BTOW Calibration Uncertainty Update


Alan, Matt, Mike, and I investigated some properties of the 2006 BTOW calibration with the ntuples that were used to generate the gains in the DB. Here are our preliminary findings:

  • The calibration has a strong dependence on momentum for HT triggers, lessened somewhat for the HT=0 sample, but still present (note that HT=0 still includes L2gamma triggers). Cutting out 4.5 < E_T < 6.5 GeV reduces but does not solve the problem. We need new data with proper trigger info. This is shaping up to be the dominant contribution to the uncertainty on the calibration.

  • E/p distributions binned versus η and ϕ show some non-statistical fluctuations. A crude analysis puts the size of these fluctuations at ~1%.

  • E/p has a negligible dependence on fiducial volume for R < 0.006. We recommend using this volume (or perhaps a box of |deta|, |dphi| < 0.006) for further analysis, as it eliminates any dependence on simulations. Beyond R of 0.006 the transverse shower leakage in the data is larger than predicted from single particle MC simulations.

  • The hadronic E/p background looks like it falls off linearly underneath the electron peak. Surprisingly, the extracted peak position does not really depend on how we model this background. A background ~flat in E/p persists regardless of where we place the dE/dx cut and does bias the peak position, but seems to be removed with our (extremely strict) isolation cut. We plan on tuning that isolation cut to something less extreme.

Nominal Cuts

We have the following fields available to us in the current version of these ntuples:

p: track (primary?) momentum
teta: track eta at tower
tphi: track phi at tower
dedx: track dedx * 10^6
np: # of fit points
id: tower id
eta: tower eta
phi: tower phi
energy: tower energy
r: sqrt((teta-eta)**2 + (tphi-phi)**2)
vz: vertex position
iso: all surrounding towers have adc-ped < 2*rms
ht: 117221 || 117212 || 127212 || 127213 || 137213 (HW only)

and we use the following cuts to select “good electrons”:

dedx>3.5 && dedx<4.5 && np>25 && r<0.006 && abs(vz)<30

Trigger/Momentum Dependence

E/p has a huge variation with momentum right around 5 GeV (approximately the HT and L2gamma trigger threshold). Figure 1 shows the variation with momentum integrated over all triggers; Figure 2 looks at HT triggers only.

Figure 1: all triggers


Figure 2: HT triggers only

η/ϕ/Crate Dependence

We looked at E/p distributions for each crate. The individual plots are attached at the bottom of the page, while Figure 3 shows the extracted peak position and the error on the extraction. Crate 12 is a known problem, although people often still include it in analyses.

Figure 3

Next we do the same procedure, but this time slicing things up by pseudorapidity. Details are attached, summary results are in Figure 4:

Figure 4

In each case the fluctuations in the extracted peak position have some non-statistical contribution. If we histogram the peak positions (tossing out crate 12) and take the RMS of that distribution, we can quantify the non-statistical contribution as

syst^2 = RMS^2 - stat^2

where stat is the median statistical precision of the individual E/p fits. This procedure yields a systematic of ~1% in each case. We haven’t decided how these contributions should be combined or if they should enter into the correlated uncertainty on the calibration. It doesn’t seem fair to toss them into an “uncorrelated tower-by-tower” uncertainty term.

Fiducial Volume Cut

Here’s a plot of the E/p peak position extracted from annuli in R:

Figure 5

The x-axis is basically 1000*R. Using R<0.006 doesn’t seem to introduce any significant bias to the calibration.

Fit Stability, Isolation

The effect of the dE/dx and isolation cuts on the E/p distribution is shown in Figure 6

Figure 6

We’re going to explore a less stringent isolation cut. In Figure 7 we show results from fitting to purer and purer samples. The x axis is the lower dEdx cut (the upper cut is 4.5), the y axis is the peak determined by the relevant models. For the most part, the purer the sample the fewer the statistics so there’s no really chance for disagreement. The one interested feature is the deviation of the Gaussian only model in the purer samples. At low purity the background is roughly flat, or at least any slope does not bias the peak position. In purer samples, however, there seems to appear an irreducible, skewed background that does bias the peak. The additional freedom of a line can handle it pretty well.

Figure 7

Next Steps

Matt is generating new ntuples with proper triggering information and more detailed isolation information (3x3 cluster energy and energy of highest neighbor). These are almost complete. Highest priority at that point will be to decide on a trigger/momentum cut.