Using the similar filtering procedure developed by Matthew Walker (see link), I examined the performance of this filter for proton-proton collisions at a center of mass energy of 500 GeV.  There are two filters implemented in this setup: a dijet filter executed at the particle level (used to create the fzd files) and a trigger filter invoked before the computationally expensive track reconstruction (makes the MuDst and geant files). Every filter performed raises concerns of biasing the data (removing events that would have reconstructed dijets) and thus possibly altering physics. However, a sucessful filter, one that doesn't give a large bias, reduces the computation time and provides a sample with more events of the desired signal. 

Therefore, I adminstered a test of the filter setup to determine the characteristics needed to make a dijet filtered simulation request. It is important to note that in this test no events were removed. Instead, I record only the result if the filter were applied. For example, I print out whether or not an event at the pythia level contained the signal.  

At the Pythia level a dijet filter is performed and it requires signal events  to have a dijet with the following characteristics:

both jet |eta| < 1.3

delta phi > 2

max pt > 10.0

min pt > 7.0

dijet M > 10.0

Once the dijet pythia filter is completed, the TriggerFilter is then applied, which accepts events that would have fired the JP1, AJP or BHT3 triggers.

The following table displays the results of this test of the filter performance. I threw 50K events in each partonic pT bin.  The first column indicates the actual events thrown and for some reason events were not reconstructed; I did not try to recover them. In column 3 and 4 we repectively see the number events that pass the dijet pythia filter and both the pythia and trigger filter. Column 5 informs us about the number of events that failed the pythia filer but had a dijet signal after the BFC was performed. However, these events did not pass the Trigger filter. Column 6 is the true indicator if there is a bias for each pT bin. It tells us the number of events that failed the pythia filter but had dijets after the BFC and passed the Trigger filter.  Column 7 tells the number of events that pass the trigger filter and column 8 shows the number of events need to obatin 1pb^-1 for that pT bin.

Now one can use this table to determine the biases of the filter for each pT bin.  This is shown in Table 2.

In order for a filter to be considered unbiased the values in column 2 of Table 2 should be << 10^-3. One can then see that the largest bias occurs at the partonic pT bin of 25-35GeV. Therefore, using the tables above as an indicator, we propose to use apply the full filter on pT bin of 3-11GeV, the Trigger Filter of pT bin 11-25GeV, and no filter on 25GeV and above.

From the analysis meeting in March of 2011, I was informed that JP1 and JP2 trigger thresholds are higher than those at 2009 200 GeV. Therefore, the asymetric pT cuts (see above) should be increased as well. I have changed this values to be

max pt > 13.0 GeV

min pt > 10.0 GeV

Using this information I began calculating the time and disk space needed for the simulation request. I threw 10K events with the filters implemented as indicated above and the results can be seen on the follwing table: