April 22-24 updates:

Correction: in the study of single-photon events we compared the energy response of converted photons vs unconverted photons. For the case of converted photons, their reconstructed energy was always taken to be the CaloCluster energy, and never - the energy of the AOD photon objects. In fact, the AOD Photon objects are never actually used (except in this study from April 21). The reconstructed AOD electron objects are, however, used in favor of calo cluster energy of the electron/positron pair only in the reconstruction of higgs mass, and not in the single-photon studies.
• (Et-E)/Et histograms have been updated. Now we are considering small eta bins of 0.2 and doing a gaussian fit for the data between -0.3 and 0.3 (to avoid tail events).
• The latest reconstructed files are listed on this page. Apparently 11.0.42 reconstruction was posted on April 19, but there is nothing for 11.5.0 yet.

April 20-21 updates:

• Concerning single-photons results, I used the single-photon AOD/ESD data for 100 GeV photons located on Castor at:
  /castor/cern.ch/atlas/csc/valiprod/sampleA/mc11/004004.Photons_e100/recon/v11004101
  • I don't think these files are listed at http://jarguin.home.cern.ch/jarguin/dc3requests_urgent.html.
  • The root file used in my ROOT scripts and obtained from the Mike's Athena package is available here.
  • Apparently, the eta for these photons was limited to 2.5 when they were generated, which explains why the conversion histogram suddenly breaks off.
• Here is a more detailed study of single-photon events where we compare the energy response of converted photons versus unconverted photons.
• In (Et-E)/Et histograms, there are now cuts on the crack region (i.e. we are removing the events in which the resulting electrons are calculated to land between eta=1.37 and eta=1.52). This has improved the (Et-E)/Et histograms in the eta cuts near the crack. See the difference for 3x7, 3x5, 3x3, and TopoCluster (don't forget to scroll down on each of these pages). Note that there is no improvement in the TopoCluster case. It appears very likely that I am not using TopoCluster cells correctly.
• There are some new <(Et-E)/Et> vs eta plots where we sum over all R (so we get more statistics):
  • 3x7 all R and R-cuts (old)
  • 3x5 all R and R-cuts (old)
  • 3x3 all R and R-cuts (old)
  • TOPO all R and R-cuts (old)
  As expected, the variations have smoothed out quite a bit thanks to larger statistics.
• The performance of TopoClusters remains a mystery. I get consistently bad results with them, so there is likely a coding error. I am paticularly suspicious of the code, because specifying TopoClusters in the jobOptions file produces a CBNT root file with different MC truth, which should not happen! There most likely explanation is that MC truth looks different simply because we are considering a smaller subset of events (the rest were filtered out by Mike's algorithm). I know how to check this, and I will try to look into it later.

April 19-20 results:

Plot type \ CaloContainer:	Single Photon LArClusterEM37 (.root)	Higgs LArClusterEM37 (.root)	Higgs LArClusterEM35 (.root)	Higgs LArClusterEM33 (.root)	Higgs EMTopoCluster (.root)
E and Et histos; cuts in R, eta	png	png	png	png	png
<(Et-E)/Et> vs eta; cuts in R	png	png	png	png	png
(Et-E)/Et histos; cuts in eta	png	png	png	png	png
E/Et histos; cuts in eta	png	png	png	png	png
Conversion rate vs eta; cuts in R	png	png	png	png	png
Conversion rate vs eta; all R	png	png	png	png	png

Notes:

• The single-photon root files are Mike's files as I didn't yet run the athena job for single photons.
• The 1 - 2 % variations as a function of eta (over small changes in eta) is present in the single-photon case. In fact, it appears that the variations are sometimes much bigger, but this can (at least in part) be attributed to a much lower statistics in the sample.
• Right now we are attributing these variations to the fact that the calibrations constants used in 11.0.41 were tuned for the Rome layout in which the material distribution was slightly different. But there could be more to it.

  ---

• Using TopoCluster calorimeter cells produces weird results: the MC truth in the resulting CBNT file is different compared to when we use 3x7, 3x5, or 3x3 cells. This is likely due to my misunderstanding of how TopoCluster cells should be used (or whether they should used at all).

  ---

• It appears that the best way to reconstruct converted photons is to use 3x7 calo clusters. This can be seen from <(Et-E)/Et> plots (second row in the table). In fact the performance of 3x3 cells is already very bad (or I am using them in a wrong way).

Source code:

• I am looking at (a) higgs->gamma,gamma and (b) single photon events and considering how well we can reconstruct converted photons depending on what part of the detector the conversion occurs in. I also look at how using 3x7, 3x5, or 3x3 calo cells affects the results. Based almost entirely on Mike Hance's work.
• A complete package with all root files is available here (~40mb). Use "tar xvfj package.tar.bz2" to extract.
• Here is the conversions.C file and its HTML-browsable version.
• NOTE: there are links to individual root files and individual root functions in the table below!
• Functions available in the code:
  • zhelp(): Prints a summary of available functions
  • z0(): Et and E histograms for 3 cuts in R and 3 cuts in eta
  • z1(): <(Et-E)/Et> as a function of eta for three different cuts in R, converted only
  • z10(): <(Et-E)/Et> as a function of eta summer over ALL R, converted only
  • z11(): <(Et-Ereco)/Et> for SINGLE PHOTONS, non-converted, as a function of eta
  • z12(): <(Et-Ecalo)/Et> for SINGLE PHOTONS, non-converted, as a function of eta
  • z2(): (Et-E)/Et histograms for 6 different cuts in eta
  • z3(): E/Et histograms for 6 different cuts in eta
  • z4(): Conversion rate as a function of eta for 3 different cuts in R
  • z5(): Conversion rate as a function of Eta, summed over all R
  • zz(): Make all plots. VERY SLOW!
• Suggestion: it is convenient to look at the plots by considering a row at a time. For instance, I like to open all links in each row in a separate tab using firefox.