Nuclear Instruments and Methods in Physics Research A 644 (2011) 18–26

Contents lists available at ScienceDirect

Nuclear Instruments and Methods inPhysics Research A

0168-90

doi:10.1

� Corr

E-m

journal homepage: www.elsevier.com/locate/nima

A highly efficient neutron veto for dark matter experiments

Alex Wright �, Pablo Mosteiro, Ben Loer, Frank Calaprice

Department of Physics, Jadwin Hall, Princeton University, Princeton, NJ 08540, USA

a r t i c l e i n f o

Article history:

Received 21 October 2010

Received in revised form

3 April 2011

Accepted 5 April 2011Available online 14 April 2011

Keywords:

Direct-detection dark matter search

Low-background techniques

Neutron veto

Boron-loaded liquid scintillator

02/$ - see front matter & 2011 Elsevier B.V. A

016/j.nima.2011.04.009

esponding author. Tel.: þ1 609 258 0154.

ail address: ajw@princeton.edu (A. Wright).

a b s t r a c t

We present a conceptual design for an active neutron veto, based on boron-loaded liquid scintillator, for

use in direct-detection dark matter experiments. The simulated efficiency of a 1 m thick veto is greater

than 99.5% for background events produced by radiogenic neutrons, while the background due to

externally produced cosmogenic neutrons is reduced by more than 95%. The ability of the veto to both

significantly suppress, and provide in situ measurements of, these two dominant sources of background

would make the next generation of dark matter experiments much more robust, and dramatically

improve the credibility of a dark matter detection claim based on the observation of a few recoil events.

The veto would also allow direct extrapolation between the background-free operation of a small

detector and the physics reach of a larger detector of similar construction.

& 2011 Elsevier B.V. All rights reserved.

1. Introduction

Astrophysical and cosmological evidence strongly imply theexistence of ‘‘dark matter,’’ a non-baryonic, non-luminous form ofmatter, at an energy density roughly five times that of baryonicmatter. One well-motivated group of theories predicts that thedark matter is composed of weakly interacting massive particlesor ‘‘WIMPs.’’ A global program of increasingly sensitive experi-ments is searching for the possible WIMP dark matter byattempting to detect nuclear recoils produced by WIMP interac-tions with nuclei.

Neutrons pose a particularly relevant background to the WIMPdark matter search, as single scatter neutron events producerecoiling nuclei which are event-by-event indistinguishable fromWIMP-induced recoils. Neutron-induced backgrounds in a darkmatter detector can be produced both by neutrons createdthrough radiogenic processes such as (a,n) reactions and sponta-neous fission and by cosmic-ray muon induced spallation.Neutron backgrounds in dark matter experiments have typicallybeen mitigated by constructing the detectors from materials withvery low levels of intrinsic radioactivity, by shielding the experi-ments using passive neutron absorbers, and by operating theexperiments in underground laboratories to minimize the cosmicmuon flux. As experiments continue to improve in sensitivity,however, these passive techniques are reaching their practical limits.

Dark matter experiments are already reaching sizes andsensitivities at which it is extremely difficult to locate or produce

ll rights reserved.

materials with levels of intrinsic radioactivity low enough to builddetectors with acceptably small rates of radiogenic neutronproduction, and only the deepest of underground laboratoriesnow have cosmogenic neutron fluxes low enough to permit thebackground-free operation of such detectors. Moreover, at thislevel it is extremely difficult to determine the expected neutronbackground, using calculations based on ex situ material assaytechniques, precisely enough that a credible claim of dark matterdetection could be made based on the observation of a few recoilevents. This difficulty arises from the fact that in very purematerials sample-to-sample variations in background levels canbe large. Thus, extrapolation from the level of contaminationmeasured in one assayed sample of a material to the contamina-tion in other samples of that same material used in detectorconstruction can be a source of significant uncertainty. Theuncertainty in the neutron background is compounded by sig-nificant uncertainty in our ability to model cosmogenic neutronbackgrounds at underground laboratories (see, e.g. [1]). Indeed, atthe present time all of the leading dark matter experiments seesome nuclear recoil-like events, so these experiments are eitherbackground limited or unable to make a detection claim based onthe few events observed.

The use of an active neutron veto is one way in whichexperiments can both lower their neutron backgrounds andprecisely assay them in situ. A high-efficiency active neutron vetowould not only significantly improve the sensitivity of an experi-ment which would otherwise be limited by neutron backgrounds,but it would also allow an experiment which has been successfulin achieving a low neutron background to demonstrate convin-cingly that the number of residual neutron-induced recoil back-ground events in the data after applying the veto is extremely

A. Wright et al. / Nuclear Instruments and Methods in Physics Research A 644 (2011) 18–26 19

small. This would help to give such an experiment the ability tomake a convincing claim of dark matter detection based on theobservation of a few events. In addition, the ability to extrapolatea well-characterized background, coupled with the direct sup-pression afforded by the neutron veto, would allow a smallerexperiment to demonstrate directly the potential for a muchlarger experiment of similar construction to operate backgroundfree. The practical utility of all of these applications, however, isdictated by the absolute efficiency of the neutron veto; as a result,the development of a highly efficient veto system is extremelydesirable.

The obvious benefit of neutron veto capability has led mostcurrent dark matter experiments to implement veto procedures.In some (e.g. [2,3]), the veto is achieved by segmenting (throughposition reconstruction and/or physical segmentation) the activevolume of the dark matter detector to look for neutrons thatproduce more than one recoil event. Other experiments (e.g. [4,5])have deployed separate, dedicated veto systems. In the future,very large, monolithic detectors could veto neutron events quiteefficiently using internal coincidences, except near the detectorwalls where there is a reasonable probability for the recoilingneutron to escape. In order to have a highly efficient neutron veto,then, these large detectors will have to either take a reasonablylarge cut in fiducial volume or install an external neutronveto system.

A neutron veto system with a very high detection efficiencycan be produced by surrounding a dark matter detector with alayer of liquid scintillator. Such a veto, with a thickness of order1 m, is sufficient to detect a very high percentage of the radio-genic neutrons produced by the inner detector, and a significantfraction of cosmogenic neutrons. However, at about 250 ms, thecapture time for thermal neutrons in a liquid scintillator is ratherlong. This means that in order to efficiently veto the (promptlyproduced) neutron-induced nuclear recoil backgrounds, vetowindows of millisecond duration are necessary. With such a longveto window, the background rate in the scintillator must bequite low, less than � 100 Hz, to keep the veto-induced dead timein the dark matter detector from becoming significant. Achievingsuch a low event rate in the veto requires bulky and expensivepassive shielding; even the use of conventional photomultipliertubes to instrument the veto would be difficult, as the event ratedue to radioactive contaminants in the PMTs themselves wouldbe excessive.

In this paper we show that relatively compact, highly efficientneutron vetoes for dark matter detectors are practically realizablethrough the use of boron-loaded liquid scintillator. We havestudied such a veto using Monte Carlo simulations in the contextof the DarkSide-50 dark matter detector. The efficiency of the vetois shown to be very high. The total event rate in the veto frominternal and external sources (with a relatively modest amount ofpassive shielding against external gamma rays) is estimated andfound to be acceptable, even using conventional PMTs. Opticalsimulations of the veto show that the very low energy thresholdnecessary to reliably detect the reaction products of neutroncapture on 10B is comfortably achievable, even under conservativeassumptions about the optical characteristics of the veto.

2. Boron-loaded scintillator

Neutron detection using boron-loaded liquid scintillator, pro-duced by adding tri-methyl borate (TMB) to standard scintillatorcocktails, was first investigated in Ref. [6] and subsequentlydeveloped into practical detectors by Ref. [7].

10B, which has a natural abundance of about 20%, capturesthermal neutrons with a very high total capture cross-section

(3837(9)b) via two channels [8]:

10Bþn-7Liðg:s:Þþa 6:4%

-7Li�þa,7Li

�-7Liþgð478 keVÞ 93:7%

Importantly for the current discussion, the nuclear recoil reactionproducts carry a significant amount of kinetic energy (in the decay tothe excited state of 7Li, EðaÞ ¼ 1471 keV and E(7Li)¼839 keV, whilefor the ground state decay EðaÞ ¼ 1775 keV and E(7Li)¼1015 keV).The light output from the nuclear recoils is heavily quenchedin liquid scintillator, to the level of 50–60 keV electron recoils(50–60 keVee) [9,10], but, as will be shown, this is still detectable.The ability to detect neutron captures via scintillation from thenuclear recoil products makes relatively compact neutron detec-tors made from boron-loaded scintillator possible, as it is notnecessary for the detector to be large enough to contain capture-induced gamma rays.

TMB-loaded scintillators have also been investigated for usein large, low background particle detectors, as reported inRefs. [11,12]. In particular, scintillator cocktails with TMB loadingof up to 80%, and with light output, optical attenuation, radio-purity, and scintillator stability properties suitable for a large,low-background neutrino experiment were identified in Ref. [11].

3. Monte Carlo modeling and validation

The Monte Carlo studies which constitute the bulk of thisreport were carried out using the GEANT4 Toolkit (version 4.9.3)[13], via a flexible physics and geometry interface developed atPrinceton University.

The energy spectra used in simulating the radiogenic neutronsproduced in different materials were determined by first calculat-ing the individual spectra expected from (a,n) and fissionprocesses due to 235U, 238U, and 232Th chain activities in eachmaterial using the SOURCES4A software package [14]. These‘‘component spectra’’ were then combined, using measured ratiosof the different radioactive species, to give the total radiogenicneutron spectrum for each detector material. Cosmogenic neu-trons were generated with the approximate energy spectrumgiven in Ref. [15] for Italy’s Gran Sasso National Laboratory(‘‘LNGS’’). Representative radiogenic and cosmogenic neutronenergy spectra are shown in Fig. 1.

In what follows, ‘‘pure scintillator’’ is pseudocumene (C9H12,r¼ 0:876 g=cm3), while ‘‘boron loaded scintillator’’ is 50% w/wTMB in pseudocumene with a composition of 62.3% C, 23.1% O,9.4% H, and 5.2% B by mass. It is assumed to have a density equalto that of pseudocumene (in fact the density will be slightlyhigher, which will improve the neutron capture performancerelative to these simulations). The wavelength shifter necessaryfor the efficient optical performance of the scintillator, which islikely to be 2,5-diphenyloxazole (PPO) at the level of a few gramsper litre in both boron-loaded and unloaded scintillator, will havea negligible effect on the overall neutron capture efficiency and isneglected in the simulation. All elements, including boron, areassumed to have natural isotopic abundances.

In order to confirm that the simulation provides a reasonablereproduction of neutron behavior, a number of ‘‘benchmark’’comparisons have been made:

1.

The simulated mean capture time for radiogenic neutrons inpseudocumene is 25371 ms. This can be compared to the256:070:4 ms neutron capture time observed by Borexino[16], which uses pseudocumene-based scintillator.

2.

In the simulation, 99.1% of thermal neutrons captured in theboron-loaded scintillator were captured by 10B. Based on theratio of the neutron capture cross-sections of the scintillator

Table 1The mean free path (MFP) of neutrons of different energies in water, scintillator

and boron-loaded scintillator, as simulated (‘‘Sim.’’) and as calculated using the

neutron interaction cross-sections in Ref. [8] and the chemical composition of the

veto (‘‘XS,’’ where data is available). In the Monte Carlo, only those interactions

resulting in the creation of a secondary particle with more than 1 eV of kinetic

energy were directly recorded. Thus, some low energy interactions, low angle

forward scattering for example, were excluded in the determination of the Monte

Carlo MPFs; this could help to account for the longer interaction lengths in the

Monte Carlo.

Neutron

energy (MeV)

MFP in

water (cm)

MFP in pure

scintillator (cm)

MFP in loaded

scintillator (cm)

Sim. XS Sim. XS Sim. XS

10 8.9 9.4 9.9 10.5 10.5 10.9

20 12.8 11.5 13.3 11.8 13.6 12.3

50 26.7 19.5 29.9 21.8 30.4 –

100 40.9 38.0 45.3 43.3 46.6 –

200 63.5 – 70.9 – 71.1 –

Energy (MeV)

0 2 4 6 8 10

Rel

ativ

e N

eutr

on P

rodu

ctio

n R

ate

Energy (GeV)

0 0.5 1 1.5 2 2.5 3 3.5

Rel

ativ

e N

eutr

on P

rodu

ctio

n R

ate

Fig. 1. The neutron energy spectra used in the simulations for (a) radiogenic neutrons from the photodetectors and (b) external cosmogenic neutrons.

A. Wright et al. / Nuclear Instruments and Methods in Physics Research A 644 (2011) 18–2620

components (from Ref. [17]), one expects that 98.770.2% ofthermal neutrons should be captured by 10B (at 19.9% 10Babundance).

3.

The fraction of neutron captures by 10B that produce 7Li in thefirst excited state is 93.6770.07% in the simulation, in goodagreement with the expected value of 93.7% [8].

4.

The mean free paths of neutrons of different energies deducedfrom the results of the simulation, as well as the mean freepaths expected based on the cross-sections in Ref. [8] areshown in Table 1.

These comparisons suggest that, although not perfect, neutronpropagation and capture are reproduced reasonably well by thesimulation code, and that the predictions of the neutron vetosimulation might, therefore, be expected to provide reasonablepredictions of the performance of the veto.

4. Neutron detection with boron-loaded scintillator

Using the Monte Carlo described above, we have simulated thethermalization and capture of radiogenic neutrons in pure andboron-loaded scintillators. The neutrons were generated at the centerof large, 2 m radius, uniform volumes of scintillator. Fig. 2 shows theradial and time distributions of the simulated neutron captures. Ascan be seen, the addition of the boron reduces the mean neutroncapture time by more than a factor of 100, from 253 to 2:3 ms.

The distributions shown in Fig. 2, while illustrative, do notcorrespond exactly to the distribution of veto event production inthe scintillator. Two main factors contribute to this difference:first, capturing a neutron does not necessarily result in a vetosignal at the point of capture, especially in pure scintillator.Instead, secondary particles, particularly gamma rays, whichpropagate some distance before depositing a detectable amountof energy in the scintillator are often produced. Second, it ispossible that the neutron deposits sufficient energy via nuclearrecoils to produce a veto signal before being captured. As theserecoils occur promptly, the recoil-induced veto signals can occursignificantly earlier than the capture signals.

To investigate these effects, a ‘‘veto signal’’ was assumed to begenerated in the simulation if 40 keVee or more was deposited inthe scintillator within any 1 ms time window (the feasibility ofoperating with such a low threshold is discussed in Section 7).The quenching of heavy particles in the TMB-loaded scintillatorwas assumed to be identical to quenching in unloaded scintillator.Quenching for protons and alpha particles were treated sepa-rately following [18,19]; all heavier recoils were quenched ascarbon [20,18]. These quenching values give 50 keVee energydeposition in the scintillator from the recoil products ofnþ10B-aþ7Li

�, in good agreement with observations in TMB-

loaded (and hence diluted) scintillator [9,10].Fig. 3 shows the time distribution for the production of the first

veto signal (if any) for each neutron event, and the detector radiusnecessary to contain the energy deposited by that event. Asexpected, in pure scintillator the spatial distribution of veto triggersis broader than the distribution of neutron captures because ofgamma-ray propagation. For boron-loaded scintillator, by contrast,this broadening does not occur because the recoil daughters fromneutron capture on 10B deposit sufficient energy at the site ofcapture to produce a veto signal. In fact, in boron-loaded scintillatorthe distribution of veto trigger production is narrower than thedistribution of neutron captures due to the generation of vetotriggers by neutrons scattering prior to capture. Tables 2 and 3contrast the radial and time windows necessary to contain, withdifferent probabilities, the veto signal after the production of aradiogenic neutron in pure and boron-loaded scintillator. As can beseen, the addition of boron decreases the windows in both radiusand time necessary to veto the neutrons with high probability.

4.1. Other loading options

We note that the reduction in the average capture time ofneutrons in boron-loaded scintillator compared to pure

Mean 130RMS 84

Radius of Capture (mm)

0

Frac

tion

of C

aptu

res

/ 10m

m

Mean 170RMS 93

Boron-loaded Scintillator

Pure Scintillator

Mean 3.3e-06RMS 2.3e-06

Capture Time (s)

10-6 10-5 10-4 10-3

Frac

tion

of C

aptu

res

/ µs

10-5

10-4

10-3

10-2

10-1

1

Mean 2.5e-04RMS 2.5e-04

Boron-loaded Scintillator

Pure Scintillator

200 400 600 800 1000 1200 1400 1600 1800 2000

10-6

10-5

10-4

10-3

10-2

10-1

Fig. 2. The simulated distributions in radius and time of the captures of centrally produced radiogenic neutrons in large, uniform volumes of boron-loaded and pure liquid

scintillator.

Mean 86RMS 68

Threshold Radius (mm)

0

Frac

tion

of E

vent

s / 1

0mm

10-6

10-5

10-4

10-3

10-2

10-1

Mean 100RMS 120

Boron-loaded Scintillator

Pure Scintillator

Mean 0.28e-06RMS 1.0e-06

Time of First Threshold Crossing (s)

10-8 10-7 10-6 10-5 10-4 10-3

Frac

tion

of E

vent

s / 0

.1 µ

s

10-6

10-5

10-4

10-3

10-2

10-1

1

Mean 0.26e-04RMS 1.1e-04

Boron-loaded Scintillator

Pure Scintillator

200 400 600 800 1000 1200 1400 1600 1800 2000

Fig. 3. The radial positions and times at which the first veto trigger associated with each neutron event was generated.The x-axis of (b) has been changed relative to Fig. 2

to emphasize the prompt veto events created by nuclear recoils during neutron thermalization: such a prompt veto is produced by approximately 90% of neutrons in both

pure and boron-loaded scintillator.

Table 2The radius required to contain the scintillator veto signal with different

probabilities.

Containment

probability (%)

Pure scintillator

radius (cm)

Loaded scintillator

radius (cm)

70 11.1 10.2

90 21.0 17.2

95 29.1 21.7

98 44.8 28.0

99 60.4 32.9

99.5 78.0 38.1

99.9 129.7 51.6

99.99 – 136.5

Table 3The time interval after neutron production (and hence any prompt recoils in the

WIMP detector) necessary to contain the veto signals with different probabilities.

Detection

efficiency (%)

Time in pure

scintillator (ms)

Time in loaded

scintillator (ms)

70 0.08 0.08

90 7.8 0.1

95 185 1.7

98 421 3.8

99 603 5.4

99.5 788 7.0

99.9 1282 10.9

99.99 – 22.0

A. Wright et al. / Nuclear Instruments and Methods in Physics Research A 644 (2011) 18–26 21

scintillator could also be achieved by loading the scintillator withother isotopes possessing large neutron capture cross-sections,among them 6Li, 113Cd, and 157Gd. We note that gadolinium-loaded scintillators have been successfully produced for reactorantineutrino experiments (see, e.g. [21,22]). Neutron capture on157Gd is detected through the emission of gamma rays, whichraises the prospect that, as with neutron captures by protons,

larger scintillator volumes would be necessary to contain thegamma rays. In our simulations, however, we find the neutroncapture performance of a veto composed of scintillator loaded to0.6% by weight with natural gadolinium offers very similarperformance, in both capture time and radius of energy deposi-tion, to the boron-loaded scintillator. The production of a cascadeof gamma rays, rather than a single photon, by neutron captures

A. Wright et al. / Nuclear Instruments and Methods in Physics Research A 644 (2011) 18–2622

on 157Gd seems to be responsible for the improved spatialperformance of gadolinium relative to pure scintillator. Gadoli-nium loading offers the advantage that the majority of neutroncaptures on 157Gd deposit more energy in the veto than dobackground gamma rays. In simulations including the innerWIMP detector from DarkSide-50 (described below), however,6% of neutron events deposited less than 2 MeVee in a vetocomposed of Gd-loaded scintillator and hence fell below theendpoint of the background from external gamma rays, while0.5% fell below 100 keV. Therefore, while gadolinium loading mayin fact be preferable for a veto of moderately high efficiency (as itcould be operated with less shielding), for very high-efficiencyvetoes there is little, if any, difference between gadolinium andboron loading in the energy threshold required for the veto, andhence no difference in the tolerable rate of external background.

Fig. 4. The simulated geometry of DarkSide-50 within the veto tank. The inner

argon volume, with photodetector arrays above and below, is housed within a

metal cryostat that is suspended in the center of the neutron veto. A dog-legged

feed through extends from the top of the cryostat through the neutron veto. The

stainless steel veto tank and the steel passive shielding surrounding it are

not shown.

Table 4The masses of the different components of the DarkSide-50 detector, as simulated.

Except for the depleted argon in the active region and the boron-loaded

scintillator, all components are passive absorbers of neutrons. The listed external

gamma-ray shield is 25 cm thick; as discussed in Section 6, in some underground

locations much less passive shielding would be required.

Component Material Mass (kg)

Active region Depleted argon 52.7

Inner vessel þ photodetectors Fused silica 25.4

Passive buffer Depleted argon 74.1

Cryostat þ inner mechanics Titanium 78.6

Neutron veto Boron-loaded liquid scintillator 1.15�104

g Shielding Steel 7.37�104

Table 5The time interval required after neutron production to

contain the veto signals with different probabilities. The

increase in the neutron capture time caused by the pre-

sence of the inner detector can be seen by comparing these

values with those in Table 3.

Detection efficiency (%) Time required (msÞ

70 0.08

90 0.37

95 2.3

98 5.5

99 9.3

99.5 21.5

99.8 57.7

5. A boron-loaded scintillator neutron veto

The discussion above suggests that a neutron veto efficiencygreater than 99.9% could be obtained using a roughly 50 cm thickboron-loaded liquid scintillator and an 11 ms time window. In actualoperation, however, the veto will surround a detector of finite massin which neutrons may capture and, therefore, not enter the veto.

This effect has been investigated extensively in the context ofthe DarkSide-50 experiment. DarkSide-50 is a proposed direct-detection dark matter experiment based on a 2-phase argon TPC,which will make use of argon depleted in 39Ar from undergroundsources [23], and the new ultra-low-background QUPID photondetectors [24]. The DarkSide-50 geometry will essentially consistof a cylindrical active volume with equal diameter and height,contained in a fused silica inner vessel, with arrays of QUPIDs onthe flat top and bottom faces. The vessel and QUPIDs areimmersed in a (passive) liquid argon buffer inside a titanium(or low background stainless steel) cryostat [25].

The simulations reported here made use of a simplified geometryin which the photodetectors were modeled as monolithic cylindersabove and below the inner detector. The neutron veto was modeledas a 1 m thick layer of boron-loaded liquid scintillator surroundingthe dewar. The veto was penetrated by a 10 cm diameter air-filledtrunk to allow communication with the inner detector.1 Thesimulated geometry is illustrated in Fig. 4, and the masses of thedifferent detector components are listed in Table 4.

It is important to note that in DarkSide-50, which wasexplicitly designed for operation in a neutron veto, the materialsused in the inner detector were chosen to minimize the numberof neutrons lost to captures by inactive components, and hence toincrease the veto efficiency. This is in contrast to dark matterdetector designs which attempt to mitigate internal cosmogenicneutrons using (passive) internal neutron absorbers. We havestudied both approaches, and have found that the increase in vetoefficiency associated with the removal of the passive absorberssignificantly outweighed the loss of the relatively modest neutronreduction they afforded.

5.1. Internal radiogenic neutrons

For the DarkSide-50 geometry described above, radiogenicneutrons generated in the inner detector, specifically in thephotosensors, produced veto events with an efficiency of99.7570.02%. The inefficiency due to neutron capture on theinner detector was small, not because neutrons were not captured

1 Tests showed that the trunk did not have a significant impact on the

efficiency of the veto.

by the inner detector (indeed, in this simulation about 21% ofprimary neutrons were captured by inner detector components),but because the majority of these inner detector captures resultedin the production of secondary particles, particularly gamma rays,which were subsequently detected by the veto.

The potential for neutrons to thermally diffuse within theinner detector, the components of which have lower neutroncapture cross-sections than the boron-loaded scintillator, meansthat the time distribution of neutron veto signals is longer withthe detector present (see Table 5) than without (Table 3), and isno longer described at longer times by a single exponential(Fig. 5). Neither the veto times nor the veto efficiency appear tovary significantly between those neutron events which depositedenergy in the active argon volume and those which did not.

Mean 0.50e-06RMS 3.9e-06

Time of First Threshold Crossing (s)

10-8 10-7 10-6 10-5 10-4 10-3

Frac

tion

of E

vent

s / 0

.1 µ

s

10-6

10-5

10-4

10-3

10-2

10-1

1

Mean 0.51e-06

RMS 4.4e-06

All Neutrons

>1keVee in Ar

Fig. 5. The simulated time distribution of veto events produced by radiogenic

neutrons from the inner detector. As can be seen, the distribution of veto times is

quite insensitive to whether or not the neutron deposited energy in the argon

active volume of the inner detector.

Table 6The mean position of the most distant nuclear recoil (‘‘MDR’’) produced by

neutrons of different energies (or any of their daughter particles). This metric

provides an ‘‘effective attenuation length’’ appropriate for use in estimating the

amount of neutron shielding necessary for direct-detection dark matter

experiments.

Neutron

energy (MeV)

MDR in

water (cm)

MDR in pure

scintillator (cm)

MDR in loaded

scintillator (cm)

10 22.2 26.3 27.5

20 27.4 30.8 32.5

50 54.8 60.5 62.7

100 92.0 98.5 102.9

200 145.7 168.2 170.4

Table 7The factor by which the rate of external cosmogenic

backgrounds is reduced by neutron vetoes of different

thicknesses.

Veto

thickness

(m)

Relative recoil rate in

DarkSide-50

0 1.0

1 (2.770.4)�10�2

2 (2.470.6)�10�3

3 (572)�10�4

A. Wright et al. / Nuclear Instruments and Methods in Physics Research A 644 (2011) 18–26 23

5.2. External radiogenic neutrons

In addition to its high detection efficiency for internallyproduced radiogenic neutrons, a boron-loaded scintillator vetois also an excellent shield against external radiogenic neutrons.Our simulations suggest that the fraction of external radiogenicneutrons that will penetrate the 1 m thick neutron vetoand produce a recoil in the active volume of DarkSide-50is t1� 10�7. It is expected that the residual external radiogenicevents will be vetoed with an efficiency at least equal to the vetoefficiency for internal radiogenic neutrons. Therefore, we expectan overall reduction in the rate of external radiogenic neutronbackgrounds of more than 109 relative to the rate in the detectorwith no neutron shielding. We have not, however, generatedsufficient Monte Carlo statistics to confirm this.

5.3. Cosmogenic neutrons

A liquid scintillator neutron veto has the potential to detectcosmogenic neutrons, rather than simply attenuating them. Thismeans that the mean free path (Table 1), rather than the attenuationlength (Table 6), gives the appropriate scale for estimating thereduction in external cosmogenic neutrons by a scintillator veto.2

This increases the effectiveness of the scintillator veto by approxi-mately a factor of two relative to water shielding. In addition, thescintillator veto has the potential to detect cosmogenic neutrons bothbefore and after they have interacted in the inner detector, asopposed to attenuation-based shielding which is effective on theincident particle only. This increases the relative efficiency of anactive veto by another factor of two relative to passive shielding. Inaggregate, therefore, 1 m of scintillator veto is about as effective as4 m of water shielding in reducing backgrounds due to externalcosmogenic neutrons.

In the DarkSide-50 based simulations, only primary cosmogenicneutrons with energies greater than 12 MeV were simulated;imposing this energy cut saved computation time, while providinga conservative estimate of the veto efficiency.3 In these simulations,

2 Note that the efficiency of the scintillator in vetoing those muons which pass

through the scintillator directly is sufficiently high that internal cosmogenic

backgrounds are strongly sub-dominant to the external cosmogenics, and as a

result internal cosmogenics are not considered here.3 From Table 1, neutrons with energies less than 12 MeV have mean free paths

of roughly 10 cm in scintillator. This means that the small number of lower energy

a 1 m thick scintillator veto reduced the number of recoil eventsproduced by cosmogenic neutrons in the active argon detector byabout a factor of 40 compared to the rate with no neutron shielding(the number of recoil events was reduced by � 25% from attenua-tion, with � 97% of the remainder being vetoed). We note that thisis an extremely conservative number, as by generating singlecosmogenic neutrons as the primary particles in our simulationswe neglected the possibility that other particles from the primarymuon shower could also enter the veto. We note also that reducingcosmogenic background by a factor of 40 is equivalent to increasingthe overburden of the experiment by more than 2 km of waterequivalent [15].

For interest, we have simulated the suppression of externalcosmogenic neutrons possible with thicker scintillator vetoes.As can be seen in Table 7, cosmogenic neutron suppressioncontinues to increase as the veto thickness increases.

6. Veto trigger rates and veto-induced dead time

The effective dead time produced by the veto in the innerdetector depends on the event rate in the veto and the veto timewindow. For a 60 ms veto window, which from Table 5 isnecessary to maintain a very high veto efficiency, a 1% dead timewould be produced by a veto rate of about 170 Hz, while a 10%dead time would be produced by a veto rate of 1800 Hz.

The background rate in the veto will be due to external back-grounds, backgrounds intrinsic to the scintillator itself, backgrounddecays in the PMTs, and backgrounds from the inner detector. Thelatter will presumably be negligible on the scale of 100 Hz, while theintrinsic rate in the scintillator will be dominated by 14C,4 which, at

(footnote continued)

neutrons which penetrate the veto to produce recoils in the dark matter detector

have less than about a 0.01% chance of doing so without depositing energy in the

veto.4 The reduction of U and Th backgrounds by distillation was demonstrated in

Ref. [11] to be as effective in TMB as in pseudocumene, so, as in Ref. [26], the rates

of these ‘‘other’’ backgrounds should be small compared to 14C.

Table 8The effective attenuation of 2.6 and 1.4 MeV gamma rays in layers of steel of

different thicknesses, as simulated using GEANT4. The attenuation factor is

calculated based on the number of events produced in the neutron veto above

the 40 keVee threshold.

Steel thickness

(cm)

Survival fraction

for 2.6 MeV g’s

Survival fraction

for 1.4 MeV g’s

0 1.0 1.0

5 0.27 0.18

10 0.061 0.027

15 0.013 3.9�10�3

20 3.0�10�3 5.9�10�4

25 6.4�10�4 7.5�10�5

30 1.8�10�4 1.4�10�5

35 1.7�10�5 1.8�10�6

Table 9The estimated rate of background events in the veto for a

‘‘typical’’ underground installation, with 25 cm of passive

steel shielding. This total veto rate corresponds to a dead

time of about 3% in the argon detector.

Background source Veto rate (Hz)

Inner detector o1

Scintillator background 3

PMTs 200

External backgrounds with

25 cm steel

150

Steel backgrounds 65

Random veto triggers (1 kHz

dark rate)

80

Total veto rate � 500

Energy of Veto Event (keVee)

102 103 104

Frac

tion

of E

vent

s / 1

0 ke

Vee

10-6

10-5

10-4

10-3

10-2

10-1

Fig. 6. The energy spectrum of veto triggers generated by radiogenic neutrons. In

each case the energy is deposited within a 1 ms window; if more than one veto

trigger was generated by a neutron event, only the earliest is plotted. The

DarkSide-50 inner detector was present in this simulation and gamma-ray lines

from neutron capture on the inner detector components, as well as energy

depositions from neutron scattering and capture within the veto itself can be

seen in the spectrum.

A. Wright et al. / Nuclear Instruments and Methods in Physics Research A 644 (2011) 18–2624

the 10�18 14C/12C ratio found in petrochemically derived scintillator[11], would give a veto rate of 2–3 Hz.

While the background rate from PMT activity and externalbackgrounds will depend on the construction and location of thedetector, it is possible to produce reasonable estimates of whatthese background rates could be. The total background decayrate in ‘‘low-background’’ 800 PMTs varies in the range of about1–6 Hz, depending on the manufacturer [27]. With 80 PMTs (seeSection 7), the background rate would thus likely be 100–500 Hz.With the PMTs mounted so that their faces are flush with theouter wall of the veto (as assumed in the optical simulations),perhaps half of this activity would be detected by the veto.

External backgrounds will be dominated by gamma raysproduced in the rock surrounding the experiment. A reasonableestimate for the activity in ‘‘typical’’ rock is 15 Bq/kg 238U þ 232Thactivity, and 300 Bq/kg 40K activity [28]; there is, of course,significant variation in these activities between different sites.These correspond to fluxes of � 2500 g=m2=s for the 2.2 and2.6 MeV U and Th chain gamma rays, and � 42,000 g=m2=s forthe 1.4 MeV gamma ray from K.

Taking DarkSide-50 as an example, the surface area of the vetowill be about 31 m2, so for these levels of external activity wewould expect � 80,000 Hz of incident 2.6 MeV photons,and � 1:3� 106 Hz of 1.4 MeV gammas.5 As these rates are toohigh for effective veto operation, the gamma rays must beattenuated. We have explored relatively inexpensive scrap steelfrom de-commissioned ships as a passive shield. The simulatedattenuation factors for 2.6 and 1.4 MeV gammas as a function ofsteel thickness are given in Table 8. For the rate of externalgamma rays described above, about 25 cm of steel would besufficient to reduce the rate of external gamma ray interactions inthe veto to � 150 Hz. It should be noted that at some under-ground sites with lower gamma-ray background rates, as little ashalf this passive shielding would be required.

An additional source of radioactive background would then bethe steel itself. Measurements of the ship-breaking steel indicatethat the K activity is less than 13 mBq/kg, while the U and Thactivity totals less than about 2.5 mBq/kg. 137Cs and 60Co arestrongly subdominant at o0:23 and o0:18 mBq=kg, respectively.At these rates (conservatively taking all U and Th decays toproduce 2.6 MeV gamma rays, and 137Cs and 60Co gamma raysto have the same attenuation lengths as 1.4 MeV gamma rays),the simulation predicts a rate of about 65 Hz in the veto fromactivity in the steel.

5 We actually plan to deploy DarkSide-50 within the CTF water tank, an 11 m

diameter�10 m high water tank at LNGS, which will significantly attenuate

external backgrounds. Here we discuss a solution which might be more generally

useful in other applications of a neutron veto.

A final source of veto-induced dead time in the inner detectorwill be random triggers produced by dark rate pile-up. The 80 800

PMTs that we assume will be used to instrument the veto canbe reasonably taken to have a 1 kHz dark rate [29]. Requiring acoincidence between at least three PMTs within a 1 ms vetotrigger window, and assuming that the dark hits are randomlydistributed in time, the rate of random veto triggers can becalculated to be 80 Hz.

Although the trigger rates induced in the veto by the differentsources discussed above will vary depending on the location ofthe detector and the construction of the veto, we have none-theless produced reasonable estimates for the rates in a ‘‘typical’’installation. As shown in Table 9, this ‘‘typical’’ rate would bequite acceptable, with the inner detector incurring a dead time ofabout 3% from the veto.

7. Veto optical efficiency

In order for the veto to achieve high neutron detectionefficiency, it is critical that the � 50 keVee energy deposition bythe recoil daughters after neutron capture on 10B be efficientlydetected. As can be seen in Fig. 6, while most events depositsignificantly more energy in the veto, approximately 2% of the

Number of PMTs Triggered

0

Frac

tion

of 4

0 ke

Vee

Eve

nts

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

Mean 7.7RMS 2.6

Number of Photo-Electrons Collected

0

Frac

tion

of 4

0 ke

Vee

Eve

nts

0

0.02

0.04

0.06

0.08

0.1

0.12 Mean 8.6RMS 3.2

2 4 6 8 10 12 14 16 18 20 2 4 6 8 10 12 14 16 18 20

Fig. 7. The distributions of (a) the number of PMTs which detected at least one photoelectron and (b) the total number of collected photoelectrons in simulations of

40 keVee events uniformly distributed through the neutron veto.

A. Wright et al. / Nuclear Instruments and Methods in Physics Research A 644 (2011) 18–26 25

observed neutron captures deposit only the nuclear recoil energy;failure to detect these recoils would thus limit the efficiency ofthe veto to 98%. The 40 keVee veto threshold assumed in theearlier analysis was chosen so as to be certain that these recoilevents are reliably detected. We must now demonstrate that thislow threshold is practically achievable.

The chief concern in detecting such low energy events, ofcourse, is the collection of a sufficient number of photons fromeach event to identify the neutron capture. We have performed astudy of the optical efficiency of the scintillator veto instrumen-ted with 80 standard 800 PMTs evenly distributed over the outer(curved) side of the veto cylinder, with no PMTs on the top andbottom. A rather crude optical simulation, independent of theGEANT4-based physics simulation, was generated to study thelight collection efficiency of the veto in such a configuration. Themodel assumed:

1.

A 20% quantum efficiency for the 800 PMTs. Photons that hit aPMT and were not detected were assumed to have a 20%probability of being reflected back into the veto.

2.

A 95% reflection probability on all surfaces other than thePMTs (i.e. the inside walls of the veto tank and the outside ofthe DarkSide-50 cryostat). No significant difference was seenbetween specular reflection (used in the results reported here)and diffuse reflection.

3.

An average scintillator light output of 6000 photons/MeVee.The statistical fluctuation in the light output was assumed tobe Gaussian, with a width of

ffiffiffiffi

Np

.

4. A 5 m optical absorption length in the scintillator, with no

probability of re-emission, and a 2 m optical scattering length.

Each of these assumptions is extremely conservative:

1.

800 PMTs with quantum efficiencies 430% are currentlycommercially available.

2.

Both Spectralon PTFE and 3M Vikuiti foils, which are commer-cially available, have reflectances in excess of 95% for therelevant wavelengths.

3.

6 For example, the KamLAND scintillator, which consists of pseudocumene

diluted to 20% in dodecane, has a light output � 80% that of pure PC [31].

Carefully prepared pseudocumene scintillator has a typicallight output of � 12,000 photons=MeVee [30], and we haveassumed that light output scales with TMB dilution. It isknown, however, that the light output of scintillator with anoptically inert dilutant, like TMB, typically decreases moreslowly than the concentration.6 The 80% (rather than 50%)

TMB-loaded scintillator with a light output of 6000 photons/MeV was reported in Ref. [11].

4.

Scattering and absorption lengths in TMB-loaded scintillatorgreater than 10 m are reported in Ref. [11].

Fig. 7 shows the simulated distributions of the total number ofcollected photoelectrons and the number of PMTs which detectedat least one photoelectron when events of exactly 40 keVee weresimulated uniformly throughout the veto volume. As discussed inSection 6, in order to maintain a reasonable rate of randomtriggers, a coincidence of three PMT hits was required to triggera veto event. As can be seen, even with the very conservativeassumptions about the optical performance of the veto discussedabove, about 98% of 40 keVee events produced a veto trigger.Given that events below 100 keVee constitute less than 5% of thetotal, the overall fraction of neutron-induced backgrounds misseddue to veto trigger inefficiency can, therefore, be expected to besignificantly less than 0.1%.

8. Conclusions

Neutron vetoes with very high efficiencies can be produced fordirect detection dark matter experiments by surrounding the WIMPdetector with a layer of liquid scintillator. Even after consideringneutron loss in the inner detector and veto inefficiency due tofeedthrough connections, our simulations suggest that a 1 m thickveto can provide greater than a 99.5% efficiency for rejectingbackground events due to internal radiogenic neutrons, whilereducing the background from external cosmogenic neutrons bymore than 95%. Loading 10B, or another isotope with a high neutroncapture cross-section, into the scintillator makes the veto practicalby reducing the neutron capture time enough that conventionalPMTs can be used in the veto and typical underground backgroundrates can be tolerated with relatively little external shielding.

Acknowledgments

The authors would like to thank the other members ofthe DarkSide collaboration and the members of the CDMS group

A. Wright et al. / Nuclear Instruments and Methods in Physics Research A 644 (2011) 18–2626

at FNAL for interesting and productive discussions aboutthe practical aspects of the deployment of liquid scintillatorneutron vetoes in dark matter experiments. This work wassupported by NSF Grants 0802646 and 0704220, DOE GrantDE-FG0291ER40671, and the Princeton University PFEP program.

References

[1] M.G. Marino, et al., Nucl. Instr. and Meth. A 582 (2007) 611.[2] E. Aprile, et al., Phys. Rev. Lett. 105 (2010) 131302.[3] A. Ahmed, et al., Science 327 (2010) 1602.[4] R. Acciarri, et al., J. Phys. Conf. Ser. 203 (2010) 012006.[5] D.Y. Akimov, et al., Astropart. Phys. 34 (2010) 151.[6] C. Muehlhause, G. Thomas, Phys. Rev. 85 (1952) 926.[7] L. Bollinger, G. Thomas, Rev. Sci. Instr. 28 (1957) 489.[8] M. Chadwick, et al., Nucl. Data Sheets 107 (2006) 2931 Accessed via /http://

t2.lanl.gov/ and w?w?w?.nndc.bnl.govS.[9] L. Greenwood, N. Chellew, Rev. Sci. Instr. 50 (4) (1979) 466.

[10] S. Wang, et al., Nucl. Instr. and Meth. A 432 (1999) 111.[11] G. Bellini, et al., Borexino at Gran Sasso. Proposal for a Real Time Detector for

Low Energy Solar Neutrinos, vol. 1, 1991.[12] C.L. Cowan, et al., Phys. Rev. 90 (1953) 493.[13] S. Agostinelli, et al., Nucl. Instr. and Meth. A 506 (2003) 250.[14] W. Wilson, et al., Prog. Nucl. Energy 51 (2009) 608.

[15] D. Mei, A. Hime, Phys. Rev. D 73 (2006) 053004.[16] M. Pallavicini, Recent Results from Borexino, 2010. Presentation at

NOW2010. Available /www.ba.infn.it/�now/now2010/S.[17] K. Shibata, et al., J. Nucl. Sci. Technol. 39 (2002) 1125 Accessed via /http://

www.ndc.jaea.go.jp/jendl/j33/j33.htmlS.[18] J. Hong, et al., Astropart. Phys. 16 (2002) 333.[19] F. Arneodo, et al., Nucl. Instr. and Meth. A 418 (1998) 285.[20] M. Steuer, B. Wenzel, Nucl. Instr. and Meth. 33 (1965) 131.[21] M. Apollonio, et al., Eur. Phys. J. C 27 (2003) 331.[22] F. Boehm, et al., Phys. Rev. Lett. 84 (2000) 3764.[23] D. Acosta-Kane, et al., Nucl. Instr. and Meth. A 587 (2008) 46.[24] A. Fukasawa, et al., Nucl. Instr. and Meth. A 623 (2010) 270.[25] D. Alton, et al., DarkSide: A Depleted Argon Dark Matter Search, 2010.

Available /http://www.fnal.gov/directorate/program_planning/Nov2009PAC

Public/DarkSideProposal.pdfS.[26] C. Arpesella, et al., Phys. Rev. Lett. 101 (2008) 091302.[27] C. Arpesella, et al., Astropart. Phys. 18 (2002) 1.[28] P. Catalano, et al., Mem. Soc. Geol. It. 35 (1986) 647.[29] P. Lombardi, The Borexino Detector: Photomultiplier System, 2008. Poster

presented at Neutrino 08, Christchurch, New Zealand. Available /www2.phys.canterbury.ac.nz.� jaa53/S.

[30] F. Elisei, et al., Nucl. Instr. and Meth. A 400 (1997) 53.[31] F. Suekane, et al., in: KEK Proceedings: KEK-RCNP International School and

Mini workshop for Scintillating Crystals and their Applications in Particle and

Nuclear Physics, vol. 4, pp. 279–290. An anthracene light output of16,600 photons/MeV was assumed in interpreting this result.