Hi all,
as discussed at the last vub-recoil meeting, I have performed a few checks that
can help un in choosing a model for our pdf.
The first check is related to the systematic variation in the event yield when
varying the fit parameters within their errors. As implementing this in VVF is
quite lengthy and bug-prone, I did the check using our standalone macros. I let
floating the parameters that I usually keep fixed, to get an idea of their
possible variation, and then performed mES fits fixing the parameters to their
values+error and values-error. I did that for leptonic cuts and all analysis cuts,
on run4 and run12 generic MC, using both models (argus+CrystalBall,
argus+Frankenstein).
All the variations I see are below the % level, with just one exception: the
variation on the parameter alpha (of the Frankenstein model) on run12, allcuts,
induces a 1% variation in the signal yield.
The CrystalBall behaves in much the same way, variations below the percent level,
just less parameters!
The second check is about the effect of generated versus reconstructed number of
events/efficiencies for different truthmatching flavours. I produced a table using
Kerstin's algorithms #3 and #6 (described in
http://www.slac.stanford.edu/~petrella/truthm/R18/truthmatching.html along with #4
and several others). Here is the result (for some reason, the run5 job crashed for
#3):
Algorithm #3:
Nu true Nu
R12 753 +- 29 799 +- 32 +- 61
R3 191 +- 15 216 +- 17 +- 35
R4 681 +- 28 685 +- 27 +- 65
R5
Nsl true Nsl
R12 130860 131672 +- 386
R3 38263 38489 +- 210
R4 134021 134584 +- 390
R5
Eps_mx*Eps_u true Eps_mx*Eps_u
R12 0.435172 0.435955
R3 0.433593 0.4312
R4 0.411164 0.410868
R5
Algorithm #6:
Nu true Nu
R12 791 +- 30 888 +- 34 +- 68
R3 203 +- 15 217 +- 16 +- 36
R4 707 +- 28 688 +- 28 +- 67
R5 664 +- 28 704 +- 30 +- 66
Nsl true Nsl
R12 149973 151190 +- 415
R3 43942 44207 +- 225
R4 154949 155944 +- 421
R5 155067 154668 +- 419
Eps_mx*Eps_u true Eps_mx*Eps_u
R12 0.421148 0.421738
R3 0.413058 0.4118
R4 0.380224 0.379475
R5 0.350562 0.351446
I would conclude by saying that these results support our idea of moving to the
argus+Frankenstein model.
All the best,
Roberto
|