Hoi, today we got the following. Keep reading, especially to the last paragraph. We should discuss this next week. For various reasons, it looks difficult before Thursday. Thursday 8:30 am PDT? Cheers, --U. > From: Ed Thorndike <[log in to unmask]> > To: [log in to unmask], [log in to unmask] > Subject: Concerns with Vub PRL > Date: Fri, 19 Sep 2003 18:02:54 -0400 (EDT) > > Hi Vera, Hi Urs, > As Dave Cassel conveyed to Vera by phone in late July, I was in shock that > you continued to use HQET OPE values of /\bar, lambda_1 in the shape function. > I apologize for being so slow in getting my concerns to you. My lateness in > writing in no way is due to a lack of concern. Let me try to be very explicit > about my concerns. > > You do explicitly say what you do, which is good. However, you bury it in a > Reference, Ref. 18, which is not so good. In Ref. 18 you say: "We assume > that the shape function parameters in Ref. [9] can be related to the HQET > parameters /\bar and lambda_1." We can all agree that this is true, but the > essential point is that nobody knows exactly what this relationship is! > Hence, the statement is misleading to the casual reader who is not aware that > the relationship between the two sets of parameters is not known. Also, > you do not include an estimate of error for the assumption that the > two sets of parameters are IDENTICAL, rather than related somehow in > a way that is not understood. I understand that not all theorists > feel as strongly as Neubert that your assumption that they are > identical is unreasonable, but don't they all feel that SOME error > must be assigned to this assumption? > > In our phone meeting of 3 July, the question was raised as to whether the > range of exponential shape function parameters (/\bar, lambda_1) gave > <E_gamma> values and errors consistent with CLEO's published values. In Email > on July 4 I gave you the numbers showing that it did. A correlary of this is > that if one calculates <E_gamma> using the values of (/\bar, lambda_1) that YOU > have been using, one will get an <E_gamma> value, with error, not in good > agreement with CLEO's measurement and error. We have carried out this > exercise, and find: > > Measured 2.346 +/- 0.032 GeV > calculated, using correct /\bar, lambda_1 2.353 +/- 0.031 GeV > calculated, using your /\bar, lambda_1 2.377 +/- 0.043 GeV > > Note the differnce in the central value, 31 MeV. Your values of /\bar and > lambda_1 come from our <E_gamma>, with its error, 2.346 +/- 0.032. Your > procedure should, MUST get that back, to an accuracy small compared to the > statistical error on the measurement (32 MeV), if using the HQET /\bar and > lambda_1 in the shape function is reasonable. You don't, and the only possible > interpretation is that HQET /\bar, lambda_1 don't mean the same thing as > Light Cone /\bar, lambda_1. This conclusion is TOTALLY INDEPENDENT of what any > theorist tells you, or us. It's an empirical fact, the output must match the > input, if what you're doing is right, and it doesn't. > (As an aside, our /\bar, lambda_1 Shape Function are obtained from a fit to the > spectrum down to 1.5 GeV, and so what we obtain (line 2) needn't perfectly match > line 1, as line 3 should. Nonetheless, it differs by only 7 Mev, and matches > the error perfectly. The latter is probably fortuitous.) > > If you choose not to use CLEO's determination of the shape function, but > instead use CLEO's determination of HQET parameters /\bar and lambda_1, as you > have done, then shouldn't you include SOME error for this approximation? I > think the answer is a clear "Yes". How big? Hard to say. Would it matter? > YES! This is your dominant error. You're clearly better off having it > determined by an objective procedure than by a guess. And you're clearly > better off having it determined by a guess than by setting it equal to zero! > > I recall, and Dave Cassel recalls, that you mentioned that the range of > uncertainty in /\bar, lambda_1 that we obtain from our b -> s gamma measurement > gives a varaiation in M_X distribution that you find in poor agreement with your > measured M_X distribution. If this is true, if you indeed can reduce the errors > on the shape function parameters, that is great! It certainly deserves mention > in your paper, since we're talking about the dominant error. > > > On a somewhat separate matter, I disagree with the first sentence in your > conclusions paragraph "This result is consistent with previous measurements [4], > but has a smaller systematic error, ..." Ref. [4] is to the four LEP > experiments, and to CLEO's endpoint result. Far be it for me to defend the LEP > experiments. However, let's compare your result with CLEO's endpoint result. > For statistical plus experimental systematic errors, you quote +/-0.28, > +/-0.27 => +/-0.39, 8.4%. We quote +/-0.34, 8.3%. Pretty comparable. Your > fourth error, +/-0.26, corresponds to our third, for which we quote +/-0.16, but > would quote +/-0.23 for a treatment consistent with yours. Those errors, going > from branching fraction to |Vub|, are 100% correlated. Finally there is your > third error, for which you quote +/-0.40, but I say you should quote +/-0.60, > 13.0%. ("I say you should" means that's what you would get if you used the > light cone /\bar, lambda_1 from our b -> s gamma determination.) > That is to be compared with our second (+/-0.44) and fourth (+/-0.24) > error, => +/-0.50, 12.3%. These errors, yours and ours, in addition to being > comparable (13% vs. 12.3%), are HIGHLY CORRELATED. So, collecting errors, > 8.4% vs. 8.3% (uncorrelated), 5.6% (completely correlated), 13.0% vs. 12.3% > (highly correlated). The correlated errors dominate. > Conclusions from this part. > 1. Your statement of smaller systematic errors is incorrect,if you > include theory errors in the systematic errors. > 2. Putting all the errors together, your analysis and our analysis have > very nearly equal errors, if theory errors are treated in a > consistent way. > 3. There is very substantial correlation between your errors and ours. > I think Point (3.) is worth your mentioning. > > > > If you are not able to show that the issues that I raise here are incorrect, > I very much hope that you will revise your paper before it appears in > PRL. If it is too late for that, then I think that you should submit an > Erratum. Failing that, we would feel forced to submit a Comment, calling > attention to the unresolved scientific contraversy concerning the key > theoretical assumption in your paper. Clearly this would be good for > none of us. > > Regards, > Ed