Offline Evaluation of L2gamma EEMC 2008 Algorithm

Abstract: We evaluate the 2008 EEMC L2gamma algorithm for run 9050015.

Summary

1. Figures 1.1 - 1.3 demonstrate that the trigger is forming clusters of adjacent towers.  They show the tower-index (or eta or phibin) of the 2nd highest tower in the event plotted versus the leading tower in the event.  The patterns in the plot are consistent with the desired clustering pattern.

2. Figure 1.2 shows the ADC spectrum of the highest tower, determined offline.  The spectrum turns on at ADC=100 channels above pedestal.  This is at an E_T value of 1.5 GeV, which is where the online cluster finder's seed threshold was set for the specified run.  Above ADC=200 channels above pedstal, the distribution follows an exponetial decay.  This is consistent with triggering on clusters of 3 GeV, where the high tower is now capable of firing the trigger by itself.

3. Figure 2.4 compares the pT of the highest 2x2 cluster found offline to the pT recorded in the L2 result array.  They are consistent w/in about 4%.  The 4% spread is consistent with the reduced resolution with which we save the online result.

We will perform the following tests:
1. Does the event topology look right?  i.e. do we see the 2x2 clusters in the event?  Do the distributions look reasonable?
2. Does the online trigger cluster match up with the offline cluster?

Database configuration (sector 1)

StInfo: mapping=S1 rot, P1 as in sect=5, swap TA4-5, QA11-B2, JB
StInfo: tower calib=Ideal EEMC tower gains, E_T 60GeV=4096ch, RF
StInfo: PMT names=oryginal 1:1 , JB
StInfo: MAPMT calib=final 2006 p+p, using MIP UxV 7vetoXX7veto, from day 89, Scott
StInfo: ped tower & MAPMT=ped, ofl, generic AuAu 2007 day 95+, from R8141062, Scott
StInfo: status tower & MAPMT=Startup ofl status table EEMC, before 2008 run, Scott

Oh yikes... peds are old! 

1.0 Test the topology of the trigger

The L2 gamma trigger algorithm (in the EEMC) behaves as follows:

1. Find all seed towers above 1.5 GeV
2. Sort seeds descending in E_T
3. Loop over seed towers
    a. Find the highest 2x2 cluster containing the seed tower
    b. Add to list of clusters
    c. Mark all towers in cluster as used / remove from further consideration
    d. Continue loop until all seeds exhausted
4. Find the highest E_T cluster in the event
5. Accept event if E_T > 3.0 GeV.

Figure 1.1 -- Plot leading tower index, phibin, etabin vs subleading tower.

We see the expected behavior: an enhancment for the 2nd highest tower adjacent to the leading tower.

Figure 1.2 -- ADC  and E_T spectrum for high tower (note that titles which say "3x3" cluster should read "2x2".

1. We see a sharp turnon around ADC = 100 which is ~right.
2. There are ~50 events / 12k which appear below threshold.  This _may_ indicate a stale-data failure rate. (?)
3. Above ~4 GeV we see an exponential falloff of the HT E_T distribution.

Figure 1.3 -- Ratio of seed energy to 2x2 cluster energy (or ET).

I think this is the correct behavior.  We'll compare with the online result later.

Figure 1.4 -- Which of the 4 possible 2x2 clusters is associated with the seed tower.

1. The enhancement for position 0 is understood.  It is the first position tested.  If there is only 1hit tower in the cluster, then position 0 will be preferred.
2. The depletion of position 4 is also understood.  If the cluster only contains 2 clusters, those clusters will be found in one of the 3 preceding positions before we test position 4.

Comparison between the online trigger and offline reconstruction

Figure 2.1 -- Comparison of the online and offline high tower position. Plots show the online phi(eta) bin vs the offline phi(eta) bin.

The online trigger is finding the highest tower in the event, as intended.

Figure 2.2 -- Comparison of the E_T of the highest 2x2 cluster found offline to the online cluster.

Looks like the Nike Swoosh (tm).  This can't be good.  The offline E_T is reconstructed systematically high.  And it tails off significantly at low pT...

Figure 2.3 -- Comparison of the E_T of the highest 2x2 cluster found offline to the online cluster, versus eta.

Y-axis shows the eta-bin in the EEMC.  In other words, 0 is closest to the beam, 12 is closest to the BEMC.  There appears to be a slight eta dependence here.

02/22/2008

• Stale pedestals are not the problem.  Jan loaded peds from day 27.  Rerunning the code, I get the same plots as above.  Shifted down by 1%.
• Need to investigate further...  There's likely a bug somewhere.

02/27/2008

Tracked down the discrepancy in figures 2.2 and 2.3.  They were due to a bug in the offline reconstruction.  I was double-counting a tower (and not the seed tower, which made it fun to debug.)

Figure 2.4 -- Plot of the (E_T offline - E_T online)/E_T offline vs E_T offline with the bug fix

References

1. Steve's notes on the test run.
2. Run browswer entry for run 9050015.
3. Link to the online QA plots.
4.