PID Stability

I plot the nSigmaPion distributions for the 859 ppProduction and ppProductionMinBias runs passing 2005 jet QA. Tracks included in each histogram passed the following cuts:

  • pT > 2.0
  • |eta| < 1.0
  • |dcaG| < 1.0
  • nFitPoints > 25

Here’s a summary of the means from each run:

PID summary

I’ve also attached a PDF of the run-by-run plots at the bottom of the page (pid-stability.pdf). The group of runs around index 800 correspond to fill 7305 (highlighted in red on page 53 of the pid-stability PDF), and the large jump around index 95 coincides with the beginning of fill 7048.

I’m not so concerned about the latter group, as fills 7048 and 7055 were already excluded from analysis by the RHIC Polarimetry group. I am wondering, though, if it’s safe to analyze fill 7305. I tried the triple-Gaussian fit for data from this fill, which results in a 0.7 sigma offset for the pions. The Chi2/dof is certainly good enough:

F7305 recalibration

I went ahead and did these fits for every fill in my analysis and posted the results at the bottom of the page (pid-by-fill.pdf). The fit parameters were configured as

fit = ROOT.TF1('fit','gaus(0)+gaus(3)+gaus(6)', -6.0, 6.0)
fit.SetParameter(0, h.GetMaximum() * 0.9)
fit.SetParameter(1, 0.0)
fit.SetParameter(2, 1.0)
fit.SetParameter(3, h.GetMaximum() * 0.5)
fit.SetParameter(4, -1.0)
fit.SetParameter(5, 1.0)
fit.SetParameter(6, h.GetMaximum() * 0.05)
fit.SetParameter(7, 2.)
fit.SetParameter(8, 1.0)

I’m a little surprised at the fit results for these fills (i.e. pion means are often around -0.2 or lower), but I’m no expert when it comes to this stuff. Here’s a summary plot of means for the pion Gaussians in each of the fill-by-fill fits:

Fill Summary


nSigmaPion distributions are well-described in all cases by a triple-Gaussian fit, and are generally stable within a given RHIC fill. A few fills have distributions which appear to be shifted relative to the remainder of the dataset, in particular F7305.

Update 2008-02-05

I decided to change the PID selection window in my analysis to use the results of these fill-by-fill fits. Now, instead of using a fixed [-1,2] window in nSigmaPion, I center the window on my calculated pion mean for that fill, and the width of the window is normalized by the width of the pion Gaussian. The overall effect is hardly noticeable: pion identification efficiency (based on these fits) improves by 0.6%, and background p/K/e contamination drops by an even smaller amount. The actual numbers using my final runlist are

Efficiency: 81.9%
Background: 9.1%

Update 2008-03-07

I had been using the RunLog_onl DB to get fill/run mappings. I believe some entries in this database are incorrect; for more information, see this HN post. I just updated the results on this page to reflect what I now believe to be the correct mappings for fills 7127, 7128, 7129, 7134, 7136, and 7138.