Online Work

Steve's caiibrations page contains most of the details:

http://www.star.bnl.gov/protected/spin/trent/calibration/calibration.html

Some files of slopes, etc. that I produced are currently stored at

http://web.mit.edu/kocolosk/www/slopes/

2006 Gain uniformity

Using the gains we calculated for 2006 tower by tower from the MIPs and then corrected with the electron eta rings, I calculated how it differed from the ideal gain assuming containment of an EM shower with 60 GeV E_T. After removing bad towers, we can fit the distribution of this ratio to a gaussian and we find there is approximately a 6% variation in the gains.

 

Calibration Uncertainty Calculation

Summary:

The uncertainty on the 2006 BTOW Calibration is 1.6%. This value is the combination of a 1.3% overall uncertainty and a 0.9% uncertainty caused by variations in the different crates. This uncertainty should be treated as a measure of the bias in the 2006 Calibration.

Plan:

Attached is a document how the calibration uncertainty for 2006 will be calculated:

• Eliminate dependence on simulations through tighter fiducial volume cuts
• Reduce trigger bias by explicitly avoiding electrons matched to trigger HTs or TPs
• Validate modeling of E/p backgrounds. Confirm that the fit is unbiased by checking
  consistency of low-background and high-background samples.
• Confirm that crate timing does not systematically bias the energy reconstruction.

The uncertainty on the calibration will be assigned as the maximum between |E/p −1.0| and the uncertainty on the peak position.

Method:

We did some initial studies to determine the magnitude of each of these effects, and then we generated calibration trees covering the entire 200 GeV pp run from 2006. The code used to generate these trees is available in StRoot/StEmcPool/StEmcOfflineCalibrationMaker.

We made the following cuts on the tracks to select good electrons and an unbiased sample.

List 1:

• 3.5 < dEdx < 4.5
• abs(vertexZ) < 30
• 1.5 < p < 15 GeV
• dR (from tower center to track projection) < 0.004 (in units of deta,dphi)
• tower_id == tower_id_exit of projection
• Energy of highest neighbor < 0.5 * track energy

After making these cuts, we fit the remaining sample to a gaussian plus a first order polynomial, based on a study of how to fit the background best.

Figure 1 uses an isolation cut to find a background rich sample to fit:

Figure 2 shows the stability of the E/p location (on the y-axis) between our fit and just a gaussian for different windows in dEdx (x-axis)

Figure 3 shows the E/p location (y-axis) for different annuli in dR (x-axis/1000), which motivated our dR cut to stay in a flat region:

After making all of these cuts, we fit E/p to the entire sample of all our electrons. We then add different cuts based on the trigger information to see how that might affect the bias. We looked at four scenarios:

List 2:

1. All electrons after stringent cuts
2. Electrons from events that were NOT HT/HTTP triggered
3. Electrons from events that were only HT/HTTP triggered
4. Electrons from HT/HTTP events with trigger turnon tracks (4.5 < p < 6.5 GeV) removed

From these scenarios we chose the largest deviation from E/p = 1.0 as the overall uncertainty on the calibration. This happens to be scenario 3, working out to 1.3%.

Figure 4: E/p for different scenarios

 We also observed a possible crate systematic by fitting E/p for each crate separately.

Figure 5 E/p for each crate:

 According to the chi^2, there is a non statistical fluctuation. To figure out how much that is, we compared the RMS of these points to that when the data is randomly put into 30 partitions. It turns out that all of it is due to that one outlier, crate 12. Since crate 12 contributes 1/15 to each eta ring that it touches, the deviation of this point from the fit causes an uncertainty of 0.9%. This additional uncertainty increases the total uncertainty to 1.6%.

Side Note - Linearity:

After removing HT/HTTP events, we took a look at this plot of p (y-axis) vs E/p (x-axis). By eye, it looks pretty flat, which we verified by splitting into p bins.

Figure 6 p vs E/p

Figure 7 E/p vs p

 Eta Dependence:

There was some question about whether there was eta dependence. This was investigated and found to be inconsequential: http://drupal.star.bnl.gov/STAR/node/14147/

Figure 8: Divided the sample into 3 separate time periods. Period 1 goes is before run 7110000. Period 2 is between runs 7110000 and 7130000. Period 3 is after run 7130000. The deviations are below 1.6%.

Figure 9: Agreement between east and west barrel:

 Figure 10: ZDC Rate vs. energy/p

Figure 11: E/p fits for three different regions in ZDC rate: 0 - 8000, 8000-10000, 10000-20000 Hz.

Comparison of Electron and MIP Calibrations

This page compares the new tower calibration performed using only electron E/p to the calibration using last year's algorithm.  The two calibrations are found to be consistent within 120 MeV in E_T.



I've also attached the tower-by-tower plots of electron E/p so you can see the results for yourself.  I'll write up a more complete description of the calibration algorithm shortly.

Comparison of Online and Offline Calibrations

Introduction:

This page compares the online and first offline calibrations.  The online calibration table was generated during data-taking using a single long run processed through fastOffline production and uploaded on March 30th.  It uses slopes to set the relative gains in an eta-ring and then normalizes the eta-rings using MIPs.  The first offline calibration uses a significant fraction of the produced transverse and late longitudinal runs.  It sets the relative gains using MIP peaks and then uses electron E/p to set the absolute scale.  It was uploaded on December 7th.

Body Counts:

138 additonal towers are masked in this first offline calibration, leaving 4517 good towers.  1 tower (2916) was masked before but is now listed as OK.  To be honest, I have no idea why it was masked in the online calib; its slope looks fine to me.

Plots:

The electron E/p scaling in the offline calibration increased the gains by an approximately uniform 10 percent (more at the edges).  This effect is seen in the following plot of offline E_T - online E_T, integrated over all towers that were good in both calibrations:


Now, the interesting thing is the relative changes of offline-online for the east side and the west side.  If I only plot the location of towers whose gains increased by more than 20% I get




There were only 12 towers whose gains decreased by 20%; all of them were on the west side. Finally here's a plot of the E_T change of the remaining towers:



I think the message here is clear:  the gains on the east side have increased more than the gains on the west side!  It's possible that the use of the online calibration in previous Run 6 jet studies is at least partially responsible for the obsereved east-west jet asymmetry.

To get quantitative about this effect we have to go to 1D.  I've attached a PDF of eta-ring by eta-ring histograms like the first one on this page.  The first two pages are the east side; the next two are the west side.  I've found it easiest to analyze if you set your Reader to view two pages at a time; then you'll be comparing towers with the same absolute value of pseudorapidity when you flip.  The conclusion is pretty clear: at midrapidity the difference in offline - online E_T floats around 5 - 8 GeV on the east side, but it's only about 2 - 5 GeV on the west side.

Gain Stability Check

I've updated my codes to do a more systematic investigation of the stability of the gains.  Instead of trying to get sufficient tower-by-tower statistics for different time periods, I'm looking at MIP peaks for single runs integrated over all towers.  Here's the plot:


Features of note include the bump covering the first couple of days after the shutdown, the bunch of runs on day 123/4 with very low average peaks, and the general decreasing slope (consistent with towers losing high voltage).  I also ran this plot for west and east separately:



I'm running jobs now to do electron selection instead of MIPs.  I think that I can probably still do this as a function of run, but certainly I'll have sufficient stats to  plot vs. fill if necessary.

---

Goal:  Compare the tower slopes and MIP peaks from the following three periods to check for stability.

1. Day < 104 (97-99)
   
2. 104 < Day < 134 (114-119)
   
3. Day > 134 (134-139)

I just chose those periods arbitrarily.  I still have to generate the slope histograms for the longitudinal running, but of course I can extend the study to do that if it proves interesting.  Now, I filled these histograms without doing any trigger selection or status table cuts, so I wouldn't expect them to be the cleanest things in the world.  Consider it a "first look", if you like.  Anyway, here are gaussian fits to histograms of (slopeA-slopeB)/slopeA for all 4800 towers for each of the three combinations.  We see a 2.7% shift towards flatter slopes between days 97-99 and 114-119.  Note that this period includes the 10 day accident shutdown (104-114).  The slopes spread out quite a bit between periods 2 and 3, and there's also a small 1% shift towards flatter slopes.



The next set of plots compare gains extracted from MIP peak positions where the MIP peaks are generated using subsets of the Run.  Comparing before and after the shutdown yields a mean difference of 110 MeV with a width of 3 GeV.  This difference is significantly less than 1 percent.  The comparison between middle and late (essentially a comparison between transverse and late longitudinal running) indicates a 1.5 percent drop in the gains with a 2.3 GeV sigma.