CMS SUSY in Karlsruhe

1
CMS SUSY in Karlsruhe
The group Prof. W.deBoer Dr. V.Zhukov PhD.
M.Niegel PhD. A.Cakir PhD. E.Ziebarth Diplm.
D.Troendle
2
The subject
mSUGRA inclusive topological search
Model dependent search rather than looking on
any deviations from SM background (which might be
related to the systematic uncertainties)

• The goal
• find any manifestation of mSUGRA model at LHC
  in allowable parameter space in different event
  topologies, or/and exclude some mSUGRA
  models
• reconstruct or constrain the mSUGRA model
  parameters
• check consistency with observations of Dark
  Matter (direct and indirect)
• and EW constraints

3
The method

• Consider general inclusive SUSY signature
  LeptonsJetMET .
• Leptons appears at the end of SUSY cascade
  c20-gtllc0 , c1-gtlnc0 , l -gt lc
• and defined by gauginos composition and the mass
  scale. Use leptons as the main parameter to
  discriminate topologies (jets can be used as
  well).
• 0lJetMET
• 1lJetMET, etc
• Analyze all possibilities simultaneously in one
  analysis,
• combine signal (if any) from all topologies.
• Thus
• increasing signal significance
  Significancesign1sign2...
• reducing systematic uncertainties in SM
  backgrounds dNbkg
  Significance Ns/sqrt(NbkgdNbkg2 ) by data
  driven calibration in the signal less regions.
• constraining model parameters from ratios of
  different topologies Ns1Ns2Ns3...

The method is complimentary to the kinematic end
points and mass scales analysis but can work in
the 'difficult' regions and at low statistics.
4
SUSY leptonic signatures
Multileptonic (Nmuons) signal topologies(MC)
in mSUGRA mass(m0-m1/2) plane (tanb50)
Nm0/total
Nm1/total
MSUGRA total cspb
Nm2/total
Nm3/total
rijNsi/Nsj - observed ratios of topologies sji
- uncertainties and bkg contamination Rij
(mSUGRA) -predicted ratios chi2Sum(r - R)2/s2
-gt the preferable mSUGRA regions can be
identified by minimizing chi2. Limitation
statistics and the topology dependent
contaminations.
Nm1/Nm2
The ratios reduce uncertainties in the selection.
5
Cross sections and backgrounds
) Only muons MC level
SUSY signature Muons JetMET
SM backgrounds(B) with real prompt leptons and
fake leptons from jets (Bfake)
SM background with prompt leptons W -gtln ,
Z/g -gt ll the leptons can be separated by
kinematics (PT, Minv, MinvT, etc)
LM11 LM1 LM9 LM others QCD fakes _ Zj --
Wj - - ttbar .....

• SM background with fakes
• from short lived particles B,D
• on fly decays pi, K
• punchtrough, ghosts...

recoMuons PT
recoMuons eta

• Fakes have similar kinematics
• Large uncertainties in the fake rates
  estimation.
• Fakes becomes more important for multileptonic
  states

6
Ingredients
1. Leptons identification. Data driven
calibration of fake rates with the fakes
enriched reference samples. Optimization
of leptons identification for different SUSY
models and different signal
topologies. Using prompt enriched reference
samples for optimization and MC
corrections for SUSY.
2. Optimization of the event selection.
Framework for the multi topological search. Use
of multi variant
methods (TMVA,NN) for selection of
observables and building discriminants .
Integration of tools into PAT.
Define a set of most significant observables and
the selector for each topology and
corresponding backgrounds.
3. Control on systematic uncertainties.
Theoretical uncertainties in the observables used
in SUSY selection,
define observables least prone to the
uncertainties Stat methods for
combination of different systematic uncertainties
and reduction of systematics.
Tuning of MC generators with reference samples.
Many related topics can be used elsewhere...
7
Leptons identification fakes origins
parton id of reco fakes
QCD jets RecoMuons PTgt5GeV/c
Most dangerous are leptons from the cascade
decays of heavy flavors b-gtc-gtu,d sinf(mB-mD
)/2plmD
b
notisoPt
94
Leptons in jets
All selected
c
Isolated Fakes
6
isoPt
100
identification
0.8
W
W
d
f
u
From b
From pi,K
From c,s,u,d
b
c
u
0.1
0.40
0.30
Main selection parameters - isolation in the dR
cone (lt0.3) by sumPT of tracks or calo
deposition - vertex (tb1.5ps, tc0.5 ps)
significance Dxy/sxy
Muons fake rate per event in Z/W/ttbar (chowder
jetmet) after SUSYAnalyzer leptons cleaning
Heavy flavors production 1. ME at large PT
(Alpgen,Sherpa,Comphep,etc) 2. soft gluon
splitting by GLAP (PYTHIA SR) Large uncertainties
due to PDF, mb, sensitivity to the matching
(MLM, CKKW), etc
allfakes b,c and short lived long lived p,K

• Calibration of fakes from soft gluon splitting
  will reduce uncertainties in the fake rate in
  SUSY search and allow to tune the soft QCD
  simulation

recoMuons PT
recoMuons eta
8
Leptons identification fakes enriched sample
'Fake' enriched sample(muons) QCD dijets
Bkg(prompt muons) Wjets ,ttbar, Zjets
) Consider muons as an example electrons are
similar
Preselection HLT jetmet stream (avoid bias on
leptons), METlt50, Nm(PTgt5)gt0, Njets2 Final
selection cuts optimized with Genetic algo
using more observables
Selection efficiencies
At L10 pb-1 - Ndijets 10 5 events -
contamination 4 10 - 4 - Systematic
uncertainties ( JES, PDF) lt10 Low HLT
efficiency (lt10) limits the signal statistics,
needs a specific low PT dijets stream.
9
Leptons identification fake rates
Fake rate per event (JetMet stream) for
different channels (SusyAnalyzer clean)
Fake rate in Wj, Zj, ttbar/10 normalized to
the fake rate in QCD events
QCDjets sample Z/j Wj ttbar LM9 LM1
reference
MC corrections to extrapolate to another channel
and selections
PT distribution off all isolated muons after
subtraction fakes according to dijets fake rate
PT distribution off all isolated muons in
inclusive 3muonsJetMET selection
Example selection JetMET HLT, METgt50,
Meffgt200, Njgt2 (100,60,30), 3 reco muons (PTgt5
GeV) no extra selection
10
Leptons identification prompt enriched sample
Prompt enriched sample Z-gtmuons MZ60-120
GeV Bkg(fake muons) QCDjets, Wjets ,ttbar
Preselection and selection similar to the 'fake'
sample
At L10 pb-1 - N (Z-gtmm) 10 4 events -
contamination 10 - 3 - Systematic
uncertainties ( JES, PDF) lt20 main
contamination from bbar
11
Lepton identification optimization

• Use leptons from 'fakes' and prompts enriched
  samples to optimize selection
• Select most significant observables using Genetic
  algorithm(GA) and NeuralNet(NN)
• Verify the selection with the MC truth SUSY LM9
  and Z/W/ttbar/QCD samples

Reduction of the muons fake rate from QCD with
different selections
Optimized selection cuts
Selected isol byPT selection SusyAnal cuts GA
samples NN samples
Selection efficiency (cuts from samples) of
prompt SUSY for all SUSY LM points

• Improvement of fakes suppression by factor with
  30 drop in efficiency using reference samples.
• MC truth gives similar within 10

12
Event selection topologies
Selection of leptonic SUSY states in steps 1.
preselection for all topologies using JetMET
parameters and triggers 2. specific selection
for each topology optimized for the
backgrounds 3. leptons related selection for each
leptonic state
PTDR JetMET incl. HLTjetMET METgt200 Njgt2,
ET180,110,30 ??lt1.7 ??(metj2)gt20o
Inclusive averaged observables in mSUGRA
m0-m1/2 plane, tanb50 preselection defines
limits of discovery reaches
13
Event selection observables and selectors
Select observables for each step of selection
according to theirs correlations(linear and non
linear) and significance ranking.
Different MVA methods for LM9-ttbar (using
parameters above)
Linear correlation matrix for the LM9-ttbar
inclusive SUSY search
Etj2 f(j2,MET) Etj1 f(j1,MET) hj1 Meff MET sumET N
j Nm
Ranking LM9-ttbar for different algos

• Factorize the least correlated observables for
  simple cuts
• Use non correlated observables for calibration
  in side regions
• Combine observables to reduce the systematic
  uncertainties
• Use Neural Net(NN,MLP) and Boosted Decision
  Trees(BDT) for correlated observables in fine
  tune selection with reference samples
• Optimization will depend on mSUGRA regions (LM
  points)
• -gtconsider regions complaint with other
  constraints

14
Systematic uncertainties

• Detector related systematics are mostly taken by
  simulation.

• For comparison of signal in different topologies
  one need optimization of Nsig/Nbkg, i.e. small
  statistics (thus also reducing systematics
  contribution)

• Consider theoretical model uncertainties coming
  from QCD simulation
• PDF's, factorization scale, matching schemes,
  UE tunes, etc.
• for the selected observables used in SUSY
  selections
• MET, Etjets, Meff, angular parameters, etc

Example Etj1 and MET for different
factorization scales in ALPGEN/SHERPA in Zjets
(MC)
15
Systematics uncertainties MET and Meff
Effect of systematic uncertainties in data
driven MET calibration using Z(mm)-gtZ(nn) .
MET and Meff for LM1, LM9 SUSY points and
calibrated Z(nn)jets simulated with ALPGEN
and SHERPA at MC level. Lint1fb-1
MET ratio ALPGEN/SHERPA

• The largest systematics are related to the
  simulations of jets ET tails(factor. scale) and
  number of soft jets (matching),
• this would require a tune of generators.
• The MET uncertainties are dominated by the
  detector

16
SUMMARY
Concentrate on the model dependent
topological search of SUSY (mSUGRA), consider
multileptonic (multijet) inclusive states in
different mSUGRA regions complaint with the DM
and EW constraints. Developed a method for
the constrained mSUGRA scan with Markov Chain
MC. Developed an algorithm for optimization
of leptons identification and fake rate
calibration with data. Can be used for the study
of soft QCD and tuning of matchings schemes.
Under development a framework based on
multivariant methods for selection of
observables, theirs data driven calibration, and
optimization of event selection. Using it for
the SUSY trileptons inclusive and exclusive
study as an example. Studied the theoretical
systematic uncertainties in the bosonjets
production at MC level. Extending the study for
the full/fast simulations in real analysis.