First Attempt

As the filter is designed to loose very few events that would have passed the trigger, it was initially attempted to use the official production as the numberator in computing the trigger efficiencies and use a pure generator level (no reconstruction) production for the denominator.  Howevever, while all options from the official production request page were incorperated, and the Pythia-reported cross section matched to several decimal places, the overall normalization was off by a factor of 15-20.  The ratio of generated to tried was also off by a factor, which happened to be the same order of magnitude.  The trials to generated ratio should be equal to the max value of the cross section divided by the total cross section.  Since the reported cross section was the same, the estimated max value must have been different.  Unfortunatly, I cannot determine this value, and it seems that the filtering somehow effected this valued.  One cannot accurately set the relative normalization if the filter effects the maximum value. 

Thus a different approach is used in the following: use the pure generator level production to estimate the filter efficiency, and use the official (filtered) production to estimate the trigger efficiency among filtered events.  The product of the two is then the full trigger efficiency, but the advantage is that numerator and denomiator come from the same production for each efficiency.

Second Attempt

The approach will be that mentioned in the last paragraph of the above section.  The following tables list the number of runs, total number of tried events, the cross section, and the luminoscity for both the official Pythia production (Table 1) and for the pure generator production (Table 2).  The Pythia-level filter was recoded (and the vertex simulated) so that each event in the pure production could be flagged whether it passed the filter.  The Y-values (expected number of pi0s per unit luminoscity per pi0 piT bin) were computed for both cases, using the formula described in this other blog.  The filter efficiency is then computed as the ratio of the Y_i for events passing the filter over the Y_i for all events (passing filter and not passing the filter).  No cuts are made on the pi0s except the true pi0 eta-value must be within the theory cut of 0.8 to 1.8.

The results for the filter efficiencies are in the following Figure 1.

The filter efficiency is fairly flat, at about 80%, after a pi0 pT value of 6 GeV.  Depending on the final systematics, it may be desirable to increase the statistics of the pure sample to further reduce the uncertainty on the filter efficiency.

To compute the trigger efficiency within the filter, the official reconstructed Pythia production is used.  The Y-values are made seperately for all generated pi0s in the StMcEvent (i.e. all but those removed by the Pythia level filter) and for the same pi0's but with the cut that the event must pass the emulated trigger.  The thresholds are set to be (3, 12, 22) for the HT DSM thresholds and (1, 17, 31) for the TP DSM thresholds, corresponding to a high tower threshold of 2.6 GeV and a trigger patch threshold of 3.8 GeV.  See Pibero's and Jamie's documentation.  These correspond to the thresholds on May 12th, or the early part of the longitudinal running.  Trigger efficiencies for the other thresholds will also be computed.  ***Correction: the thresholds are actually those of the latter part of the longitudinal running, relevant starting from run 7133052.***

The results for the trigger efficiency with in the filter is given in the following figure.

Note: the uncertainty is currently lower on the data from the official production.  The product of the two efficiencies gives the total trigger efficiency, which is given in the following figure:

Conclusions

The trigger efficiency can now be reliable computed.  We need to additionally compute the trigger efficiencies for the different trigger thresholds, so that the appropriate efficiency is used for each data run.  Depending on the final uncertainties, we may need to increase the number of Monte Carlo statistics for the non-reconstructed case to decrease the uncertainty on the overall trigger efficiencies.