Hi,
this will be a little lengthy, but it I guess it is interesting for the
decision about the kinematic fitter (for both the (mX) BRBR and the
mX unfolding).
I ran the VirVubFitter jobs on generic MC using the same genMC files for
the generic MC and as data (as we said on Friday). This uses the ichep
ntuples. You can find the VirVubFitter output in the following directory
at SLAC: ~kerstin/scra.
Taking the numbers from the results.dat for the fitted numbers of events
in the first bin (0..1.55GeV) I find
S/N(mxhadfit) = 1.04
S/N(mxhad) = 0.98
so the difference is a lot smaller than what Roberto saw on data if I
remember correctly. Maybe someone could check my numbers to make
sure I did not mess it up?
Also I computed the moments for the unfolded spectra. Please find the
tables in here:
http://www.slac.stanford.edu:~/public_html/vubfit/moments_data.pdf
The systematics evaluated are the same we had on Friday:
statistical (on spectrum and detector matrix), multiplicity category
reweighting, B>Xclnu BFs, exclusive D BFs, mb and a from Belle, B>Xulnu
BFs (as done for ichep).
The first four pages have tables for the uncorrected and the bias
corrected moments as obtained on data for mxhad and mxhadfit. Comparing
the uncertainties it looks like we see quite larger uncertainties on some
of the moments using mxhad. This is coming mostly from the systematics
(see the tables where the uncertainties are split up). I tried to figure
out where this comes from. First having a little closer look at the error
bars on the left plot on page 4 of VR011105.pdf from last Friday you can
see that we have larger error bars for mxhad than for mhadfit, so this is
consistent.
The corresponding covariance matrices for the theo and the B>Xulnu
branching fraction uncertainties on the MEASURED spectrum show somewhat
larger uncertainties using mxhad than unsing mxhadfit. The difference is
larger when comparing the covariance matrices on the UNFOLDED spectra and
results in the larger uncertainties on the moments, which you see in the
tables. This cannot be due to "bad toys", since the evaluation of these
two uncertainties does not use toys.
We also see enlarged uncertainties on some moments from the B>Xclnu and D
branching fraction uncertainties when using mxhad. This does not seem to
be due to "bad toys" either. I used 1/4 of the respective covariance
matrices for tests and so not see a change in the relative size (bad toys
would become less likely when using smaller covariance matrices).
My current guess is that the worse resolution in mxhad is the reason for
the enlarged uncertainties, but I am not sure how to make a meaningful
test for this.
What I tried is to just use the detector response matrix using mxhadfit
when evaluating the B>Xclnu and D BF uncertainty for the mxhad spectrum
and I see that the effect gets smaller. So this backs the guess at least.
The following pages show the moments of the unfolded spectra using generic
MC, first for the Set 1 and the Set2 for which we showed the plots on
Friday and then for the case where we do not split up the MC sample but
rather use the full sample for both genMC and as data. Please have a look
and check if you think they are close enough for mxhadfit and mxhad.
There does not seem to be enlarged uncertainties for mxhad with respect to
mxhadfit, but these uncertainties are primarily statistical (and we do not
see a big enlargement for the statistical uncertainties on the moments for
the data either, compared to what we see for the systematics).
I compared the theo and B>Xulnu BF systematic uncertainties on BRBR for
mxhad and mxhadfit, but there I do not see a large difference:
for mxhadfit: +8.2% 7.5% (theo) + 6.5% (B>Xulnu)
for mxhad : +8.6% 7.5% (theo) + 6.6% (B>Xulnu)
I assume someone has something set up to do the same for the B>Xclnu and
D BFs since it was done for the summer results. I copied my results.dat
files for this to ~kerstin/ebr/. There is also the file giving the
corresponding mean value (mxhadcleo).
With the Run1+2 statistics those systematics we evaluated for mxhad seem
to be larger than the same ones for mxhadfit when looking at the moments
(I also see this in the covariance matrices).
At least for the theo and B>Xulnu BF systematics I do not see such a
large difference for BRBR between mxhad and mxhadfit. Maybe someone should
check the B>Xclnu and D BFs?
Cheers,
Kerstin
