Radioactivity Backgrounds in ZEPLIN-III
H. M. Araujo, D. Yu. Akimov, E. J. Barnes, V. A. Belov, A. Bewick, A. A. Burenkov, V. Chepel. A. Currie, L. DeViveiros, B. Edwards, C. Ghag, A. Hollingsworth, M. Horn, G. E. Kalmus, A. S. Kobyakin, A. G. Kovalenko, V. N. Lebedenko, A. Lindote, M. I. Lopes, R. Luscher, P. Majewski, A. StJ. Murphy. F. Neves, S. M. Paling, J. Pinto da Cunha, R. Preece, J. J. Quenby, L. Reichhart, P. R. Scovell, C. Silva, V. N. Solovov, N. J. T. Smith, P. F. Smith, V. N. Stekhanov, T. J. Sumner, C. Thorne, R. J. Walker
aa r X i v : . [ phy s i c s . i n s - d e t ] A ug Radioactivity backgrounds in ZEPLIN–III
H.M. Ara´ujo a, ∗ , D.Yu. Akimov b , E.J. Barnes c , V.A. Belov b , A. Bewick a , A.A. Burenkov b , V. Chepel d ,A. Currie a , L. DeViveiros d , B. Edwards e , C. Ghag c , A. Hollingsworth c , M. Horn a , G.E. Kalmus e ,A.S. Kobyakin b , A.G. Kovalenko b , V.N. Lebedenko a , A. Lindote d,e , M.I. Lopes d , R. L¨uscher e ,P. Majewski e , A.St J. Murphy c , F. Neves d,a , S.M. Paling e , J. Pinto da Cunha d , R. Preece e , J.J. Quenby a ,L. Reichhart c , P.R. Scovell c , C. Silva d , V.N. Solovov d , N.J.T. Smith e , P.F. Smith e , V.N. Stekhanov b ,T.J. Sumner a , C. Thorne a , R.J. Walker a a High Energy Physics group, Blackett Laboratory, Imperial College London, UK b Institute for Theoretical and Experimental Physics, Moscow, Russia c School of Physics & Astronomy, SUPA University of Edinburgh, UK d LIP–Coimbra & Department of Physics of the University of Coimbra, Portugal e Particle Physics Department, STFC Rutherford Appleton Laboratory, Chilton, UK
Abstract
We examine electron and nuclear recoil backgrounds from radioactivity in the ZEPLIN-III dark mat-ter experiment at Boulby. The rate of low-energy electron recoils in the liquid xenon WIMP target is0.75 ± ± β activity internal to photomultipliers, which can increase the size andlower the apparent time constant of the scintillation response. Another challenge is posed by multiple-scatter γ -rays with one or more vertices in regions that yield no ionisation. If the discrimination power achievedin the first run can be replicated, ZEPLIN-III should reach a sensitivity of ∼ × − pb · year to the scalarWIMP-nucleon elastic cross-section, as originally conceived. Keywords:
Liquid xenon detectors, dark matter searches, ZEPLIN-III, radioactivity, background studies
1. Introduction
The ZEPLIN-III direct dark matter search hasbeen operating at the Boulby underground labora-tory (UK) under a rock overburden of 2,800 m wa-ter equivalent. The WIMP target is a xenon emis-sion detector [1] which discriminates between elec-tron and nuclear recoils by measuring the relativeamount of prompt scintillation light and ionisationcharge extracted from particle interactions in theliquid xenon (LXe) phase; the ionisation is trans- ∗ Corresponding author
Email address: [email protected] (H.M. Ara´ujo) duced into a second optical signal via electrolumi-nescence in the thin vapour layer above the liquid.Both the primary (S1) and secondary (S2) scintilla-tion pulses are detected by an array of 31 photomul-tipliers immersed in the liquid. Key components ofthe experiment are shown in Figure 1; for details onthe design and construction of the system we referthe reader to Refs. [2, 3, 4, 5].After deployment underground in early 2007, an83-day run produced competitive limits for WIMP-nucleon scattering cross-sections [6, 7, 8]. In thisfirst phase the background of the experiment wasdominated by the photomultiplier array by a signif-icant factor – both in electron recoils from γ -raysand in nuclear recoils from neutrons. A develop- Preprint submitted to Astroparticle Physics September 17, 2018
998 7 65 634215
Figure 1: Schematic diagram of the ZEPLIN-III experiment.1: WIMP target (LXe shown in blue); 2: photomultiplierarray; 3: xenon chamber; 4: vacuum cryostat; 5: Gd-loaded polypropylene; 6: veto scintillator modules; 7-barepolypropylene; 8-copper flooring; 9-lead shielding. ment programme was thus undertaken with indus-try in order to develop new phototubes targeting anorder of magnitude reduction in radioactivity. Afterupgrade of the photomultipliers and the addition ofa veto system around the main target [4, 5], the ex-periment returned to taking WIMP search data inmid 2010. The first aim of this article is to describethe radioactivity backgrounds which determine thesensitivity of the rare event search ahead of forth-coming results from the second run.We highlight also that many background stud-ies focus solely on simple event topologies quan-tifiable by Monte Carlo simulation, but often over-look other important factors. Although the require-ment of negligible neutron background drives thedesign of most dark matter experiments, in factthis threat materialises only rarely. In two-phasexenon detectors, multiple scattering γ -rays can bechallenging when one of the vertices occurs in a re-gion which yields no ionisation; in this instance theprompt scintillation signals for the multiple verticesare time-coincident and difficult to distinguish, andthe charge-to-light ratio (S2/S1) is low, more typi-cal of nuclear recoils. In another example, when es-timating photomultiplier backgrounds, many stud-ies consider only high-energy γ -rays as contribut- ing significantly to electron recoil background; wepresent evidence here that β -induced signals gener-ated internally (involving, for example, Cherenkovemission from the phototube envelopes) should alsobe assessed carefully.Besides their sensitivity in the scintillation andionisation channels (see, e.g., Ref. [9] for an em-phatic demonstration of the latter), two-phasexenon detectors hold great promise for next gen-eration dark matter searches due to the ability toself-shield against external backgrounds. The accu-rate reconstruction of the position of interactionsin three dimensions allows a sacrificial LXe vol-ume to be defined which shields an inner ‘fiducial’mass from most sources of radioactivity. ZEPLIN-II was the first WIMP experiment to operate usingthis technology [10], soon followed by XENON10at Gran Sasso [11] and ZEPLIN-III at Boulby [6].XENON100 [12] already benefits from self-shieldingvery significantly, and even more so will the up-coming LUX350 experiment at SUSEL [13]. Two-phase xenon experiments have been at the forefrontof WIMP sensitivity and systems with tonne-scalefiducial mass could probe most of the parameterspace favoured by constrained minimal supersym-metry (cMSSM) and by other extensions to thestandard model. With this work we wish to sum-marise a decade of experience in designing, buildingand operating ZEPLIN-III, and thus inform the de-sign of next-generation WIMP experiments.This article is organised as follows. Ordi-nary, single-vertex electron recoil backgrounds aretreated in Section 2, where ZEPLIN-III data areconfronted with an overall prediction built up fromcomponent-level simulations. In Section 3 we dis-cuss event topologies that may degrade the discrim-ination power of these instruments, such as γ -raymultiple scatters and β − background internal to thephotomultipliers. Having validated U/Th contami-nation levels in key components, we present calcula-tions of neutron background in Section 4. Finally,we summarise our findings in Section 5, with thedesign of future systems in mind.
2. Single-vertex electron recoil backgrounds
Most low-energy electron recoils in the LXe re-sult from interactions of γ -rays and β -particles fromnatural radioactivity. We begin by calculating ab-solute spectra of deposited energy from dominantsources, and compare these with ZEPLIN-III dataacquired at Boulby in fully-shielded configuration.2hese signals can be discriminated from nuclear re-coils to a large degree; it is reasonable to expectthat a ‘leakage’ (mis-identification probability) oforder 1:10,000 may be achieved for single-site inter-actions such as those from Kr β -particles withinthe LXe (the value eventually achieved in the firstrun for all interactions was 1:7,800 [14]). However,more complicated topologies must be considered,and these may not be rejected so efficiently. Weanalyse two such types in Section 3.The γ -ray background is dominated by primor-dial U, Th and K. We assume natural uranium(0.956
U, 0.044
U) and that all chains are insecular equilibrium, except where indicated other-wise. To determine contamination levels, a largenumber of materials were radio-assayed over thelifetime of the project. All HPGe γ -ray spec-troscopy was conducted at Boulby with an OR-TEC GEM detector with a 2 kg crystal, reach-ing down to ∼ γ -rays in each chain and the1,461 keV line from K are propagated using theGEANT4 Monte Carlo toolkit [16] to interactionsin the LXe target. For internal backgrounds theZEPLIN-III simulation model was used [2]; this wascomplemented by importing a detailed CAD solidmodel for calculations involving the veto, shieldingand the laboratory rock.
A 24-day long dataset was analysed to study thespatial and energy distribution of electron recoilbackground. This followed a neutron calibrationby only a few weeks, and so isomeric transition(IT) γ -rays from activation of the xenon are stillvisible (mainly from Xe, T / =8.88 days, and Xe, T / =11.8 days). The data were collectedas part of the WIMP search dataset and, for thisanalysis, the nuclear recoil signal region remainedblinded up to 40 keVee. However, this has little im-pact on the electron recoil population. Events with LXe energy deposits are expressed in electron-equivalent‘keVee’ units when calibrated by Co 122 keV γ -rays, or in‘keVnr’ when referring to nuclear recoil energy. -1
110 0 50 100 150 200 250 300 350 400energy, keVee d i ff e r e n ti a l r a t e , ev t/ kg / da y / k e V ee ( d r u ) datasimulationphototubespolypropyleneveto plasticfeedthroughs U Th Xe d i ff e r e n ti a l r a t e , d r u datasimulationphototubespolypropyleneveto plasticfeedthroughs00.511.522.53 0 5 10 15 20 25 30 35 40depth, mm eve n t r a t e , d r u datasimulationphototubespolypropyleneveto plasticfeedthroughs Figure 2: Electron recoil background in ZEPLIN-III. Dataare shown in yellow and simulated components are summedby the solid blue line, with individual contributions orderedby importance at low energy. Top: energy spectrum in fidu-cial volume, showing good agreement below ∼
150 keVee,the dynamic range of the dataset (the thin black histogramretains only unsaturated events); U/Th and Xe activation γ -rays are indicated. Middle: radial distribution below100 keVee, confirming very good agreement beyond the cen-tral channel. Bottom: depth distribution for fiducial region(bright yellow) and full depth (hatched yellow); contami-nation of the cathode grid (37 mm) with radon progeny leftover from the first run is clearly visible (but easily fiducialisedaway). ∼
150 keVee in the WIMP-search dataset.The measured energy and spatial distributionsare shown in Figure 2 – along with a component-level breakdown from simulated data calculatedas described below. These contributions are alsolisted in Table I. The measured low-energy back-ground is 0.75 ± ∼
150 keV and even beyond, despite the xenon acti-vation which is still prominent in this dataset. Thisrepresents a ∼ ± ∼ Pb in some detector sur-faces. Notwithstanding, both energy and spatialdistributions of electron recoils are well reproducedby the simulation within the fiducial volume.
Although other primordial decay chains also in-clude radon isotopes,
Rn (
U chain) is thelongest lived (T / =3.824 days). The issue ofgetter emanation had been previously identifiedin ZEPLIN-II [10], which relied on continuousgas recirculation through the same getter model(SAES PS11-MC500). Upon interruption of purifi-cation, the decay of the α activity was measured atT / =3.83 ± Rn contam-ination [17]. By then the same getters had beenused to purify ZEPLIN-III xenon before the firstrun, which led to internal plating with
Pb. Thisdid not pose a problem then, and we opted not toetch it away during the upgrade; we replaced thegetter with a new model (SAES PS4-MT3). Theeffect of internal radon decay can be understoodwith reference to the chain sequence:
Rn ( α, . → Po ( α, . → Pb ( β − , → Bi ( β − , → Po ( α, µ s) → Pb ( β − , ) → Bi ( β − , → Po ( α, → Pb (stable) (1)
Once washed into the target the radon mixes uni-formly in the LXe. The first two α decays producenegatively charged atoms which promptly lose ex-cess electrons (the α -particle drifts to the cathodewhere it reduces to He). The next two β − decayscreate Bi + and Po + ions which cannot trans-fer their ionisation state to the rare gas and thusdrift to the cathode. Finally, the α decay of Poimplants
Pb (T / =22.3 yr) within the wire grid,which bottlenecks the chain. This Pb activity isstill present on the cathode in the second run andhas remained approximately stable at 19.4 ± Pb,
Bi and
Po, as measured by the rate of
Ponuclear recoils).The electron-recoil background arising in thefiducial volume from this contamination is insignifi-cant in the context of the overall radioactivity bud-get (and was not included in the simulation); thesame is true of nuclear recoils from the ( α, n ) reac-tion from
Po on the stainless steel wires. In fact,these populations can be useful as stable sourcesof ionisation; they allow, for example, the ionisa-tion yield for the deepest events to be monitoredas a function of time. The Bateman decay equa-tions for the
Pb sub-chain indicate that the ra-tios of
Pb,
Bi and
Po become constant af-ter ∼ able 1: Component-level analysis of electron recoil background and nuclear recoils from neutron elastic scattering inZEPLIN-III. The sum of simulated contributions (excluding components without measured contamination, indicatedin brackets) is labelled ‘SSR total’. The additional rejection efficiency provided by the veto is indicated: ‘ptag’ forprompt tags, which remove γ -rays, and ‘dtag’ for delayed tags, which reject neutrons [5] (weighted averages areindicated under ‘SSR total’; these agree with the measured tagging efficiencies). After veto tagging, the neutronevent expectation in a 1 year long dataset with typical signal acceptance is some 10 times lower ( ≃ Material mass, kg e-recoil, dru † ptag n/year ‡ dtagKrypton-85 – 0.007 ∼ ∼ ∼ ∼ ∼ ∼
400 ( < < ∼ ∼ ∼ ∼ ∼ ± ± ± ± ± ∗ – ) † events/kg/day/keVee at 10 keVee ‡ single scatters in 2,370 kg · days over 5–50 keVnr (unity detection efficiency) ∗ FSR dataset expectation was 1.2 ± · days )elapsed between the start of the two runs); whenthis so-called transient equilibrium is reached, therelative spectrum of decay products becomes con-stant, and so does the mean ionisation yield perevent. Monitoring the S2/S1 ratio for these low-energy events allows, for example, the accuracy ofthe electron lifetime correction to be checked. γ -rays The original phototubes (2-inch ETEL D730Q)were purchased a decade ago and had relativelyhigh contamination levels (250 ppb U, 290 ppb Thand 1350 ppm K), releasing 1,400 mBq per unitin γ -ray activity. This dominated by far the back-ground budget in the first run (14.5 dru). For thisreason a second run of the experiment with up-graded devices was planned from the outset. Alow-background product (D766Q) was developedby ETEL in collaboration with the project to en-able pin-by-pin compatibility with the existing ar-ray. All materials and sub-components were radio-assayed with HPGe and/or ICP-MS. The best pho-totubes have 35 mBq in γ -rays, which was achievedthrough a complete redesign of the device. Most ofthis activity is concentrated at the rear of the en-velope, away from the fiducial volume. Simulations of background contribution to ZEPLIN-III take thespatial distribution of activity into consideration.A value of 0.40 dru is predicted from internal γ -rays, which represents a 40-fold improvement overthe previous array and makes this component onlymarginally dominant. The equivalent simulationsfor the first run reproduced data well, which givesadded confidence in this calculation. Backgroundsarising from the β − activity internal to the pho-totubes, which are often neglected in this type ofstudy, are discussed in Section 3.1. Several LXe-intrinsic backgrounds can affect thisand similar experiments.
Pb from
U and
Rn chains and
Ra from the
Th chain, forexample, undergo low-energy β − decay occasionallylacking significant x- and γ -rays to help flag theseevents (‘naked betas’). Tritium (T / =12.3 yr, β max =19 keV) is produced cosmogenically and, onsome accounts, it could generate as much as ∼ Kr,long recognised as a serious background for WIMPsearches. Kr is an anthropogenic β − emitter(T / =10.76 yr, β max =687 keV); its presence inthe atmosphere is mainly due to the reprocessingof spent nuclear fuel. The pre-nuclear Kr/Kr ra-tio was as low as 3 × − in the early 1950s, butthe present-day ratio is ∼ × − [19].The xenon used in ZEPLIN-III was sourced inthe 1970s from an underground origin. An accu-rate measurement of its Kr content was obtainedin 1997 from a 5.9 g sample [20], from which we de-rive an age-corrected Kr/Kr ratio of 1.5 × − .The overall Kr content was subsequently reducedby cryogenic distillation to ∼
50 ppb Kr (by weight),corresponding to 0.2 dru at low energy in the xenontarget. This was, however, a conservative estimateand the present data suggest much lower contami-nation, which can be confirmed independently.A minor β − decay (0.434%, β max =173 keV) is ac-companied by a 514 keV γ -ray from the Rb 9/2 + level (T / =1.015 µ s) [21]. This allows a β − γ de-layed coincidence analysis, with the γ -ray searchedin the veto scintillator or in the LXe target. A 6-month dataset revealed ∼ × −
21 85
Kr/Xe ratio(in present-day xenon this would represent a verylow 150 ppt Kr). The decay rate is 0.007 ± Along with the photomultipliers, two materialscould make up a significant fraction of the radioac-tivity budget of the experiment from the outset: the ∼
400 kg of copper (from which all structural ele-ments, vessels, fastening hardware parts, etc., weremanufactured) and ceramic feedthroughs. Otheritems were screened and their locations chosen tominimise their impact: stainless steel parts (e.g.vacuum ports), indium seals, LN in the 30-litrereservoir.For copper (OFHC C103), conservative upperlimits of 0.5 ppb for U/Th contamination were ini-tially adopted, although lower values existed in ourdatabase for similar material [15]. Simulation ofthis contribution indicated an electron recoil back-ground of 0.35 dru at low energy. However, themeasured energy and spatial distributions show no evidence for such a significant rate, and we revisedthe upper limit down to 0.1 dru for the primordialradioactivity.Cosmogenic activation also produces severalradio-isotopes, of which Co (T / =5.3 yr) willretain the highest activity a few years after ex-posure to atmospheric muons ceases. We adopta production rate for Co in natural copper of ∼
50 atoms/kg/day [18]. Since the exact exposuretime on the surface is uncertain we assume secularequilibrium, with a subsequent cool-off period of4 years (preceding the second run after deploymentunderground). Simulation of the resulting γ -ray ac-tivity within the copperware produces a negligible0.01 dru in the fiducial volume.There are nearly 100 ceramic feedthroughs forsignal and high voltage connections, weighing 900 gin total; these are distributed among the xenonchamber baseplate and the vacuum vessel flange.They are alumina based and their average contam-ination was measured at 105 ppb U, 270 ppb Thand 880 ppm K. The γ -ray background is not severe(0.08 dru), but they dominate the neutron back-ground in the detector as discussed later. The ZEPLIN-III veto consists of a tight-fittingpolypropylene shield enveloping the main instru-ment, itself surrounded by 52 plastic scintilla-tor modules forming barrel and roof sections.The gadolinium-loaded hydrocarbon moderatesand captures internal neutrons very efficiently; theGd capture γ -rays are detected by the scintillators.The neutron detection efficiency (delayed tagging ina 70 µ s window) is 58%; internal γ -rays are detectedwith an average 28% efficiency (prompt tagging in ± µ s window). The prompt tag adds a further2% to the neutron efficiency. A detailed discussionof the hardware and physics performance is givenelsewhere [4, 5]; there, we analyse the backgroundscontributed to the WIMP target (a prime designconsideration) as well as background measurementsin the veto detector itself. A tonne of scintillatorwithin the lead castle provides confirmation of thelocal radiation environment and we find that theveto measurements agree with the background mea-sured within ZEPLIN-III itself.The readout (photomultipliers, electronics, ca-bling, etc.) are external to the hydrocarbon shield-ing to mitigate neutrons, which is achieved very suc-cessfully. These also have a negligible contribution6o the γ -ray background in the LXe target. How-ever, owing to their large mass, the polypropyleneand the plastic scintillator do contribute apprecia-bly to the ZEPLIN-III budget.Contamination levelshave been measured for the polypropylene (1 ppbU, 1 ppb Th and 5 ppm K) and the plastic scin-tillator (0.2 ppb U, 0.1 ppb Th and 0.2 ppm K).These values are close to the sensitivity of the HPGeand ICP-MS measurements of those samples, butthis radiological contamination is entirely consis-tent with the interaction rate recorded in the vetoitself. Furthermore, these have distinct energy andspatial distributions in ZEPLIN-III, as shown inFigure 2, which confirm these contributions: the en-ergy spectra are generally softer and they increasethe event rate especially at the top of the xenon tar-get, in contrast with other dominant components,such as the phototubes and ceramics, which are lo-cated below the xenon volume.We note that radioactivity in the plastic scintil-lator does not contribute to the overall backgroundsince the Q -value of the radioactive decay, most ofwhich is deposited locally, is nearly always suffi-cient to trigger the veto (even if the highest energy γ -rays are able to escape). On the other hand, thepolypropylene generates 0.25 dru in the xenon tar-get and has a low veto probability of 4%. γ -rays inside lead castle The γ -ray flux inside the empty ZEPLIN-III cas-tle includes external contributions from the labo-ratory walls, natural as well as cosmogenic activ-ity in the shield, air-borne radon and its metallicprogeny plated out on the inner castle walls. γ -rays from Boulby rock (halite) are attenuated by afactor of ∼ by the 20-cm thick lead shield [22]and are insignificant as a background in the WIMPtarget. The activity of the Rn chain measuredwith a commercial monitor inside the empty castlewas 2.4 Bq/m , which is typical at Boulby. Simu-lation of γ -rays from this activity leads to 0.03 druin ZEPLIN-III. Further in situ measurements witha HPGe detector place an upper limit on the back-ground from the castle (primordial and cosmogenicactivity in the bulk shield and inner walls). Thecastle walls and roof are made from lead smeltedinto thin stainless steel containers to make 1.2-tonne chevron-shaped blocks; the lead had beenunderground for some two decades and is knownto have low Pb content (it was used to shieldprevious dark matter experiments at Boulby). Theshield flooring is OFHC copper, which was moved underground over a decade ago as ‘new copper’.The HPGe spectra show only a vestigial amountof Co. The combined contribution of the shieldto ZEPLIN-III is ∼
3. Rarer types of electron recoil event β - and γ -induced photomultiplier backgrounds A significant amount of potassium is involvedin sensitising internal photomultiplier surfaces,namely the bialkali photocathodes. The K thuslocated close to the fiducial volume can generate aresponse in the detector in several ways, with the1,461 keV EC γ -rays being perhaps the most ob-vious. However, the 89% branching ratio β − de-cay ( β max =1,311 keV) can be as problematic (thisrepresents 30 mBq in the D766Q). VUV-sensitivephototubes have thin quartz windows ( ∼ β -particles to reach the LXe.Others will generate bremsstrahlung photons in thequartz. Cherenkov photons are also produced in thequartz and can be detected in coincidence with theprompt scintillation. Other luminescence processesarise internally in the phototubes, such as ‘dynodeglow’ from electron impact, which is signal coinci-dent. Besides contributing some 10–20% to the de-tector trigger rate, these β -induced events can leadto backgrounds which are harder to discriminatethan single scatters in the fiducial volume. Thesemight combine, for example, bremsstrahlung pho-tons interacting in the main volume with Cherenkovlight from the quartz, thus decreasing the observedS2/S1 ratio of the fiducial interaction. Cherenkovemission from Compton electrons created by γ -raybackgrounds may be even more damaging as fewerphotoelectrons may be involved in this instance.Figure 3 illustrates some of these processes; itshows Monte Carlo data for K decays generatedinternally in the central phototube in ZEPLIN-III.The quartz windows are curved and < γ -rays). Transmitted β -particles andbremsstrahlung in the quartz window clearly dom-inate at low energies for interactions just above thephototube. In the fiducial volume, some 10 mmaway, these contributions become negligible.As mentioned above, Cherenkov light generatedinside the quartz may also contribute to S1, yield-ing potentially tens of photoelectrons. For exam-ple, a 1 MeV electron – e.g. a K β -particle or a7 -1
110 0 200 400 600 800 1000 1200 1400energy deposited, keVee f i d u c i a l r a t e , ev t s / kg / da y / k e V ee g -raystransmitted b -particlesbremsstrahung g -rays (quartz) g -rays in fiducial Figure 3: Energy spectrum in LXe from simulated K de-cays internal to central photomultiplier; curves are scaled( ×
31) for correct fiducial rate from this contribution. Thesolid histograms are cumulative and represent interactionsabove the phototube window; these include bremsstrahlungphotons generated in the quartz, transmitted β -particles andnon-bremsstrahlung γ -rays. The spectrum above the cath-ode is shown in red, mostly due to 1,461 keV γ -rays. Compton electron from the 1,461 MeV γ -ray – pro-duces ∼
100 photons/mm of quartz within the spec-tral response of these photomultipliers; some gen-erate photoelectrons even without leaving the win-dow, while others suffer total internal reflection atthe liquid surface and strike the array again; there-fore, the photoelectron yield can be significantlyhigher than suggested by the nominal QE. This op-tical signal can both increase the size of a time-coincident scintillation pulse as well as reduce itsapparent time constant, threatening single-phase(scintillation only) as well as two-phase detectors.We see evidence of Cherenkov emission in ZEPLIN-III. This is illustrated in Figure 4, where the pulsetiming parameter τ , which measures the mean ar-rival time of the S1 signal, is shown for fiducial scin-tillation pulses and for sub-cathode events (no S2)with the same range of apparent energies (30–60keVee). The latter are mainly due to β -particlestransmitted through the phototube windows – buta faster population is also clearly seen which weattribute to Cherenkov emission. We can rule outan explanation based on nuclear recoils on the un-derside of the cathode grid, which would indeed befaster than electron recoils and yield no ionisation,by comparing with the corresponding timing spec-trum for the upper grid surface. We conclude that pulses in this population are indeed faster than nu-clear recoil scintillation. More extreme examplescan be identified if signals triggering in a singlechannel are included (such as the blue pulse shownin inset), but in this instance it is harder to provean optical, rather than electrical, origin.Such Cherenkov processes can be dangerous if as-sociated with low-energy electron recoils in the fidu-cial volume. Besides Compton electrons generatedby background γ -rays, this could involve, more gen-erally, β − decays with coincident γ -rays (many ex-amples exist in the U and Th chains). In ZEPLIN-III, the timing distribution of fiducial events withhigher than average scintillation (lower S2/S1 ra-tio) at energies above 30 keVee does not correlatewith shorter than average pulses; rather, it is fullycompatible with that of electron recoils – althoughthis conclusion stems from a relatively small expo-sure. In any case, this effect should be carefullyassessed in future noble liquid instruments wherescintillation is detected by photomultipliers.For this and other reasons optical cross-talk be-tween phototubes must be avoided, so that theseeffects remain localised to a single channel and aretherefore easier to identify. It is also important toreduce the amount of LXe allowed around their en-velopes to minimise scintillation from internal β -particles, and to prevent those photons from reach-ing the photocathode either by external or inter-nal paths. The copper structures which house theZEPLIN-III phototubes allow a thicker LXe layeraround their envelopes than with the original ones,since the upgrade photomultipliers are slightly nar-rower. However, there is no significant direct pathinto the main volume and the quartz envelopes weremetallised to prevent light from reaching the pho-tocathode directly. γ -ray interactions: MSSI events Multiple interactions where one vertex occursin a detector region which is optically coupled tothe phototube array but from which no chargecan be extracted can lead to a challenging back-ground. Multiple-scintillation single-ionisation(MSSI) events produce lower S2/S1 ratios than sin-gle fiducial scatters since the multiple S1 pulses aretime-coincident but there is only one S2. This wasidentified early on as a background mechanism intwo-phase detectors. In ZEPLIN-III an initial as-sessment during construction showed that efficientrejection based on S1 light pattern and S1-S2 ver-tex consistency would be necessary [2]. In the first8 -350-300-250-200-150-100-500 0 50 100time, ns v o lt ag e , m V x0.5 Figure 4: Distribution of photoelectron mean arrival times( τ ) in S1-like pulses for fiducial interactions (yellow his-togram) and for S1-only events occurring between the pho-tomultiplier windows and the cathode grid. The apparentenergies are 30–60 keVee for both histograms. A popula-tion of fast pulses is clearly visible, which we attribute toCherenkov photons generated in the phototube windows byinternal β − emission. Inset: array time response for threeevents with 30 keVee reconstructed energy (approximately55 photoelectrons); the red trace is a fiducial event with τ =35 ns; the black trace, probably a scintillation event con-taminated by significant Cherenkov light, is chosen from theblack histogram with τ =13 ns. The pulse in blue has τ =7 nsand occurs in only one channel (the other cases both trigger7 channels). run this event topology became perhaps the mostchallenging, in particular because these populationscannot be reproduced accurately with calibrationpoint sources.A detailed calculation is difficult when the lightcollection for the dead vertex is not known pre-cisely. In ZEPLIN-III, this is the case in two re-gions which yield no charge: i) the reverse field re-gion below the cathode grid, where light collectionis high but variable due to the proximity to thephotomultiplier windows; ii) the peripheral regionnear the side walls, where the light collection is lowand also uncertain. In any case, it is instructiveto assess what effect these events will have on dis-crimination if not successfully rejected. Considertwo γ -ray interactions, one in the fiducial regionand one in a dead region, which we assume to haveequal light collection and a constant spectrum of en- ergy deposited (as suggested by the Klein-Nishinaformula for low-angle scattering of high energy γ -rays). The energy is reconstructed from the coin-cident S1 components, but there is only one S2.The number of such double scatters adding up toenergy E is P ( E ) dE = P E dE , where P is thesingle-scatter probability per unit energy. So, thiseffect becomes more frequent with increasing recon-structed energy, as might be expected. The discrim-ination parameter, S2/S1, will suffer most when the‘good’ vertex produces the smallest detectable S2signal ( E th ) and S1 takes its apparent energy fromthe dead vertex only (S2/S1 >E th /E ). However,the ( x, y ) positions recovered from S1 and S2 in-dependently are likely to be inconsistent for theseevents, leading to their probable rejection. There-fore, the good S1 vertex must produce a minimumviable signature and, in practice, E th ∼ . E , then the spectrum of double-vertex events is P ( E ) dE ∝ E exp( − E/E ) dE (i.e. the relative num-ber of such events still increases with E ).
4. Radioactivity neutrons
The agreement between data and simulationsobtained for the spectrum of electron recoils atlow energy confirms the radiological contamina-tions adopted for each component; these were sub-sequently used to derive neutron production rates inthose materials. Muon-induced showers, in partic-ular photo-production in electromagnetic cascades,are the only noteworthy source of non-radioactivityneutrons. This contribution is not dominant in adetector of the size of ZEPLIN-III [23, 24] and, inaddition, these neutrons are vetoed effectively. Thedominant sources of fast neutrons are, therefore,spontaneous fission (mostly of
U) and ( α ,n) re-actions arising from the U and Th chains. Absolute neutron spectra were calculated withSOURCES-4A and -4C [25, 26] with ( α ,n) cross-sections extended to 10 MeV as described in An α activity of 10 Bq is present in the polypropylenedue to Gd, but this is not a significant source of neutrons( E α =2.8 MeV). α ,n) reaction domi-nates in light materials, producing, e.g., nearly allneutrons emitted by the Boulby rock (consideredpure NaCl here), very few in copper ( < α -particles.Primary neutron spectra are generated isotropi-cally from an appropriate position in the GEANT4simulation geometry and tracked to the LXe tar-get. In the case of the photomultipliers this is doneat sub-component level, since some materials arefurther away from the fiducial volume than oth-ers. Each photomultiplier emits just over 2 neu-trons per year; this is dominated by a small amountof borosilicate glass still present in this model; al-though there is scope for further improvement, thisvalue represents a 50-fold reduction relative to thoseused in the first phase of the experiment.The most challenging calculation is the contri-bution from the laboratory rock. It is computa-tionally intensive and requires accurate modellingof the shield: in spite of the high overall attenuationfactor of ∼ , any gaps for pipework, cabling andalignment tolerances need to be taken into account.To ensure an accurate result a detailed CAD solidmodel of the veto and shield was imported into theGEANT4 model. The Boulby rock has 65 ppb Uand 130 ppb Th assumed in secular equilibrium [30].Neutrons are generated in the rock to a depth of3 m and tallied initially on a test surface in the lab-oratory. The simulated integral flux is adjusted atthis stage to measurements at Boulby [31], includ-ing re-entrant (i.e. backscattered) neutrons. Thedifferential spectrum is released with a cosine biasfrom a generator sphere containing the experimentto generate an isotropic and uniform flux within it.Yearly rates of single scatters in 5–50 keVnr inthe fiducial volume and the component-level break-down are given in Table I and shown in Figure 5.The total of 3.05 events per year is dominated bythe ceramic feedthroughs (44%), followed by thephotomultipliers and rock neutrons. We note thatthe background expectation predicted for a one-year dataset with realistic signal acceptance is con-siderably lower; typical signal detection efficiencies(including 90% operational duty cycle, 50% accep- tance in S2/S1 discrimination parameter, a further25% loss from software selection cuts), bring thisdown to 1.0 n/year; the rate in anti-coincidencewith the veto is 60% lower, at 0.4 n/year. Finally,the detection threshold is likely to be a little higherthan 5 keVnr, so a value just below 0.3 n/year is ex-pected. This is very close to the original goal of theexperiment. The veto-coincident data analysed sofar contains zero delayed coincidences (dtag) withinthe signal acceptance region, but this constrains thelatter event rate only to < -8 -7 -6 -5 -4 -3 -2 nuclear recoil energy, keVnr d i ff e r e n ti a l r a t e , ev t/ kg / da y / k e V n r t o t a l f i d u c i a l r a t e , eve n t s / ye a r Integral rate (right)Differential rate (left)FeedthroughsPhototubesRock (halite)Polypropylene
Figure 5: Simulated differential and integral rates of neutronsingle elastic scattering in the fiducial volume of ZEPLIN-III.
The event rates discussed above are for singlescatters in the 6.5 kg fiducial volume with no ad-ditional interactions (elastic or otherwise) in the12.5 kg active volume. Where more than one elasticscatter occurs, the low S2/S1 ratios (and short scin-tillation time constant) characteristic of nuclear re-coils are essentially preserved. Although the eventcan still be rejected from the multiple S2 pulses, itis often useful to establish the distribution of scat-tering multiplicity. This may provide additionalconstraints to the expected rate of single scatters,which are an irreducible background for WIMPsearches. In ZEPLIN-III, some 50% of events withenergies reconstructed in the 5–50 keVnr regionare single interactions. The mean scattering lengthvaries quite significantly with neutron energy, butit is much less sensitive as a function of recoil en-10rgy ( λ ∼
30 cm). For systems much larger thanthis value, self-shielding can render the fraction ofsingle scatters very small. Multiple scatters can beidentified by additional S2 pulses in two-phase sys-tems (although they will become MSSI events if thesecond vertex yields no charge). However, the time-of-flight for radioactivity neutrons should not be ne-glected as a valuable discriminant. The mean timebetween scatters (above threshold) is only ∼
15 nsin ZEPLIN-III, but this effect can be readily seen inthe structure of S1 pulses during neutron calibra-tion, which points to its usefulness in large single-phase detectors.
5. Discussion
In this study we analysed electron and nuclearrecoil backgrounds of radiological origin in theZEPLIN-III experiment now operating in its sec-ond phase at Boulby. Energy spectra and spa-tial distributions are presented for electron recoilsin the fiducial volume. The low energy rate is0.75 ± Kr, identified early on as possible concerns, con-tribute very little. Similar studies for other experi-ments, namely XENON100 [32], have been equallysuccessful, confirming that this methodology pro-duces accurate results.The good agreement achieved for electron re-coils also confirms the primordial U/Th contamina-tion of key components, lending more certainty to the calculation of neutron backgrounds from nat-ural radioactivity. The total rate of single elasticscatters in the fiducial volume is 3 events/year inthe energy range 5–50 keVnr, which translates to0.4 events/year in anti-coincidence with the vetofor a realistic signal acceptance. This will not limitthe sensitivity of the experiment. We also pointedout that neutron time-of-flight can be important toestablish scatter multiplicity in large single-phaseexperiments.We examined event topologies which chal-lenge the high discrimination power which canbe achieved for single-vertex electron recoils.Cherenkov light produced in photomultiplier win-dows by internal β − activity or by Compton elec-trons from background γ -rays can be detected incoincidence with γ -ray fiducial scatters, therebymaking scintillation pulses appear larger as well asfaster – i.e. more similar to nuclear recoils. Dynodeglow is another cause of signal-coincident light, al-though it should affect both S1 and S2 pulses. Inthis instance photon emission is delayed with re-spect to the original optical stimulus depending onthe input optics configuration and its voltage bias.These mechanisms were identified several decadesago (see, e.g., [33] and references therein). In thecontext of dark matter searches, tests conducted atImperial College in the early 1990s confirmed thatdynode glow and Cherenkov light were observedwhen two photomultipliers face each other withoutan intervening scintillator [34]; subsequently, theNAIAD experiment at Boulby documented a pop-ulation of fast noise events ( τ <
100 ns) which wereattributed to dynode glow (see Fig. 1 in Ref.[35]).More recently, a WIMP search with CaF (Eu) scin-tillator at Kamioka Observatory found that the sen-sitivity below 10 keVee was limited by Cherenkovphotons produced in light guides by Compton elec-trons caused by background γ -rays [36]. This typeof event may be difficult to identify if the light pat-tern is not too peaked in a single channel. We notethat these effects should also concern single-phaseexperiments such as XMASS [37], MiniCLEAN [38]and DEAP-3600 [39].We discussed also γ -ray MSSI events and howthese degrade the discrimination power of two-phase detectors. They pose a real challenge tocurrent experiments and their mitigation shouldattract significant design effort in future systems.Whereas single-vertex electron recoils have a steepdistribution in the S2/S1 discrimination parame-ter, and their leaking into the nuclear recoil re-11ion is progressive, this is not the case for MSSIevents. Fortunately, both the frequency and sever-ity of these events decreases for lower energies,where more signal is expected.Under the assumption that the discriminationpower eventually observed is not hindered signifi-cantly by poor photomultiplier performance or in-complete rejection of MSSI events, ZEPLIN-IIIshould achieve a sensitivity of ∼ × − pb · year tothe scalar WIMP-nucleon elastic cross-section.
6. Acknowledgements
The UK groups acknowledge the support of theScience & Technology Facilities Council (STFC)for the ZEPLIN-III project and for maintenanceand operation of the Boulby underground labora-tory. LIP-Coimbra acknowledges financial supportfrom Funda¸c˜ao para a Ciˆencia e a Tecnologia (FCT)through project grant CERN/FP/116374/2010and postdoctoral grants SFRH/BPD/27054/2006,SFRH/BPD/47320/2008 and SFRH/BPD/63096/-2009. The ITEP group acknowledges supportfrom the Russian Foundation of Basic Research(grant 08-02-91851 KO a) and Rosatom (contractH.4e.45.90.10.1053 from 03-02-2010). We are alsograteful for support provided jointly to ITEP andImperial from the UK Royal Society. ZEPLIN-IIIis hosted by Cleveland Potash Ltd (CPL) at theBoulby Mine and we thank CPL management andstaff for their long-standing support. We also ex-press our gratitude to the Boulby facility staff fortheir dedication. The University of Edinburgh is acharitable body registered in Scotland (SC005336).
References α ,n), spontaneous fission, and delayed neu-tron sources and spectra, Tech. Rep. LA-13693, LANL(1999).[26] W. B. Wilson, et al., SOURCES 4C: A code for calcu-lating ( α ,n), spontaneous fission, and delayed neutronsources and spectra, Tech. Rep. LA-UR-02-1839, LANL(2002).[27] M. Carson, et al., Astropart. Phys. 21 (2004) 667.[28] M. Herman, et al., NDS 108 (2007) 2655.[29] V. Tomasello, V. A. Kudryavtsev, M. Robinson, Nucl.Instrum. Meth. A 595 (2008) 431.[30] P. F. Smith, et al., Astropart. Phys. 22 (2005) 409.[31] E. Tziaferi, et al., Astropart. Phys. 27 (2007) 326.[32] E. Aprile, et al.ArXiv:1101.3866v1.[33] A. G. Wright, J. Phys. E: Sci. Instrum. 16 (1983) 300.[34] J. Li, Tech. rep., Imperial College London (1991).[35] G. J. Alner, et al., Phys. Lett. B 616 (2005) 17.[36] C. E. Shimizu, et al., Phys. Lett. B 633 (2006) 195.[37] Y. Suzuki, arXiv:hep-ph/0008296v1.[38] D. N. McKinsey, et al., Nuc. Phys. B (Proc. Suppl.) 173(2007) 152.[39] M. Boulay, B. Cai, J. Phys.: Conf. Ser. 136 (2008)042081.,n), spontaneous fission, and delayed neutronsources and spectra, Tech. Rep. LA-UR-02-1839, LANL(2002).[27] M. Carson, et al., Astropart. Phys. 21 (2004) 667.[28] M. Herman, et al., NDS 108 (2007) 2655.[29] V. Tomasello, V. A. Kudryavtsev, M. Robinson, Nucl.Instrum. Meth. A 595 (2008) 431.[30] P. F. Smith, et al., Astropart. Phys. 22 (2005) 409.[31] E. Tziaferi, et al., Astropart. Phys. 27 (2007) 326.[32] E. Aprile, et al.ArXiv:1101.3866v1.[33] A. G. Wright, J. Phys. E: Sci. Instrum. 16 (1983) 300.[34] J. Li, Tech. rep., Imperial College London (1991).[35] G. J. Alner, et al., Phys. Lett. B 616 (2005) 17.[36] C. E. Shimizu, et al., Phys. Lett. B 633 (2006) 195.[37] Y. Suzuki, arXiv:hep-ph/0008296v1.[38] D. N. McKinsey, et al., Nuc. Phys. B (Proc. Suppl.) 173(2007) 152.[39] M. Boulay, B. Cai, J. Phys.: Conf. Ser. 136 (2008)042081.