Comments/Responses on the D0 note 4324
"Observation of B -> D** mu nu X decays"

12/16/2004

mass resolution 8.5 MeV +- 30%
8.5 MeV
8.5 MeV + 30%
8.5 MeV - 30%
D1 width free, D*2 width 6.5 MeV





MC efficiency can be cross checked using known Br for D0 and D*.

Br(B->D*mu X) = Br(B->D0 mu X) * N(D*)/N(D0) * 1/eff (D*/D0)_MC * 1/Br(D*+->D0pi+) * 1/eff(d0mess) =
                           =  6.9% *  25680/103160 * 1/0.76 * 1/0.677 * 1/0.98 = 3.40 +/- 0.17%
PDG gives Br(B->D*mu X) = 2.75 +/- 0.19%

the discrepancy can be explained by the wide D0 mass window [1.75 - 1.95 Gev] which allows D0 decay modes
other than D0->Kpi (for ex. Kpipi0) to contribute.

indeed using tigher D0 mass window [1.8 - 1.9 GeV] and determining the # of events as integral of the signal part
in this window  (now it's important because the tails go beyond the window borders):
Br(B->D*mu X) = Br(B->D0 mu X) * N(D*)/N(D0) * 1/eff (D*/D0) * 1/Br(D*+->D0pi+) * 1/eff(d0mess) =
                           =  6.9% *  20889/94536 * 1/0.76 * 1/0.677 * 1/0.98 = 3.02 +/- 0.15%

the above arguments allow to estimate the relative efficiency D*/D0 from the data. We used the tighter D0 window
for this estimate since it is less biased by contributions of other channels.
eff (D*/D0)_data =  83 +/- 7 %  in agreement with eff (D*/D0)_MC = 76 +/- 2 %.

the narrower D0 mass window was used to estimate the D** Br since it gives smaller bias.

the result did not change much

Br(B->narrow D** mu nu X -> D* pi mu nu X) = 0.226 +/- 0.024(stat)  %    (to be compared with 0.241 +/- 0.022(stat) % in the note)
Comparison of muon pt spectrum for D0, D* & D** MC


the muon pt spectrum is relevant to estimate the trigger bias. Any difference in the spectrum could translate into different behavior
wrt trigger pt thresholds and cause different trigger efficiency. Correct way to estimate it will be to run trigsim over D0/D*/D** MC.
However at the moment we estimated that this effect must be small because :
1) Pt spectra above look similar.  Efficiency does not differ by more than 10% for any value of pt cut.
2) Br (B-> narrow D** mu X -> D*pi mu X ) has been recalculated with raised pt muon (pt muon > 4 GeV). Relative difference was
found to be 7%.
Signal was approximated by two interfering non-relativistic Breit Wigners (BW) with parameters from the Belle paper smeared with a 3 Mev sigma gaussian 
Background is taken from the linear fit of the wrong sign combination and is shown as a line on the plot.

Free parameters of this fit :  2 normalizations of BW, phase between them.
Result of the fit : N(2420)+N(2460) = 541 events, phase = 2.52 +/- 0.29, chi^2 = 65.
Number of events is calculated as integral of the signal part  in the [2.3-2.6 GeV] invariant mass window.


Free parameters of this fit :  2 normalizations of BW, phase between them fixed at pi = 3.1415
Result of the fit : N(2420)+N(2460) = 518 events, chi^2 = 68. Note that the tails are minimized at this phase (cancellation of
tails from two BW).



Free parameters of this fit :  2 normalizations of BW, phase between them fixed at 0
Result of the fit : N(2420)+N(2460) = 518 events, chi^2 = 94. Note that the tails are maximized at this phase.
The chi^2 is bad because the fit does not find the second resonance. It happens most likely because there is a dip
at 2.34 GeV in the data and the tail from the second resonance would make disagreement there more dramatic. 


the same data points with three different parametrization of the signal. The relative contributions of two BW correspond to
the result of the fit (first plot in this section). Mass and widths fixed to the Belle numbers.


apparently further work is required to understand shapes of the mass peaks - the fits as they are above does not look nice.
however in general the feeling is that the variation of the shape of the observed mass peaks due to the interference effects cannot
account for more than 10% change in the number of total events in the two mass peaks. 10% is factor of two larger than
the variation observed above in the total number of events : (541-518) / 518 = 4.4%

This procedure can be used in the future to estimate contribution from two states separately.


 all cuts are as described in the note 4324. Changes are visible but small.


Some comments/questions on the B->D** draft analysis note:

o Which quality muons were used? loose/medium/tight?

in this analysis the selections for the muons are very close to the 'tight muons' as defined 
by the muon ID group but do not coincide with them 100%. the full list of the muon 
selections is described in the note. historically we settled on the muon selections 
after quite some studies very specific for the B physics requirements. this happened before 
the muon ID group released their definitions which as we understand them were tuned more
for the high Pt physics. we are open for discussion but need to be convinced to switch over 
to their objects.   

o "both hits in SMT and CFT present" means
N_SMT > 0 and N_CFT > 0?

yes

o Requiring muons to be in the same jet means that the central
track matched was part of the Durham jet clustering?

yes

o In each case of D0, D*, and D**, it would be very useful to see
plots of cut variables, and also how peaks improve as the various
impact parameter and decay length significance criteria are
applied.  How were cut values determined?

the plots will be provided. the cuts were tuned on data for 
a smaller (47 pb-1)sample. the reoptimization/crosschecks on MC 
will be possible after larger D** MC samples are available.

Pt of pi** after final cuts except this Pt>0.3 cut



and corresponding mass plot. here and below i did not pay attention to the fit - disregard it.



significance of Lxy for B, all lifetime cuts removed. cut at > 3



mass plot with Lxy significance cut removed




B vertex chi^2. all lifetime cuts removed.  cut  c2B<25. the plot is sculptured by D** chi^2 vertex
cut at 36. 



mass plot with B vertex chi^2 cut removed



significance (IP wrt PV) / significance (IP wrt SV). all lifetime cuts removed. cut at ratio > 4



mass plot with this ratio cut removed. as you see this is the most important cut.



o Is it really useful to go all the way up to 5.2 GeV on
M(mu+ D0)?  With two more pions and a neutrino "still to go", it
would be surprising to be within only 80 MeV of the B mass.
What does the MC say for this mass?

for this analysis we started from the 'mu + tight D0' sample used
for the lifetime ratio analysis. we agree that some cuts for 
that sample can be relaxed to optimize better D** selections. 
will look at MC when it's ready.

o Will need a systematic error on the fitted D0 yield,
i.e., varying functional form for background, fit range, etc.

agree, will do

o I realize that you don't use the D* yield in the analysis
(although it can be used as a cross check), the functional
forms for the signal and background are not providing a
very good model.  There's definitely a high-mass side
tail on the mass difference peak.

yes, the fit is not good.. this tail is characteristic for the Kpipi0 channel 
which is probably sneaking to our sample because of a rather loose mass window
1.75-1.95 Gev. anyway we know how to make it look nice, will do it.

o What does MC say about the expected width of the D0 mass,
mass difference, D** mass?

will add these plots to the note. the numbers are in the right ballpark.

here they are for D**(2420): mass 2423 MeV, sigma 19.7 MeV



and for D**(2460) : mass 2462 , sigma 13.3



o When adding another pion to the D*, why is it necessary to
require large impact parameters and a decent chi-squared when
vertexing?  With the D0 required to have a significant decay
length, wouldn't just requiring a consistent vertex do?

some selections have strong correlation indeed and reoptimization of 
the cuts for the publication should address this as well.

o After adding the pion to the D*, is the invariant mass of the
whole system required to be in some range?  Show the plot of the
B mass "pseudomass" in this case (missing the neutrino and X still).
Since MC is used to get ratio of efficiencies, would be good to see
the same pseudomass in MC.

we don't have any B mass requirement. will prepare the plot.

o Need more detail about the double gaussian fit to the D**: were
the peak central values locked, widths varied, etc.?

we fixed the distance between the two peaks at its PDG value of 37 MeV. all other
parameters were left free. as we discussed in the note we did not want to release
the fit results because we'd like to understand how the interference between these two 
states affect them (but the most curious ones could figure it out from the fit 
window in the plot). For the total number of events which we quote in the note these details are 
not important since the overall normalization is a single parameter and the fit looks
adequate. 

o Could you show the D** mass peak before and after the D0 mass
constraint?

ok, here it is. looks ugly.




for your reference here is the plot for the same channel but when D0->Kpipi0. 
here the mass resolution is poor because of the lost pi0



o Will also need a systematic error on yield of D**: using
wrong-sign to estimate background, different background functional
form, etc.

ok

o One of the biggest concerns I think is the systematic uncertainties in the
epsilon = N^MC(D**)/N^MC(D0) since it involves a soft pion where
tracking efficiencies may be different in data vs. MC, and the second
pion where cuts are made on the fit chisquared.  How well does
the vertex chisquared in data match MC?

this is a good question for all our analyses using this sample. we've done some 
crosschecks on MC (like found no dependence on the jet multiplicity) and we are 
thinking about additional checks using data.

o Systematic uncertainties in e_gen are also a concern as you already
point out in the note.

better estimate should be possible with the existing MC.

o Need checks comparing data and MC for D** signals.

more MC statistics would help here..

o What about B -> D* D_s contributions where the D_s decays
semileptonically to a muon?  Br(b -> cc(bar)s) =  (22 +/- 4)%,
certainly not negligible.

agree, we have plans to generate D*D_s MC, not ready yet. the expected effect
should be small because of the softer muon kinematics but we need an estimate you are right.
this applies to all our analyses. another channel we plan to generate is B -> tau D(*)X which
should give similar yields.

o What about contributions from B_s?

the contribution from B_s to the D** sample should be negligible. we estimated the 
corresponding contributions to the D0 and D* at 2% for both samples. so 2% will be 
an upper limit for D** as well.

o As Andrzej points out: there has to be some level of prompt
charm still getting by the cuts and in to the sample.  Need charm
MC to estimate (and also need relative rate of charm production
to b production).

this is a good question - we have ideas how to estimate this from the data. for example 
we could follow the CDF Run1 procedure and compare positions of B and D vertices. if it's charm
then they will coincide. CDF estimated this contribution at a fraction of a percent level in their
Run1 oscillation analysis. of course our cuts are different so we need our own estimate but
the feeling is this will be small. 

o Is it possible to instead normalize to the number of observed
D* instead since the PDG also has Br(B->D*pi ell nu X), albeit with
larger errors, but as a useful cross check (and not having the soft pion
efficiency data/MC mismatch cause a problem in epsilon)?

yes, this should be done indeed as a crosscheck. we have ideas also how
use D* for normalization of the topologically searched D** which will
include resonant and non-resonant cases, see our SST oscillation analysis. 

o Where did you find that Br(B->D*pi mu nu X) = (0.48 +/- 0.10)%
number?  It says the PDG, but I couldn't find it in the full version
on the web (but could find Br(B->D*pi ell nu X) = (1.00 +/- 0.34)%)

it's on the 'B0/B+/Bs/b-baryon admixture = LEP results' page

o I'm not so sure that you can claim that this covers the
resonant components of B->D*pi mu nu X, the wide states are still
resonances, but are not fit to.

agree, we are covering indeed only narrow states

o Important systematic checks will be varying cuts for different
levels of background (notably lifetime cuts - tests for getting
rid of prompt charge getting by, pT cut on B candidate, etc),
refinding efficiencies, should get same result.

yes

  Looking forward to a separation of the narrow D** states!

                Regards,
                              Rick


        Hi Guennadi, Sergey and Andrei,

        Congratulations on your effectivness in producing interesting
results. I am writing to you in my capacity as a member of the B-physics
group.

        I have three types of comments:

        (1) editorial

        I suggest that you enumerate your cuts, rather than list them in
the text. Having lists of cuts at each stage: D0, D*, D** would help the
reader and could later serve as a reference when you start discussing
systematic errors. I know that you are not yet ready with systematic
uncertainties, however, even a brief, quantitative discussion, which of
the cuts are expected to influence your measurement would be useful.

will be done

        (2) analysis


        Your efficiency calculations are very dependent on MCarlo, in
particular on the slope of the pt distribution and on event multiplicity.
I know from my experience that the LO b-bbar MCarlo has a much steeper pt
dependence for b-quarks and their decay products than observed in the
data.

        I suggest that you compare observed D0 pt spectra with those
predicted by MCarlo and, if necessary, weight MCarlo events accordingly.

this is a good idea. we should try this 

        (3) result

        There is an issue of direct charm contamination in your "B"
sample, that is not addressed.

please see response to the same question by Rick above 

        A minor point.
        I am in favor of quoting a tripple product of BR as your result,
in other words, I would move Br(B*+ ->D0pi+) to the left side of Eq.(2)
This is closer to what we measure, and does not depend on the PDG value.

you probably meant D*+->D0pi+. this Br is known pretty well both experimentally 
and theoretically. for our present precision it perhaps does not matter. we can quote
both cases.

                                regards, Andrzej


***************************************************************************

relevant comments from the lifetime analysis by the same authors
Dear Guennadi, Andrei and Serguei,

Here are my questions/comments on your lifetime ratio draft v1.0.

1. Event selection:
   p.2 "...muons are required to have hits in at least two out of
   three muon chambers"
   Just to confirm: it means that you allow: ABC, AB, AC, and BC.

we require nseg>1. this will be reworded in the note

2. MC simulation
   This analysis relies on the correct simulation of the ratio r.
   We should see more discussion that the simulation is realistic.
   You have generated LO (Leading Order) bbbar events.
   This may be too optimistic. "Gluon splitting" producess the b-bbar
   system forming a single jet, with a higher track multiplicity.
   Questions:
   - What is the mean track multilicity in the jet containing the analysed
    (mu,K,pi) system in data and MC? do they agree?
   - what is the soft pion efficiency vs jettrack multiplicity?

the last distribution is included in the new revision of the note 4280 (lifetine ratio analysis).
the efficiency does not depend on the track multiplicity.

3. On p.6 you introduce Res_j, Eff_j - but later you ignore the possible
   channel dependence of the efficiency and resolution without a comment.

We check and take into account the difference in resolution 
between D0 and D* channels, it is described in the note. 
We don't see any difference in resolution between channels
contributing to D0 and D* samples, at least within available
MC statistics. The variation of resolution in reasonable limits
will be taken into account in systematics.

We take into account the differences in reconstruction efficiencies 
for all channels. this is described now in the note.


4. N - the global normalization coefficient is a free parameter.
   What is its fitted value? Comment?

Fit gives N = 0.957 +/- 0.012 as described now in the note

5. Fit results:
   There are three fitted numbers per VPDL bin. They should all be
   provided in Table 1. I would also like to see all the 8 X 3 plots.

done

6. Systematic errors - yes, they cancel out to a large extent;  I believe
   you are working on the estimates.

new revision of the note 4280 includes the systematic errors
--

we thank everybody for very useful comments!