Data / MC comparisons for isolation cut

Abstract: An isolation cut for the endcap is formed by taking the ratio of the transverse energy recorded by a single tower, correcting it for expected shower leakage, and dividing it by the sum of the energy of towers w/in R<0.3 of the candidate photon.  Data and MC are compared.

1.0 Absolute Normalization for Histograms

Let RI be the isolation ratio, defined as the single tower ET, corrected for shower leakage, divided by ET summed over radius of 0.3.
Figure 1 -- Single tower ET, corrected for shower leakage, divided by ET summed over radius of 0.3.  Left -- etabins 1-4.  Center etabins 5-8.  Right etabins 9-12.  Black triangles are the data (approx 1 pb-1 integrated luminosity).  Black histogram is the sum of jet background (blue) and prompt photon (red) monte carlo samples, weighted for 1 pb-1 integrated luminosity.  Bottom panels show the ratio of data / MC.  The scale is difficult to read... but ranges from 0 to 2.5.


0. Data sample: runs from pp2006 longitudinal after L2gamma trigger thresholds stabalized (ID = 137641)

1. Pythia's absolute normalization is used to create figure 1.

2. A pT cut of 5.2 Gev has been imposed on both data and Monte Carlo.

3. The only other cuts which are applied are (1) the event satisfied the L2gamma trigger (ID = 137641)

4. The agreement between data and MC is poor right where we need it to be good at RI ~ 1, where the photons are.  The origin of this effect is probably related to the sampling fraction     problem... smaller energy deposits require a larger "fudge" factor to get the thrown energy back.

5. Data and MC agree "well" (in shape if not normalization) in the region of RI < 0.5, for etabins 5-12.

  1. There is a peak in the data, reproduced  by the MC spectra, around RI ~ 0.4.
  2. This peak originates from photons which convert having the energy of the e+ and/or e- counted once in the numerator (the calorimeter tower), and multiple counted in the denominator (the e+ and/or e- track pT plus the calormiter ET).
  3. We cannot make a definitive statement about the normalization of this peak because
    1. Pythia cross section is not NLO (its not even LO)
    2. We do not know if TPC tracking efficiency or resolution is well modeled here, and
    3. The amount of material in front of the endcap is somewhat uncertain.
  4. The flat shape suggests that the distribution of material in front of the EEMC is described well.  However, the normalization changes from outer four etabins to middle four etabins.  Suggests either excess or deficiet of material in front of these areas.

2.0 Common Yield Normalization

In figure 2 below, scale the MC histograms so that the data and jet+prompt distributions have equal counts.


1. Things look "better" with common normalization.  Discrepancies between data and MC are reduced to the ~20% level for RI<0.8.  Of course, its still quite awful where it needs to be good. 

2. With this normalization we see the same level of discrepancy in all three eta ranges for RI < 0.8.  Details / statistics change, but the magnitude of the discrepancy remains the same.

3.0 Plans

We need to have good agreement between data and MC to hope to extract a signal... we especially need it where the photon peaks around 0.9 to 1.0, otherwise we would be insane to make a tight isolation cut.
Hal has identified a known issue: the sampling fraction in the MC appears to vary as a function of pT and eta, resulting in a pT and eta dependence to the energy of a reconstructed photon candidate and/or the invariant mass of a neutral pion decay.  Hal sent out a note to a small list this morning stating that he's tracked it down to the geant level... so this is likely a problem in the MC itself... so we're stuck with the MC response with the current data.
o Add an ad-hoc correction to the energy response of single towers when analyzing the Monte Carlo, and reproduce the gamma trees with it.
o See whether this improves the data / MC comparison
o Estimate material and tracking uncertainties based on the two different assumptions for normalizing the histograms.