Search for di-Higgs resonances decaying to 4 b-jets onCMS at 13 TeV
F. Nechansky1, Supervisor Caterina Vernieri1;
Consultants: Silvio Donato3, Jacobo Konigsberg4;Additional members of analysis team: Souvik Das4, Andrea Rizzi2
1FNAL 2U Pisa & SNS 3U Zurich 4U Florida
9/22/2016
Final report presentation
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 1 / 45
Outline
I Motivation
I Trigger efficiency (data, tt)
I Preselection
I Signal Region
I Regression
I Background
I Limits
I Summary
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 2 / 45
Motivation
I 2012 Higgs discovery completes theStandard modelmH = 125.7± 0.4 GeV
I SM incomplete, does not explain e.g.:I Neutrino massI Dark matterI Dark energyI Gravity
I Necessary to go beyond standard model
I Many theories with various predictions and freeparameters - e.g. SuperSymmetry
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 3 / 45
Experimental Point of View
I New particle Higgs Boson- in which ways it can be produced?
I Many modern BSM theories- which one is true?
I Some theories predict particle (X )decaying to pair of Higgs bosons(Randall-Sundrum radion,massive KK graviton[1], 2HDM[2])
I Possible channels of Higgs decay, e.g.:I H → γγ, BR: 0.23%, high res. (1-2%)I H → bb, BR: 57.7%, low res. (≈ 10%)
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 4 / 45
Current search
I This presentation reports on channel X → HH → bbbb:
I Best sensitivity for X mass mX > 400 GeV
I Done on CMS experiment at CERN for√
s = 13TeVusing data from 2016 (currently 9.3 fb−1)
I Similar study was done at 8 TeV ( arXiv:1503.04114 )and at 13 TeV ( 2015 data, CMS-PAS-HIG-16-002 )
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 5 / 45
Analysis work-flow
1. Four jet events dominated by e.g. multi-jet background, neccessary to selectonly events with b-jets =⇒ b-tagging
2. Triggers not modelled perfectly =⇒ study of trigger behaviour
3. After selection of four b-jets need to check if they originate from Higgs decay
4. Corrections for imperfection of detector on jet energy/momentum
5. Subtraction of 4b multi jet background
6. Estimation of cross-section for new processes
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 6 / 45
Compact Muon Solenoid (CMS)
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 7 / 45
Identification of b-jets (b-tagging)
I b-quark hadronize to b-hadronswith relatively large lifetime
I Identification using secondary vertex(few mm from PV)
I Tracks often have none-zero andpositive impact parameter
I Multivariate discriminant exploitthis information to identify b-jets:e.g. CSV (online) and CMVA(offline)
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 8 / 45
CMVA performance
I Performance of discriminants characterized by:I b-jet efficiency - fraction of b-jets correctly identifiedI misidentification prob. - probability of identifying non-b-jet as b-jet
I Performance of CMVA as function of jet pT :
I Medium working point used
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 9 / 45
Trigger efficiency
Triggers
I LHC produces large amount of collisions (≈MHz), impossible to record all
I Preference for interesting events =⇒ implementation of triggers
I Triggers decide based on portion event information to keep the event or not
I Search for 4b resonance =⇒ necessary reduction of background (multi jet)=⇒ online b-tagging
Quad Jet trigger (QJ):HLT BIT HLT QuadJet45 TripleBTagCSV p087 v
I L1 jet activity
I 4 jets |η| < 2.6, pT > 45 GeV(Calorimeter and Particle flow level)
I three b-tagged jets
Double Jet trigger (DJ):HLT BIT HLT DoubleJet90 Double30 TripleBTagCSV p087 v
I L1 jet activity
I 4 jets |η| < 2.6, pT > 30 GeV
I 2 jets |η| < 2.6, pT > 90 GeV(Calorimeter and Particle flow level)
I three b-tagged jets
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 11 / 45
Trigger efficiency
I Not all events detected by the trigger - necessary to derive correction
I Complicated triggers - usage of data driven technique
I Consider trigger, that requires event to pass specific selections A,B,C ,D...,then it can be shows that efficiency can be rewritten:
P(A&B&C &...) = P(A) · P(B|A) · P(C |A&B) · · · ·
I Trigger divided in stages, each studied as function of some relavant variable
I e.g. Quad Jet trigger:I L1 as function of sum of pT of four leading jets (
∑4 pT )I Four Calorimeter-jet selection as function of pT of the fourth jet (pT ,4)I Three b-tagged jets as function of discriminant of the third jet (CSV3)I Four Particle-flow-jets selection as function of pT of the fourth jet (pT ,4)
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 12 / 45
I Efficiency of each step estimated from data (turn-on function)
I Trigger efficiency can be written as:
Eff (4∑
pT , pT ,4,CSV3) = Turn on L1(4∑
pT ) · Turn on Calo pt4(pT ,4)·
·Turn on btag(CSV3) · Turn on PF pt4(pT ,4)
I Preselection designed to resemble final selectionI Preselection (orthogoal) trigger HLT BIT HLT IsoMu24 vI diHiggs cut - two pairs of b-jets compatible with Higgs massI Only considering jets with |η| < 2.6, CMVA >0.185 and PUId >= 4
Turn-on:
Calot pt4:
4Tp
40 50 60 70 80 90 100 110
Effi
cien
cy
0
0.2
0.4
0.6
0.8
1
CSV3:
3CSV0.8 0.85 0.9 0.95 1
Effi
cien
cy
0
0.2
0.4
0.6
0.8
1
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 13 / 45
Validation
I It is necessary to validate derived efficiencies (closure tests)
I Done for events after preselection
I Compare distributions of events weighted by the efficiency and of eventswhich pass the studied trigger
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
20
40
60
80
100
120
140
160
180
200
220
QuadJet Trigger Closure
TriggeredWeightedInternal CMS
QuadJet Trigger Closure
3CSV0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Rat
io
0.6
0.8
1
1.2
1.4
CSV3
0 20 40 60 80 100 120 140 160 180 2000
20
40
60
80
100
120
140
160
QuadJet Trigger Closure
TriggeredWeightedInternal CMS
QuadJet Trigger Closure
4T
p0 20 40 60 80 100 120 140 160 180 200
Rat
io
0.6
0.8
1
1.2
1.4
pT ,4
4− 3− 2− 1− 0 1 2 3 40
20
40
60
80
100
120
QuadJet Trigger Closure
TriggeredWeightedInternal CMS
QuadJet Trigger Closure
1η4− 3− 2− 1− 0 1 2 3 4
Rat
io
0.6
0.8
1
1.2
1.4
η1
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 14 / 45
Comparison between data and signal MC efficiency
I Done for events after analysis preselection and diHiggs selection
I Comparison between events weighted by data driven efficiencyand event passing the triggers
Runs B-F
I Problem during data-taking for Runs B-F - tracking inefficiency not present in oursimulation
I Results consistent nevertheless (within stat. uncertainty)
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 15 / 45
HbbHbb analysisof 2016 data
Event selection
I Trigger: Quad Jet OR Double Jet Trigger
I At least 4 jets with:I pT > 30 GeV, |η| < 2.5,I CMVAV2>0.185
|<2.5 (GeV)η Jet pT 4 for jets with |0 50 100 150 200 250
Eve
nts
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2 = 300 GeV
XSignal m
= 600 GeVX
Signal m
= 900 GeVX
Signal m
13 TeV Data
pT ,4
> 30 GeVT
|<2.5, pη Jet CMVA 4 for jets with |1− 0.8− 0.6− 0.4− 0.2− 0 0.2 0.4 0.6 0.8 1
Eve
nts
0
0.1
0.2
0.3
0.4
0.5 = 300 GeVX
Signal m
= 600 GeVX
Signal m
= 900 GeVX
Signal m
13 TeV Data
B-tagging discriminant
> 30 GeV, CMVA > CMVAMT
|<2.5, pη #Jets with |0 1 2 3 4 5 6 7 8 9 10
Eve
nts
0
0.1
0.2
0.3
0.4
0.5 = 300 GeV
XSignal m
= 600 GeVX
Signal m
= 900 GeVX
Signal m
13 TeV Data
Number of selected jets
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 17 / 45
Higgs selection
I Four b-jets from two Higgs bosons - looking for two Higgs candidates (H1,H2)
I Several regions (based on invariant mass of the original particle X )due to different event topology:
I Low mass region (LMR): two dijets with mass compatible with Higgs bosonI Medium mass region (MMR): b-jets are more boosted → requirement on
distance between the b-jets (∆R < 1.5,∆R =√
∆η2 + ∆φ2)
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 18 / 45
LMR: MMR (Gen):
}b R(b∆ 0 0.5 1 1.5 2 2.5 3 3.5 4
Eve
nts
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
= 300 GeVX
Signal m
= 600 GeVX
Signal m
= 900 GeVX
Signal m
13 TeV Data
left Comparison of dijet mass after Higgs selection for LMR
right Distance between two jets from Higgs decay (generated level)
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 19 / 45
Signal selection
I χ2 = (mH1 − mH)2/σH2 + (mH2 − mH)2/σH
2
I Reconstruct two Higgs Boson candidates with lowest value of χ2
I Signal region (SR): χ < 1I Sideband region (SB): 1 < χ < 2, (mH1 − mH) · (mH2 − mH) < 0
MC mX = 350 GeV MC mX = 650 GeV
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 20 / 45
Regression
I Presence of neutrinos + imperfection of the detector=⇒ lower recorded energy of jets
I b-jet pT correction using regression: uses multivariate algorithm trained onsignal Monte Carlo
I e.g. LMR:
Improvement 10.06% Improvement 11.94%Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 21 / 45
Signal region optimization
I Signal region is optimized for both before and after the regressionI Regression reduces radius of the SR and leads to better signal significance
LMR MMR
center radius center radius
baseline 115 25 120 20after regression 120 20 125 20
LMR: MMR:
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 22 / 45
Kinematic fit
I We expect mbb = 125 GeV (Higgs mass)
I Correction of momentum to achieve theinvariant mass
I Modification based on resolution of jetvariables (pT , η, φ) to achieve the lowestpossible χ2
I Resolution measured in Monte Carlo
mx 300 600
baseline µ 276.3 572.8σ 23 32.2σ/µ 8.32 5.62
kinematic fit µ 300.6 604.3σ 7.6 17.7σ/µ 2.53 2.93
kinematic fit + regression µ 301.1 606.9σ 7.7 17.5σ/µ 2.56 2.88
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 23 / 45
Signal extraction
I Signal as a peak on smooth background
I Fit done for LMR/MMR, components:I Signal - fit shape determined from MCI Multi-jet background - determined
from data (poor modelling)I tt - negligible
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 24 / 45
Signal fits
I Low mass: Gaussian signal + Gaussian combinatoric backgroundI High mass: ExpGaussExp function
Signal 260 GeV (LMR): Signal 800 GeV (MMR):
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 25 / 45
Background modeling
I Signal region blinded- fit in Sideband (SB), using Gauss-Exp function
Example: SB+LMR
Without regression: With regression:
(GeV)X m260 280 300 320 340 360 380 400 420 440 460
Eve
nts
/ 5 G
eV
0
20
40
60
80
100
120
140
/n = 0.862χSB
Data in SB
(2016) (13 TeV)-19.2 fb
CMSPreliminary
(GeV)X m260 280 300 320 340 360 380 400 420 440 460
Pul
l
5−4−3−2−1−012345
(GeV)X m260 280 300 320 340 360 380 400 420 440 460
Eve
nts
/ 5 G
eV
0
20
40
60
80
100
120
140
160
180
200
220
/n = 1.142χSB
Data in SB
(2016) (13 TeV)-19.2 fb
CMSPreliminary
(GeV)X m260 280 300 320 340 360 380 400 420 440 460
Pul
l
5−4−3−2−1−012345
SB definition:(H1 vs H2)
I Regression does not sculpt the background
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 26 / 45
Expected upper limits
LMR: MMR:
I Expected upper limits on the signal cross sections at 95% confidence level(computed using Asymptotic CLS method)
I Already better exclusion with only 9 fb−1 than at 8 TeVI With regression improvement 3.3-4.1% for LMR and 9.4-19.0% for MMR
(mX > 400 GeV)Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 27 / 45
Summary
I This talk summarizes current status of the 2016 search for two Higgs bosonresonance decaying into four b quarks
I My contribution over the summer:I Bulk of work on trigger efficiencyI Optimization of Signal regionI Study of effects of regression
I Corrections already applied: regression of jet pT , kinematic fit on Higgs mass;improvement of mass resolution
I Optimization of signal region, background estimation and expected upperlimits finished
Thank you for your attention!
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 28 / 45
L. Randall and R. Sundrum, “A Large mass hierarchy from a small extradimension,” Phys. Rev. Lett. 83 (1999) 3370doi:10.1103/PhysRevLett.83.3370[hep-ph/9905221].
R. Barbieri, D. Buttazzo, K. Kannike, F. Sala and A. Tesi, “Exploring theHiggs sector of a most natural NMSSM,” Phys. Rev. D 87 (2013) no.11,115018 doi:10.1103/PhysRevD.87.115018[arXiv:1304.3670 [hep-ph]].
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 29 / 45
Backup slides
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 30 / 45
Turn-ons (e.g. Double Jet, data driven)
L1:
4T+p
3T+p
2T+p
1Tp
200 250 300 350 400 450 500 550 600
Effi
cien
cy
0
0.2
0.4
0.6
0.8
1
Calo pt4:
4Tp
40 50 60 70 80 90 100 110 120
Effi
cien
cy
0
0.2
0.4
0.6
0.8
1
Calo pt2:
2Tp
40 60 80 100 120 140 160 180 200 220
Effi
cien
cy
0
0.2
0.4
0.6
0.8
1
CSV3:
3CSV0.75 0.8 0.85 0.9 0.95 1
Effi
cien
cy
0
0.2
0.4
0.6
0.8
1
PF pt4:
4Tp
40 60 80 100 120
Effi
cien
cy
0
0.2
0.4
0.6
0.8
1
PF pt2:
2Tp
60 80 100 120 140 160 180 200
Effi
cien
cy
0
0.2
0.4
0.6
0.8
1
Bands visualize uncertainty of the fit
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 31 / 45
Closure test
Double Jet trigger (DJ):
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
20
40
60
80
100
120
140
160
180
DoubleJet Trigger Closure
TriggeredWeightedInternal CMS
DoubleJet Trigger Closure
3CSV0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Rat
io
0.6
0.8
1
1.2
1.4
0 20 40 60 80 100 120 140 160 180 2000
10
20
30
40
50
60
70
80
90
DoubleJet Trigger Closure
TriggeredWeightedInternal CMS
DoubleJet Trigger Closure
4T
p0 20 40 60 80 100 120 140 160 180 200
Rat
io
0.6
0.8
1
1.2
1.4
4− 3− 2− 1− 0 1 2 3 40
10
20
30
40
50
60
DoubleJet Trigger Closure
TriggeredWeightedInternal CMS
DoubleJet Trigger Closure
1η4− 3− 2− 1− 0 1 2 3 4
Rat
io
0.6
0.8
1
1.2
1.4
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 32 / 45
Runs B-F Run G
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 33 / 45
Kinematic compatibility
I Comparison of tt sample and several signal masses:
0 50 100 150 200 250 300 350 400 450 5000
0.02
0.04
0.06
0.08
0.1
0.12
Jet_pt[0]+Jet_pt[1]+Jet_pt[2]+Jet_pt[3] {CMVAVsorted[3]>0.185 && diHiggs}
hEntries 7105Mean 316.7RMS 70.95
Signal (300 GeV)Signal (500 GeV)Signal (700 GeV)Signal (900 GeV)TTbar
Jet_pt[0]+Jet_pt[1]+Jet_pt[2]+Jet_pt[3] {CMVAVsorted[3]>0.185 && diHiggs}
0 20 40 60 80 100 120 140 160 180 2000
0.05
0.1
0.15
0.2
0.25
Jet_pt[3] {CMVAVsorted[3]>0.185 && diHiggs}h
Entries 7105Mean 64.9RMS 22.43
Signal (300 GeV)Signal (500 GeV)Signal (700 GeV)Signal (900 GeV)TTbar
Jet_pt[3] {CMVAVsorted[3]>0.185 && diHiggs}
∑4 JetpT JetpT ,4
I Low mass kinematically more similar than high mass= same behavior as in efficiency comparison
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 34 / 45
Regression (MMR)
I = bjet pT correction
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 35 / 45
Signal region
MC mX = 300 GeV< LMR
Data
MC mX = 900 GeV< MMR
DataFilip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 36 / 45
pt(1,2,3,4) before/after regression
|<2.5 (GeV)η Jet pT 1 for jets with |0 100 200 300 400 500 600 700 800
Eve
nts
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
0.22
0.24 = 300 GeVX
Signal m
= 450 GeVX
Signal m
= 600 GeVX
Signal m
= 750 GeVX
Signal m
= 900 GeVX
Signal m
13 TeV Data
|<2.5 (GeV)η Jet pT 2 for jets with |0 50 100 150 200 250 300 350 400 450 500
Eve
nts
0
0.05
0.1
0.15
0.2
0.25
= 300 GeVX
Signal m
= 450 GeVX
Signal m
= 600 GeVX
Signal m
= 750 GeVX
Signal m
= 900 GeVX
Signal m
13 TeV Data
|<2.5 (GeV)η Jet pT 3 for jets with |0 50 100 150 200 250 300 350
Eve
nts
0
0.05
0.1
0.15
0.2
0.25 = 300 GeV
XSignal m
= 450 GeVX
Signal m
= 600 GeVX
Signal m
= 750 GeVX
Signal m
= 900 GeVX
Signal m
13 TeV Data
|<2.5 (GeV)η Jet pT 4 for jets with |0 50 100 150 200 250
Eve
nts
0
0.05
0.1
0.15
0.2
0.25
= 300 GeVX
Signal m
= 450 GeVX
Signal m
= 600 GeVX
Signal m
= 750 GeVX
Signal m
= 900 GeVX
Signal m
13 TeV Data
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 37 / 45
Purity
Purity is equal to number of b-jets matched to generated b-quark. When two jetsare matched to one quark, the purity is -1.
(GeV)X m50 100 150 200 250 300 350 400 450 500 550
1
10
210
310Purity = 4Purity = 3Purity = 2Purity = 1Purity = 0Purity = -1
(GeV)X m250 300 350 400 450 500 550 600 650 700 750
1
10
210
Purity = 4Purity = 3Purity = 2Purity = 1Purity = 0Purity = -1
(GeV)X m500 550 600 650 700 750 800 850 900 950 1000
1
10
210
310Purity = 4Purity = 3Purity = 2Purity = 1Purity = 0Purity = -1
(GeV)X m650 700 750 800 850 900 950 1000 1050 1100 1150
1
10
210
310Purity = 4Purity = 3Purity = 2Purity = 1Purity = 0Purity = -1
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 38 / 45
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 39 / 45
Signal significance
I Study of signal significance to determine best signal regionI e.g. LMR: (TODO update or remove)
Before Reg.
mH [GeV] 115 110 120 115 115σH [GeV] 30 30 30 25 35
mX 260 46.82 48.40 42.91 45.17 48.14300 59.35 58.49 57.33 58.41 59.49350 97.18 93.48 96.47 97.02 93.20400 144.53 137.74 144.25 143.30 140.28450 74.28 73.94 72.57 76.10 70.49500 52.61 52.30 50.85 55.16 48.98550 147.56 144.60 142.90 155.50 137.96
After Reg.
mH [GeV] 120 125 120 120 120σH [GeV] 25 25 20 30 35
mX 260 49.89 47.30 46.44 51.50 53.43300 64.87 63.31 61.33 65.54 65.91350 112.73 112.73 107.80 113.06 109.79400 190.34 188.10 186.58 187.89 176.66450 91.47 87.40 91.64 88.06 82.30500 63.74 61.11 65.67 60.40 56.87550 178.99 170.77 184.70 168.21 152.47
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 40 / 45
mx fits and resolution
I Low mass: Gaussian signal + Gaussian combinatoric backgroundI High mass: ExpGaussExp function
Signal 260 GeV (LMR): Signal 800 GeV (MMR):
mx 260 300 350 450 600 700 800 900
baseline µ 243.4 276.3 318.5 425.3 572.8 673.5 771.9 870.6σ 18.4 23 28.7 25.2 32.2 35.5 38.9 42.1σ/µ 7.56 8.32 9.01 5.93 5.62 5.27 5.04 4.84
kinematic fit µ 260.7 300.6 350.4 452.4 604.3 706.6 807.9 908.3σ 4.1 7.6 10.4 12 17.7 21.5 27.3 31.1σ/µ 1.57 2.53 2.97 2.65 2.93 3.04 3.38 3.42
kinematic fit + regression µ 260.9 301.1 351.6 453.4 606.9 710 812.2 913.9σ 3.9 7.7 10 12.3 17.5 21.3 27.1 31.4σ/µ 1.49 2.56 2.84 2.71 2.88 3.00 3.34 3.44
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 41 / 45
mx fits and resolution - ExpGaussExp function
I High mass: ExpGaussExp function:
f (x ; x , σ, kL, kH) = exp
(k2H
2− kH(x − x)
σ
), for
x − x
σ> kH
= exp
(− (x − x)2
2σ2
), for kL ≤
x − x
σ≤ kH
= exp
(k2L
2+
kL(x − x)
σ
), for
x − x
σ< kL
(1)
I x : The mean of the Gaussian core,I σ: The standard deviation of the Gaussian core,I kL: The decay-coefficient of the lower exponential tail. This is also the number
of standard deviations, on the low side, beyond which the Gaussian inflectsinto the exponential.
I kH : The decay-coefficient of the higher exponential tail. This is also thenumber of standard deviations, on the high side, beyond which the Gaussianinflects into the exponential.
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 42 / 45
Background modeling - ExpGaus function description
I Signal region blinded - fit in Sideband, using Gauss-Exp function:I x : The mean of the Gaussian core,I σ: The standard deviation of the Gaussian core,I k: The decay-coefficient of the exponential tail. This is also the number of
standard deviations beyond which the Gaussian inflects into the exponential onthe high side.
f (mX ; x , σ, k) = exp
(−1
2(
x − x
σ)2
), for
x − x
σ≤ k (2)
= exp
(k2
2− k
x − x
σ
), for
x − x
σ> k
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 43 / 45
Background modeling: SB+MMR
Without regression: With regression:
(GeV)X m400 600 800 1000 1200 1400 1600 1800
Eve
nts
/ 20
GeV
0
10
20
30
40
50
/n = 0.442χSB
Data in SB
(2016) (13 TeV)-19.2 fb
CMSPreliminary
(GeV)X m400 600 800 1000 1200 1400 1600 1800
Pul
l
5−4−3−2−1−012345
(GeV)X m400 600 800 1000 1200 1400 1600 1800
Eve
nts
/ 20
GeV
0
5
10
15
20
25
30
35
40
/n = 0.442χSB
Data in SB
(2016) (13 TeV)-19.2 fb
CMSPreliminary
(GeV)X m400 600 800 1000 1200 1400 1600 1800
Pul
l
5−4−3−2−1−012345
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 44 / 45
Regression: improvement of limits
Kin. Fit KF+regr. Improv.(%)
LMR
260 3085 2976 3.7300 1976 1898 4.1350 808 781 3.5400 470 455 3.3
MMR
350 - 925.8 -400 315.9 321.8 -1.8450 140.6 118.2 19.0500 108.9 92.8 17.3550 89.4 76.7 16.6650 59.1 49.3 19.9700 52.2 45.4 15.0800 40.5 35.6 13.8900 35.6 31.7 12.3
1000 33.7 30.8 9.41200 38.6 34.7 11.2
I With regression improvement3.3-4.1% for LMR and9.4-19.0% for MMR(mX > 400 GeV)
Filip Nechansky (FNAL) HbbHbb search on CMS 9/22/2016 45 / 45