CAL Shower Shape Analysis Pass7 vs Pass8

This is a first attempt at comparing the performance of the CAL shower shape analysis between pass7 and pass8.

The file used for the work are:

For Pass7:

AG-GR-v17r31-StdCuts-merit.root

BKG-GR-v17r31-StdCuts-merit.root

where StdCuts are: FswGamState == 0&&  CalEnergyRaw>  5&& TkrNumTracks>  0&&  CalCsIRLn>  4.

For Pass8:

AG-GR-v20r3p4-OVL

BKG-GR-v20r3p4-OVL

I include the following precuts to the allGamma sample for both Pass7 and Pass8:

FswGamState==0 && TkrNumTracks > 0 && (CalEnergyRaw > 5 && CalCsIRLn > 4) && CTBCPFGamProb>0

note for Pass8 CalEnergyRaw is substutited by CalUberEnergy and I have added the cut on CTBCPFGamProb because it is necessary that the sample of events pass the

Cpf selection before I can compare the Pass7 and Pass8 selection in the CAL.

To the BKG sample I include in addition to the above selection the condition that there not be gamma's in the sample, i.e. a cut on

McSourceId != 7000

Useful References

• Bill's Primer on Background rejection: Cal Shower shape is describe from slide 33 (pdf)
• Old CAL talk in Bari ppt

Here is a snapshot of the CAL shower shape worksheet just for reference.

The selection conditions assigned to each node are defined below.

Note that all the CTB variables have been redefined based on their definition found in the workbook. N.B for variables such as XtalRmsTkrCoreDoca, TkrSHRCalAngle, and TkrCalTrackDoca which are based on the merit variable CalTrackDoca in Pass8 this variable is for the first cluster and not the uber one. While the CalEdgeEnergy variable (which is also used to define CalEdgeEnergyRatio) is for the uber cluster, therefore there is a inconsistency for the Pass8 selection. We have looked into creating the CalEdgeEnergy variable for the first cluster and stumbled across the issue that the getEnergy() method from recon is different from that in the CalValsTool. The variables used for the BackSeal veto (i.e. Cal67Sum and Cal67Diff ) are for the first cluster, the variable CalLLRms also used in the BackSeal veto is for the uber cluster therefore this condition needs to be fixed in order to make it consistent. And finally also CTBBestEnergy in Pass8 is for the first cluster.

veto BackSeal: ((Cal67Sum - 66.)/33.5)^2 + ((Cal67Diff + .4)/7.2)^2 < 1.2 && CalLLRms < 30

veto LowE: CalEdgeEnergy>0 && CalEdgeEnergyRatio >  (LogCalEnergyRaw -.5)/2.

filter VTXSplit: VtxAngle > 0

veto VtxSideSeal: (CalDeltaEdgeEnergy < .6  && CalDeltaEdgeEnergy < 6/CalEdgeEnergy & CalEdgeEnergy < 190)

filter Vtx: (CTBTkrCoreCalDoca<32.3 || (CTBTkrCoreCalDoca<152 && XtalRmsTkrCoreDoca<-10)) && XtalRmsTkrCoreDoca < (2- .15*CTBTkrCoreCalDoca)  && Tkr1CoreHC > 0

filter notVTXSplit: VtxAngle == 0

veto TkrSideSeal: (CalDeltaEdgeEnergy < .75 || (CalDeltaEdgeEnergy < .95 - (30./(CalEdgeEnergy+20))^2 & CalEdgeEnergy < 300)) && CalDeltaEdgeEnergy < (1.1 - CalEdgeEnergy/200)

filter nTkrFront: TkrNumTracks > 1 && Tkr1FirstLayer < 5 && (CTBTkrCoreCalDoca<40.56 || (CalXtalRatio<0.105 & EvtECalTransRms<3.45))

filter nTkrBack: "TkrNumTracks > 1 && Tkr1FirstLayer > 5 && ((CTBTkrSHRCalAngle>0.142 && CTBTkrCoreCalDoca<16.7) || (CTBTkrSHRCalAngle<0.142 && XtalRmsTkrCoreDoca<-15.4))

filter 1TkrFront: (TkrNumTracks == 1) && Tkr1FirstLayer < 5 && (EvtECalTrackDoca < 2.13 || CTBTkrEnergyFrac > 0.285)

filter 1TkrBack: (TkrNumTracks == 1) && Tkr1FirstLayer > 5 && !(XtalRmsTkrCoreDoca > -3 && CalTrkXtalRmsE > 12)

Selection efficiency per node Pass7 vs Pass8.

N.B. There is a typo on the y-axis for the 2d plots!! Both histograms are plotted for McZDir.

VtxAngle issue

The number of vertexed events (VtxAngel >0) is roughly a factor of 10 less in Pass8. The issue is that the extraction of tracks from trees does not result in 2 tracks which vertex very often, Tracy has been working on this. Below is the comparison of VtxAngle for Pass7 (v17r31), v19r4p1gr14, and v20r3p4 which shows the difference in this variable between the three passes.

FirstEdgeEnergy

I took the recon files and created the CalEdgeEnergy variable for the first cluster to see if there was any impact on the selection. Unfortunately the recon files for v20r3p4 were somehow corrupt therefore in order to try this out I needed to use the recon files for the v19r4p1gr14 series. Below are the distributions for the Pass7 and Pass8. The statistics are limited (there are only 100 recon files available) thus it is hard to say anything however it looks like there is no substantial difference. I applied the same precuts as described above except the CTBCPFGamProb>0 because it removed too many background events and I wanted to see the overall effect of using the first cluster was.