Hi all,
I performed the optimization of the cut on the integrated purity of
Breco sample. I ran on 20fb-1 of Run1 data. I extracted the yields and
the the number of background events fitting the mes plot after all the
analysis cuts. Here you find the plot of S/sqrt(S+B) versus the cut on the
integrated purity for each seed using a Cristall ball + argus fit:
http://www.slac.stanford.edu/~daniele/vub/Dstaropt.gif
http://www.slac.stanford.edu/~daniele/vub/Dcopt.gif
http://www.slac.stanford.edu/~daniele/vub/Dstar0opt.gif
http://www.slac.stanford.edu/~daniele/vub/D0opt.gif
(due to a bug in the tables D0's have S/sqrt(S+B) versus the
purity of the single mode instead of the intergrated purity)
and the optimal cut is (I put the yield and the background events with
that cut):
Dstar 44% S 89
B 10
Dc 36%(no cut) S 117
B 50
Dstar0 38%(no cut) S 168
B 48
D0 35%(no cut) S 170
B 70
total S 544
B 178
S/sqrt(S+B) 20.25
with 50 fb-1 we will have S ~ 1500 and B ~ 500 and comparing with the MC,
assuming a the PDG BR(b->ulnu), we will have an error on Vub ~14%.
With a gaussian fit for the signal I got similar optimal cuts, see
http://www.slac.stanford.edu/~daniele/vub/Dstaropt_gauss.gif
http://www.slac.stanford.edu/~daniele/vub/Dcopt_gauss.gif
http://www.slac.stanford.edu/~daniele/vub/Dstar0opt_gauss.gif
http://www.slac.stanford.edu/~daniele/vub/D0opt_gauss.gif
Once again the choice of the fit function for the signal introduce large
systematics. We need to study MC mes shapes more in detail.
I also noticed that the efficiency of the cuts is different from the MC,
as already found by Urs. With an integrated purity of 50% I get an
efficiency of 0.77%. On Cocktail MC I get an efficiency of 1.02%. Then we
have a discrepancy of 25%.
Finally with a cut of 50% on the integrated purity, now I get 47000
events and the purity is lower than 50%. We have to investigate this.
Daniele
|