• from Vivek Jain
  

Hi Vivek,

Find below the answers on your questions:

> 1) Guennadi wrote
> 
> "The main difference is that now we are able to estimate
> all branching fractions of B->D** mu nu from the existing data,
> which were not available at the time of CDF paper"
> 
> Is this the same as what you say below?
> 

Yes, it's the same. We used inclusive measurements B->D** l nu,
which became available recently (ALEPH, DELPHI). These measurement
include both resonant and non-resonant production.

By the way, we can repeat this measurement in Dzero with much
higher statistics. At some point we will need to do this,
because sooner or later this systematics will become dominating.
Anybody iterested? May be somebody has a free student? 
It could be a nice subject for thesis.

> 2) Also, when you say use iso-spin, do you mean you use it to correlate
>    B0 -> D**+/mu-/X to B- -> D**0/mu-/X?
> 
> Do you set Gamma(B0->D**+/mu-/X) = Gamma(B- -> D**0/mu-/X)? Or, 
> something else?
> 
> It is not obvious you can use iso-spin, since it is not a conserved 
> quantity in weak decays?

Yes, we suppose this equality, corrected, in fact, for lifetimes.
It is a usual suggestion in all such measurements, it is included
in our simulation and it is valid for D and D* within errors.
Here the isospin symmetry is supposed for the spectator quark.

I have a couple of questions (details).
> 
> 1) On page 7 of your note, you make the comment that you use 
> Br(B->mu+/nu/D**)=2.7+-0.7% (PDG) to estimate
> Br(B+ -> mu+/nu/D**0). However, looking at the equations, it is not 
> clear how you use this input.
> 
> you say, Br(B+->mu+/nu/D**) = Br(B->mu+/nu/X) - (B->D*) - *B->D)
> 
> So, where exactly does 2.7% come in?

We use both estimates of Br(B->mu+/nu/D**) to improve its error.
I checked in PDG that they came from independent sources.

> 
> Also, why do you need to assume that B0 -> D**- = (tau0/tau+)*B+ -> 
> D**0)? Why don't you use the above equation
> separately for B0 and B+?

> 
> 2) In eqn. 5, what exactly does (Integral_i dx^M) imply? Is this where 
> you are averaging the data points within one of
> your VPDL bins?

> 
> 3) What quantities change when you go from one VPDL bin to another?
> 
>       a) I would guess Eff_j(x) will change?
>       b)When you integrate over K to get  P_j(x), what does dK 
> represent? Is this over pT(D+mu) or x?
>       c) Anything else?

thanks for the detailed note. I took a quick look at it and here are some
questions/comments.

1)  Section 4.

   a) Why do you make a cut on the total mom. of the muon? What does it buy
you?

this was a part of tuning of our cuts we went through in the very beginning to get a clean D0 sample. 
we found it helps to clean up the sample. it's not the main cut but it helps. 

   b) Please provide a brief description of the Durham algorithm, as well as
a reference.

will do

   c) How do you set the values for the various cuts in section 4, e.g.,
chi2 of the D, B vertex, impact param. significance cuts?

we optimized the cuts to have the best signal to background while not loosing 
much statistics. it was not strictly speaking the S/sqrt(B) optimization (and it's not really clear 
how to apply it for this analysis) but something along these lines 

2) Section 5.

  a) What do you mean by back-to-back b/bbar production? Did you ask for
MSEL=5 or =1 in your samples?

MSEL=5

3) Section 6.

  a) In eqn. 2, I don't understand the implication of the constant C. Since
you do a fit to the mass peaks, why do you
need a fudge factor?

for the D* sample there is a combinatorial background under the D* peak which needs to be subtracted. 
this is how we do it. since we fit the corresponding D0 peak (and not the D* peak) for the D* sample 
we have to do it this way. otherwise this background will be present in our estimate of the number 
of D* events. if we would fit the D* peak we would not need to do it but we've chosen to fit the D0 
peak for both samples for the reasons described in the note. 

4) Section 7.

  a) What is the global constant N in eqn 7?

N is the global normalization constant which is a free parameter in the fit. in principle it 
should be 1 by construction. leaving it floating uses the fact that only the "r" carries 
the lifetime information and may help to reduce some biases, for ex. from the fitting procedure 
as explained below in the Brendan's comments. we will include the fit value of N in the note.

5) Section 8.

  a) Didn't you consider Bs -> Ds(**)/mu/X decays? The two high mass Ds**
states decay to D(*)K.

done. we prepared all necessary information and included the Bs contributions into the fit - 
as expected the effect is small. will write it up in the next revision of the note.

  b) In section 8.3, you make a comment that the cut of 5 GeV on pT_D0
reduces the ccbar contamination. Could you quantify this?

we will try to quantify it.

 c) In section 8.4, could you provide the number of the ratio in simulation
for the -ve lifetime bin? It is not clear from the discussion if you used
the resolution from MC in your fit. Could you clarify?

we used the MC resolution in the fit. the measured resolution has been used to estimate 
the corresponding systematic error. the study of the resolution has been presented at the 11/13/03 
Bmixing meeting by Guennadi. Table 1 has the ratio in the negative lifetime bin.

6) Section 11

   a) You should also mention the single best measurement. Belle's latest
number is 0.091+-0.023+0.014. Let's see what our syst. error will be?

will do

vivek

• from Brad Abbott
  

• 2 more additional comments.
  
  I section 8.2 K-factors.
  You make the statement that the K-factors cannot exceed 1.
  But if you look at Figure 4, there are some events with the K-factor
  larger than 1

• from Brendan Casey

One more thing:

I just noticed that your bin width increases with flightlength.  My 
understanding is that when this is done, one usually changes the x 
value from the center of the bin to a location that is weighted by 
where the data is in that bin.

I know that when you display the results of an unbinned fit as an 
overlay on a binned plot, it really can make a big difference in 
whether or not the overlay agrees with the wider bins.  
I dont know what effect it has on the fitting proceedure here.  But it 
would be worth investigating.

Brendan

it's an important point indeed. we integrate over the bin width so all dependencies on VPDL should be taken into account correctly.


As a cross check to demonstrate that there is no flight length 
dependence to the efficiency, you could look at the relative yields of 
cascade to lambda in the same bins you are using.  

I believe you already have these samples, right?

we don't have these samples ready but anyway we think it's a good idea. we'll look into it.

Hi Guys,

I read the note.  Again, I'm surprised we can do so well with B+ and 
Bd.  I never thought we would be competetive with these modes.

I think the analysis needs a few more things before it goes public:

1) Apply the analysis to MC and verify that the output equals the input.

will do

2) At least have an estimate of the error due to Bs contamination as 
Vivek points out.

done, see above 

3) Try and get an idea of the systematic error from the background 
model in the fit.  Since the D0 sample has larger background than the 
D* sample, I expect this systematic may not cancel.  I suggest some 
simple things like changeing from a polynomial to an exponential...
What would be nice would be to have one proceedure that clearly 
overestimates (or understimates) the background in each sample and then 
seeing how much the final answer changes if you apply this fit to each 
bin in both samples.

this will be an important systematic error - we agree with you. we've already done this study, 
tryed several different fitting functions and showed a preliminary contribution to the 
systematic error in our ADM talk on 1/9/04. we'll continue to study this. note that the fact that we take the ratio 
and leave the overall normalization N floating in the fit will take care of this bias to some extent. 
imagine you systematically underestimate your D* sample in all points by 5%. then this will be taken out 
by the floating normalization constant N which would come out as 1.05 instead of 1.00

I think before it is ready to publish you need:

1) MC study that shows input equals output with more than one set of 
values for the input lifetime.  If you generate at least 1 more B+ and 
1 more Bd that at least gives you 4 points for the ratio.

could be useful. we'll have a better feeling about this when we do it for one point.
to do this we don't really need to redo the MC - it will be enough to use it with appropriate weights.

2) Full list of systematics.  As I mentioned, its probably best to 
contact the HFAG and ask them what their standard classification of 
error categories is.  It may even be on their web page.

will do 

3) Overall cross-check in data to show that their is no flight length 
dependent efficiency for the slow pion.  As I suggested Cascade over 
lambda in the bins you are using is probably good.  Its not the same 
vertex but since its fully reconstructed, you know that the only thing 
you are testing is the slow pi efficiency.

answered above - will try

Two comments:

When people use D*'s to calibrate particle ID at the 4S, they usually 
make a cut that the kaon and pion from the D do not have the same 
momentum.  If they have the same momentum, the wrong mass assignment 
still peaks under the D.  We also typically made the cut that if you 
reverse the mass assignment, its outside the D window.  These are just 
some cuts you may want to try.

no comment so far. we think what we've done for the event selection is good but we'll think about it.

Also, I dont understand what you are doing in equation 2 when you 
subtract the N_i^W from N_i^R.  I'm not sure this is necessary.  But we 
can discuss it.  I dont think thats something that can be explained in 
email.

answered in the Vivek's part

I think you should shoot for the bottle of wine Jianming is offering 
for the first publication.

what wine is he offering by the way?!

• from Rick Van Kooten
     As far as I know, the document at:
  
  http://arxiv.org/abs/hep-ex/0112028  (long, > 90 pgs)
  
  is the latest from the HFAG group regarding:
  
    i) agreed on common inputs (lots of stuff for D* and D**), and
  
   ii) dominant systematic errors
  
  although both categories are definitely geared towards e+e- results.

we thank everybody for the comments - we think it will help to improve this analysis!