HIGH-THRESHOLD ANALYSIS: DATA SELECTION
3.5 Physics cuts
(a) (b)
Figure 3.7: The LF noise cut is very similar to the glitch cut described in figure 3.6.
An LF template is constructed (a) and events that reconstruct as more LF-noise than true event are rejected (b). [69, 51]
Good Start Time (cGoodPStartTime_v53)
During the OF fit, the start time of a pulse is allowed to float within a window around the trigger time. Ideally an event would be reconstructed near the center of the window. Some low energy pulses will have rise-times that are slow enough that the true start time of the pulse falls outside this window, causing the fit to “rail”.
Similarly, due to the high event rate in some calibration datasets, “cross-detector pileup” events can occur that “rail” at the tail end of the OF search window. This occurs when the secondary trigger is issued just after the OF search window. These are excluded by ensuring that
(EPhonontotal > Emin)∩(−190µs<tOF <tthreshold) (3.1) Where Emin and tthreshold are detector-dependent thresholds. For more on this cut see [70].
efforts. As a result, they act to reduce the size of our dataset, and are generally applied uniformly with the above time-period and reconstruction cuts.
Muon Veto (cVTStrict_133)
Any event that is within 50µs of a muon-veto trigger is removed.
Triggered event (cTriggeredEvent_133_HT)
The correct functional detector response includes an estimation of the trigger effi- ciency. This is calculated based on the number of events that actually trigger the detector, so for it to be valid, we can not include any pulses from a given detector that didn’t actually trigger the experiment. This cut requires that any event under study for a given detector must have issued a trigger in that detector in the time window from−200µs to+100µs of the event.
Phonon Consistency Region cut (cPCfSVM_2pct_Sandbox_v53_HT)
Figure 3.8: One-class support vector machines (SVM) can be used as an outlier rejection tool, which is the basis for the phonon consistency region cut. Plotted is the phonon radial partition for side 1 vs side 2 of 252Cf calibration events for detector IT2Z1. The cut removes the outlying 2% of events (blue) from the phonon partition distribution [71].
sification algorithm. The phonon partition quantities are combined in the SVM machinery and the outlying 2% of events were removed to make a clean sample.
Localized Low-Yield Phonon cut (∼cSpot_v53_HT)
During the investigation of the γ-sourced backgrounds, there was some surprise expressed at the large numbers of single-scatterγ’s (from133Ba calibration) leaking into the nuclear-recoil band. When examining ionization quantity planes such as qzpartOFvsqrpartOFit was generally seen that these events had a preference for qzpartOFvalues larger than 0 and did typically occur at higherqrpartOFvalues.
Unfortunately, they did not occur in a very tight cluster that would suggest easy removal. During some cross-checks, we noticed that these events were relatively tightly clustered in the phonon partition plane on both sides of the detector. These events occur along the flats of the wafer, near the DIB on each side of the detector, though the phenomena is more pronounced on Side 1. It is believed that these events are caused by the voltage on the DIB that is adjacent to the flat causing events to preferentially drift into the sidewall of the device. While this is a design flaw in the electronics, since the events are relatively localized, we can simply remove that portion of the detector. To accomplish this the “Spot cut”6 was created to selectively remove these events. The variable in which the spot is most local is pthetaOF. Figure3.9 shows histograms of pthetaOF for both 252Cf- as well as
133Ba-calibration data. If we take the 252Cf data to represent “normal” detector response we can use the ratio of133Ba to252Cf to gain an insight into where there is an excess ofγ-sourced events.
pxpartOFandpypartOFcan be dealt with in a similar fashion topthetaOFexcept the 1d marginal distributions globally are hard to work with due to the triangular shape of the distribution inpxpartOF/pypartOFspace. To simplify things we first
6So named because it appears that there is a “bad spot” in our detectors.
(a) (b)
(c)
Figure 3.9: The “spot” of low-yield γ-sourced interactions is most localized in phonon θ-partition space. To find it the distribution of 252Cf-sourced events (a) (assumed to represent uniform detector response) is subtracted from the same dis- tribution constructed of low-yield 133Ba-sourced events (b). The residual is fit to a Gaussian (c). This process is repeated in the X and Y phonon partitions and the intersection of these three Gaussians are removed.
constructed 2d histograms in pxpartOF/pypartOF space, found the peak and re- stricted ourselves to its local neighborhood. After restricting the domain to around the peak, we use the same re-weighting and fitting method as described above with thepthetaOFdistribution. The intersection of the 2 sigma bands in all three vari- ables calculated from the Gaussian fits to define the “spot”, and can be seen in figure 3.10.
Ionization Threshold cut (cQthresh_v53_HT)
To remove the zero charge events, a charge threshold was included in the analy- sis. This cut was designed to remove any charge signals within 4σ of either the
(a) Side 1: before cut (b) Side 2: before cut
(c) Side 1: after cut (d) Side 2: after cut
Figure 3.10: Plots showing the action of the spot cut for detector IT2Z1. Each is the phonon Y vs X partition plane and depicts133Ba-sourced events with those that fall inside the 3σNR-band highlighted in blue.
run averaged zero-energy noise distribution or the series-specific zero-energy noise distribution. The overall effect of this approach is to apply the average threshold in most cases, but to raise the threshold for more problematic series throughout the course of the run.
Energy Threshold (cAnalysisThreshold_v53_HT)
For each detector, the analysis threshold is the largest of the 95% trigger efficiency energy, the blinding lower energy limit, or a fixed value of 4 keV.