Inclusive mSUGRA search at CMS Topology Selector

1
Inclusive mSUGRA search at CMSTopology Selector
V.Zhukov M.Niegel

• Inclusive and Exclusive SUSY search at LHC
• Overview of the CMS PTDR results
• Regions in mSUGRA parameter space
• Topology selector for mSUGRA

22/0672006 Progress report since may 2006
2
MSUGRA 5 parameters
Five universal (at the GUT) parameters ultimitly
describes masses and couplings mo, m1/2, A0,
tanb , sgnm . The masses at EW scale are
obtained from RGE (-corrections)
Useful mass relations m(g)2.7m1/2
m(c10)0.4m1/2 m(c20)m(c1)0.8m1/2
m2(q(u,d))mo2 6.2m1/22 m2(lL(lR))mo2
0.5(0.15)m1/22 m2(t)m(u)2 mt2-Am2o -Bm1/22
Also wotk without universality(MSSM) but less
spectacular
mo60 m12250 tanb10 mo230 m12360
tanb10 mo175 m121450 tanb50 mo189 m12850
tanb10
3
MSUGRA signal signatures
Cascade decays

• General Signature
• MET
• Jets
• Leptons

depends upon mass spectrum and couplings
4
Inclusive and Exclusive search
Inclusive any destortion from SM bkg
prediction at some selection. works at low
statistics. First days physics. - Counting like
experiment ,very sensitive to the systematic
uncertainties (theoretical, reconstruction,
selection), low significance Sinc
Can the inclusive search be improved?
Exclusive Why it is better? Look for a
particular channel where some kinematic variables
are very different from the bkg (f.ex. Invariant
mass of OSSF dileptons), i.e first inclusive
search to select events, and second look for
the distinctive parameter. increase signal
ignificance S2excl Sinc2Sobs2 and
hypothesis separation, less sensitive to uncert.
Can reconstruct model parameters. - require
relatively large statistics.
10 fb-1
SUSY spectroscopy from the kinematic end points.
Works well mostly for 2 body decays at Lgt30
fb-1. ATLAS discovery reach with kinematric end
points
5
MSUGRA at CMS
MSUGRA has five ' free' parameters mo, m1/2, Ao,
tanb, sgnm CMS tune analysis to the defined test
points LM(low mass) and HM located in the 'bulk'
region at low m0
Relic DM density constraints exclude most of the
mSUGRA Large m0 , low m1/2 region can be
important.
tan?50 A00
At tanb10 only coannihilation and Focuspoint
6
CMS analysis summary
Use PYTHIA (LO) for the signal and bkg full
simulation for the test point, fast(FAMOS) for
the scan
Main selection METgt200 GeV Main bkg
ttbar, wjets, qcd Tuned mostly to the bulk
region (low mo) top, t , ho, Z, 3l, reconstruct
invariant masses (exclusive)
7
CMS discovery reaches from PTDR
Very different signal processings, bkg
treatment independent cross check (different
triggers can be treated as different exp.) -
difficult to compare results!
8
SUSY production at LHC
Total mSUGRA cross section (ISASUSY LO) and
channel fractions. KNLO1.6
s tot pb
pp-gtgg (excl.)
pp-gtc20c1
pp-gtqq
tan?50 A00
9
MSUGRA observables Jets
Reconstructed (FAMOS) averaged observables
ltNjgt ETgt30GeV
ltNBjgt B jets
mgmq long cascades
c02-gt H0 -gtbbar
leptons
Second jet ltETgt
Highest jet ltETgt
200
250
Hardest jet in q decays
120
Third jet ltETgt
Cosf between first and second
Cosf between first and MET
Cosf between second and MET
10
more observables MET
S ET from CaloTowers
MET from CaloTowers
2 body
Low MET region
MinvminvH1minvH2/2 invariant mass per hemisphere
Heff METS ET jets S ETleptons
11
And more Leptons
ltN gt electrons
ltN gt muons
ltN gt SameSignSameFlavor
ltN gt OppositeSignSameFlavor
Highest muon ltPTgt m1/2
Cosf OSSF and MET
back2back
Nss/Nss-
ltMinvgt OSSF
Expect assymetry
Zpeak
12
Trigger on MSUGRA
Channels and thresholds defined by allowable
rate L1(100 kHz) HLT(100Hz)
L1 trigger streams (hardcoded) can be treated as
different experiments!
HLT streams (flexible alorithms)
Inclusive efficiencies
HLT/L1
Low MET low PT region
13
SM background
How well SM is known( cross section, kinematics)
? EW relatively well, QCD -not reall y
LO PYTHIA CTEQ5L 2-gt2 channels (ex. Zbbar) . NLO
by K factors if exist.
Most dangerouse with highest MET ttbar, Wjets,
Zjets and highset multijet cross section QCD
14
SM background observables
Typical SUSY cuts
Njets ETgt30
MET
Minv(OSSF)
Heff
Muon PT1
Jet ET1
15
Theoretical uncertainties
1. Jets kinematics. Parton shower (PS) in PYTHIA
and matrix element (ME) in ALPGEN differs for
high PT in multijet states
This affects jets ET and MET strongly (since
sMETsqrt(sumET)) -SM bkg increases by 2-3 times
- The uncertainties on the Nbkg are increasing
reducing significance further
PYTHIA PS 6.2 ALPGEN ME
PT of highest emmited jet in ttbar production
2. Cross sections, PDF, KNLO uncertainties can
amount up to 10.
3. QCD effects, ISR/FSR depend on (as). Can
change jets multiplicity and jets energy spectrum
4. Underlaying events, minimubias events tend to
increase MET.
Find/use parameters which are least sensitive to
these.
16
Reconstruction uncertainties
Jets and Jet Energy Scale(JES) two component in
the errors - energy scale, can be improved with
statistics - stochastic term 1/sqrt(E) - fake
jets (jets algorithms)
MET CDF expirience, will be hard before MET can
be used. - detector hermeticy (dead calocell
fake MET) - JES calibration and monitoring. -
still sMETsqrt(SET)
Leptons JetMETLeptons can have smaller
uncertaint. But. .Fakes are the main
problem Fe10-4 Fm10-5 depends on the
channel (see Martins talk)
17
Event selection
Inclusive analysis in CMS
Bkg hypothesis Hb
Signal hypothesis Hs
signal
Selector
ns
S
nb
sb
backgrounds
a
/- uncert.
S
nb
nbns
Selector is optimised for some particular test
point to get maximum significance S. For small
statistic regime at S5 (discovery reach) ns5,
nb1 shape of Hb distribution is important!
a -probability to accept wrongly the Hs P-value
, probablity to observe gtnbns
Significance Sns/sb , where sb
sqrt(nbdnb2) nb -Poisson term dnb - systematic
uncertainties
- Only one parameter (S) to characterize a
hypothesis, that the bkg is not enough to
explain data. - Can not say much about the
signal hypothesis itself. - Systematic of the
background is important
18
Event selection
Can we do better?
-The mSUGRA topologies depends on parameters ,
can define n- region dependant selectors Seli
-There is only one solution , i.e. Selk the
Sel n-m should give no signal at the defined
threshold (5s) . Probability to observe a signal
in k -region Ppk Si?k (1-pi)
Multichannel inclusive analysis
signal
Selector i optimized for i-topology
Si
ns
ns
S
Si
Selector i optimized for i-topology
ns
nb
backgrounds
Preselector
nb
/- uncert.
Preselector suppresses most of the SM
background Selectors - tunes to different
mSUGRA topologies
m1/2
k
Problem The bkg suppression efficiency should
be mSUGRA model independant, i.e. selection
efficiency Ei(Dbkg)const.
mo
identify mSUGRA region, increases the model
significance. Can do a BLIND analysis, i.e.
exclude the signal topology and test the signal
and bkg hypothesis in neighbore regions. - very
model dependant . How to control bkg?
19
Selection efficiency for the PTDR analysis
Use selection cuts defined in PTDR analysis. Most
of them are tuned to mo60, m12250 tanb10 bulk
region
tan?50 A00
20
Discovery reaches 10 fb-1
Taking the Nbkg from analysis, the results are
very close... PTDR is tuned to tanb10, here is 50
21
Region separation.
Use Neural Network (NN) for region separation.
Simple preselecion(not optimized for bkg) train
NN for region i against other regions j-i
Sig2/bkg13
Preselection efficiency
Preselect METgt50, Njgt2 (gt30GeV)
NN parameters MET, Heff, ETj1, ETj2, ETj3, Ptl1,
Nj,Njb cosf(j1j2), cosf(j1met),cosf(j2met)
3 test points (m0,m1/2,tanb50) Sel1(500,450) Sel
2 (1100,300) Sel3(2500,500)
Sel Efficiency
ExampleTeacher out
Selection efficencies of 3 NNs trained for
Sig3/bkg12
Sig1/bkg23
Sig2/bkg13
Can constrain mSUGRA regions
22
MSUGRA parameters
Can we separate channels and identify the
fractions(f.ex. gg, qq, etc)? Train NNs for
individual channel agins others for one test
point (m0500 m1/2200)
qq
gg
cc
MC and Rec fractions
NN selection efficiencies ( m0500 m1/2200)
The selection efficiency E is a convolution of
NN and preselection efficiencies Eie (presl)ki
e (NN)k . Find the fractions F i from the
observed Di DiEi X Fi And compare with the
expectation from Mc(normalized)
Another point(m02500 m1/2600) with the same
NN( m0500 m1/2200)
Larger errors, but still can constrain the ratios.
The s(gg)s(qq)s(cc) ratios also
constarins the mSUGRA
23
Bkg suppression
Preselect data samples and use NN to train ttbar
against mSUGRA regions, produce many Nns and
check efficiency for the bkg and signal
Preseelction METgt50, Njgt2(ETgt30), 2m(PTgt10)
NNs (preselection) efficiency for
bkg(ttbar)Best is const, i.e. region
independant.
NNs preselection efficiency for two mSUGRA points
m0500 m1/2200
m02500 m1/2600
Preselection efficiency dominates, similar
pattern. Are there improvement in discovery?
Up to 10 spread, is it acceptable?
24
Summary and Plans
-The mSUGRA topology selection and region
separation is possible ininclusive search. -The
multiregion optimization allows to identify
mSUGRA region and can be used to separate SUSY
from other models (UED,etc)
Under investigation - produce more backgrounds
(ALPGEN) and optimize bkg suppresion -
sensitivity to the uncertainties, selection of
parameters - BLIND analysis example - extraction
of the channel fractions for the whole plane -
Use of the Genetic Algorithm instead NN. - check
with other, non susy models.
Plans - CMS IN (almost ready) by July - present
at CMS in July? - analysis note follows in
October.