The above left plot shows the two-gamma invariant mass distribution for L2 gamma events for one particular Pt bin (6.75 - 8.25 GeV.)  The x axis is invariant mass in units of GeV/c² (sorry for the poor labeling.)  The black curve is data taken with the L2 gamma trigger.  The red curve is the combinatoric background calculated from mixed events using BEMC points that have been rotated so the jet axes from the different events are aligned.  A detailed description of the procedure is given below.  This gives the shape of the red curve, which has been normalized to the region between 0.8 and 1.2 GeV/c².  While not perfect, this shape recreates the background much more faithfully than mixing random hits from random events.  As of right now, the red curve includes all Pt values above 5.2 GeV/c.  I plan on recreating this shape for all Pt bins to see if the fits are more faithful to the data.  The plot on the right side is just a magnification of the left plot to better show the fit and the differences between model and data.

---- procedure ----

First, I assigned, to each BEMC point (hit) a 'jet eta' and 'jet phi,' which are the coordinates of axis of a jet in the event from which the hit was taken.  For events with more than one jet, I used the first jet in as it appears in Adam's combined spintrees (i.e. calling event->GetJet(0).) 

Next, I split the events into 12 z-vertex bins, each 10 cm wide, from -60 cm. to 60 cm.  Only events from the same z-vertex bins are mixed together.  Since real combined hits would originate from a single vertex, we want to ensure that our mixed hits come from (at least) nearby z vertices.

Next, I pick out a single event, and 'mix' (see below) each hit in that event with each photon in the previous 24 events originating in the same vertex bin.
Mixing consists of taking the hit from the previous event (mix hit) and rotating it by an angle phi which is the difference between 'jet phi's' from each of the events.  Then the mix hit's eta is shifted by the difference in 'jet eta's' from the two events.  If a hit   Then I take the mix hit, and together with the hit from the current event (current hit) calculate all of the typical pion quantities (mass, Zgg, Eta, Phi, etc.) Currently I get the angle of separation by pointing both the mix hit and the current hit back to the primary vertex of the event from which the current hit is taken.

The philosophy is that if both hits were coincident with their respective jet axie, after the rotation and shifting the hits would be perfectly aligned.  In the real data, combinatoric background comes from 'fake' pion signatures predominantly from two photons from the same jet.  The mixing procedure allows the proper mixing by simulating hits coming from one jet.

After the mixing is done for each hit in each event, I end up with a sample of pion candidates that should mimic the pion candidates I get from falsely mixed photons in the data.  I run these candidates through my regular pion finding cuts (Zgg < .8; Good SMD strips in each plane; Charge track veto; etc.)  Any candidate that passes my regular cuts is saved.  The masses are potted out to give me the shape in the above plots.

----------------------

This background modeling is not finished yet, but these preliminary findings are encouraging.