11/2/2009 Ulrich Heintz - seminar - Stony Brook 1
single top quark production
Ulrich HeintzBrown University
Ulrich Heintz - seminar - Stony Brook 2
outlinetop quark introductionTevatron and DØ experimentevent selectionmatrix elementsboosted decision treesbayesian neural networkscross section and |Vtb|other measurementssummary
11/2/2009
Ulrich Heintz - seminar - Stony Brook 3
outlinetop quark introductionTevatron and DØ experimentevent selectionmatrix elementsboosted decision treesbayesian neural networkscross section and |Vtb|other measurementssummary
11/2/2009
11/2/2009 Ulrich Heintz - seminar - Stony Brook 4
top-antitop quark pair productionobserved first in 1995 by CDF and DØ
“easy” to see
11/2/2009 Ulrich Heintz - seminar - Stony Brook 5
top-antitop quark pair productionobserved first in 1995 by CDF and DØ
measure strong coupling of top quarkpp = 7.6 0.9 pb mtop = 173.11.3 GeV q = +2/3 (tW+b) preferred over -4/3 (tW-b)
top quark decayweak interaction
tWb’b’ = Vtd d + Vts s + Vtb b
tt production B(tWb) > 0.79 @ 95% CL |Vtb| >0.89 @ 96% CLtop << experimental resolution
B decays |Vub|2 + |Vcb|2 + |Vtb|2 = 1|Vub|=0.00393, |Vts| = 0.0412 |Vtb| = 0.9991
11/2/2009 Ulrich Heintz - seminar - Stony Brook 6
assume unitarity of 33 CKM matrix
t
b
W
bsd
VVVVVVVVV
bsd
tbtstd
cbcscd
ubusud
'''
single top quark productionweak interaction
|Vtb|2
no assumptions on number of generations or unitarity of CKM matrix
NLO = 1.120.05 pb = 2.340.13 pb Kidonakis and Vogt, PRD 68, 114014 (2003) for mt =170 GeV
11/2/2009 7Ulrich Heintz - seminar - Stony Brook
s-channel t-channel
11/2/2009 Ulrich Heintz - seminar - Stony Brook
8
single top quark productionsensitive to new physics 4th quark generation anomalous Wtb vertex new particles (H+, W’) FCNC
important benchmark in understanding the backgrounds to Higgs search in WH channel
11/2/2009 Ulrich Heintz - seminar - Stony Brook 9
single top quark production
s-channel
t-cha
nnel
Ulrich Heintz - seminar - Stony Brook 10
single top quark production2006: D0 announces evidence for single top production
11/2/2009
DØ Evidence paperPRL “Editor’s Suggestion”110 SPIRES citations
Ulrich Heintz - seminar - Stony Brook 11
outlinetop quark introductionTevatron and DØ experimentevent selectionmatrix elementsboosted decision treesbayesian neural networkscross section and |Vtb|other measurementssummary
11/2/2009
11/2/2009 Ulrich Heintz - seminar - Stony Brook 12
the Tevatron
counter rotating beams of protons and antiprotonsradius = 1 kmbeam energy = 980 GeV21011 protons in 36 bunches 21010 antiprotons in 36 bunchesenergy stored in beams = 35 kJtime for one revolution = 21 stime between collisions = 396 nspeak luminosity = 2.81032cm-2s-1
11/2/2009 Ulrich Heintz - seminar - Stony Brook 13
the Tevatron… still the only place to find top quarks
15
090.0878.128/1)( 2
t
Z
mM
2 km
CDF DØ
the Tevatron
11/2/2009 14Ulrich Heintz - seminar - Stony Brook
0.9 fb-1 evidence
2.3 fb-1 observation
11/2/2009 Ulrich Heintz - seminar - Stony Brook 15
the DØ detector
beam pipe
calorimeter
muon toroid
11/2/2009 Ulrich Heintz - seminar - Stony Brook 16
➔ 19 countries ➔ 80 institutions➔ 700 physicists
D0 Collaboration
Ulrich Heintz - seminar - Stony Brook 17
outlinetop quark introductionTevatron and DØ experimentevent selectionmatrix elementsboosted decision treesbayesian neural networkscross section and |Vtb|other measurementssummary
11/2/2009
11/2/2009 Ulrich Heintz - seminar - Stony Brook 18
a needle in a hay stacksingle top
dominant background: Wbb, W+jets
Ulrich Heintz - seminar - Stony Brook 19
event selection
11/2/2009
muon pT > 15 GeV, || < 2.0 electron pT > 15 GeV, || < 1.1
20 < missing pT < 200 GeV
2-4 jetsleading jet pT > 25 GeV, || < 3.4other jets pT > 15 GeV, || < 3.4
1 b-tagged jetleading b-jet pT > 25 GeV, || < 3.4
24 channels: 2 running periods 2 lepton flavors 3 jet multiplicities 2 b-tag multiplicities
11/2/2009 Ulrich Heintz - seminar - Stony Brook 20
event selectionb-jet taggingb lifetime 1.6 ps
travels a few mm before decaying
secondary
vertex
large impact parameter
primary vertex
Ulrich Heintz - seminar - Stony Brook 21
event selectionseparate b-jets from light-quark and gluon jets to
reject most W+jets backgroundneural network algorithm
based on impact parameter and reconstructed vertexleading b-jet pT > 20 GeVdefine two mutually exclusive samplesone tight tag (eb = 40%, ec = 9%, el = 0.4%)two loose tags (eb = 50%, ec = 14%, el = 1.5%)
11/2/2009
Ulrich Heintz - seminar - Stony Brook 22
signal and background modelssingle top quark production
modeled using SINGLETOPbased on COMPHEP reproduces NLO kinematic distributions
PYTHIA for hadronization
top-antitop pair productionmodeled using ALPGEN
parton-jet matching to avoid double-counting final statesPYTHIA for hadronizationnormalized to σ = 7.91pb
Kidonakis and Vogt, PRD 68, 114014 (2003) uncertainty +7.7% −12.7% (theory, pdf, mtop)
11/2/2009
Ulrich Heintz - seminar - Stony Brook 23
signal and background modelsW+jets production
modeled using ALPGEN + PYTHIA w/ matchingjet , , between leading jets corrected to match
dataZ+jets production
modeled using ALPGEN + PYTHIA Z+ heavy flavor corrected to theory, with ±14%
uncertaintydiboson production
modeled using PYTHIA Normalized to expected cross sections
11/2/2009
Ulrich Heintz - seminar - Stony Brook 24
signal and background modelsmultijet background
jets mimic e, from semileptonic b-decaysestimates data drivenkeep small with selection cuts( 5%)
11/2/2009
3/24/2009 Meenakshi Narain 24
Ulrich Heintz - seminar - Stony Brook 25
background normalizationbefore b-tagging
iterative fits to data in three variableslepton pT, MT, and missing pT
subject to constraint30% to 54% (multijet), 1.8% to 3.9% (W+jets)from max difference with 1-variable fit result
11/2/2009
multijetsjetsWbkgdata NNNN
Ulrich Heintz - seminar - Stony Brook 26
background normalizationafter b-tagging
W + heavy flavor normalized to theory (MCFM @ NLO)
• 1.47 (Wbb,Wcc), 1.38 (Wcj)empirical correction from two-jet data and simulation
• 0.95 ± 0.13 (Wbb, Wcc)
11/2/2009
event yield (before b-tagging)
11/2/2009 Ulrich Heintz - seminar - Stony Brook 27
expected signal
backgrounds
s:b 1:250
observed
acceptance: 3.70.5% (tb) 2.50.3% (tqb)
event yield (after b-tagging)
11/2/2009 Ulrich Heintz - seminar - Stony Brook 28
expected signal
backgrounds
s:b 1:20
observed
Ulrich Heintz - seminar - Stony Brook 29
Data/MC comparison(all channels combined)
11/2/2009
Ulrich Heintz - seminar - Stony Brook 30
signal and background models
11/2/2009
pre tag
1 b-tag
2 b-tags
2 jets 3 jets 4 jets
signal and background modelstest background model in regions dominated by
one type of background
GeV 175tag-b 1 jets, 2
,,
jetsplepton
TT
p GeV 300tag-b 1 jets, 4
,,
jetsplepton
TT
pW+jets: tt pairs:
11/2/2009 31Ulrich Heintz - seminar - Stony Brook
Ulrich Heintz - seminar - Stony Brook 32
outlinetop quark introductionTevatron and DØ experimentevent selectionmatrix elementsboosted decision treesbayesian neural networkscross section and |Vtb|other measurementssummary
11/2/2009
Ulrich Heintz - seminar - Stony Brook 33
matrix elementsmethod pioneered by DØ for top quark mass measurement use 4-vectors of all reconstructed leptons and jets use matrix elements of main signal and background processes compute a discriminant
define Psignal as a normalized differential cross section:
performed in 2-jets and 3-jets channels only split sample in high and low HT to improve performance (W+jets
and top quark pair dominated regions)
11/2/2009
matrix elements
11/2/2009 Ulrich Heintz - seminar - Stony Brook 34
2-jet channels
3-jet channels
Ulrich Heintz - seminar - Stony Brook 35
matrix elements2-jet channels
11/2/2009
tb discriminant
tqb discriminant
Ulrich Heintz - seminar - Stony Brook 36
matrix elementsstarting from 2dimensional s vs t-channel discriminant
rebin to ensure enough background events in each bin re-order bins according to highest-to-lowest signal:background
to obtain the 1dim tb+tqb discriminant split according to HT
11/2/2009
HT < 175 GeV HT > 175 GeV
Ulrich Heintz - seminar - Stony Brook 37
outlinetop quark introductionTevatron and DØ experimentevent selectionmatrix elementsboosted decision treesbayesian neural networkscross section and |Vtb|other measurementssummary
11/2/2009
11/2/2009 Ulrich Heintz - seminar - Stony Brook 38
boosted decision treesdecision trees
widely used in social sciences idea: recover events that fail a cut find cuts with best separation between
signal and background repeat recursively on each branch stop when no further improvement or when too few events left terminal node is called a “leaf” decision tree output = leaf purity
adaptive boosting technique to improve any weak classifier used with decision trees by GLAST and MiniBooNE train a tree increase weight of misclassified events train again average over 20 boosting cycles dilutes discrete nature of output and improves performance
boosted decision trees64 input variables
rank variables to select the 50 most sensitive variables for each channel
adding more variables does not degrade the performance
reducing the number of variables reduces the sensitivity of the analysis
use 1/3 of all signal and background events as training sample
train 24 treese,2,3,4 jets1,2 b-tags2 detector configurations
11/2/2009 Ulrich Heintz - seminar - Stony Brook 39
boosted decision treeskinematics angular correlationsjet characteristics
top reconstruction
11/2/2009 40Ulrich Heintz - seminar - Stony Brook
boosted decision treesvariables
11/2/2009 Ulrich Heintz - seminar - Stony Brook 41
Ulrich Heintz - seminar - Stony Brook 42
boosted decision treesapply transformation to discriminant to ensure sufficient
number of background events in each bin provides stability in the final cross section measurement
11/2/2009
Ulrich Heintz - seminar - Stony Brook 43
outlinetop quark introductionTevatron and DØ experimentevent selectionmatrix elementsboosted decision treesbayesian neural networkscross section and |Vtb|other measurementssummary
11/2/2009
Ulrich Heintz - seminar - Stony Brook 44
Bayesian neural networks neural networks are nonlinear functions
defined by weights associated with each node weights are determined by training on signal and
background samples Bayesian neural networks improve on this
average over many networks weighted by the probability of each network given the training samples
less prone to over-training network structure is less important – can use larger
numbers of variables and hidden nodes for this analysis:
18-28 input variables in each channel 20 hidden nodes 100 training iterations each iteration is the average of 20 training cycles backgrounds are combined for training
11/2/2009
Network output
tqb
Network output
Wbb
Bayesian neural networkslist of variables (example from one channel)
11/2/2009 Ulrich Heintz - seminar - Stony Brook 45
final discriminant after binning transformation similar to BDT
Ulrich Heintz - seminar - Stony Brook 46
outlinetop quark introductionTevatron and DØ experimentevent selectionmatrix elementsboosted decision treesbayesian neural networkscross section and |Vtb|other measurementssummary
11/2/2009
Ulrich Heintz - seminar - Stony Brook 47
cross section measurement verify that calculation methods work as expected using ensembles
of pseudo-experiments select subsets of events from total pool of MC events randomly sample a Poisson distribution to simulate statistical
fluctuations background yields fluctuated according to uncertainties to reproduce
correlations between components from normalization each pseudo-experiment simulates one DØ experiment
11/2/2009
Ulrich Heintz - seminar - Stony Brook 48
cross section measurementcheck discriminant in background dominated regions
W+jets: 2 jets, 1 b-tag, HT < 175 GeV
ttbar : 4 jets, 1-2 b-tags, HT > 300 GeV
11/2/2009
DØ
DØ
DØ
DØDØ
Ulrich Heintz - seminar - Stony Brook 49
cross section measurement
11/2/2009
cross section is given by posterior density peak with 68% interval as uncertainty
Ulrich Heintz - seminar - Stony Brook 50
cross section measurementbefore looking at the data
how well can we rule out the background-only hypothesis?fraction of the ensembles without single top signal that give a cross
section at least as large as the expected sm valueconvert p-value to “expected significance”
from the data how well do we rule out the background-only hypothesis?
fraction of the ensembles without single top signal that give a cross section at least as large as the observed value
convert p-value to “measured significance” what cross section do we measure? how consistent is the measured cross section with the SM?
fraction of the ensembles with SM-signal pseudo-datasets that give a cross section at least as large as the measured value to get “consistency with SM”
11/2/2009
systematic uncertainties
11/2/2009 51Ulrich Heintz - seminar - Stony Brook
final discriminant outputs
11/2/2009 52Ulrich Heintz - seminar - Stony Brook
cross section results
11/2/2009 53Ulrich Heintz - seminar - Stony Brook
correlations between methodseven though all analyses use the same data, they are not
100% correlated
11/2/2009 54Ulrich Heintz - seminar - Stony Brook
Ulrich Heintz - seminar - Stony Brook 55
combined DØ resultsuse BNN to combine the three methods
input variables are output discriminants of individual analyses
11/2/2009
observed cross section = 3.90.9 pb
observed(expected) significance = 5.0σ (4.5σ)
DØ, PRL 103 092001 (2009)
s- and t-channel cross sections
11/2/2009 Ulrich Heintz - seminar - Stony Brook 56
pb 14.3
pb 81.005.194.080.0
t
s
significance = 4.8
submitted to Phys. Lett. B arXiv.org:0907.4259
11/2/2009 Ulrich Heintz - seminar - Stony Brook 57
measurement of |Vtb| use the measurement of the single top cross section to make the
first direct measurement of |Vtb| calculate a posterior in |Vtb|2 ((tb, tqb) |Vtb|2) general form of Wtb vertex:
assume sm top quark decay : |Vtd|2 + |Vts|2 << |Vtb|2 pure V–A : f1
R = 0 CP conservation : f2
L= f2R = 0
do not assume three quark families CKM matrix unitarity
(unlike for measurements using tt decays)
measure the strength of the V–A coupling |Vtb f1L|, which can be > 1
Ulrich Heintz - seminar - Stony Brook 58
measurement of |Vtb|
11/2/2009
assuming f1L =1
Ulrich Heintz - seminar - Stony Brook 59
combined Tevatron resultsuse BNN to combine CDF and DØ results
11/2/2009
pb 76.2 58.047.0
07.088.0|| tbV
08.091.0|| tbV
single top productioncross section
for cross section byKidonakis (NLO+soft gluon)
Harris & Sullivan (NLO)
Ulrich Heintz - seminar - Stony Brook 60
outlinetop quark introductionTevatron and DØ experimentevent selectionmatrix elementsboosted decision treesbayesian neural networkscross section and |Vtb|other measurementssummary
11/2/2009
Ulrich Heintz - seminar - Stony Brook 61
Wtb couplingsgeneral Lagrangian
in standard model: f1L = 1, f1
R = f2L = f2
R = 0anomalous couplings can change
kinematicsangular distributionscross section
11/2/2009
coupling cross section s:t channelf1
L = 1 or f1R = 1 3 pb 1:2
f2L = 1 or f2
R = 1 10 pb 6:1
Boos, Dudko, Ohl, Eur. Phys. J. C11, 472 (1999)
Wtb couplings train boosted decision trees
on models with f1L and one
other coupling >0compute 2dim posterior first limits on tensor
couplings f2L and f2
R
11/2/2009 Ulrich Heintz - seminar - Stony Brook 62
sm expectation measurement
DØ, PRL 101 221801 (2008)
Wtb couplings helicity of W from top decay
= 0 70% = 1 30% = 1 0
angle between down-type fermion and and top quark in W rest frame ()
constrains only ratio of couplings
combine with W helicity
11/2/2009 Ulrich Heintz - seminar - Stony Brook 63
W helicity prior combined results
DØ, PRL 102 092002 (2009)
LongitudinalLeft-handed
Right-handed
search for W’tb
11/2/2009 Ulrich Heintz - seminar - Stony Brook 64
DØ, PRL 100 211803 (2008)
W’L W’R
search for H+tb
11/2/2009 Ulrich Heintz - seminar - Stony Brook 65
DØ, PRL 102 191802 (2009)
non minimal Higgs sector 2 weak isospin doublet fields (e.g. MSSM) 5 physical Higgs bosons: h0, H0, A0, H
if mH < mtop then t H b affects top quark branching fractions in top-antitop quark pair decays
if mH > mtop then H tb
Ulrich Heintz - seminar - Stony Brook 66
outlinetop quark introductionTevatron and DØ experimentevent selectionmatrix elementsboosted decision treesbayesian neural networkscross section and |Vtb|other measurementssummary
11/2/2009
summarysingle top production has been observed by the
DØ and CDF experimentsthe cross section is measured to be
consistent with the standard model implies that |Vtb| > 0.77 @ 95% CL
opens a new window to studying the top quark
11/2/2009 Ulrich Heintz - seminar - Stony Brook 67
pb 76.2 58.047.0
CKM matrix
4/16/2009 Ulrich Heintz 68
tbtstd
cbcscd
ubusud
VVVVVVVVV
tbV039.0008.0041.004.1230.0004.02255.09742.0
R
R
L
R
R
L
R
R
L
R
L
R
L
R
L
e
btbt
scsc
dudu
ee
quarksleptons
w
w
w
bsd
bsd
1222 tbcbub VVV 9991.0tbV
unitarity
CKM matrix
4/16/2009 Ulrich Heintz 69
'''''
'
'
'
039.0008.0041.004.1230.0004.02255.09742.0
btbtstdt
tbtb
cb
ub
VVVVVVVV
w
w
w
w
bbsd
'
'bbsd
12'
222 tbtbcbub VVVV ?tbV
unitarity
R
R
L
R
R
L
R
R
L
R
L
R
L
R
L
e
btbt
scsc
dudu
ee
quarksleptons
evidence versus observation
4/16/2009 Ulrich Heintz 70
68%
|x-x0| probability
>1 0.32>2 0.046>3 0.0027>4 6.3 x 10-5
>5 5.7 x 10-7
nothing to write home about
claim evidencestart a rumor, write to NY Times
it’s getting seriousdiscovery – collect Nobel Prize…
95.4%
99.7%
99.994%
99.99994%
0xx