search for single top at cdf
DESCRIPTION
Search for Single Top at CDF. Bernd Stelzer, UCLA on behalf of the CDF Collaboration Fermilab, December 1st 2006. Outline. Single Quark Production at the Tevatron Motivation for Single Top Search The Experimental Challenge Analysis Techniques at CDF - PowerPoint PPT PresentationTRANSCRIPT
Search for Single Top at CDF
Bernd Stelzer, UCLA
on behalf of the CDF Collaboration
Fermilab, December 1st 2006
2
Outline
1. Single Quark Production at the Tevatron
2. Motivation for Single Top Search
3. The Experimental Challenge
4. Analysis Techniques at CDF
• Likelihood Function Analysis (955 pb-1)
• Neural Network Analysis (700 pb-1)
• Matrix Element Analysis (955 pb-1)
5. New Results
6. Conclusions
3
The Tevatron Collider
•Tevatron produces per day: ~ 40 top pair events~ 20 single top events
Cross Sections at s = 1.96 TeV
4
Top Quark Production
s-channelNLO = 0.88±0.07 pb
t-channelNLO = 1.98±0.21 pb
Observed
1995!
Wanted!
2006/7?
B.W. Harris et. al, hep-ph/0207055, Z. Sullivan hep-ph/0408049
Quoted cross-sections at Mtop=175GeV/c2
Vtb•Directly measure Vtb
Single Top ~ (Vtb)2
•Source of ~100% polarized top quarks
NLO = 6.7±0.8 pb
Mtop = 171.4 2.1 GeV/c2Current World average:
5
Sensitivity to New Physics•Single top rate can be altered due to the presence
of new Physics
-Heavy W boson, charged Higgs H+, Kaluza Klein excited WKK (s-channel signature)
-Flavor changing neutral currents: t-Z/γ/g-c couplings (t-channel signature)
Tait, Yuan PRD63, 014018(2001)
s-channel and t-channel have
different sensitivity to new physics
Z
ct
W,H+
s (pb)
1.25 t (pb)
6
ExperimentalChallenge
7
Event Signatures
Jet1
Jet2
Electron
Jet4
Jet3
MET
Top Pair Production with decayInto Lepton + 4 Jets final stateare very striking signatures!
Single top Production with decayInto Lepton + 2 Jets final stateIs less distinct!
8
Data Collected at CDF
Delivered : 2.1 fb-1
Collected : 1.7 fb-1
This analysis: 955/pb (All detector components ON)
CDF is getting faster, too!6 weeks turnaround time to calibrate, validate and process raw data
Tevatron people are doing a fantastic job!2fb-1 party coming up!
Design goal
9
Single Top Selection
Event Selection:•1 Lepton, ET >15 GeV, ||< 2.0
•Missing ET (MET) > 25 GeV
•2 Jets, ET > 15 GeV, ||< 2.8•Veto Fake W, Z, Dileptons,
Conversions, Cosmics
•At least one b-tagged jet, (secondary vertex tag)
CDF W+2jet Candidate Event:CDF W+2jet Candidate Event:
Close-up View of Layer 00 Silicon DetectorClose-up View of Layer 00 Silicon Detector
Run: 205964, Event: 337705Electron ET= 39.6 GeV, MET = 37.1 GeVJet 1: ET = 62.8 GeV, Lxy = 2.9mmJet 2: ET = 42.7 GeV, Lxy = 3.9mm
Jet2
Jet1
Electron
12mm
Number of Events / 955 pb-1 Single Top
Background
S/B S/B
W(l) + 2 jets 74 15500 ~1/210
~ 0.6
W(l) + 2 jets + b-tag 38 540 ~1/15 ~ 1.6
10
Mistags (W+2jets)
• Falsely tagged light quark or gluon jets
• Mistag probability parameterization obtained from inclusive jet data
Background Estimate
W+HF jets (Wbb/Wcc/Wc)
•W+jets normalization from data and
heavy flavor (HF) fraction from MC
Top/EWK (WW/WZ/Z→ττ, ttbar)
•MC normalized to theoretical cross-section
Non-W (QCD)
•Multijet events and jets with semileptonic b-decay
•Fit low MET data and extrapolate into signal region
Wbb
WccWc
non-W
Z/DibMistags
tt
W+HF jets (Wbb/Wcc/Wc)
•W+jets normalization from data and heavy flavor (HF) fractions from ALPGEN Monte Carlo
11
Signal and Background Event Yield
CDF Run II Preliminary, L=955 CDF Run II Preliminary, L=955 pbpb-1 Event yield in W+2jetsEvent yield in W+2jets
Single top hidden behind background uncertainty! Makes counting experiment impossible!s-channel 15.4 ± 2.2
t-channel 22.4 ± 3.6
tt 58.4 ±13.5
Diboson 13.7 ± 1.9
Z + jets 11.9 ± 4.4
Wbb170.9 ± 50.7
Wcc63.5 ± 19.9
Wc68.6 ± 19.0
Non-W26.2 ± 15.9
Mistags136.1 ± 19.7
Single top 37.8 ± 5.9
Total background
549.3 ± 95.2
Total prediction
587.1 ± 96.6
Observed 644
12
Jet Flavor Separation
• Distinguish b-quark jets from charm / light jets using a Neural Network trained with secondary vertex information
–Applied to b-tagged jets with secondary vertex
–25 input variables: Lxy, vertex mass, track multiplicity, impact parameter, semilepton decay information, etc...
• Good jet-flavor separation!
• Independent of b-jet source
• Used in all three single top analyses
13
Jet Flavor Separation II
• Fit to W+jets data shows good shape agreement
• Fit result consistent with background estimate
W + 2 jet events with ≥1 b-tag
Background
Estimate
Neural Network Fit
W+bottom 299.0 56.8
292.8 26.3
W+charm 148.1 39.4
171.6 53.8
Mistags 140.0 19.8
179.5 42.5
Sum 587.1 96.6
644.0
14
Analysis Techniques
15
Analysis Flow Chart
Analysis Event
Selection
Analysis Event
Selection
CDF Data
CDF Data
Monte CarloSignal/
Background
Monte CarloSignal/
Background
Apply MCCorrection
s
Apply MCCorrection
s
3 Analysis Techniques
3 Analysis Techniques
Result
Template Fit to Data
Template Fit to Data
Discriminant
Signal
Background
Cross Section
16
Analysis Techniques
Likelihood Analysis
Neural Network Analysis
Matrix Element Analysis
17
The Likelihood Function Analysis
t-channel LF Input Variables:•total transverse energy: HT
•Mlb (neutrino pz from kin. fitter)•Cos(lepton,light jet) in top decay frame•Qlepton*untagged jet aka QxEta•mj1j2
•log(MEtchan) from MADGRAPH•Neural Network b-tagger•LF=0.01 for double tagged events
s-channel LF Input Variables:•Mlb
•log(HT* Mlb )•ET(jet1)•log(MEtchan) •HT
•Neural Network b-tagger
pisig
N isig
N isig N i
bkg
Nsig
Nbkg
i, indexes input variable
L(x )
psigi (x i)i1
nvarpsig
i (x i)i1
nvar pbkgi (x i)i1
nvar
18
Likelihood Function Analysis
Background Background SignalSignal Background Signal
Unit area
Wbbttbar
Wbbttbar
Wbbttbar
tchanschan
tchanschan
tchanschan
19
Likelihood Function Discriminants
t-channel s-channel
Background SignalBackground Signal
Unit Area
Wbbttbar
tchanschan
tchanschan
Wbbttbar
Templates normalized to prediction
Templates normalized to prediction
20
Analysis Techniques
Likelihood Analysis
Neural Network Analysis
Matrix Element Analysis
21
Neural Network Analysis - Combined Search
•Single Neural Network trained with SM combination of s- and t-channel as signal
•14 Variables: top and dijet invariant masses, Qlxq, angles, jet ET1/2 and j1+ j2, W-boson , lepton pT, kinematic top mass fitter quantities, Neural Network b-tag output etc..Current result using 695/pb (update with 955/pb expected shortly!)Yield Estimate [695/pb]: Single-Top: 28±3 events, Total Background: 646±96 events
22
Neural Network Analysis - Separate Search
• Two NN’s trained separately for s-channel and t-channel (similar variables)
t-channel
W+heavy flavor
ttbar
s-channel
23
Analysis Techniques
Likelihood Analysis
Neural Network Analysis
Matrix Element Analysis
24
Matrix Element Approach
P(x) d (pi
)
1
M
2d
• Inspired by D0/CDF Matrix Element top mass analyses
• Here, we apply the method to a search!
• Attempt to include all available kinematic information:
Calculate an event-by-event probability (based on fully differential cross-section calculation) for signal and background hypothesis
25
Matrix Element Method
P( pl, p j1
, p j 2 )
1
d j1d j 2dp
z | M(pi ) |2
f (q1) f (q2)
| q1 ||q2 |4 W jet (E jet,E part)
comb
Parton
distribution function (CTEQ5)
Leading Order matrix element (MadEvent)
W(Ejet,Epart) is the probability of measuring a jet energy Ejet when Epart was produced
Integration over part of the phase space Φ4
Event probability for signal and background hypothesis:
Input only lepton and 2 jets 4-vectors!
c
26
Event Probability Discriminant (EPD)
EPD bPsin gletop
bPsin gletop bPWbb (1 b)PWcc (1 b)PWcj
;b = Neural Network b-tagger output
•We compute probabilities for signal and background hypothesis per event
Use full kinematic correlation between signal and background events
•Define ratio of probabilities as event probability discriminant (EPD):
27
Event Probabilty Discriminant
S/B~1/3S/B~2.5
In most sensitive bins!(EPD>0.8)
S/B~1/15, S/B~1.6All events
Templates normalized to prediction
28
Cross-Checks
29
Cross-Checks in Data Control Samples•Validate method using data without looking at single
top candidates
•Compare the Monte Carlo prediction of the discriminant shape to various control samples in data
•W+2 jets data (veto b-jets, orthogonal to our candidate sample)
30
Cross-Checks in Data Control Samples
CDF Run II Preliminary
CDF Run II Preliminary
•b-tagged dilepton + 2 jets sample•Purity: 99% ttbar•Discard lepton with lower
pT
•b-tagged lepton + 4 jets sample•Purity: 85% ttbar•Discard 2jets with
lowest pT
31
Template Fitto the data
32
Analysis Flow Chart
Analysis Event
Selection
Analysis Event
Selection
CDF Data
CDF Data
Monte CarloSignal/
Background
Monte CarloSignal/
Background
Apply MCCorrection
s
Apply MCCorrection
s
Result
Likelihood
Fit to Data
Likelihood
Fit to Data
Discriminant
Signal
Background
Cross Section
Multivariate
Analysis Technique
Multivariate
Analysis Technique
33
Likelihood Fit to Data•The distribution of the discriminant in data is a
superpositionof the single top and several background template
distributions
Obtain most probable single top content in data by performing abinned maximum likelihood fit
Background templates are allowed to float in the fit within their rate uncertainties (Gaussian constrained)
Other sources of systematic uncertainty (rate and shape) are included as nuisance parameters in the
likelihood function andare also allowed to float within their
uncertainties
34
Rate vs Shape Systematic Uncertainty
Discriminant
•Rate systematics give fit templates freedom to move vertically only•Shape systematics allow templates to ‘slide
horizontally’ (bin by bin)
Rate systematics
Shape systematics
Systematic uncertainties can affect rate and template shape
35
Binned Likelihood Fit
Binned Likelihood Function:
Expected mean in bin k:
All sources of systematic uncertainty included as nuisance parameters
Correlation between Shape/Normalization uncertainty considered (δi)
βj = σj/σSM parameter
single top (j=1)
W+bottom (j=2)
W+charm (j=3)
Mistags (j=4)
ttbar (j=5)
k = Bin index
i = Systematic effect
δi = Strength of effect
εji± = ±1σ norm. shifts
κjik± = ±1σ shift in bin k
36
Sources of Systematic Uncertainty
Single Top Rate Variations
Shape Variations
Jet Energy Scale
Initial State Radiation
Final State Radiation
Parton Dist. Function
Monte Carlo Generator
Efficiencies / b-tagging SF
Luminosity
Total Rate Uncertainty
10.5% N/A
CDF RunII Preliminary, L=955pb-1
Background
Rate Unertainty
W+bottom 25%
W+charm 28%
Mistag 15%
ttbar 23%
Backgrounds Rate Variations
Shape Variations
Jet Energy Scale
Neural Net b-tagger
Mistag Model
Non-W Model
Q2 Scale in Alpgen MC
37
Discovery Potential
38
Signal Sensitivity
We use the CLs Method developed at LEP L. Read, J. Phys. G 28, 2693 (2002)T. Junk, Nucl. Instrum. Meth. A 434, 435 (1999)http://www.hep.uiuc.edu/home/trj/cdfstats/mclimit_csm1/
•Compare two models at a time
•Define Likelihood ratio test statistic:
•Systematic uncertainties included in pseudo-experiments
•Use median p-value as expected sensitivity
Likelihood Function Analysis:
Median p-value = 2.3% (2.0)
Matrix Element Analysis:
Median p-value = 0.6% (2.5)
Q L(data | s b)
L(data | b)
Median
More signal likeLess signal like
39
Results
40
Neural Network Results
Best fit Separate Search:
Best fit Combined Search:
•Analysis very correlated with Likelihood Function analysis•Expected sensitivity similar to
Matrix Element
t ch 0.6 0.61.9 (stat) 0.1
0.1(syst)pb
s ch 0.3 0.32.2 (stat)-0.3
0.5 (syst)pb
s+t 0.8 0.81.3 (stat) 0.3
0.2 (syst)pb
41
Likelihood Function Results
t ch 0.2 0.20.9 pb
s ch 0.1 0.10.7 pb
Best fit Separate Search:
s+t 0.3 0.31.2 pb
Best fit Combined Search:
95 s+t channel
Expected 2.9 pb
Observed 2.7 pb
95% upper limit on combined single top cross section
Note: Expected limit assumes no single topCurrent result excludes modelsbeyond the Standard Model
( stSM 2.9 pb)
42
Matrix Element Technique - Result
• Matrix Element analysis observes excess over background expectation
• Likelihood fit result for combined search:
Single Top 2.7 1.31.5 pb
Single Top 2.7 1.31.5 pb
43
Observed p-value
Observed p-value = 1.0% (2.3)
bs+b
CDF RunII Preliminary, L=955pb-1
Observed p-value = 51.3%
CDF RunII Preliminary, L=955pb-1
-2lnQ
44
Central Electron CandidateCharge: -1, Eta=-0.72 MET=41.85, MetPhi=-0.83 Jet1: Et=46.7 Eta=-0.61 b-tag=1 Jet2: Et=16.6 Eta=-2.91 b-tag=0QxEta = 2.91 (t-channel signature)EPD=0.95
Single Top Candidate Event
Jet1
Jet2
Lepton
Run: 211883, Event: 1911511
u,d
d,u
45
QxEta for Candidate Events in Signal Region
1) EPD>0.60
2) EPD>0.80
3) EPD>0.90
4) EPD>0.95
Look for signal features(QxEta) in signal region
46
QxEta Distributions in Signal Region
1) 2)
3) 4)
47
Compatibility of the New Results
•Performed common pseudo-experiments – Use identical events
– ME uses only 4-vectors of lepton, Jet1/Jet2
– LF uses sensitive event variables
– Correlation among fit results: ~53%
– 6% of the pseudo-experiments had a difference in fit results at least as large as the difference observed in data
CDF II data
The result we observe in the data is compatible at the ~6% level
48
Candidate Events in 2D LF and ME Discriminant Space
•Divide the 2D discriminant space of the Matrix Element and Likelihood Function analysis into 4 regions •Define combined background region (1) and combined signal region (1)•Look also at mixed regions (2,3)
LF
Signal Hypothesis Preferred
2 prob 33.7% 2 prob 49.8%
1 2 3 4 1 2 3 4
Null hypothesis Signal hypothesis
49
Conclusions• Single top production probes Vtb and is sensitive new physics
• We improved sensitivity by a factor of 3-4 compared to published results
• We now have 2 - 2.5 sensitivity to single top per analysis!
• Presented three analyses using different techniques to separate signal from huge background
• Results consistent at 6% level but it's interesting that they show differences
• With more data and further improvements we learn what the data is telling us
• Exciting times! Back to work!
Technique s+t cross-section
Expected p-value
Observed p-value
Likelihood Function (955/pb)
0.3(+1.2/-0.3)pb
2.3% 51.3%
Neural Network (695/pb)
0.8(+1.3/-0.8)pb
coming soon coming soon
Matrix Element (955/pb)
2.7(+1.5/-1.3)pb
0.6% 1.0%
Combined Analysis (955/pb)
coming soon coming soon coming soon