Measuring the top production cross
section using dilepton events
Peter Wittich
February 22nd, 2005
Fermilab - Peter Wittich
1
Why study the top quark? (I)
yt 
• New particle, barely
characterized
2mt
1
• Top is extremely heavy
v
“Yukawa scale”
• CDF & DØ discovery in 1995
(mtop≈178 GeV):
• Special relation to
missing Higgs boson
CDF/DØ
2 fb-1
goal
February 22nd, 2005
• Strategy: look for
discrepancies with
Standard Model
Fermilab - Peter Wittich
2
Why study the top quark? (II)
• Open Questions:
• Is top production
described by QCD?
Resonant production?
• Is BR(t→Wb)≈100%?
Non-SM decays?
• Is what we call top really
top or top plus X, X
possibly exotic?
• Today’s signal is
tomorrow’s background
• Understanding of sample
of high momentum
leptons + missing energy
is crucial for exotic
searches, e.g., SUSY
February 22nd, 2005
~
t  t1  LSP
CDF: Phys. Rev. D63, 091101 (2001)
Fermilab - Peter Wittich
3
_
Fermilab’s Tevatron pp Collider
• Tevatron is world’s highest
energy collider:
√s = 1.96 TeV
• Run 2 data-taking started
in 2001
•
•
•
•
CDF and DØ upgraded
Ultimate goal: >40x Run 1
Currently: >350/pb
This analysis: 200/pb
• Tev is at energy frontier
until LHC turn-on (~2007)
• Tev is the only lab making
top until then!
February 22nd, 2005
Fermilab - Peter Wittich
4
Top production & decay @ Tevatron
• In p-pbar collisions, top
quarks are produced
mainly in pairs
• At √s=1.96 TeV, qqbar
production is dominant
process (~85%)
• Due to large mtop, no
toponium states
• Standard-model top
decays mainly via
tWb
• W daughters label
decay mode
February 22nd, 2005
85%
•qqbar
(NB: qq, gg fractions reversed at LHC)
15%
•gg
BR(%)
di-l
10
l+j
44
all-j
46
Fermilab - Peter Wittich
5
Why dilepton events?
• Dilepton channel is
• Clean – small SM backgrounds
• Different experimental
requirements from higher-stats
channel (l+jets)
• Don’t have to identify b quarks
• Different backgrounds
• Part of a exp. program
• SM top: dil, l+jet, all-hadronic
all agree w/ one-another
• Run I dilepton kinematics
caused some excitement
• Reminder: CDF (109/pb)
• 7 em events out of 9 total
• Small event sample intriguing –
check with larger sample (×2
larger)
February 22nd, 2005
 ( tt )8.243..44 pb
SM :  (tt )  5.2  0.3 pb @1.8TeV
Fermilab - Peter Wittich
6
New physics hints in Run 1 dileptons?
Last time we looked, we saw something intriguing here….
tt
Standard Model top or
top + something else?
Total event energy
Kinematics in Dilepton Events
February 22nd, 2005
Fermilab - Peter Wittich
7
Supersymmetry – one possibility
• Based on fundamental
symmetries
• Many string theories are
supersymmetric
• Solves some technical problems
of Standard Model
• How: double particle spectrum!

e e

• Worked before: postulate positron for
quantum mechanics
• Introduce “super-partners” with
different spin
• Makes theory self-consistent
• Also provides dark matter candidate
• But: where are they?
• M(positron)=M(electron)
• But not so for ~e
• SUSY is broken!
Dark Matter
Candidate
• Should be visible in near future…
February 22nd, 2005
Fermilab - Peter Wittich
Particle
Superpartner
e,n,u,d
~
~
~
~
e ,n , u , d
 ~
~
g,W,Z,h 1 ,  2 ,
~ 0 ...~ 0
1
4
8
Event topology
pp 

• Two high-momentum,
opposite-sign leptons
• (e,µ, some t) from W decay
• 2 b jets
• Heavy flavor ID not used for
this analysis
tt


W W bb
 l l nn bb
 l  l  ETmiss jj
• Large missing energy (ETmiss)
• Two escaping neutrinos
February 22nd, 2005
Fermilab - Peter Wittich
9
CDF detector
   ln(tan
General, multi-purpose detector
•Focus on charged-particle tracking
Features:
• Silicon tracker
• large radius tracking
wire chamber
(COT)
• Magnet (1.4T)
• Calorimeter
ẑ

2
)
ŷ
x̂
• (|η| < 3.6)
• muon chambers
• (|η| < 1)
February 22nd, 2005
Fermilab - Peter Wittich
10
Measurement strategy
 (tt ) 
• For this measurement we need to
•
•
•
•
•
N obs  N bgnd
(  A)  Ldt
Collect lots of data (Ldt)
Select signal events (Nobs)
Understand our signal acceptance (A)
Understand corrections to this acceptance (ε)
Estimate our backgrounds (Nbgnd)
• Consider control region (njet < 2) to test background
estimates
• We pursue two independent and complementary
analysis approaches
February 22nd, 2005
Fermilab - Peter Wittich
11
Collecting the data
• Events are triggered on
one high-momentum
lepton (e or m)
• We use data collected up
to 9/2003, corresponding
to 197 pb-1
• (cf ~108 pb-1 Run I)
Trigger
e
Central (|η|<1)
Forward
(1.0<|η|<2.6)
µ
Central (|η|<1)
ET>18 GeV,
PT>8 GeV
ET>20 GeV
ETmiss>15 GeV
PT>18 GeV
Currently updating
to 9/2004
• Additionally
• √s 1.8 TeV→ 1.96 TeV
• Electron acceptance
• |η| < 1 → |η| < 2
 expect more than twice
Run I yield
February 22nd, 2005
Fermilab - Peter Wittich
12
Aside: Triggering at hadron colliders
Like drinking out of a fire hydrant…
• Interactions occur at ≥ 1 MHz (every 396 ns)
• Have to sort out ≈ 100 Hz to save
• This is the job of the trigger system
• Combination of custom electronics and computer farms
• Even top physics profits from a capable trigger…
• CDF’s inclusive trigger a big benefit
1.7 MHz bunch
crossing rate
Detector
Hardware tracking for pT 1.5 GeV
Muon-track matching
46 crossing
L1 pipeline
L1 trigger
Electron-track matching
Missing ET, sum-ET
20-30 kHz L1 accept
Silicon tracking
Process
Inelastic ppbar
b anti-b quark
W boson
ttbar
into2005
lljjETMiss
February 22nd,
Rate
(45E30)
2.7 MHz
450 Hz
10-1 Hz
10-5 Hz
4 L2
buffers
L2 trigger
~300 Hz L2 accept
100’s of CPU’s
L3 trigger
Jet finding
Refined electron/photon finding
Full event reconstruction
70 Hz L3 accept
tape
≈ 100Hz with
data compression
This analysis
Fermilab - Peter Wittich
13
Event Selection: Two independent analyses
• Each analysis is seeded by a single, isolated high-momentum
lepton (e or m)
• Electron is track plus matching calorimeter cluster consistent with
electron test beam
• Muon is track plus matching stub in muon chambers
• Split occurs when we look for the second lepton
“LTRK” analysis
“DIL” analysis
•Two well-identified leptons
•Second lepton uses
traditional lepton ID in
calorimeter, muon chambers
•Possibly looser requirements
•Higher purity, lower
statistical significance
•Second lepton is just track isolated in
drift chamber (“tl”)
•Increase acceptance at expense of purity
•Get ~hadronic t & holes in lepton ID
coverage
•Lower purity, higher statistical
significance
Two independent, complementary approaches
February 22nd, 2005
Fermilab - Peter Wittich
14
Event selection details (I)
• Identify leading lepton
• Central muon (|η|<1) or central or forward electron (|η|<2)
• ET,pT>20 GeV, isolated in the calorimeter
• Re-cluster jets, recalculate missing energy (ETmiss )
with respect to lepton
• Luminous region is extended along direction of beam
• Defines event vertex in along luminous axis (Z axis)
• Look for second lepton
• DIL, LTRK as described previously
February 22nd, 2005
Fermilab - Peter Wittich
15
Event selection details (II)
• Require ETmiss>25 GeV
• Reject events with ETmiss co-linear with jets or lepton
• LTRK: Reject events with ETmiss parallel or antiparallel to isolated track
• If lepton pair mass is in Z boson mass region,
apply additional rejection
• Require ≥2 jets
• Require leptons to have opposite charge
• DIL:
• Increase purity by requiring HT ≡ (scalar sum of
event energy) > 200 GeV
February 22nd, 2005
Fermilab - Peter Wittich
16
Signal: Top acceptance
all hadronic, 46%
ee, emu, mumu,
2%
t+X, 20%
single prong, 3%
LTRK
DIL
multiprong,
tau + jet 16%
direct e,mu, 5%
l + jets , 29%
• Determine from PYTHIA Monte Carlo (mT=175 GeV)
• Apply trigger efficiencies, lepton ID Monte Carlo correction
factors, luminosity weights for different detector categories
• DIL:
(0.62 ± 0.09)%
• LTRK:
(0.88 ± 0.14)%
[ This acceptance includes the BR(Wlν)=10.8% ]
February 22nd, 2005
Fermilab - Peter Wittich
17
Estimate efficiencies from data
• Estimate differences between simulation and
data for Lepton ID using Z boson data
• Select Z boson candidates with one tight lepton leg
and one probe leg
• Measure efficiencies of probe leg, compare to
Monte Carlo efficiencies
• LTRK: Estimate “track lepton” (tl) selection
efficiencies:
• Use W boson sample selected w/o tracking
requirement
• Compare track efficiency with efficiency for W track
in top Monte Carlo simulation
February 22nd, 2005
Fermilab - Peter Wittich
18
Backgrounds
Categorize backgrounds as instrumental or physics
• Instrumental backgrounds
• False ETmiss or leptons
• Drell-Yan Z →ee, µµ
• Mismeasurement gives false ETmiss
• W+jets
• Jet is mis-id’d as lepton
• Physics backgrounds:
LTRK
dominant
DIL:
equal
• Real leptons, ETmiss
• Diboson (WW, WZ, ZZ)
• Drell Yan Z →tt
February 22nd, 2005
Fermilab - Peter Wittich
19
Instrumental: Drell-Yan background (I)
pp  Z / g  l l
BR    252 pb
*
 
q
_
q
Z/g*
ee+
miss resolution:
•
For
optimal
E
T
• Large cross section but no
• Correct jets for uniform
intrinsic ETmiss
calorimeter behavior
• False ETmiss
• Detector coverage isn’t 4π
• Correct ETmiss with these
• Reconstruction isn’t perfect
jets
• Tails of ETmiss resolution
• LTRK: for “tl”, correct
critical
ETmiss if tl is min ionizing
• Simulation doesn’t accurately
model this
February 22nd, 2005
• μ’s
• Undermeasured e’s
Fermilab - Peter Wittich
20
Drell Yan background (II)
ETMiss
  Miss  8
j
 ET ET
hemi
“Jet significance”
Data-based scale
• DIL and LTRK use special cuts
to suppress DY background
• Require min δΦ(j or l, ETMiss)
• Additionally in “Z window”
(76<mll<106 GeV)
• DIL: “Jet significance”
• LTRK: Boost ETMiss requirement:
>40 GeV
• Estimate residual contamination
ETmiss>25
• Overall normalization: Z-like data
set: Two leptons, high ETMiss, in Z
window
• Remove signal contribution
• Use simulation to estimate
background outside of Z window
• Low data counts dominant
uncertainty
February 22nd, 2005
Fermilab - Peter Wittich
21
Instrumental: fake lepton background
Illustrative fake rate example for LTRK
•Fakes from W+1p
One or both leptons are not real
• LTRK
• Single pions from jets
•Fake rate as fcn(ET)
• DIL:
• Jet that passes electron
requirements
• π punch-through to muon
chambers
• Measure probability from jet
sample
• Sample with triggered on one jet
with more than 50 GeV of energy
(a.k.a. ‘Jet50’)
• Apply to our data sample
February 22nd, 2005
Fermilab - Peter Wittich
•Fake rate as fcn()
22
# observed (predicted) fakes
Instrumental: fake lepton background
• Cross-check
technique on
different
samples
• Check scalar
prediction and
shapes
Jet20
Jet70
ET of observed (predicted) fake tracks in green (black)
•Scalar prediction of rates
LTRK
Obsv
Pred
DIL
Obsv
Pred
j20
74
70 ± 14
j20
51
49 ± 6
j70
316
314 ± 73
j70
75
65 ± 9
inc lep
231
189 ± 37
j100
69
114 ± 31
February 22nd, 2005
Fermilab - Peter Wittich
23
Physics backgrounds
• Diboson:
• WW, WZ, ZZ
• Drell Yan Z/g*→tt
• In both cases:
• Real missing energy
• Jets from decays or initial/final
state radiation
• Estimates derived from
Monte Carlo calculations
• ALPGEN, PYTHIA, HERWIG
• Cross sections normalized to
theoretical calculations
• Correct for underestimation
of extra jets in MC
• Determine jet bin reweighting
factors for Z → tt from Z →ee,
mm data
• Reweight WW similarly
February 22nd, 2005
Fermilab - Peter Wittich
24
Signal acceptance systematics (I)
Systematic
Lepton ID efficiency
LTRK(%)
DIL(%)
5
5
6
-
6
5
- Variation of data/MC correction factor with isolation
tl efficiency
- iso efficiency difference btw W+2j data and ttbar MC
Absolute jet energy scale
- vary jet corrections by ±1σ, δ(acc) for evts w/≥2 jets
February 22nd, 2005
Fermilab - Peter Wittich
25
Signal acceptance systematics (II)
Signal Systematics Continued
Initial- and Final-state radiation
LTRK(%)
DIL(%)
7
2
6
6
5
6
- ISR: difference from no-ISR sample
- FSR: parton-matching method, different PYTHIA tune
Parton Distribution Functions (PDF’s)
- default CTEQ5L vs MRST PDF’s, different as samples
Monte Carlo Generators
- compare acceptance of PYTHIA to HERWIG
February 22nd, 2005
Fermilab - Peter Wittich
26
Systematic Uncertainty on Background Estimate
LTRK(%)
DIL(%)
5 (6)
5(-)
Jet Energy Scale – same procedure on bkgnd acc.
10
18-29
WW, WZ, ZZ estimate
20
20
30
51
12
41
Systematic
Lepton (track) efficiency – same as signal
- Compare scaled MC jet estimate to straight calculation
Drell-Yan Estimate
- Absolute scale (data driven), Monte Carlo shape
Fake Estimate
- J20, J50, J70, J100 x-check
- DIL: statistical uncertainty
•Total syst uncertainty due to background is ±0.6 pb for DIL,
±1.0 pb for LTRK
(NB: these are applied to background, not final x-section)
February 22nd, 2005
Fermilab - Peter Wittich
27
Signal and background vs data
DIL
LTRK
Good agreement in
background region
February 22nd, 2005
(tt) signal region
Fermilab - Peter Wittich
28
Cross section results
 (tt ) 
N obs  N bgnd
SM :  (tt )  6.7 00..79 pb
(  A)  Ldt
At NLO @ √s=1.96 TeV for mtop = 175 GeV:
hep-ph/0303085 (Mangano et al)
DIL:
 (tt )  8.4
3.2
 2.7
( stat)
1.5
1.1
( syst )  0.5(lum ) pb
LTRK:
 (tt )  7.022..73 ( stat) 11..35 ( syst )  0.4(lum ) pb
(Assume BR(W→lν)=10.8%)
• Both measurements consistent with SM calc
• Error is statistics-dominated
• Combining results makes for better measurement
February 22nd, 2005
Fermilab - Peter Wittich
29
Double-tagged top dilepton candidate
• An event with 2 jets and 2 muons
jet
m
m
jet
ET
• Both jets show displaced secondary vertices
from the interaction point: b jet candidates
February 22nd, 2005
Fermilab - Peter Wittich
30
Cross-checks: W, Z cross sections
LTRK, DIL
BR ( Z / g *  ll )   ( pp  Z / g * )  252 pb
BR (W  ln )   ( pp  W )  2.69nb
Gauge bosons as “standard candle”
• Measure W, Z cross sections
• Use analysis tools, selections
and samples
• Validates acceptance,
efficiencies, luminosity
estimate
• Good agreement with other
measurements in all cases
February 22nd, 2005
Fermilab - Peter Wittich
E/p in W boson decays
31
Cross-checks: b content
•Signature of a B decay is a displaced
vertex:
•Long lifetime of B hadrons (ct ~ 450
mm)+ boost
•B hadrons travel Lxy~3mm before
decay with large charged track
multiplicity
• Top decays should have two
b quarks per event
• We don’t use this info here
• Consider b quark content as
check
• Look for jets consistent with
long-lived particles detected in
Silicon Vertex tracker
 Number of events with
detected b-like quark jets
consistent w/expectation
• DIL: 7, LTRK: 10
Top Event Tag Efficiency
False Tag Rate (QCD jets)
February 22nd, 2005
55%
0.5%
Fermilab - Peter Wittich
32
Cross-checks: Tighten 2nd lepton ID
0 jets
#
1 jet
uncert
#
≥2 jets
uncert
#
uncert
Top dilepton
0.11
0.03
1.40
0.24
4.52
0.74
Diboson
8.44
2.15
2.54
0.67
0.48
0.15
Drell-Yan
3.21
2.31
2.59
1.51
0.75
0.50
Fakes
1.24
0.26
0.37
0.08
0.11
0.02
Total bkgnd
12.88
3.17
5.50
1.65
1.33
0.52
Total pred.
12.99
3.17
6.89
1.67
5.85
0.90
Observed
14
4
Signal
7
background
• LTRK sample: apply lepton ID on track lepton
• Very few fake leptons, no hadronic t’s
 (tt )  8.5
4.5
3.5
( stat )
1.8
1.4
( syst )  0.5(lum ) pb
• Good agreement with DIL, LTRK
February 22nd, 2005
Fermilab - Peter Wittich
33
Cross-check: vary Jet, 2nd lepton thresholds
Central value
LTRK
Cross Sections - Summary
14
12
pb
10
8
6
15,20
15,25
20,20
20,25
25,25
10,20
4
2
0
Kinematical Region
(Jet Et, Trk Pt)
• Vary jet threshold, 2nd lepton momentum threshold
• Changes background composition
• Fake dominated → physics dominated
→ See consistent results (NB Uncerts very correlated)
• Choose value with best a-priori significance as central
value
February 22nd, 2005
Fermilab - Peter Wittich
34
Combining the cross sections
• Combining two
measurement reduces
3.0
the largest uncertainty
(statistics)
1
2.6
• Strategy: divide signal,
background expectation Acceptance ratio:
and data into three
DIL:LTRK:Common 1:2.6:3.0
disjoint regions
• Use extra information
about events (high
purity, low purity) and
higher stats to get better
measurement
February 22nd, 2005
Fermilab - Peter Wittich
35
Combination technique
L
subgroup
P(n

a
a
obs
a
| n pred ( tt ))
α=product over DIL only, LTRK only, overlap
•For three regions, maximize combined Poisson likelihood
•Be conservative w/ systematics between regions
•Treat as 100% correlated, distribute to give largest total
systematic
 (tt )  7.0
2.4
 2.1
( stat )
1.6
1.1
( syst )  0.4(lum ) pb
12% reduction in
statistical error
February 22nd, 2005
Fermilab - Peter Wittich
36
Kinematical distributions
With larger statistics, we can start going beyond counting
experiments to do shape tests on our selected sample.
Use larger statistics of LTRK to examine sample kinematics
KS = 75%
KS = 66%
Data follow expected distribution of top + background
February 22nd, 2005
Fermilab - Peter Wittich
37
Flavor distribution
Use sample with two identified lepton (DIL) to
look at flavor distribution
channel
Expected
(scaled to 13
total obsv’d)
Observed
ee
3.3 ± 0.5
1
mm
2.8 ± 0.5
3
em
6.8 ± 0.8
9
 Flavor distribution is consistent with expectation.
February 22nd, 2005
Fermilab - Peter Wittich
38
What’s next for top?
• CDF has many other
measurements that use the
dilepton channel
• Dedicated hadronic tau
measurement
• Detailed kinematic studies
• W helicity (dil + l + jets)
• Top mass in dilepton channel
• Combined cross section
dilepton and l + jets channels
• Other cross-sections en
route to publication
• Mass results
• Dilepton & lepton + jets
• See CDF public top results
page for full details
http://www-cdf.fnal.gov/physics/new/top/top.html
February 22nd, 2005
Fermilab - Peter Wittich
39
Conclusions
• We have measured the tt production cross section in
the dilepton decay channel using 197 pb-1 of Run II
data
• Our result is (mt = 175 GeV/c2, BR(W→lν)=10.8%):
 (tt )  7.0
2.4
 2.1
( stat)
1.6
1.1
( syst )  0.4(lum ) pb
consistent with SM prediction of   6.7 00..79 pb
• Kinematics, flavor distribution of data also consistent
with Standard Model expectation
• Paper published
• Phys. Rev. Lett. 142001 93 (2004)
• First published Run 2 high momentum physics paper from
Tevatron!
February 22nd, 2005
Fermilab - Peter Wittich
40