Hello all,
we made some plots concerning the mX resolution in signal MC in the
ntuples we used for ICHEP. Below are the links. I recognize this became
quite lengthy and hope that is not too deterring... It would be
interesting to compare this to what we see with the VtxTreeFitter, I
guess.
The plots are done separately for the nonres and the mix MC, where for mix
we use the mix as it is in the files, meaning that there is no nonres
component below 1.264 GeV. So for the mix the first four bins we show here
show what happens to the resonances.
For all those plots we applied the following cuts:
pcms<1, mm2<1, mes>5.27, (brecoqual==1brecoqual==5brecoqual==7)
The latter is the cut on the Breco quality as we have been using it for
more than a year and cleans up the mes sideband well (the mes cut does the
tiny rest).
As truth variable we used mxhadgenwoph, which we use for the unfolding.
The difference to mxhadgen is that photons which are B daughers
according to MC truth and are likely to come from the electron have been
taken out (looking at the angle between the photon and the e).
You can find the plots here:
http://www.slac.stanford.edu/~kerstin/vubfit/mXreso16_vubmix.eps
http://www.slac.stanford.edu/~kerstin/vubfit/mXreso16_vubnre.eps
http://www.slac.stanford.edu/~kerstin/vubfit/mXreso16_genbch.eps
http://www.slac.stanford.edu/~kerstin/vubfit/mXreso16_genbnu.eps
http://www.slac.stanford.edu/~kerstin/vubfit/mXreso16_logy_vubmix.eps
http://www.slac.stanford.edu/~kerstin/vubfit/mXreso16_logy_vubnre.eps
The latter ones are the same as the first ones up to having a log yaxis
rather than a linear one.
Each of the files has 12 plots. They differ by the region of mxhadgenwoph
that is considered. The upper left plots has mxhadgenwoph<0.31, the upper,
second left plots is 0.31<mxhadgenwoph<0.62 and so on. The bottom right
plot is the sum of the three other bottom plots.
This is the 310 MeV binning we are using for the unfolding.
The black histos are mxhadmxhadgenwoph and the orange dash histos are
mxhadfitmxhadgenwoph on the xaxes.
In these plots it looks as if the tails in the distributions do not really
get a whole lot better by using the kinematic fit, but the region close to
the peak at 0 gets somewhat narrower.
We looked at two features in more detail.
First there is a long tail to the right in the vubmix for
mxhadgenwoph<0.31 (top left plot). We checked the Gvxbtyp of the events in
this tail and it turns out all are B>pi l nu decays (which is exactly
what they should be since there is no nonres down here in the mix, so this
is just a sanity check).
Here is one more plot:
http://www.slac.stanford.edu/~kerstin/vubfit/mXpitail.eps
We use the same cuts as given above and focus on the region
mxhadgenwoph<0.31 and (mxhadmxhadgenwoph)>0.1 for the top plots,
(mxhadmxhadgenwoph)<0.1 for the bottom plots.
We plot the following quantity:
nchgchgdaugen:nneuneudaugen, so the yaxis has the difference
in numbers in reco and true charged daughters of the X and the xaxis the
difference of reco and true neutral daughters of the X.
The top left plot is for charged B (Gvxbtyp<0) (and the tail region), the
top right plot is for neutral B (and the tail region), the bottom left
plot is for charged B (and not the tail region) and the bottom right plot
is for neutral B (and not the tail region).
This tells us that for charged B in the tail region, we are most likely to
pick up an additional neutral particle (not from the X) or to lose a
neutral and pick up an extra charged particle. For B0 it looks like we
pick up extra neutral particles.
It would be interesting to know where these extra neutrals come from 
unrecovered bremsstrahlung, FSR, from the true Breco?
Second we had a look at the small peak we see left of the peak at zero in
the third plot (0.62>mxhadgenwoph>0.93). It contains 39 events, 36 of
which are B>rholnu according to Gvxbtyp. Here are the plots of
nchgchgdaugen:nneuneudaugen for this peak region:
http://www.slac.stanford.edu/~kerstin/vubfit/Peakinrhowindow.eps.
The left plot is for charged B, the right plot for neutral B.
This tells us that for neutral B in this peak we are most likely to miss
2 neutral daughters (or miss both a charged and a neutral daughter) and
for charged B in this peak we are most likely to miss one of the charged
daughters.
Finally, here are the bias and resolution plots we showed earlier, but
these are in this 310 MeV binning, using the plots given above:
http://www.slac.stanford.edu/~kerstin/vubfit/mXreso16_nre.eps
http://www.slac.stanford.edu/~kerstin/vubfit/mXrms16_nre.eps
http://www.slac.stanford.edu/~kerstin/vubfit/mXreso16_mix.eps
http://www.slac.stanford.edu/~kerstin/vubfit/mXrms16_mix.eps
I had a very brief look at generic MC from the ICHEP ntuples to see if I
find the same broad structure as in Roberto's plots. I see this structure,
but it seems that these are not B>Xlnu events, at least they have vcb==0
and vub==0. Do you see the same, Roberto?
Thanks for your patience with this long mail,
Kerstin
