Hi Oliver,

 may you repeate your study (I mean the MEAN, the RMS and the BIAS) as a
function of Mx instead of Mv, applying some loose cut in Mv?
 Yes, the study was performed with a tight cut in Mv. I'll give you the
result of the complete optimization.

 Thanks a lot,

 Daniele

On Wed, 2 Jan 2002, Oliver Buchmueller wrote:

> Hi Daniele,
>
> here are a few comments/remarks concerning your study.
>
> Attached to this mail you will find a plot showing the MEAN,RMS
> and BIAS(=MEAN*RMS) for the MX distribution as a function of
> the cut in the missing mass. It is in principal the same plot
> send around roughly two weeks ago but now the relevant region in Mmiss
> (around zero) is magnified. I hope this presentation is more illustrative.
>
> First of all I assume that you still apply a very hard
> cut in the Missing mass of the event (ABS(Mmiss)<0.6 or something similar)
> - please correct me If I am wrong.
>
> Such a cut would corresponds to the first few bins around zero (e.g. the
> two bins -0.5,0.0 ; 0.0,0.5) in my plot. There it is obvious that MEAN and
> RMS for the fitted and raw distribution are not so much different (still
> slightly better for the fit thought).
> This behavior, however, is easy to understand. Your very tight cut
> in Mmiss selects already events that are fulfilling the global
> constraints (e.g. Energy conservation) reasonably well.
> Hence the improvements from the kinematic fit can not be large for those kinds
> of events.
> However, as you also learn from the plot, the real improvement from the
> fit stems from regions where Mmiss is significantly away from zero. In
> this region the fit still gives you an unbiased estimate of the MEAN with
> a "stable" RMS.
> This is not anymore true for MX raw!
>
> So -to summarize- the improvements from the kinematic fit stems from
> region where the global constraints are not perfectly fulfilled
> - thats why we need a cfit :-). Therefore, for further studies (please!)
> try to soften your cut in Mmiss and figure out where the optimum value
> of this cut this is supposed to be.
>
> => Since you have now set up the "extracting machinery", it probably would
>    be a good idea to perform an optimization of the Mmiss cut by just
>    scanning the expected error on Vub as function of Mmiss.
>    Similar to what I have done for the MEAN and RMS
>    of the Mx distribution.
>
> Looking forward to see this results from you.
>
> Regards,
>
> Oliver
>
>
> On Sun, 23 Dec 2001, Daniele del Re wrote:
>
> >
> > Hi all,
> >
> >  using the last version of recoilNtp and abcFit I started to look at the
> > effect of the kinematic fit on the Vub-Vcb separation.
> >  In:
> >
> >  http://www.slac.stanford.edu/~daniele/vub/mx.gif  with kin fit
> >
> >  http://www.slac.stanford.edu/~daniele/vub/mxfit.gif  without kin fit
> >
> >  you see the "classic" plot for each component after all cuts
> > with and without the kin fit (B0 Cocktail MC). The D/D* peak has more
> > events and the Vcb tail at low Mx seems to be less evident after the fit
> > but the difference is not dramatic.
> >
> >  This observation is confirmed by the fit result for the BR(btoulnu)
> > (error basically unchanged) and by the plot:
> >
> >  http://www.slac.stanford.edu/~daniele/vub/mxfitvsmx.gif
> >
> >  that shows Mxhadfit vs Mxhad after all analysis cuts. Below 1.5 GeV
> > Mxhadfit is almost identical to mxhad and this implies that the separation
> > Vub-Vcb remains the same.
> >
> >  Further MC studies are needed to understand the leakage at low Mx for
> > Vcb (Guglielmo is studing MC truth for Vcb in Mx<1.5GeV).
> >
> >  Daniele
> >
> >
> >
> >
> >
> >
>