A measurement of absolute efficiency of the ARAPUCA photon detector in Liquid Argon
Dante Totani, Gustavo Cancelo, Flavio Cavanna, Carlos O. Escobar, Ernesto Kemp, Franciole Marinho, Laura Paulucci, Dung D. Phan, Stuart Mufson, Chris Macias, David Warner
AA measurement of absolute efficiency of theARAPUCA photon detector in Liquid Argon
Dante Totani a , Gustavo Cancelo b , Flavio Cavanna b , Carlos O. Escobar b , ErnestoKemp c , Franciole Marinho d , Laura Paulucci e , Dung D. Phan f , Stuart Mufson g ,Chris Macias g , and David Warner ha Università degli Studi dell’Aquila, L’Aquila, 67100, Italia b Fermi National Accelerator Laboratory, Batavia, IL 60510, USA c Universidade Estadual de Campinas, Campinas - SP, 13083-970, Brazil d Universidade Federal de São Carlos, Araras - SP, 13604-900, Brazil e Universidade Federal do ABC, Santo André - SP, 09210-580, Brazil f University of Texas at Austin, Austin, TX 78712, USA g Indiana University, Bloomington, IN 47405, USA h Colorado State University, Fort Collins, CO 80523, USAAugust 13, 2020
Abstract
In the Fall of 2017, two photon detector designs for the Deep Underground NeutrinoExperiment (DUNE) Far Detector were installed and tested in the TallBo liquid argon(LAr) cryostat at the Proton Assembly (PAB) facility, Fermilab. The designs includetwo light bars developed at Indiana University and a photon detector based on theARAPUCA light trap engineered by Colorado State University and Fermilab. Theperformance of these devices is determined by analyzing 8 weeks of cosmic ray data.The current paper focuses solely on the ARAPUCA device as the performance of thelight bars will be reported separately. The paper briefly describes the ARAPUCAconcept, the TallBo setup, and focuses on data analysis and results.
The efficiency of photon detectors is of paramount importance for large volume LAr exper-iments. The detection of scintillation light generated as charged particles traverse a largeliquid argon time-projection chamber (LArTPC) adds valuable information to the study ofweakly-interacting particles. Most importantly, the leading edge of the scintillation lightpulse yields sub-mm precision in reconstructing the absolute position of the event in thedrift direction [1]. In addition, the scintillation light can provide the trigger for baryonnumber violation events, such as proton decay and neutron-antineutron oscillations, and su-pernova neutrinos, as well as improve rejection of uncorrelated cosmic backgrounds. Giventhe enormous volume of future experiments such as DUNE (Deep Underground NeutrinoExperiment) [2], the photon detectors must cover a large surface area in a cost-effectivemanner. Several technologies and implementations have been proposed and tested. A rel-atively novel technology is based on a light trapper device designed by A. Machado andE. Segretto named ARAPUCA [3]. ARAPUCAs are able to increase the effective photoncollection area while keeping the sensor area small. The latter property allows a relativelyhigh photon collection efficiency at a reasonable cost. The following paper reports on the1 a r X i v : . [ phy s i c s . i n s - d e t ] A ug ichroic filter4 6x6 mm^2 SiPM (a) ARAPUCA mechanical design (b) ARAPUCA conceptual design Figure 1: ARAPUCA light trapper device.design and evaluation of a photon detector unit composed of eight ARAPUCA cells. Thedetector was placed along a photon detector plane that also included two wavelength shift-ing light guides from the photon detector group at Indiana University [4]. The combinedphoton detector set operated for eight weeks in the TallBo LAr cryostat at the PAB facility,Fermilab.
The nm scintillation light from interactions of charged particles in LAr is not detectabledirectly by affordable sensors such as the traditional photomultipliers (PMTs) or the newersilicon photomultipliers (SIPMs) without appropriate wavelength shifter coating. Amongavailable photon detectors, SIPMs are gaining popularity due to its superior quantum effi-ciency (QE), desired physical and electrical properties as well as a low cost of production.However, due to their small size, their effective light collection area per unit is easily sur-passed by PMTs. The ARAPUCA concept [3] brings up a solution to this problem. Asshown in Figure 1b the ARAPUCA consists of two wavelength shifters, a dichroic filter, ahighly reflective box and photosensors (SIPMs). The face of the box is used to augment thephoton collection area. The VUV photons that enter the light collecting surface of an ARA-PUCA device are wavelength-shifted by µg/cm of p-terphenyl deposited on the externalface of a dichroic filter. The dichroic filter allows the converted photons to go through and (a) Dichroic filter transmission andp-terphenyl emission spectra (b) Dichroic filter reflection andTPB emission spectra Figure 2: Double coated dichroic filter and wavelength shifter spectra2nter the box. Those photons are wavelength shifted a second time using µg/cm oftetra-phenyl butadiene (TPB) deposited on the internal face of the dichroic filter. Since thewavelength cutoff of the dichroic filter is in between the emission spectrum of p-terphenyland the emission spectrum of TPB, the twice shifted photons remain trapped inside thereflective box and bounce off the walls until they hit the sensors. Figure 1a shows one of themechanical designs of the ARAPUCA. Several iterations of ARAPUCAs have been studiedwith variable box sizes, number of SIPMs and location of the SIPMs. In one of the experi-ments with previous versions of the ARAPUCAs, also performed at TallBo during the Spring2017 with a radioactive source to excite the LAr, we measured an efficiency ∼ . [5], con-siderably lower than the values reported with the new configuration used in this work (seesection 3.6.2). The Fall 2017 TallBo experiment used 8 ARAPUCA units of . cm × . cm with 4 SensL mm × mm SIPM biased at . V . Figure 2a and 2b show that the emissionspectra of p-terphenyl and TPB do not overlap; and the wavelength cutoff of the dichroicfilter is located between the two spectra. The peak of the p-terphenyl spectrum is nm ,the peak of the TPB spectrum is nm , and the cutoff of the filter is nm .Figure 3: 3D view of the TallBo LAr cryostat. The vessel’s external diameter is . cm and the inner diameter is . cm . The fullcapacity of TallBo is l . For this experiment TallBo was filled with approximately l of liquid argon. The total thickness of stainless steel presented to a cosmic ray particle thatcrosses the cryostat is . cm . The liquid Argon purity in TallBo was monitored usingcommercial gas analyzers. During the period of the data acquisition the levels of N , H O ,and O were well below ppm ( ppm = part per milion). Tipical values were: N ∼ ppb , H O ∼ − ppb and ∼ ppb ( ppb = part per bilion).The experiment’s trigger used a set of two scintillation paddles and a tracking mechanismbased on scintillation paddles and hodoscopes. The hodoscopes were used before in theCREST baloon flight experiment [6, 7]. The hodoscope modules were installed on oppositesides of the TallBo cryostat to select single-track cosmic-ray muons passing through the3 a) Top view.(b) Front view. Figure 4: Cross-sectional views of the TallBo LAr cryostat.4igure 5: Detector plane configuration.LAr volume [7]. Figure 3 shows a 3D view of the photon detector plane inside the cryostatand the two hodoscope blocks (in white), one of which was moved from a high position toa low position to trigger on high angle and low angle minimum ionizing particles (MIPs)respectively. Each hodoscope module consists of 64 2-inch diameter barium-fluoride crystals,coated with TPB. Each crystal is monitored by a 2-inch PMT. The crystals are arranged as × matrix. Since the hodoscope matrix elements are very sensitive to extraneous photonactivity and have a high dark count rate, to remove extraneous events two scintillatorpanels covering the entire hodoscope face were placed between each hodoscope module andthe TallBo dwear. These panels are individually read out by PMTs. The readout systemwas then triggered by four-fold coincidence logic that required at least one hit in bothhodoscope modules as well as one hit in their adjacent scintillator planes in a coincidencewindow of 150 ns. Events were further filtered offline by requiring one and only one hitin each hodoscope module to reject showers. Together triggering on cosmic rays rejectingshowers the requirement of having a single crystal fired per hodosocpe module gave thegeometric position of the tracks. Figure 4a and 4b show, from the top and from the frontof the cryostat, how a track is identified and located by the hodoscope. As shown in Figure 5, the 8 ARAPUCAs are divided into two frames of 4 ARAPUCAseach and occupy one third of the photon detector plane. They were located in a sideposition with respect to the vertical central axis plane. Each ARAPUCA has dimensions of cm × . cm × cm . The filter windows are cm × cm . Two sets of four contiguousARAPUCAs are placed as shown separated by a . cm gap. The values in Figure 5 aregiven in inches. 5 .3 Data Acquisition System Signals are processed by the SiPM Signal Processor (SSP) module designed by the Elec-tronics Group of the High Energy Physics division at Argonne National Laboratory [8]. TheSSP is a bit , M S/s , 12-channel waveform digitizer DAQ. The ADC has a full-scaledynamic range of V and a preamplifier gain of . V /V . For typical SiPM gains andganging configurations the SSP allows large signals equivalent ∼ PEs before the ADCsaturates. The SSP input noise is ∼ µV . Each acquired waveform contains 1950 ADCvalues sampled at M S/s , aggregating to an acquisition time of µs Each sample bin, . ns long, is called tick . The SSP can trigger internally on each individual channel orexternally via external trigger input. The self-trigger mode was used to take calibrationdata while the external trigger mode with the hodoscope as the trigger source was employedfor the cosmic ray run. The hodoscope trigger malfunctioned during the experiment. A mistake in the way thecoincidence logic was set up caused the trigger to fire more often than expected. The logicperformed an OR instead of an AND in two of the channels. Therefore, the trigger notonly fired on valid tracks but also on coincidental PMT dark counts of the hodoscope arrayslocated at the left and right of the dewar during the defined trigger window ( ns ). Thecollected data set contained a large amount of unusable data that did not represent validtracks. As shown in the next sections, most of the events were triggered by coincidenceof PMT dark counts and contained no signals from the photon detectors, neither in theARAPUCAs nor the IU light guides. Other spurious events contained partial tracks that didnot generate a trigger but were captured by chance during a dark count coincidence event.The hodoscope trigger problem made the analysis of the data challenging, since offline filtershad to be devised and applied to the data. However, it also showed the potential of theARAPUCA architecture given that those filters could have not been applied without thesegmentation property of the ARAPUCAs, as will become clear in the next sections of thepaper. During calibration runs the ARAPUCA signals were acquired by the SSP single channelself trigger mode, using an amplitude threshold slightly above 1 PE. Since the thresholdwas higher than the single photon electron the histogram is able to exhibit the nd peakand higher (Figure 6a blue histogram). The first PE was recovered looking at dark countsin a different integration window (Figure 6a green histogram), far from the trigger point.The first integration window, containing the self trigger point, was in the first half of thewaveform (from 100 to 900 ticks ) and the second window, containing possible dark counts,in the second half of the waveform (from 900 to 1700 ticks ). Figure 6b shows the linearityof the calibration. The linear fit in Figure 6b is obtained using the means of the Gaussiandistribution peaks of Figure 6a. The integrated charge value ( ADC · T icks ) due to one PEcan be obtained by the slope of the straight line.A second procedure was developed to cross check values using the calibration constantsobtained from the calibration run. A smaller set of data triggered with the hodoscope wasused. Baseline and noise levels were obtained as the mean and root mean square (RMS)for the first 100 points of the signal amplitude for each channel per event. After baselinesubtraction a filtering algorithm was used that suppresses random noise but keeps the main6 a) Self triggered and dark count data spec-tra. In red is shown the multi gaussian fit. (b) Linear fit for the peak positions. Statis-tical error bars are smaller than symbols.(c) Original and denoised waveformcomparison. (d) Hodoscope triggered data spectrumand multi gaussian fit.
Figure 6: Calibration procedures for ARAPUCA 1. The slope of linear fit (b) gives thecalibration in ( ADC · T icks ) /P E .features of the signal, in particular the fast transitions [9]. A peak finding algorithm wasthen devised to find peaks and integrate the charge under a peak. Figure 6c shows anexample of a peak found. The red line shows the raw data and the black line indicates thefiltered signal. The threshold for detecting a pulse has been set to σ of the baseline noise(e.g. 8.2 ADC). Figure 6d shows the spectrum obtained with peaks from ARAPUCA 1(the same used to show the self trigger events analysis). Notice that the pedestal is almostsuppressed by the denoising procedure and threshold requirement. A multi Gaussian fitwas performed to obtain the charge value for each peak. A linear fit of the peaks as afunction of the photon electron number provided the average calibration constants whichare in agreement with the self trigger events analysis. As mentioned in subsection 2.4 most of the data recorded by the experiment was due to thebackground, since the majority of the triggers were generated by coincidental dark countsin the hodoscope PMTs. Nevertheless, the system has also properly triggered on real tracks(i.e. MIPs), although these events only constituted a small fraction of the stored data, ofthe order of to .Since the dark counts are uncorrelated with light in the cryostat most events of that type7igure 7: Log(PE) spectra for events triggered as "front" (blue) and "back" (red). Thesame number of events for both spectra are used, this normalization is reflected in havingthe same amount of empty events (first peak in both spectra). The excess of events in thesecond peak of the red spectrum respect the blue one is interpreted as back tracks associatedto the right trigger information. In the same way the excess of events in the third peak ofthe blue spectrum respect the red one is interpreted as front tracks associated to the righttrigger information. All common parts of the spectra are interpreted as background: emptyevents (first peak) and spurious hodoscope coincidences.were empty or captured incidental background light. Most of these events collected few PEsfrom radiogenics or very low light background in the TallBo cryostat. A fraction of thoseevents also captured an incidental track that fell within the ns trigger window. Mostof those tracks are partial tracks that only passed through one hodoscope arm or none atall. In either case the track position information given by the hodoscope is incorrect, sinceit corresponds to the hot PMTs which generated the trigger and not to the passage of thetrack.The tested version of ARAPUCAs were single sided detectors having filters on only oneside. Therefore only tracks passing in front of the ARAPUCA plane could be seen. In thisparticular occasion, due to the high trigger rate, the comparison between the spectra ofevents passing in front of the single sided ARAPUCAs and behind them was useful to doevent discrimination. It was possible to get information comparing the spectra of eventstriggered as tracks located entirely in front of the plane containing the windows and behindit. That allowed tracks uncorrelated with trigger position information to be filtered out.As a first step we considered the spectrum of the number of PEs seen by all ARAPUCAsin the detector, for a given event ( P E tot ). We obtained individual spectra for two groups ofevents. Group 1 (blue spectrum of Figures 7) has all events for which the trigger geometrygiven by the hodoscope labeled a track as passing in front of the ARAPUCA window plane(front events) and Group 2 (red spectrum of Figures 7) with all events passing behind theARAPUCA window plane (back events). Events that have an entry point in the front andthe other one in the back or vice-versa were discarded, like the example in Figures 4a. The8ifference between the back and front spectra is the only feature which can be associated tothe information coming from the hodoscopes. To enhance the peak display in the spectra,instead to use the PE number, it has been chosen to show the common logarithm of the PEnumber in plots.The spectra, reported in Figures 7, show three peaks: • the first, the largest one, is composed of events for which two crystals fired but negligi-ble light is seen by any ARAPUCA, indicating these are completely random triggers,due to the high hodoscope rates. The fact that both back and front spectra show thefirst peak identically confirms its origin. The same number of events for both spectraare used. • the excess of events in the second peak of the red spectrum respect the blue one isinterpreted as back tracks associated with a MIP triggering the crystals for a tracklocated entirely behind the ARAPUCAs plan (back track). • in an equivalent way, the excess of events in the third peak of the blue spectrumrespect the red one is interpreted as front tracks associated with a MIP triggering thecrystals for a track located entirely in front of the ARAPUCAs plan (front track).All common parts of the spectra are interpreted as background: empty events (first peak)and spurious hodoscope coincidences.Similar considerations can be made for each single ARAPUCA. Applying cuts on the min-imum and maximum number of PEs collected by each ARAPUCA, it is possible to get aselected dataset of events. Single ARAPUCA cuts, shown as black lines in 9, consist inremoving events for which the back spectrum (red in 9) has more events respect to the frontspectrum (blue in 9). The front and back spectra show a different behavior for each ofthe eight ARAPUCAs. The reason of that is the geometrical effect due to the position ofthe ARAPUCA cells respect the tracks selected by the hodoscope. For this reason the cutneeded is different for each ARAPUCA, black lines in Figure 9.Looking at the sum of the detected PE over the 8 ARAPUCA for the selected events, redspectrum in Figure 8b, they are compatible with the excess of events in the third peak of theblue spectrum in Figure 7. Events remained, black spectrum in Figure 8a, are compatiblewith the common parts of the spectra in Figure 8b, interpreted as background. (a) Background. (b) Selected events. Figure 8: Background and selected events obtained through cuts applied on the number ofPEs collected by each ARAPUCA, got comparing front and back spectra.9igure 9: Front (blue) and back (red) Log(PE) spectra for the eight ARAPUCAs, and thecut applied (black). The cut consists in removing events for which the back spectrum (red)has more events respect to the front spectrum (blue).10 .3 ARAPUCAs multi-channel information
Each ARAPUCA is read out by an independent SSP channel, the segmented information canbe used to study the track geometry. The number of PEs measured by each ARAPUCA mustmatch the profile of the expected number of photons hitting the ARAPUCA array given atrack geometry, with the track geometry provided by the hodoscope. The ratio between thenumber of PEs collected and the number of photons hitting each ARAPUCA depends onlyon their intrinsic features (geometrical dimensions, number of SiPM, wavelength shifter,dichroic filter). It should be constant and independent of the track geometry and thenumber of photons generated by a track in the LAr. Inconsistency of these ratio means weare observing a track not compatible with the geometry provided by the hodoscope. In thatexperiment the 8 cells of which the ARAPUCA array is composed, have the same featuresso we expect them to behave in the same way. The determination of the number of photonshitting each ARAPUCA is shown in the next section (3.4).Figure 10: Track angular acceptance.
A charged particle crossing the volume of liquid argon will generate an amount of scintillationphotons per unit of track length given by: dN γ dl = (cid:28) dEdx (cid:29) ρY γ q p , (1)where (cid:10) dEdx (cid:11) is the specific deposited energy ( M eV cm /g ), ρ is the density of the liquidin which the particle travels ( g/cm ), Y γ is the number of photons emitted per unit de-posited energy in the medium ( Y γ = 5 . × γ/M eV ), and q p = 0 . is the quenchingfactor [10, 11]. A crossing muon was assumed to be a MIP with a photon yield in liquidargon Y γ q p = 4 × photons/M eV and an energy deposit (cid:10) dEdx (cid:11) ρ = 2 . M eV /cm .The number of photons that arrive at the detector window due to a muon passing the argonvolume is estimated as the product of the track integrated angular acceptance, A Ω , and thenumber of emitted photons per unit length. This acceptance is defined numerically as: A Ω = d N (cid:88) i =1 Ω i , (2)11 a) 3 (b)(c) (d) Figure 11: Comparison between the amount of light expected to arrive in each ARAPUCA(red), the number of PEs extracted from the waveforms (black) and the ratio between them(blue). Figures (a) and (b) show events following the pattern expected, Figures (c) and(d) show events not correlated with the trigger information. Blue points show the ratio ofmeasured to expected number of photons. Statistical error bars are smaller than symbols.where d is the length of a small (i.e. differential) segment of the muon trajectory and Ω i is the solid angle of the pyramid with its apex at the center of that segment and its basegiven by the sensor window. Ω i is calculated using a set of formulas given in reference [12]. Figure 10 illustrates tracksegments and their correspondent Ω i .The total number of photons arriving (PH) at the ARAPUCA window is obtained as: P H = 14 π A Ω dN γ dl . (3)The expected light calculated by the formulas and the number of PEs measured were com-pared, as displayed in Figure 11 for a few example tracks. There is a reduced set of eventsfor which the expected and detected light follow the same shape pattern. It is expected thatthe absolute efficiency per ARAPUCA is a constant number smaller than 1. As mentionedpreviously, the efficiency is defined as the ratio between the number of PEs collected bythe ARAPUCA and the number of photons arriving at the ARAPUCA window (PH) fora valid track event. Figure 11 demonstrates, for a valid track Figure 11a, the remarkablediscrimination power of the ARAPUCA showing how the distribution of the collected lightclosely follows that expected for the illumination light, scaled down by a constant smallerthan one that is directly related to the efficiency of the device. Also, for a different validtrack such as in Figure 11b, that constant must remain the same within errors. Further-more, Figure 11a and Figure 11b show that the ratios are similar for the 8 ARAPUCAs12sed in the experiment. Figure 11c and Figure 11d also show that the same discriminationpower allows filtering events where the shape of the distribution of the light collected bythe ARAPUCAs do not follow that of the expected illumination. These tracks are due toMIPs that partially illuminated the ARAPUCAs and were mistakenly catalogued by thehodoscope with the wrong geometry due to hot PMTs and coincidental PMT events.The criteria to separate valid from invalid tracks followed a χ requirement as explained insection 3.6.1. - - C oun t s Front EventsTotalSelected
Figure 12: Correlation between
P E i and P H i for the eight ARAPUCAs. Blue is for all theevents, red for the events with the PE numbers, recorded by each ARAPUCA, in the rangedetermined from the back and front spectra analysis.After the selection based on front versus back spectrum discrimination, made in sec-tion 3.2, the events left are 14,811. The cut applied is crude and the data set is stillcontaminated with background. A second background rejection can be made utilizing thesegmentation of the detector given by the 8 ARAPUCA cells shown in section 3.4.Comparing the amount of expected light coming from the analytical formula and the num-ber of PEs from waveforms, leads us to the determination of good tracks. In Figure 11 thenumber of PEs measured (black), the number of landing photons (PH, in red), and theirratio (blue) are displayed for each ARAPUCA, for a few tracks. There is a reduced set ofevents for which the expected and detected light follow the same pattern Figure 11a andFigure 11b. As an estimator for light pattern information we used the correlation betweenthe measured PE number and the calculated number of photons landing on the ARAPUCAfor each ARAPUCA in the event. The measure of correlation used is given by the Pearsoncorrelation coefficient defined as C = (cid:80) ( P E i · P H i ) − (cid:80) P E i · (cid:80) P H i / (cid:113)(cid:80) ( P E i ) − ( (cid:80) P E i ) / · (cid:113)(cid:80) ( P H i ) − ( (cid:80) P H i ) / (4)where P E i is the number of collected photo electrons in the i − th ARAPUCA and
P H i isthe number of expected photons arriving on the i − th ARAPUCA surface. The i index gofrom 1 to 8.Figure 12 shows the correlation parameter for the total number of events (blue) and for theevents passing the cut on the spectra (red). Most of the events passing the cut are peakedaround C = 1 indicating a good correlation between the eight ARAPUCAs. However thereare some events with C ≤ . A second requirement on the correlation parameter is done,selecting events with C ≥ . . After that the total number of selected events is 13005. Thisnumber is the total for 366 hours of running time at a rate of . Hz .13he number of events selected is in a very good agreement with the rate expected fromcosmic ray muon flux through the hodoscopes, obtained analytically, calculating the cosmicray flux [14] through the hodoscope geometry getting . Hz and doing a simulation ofthe muons crossing the hodoscope: . Hz .Figure 14a shows the common logarithm of the ratio between the sum of the photo electronsmeasured and the sum of the estimated number of arriving photons on each ARAPUCA,comparing all the data (blue spectra) and the selected dataset (red spectra) using thecombination of both spectra selections and correlation requirement. The selected data set is then used to perform the efficiency analysis. The efficiency is definedas the ratio between the number of measured photons and the estimated number of arrivingphotons for each ARAPUCA: R i = P E i P H i (5)Figure 13 shows the ratio distribution for the eight channels. Furthermore, the ratio betweenthe sum of the number of photoelectrons collected by all ARAPUCAs divided by the sumof the number of expected photons landing on all ARAPUCAs is considered. R T OT = (cid:80) i =1 P E i (cid:80) i =1 P H i (6)In Figure 14a is reported Log ( R T OT ) (for better visualization) for all the data (blue) andthe data selected through spectra analysis and correlation cut (red), called " Dataset • log-normal fit • robust statistic • bootstrap procedure on a reduced datasetThe frist two methods were used on the data selected " Dataset
Dataset % ) Median ( % ) Median ( % ) MAD ( % )Fit result Robust stat.TOT 0.78 ± ± ± ± ± ± ± ± ± Dataset .6.1 χ requirement on light patterns (a) Original data set in blue and after thespectra selection and correlation cut in red. (b) Original data set in blue and after a χ se-lection cut in black using Σ < . of Eq. (7).(c) Original data set in blue, after the spectraselection and correlation cut in red, and usinga further restriction in green. (d) Original data set in blue and after a χ selection cut of Eq. (7), black using Σ < . and purple using Σ < . . Figure 14: Comparison between the two kinds of cut applied to the original data set (bluespectra) of the ratio between the photo electrons collected sum and the estimated number ofarriving photons sum. In Figure (a) and (c) are reported the selected datasets using the cutbased on the spectra analysis and correlation condition, in figure (b) and (d) are reportedthe selected datasets using the χ approach, described in section 3.6 Eq. (7).A second criteria of data selection based on a χ requirement on the ratio R i , was appliedto the original dataset before any cuts (all the events), in order to check the goodness ofthe spectra selection and correlation requirements. Σ = 18 (cid:88) i =1 (cid:18) R i − R T OT R i + R T OT (cid:19) (7)Requiring a condition on Σ we can select a certain amount of events. A smaller Σ means abetter match between collected and expected photons. The conditions on Σ were determinedin order to get the same amount of data of the other cuts, and the same three analysis wereperformed: robust statistic and log normal fit for a dataset using Σ < . (" Dataset Σ < . (" Dataset
The data sets got through the two selection methods were tested incresing the requirementson the cuts, in the first case using the condition shown in Appendix A.3, in the second case16RAPUCA Hi-Low Low-Low Hi-Low Low-Low
Dataset Dataset Dataset Dataset ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± R T OT (Eq. 6) and R i (Eq. 5) for all the cells. Valuesfor the two hodosocope configurations (Hi-Low ad Low-Low) are reported, using the twoselection criteria: spectra analysis and cut on the χ (Eq. 7).using Σ < . from Eq. (7). These procedures reduced to ∼ both the selected datasets (shown in Figure 14). In both cases the mean values of each ARAPUCA ratio becamecompatible with the median values found.A similar analysis was made for the Low-Low hodoscope configuration using 3 MeV/cmof energy loss, reflecting the fact that the average muon energy increases for more horizontalevents [13]. Due to the geometry of the tracks only the lower four ARAPUCAs are takeninto account in the Low-Low configuration. The upper four have a small acceptance and theratio R i (Eq. 5) is affected by large fluctuations because the number of photons detected isdominated by Poisson statistics. Table 2 shows the results from the two selection criteriafor the two configurations. The ARAPUCA efficiency given by the median of the individualcells is R = (0 . ± . . The TallBo7 test did not provide a measurement of afterpulse and cross talk. The efficiency values reported here do not take into account theircontribution. However, an estimation of (31 ± of after pulse and cross talk is obtainedfrom ProtoDUNE [15] preliminary analysis, where the same kind of SiPM were used andrun in similar bias conditions. Adjusting by the detection efficiency of the ARAPUCAdetectors becomes (cid:15) = (0 . ± . . The Fall 2017 TallBo experiment has been successful in testing ARAPUCA photon detec-tors. Two of the features of the current ARAPUCA photon detector design were used tobe able to filter the data offline. The single face detection and the fine segmentation of theARAPUCAs allowed the offline analysis to separate events from background. The segmen-tation of the ARAPUCA detector validated the geometry given by the hodoscope. In theend an accurate absolute efficiency of 0.6% was determined by three different methods. Thatabsolute efficiency is higher than efficiency measurements reported by previous experimentsat TallBo using scintillation bars and wavelength shifters.It is also worth mentioning that the ARAPUCAs only used 4 SiPMs per board, the filter tosensor aspect ratio was 35 which shows a remarkable improvement in the equivalent photoncollection area given by the light-trap effect in the ARAPUCA.Future work will study improvements in the internal reflective surfaces of the ARAPUCA,wavelength shifter thickness and adherence which should help increasing the light collection.A step forward in the development of ARAPUCA is the active ganging of SiPMs to lower17he number of readout channels per module and the use of the two faces of the detector forphoton detection, having filters on both sides.
A Analysis metrics
A.1 Log-normal fit
A first analysis is made by fitting the ratio distribution with a Log-Normal distribution: f ( x ) = 1 xσ √ π e − (ln( x ) − µ )22 σ (8)This choice is driven by the Log ( R T OT ) which seems to follow a Gaussian distribution.The values of the Mean, Median, Mode, Variance and Standard Deviation are obtained bythe relations satisfied by the parameters of the Log-Normal distribution: • Mean: e (cid:16) µ + σ (cid:17) • Median: e µ • Mode: e ( µ − σ ) • Variance: (cid:16) e σ − (cid:17) e ( µ + σ ) • Standard Deviation: (cid:113)(cid:0) e σ − (cid:1) e (cid:16) µ + σ (cid:17) The error propagation, from the error in the fit, gives for the median the value: ∆ Median = e µ · ∆ mu (9) A.2 Robust statistic
Because of the presence of outlier events, such as the ones in the far tail of the spectrum,a robust statistic data analysis is made for the data relative to the single ARAPUCA ( R i ) and their sum ( R T OT ) , using median and the median absolute deviation (MAD) to build arobust score defined as: S i = ( R i − median j =1 ,...,n ( R j )) M AD (10)Then the median of the data set composed by the data which passes the requirement S i < . is calculated. A.3 Strong correlation
Finally a restricted data set, obtained with a further strong condition on the light pattern,is analyzed.The strong condition consists in requiring that all ARAPUCAs have the ratio R i similar toeach other: · R j ≤ R i ≤ · R j (11)This condition reduces the data set to be ∼ of the one used in the previous twoanalysis.A bootstrap procedure is used to find the average values of the ratio and their errors,generating 10 thousand data sets, each of them composed by random extraction of events,from the original data set. 18 Acknowledgements
The authors would like to thank A. Hahn, R. P. Davis, W. Miner, K. Harding for theirtechnical support at PAB. We also thank the whole Indiana University team with whomthis experiment was performed. We also thank Eileen Hahn for her invaluable knowledgeand support for the wavelength shifter coatings of filters and reflectors, and also KennethTreptow for his technical support in assembling the ARAPUCA’s module. This work hasbeen partially supported by the Brazilian agency FAPESP under grant no 2017/13942-5.Fermilab is Operated by Fermi Research Alliance, LLC under Contract No. De-AC02-07CH11359 with the United States Department of Energy.
References [1] B. Baller et al.,
Liquid argon time projection chamber research and development in theunited states , Journal of Instrumentation, 9(05):T05005–T05005, may 2014.[2] DUNE Collaboration, B. Abi et al.,
The DUNE Far Detector Interim Design ReportVolume 1: Physics, Technology and Strategies , arXiv e-prints, July 2018.[3] A. A. Machado and E. Segreto.
ARAPUCA a new device for liquid argon scintillationlight detection , JINST, 11(02):C02004, 2016.[4] B. Howard, S. Mufson, D. Whittington, B. Adams, B. Baugh, J.R. Jordan, J. Karty,C.T. Macias, and Pla-Dalmau.
A novel use of light guides and wavelength shiftingplates for the detection of scintillation photons in large liquid argon detectors , NuclearInstruments and Methods in Physics Research, 907:9–21, 2018.[5] G. Cancelo, F. Cavanna, C. O. Escobar, E. Kemp, A. A. Machado, A. Para, E. Segreto,D. Totani, and D. Warner.
Increasing the efficiency of photon collection in LArTPCs:the ARAPUCA light trap , JINST, 13(03):C03040, 2018.[6] S. Coutu et al.
Searching for TeV cosmic electrons with the CREST experiment , Nucl.Phys. Proc. Suppl., 215:250–254, 2011.[7] D. Whittington, S. Mufson, and B. Howard.
Scintillation Light from Cosmic-Ray Muonsin Liquid Argon , JINST, 11(05):P05016, 2016.[8] G. Drake P. DeLurgio, Z. Djurcic and M. Oberling.
SSP User Manual , Technical report,Argonne National Laboratory, 9700 S. Cass Avenue, Lemont, IL 60439 USA, 2016.[9] L. Condat.
A direct algorithm for 1-D total variation denoising , IEEE Signal Proc.Letters, 20(11):1054–1057, 2013.[10] T. Doke, H. J. Crawford, A. Hitachi, J. Kikuchi, P. J. Lindstrom, K. Masuda, E. Shiba-mura, and T. Takahashi.
Let Dependence of Scintillation Yields in Liquid Argon , Nucl.Instrum. Meth., A269:291–296, 1988.[11] T. Doke, A. Hitachi, J. Kikuchi, K. Masuda, H. Okada, and E. Shibamura.
AbsoluteScintillation Yields in Liquid Argon and Xenon for Various Particles , Japanese Journalof Applied Physics, 41:1538, March 2002.[12] R. J. Mathar.
Solid angle of a rectangular plate , Technical report, Max-Planck Instituteof Astronomy, Königstuhl, 17, 69117 Heidelberg, Germany, 2015.[13] M. Tanabashi et al. (Particle Data Group).
The review of particle physics , Phys. Rev.D, 98:030001, 2018. 1914] Prashant Shukla.