W+jets - I. Physikalischen Instituts B - RWTH

Semileptonic tt decays
with 0.1/fb
Stefan Kasselmann
III. Physikalisches Institut B, RWTH Aachen
Stefan Kasselmann
Bad Honnef, August 2006
RWTH Aachen, III. Phys. Inst. B
LHC/CMS schedule
• CMS closes at 31/08/07
• First beam: November 2007:
- 0.9 TeV (CM)
- 43 vs. 43 bunches
- 1028-1030 cm-2s-1
• Debugging machine/detector
• Then: Commissioning of all
8 sectors for full energy in
winter 2008 shutdown
• First Physics: Spring 2008
- 14 TeV
- 156 vs. 156 b.
- 1032 cm-2s-1
• 0.1/fb : “A few weeks of data
taking"
Picture of the Tracker Inner/
Outer Barrel from July 2006
Stefan Kasselmann
Bad Honnef, August 2006
http://lhc-commissioning.web.cern.ch
RWTH Aachen, III. Phys. Inst. B
1/19
“1st physics run” scenario
• 0.1/fb corresponds to about 48.800 ttbar (inclusive) events,
taking the LO cross section from PYTHIA (CTEQ 5L)
(http://cmsdoc.cern.ch/cms/PRS/gentools/www/xsec/cmsxsec.html)
• This analysis is (so far) based on the following assumptions:
• no pixel detector  No b jet tagging were used!
• no ECAL endcaps  Electron identification only in || < 1.47
 No cut on MET used
• Ideal: Use new MC data (CMSSW) in this specific detector configuration:
• Tracking algorithms work different without pixel detector (seeds)!
• Less material in front of silicon detector (particle interactions)
• …
• So far: Data (Pythia) from 2004 used (with pixel). Ongoing: Converting data files
(ALPGEN) which better simulates gluon radiation processes
• Goal: Develop analysis to "see" tops in this scenario (e.g. invar. mass spectrum)
Stefan Kasselmann
Bad Honnef, August 2006
RWTH Aachen, III. Phys. Inst. B
2/19
CMS detector
y
Electromagnetic
calorimeter
(ECAL)
x
Myon chambers


Silicon Tracker
(Pixel+Strip)
z
Hadronic
calorimeter
(HCAL)
Superconductive coil
(4 Tesla)
Stefan Kasselmann
Bad Honnef, August 2006
Forward calorimeter
RWTH Aachen, III. Phys. Inst. B
3/19
top pairs @ LHC
• tt production: produced via two
processes (strong interaction):
87%: gg
 tt
13%: qq  tt
• 10 top pairs/s @ 1034 cm-2s-1
• But: About 20 pile up events!
• Main background: W+jets, Z+jets,
Dileptonic ttbar decay
Stefan Kasselmann
Bad Honnef, August 2006
RWTH Aachen, III. Phys. Inst. B
4/19
Top pair decay
BR(tt  bW+bW-)
2/3
~ 100%
BR(W+W-  l11 l22) ~ 11% (9/81)
BR(W+W-  q1q2q3q4) ~ 44% (36/81)
BR (W+W-  q1q2l) ~ 44% (36/81)
Stefan Kasselmann
1/3
Bad Honnef, August 2006
W+  u , d / c , s (3 colours)
W-  u , d / c , s (3 colours)
RWTH Aachen, III. Phys. Inst. B
5/19
Lepton identification
• Only electrons (pT > 10 GeV, || < 1.47) and muons (pT > 10 GeV, || < 2.4) are used
Electrons: Likelihood based selection of electrons from candidates
Muons: Are taken as they come out of the GlobalMuonReconstructor
• Lepton isolation consists of calorimeter and tracker isolation
For both a cone of R = srqt(2 + 2) = 0.2 is used around the track of the particle
Calorimeter isolation: No energy deposits > 15% of lepton energy
Tracker isolation: No tracks > 10% of lepton momentum
Some of the input variables for electron likelihood:
• E / P: super cluster energy / track momentum
•
•
•
•
(for electrons close to 1)
H / E: energy in HCAL (behind super cluster) /
super cluster energy (for electrons close to 0)
 = | SC - track | : Difference between super
cluster position and extr. track pos. at ECAL
E9 / E25 : ECAL energy 3x3 cell / 5x5 cell
…
Stefan Kasselmann
Bad Honnef, August 2006
R = 0.2
RWTH Aachen, III. Phys. Inst. B
6/19
Generic preselection
Typical preselection:
• L1 & HLT Trigger
• 4 jets with pT > 10 GeV, || < 2.5 (low pT cut to be able to run different scenarios)
• At least one (tracker & calo) isolated lepton with pT > 10 GeV
Dataset
Events 0.1/fb
L1
HLT
4 jets
lepton
eff. (% )
Signal:
semilept. (e/mu)
Background:
semilept. (tau)
dileptonic
fully hadronic
W + jets
Z + jets
QCD
WW + jets
ZW + jets
ZZ + jets
14.331
12.907
9.997
9.987
5.798
40,46
7.176
5.094
22.199
4.871
4.593
12.684
2.303
3.767
6.185
2.302
3.756
6.185
358
2.622
13
4,99
51,47
0,06
34.823.250 11.274.140
1.105.530
296.842
230.938.958.000 481.056.332
7.002
3.343
2.700
8.056
1.100
306
9.539.255
200.770
17.475.944
2.124
446
153
7.275.019
188.370
17.341.508
2.086
432
150
4.736.267
128.650
797.681
1.066
256
86
13,60
11,64
3,5E-04
15,22
9,48
7,82
Signal
Background (scaled)
14.331
2,31E+11
12.907
4,93E+08
9.997
2,72E+07
9.987
2,48E+07
5.798
5,67E+06
S/B
6,20E-08
2,62E-05
3,67E-04
4,02E-04
1/1000
Stefan Kasselmann
Bad Honnef, August 2006
RWTH Aachen, III. Phys. Inst. B
7/19
Selection
• First selection cut:
Exactly one lepton
• Less efficient cuts (not used):
- Two leptons with diff. charge
- Two lepton mass (Z peak)
W+jets
• The „one lepton cut“ is most
efficient against Z+jets and
dileptonic TTbar events
Z+jets
dileptonic
• Z+jets suppression:
35%
• Dileptonic suppression: 23%
• Signal loss:
< 1‰
logarithmic scale!
• In about 98.1% of the selected semileptonic events the lepton taken is the one
from W decay (that means it matches the MC signal lepton with R < 0.01 and
has correct charge)
Stefan Kasselmann
Bad Honnef, August 2006
RWTH Aachen, III. Phys. Inst. B
8/19
Selection
• Second selection cut:
3rd jet pT > 45 GeV
• Tried many cut variations on
the (pT sorted) four highest pt
jets
• Most efficient against W+jets
and dileptonic ttbar events
(which only have two high
energetic b jets from hard
interaction)
W+jets
dileptonic
• Dileptonic suppression: 67%
W+jets suppression: 99,9%
Signal loss:
40%
• In addition all other jets (4th, 5th) in the event must fullfill: pT > 30 GeV to
reduce the jet combinations for the Jet Parton Matching (JPM)
Stefan Kasselmann
Bad Honnef, August 2006
RWTH Aachen, III. Phys. Inst. B
9/19
Selection
• Third selection cut:
Circularity > 0.3
• This variable has small values
for planar events and high
values for circular events.
 (pxy nˆ )
C  2 min
 (p )
i
T
2
i
i 2
T
i
• This cut is most efficient
against QCD events
• QCD suppression:
99%
W+jets suppression: 45%
Signal loss:
40%
l
• (But: Low statistics of QCD!)
• Result: After these three selection cuts one gets an S/B of about 0.9
• Now one has to find the three jets from top out of 4 or 5 jets. Therefore a
likelihood was developed.
Stefan Kasselmann
Bad Honnef, August 2006
RWTH Aachen, III. Phys. Inst. B
10/19
Selection overview
• 4 or 5 jets with pT > 30 GeV, 3rd jet pT > 45 GeV (pT sorted)
• Exactly one (tracker & calorimeter) isolated lepton with pT > 10 GeV
• Circularity > 0.3
Dataset
Preselected
Events
4 or 5 jets
Exactly
pt > 30 GeV one lepton
3rd pt jet >
45 GeV
circularity >
0.3
eff. (% )
preliminary
Signal:
semilept. (e/mu)
5.798
2.743
2.656
2.122
1.330
22,94
358
2.622
13
170
620
6
164
448
6
134
319
5
84
199
2
23,46
17,00
15,38
W + jets
Z + jets
QCD
WW + jets
ZW + jets
ZZ + jets
4.736.267
128.650
797.681
1.066
256
86
6.730
1.261
1.982
86
10
8
6.705
869
1.956
83
8
6
1.482
327
1.913
56
4
4
962
204
2
23
3
2
0,02
0,16
0,00
2,16
1,17
2,33
Signal
Background (scaled)
5.798
5.666.999
2.743
10.873
2.656
10.245
2122
4.244
1.330
1.481
0,0010
0,2523
0,2592
0,5000
0,90
Background:
semilept. (tau)
dileptonic
fully hadronic
S/B
Stefan Kasselmann
Bad Honnef, August 2006
RWTH Aachen, III. Phys. Inst. B
11/19
Jet Parton Matching
• For early top physics JPM, I use six (simple) variables which distinguish between
right and wrong jet pairings, namely angles, masses and pT of jets.
• JPM criteria (All 4-jet-combinations out of 4 or 5 jets are used)
The sum of R(jet, parton) of all 4-jet-comb. is calculated, the lowest taken
( -> best global matching)
Each jet then must fulfill: R(jet, parton) < 0.25 and
|PTMC – PTRec| / PTMC < 0.5
( -> definition of matching jet)
The top candidate itself must fulfill: R(Rec. top, MC top) < 0.25
( -> reject badly reconstructed events)
The selected lepton must fulfill: R(Rec. lep., MC lep.) < 0.01
( -> the right lepton must have been found)
• The permutation that fulfills all these requirements for 4 jets is declared as true
jet pairing (black curves). All others are filled as wrong pairings (red curves).
The normalized distributions are used as probability density functions (PDFs)
Stefan Kasselmann
Bad Honnef, August 2006
RWTH Aachen, III. Phys. Inst. B
12/19
JPM PDFs
• Mass of 2-jet-permutations:
True combinations: Both jets from W
False combinations: All other permutations
(Right combinations of two jets
peak at W mass)
• Angle between 2-jet-permutations:
True combinations: Both jets from W
False combinations: All other permutations
(The jets from a W tend to have a smaller
angle)
Stefan Kasselmann
Bad Honnef, August 2006
RWTH Aachen, III. Phys. Inst. B
13/19
JPM PDFs
• Angle sum of 3-jet-permutations:
True combinations: All jets from had. top
False combinations: All other permutations
(Right combinations of three jets tend to
have a smaller angle sum due to boost)
•  between top and anti top:
True combinations: 3 jets from top /
one jet/lepton from other top
False combinations: All other permutations
of 3 to 1 jet+lepton
(Right combinations tend to be antiparallel
in )
Stefan Kasselmann
Bad Honnef, August 2006
RWTH Aachen, III. Phys. Inst. B
14/19
JPM PDFs
• Angle between lepton and b jet:
True combinations: b jet lep. side / lepton
False combinations: All other permutations
(Right comb. of lepton and b jet tend to
have a smaller angle due to boost)
• pT sum of 2-jet-permutations:
True combinations: Both jets are b jets
False combinations: All other permutations
(The b jets tend to have a tiny higher
transverse momentum than other jets)
Stefan Kasselmann
Bad Honnef, August 2006
RWTH Aachen, III. Phys. Inst. B
15/19
JPM (likelihood cut)
• Final selection cut:
Likelihood > 0.85
• This cut is mainly to have a
good probability to choose
the right three jets (from top)
• This cut obviously also
reduces much of the
remaining background
• After the final selection one gets an S/B of about 6
• For the following top mass plots only events with a LR of more than 0.85 are
taken (207 semileptonic events remain).
Stefan Kasselmann
Bad Honnef, August 2006
RWTH Aachen, III. Phys. Inst. B
16/19
Top signal 0.1/fb
with in situ cal.
(Not stacked)
• Top signal clearly visible!
• But high combinatorial background:
In about 50% the correct W was found
In about 35% the correct top was found
w/o in situ cal.
• Problem: Higher purity needs higher cut
on JPM likelihood, but too less statistics!
Stefan Kasselmann
Bad Honnef, August 2006
RWTH Aachen, III. Phys. Inst. B
17/19
Outlook- Use ANN?
• Artificial Neural Networks (ANN) uses correlations between input variables!
But: Need three times more MC (training and validation)
• First look at different network topologies (1/2/3 hidden layers and different
number of perceptrons) using SNNS http://www-ra.informatik.uni-tuebingen.de/SNNS/
• Can an ANN improve the JPM (likelihood) efficiency? -> Studies ongoing…
• As an example:
Training (black) and
validation (red) of an ANN:
Two important issues:
1.) For each net take configuration
with minimum of validation error
2.) Of all nets take the one with
the smallest validation error
Use net parameters at this point of training
(empirically search for best net)
Stefan Kasselmann
Bad Honnef, August 2006
RWTH Aachen, III. Phys. Inst. B
18/19
Summary
• The top quark can clearly be identified with
0.1/fb of data (within the „1st physics run“ )
which can be collected in a couple of weeks
(1032 cm-2s-1) without using any b tagging
• Background can almost be eliminated using
lepton isolation, jet pt, event shape variables
like circularity and the JPM likelihood
SIM
• A final S/B of about 6 was achieved with
the use of a likelihood
Real: MTCC
• Remaining problem so far: Combinatorical
background is high (Can an ANN help?)
http://www.physik.rwthaachen.de/~cmsmgr/analysis/
Stefan Kasselmann
Bad Honnef, August 2006
RWTH Aachen, III. Phys. Inst. B
19/19