Identification efficiency vs η

The identification efficiencies are acquired through calculating the binomial mean of the estimated signal after the background has been subtracted at each level. (The background is estimated with the OS-SS sideband method as was presented in Table 5.8 above.) The η dependence of the resulting electron identification efficiency with respect to track quality cuts can be seen in Figure 5.7 for loose, medium and tight, together with the true efficiencies obtained from MC. The efficiencies are measured

for opposite sign tag and probe pairs with an invariant mass within [80,100] GeV.

Only probes with transverse energy within [20,50] GeV are considered.

(a)

Figure 5.7: Electron efficiencies (upper plots) and data/MC scale factors (lower plots) for (a) loose, (b) medium and (c) tight identification. The data-driven efficiencies are obtained with the OS-SS sideband background subtraction method on Z tag-and-probe, with 20< ET <50 GeV probes.

The identification efficiencies and scale factors with their statistical uncertainty in eachηbin are explicitly stated in Table B.2, B.3 and B.4 in Appendix B. The overall efficiencies and scale factors (integrated over all η except the crack regions) obtained with the OS-SS sideband method are shown in Table 5.10.

[%] Efficiency Data/MC scale factor Loose ID Data 97.89 ± 0.33

99.51± 0.33 MC truth 98.37± 0.01

Medium ID Data 95.14 ± 0.35

98.86± 0.37 MC truth 96.24± 0.01

Tight ID Data 81.09 ± 0.42

103.34 ± 0.54 MC truth 78.47± 0.03

Table 5.10: Loose, medium and tight identification efficiencies, integrated over η, with respect to track quality forZ tag-and-probe, MC truth and their data/MC scale factor. The data-driven efficiencies are obtained with the OS-SS sideband method for probes within 20< ET <50 GeV and outside crack regions.

Tight versus medium scale factor

Studying the different plots in Fig. 5.7 shows that both loose and medium identifica-tion efficiencies are overall somewhat lower in data than in MC, yielding SFs below unity. This has been found to be due to broader shower shape distributions in data than in the Monte Carlo description [81]. The scale factors are however reasonably well-behaved as a function of η.

Taking a closer look at the tight identification efficiency in data and MC shows that the efficiency is significantly higher in data. In addition, the data/MC scale factor displays a strongη dependence. One or several of the tight identification cuts must be responsible for this large increase in the efficiency SF from medium to tight as well as the fluctuations of the scale factors inη³. An investigation is performed by adding one tight cut at the time to the medium identification and then computing the new SF normalized to the medium SF; SF(medium + tight cut)/SF(medium). This double-double ratio is shown in Figure 5.8 as a function ofη. In addition, the ratio of tight and medium scale factors is included in black. It is trivial to conclude from the figure that the cut on TR ratio is causing both the increase of the overall SF as well as the large fluctuations inη. Note that the TRT only extends up to|η|= 2.0 and the SF of the TR ratio as well as the cut on TRT hits is thus unity for 2.0<|η|<2.47.

The mis-modeling of the signal efficiency of the TR ratio in simulated data is largely due to the fact that the optimization of the cut is based on background rejection rather than signal efficiency. Figure 5.9 shows the TR ratio for the probe at medium identification level for data and MC. The MC distribution is shifted towards lower values with respect to data, implying that placing a selection cut on the variable rejects more signal events in MC than in data.

3The cuts involved at the different identification levels are outlined in Table 5.1

η

-2 -1 0 1 2

(Medium + bitX SF)/Medium SF

0.98 1 1.02 1.04 1.06 1.08 1.1

tight Conversion Blayer E/p

TRT hits TR ratio

η Track match

Figure 5.8: The ratio of efficiency scale factors for medium + tight cuts over medium.

The ratio of the tight and medium scale factor is also included in black. The cut on the TR ratio is causing the increase of the efficiency scale factor between medium and tight level, as well as the strong η dependence.

The end-cap efficiency asymmetry

Another observation that can be made from Figure 5.7 is that the efficiency scale factors for the positive end-cap (end-cap A) is consistently a couple of percent higher than for the negative end-cap (end-cap C). There is no clear explanation for this behavior, but the effect has been found to be larger for period D-H (especially period G) than period I.

To investigate the origin of this effect, the efficiency of the different individual identification cuts for loose, medium and tight have been studied. No single cut is however responsible for the end-cap asymmetry in period D-H since the effect is already present at denominator level.

Table 5.11 shows the number of probes at track quality (denominator) level, before and after background subtraction, for period D-H and I as a function of η. After background subtraction, there are approximately 8% more probes in end-cap C than in end-cap A for period D-H. In period I the number of probes at track quality level is equivalent in both end-caps within the statistical uncertainties. Before background subtraction there is already approximately 7% more probes in end-cap C in period D-H. This, combined with about 1% less background in end-cap C yield the observed end-cap asymmetry.

The end-cap asymmetry is much reduced if the probes in the 20-30 GeV bins are excluded and the efficiency is only estimated for probes with 30 < ET < 50 GeV.

TR ratio

0 0.2 0.4 0.6 0.8 1

Entries

1 10 102

103

Data MC

TR ratio

0 0.2 0.4 0.6 0.8 1

Entries

1 10 102

103

Figure 5.9: Electron TR ratio for data and MC obtained for probes at medium level.

The MC distribution is shifted towards lower values with respect to data. The plot is in logarithmic scale and the MC is normalized to the data.

The cause of this effect must therefore be an excess of worse quality probes in end-cap C, that are only present in early data and mainly at lower ET. Whether or not these probes are signal or background is not entirely clear, but they do fall in the narrow defined invariant mass range of 80-100 GeV and is thus treated as signal by the background subtraction method. There is also no excess of same sign events in the sideband region, since the fraction of background at the denominator level is consistent between the two end-caps, as was seen in Table 5.8.

To confirm that the end-cap asymmetry is not simply due to the OS-SS sideband method failing to classify the excess of probes as background, the background is also estimated by fitting the invariant mass at container level. The fitting method (which is further discussed in section 5.4.1) finds compatible background fractions in the two end-caps (9.9% in end-cap C and 10.2% in end-cap A) and the excess of probes in end-cap C is thus treated as signal.