Unfolding Neutron Spectrum with Markov Chain Monte Carlo at MIT Research Reactor with He-3 Neutral Current Detectors
A. F. Leder, A. J. Anderson, J. Billard, E. Figueroa-Feliciano, J. A. Formaggio, C. Hasselkus, E. Newman, K. Palladino, M. Phuthi, L. Winslow, L. Zhang
PPreprint typeset in JINST style - HYPER VERSION
Unfolding Neutron Spectrum with Markov ChainMonte Carlo at MIT Research Reactor with He-3Neutral Current Detectors
A. Leder a , g , A. J. Anderson b , c , J. Billard d , E. Figueroa-Feliciano e , J. A. Formaggio a ,C. Hasselkus f , E. Newman a , K. Palladino f , M. Phuthi a , L. Winslow a , L. Zhang a a Massachusetts Institute of Technology, 77 Massachusetts Ave. Cambridge, MA 02139, USA b Fermi National Accelerator Laboratory, Batavia, IL 60510, USA c Kavli Institute for Cosmological Physics, University of Chicago, Chicago, IL, USA 60637 d Univ Lyon, Université Claude Bernard Lyon 1, CNRS/IN2P3, Institut de Physique nucléaire deLyon, 4 rue Enrico Fermi, F-69622 France e Northwestern University Department of Physics & Astronomy 2145 Sheridan Road, Evanston, IL60208-3112 f Dept. of Physics and Astronomy, University of Wisconsin, Madison, WI 53706, USA g Corresponding AuthorE-mail: [email protected] A BSTRACT : The Ricochet experiment seeks to measure Coherent (neutral-current) ElasticNeutrino-Nucleus Scattering (CE ν NS) using dark-matter-style detectors with sub-keV thresholdsplaced near a neutrino source, such as the MIT (research) Reactor (MITR), which operates at 5.5MW generating approximately 2 . × ν /second in its core. Currently, Ricochet is character-izing the backgrounds at MITR, the main component of which comes in the form of neutronsemitted from the core simultaneous with the neutrino signal. To characterize this background, wewrapped Bonner cylinders around a He thermal neutron detector, whose data was then unfoldedvia a Markov Chain Monte Carlo (MCMC) to produce a neutron energy spectrum across severalorders of magnitude. We discuss the resulting spectrum and its implications for deploying Rico-chet at the MITR site as well as the feasibility of reducing this background level via the addition ofpolyethylene shielding around the detector setup.K
EYWORDS : Neutron detectors (cold, thermal, fast neutrons); Neutrino detectors; Gaseousdetectors;Detector modeling and simulations I . a r X i v : . [ phy s i c s . i n s - d e t ] F e b ontents
1. Introduction 12. Experimental Setup 2
3. MITR flux reconstruction 9
4. Conclusion 11
1. Introduction
Coherent (neutral-current) Elastic Neutrino-Nucleus Scattering (CE ν NS ) is a phenomenon whoseobservation and characterization would probe physics beyond the Standard Model. The recentdiscovery of CE ν NS by the COHERENT collaboration [1] has proven the feasibility of the idea,opening the door to the characterization of CE ν NS . CE ν NS has recently attracted attention as agateway to understanding non-standard interactions [2] in the neutrino sector as well as contribut-ing to our understanding of supernova dynamics. Several proposed experiments [3, 4], includingRicochet [5], seek to take advantage of the low-threshold and high-mass detectors already deployedfor dark matter direct detection experiments. The Ricochet proposal seeks to utilize a CDMS-stylesuperconducting Zn and/or Ge detector with a target threshold of 100 eV and an initial payloadof 1 kg. Zinc superconducting detectors, the subject of ongoing research and development, offerthe possibility of intrinsic background rejection due to the difference in quasi-particle interactiondynamics between nuclear and electron recoils [6]. In this paper, we explore the possibility ofdeploying the first phase of the Ricochet experiment 7 meters from the 5.5 MW MIT (research)Reactor (MITR) core. The MITR has an expected core neutrino flux of 2 . × ν /second cor-responding to a CE ν NS signal event rate of approximately 1 events/kg/day with low threshold Zndetectors. The proximity to the reactor core comes at the cost of an additional intrinsic backgroundin the form of neutrons, which could mimic a CE ν NS signal in Ge/Zn detectors. High-energyreactor neutrons (above 1 MeV) in particular make it through the concrete shielding surround-ing the reactor core and can then interact with our detectors. This creates the need for additionalpolyethylene shielding around the detector to absorb this high-energy neutron background. Thisunderscores the need for accurate information on both the shape and overall normalization of the– 1 –eutron energy spectrum over a wide range of energies. In order to achieve this neutron monitor-ing, we deployed a He detector that had previously been used to monitor neutrons produced inneutral-current reactions at the Sudbury Neutrino Observatory (SNO) together with incrementallythicker PVC layers in a Bonner cylinder approach.The He detector as shown in Figure 1, has a high neutron sensitivity at thermal neutronenergies, but this drops off quickly at higher neutron energies [7]. Akin to a Bonner sphere, byincreasing the amount of shielding around the detector one samples progressively higher energycomponents of the incoming neutron spectrum, which have been thermalized to the sensitive regionof the He detector. For each shielding configuration, we then record energy deposited in the He detector as well as the rise time of each pulse for use as a pulse-shape discriminator (discussedin Section 2). He-CF Gas FillAnode WireDelay Line TerminationFused Silica InsulatorNickel Counter BodyReadout Cable 2 m5.08 cm Coaxial View He Hen ntp tpE <764 keVE = 764 keVFigure 3-1: A diagram of the NCD including a coaxial view illustrating the processof neutron captures.
The MITR is powered by fission of U , which produces (among other things) mostlythermal neutrons (as opposed to high energy neutrons). Some higher energy neutronsare created in the process but most of them are successfully ‘cooled’ to thermal energylevels such that they can contribute to the chain reaction of fission [6]. The neutronspectrum from the reactor incident on our experimental setup is not known. In orderto measure it accurately, and thus more accurately measure the neutrino flux, we25 Figure 1:
Labeled schematic of the NCD detec-tors taken from [8]
After collecting the event rates from the vari-ous shielding configurations, we then unfold boththe neutron spectral shape and the normalization.This is achieved by creating a series of trans-fer functions for each shielding configuration inGeant4 by simulating the detector response to var-ious mono-energetic neutron sources as discussedin Section 3. This suite of transfer functions isthen used to calculate the likelihood of variousbinned neutron spectral shapes, which allows usto then optimize the spectral shape and normal-ization that maximizes the likelihood. For calibra-tion and as a cross-check of our unfolding proce-dure we also deployed a
Cf source at the MITRsite. Finally, we ran additional Geant4 simula-tions, using our calculated neutron spectrum as aninput, which sought to measure the possibility ofreducing the overall neutron background to a levelbelow the expected CE ν NS signal rate.
2. Experimental Setup
In order to monitor the neutron flux, we used oneNeutral Current Detector (NCD) taken from theSNO experiment [9], which had previously beenused to tag neutrons produced in neutral-currentinteractions. The NCD consists of one 2-meter-long cylinder filled with a 85:15 percent mixture of He and CF , ∗ with a collection anode wire running coaxially through the gas. The detectors were ∗ CF operates as a quenching gas which stabilizes the avalanche process that occurs when the induced charge getsto within tens of microns of the anode collection wire – 2 –iased at 1650 V, which represented the highest gain we could use before screening effects fromspace charge degraded the NCD energy resolution † . Incoming neutrons interact with the He viathe following process: n + He → p + H +
764 keV (2.1)The proton and the triton molecules share the 764 keV kinetic between themselves and then ionizethe He /CF gas mixture as they lose energy passing through the cylinder. While the ions getcollected on the outer grounded shell of the NCD, the resulting induced charge is collected on theanode wire which is then read out using a charge-sensitive pre-amplifier (Canberra Model 2006)fed into a National Instruments ADC (model number: USB-5132). We performed a series of SRIMsimulations [10] to determine the proton and triton track length in the He , determining an averagetrack length of 7.5 and 2.5 mm respectively [11]. We tested 6 different shielding configurationscorresponding to an overall shielding thickness that ranged from 0 up to 7.23 cm radially. For eachof the 6 shielding configurations we collected data with the reactor on/off and with a
Cf neutroncalibration source.
Calibrated Event Energy [keV]200 400 600 800 1000 1200 s ] m R i s e T i m e [ (a) The trapezoid plot showcasing the four eventpopulations present and their relative densities inour detector setup. The cut on Neutron Cap-ture (NC) events was defined by a series of 4linear cuts over the NC region, with particularcare taken on the region between the LIE and NCevent regions s] m Time [ R a w A m p li t ude [ A DC ] -1000100 LIE EventGlitch Event764 keV Peak Event Event a (b) Each of the example pulses shown herewas drawn from one of the four event re-gions/topologies visible in figure 2a. The largedifference in rise times stem from differences inthe orientation of event tracks while the differ-ence in energies for NC events stem from thecapture of one or both of the He -Neutron re-action products
Figure 2:
The lower trapezoid corresponds to the signal neutron captures while the band on the left corre-spond to Low Ionizing Events (LIE). The band of events seen at approximately 1200 keV corresponds tohigh energy alpha energy depositions saturating the readout electronics. Events with extremely short risetimes form a population of glitch events which were discarded as background in this analysis
When an event interacts with the NCDs and ionizes the He / CF mixture, the dE / dx event plus the track orientation determines the event time length, while the overall event amplitude isdetermined by the total energy of the proton/triton pair collected by the anode wire. † Measured to be 44 keV (FWHM) at 764 keV – 3 –rom each event trace, two variables were estimated: the rise time ‡ and the event amplitude.From these data, we could then produce a "trapezoid plot" (as seen in Figure 2a) with two clearpopulations of events visible: with fast, high-energy events defining neutron captures and generallyslower lower energy background events. The main source of background events for the NCDsstem from Compton scattered light particles such as electrons and muons with longer rise timescoupled with lower event amplitudes allowing for their classification as Low-Ionizing Events (LIE).We also observed two additional background signal populations with extremely short rise times,corresponding to micro-discharges/glitch events and events that saturated the DAQ system at 1.2MeV corresponding to α deposition events. For a given energy of event, one can also note therange of rise times which stems from the particular orientation of the event track, with perfectlyparallel events resulting in very short rise times and perpendicular events having the longest risetimes. Lower energy neutron events produce a smaller range of rise times because of incompleteevent collection due to the walls of the NCDs. The NCDs were calibrated by using a three-pointcalibration fitting scheme using the 191 keV, 564 keV and 764 keV event peaks/shoulders whichcorrespond to: full energy of just the proton collected, full energy of just triton collected and fullreconstruction of the entire event-with both proton and triton induced ionization collected on anodewire respectively. The event-selection criteria were defined by a trapezoidal region defined by aseries of four linear boundaries as shown in Figure 3a in this rise-time/energy-deposition space.The four linear boundaries were defined by looking at the data collected using the maximum PVCthickness as these data had the clearest separation between LIE and neutron events due to the thickshielding around the NCDs.The gain and shape of the trapezoid region was stable from run to run and over the courseof 24+ hour long runs, which justified our use of the same selection criteria for all the data setscollected. The rounded edge of the nuclear recoil band at 764 keV visible on Figure 3b was dueto space charge screening effects. The space charge effects were simulated in GEANT4 usingthe framework established by [13]. At lower anode voltages, space charge effects are minimizedas evidenced by the sharp cutoff in the NCD spectrum at 764 keV (visible in Figure 3), while atlower event energies the space charge effect is negligible. Reasonable agreement was achievedbetween the simulated and the collected NCD spectrum even at higher anode voltages. Spacecharge effects were determined to only have a negligible effect on the systematics on the overallevent rates extracted from the NCDs due to the excellent signal/background ratio at 764 keV. Inorder to calculate the event rate accounting for dead time and small changes in the event rateover time, the time between events was measured and fit to an exponential distribution. FromFigure 4, one can see that the data is well-described by an exponential distribution and that thedetectors had a dead time of approximately 30 ms. These extracted event rates were then fed intothe deconvolution procedure outlined below. Background NCD rate measurements, which includedcosmic-ray induced neutrons, with the reactor off were subtracted from the reactor on NCD data. Inorder to determine the effective cosmic ray overburden at the MITR for a possible future Ricochetdeployment, we collected cosmic ray background data using desktop muon counters [14] during a24 hour period where the reactor was turned off. We determined the effective overburden to be 1.5 ‡ We defined the rise time as the time it takes for the event trace to go from 10 to 70 percent of the maximumamplitude – 4 – alibrated Event Energy [keV]200 400 600 800 1000 1200 s ] m R i s e T i m e [ (a) Rise-time versus Energy plot taken from atypical NCD run. Red events indicate thoseevents that passed the data quality cuts thatidentify these events as neutron induced (NR)events, while all black events represent back-ground LIE/Glitch/ α events. Calibrated Energy [keV]
200 400 600 800 E v en t s / k e V All DataLIE EventsNC EventsNCD Simulated Events (1650 V) (b)
Calibrated spectrum of events in the neutroncapture region. The 764 peaks together with the564 and 191 keV shoulders were fitted to cali-brate the NCD response. The 764 keV peak wasfitted using an Exponentially Modified Gaussian(EMG) [12], while the 564 and 191 keV shoul-ders were fitted with error functions. Note thatthe error bars on the NR/LIE spectrum may betoo small to see
Figure 3:
NCD nuclear capture event criteria and calibrated NCD spectra (after selecting nuclear captureevents) as well as a simulated NCD spectrum. In both plots one can see the 764, 573 and 191 keV featuresthat are the hallmarks of NCD NR events m.w.e.
Time Between Events [ms] E v en t s / m s -6 -5 -4 -3 Layer 1 DataLayer 2 DataLayer 3 DataLayer 4 DataLayer 5 DataLayer 6 Data
Time Between Events [ms] E v en t s / m s -4 -3 -2 Figure 4:
Time between subsequent nuclear recoil events for all the shielding configurations we ran, nor-malized to 1. One can clearly see the monotonic decrease in event rate as a function of layers, consistent witha strong thermal component to the neutron spectrum.
Inset : Same plot focusing on the short time betweenevents to highlight the 30 ms dead time in our detector setup – 5 – .3 Geant4 simulation
Given measurements of the reactor neutron rate with various numbers of PVC layers surround-ing the detector, we are interested in recovering both the shape and normalization of the neutronspectrum which requires unfolding techniques. Although there are various approaches advocatedin the literature, nearly all require the calculation of a transfer function T j ( E i ) , which relates theincident neutron flux φ to the event rate R measured by the NCD. In our case, for each shieldingconfiguration j , we require a transfer function, T j , defined over the same interval as the binnedenergy spectrum E i , in order to estimate the theoretical event rate: R jtheory (cid:39) ∑ i = (cid:90) E i + E i φ ( E i ) T j ( E , n ) dE . (2.2)We assume that the neutron flux is flat in lethargy space, § while showing an inverse energydependence in energy space throughout a bin. The transfer functions were calculated over a neutronenergy range encompassing thermal neutrons up to high energy reactor neutrons. The high-energycutoff (10 MeV) for the reconstruction bins was selected because less than 1 percent of the emit-ted neutrons have energies greater than 10 MeV for a thermal reactor core spectrum centered at1.4 MeV. We then use maximum likelihood to fit the neutron energy spectrum in energy bins to theobserved rates for each shielding configuration as discussed further in Section 4.We estimate the transfer function of the detector and shielding setups using a Monte Carlosimulation based on Geant4 10.00.p02 [9]. To simulate the hadronic physics, including neutroninteractions, we use a modular physics list including the QGSP_BERT_HP model containing the”high-precision” neutron physics simulation that uses version 4.4 of the G4NDL data. In addi-tion, our physics list incorporates the “G4ScreenedNuclearRecoil” process that models screenedelectromagnetic nuclear elastic scattering and is important for accurately simulating the propaga-tion of the proton and triton after a neutron capture on He [15]. Unlike the default models in theQGSP_BERT_HP physics list, the G4ScreenedNuclearRecoil processes give physically realisticresults for ion track lengths and energy loss in the detector gas, which were also found to be inreasonable agreement with results from the widely-used SRIM software [10].The simulation includes the layers of PVC shielding as described in the section 2, correctlyaccounting for the geometry of the PVC pipes when resting on the ground. Although it is notneeded for the transfer function analysis, the simulation also implements models of the chargepropagation, space-charge effects, and the preamplifier electronics, similar to [11]. To simulatethe transfer functions needed for the unfolding analysis, we simulate 10 million monoenergeticneutrons with each of the 6 shielding configurations and at 34 logarithmically-spaced energiesbetween 10 − MeV and 10 MeV. The neutrons in the simulation are generated on a cylindricalsurface immediately surrounding the outermost layer of shielding, with a direction that is chosenisotropically from the surface of the shielding directed inward toward the detector. We have alsodetermined that bias in our calculations due to non-concentric compared to perfectly concentric § Lethargy is the logarithmic ratio of energy before/after a collision, given by the relation u = ln ( E E ) (2.3) – 6 –lignment of the various shells is small due to the approximate radial symmetry of the NCD setup.Another effect of this radial symmetry is that the NCD setup is insensitive to the initial direction ofthe reactor background neutrons. In addition, the data collected by the NCDs pointed to a strongthermalized component to our neutron spectrum at the MITR, which supports the assumption thatour neutron flux will be closer to isotropic. Log (keV)0100200300400500600700800900 T r a n s f e r F un c t i o n s ( c m ) Layer 1 = 0.8 cm of Shielding Layer 2 = 1.6 cm of ShieldingLayer 3 = 2.7 cm of ShieldingLayer 4 = 4.0 cm of ShieldingLayer 5 = 5.5 cm of ShieldingLayer 6 = 7.2 cm of Shielding
Figure 5:
Transfer functions for each shielding configuration as a function of incident simulated monoener-getic neutron energy - the large degree of overlap between these transfer functions prompted our use of anMCMC to deconvolve the data. The transfer functions are normalized to the surface area of the shieldingconfiguration used to calculate them.
Figure 5 shows the transfer functions for each layer of shielding as a function of energy. Fromthese plots one can note a few important features, namely that we only gain significant sensitivityfor fast "core" neutrons with 5 or 6 layers of shielding. In addition, we note a significant degeneracybetween the transfer functions for each shielding configuration, which limits the sensitivity of theunfolding analysis, which motivated the idea of using a MCMC analysis as discussed below.
Using the transfer functions (see Figure 5) calculated from the Geant4 simulations in the previoussection, we can calculate an expected theoretical event rate for each shielding configuration, given abinned neutron spectrum shape. By minimizing the difference between the theoretical and observedevent rates we can then extract a binned neutron spectrum. Due to the high correlation observedbetween the bins (see Figure 6) and to ensure that the best fit χ f it that we found truly represents aglobal best fit we use the maximum entropy method.First, we define a likelihood L function for the theoretical event rates for a given binnedspectrum ( φ i ) via the following relations: – 7 – f it = n ∑ i = ( R obsi − R theoryi [ φ ( (cid:126) E )]) σ i (2.4) S = − m ∑ i = p i log ( p i ) with p i = φ i ( E ) ∑ i φ i ( E ) (2.5) L φ = exp (cid:32) − χ f it + S w (cid:33) (2.6)where i corresponds to the particular NCD shielding configuration and S corresponds to the en- . . . . L o g ( L a y e r F l u x ) . . . . L o g ( L a y e r F l u x ) . . . . L o g ( L a y e r F l u x ) . . . . L o g ( L a y e r F l u x ) .
715 2 .
710 2 .
705 2 . Log ( Layer Flux ) . . . L o g ( L a y e r F l u x ) .
60 3 .
54 3 .
48 3 . Log ( Layer Flux ) .
20 4 .
05 3 .
90 3 . Log ( Layer Flux ) .
20 4 .
05 3 .
90 3 . Log ( Layer Flux ) .
05 3 .
90 3 .
75 3 . Log ( Layer Flux ) . . . Log ( Layer Flux ) Figure 6:
Correlation plots for accepted 6-bin MCMC points. Note that the values of each of the MCMCpoints are shown on a log scale here and in particular bins 1 and 2 have strongest correlations to the other,consistent with the large overlap in the transfer functions tropy. In this context, S represents another quantity that needs to be simultaneously minimizedby maximizing the likelihood function. In Equation 2.6, w corresponds to a regulation parameterwhich determines if our extracted spectrum is prior- (small w ) or data-dominated (large w ). A prior-dominated deconvolution would result in essential features smoothed out of the unfolded spectrumas the unfolding algorithm would converge to the prior ¶ . On the other hand, a data-dominated un-folding can result in large unphysical fluctuations in the resulting unfolded spectrum. We wish to ¶ In this analysis, our prior is the result of the Minuit fitting (maximum entropy) plus our choice of regularizationparameter – 8 –nd a value for w that minimizes the resulting χ f it with the additional condition that small changesin χ f it only result in minimal changes to S. In practice, this means finding the point on the χ − S curve (see Figure 7) with the largest curvature. It should be noted that this prescription only givesus an order-of-magnitude estimate for the best value for w . By varying the regularization parameteralong this curve we were also able to calculate the systematic error due to this method, which isincluded in the quoted systematic error in Table 1 χ [dless] E n t r o p y [ d l e ss ] Regularization Parameter = (Data Dominated)Regularization Parameter = − (Prior Dominated) Figure 7: χ -Entropy Plot used to calculate the optimal value of the regularization parameter. The optimalvalue for the regularization parameter was determined to be 10 − by examining the point with the largestcurvature. The regularization parameter controls whether the fitting procedure we are using is data or priordominated. We assume Poisson fluctuations in the measured values of the R obsi , which allows us to ap-proximate the likelihood as a multivariate Gaussian. We also assume that the neutron flux followsan exponential spectral shape within a bin (flat in lethargy space). By finding the binned spec-tral shape that minimizes L via Minuit [16], we generate a preliminary binned neutron spectrum;however, due to the high degree of overlap between the various transfer functions this results inhighly-correlated spectral neutron bins, and additional minimization is needed to find a global bestfit. In order to better characterize this strong correlation between neutron bins and determine thedistribution of expected neutron background events at the MITR, we employed a Markov ChainMonte Carlo (MCMC) analysis based on the framework described in Billard et al. [17].
3. MITR flux reconstruction
Over the course of May-October 2015, we undertook three sets of measurements at the MITRreactor corresponding to the reactor on/off together with a
Cf calibration neutron source. Usingthe unfolding techniques discussed above we were able to generate the following neutron spectra- see Figure 8. In particular, for the
Cf calibration data we noticed the peak position shifteddownward in energy, which we attribute to thermalization of the neutrons coming from the source.It is also important to note that strong thermal component inherent in all the unfolded spectra. This– 9 –uggests that the surrounding concrete/shielding converts the raw neutron flux into an isotropicneutron gas; however, for all spectra there still exists a strong high energy (> 1 MeV) componentto the neutron spectrum, and is it these high energy neutrons that then will contribute the most toany neutron background for a possible Ricochet deployment at the MITR. [MeV]) n (E Log -8 -6 -4 -2 0 ] - . c m - N eu t r on f l u x pe r un i t o f l e t ha r g y [ s Log -5-4-3-2-1
Cf Data at MITRReactor On DataReactor Off Data
Figure 8:
Deconvolved spectra using the NCD data collected with reactor on, off and with a
Cf neutronsource deployed at the MITR. The vertical error bars here correspond to the a ± As discussed in [ ] , Zn superconducting detectors offer the ability to distinguish between nuclearand electromagnetic recoils due to the differences in quasi-particle propagation inside the crys-tal lattice. Throughout these calculations, we foresee deploying a 1 kg Zn detector with a 100 eVthreshold, which represents an achievable target threshold set by the EDELWEISS experiment [18].Based on the calculations in [5], we determine the expected CE ν NS event rate in Zn at the MITRfacility to be approximately 1 events/kg/day at 7 meters from the core. Using the spectra extractedfrom the NCD data we performed another series of Geant4 simulations with a 1 kg Zn payloaddeployed in a Adiabatic Demagnetization Refrigerator (ADR) configuration. We simulate neutronsisotropically at a radius of 64 cm from the center of the ADR, drawing from an exponential distri-bution in energy space across each of the bins used in the MCMC deconvolution. These primaryevents are then passed through a Geant4 model of the ADR setup to record the energy depositionsof any neutrons. By then examining the number of events that fall in the Region of Interest (ROI)between 100 eV NR and 1 keV NR we then weigh each event by its corresponding flux value from theMCMC fit to then extract an expected neutron background event rate. The high neutron backgroundevent rate with no shielding highlights the need to include shielding around our detector setup forany deployment at the MITR. This prompted us to perform an additional simulation surroundingthe ADR with 30 cm, both radially and on the top/bottom, of bare and boron-doped (5 % by mass)polythene shielding to determine the feasibility of reducing the neutron background rate to the level– 10 –f the expected CE ν NS event rate. Systematic errors for this shielding simulation were calculatedby comparing the neutron stopping power for 0 to 30 cm of poly shielding between 400 keV and10 MeV to the reduction in flux as seen in [19]. Our systematic error indicates that the calculatedneutron event rate represents more of an upper limit on the actual neutron event rate that we couldexpect for a Ricochet deployment at MITR. With 30 cm of polyethylene shielding around the de-tector, our calculated background neutron event rate becomes comparable to the CE ν NS event rate,with the actual neutron background event rate being possibly even lower. By taking advantage ofMITR duty cycle, coupled with knowing the precise power levels over the course of 1 to 5 years,we have shown it is still possible to extract a CE ν NS discovery signal even when the signal tobackground ratio is as high as 1:5 (see Table 3 in [6]). With additional shielding, our neutronbackground rate could be brought down still more, improving our CE ν NS sensitivity further.
Rates (per kg per day) CE ν NS Rate MITR On MITR Off Cf0 cm poly shielding 1.0 36 . + . ( stat )+ . ( sys ) − . ( stat ) − . ( sys ) . + . − . ( . + . − . ) ×
30 cm poly shielding 1.0 3 . + . ( stat )+ . ( sys ) − . ( stat ) − . ( sys ) — —30 cm borated shielding 1.0 2 . + . ( stat )+ . ( sys ) − . ( stat ) − . ( sys ) — —90 % CL (0 cm borated shielding) 1.0 < 41 < 7 < 5 ×
90 % CL (30 cm borated shielding) 1.0 < 2.9 — —90 % CL (30 cm bare poly shielding) 1.0 < 3.8 — —
Table 1:
Summary of the neutron background rate expected by a Ricochet deployment at the MITR forthe three experimental configurations tested in this paper. All borated tests conducted with 5% (by mass)boron-doped poly shielding
4. Conclusion
We successfully deployed NCDs at the MITR site in order to measure the neutron backgroundacross several orders of magnitude to lay the groundwork for a possible future stage 1 deploymentof the Ricochet experiment. After applying pulse shape discrimination event-selection criteria weextracted an event rate for each of the 6 different NCD Bonner cylinder shielding configurations,each of which were used to probe different components of the overall neutron spectrum. We ap-plied an optimized maximum likelihood coupled with a MCMC analysis framework to character-ize the high correlations between the various neutron bins in the final reactor on, reactor off and
Cf calibration data. Using the final unfolded neutron spectrum we simulated a 1 kg Ricochetdeployment with a Zn target at the MITR with and without 30 cm of polyethylene shielding. Whilethe raw unfolded spectrum points to an intrinsic neutron background rate higher than the expectedCE ν NS signal rate. With the addition of 30 cm of shielding this background event rate was broughtdown close to the CE ν NS signal level as shown in Figure 9. While the low signal to backgroundratio makes a CE ν NS signal detection search challenging, we have shown in [6], that it is possibleto extract a CE ν NS discovery signal even when the signal to background rates are close to 1 to 5.With a similar analysis, coupled with additional passive shielding, the MITR represents not just agood location for continued Ricochet Zn Bolometer R&D testing, but also a potential additionalCE ν NS signal site. – 11 –
Radial Shielding Thickness (cm) E v e n t R a t e ( e v e n t s / k g / d a y ) CNS Rate at MITR (7 meters from core)ROI Event Rate (Poly Shielding)ROI Event Rate (Boron-doped Shielding)
Figure 9:
Event rate in the Region of Interest (between 0.1 and 1 keV) as a function of various thicknessesof Poly shielding (Normal and Boron-doped) around the sensitive detector region. Note: the error bars onthis plot only correspond to statistical errors
Acknowledgments
We wish to thank all the personnel, in particular Thomas Bork and Dane Kouttron, of the MITRcomplex for all their help in setting up and running our experiment. We also wish to thank theSNO collaboration and in particular Hamish Robertson for lending us the NCD detectors. Thismaterial is based upon work supported by the U.S. Department of Energy, Office of Science, Officeof Nuclear Physics under Award Numbers FG02-97ER41041 and DE-FG02-06ER-41420.– 12 – eferences [1] D. Akimov, J. B. Albert, P. An, C. Awe, P. S. Barbeau, B. Becker, V. Belov, A. Brown, A. Bolozdynya,B. Cabrera-Palmer, M. Cervantes, J. I. Collar, R. J. Cooper, R. L. Cooper, C. Cuesta, D. J. Dean, J. A.Detwiler, A. Eberhardt, Y. Efremenko, S. R. Elliott, E. M. Erkela, L. Fabris, M. Febbraro, N. E. Fields,W. Fox, Z. Fu, A. Galindo-Uribarri, M. P. Green, M. Hai, M. R. Heath, S. Hedges, D. Hornback, T. W.Hossbach, E. B. Iverson, L. J. Kaufman, S. Ki, S. R. Klein, A. Khromov, A. Konovalov, M. Kremer,A. Kumpan, C. Leadbetter, L. Li, W. Lu, K. Mann, D. M. Markoff, K. Miller, H. Moreno, P. E.Mueller, J. Newby, J. L. Orrell, C. T. Overman, D. S. Parno, S. Penttila, G. Perumpilly, H. Ray,J. Raybern, D. Reyna, G. C. Rich, D. Rimal, D. Rudik, K. Scholberg, B. J. Scholz, G. Sinev, W. M.Snow, V. Sosnovtsev, A. Shakirov, S. Suchyta, B. Suh, R. Tayloe, R. T. Thornton, I. Tolstukhin,J. Vanderwerp, R. L. Varner, C. J. Virtue, Z. Wan, J. Yoo, C.-H. Yu, A. Zawada, J. Zettlemoyer, andA. M. Zderic, “Observation of coherent elastic neutrino-nucleus scattering,”
Science , 2017.[2] K. Scholberg, T. Wongjirad, E. Hungerford, A. Empl, D. Markoff, P. Mueller, Y. Efremenko,D. McKinsey, and J. Nikkel, “The CLEAR Experiment,” in
Particles and fields. Proceedings, Meetingof the Division of the American Physical Society, DPF 2009, Detroit, USA, July 26-31, 2009 , 2009.[3] P. S. Barbeau, J. I. Collar, and O. Tench, “Large-mass Ultralow Noise Germanium Detectors:Performance and Applications in Neutrino and Astroparticle Physics,”
Journal of Cosmology andAstroparticle Physics , vol. 2007, no. 09, p. 9, 2007.[4] H. T.-K. Wong, “The TEXONO Research Program on Neutrino and AstroParticle Physics,”
ModernPhysics Letters A , vol. 19, no. 13n16, pp. 1207–1214, 2004.[5] J. A. Formaggio, E. Figueroa-Feliciano, and A. J. Anderson, “Sterile neutrinos, coherent scattering,and oscillometry measurements with low-temperature bolometers,”
Phys. Rev. D , vol. 85, p. 013009,Jan 2012.[6] J. Billard, R. Carr, J. Dawson, E. Figueroa-Feliciano, J. A. Formaggio, J. Gascon, S. T. Heine, M. D.Jesus, J. Johnston, T. Lasserre, A. Leder, K. J. Palladino, V. Sibille, M. Vivier, and L. Winslow,“Coherent Neutrino Scattering with Low Temperature Bolometers at Chooz Reactor Complex,”
Journal of Physics G: Nuclear and Particle Physics , vol. 44, no. 10, p. 105101, 2017.[7] L. D. P. King and L. Goldstein, “The Total Cross Section of the He-3 Nucleus for Slow Neutrons,”
Phys. Rev. , vol. 75, pp. 1366–1369, May 1949.[8] E. Newmann, “Neutron Backgrounds in the Ricochet Experiment,” Master’s thesis, MassachusettsInstitute of Technology, 77 Massaschusetts Avenue, Cambridge, MA 02139.[9] J. Allison et al. , “Geant4 Developments and Applications,”
IEEE Trans. Nucl. Sci. , vol. 53, p. 270,2006.[10] J. F. Ziegler, M. Ziegler, and J. Biersack, “SRIM – The Stopping and Range of Ions in Matter,”
Nucl.Instrum. Meth. , vol. B268, pp. 1818 – 1823, 2010.[11] B. Beltran et al. , “Full simulation of the Sudbury Neutrino Observatory proportional counters,”
New J.Phys. , vol. 13, p. 073006, 2011.[12] E. Grushka, “Characterization of exponentially modified gaussian peaks in chromatography,”
Analytical Chemistry , vol. 44, no. 11, pp. 1733–1738, 1972. PMID: 22324584.[13] H. S. W. C. Tseung,
Simulation of the Sudbury Neutrino Observatory Neutral Current Detectors . PhDthesis, University of Oxford, 2008. – 13 –
14] S. N. Axani, J. M. Conrad, and C. Kirby, “The desktop muon detector: A simple, physics-motivatedmachine- and electronics-shop project for university students,”
American Journal of Physics , vol. 85,no. 12, pp. 948–958, 2017.[15] M. H. Mendenhall and R. A. Weller, “An algorithm for computing screened coulomb scattering ingeant4,”
Nucl. Instrum. Meth. , vol. B227, pp. 420–430, 2005.[16] W. Verkerke and D. P. Kirkby, “The RooFit Toolkit for Data Modeling,” eConf , vol. C0303241,p. MOLT007, 2003. [,186(2003)].[17] J. Billard, F. Mayet, and D. Santos, “Markov chain monte carlo analysis to constrain dark matterproperties with directional detection,”
Phys. Rev. D , vol. 83, p. 075002, Apr 2011.[18] EDELWEISS Collaboration, E. Armengaud, Q. Arnaud, C. Augier, A. Benoît, A. Benoît, L. Bergé,T. Bergmann, J. Billard, J. Blümer, T. de Boissière, G. Bres, A. Broniatowski, V. Brudanin, P. Camus,A. Cazes, M. Chapellier, F. Charlieux, L. Dumoulin, K. Eitel, D. Filosofov, N. Foerster, N. Fourches,G. Garde, J. Gascon, G. Gerbier, A. Giuliani, M. Grollier, M. Gros, L. Hehn, S. Hervé,G. Heuermann, V. Humbert, M. De Jésus, Y. Jin, S. Jokisch, A. Juillard, C. Kéfélian, M. Kleifges,V. Kozlov, H. Kraus, V. A. Kudryavtsev, H. Le-Sueur, J. Lin, M. Mancuso, S. Marnieros,A. Menshikov, X. F. Navick, C. Nones, E. Olivieri, P. Pari, B. Paul, M. C. Piro, D. V. Poda,E. Queguiner, M. Robinson, H. Rodenas, S. Rozov, V. Sanglard, B. Schmidt, S. Scorza,B. Siebenborn, D. Tcherniakhovski, L. Vagneron, M. Weber, E. Yakushev, and X. Zhang, “Constraintson Low-Mass WIMPs from the EDELWEISS-III Dark Matter Search,” 2016.[19] V. Tomasello, M. Robinson, and V. Kudryavtsev, “Radioactive background in a cryogenic dark matterexperiment,”
Astroparticle Physics , vol. 34, no. 2, pp. 70 – 79, 2010., vol. 34, no. 2, pp. 70 – 79, 2010.