Hoi, the Fermi motion is parametrized with a function of the form f(x) = N * (1-x)**a * e**((1+a)*x) where x = kPlus/lambdaBar and lambdaBar = mB - mb. kPlus is in the 'events' tree. ~ursl/macros/FsxUtil/anaVub.cc::sfDependency() contains my implementation (just in case ...). The default parameters in the generation of the nres signal MC were BMASS = 5.2792; BQMASS = 4.800; A0 = 1.29; In the following eps files, the first page contains size plots: Mxfit Mxgen efficiency for Mxfit cut efficiency for Mxgen cut f(kPlus) weights The second page contains the relative error on the efficiency (NOT Vub) for Mxfit (left) and Mxgen (right), fitted with a polynomial of fourth order (or degree, I always forget). mb = 4.800 +/- 90 MeV --------------------- http://www.slac.stanford.edu/~ursl/talks/062002/sfDependency-0.09-0.00.eps http://www.slac.stanford.edu/~ursl/talks/062002/sfError-0.09-0.00.eps mb = 4.800 +/- 150 MeV ---------------------- http://www.slac.stanford.edu/~ursl/talks/062002/sfDependency-0.015-0.00.eps http://www.slac.stanford.edu/~ursl/talks/062002/sfError-0.15-0.00.eps a = 1.29 +/- 0.5 ---------------- http://www.slac.stanford.edu/~ursl/talks/062002/sfDependency-0.00-0.50.eps http://www.slac.stanford.edu/~ursl/talks/062002/sfError-0.00-0.50.eps a = 1.29 +/- 1.0 ---------------- http://www.slac.stanford.edu/~ursl/talks/062002/sfDependency-0.00-1.00.eps http://www.slac.stanford.edu/~ursl/talks/062002/sfError-0.00-1.00.eps Summary: -------- o Sven's error is reproduced with the Mxgen cuts when varying mb by 150 MeV. (Remembering a factor of two.) o Mxfit seems much less affected than Mxgen. o 150 MeV as uncertainty on mb might be (or not) a sensible number. Maybe 90 MeV is as good? Maybe we trust Uraltsev? o Compared to the uncertainty in mb, the error from a is small. Again, cutting on Mxfit seems to reduce the dependence on this parameter. The interesting question to be settled: What uncertainties do we take? Cheers, --U.