MSUGRA search at large m0 review and plans of Karlsruhe group

1
MSUGRA search at large m0review and plans of
Karlsruhe group
Content People Motivations Trileptons
from direct neutralino-chargino production
experience from PTDR 2006 encountered
problems and plans for 2007 Inclusive search at
large m0 selection of observables for SUSY
search SM bkg validation models
identification
2
People at Karlsruhe
2006 Prof. W.de Boer Dr. V.Zhukov PhD students
M.Niegel
2007 (new) Prof. W.de Boer Dr. V.Zhukov PhD
students M.Niegel, A.Cakir Diploma students
D. Daeuwel, S.Reisser, D.Stricker, E.Ziebarth
Other CMS PRS groups in Karlsruhe Higgs, Top,
btagging Prof. T.Mueller, W.Wagner et
al. Grid, EW, statistics Prof. G.Quast et
al. QCD Dr.
K.Rabbertz et al.
Seems enough manpower, will collaborate with
other groups sharing same SM bkg and analysis.
3
Motivations constraints
MSUGRA constraints and Wh2 band
tanb50

• Indirect DM (EGRET) m1/2lt250 , tan?gt50
• light neutralino-gt large m0 (mc m1/2)

Scan of models compatible with Wh2
Region m0gt1000 is complimentary to the indirect
and direct DM search
MSUGRA is a reference, other models with the
light LSP are similar.
4
Motivations SUSY production
Contributions of different production channels
Main production channels
? tot pb
tan?50 A00
NO EWSB
Total mSUGRA cross section(LO)
LO ISAJET PYTHIA NLO(PROSPINO) KNLO1.3-1.8
(m0100-2000)
Dominant gaugino production at large m0 c1 ?c1? ,
c2 0c1? , c2 ?c2?, ....
5
Motivations Focus Point
Narrow region along EWSB at large m0 with low m,
compatible with Wh2 (?)
See backup slide for details...
Relic density Wh2
lts v gt DM annihilation
Production channels along FP
H.Baer et al
cc
mNLSP / mLSP
mSUGRA m
gg
NOEWSB
mco1
gg
Mllmc20-mc10
c20c1?
tan?50 A00
6
Motivation event topologies
Large m0 (low m1/2) region low MET, soft jets
and leptons -gt difficult region Start with
trileptons...
Inclusive mSUGRA averaged observables in
m0-m1/2 plane
tan?50 A00
7
Trileptons in PTDR 2006
Selection with cuts
Cross section c20c1? -gt 3l(e,m)
3 isolated leptons OSSF l (PTmgt10, PTegt17,
PT3gt10 GeV/c ) NO hard Jets (ETgt30 GeV) MET?
Analysis optimized for the lightest mc LM9
(m01450, m1/2175, tanb50)
Trileptons(3l) MET is small and comparable with
SM. Somewhat bigger for the gg -gt2l..
Trigger
L1HLT efficiency
streams L1 , HLT, m 90
82 2m 74 70 e/g 45
38 2e 25 21
Reconstruction
Leptons isolation(Tracks, Ecal) and
identification (Eeld) use all available tools in
ORCA/FAMOS
muons
electrons
Efficiency (LM9_3l) L1 86 HLT
91
8
Trileptons backgrounds
Trileptons backgrounds
Main bkg is Z/gjets due to fake leptons, very
dependent on MC model.
Most of fakes are from b jets (hard or SR)
Fake rates per event with standard selections and
Nbkg due to fakes
9
Trileptons Minv OSSF at LM9
Use Neural network to improve results NNMET,
PTl, ETj, angles, etc, 20 observables
Minv(OSSF) selection with cuts
Model dependent selection
10
Trileptons discovery reaches PTDR
Systematics reconstruction 2 PDF (re
weighting ) 1.7 But dominant are
uncertainties in the fake rate Nbkg/Nbkgfake0.68
/-?
Without fake rates uncertainties
Hardly can see trileptons at Lgt10 fb-1
Number of selected events with different
PDF Normalized to selection efficiencies
11
Trileptons summary
Trilepton 2006 not a discovery channel, does not
compete with gg cascades for most of SUSY
regions due to low MET and fakes leptons. BUT,
can be used in addition to identify/narrow the
mSUGRA region. Moreover trileptons is the
dominant channels for FP. The Z/g jets bkg is
dominant and probably can be reduced.
For 2007 plans 1. Backgrounds PYTHIA
DY-Zjets replace it by ALPGEN or SHERPA
Z/gjets. Cross section? NLO(MCFM)
missed in PYTHIA Wg validation with
data 2. Fake leptons detailed study
classification, lepton Id, data driven
validation 3. General validation with
CMSSW/Fast, new data samples, selections
(relax jets veto?), optimization of observables
systematics uncertainties analysis
framework
Use trileptons as an example for the low MET
studies Update of trileptons by the october 2007
12
Trilepton backgrounds revisited
Z/g jets
Zjets
DYSR

• PYTHIA(PS) s(DY)15nb(gt20), s(Zj)14nb(gt20GeV)
• inclusive DY (Z/g) for PTltMz, but SR is
  probably not correct
• Z/gf does not reproduce well the low PT part
  
• The jets ET tails are wrong.
• The multijets are absent.

G
/g
/g
QED FSR is also important!
g-gtee
ALPGEN (ME up to 5 jets) or SHERPA can be a
solution. s(Z/g1j) 4 nb, more reliable jets
in collaboration with F.Moortgat, F.Krauss,
M.Mangano,etc
Need more validation before starting mass
production.
Wjets ALPGEN/SHERPA
A.Cakir, M.Niegel,...
13
Trilepton backgrounds revisited
Z/gW(3l)
Considered s(WZ-gt3m) 1.7 pb (LO) Missed
W/g ? similar to Z/g jets but with a real
W lepton significantly contribute to MllltMz
where signal is expected.
CompHEP
Calibration with data Use reference SM
channels. For.ex. Z/g jets NjetsgtN0,
METltMET0 -gtno SUSY or ZW (MllMz,
MjjMw) validatation with Mz peak.
g and interference region with large MC
uncertainties
Zpeak- can be used for the absolute calibration
Minv-gt0 for FPregion mc20mc10
In collaboration with V.Brigljevic et al.
(Zagreb) CERN EW group
14
Fake leptons
M.Niegel et al.
First step deep tracing of all fakes origins
Example origins of fake electrons in DY and bbar
PYTHIA
developed based on CMSSW Btag/McTools.
Second step improvement of identification and
isolation -use NN for improvement of soft
electrons and muons isolation. Need large data
samples gt10 6 (CMSSW and FAMOS)
In collaboration with E/gamma and Muons
Z -gtll- MllMZ
Third step data driven calibration Use
Z1jet as a reference channel. Validate with
real data using pure Zj sample. Currently
emulate with smeared data.
One Jet Pj-Pll
Fake/jet
'fake'
15
Inclusive search
At low luminosity or/and low cross section
-gtinclusive topological search in low statistics
regime, complimentary to end points analysis,
constrains the model.
1. Selection and conditioning of observables
(general) - most significant observables for a
particular signal-bkg, ranking of observables. -
least sensitive to the systematic
uncertainties - preprocessing, get rid of linear
correlation, smearing and flattening avoid local
mins Need tools Here use NeuralNet
_at_phit(Karlsruhe), also have a look at TMVA
ROOT. Goal - write a general use public
package. 2. Validation of Simulation model (MC
and Reco) - data driven validation 3. Event
selection 4. Model constraints - mSUGRA
region separation by event topology
in collaboration with H.Prosper, K.Johnson,
R.Cavanaugh, H.Baer, etc (Florida)
16
Observables
Example LM1(60,250,10)-ttbar
Observables I METx, METy Njets(gt30GeV),
ETj1,ETj2,ETj3, h1,2,3 Correlation matrix-gt
ranking of observables
Observables I decorelated no prerpocessing
metx mety Etj1 Etj2 Etj3 h1 h2 h3 Njets
Remove linear correlations, non linear are taken
by NN
With preprocessing (flatenning, splines)
metx mety Etj1 Etj2 Etj3 h1 h2 h3
See backup for details...
Observables II MET,Meff,sumET,Njets Nj,ETj1,ETj2,
ETj3,h1,2,3 cos(j1j2), cos(j1met),cos(j2met) Nl
ept, Ptl1, Ptl2, Ptl3,MinvOSSF, Assym, h more
30 variables. De correlated and preprocessed
Observable II preprocessed
17
Selecting observables
NN out SUSY-ttbar
NN out LM9-bkg
efficiency
Worst Focus Point
contamination
Best- High Mass
The significance of each observable for a
particular sig/bkg pair depends upon preselection
and systematics.
18
Validation of the Simulation Model
Data -driven validation 1. preselect a data
sample for a particular channel(class of
events) for .ex. large sumET- small MET , or
ttbar (use KineFit to reconstruct top) , or Zjets
(select Z peak) 2. setup a NN with observables
selected for the SUSY discovery. Use Sim data for
the tested channel as a signal and Real Data as
bkg.
Example ttbar and ttbar smeared with
uncertainties in order to emulate the real data
Jets Energy scale (JES) s(Ejets)-10 METMET-ETj
corrected s(Eelectron)-10 stochastic s(ETjets)
10 s(PTleptons)5
Use FAMOS for a moment
Can do
These events have to be studied in details, they
can fake the signal
Find differences important for the SUSY search
Re weight the Simulated Data to match the real
one. Or use the weighted real (strongly
preselected) data for the NN training.
Use the NN to veto region of interest (blind
analysis)
NN out for truly ttbar and the smeared ttbar.
19
Event selection and bkg suppression
Ideal(conservative) world event selector does
not depend on the signal model. Assume only the
SM is known (really?). Remove the SM events and
see what's remaining.
-gt Selection with cuts
Can try to use model dependent selector based on
NN for each SUSY region. Drawbacks selection
depends on the model
Selection efficiencies for PTDR2006 analysis
m0-m1/2 mSUGRA
JetsMET
JetsMETm
5s reach for cuts and NN
430 NNs
SS 2mJetsMET
OSSFJetsMET
OS 2tJetsMET
Trileptons
NN with observables MET, Heff, ETj1, ETj2, ETj3,
Ptl Nl, Nj,Njb cos?(j1j2), cos?(j1met),
cos?(j2met)
PTDR incl METJets METgt200 Njgt2, ET180,110,30
h lt1.7 f(metj2)gt20
More work to build a selector for large m0 .....
20
SUSY region separation
Can constrain the mSUGRA model by the event
topology
S2-(S1S3)
For.ex Consider 3 mSUGRA regions. train NNs for
each combination
NN out
70 efficiency with 30 contamination In reality
have to add bkgs.
Selection efficiencies of 3 NNs in mSUGRA
plane(fixed NNout cut)
S2/S1S3
S3/S1S2
S1/S2S3
Can reconstruct fraction of the production
channels gg, qq, cc for a patuclar
point Use 3 NNs trained for each channel against
others
Ex( m0500 m1/2200 tb50) MC and reconstructed
fractions
21
Summary
Manpower 1 Prof, 1 Dr, 2PhD, 4 students
Physics concentrate on SUSY scenario with
light LSP (split SUSY like)
BSM physics group Leptonic searches SUSY
trileptonMETjets (exclusive) gt0
leptonMETjets (inclusive) (new, join)

Collaborate Egamma, Muon, JetMET, Btag
EWK
Bkg and Data samples mSUGRA(LM7,9,10, scans)
SUSPECTPYTHIA Z/gjets
ALPGEN,SHERPA, NLO? MC tuning,
validation Z/gWlets
CompHEP, NLO(MCFM?) validation ttbar, QCD,Wj
general interest
22
Summary (cont.)
Trigger 2e, 2m, e, m, mjet, mMET, eMET,
ejet low Pt leptonic triggers
validation

• Reco objects
• electrons and muons isolations and cleaning,
  efficiencies
• fake electrons and muons, data driven
  calibration
• MET calibration at low ET

Analysis

• Optimization of observables in multiparameters
  search
• Use of stat methods (NN/LH/GA,etc ) for event
  selections and model identification
• MC and simulation model validation with data

• Proposal?
• Create one more working group/taskforce SUSY
  BSM tools
• analyzing framework (susyAnalyzer), technical
  aspects
• stat methods for analysis and discovery
• data format and access, etc...

Will work in clode contact with AnalysisTools and
stat group
23
BackUp slides
24
Focus point details
K.Matchev at el.
At large tanb m2-m2H2 When the m0 is large
enough the mH2 and therefore m decreases until
m2 lt0 i,e, no EWSB is possible. This Focus point
is at large m0 and low m where all usial SUSY
mass relations does not hold (mltM1,M2),
neutralino gets a large higgsino component.
mc 0.4 m1/2
H.Baer et al
Wh2
Higgsino like
Higgsino component is increasing for large m
25
Qualify the observables details
Processing MET in LM1-ttbar (inclusive)
Same for the cos(j1j2) not as good
Flattening sigbkgconst
Purity Nsig/Ntrue
Input for NN norm.to mean
Observables sorted by significance (contribution
to the NNout in )
MET(23), Meff(5), Nl(9), sumET(8), Nj(4),
cos(j2met) (5), ETj1(3), hj3(3), cos(j1met)
(3), ETj3(2),cos(j1j2)(2),hj2(1.5) hj3,
Etj2, PTl2, PTl1,.....
Enough to keep only first 10 in NN...
Final NN purity-efficiency(red) and with only
MET(black)
26
Analysis framework details
Based on SusyAnalyzer and AnalysisTools
MrMET
RecoCandidate
MrRecoParticle clean0
CMSAnalyserEDAnalyzer
Calibration
MrJet
Setup
HEPMCCandidate
MrTau
ConfigParams
MrElectron
MCParton
MCParticle fromSusy() fromSR() ...
MyAnalyzer RecoAnalyzer ra MCAnalyzer
mca RootTree tree
MrMuon isolate(vectorltMrRecoParticlegt) clean(vect
orltMrRecoParticlegt) matchMC(vectorltMCParticlegt)
......
RecoAnalyzer vectorltMrRecoparticlegt
MCAnalyzer vectorltMCParticlegt
RootTree double R_jp double R_mup .....
Histos
External packages working on custom tree
namespaceTools
TMVAfactory
susyNNAnalyzer
endpointAnalyzer