From: John Sullivan Subject: Re: comments on MVD multiplicity at L1 To: wohn@iastate.edu (Fred K Wohn) Date: Fri, 04 Dec 1998 17:13:13 MST Cc: hubert@lanl.gov (Hubert van Hecke), jsimon@lanl.gov (Jehanne Simon-Gillo), schlei@lanl.gov (Bernd Schlei), mjbennett@lanl.gov (Mike Bennett), young@orph01.phy.ornl.gov (Glenn Young), petridis@iastate.edu (Athan Petridis), lajoie@iastate.edu (John Lajoie), lwood@iastate.edu (Lynn Wood), jhill@iastate.edu (John Hill), mrosati@iastate.edu (Marzia Rosati), lebedev@iastate.edu (Sasha Lebedev), wohn@iastate.edu (Fred Wohn), Soren-Sorensen@utk.edu (= Soren Sorensen) In-Reply-To: <199812032059.OAA19835@pv7453.vincent.iastate.edu>; from "Fred K Wohn" at Dec 03, 98 2:59 pm X-Mailer: Elm [revision: 212.2] Hi, This is a response to Fred's note 1 or 2 days ago. I have not done any new calculations, but thought I would at least answer Fred's direct question. > > Since I have plots from Athan on his work, one of the first things > that struck me was the similarity between his plots of Nstrip vs z > and John's. However Athan did one thing that I think (and I'm asking > John with this e-mail) is different from what John plotted. The data > points that Athan showed were mean values (averages over the events > and the 6 L1 strip sums with the same value of z), but his error bars > were not the uncertainty in the mean but the rms deviations. We wanted > the rms since it gives the event-to-event fluctuations. If all is > gaussian, the the error in the mean is the rms/N^(1/2) - this roughly > agrees to my eyeball since John's N = 150, so the rms should be about > 12 times larger than the error bars in John's plots. With this > difference in mind John's and Athan's plots look quite similar. Fred is correct about the definition of my error bars. However, I like the way they are defined. A sufficiently ambitious person could have figured out the defintion of the error bars out from my web page because the root commands to make the plot are on the page. However, root seems to be closely related to Welsh as far as I'm concerned, so you have to be very ambitious. My error bars are rms/N^(1/2), which means they are the estimated uncertainty on the mean. I say estimated uncertainty because it is relevant in this case -- the number of data points is too small to determine the rms in some cases. I think this is the correct approach for estimating this correction, because we can not do much better than make a correction so that the mean value is correct. When attempting to correct the mean value vs. the vertex position, the uncertainty on the mean seems like the correct error bar to use in the fit. I had not thought about the multiplicity fluctuations much. I still believe that it would be unfortuate to design away the ability to make a vertex correction of the MVD's signal to level-1. I think this because I can imagine a dN/deta distribution (perhaps unrealistic) for which this effect would be much larger. In a year or so we will no longer need to speculate. The good news seems to be that there is no discrepency in the calculations themselves (Athan and myself). Therefore, I think you guys can safely argue about the consequences of the calculations at BNL next week. I will not be there. My opinion is summarized in the paragraph above and I am content to let Mike Bennett represent me in this matter. My web page on this subject is at: http://p25ext.lanl.gov/people/sullivan/notes/phenix/trigger/trig_vs_zvert.html Regards, John