Thanks Daniele for the clarification. I never thought that we could use the data directly for the subtraction, since the total number of missed K0 is always larger than the detected Ks. The idea was to use the MC to do the subtraction, but to do a careful comparison between data and MC to ascertain that the MC describes the data correctly. For this and other checks we should have four subsamples: 1) Mx < Mxcut and b--> u enhanced signal Mu 2) Mx < Mxcut and b--> c enhanced 3) Mx > Mxcut and b--> u enhanced 4) Mx > Mxcut and b--> c enhanced We need to compare the data to the B --> c MC samples, for instance the observed kaon spectra. If they agree, then we can hope to use these sample to see how the inefficiency in the kaon detection impacts the Mx distribution overall, and specifically at low Mx. The more of the b--> c background we can understand, both in shape and normalization, the more we can trust this subtraction, and thereby reduce the systematic uncertainty in the subtraction. For instance, does this subtraction work for the b--> c enhanced sample, i.e. zero events when we subtract the MC from the data? I would also like to see the lepton spectra and charge multiplicity for the four subsamples, and see what we get after we perform the subtraction from the observed momentum spectrum, based on the same bg normalization as for Mx. We know that unlike the hadron mass spectrum, the lepton spectrum should be smooth! There is quite a bit more to learn, but we now have the data and MC to check all this, I need to study your presentation from today a little more, before I can comment, Ciao Vera -----Original Message----- From: Del Re, Daniele Sent: Monday, March 25, 2002 11:25 AM To: vub-recoil Subject: can we use Ks to model Kl background distribution? Hi all, finally I have some doubts about the use of Vub depleted sample with reconstructed Ks to get the right Kl background subtraction. In our data sample we expect to have ~70 signal events for Mx<1.6 Gev. Since the ratio S/B is 1/2 at the moment, we will have 140 background events. From Guglielmo's studies we know that the 50% of background events are due to Kl => we will have ~70 background event with missed Kl's. Since for physical reasons the number of Ks's has to be ~the same and since the efficiency on Ks reconstruction (BR(Ks->pi+pi-)*eff(pi+pi-) is aroud 30% we will end up with ~22 Ks events from the depleted sample. This implies that the relative statistical error on this component will be ~20% => the background subtraction related to Kl will have an error of ~14 events and this is already similar to the statistical error. The conclusion is that we can't use Ks. Then my proposal is the following: - use the depleted sample for all the remaining background component. We have a lot of events with K+ from the depleted sample and the comparison looks fine. - use MC for Kl with the following correction: * resolution correction looking at the difference in the depleted sample for a subsample with similar multiplicities * use control sample to evaluate the difference in the calorimeter deposits for Kl. This is actually the only check we have to do and we don't need too much statistics. Suppose for instance to find a difference of 10-20 MeV in the calorimeter deposit. This should be a small systematic and we could use the MC shape. phi gamma control sample could be very nice for us. We should discuss about this further. Daniele