertainty
• small systematic for scale of b-jets in dilepton vs. l+jets events
• residual systematic uncertainty to encompass Pt and η
dependence of JES in dilepton
Jet energy scale (JES)
from l+jets channel
Precise measurement of the top-quark mass from lepton+jets events at D0
Phys. Rev. D 84, 032004 (2011)
Likelihood L(mt, kjes) in l+jets with
contours of equal probability
37
Adopting the l+jets channel JES
Event topology in l + jets differs from that in dilepton channel
Higher jet multiplicity affects reconstruction and so jet energies
Hadrons can be misrecognized by algorithm as belonging to an
incorrect jet
Different color flow scenarios l + jets and dilepton have an impact
on reconstructed jet energy
in l+jets, singlet W boson produces two quarks forming a color
dipole
color dipole produces more radiation in the region between
the quarks
The difference between two channels can be evaluated using the
following response double ratio b-jet responses in l+jets and
dilepton in MC and Data
38
Double ratio shows that
the response for the b-
quarks in dilepton channel
and l + jets channel are
very close
Asymptotic value is
shifted by 0.3% from
unity
b-jets responses in
l+jets and dilepton are
almost same
Double ratio
If the event topologies for both channels were equal, the double ratio would
equal unity
Double ration for b-jets as function of jet pt
39
Double ratio
Small variation of the double ratio from one can be a result of different jet
particle multiplicity
Average number of particles in a jet in l+jets and dilepton channels
Particle multiplicity in l+jets few percent higher than that in the
dilepton sample
Enough to produce observed shift in the double ratio up to 0.3%
=> error on l+jets jes is 0.3%, added as systematic uncertainty
40
Residual systematic
Arises from the dependence of kJES on
Pt and η
Estimated from shape of 𝜎𝐽𝐸𝑆
𝐽𝐸𝑆 in
photon+jet sample
Shift jet energy by the correction
factor from distribution minus
correction average (up and
down)
𝜎𝐽𝐸𝑆
𝐽𝐸𝑆 by jet pt in η: [0,0.5]
correction average over all jets’ pt in all
η regions
To estimate residual uncertainty
New MC samples are produced
JES in the samples are shifted up
and down
41
Jet specific correction for light quarks, gluons
and heavy quarks
Jets initiated from different flavors of
partons
different kinematic
characteristics
different particles compositions
=> b and light jets have different
response
=> can result in a systematic shift on mt
Jet multiplicity for gluon and quark jets Previous iteration of this analysis
standard JES calibration to b
jets caused 1.8% shift in jet pt
42
Discrepancy in energy between Data and MC
Define correction factor for jet of flavor β
flavor-averaged in γ+jets events
Ei, Ri – single particle
energy and response
Systematical uncertainty is significantly reduced!
(Fcorr-1) for light jets in |η|<1.4 Correct jet energies based on their flavor
Jet specific correction for light quarks, gluons
and heavy quarks
New approach
• Correct b and light jet response in MC independently
• based on Single Particle responses (separately done on MC and Data)
• b/light systematic has been replaced by sample dependent systematic
(analogous to lepton+jets channel)
43
Systematic uncertainties
Major improvement –
lepton+ jets calibration
reduces standard JES
systematic that was 1.5
GeV
Major systematic effects
are coming from theoretical
modeling and systematic of
the method
Total combined systematic
uncertainty is 1.5 GeV
44
Major modeling uncertainties
Higher order effects – [0.6 GeV]
Default samples – ALPGEN+PYTHIA (no such higher-order effects as
gg initial state, additional radiation of hard jets)
Compared to MC@NLO+HERWIG
Color reconnection – [0.5 GeV]
Color reconnection – strong interaction between underlying event and
hard-scattering process
Sample with and without explicit color reconnection in PYTHIA are
compared
PDF uncertainty – [0.5 GeV]
Default samples – use set of PDF from CTEQ6L1
Compared to reweighted sample that matches CTEQ6M set of PDFs
Template Statistics – [0.5 GeV]
Arise from limited MC statistics
Bins in templates are varied within their uncertainties
New measurements performed 1000 times
45
Conclusion (I)
In 4.3 fb-1 the top quark mass estimate:
Combining with previous measurement 1 fb-1
(total of 5.3 fb-1) using BLUE method (Best
Linear Unbiased Estimator)
=> error on top quark mass is 1.6%
mt = 174.0 ± 2.4 (stat.) ± 1.4 (syst.) GeV
mt = 174.0 ± 2.8 (stat.+ syst.) GeV
mt = 173.7 ± 2.8 (stat.) ± 1.5 (syst.) GeV
Most precise measurement in dilepton
channel
First time JES systematics done to take
advantage of l+jets scale
Published in PRD (RC) 86, 051103 (2012)
Current D0 top quark mass average
Top quark mass measurement with
neutrino weighting method
46
Conclusion (II)
Constraints on mt and mW @ 68% CL
(2012).
Δχ2 of a global fit to electroweak data as a
function on the Higgs mass (2012).
mH = 94 +29 -24 GeV
Tevatron (World) average top quark mass
Contributions
47
D0 experiment
Top quark mass measurement (PRD (RC) 86, 051103 (2012))
γ+jet purity estimation (D0 internal Note 6327)
P20/p17 MC verification
ATLAS experiment
LAr Panel and commissioning
DQM Display - Y. Ilchenko et al` `Data quality monitoring display for ATLAS
experiment at the LHC,'‘ J.Phys.Conf.Ser. 219, 022035 (2010)
48
Back-up slides
49
Oversampling
Number of pseudoexperimnents (PE) is large
• Events are selected in ensembles more
than once
• A correction for correlation among
ensembles is performed
The uncertainty on average top quark mass
is corrected by following facor
• Nsig - # of signal events in PE
• Npe - # of PEs tried
• NMC - # of events in MC sample after
selection
The uncertainty on pulls are corrected with
Background MC Sample
Signal MC Sample
same event
…
Pse
ud
oe
xp
mn
ts
DQMD – Graphical View
50 DQMD Detailed Panel
DQMD – Graphical View
51 Graphical representation of Muon Spectrometer of ATLAS in DQMD