Observation of Excess Electronic Recoil Events in XENON1T

E. Aprile,1 J. Aalbers,2 F. Agostini,3 M. Alfonsi,4 L. Althueser,5 F. D. Amaro,6 V. C. Antochi,2

E. Angelino,7 J. R. Angevaare,8 F. Arneodo,9 D. Barge,2 L. Baudis,10 B. Bauermeister,2 L. Bellagamba,3

M. L. Benabderrahmane,9 T. Berger,11 A. Brown,10 E. Brown,11 S. Bruenner,8 G. Bruno,9 R. Budnik,12, ∗

C. Capelli,10 J. M. R. Cardoso,6 D. Cichon,13 B. Cimmino,14 M. Clark,15 D. Coderre,16 A. P. Colijn,8, †

J. Conrad,2 J. P. Cussonneau,17 M. P. Decowski,8 A. Depoian,15 P. Di Gangi,3 A. Di Giovanni,9 R. Di Stefano,14

S. Diglio,17 A. Elykov,16 G. Eurin,13 A. D. Ferella,18, 19 W. Fulgione,7, 19 P. Gaemers,8 R. Gaior,20

M. Galloway,10, ‡ F. Gao,1 L. Grandi,21 C. Hasterok,13 C. Hils,4 K. Hiraide,22 L. Hoetzsch,13 J. Howlett,1

M. Iacovacci,14 Y. Itow,23 F. Joerg,13 N. Kato,22 S. Kazama,23, § M. Kobayashi,1 G. Koltman,12 A. Kopec,15

H. Landsman,12 R. F. Lang,15 L. Levinson,12 Q. Lin,1 S. Lindemann,16 M. Lindner,13 F. Lombardi,6

J. Long,21 J. A. M. Lopes,6, ¶ E. Lopez Fune,20 C. Macolino,24 J. Mahlstedt,2 A. Mancuso,3 L. Manenti,9

A. Manfredini,10 F. Marignetti,14 T. Marrodan Undagoitia,13 K. Martens,22 J. Masbou,17 D. Masson,16

S. Mastroianni,14 M. Messina,19 K. Miuchi,25 K. Mizukoshi,25 A. Molinario,19 K. Mora,1, 2 S. Moriyama,22

Y. Mosbacher,12 M. Murra,5 J. Naganoma,19 K. Ni,26 U. Oberlack,4 K. Odgers,11 J. Palacio,13, 17 B. Pelssers,2

R. Peres,10 J. Pienaar,21 V. Pizzella,13 G. Plante,1 J. Qin,15 H. Qiu,12 D. Ramırez Garcıa,16 S. Reichard,10

A. Rocchetti,16 N. Rupp,13 J. M. F. dos Santos,6 G. Sartorelli,3 N. Sarcevic,16 M. Scheibelhut,4 J. Schreiner,13

D. Schulte,5 M. Schumann,16 L. Scotto Lavina,20 M. Selvi,3 F. Semeria,3 P. Shagin,27 E. Shockley,21, ∗∗

M. Silva,6 H. Simgen,13 A. Takeda,22 C. Therreau,17 D. Thers,17 F. Toschi,16 G. Trinchero,7 C. Tunnell,27

M. Vargas,5 G. Volta,10 H. Wang,28 Y. Wei,26 C. Weinheimer,5 M. Weiss,12 D. Wenz,4 C. Wittweg,5

Z. Xu,1 M. Yamashita,23, 22 J. Ye,26, †† G. Zavattini,3, ‡‡ Y. Zhang,1 T. Zhu,1 and J. P. Zopounidis20

(XENON Collaboration)

X. Mougeot29

1Physics Department, Columbia University, New York, NY 10027, USA2Oskar Klein Centre, Department of Physics, Stockholm University, AlbaNova, Stockholm SE-10691, Sweden

3Department of Physics and Astronomy, University of Bologna and INFN-Bologna, 40126 Bologna, Italy4Institut fur Physik & Exzellenzcluster PRISMA, Johannes Gutenberg-Universitat Mainz, 55099 Mainz, Germany

5Institut fur Kernphysik, Westfalische Wilhelms-Universitat Munster, 48149 Munster, Germany6LIBPhys, Department of Physics, University of Coimbra, 3004-516 Coimbra, Portugal

7INAF-Astrophysical Observatory of Torino, Department of Physics,University of Torino and INFN-Torino, 10125 Torino, Italy

8Nikhef and the University of Amsterdam, Science Park, 1098XG Amsterdam, Netherlands9New York University Abu Dhabi, Abu Dhabi, United Arab Emirates

10Physik-Institut, University of Zurich, 8057 Zurich, Switzerland11Department of Physics, Applied Physics and Astronomy, Rensselaer Polytechnic Institute, Troy, NY 12180, USA

12Department of Particle Physics and Astrophysics, Weizmann Institute of Science, Rehovot 7610001, Israel13Max-Planck-Institut fur Kernphysik, 69117 Heidelberg, Germany

14Department of Physics “Ettore Pancini”, University of Napoli and INFN-Napoli, 80126 Napoli, Italy15Department of Physics and Astronomy, Purdue University, West Lafayette, IN 47907, USA

16Physikalisches Institut, Universitat Freiburg, 79104 Freiburg, Germany17SUBATECH, IMT Atlantique, CNRS/IN2P3, Universite de Nantes, Nantes 44307, France

18Department of Physics and Chemistry, University of L’Aquila, 67100 L’Aquila, Italy19INFN-Laboratori Nazionali del Gran Sasso and Gran Sasso Science Institute, 67100 L’Aquila, Italy

20LPNHE, Sorbonne Universite, Universite de Paris, CNRS/IN2P3, Paris, France21Department of Physics & Kavli Institute for Cosmological Physics, University of Chicago, Chicago, IL 60637, USA

22Kamioka Observatory, Institute for Cosmic Ray Research, and Kavli Institute for the Physics and Mathematicsof the Universe (WPI), the University of Tokyo, Higashi-Mozumi, Kamioka, Hida, Gifu 506-1205, Japan

23Kobayashi-Maskawa Institute for the Origin of Particles and the Universe, and Institute for Space-EarthEnvironmental Research, Nagoya University, Furo-cho, Chikusa-ku, Nagoya, Aichi 464-8602, Japan

24Universite Paris-Saclay, CNRS/IN2P3, IJCLab, 91405 Orsay, France25Department of Physics, Kobe University, Kobe, Hyogo 657-8501, Japan

26Department of Physics, University of California San Diego, La Jolla, CA 92093, USA27Department of Physics and Astronomy, Rice University, Houston, TX 77005, USA

28Physics & Astronomy Department, University of California, Los Angeles, CA 90095, USA29CEA, LIST, Laboratoire National Henri Becquerel, CEA-Saclay 91191 Gif-sur-Yvette Cedex, France

(Dated: July 1, 2020)

We report results from searches for new physics with low-energy electronic recoil data recordedwith the XENON1T detector. With an exposure of 0.65 tonne-years and an unprecedentedly low

arX

iv:2

006.

0972

1v2

[he

p-ex

] 3

0 Ju

n 20

20

2

background rate of 76± 2 stat events/(tonne × year × keV) between 1–30 keV, the data enables com-petitive searches for solar axions, an enhanced neutrino magnetic moment using solar neutrinos, andbosonic dark matter. An excess over known backgrounds is observed below 7 keV, rising towardslower energies and prominent between 2–3 keV. The solar axion model has a 3.5σ significance, and athree-dimensional 90% confidence surface is reported for axion couplings to electrons, photons, andnucleons. This surface is inscribed in the cuboid defined by gae < 3.7× 10−12, gaeg

effan < 4.6× 10−18,

and gaegaγ < 7.6× 10−22 GeV−1, and excludes either gae = 0 or gaegaγ = gaegeffan = 0. The neutrino

magnetic moment signal is similarly favored over background at 3.2σ and a confidence interval ofµν ∈ (1.4, 2.9) × 10−11 µB (90% C.L.) is reported. Both results are in strong tension with stellarconstraints. The excess can also be explained by β decays of tritium at 3.2σ significance with acorresponding tritium concentration in xenon of (6.2± 2.0)× 10−25 mol/mol. Such a trace amountcan be neither confirmed nor excluded with current knowledge of production and reduction mecha-nisms. The significances of the solar axion and neutrino magnetic moment hypotheses are decreasedto 2.1σ and 0.9σ, respectively, if an unconstrained tritium component is included in the fitting.With respect to bosonic dark matter, the excess favors a monoenergetic peak at (2.3 ± 0.2) keV(68% C.L.) with a 3.0σ global (4.0σ local) significance over background. This analysis sets themost restrictive direct constraints to date on pseudoscalar and vector bosonic dark matter for mostmasses between 1 and 210 keV/c2.

PACS numbers:

Keywords:

Keywords: Dark Matter, Direct Detection, Xenon

I. INTRODUCTION

A preponderance of astrophysical and cosmological ev-idence suggests that most of the matter content in theUniverse is made up of a rarely interacting, non-luminouscomponent called dark matter [1]. Although several hy-pothetical dark matter particle candidates have been pro-posed with an assortment of couplings, masses, and de-tection signatures, dark matter has thus far eluded directdetection. The XENON1T experiment [2], employing aliquid-xenon time projection chamber (LXe TPC), wasprimarily designed to detect Weakly Interacting MassiveParticle (WIMP) dark matter. Due to its unprecedent-edly low background rate, large target mass, and lowenergy threshold, XENON1T is also sensitive to inter-actions from alternative dark matter candidates and toother physics beyond the Standard Model (SM). Here wereport on searches for (1) axions produced in the Sun,(2) an enhancement of the neutrino magnetic momentusing solar neutrinos, and (3) pseudoscalar and vectorbosonic dark matter, including axion-like particles anddark photons.

∗ Also at Simons Center for Geometry and Physics and C. N. YangInstitute for Theoretical Physics, SUNY, Stony Brook, NY, USA† Also at Institute for Subatomic Physics, Utrecht University,

Utrecht, Netherlands‡ galloway@physik.uzh.ch§ Also at Institute for Advanced Research, Nagoya University,

Nagoya, Aichi 464-8601, Japan¶ Also at Coimbra Polytechnic - ISEC, Coimbra, Portugal∗∗ ershockley@uchicago.edu†† jiy171@ucsd.edu‡‡ Also at INFN, Sez. di Ferrara and Dip. di Fisica e Scienze della

Terra, Universita di Ferrara, via G. Saragat 1, Edificio C, I-44122Ferrara (FE), Italy

The XENON1T experiment operated underground atthe INFN Laboratori Nazionali del Gran Sasso (LNGS)from 2016–2018, utilizing a dual-phase LXe TPC with a2.0-tonne active target to search for rare processes. Aparticle interaction within the detector produces bothprompt scintillation (S1) and delayed electrolumines-ence (S2) signals. These light signals are detected byarrays of photomultiplier tubes (PMTs) on the top andbottom of the active volume, and are used to determinethe deposited energy and interaction position of an event.The latter allows for removing background events nearthe edges of the target volume (e.g., from radioactivityin detector materials) through fiducialization. The S2/S1ratio is used to distinguish electronic recoils (ERs), pro-duced by, e.g., gamma rays (γs) or beta electrons (βs),from nuclear recoils (NRs), produced by, e.g., neutrons orWIMPs, allowing for a degree of particle identification.The ability to determine scatter multiplicity enables fur-ther reduction of backgrounds, as signals are expected tohave only single energy deposition.

In this paper, we report on searches for ER signals withdata acquired from February 2017 to February 2018, atime period referred to as Science Run 1 (SR1) [3]. As thevast majority of background comes from ER events, wesearch for excesses above a known background level. Theanalysis is carried out in the space of reconstructed en-ergy, which exploits the anti-correlation of S1 and S2 sig-nals by combining them into a single energy scale [4], thusreducing the statistical fluctuations from electron-ion re-combination [5]. Both S1 and S2 signals are correctedto disentangle position-dependent effects, such as lightcollection efficiency (LCE) and electron attachment toelectronegative impurities. S1 is reconstructed using cor-rected signals from all PMTs (cS1). For the S2 recon-struction, only the bottom PMT array is used (cS2b) be-cause it features a more homogeneous light collection [3].

3

The energy region of interest (ROI) is (1, 210) keV.The paper is organized as follows. In Sec. II we

present the theoretical background and signal model-ing of the beyond-the-SM channels considered in thissearch. We describe the data analysis in Sec. III, in-cluding the data selection, background model, and sta-tistical framework. In Sec. IV, upon observation of alow-energy excess in the data, we present a hypothesisof a previously-unconsidered background component, tri-tium, and then report the results of searches for solaraxions, an anomalous neutrino magnetic moment, andbosonic dark matter. We end with further discussionof our findings and a summary of this work in Secs. Vand VI, respectively. The presence of the excess mo-tivated further scrutiny of the modeling of dominantbackgrounds, the details of which we present in the Ap-pendix.

II. SIGNAL MODELS

This section describes the physics channels we searchfor in this work. In Sec. II A, we motivate the search ofsolar axions, presenting their production mechanisms inthe Sun and the detection mechanism in LXe TPCs, andsummarize two benchmark axion models. In Sec. II B, weintroduce the search for an anomalous neutrino magneticmoment, which would enhance the neutrino-electron elas-tic scattering cross section at low energies. In Sec. II C,we discuss the signals induced by bosonic dark matterincluding pseudoscalar and vector bosons, examples ofwhich are axion-like particles and dark photons, respec-tively. Expected energy spectra of these signals in theXENON1T detector are summarized at the end of thissection.

For all signal models presented below, the theoret-ical energy spectra in a LXe TPC were converted tothe space of reconstructed energy by accounting for de-tector efficiency and resolution, summarized in Fig. 1.The efficiency is shown in Fig. 2 and discussed inSec. III A. For the energy resolution, the theoretical spec-tra were smeared using a Gaussian distribution withenergy-dependent width, which was determined usingan empirical fit of mono-energetic peaks as describedin [2, 4]. The energy resolution σ is given by

σ(E) = a ·√E + b · E, (1)

with a = (0.310± 0.004)√

keV and b = 0.0037 ± 0.0003.

A. Solar Axions

As a solution to the strong CP problem in quan-tum chromodynamics (QCD), Peccei and Quinn postu-lated a mechanism that naturally gives rise to a Nambu-Goldstone boson, the so-called axion [6–8]. In additionto solving the strong CP problem, QCD axions are also

well-motivated dark matter candidates, with cosmologi-cal and astrophysical bounds requiring their mass to besmall (typically � keV) [9–13]. On account of this massconstraint, dark matter axions produced in the early Uni-verse cannot be observed in XENON1T. However, so-lar axions would emerge with—and in turn deposit—energies in the keV range [14–16], the precise energiesto which XENON1T was designed to be most sensi-tive. An observation of solar axions would be evidence ofbeyond-the-SM physics, but would not by itself be suffi-cient to draw conclusions about axionic dark matter.

We consider three production mechanisms that con-tribute to the total solar axion flux: (1) Atomic recom-bination and deexcitation, Bremsstrahlung, and Comp-ton (ABC) interactions [14, 17], (2) a mono-energetic14.4 keV M1 nuclear transition of 57Fe [15], and (3)the Primakoff conversion of photons to axions in theSun [18, 19]. The ABC flux scales with the axion-electroncoupling gae as

ΦABCa ∝ g2

ae (2)

and was taken from [14]. The 57Fe flux scales with an ef-fective axion-nucleon coupling geff

an = −1.19g0an + g3

an andis given by [20, 21]

Φ57Fea =

(ka

kγ

)3

× 4.56× 1023(geffan )2 cm−2s−1, (3)

where g0/3an are the isoscalar/isovector coupling constants

and ka and kγ are the momenta of the produced axionand photon, respectively. The Primakoff flux scales withthe axion-photon coupling gaγ and is given by [22]

dΦPrima

dEa=

(gaγ

GeV−1

)2(Ea

keV

)2.481

e−Ea/(1.205 keV)

× 6× 1030 cm−2s−1keV−1,

(4)

where Ea is the energy of the axion. All three flux com-ponents could be detected in XENON1T via the axioelec-tric effect – the axion analog to the photoelectric effect –which has a cross section that scales with axion-electroncoupling gae and is given by [20, 23–25]

σae = σpeg2

ae

β

3E2a

16παm2e

(1− β2/3

3

), (5)

where σpe is the photoelectric cross section, β and Ea

are the velocity and energy of the axion, respectively, αis the fine structure constant, and me is the mass of theelectron. Combining the production and detection mech-anisms, we are able to constrain the values of |gae| (ABC),∣∣gaeg

effan

∣∣ (57Fe), and |gaegaγ| (Primakoff)1. We consider

1 We drop the absolute value notation for the remainder of thispaper.

4

these three observables independently in the analysis, lestwe implicitly assume any particular axion model. Still,it is important to note that these values are indeed re-lated to each other and to the axion mass under differentmodels.

For QCD axions, the mass ma is related to the decayconstant fa via

ma '6× 106 GeV

faeV/c

2, (6)

and the axion couplings to matter are mostlymodel-dependent. We describe here two benchmarkclasses of QCD axion models: Dine-Fischler-Srednicki-Zhitnitsky (DFSZ) [26, 27], in which axions couple toelectrons at tree level, and Kim-Shifman-Vainshtein-Zhakharov (KSVZ) [28, 29], where couplings to leptonsoccur only at loop level. For this reason the ABCflux is dominant in DFSZ models, while the Primakoffflux is dominant in KSVZ models. Since the axioelec-tric cross section scales with the axion-electron coupling,XENON1T is in general more sensitive to DFSZ-type ax-ions.

In DFSZ models the axion-electron coupling is givenby

gae =me

3facos2 βDFSZ, (7)

where

tan(βDFSZ) =

(Xu

Xd

)1/2

, (8)

and Xu and Xd are the Peccei-Quinn (PQ) charges ofthe up and down quarks, respectively [20, 30, 31]. Thecouplings to quarks take on a similar expression withrespect to βDFSZ. The axion-nucleon couplings g0

an andg3

an are functions of Xu, Xd, and fa, and can be foundin [30, 32]. For a DFSZ axion, it follows that gae andgeff

an are both non-zero in general, as they are connectedvia βDFSZ and fa. The axion-photon coupling does notdepend on the PQ charges but is directly related to theaxion decay constant (and thus the mass):

gaγ =α

2πfa

(E

N− 2

3

4 + z

1 + z

), (9)

where z = mu/md, mu/d are the respective massesof the up/down quarks, and E/N represent themodel-dependent electromagnetic/color anomalies of theaxial current associated with the axion field [33]. It istypically assumed that E/N = 8/3 in DFSZ models.

In KSVZ models, the PQ charges of the SM quarksvanish, and there is no βDFSZ-like parameter. Theaxion-electron coupling strength, induced by radiativecorrections, depends on the axial current [30, 34]:

gae =3α2Nme

4π2fa

(E

Nlnfa

me− 2

3

4 + z + w

1 + z + wln

Λ

me

),

(10)

where w = mu/ms, ms is the mass of the strange quark;Λ is the cutoff of the QCD confinement scale. The

isoscalar/isovector axion-nucleon couplings g0/3an do not

depend on the PQ charges and are also found in [30, 32].The axion-photon coupling is given by Eq. (9). For KSVZmodels a benchmark value of E/N = 0 is often used, butmany values are possible [35].

As mentioned above, no particular axion model is as-sumed in the analysis itself; the three flux components areconsidered completely independent of each other. Since,in principle, it is possible for all three components tobe present at the same time, our solar axion model in-cludes three unconstrained parameters for the differentcomponents. Were a signal observed, the results of thethree-component analysis could then be used to constraindifferent axion models and possibly infer the axion mass.This approach also implies that the results hold gen-erally for solar axion-like particles, which do not havestrict relationships between the couplings, as describedin Sec. II C.

The expected spectra from solar axions withgae = 5× 10−12, gaγ = 2× 10−10 GeV−1, andgeff

an = 10−6 are shown in Fig. 1 (left) with before/afterdetector effects indicated by unshaded/shaded curves,respectively. The rate of the ABC component is pro-portional to gae

4; the 57Fe component is proportional to(gaeg

effan )2; and the Primakoff component is proportional

to (gaegaγ)2.

B. Neutrino Magnetic Moment

In the SM, neutrinos are massless, and therefore with-out a magnetic dipole moment. However, the observationof neutrino oscillation tells us that neutrinos have massand the SM must be extended, thus implying a magneticmoment of µν ∼ 10−20 µB [36, 37], where µB is the Bohrmagneton. Larger values of µν have been considered the-oretically and experimentally [37–39]. Interestingly, inaddition to providing evidence of beyond-SM physics, theobservation of a µν & 10−15 µB would suggest that neu-trinos are Majorana fermions [37]. Currently the moststringent direct detection limit is µν < 2.8× 10−11 µBfrom Borexino [39], and indirect constraints based on thecooling of globular cluster and white dwarfs are an orderof magnitude stronger at ∼ 10−12 µB [31, 40, 41].

An enhanced magnetic moment would increase theneutrino scattering cross-sections at low energies (onboth electrons and nuclei), and thus could be observableby low-threshold detectors such as XENON1T. Here weonly consider the enhancement to elastic scattering onelectrons, given by [42]

dσµdEr

= µ2να

(1

Er− 1

Eν

), (11)

where Er is the electronic recoil energy and Eν is theenergy of the neutrino. Note that Eq. (11) assumes free

5

0 5 10 15Energy [keV]

0

100

200

300

400E

vent

s/(t

yke

V) gae = 5×10-12

geffan = 1×10-6

ga = 2×10-10 GeV-1

ABC axion57Fe axionPrimakoff axion

0 5 10 15 20 25 30Energy [keV]

= 7×10-11 Bgae = 2×10-13

20 keV/c2 ALP

FIG. 1. Left: Expected signal in energy space for ABC solar axions with a coupling gae = 5 × 10−12 (blue), for solar axionsproduced from the de-excitation of 57Fe with coupling geff

an = 1× 10−6 (red), and for solar axions produced from the Primakoffeffect with coupling gaγ = 2 × 10−10 (orange). Right: Signature of an enhanced neutrino magnetic moment with magnitude7 × 10−11 µB (green) and a 20 keV/c2 ALP with coupling constant gae = 2 × 10−13 (purple). Both the true deposited energyspectra in a xenon detector without efficiency loss (unshaded) and the expected observed spectra in XENON1T including thespecific detector resolution and efficiency (shaded) are shown.

electrons; small corrections need to be made for the elec-tron binding energies at O(keV) energies.

We search for an anomalous magnetic moment usingsolar neutrinos, predominantly those from the proton-proton (pp) reaction [43]. The expected energy spectrumfor µ = 7× 10−11 µB is shown in Fig. 1 (right), which wascalculated by folding the expected solar neutrino flux [43]with Eq. (11) and applying a step-function approxima-tion to account for the electron binding energies. In theenergy range considered here, this approximation agreeswell with more detailed calculations [44]. Note that thissignal would be added to the SM neutrino elastic scat-tering spectrum, which we treat as a background as de-scribed in Sec. III B.

C. Bosonic Dark Matter

Axion-like particles (ALPs), like QCD axions, arepseudoscalar bosons, but with decay constant and parti-cle mass (Eq. (7)) decoupled from each other and insteadtaken as two independent parameters. This decouplingallows for ALPs to take on higher masses than QCD ax-ions; however, it also implies that ALPs do not solve thestrong CP problem.

ALPs are viable dark matter candidates [45], andcould be absorbed in XENON1T via the axioelectric ef-fect (Eq. (5)) like their QCD counterparts. AssumingALPs are non-relativistic and make up all of the lo-cal dark matter (density ρ ∼ 0.3 GeV/cm3 [46]), the ex-pected signal is a mono-energetic peak at the rest mass ofthe particle, ma, with an event rate given by (see [25, 47])

R ' 1.5× 1019

Ag2

ae

(ma

keV/c2

)(σpe

b

)kg−1d−1, (12)

where A is the average atomic mass of the detectormedium (A ≈ 131 u for xenon). The rate coefficient fromour calculation is consistent with [48] for the dark matterdensity used in this work.

In addition to the pseudoscalar ALPs, XENON1T isalso sensitive to vector bosonic dark matter, of whichdark photons are a common example. Dark photons cancouple weakly with SM photons through kinetic mix-ing [49] and be absorbed with cross section σV givenby [50]

σV 'σpe

βκ2, (13)

where σpe, α, and β are the same as in Eq. (5), and κparameterizes the strength of kinetic mixing between thephoton and dark photon. Similarly to Eq. (12), by fol-lowing the calculation in [25], the rate for non-relativisticdark photons in a detector reduces to

R ' 4.7× 1023

Aκ2

(keV/c2

mV

)(σpe

b

)kg−1d−1, (14)

where mV is the rest mass of the vector boson. Like thepseudoscalar above, absorption of a vector boson wouldalso result in a monoenergetic peak broadened by theenergy resolution of the detector, but with a rate that isinversely proportional to the particle mass. The expectedspectrum for a 20 keV/c2 ALP with gae = 2 × 10−13 isshown in Fig. 1 (right). Vector bosons have the samesignature as ALPs, but the rate scales differently withmass (see Eqs. (12, 14)).

6

III. DATA ANALYSIS

This section describes the data-analysis methods em-ployed to search for the aforementioned signals. Theevent-selection criteria and their overall efficiency, thedetection efficiency, as well as the determination of fu-dicialization and ROI are given in Sec. III A. Sec. III Bdetails each component of our background model, thepredictions of which are consistent with the results of abackground-only fit to the data. In Sec. III C, we definethe likelihood used for the fitting and discuss the statis-tical framework.

A. Data Selection

The data-selection criteria for this search are similarto [3], with the selections and efficiencies optimized andreevaluated for the different parameter space and ex-tended energy range. For an event to be considered valid,an S1-S2 pair is required. A valid S1 demands coinci-dent signals in at least 3 PMTs, and a 500 photoelectron(PE) threshold is imposed on the S2 size. This S2 thresh-old is more stringent than that in [3] in order to rejectbackground events originating from radon daughters onthe TPC surface [51]. Since signal events are expectedto deposit energy only once in the detector, events withmultiple interaction sites are removed. A variety of se-lection criteria are applied to ensure data quality and acorrect S1 and S2 pairing, which is detailed in [52]. Theefficiencies and uncertainties of the selection criteria areestimated in a procedure similar to [52], and the cumula-tive selection efficiency is determined using an empiricalfit of the data. The average cumulative selection effi-ciency over the (1, 210) keV region is (91.2± 0.3)%.

The combined efficiency of detection and event selec-tion with uncertainties is shown in Fig. 2. The detec-tion efficiency, dominated by the 3-fold coincidence re-quirement of S1s, was estimated using both a data-drivenmethod of sampling PMT hits from S1s in the 20–100 PErange and an independent study based on simulation oflow-energy S1 waveforms [52]. The difference betweenthe two methods (∼ 3% average relative difference in thedrop-off region) was considered as a systematic uncer-tainty. This efficiency was then converted from S1 toreconstructed energy using the detector-response modeldescribed in [51], accounting for additional uncertaintiessuch as the photon yield. The S2 efficiency can be as-sumed to be unity for the energies considered here [52].

Events with energies between (1, 210) keV are selectedfor this search, with the lower bound determined by re-quiring the total efficiency be larger than 10%, shown inFig. 2, and the upper bound chosen due to an increas-ing γ−ray background from detector materials, whichis difficult to model due to large uncertainties on itsspectral shape. The same 1042 kg cylindrical fiducialvolume as in [53] was used to reduce the surface andmaterial backgrounds. After event selection and strict

100 101 102

Energy [keV]

0.0

0.2

0.4

0.6

0.8

1.0

Eff

icie

ncy

SelectionDetection

FIG. 2. Efficiency as a function of energy. The dashed (dot-ted) line refers to detection (selection) efficiency, while theblue curve and band illustrate the total efficiency and the as-sociated 1-σ uncertainty, respectively. The detection thresh-old is indicated by the right bound of the gray shaded region.

fiducialization, the surface backgrounds, accidental coin-cidences, and neutrons make up less than 0.003% of thetotal events (< 0.3% below 7 keV), and thus are negligiblefor this search. Additionally, events within 24 hours fromthe end of calibration campaigns using injected radioac-tive sources were removed due to residual source activity.The final effective SR1 live time is 226.9 days and thusthe total exposure is 0.65 tonne-years.

B. Background Model

Within the (1, 210) keV ROI and the 1042 kg fiducialvolume, ten different components were used to model thebackground and fit the data, as listed in Tab. I and illus-trated in Fig. 3.

Six components, numbers i–vi in Tab. I, exhibit con-tinuous energy spectra and were modeled based on eithertheoretical predictions or GEANT4 Monte Carlo simula-tions, and the rest are mono-energetic peaks that weremodeled as Gaussian functions of known energies andresolution. The spectrum of each background componentconsiders the detector energy resolution and efficiencyloss in the same way as the signal model construction inSec. II. The rates of the background components are con-strained, when possible, by independent measurementsand extracted by the fit.

The β-decay of 214Pb, the dominant continuous back-ground, is present due to 222Rn emanation into theLXe volume by materials. An additional backgroundcomes from intrinsic 85Kr, which is subdominant dueto its removal via cryogenic distillation [54, 55]. Theshape of these spectra, particularly at low energies, canbe affected by atomic screening and exchange effects,as well as by nuclear structure [56, 57]. The β de-

7

No. Component Expected Events Fitted Events

i 214Pb (3450, 8530) 7480 ± 160

ii 85Kr 890 ± 150 773 ± 80

iii Materials 323 (fixed) 323 (fixed)

iv 136Xe 2120 ± 210 2150 ± 120

v Solar neutrino 220.7 ± 6.6 220.8 ± 4.7

vi 133Xe 3900 ± 410 4009 ± 85

vii 131mXe 23760 ± 640 24270 ± 150

viii

125I (K) 79 ± 33 67 ± 12125I (L) 15.3 ± 6.5 13.1 ± 2.3125I (M) 3.4 ± 1.5 2.94 ± 0.50

ix 83mKr 2500 ± 250 2671 ± 53

x

124Xe (KK) 125 ± 50 113 ± 24124Xe (KL) 38 ± 15 34.0 ± 7.3124Xe (LL) 2.8 ± 1.1 2.56 ± 0.55

TABLE I. Summary of components in the backgroundmodel B0 with expected and fitted number of events in the0.65 tonne-year exposure of SR1. Both numbers are withinthe (1, 210) keV ROI and before efficiency correction. Seetext for details on the various components.

cays of 214Pb and 85Kr are first forbidden non-uniqueand first forbidden unique transitions, respectively; how-ever spectra from the IAEA LiveChart (Nuclear DataServices database) [58] are based on calculations of al-lowed and forbidden unique transitions, neither of whichincludes exchange effects [59]. Likewise, models fromGEANT4 [60] include only the screening effect; however,its implementation displays a non-physical discontinuityat low energies [59, 61]. For this work, we performeddedicated theoretical calculations to account for possiblelow-energy discrepancies from these effects in 214Pb and85Kr spectra. These calculations are described in detailin Appendix A.

The activity of 214Pb can be constrained using in situmeasurements of other nuclei in the same decay chain.These constraints, described in [51], place a lower boundof 5.1 ± 0.5µBq/kg (from coincident 214BiPo) and up-per bound of 12.6 ± 0.8 µBq/kg (218Po α-decays). Forthis analysis, we leave the normalization of 214Pb rateunconstrained and use the fit to extract the activ-ity. The background-only fit results give an event rateof 63.0 ± 1.3 events/(tonne×year×keV) (abbreviated asevents/(t·y·keV) for the rest of paper) over the ROI af-ter efficiency correction. With the 11% branching ra-tio (from [62]) and the spectrum of 214Pb decay to theground state (calculated in Appendix A), the 214Pb ac-tivity is evaluated to be 11.1± 0.2stats ± 1.1sys µBq/kgthroughout SR1 and is well within the upper/lowerbounds. The 10% systematic uncertainty is mainly fromthe aforementioned branching ratio [62].

The 85Kr decay rate is inferred from dedicated mea-surements of the isotopic abundance of 85Kr/natKr(2 × 10−11 mol/mol) and the natKr concentration evo-lution in LXe [63]. The same measurements also al-

10-1

100

101

102

103

104

Eve

nts/

(ty

keV)

SR1 (226.9 days)

214Pb85KrMaterials136Xe

Solar 133Xe131mXe125I

83mKr124XeB0

SR1 data

Energy [keV]

22

10-1

100

101

102

103

104

Eve

nts/

(ty

keV)

SR1a (55.8 days)

22

10-1

100

101

102

103

104

Eve

nts/

(ty

keV)

SR1b (171.2 days)

0 25 50 75 100 125 150 175 200Energy [keV]

22

FIG. 3. Fit to the SR1 data set using the likelihood frame-work described in Sec. III C and the background model B0

in Sec. III B. The top panel shows the entire SR1 spectrum,the sum of the two spectra below it. The middle (bottom)panel shows SR1a (SR1b), which contains more (less) neutron-activated backgrounds. SR1a and SR1b are fit simultaneously.The light green (yellow) band indicates the 1-σ (2-σ) residu-als. The summed fit results are listed in Tab. I.

8

low for the time-dependence of the 85Kr decay rate tobe taken into account. The average rate of 85Kr is7.4± 1.3 events/(t·y·keV) over the ROI in SR1.

An additional background arises from γ emissionsfrom radioimpurities in detector materials that induceCompton-scattered electrons; however, this backgroundis subdominant in the ROI due to the strict fiducial vol-ume selection. The rate from materials is constrainedby radioassay measurements [64] and predicted by simu-lations [65] to be 2.7± 0.3 events/(t·y·keV). This back-ground is modeled by a fixed, flat component in the fit.

One of the continuous backgrounds considered was136Xe, a 2νββ emitter intrinsic to xenon. This com-ponent has an increasing rate as a function of energyover the ROI. It was constrained in the fit according tothe predicted rate and associated uncertainties on (1) a136Xe isotopic abundance of (8.49± 0.04stat ± 0.13sys)%as measured by a residual gas analyzer [66], (2) the re-ported half-life [67], and (3) the spectral shape [68, 69].

The first observation of two-neutrino double electroncapture (2νECEC) of 124Xe was recently reported us-ing mostly the same SR1 dataset (but different selectioncuts) as used in this analysis [70] and is treated as abackground here. In [70] we considered the dominantbranching ratio of 2νECEC, the capture of two K-shellelectrons inducing a peak at 64.3 keV. It is also possibleto capture a K-shell and L-shell electron (36.7 keV) or twoL-shell electrons (9.8 keV) with decreasing probability, ascalculated in [71]. For this analysis, the event selectionand consideration of time dependence allow us to includeall three peaks in the background model. The predictedrates of the peaks are taken from an updated half-life [72]with fixed branching ratios from [71]; the overall rate wasnot constrained in the fit since the half-life was derivedfrom the same dataset.

Three additional backgrounds were included forneutron-activated isotopes: 133Xe (β), 131mXe (internalconversion (IC)), and 125I (electron capture (EC)). Theseisotopes were produced after neutron calibrations and de-cayed away with half-lives of O(10) days. The IC decay of131mXe produces a mono-energetic peak at 164 keV [73],which, along with the other mono-energetic backgrounds,has the same signature as a bosonic dark matter signal.It was well-constrained using its half-life and known datesof neutron calibration. 133Xe decays to an excited statewith a dominant branching ratio and emits an 81 keVprompt γ upon de-excitation [74], resulting in a continu-ous spectrum starting at ∼ 75 keV, given the energy reso-lution. The rate was also constrained in the fit with pre-diction obtained using time dependence. The third acti-vated isotope 125I, a daughter of 125Xe, decays via EC ofK-shell, L-shell, and M-shell with decreasing probabilityand produces peaks at 67.3 keV, 40.4 keV, and 36.5 keV,respectively [75]. Similar to 124Xe 2νECEC, all threepeaks of 125I EC are included in the background modelwith the fixed branching ratios from [75]. The 125I contri-bution was constrained using a model based on the timeevolution of 125Xe throughout SR1, as detailed in [70].

During SR1, a background from 83mKr (IC) waspresent due to a trace amount of 83Rb (EC,T1/2 ∼ 86 days) in the xenon recirculation system, whichpresumably was caused by a momentary malfunction ofthe source valve and confirmed using half-life measure-ments. 83mKr decays via a two-step scheme (second stepT1/2 ∼ 154 ns) [76] resulting in many of these eventsbeing removed by the multi-site selections mentioned inSec. III A; however, due to the short half-life of the secondstep, these decays are often unresolved in time and hencecontribute as a mono-energetic peak at 41.5 keV. Thiscomponent was also constrained using a time-evolutionmodel.

Elastic scattering of solar neutrinos off electrons isexpected to contribute subdominantly over the entireROI. The expected energy spectrum was obtained usingthe standard neutrino flux in the Large Mixing AngleMikheyev-Smirnov-Wolfenstein (LMA-MSW) model andcross section given by the SM [43, 77]. Based on ratecalculations of neutrino-electron scattering in xenon asgiven in [78], a 3% uncertainty was assigned and used toconstrain the solar neutrino rate in the fit.

We denote the background model described aboveas B0. This model was used to fit the SR1 data in(1, 210) keV by maximizing the likelihood constructed inSec. III C. The fit results are consistent with predictions,as summarized in Tab. I. The best fit of B0 is shownin Fig. 3, where the top panel is the full SR1 data setand the bottom two panels are partitions of SR1, whichwere fit simultaneously to include the temporal infor-mation of several backgrounds (see Sec. III C). This fitgives a background rate of 76±2 events/(t·y·keV) withinthe (1, 30) keV region after efficiency correction with theassociated uncertainty from the fitting. Fig. 4 shows azoom in (0, 30) keV region of Fig. 3 with a finer binning.

In Sec. IV we raise the possibility of an additionalbackground component, the β decay of tritium, thatwe did not include while constructing the backgroundmodel. A validated β−decay spectrum from the IAEALiveChart [58, 79] was used for the 3H model, as de-scribed in Appendix A. We treat the possible tritiumcontribution separately from B0 for reasons discussed inSec. IV A.

C. Statistical Method

An unbinned profile likelihood method is employed inthis analysis. The likelihood is constructed as

9

L(µs,µb,θ) = Poiss(N |µtot)

×N∏i

∑j

µbjµtot

fbj (Ei,θ) +µsµtot

fs(Ei,θ)

×∏m

Cµm(µbm)×

∏n

Cθn(θn), (15)

µtot ≡∑j

µbj + µs,

where µs and µb are the expected total signal and back-ground events. Both µb and θ are nuisance parame-ters, where θ includes shape parameters for the efficiencyspectral uncertainty (see Fig. 2), as well as peak loca-tion uncertainties, specifically for 124Xe (3 peaks), 83mKr,and 131mXe. Having largely subdominant event rates,the 3 peak locations from 125I EC are fixed at theirexpected positions to save computation time. Index iruns over all observed events with the total number ofN (=42251 events), and Ei corresponds to the energy ofthe ith event. fb and fs are the background and sig-nal probability distribution functions, and index j runsover all the background components. Cµ and Cθ are con-straints on the expected numbers of background eventsand the shape parameters. Index m runs over back-grounds including 85Kr, solar neutrino, 136Xe, 83mKr,125I, 133Xe, and 131mXe, while index n is for all six shapeparameters.

Due to time-dependent backgrounds, the SR1 dataset is divided into two partitions: SR1a consisting ofevents within 50 days following the end of neutron cal-ibrations and SR1b containing the rest, with effectivelive times of 55.8 and 171.2 days, respectively. Includingthis time information allows for better constraints on thetime-independent backgrounds and improves sensitivityto bosonic dark matter, especially as the time-dependentbackground from 133Xe impacts a large fraction of itssearch region. The full likelihood is then given by

L = La × Lb, (16)

where La and Lb are evaluated using Eq. (15) in eachpartition. Nuisance parameters that do not change withtime, along with all of the signal parameters, are sharedbetween the two partitions. The constant nuisance pa-rameters are:

• the efficiency parameter, which is dominated by de-tection efficiency and does not change with time.

• The 214Pb component, which was determined tohave a constant rate in time using detailed studiesof the α-decays of the 222Rn and 218Po as well asthe coincidence signature of 214Bi and 214Po.

• The solar neutrino rate, which would vary by ∼3 %between the two partitions on account of Earth’sorbit around the Sun. This is ignored due to thesubdominant contribution from this source.

• The decay rates of the intrinsic xenon isotopes136Xe and 124Xe, as well as the Compton contin-uum from materials.

The remaining parameters all display time dependenciesthat are modeled in the two partitions.

The test statistic used for the inference is defined as

q(µs) = −2lnL(µs, ˆµb,

ˆθ)

L(µs, µb, θ), (17)

where (µs, µb, θ) is the overall set of signal and nuisance

parameters that maximizes L, while L(µs, ˆµb,ˆθ) is the

maximized L by profiling nuisance parameters with aspecified signal parameter µs. The statistical signifi-cance of a potential signal is determined by q(0). Forthe neutrino magnetic moment and bosonic dark mat-ter searches, a modified Feldman-Cousins method in [80]was adopted in order to derive 90% C.L. bounds with theright coverage. We report an interval instead of an upperlimit if the global significance exceeds 3σ. For bosonicdark matter this corresponds to 4σ local significance onaccount of the look-elsewhere effect, which is not presentfor the neutrino magnetic moment search. The 3σ signif-icance threshold only serves as the transition point be-tween reporting one- and two-sided intervals, and wasdecided prior to the analysis to ensure correct coverage.A two-sided interval does not necessarily indicate a dis-covery, which in particle physics generally demands a 5σsignificance and absence of compelling alternate explana-tions.

Since the solar axion search is done in the space ofgae, gaegaγ, and gaeg

effan , we extend its statistical analy-

sis to three dimensions. For this search, we use a stan-dard profile likelihood construction where the true 90th-percentile of the test statistic (Eq. (17)) was evaluatedat several points on a three-dimensional grid and in-terpolated between points to define a 3D ‘critical’ vol-ume of true 90-percent threshold values. By construc-tion, the intersection of this volume with the test statis-tic q(gae, gaegaγ, gaegan) defines a three-dimensional 90%C.L. volume in the space of the three axion parameters.In Sec. IV we report the two-dimensional projections ofthis volume, found by profiling over the third respectivesignal component.

IV. RESULTS

When compared to the background model B0, thedata display an excess at low energies, as shown inFig. 4. The excess rises with decreasing energy, peaksnear 2–3 keV, and then subsides to within ±1σ of thebackground model near 1–2 keV. Within 1–7 keV, thereare 285 events observed in the data compared to an ex-pected 232± 15 events from the background-only fit, a3.3σ Poissonian fluctuation. Events in this energy re-gion are uniformly distributed in the fiducial volume. Al-though the statistics are limited, the excess events do not

10

Energy [keV]0

20

40

60

80

100

120E

vent

s/(t

yke

V)

B0

SR1 data

0 5 10 15 20 25 30Energy [keV]

22

FIG. 4. A zoomed-in and re-binned version of Fig. 3 (top),where the data display an excess over the background modelB0. In the following sections, this excess is interpreted undersolar axion, neutrino magnetic moment, and tritium hypothe-ses.

exhibit a clear time dependence. More detailed studiesof the temporal distribution of these events are describedin Sec. IV E.

Several instrumental backgrounds and systematic ef-fects were excluded as possible sources of the excess.Accidental coincidences (AC), an artificial backgroundfrom detector effects, are expected to be spatially uni-form, but are tightly constrained to have a rate of< 1 event/(t·y·keV) based on the rates of lone signalsin the detector, i.e., S1s (S2s) that do not have a corre-sponding S2 (S1) [51]. Surface backgrounds have a strongspatial dependence [51] and are removed by the fiducial-ization (1.0 tonne here vs. 1.3 tonnes in [3], correspond-ing to a radial distance from the TPC surface of & 11 cm)along with the stricter S2 threshold cut. Both of thesebackgrounds also have well-understood signatures in the(cS1, cS2b) parameter space that are not observed here,as shown in Fig. 5.

The detection and selection efficiencies were verifiedusing 220Rn calibration data. The β decay of 212Pb, adaughter of 220Rn, was used to calibrate the ER responseof the detector, and thus allows us to validate the effi-ciency modeling with a high-statistics data set. Similarlyto 214Pb, the model for 212Pb was calculated to accountfor atomic screening and exchange effects, as detailed inAppendix A. A fit to the 220Rn data with this model andthe efficiency parameter described in Sec. III C is shownin Fig. 6 for a 1-tonne fiducial volume, where good agree-ment is observed (p-value = 0.50). Additionally, the S1and S2 signals of the low-energy events in backgrounddata were found to be consistent with this 220Rn dataset, as shown in Fig. 5. This discounts threshold effectsand other mismodeling (e.g., energy reconstruction) aspossible causes for the excess observed in Fig. 4.

0 10 20 30 40 50 60 70cS1 [PE]

100

200

400

1000

2000

4000

8000

cS2 b

[PE

]

108

54

9

3

13

1

11

6

12

7

2

keV

FIG. 5. Distribution of low energy events (black dots) inthe (cS1, cS2b) parameter space, along with the expectedsurface (purple) and AC (orange) backgrounds (1σ band).220Rn calibration events are also shown (density map). All thedistributions are within the one-tonne fiducial volume. Graylines show isoenergy contours for electronic recoils, where theexcess is between the 1 and 7 keV contours, highlighted inblue.

0

200

400

600

800

Eve

nts/

keV

Best-fitSR1 220Rn data

0 2 4 6 8 10 12 14Energy [keV]

-2

2

FIG. 6. Fit to 220Rn calibration data with a theoretical β-decay model (see Appendix A) and the efficiency nuisanceparameter, using the same unbinned profile likelihood frame-work described in Sec. III C. This fit suggests that the effi-ciency shown in Fig. 2 describes well the expected spectrumfrom 214Pb, the dominant background at low energies.

Uncertainties in the calculated spectra were consid-ered, particularly for the dominant 214Pb background.More details can be found in Appendix A, but we brieflysummarize them here. A steep rise in the spectrum at lowenergies could potentially be caused by exchange effects;however this component is accurate to within 1% andtherefore negligible with respect to the observed excess.The remaining two components, namely the endpoint en-

11

ergy and nuclear structure, tend to shift the entire β dis-tribution, rather than cause steep changes over a range of∼ 10 keV. Conservatively, the combined uncertainty fromthese two components is +6% in the 1–10 keV region, asdescribed in the Appendix A. In comparison, a +50% un-certainty at 2–3 keV on the calculated 214Pb spectrum,as constrained by the higher energy component, wouldbe needed to make up the excess.

We also considered backgrounds that might in princi-ple be present in trace amounts. First, low-energy X-rays from 127Xe EC, as seen in [81] and [82], are ruledout for a number of reasons. 127Xe is produced from cos-mogenic activation at sea level; given the short half-lifeof 36.4 days and the fact that the xenon gas was under-ground for O(years) before the operation of XENON1T,127Xe would have decayed to a negligible level. Indeed,high-energy γs that accompany these X-rays were notobserved, and with their O(cm) mean free path in LXethey could not have left the O(m)-sized TPC undetected.For these reasons, we conclude that 127Xe was no longerpresent during SR1.

Another potential background is 37Ar, which decaysvia EC to the ground state of 37Cl, emitting a 2.8 keVX-ray with a 0.90 branching ratio [83]. We consider twopossible 37Ar contributions: one from its presence in thexenon gas before filling, and another from a possible airleak that could provide a constant source of argon.

Argon, a noble gas, would not be removed by thegetter in the purification system; however, it would beremoved by the online 85Kr distillation that occurredbefore SR1 (see Sec. III B) and by its decay (T1/2 =

35.0 days [83]). Based on the injection of 37Ar as a cal-ibration source at the end of XENON1T (∼ 8 monthsafter the end of SR1) [84], its removal via distillationand decay was determined to have an effective time con-stant of 1.8 days [85]. With a 90-day distillation cam-paign during commissioning, natAr was therefore greatlysuppressed in the early days of XENON1T. The isotopicabundance of 37Ar is ∼ 10−20 [86], implying that evenif there were 1 ppm of natAr in the xenon originally, the37Ar concentration would have been reduced to a negli-gible level before the acquisition of science data.

As far as the air leak possibility is concerned,based on the number of excess events observedand assuming a (sea-level) activation rate of(92 ± 13) atoms/(kgAr ·day) [86], we estimate thatany such leak would have to introduce ∼ 10−4 kg ofargon per day, corresponding to a total air leak of∼ 3 L/day. This is ruled out by the natKr concentration,which increased by < 1 ppt/year during SR1 as informedby RGMS measurements [63]. Based on a similar studyin XENON100 [87] and scaling for the larger activevolume, a 1-ppt/year increase in natKr would correspondto an air leak of ∼ 1 L/year in XENON1T, 3 ordersof magnitude lower than the ∼3 L per day required toaccount for the excess observed. We thus exclude 37Aras an explanation for the excess.

We also considered an additional background that has

never been observed before in LXe TPCs: the β emis-sion of tritium2, which has a Q-value of 18.6 keV and ahalf-life of 12.3 years [91]. Tritium may be introducedfrom predominantly two sources: cosmogenic activationof xenon during above-ground exposure [92] and emana-tion of tritiated water (HTO) and hydrogen (HT) fromdetector materials due to its cosmogenic and anthro-pogenic abundance. In contrast to 127Xe and 37Ar, thetritium hypothesis cannot be ruled out. Below we dis-cuss several possibilities of introducing tritium into thedetector and the uncertainties involved in its productionand reduction processes in an attempt to estimate itsconcentration.

A. Tritium Hypothesis

In order to determine the hypothetical concentration oftritium required to account for the excess, we search fora 3H ‘signal’ on top of the background model B0. Whencompared to B0, the tritium hypothesis is favored at 3.2σand the fitted rate is 159 ± 51 events/(t·y) (68% C.L.),which would correspond to a 3H/Xe concentration of(6.2 ± 2.0) × 10−25 mol/mol. As tritium is expected tobe removed by the xenon purification system, this con-centration would correspond to an equilibrium value be-tween emanation and removal. This analysis is summa-rized in Fig. 7 (a), where the best fits under both thealternate hypothesis (B0 + 3H) and null hypothesis (B0)are shown.

Due to its minute possible concentration, long half-lifewith respect to our exposure, and the fact that it decaysthrough a single channel, we are unable to confirm thepresence of tritium from SR1 data directly. We thereforetry to infer its concentration from both initial conditionsand detector performance parameters.

A tritium background component from cosmogenic ac-tivation of target materials has been observed in sev-eral dark matter experiments at rates compatible withpredictions [93], although it has never before been de-tected in xenon. From exposure to cosmic rays duringabove-ground storage of xenon, we estimate a conser-vative upper limit on the initial 3H/Xe concentrationof < 4 × 10−20 mol/mol, based on GEANT4 activationrates [92] and assuming saturation activity. At this stage,tritium will predominantly take the form of HTO, giventhe measured ppm water impurities in the xenon gas andequilibrium conditions [94, 95]. Through xenon gas han-dling prior to filling the detector (i.e., condensation ofH2O/HTO on the walls of the cooled xenon-storage ves-sel) and purification via a high-efficiency getter with a

2 Tritium in the form of tritiated methane has been used for cal-ibration of LXe TPCs [88–90], including XENON100, but wasnot used as a calibration source in XENON1T. Following theXENON100 tritium calibration, neither the xenon gas nor thematerials that came into contact with the tritiated methane wereused in XENON1T.

12

hydrogen removal unit [2, 96], we expect the concentra-tion to be reduced to < 10−27 mol/mol, thus reachingnegligible levels with respect to the observed excess.

Tritium may also be introduced as HTO and HT viatheir respective atmospheric abundances. Water and hy-drogen, and therefore tritium, may be stored inside ma-terials, such as the TPC reflectors and the stainless steelof the cryostat. This type of source is expected to em-anate from detector and subsystem materials at a ratein equilibrium with its removal via getter purification.Tritium can be found in water at a concentration of(5 − 10) × 10−18 atoms of 3H for each atom of hydro-gen in H2O [97–99]. Here we assume the same abun-dance of 3H in atmospheric H2 as for water3. Usingthe best-fit rate of tritium and the HTO atmosphericabundance, a combined (H2O + H2) impurity concentra-tion of & 60 ppb in the LXe target would be required tomake up the excess. Since water impurities affect opti-cal transparency, the high light yield in SR1 indicates anO(1)-ppb H2O concentration [65, 102], thus implying amaximum contribution from HTO to the 3H/Xe concen-tration of ∼ 1× 10−26 mol/mol. With respect to H2, wecurrently have no direct or indirect measurements of itsconcentration in the detector. Instead, we consider thatO2-equivalent, electronegative impurities must reach sub-ppb levels in SR1, given the achieved electron lifetime of∼ 650µs (at 81 V/cm) [3, 103]. Thus for tritium to makeup the excess requires a factor ∼ 100 higher H2 concen-tration than that of electronegative impurities. Underthe above assumptions, tritium from atmospheric abun-dance appears to be an unlikely explanation for the ex-cess. However, we do not currently have measurementsof the equilibrium H2 emanation rate in XENON1T, andthus the HT concentration cannot be sufficiently quanti-fied.

In conclusion, possible tritium contributions from cos-mogenic activation or from HTO in SR1 appear too smallto account for the excess, while it is not possible to in-fer the concentration of HT. In addition, various fac-tors contribute further to the uncertainty in estimat-ing a tritium concentration within a LXe environment,such as its unknown solubility and diffusion properties,as well as the possibility that it may form molecules otherthan HT and HTO. Since the information and measure-ments necessary to quantify the tritium concentrationare not available, we can neither confirm nor exclude itas a background component. Therefore, we report re-sults using the background model B0, and then sum-marize how our results would change if tritium were in-cluded as an unconstrained background component. Allreported constraints are placed with the validated back-ground model B0 (i.e., without tritium).

3 Although geographical and temporal HT abundances in theatmosphere vary due to anthropogenic activities, HT thatreaches the Earth’s surface undergoes exchange to HTO within5 hours [100, 101].

B. Solar Axion Results

We search for ABC, 57Fe, and Primakoff axions simul-taneously. Under this signal model, B0 is rejected at3.5σ, a value determined using toy Monte Carlo meth-ods to account for the three parameters of interest inthe alternative hypothesis. A comparison of the best fitsunder the alternative hypothesis (B0 + axion) and nullhypothesis (B0) can be found in Fig. 7 (b).

A three-dimensional confidence surface (90% C.L.) wascalculated in the space of gae vs. gaegaγ vs. gaeg

effan . This

surface is inscribed in the cuboid given by

gae < 3.7× 10−12

gaegeffan < 4.6× 10−18

gaegaγ < 7.6× 10−22 GeV−1.

While easy to visualize, this cuboid is more conservative(it displays over-coverage) than the three-dimensionalconfidence volume it encloses and does not describethe correlations between the parameters. The correla-tion information can be found in Fig. 8, which showsthe two-dimensional projections of the surface. Forthe ABC–Primakoff and ABC–57Fe projections (Fig. 8top and middle, respectively), gae can be easily fac-tored out of the y-axis to plot gaγ vs gae (top) andgeff

an vs gae (middle). This is not as straightforward forthe 57Fe-Primakoff projection (Fig. 8 bottom). Alsoshown in Fig. 8 are constraints from other axion searches[81, 82, 104–108] as well as predicted values from thebenchmark QCD models DFSZ and KSVZ.

Fig. 8 (top) is extracted from the projection onto theABC–Primakoff plane. Since the ABC and Primakoffcomponents are both low-energy signals, the 90% con-fidence region is anti-correlated in this space and — dueto the presence of the low-energy excess — suggests ei-ther a non-zero ABC component or non-zero Primakoffcomponent. Since our result gives no absolute lowerbound on gae, the limit on the product gaegaγ cannotbe converted into a limit on gaγ on its own; i.e., with

gaegaγ=7.6 × 10−22 GeV−1, gaγ → ∞ as gae → 0, asshown in Fig. 8 (top).

Fig. 8 (middle) is taken from the projection onto theABC–57Fe plane. Unlike the ABC-Primakoff case, thesetwo signals are not degenerate; however, they still displayanti-correlated behavior. The reason for this is that thetest statistic q (Eq. (17)) is relatively large with smallgae, meaning small changes in the 57Fe rate about thebest-fit make q cross the 90% threshold value and thusbe excluded by our 90% confidence surface. There isno statistical significance (< 1σ) for the presence of a14.4 keV peak from 57Fe axions.

Lastly, Fig. 8 (bottom) shows the projection onto thePrimakoff-57Fe plane, where no correlation is observed.The Primakoff and 57Fe components are both allowed tobe absent as long as there is a non-zero ABC component.This means that, of the three axion signals considered,

13

0 5 10 15 20 25 30Energy [keV]

0

20

40

60

80

100

120

140E

vent

s/(t

yke

V)(a) Tritium

H0: B0

H1: B0 + 3H3H

0 5 10 15 20 25 30Energy [keV]

0

20

40

60

80

100

120

140

Eve

nts/

(ty

keV)

(b) Solar axionH0: B0

H1: B0 + axionABC axion57Fe axionPrimakoff axion

0 5 10 15 20 25 30Energy [keV]

0

20

40

60

80

100

120

140

Eve

nts/

(ty

keV)

(c) Neutrino magnetic momentH0: B0

H1: B0 +

0 5 10 15 20 25 30Energy [keV]

0

20

40

60

80

100

120

140

Eve

nts/

(ty

keV)

(d) Solar axion vs tritium backgroundH0: B0 + 3HH1: B0 + 3H + axion

ABC axion57Fe axionPrimakoff axion3H

FIG. 7. Fits to the data under various hypotheses. The null and alternative hypotheses in each scenario are denoted by gray(solid) and red (solid) lines, respectively. For the tritium (a), solar axion (b), and neutrino magnetic moment (c) searches,the null hypothesis is the background model B0 and the alternative hypothesis is B0 plus the respective signal. Contributionsfrom selected components in each alternative hypothesis are illustrated by dashed lines. Panel (d) shows the best fits for anadditional statistical test on the solar axion hypothesis, where an unconstrained tritium component is included in both nulland alternative hypotheses. This tritium component contributes significantly to the null hypothesis, but its best-fit rate isnegligible in the alternative hypothesis, which is illustrated by the orange dashed line in the same panel.

the ABC component is the most consistent with the ob-served excess.

The three projections of Fig. 8 can be used to recon-struct the three-dimensional 90% confidence surface forgae, gaegaγ, and gaeg

effan . Due to the presence of an ex-

cess at low energy, this surface would suggest either anon-zero ABC component or a non-zero Primakoff com-ponent. However, the coupling values needed to explainthis excess are in strong tension with stellar cooling con-straints [106–110]. The CAST constraints [104] as shownare valid for axion masses below 10 meV/c2 while thosefrom XENON1T and similar experiments hold for all ax-ion masses up to ∼ 100 eV/c2.

As described above, we cannot exclude tritium as anexplanation for this excess. Thus, we report on an ad-ditional statistical test, where an unconstrained tritiumcomponent was added to the background model B0 andprofiled over alongside the other nuisance parameters. Inthis case, the null hypothesis is the background modelplus tritium (B0 + 3H) and the alternative includes thethree axion signal components (B0 + 3H + axion), where

tritium is unconstrained in both cases. The solar axionsignal is still preferred in this test, but its significanceis reduced to 2.1σ. The fits for this analysis are shownin Fig. 7 (d). The tritium component is negligible in thealternate best-fit, but its presence allows for a better fitunder—and thus a reduced significance of rejecting—thenull hypothesis.

C. Neutrino Magnetic Moment Results

When compared to the neutrino magnetic moment sig-nal model, the background model B0 is rejected at 3.2σ.The best-fits of the null (B0) and alternative (B0 + µν)hypotheses for this search are shown in Fig. 7 (c).

The 90% confidence interval for µν from this analysisis given by

µν ∈ (1.4, 2.9)× 10−11 µB ,

and is shown in Fig. 9 along with the constraints fromother searches. The upper boundary of this interval is

14

0 1 2 3 4 5gae 1e 12

10-12

10-11

10-10

10-9

10-8

10-7g a

[GeV

-1]

XENON1T(this work)

solar

LUX

PandaX-II

CAST (ma < 10 meV)

red giants

white dw

arfs

HB stars

DFSZ

KSVZ

0 1 2 3 4 5gae 1e 12

10-10

10-9

10-8

10-7

10-6

10-5

10-4

10-3

geff

an

XENON1T(this work)

solar

LUX

PandaX-II

red giants

white dw

arfs DFSZ

KSVZ

10-23 10-22 10-21

gaega [GeV-1]

10-21

10-20

10-19

10-18

10-17

10-16

10-15

g aeg

eff

an

XENON1T(this work)

CAST (ma < 10 meV)

PandaX-II

DFSZ

KSVZ

FIG. 8. Constraints on the axion-electron gae, axion-photongaγ, and effective axion-nucleon geff

an couplings from a searchfor solar axions. The shaded blue regions show the two-dimensional projections of the three-dimensional confidencesurface (90% C.L.) of this work, and hold for ma < 100 eV/c2.See text for more details on the three individual projections.All three plots include constraints (90% C.L.) from otheraxion searches, with arrows denoting allowed regions, andthe predicted values from the benchmark QCD axion mod-els DFSZ and KSVZ.

very close to the limit reported by Borexino [39], whichis currently the most stringent direct detection constrainton the neutrino magnetic moment. Similar to the solaraxion analysis, if we infer the excess as a neutrino mag-netic moment signal, our result is in strong tension withindirect constraints from analyses of white dwarfs [111]and globular clusters [41].

As in Sec. IV B, we report on the additional statisticaltest where an unconstrained tritium component was in-cluded in both null and alternative hypotheses. In thistest the significance of the neutrino magnetic momentsignal is reduced to 0.9σ.

This is the most sensitive search to date for an en-hanced neutrino magnetic moment with a dark matterdetector, and suggests that this beyond-the-SM signalbe included in the physics reach of other dark matterexperiments.

Borexino Gemma Globularclusters

Whitedwarfs

XENON1T(this work)

10-12

10-11

10-10

[B]

FIG. 9. Constraints (90% C.L.) on the neutrino magnetic mo-ment from this work compared to experiments Borexino [39]and Gemma [112], along with astrophysical limits from thecooling of globular clusters [41] and white dwarfs [111]. Ar-rows denote allowed regions. The upper boundary of the in-terval from this work is about the same as that from Borexinoand Gemma. If we interpret the low-energy excess as a neu-trino magnetic moment signal, its 90% confidence interval isin strong tension with the astrophysical constraints.

D. Bosonic Dark Matter Results

For bosonic dark matter, we iterate over (fixed) massesbetween 1 and 210 keV/c2 to search for peak-like ex-cesses. The trial factors to convert between local andglobal significance were extracted using toy Monte Carlomethods. While the excess does lead to looser constraintsthan expected at low energies, we find no global signifi-cance over 3σ for this search under the background modelB0. We thus set an upper limit on the couplings gae andκ as a function of particle mass.

These upper limits (90% C.L.) are shown in Fig. 10,along with the sensitivity band in green (1σ) and yel-

15

low (2σ). The losses of sensitivity at 41.5 keV and164 keV are due to the 83mKr and 131mXe backgrounds,respectively, and the gains in sensitivity at around 5and 35 keV are due to increases in the photoelectriccross-section in xenon. The fluctuations in our limitare due to the photoelectric cross-section, the logarith-mic scaling, and the fact that the energy spectra dif-fer significantly across the range of masses. For mostmasses considered, XENON1T sets the most stringentdirect-detection limits to date on pseudoscalar and vec-tor bosonic dark matter couplings.

10-1 100 101 102

ma [keV/c2]

10-14

10-13

10-12

10-11

10-10

g ae

XENON1T(this work)

XENON1T(S2-only)

Majorana

XENON100

SuperCDMS

GERDACDEX-1B

PandaX-II

XMASS

EDELWEISS-III

LUX

10-1 100 101 102

mV [keV/c2]

10-17

10-16

10-15

10-14

10-13

10-12

10-11

10-10

kine

tic m

ixin

g

XENON1T(this work)XENON1T

(S2-only)

Majorana

XENON100

SuperCDMS

GERDA

CDEX-1B

XMASS

stellar bounds

EDELWEISS-III

FIG. 10. Constraints on couplings for bosonic pseudoscalarALP (top) and vector (bottom) dark matter, as a function ofparticle mass. The XENON1T limits (90% C.L.) are shown inblack with the expected 1 (2)σ sensitivities in green (yellow).Limits from other detectors or astrophysical constraints arealso shown for both the pseudoscalar and vector cases [50, 81,82, 113–120].

Due to the presence of the excess, we performed an ad-ditional fit using the bosonic dark matter signal model,with the particle mass allowed to vary freely between1.7–3.3 keV/c2. The result gives a favored mass value of(2.3 ± 0.2) keV (68% C.L.) with a 3.0σ global (4.0σ lo-cal) significance over background. A log-likelihood ratiocurve as a function of mass is shown in Fig. 11, along withthe asymptotic 1-σ uncertainty. Since the energy recon-

struction in this region is validated using 37Ar calibrationdata, whose distribution has a mean value within < 1%of the expectation at 2.82 keV [83], this analysis can alsobe used to compare the data to potential mono-energeticbackgrounds in this region.

1.8 2.0 2.2 2.4 2.6 2.8 3.0mass [keV/c2]

0

1

2

3

4

5

6

7

8

qFIG. 11. The -2 logarithmic likelihood ratio for differentbosonic dark matter masses with respect to the best-fit massat 2.3 keV/c2. At each mass, we show the result for the best-fit coupling at that mass. The green band shows an asymp-totic 68% C.L. confidence interval on the bosonic DM mass.As noted in the text, the global significance for this model is3.0 σ.

E. Additional Checks

Here we describe a number of additional checks to in-vestigate the low-energy excess in the context of the tri-tium, solar axion, and neutrino magnetic moment hy-potheses.

The time dependence of events in the (1, 7) keV regionin SR1 was investigated. The rate evolution does notshow a clear preference for one hypothesis over the othersfor several reasons. For one, the event rates have largeuncertainties as a result of the limited statistics and shortexposure time. Additionally, the expected time evolutionof the solar signals (axion and ν magnetic moment) is asubtle ∼ 7 % (peak-to-peak) rate modulation from thechange in Earth-Sun distance; such a small effect is notobservable with our exposure. Similarly, the expectedexponential decay of the tritium rate cannot be observeddue to its long half-life with respect to the duration ofSR1. Therefore, none of the hypotheses is rejected onthe grounds of time dependence.

Since the excess events have energies near our 1 keVthreshold, where the efficiency is ∼ 10%, we consid-ered higher analysis thresholds to check the impact ofthis choice on the results. With the excess most promi-nent between 2 and 3 keV, where the respective detec-tion efficiencies are ∼ 80% and 94%, changing the analy-

16

sis threshold has little impact unless set high enough soas to remove the events in question. This is not well-motivated, given the high efficiency in the region of theexcess. For all thresholds considered (namely, 1.0, 1.6,2.0, 3.0 keV), the solar axion model gives the best fit tothe data. We hence conclude that our choice of analysisthreshold impacts neither the presence nor interpretationof the low-energy excess.

We also checked data from Science Run 2 (SR2), anR&D science run that followed SR1, in an attempt tounderstand the observed excess. Many purification up-grades were implemented during SR2, including the re-placement of the xenon circulation pumps with units that(1) are more powerful, leading to improved purificationspeed, and (2) have lower 222Rn emanation, leading toa reduced 214Pb background rate in the TPC [121, 122],which is further decreased by online radon distillation.The resulting increased purification speed and reducedbackground make SR2 useful to study the tritium hy-pothesis. If the excess were from tritium (or another non-noble contaminant), we would expect its rate to decreasedue to the improved purification; on the other hand, therate of the signal hypotheses would not change with pu-rification speed.

While the SR2 purification upgrades allowed for animproved xenon purity and a reduced background level,the unavoidable interruption of recirculation for the up-grades also led to less stable detector conditions. Thus,in addition to a similar event selection process as SR1in Sec. III A, we removed several periods of SR2 for thisanalysis to ensure data quality. Periods where the elec-tron lifetime changed rapidly due to tests of the purifi-cation system were removed to reduce uncertainty in theenergy reconstruction. We also removed datasets duringwhich a 83mKr source was left open for calibration. Datawithin 50 days of the end of neutron calibrations werealso removed to reduce neutron-activated backgroundsand better constrain the background at low energies. Af-ter the other selections, this data would have only added∼ 10 days of live time; thus, for simplicity, it was re-moved rather than fit separately like the SR1 dataset.With these selections, the effective SR2 live time for thisanalysis is 24.4 days, with an average ER background re-duction of 20% in (1, 30) keV as compared to SR1.

A profile likelihood analysis was then performed onSR2 with a similar background model as SR1, denotedas BSR2. Since we are primarily interested in using thisdata set to test the tritium hypothesis, we focus on thetritium results.

Similarly to SR1, we search for a tritium signalon top of the background model BSR2, and find thatthe background-only hypothesis is slightly disfavored at2.3σ. The SR2 spectrum, along with the fits for the null(BSR2) and alternative (BSR2 + 3H) hypotheses, can befound in Fig. 12. A log-likelihood ratio curve for the tri-tium component is given in Fig. 13, which shows thatthe fitted tritium rate is 320 ± 160 events/(t·y), higherthan that from SR1 but consistent within uncertainties.

0 5 10 15 20 25 30Energy [keV]

0

50

100

150

200

250

Eve

nts/

(ty

keV)

SR2 (24.4 days)H0: BSR2

H1: BSR2 + 3HSR2 data

FIG. 12. A fit to SR2 data if tritium is treated as a signal.The red (gray) line is the fit with (without) tritium in thebackground model.

0 100 200 300 400 500 6003H rate [events/(t y)]

0

1

2

3

4

5

6

q

SR2 LLRSR1 fit (1 )

FIG. 13. The log likelihood ratio curve for the tritium rate inSR2. The orange line and band indicate the best-fit and 1σuncertainty for the tritium rate in SR1. The SR2 fit resultis consistent with SR1, but with a large uncertainty due tolimited statistics.

The rate uncertainty in SR2 is much larger than thatin SR1 due to limited statistics. The solar axion andmagnetic moment hypotheses give similar results, withsignificances ∼ 2σ and best-fit values larger than, butconsistent with, the respective SR1 fit results. Thus theseSR2 studies are largely inconclusive.

Lastly, we also checked these hypotheses in a differ-ent energy region using the so-called ‘S2-only’ approach,where the requirement for an S1 signal is dropped, al-lowing for a ∼ 200 eV energy threshold. XENON1T’s S2-only analysis [115] was used to place limits on the tritiumrate (< 2256 events/(t·y)) and gae (< 4.8×10−12) that arefar greater than, and therefore consistent with, the con-straints derived here. The S2-only analysis is not as sensi-tive to the tritium and axion signals because both spectra

17

peak above keV. On the other hand, many of the pre-dicted signal events from neutrino magnetic moment fallbelow 1 keV as the rate increases with falling energy, sothe S2-only search is more relevant for this hypothesis. Ityields a 90% C.L. one-sided limit of µν < 3.1×10−11 µB ,consistent with the upper boundary of the 90% confi-dence interval obtained in Sec. IV C. Therefore, none ofthe discussed hypotheses are in conflict with the S2-onlyresult.

V. DISCUSSION

We observe an excess at low energies (1–7 keV),where 285 events are observed compared to an expected232± 15 events from the background-only fit to the data.The β decay of tritium is considered as a possible expla-nation, as it has a similar spectrum to that observedand is expected to be present in the detector at somelevel. We are unable to independently confirm the pres-ence of tritium at the O(10−25) mol/mol concentrationrequired to account for the excess, and so treat it sep-arately from our validated background model. If elec-tronic recoils from tritium decay were the source of theexcess, this would be its first indication as an atmosphericsource of background in LXe TPCs. The tritium hypoth-esis clearly represents a possible SM explanation for theexcess, but — based on spectral shape alone — the solaraxion model is the most favored by the data at 3.5σ, al-beit at only ∼ 2σ if one considers tritium as an additionalbackground.

If this excess were a hint of a solar axion, our resultwould suggest either (1) a non-zero rate of ABC axionsor (2) a non-zero rate of both Primakoff and 57Fe axions.If we interpret the excess as an ABC axion signal (i.e.,take gaγ and geff

an to be zero), the required value of gae

is smaller than that ruled out by other direct searchesbut have a clear discrepancy with constraints from indi-rect searches [41, 123]. These constraints are a factor of∼ 5–10 lower than reported here, although subject to sys-tematic uncertainties. It is noteworthy that some of theseastrophysical analyses, while their constraints are stillstronger than direct searches, do in fact suggest an addi-tional source of cooling compatible with axions [109, 123].If the indirect hints and the XENON1T excess were in-deed explained by axions, the tension in gae could berelieved by underestimated systematic uncertainties in,e.g., stellar evolution theory [41] or white dwarf luminos-ity functions [124], or by a larger solar axion flux thanthat given in [14].

Continuing to interpret the excess as a hypotheticalQCD axion signal, we can extend the analysis to makestatements on the axion mass ma under assumptions ofdifferent models, as outlined in Sec. II A. As examples, weconsider a DFSZ model with variable βDFSZ and KSVZmodel with variable electromagnetic anomaly E (for sim-plicity we fix the color anomaly N = 3). Comparingthese two classes of models with our 90% confidence sur-

face, we find that both are consistent with our result fora subset of parameters. For the DFSZ model, we findma ∼ 0.1− 4.1 eV/c2 and cos2 βDFSZ ∼ 0.01− 1 wouldbe consistent with this work. Alternatively, under theKSVZ model ma ∼ 46− 56 eV/c2 and E = 6 would besimilarly consistent. These model-specific mass rangesare not confidence intervals, as their specific assumptionswere not included when constructing Fig. 8. We insteadreport a single, model-independent confidence region onthe couplings to allow comparison with a variety of mod-els, not just the examples mentioned here.

Additionally, we describe a direct search for an en-hanced neutrino magnetic moment. This signal alsohas a similar spectrum to the excess observed, butat 3.2σ displays a lower significance than that fromsolar axions. We report a confidence interval ofµν ∈ (1.4, 2.9)× 10−11 µB (90% C.L.), the upper bound-ary of which is very close to the world-leading direct limitreported by Borexino [39]. This shows that dark mat-ter experiments are also sensitive to beyond-SM physicsin the neutrino sector. Here we only search for an en-hanced neutrino-electron cross-section due to an anoma-lous magnetic moment, but a similar enhancement wouldalso occur in neutrino-nucleus scattering [125]. With thediscrimination capabilities of LXe TPCs to ER and NRevents, it would be interesting to consider this channel infuture searches as well.

If from an astrophysical source, the excess presentedhere is different from the result reported by the DAMAexperiment, which claims that an observed annual mod-ulation of events between 1 and 6 keV might be due toa dark matter signal [126, 127]. We present here a lep-tophilic dark matter model, where WIMPs couple withelectrons through an axial-vector interaction [128]. Thismodel was used to explain the DAMA signal but was re-jected already by the XENON100 experiment [129]. In-terpreting the modulating source observed by DAMA un-der this model, the expected signal rate in the XENON1Tdetector would be more than 2 orders of magnitudehigher than the total event rate we observed, as shown inFig. 14. Consequently, the excess observed in this workis unrelated to the one observed by DAMA.

VI. SUMMARY

We report on searches for new physics using low-energyelectronic recoils in XENON1T. In a search for bosonicdark matter, world-leading constraints are placed on theinteraction strengths of pseudoscalar and vector parti-cles. An excess is observed between 1 and 7 keV thatis consistent with a solar axion signal, a solar neutrinosignal with enhanced magnetic moment, or a possible tri-tium background. We are unable to confirm nor excludethe presence of tritium at this time.

In an attempt to understand the low-energy excess, weperformed a number of additional studies. The analysisof an additional data set called SR2 — which displays a

18

0 2 4 6 8 10Energy [keV]

101

102

103

104E

vent

s/(t

yke

V)DAMA interpreted asleptophilic dark matter(axial-vector coupling)

2 GeV20 GeV200 GeVB0

SR1 data

FIG. 14. Comparison between DAMA expected signals andXENON1T data. Dotted lines represent the expected signalspectra of selected masses in the XENON1T detector if theDAMA modulated signals are interpreted as WIMPs scatter-ing on electrons through axial-vector interactions. XENON1Tdata are indicated by black points and the backgroundmodel B0 is illustrated by the red line. The right bound ofthe shaded region shows the threshold in this analysis.

∼20% lower background rate but only ∼10% statisticscompared to SR1 — is consistent with the SR1 analysisbut largely inconclusive about the nature of the excess.An S2-only search, which is able to probe sub-keV en-ergies, similarly yielded consistent constraints for all thediscussed hypotheses. Compared to the excess observedby DAMA, it is much lower in rate and thereby unrelated.

The signals discussed here can be further exploredin the next-generation detectors, such as the upcom-ing PandaX-4T [130], LZ [131] and XENONnT [132] ex-periments. The next phase of the XENON program,XENONnT, featuring a target mass of 5.9 tonnes anda factor of ∼6 reduction in ER background, will enableus to study the excess in much more detail if it persists.Preliminary studies based on the best-fit results of thiswork suggest that a solar axion signal could be differen-tiated from a tritium background at the 5σ level afteronly a few months of data from XENONnT.

Acknowledgements. We thank Dr. Robin Grossleof the Institute for Nuclear Physics - Tritium Laboratoryat Karlsruhe Institute of Technology (KIT) for informa-tive discussions. We gratefully acknowledge support fromthe National Science Foundation, Swiss National ScienceFoundation, German Ministry for Education and Re-search, Max Planck Gesellschaft, Deutsche Forschungs-gemeinschaft, Netherlands Organisation for Scientific Re-search (NWO), Weizmann Institute of Science, ISF, Fun-dacao para a Ciencia e a Tecnologia, Rgion des Pays dela Loire, Knut and Alice Wallenberg Foundation, KavliFoundation, JSPS Kakenhi in Japan, the Abeloe Grad-uate Fellowship and Istituto Nazionale di Fisica Nucle-

are. This project has received funding or support fromthe European Unions Horizon 2020 research and inno-vation programme under the Marie Sklodowska-CurieGrant Agreements No. 690575 and No. 674896, respec-tively. Data processing is performed using infrastructuresfrom the Open Science Grid, the European Grid Initia-tive and the Dutch national e-infrastructure with the sup-port of SURF Cooperative. We are grateful to LaboratoriNazionali del Gran Sasso for hosting and supporting theXENON project.

Appendix A: β Spectra Modeling

This appendix briefly describes the different theoreticalmodels used in the present work to compute the β spectrafor 214Pb, 212Pb, and 85Kr.

1. GEANT4 Radioactive Decay Module

The Radioactive Decay Module (RDM) in GEANT44

simulates the decay of a given radionuclide using the nu-clear data taken from an Evaluated Nuclear StructureData File (ENSDF) [133]. The required β spectra aregenerated in a dedicated class using an analytical model.The β spectral shape, i.e. the unnormalized emissionprobability per electron energy, is derived from Fermi’sgolden rule as:

dN

dW∝ pWq2F (Z,W )C(W )S(Z,W ), (A1)

with Z the atomic number of the daughter nucleus. Here,W is the total energy of the β particle and is related toits kinetic energy E by W = 1+E/me, with me the elec-tron rest mass. The maximum energy W0 is defined iden-tically from the energy of the transition E0. The β par-ticle momentum is p =

√W 2 − 1 and the (anti)neutrino

momentum is q = W0 −W , assuming a massless parti-cle (mν = 0).

The Fermi function F (Z,W ) corrects for the staticCoulomb effect of the nucleus on the β particle. Con-sidering the Coulomb field generated by a point-like nu-cleus, the Dirac equation can be solved analytically andthe well-known expression of the Fermi function can bederived. GEANT4 follows the approximate expression ofthe Fermi function from [134].

The shape factor C(W ) takes into account the nuclearand lepton matrix elements. Assuming constant valuesof the lepton wave functions within the nuclear volume,one can demonstrate that allowed and forbidden unique

4 Here we refer specifically to the current version 10.6; however thecorrections described in this work have been implemented sinceat least version 9.5 [60].

19

transitions can be calculated without involving the struc-ture of the nucleus. For an allowed transition, the shapefactor is constant: C(W ) = 1. In GEANT4, first, secondand third forbidden unique transitions are calculated fol-lowing the approximate expressions given in [135] thatwere established by considering the analytical solutionsof the Dirac equation, the same as for the Fermi function.In any other case, the decay is treated as allowed.

The atomic screening effect corresponds to the influ-ence of the electron cloud surrounding the daughter nu-cleus on the β particle wave function. GEANT4 takesthis into account following the most widespread approachset out by Rose in [136] almost a century ago. For aβ electron, this effect is evaluated by subtracting fromthe particle energy W a constant Thomas-Fermi poten-tial V0 which only depends on Z. This corrected energyW ′ = W − V0 replaces W in all the quantities requiredfor the calculation of the spectral shape, except in the(anti)neutrino energy q because this neutral particle isnot affected by the Coulomb field. The parameterizationof the potential used in GEANT4 is close to the prescrip-tion given in [137]. The screening correction is then givenby:

S(W,Z) =p′W ′

pW× F (Z,W ′)

F (Z,W ). (A2)

It is noteworthy that this correction can only be appliedfor W ≥ V0, which creates a non-physical discontinuityin the spectrum at W = V0, as seen in Fig. 15.

2. IAEA LiveChart

The β spectra available on the IAEA LiveChart web-site [58] are produced with the first version of the Be-taShape program [138]. The required information foreach transition is taken from the most recent ENSDF filewith results from the latest nuclear data evaluation [56].

The physics model in BetaShape has already been de-tailed in [59], except for the atomic screening effect. Theβ spectral shape is described in the Behrens and Buhringformalism [139] by:

dN

dW∝ pWq2F (Z,W )C(W )S(Z,W )R(Z,W ), (A3)

with all quantities as defined before. The quantityR(Z,W ) are the radiative corrections described below.

In this formalism, the Fermi function is defined fromthe Coulomb amplitudes αk of the relativistic electronwave functions:

F (Z,W ) = F0L0 =α2−1 + α2

+1

2p2. (A4)

These wave functions are numerical solutions of the Diracequation for the Coulomb potential of a nucleus modeledas a uniformly charged sphere. Indeed, no analytical solu-tion exists even for such a simple potential; however, the

method from [139] allows for a precise, and fast, calcula-tion of the Coulomb amplitudes. The method inherentlyaccounts for the finite nucleus size while other methodsusually require an analytical correction (L0 in Eq. (A4)).

The total angular momentum change ∆J = |Ji − Jf |and the parity change πiπf between the initial and finalnuclear states are from the input ENSDF file and deter-mine the nature of the transition. Given that L = 1 if∆J = 0 or 1 for an allowed transition, and L = ∆J forany (L−1)th forbidden unique transition, the theoreticalshape factor can be expressed as:

C(W ) = (2L−1)!

L∑k=1

λkp2(k−1) q2(L−k)

(2k − 1)![2(L− k) + 1]!. (A5)

The λk parameters are defined from the Coulomb ampli-tudes αk by:

λk =α2−k + α2

+k

α2−1 + α2

+1

. (A6)

In the case of forbidden non-unique transitions, thestructures of the initial and final nuclear states must betaken into account, which greatly complicates the cal-culation. The usual approximation consists of treatingsuch a transition as a forbidden unique transition of iden-tical ∆J . The validity of this approximation, minutelytested in [59], can be demonstrated only for some firstforbidden non-unique transitions, which are then calcu-lated as allowed. Its generalization to every forbiddennon-unique transition is implemented in BetaShape.

The radiative corrections are non-static Coulomb cor-rections from quantum electrodynamics. They can besplit into two parts: the inner corrections, which areindependent of the nucleus; and the outer corrections,which depend on the nucleus. Only the latter dependon the β particle energy. The outer radiative correctionsR(Z,W ) take into account the internal bremsstrahlungprocess, by which the β particles lose energy in the elec-tromagnetic field of the nucleus. For allowed transitions,analytical corrections were derived in [140, 141] and areimplemented in the first version of BetaShape as de-scribed in [59].

Finally, the spectral shape is modified by applying thescreening correction S(W,Z). The BetaShape programincludes an analytical correction based on the work ofBuhring [142] that is more precise than Rose’s correction.The most realistic, spatially varying screened potentialsat the time were of Hulthen type (see [143] and referencestherein). Buhring first developed a version of the Diracequation that correctly includes Hulthen’s potentials butsimplified the angular momentum dependency, allowinganalytical solutions to be established [144]. He then per-formed in [142] a radial expansion at the origin of boththe wave functions and the Coulomb potential, includ-ing Hulthen’s screened potential, and retained only thedominant term. This procedure allows the determinationof screened-to-unscreened ratios for the Fermi function

20

F0L0 and the λk parameters, which are then used to cor-rect for screening in Eq. (A3). Therefore, the quantityS(Z,W ) in Eq. (A3) is more a symbolic notation. InBetaShape, this approach is used with Salvat’s screenedpotentials [145], which can be expanded at the origin as:

V (r) = −αZr

+αZ

2β +O(r), (A7)

where β is determined from the parameters Ai and αigiven in [145]:

β =

3∑i=1

Aiαi. (A8)

These potentials are widely used for their precision andcompleteness. It is noteworthy that Buhring’s correc-tion does not create any non-physical discontinuity in thespectrum as in Rose’s correction. However, it tends togreatly decrease the emission probability at low energy.

3. Improved Calculations

When high precision at low energy is required, themodeling of β− decays must include the atomic screen-ing and exchange effects. The two approximate screeningcorrections previously described are not sufficient. Theexchange effect is even more significant and comes fromthe indistinguishability of the electrons. The regular, di-rect decay corresponds to the creation of the β electronin a continuum orbital of the daughter atom. In the ex-change process, the β electron is created in an atomic or-bital of the daughter atom and the atomic electron whichwas present in the same orbital in the parent atom isejected to the continuum. This process leads to the samefinal state as the direct decay, i.e. one electron in thecontinuum, and is possible because the nuclear chargechanges in the decay.

Precise relativistic electron wave functions are neces-sary to calculate such effects. The numerical procedurewas described in detail in [146], with the nucleus mod-eled as a uniformly charged sphere. For the continuumstates, the Coulomb potential includes the appropriateSalvat screened potential. The wave functions, and there-fore the Fermi function F0L0 and the λk parameters, in-herently take into account the screening effect. For thebound states, an exchange potential has to be added tothis Coulomb potential and a specific procedure was im-plemented to ensure good convergence to precise atomicenergies. In [146], the one-electron energies from [147]were considered while in the present work, the more ac-curate orbital energies from [148] that include electroncorrelations are used.

A precise description of the exchange effect was setout in detail in [57, 149], but only for the allowed tran-sitions. In such a case, β electrons are created in con-tinuum states with quantum number κ = ±1 and theselection rules imply that exchange can only occur with

atomic electrons of identical κ, i.e. in s1/2 (κ = −1) andp1/2 (κ = +1) orbitals. The influence of the exchange ef-fect can then be taken into account through a correctionfactor on Eq. (A3):

dN

dW−→ dN

dW×(1 + ηTex

). (A9)

The total exchange correction is defined by:

ηTex(E) = fs(2T−1 +T 2−1) + (1− fs)(2T+1 +T 2

+1), (A10)

with:

fs =gc

′

−1(R)2

gc′−1(R)2 + f c

′+1(R)2

. (A11)

All primed quantities refer to the daughter atom, and tothe parent atom otherwise. The large and small compo-nents of the relativistic electron wave functions, respec-tively gcκ and f cκ for the continuum states and gbn,κ and

f bn,κ for the bound states, respectively, are calculated atthe nuclear radius R. The quantities T−1 and T+1 de-pend on the overlaps between the bound states of theparent atom and the continuum states of the daughteratom with energy E,

T(κ=−1) = −∑

(n,κ)′

〈(Eκ)′|(nκ)〉gb

′

n,κ(R)

gc′κ (R)(A12)

and

T(κ=+1) = −∑

(n,κ)′

〈(Eκ)′|(nκ)〉f b

′

n,κ(R)

f c′κ (R). (A13)

The sums are running over all occupied orbitals of thedaughter atom of same quantum number κ.

It is noteworthy that in [146], only the s1/2 orbitalswere taken into account, following the prescription in[57]. The “new screening correction” proposed in [146]was necessary to reproduce the experimental β spectraof 63Ni and 241Pu, but was later found to be incompat-ible with a rigorous derivation of the β spectrum start-ing from the decay Hamiltonian and the correspondingS matrix. If correct screening and exchange effect withs1/2 and p1/2 orbitals are considered, together with pre-cise atomic orbital energies, excellent agreement over theentire energy range of the two spectra is obtained.

Finally, more precise radiative corrections have beenconsidered compared with those previously described.They were developed using more recent mathematicaltechniques and a significant change in the correctionterms was found [150]. Describing the various changes isout of the scope of the present work; however, many de-tails can be found in [151]. The influence of these new ra-diative corrections on the integrated β spectrum is givenfor twenty superallowed transitions in [152], for which anexcellent agreement is obtained with the present imple-mentation. It appears that these corrections are signifi-cantly smaller than the previous ones, especially for highatomic numbers.

21

4. Application to the Transitions of Interest

These different models have been applied to theground-state to ground-state transitions in 212Pb, 214Pb,and 85Kr decays. The resulting spectra are similar toeach other in the major part of the energy range, exceptat low energy.

The differences are illustrated in Fig. 15 for the lowenergy region of the 214Pb β spectrum. The yellow curveis the GEANT4 RDM model as described in A 1 and thenon-physical discontinuity due to the screening correc-tion is clearly visible at 12 keV. The red curve is fromIAEA LiveChart, thus generated by the first version ofthe BetaShape program as described in A 2. One cansee the effect of Buhring screening correction that tendsto decrease the emission probability. The cyan and bluecurves were determined as described in A 3, without andwith the atomic exchange correction, respectively. Thescreening effect is found to have a much smaller influenceon the spectral shape when determined using a full nu-merical procedure than when applying an analytical ap-proximation. However, the atomic exchange effect has astrong influence, as expected from previous studies [146].

0 10 20 30 40 50Energy [keV]

0.86

0.88

0.90

0.92

0.94

0.96

0.98

1.00

Rat

e [a

rb. u

nits

]

GEANT4 10.6IAEA LiveChartThis work w/o exchange effectsThis work final 214Pb model

FIG. 15. Low energy part of the β spectral shape of theground-state to ground-state transition in 214Pb decay. Thisfirst forbidden non-unique transition was calculated as al-lowed in every case but with different levels of approximations,as described in the text. The four spectra are normalized byarea over the full energy range. See text for details on theshape of each spectrum.

Both transitions in 212Pb and 214Pb ground-state toground-state decays were calculated as allowed, accord-ingly with the approximation described in A 2. It is im-portant to keep in mind that formally, such first forbid-den non-unique transitions should be determined includ-ing the structure of the initial and final nuclear states,a much more complicated calculation that is beyond thepresent scope.

The transition in 85Kr decay is first forbidden uniqueand can thus be calculated accurately without nuclearstructure. The description of the exchange effect from

[57, 149] used here is only valid for allowed transitions.For a first forbidden unique transition, one can expect acontribution of the κ = ±2 atomic orbitals but the exactsolutions have still to be derived. However, the spectralshape is derived from a multipole expansion of the nu-clear and lepton currents, as shown in the shape factorin Eq. (A5). Therefore, one can expect that the allowedexchange correction should give the main contribution,and this was done to determine the 85Kr β spectrum.

The tritium β spectrum used in this work was ob-tained from the IAEA LiveChart [58], thus calculatedusing the standard Fermi function without corrections.As 3H decays via an allowed transition, this spectrumis sufficiently precise at energies above 0.5 keV, as con-firmed experimentally in [79].

5. Uncertainties

The dominant contribution to the continuousXENON1T low-energy background comes from 214Pbβ decay. We thus focus the uncertainty discussionon the 214Pb ground-state to ground-state transition,calculated for the final model (blue curve in Fig. 15).

The transition energy is directly given by the Q-value [153]: Qβ = 1018(11) keV. This uncertaintycan be propagated by calculating the spectrum at(1018 ± 11) keV, namely at 1σ. The result is an enve-lope centered on the spectrum calculated at the Q-value,which provides an uncertainty on the emission proba-bility for each energy bin. The relative uncertainty is1.7% below 10 keV and 1.1% at 210 keV. However, mostof this uncertainty is removed because the 214Pb spec-trum is left unconstrained in the fitting procedure. Theremaining uncertainty component on the emission prob-ability is ∼ 0.5% for each energy bin, in which the shapeof the spectrum cannot vary steeply.

The atomic screening effect only slightly modifies theshape of the β spectrum. Its uncertainty contributioncan thus be safely ignored. The atomic exchange effectstrongly affects the spectral shape below 5 keV, and itsaccuracy depends on the atomic model used. For the βspectra of 63Ni and 241Pu, the residuals between theirhigh-precision measurement and the improved calcula-tion in A 3 showed that the agreement is better thanthe statistical fluctuations due to the number of countsin each energy channel, from 0.5 keV to the endpoint en-ergy. A conservative value of 1% for each energy bin is themaximum relative uncertainty and is the value adoptedhere.

The 214Pb transition of interest is first forbidden non-unique. As explained in A 4, the nuclear structure shouldbe taken into account for such a transition because it hasan influence on the spectral shape. Treating it as an al-lowed transition induces an inaccuracy which cannot beestimated by comparison with a measured spectrum –no measurement has been reported so far. In the samemass region, the 210Bi decay exhibits also a first forbid-

22

den non-unique, ground-state to ground-state transitionwith a comparable Q-value, and an experimental shapefactor is available. As can be seen in [154], treating thistransition as allowed leads to an important discrepancywith measurement. The question is then how this ob-servation can be used for assessing an uncertainty to the214Pb spectral shape.

First, allowing the rate normalization to be free inthe (1, 210) keV region absorbs the vast majority ofany difference. Second, the nuclear structures of 210Biand 214Pb are not identical. The 210Bi decay can beseen as two nucleons in the valence space above thedoubly-magic 208Pb core, with the initial configuration(p, 1h9/2)(n, 2g9/2) and two protons in the 1h9/2 orbitalin the final state. However, this picture is too simple tobe accurate because the core is not really inert. Nucle-ons from the core can give contributions to the β decaymatrix elements, mainly through meson exchange effectsand core polarization effects [155]. In 214Pb decay, a

single proton in the 1h9/2 orbital is present in the fi-nal state and in the initial state, six neutrons are spreadover the orbitals of the valence space but tend to coupleto each other through pairing and dominantly occupythe 2g9/2 orbital. Contributions from the core nucleonscan be expected to be relatively small compared to themain (n, 2g9/2)→ (p, 1h9/2) transition. In addition, eventhough it is difficult to predict if the nuclear structurecomponent shifts the spectrum to lower energies, as for210Bi, or to higher energies, a steep variation at low en-ergy is not realistic.

To conclude, we conservatively estimate a relative un-certainty on the spectral shape of 5% due to the nuclearstructure component and an additional 1% for the en-ergy dependency of the relative uncertainty on the maxi-mum energy. Thus a 6% total uncertainty on the spectralshape is estimated for the 214Pb β-decay model in thiswork.

[1] G. Bertone, D. Hooper, and J. Silk, Physics Reports405, 279 (2005).

[2] E. Aprile et al. (XENON Collaboration), The EuropeanPhysical Journal C 77, 881 (2017).

[3] E. Aprile et al. (XENON Collaboration), Phys. Rev.Lett. 121, 111302 (2018).

[4] E. Aprile et al. (XENON Collaboration),arXiv:2003.03825 [physics.ins-det].

[5] E. Aprile, K. L. Giboni, P. Majewski, K. Ni, and M. Ya-mashita, Phys. Rev. B 76, 014115 (2007).

[6] R. D. Peccei and H. R. Quinn, Phys. Rev. Lett. 38, 1440(1977).

[7] S. Weinberg, Phys. Rev. Lett. 40, 223 (1978).[8] F. Wilczek, Phys. Rev. Lett. 40, 279 (1978).[9] J. Preskill, M. B. Wise, and F. Wilczek, Physics Letters

B 120, 127 (1983).[10] L. Abbott and P. Sikivie, Physics Letters B 120, 133

(1983).[11] M. Dine and W. Fischler, Phys. Lett. B 120, 137 (1983).[12] D. Cadamuro and J. Redondo, Journal of Cosmology

and Astroparticle Physics 2012, 032032 (2012).[13] G. G. Raffelt, Astrophysical Axion Bounds (Springer

Berlin Heidelberg) p. 5171.[14] J. Redondo, Journal of Cosmology and Astroparticle

Physics 2013, 008 (2013).[15] S. Moriyama, Phys. Rev. Lett. 75, 3222 (1995).[16] K. van Bibber, P. M. McIntyre, D. E. Morris, and G. G.

Raffelt, Phys. Rev. D 39, 2089 (1989).[17] S. Dimopoulos, J. A. Frieman, B. Lynn, and G. Stark-

man, Phys. Lett. B 179, 223 (1986).[18] H. Primakoff, Phys. Rev. 81, 899 (1951).[19] D. A. Dicus, E. W. Kolb, V. L. Teplitz, and R. V.

Wagoner, Phys. Rev. D 18, 1829 (1978).[20] F. Alessandria et al. (CUORE Collaboration), Jour-

nal of Cosmology and Astroparticle Physics 2013, 007(2013).

[21] S. Andriamonje et al. (CAST Collaboration), Journal ofCosmology and Astroparticle Physics 2009, 002 (2009).

[22] M. Kuster, G. Raffelt, and B. Beltran, Axions

(Springer-Verlag Berlin Heidelberg, 2008).[23] S. Dimopoulos, G. Starkman, and B. Lynn, Mod. Phys.

Lett. A 1, 491 (1986).[24] S. Dimopoulos, G. Starkman, and B. Lynn, Phys. Lett.

B 168, 145 (1986).[25] M. Pospelov, A. Ritz, and M. Voloshin, Phys. Rev. D

78, 115012 (2008).[26] M. Dine, W. Fischler, and M. Srednicki, Physics Letters

B 104, 199 (1981).[27] A. Zhitnitskii, Sov. J. Nucl. Phys. 31, 260 (1980).[28] J. E. Kim, Phys. Rev. Lett. 43, 103 (1979).[29] M. Shifman, A. Vainshtein, and V. Zakharov, Nuclear

Physics B 166, 493 (1980).[30] M. Srednicki, Nuclear Physics B 260, 689700 (1985).[31] M. Tanabashi et al. (Particle Data Group), Phys. Rev.

D98, 030001 (2018).[32] D. B. Kaplan, Nuclear Physics B 260, 215226 (1985).[33] G. G. di Cortona, E. Hardy, J. P. Vega, and G. Vil-

ladoro, Journal of High Energy Physics 2016 (2016),10.1007/jhep01(2016)034.

[34] S. Chang and K. Choi, Physics Letters B 316, 5156(1993).

[35] L. Di Luzio, F. Mescia, and E. Nardi, Phys. Rev. D 96,075003 (2017).

[36] K. Fujikawa and R. E. Shrock, Phys. Rev. Lett. 45, 963(1980).

[37] N. F. Bell, M. Gorchtein, M. J. Ramsey-Musolf, P. Vo-gel, and P. Wang, Physics Letters, Section B: Nu-clear, Elementary Particle and High-Energy Physics642, 377383 (2006).

[38] N. F. Bell, V. Cirigliano, M. J. Ramsey-Musolf, P. Vo-gel, and M. B. Wise, Physical Review Letters 95, 14(2005).

[39] M. Agostini et al. (Borexino collaboration), Phys. Rev.D 96, 091103 (2017).

[40] S. Arceo-Daz, K. P. Schrder, K. Zuber,and D. Jack, Astroparticle Physics (2015),10.1016/j.astropartphys.2015.03.006.

[41] S. A. Daz, K.-P. Schroder, K. Zuber, D. Jack, and

23

E. E. B. Barrios, arXiv:1910.10568.[42] P. Vogel and J. Engel, Phys. Rev. D 39, 3378 (1989).[43] J. N. Bahcall and C. Pena-Garay, New Journal of

Physics 6, 63 (2004).[44] C.-C. Hsieh, L. Singh, C.-P. Wu, J.-W. Chen, H.-C. Chi,

C.-P. Liu, M. K. Pandey, and H. T. Wong, Phys. Rev.D 100, 073001 (2019).

[45] P. Arias, D. Cadamuro, M. Goodsell, J. Jaeckel, J. Re-dondo, and A. Ringwald, Journal of Cosmology andAstroparticle Physics 2012, 013 (2012).

[46] P. de Salas, K. Malhan, K. Freese, K. Hattori, andM. Valluri, Journal of Cosmology and AstroparticlePhysics 2019, 037 (2019).

[47] K. Arisaka, P. Beltrame, C. Ghag, J. Kaidi, K. Lung,A. Lyashenko, R. Peccei, P. Smith, and K. Ye, As-troparticle Physics 44, 59 (2013).

[48] I. M. Bloch, R. Essig, K. Tobioka, T. Volansky, andT.-T. Yu, JHEP 06, 087 (2017).

[49] P. Galison and A. Manohar, Physics Letters B 136, 279(1984).

[50] H. An, M. Pospelov, J. Pradler, and A. Ritz, PhysicsLetters B 747, 331 (2015).

[51] E. Aprile et al. (XENON Collaboration), Phys. Rev. D99, 112009 (2019).

[52] E. Aprile et al. (XENON Collaboration), Phys. Rev. D100, 052014 (2019).

[53] E. Aprile et al. (XENON Collaboration), Phys. Rev.Lett. 119, 181301 (2017).

[54] E. Aprile et al. (XENON Collaboration), The EuropeanPhysical Journal C 77, 275 (2017).

[55] M. Murra, Intrinsic background reduction by cryo-genic distillation for the XENON1T dark matter exper-iment, Ph.D. thesis, Westfalische Wilhelms-UniversitatMunster (2019).

[56] M. R. Bhat, Evaluated Nuclear Structure Data File(ENSDF), edited by S. M. Qaim, Nucl. Data Sci. Tech.(Springer-Verlag, Berlin, Germany, 1992) p. 817, Dataextracted using the NNDC On-Line Data Service fromthe ENSDF database, files revised as of October, 2019.

[57] M. R. Harston and N. C. Pyper, Phys. Rev. A 45, 6282(1992).

[58] M. Verpelli and L. Vrapcenjak, LiveChart of Nuclides,https://www-nds.iaea.org/livechart/ (IAEA, Nu-clear Data Section, 2020).

[59] X. Mougeot, Phys. Rev. C 91, 055504 (2015).[60] S. Agostinelli et al. (GEANT4), Nucl. Instrum. Meth.

A506, 250 (2003).[61] S. Hauf et al., IEEE Trans. Nucl. Sci. 60, 2966 (2013).[62] S.-C. Wu, Nuclear Data Sheets 110, 681 (2009).[63] S. Lindemann and H. Simgen, The European Physical

Journal C 74, 2746 (2014).[64] E. Aprile et al. (XENON Collaboration), The European

Physical Journal C 77, 890 (2017).[65] E. Aprile et al., Journal of Cosmology and Astroparticle

Physics 2016, 027 (2016).[66] A. Fieguth, First observation of double electron cap-

ture in Xe-124 and detection prospects for underlyingnuclear interaction mechanisms in direct dark mattersearch, Ph.D. thesis, Westfalische Wilhelms-UniversitatMunster (2018).

[67] J. B. Albert et al. (EXO Collaboration), Phys. Rev. C89, 015502 (2014).

[68] J. Kotila and F. Iachello, Phys. Rev. C 85, 034316(2012).

[69] http://nucleartheory.yale.edu.[70] E. Aprile et al. (XENON Collaboration), Nature 568,

532 (2019).[71] M. Doi and T. Kotani, Progress of Theoretical Physics

87, 1207 (1992).[72] C. Wittweg, B. Lenardo, A. Fieguth, and C. Wein-

heimer, “Detection prospects for the second-order weakdecays of 124xe in multi-tonne xenon time projectionchambers,” (2020), arXiv:2002.04239 [nucl-ex].

[73] Y. Khazov, I. Mitropolsky, and A. Rodionov, NuclearData Sheets 107, 2715 (2006).

[74] Y. Khazov, A. Rodionov, and F. Kondev, Nuclear DataSheets 112, 855 (2011).

[75] M.-M. Be, V. Chiste, C. Dulieu, X. Mougeot,V. Chechev, N. Kuzmenko, F. Kondev, A. Luca,M. Galan, A. Nichols, A. Arinc, A. Pearce, X. Huang,and B. Wang, Table of Radionuclides, MonographieBIPM-5, Vol. 6 (Bureau International des Poids etMesures, Pavillon de Breteuil, F-92310 Sevres, France,2011).

[76] A. Manalaysay, T. M. Undagoitia, A. Askin, L. Baudis,A. Behrens, A. D. Ferella, A. Kish, O. Lebeda, R. San-torelli, D. Venos, and A. Vollhardt, Review of ScientificInstruments 81, 073303 (2010).

[77] W. C. Haxton, R. G. Hamish Robertson, and A. M.Serenelli, Annual Review of Astronomy and Astro-physics 51, 21 (2013).

[78] J.-W. Chen, H.-C. Chi, C.-P. Liu, and C.-P. Wu,Physics Letters B 774, 656 (2017).

[79] J. Simpson, Phys. Rev. Lett. 54, 1891 (1985).[80] K. Mora, Journal of Instrumentation 14, P02003 (2019).[81] D. S. Akerib et al. (LUX Collaboration), Phys. Rev.

Lett. 118, 261301 (2017).[82] C. Fu et al. (PandaX-II Collaboration), Phys. Rev. Lett.

119, 181806 (2017).[83] V. Barsanov et al., Phys. Atom. Nucl. 70, 300 (2007).[84] M. Alfonsi, “Injection and removal of the calibration

isotope 37ar in the xenon1t detector,” (2019).[85] E. Aprile et al. (XENON Collaboration), (in prepara-

tion).[86] R. Saldanha, H. O. Back, R. H. M. Tsang, T. Alexander,

S. R. Elliott, S. Ferrara, E. Mace, C. Overman, andM. Zalavadia, Phys. Rev. C 100, 024608 (2019).

[87] E. Aprile et al. (XENON Collaboration), Phys. Rev.Lett. 118, 101101 (2017).

[88] E. Aprile et al. (XENON Collaboration), Phys. Rev. D97, 092007 (2018).

[89] D. S. Akerib et al. (LUX Collaboration), Phys. Rev. D93, 072009 (2016).

[90] X. Cui et al. (PandaX-II Collaboration), Phys. Rev.Lett. 119, 181302 (2017).

[91] L. L. Lucas and M. P. Unterweger, Journal of researchof the National Institute of Standards and Technology105, 541 (2000).

[92] C. Zhang, D.-M. Mei, V. Kudryavtsev, and S. Fiorucci,Astroparticle Physics 84, 62 (2016).

[93] J. Amare et al., Astropart. Phys. 97, 96 (2018),arXiv:1706.05818 [physics.ins-det].

[94] C. Hasterok, Gas Purity Analytics, Calibration Studies,and Background Predictions towards the First Results ofXENON1T., Ph.D. thesis, Ruprecht-Karls-Universitt,Heidelberg (2017).

[95] T. Ishida, Journal of Nuclear Science and Technology39, 407 (2002).

24

[96] A. Dobi, D. S. Leonard, C. Hall, L. Kaufman, T. Lang-ford, S. Slutsky, and Y. R. Yen, Nucl. Instrum. Meth.A620, 594 (2010).

[97] W. Plastino et al., Radiat. Meas. 42, 68 (2007).[98] W. Plastino et al., Journal of Radioanalytical and Nu-

clear Chemistry 288, 101 (2011).[99] IAEA/WMO, “Global Network of Isotopes in Precipi-

tation. The GNIP Database.” https://nucleus.iaea.

org/wiser (2015).[100] J. Happell et al., Tellus 56(3), 183 (2004).[101] J. Mishima and C. M. Steele, Oxidation of Tritium Gas

Under Accident and Transport Conditions, Report LA-UR-02-3803 (U.S.. Department of Energy Los Alamos,NM, 2002).

[102] E. Aprile and T. Doke, Rev. Mod. Phys. 82, 2053(2010).

[103] E. Aprile, R. Mukherjee, and M. Suzuki, Nucl. Instrum.Meth. A300, 343 (1991).

[104] K. Barth et al. (CAST Collaboration), Journal of Cos-mology and Astroparticle Physics 2013, 010 (2013).

[105] P. Gondolo and G. G. Raffelt, Phys. Rev. D 79, 107301(2009).

[106] M. M. Bertolami, B. Melendez, L. Althaus, and J. Isern,Journal of Cosmology and Astroparticle Physics 2014,069 (2014).

[107] A. Ayala, I. Domınguez, M. Giannotti, A. Mirizzi, andO. Straniero, Phys. Rev. Lett. 113, 191302 (2014).

[108] N. Viaux, M. Catelan, P. B. Stetson, G. G. Raffelt,J. Redondo, A. A. R. Valcarce, and A. Weiss, Phys.Rev. Lett. 111, 231301 (2013).

[109] M. Giannotti, I. G. Irastorza, J. Redondo, A. Ringwald,and K. Saikawa, Journal of Cosmology and Astroparti-cle Physics 2017, 010010 (2017).

[110] L. Di Luzio, M. Giannotti, E. Nardi, and L. Visinelli,(2020), arXiv:2003.01100 [hep-ph].

[111] A. Corsico, L. Althaus, M. M. Bertolami, S. Kepler, andE. Garcıa-Berro, Journal of Cosmology and Astroparti-cle Physics 2014, 054 (2014).

[112] A. G. Beda, V. B. Brudanin, V. G. Egorov, D. V.Medvedev, V. S. Pogosov, E. A. Shevchik, M. V.Shirchenko, A. S. Starostin, and I. V. Zhitnikov,Physics of Particles and Nuclei Letters 10, 139143(2013).

[113] Y. Wang et al. (CDEX Collaboration),arXiv:1911.03085 [hep-ex].

[114] T. Aralis et al. (SuperCDMS Collaboration),arXiv:1911.11905 [hep-ex].

[115] E. Aprile et al. (XENON Collaboration), Phys. Rev.Lett. 123, 251801 (2019).

[116] K. Abe et al. (XMASS Collaboration), Physics LettersB 787, 153 (2018).

[117] E. Armengaud et al. (EDELWEISS Collaboration),Phys. Rev. D 98, 082004 (2018).

[118] E. Aprile et al. (XENON Collaboration), Phys. Rev. D96, 122002 (2017).

[119] M. Agostini et al. (GERDA), (2020), arXiv:2005.14184[hep-ex].

[120] N. Abgrall et al. (Majorana Collaboration), Phys. Rev.Lett. 118, 161801 (2017).

[121] E. Brown et al., The European Physical Journal C 78,604 (2018).

[122] E. Aprile et al. (XENON Collaboration), (in prepara-tion).

[123] A. Corsico, L. Althaus, M. Bertolami, and S. Kepler,

The Astronomy and Astrophysics Review 27 (2019),10.1007/s00159-019-0118-4.

[124] M. M. Bertolami, B. Melendez, L. Althaus, and J. Isern,Journal of Cosmology and Astroparticle Physics 2014,069 (2014).

[125] R. Harnik, J. Kopp, and P. A. Machado, Journal ofCosmology and Astroparticle Physics 2012, 026 (2012).

[126] R. Bernabei et al., The European Physical Journal C56, 333 (2008).

[127] R. Bernabei et al., Nucl. Phys. At. Energy 19, 116(2018).

[128] J. Kopp, V. Niro, T. Schwetz, and J. Zupan, Phys. Rev.D 80, 083502 (2009).

[129] E. Aprile et al. (XENON Collaboration), Science 349,851 (2015).

[130] H. Zhong et al. (PandaX Collaboration), Science ChinaPhysics, Mechanics & Astronomy 62, 31011 (2018).

[131] D. Akerib et al., Nuclear Instruments and Methods inPhysics Research Section A: Accelerators, Spectrome-ters, Detectors and Associated Equipment 953, 163047(2020).

[132] E. Aprile et al. (XENON Collaboration), (in prepara-tion).

[133] J. Tuli, Nucl. Instr. Meth. A 369, 506 (1996).[134] D. Wilkinson, Nucl. Instr. Meth. 82, 122 (1970).[135] E. Konopinski, The Theory of Beta Radioactivity

(Clarendon, Oxford, 1966).[136] M. E. Rose, Phys. Rev. 49, 727 (1936).[137] D. Wilkinson, Nucl. Phys. A 150, 478 (1970).[138] X. Mougeot, EPJ Web Conf. 146, 12015 (2017).[139] H. Behrens and W. Buhring, Electron Radial Wave

Functions and Nuclear Beta Decay (Clarendon, Oxford,1982).

[140] A. Sirlin, Phys. Rev. 164, 1767 (1967).[141] W. Jaus, Phys. Lett. B 40, 616 (1972).[142] W. Buhring, Nucl. Phys. A 430, 1 (1984).[143] H. Behrens and J. Janecke, in Numerical Tables for

Beta-Decay and Electron Capture, edited by H. Schop-per, Landolt-Bornstein, New Series, Group I, Vol. 4(Springer Verlag, Berlin, 1969).

[144] W. Buhring, Z. Phys. A 312, 11 (1983).[145] R. M. F. Salvat, J. D. Martınez and J. Parellada, Phys.

Rev. A 36, 467 (1987).[146] X. Mougeot and C. Bisch, Phys. Rev. A 90, 012501

(2014).[147] J. P. Desclaux, At. Data Nucl. Data Tables 12, 311

(1973).[148] S. Kotochigova, Z. H. Levine, E. L. Shirley, M. D. Stiles,

and C. W. Clark, Phys. Rev. A 55, 191 (1997).[149] N. C. Pyper and M. R. Harston, Proc. Roy. Soc. Lond.

A 420, 277 (1988).[150] A. Czarnecki, W. J. Marciano, and A. Sirlin, Phys. Rev.

D 70, 093006 (2004).[151] L. Hayen, N. Severijns, K. Bodek, D. Rozpedzik, and

X. Mougeot, Rev. Mod. Phys. 90, 015008 (2018).[152] I. S. Towner and J. C. Hardy, Phys. Rev. C 77, 025501

(2008).[153] M. Wang, G. Audi, F. Kondev, W. Huang, S. Naimi,

and X. Xu, Chin. Phys. C 41, 030003 (2017).[154] X. Mougeot, Applied Radiation and Isotopes 109, 177

(2016).[155] H. Ejiri and J. Fujita, Physics Reports 38, 85 (1978).