Run 68 -125V Cut Efficiencies

The cuts used are described here but are applied in the following order:

//Order of cuts
TCut cuts_bkg[]={!crand, cFstd, !cBShi, cofdel, cofwktiming, cofint_bkg, cchit_bkg, !cspikey};
TCut cuts_Na[]={!crand, cFstd, !cBShi, cofdel, cofwktiming, cofint_Na, cchit_Na, !cspikey};
TCut cuts_PuBe[]={!crand, cFstd, !cBShi, cofdel, cofwktiming, cofint_PuBe, cchit_PuBe, !cspikey};

The sections below describe the efficiency of each cut. The efficiency curves in each section show the relative efficiency of adding that cut after all previous cuts. A discussion of all cut effects is in the final section. Calibrated PTOFamps is used as the energy quantity.

Note that all of the work below is with a sparse subset (only ~10%) of the total dataset.

This just removes random events. We use it to consider only triggered events for the analysis. The trigger efficiency (studies elsewhere) determines which true physics events meet the trigger conditions.

This was implemented specifically to remove retriggered events due to phonon oscillations on the falling edge. It is assumed that these are 'duplicate' events, in that we've also triggered in the proper location on the pulse in a prior event. As such, this cut is not expected to change the spectrum of good, measured events.

Relative Efficiency	1 keV	30 keV
Bkg
Na
PuBe

N.B. This uses v2 of the cBScut.
This cut is based purely on the prepulse baseline value of PF. As such, it should have no correlation with the true energy of events in the detector. It should simply be a flat efficiency related to the rate of high energy events, the likelihood of being able to trigger on events upon an elevated baseline, etc.

Relative Efficiency	1 keV	30 keV
Bkg
Na
PuBe

This cut removes events for which PTOFdelay is at an extreme position in the allowed window. This likely indicates that there would have been a better OF fit outside the window. This occurs for mostly noise events, so again the Bkg takes the largest hit due to the relatively high fraction of noise events in data.

Non-noise events can also fail this cut if there is a second pulse (pileup) elsewhere in the trace. There is then expected to be some energy dependence to this cut for the following reason. If there is pileup in a trace, the OF delay will be most affected if the pileup event is larger than the triggered event. Roughly speaking, then, the likelihood that an event fails this cut goes like the likelihood of it having a larger pileup event. For a flat spectrum (again, a crude approximation) this probability will go as

<m>P(cut|pileup)~ ∝ ~P(E_pileup>E_1)~ ∝~ (1 ~- ~E_1/E_max)</m>

i.e. it ranges linearly from 100% for <m>E_1→0</m> to 0% for <m>E_1 → E_max</m> for some <m>E_max</m>. This will result in a component of the efficiency which is linear and increasing with increasing event energy. This is of course a gross oversimplification, but this shape can be observed in both this cut and the other 2 pileup cuts, ofwktiming and ofint.

Update: Looking more closely at the events which are cut by this, there are 2 main classes; pulse triggers w/pileup and noise triggers w/pileup. We'll need to distinguish between these for efficiency studies.

Relative Efficiency	1 keV	30 keV
Bkg
Na
PuBe

This cut is designed to remove event where the OFdelay and RTFTWK algorithm timing quantities are inconsistent. This can occur if there is pileup even when the pileup event is in the last few bins of the trace. This is rare, but does occur. The OFdelay and RTFT can disagree for pure noise traces as well. This makes up most of the events removed by this cut. The Bkg data, with the lowest event rate, is hit hardest by this cut because a higher proportion of triggered events are in the noise for this dataset.

Relative Efficiency	1 keV	30 keV
Bkg
Na
PuBe

This cut is designed to remove pileup events. While the presence of a pileup event should be uncorrelated with triggered event energy, this cut is expected to have energy dependence because it is defined in terms of energy rqs PTOFamps and PTINTall. The cut was initially set by eye to retain the majority of events lying in a band of consistent PTOFamps and PTINTall. The lower passage rates for PuBe and Na data are attributed to the higher event (and therefore pileup) rates in those datasets.

Relative Efficiency	1 keV	30 keV
Bkg
Na
PuBe

Given the arguments above for the linear shape of these efficiencies, we can actually make an estimate of the relative event rates in these datasets. If we extrapolate to E=0 keV, the probability of being cut if there is pileup (and the cut is working) is 100%. Then, the efficiency at E=0 is just 100% minus the probability of a pileup event occurring. From the plots below, these probabilities are roughly 0.2, 0.65, and 0.4 for Bkg, Na, and PuBe respectively. The events are Poisson distributed, so the fraction of traces with any event which have pileup is

where r=RT, the event rate, R, times the trace length(?), T. We can then find that the corresponding rates for the datasets are r = 0.4, 1.8, 0.9 for Bkg, Na, PuBe respectively. This allows us to make an estimate that the relative event rates of the datasets are the ratio of r's. i.e. the Na dataset will have a rate ~2× that of the PuBe and ~4.5× that of the Bkg. It would be interesting to compare this to the rate studies being performed by Derek.

The PTOFchisq cut removes events which are inconsistent with the template in frequency space. This can occur for pileup, saturated events, or noise. This cut shows high efficiency for most energies, indicating that the previous cuts successfully removed the majority of pileup events. Near the noise wall, efficiency of this cut start to fall, presumably as it removes noise traces which should have a different spectral shape from events.

Relative Efficiency	1 keV	30 keV
Bkg
Na
PuBe

The spike cut (which includes the glitch cut) removes known digitization glitches and an unexplained class of spikey pulse trains observed during a particular time period of the PuBe data. The cut also removes a significant amount of the noise wall, so the cut is applied to all datasets for consistency. Events in the noise wall are the bulk of those removed by this cut.

Relative Efficiency	1 keV	30 keV
Bkg
Na
PuBe

The number and total fraction of events surviving after each cut are shown in the table below. As noted above, these counts represent only ~10% of the total -125 V dataset.

TODO: It looks like most of the events we're cutting are pileup and pulse tails. Try to calculate the expected rate of those things and compare it to what is being cut.

The efficiency as a function of energy is shown for each cut. The passage fractions in the table above may sometimes seem in conflict with the efficiency curves. This can occur because there are a significant number of events which reconstruct to an energy outside of the ranges shown below (i.e. <0 keV). Such events are counted in the table above, but will not appear in spectra of physical energies. The efficiency curves below are all compare to the data after the first 3 cuts (ending in ncBShi) are applied. This is because there are arguments for that first set of cuts being independent of energy.

Efficiency	1 keV	30 keV
Bkg
Na
PuBe

The effect of these cuts on the various spectra can be seen below. Again, much of the perceived energy dependence of the first few cuts is simply due to incorrectly reconstructed event energies for problem events, such as retriggers, which are removed by those cuts. This is also a confounding factor for the other cuts. The whole reason for removing these events is that we suspect their energy reconstructions are incorrect. Without a population of known events which we can also pass through these cuts, it is difficult to infer what effect these cuts will have on the true energy spectrum.

Spectra	1 keV	30 keV
Bkg
Na
PuBe

Here is a closer look at the final spectra in the ROI using these cuts. The energy quantity is PTOFamps calibrated to the 14, 18keV lines and the spectra have not been normalized to livetime. The background spectrum appears very flat as is expected for a compton continuum at these energies. The Na spectrum is also flat, except for a small excess below 100eVee. It is not known what this might be. The nuclear recoil scattering of the 22Na 1.3MeV gamma should come in around 12 eVee. The now familiar n-on spectrum is observed with a hump around 150eVee (PuBe 4.4 MeV gamma-nuclear scatters?) and a spike at ~75eVee.