Hi, in the absence of Piotr I took over (temporary) the MIP calibration with L3 tracks. (eLog entry #406) Below, I'll comment on Piotr's results and present next steps in a logical (for me) order. (Do not skip #6 !) 1) The position of the MIP peak looks very reasonable for most of spectra . It would be good to have a factor two more events, but this 6M sample cost Piotr ~3 weeks of baby siting and 1 week of my time. We can't afford more at the moment. 2) The automatized MIP finder made by Piotr detected correctly all dead towers , seen previously with the slope method ! I updated the web page : http://www.star.bnl.gov/eemc/ --> calibration-->* history of all dead towers (first entry on the page) 3) In the case the Landau fit failed the scaled down average over the same ADC range was used instead. The formula (possibly a crude one) is: xscal = 1.03 * xmean - 7.98; Piotr, could you post for the record the exact track selection criteria you used, when you have time. 4) A possible improvements of the fitting procedure (if we do another iteration): tower 6TD11 : there is too many ADC >60 algo: if integral for ADC[0-60] <50% total then change integration interval tower 7TC10 : huge discrepancy between Landau peak and mean. Algo choose the (higher) mean, but it may be a low gain tube. algo : verify Landau rejection criteria tower 8TD09 : this is a very noisy tube (see history). Landau failed but not due to the statistical fluctuation, but different shape. algo: fit the pedestal shape as an alternative We may not need any improvements for this year data (perhaps just run the same algo once more over data from last few days to track any time dependence). However, me may think of it as a good tool to detect problems with 720*3-4 towers next year. We will try to strip the L3+EEMC data from .daq files in a real time, so it all have chance to work with a latency of 1 day. One more general observation: it looks like the leftovers from the pedestal shows up in the first few bins, e.g. :5TA09, 5TE09, 6TA06, 6TA09,... (note, only ADC values 3 sigma above the pedestal are plotted). One could think of a more advanced algo: - use all ADC range above pedestal (or ped -2 sigma) - record in parallel pedestal plots for towers matched to tracks rotated by 90, 180, 270 deg. - subtract pedestal peak using ADC<5 for normalization. 5) Despite all remarks above, I decided to use the MIP position from Piotr analysis as is and combine it with the slope analysis for eta bins [1-4]. The goal was to store in DB the relative gains for 240 towers and apply (change) the overall scaling factor in flight. 5.a) eta bins 5-12 Rather arbitrary the tower 05TB09 has been chosen as the reference one. It has a nice Landau shape and 'healthy' neighbors. In general, the MIP position depends on: MIP=G*d/cos(theta) where G=gain, d=thickness of scint along Z-axis, theta=acos(pz/p). The relative gains g_i=G_i/G_0 are given by the relation: g_i=MIP_i*tanh(eta_i)/MIP_0/tanh(eta_0) where index '0' is tower 05TB09 For completeness, the electromagnetic energy is calculated with the following formula: Energy/GeV = (rawAdc-ped_i)/( g_i * SF ) SF is in ch/GeV, but in principle it should be divided by tanh(etaBin=9)=tanh(1.3)=0.86 if we want to refer to the nominal thickness of EEMC (along Z-axis). I'm not sure we should put this overall factor in to DB, so I left it out for the moment. 5.b) eta bins 1-4 the gains from slopes are used, rescaled to have the gain=1 for 05TB09 . g_i=slope_0(HV-source)/slope_i(HV_H) 5.c) the following towers are taken out in the off-line analysis: Based on L3-MIP analysis: 5TC12, 5TD04, 5TD10, 6TA12, 6TB04, 6TC09, 7TB09, 7TD08, 8TD04, 8TD08, 8TE12 no slope 8TE04 The new gains are attached. (format : name, gain, error_gain, dum) They are loaded to DB with the time stamp April 30 0:20:00 (beginning of the Fill 3587 ) 6) While combining gains from slopes with L3-MIP I found the cause of the 'factor 2' gain problem. In short, I used in the past formula: g_i=slope_0(HV_source)/slope_i(HV_H)/cosh(eta_i) which lowers gains by a factor 1.5-3.4 in the eta-dependent way. It happened in the piece of my code producing final tables, loaded later to DB. This bug did not affected calculations of new HV for subsequent iterations. More details and new scan of the scale factor is documented at: http://www.star.bnl.gov/protected/spin/eemc/SFbugFound/ Now, the best SF is 60 ch/GeV. Also pi0 invariant mass peak is sharper. 7) next step is to look for pi0/eta peak as function of towers With regards Jan