Minutes of the NFS-meetings held on 17/5/94 and 31/5/94:
========================================================
 
In these meetings we had the following reports:
 
* Olaf Behnke  : Study of correlations between Asymmetry fit paramters
* Bernd Pagels : The pi0pi0 analysis; other background
* Thomas Ruf   : News about analysis with converted photons
* Marion Schaefer : K0-->3pi0 reconstruction using a 19C-fit
 
The summary of M.Schaefer will be added to the next minutes.
 
 
Study of correlations between Asymmetry fit paramters: (O.Behnke)
======================================================
 
PRECEDING REMARKS:
By fitting the asymmetry one gets not only the errors
for phi, eta, normalisation and background but also
the correlation coefficients of any two parameters.
These coefficients are interesting because they indicate
how a systematic shift in one paramter translates into
a shift of another.
The correlation coefficients
are independant of the statistics.
As all the analysis groups are more or less having
normalisation problems, especially the correlations
between normalisation and cp-paramters are of importance.
*
WORK:
A MC-simulation with 10^12 events (total sum of k0 and k0b)
was done with a relative normalisation K0b/K0 of 1.16, perfect
resolution and no background. The CP-paramters were taken
from pdg.
Fitting this "data" the errors and correlations of phi, eta
and normalisation are obtained for various START- and
ENDPOINTS of the fit.
*
OBSERVATIONS:
1.Errors:
- the endpoint of the fit can be varied from ~12-20 ts
  without changing the errors on phi and eta significantly
  (starting always at 0 ts). -> this region is unimportant!
- now fix the endpoint at 20 ts. The error on phi (eta) increases
  50% (20%) if one starts the fit at 4 ts and not at 0 ts.
  -> phi dependance is much bigger as for eta
2.Correlations:
  (endpoint fixed at 20 ts)
- The correlations between the paramters are oscillating
  functions of the startpoint.
  THEY ALL GO AT LEAST IN ONE POINT THROUGH ZERO and change
  sign !!!!
_ The coefficient for phi-normalisation is zero at 0.5 ts
  and at 7 ts. At 4 ts it reaches the maximum value of 0.6,
  which means this is an unfavorable startpoint.
- eta-normalisation is high (0.6) at 0 and goes through
  zero at ~3.
*
CONCLUSIONS:
*
- The variation of the correlations with different startpoints
  gives a tool for testing systematic
  shifts in the cp-paramters due to wrong normalisation.
*
*
ACKNOWLEDGEMENT:
The idea to study the correlations comes mainly from Prof. Gerber
who has a different technical approach to evaluate them.
Nevertheless we get agreement in the principle results and
conclusions.
*
*
SPECIAL TEST FOR PI+PI- people:
*
I tested if their approach to fit first the normalisation
from 1-3 ts and then the cp paramters from 4-20 ts cannot
introduce a certain systematic error.
The problem is that for the normalisation fit they already
use fixed input cp paramters from pdg.
So i checked with MC-simulation and fits what the dependance
is for the fitted cp parameters as function of the assumed input
cp paramters used for normalisation.
The result is that a shift of 1 pdg sigma in eta input leads to
a shift of less then 1/10 pdg sigma in the fitted cp paramters
via normalisation.
The dependance on the input phi is even much weaker.
So by this it is proved that at least this possible source
for systematic error is negligible.
*
 
 
The pi0pi0 analysis: other background: (B.Pagels)
======================================
Three other background channels are studied now:
 (1) k+k-pi0pi0:
     One exabyte of data exists now and was investigated.
     The background is absolutly negligible.
 (2) Golden + pi0:
     The production of one exabyte runs now. Results coming soon.
 (3) pi+pi- + some pi0:
     A new idea of looking at this important background channel looks
     promising: selecting events, where the kaon candidat passes the defect
     PID22. Doing this, one enriches the pionic contamination to our
     golden pi0pi0 event sample (de/dx-spectrum of the kaon candidat).
     It seems, that the pionic events in our lifetime distribution are mainly
     concentrated betweeen -2 and 6 taus (peaking at about 3 taus). The
     absolute rate of this events is yet unknown.
 
 
 
News about analysis with converted photons: (T.Ruf)
===========================================
 
1. Neural network based on 10 kinematic parameters
   tested for background reduction of 3pi0 in 2pi0 events.
   Background contamination found by neural net agrees with
   lifetime plot and asymmetry fit concerning the order of
   magnitude. After cutting the background identified by the
   NN, a fit to the asymmetry give a reasonable phi: 46\pm 12.
   Future work is to include more variables into the NN in order
   to suppress pi+pi-p0 and pi+pi- (wrong identified) background
   and to release other cuts which we are using at the moment.
 
2. Golden pi0 background to the 3pi0 reconstruction was studied.
   Result: Compared to the golden 2pi0 + fake showers, this
   background is neglible, less than 10% of the final sample.
   NN for 3pi0 analysis based on 7 kinematic variables is not
   yet efficient enough to suppress the 2pi0 background to an
   acceptable level.
   3pi0 decays can only be reconstructed above 4 taus with almost
   no background (has to be verified).
   The idea is to include more parameters into the NN.
 
   The background produced by fake showers seems to be more
   important in the analysis using converted photons (one fake
   shower enough) as in  the 6 gamma analysis (needs two fake shower).
   The cut used by Marion to reduce fake showers needs to be
   studied in detail.
 
 
 
Next meeting:
=============
 
 ***********************************************************************
* We agreed to have the next NFS-meeting on Tuesday 14/6/94 at 10.30 h  *
* in the CPLEAR meeting room.                                  =======  *
 ***********************************************************************