Dear Bjorn,
I just read "Vertex-determination from TPM1". For the cluster
algorithm I think that a better estimate of the uncertainty of the vertex
is the error on the mean of the gaussian rather than the RMS.
Imagine that you have 2 events with the same set of tracks. However
the first event produces only 3 clusters per track while the second
produces 10 clusters per track. Both events should have the same RMS of
the vertex histogram but the second event will have many more events and
so a much smaller uncertainty in the mean.
For the tracking algorithm you use a cut of Chi**2<5. It seems to me
that this cut should depend on the number of degrees of freedom in
your fit. I think it would be better to use the confidence level. This
is a statistical measure of the likelyhood that a fit to a
distribution with gaussian errors and N degrees of freedom would have a given
Chi**2. There was a CERNLIB routine in FORTRAN whose syntax was
conf_level = PROB(Chi**2, N_Degrees_of_freedom)
I'm sure this must exist in ROOT.
If you have only good events in your sample a plot of confidence level
should be flat. However "bad" events, eg where this is no vertex, will tend
to have a larger Chi**2 and show up a a peak at low confidence level.
You can then cut at confindence levels just above the peak. Another nice
feature is that if you cut on confidence level .15 say then you know
that you have lost 15% of your "true" events as well as the garbage.
It is important to know the efficencies of your cuts in order to costruct
dN/dY etc.
Yours Michael
:
: Michael Murray murray@comp.tamu.edu
: Cyclotron Institute, Texas A&M University, TX 77843-3366
: Telephone (979) 845 1411, FAX 1899
This archive was generated by hypermail 2b29 : Tue Sep 12 2000 - 12:17:49 EDT