here it is
- Simultaneous fit of D0 and D* samples
Here are the results of the combined (D0+D* samples) fit for opposite
side muon tagging:
1) purities for B0 and B+ are different:
FCN = 7.261267
dM = 0.495362 +- 0.0453544
pur0 = 0.74696 +- 0.0239902
pur+ = 0.695649 +- 0.0185146
2) purities for B0 and B+ are the same:
FCN= 9.677110
dM = 0.498221 +- 0.0496028
pur0 = pur+ = 0.715805 +- 0.0131498
the results are consistent and the changes are small
Then we worked our way through the note (comments with a - are author replies except
where specified differently):
Action items are in purple
-event selection:
*Kazu: some cuts can introduce a lifetime bias, impact on systematics?
-affects statistics in different bins in Fig26, but should affect both
oscillated and non-oscillated in the same way
-affects eff. of D0 selection (lifetime), verify if this has the same impact
on mixed and unmixed samples - check with MC
indeed some cuts for D0 selection have lifetime bias, however as we take ratio of tagged and not-tagged
events (and the tagging is not lifetime biased) it should cancel out. did did not get to the MC yet to verify this but
control samples without expected oscillations (B+ samples) behave resonably.
*Kazu: fit functions
-working on trying different fits, will include effect of choice of fit
function in systematics
- tryed 3 other approaches, this indeed gives the biggest uncertainty so far
- Systematics from fitting procedure
Changed procedure to determnine # of D* events in the mass difference peak. Used # of events in the window [0.141-0.149 Gev].
Bkg was estimated as # of events with wrong sign and subtracted. Results
chi^2 = 1.15/5
dM 0.47595 +/- 0.05271
pur(IST) 0.73204 +/- 0.02194
Asymmetry :
Bin: 0 0.37037 0.161499 0.453449
Bin: 1 0.452991 0.0659028 0.445257
Bin: 2 0.3829 0.0595146 0.404907
Bin: 3 0.383784 0.0764083 0.329159
Bin: 4 0.209677 0.0947979 0.228584
Bin: 5 0.149425 0.119254 0.116679
Bin: 6 -0.109677 0.0886623 -0.0803413
Variation of the above with different normalization for the bkg. the bkg is taken from the wrong slow pion sign distribution
as before but there is a normalization factor applied to the bkg. the normalization factor is determined as ratio of bkg off
mass difference peak in the window [0.16 - 0.18 GeV].
chi^2 = 2.025985/5
dM 0.46090 +/- 0.05358
pur(IST) 0.73262 +/- 0.02228
Asymmetry
Bin: 0 0.451524 0.201277 0.454791
Bin: 1 0.443057 0.0655189 0.447077
Bin: 2 0.382375 0.0599601 0.409042
Bin: 3 0.424946 0.0850637 0.337388
Bin: 4 0.189265 0.0952682 0.241627
Bin: 5 0.216802 0.138732 0.133975
Bin: 6 -0.0917272 0.0883765 -0.0629367
Changed D* fitting function to "gaus + (1-exp+p1)" (originally "gaus+sqrt")
chi2 = 2.205614
dM 0.50611 +/- 0.05482
pur(IST) 0.73116 +/- 0.02179
Asymmetry
Bin: 0 0.304334 0.12838 0.45132
Bin: 1 0.44119 0.0683116 0.442121
Bin: 2 0.400782 0.0584471 0.396915
Bin: 3 0.368547 0.0667766 0.312683
Bin: 4 0.192028 0.0918918 0.202385
Bin: 5 0.0368326 0.122242 0.0823411
Bin: 6 -0.129667 0.0901646 -0.111856
max variation from nominal 0.502 declared the "fitting procedure systematic error" = 0.502 - 0.461 = 0.041 ps^-1
*Lars & Gustaaf: bad runs are included, including runs from bad muon r/o period
-yes, it might be best to exclude those runs
- Redone analysis with MU and SMT bad runs removed
N(D0) = 99394.1: +/- 609.786
N(D*) before tag = 25718.4: +/- 167.748
N(D*) after tag = 1222.03: +/- 36.2998
N(D*) after tag & fiducial cuts = 1091.94: +/- 34.2717
New Asymmetry (y) vs. VPDL (x)
Double_t x[7] = {-0.0125,0.0125,0.0375,0.0625,0.0875,0.1125,0.1875};
Double_t ex[7] = {0.0125,0.0125,0.0125,0.0125,0.0125,0.0125,0.0625};
Double_t y[7] = {0.323095,0.431907,0.424368,0.387412,0.197254,0.0626717,-0.120423};
Double_t ey[7] = {0.139223,0.0628955,0.058983,0.0707609,0.0937261,0.11207,0.0839216};
New Minimization output
chi^2 = 2.12/5
dM 0.49614 +/- 0.05060
pur(IST) 0.73610 +/- 0.02117
the change is small.
*Mossadek: p2, par4: is the cut on the angle only done if the distance is more
than 4 sigma? What do you do is the distance is less than 4 sigma?
-make no cut, will illustrate with distributions
did not get to it yet. note that this is the D0 selections which may affect the efficiency but not the
asymmetry. the efficiency value is not used in the analysis.
*Mossadek: how is the number of events estimated in Fig 2?
-number of events in peak is determined by single gaussian so far, plan to
improve as investigate different fitting functions
-more from Andrei (couldn't follow)
-(Vivek): note that fig 2 is a PR plot, and is not really used in the analysis
fit improved, the signal is described now by two gaussians
*Christos: in Fig7 error on number of D* candidates seems small (about sqrt(N)),
what about effect of background subtraction?
-yes, systematics due to background subtraction is being worked on
see above in the fitting procedure part
*Kazu: what are the errors in table 1?
-use normalization of gaussian/bin width, will explain in note
will explain in the next rev of the note
*Christos: is background handled differently for each lifetime bin? If so, may
increase statistical error because lose correlations
-yes, that is correct
-initial state tagging
*Kazu:how are tagging efficiencies optimized? (eff*dilution)
-cuts have not been optimized, this is a common algorithm extensively studied
by others in j/psi K+
-some of the cuts verified by picking a few points in parameter space
*Kazu:is there an asymmetry in positive vs negative?
-will x-check, but think it's 2-3%
- Charge asymmetry (consider tag muon+ and tag muon- separately)
0 & -0.025 & 0 & 14.3697 & 3.99822 & 6.55792 & 2.89528 & 0.373275 & 0.224572
1 & 0 & 0.025 & 90.8665 & 9.94031 & 31.3914 & 5.89613 & 0.486473 & 0.0829607
2 & 0.025 & 0.05 & 93.8083 & 9.94785 & 31.2276 & 5.9155 & 0.500502 & 0.0813557
3 & 0.05 & 0.075 & 69.0866 & 8.49243 & 24.9203 & 5.19967 & 0.46982 & 0.0943577
4 & 0.075 & 0.1 & 31.1442 & 5.87808 & 21.6321 & 4.91788 & 0.180235 & 0.142939
5 & 0.1 & 0.125 & 20.9815 & 4.86694 & 20.2419 & 4.65962 & 0.0179404 & 0.163347
6 & 0.125 & 0.25 & 32.0094 & 5.88752 & 41.0786 & 6.65545 & -0.124086 & 0.120669
: N(D0) = 99394.1: +/- 609.786
: N(D*) before tag = 25718.4: +/- 167.748
: N(D*) after tag = 589.494: +/- 25.1657
: N(D*) after tag & fiducial cuts = 529.339: +/- 23.8126
Double_t x[7] = {-0.0125,0.0125,0.0375,0.0625,0.0875,0.1125,0.1875};
Double_t ex[7] = {0.0125,0.0125,0.0125,0.0125,0.0125,0.0125,0.0625};
Double_t y[7] = {0.373275,0.486473,0.500502,0.46982,0.180235,0.0179404,-0.124086};
Double_t ey[7] = {0.224572,0.0829607,0.0813557,0.0943577,0.142939,0.163347,0.120669};
chi2 = 2.373372/5
dM 0.50176 +/- 0.06425
pur(IST) 0.77505 +/- 0.02920
muon- case
0 & -0.025 & 0 & 21.7907 & 4.9632 & 11.8455 & 3.73724 & 0.29567 & 0.177553
1 & 0 & 0.025 & 70.5269 & 8.72023 & 35.3065 & 6.22004 & 0.332791 & 0.0956974
2 & 0.025 & 0.05 & 89.6055 & 9.74431 & 46.2482 & 7.10697 & 0.319147 & 0.0845406
3 & 0.05 & 0.075 & 58.0137 & 7.91546 & 34.3346 & 6.11287 & 0.25641 & 0.10478
4 & 0.075 & 0.1 & 40.0559 & 6.56125 & 27.9521 & 5.45995 & 0.177976 & 0.123424
5 & 0.1 & 0.125 & 24.5102 & 5.08236 & 20.0732 & 4.67318 & 0.0995209 & 0.154337
6 & 0.125 & 0.25 & 35.5352 & 6.37075 & 44.6143 & 6.8739 & -0.113277 & 0.116678
: N(D0) = 99394.1: +/- 609.786
: N(D*) before tag = 25718.4: +/- 167.748
: N(D*) after tag = 630.494: +/- 25.9885
: N(D*) after tag & fiducial cuts = 559.975: +/- 24.4526
Double_t x[7] = {-0.0125,0.0125,0.0375,0.0625,0.0875,0.1125,0.1875};
Double_t ex[7] = {0.0125,0.0125,0.0125,0.0125,0.0125,0.0125,0.0625};
Double_t y[7] = {0.29567,0.332791,0.319147,0.25641,0.177976,0.0995209,-0.113277};
Double_t ey[7] = {0.177553,0.0956974,0.0845406,0.10478,0.123424,0.154337,0.116678};
chi2 = 0.2585747
dM 0.49550 +/- 0.09009
pur(IST) 0.67953 +/- 0.03058
two cases are (very) consistent, charge asymmetry is 560-529/560+529 = 2.8 +- 3.2%
*Sijbrand: how many muons not from B in sample? Where do they come from? Do they
have specific charge correlation? Since there is overall charge conservation
there is potential for bias.
-is incorporated in dilution for ccbar
-agree that this needs to be investigated more thoroughly
done a study for ccbar contamination in the sample, see
http://d0server1.fnal.gov/users/burdin/www/myweb/bphysics/osc/ccbar.html
bottom line : did not find the ccbar contribution into our muD0 sample and set the upper limit <16%. this limit will be improved in the future.
however for the OS muon tagging there will be additional suppression of the ccbar contribution since we require a muon
on the other side. ccbar coming from the gluon splitting (which could mimic mu D0(D*) signal) it is not expected to have this muon present.
so for the purposes on this analysis we consider this question answered.
fake muons can also contribute to the sample but they are expected to do so equally for the different charge combinations and
their contribution is adequately described by the background under the D* mass peak.
*Lars: last sentence in section 3.1: should this be increase rather than decrease?
-will get rid of word dilution which is used erroneously in b-tagging due to historical reasons
will do in the next rev of the note
*Mossadek: are you double counting? What happens to events with multiple D*?
-don't cut, allow all. 3-4% of the events have multiple D*
-will investigate if there's a bias with these events in right-wrong tags
we don't allow more than one muon in the jet with mu D0 candidate => there are no cases when the same D* is assigned to
different muons. we allow full combinatorics for the D* candidates and this bkg is included in the bkg fit function. there are more than one D* assigned
to one muon in 8.3 % of events.
the number of right/wrong tags for events with more than one D* per mu is 103/70 = 1.47 (to be compared with 967/590 = 1.64 for the full tagged sample).
cross checked if there is a bias redoing analysis with veto of multiple D* per muon candidates.
results:
dM 0.48040 +/- 0.05582
pur(IST) 0.72582 +/- 0.02231
dm is changed by 0.022 ps-1. for now we propose to leave it as is but we'll investigate how to decrease
possible bias here. it may be indeed important. if the second D* comes from combinatoric bkg then it's included
in the bkg fit. however if it comes from physics (ex. bbar from gluon splitting) then we should be careful. both b
and bbar from gluon splitting can end up in the same jet. the second b can fragment into B0 and B0 can oscillate
(or not) and decay to D*. so in principle this term should be estimated and taken into account in the formulas. we have ideas how to estimate
it but for now propose to leave it as is (or assign a systematic to it). the above cross check showed that with the current sample
the possible effect is covered by the assigned errors.
-MC samples:
*Lars: did you verify the turn-on curve due to the d0mess cuts?
-have verified that the cuts are far enough below selection cuts
-vivek: note that the trigger/reconstruction cuts are generally harder
-should have no effect on oscillation vs non-oscillation
*Christos: do we care about the muon spectrum being different in MC than data?
-investigated this as part of the study of the systematic uncertainty on
the K factor, led to the 2% systematic used on the K-factor derived by
using harder muon cuts
-Christos: note that you could do this on an event-by-event basis
-will add plots to show this
-Andrei: but evtgen uses a HQET model for the decays - CDF says the effect
of this is small
think that 2% variation of k-factors is adequate.
-also suffer from the fact that are running pythia with MSEL 5, but only
process with massive b's
-experimental observables:
*Kazu: where does sigma(Ai) come from, dA = sqrt((1-A*A)/N)?
-take asymmetry, and propagate errors from components in the asymmetry
-Christos: error is larger because there's background
-will add formula used to the note
next rev of the note
*Mossadek: why do you have this upper cut on L_xy?
-fiducial cut so don't have to worry about scattering in the beampipe
-will look at events that fail the 18 mm cut
- Removed fiducial cut DL<1.8 cm
1 & 0 & 0.025 & 161.298 & 13.2199 & 66.6762 & 8.56855 & 0.415054 & 0.0630816
2 & 0.025 & 0.05 & 183.417 & 13.925 & 77.3423 & 9.25024 & 0.406793 & 0.0591101
3 & 0.05 & 0.075 & 127.26 & 11.6175 & 59.1881 & 8.01713 & 0.365098 & 0.070785
4 & 0.075 & 0.1 & 71.2529 & 8.81503 & 49.6743 & 7.34371 & 0.178443 & 0.093317
5 & 0.1 & 0.125 & 45.6823 & 7.0466 & 40.2941 & 6.6016 & 0.0626717 & 0.11207
6 & 0.125 & 0.25 & 67.2523 & 8.64769 & 85.6674 & 9.56725 & -0.120423 & 0.0839216
N(D0) = 99394.1: +/- 609.786
: N(D*) before tag = 25718.4: +/- 167.748
: N(D*) after tag = 1222.03: +/- 36.2998
: N(D*) after tag & fiducial cuts = 1091.94: +/- 34.2717
Double_t x[7] = {-0.0125,0.0125,0.0375,0.0625,0.0875,0.1125,0.1875};
Double_t ex[7] = {0.0125,0.0125,0.0125,0.0125,0.0125,0.0125,0.0625};
Double_t y[7] = {0.324492,0.415054,0.406793,0.365098,0.178443,0.0626717,-0.120423};
Double_t ey[7] = {0.139352,0.0630816,0.0591101,0.070785,0.093317,0.11207,0.0839216};
no change
-expected values:
*Kazu: should you add charm to the mix?
-will try to evaluate ccbar contribution, gut feeling is it's small
see above
*Lars: confirm that formula 3 is true (not just should be)
-is being checked now
simultaneous fit of D0 nd D* samples gave consitent results for the purity, see above
-inputs
*Lars: 7.4: should you be worried about the effect of the MC model on
the determination of the reconstruction eff?
-are studying this, is MC efficiency in table 2
varied MC efficiencies, now included into the sys error
*Lars: form of fit function in section 7.5
-small effect because bins are big
-Kazu: what about the relative contribution of each of the 3 gaussians
gaussians are 26 , 56 and 141 um, with relative normalizations 0.423/0.505/0.072
this has been varied in (very) wide range for teh systematic estimate
-results/systematics:
*Kazu: can you use MC to verify the size of the statistical error?
What is the pull? (Show the mixed & unmixed events separately,
show lifetime plot)
-will do
did not have time for this yet
*Makoto: is B lifetime error taken into account
-no, will do
done, gives small systematic error
*Sijbrand: detector alignment: what is the effect of overall scale on decay length?
-rick: investigated by Andre Sopczak, smeared geometry file available
-authors will use that or take effect from an existing lifetime measurement, or
take conservative estimate and include in systematics
done, gives small systematic error
*Sijbrand: is experimental momentum resolution factored into the K-factor?
-taken from MC
-we use spatial resolution in eq 11
-vivek: error from k-factor distribution is much worse than error due
to mom resolution
-Sijbrand: point is momentum resolution is ~gaussian while K factor
distribution is skewed
-authors: will include this in systematic studies
compared k-factor for measured vs. generated paramaters - change is small and is covered by the 2% variation by far.