In my previous post I mentioned that I was getting very low efficiency in identifying 2-photon clusters with the new asymmetric shower fit, even in comparsion with the old symmetric

shower fit. And there are at least 2 places where the separation algorithm could be wrong ( or need to be improved ) -- cuts in the cluster moment and/or chi2 calculations and related

operations (check for bogus photons,etc.)

Since I dont have direct information that tells me whether a specific cluster contains 1 or 2 photons in simulation, it seems necessary to have a large MC sample with pions from a wide

energy range in order to assess the effectiveness of the moment-based cuts. And I dont have that big MC sample yet. So I turned to look at data first and compared different cuts.

                          figure 1. a) cluster moment cuts used in current data, small cells                   figure 1. b). cluster moment cuts used in Run6 FPD analysis, on top of FMS data, small cells

                       figure 2. a). same as above, but from large cells                                                                              figure 2. b). same as above, but from large cells

Basically I looked at two kinds of cuts, 1). linear cuts on cluster energy vs product of energy and sigmaMax, which is what we are currently using on FMS data. 2). non-linear cuts on

sigmaMax vs cluster energy, which is used by Len on FPD data. At first glance, it seems both of these 2 cuts are doing a pretty good job in separating the 2 types of clusters within

reasonable energy range, although I dont have a quantitive comparison of how many clusters of a speciific type they would produce yet. And the general conclusion is that the

moment-base analysis will discriminate 1- vs. 2-photon clusters up to ~70GeV in the small cells and ~45GeV in the large cells. Further separation power must come from the goodness

of shower fit in the subsequent stage.

Then I checked my calculation of the chi2 for asymmetric shower fit, and related operations in terns of identifiying bogus photons ( photons centered on low energy towers, abnormally

high fiited energy on the central tower than the actual tower energy, etc.) I found that the chi2 calculation has no problem but there was one more step I need to modify in the bogus

photon code. And that was the reason that I got abnormally large # of 1-photon clusters compared to the symmetric shower fit. The following plot shows chi2/ndf of 2-photon fit for

catag=0 (ambiguous) clusters in small cells (40GeV pion simulation).

                                                            figure 2. a)                                                                                                                             figure 2.b)      

Here I have labeled 3 scenarios regarding how the ambiguous cluster was processed based on the chi2/ndf of 1-photon and 2-photon fit.

1) The bin at  -2 tags the clusters that have been categorized as 1-photon cluster because 1-photon shower fit is already "good enough", where good enough means "chi2/ndf of

1-photon fit is less than 5.

2) The bin at -3 represents the clusters that will produce a "bogus photon" when applying 2-photon fit. Currently the routine that checks for bogus photon consists of the following

     steps:

     a).If the fitted position of the photon is located at a tower which does not belong to the cluster, the photon is a bogus photon.

     b).If the energy of the central tower is too low compared to the fitted photon energy ( < 0.25*photonE ), the photon is a bogus photon.

     c).If the energy of the central tower is too high compared to the fitted tower energy ( > 1.5* fitted towerE), the photon is a bogus photon.

3) The rest of the bin represents the clusters that have made their way to the last step where the chi2/ndf of 2-photon fit will be compared to that of 1-photon fit, to determine the number

of photons in the cluster.

The next step is to generate a large pion sample with a wide energy range, either by Pythia or just toy single particle generators. And compare moment-based cuts in simuation and

data. And compare chi2/ndf of 1- vs. 2-photon fit for ambiguous clusters in simulation and data, to see how far we can go in identifying high energy pions.

Trigger Simulator

I have followed up on Saroj's work in developing the trigger simulator for Run11 DSM logic. Saroj has finished most part of the L0/1/2 DSM algorithm. But there seemed to be some

minor problems. For example,  his starting point in calculating the board sum is the QT daugher card sum that he read out from L0 DSM inputs, where I think it should be started from

the QT channels themselves. So in my version the daughter card sums are calculated from the QT channels, and then they are subject to 5 bit suppression and truncation. And in Saroj

's code the mapping between North/South and QT# is incorrect. Everything else looks ok.

I also add more functions to produce QA plots, almost the same as those from Pibero's L0 monitoring page, and other interfaces to work with simluation (produce trigger bits from

simulated data)

 ps. Figure 2 b) without mass cut