Hi,
in my last posting I noticed that
> I tried to run the using the generic MC as a b->clnu model instead of the
> mixture generic-cocktail. This is the result
>
> BRBR :0.0144+-0.0036 (it was BRBR :0.0154+-0.0031)
>
> The error is bigger since the error due the MC statistics is larger. The
> difference between the two results (here I am forgetting about the
> correlations, the generic MC is 1/3 of the mixture) is:
>
> D(BRBR) = 0.0010 +- 0.0018
>
> then the two results are compatible.
>
>
> If I add the crossfeed mc to the generic MC I get
>
>
> BRBR :0.0155+-0.0036
I looked in detail the source of this not negligible difference.
Most of the difference comes from the different shapes in Mx and from the
different fraction of events in the first bin (Mx<1.55GeV).
No effect in the denominator (factor Nsl/(Nsl+BGsl)), <1%.
Since we get more events in each bin after adding the crossfeed MC, I
tried to understand if this change in shape, and in particular in the
very first bin, is statistically significant or not.
In
http://www.slac.stanford.edu/~daniele/vub/diffgene.eps
you can find the ratio (Nevent(nocross)-Nevent(cross))/Nevent(nocross)
and the error on this quantity is calculated properly (difference in
quadrature of the two errors).
As you see, the points are well compatible with a flat distribution. Then
the Mx distribution should be just multiplied by a factor but there is not
evidence of a distortion.
Conclusions:
* we can use the MC with no crossfeed to get the Mx shape.
* using crossfeed MC is otherwise crucial in order to get the
right efficiencies
Daniele
