1. Search for Lepton Flavour Violation in the decay  →  @ BaBar
ELISA MANONI
INFN & UNIVERSITY OF PERUGIA
LNF SUMMER SCHOOL
MAY
LNF SUMMER SCHOOL 2005
MAY
ELISA MANONI
INFN & UNIVERSITY OF PERUGIA

2. INFN & UNIVERSITY OF PERUGIA
Outline
Theoretical Motivations
The BaBar Detector
Data Sample
Cut-Based Event Selection
Likelihood Approach and
Neural Network Method
Results and Statistical Interpretation
Conclusions
LNF SUMMER SCHOOL 2005
MAY
ELISA MANONI
INFN & UNIVERSITY OF PERUGIA

3. Theoretical Motivations (1)
Lepton Flavour Violation observed in athmosferic and solar
neutrino oscillations
LFV processes in the charged lepton sector predicted
within Standard Model scenario extended to include finite mn
lepton-neutrino couplings
strongly suppressed
because of
mni ~ eV
l ni l2
LNF SUMMER SCHOOL 2005
MAY
ELISA MANONI
INFN & UNIVERSITY OF PERUGIA

4. Theoretical Motivations (2)
LFV processes with higher BR evidence of physics beyond SM
Prediction of in many SUSY models
Within SUSY SU(10) Right Handed  model:
(depending on model parameters)
Predictions with l1 l2
l ni l2
_Yukawa coupling
of the order of
top_Yukawa coupling
LNF SUMMER SCHOOL 2005
MAY
ELISA MANONI
INFN & UNIVERSITY OF PERUGIA

5. INFN & UNIVERSITY OF PERUGIA
BaBar Detector
B-factory: pairs of mesons are produced almost at rest in the CM frame from: Y(4S) → B+B-, B0B0
Asymmetrical: the CM is boosted forward by bg ~ 0.55
1.5 T Magnet
EM Calorimeter
6580 CsI crystals
e+ ID, p0 and g reco
LNF SUMMER SCHOOL 2005
MAY
ELISA MANONI
INFN & UNIVERSITY OF PERUGIA

6. INFN & UNIVERSITY OF PERUGIA
Sample
229.4 fb-1 data collected by BaBar
center of mass energy ≈ 10.5 GeV
s(e+e-  t+t-) ≈ 0.89 nb, similar to s(e+e-  bb)
~ 360M t-pairs
Signal events:
e+e-  t+t- with t mg
IFR-identified m
EMC-reconstructed g or reconstructed
g->e+e- (converted g)
Main background sources:
e+e-  t+t- with t  anything else
e+e-  m+m-g
e+e-  qq
e+e-  e+e-
e+e- B+B-
e+e- B0B0
LNF SUMMER SCHOOL 2005
MAY
ELISA MANONI
INFN & UNIVERSITY OF PERUGIA

7. Cut-Based Events Selection
Charged track:
none of the BaBar e-Id selectors allowed
Neutral cluster:
Photon-quality cuts and p0 veto for EMC
reconstructed g
Photon invariant mass and opening angle cuts for
converted g
signal side m- g t- e+ e-
1-1 topology:
1 charged track not identified as m
3-1 topology:
3 charged tracks
Mrecoil < 2 GeV/c2, |cosqCMmiss| <0.95,
pTCMmiss>300MeV/c
tag side t+ nt
Preliminary cuts on |Minvmg-mt|, DE≡√s/2-Ecmmg , cosqcmmg
Rejection of “fake missing momentum event” (e+e-  m+m-(n)g ):
EEMC cluster > 1 GeV
#IFR hit layers > 5 m g g m
LNF SUMMER SCHOOL 2005
MAY
ELISA MANONI
INFN & UNIVERSITY OF PERUGIA

8. Kinematic Fit and Final Selection
Kinematic Fit in order to remove background and
improve mass resolution
two constraints:
mg vertex
beam energy
cut on distribution of c2 variable
Final Selection: cuts optimization
to improve
the ratio
Efficiency = (8.46 ± 0.85)%
Background events:
expected ± 1.15
observed 12 (111-1,13-1)
1-1 topology
3-1 topology
CM m momentunm
GeV/c
GeV/c
Minimum CM g energy
0.9 GeV
minimum pT CMmiss
1.2 GeV/c
0.6 GeV/c
cosqCMmiss
-0.90 – 0.80
maximum Mrecoil
1.65 GeV/c2
1.75 GeV/c2 DE
-0.06 – 0.10 GeV
-0.09 – GeV
Mmgfit window
±2.0 s (Mmgfit)
LNF SUMMER SCHOOL 2005
MAY
ELISA MANONI
INFN & UNIVERSITY OF PERUGIA

9. INFN & UNIVERSITY OF PERUGIA
Likelihood method
Four variables chosen: DE
pT missCM
a ≡fmg-ftag-p
DpmgCM≡|pmCM|-|pgCM|
to set up a likelihood ratio variable (R) defined as :
signal
background
Number of events determined from
an Unbinned Maximum
Likelihood fit to the variables:
Mmgfit, R for topology 1-1
Mmgfit for topology 3-1
Efficiency = 10.71%
Background events:
expected 28.5 ± 2.3
observed 27 (261-1,13-1) 
LNF SUMMER SCHOOL 2005
MAY
ELISA MANONI
INFN & UNIVERSITY OF PERUGIA

10. Neural Network approach
5 discriminant variables as inputs for NN:
Mmiss
pCMtag
cosqH
pmissT
mn2
Different selection criteria for different
tag modes:
ttag -> enn
ttag -> egnn
ttag -> mgg
ttag -> hn
ttag -> h≥1p0n
ttag -> 3h≥p0n
Unbinned Maximum Likelihood Fit to Mmgfit
Efficiency = (7.42±0.65)%
Background events:
expected 6.2 ± 0.5
observed 4
LNF SUMMER SCHOOL 2005
MAY
ELISA MANONI
INFN & UNIVERSITY OF PERUGIA

11. Results and Statistical Interpretation
Likelihood Method
Definition of a test-statistic variable:
Evaluation of Q in binned 2-dimension
Mmgfit vs R plane
Measurement of Qobs on the data sample
Generation of toy MC experiments with an
arbitrary value, N’signal, of signal events
Evaluation of:
CLS+B=PS+B(Q<Qobs) (pure frequentist method)
CLB=PB(Q<Qobs)
CLS=CLS+B/CLB (modified frequentist method)
Iterating by varying N’signal till 1-CLS is 90%
Neural Network Method
Generation of 10k toy MC with
Poisson-distributed number of
background (fixed)
and segnal events (s,floating)
Iterating by varying s till fitted
number of signal events greater than
that observed
# of signal events
Sensitivity
Upper Limit
13x10-8
6.8x10-8
Sensitivity
Upper Limit
CLS+B
11.9x10-8
9.4x10-8
CLS
16.1x10-8
14.6x10-8
LNF SUMMER SCHOOL 2005
MAY
ELISA MANONI
INFN & UNIVERSITY OF PERUGIA

12. INFN & UNIVERSITY OF PERUGIA
Conclusions
229.4 fb-1 data analyzed
Cut Based Selection +
Upper 90% CL:
Neural Network approach
Likelihood method
Improvement with
respect to previous
measurements
Some SUSY scenarios
excluded by this
measurement
2002
BaBar 2005
LNF SUMMER SCHOOL 2005
MAY
ELISA MANONI
INFN & UNIVERSITY OF PERUGIA