Lawrence Berkeley National Laboratory
Recent Work
Title
Results from a Search for Dark Matter in the Complete LUX Exposure.
Permalink
https://escholarship.org/uc/item/11b922t0
Journal
Physical review letters, 118(2)
ISSN
0031-9007
Authors
Akerib, DS
Alsum, S
Araújo, HM
et al.
Publication Date
2017
DOI
10.1103/physrevlett.118.021303
Peer reviewed
eScholarship.org Powered by the California Digital Library
University of California
Results from a search for dark matter in the complete LUX exposure
D.S. Akerib,
1, 2, 3
S. Alsum,
4
H.M. Ara´ujo,
5
X. Bai,
6
A.J. Bailey,
5
J. Balajthy,
7
P. Beltrame,
8
E.P. Bernard,
9, 10
A. Bernstein,
11
T.P. Biesiadzinski,
1, 2, 3
E.M. Boulton,
9, 10
R. Bramante,
1, 2, 3
P. Br´as,
12
D. Byram,
13, 14
S.B. Cahn,
10
M.C. Carmona-Benitez,
15
C. Chan,
16
A.A. Chiller,
13
C. Chiller,
13
A. Currie,
5
J.E. Cutter,
17
T.J.R. Davison,
8
A. Dobi,
18
J.E.Y. Dobson,
19
E. Druszkiewicz,
20
B.N. Edwards,
10
C.H. Faham,
18
S. Fiorucci,
16, 18
R.J. Gaitskell,
16
V.M. Gehman,
18
C. Ghag,
19
K.R. Gibson,
1
M.G.D. Gilchriese,
18
C.R. Hall,
7
M. Hanhardt,
6, 14
S.J. Haselschwardt,
15
S.A. Hertel,
9, 10, ∗
D.P. Hogan,
9
M. Horn,
14, 9, 10
D.Q. Huang,
16
C.M. Ignarra,
2, 3
M. Ihm,
9
R.G. Jacobsen,
9
W. Ji,
1, 2, 3
K. Kamdin,
9
K. Kazkaz,
11
D. Khaitan,
20
R. Knoche,
7
N.A. Larsen,
10
C. Lee,
1, 2, 3
B.G. Lenardo,
17, 11
K.T. Lesko,
18
A. Lindote,
12
M.I. Lopes,
12
A. Manalaysay,
17, †
R.L. Mannino,
21
M.F. Marzioni,
8
D.N. McKinsey,
9, 18, 10
D.-M. Mei,
13
J. Mock,
22
M. Moongweluwan,
20
J.A. Morad,
17
A.St.J. Murphy,
8
C. Nehrkorn,
15
H.N. Nelson,
15
F. Neves,
12
K. O’Sullivan,
9, 18, 10
K.C. Oliver-Mallory,
9
K.J. Palladino,
4, 2, 3
E.K. Pease,
9, 18, 10
P. Phelps,
1
L. Reichhart,
19
C. Rhyne,
16
S. Shaw,
19
T.A. Shutt,
1, 2, 3
C. Silva,
12
M. Solmaz,
15
V.N. Solovov,
12
P. Sorensen,
18
S. Stephenson,
17
T.J. Sumner,
5
M. Szydagis,
22
D.J. Taylor,
14
W.C. Taylor,
16
B.P. Tennyson,
10
P.A. Terman,
21
D.R. Tiedt,
6
W.H. To,
1, 2, 3
M. Tripathi,
17
L. Tvrznikova,
9, 10
S. Uvarov,
17
J.R. Verbus,
16
R.C. Webb,
21
J.T. White,
21
T.J. Whitis,
1, 2, 3
M.S. Witherell,
18
F.L.H. Wolfs,
20
J. Xu,
11
K. Yazdani,
5
S.K. Young,
22
and C. Zhang
13
(LUX Collaboration)
1
Case Western Reserve University, Department of Physics, 10900 Euclid Ave, Cleveland, OH 44106, USA
2
SLAC National Accelerator Laboratory, 2575 Sand Hill Road, Menlo Park, CA 94205, USA
3
Kavli Institute for Particle Astrophysics and Cosmology,
Stanford University, 452 Lomita Mall, Stanford, CA 94309, USA
4
University of Wisconsin-Madison, Department of Physics,
1150 University Ave., Madison, WI 53706, USA
5
Imperial College London, High Energy Physics, Blackett Laboratory, London SW7 2BZ, United Kingdom
6
South Dakota School of Mines and Technology, 501 East St Joseph St., Rapid City, SD 57701, USA
7
University of Maryland, Department of Physics, College Park, MD 20742, USA
8
SUPA, School of Physics and Astronomy, University of Edinburgh, Edinburgh EH9 3FD, United Kingdom
9
University of California Berkeley, Department of Physics, Berkeley, CA 94720, USA
10
Yale University, Department of Physics, 217 Prospect St., New Haven, CT 06511, USA
11
Lawrence Livermore National Laboratory, 7000 East Ave., Livermore, CA 94551, USA
12
LIP-Coimbra, Department of Physics, University of Coimbra, Rua Larga, 3004-516 Coimbra, Portugal
13
University of South Dakota, Department of Physics, 414E Clark St., Vermillion, SD 57069, USA
14
South Dakota Science and Technology Authority,
Sanford Underground Research Facility, Lead, SD 57754, USA
15
University of California Santa Barbara, Department of Physics, Santa Barbara, CA 93106, USA
16
Brown University, Department of Physics, 182 Hope St., Providence, RI 02912, USA
17
University of California Davis, Department of Physics, One Shields Ave., Davis, CA 95616, USA
18
Lawrence Berkeley National Laboratory, 1 Cyclotron Rd., Berkeley, CA 94720, USA
19
Department of Physics and Astronomy, University College London,
Gower Street, London WC1E 6BT, United Kingdom
20
University of Rochester, Department of Physics and Astronomy, Rochester, NY 14627, USA
21
Texas A & M University, Department of Physics, College Station, TX 77843, USA
22
University at Albany, State University of New York,
Department of Physics, 1400 Washington Ave., Albany, NY 12222, USA
(Dated: January 17, 2017)
We report constraints on spin-independent weakly interacting massive particle (WIMP)-nucleon
scattering using a 3.35×10
4
kg day exposure of the Large Underground Xenon (LUX) experiment. A
dual-phase xenon time projection chamber with 250 kg of active mass is operated at the Sanford Un-
derground Research Facility under Lead, South Dakota (USA). With roughly fourfold improvement
in sensitivity for high WIMP masses relative to our previous results, this search yields no evidence
of WIMP nuclear recoils. At a WIMP mass of 50 GeV c
−2
, WIMP-nucleon spin-independent cross
sections above 2.2×10
−46
cm
2
are excluded at the 90% confidence level. When combined with the
previously reported LUX exposure, this exclusion strengthens to 1.1×10
−46
cm
2
at 50 GeV c
−2
.
∗
hertel@berkeley.edu
†
aaronm@ucdavis.edu
arXiv:1608.07648v3 [astro-ph.CO] 14 Jan 2017
2
The Large Underground Xenon (LUX) experiment
searches for direct evidence of weakly interacting mas-
sive particles (WIMPs), a favored dark matter candidate.
The LUX search is performed with a dual-phase (liquid-
gas) xenon time projection chamber (TPC) containing
250 kg of ultrapure liquid xenon (LXe) in the active de-
tector volume [1]. Energy deposited by particle interac-
tions in the LXe induces two measurable signal channels:
prompt VUV photons from scintillation (S1), and free
electrons from ionization. The S1 photons are emitted
from the interaction site and detected by top and bot-
tom arrays of photomultiplier tubes (PMTs). Electrons
liberated by the interaction drift to the surface of the liq-
uid via an applied electric field. They are extracted into
the gas and accelerated by a larger electric field, pro-
ducing secondary electroluminescence photons collected
in both arrays with localization in the top PMTs (S2).
The PMT signals from both light pulses, S1 and S2, al-
low for the reconstruction of interaction vertices in three
dimensions.
The ratio of the S1 and S2 signals is used to discrimi-
nate between electronic recoils (ER) and nuclear recoils
(NR). WIMP interactions in the detector would primar-
ily appear as nuclear recoils of energy . 100 keV [2]. In
order to reduce backgrounds from external sources, the
detector is immersed in a 7.6 m diameter and 6.1 m high
ultrapure water tank, which itself is located underground
at the Sanford Underground Research Facility (SURF) in
Lead, SD, USA. The ∼1.5 km of rock overburden (4300
m.w.e.) provides a reduction in the rate of cosmic muons
of O(10
−7
). ER background populations arise from
40
K
and the
238
U/
232
Th decay chains present as contami-
nants in materials other than LXe, as well as from trace
amounts of
222
Rn and
85
Kr in the LXe itself. The
85
Kr is
largely removed from the xenon prior to filling by chro-
matographic separation in activated charcoal [3]. Addi-
tional information on the experimental setup [4–6] and
backgrounds [7] has been previously published.
The first LUX WIMP search (WS2013) collected
95 live-days of data from April to August, 2013 [8–
10]. Extensive periods of calibration under the same
WS2013 running conditions followed, including NR cal-
ibrations using neutrons from a deuterium-deuterium
(DD) beam [11, 12], and low-energy ER calibrations using
3
H beta decay [13]. This novel calibration program has
markedly extended the understanding of the LXe detec-
tion medium for low-energy interactions; the sensitivity
of the WIMP searches has consequently improved, par-
ticularly for low-mass WIMPs.
In preparation for the WIMP-search exposure reported
here (WS2014–16), the anode, gate, and cathode grid
electrodes underwent a campaign of “conditioning” in
cold Xe gas, during which each electrode’s applied volt-
age was elevated just above the onset of sustained dis-
charge and maintained for a multiday period, akin to
the burn-in period often employed in room-temperature
proportional counter commissioning [14–18]. The goal of
this campaign was to improve the voltages at which the
electrodes could be biased. As a result, the measured
electron extraction efficiency (i.e. the fraction of electrons
which promptly cross the liquid–gas interface) increased
from (49±3)% in WS2013 to (73±4)% in WS2014–16.
Following the conditioning campaign and extensive cali-
brations at the new operating voltages, WS2014–16 ran
from September 11, 2014 until May 2, 2016, during which
time the detector conditions were kept uniform. The
long-term behavior of the LXe temperature and pres-
sure varied by less than 0.5 K and 10
−2
bar. The elec-
tron lifetime in the LXe was typically stable for long
durations and above 1 ms, significantly longer than the
maximum electron drift time of ∼400 µs. Periods of low
(<500 µs) electron lifetime are excluded from this analy-
sis (including an extended period from March 24 to June
2, 2015), as were periods in which detector-stability pa-
rameters (e.g. pressure, temperature, liquid level, recir-
culation flow rate) deviated by more than a few percent
over short time scales. The WS2014–16 exposure consists
of 332.0 live days.
Though the grid conditioning campaign achieved the
goal of an increased electron extraction efficiency, it was
observed during calibrations that electron drift trajec-
tories were significantly altered from the near-vertical
paths seen in WS2013. In WS2013, due to field cage
geometry alone (similar to [19]), electrons emitted near
the periphery of the cathode grid, at a starting radius
of ∼24 cm, were directed slightly radially inwards during
their upward drift, exiting the liquid surface at an S2 ra-
dius (r
S2
) of ∼20 cm. In WS2014–16, a stronger radial
effect is seen. Electrons of the same cathode-edge start-
ing radius (∼24 cm) exit the liquid surface at ∼10 cm;
the strength of this effect varies with both azimuth and
date. These observations are consistent with a nonuni-
form and time-varying negative charge density in the
polytetrafluoroethylene (PTFE) panels which define the
radial boundary of the active volume. This PTFE charge
is understood as resulting from exposure to coronal dis-
charge during the grid conditioning. The VUV photons
produced in this process can liberate PTFE electron-hole
pairs. As the holes in PTFE have a significantly higher
mobility than the electrons [20, 21], the applied electric
field preferentially removes holes, resulting in a buildup of
net negative charge over long time scales. The observed
charge densities and transport time scales are consistent
with values in the literature [22, 23].
A time-dependent mapping between true recoil posi-
tion and the “observed S2 coordinates” of {x
S2
, y
S2
, and
drift time z
S2
} is required for interpretation of the data,
necessitating the construction of an electric field model.
The comsol Multiphysics package [24] is used to build
a 3-D electrostatic model of the LUX detector, including
a heterogeneous and date-specific charge density in the
PTFE panels. This charge density is fit to data from
regular (∼weekly)
83m
Kr [25–27] calibrations, each pro-
ducing ∼10
6
events of uniform true recoil position within
the active volume. The heterogeneous PTFE charge den-
sity is modeled by dividing the PTFE surface into a grid
3
of 42 sections (seven sections in height, six in azimuth),
each section having a variable uniform charge density.
These 42 individual electrostatic charge densities are var-
ied through a Metropolis-Hastings Markov Chain Monte
Carlo algorithm fitting procedure [28, 29], minimizing
the difference in {x
S2
, y
S2
, z
S2
} distribution boundaries
between simulation and data. Field and charge maps
are updated on a monthly basis, although the variation
time scale is observed to be longer. The average PTFE
charge density is observed to increase in magnitude over
the course of the exposure, starting at −3.6 µC/m
2
and
asymptotically approaching −5.5 µC/m
2
. In the WIMP-
search analysis, comparisons of data to models of signal
and background are most naturally performed in the ob-
served S2 coordinates of {x
S2
, y
S2
, z
S2
}. Data are kept in
these observed S2 coordinates, while the true recoil po-
sitions of simulated data are mapped into this space us-
ing field models mentioned above. Comparisons between
the observed and modeled spatial distributions (see Ap-
pendix A), discussed later, show excellent agreement.
A generic feature of dual-phase TPCs is that measured
S1 and S2 signals from a monoenergetic source will vary
according to the vertex position of the interaction. For
S1, this is due to spatially varying geometrical condi-
tions that affect the efficiency for detecting S1 photons.
In LUX, this detection efficiency is larger for photons
emitted close to the cathode, and smaller for photons
emitted close to the liquid surface (a variation of around
30%). For S2, a similar variation results from the loss
of electrons to electronegative impurities in the LXe (a
date-dependent variation of around 20%–50%). The vari-
ations in S1 and S2 due to these geometrical effects are
independent of the incident particle type and deposited
energy. In WS2013, a position-dependent correction map
for these effects was derived in a straightforward manner,
by measuring the spatial variation in S1 and S2 from a
monoenergetic
83m
Kr calibration source.
In WS2014–16, this picture is complicated because the
spatially varying electric field magnitude influences the
recombination of electron-ion pairs, changing the yields
of photons and electrons emitted at an interaction ver-
tex before the geometrical effects come into play. As the
electric field magnitude is increased, fewer photons and
more electrons escape the interaction [30]. For the 50
to 600 V/cm field variation over the fiducial region rele-
vant to this analysis, the average light yield for a 5 keV
ER event falls by 15%, while average charge yield rises
by the same amount. The scale of variation is less pro-
nounced for lower-energy ER events [31, 32]. For a 5 keV
NR event, the field-induced changes in light and charge
yield are smaller, at the level of 5% [33]. The observed
total spatial variation in S1 and S2 from a monoener-
getic calibration source is therefore a combination both
of field effects and geometrical effects. The geometrical
effects are independent of particle type and energy de-
position, but the field effects depend strongly on these
factors. Therefore, a position-dependent correction map
can only be universally applied to all observed signals if
it corrects for geometrical effects only.
Several techniques are employed to separate the ge-
ometrical effects from the field effects, enabling the de-
sired correction of observed signals for geometrical effects
alone. The field effects remain in the observed science
data, and are similarly included in the background and
signal models for interpretation.
Two calibration tools enable the construction of
geometry-only correction maps. The first is
83m
Kr, which
decays in two steps: 32.1 keV and 9.4 keV. These steps
are separated by a decay constant of 154 ns, thereby pro-
ducing two S1 pulses. While the variation in size of these
S1 pulses depends on several factors, the variation in the
ratio of the two depends only on the applied field [27].
Second, the field effect for low-energy electronic recoils
is extremely weak [34]. Observed variations in the posi-
tion of the
3
H spectral maximum (2.5 keV) are therefore
almost entirely due to geometrical effects alone. Lever-
aging the
83m
Kr S1 ratio that depends on field alone, and
low-energy
3
H response that depends nearly on geometry
alone, geometry-only corrections are constructed.
The italicized quantities S1 and S2 indicate signal
amplitudes that have been corrected for geometrical ef-
fects; S1 is normalized to the center of the active xenon,
while S2 is normalized to the top of the active xenon.
Using these quantities, gain factors g
1
and g
2
are de-
fined through the expectation values hS1 i = g
1
n
γ
and
hS2 i = g
2
n
e
, given n
γ
initial photons and n
e
initial elec-
trons leaving the interaction site. The g
1
and g
2
values
in WS2014–16 are found using a set of monoenergetic
electronic-recoil sources as in [9], and are observed to vary
slightly over the course of the exposure, independent of
the field variation. The g
2
value varies within the range
of 18.92 ±0.82 to 19.72 ±2.39 phd per liquid electron; g
1
gradually falls from 0.100 ±0.002 to 0.097 ±0.001 phd
per photon. Here, “phd” indicates “photons detected,”
differing from the more commonly used unit of photo-
electrons (phe) through a small factor representing the
probability of a single photon to produce multiple phe in
a PMT cathode [35]. Using ˆn
γ
≡ S1/g
1
and ˆn
e
≡ S2/g
2
,
the ER combined energy scale (CES) is constructed as
E
ces
≡ (ˆn
γ
+ ˆn
e
) × 13.7 eV [36]. This observable is inde-
pendent of electric field because of the experimentally ob-
served anticorrelation of n
γ
and n
e
[37–39]. Spatial varia-
tions in the E
ces
peak position of a monoenergetic source
are therefore due to geometrical effects only, and are used
as a cross-check to verify the accuracy of the geometry-
only corrections. For all dates during the WS2014–16
run, the E
ces
peak position of
83m
Kr (41.5 keV) varies by
less than 1% within the fiducial volume. A cross-check
using the E
ces
peak of
131m
Xe (164 keV) gives a spatial
variation of 1.8%.
The electric field dependencies of S1 and S2 yields are
included in the analysis by dividing the WS2014–16 ex-
posure into “exposure segments”, each having its own
ER and NR detector-response model. There are 16 such
segments, constructed by dividing the exposure into four
bins of drift time (related to event depth) and four bins of
4
date. Within each exposure segment, the field magnitude
is considered to be constant and uniform. Boundaries in
date are September 11, 2014; January 1, 2015; April 1,
2015; October 1, 2015; May 2, 2016. Boundaries in drift
time are 40, 105, 170, 235, 300 µs. Periodic
3
H calibra-
tions provide each of the 16 exposure segments with a
unique calibration set from which to construct a unique
individual response model. These 16 response models
take the form of parameter variations of the Noble El-
ement Simulation Technique (NEST) model [33], which
captures both the LXe microphysics of signal production
and the detector physics of signal collection. Fits are
performed by comparing the measured ER band (median
and 10–90 percentile width in the {S1, log
10
(S2)} plane
as in Fig. 1) with that predicted by the response model,
in the range of 0–50 phd, which roughly corresponds to
an energy range of 0–10 keVee. Specific to each exposure
segment, two model parameters are varied during these
fits: the electric field magnitude, and the recombination
fluctuation parameter F
r
(see [31, 33, 34, 40]). Parame-
ters that describe the detector as a whole (e.g., g
1
, elec-
tron extraction efficiency, and S2 gas gain), are allowed
to vary while constrained to be equal for all exposure
segments within a given date bin. In each exposure seg-
ment, the measured ER band median differs from the
model band median by less than 1% for all S1. The
16 electric field magnitudes found through these fits are
consistent with the values earlier obtained from the elec-
trostatic field models. This last point deserves emphasis,
because the two techniques for estimating electric field
magnitude are completely independent: the electrostatic
field model is based on the observed electron drift paths
alone, while the NEST fits are based on the S1 and S2
amplitudes alone.
Neutron calibrations with the DD source were per-
formed in each date bin. For each individual exposure
segment, the best-fit parameters from the corresponding
ER calibration are applied to the NEST NR model. The
resulting NR models show excellent agreement with cali-
brations, such that the NR band medians of correspond-
ing models and calibrations differ by less than 2.6% for
all S1. As in [9], the overall energy scale in the response
models is fixed by fitting the NEST NR model to a sepa-
rate in situ energy calibration using tagged neutron mul-
tiple scatters [11, 12]. As before, we conservatively as-
sume NR light yield to be zero below 1.1 keV, the lowest
energy at which NR light yield was measured in [11]. The
16 ER and 16 NR models are then used within a profile
likelihood ratio (PLR) method [41] to search for evidence
of dark-matter scattering events. It can be seen from the
light-dashed curves in Fig. 1, representing extrema of the
16 ER and NR models, that the scale of model variation
is small and diminishes towards the energy threshold.
Events consisting of a single scatter within the active
LXe are selected according to several criteria: a single S2
preceded by a single S1, an S1 threshold of 2 PMT coinci-
dence, and an upper threshold for the summed pulse area
outside S1 and S2 within the trigger window. This last
selection removes triggers during high single-extracted-
electron activity following large-S2 events [9, 42], and
results in 99.0% efficiency when applied to
3
H calibra-
tion data for WS2014–16. The S2 threshold is set to
200 phd (raw uncorrected pulse area) to avoid events
for which the {x
S2
, y
S2
} position uncertainty is high.
Events for which S2 > 10
4
phd, S1 > 50 phd, log
10
(S2) <
median
NR
− 5σ
NR
or log
10
(S2) > median
ER
+ 3σ
ER
(boundaries evident in Fig. 1) are considered far from
the region of interest and are ignored.
A fiducial volume in drift time is defined as 40–300 µs
(date-independent). Each of the four date bins has
a uniquely defined radial fiducial selection boundary,
3.0 cm radially inward from the measured PTFE sur-
face position for that date bin in observed S2 coordi-
nates, {x
S2
, y
S2
, z
S2
}. The wall position, a function of
{φ
S2
, z
S2
}, is measured with
210
Pb subchain events that
originate on the PTFE surface. The fiducial mass is de-
termined by scaling the 250 kg of active LXe by the ac-
ceptance fraction of
83m
Kr events through the fiducial-
selection criteria. The time-averaged fiducial masses for
the date bins are 105.4, 107.2, 99.2, and 98.4 kg, in
chronological order. A 3% systematic uncertainty across
all dates is estimated through comparison with accep-
tance fractions of
3
H calibration data, of similarly uni-
form distribution in true recoil position.
We apply additional pulse-quality cuts to eliminate
pathological pulses which would otherwise be incorrectly
classified as single-scatter interactions. The first of these
populations is a class of energy depositions in the gaseous
xenon (“gas events”), in which the entire gas event is
classified as an S2 pulse. A cut targeting these pulses
is formed by requiring σ
S2
> 0.4 µs, where σ
S2
is the
width resulting from a Gaussian fit to the pulse wave-
form. This cut has an acceptance of >90% at the S2
threshold of 200 phd, rising to >99% for S2 > 800 phd.
The second pathological population is events in which
two S2 pulses occur close together and are classified as
a single S2 pulse (“merged multiple scatters”). Merged
multiple scatters are rejected with cuts on the {x
S2
, y
S2
}
position reconstruction goodness-of-fit (this cut flagging
multiple scatters separated in x, y) and on the ratio of
σ
S2
to the time between the cumulative 1% and 50% area
fraction (this cut flagging vertices separated in z). The
combined efficiency for these cuts, calculated by applying
these cuts to a population of known single-scatter
3
H S2
waveforms, is >70% at the 200 phd S2 threshold, rising
to >95% for detected S2 > 1000 phd. A summary of all
efficiencies is shown as a function of NR energy in Fig. 2.
For implementation in the PLR, a background model
consisting of three classes of events is constructed: events
of typical LXe charge and light yield, events affected by
proximity to the PTFE surface, and accidental coinci-
dences of isolated S1 and S2 pulses.
A background model representing recoils of typical
charge and light yield is constructed much as in [9]. A
counting of detector materials [7] informs a Geant4-
based LUXSim [43] simulation. Two types of ER back-