- My blog
- Post new blog entry
- Top bloggers
- Recent posts
Sort by:[Date]
test
Updated on Tue, 2009-03-17 08:44. Originally created by xpzhang on 2009-03-17 08:43.mcVertexZ.gif
QM09 v1
Updated on Wed, 2009-03-25 08:59. Originally created by cperkins on 2009-03-17 00:04.QM09 Talk (v1) - Chris Perkins - Run8 p+p and d+Au
New PID Asymmetries II
Updated on Mon, 2009-03-16 20:22. Originally created by kocolosk on 2009-03-16 20:21.This post is basically a correction to New PID Asymmetries.
Run 8 Pions - Correlations
Updated on Wed, 2009-06-03 14:36. Originally created by stevens4 on 2009-03-16 15:58.Correlations with Invariant Mass
Data Set1: STAR 2008 pp data
- I used the following query conditions in the file_catalog for this study:
EEMC Tower and Mapmt Channel Prinouts
Updated on Mon, 2009-03-16 15:50. Originally created by aliceb on 2009-03-16 15:50.Attached are prinouts with the ratio of ADC-ped=20 to ADC-ped=100/total at what appears to be the optimal time delay for the crates. For towers I have everything at 40 and 60 ns and for the m
Analyzing Day 74 ZDC data
Updated on Mon, 2009-03-16 17:03. Originally created by aliceb on 2009-03-16 14:18.Runs 10074004,5
New PID Asymmetries
Update: I coded up the statistical uncertainty calculation incorrectly in this post. I forgot to divide by the purity! The optimal cuts changed by a lot when I fixed that error. I also generalized the formula to account for the presence of signal in the sidebands, and I included the lowest p_{T} bin in the analysis. For more details see New PID Asymmetries II. Bottom Line — the results on this page are wrong!
In an earlier post I explained the new method for calculating identified particle yields that I’m using in my A_{LL} analysis. I began that study because I planned to calculate A_{LL} differently than I had been in the past. Specifically, I wanted to incorporate the proton/kaon/electron backgrounds into the statistical uncertainty instead of assigning a separate (statistics-limited) systematic uncertainty to account for their presence. I’m using the following formulas for A_{LL} and its statistical uncertainty:
where the p_{T}-dependent background fractions are defined as:
I wrote a small function to estimate the statistical precision on A_{LL} given the p_{T} bin, pion acceptance window, and sideband acceptance windows. I didn’t care about the absolute statistical precision, so I just used 1/sqrt(N) for the uncertainty on each A_{LL}. I used Minuit2 to minimize this function and extract the optimal acceptance windows, with the constraint that the purity in each sideband is never below 90%. In principle, this approach would yield four momentum-dependent cuts. In practice, the p+K sideband cut and the left side of the pion acceptance window only had a small momentum dependence, so for the sake of simplicity I keep them fixed. I also choose to fix the other two cuts in each p_{T} bin instead of letting them vary with momentum. In the end I employ the following cuts
| pT bin | π window | max p+K | min electron |
|---|---|---|---|
| 3.18 - 4.56 | (-1.90, 2.40) | -1.90 | 2.40 |
| 4.56 - 6.32 | (-1.90, 2.25) | -1.90 | 2.50 |
| 6.32 - 8.80 | (-1.90, 2.00) | -1.90 | 2.60 |
| 8.80 - 12.84 | (-1.90, 1.50) | -1.90 | 2.60 |
These cuts are significantly wider than the (-1.0, 2.0) acceptance window I had been using in the past. Apparently the reduction in purity is more than offset by the extra efficiency.
The electron side of the acceptance window is interesting. As momentum increases the pion band moves closer to the electron band. As a result, we need to move the electron sideband cut further out to maintain the 90% purity. This cuts down on the electron background A_{LL} statistics. The minimizer compensates for that uncertainty by restricting the right side of the pion acceptance window and thus reducing the electron background fraction.
I compared the uncertainties obtained by the minimizer with the uncertainties from my old method (a flat (-1.0, 2.0) cut that does not subtract out the background asymmetries). It turns out that the uncertainties from the new method are actually smaller in every p_{T} bin. I haven’t calculated a systematic uncertainty for this method, but if there is one it will be far smaller than the systematic from the old method (~ background fraction * sigma of background A_{LL}). In other words, using the new method is a no-brainer.
Oh, and one plot, just because I think it’s pretty:
New PID Cuts
Updated on Sun, 2009-03-22 10:56. Originally created by kocolosk on 2009-03-16 09:09.Update:I determined the electron peak positions in the p < 3.00 bins by letting that parameter float instead of fixing it to the Bichsel value. The difference was about 0.75σ in the lowest bin. The χ2 for that bin is dramatically better. Some of the other higher momentum slices in the first bin still have large χ2 values, but for the moment I’m working under the assumption that the background fractions are basically correct. So that bin’s back in. I’ve updated the PDF with the new fits.
Historically, I’ve used separate triple-Gaussian fits for positively- and negatively-charged tracks (18 free parameters in total) to extract the pion, proton/kaon, and electron yields for my A_{LL} analyses. There are several problems with that approach
- To first order, dE/dx resolution is independent of particle species.
- The peak positions for pions, kaons, protons, and electrons are not charge-dependent.
- Those peak positions (or at least the separations between them) have been determined with a great deal of accuracy in other analyses.
- dE/dx scales with momentum, not p_{T}. At y = 1.0, p ~ 1.5*p_{T}. This difference was not being taken into account.
I set out to address those three points by redoing the fits in the manner employed by the lfspectra working group. I fill a 3D histogram of p_{T} vs. p vs. (nσ(π) + 6*track.charge()), and then fit individual xy slices of this histogram with a function comprised of 8 Gaussians. That function is subject to the following constraints:
- all Gaussian widths are identical
- π+ mean == π- mean (after subtracting imposed offset of 12)
- K+ mean == K- mean (ditto)
- proton mean == pbar mean (ditto)
- positron mean == electron mean (ditto)
- π mean - K mean is a fixed function of momentum
- π mean - p mean is a fixed function of momentum
- π mean - e mean is a fixed function of momentum
That leaves 24-14 = 10 free parameters for each fit. I received the momentum-dependent particle separations from Yichun Xu. Her work is documented in this NIM draft, and the actual values for the separation are posted at (columns are p, p/mass, and separation):
π - electron separations
π - kaon separations
π - proton separations
The results of the fits are attached as a PDF. The fits in the first p_{T} bin (2.00 - 3.18) are not good. Part of the reason is that Yichun’s analysis doesn’t go below 3.0 GeV/c, so I used a Bichsel parameterization for the particle separations instead of taking them from data. This bin is complicated because at the low momenta the protons are entering the 1/β^2 region. I think the sensible choice may be to drop the bin from the A_{LL} analysis.
The fits for the higher p_{T} bins are much better, particularly at mid-rapidity. I don’t really believe in the p/π and K/π ratios that come out, but in the end I think the only important thing for my analysis is that the (p+K)/π ratio is correct. I’ll explore the effect of these new fits on A_{LL} in a separate post.
Draft of Junior's Talk 2009
Updated on Fri, 2009-03-20 16:00. Originally created by pagebs on 2009-03-15 02:40.Here is the draft of my talk for the Juiors meeting at the 2009 collaboration meeting.