1. Parameterization of EMC response for
low energy p m e
Marianna Testa
for CP + charged kaon WG
KLOE general meeting

2. Introduction
PID with EMC (TOF, energy deposit)
important for several analyses KL->pln,
K->pln, KL->p+p-, KS->pe(m?)n….
calorimeter response for p m (e) not
reliable in MC
need for parameterization from data
clustering efficiency
energy related variable (see later)
TOF (should be OK with new MC)

3. data sample
Kl3, KL-> p+ p- p0 (both with KS-> p+p-)
using 2002 data
selection: kinematics and TOF cuts
K-> p  p0 , K -> m  n
selection: easy kinematics cuts

4. Clustering efficiency
Defined as  =
N extrap: # particles whose track has been
extrapolated on the EMC
N cluster: # particles extrapolated with a cluster in the EMC associated
Extrapolation well reproduced by MC
N extrap =the normalization sample
The clustering efficiency for e, p, m is parameterized with respect to the momentum of the particle and the impact angle on the EMC

5. Cuts used for the PID
KL->pen:
kinematics cuts
on the value of the
variable ‘lepton mass’
calculated in the
hypothesis of positive
or negative track in
a decay with a pion
and a massless
particle -
KL->p+ e- n
KL->p- e+ n
Contamination<1%

6. difficult kinematics cuts on the variable ‘lepton mass’
KL->pmn:
difficult kinematics cuts
on the variable
‘lepton mass’
need for TOF identification of the
associated p
(TOF>-0.5 ns) contamination  5.5%
(<1% if TOF from m is used: calo. response)
TOF distribution
KL->p-m+ n
background -
KL->p+ m- n
KL->p- m+ n
Dt(ns)

7. KL->p+p-p0: kinematics cuts on the value of. the missing mass (Mm)
KL->p+p-p0: kinematics cuts on the value of the missing mass (Mm) |Mm-mp0|< 4 MeV
contamination  0.8%
missing mass distribution

8. Clustering efficiency
Electrons:
Efficiency
dependence
most on the
impact angle
Muons:
Stronger correlation
between momentum and impact angle

9. Pions from KL->pen and KL->p+p-p0
Different e for p- and p+ at low energies
e versus momentum e m p p- p+

10. Data MC comparison m - MC sample  7pb-1 MC response
edata/emc
vs momentum
edata/emc
vs momentum
m - e-
edata/emc
vs momentum
MC sample  7pb-1
MC response
not well reproduced
at low energies
Wait for new MC p-

11. PID with calorimeter
full calorimeter response parameterization
almost impossible
Use few significant variables, keep correlations
Total energy (E), 3d cluster centroid (xbar)
Parameterization in
10 MeV bins in
Momentum (P)
Use Neural net
both for
parameterization
and PID

12. NN basic Multi-Layer Perceptrons:
Input quantities are processed through successive layers of "neurons" IN I1 I2
N input
1st layer
N1 "neurons"
X1= C01+S Ci1 Ii;
Y11= 1/(1+e-x1)
nth layer
O1= F01+S Fi1 Yni

13. Results: electrons at [140, 150] MeV 2 - 15 - 2 – 1 MLP
Input # neu 1st 2nd layer output
Integrated 1d distribution
2d distribution of the xbar and the ratio P/E
P/E distribution
xbar distribution
xbar
P/E

14. Results: p+ at [140, 150] MeV P/E distribution xbar distribution xbar

15. Results: m+ at [140, 150] MeV P/E distribution xbar distribution xbar

16. Xbar (cm)
p+ vs e
[90-230] MeV
P/E

17. Xbar (cm)
m+ vs p+
[ ] MeV
P/E

18. Results: p + e+ separation at 200 MeV 2 - 5 - 10 – 1 MLP
Input # neu 1st 2nd layer output e
purity p
Efficiency
NN output

19. Results: p + m+ separation at 200 MeV
purity p
Efficiency
NN output

20. Conclusions Parameterization of EMC response for
low energy p m e well advanced
Clustering Efficiency:
data almost done
MC need more MC statistics
and new MC
PID:
parameterization done (to be included in
KCP-KPM lib)
good results for p/e and p/m separation