The calibration of the tower gains in the BEMC is done in two stages: first the location of the MIP peak is identified in the ADC distribution for each individual tower, to obtain a relative calibration between towers, and then electrons are used to find the E/p peak for groups of towers, to obtain an absolute calibration scale. 

I follow the procedure described in the 2006 report (drupal.star.bnl.gov/STAR/system/files/2006-Calibration-Report_3.pdf). 

Analyzing MuDSTs

We select events from the Run 11 500 GeV pp run with the following trigger IDs:
minimum bias....
VPDMB: 320000, 320001, 320011, 320021, 330021
ZDCMB: 320100, 320101, 320102, 320102, 320111, 330100, 330102, 330111
BBCMB: 320103, 320113, 320123, 330123
TOF0*VPDMB: 320300, 320302, 330322, 320312, 320322
TOF1: 320301, 320311, 330311
high tower....
BHT0*VPDMB: 320500, 320504, 320514, 320524, 330524
BHT1: 320501, 330501
BHT2: 320503, 330503
BHT3*L2BW: 320801, 330801

For each primary track we write out the track information from the TPC, the EMC information for the 3x3 tower cluster around the track, the TOF information, and the trigger information. 

MIP calibration

We select events with |vz| < 30 cm and exclude any towers that have multiple tracks associated with them.  We select tracks with p > 1 GeV/c, and require that they enter and exit the same tower.  We require that the towers surrounding the central tower do not contain a large energy deposition.  We require that ADC-ped > 1.5*pedRMS.  After these track quality cuts we fill histograms with the ADC-ped values for each tower.  These histograms are shown in mip.pdf (attached). 

We fit each histogram with a gaussian on a pedestal.  If the fit values fail basic quality cuts (such as if the mean is < 5), then the tower is assigned a bad status (!=1).  These fits are marked as red in mip.pdf.  The bad towers are: 
34 106 139 160 208 267 286 287 410 426 533 541 561 615 649 657 693 813 821 822
823 824 829 830 831 841 842 843 844 846 850 851 852 857 875 939 953 993 1026
1048 1080 1100 1180 1197 1198 1199 1200 1217 1220 1221 1222 1223 1224 1237
1238 1240 1241 1242 1243 1244 1257 1258 1259 1348 1353 1388 1407 1409 1433
1434 1448 1567 1574 1668 1753 1761 1765 1856 1877 2032 2077 2092 2093 2107
2168 2214 2299 2305 2424 2589 2632 2961 2972 3017 3070 3493 3495 3508 3588
3604 3611 3678 3679 3737 3900 4018 4019 4331 4432 4459 4646 4677 4684 4768
Additionally, I checked mip.pdf and picked out other bad towers by eye.  They are: 
240 266 431 483 484 504 539 616 633 637 638 653 671 673 674 757 758 760 790
832 837 873 1171 1183 1187 1207 1218 1219 1304 1312 1397 1405 1427 1612 1654
1762 1766 1773 1976 2073 2097 2160 2415 2590 2811 2834 2969 3071 3494 3668
3690 3718 3738 4057 4059 4223 4500
Note that most of the towers with bad MIP peaks were marked as bad (cold/hot/stuck bit) towers when the status/pedestal tables were originally computed. 

For each tower we record the mean and sigma of the gaussian fit, and the status. 

E/p calibration

To isolate electron tracks we apply the following cuts: 
-- |vz| < 60
-- track must be matched to a reconstructed vertex
-- nhits >= 10
-- ADC-ped > 2.5*pedRMS
-- track must enter and exit the same tower
-- 1.5 < p < 6 GeV/c
-- track does not point towards a tower which fired the HT trigger
-- dR < 0.025 (distance from the center of the tower)
-- dE/dx > 3.4e-6
-- the maximum Et in the 3x3 cluster of towers must be in the central tower
-- there are no other tracks pointing to the central tower

We can compare all towers at a given pseudorapidity (eta) by calculating the calibration constant C = 0.264*(1+0.056*eta*eta)/(sin(theta)*MIPpeak), where MIPpeak is the mean of the gaussian fit to the MIP histograms above.  For each track we then calculate E/p by evaluating C*(ADC-ped)/p.  We fill E/p histograms for 40 rings in eta (-1 < eta < 1).  We fit these histograms with a gaussian on a first-order polynomial to determine the location (mean) of the E/p peak.  The deviation of the mean from 1 gives an additional overall adjustment to the gain to obtain the absolute energy scale. 

In electrons.pdf I show the E/p histograms and fits for each of the 40 eta rings (pages 1-20).  I then repeat the procedure for each of 30 crates (pages 21-35).  On page 36 I show the location of the E/p means for each eta ring.  (In the top plot the error bars are given by the gaussian sigma, and in the bottom plot the error bars are defined by the error on the fit parameter).  It is clear that for |eta| < 0.85 the means of the E/p peaks are approximately 1 and do not vary as a function of pseudorapidity.  For |eta| > 0.85 the correction is large. 

Using the TOF

It was suggested that I use the TOF information to improve our electron identification.  I impose the additional requirement that tracks have valid TOF information, and that |(DeltaInvBeta/InvBeta)| < 0.04.  This removes some hadron contamination, particularly at low momentum.  The E/p histograms with the additional TOF requirements are shown as purple points/lines in electrons.pdf.  Clearly the TOF requirement removes some background, particularly at low E/p (below 0.5).  On page  36 of electrons.pdf the location of the E/p peaks are compared with and without the TOF requirements (purple and black points, respectively).  I attribute the deviation between the purple and black points at large eta to the TOF acceptance (there is no TOF coverage at all in the first and last eta rings).  Otherwise the deviations are very small, and could be used as a systematic uncertainty on the energy scale. 

Some QA checks

1) When I split the histograms by charge, I see a difference between the location of the E/p peaks for electrons and positrons (as was seen in previous years).  This difference remains when I add the additional TOF requirements.  
2) I see a dependence of the calibration on momentum that I will explore further. 
3) I do not see a large crate-dependence in the calibration.