Experiment to Demonstrate Separation of Cherenkov and Scintillation Signals

J. Caravaca,1, 2 F.B. Descamps,1, 2 B.J. Land,1, 2 J. Wallig,2 M. Yeh,3 and G.D. Orebi Gann1, 2

1University of California, Berkeley, CA 94720-7300, USA2Lawrence Berkeley National Laboratory, CA 94720-8153, USA3Brookhaven National Laboratory, Upton, NY 11973-500, USA

The ability to separately identify the Cherenkov and scintillation light components produced inscintillating mediums holds the potential for a major breakthrough in neutrino detection technology,allowing development of a large, low-threshold, directional detector with a broad physics program.The CHESS (CHErenkov / Scintillation Separation) experiment employs an innovative detectordesign with an array of small, fast photomultiplier tubes and state-of-the-art electronics to demon-strate the reconstruction of a Cherenkov ring in a scintillating medium based on photon hit timeand detected photoelectron density. This paper describes the physical properties and calibration ofCHESS along with first results. The ability to reconstruct Cherenkov rings is demonstrated in awater target, and a time precision of 338±12 ps FWHM is achieved. Monte Carlo based predictionsfor the ring imaging sensitivity with a liquid scintillator target predict an efficiency for identifyingCherenkov hits of 94 ± 1% and 81 ± 1% in pure linear alkyl benzene (LAB) and LAB loaded with2 g/L of PPO, respectively, with a scintillation contamination of 12 ± 1% and 26 ± 1%.

I. INTRODUCTION

Optical photon detection has been a commondetection mechanism in neutrino experiments formany decades [1–6] and the technology is well de-veloped. Experiments have historically been opti-mized to detect one of two types of optical radiation:Cherenkov [7] or scintillation [8] light. DirectionalCherenkov light has been successfully used in bothhigh energy and nuclear physics by large ultra-purewater experiments such as SuperKamiokande [2] andSNO [3], while the more abundant, isotropic scintil-lation is more commonly used in low-energy detec-tors such as KamLAND [4] and Borexino [5]. A briefsummary of the advantages and limitations of eachtechnique can be found in Table I.

Cherenkov Scintillation

Directional Isotropic

Very well understoodStrong dependence on material

properties

Good shower/MIP separation Reasonable particle ID

Minimum energy threshold forlight production

No energy threshold for lightproduction

Low light yieldHigh light yield results in

lower detector threshold andimproved energy resolution

Occurs in all dielectricmaterials, some with very

good optical properties

Scintillating materials tend tohave substantially shorter

attenuation lengths

Cost-effective More expensive materials

TABLE I. Comparison of Cherenkov and scintillationlight in the context of optical particle detection.

The LSND experiment [6] used a mineral-oil-basedscintillator designed to allow detection of both scin-tillation and Cherenkov light, opening up sensitiv-ity to low-energy particles below Cherenkov thresh-old. The ratio of created photoelectrons for the twosources of light was approximately 5 to 1 (scintilla-tion to Cherenkov) for a 45 MeV electron created inthe detector. Photon timing and charge was used forparticle identification and to reconstruct vertex andangle information. Particles emitting Cherenkovradiation produced significant fractions of promptlight and thus particle identification techniques re-lied on the fraction of prompt light as a tool fordiscrimination.Separation of Cherenkov- and scintillation-lightcomponents on an event-by-event basis enables a sin-gle detector to exploit the advantages of each tech-nique. If this separation can be achieved in a pureliquid scintillator (LS) detector it would substan-tially improve the sensitivity of low-energy physicsprograms by adding directional reconstruction capa-bility to a low-threshold detector. Searches for neu-trinoless double beta decay (NLDBD) would benefitfrom rejection of the directional solar neutrino sig-nal, the dominant background in experiments suchas SNO+ [9]. Precision solar neutrino measurementswould benefit from separation of the directional sig-nal from isotropic radioactive background events.However, the high scintillation light yield and fasttiming of commonly used LS such as LAB (linearalkyl benzene) makes this separation extremely chal-lenging. In addition, liquid scintillator detectors arelimited to kiloton scales by the attenuation of lightin the scintillator material and also by cost.

arX

iv:1

610.

0202

9v3

[ph

ysic

s.in

s-de

t] 7

Oct

201

7

The recent development of Water-based LiquidScintillator (WbLS) [10] allows both the scintilla-tion light yield and timing profile of the target ma-terial to be tuned, thus increasing sensitivity to theCherenkov component. The admixture of water in-creases the attenuation length, thus improving lightcollection. For the first time one can thus envis-age a large-scale, low-threshold detector with direc-tional sensitivity. The concept for a large monolithicWbLS detector capable of a very broad program ofphysics, such as the Theia experiment, is presentedin [11] and [12]. The unique capabilities of a large-scale WbLS detector enable both a broad low-energyprogram, including a sensitive NLDBD search, solarneutrino studies, supernova neutrinos, and diffusesupernova neutrinos (DSNB), as well as sensitivityto nucleon decay and, if sited in a high-energy neu-trino beam, to long-baseline physics (neutrino masshierarchy and CP violation). The high-energy pro-gram benefits from the potential to image Cherenkovrings, thus improving particle ID, and the ability todetect particles below Cherenkov threshold, whichimproves background rejection.

The goals of the CHErenkov / Scintillation Sepa-ration (CHESS) experiment are several: to demon-strate Cherenkov / scintillation separation in a pureLS target by deploying ultra-fast-timing photosen-sors, and to study how the separation improves inWbLS and quantify the efficiency as a function ofthe scintillator fraction. These studies will allowoptimization of the Theia detector configuration(WbLS target cocktail, photosensor requirements,detector scale) in order to maximize the physicsreach across a broad program.

Separation of scintillation and Cherenkov compo-nents can occur in three domains:

• Wavelength: Cherenkov and scintillation lighthave different emission spectra (Fig. 1). TheCherenkov signal could be enhanced by select-ing photosensors with improved sensitivity inthe high wavelength region above the scintilla-tion emission cutoff.

• Photon density: depending on the ratio ofCherenkov to scintillation light yield, it maybe possible to distinguish the Cherenkov ringstructure on top of the background of isotropicscintillation light.

• Time separation: scintillation light, originat-ing from molecular de-excitation, is typicallydelayed from the fast Cherenkov light pulse by1 to 10s of ns [7, 8]. By combining fast photondetectors and electronics it should be possible

(nm)λ300 350 400 450 500 550 600

PMT

QE

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1PMT (H11934-200) QE

Acrylic Absorption Length

Cherenkov Emission

LAB/PPO(2g/L) Emission

LAB Emission

Absorption (m

m)

5000

10000

15000

20000

25000

30000

35000

40000

FIG. 1. PMT H11934 quantum efficiency [13] and UVTacrylic absorption length compared to the Cherenkovand scintillation emission spectra for pure LAB [14] andLAB loaded with 2 g/L PPO [15]. The normalization ofthe emission spectra are shown in arbitrary units.

to achieve some signal separation using photondetection times.

Each option has the potential to enhance signalseparation, but each comes with a cost, such as: re-duced light yield, reduced light collection efficiency,high-cost photosensors, or limitations on detectorsize. The challenge for time-based separation is thetiming precision of available photosensors. Largephotomultiplier Tubes (PMTs) typically have worsethan ns time precision, making Cherenkov light iden-tification in a scintillation medium extremely diffi-cult. However, newly developed small PMTs [13] canachieve timing precisions of . 300 ps and new micro-channel plate (MCP) photosensor technology [16–18]can achieve . 100 ps. In addition, fast and precisecommercial electronics are available to digitize thesesignals without significantly degrading the time res-olution. This makes separation in the time domaina theoretical possibility.

This paper describes the CHESS detector and itssensitivity to Cherenkov / scintillation separation inthe time domain. Section II describes the detectoritself. Section III describes the detailed Monte Carlosimulation used to model the detector. Section IVdescribes the general approach to data analysis. Sec-tion V describes detector calibrations. Section VIdescribes event selection and techniques for rejectionof instrumental and physics background events. Sec-tion IX describes the predicted sensitivity in a pureLS target: for both pure LAB and LAB loaded with2 g/L of PPO (hereafter referred to as LAB/PPO);and Section X concludes the paper.

2

II. THE CHESS DETECTOR

The primary goal of the CHESS experiment is todemonstrate Cherenkov / scintillation separation byCherenkov ring imaging in both charge (detectedphotoelectron density) and time, for different scin-tillating liquids. A schematic of CHESS is shownin Fig. 2. An acrylic target vessel is viewed by anarray of small, fast PMTs. The setup is designed todetect either cosmic muons or events from deployedradioactive sources. The primary ring imaging mea-surement is performed using through-going muons.Vertical-going events are selected via a 1-cm diame-ter coincidence tag, ensuring a population of eventswith known orientation and thus a known expecta-tion for the position of the Cherenkov ring. Themuons produce Cherenkov and scintillation light inthe target material, which is detected on the PMTarray. The apparatus has been optimized using afull Monte Carlo simulation (Sec. III), complete withoptical ray tracing, such that direct Cherenkov lightfrom vertical muons falls on a distinct set of PMTs,forming a clear ring in the PMT array. This yieldstwo distinct groups of PMTs by construction: thosewith pure scintillation hits and those with both scin-tillation and Cherenkov hits. The earliest hits oneach PMT can thus be identified as being causedby either Cherenkov or scintillation photons, anddemonstrate the time separation between these twosignals to high precision. A measurement of the clar-ity of the Cherenkov ring imaged in charge on topof the isotropic scintillation light background is alsopossible.

A. Target Vessel

The target vessel consists of a cylinder 5 cmin radius and 3 cm in height, constructed fromultraviolet-transmitting (UVT) acrylic. Three flatfaces on the outer surface of the cylinder, each 3 cmin diameter, provide surfaces for attaching a radioac-tive button source or optical coupling of a PMT asa source tag. A lid made of the same UVT acrylicencloses the target material in an air-tight environ-ment using an FFKM o-ring [19].

Distinct but similarly designed target vessels areused for water, pure LS and WbLS target materialsin order to minimize the risk of cross contamination.

( b)

FIG. 2. The CHESS apparatus. (a) Detailed schematicview with dimensions and (b) demonstration of ring-imaging concept. The PMT array is designed to holdup to 53 PMTs; the dozen slots occupied for this studyare color coded by radius: red and orange for those hitprimarily by scintillation photons, and blue for those inthe expected Cherenkov ring for LAB and LAB/PPO.Due to the lower refractive index, the ring from a watertarget is detected in the middle (orange) PMTs.

3

B. PMT Array

In the initial phase of CHESS, one dozen Hama-matsu H11934-200 PMTs [13] were deployed in across shape beneath the target, providing threeradial rings of four PMTs each for detection ofCherenkov and scintillation light. An additionaltwelve similar PMTs are available for deployment ina future upgrade. The H11934-200 PMT is a small1-inch cubic PMT with superb quantum efficiency(QE) peaked at 42% (Fig. 1) and excellent transittime spread (TTS) of 300 ps (FWHM for single pho-toelectrons).

The PMTs are held in an array that consists ofa 7x7 grid, with four additional slots at each com-pass point, as shown in Fig. 2. The holder is 3Dprinted from black ABS plastic. The initial deploy-ment positions for the twelve PMTs used in phaseI of CHESS are shown in Fig. 2. The unfilled loca-tions on the grid can be populated with additionalPMTs for future expansion.

C. Optical Propagation

The target vessel is optically coupled toa propagation medium made of UVT acrylic(30 cm×30 cm×6.5 cm). This coupling eliminates anacrylic/air boundary which would otherwise blockall Cherenkov light due to total internal reflection.The propagation medium acts as a light guide to al-low photons to propagate from the bottom of thetarget vessel towards the PMT array. The PMTs inthe array are similarly optically coupled to the bot-tom of the propagation medium. The propagationmedium and target vessel have been polished for abetter optical coupling and a more accurate simula-tion of refracted light at the boundaries.

A vertical cylindrical hole 1 cm in diameter ismachined through the center of the propagationmedium aligned with the center of the target ves-sel and cosmic tags in order to allow vertical-goingmuons to interact with the target material, but passthrough the propagation medium without producingadditional Cherenkov light in the acrylic block thatwould contaminate the measurement.

Throughout the setup, EJ-550 optical grease [20]is used for optical coupling of components.

D. Cosmic Muon Tags

Two custom-made cylindrical scintillator tags arepositioned above and below the target vessel (Fig. 2)in order to trigger on vertical cosmic muons. Analuminum arm maintains the alignment of the up-per cosmic muon tag while the bottom is fixed inthe PMT array (described in Sec. II B). Each tagconsists of a cylindrical 1-cm diameter HamamatsuPMT [21] optically coupled to EJ-200 plastic scin-tillator [22] shaped into a cylinder of 1-cm diame-ter and 5-cm height. The scintillator is coated withwhite paint to reflect light into the tag PMT, andthen coated with matte black paint (except for theend that is coupled to the PMT) to shield the re-mainder of the setup from light contamination. Thisassembly is held in a custom-built mount.

The small size of the tags results in a low angularacceptance (6 from vertical), ensuring a populationof events with known orientation and thus a knownexpectation for the position of the Cherenkov ring.Given the typical muon flux at the Earth surface(1 cm−2min−1) and the angular acceptance of thetags, a coincidence rate of ∼ 4 µ/day is predicted.

E. Veto Panels

The apparatus is surrounded by four scintillatorpanels (50 cm×100 cm×5.3 cm), as shown in Fig. 3,fabricated from EJ-200 plastic scintillator [22] (twoon the floor and two on the sides in a corner distri-bution) providing effective 4π coverage. Each panelis instrumented with a PMT [23] which is read outfor each event and used for an offline veto. The scin-tillator panels are used to veto cosmic events duringcalibration source deployment, and to reject cosmicshower events and coincidence of multiple cosmicmuons in the ring imaging analysis.

F. Shielding

CHESS is enclosed in a light-tight darkbox ofabout 1.5 m×1.5 m×1 m wrapped in FINEMET R©

sheets [24] for magnetic field isolation and paintedmatte black on the inside to minimize reflections.

G. Radioactive Source Deployment

A 90Sr button source is used to calibrate vari-ous aspects of the detector response (Sec. V). The

4

FIG. 3. Layout of veto panels around the CHESS appa-ratus.

source consists of a 1-inch diameter acrylic disk witha thickness of 0.125 inches ([25]). A 0.25-inch diam-eter cylindrical well in the center of the button con-tains a deposit of 0.1µCi of 90Sr combined with aresin. The well containing the radioactive materialis encapsulated with a 0.02-inch thick acrylic win-dow. This source is attached to one face of the targetvessel with the window facing the target material.

To trigger on source events a 1-inch cubic H11934-200 PMT [13] is optically coupled to another of theflat faces on the target vessel. This is the same typeof PMT used in the PMT array and was chosen forits high QE and low TTS.

H. The DAQ System

A schematic diagram of the data acquisition(DAQ) hardware is shown in Fig. 4 (high voltageomitted) and described in detail in the following sec-tions. The DAQ software [26] utilizes the CAENVME library [27] to configure and read out thedigitizers and high voltage supplies, and outputsHDF5 [29] formatted files containing raw digitizeddata and metadata for each triggered event.

1. Readout Electronics and High Voltage

Three six-channel high-voltage power supplies(CAEN V6533 [30]) power the PMTs. The PMToutput signals are connected to two CAEN digitiz-ers: a high-precision V1730 [31] digitizes the vetopanels and source tag PMT signals, while a fastV1742 [32] based on the DRS4 [33] chip digitizes the

PMT Array

Upper Cosmic Tag

Lower Cosmic Tag

SourceTag

FIG. 4. A schematic diagram of how the DAQ and hard-ware are connected together with signal and data pathslabeled. Omitted are the high voltage supplies, whichare connected to the VME backplane, and all PMTs.

PMT array and cosmic tag signals. The hardwareis housed in a VME crate, and a CAEN V1718 [28]VME to USB bridge is used for communications.

The V1742 card is capable of sub-100 ps resolu-tion, which exceeds the TTS of the PMT array andis therefore not a limiting factor in the time preci-sion. However, on-board buffer size limits the ac-quisition to a maximum of 1024 samples or 200 ns.This shallow buffer necessitated a low-latency trig-gering scheme in order to contain the pertinent datain the available event window. The V1730 digitizeris deadtimeless, however the V1742 introduces deadtime on the order of 100 µs. This is neither a limita-tion for measurements with cosmic particles, wherethe trigger rate is approximately 0.2 Hz, nor for mea-surements with radioactive source data where thetrigger rate is approximately 30 Hz.

2. Trigger

A LeCroy 606Zi [34] oscilloscope is used to pro-duce low-latency trigger signals for the setup withprogrammable coincidence logic. Depending on theoperating mode one of the following three trigger

5

configurations are used:

Bottom-only Trigger : A threshold condition is ap-plied to the lower cosmic tag only. This is thenormal operating mode for cosmic data, withthe coincidence requirement applied offline. Arate of 4 µ/day is expected after the coinci-dence trigger.

Or Trigger : A threshold condition is applied to bothof the cosmic tags, and the logical OR of thetwo is allowed to generate a trigger. This modeis used to acquire unbiased charge distribu-tions for each cosmic tag simultaneously.

Source Trigger : A threshold condition is applied tothe source tag. This is the configuration usedduring radioactive source deployment.

The oscilloscope trigger signal is fanned out to theexternal trigger input on the V1730 and the low-latency trigger inputs on the V1742. The V1742consists of four DRS4 analog sampling chips record-ing eight channels each. As these four chips operateon independent high-speed sampling clocks, the trig-ger signal is also digitized by each chip so that finetime offsets between channels on different chips canbe reliably calculated offline.

III. MONTE CARLO SIMULATION

The entire setup is simulated in the RAT-PAC suite [35]. Primary particles are produced(Sec. III A) and tracked in a complete geometryimplemented in GEANT4 [36]. When Cherenkovand scintillation processes take place (Sec. III B),individual photons are tracked from the productionpoint to the PMTs (Sec. III C). Once the photonreaches the PMT, a custom model decides whetherto produce a photoelectron (PE) based on the pho-ton properties and PMT features (Sec. III D). If a PEis produced, a custom DAQ simulation (Sec. III E)produces a pulse per PE and the total waveform isdigitized and stored for posterior analysis. Furtherdetails on each of the points are given in the follow-ing sections.

A. Cosmic Muon Generator

Muons and anti-muons are generated at 50% ratiofollowing the semi-empirical modified Gaisser distri-bution [37]. The hadronic component at surface is

expected to be sub-dominant by a factor of approx-imately 50, resulting in a predicted 3 events withina month’s worth of data, thus this component is notconsidered. Muons are constrained to pass throughthe bottom tag volume. There exists the possibilitythat high energy electrons produced by muon ioniza-tion trigger the bottom cosmic tag. This is repro-duced in simulation by including a complete modelof the holder material and geometry for both topand bottom tags.

B. Photon Production

Cherenkov production is simulated by GEANT4by the standard class G4Cerenkov. The typicalCherenkov emission spectrum is shown in Fig. 1.Cherenkov photons emitted with a wavelength belowa certain value are immediately reabsorbed by themedium and hence are not propagated in the simu-lation. Scintillation emission is handled by a mini-mally modified version of the GLG4Sim model [38].A charged particle passing through a medium de-posits an energy E due to ionization. The totalnumber of scintillation photons generated is not ex-pected to behave linearly with E due to quenchingeffects. This is taken into account by using Birk’slaw [8] which states that the deposited energy afterquenching, Eq, is:

dEqdx

=dE/dx

1 + kBdE/dx, (1)

where kB is Birk’s constant. The total number ofphotons produced, Nγ , is the direct product of Eqand the light yield of the material in question inphotons/MeV. Ex-situ measurements for the scintil-lation emission spectrum, light yield and time profilefor LAB and LAB/PPO are included in the simu-lation. Nγ photons are drawn randomly from theemission spectrum of the scintillator under investi-gation. The spectra for LAB [14] and LAB/PPO [15]are shown in Fig. 1.

The short term components of the timing profileρ(t) for the scintillation process is well described by adouble exponential model [39], including both a risetime and a decay time. Two further decaying expo-nentials are included for LAB/PPO, based on [40].The simulated time profile is:

ρ(t) ∝ (1− e−t/τr )×3∑i

Aie−t/τi , (2)

where τr is the rise time and the parameters τi andAi are the decay times and their scale factors.

6

C. Optical Propagation

Optical photons are propagated through differentmedia by GEANT4. Absorption, refraction and re-flection are taken into account according to the ma-terial’s optical properties. The absorption length ofthe UVT acrylic used in CHESS (Fig. 1) has beenmeasured with a spectrometer and is included di-rectly in the simulation. The refractive indexes forLAB and LAB/PPO are considered to be the sameand are taken from [15].Photon reemission from wavelength shifting mate-rials are simulated in a similar fashion as the scin-tillation process. Once a photon is absorbed in amedium, there is a finite probability that it will bereemitted following a specified reemission spectrum.

D. PMT Model

A full and precise simulation of the different PMTsis implemented in RAT-PAC. It contains a detailedPMT geometry (glass, photocathode, dynode stackand case) as well as a dedicated photon trackingsimulation inside the PMT volume. Once an op-tical photon hits the external boundary of the PMTglass, it is propagated through the different inter-nal PMT surfaces according to the relevant materialoptical properties. The QE of the PMTs is takenfrom Hamamatsu specifications [13] and used as aninput to the simulation to determine whether or notto create a PE for an incident photon at a partic-ular wavelength. An individual normalization canbe applied to the efficiency of each tube to allow fora finite collection efficiency. These are set to 90%for each tube [41]. If the incident photon createsa PE, its associated charge and time are extractedfrom Gaussian distributions that have been previ-ously calibrated to take into account the PMT gainand electronic delays (Sec. V).

1. Charge Distribution Model

The single-PE (SPE) charge distribution is mod-eled as a Gaussian truncated for negative chargevalues. The SPE charge distribution for individualPMTs has been measured (Sec. V) and Fig. 5 showsthat the Gaussian model agrees well with the data.

2. Time Profile Model

Each PMT has a characteristic transit time andTTS that depends on the PMT design. These num-bers are taken from the PMT specifications and in-cluded in the simulation assuming the time is Gaus-sian distributed. The PMT specifications [13] showthat the Gaussian model is a very good approxima-tion to the time profile. The mean is set to theprovided transit time and the width to the providedTTS. Extra time offset parameters allow for individ-ual delays per electronics channel (referring to thePMT, cable, and readout electronics) primarily dueto different cable lengths. These are measured on achannel-by-channel basis (Sec. V).

E. DAQ Simulation

For each event an analog waveform is generatedper channel by summing the individual pulses cre-ated by each PE (Sec. V B). The waveform is thendigitized via a process that mimics the character-istics of the two models of CAEN digitizers usedin the detector. High frequency electronics noise isadded to these digitized waveforms following a Gaus-sian distribution centered at zero and with a widthof 0.35 mV and 0.88 mV for the V1730 and V1742cards, respectively. These widths are measured fromdata by performing a Gaussian fit to the residuals ofpedestal-corrected noise data. In this way, any ef-fects introduced into the dataset by the digitizationprocess are reproduced in the simulation.

The trigger process implements the conditions de-scribed in Sec. II H 2 and decides whether to create atriggered detector event. When an event is created,acquisition windows corresponding to the buffer sizeof the appropriate digitizer are captured from thedigitized waveforms. These windows are used to de-termine photon hit times and charge using the sameprocess as applied to data in Sec. IV. The triggerprocess then scans the remainder of the digitizedwaveform to look for additional triggers within thesame simulated physics event.

IV. WAVEFORM ANALYSIS

The full waveform is analyzed to extract individ-ual PMT charges and hit times. An analysis windowof 135 ns (675 samples) is chosen starting 160 ns(800 samples) prior to the acquisition trigger. Thisis defined based on when light from the target is

7

expected to hit the PMT array. The charge in pico-coulombs is defined as the integral within that win-dow multiplied by a 50 Ω resistance, corrected bythe pedestal charge. The pedestal charge is calcu-lated on a trace-by-trace basis by taking an averageof sample values across the pedestal region, which isdefined by a 25 ns window right before the analysiswindow. If fluctuations of more than 5 mV peak-to-valley are detected in the pedestal region, the wave-form is not analyzed. The total integrated charge isconverted to an estimated number of detected PEsby normalizing by the mean value of the SPE chargedistribution measured in Sec. V.

The time of an individual PMT pulse is definedas the time at which the waveform crosses a thresh-old given by 25% of the height of a single PE pulse.This threshold is determined by modeling the PMTpulse as a log-normal pulse with parameters ex-tracted from the fit in Sec. V B and an integratedcharge equivalent to the mean of the SPE distribu-tion for each PMT (Sec. V A). This threshold is wellclear of the per-channel noise levels, as can be seenin Fig. 6.

Linear interpolation between ADC samples is usedto maximize the time precision. The measured hittime at the PMT array is corrected by the photontime of flight (ToF), assuming that the photons areproduced in the center of the target.

V. CALIBRATION

The setup is calibrated using a 90Sr beta sourcein combination with a water-filled target, an LEDdeployed in the dark box, and a control sample ofcosmic muons passing through the acrylic propaga-tion medium. For source data, the veto panels areused to reject coincident cosmic muon events. Betadecay, muon ionization and Cherenkov emission arewell understood processes that help to calibrate theoptical aspects of the detector. The following sec-tions detail the calibration techniques.

A. PMT Gain

PMT gains depend on the supplied voltage andvary tube-by-tube, so the gain is measured individ-ually for each PMT in the array. The PMT voltagesare kept constant between data-taking periods. Adirect measurement of the gain is given by the SPEcharge distributions. A dim source of light is pro-vided by the 90Sr beta source (Sec. II G) attached

to the water target. A small PMT optically cou-pled to the target is used to trigger the acquisition(Sec. II H 2). The SPE charge distribution for one ofthe array PMTs is shown in Fig. 5, where a clear SPEpeak is identifiable. There is a noticeable populationof multi PE events, which may be due to byprod-ucts of muon cosmic events crossing the propagationmedium. The muons themselves are easily vetoedusing the scintillator panels (Sec. II E), but high en-ergy electrons and gammas produced by these muonsare more difficult to veto and thus cause irreducibletails in the charge distribution.

A Gaussian fit is used to model the charge dis-tributions. To precisely extract the SPE parame-ters, the noise distribution and the multi-PE eventdistributions with up to 3 PEs are included and re-weighted by nuisance parameters. A Gaussian is as-sumed for the noise peak distribution with a mean,width and integral floated in the fit. For multi-PEevents only the event rates with respect to SPE arefloated in the fit, since the mean and width are de-termined by the SPE distribution. As a result, afit with 7 parameters (SPE mean and width, noisemean, width and integral, multi-PE event rates for2 and 3 PEs) driving a multi-Gaussian model is per-formed and the mean and width for the SPE of eachsingle PMT is extracted. Fig. 5 shows a sample fit.Before inclusion in the Monte Carlo model the SPEwidth is corrected for noise, since this is indepen-dently modeled, by subtracting in quadrature thewidth of the noise peak from the fitted SPE width.

The stability of the PMT gains was checked bytaking a second set of water calibration data after LSdata taking was complete. The gains for all PMTsin the array were observed to be very stable withinmeasured uncertainties.

B. PMT Pulse Shapes

The characteristic SPE pulse shapes need to bewell modeled in order to accurately reproduce thePMT time response in simulation. Events with anintegrated charge from −σ2 to +σ about the meanof the SPE charge were extracted from the SPEcalibration data for each PMT independently. Thesmaller bound on the negative range was introducedto avoid including noise events. These pulses werethen normalized to unit area and a constant frac-tion threshold was applied to determine a commonpoint of reference between all pulses from a singlePMT. This threshold was used to align pulses on aper-PMT basis, and a mean value along with upperand lower RMS values were calculated for the nor-

8

/ ndf 2χ 86.33 / 34

p0 5.191e+03± 1.612e+06

p1 0.0761± 0.4006

p2 0.06± 22.37

p3 1.947e+03± 1.704e+05

p4 0.9± 152.7

p5 0.79± 56.76

p6 7.741e+02± 1.495e+04

Charge (pC)0 200 400 600 800 1000 1200 1400

Ev

en

ts

1−10

1

10

210

310

410

/ ndf 2χ 86.33 / 34

p0 5.191e+03± 1.612e+06

p1 0.0761± 0.4006

p2 0.06± 22.37

p3 1.947e+03± 1.704e+05

p4 0.9± 152.7

p5 0.79± 56.76

p6 7.741e+02± 1.495e+04

Charge (pC)50− 0 50 100 150 200 250 300 350 400

Events

10

210

310

410

510

χ2 / ndf

Noise integralNoise μNoise σSPE integralSPE μSPE σ2PE integral

4754/60(3544±4)x103

-3.534±0.0076.455±0.005(555±2)x103

64.8±0.129.01±0.09(77±1)x103

FIG. 5. SPE charge distribution for a single PMT. Datais shown as black points with error bars, and the reddashed line is the result of the fit used to extract theshape of the SPE distribution.

malized voltage in 100 ps time bins. The result isshown for one PMT in the top panel of Fig. 6. Theuncertainty in the pulse region is consistent with thenoise of the ADCs; this was true for all PMTs. Theextracted mean PMT pulse shapes are shown for all12 array PMTs in the middle panel of Fig. 6. ThePMT-to-PMT variation in pulse shape is attributedto small variations in voltage dividers, signal cou-pling circuits, and individual PMT construction. Alog normal of the form

1

σ√

2πexp

[− (log[t− t0]− µ)

2

2σ2

](3)

was fit to the average pulse shape across all 12 PMTs(lower panel of Fig. 6) and the resulting fit parame-ters were used in the simulation to model PMT pulseshapes.

C. Per-Channel Time Delays

The total time delay due to PMT transit times,electronics, and cable lengths is measured for eachindividual channel and then corrected for in thedata. Time delays are measured relative to the av-erage across all PMTs in the array. Typical relativedelays are on the order of 100s ps. The primary cal-ibration is performed by deploying an LED at thetop of the dark box, such that the light path to eachPMT is similar. A full Monte Carlo simulation of

HaL

-20 0 20 40 60 80 100-0.30

-0.25

-0.20

-0.15

-0.10

-0.05

0.00

0.05

Time HnsL

VHar

b.

un

itsL

HbL

-20 0 20 40 60 80 100-0.30

-0.25

-0.20

-0.15

-0.10

-0.05

0.00

0.05

Time HnsL

VHar

b.

un

itsL

Log-Normal Fit

μ

σ

FIG. 6. PMT pulse shape characterization. (a) Meanand RMS spread for one PMT, showing excellent stabil-ity in the pulse region. (b) Averaged pulse shapes forSPE pulses for each individual PMT. (c) Log-normal fitto the average pulse shape across all PMTs. In all casesthe pulses are normalized to unit area, so the voltage isreported in arbitrary units.

this configuration is run and used to correct the datain order to take into account PMT-to-PMT varia-tions in photon ToF and geometry effects. Severaldatasets are generated by removing and reattach-ing the LED in order to quantify any uncertainty

9

∆ t(ns)

2− 1.5− 1− 0.5− 0 0.5 1 1.5 2

Eve

nts

0

1000

2000

3000

4000

5000

6000

χ2 / ndf

Constant

μ

σ

(a)1142/47

5290±21

-0.050±0.002

0.596±0.001

PMT ID0 1 2 3 4 5 6 7 8 9 10 11T

ime

Del

ays

(ns)

0.2−0.1−

00.10.20.3

LED

Muon

(b)

FIG. 7. (a) Time distribution for a single channel relativeto a reference channel, overlaid with the Gaussian fitused to extract the relative time delay. (b) Mean anduncertainty on the mean for the per-channel time delays.

in the LED position with respect to the simulatedone. Cosmic muons passing through the propagationmedium are used to evaluate systematics uncertain-ties in the calculation of the time delays, as theyprovide an abundant source of prompt Cherenkovlight. Events are selected by requiring a coincidencebetween the lower muon tag and the bottom scintil-lator panel. Again, a full Monte Carlo simulation isused to correct for ToF and geometry effects in thecalibration.

The time distribution for one channel is shown inthe upper panel of Fig. 7, where the offset representsthe time delay with respect to the average. The mea-sured time delays with uncertainties for each channelfrom each calibration are shown in the lower panel ofFig. 7. The two data sets show the same trend acrossthe PMT array. The MC-corrected LED data set isused to define the time delays for the final separa-tion analysis. The difference between the two datasets is taken as a measure of systematic error in thiscalibration.

VI. COSMIC MUON EVENT SELECTION

A deionized water target was used to optimizeevent selection criteria in order to select Cherenkovring events, and to reject backgrounds caused by in-strumentals and by other physics events. These cri-teria are then used to determine the sensitivity toidentifying ring candidates in LS. The criteria weredefined by classifying all events in the water dataset according to the clarity of a visible Cherenkovring (Sec. VI A), and then adjusting relevant cut val-ues to maximize acceptance of good ring candidatesand minimize contamination by non-Cherenkov-likeevents (Sec. VI B).

A. Event Classification

Events in the pure water data set were handscanned in order to understand the different typesof event topology and their sources. A classifica-tion scheme was developed to sort them according tohow well they matched the expected Cherenkov ringgeometry, for the purposes of defining a quantita-tive set of event-selection criteria (Sec. VI B). PMTswith an integrated charge greater than one third ofthe SPE charge for that PMT were counted as hit.Hits were grouped according to the radius of the hitPMT. The cross-shape PMT deployment results inthree radial groupings: the inner, middle and outerPMTs. The CHESS apparatus was designed suchthat the Cherenkov ring falls on the middle ring fora water target, and the outer ring for a pure LStarget. The total number of hit PMTs within eachgrouping, NHitinner, NHitmiddle, and NHitouter, wasdetermined by summing hits across all PMTs withinthat group. A perfect Cherenkov ring in water is ex-pected to have NHitmiddle = 4 while NHitinner =NHitouter = 0. “Ring” events were selected with thecriteria NHitmiddle − NHitinner − NHitouter > 2 toallow either one expected PMT to be missed or oneadditional PMT to be hit (but not both) to allowfor minor noise contamination and increased accep-tance.

“Background” events were sorted into categoriesaccording to event topology, in order to understandthe primary sources of background. These includedinstrumental events, which had no clear ring, so-called “follower” events, in which a secondary par-ticle generated Cherenkov light in the propagationmedium, causing unusually high charge on the innerPMTs, and events in which a cosmic muon showerlit up the majority of PMTs within the array.

10

B. Cut Development

Selection of vertical-going cosmic muon events wasachieved using a triple coincidence trigger. Thehardware trigger threshold (Sec. II H 2) applied tothe lower muon tag was set conservatively low tomaximize event acceptance. A software trigger wasapplied offline by applying a threshold to the chargeon both the upper and lower muon tags and to themuon panel directly below the setup in order to se-lect coincidence events.

Rejection of events containing either multiplemuons or secondary particles was achieved by re-quiring the charge on all veto panels except the paneldirectly below the setup to be consistent with zero.

The threshold values applied to the charge seen onthe cosmic tags (QU and QL for the upper and lowertag, respectively), and for each of the muon pan-els (QV 1 – QV 4) were selected by optimizing accep-tance of “ring” events and rejection of “background”events, according to the classification described inSec. VI A. The data before and after application ofthese cuts is shown in Fig. 8.

Cuts were also applied to remove so-called “fol-lower” events, in which muons or muon followersgenerated Cherenkov light in the acrylic propagationmedium, which contaminated the sample of singleCherenkov ring events. This occurred in two cases:

• Electron contamination: The cosmic muontriggered the acquisition, but a secondary par-ticle passed through the propagation medium.

• Muon contamination: The secondary particletriggered the lower muon tag and the muonitself passed through the propagation medium.

The secondary particles do not always make it tothe muon panels and therefore cannot be vetoed di-rectly, thus the only information available for reject-ing these events is the PMT array. Clean cosmicmuon events are expected to produce hits only onthe middle PMTs (for a water target, or outer PMTsfor LS). In both cases of follower events, the major-ity of Cherenkov light generated in the propagationmedium was observed to fall primarily on the inner-most PMTs within the array. This was confirmed byMonte Carlo simulations of each event type, whichdemonstrated that muon followers do produce eventswith this topology. Simulations of these events showa clear tail in the PE distribution observed on theinner PMT group for both water and LAB/PPO tar-gets (Fig. 9). These events typically create between30 and 400 PEs in the innermost PMTs, making

their identification in both water and LS possible byanalysis of the charge on these PMTs.

A cut was designed to reject these events based onevent topology: since the Cherenkov ring geometryis not expected to produce hits on the inner PMTsfor either a water or LS target, the total number ofestimated PEs on the inner PMTs was used to iden-tify clean rings. In water this charge is expected tobe very low (consistent with noise), whereas in LSthe total charge will be higher due to scintillationphotons. However, as demonstrated in Fig. 9, a largefraction of follower events can still be removed with ahigh-charge cut. A cut was placed at a summed PEcount of 40 for water and LAB targets, and 500 forLAB/PPO. These cuts were conservatively chosenbased on the simulations to reject events with highlight contamination due to secondary particles witha minimal impact on the efficiency. Performance ofthis cut in water is shown in the top panel of Fig. 9.The high-charge tail due to follower events is simi-lar to that seen in the simulation, supporting use ofthe simulation for defining this cut for both waterand LS targets. The cut value selected for a wa-ter target removes a large fraction of the remainingbackground events, with zero sacrifice in the controlsample.

VII. EVENT-LEVEL ANALYSIS

Time separation of the Cherenkov and scintilla-tion photon populations is based on the hit-timeresidual distributions for each radial PMT grouping(inner, middle and outer). The hit-time residualsare evaluated as the PMT hit times with respectto the event time, corrected by per-channel delays(Sec. V C) and by the photon ToF. The ToF de-pends on the distance between the target and thePMTs, and on the refractive indices of the differentmedia. It is estimated for each PMT radial group tobe 626 ps, 536 ps and 473 ps for the inner, middleand outer PMT rings, respectively.

The time at which the cosmic muon passesthrough the target, referred as the event time, iscalculated using two different approaches. The moststraightforward is using the time at which the lowermuon tag goes above threshold, since this triggersthe acquisition. Nevertheless, the cosmic tags sufferfrom a poorer time resolution than the fast PMTsin the array due both to the scintillator responseand the larger PMT TTS. Hence, a higher preci-sion event time is defined using the PMT hit timesby using the median of the four earliest hits in theevent, after time calibration and photon ToF correc-

11

HaL

1 10 100 1000 104

1

10

100

1000

104

QU HmV nsL

Counts

HbL

1 10 100 1000 104

1

10

100

1000

104

QL HmV nsL

Counts

Before SelectionBackground

Ring

After SelectionBackground

Ring

HcL

1 10 100 1000 104 105 106

1

10

100

1000

104

QV1HmV nsL

Counts

HdL

1 10 100 1000 104 105 106

1

10

100

1000

104

QV2 HmV nsL

Counts

HeL

1 10 100 1000 104 105 106

1

10

100

1000

104

QV3 HmV nsL

Counts

HfL

1 10 100 1000 104 105 106

1

10

100

1000

104

QV4 HmV nsL

Counts

FIG. 8. (Top to bottom, left to right) Charge distribution of events on the upper and lower cosmic tags (QU andQL) and the four muon panels (QV 1 – QV 4). Panel 1 is located directly below the CHESS apparatus. Events areseparated into ring (blue, lower line) and background (red, upper line) according to the criterion in Sec. VI A. Verticalblack dashed lines show the cut values in each case with arrows indicating the acceptance region. Distributions areshown before (dashed) and after (solid) application of cuts on these 6 parameters.

tion. This provides a robust time reference since theprompt hits are due to Cherenkov light, whose timeprofile is very sharp.

The hit-time residual distributions evaluated us-ing each option for the event time are shown inFig. 10, overlaid with the Monte Carlo prediction

in each case. The residual distribution with respectto the lower muon tag time is extremely well repro-duced by the Monte Carlo, demonstrating the preci-sion of the model. The simulation slightly under pre-dicts the width of the higher precision distributionof residuals with respect to the reconstructed event

12

Σ NPEinner

0 50 100 150 200 250 300 350 400

Eve

nts

2−10

1−10

1

10

210DataNo contaminationMuon contaminationElectron contaminationCombination

( a )

innerNPEΣ0 200 400 600 800 1000 1200

Eve

nts

1−10

1

10

210

310( b )

FIG. 9. Monte Carlo simulation of the summed PEdistribution on the inner PMT group due to cosmicmuon events with perfect rings (turquoise, right diag-onal hatching), muon contamination (orange, left diago-nal hatching), electron contamination (purple, right di-agonal dashed hatching), and the total (red, solid). (a)Water target, with water data overlaid to show the agree-ment, and (b) LAB/PPO target. The vertical blackdashed line represents the chosen cut value, with arrowsto illustrate the acceptance region.

time. This could be explained by multi-photon ef-fects, small differences in the PMT pulse shape, andunderestimated PMT TTS. A Gaussian correctionof σ = 214 ps is included in the Monte Carlo in or-der to take these effects into account, and to betterreproduce the time resolution seen in data.

VIII. CHERENKOV RING IMAGING INWATER

After application of the event selection criteria de-scribed in Sec. VI, 137 ring candidates were selectedin the water dataset. The number of detected PEsand first photon hit-time residuals for a typical eventare shown in Fig. 11. The averages across the data

Hit Time Residuals (ns)1.5− 1− 0.5− 0 0.5 1 1.5

Eve

nts

0

5

10

15

20

25

30

35

40

45DataNominal MC

( a )

Hit Time Residuals (ns)1.5− 1− 0.5− 0 0.5 1 1.5

Eve

nts

0

20

40

60

80

100

120

Data

Nominal MC

Corrected MC

( b )

FIG. 10. Distribution of hit-time residuals for ring can-didate events in water for middle PMTs, where theCherenkov ring is expected. Data points are shown withstatistical errors, with the Monte Carlo prediction over-laid (dashed lines). (a) Hit times with respect to thelower muon tag time and (b) hit times with respect tothe reconstructed event time.

set for both the number of detected PEs per PMTand the hit-time residuals are shown in Fig. 12; bothshow a clear ring structure.

The distributions of hit-time residuals for eachPMT radial group (inner, middle and outer PMTs)are shown in Fig. 13, with the Monte Carlo pre-diction overlaid (with the additional smearing de-scribed in Sec. VII). The middle PMTs are the onlyones detecting a sizable amount of light, and theirtime distribution is very sharp, compatible with the

13

1

2

3

4

5

6

7

8

Num

ber o

f pho

toel

ectro

ns

1

2

3

4

5

6

7

8( a )

Hit

Tim

e R

esid

uals

(ns

)

0.5−

0

0.5

1

1.5

2

Hit

Tim

e R

esid

uals

(ns

)

0.5−

0

0.5

1

1.5

2( b )

FIG. 11. Typical ring event in the water data set. (a) Es-timated number of detected PEs and (b) hit time resid-uals.

expected Cherenkov rings in water. The inner andouter PMTs are rarely hit.

The characteristic sharp time distribution ofCherenkov light provides an excellent source for es-timating the CHESS time resolution. The time dis-tribution of the observed Cherenkov rings provides ameasurement of 338±12 ps FWHM for the time pre-cision of the CHESS detector. The limiting factor inthis precision is the PMT TTS (300 ps FWHM).

IX. PROSPECTS WITH LIQUIDSCINTILLATOR TARGETS

An analysis of Monte Carlo data has been per-formed to predict the performance of CHESS for

0

1

2

3

4

5

6

7

8

9

10

0

1

2

3

4

5

6

7

8

9

10( a )

Num

ber o

f pho

toel

ectro

nsH

it T

ime

Res

idua

ls (

ns)

0.4−

0.3−

0.2−

0.1−

0

0.1

0.2

0.3

0.4

Hit

Tim

e R

esid

uals

(ns

)

0.4−

0.3−

0.2−

0.1−

0

0.1

0.2

0.3

0.4

( b )

FIG. 12. (a) Averaged number of detected PE and (b)averaged first-photon hit time residual per individualPMT for ring candidates in water data.

ring imaging and Cherenkov / scintillation separa-tion with a pure LS target: using both LAB andLAB/PPO. The targets are simulated using proper-ties from [14, 15, 40, 42, 43]. The scintillation lightyield is set to 1010 photons/MeV for LAB [42] and10800 photons/MeV for LAB/PPO [15]. Quench-ing is modeled using values for Birk’s constant ofkB = 0.0798 mm/MeV in LAB and LAB/PPO [43].The emission spectra are shown in Fig. 1. The timeprofile is modeled in the simulation as described inSec. III B, with a rise time of τr = 1 ns for LAB/PPOand and τr = 7.7 ns for LAB, and decay timesof τ1 = 4.3 ns, τ2 = 16 ns, τ3 = 166 ns [40] forLAB/PPO, and τ1 = 36.6 ns, τ2 = τ3 = 0 [42] forLAB.

Several thousand cosmic muon events are simu-

14

Hit Time Residuals (ns)1.5− 1− 0.5− 0 0.5 1 1.5

Eve

nts

0

20

40

60

80

100 Outer PMTs - Data MC

Mid PMTs - Data MC

Inner PMTs - Data MC

FIG. 13. Distribution of hit time residuals for ring can-didate events in water for each PMT grouping. Datapoints are shown with statistical errors, with the MonteCarlo prediction overlaid (dashed lines).

Hit Time Residuals (ns)2− 0 2 4 6 8

Eve

nts

1−10

1

10

210

310

Outer PMTs - MC

Mid PMTs - MC

Inner PMTs - MC

FIG. 14. Projected hit-time residual distributions forLAB as predicted from the Monte Carlo simulation.

lated for each target, producing the hit-time residualdistributions shown in Fig. 14 for LAB and Fig. 15for LAB/PPO. The small peak on the inner and mid-dle PMTs seen in LAB is due to Cherenkov lightcontamination from follower events. The earliesthits are registered in the outer PMTs, where theCherenkov ring is expected, while later features are

Hit Time Residuals (ns)2− 1− 0 1 2 3 4 5

Eve

nts

0

50

100

150

200

250

300

350

Outer PMTs - MC

Mid PMTs - MC

Inner PMTs - MC

FIG. 15. Projected hit-time residual distributions forLAB/PPO as predicted from the Monte Carlo simula-tion.

due to scintillation light. Between 5 and 10 PE areexpected on each PMT within the Cherenkov ringdue to Cherenkov light alone. The PMT hit-timeresidual is based on the first photon hit time foreach channel, thus with high confidence the hit-timeof each PMT within the Cherenkov ring can be as-signed to Cherenkov photon hits. Hits outside theCherenkov ring are due to scintillation light. Theseparation of Cherenkov and scintillation photonscan thus be defined by comparing the distributionsof hit-time residuals on PMTs within the expectedCherenkov ring location (the outer PMTs for LABand LAB/PPO) and those outside the ring (innerand middle PMTs).

A timing cut can be developed to optimizethis separation, selecting hits that occur before atime threshold, tc. The efficiency of identifyingCherenkov hits is defined as the fraction of outerPMT hits (Cherenkov hits) that occur before tc. Thecontamination due to scintillation photons is evalu-ated as the fraction of all PMT hits occuring beforetc that are due to scintillation photons i.e. the num-ber of inner and middle PMT hits occurring beforetc divided by the total number of hits before tc.

A value of tc = 0.4 ns results in a Cherenkov effi-ciency of 94±1% in LAB and 81±1% in LAB/PPO,with contaminations of 12±1% and 26±1%, respec-tively.

A study has been performed in [11] of the impact

15

of such separation on a potential large-scale NLDBDsearch. This paper finds that solar neutrinos becomea limiting background in a next-generation detec-tor. This dominant background can be rejected byuse of the directional Cherenkov component, if sep-aration of Cherenkov and scintillation light can beachieved. Assuming a conservative rejection factorof two for these events, the study shows that sen-sitivity can be achieved down to the bottom of theinverted hierarchy (15 meV). This study assumed aconservatively low light yield, based on the assump-tion that a WbLS target would be required in or-der to achieve the Cherenkov separation. However,the work in this paper demonstrates the potentialto identify the Cherenkov component even in a pureLS target, thus allowing directional sensitivity in ahigh light yield detector. Such an experiment couldhave increased sensitivity to NLDBD, even into thenormal hierarchy [44]. Ref. [45] discusses an idea forhow to use this separation to extract particle direc-tion in a scintillator detector.

X. CONCLUSIONS

The time resolution of the CHESS experimenthas been demonstrated to achieve the sub-ns preci-sion required for successful separation of Cherenkovand scintillation light. The principle of Cherenkovring imaging has been demonstrated using a deion-ized water target. A time resolution of 338 ± 12 psFWHM has been achieved and clear Cherenkov ringshave been detected in both charge and time on anevent-by-event basis. A detailed simulation withLAB and LAB/PPO shows that, with this time reso-lution, time-based separation of the Cherenkov com-ponent from the scintillation light is possible with anefficiency of 94% and 81%, respectively, with scintil-lation contamination of 12% and 26%.

If this separation can indeed be demonstrated inthe experimental data then it would have signifi-cant implications for next-generation neutrino ex-

periments. Successful Cherenkov / scintillation sep-aration would benefit a broad physics program, in-cluding: low-energy physics such as NLDBD and so-lar neutrinos; astrophysics topics such as supernovaneutrinos and the diffuse supernova neutrino back-ground; and high-energy physics such as nucleondecay, neutrino mass hierarchy, and CP violation.Understanding the degree of separation that can beachieved, and quantifying how this varies with thespecifics of the target cocktail will be critical stepsin developing the program for a future large-scaleexperiment such as Theia [12].

The next phase of CHESS will deploy pure LS inthe target, followed by WbLS samples with varyingLS fractions. This will enable a full understand-ing of how the signal separation varies with cock-tail. In a future phase CHESS will be upgraded bypopulating the array with 12 additional fast PMTs.CHESS can also be used as a test bench for studies ofnext-generation MCP-based photon detectors [16–18]. The narrow pulse width of these detectors (10sof ps, compared to 5–10 ns for a typical PMT) couldsubstantially improve the potential for charge-basedseparation by allowing much higher precision PEcounting.

ACKNOWLEDGMENTS

This work was supported by the LaboratoryDirected Research and Development Program ofLawrence Berkeley National Laboratory under U.S.Department of Energy Contract No. DEAC02-05CH11231. The work conducted at BrookhavenNational Laboratory was supported by the U.S.Department of Energy under contract DE-AC02-98CH10886. The authors would like to thank theSNO+ collaboration for providing data on the op-tical properties of LAB/PPO, including the lightyield, absorption and reemission spectra, and refrac-tive index.

[1] C. Bratton et al., Phys. Rev. D 37, 3361 (1988).[2] Y. Fukuda et al. (Super-Kamiokande Collabora-

tion), Phys. Rev. Lett. 81, 1562 (1998).[3] Q. R. Ahmad et al. (SNO Collaboration), Phys.

Rev. Lett. 89, 011301 (2002).[4] S. Abe et al. (KamLAND Collaboration), Phys.

Rev. Lett. 100, 221803 (2008).[5] Borexino Collaboration, Nature 512, 383 (2014).

[6] A. Aguilar-Arevalo et al. (LSND Collaboration),Phys. Rev. D 64, 112007 (2001).

[7] P. A. Cerenkov, Phys. Rev. 52, 378 (1937).[8] J. B. Birks, D. Fry, L. Costrell, and K. Kandiah,

The Theory and Practice of Scintillation Counting(Pergamon, 1964).

[9] S. Andringa et al. (SNO+ Collaboration), Adv.High Energy Phys. 2016, 6194250 (2016).

16

[10] M. Yeh, S. Hans, W. Beriguete, R. Rosero, L. Hu,R. Hahn, M. Diwan, D. Jae, S. Kettell, and L. Lit-tenberg, Nucl. Instrum. Methods A660, 51 (2011).

[11] J. R. Alonso et al., arXiv:1409.5864v3 [physics.ins-det] (2014).

[12] G. D. Orebi Gann for the THEIA interest group, inProceedings of the Prospects in Neutrino Physics,arXiv:1504.08284 [physics.ins-det] (2015).

[13] https://www.hamamatsu.com/resources/pdf/

etd/R11265U_H11934_TPMH1336E.pdf.[14] C. Buck & M. Yeh, J. Phys. G: Nucl. Part. Phys.

43 093001 (2016).[15] SNO+ Collaboration (private communications).[16] B. Adams, A. Elagin, H. Frisch, R. Obaid,

E. Oberla, A. Vostrikov, R. Wagner, and M. Wet-stein, Nucl. Instum. Methods A732, 392 (2013).

[17] B. W. Adams et al. (LAPPD Collaboration),arXiv:1603.01843 [physics.ins-det] (2016), [Submit-ted to: JINST].

[18] O. H. W. Siegmund, J. B. McPhate, J. V. Vallerga,A. S. Tremsin, H. E. Frisch, J. W. Elam, A. U.Mane, and R. G. Wagner, J. Instrum. 9, C04002(2014).

[19] http://www.cog.de/fileadmin/downloads/COG_

Resist_Folder_EN_Einzelseiten.pdf.[20] http://www.eljentechnology.com/index.php/

products/accessories/ej-550-ej-552.[21] http://www.hamamatsu.com/us/en/product/

alpha/P/3002/H3164-10/index.html.[22] http://www.eljentechnology.com/index.

php/products/plastic-scintillators/

ej-200-ej-204-ej-208-ej-212.[23] http://www.et-enterprises.com/

photomultipliers.[24] http://www.hitachimetals.

com/materials-products/

amorphous-nanocrystalline/

magnetic-shielding-sheet/documents/fm_

shield_ms-f_ms-fr.pdf.[25] http://www.spectrumtechniques.com/products/

sources/strontium-90/.

[26] B. J. Land, http://github.com/BenLand100/

WbLSdaq, (2016).[27] http://www.caen.it/csite/CaenProd.jsp?

idmod=689&parent=38.[28] http://www.caen.it/csite/CaenProd.jsp?

idmod=417&parent=11.[29] The HDF Group, http://www.hdfgroup.org (1997-

2016).[30] http://www.caen.it/csite/CaenProd.jsp?

idmod=657&parent=23.[31] http://www.caen.it/csite/CaenProd.jsp?

parent=11&idmod=779.[32] http://www.caen.it/csite/CaenProd.jsp?

idmod=661&parent=11.[33] https://www.psi.ch/drs/.[34] http://teledynelecroy.com/oscilloscope/

oscilloscopemodel.aspx?modelid=4782&capid=

102&mid=504.[35] S. Seibert et al., http://rat.readthedocs.io/en/

latest/.[36] S. Agostinelli et al., Nucl. Instrum. Methods Phys.

Res. A506, 250 (2003).[37] A. Tang, G. Horton-Smith, V. A. Kudryavtsev &

A. Tonazzo, Phys. Rev. D 74, 053007 (2006); M.Guan et al., arXiv:1509.06176 [hep-ex] (2015); G.Mengyun, C. Jun, Y. Changgen, S. Yaxuan & K.-B.Luk, LBNL-4262E report.

[38] G. Horton-Smith, http://neutrino.phys.ksu.

edu/~GLG4sim/.[39] R. L. McGuire and R. C. Palmer, IEEE Trans. Nucl.

Sci. 14, 217 (1967).[40] H. M. O’Keeffe, E. O’Sullivan, and M. C. Chen,

Nucl. Instum. Methods A640, 119 (2011).[41] Hamamatsu (private communication).[42] M. Li et al., arXiv:1511.09339 [physics.ins-det]

(2015).[43] B. von Krosigk, L. Neumann, R. Nolte, S. Rottger

and K. Zuber, Eur. Phys. J. C 73, 2390 (2013).[44] S. D. Biller, Phys. Rev. D 87, 071301 (2013).[45] C. Aberle, A. Elagin, H. J. Frisch, M. Wetstein, &

L. Winslow, JINST 9 (2014) P06012.

17