The purpose of creating a jet filter is to increase the sampled luminosity for the jet simulation sample. For example, the sampled luminosity of the 2006 jet simulation sample in the partonic pT bin 3-4 GeV was 5.3 x 10^-4. Aside from being much smaller than our data luminosity, these events contribute funny shapes to the corrections due to the weighting and poor sampling statistics. The sample doesn't even address the missing 2-3 GeV bin which also contributes to the signal region. To keep the same number of events as, but to get 1 pb^{-1} to disk would require a filter that only accepts 1 in 2000 events.

The demand of a filter is that any event that belongs in the signal sample needs to pass along with any events that would falsely be added to the simulation sample. Originally, I was told that the signal meant jets with pT > 5.0 GeV from the midpoint cone algorithm, so that means any events with a 5.0 GeV jet at the pythia level is signal and any event that reconstructs a jet after reconstruction with pT > 5.0 GeV also needs to be accepted.

I wrote a filter using the StMCFilter framework that implements the midpoint cone jetfinding algorithm with split merge. To have minimal bias, the idea is to accept any event that would be reconstructed after BFC, so the cuts have to be loosened. I ran with the cone radius at 0.7, the seed at 0.5, the association cut at 0, and split merge at 0.5. The jet pT cut was set at 3.8 GeV.

I ran 1000 events in the 2-3 GeV bin and the 3-4 GeV bin. I ran the StJetMaker with the 100% subtraction scheme.

 2-3 bin:

154 events accepted by filter and 22 had jets reconstructed after BFC.

additional 7 events had jets reconstructed above 5 GeV after BFC, but did not pass filter.

These results indicate an acceptance of 1 in 6.5 and a significant bias.

 3-4 bin:

329 events accepted by filter and 52 had jets reconstructed after BFC.

additional 19 events had jets reconstructed above 5 GeV after BFC, but did not pass filter.

These results an acceptance of 1 in 3 and a significant bias.

One way to "solve" the problem is to raise the signal threshold to get out of the 2-3 bin and the 3-4 if possible. Integrating enough statistics is basically a nonstarter if we can't get out of the 3-4 bin. I used the PYTHIA production created to calculate the hadronization and underlying event for the 2005 dijet analysis to estimate the contributions of higher pT partons from the 3-4 partonic pT bin (we didn't run 2-3).

Fig 1 The maximum pT of partons saved in the production (events had to have a 5 GeV jet):

Fig 2 PYTHIA partonic pT for the same:

Basically, an inclusive jet filter won't work. A dijet filter could still work however.

Note: some people have suggested using a BFC filter to filter events that don't pass the trigger. This does not work if any type of detector unfolding is to be used because the trigger dependence of the jet spectrum will be included in the unfolding. This effect is especially dangerous since the trigger effects are heavily sub process dependent and the uncertainty of the subprocess mixtures from PYTHIA can be a significant contributor to uncertainties in corrections.