Figures 1-3 show invariant mass from 15 GeV pT single-particle simulated pi0s plotted vs. the ratio of reconstructed to true opening angle, energy, and zgammagamma respectively.  As per usual, each figure contains four plots: the upper left shows the plot for reconstructed pions with both decay photons in the same tower, the upper right for the decay photons in the same pod but different towers, the lower left for decay photons in different pods but adjacent towers, and the lower right for decay photons in different pods and non-adjacent towers.

Figure 1: Invariant mass vs. reco/true opening angle ratio

Figure 2: Invariant mass vs. reco/true energy ratio

Figure 3: Invariant mass vs. reco/true zgg ratio

These plots strongly suggest that the shoulder phenomenon seen in the two adjacent tower invariant mass distributions (upper right and lower left) is due to overlarge opening angles, as reco/true opening angle ratio is much more strongly correlated with invariant mass than either of the other two ratios.  One possible way to get large opening angles is to have one or more of the decay photons convert before the calorimeter.  Figure 4 shows the distribution of reconstructed invariant masses for cases in which at least one decay photon converts.

Figure 4: Reconstructed invariant mass if at least one decay photon converts before the BEMC

While the shoulder is clearly more prominent here than in Figure 5, below, it's not large enough that conversions could account for the entire shoulder, as a comparison with Figure 5 will show.  However, some gain could probably be made by loosening the current quality cuts on track vetos.  Figure 6 shows the distribution for candidates in which the photon converted at less than 200 cm, i.e. inside the TPC.  While there are overall not that many candidates here, removing them would reduce the shoulder without losing too many signal candidates.

Figure 5: Invariant mass distribution of pi0 candidates from 15 GeV pT single pion simulations

Figure 6: Invariant mass distribution of pi0 candidates where one decay photon converted inside the TPC

Next, a look at some data results, using ~5% of the 2009 pp200 data.  First, Figure 7 shows current results: candidates are obtained in the same way as simulation except that candidates with pT<4 GeV are rejected.

Figure 7: Invariant mass distributions from data

Now, compare the lower right plot in Figure 7 to the lower right plot in Figure 8, in which no lower pT cut is applied.  It appears that essentially all the signal in the lower right plot was at pT of 4 or below.  If we choose pT above that, or make the cut slightly stricter, we can simply discard any candidates where the decay photons are not in adjacent towers.  This would remove essentially all of the combinatorial background, though the downside is that we would also lose the eta peak.

Figure 8: Invariant mass distributions from data without the lower pT cut.