Minutes of the NFS-meetings held on 22/2/94:
============================================
Reports from the following people are summarized below:
-------------------------------------------------------
- Marcin Wolter
- Olaf Behnke
- Bernd Pagels
Monte Carlo generation for P25: (M.Wolter)
===============================
New organization of the MC production on the ETH cluster:
- special MC queues with a very low priority
- MC is running constatly in the background without disturbing
other users
In order to produce golden 2 PI0 and 3 PI0 efficiently the following
generation conditions are proposed:
For the events generated with flat lifetime:
- true K0 lifetime 0-50 tau_s
- R of the true secondary vertex 0-75 cm
- Z of the true secondary vertex -75 cm - +75 cm
For all events
- true momentum of the charged kaon greater then 350 MeV
- if possible asking for all photons pointing to the calorimeter
Present production:
- golden 2 PI0, exponential lifetime distribution,
all true photons pointing to the calorimeter.
- golden 3 PI0, flat lifetime distribution (no selection on the
generation level asking for four photons pointing to the calorimeter
would allow to use this sample also for the study of 3 PI0 background
in the 2 PI0 analysis).
phi00-significance tests: (O.Behnke)
=========================
I tried to find out, if by varying cuts on
- loopmethod
- k0-momentum
- showerangles
you can improve the significance on phi00 (that means to determine
it with smaller error).
For this a Monte-Carlo-simulation of k0- (k0bar) rates and asymmetry
corresponding to 5-million events was made.
This is roughly ten times the data we have now, and may
be in the order of magnitude of what is reachable at the end.
In the simulation the rates are smeared with resolutions I took
from full Geant MC-data.
I added a constant background of 3 permille.
The main procedure was now:
1.apply cuts on the full MC-data
-> gives improved resolution
and decreased statistics in this data
2. simulate the rates and asymmetry
with this improved resolution
and decreased statistics
3. fit the asymmetry respective
to phi00 and watch the fit-
error on phi00
results:
Starting with a standard-cut (no cut on loopmethod,
pk0>400, cut on showerangles pulltsh>-1.)
you get an error of 5.1 degrees for phi00.
- the loopmethod cuts show a minimum for the error on phi00 of
4.5 degrees at dtligr<0.5
- the pk0-momentum cuts above 400 MeV dont improve anything,
above 550 MEV the error goes up
- the showerangels cut show a weak minimum
at the standardcut. without cutting on them
you get an error of 5.4 degrees for phi00
Afterwards I tested, what you could gain, if
you could cut the resolution at 10,8,6,4,2,0 Ts
without loosing any statistics.
So you have to cut at least at 6 to see a big improvement (4.5 degrees).
At 0 you reach the statistical limit of 2.6 degrees.
conclusion:
In the sets of tested variables
there are no obvious cuts, that improve the
significance majestically, but for the loopmethod
an improvement is clear and the showerangles are
also important because they make the asymmetry
much more visible at high-lifetimes and reduce
the background there.
Possible ways to study & improve the lifetime resoulution:
==========================================================
(B.Pagels, O.Behnke)
In the past the lifetime resolution was studied with a mini MC. The
availibility of full CPGEANT simulations offers now the possibility to
investigate several topics in more detail:
(1) Which variables contribute mostly to tails in the lifetime
resolutions ?
(2) What is responsible for the shape of the lifetime resolution:
the non-linearity of the 6c-fit constraints or the non-gaussian
distribution of errors of measured variables or both ?
To improve the lifetime resolution and to increase statistics there
exists the natural way of tuning our cfits. l
(1) Primary vertex and tracks:
o Use of the helix method
o An additional constraint on the z-component of the primary
vertex (see Saclay 1c-fit vs 2c-fit)
(2) Photon impact points:
o Unknown resolutions (e.g. what about PID conversions ?)
o Take resolutions from MC (or even data) ?
(3) Photon energies:
o Switch from (E,sigma_E) to (NHIT,sigma_NHIT)
o But cut-off at low hit multiplicities: how to handle this
error distributions in cfit correctly or 'sufficiently' ?
(4) Photon shower directions:
o '14c-fit'
o But also here some asymmetric error distributions exist !
We have to investigate which of the described cfit improvements gives the best
performance improvement.
NN's in the pi0-pi0 analysis: (B.Pagels, A.Chevrier)
=============================
A NN was build to replace the q'-cut. The network architecture is the
following:
3 layers
3 input nodes
3 hidden nodes
1 output node (output between 0 and 1)
The input consists of the three PCA variables, which are also used to
calculate q'. The output for bad events lies close to 0, for the good events
close to 1. The generalization performance of the net is about 93.5%.
The NN gives a slightly better efficiency for good pairings compared with
the q'-cut. It offers also the possibility to re-pair some of the
bad pairings (this must be studied in more detail).
Next a net for the identification for 3pi0 background will be constructed,
using exactly the same input.
A.O.B.
======
+ We agreed not to change anything to the P25 NFSPROD (production
program) as compared to P24.
+ People should think about what we should eventually change in the
FILTERING of the coming run (P26) (e.g. definition of photon) !
+ I would like to remind you, that all the NFS-MINUTES are also
available via WWW (with a delay of a few days).
+ All NFS-NOTES (like E-calibration, alignment ...) can be viewed on
WWW as PS files (if you have a viewer for that) including the figures.
Next meeting:
=============
We had foreseen our next NFS-meeting on Tuesday 8/3/94 at 14.00 h
in the CPLEAR meeting room, but from now on the CPLEAR group meeting
takes again place on Tuesday afternoon (and not anymore on Wednesday
afternoon as in the last months).
***********************************************************************
* I therefore suggest, that we either meet on Tuesday 8/3/94 in the *
* morning (at 10.30 h) or on Monday afternoon 7/3/94 (at 14.00 h) ! *
* Please let me know, if one of the 2 dates (or in the worse case even *
* both dates) do not fit to you !!! *
***********************************************************************