Search for supersymmetry in events with photons and missing
transverse energy
Garyfallia (Lisa) Paspalaki
on behalf of CMS colaboration
N.C.S.R ’Demokritos’
NTUA
April 8, 2017
Lisa Paspalaki (NCSR ’Demokritos’, NTUA)
HEP 2017, Ioannina
April 8, 2017
1 / 20
Introduction
Analysis Overview
Search for General gauged mediated (GGM) sypersymmetry breaking in final state
involving photons. The data sample
corresponds to an integrated luminosity of 2.32 fb −1
√
of proton proton collisions at s = 13TeV was collected with the CMS detector at the
LHC in 2015.
GGM supersymmetry breaking can produce events with double photons, jets and
significant missing energy (ETmiss ).
q
q
P2
γ
ge
χe01
χe01
e
G
e
G
γ
ge
P1
q
q
We assume gluino pair production where the NLSP neutralino decays to a gravitino and
photon ( χ˜01 → G̃ γ ), resulting in characteristic events with jets, two photons and large
ETmiss
Lisa Paspalaki (NCSR ’Demokritos’, NTUA)
HEP 2017, Ioannina
April 8, 2017
2 / 20
Introduction
Backgrounds
Quantum Chromodynamics (QCD) background
Most significant background due to huge QCD cross section
Can have real photons in the final state or we can get electromagnetically-rich jet
fragmentation mimicking the response of a photon
ETmiss comes from mis-measured hadronic activity.
Electroweak (EWK) background
Includes W γ and W + jet events
W → eν and the electron is misidentified as a photon
W + jet events, one of the jets fakes a photon
Genuine ETmiss from the neutrino
Other backgrounds-small and studied with Monte Carlo (MC).
Z γγ → ννγγ
W γγ → lνγγ
t t̄γγ
Lisa Paspalaki (NCSR ’Demokritos’, NTUA)
HEP 2017, Ioannina
April 8, 2017
3 / 20
Trigger and Event Selection
Trigger
Primary analysis trigger:
HLT Diphoton30 18 R9Id OR IsoCaloId AND HE R9Id Mass95
Lead Photon PT > 30GeV and trail for photon PT > 18GeV
Mγγ > 95GeV
-1
CMS
Leading Efficiency
2.26 fb
CMS Sub-Leading
-1
1
0.8
0.8
0.6
0.6
0.6
Prob
2
2.26 fb
CMS Total
-1
1
0.8
0.4
Efficiency
Efficiency
1.28 fb
Efficiency
Leading Efficiency
Preliminary
Efficiency
Efficiency
CMS
1
Trigger Efficiency
2.26 fb
-1
1
0.8
0.6
χ / ndf 87.09 / 91
0
0.4
0.4
0.4
2
χ / ndf 93.19 / 91
0.2
0
0
0.2
10
20
30
40
50
60
70
80
Leading Photon P (GeV)
0
Prob 1.279e-05
0.2
20
T
40
60
80
100
120
Leading Photon P (GeV)
0
0.2
20
40
T
60
80
100
120
Leading Photon P (GeV)
T
0
80
85
90
95
100
105
110
115
120
Diphoton Invariant Mass P (GeV)
T
HLT efficiency is calculated by a tag and probe method and is 98.6% for photon
PT > 40GeV
Trigger requires two photons passing the sub-leading filter and one photon passing the
leading filter, so that the total efficiency tot = lead,lead x lead,sub x sub,sub
Lisa Paspalaki (NCSR ’Demokritos’, NTUA)
HEP 2017, Ioannina
April 8, 2017
4 / 20
Event Selection
Object Selection
Muons
Photons
PT > 30GeV
From PF photon Collection
|η| < 1.4442
PT > 40GeV
Passes medium muon ID
|η| < 1.4442
Passes loose muon isolation
Passes medium photon ID
Passes Pixel seed veto
Electrons
Jets
From PF photon collection
PT > 40GeV
PT > 30GeV
|η| < 1.4442
|η| < 2.4
Passes medium photon ID
Passes PF Loose ID
Fails pixel seed veto
Remove jets that overlap within
∆R < 0.4 of a muon, electron, or a
photon
Remove electrons that overlap within
∆R < 0.4 of a muon
Lisa Paspalaki (NCSR ’Demokritos’, NTUA)
HEP 2017, Ioannina
April 8, 2017
5 / 20
Event Selection
Fake Selection
‘Fake Photons‘ are photons that fail isolation or shape requirements.
Primarily composed of electromagnetically rich jets reconstructed as photons
Control Sample with fakes are used to model the QCD background
Make Fakes orthogonal to Photons by inverting charged hadron isolation or σiηiη
requirements of the medium photon ID.
( 1.33GeV < charged hadron isolation < 15GeV XOR 0.0102 < σiηiη < 0.0150 )
Events are sourted int four categories depending on the selection of their hightest-pT
electromagnetic objects:
γγ
ee
ff
eγ
Lisa Paspalaki (NCSR ’Demokritos’, NTUA)
HEP 2017, Ioannina
April 8, 2017
6 / 20
Event Selection
Control and Candidate Sample
Double electron Sample (ee)
Used to model EM objects in the QCD background
Require 75GeV < mee < 105GeV to collect Z → ee events
Require ∆R > 0.3 between electrons
Double fake sample (ff)
Passes the primary trigger
Used to model EM objects in the QCD background
require mff > 105GeV
require ∆R > 0.3 between fakes.
Candidate DiPhoton Sample (γγ)
passes primary trigger
Require mγγ > 105GeV
∆R > 0.3 between photons
Signal Region is ETmiss > 100GeV
Lisa Paspalaki (NCSR ’Demokritos’, NTUA)
HEP 2017, Ioannina
April 8, 2017
7 / 20
QCD Background Estimation
QCD Background Estimation - Backgrounds without true ETmiss
Strategy
Processes that lack from genuine ETmiss , but can emulate GGM signal topologies if
the hadronic activities in the event are poorly measured.
Used double electrons and double fakes control samples to estimate the ETmiss
distribution of QCD backgrounds.
But this samples have different amounts of hadronic activity than the candidate γγ
sample.
2.3 fb-1 (13 TeV)
Events/GeV
CMS Preliminary
γγ
10−1
ee
ff
10−2
10−3
Ratio
10−4
3
2
γ γ /ee
γ γ /ff
1
0
0
50
100
150
200
250
300
350
400
di-EM p (GeV)
T
Model the hadronic recoil of the event with the di-EM pT of the event, where di-EM
pT is the vector sum of the PT of the two electromagnetic object.
Reweight the control samples to correct for the differences in hadronic activity.
Lisa Paspalaki (NCSR ’Demokritos’, NTUA)
HEP 2017, Ioannina
April 8, 2017
8 / 20
QCD Background Estimation
Reweight ETmiss backgrounds
Unweighted (left) and di-EM pT reweighted (right) ETmiss distributions of the ee and ff
samples
2.3 fb-1 (13 TeV)
CMS Preliminary
103
Events/GeV
Events/GeV
103
102
102
10
10
ee
1
10−1
10−1
ff
2
ee/ff
2
ee/ff
10−2
1
0
10
20
30
reweighted ee
1
10−2
0
2.3 fb-1 (13 TeV)
CMS Preliminary
40
50
60
70
80
90
100
Emiss
(GeV)
T
reweighted ff
1
0
0
10
20
30
40
50
60
70
80
90
100
Emiss
(GeV)
T
Samples are normalized to ETmiss < 50 GeV of the γγ (signal contamination < 1% )
Lisa Paspalaki (NCSR ’Demokritos’, NTUA)
HEP 2017, Ioannina
April 8, 2017
9 / 20
QCD Background Estimation
Comparison to Candidate γγ Sample
Comparing the candidate ETmiss distribution to the distributions for the ee (left) and ff
(right) control samples.
2.3 fb-1 (13 TeV)
CMS Preliminary
103
Events/GeV
Events/GeV
103
102
102
10
10
γγ
1
2
γ γ /ff
2
γ γ /ee
10−2
1
10
20
30
40
50
60
reweighted ff
10−1
10−2
0
γγ
1
reweighted ee
10−1
0
2.3 fb-1 (13 TeV)
CMS Preliminary
70
80
90
100
Emiss
(GeV)
T
1
0
0
10
20
30
40
50
60
70
80
90
100
Emiss
(GeV)
T
Samples are normalized to ETmiss < 50 GeV of the γγ
Lisa Paspalaki (NCSR ’Demokritos’, NTUA)
HEP 2017, Ioannina
April 8, 2017
10 / 20
QCD Background Estimation
Estimation of pure QCD background
The primary background estimate comes from the double electron sample.
We use the prediction from double fake as a systematic uncertainty on the estimate
from the ee sample.
Low statistics from ff sample in ETmiss >100 GeV signal region.
2.3 fb-1 (13 TeV)
CMS Preliminary
10−1
Events/GeV
Events/GeV
10−1
ff
loose ff
10−2
ff
loose ff
10−2
10−3
10−4
10−4
10−5
10−5
ff/loose ff
10−3
ff/loose ff
2.3 fb-1 (13 TeV)
CMS Preliminary
2
1
0
10
20
30
40
50
60
70
80
90
100
2
1
0
Emiss
(GeV)
T
50
100
150
200
250
300
Emiss
(GeV)
T
Tight fake and loose fake samples match well in ETmiss < 100 GeV region.
Loose fake samples can be used to model the shape in high ETmiss region.
Determine ff sample in signal region from looser fake definition.
Lisa Paspalaki (NCSR ’Demokritos’, NTUA)
HEP 2017, Ioannina
April 8, 2017
11 / 20
QCD Background Estimation
Systematic Uncertainty from Shape Difference
Fit reweighted ee and loose ff samples for 70 GeV < ETmiss < 300 GeV with a
function χp0 exp(p1χp2 ) with fit parameters p0, p1, p2
Integrate fits in each ETmiss bin for ETmiss > 100 GeV.
Difference between ee and loose ff result gives shape uncertainties in that bin.
Largest uncertainty in the last bin due to the large bin width.
-1
2.3 fb (13 TeV) χ2 / ndf
Prob
p0
102
fitting function
pow(x, p0) * Exp(p1 * pow(x, p2))
0.6745 / 4
0.9544
−3.542 ± 0.238
p1
25.26 ± 4.93
p2
−0.1149 ± 0.0067
reweighted ee
10
CMS Preliminary
Events/GeV
Events/GeV
CMS Preliminary
103
102
1
10− 1
10− 1
−2
10− 2
10− 30
50
100
150
200
250
300
10− 30
Emiss
(GeV)
T
Lisa Paspalaki (NCSR ’Demokritos’, NTUA)
fitting function
pow(x, p0) * Exp(p1 * pow(x, p2))
HEP 2017, Ioannina
1.081 / 4
Prob
p0
p1
− 1.001 ± 0.074
6339 ± 1252.2
0.8972
p2
− 1.636 ± 0.049
reweighted loose ff
10
1
10
-1
2.3 fb (13 TeV) χ2 / ndf
103
50
100
150
200
250
300
Emiss
(GeV)
T
April 8, 2017
12 / 20
QCD Background Estimation
Final QCD Estimate
ETmiss (GeV)
100-110
110-120
120-140
> 140
Bkg Prediction
1.85 ± 0.96
1.53 ± 0.63
0.97 ± 0.62
0.61 ± 2.15
ETmiss (GeV)
100-110
110-120
Di-EMpT reweighting
uncertainty comes from
propagating toy Di-EMpT
plots to the reweighted
ETmiss distribution.
Additional systematic
comes from the differences
between reweighting with
the di-EM pT only and
reweighting with di-EM pT
and jet multiplicity.
Lisa Paspalaki (NCSR ’Demokritos’, NTUA)
120-140
> 140
Systematic Uncertainty
Di-EM pT reweighting
Jet multiplicity reweighting
Shape difference between ee and ff
Statistical uncertainty of ee sample
Di-EM pT reweighting
Jet multiplicity reweighting
Shape difference between ee and ff
Statistical uncertainty of ee sample
Di-EM pT reweighting
Jet multiplicity reweighting
Shape difference between ee and ff
Statistical uncertainty of ee sample
Di-EM pT reweighting
Jet multiplicity reweighting
Shape difference between ee and ff
Statistical uncertainty of ee sample
HEP 2017, Ioannina
Value (%)
15.11
33.77
18.18
30.81
16.60
14.87
12.07
33.33
33.31
29.39
14.40
41.75
39.75
20.34
150.36
70.98
April 8, 2017
13 / 20
EWK Background Estimate: Fake Rate
EWK background in the signal region (ETmiss > 100GeV ) comes mainly from W γ → eνγ
where the electron is misidentified as a photon.
Calculate the ratio of the electrons faking photons, fe→γ using the Z → ee invariant
mass peak in both an ee sample and an eγ sample.
Events/GeV
To get the final EWK background estimate, we weight the eγETmiss distribution by
fe→γ /(1 − fe→γ ) to get the number of eγ events that sneak into the candidate γγ
sample.
CMS Preliminary
2.3 fb-1 (13 TeV)
uncertainty from fake rate
1
Table: Estimation of total EWK background
for ETmiss > 100 GeV
total uncertainty
ETmiss bin (GeV )
100 − 110
110 − 120
120 − 140
140 − Inf
10−1
10−2
0
50
100
150
Lisa Paspalaki (NCSR ’Demokritos’, NTUA)
200
Expected
0.11 ± 0.09
0.07 ± 0.06
0.14 ± 0.11
0.27 ± 0.22
250
300
Emiss
(GeV)
T
HEP 2017, Ioannina
April 8, 2017
14 / 20
Results
Observed vs Expected
ETmiss (GeV)
100-110
110-120
120-140
> 140
Observed
4
2
2
1
2.3 fb-1 (13 TeV)
CMS Preliminary
103
Events/GeV
Expected
2.26 ± 0.96
1.79 ± 0.64
1.51 ± 0.64
1.64 ± 2.16
Data
QCD
EWK
combined uncertainty
T5gg, M~ = 1.4 TeV, M 0 = 0.6 TeV
χ
g
T5gg, M~ = 1.6 TeV, M 20 = 0.6 TeV
102
10
g
χ
2
1
10−1
10−2
10−3
data/bkg
0
50
100
150
200
50
100
150
200
2
250
300
Emiss
(GeV)
T
1
0
Lisa Paspalaki (NCSR ’Demokritos’, NTUA)
HEP 2017, Ioannina
250miss
300
ET (GeV)
April 8, 2017
15 / 20
Results
Limits
CMS Preliminary
An expected exclusion reach for the
analysis was done using the modified
frequentist CLs methods
2.3 fb-1 (13 TeV)
Mχ∼0 (GeV)
~
0
0
pp → ~
g~
g, ~
g → qq∼
χ ,∼
χ → γG
1
1400
95% CL limit on σ (pb)
1
1
1600 NLO + NLL exclusion
Obs. limit
Obs. limit ± 1 σth
Exp. limit
Exp. limit ± 1 σex
1200
10−2
1000
800
600
400
200
1000
1200
1400
1600
1800
M~g (GeV)
Lisa Paspalaki (NCSR ’Demokritos’, NTUA)
This is based on long-likelihood test
statistic that compares the likelihood of
the SM-only hypothesis to the
likelihood of the presence of signal in
addition to the SM conditions
The likelihood functions are based on
the expected shape of the ETmiss
distribution for signal and background
in four separate bins.
For typical values of neutralino mass,
we expect to exclude gluino masses out
of 1.5 TeV, improving the reach of
previous searches performed at
center-of-mass energies of 8 TeV.
HEP 2017, Ioannina
April 8, 2017
16 / 20
Conclusions
Conclusions
Full analysis has been done on 13 TeV data
Even with 2.3fb −1 of data, we are able to extend our reach from 8TeV analysis
No evidence of SUSY seen, but we expect to probe even higher mass scales this year
with a larger dataset.
Lisa Paspalaki (NCSR ’Demokritos’, NTUA)
HEP 2017, Ioannina
April 8, 2017
17 / 20
Back Up
Back Up
Lisa Paspalaki (NCSR ’Demokritos’, NTUA)
HEP 2017, Ioannina
April 8, 2017
18 / 20
Back Up
Jet multiplicity Reweighting
Events/GeV
The difference in jet multiplicity between the candidate sample and the control sample
can also affect the overall ETmiss resolution of the background estimation .
The jet multiplicity distribution for the candidate and the ff samples are similar, but the
candidate and ee are different.
To investigate the effect of jet multiplicity reweighting, we plotted the eeETmiss ditribution
reweighting by the jet multiplicity in bins of di-EMpT .
103
2.3 fb-1 (13 TeV)
CMS Preliminary
ee, reweighted by njet in bins of Di-EM p
T
102
ee, reweighted by Di-EM p
T
10
1
10−1
ee(1)
ee(2)
10−2
2
1
0
0
50
100
150
200
250
300
Emiss
(GeV)
T
The difference bewtween them is small.
Choose not to reweight by the jet multiplicity, but we take the difeerence as a systematic
uncertainty.
Lisa Paspalaki (NCSR ’Demokritos’, NTUA)
HEP 2017, Ioannina
April 8, 2017
19 / 20
Back Up
Contributions to control samples
There is a small Contribution to the ee control sample from t t̄ events and the
contribution to the ff sample from Z → ν ν̄
t t̄ events will contribute 17.27 ± 0.98. and Z → ν ν̄ contribution is almost negligible.
To get rid of the contamination we substract the shape of t t̄ from our ee control sample.
Lisa Paspalaki (NCSR ’Demokritos’, NTUA)
HEP 2017, Ioannina
April 8, 2017
20 / 20