Axion Dark Matter: What is it and Why Now?

Francesca Chadha-Day1,∗ , John Ellis2,3,∗ , David J. E. Marsh2,∗

October 18, 2022

Abstract
The axion has emerged in recent years as a leading particle candidate to provide the mysterious dark
matter in the cosmos, as we review here for a general scientific audience. We describe first the historical
roots of the axion in the Standard Model of particle physics and the problem of charge-parity invariance of
arXiv:2105.01406v2 [hep-ph] 17 Oct 2022

the strong nuclear force. We then discuss how the axion emerges as a dark matter candidate, and how it is
produced in the early Universe. The symmetry properties of the axion dictate the form of its interactions
with ordinary matter. Astrophysical considerations restrict the particle mass and interaction strengths to a
limited range, which facilitates the planning of experiments to detect the axion. A companion review dis-
cusses the exciting prospect that the axion could indeed be detected in the near term in the laboratory.

A review prepared for Science Advances

1 Institute for Particle Physics Phenomenology, Department of Physics, Durham University,

Durham DH1 3LE, United Kingdom
2 Theoretical Physics and Cosmology Group, Department of Physics, King’s College London, Strand,

London WC2R 2LS, United Kingdom

3 Theoretical Physics Department, CERN, CH-1211 Geneva 23, Switzerland;

National Institute of Chemical Physics & Biophysics, Rävala 10, 10143 Tallinn, Estonia
∗ To whom correspondance should be addressed

KCL-PH-TH/2021-20, CERN-TH-2021-045, IPPP/20/91

1 Introduction
The Standard Model of particle physics provides a successful description of the visible matter in the Uni-
verse, from stars to the inner workings of atoms and nuclei. It categorises the fundamental constituents of
matter, the quarks and leptons, and the electromagnetic, weak and strong nuclear forces between them. The
Standard Model is a mathematically consistent quantum field theory that allows theorists to calculate accurate
predictions, which have in many cases been verified experimentally with a precision below the per-mille level
at the Large Hadron Collider (LHC) and other particle accelerators. The crowning success of the Standard
Model was the discovery of the Higgs boson in 2012 [1, 2], a particle of a novel type whose existence was
predicted in 1964 [3] in order to solve theoretical problems associated with the masses of vector bosons, see
also [4]. Precise experimental measurements in the decades preceding 2010 verified many other predictions
of the Standard Model, including the existence and mass of the top quark. They also confirmed the necessity
of the Higgs boson and enabled its mass to be estimated numerically. Following its discovery, experiments
have confirmed that it has zero spin, unlike any fundamental particle known previously, and interactions with
other particles that are proportional to their masses, as predicted by the Standard Model.
Despite these manifold successes, particle physicists are restless and dissatisfied with the Standard Model,
because it has many theoretical shortcomings and leaves many pressing experimental questions unanswered.
Here we review the axion [5–7], a hypothetical particle beyond the Standard Model that addresses some of
these unresolved issues and is the focus of growing experimental interest.
One of these issues is how matter and antimatter are (and are not) distinguished at the level of fundamental
particles. For each matter particle in the Standard Model, special relativity and quantum mechanics require [8]
the existence of a corresponding anti-matter particle with identical mass and spin but opposite charge. Un-
charged particles such as photons may be their own anti-particles. The existence of every anti-particle in the
Standard Model has been confirmed experimentally by observing their production in the collisions of ordinary
particles. However, astrophysical and cosmological observations tell us that most of the visible material in

1
the Universe is composed of the same matter particles as us on Earth, and that there are no large concentra-
tions of antimatter. This cosmological matter-antimatter asymmetry is thought to be due to differences in the
interactions of elementary particles and antiparticles [9], which violate certain symmetries that distinguish
particle from antiparticle. These are charge conjugation (denoted by C) and its combination with parity rever-
sal (denoted by P, also known as spatial inversion symmetry). A form of CP violation was discovered in the
laboratory over 50 years ago [10], and by now it has been observed in many decays via the weak interactions.
These observations can be accommodated within the Standard Model with six quarks [11], albeit without a
profound explanation. However, this Standard Model mechanism is insufficient to explain the cosmological
matter-antimatter asymmetry and, moreover, it is a puzzle that CP violation does not appear also in the strong
nuclear interaction: this is known as the “strong-CP problem”. As we discuss below, this puzzle could be
resolved via the hypothetical axion particle. In this review, we reserve the term ‘axion’ for the particle arising
from the solution to the strong-CP problem described below, also known as the ‘QCD axion’. This will be
the main topic of this review. We will also touch briefly on ‘axion-like particles’, which do not solve the
strong-CP problem. As a result, the properties of axion-like particles are less theoretically constrained.
The failure to explain the cosmological matter-antimatter asymmetry and the CP symmetry of the strong
force are just two of the Standard Model’s failures to describe the Universe. Some might regard them as
“cosmetic” problems, since they could be solved by fiat within the context of the Standard Model. There is,
however, a much more pressing existential problem.
Multiple astronomical observations tell us that the visible matter described by the Standard Model provides
just a small part of the total density of the Universe. There is much more invisible “dark matter” out there [12]
that remains to be explained. This is gross inadequacy of the Standard Model, since the dark matter cannot be
explained without appealing to substantial amounts of new physics. The existence of dark matter is inferred
from its gravitational effects, and astrophysical observations suggest that it is “cold”, i.e., it has been moving
very slowly for much of the history of the Universe, and there are experimental upper limits on how strongly
it interacts with the visible matter. Moreover, the agreement between independent cosmological observations
(including galaxy clustering and the temperature anisotropies in the cosmic microwave background, CMB,
radiation [13]) and the corresponding theory of structure formation also tells us that dark matter must have
been present since early in the history of the Universe, a year or so after the Big Bang or even earlier. Jim
Peebles’ share of the 2019 Nobel Prize in Physics was for the theory of cosmological structure formation
behind this understanding, and the evidence it provides for the existence of cold dark matter [14]. Beyond
these basic facts relating to the temperature and longevity of the dark matter, we have very little information
about its nature and properties. It may well consist of one or more types of fundamental particle, though part
or all of it might consist of macroscopic lumps of some invisible form of matter such as black holes. The
axion has become one of the prime particle candidates for providing dark matter.
Historically, the first type of particle dark matter that was suggested was some sort of weakly-interacting
massive particle, abbreviated to WIMP, such as a massive neutrino. Minimal versions of this hypothesis within
the Standard Model were excluded experimentally, leading particle theorists to consider possible candidates in
extensions of the Standard Model, e.g., based on supersymmetry [15]. However, this is not the only possibility.
If the dark matter particle is a low-mass boson, it may populate the Universe in a coherent wave-like state that
is also slow-moving and hence “cold”. The axion is the prototype for such wave-like dark matter.
For many years following the proposals of the axion and WIMP in the late 1970s, axion dark matter
was a minority interest among particle physicists, who were mainly focused on high-energy collider physics.
Their primary motivation was the search for the Higgs boson, but WIMP searches were also an important
consideration that could often be pursued in parallel with the Higgs search. The search for the Higgs boson
culminated in its discovery at the LHC in 2012, but WIMP searches at the LHC and elsewhere [16] have
proved fruitless so far. These two developments have contributed to growing theoretical and experimental
interest in the axion. On the one hand, the Higgs is an existence proof for an apparently elementary spin-zero
boson undergoing spontaneous symmetry breaking, something that was somewhat controversial and unknown
previously. This proof of principle invalidated one possible theoretical prejudice against the existence of the
axion, which shares these properties with the Higgs. Perhaps more importantly, the non-appearance of any
WIMP at the LHC and in other direct searches has diminished enthusiasm for that candidate for particle dark
matter, though some hope still springs eternal. In parallel, there has been growing realization that axion-like
particles (ALPs) appear quite generically in extensions of the Standard Model, e.g., those with their roots in
string theory, see Fig. 1. For these reasons, the axion has now become a favoured theoretical candidate for
dark matter, motivating this review.
A companion article reviews the prospects for axion and ALP searches driven by experimental and tech-
nological advances at the so-called “precision frontier” of particle physics. These advances make axion dark

2
matter a viable candidate for discovery, which until recently appeared impossible, and are important factors
driving the growing interest in the field.
What is known about dark matter (DM)?

• Cosmic density (strong evidence: cosmic microwave background anisotropies [13]).

Expressed as a fraction of the total density of the Universe, DM makes up 26% of
the Universe, compared to 6% in ordinary matter, and 68% in vacuum energy.
• Local density (strong evidence: Milky Way stellar motions). The local density of
dark matter is around 0.3 to 0.4 GeV cm−3 , equivalent to one proton every few
cubic centimetres, or one solar mass per cubic lightyear. The density is measured on
average over a relatively large fraction of the galaxy. The actual density at the precise
location of the Earth could be significantly different. This is particularly relevant to
axions, as discussed below. The local density is around 105 times the average cosmic
density.

• Local velocity dispersion (strong evidence: Milky Way stellar motions). The velocity
dispersion of DM is around σv = 200 km s−1 , and our local motion with respect to
the galactic rest frame is in the direction of the constellation Cygnus.
• No preferred galactic length scale (strong evidence: galaxy clustering and evolution).
DM must be non-relativistic (v ∼ c would allow DM to move significant distances
during galaxy formation), and have negligible pressure (which would imprint sound
waves during galaxy formation). This discounts standard model neutrinos and other
“hot” or “warm” DM. For bosons, the de Broglie wavelength (which can be modelled
as an effective pressure) must be small compared to the galaxy clustering scale.

• Early appearance of DM (strong evidence: galaxy clustering). DM had to be present,

and gravitating, in the Universe long before the cosmic microwave background
formed, and its gravitational influence began before the Universe was one year old.
For light bosonic DM (like the axion) this corresponds to the latest epoch of particle
creation (tcold in Fig. 4).
• Lack of significant interactions (strong evidence: the “Bullet Cluster” [17]). DM
cannot interact with itself or ordinary matter too strongly.

2 Enter the Axion

2.1 The Strong CP Problem
The strong CP problem may be posed simply as follows: why have we observed no electric dipole moment
(EDM) for the neutron? The neutron EDM is a quantity that would violate Charge Parity (CP) symmetry.
This is in contrast to most of the laws of fundamental physics, which look the same after a CP transformation.
To understand the effect of a CP transformation on the neutron, we can instead think about how a neutron
would behave under a time reversal transformation. This is because if quantum field theory is Lorentz invariant
(i.e., if it is consistent with Einstein’s theory of Special Relativity), then all particles and processes must be
invariant under the CPT combination of discrete transformations, where T reverses the arrow of time (this is
known as the CPT theorem of quantum field theory). So, if we find something that violates T symmetry, it
must also violate CP symmetry in such a way that the combination CPT is not violated.
As an illustration, let us consider the effect of time reversal on the energy of a neutron with an electric
dipole moment d in an electric field E. This energy is given by the Hamiltonian

H = −d · E. (1)
If the neutron has a permanent EDM, this EDM must be aligned with the neutron’s spin. (This follows from
the Wigner-Eckart theorem in quantum mechanics.) When we reverse the direction of time, the direction of the
neutron’s spin is reversed, and therefore the direction of the EDM must also be reversed under a time reversal

3
 Massive Bosons
(integer spin)
Ultralight
Bosons Axions &
Moduli & ALPs
Dilatons QCD
axion
Higgs, H
Scalars Pseudo-scalars
(spin 0, CP even) (spin 0, CP odd)

Hidden photon
Vectors
(spin 1)

Figure 1: Many extensions of the Standard Model predict additional massive bosons, beyond the W, Z and
Higgs bosons of the Standard Model. They might be scalar (even under both P and CP transformations),
pseudoscalar (odd under both P and CP) or vector particles. The prototype for a pseudoscalar boson is the
axion, which is “ultralight” with mass m  1 eV. Other proposals include pseudoscalar axion-like particles
(ALPs), hidden photons, and scalar particles such as moduli and dilatons. Any of these might provide the
astrophysical dark matter.

transformation T . However, the electric field direction remains the same under a time reversal transformation.
So we can see from equation (1) that if the neutron has a non-zero permanent EDM, then the energy of a
neutron in an electric field will change under a T transformation. Hence a neutron electric dipole moment
would violate T symmetry and therefore also CP symmetry.
Figure 2 shows the P and T transformations of a permanent electric or magnetic dipole moment (which
must be aligned with a particle’s spin), of an electric field and of a magnetic field. We see that an electric dipole
moment violates both T and P symmetry, as the spin transforms with the opposite sign to the electric field
under both these transformations. Conversely, a magnetic dipole moment does not violate P or T symmetry
as the spin transforms in the same way as the magnetic field.
Some experimentally observed processes within the Standard Model do violate CP symmetry, i.e., the
process and the corresponding CP transformed process occur at different rates. These CP-violating processes
are generated by the weak interaction. However, CP violation has not been observed in any experiment on the
strong interaction, which describes the forces that bind quarks together to form protons, neutrons and other
hadrons in the theory of Quantum Chromodynamics (QCD). This is very surprising because the Standard
Model predicts that the structure of the QCD vacuum itself should violate CP. The CP violation of the QCD
vacuum can be parameterized by an angle denoted θ .
Furthermore, a neutron EDM should also receive a contribution from the effect of the weak interaction (the
interaction responsible for radioactive decays) on the quarks. Thus the total neutron EDM can be expressed
using the total CP-violating angle θ̄ :

θ̄ = strong interaction contribution + weak interaction contribution (2)

Within the Standard Model, the electric dipole moment of the neutron is proportional to θ̄ , with |d| ∼
3.6 × 10−16 θ̄ e cm. The two contributions making up θ̄ are entirely unrelated within the Standard Model
and are simply set by hand. Note that, although θ̄ is the sum of these two parameters, only the total θ̄ is ex-
perimentally observable. There is no way to measure the strong interaction and weak interaction contributions
to θ̄ individually. Experiment tells us that |d| < 1.8 × 10−26 e cm [18], implying that θ̄ < 5 × 10−11 . Thus,
a very precise cancellation between unrelated parameters is required to explain the observations. This could

4
Figure 2: Effect of time and parity transformations an electric or magnetic dipole moment (proportional to the
spin shown by the red arrow), on an electric field and on a magnetic field.

merely be a coincidence, or it could be a first hint of new fundamental physics.

2.2 The Axion Solution

Strong CP Problem and the Axion
• The Standard Model contains a constant, θ̄ , which is known to be an angle and thus
take values between 0 and 2π. Symmetry specifies no preferred angle.
• The value of this angle determines the neutron electric dipole moment, |d| = 3.6 ×
10−16 θ̄ e cm. Measurements are consistent with θ̄ . 10−10 , which suggests some
missing symmetry principle.

• The axion theory introduces a new field, a(t, x), such that θ̄ ∝ a(t, x), and for which
the value a = 0 is energetically favourable.

The axion is the most popular solution to the strong CP problem. It appeared first in a model proposed
by Roberto Peccei and Helen Quinn in 1977 [5]. Their idea was to promote θ̄ to a dynamical quantity, rather
than simply a constant parameter as in the Standard Model. In practice, this means adding a new field - the
axion field - to the Standard Model that couples to the strong nuclear force in the same way as θ̄ . As we shall
see, the axion field relaxes to a value such that θ̄ is zero and hence the total neutron EDM vanishes.
The behaviour of particles is determined by their symmetries. Emmy Noether showed that these sym-
metries correspond to conservation laws, such as conservation of energy or charge. In the Standard Model,
symmetries lead to forces between particles: the force “communicates” the symmetry from place to place
between particles. Forces are in turn mediated by bosonic particles: the force is a field, and quantum mechan-
ics associates a particle with every field (e.g. the photon for the electromagnetic field and electromagnetic
gauge symmetry). Thus, to add the axion to the Standard Model, we introduce a new symmetry, called the
Peccei-Quinn (PQ) symmetry [5], which is a global axial U(1) symmetry. ‘Global’ means that the symmetry
transformation is the same everywhere and for all time - in contrast to the gauge symmetries associated with
interactions in the Standard Model, whose transformations can be made independently at different points and
at different times. ‘Axial’ means that the symmetry transformation acts differently on left-handed and right-
handed particles. Roughly speaking, a particle’s handedness tells us about the relative orientation of its spin
and its velocity. Finally, a U(1) symmetry transformation is one that is mathematically equivalent to a rotation

5
Figure 3: “Sombrero” potential of the Peccei-Quinn field Φ is shown schematically before (left) and after
(right) the QCD phase transition. The axion corresponds to the angular direction of this potential, shown
by the orange line. The state of the field is given by a point in the potential. Low energy configurations
are favoured. For illustration, the potential on the right is shown for a scenario with a large amount of PQ
symmetry breaking. More details are given in appendix A.2.

about a single axis.

Introducing the PQ symmetry corresponds to introducing a new bosonic field beyond the single Higgs
boson of the Standard Model. At high temperatures, such as in the early universe, the PQ symmetry is
evident in all particle interactions. However, there is a phase transition that “hides” the PQ symmetry when
the temperature falls low enough (the bosonic force carriers of the symmetry become heavy and cannot be
thermally excited). Similar phenomena of “spontaneous symmetry breaking” are central to our understanding
of many areas of macroscopic physics, such as superconductivity, and there are also examples in particle
physics, such as the Higgs mechanism for generating particle masses.
To visualize spontaneous symmetry breaking, consider a ball rolling in the “sombrero” potential shown
on the left of Fig. 3. If the ball has enough energy (as at high temperature), it is able to roll over the hill in the
centre of the potential and occupies equally all areas of the circular valley in the “brim” of the potential. Now
imagine that the ball loses energy (as at low temperature), slows down and comes to rest. It will choose at
random to sit in one particular position in the potential well - even though every position in the circular min-
imum of the potential well is exactly equivalent. This arbitrary choice (a “choice” made the random thermal
state of the early Universe) is spontaneous symmetry breaking. Notice now that, while it would take a lot of
energy to get the ball over the potential hill again, we can push the ball around the circle of the potential well
with the smallest of nudges. This is a generic feature of spontaneous symmetry breaking. In particle physics,
it corresponds to the appearance following spontaneous symmetry breaking of a massless particle, which is
called a Nambu-Goldstone boson [19, 20]. The Nambu-Goldstone boson of the spontaneously broken Peccei-
Quinn symmetry is the axion [6, 7]. It is represented by a field, a, which is proportional to the problematic
θ̄ angle of the strong-CP problem, making the angle dynamical rather than a fixed and mysterious constant.
(Achieving this remarkable theoretical sleight of hand to make a ∝ θ̄ is described briefly in what follows, and
in more detail in the Supplemental Material.)

We have not yet solved the Strong CP problem, as the massless axion field could a priori take any value.
The next part of the story is the QCD phase transition (strictly speaking, a cross over) that occurs as the
temperature falls. When the temperature is sufficiently low, QCD becomes strongly-coupled and confines
quarks and gluons into the bound-state protons, neutrons and other hadrons that we see today. This phase
transition breaks the PQ symmetry by distorting the sombrero potential as seen on the right in Fig. 3. The
potential now has discrete minima and the energy is minimised by the axion field taking the value of one
of these minima. Thus, after the QCD cross over, the axion field rolls to the newly created minimum point,
which is where the contribution of θ̄ to the neutron EDM vanishes, setting the net neutron EDM to zero. (The
reason this minimum has the right CP properties is discussed in the Supplemental Material.) Notice that now,

6
to make the ball move around the sombrero potential we would need to push it away and up from its minimum
point. The energy required to move the classical axion field a small distance away from the minimum can be
modelled as an effective potential V (a) = m2a a2 /2. Upon quantisation, we interpret the parameter ma in the
classical potential as the mass of the axion particle.
The axion mass can be computed in terms of well-understood physics of the strong nuclear force by
considering the axion mixing with the neutral pion - a bound state of quarks with the same quantum numbers
as the axion. The axion’s interactions with the neutral pion mean that the pion’s mass generates a small mass
for the axion - this effect is only possible because the axion and the pion have the same quantum numbers.
This leads to the following relation for the axion mass ma :

ma f a ∼ mπ f π , (3)

where fa is proportional to the energy at which the PQ symmetry is spontaneously broken, mπ is the pion
mass, and fπ is a constant that is known from the rate of decay of the pion via the weak interaction. (A deeper
understanding of the origin of the axion mass relies on the theory of “instantons”, strong-interaction effects
that are discussed briefly in the Supplemental Material.) Using the experimental measurements of these pion
properties we can calculate the axion mass. The axion mass, as most scales in particle physics, is typically
given in electronvolts (eV), where 1 eV ≈ 1.8 × 10−36 kg which assumes } = c = 1, but we restore units of
} and c in formulae for clarity. Another useful conversion is between mass in electronvolts and Compton
frequency, such that 0.25 Hz ≈ 10−15 eV. The axion mass is:
 12 
10 GeV
ma = (5.70 ± 0.007) µeV , (4)
fa

where the error includes experimental and theoretical contributions appearing in a detailed calculation of
axion-pion mixing [21]. The constant fa is related to the scale of spontaneous breaking of the PQ symmetry
(see Supplemental Material), and is measured in GeV. What values could this take? The electroweak scale,
≈ 250 GeV, was one natural choice, but is excluded experimentally. Other scales in particle physics include
the grand unified scale, O(1016 ) GeV, and the Planck scale, ∼ 1019 GeV, which is the highest scale where
ordinary quantum field theory could possibly remain valid before quantum gravity becomes important. These
considerations give only very rough guidance as to the value of fa and the mass of the axion: below we narrow
their possible ranges.

The axion mechanism may be implemented in a wide variety of extensions to the Standard Model, which
are the subject of much current research [22]. In any axion model, we must introduce a new complex field
Φ = χeiθ , which gains a non-zero vacuum expectation value that breaks the PQ symmetry spontaneously.
After this spontaneous symmetry breaking, the axion is related to the phase of Φ by a = N fa θ , where N
is an integer, the “colour anomaly”, which varies depending on the detailed realisation of the Peccei-Quinn
mechanism. Figure 3 is shown for N = 4.
For the axion to solve the strong CP problem, the model must also include quarks that are charged under
the PQ symmetry, which in turn mediate an interaction between the axion and the gluon force carriers of the
strong nuclear force. There are two ways of achieving this. In one class of axion models, the Standard Model
quarks are charged under the PQ symmetry [23, 24]: these models have N = 6. This means we have to add
an extra Higgs doublet to the model in order to allow all of the Standard Model interactions to obey the PQ
symmetry. In another class of models, we instead add extra heavy, electrically neutral quarks to the theory
[25, 26]. Only these extra quarks are charged under the PQ symmetry: the canonical version of this model has
N = 1. The precise values of the axion’s mass and interaction strengths depend on these details of the model
(see Supplemental Material).

2.3 Interactions
The axion does not interact only with quarks and gluons, but also with the other particles in the Standard
Model. We know that these interactions must be very weak for the axion to have evaded detection so far.
Particle interactions are governed by their symmetries, and the axion’s interactions are set by its pseu-
doscalar nature. A pseudoscalar field changes sign under a parity transformation, i.e., when looking at the
Universe in a mirror. If we assume that the overall interaction is unchanged by a parity transformation, we find
that only certain interactions are allowed for a pseudoscalar particle. These can be expressed (schematically)
by the non-relativistic Hamiltonian:

7
 r Z q
ε0
H = gaγγ aE · BdV + ga f f }cOa · Ŝ + ε0 (}c)3 gEDM aŜ · E, (5)
µ0
where a is the axion field measured in units energy; gaγγ is the axion’s coupling to photons, γ (i.e. to electro-
magnetism), measured in units of inverse energy, ga f f is the axion’s coupling to matter particles, which are
fermions, f (the couplng depends on the particle in question, we write just one case for simplicity); gEDM is
the strength of a nuclear EDM induced by the axion; E and B are the electric and magnetic fields, and Ŝ is
the direction of the spin of the matter particle in question. ε0 and µ0 are the permittivity and permeability
of free space respectively - constants associated with electromagnetism, } is the reduced Planck’s constant,
which parameterizes the size of quantum effects, and c is the speed of light.
These interactions are very different from the interactions of a scalar particle, which can couple directly
E2 − B2 (the scalar Maxwell term) and to the masses of matter particles. This means that light scalar fields
can mediate extra long-range forces, and their interactions with Standard Model particles are therefore very
tightly constrained by the non-observation of such extra forces. In contrast, the couplings of pseudoscalar
particles to E · B and to the spins of matter particles [27, 28] make them much harder to detect.

What would the axion’s interactions with photons and with matter particles look like experimentally? Via
its interaction with E · B, axion DM would look like an additional electrical current or anomalous magnetic
field [29]. More generally, Maxwell’s equations of electromagnetism are modified by the addition of the axion
field. Further details on how axions modify Maxwell’s equations of electromagnetism are discussed in the
companion experimental review. Via its coupling to nuclear and electron spins, axion DM would cause these
spins to precess, as they would in a magnetic field, but now with an anomalous magnetization caused by the
invisible presence of the axion field [30, 31].
What are the strengths of the axion’s interaction with Standard Model particles? How large are the cou-
pling constants gi in the Hamiltonian Eq. (5)? We know that the interactions must be very weak, or we would
have found axions already. In fact, we expect the axion’s couplings to be inversely proportional to the scale
of symmetry breaking:
1
gi ∼ , (6)
fa
This is a consequence of one of the most fundamental ideas in particle physics, namely Effective Field The-
ory [32], which tells us to expect the axion’s interactions to scale inversely with the energy scale at which
the symmetry giving rise to it originates. As this energy fa could be very high, the axion’s interactions could
be very weak, as required by the experimental constraints. Comparing equations (3) and (6), we see that the
axion’s couplings obey:
g i ∝ ma . (7)
This is true for most models of the QCD axion, whose couplings are generally proportional to its mass, though
this relationship can be broken in some specific models of the QCD axion. Also, as we will see later, the axion
is just one particle in the broader class of ALPs discussed in more detail below. Such ALPs need not in general
solve the strong CP problem or couple to gluons. This means their mass could take any value and need not
be proportional to their couplings, and the constants of proportionality differ wildly for different ALPs. ALPs
could therefore be very weakly coupled and hence extremely difficult to detect experimentally. Alternatively,
Nature may provide us with more strongly coupled ALPs that can be detected more readily, as discussed
below. By contrast, for a canonical QCD axion of a given mass, we can predict the approximate size of the
couplings to Standard Model particles, providing a definite target for experimental searches. Further details
are given in the supplemental materials.

The interaction between DM axions or ALPs with particles and forces in the Standard Model leads to
a wide variety of ways to search for them. Initially, in the 1980’s up to 2010 or so, experiments were few
and far between. There was just one viable method, the microwave cavity haloscope, and the axion in-
teractions are too feeble to produce a signal measurable with technology of the time. Now, the landscape
has changed. Microwave photon detection and cavity design have allowed the haloscope concept to break
ground to exclude regions of the QCD axion parameter space, and expand the search over a wider frequency
range [33–35]. In tandem, new methods to detect axions have been conceived of and developed, including
magnetic resonance [31, 36–39], broadband antennas [40], dielectrics and metamaterials [41, 42], and lumped
circuit technology [43, 44]. These new technologies are at various stages of maturity, with some only existing
on paper, others being prototyped, and others already making competitive measurements of axion parameter

8
space and excluding theoretical models. Technologies now in development should be able to cover almost all
of the viable axion parameter space in the coming decades: the fuel for the growth of interest in this field. The
companion experimental review covers this topic in detail.
As well as searching directly for axion dark matter in the experiments described briefly above and in the
companion experimental review, the effects of axions and ALPs may be seen indirectly in telescope obser-
vations of stars, galaxies, and galaxy clusters. These astrophysical systems offer extreme environments that
would be impossible to replicate on Earth, and are therefore ideal places to search for new physics.
The interaction aE · B means that axions or ALPs can interconvert with photons in the presence of a
background magnetic field. This process is mathematically similar to neutrino oscillations, with the crucial
difference that the strength of the mixing depends on the size of the external magnetic field. The possibility
of interconversion between ALPs and photons in space places the strongest current bounds on gaγγ for very
low mass ALPs (ma . 10−8 eV).
One such bound arises from observations of distant point sources such as Active Galactic Nuclei shining
through foreground galaxy clusters. Galaxy clusters host strong magnetic fields over very large distances. If
ALPs exist, some of the photons from these point sources would convert into ALPs as they move through the
galaxy cluster. From our point of view, some of the light from point sources shining through galaxy clusters
would go missing. The non-observation of this effect can be used to place bounds on the axion photon
coupling (e.g. [45]), requiring gaγγ . 10−12 GeV−1 for ALP masses ma . 10−11 eV. However, it should be
noted that this method relies on the accuracy of our knowledge of the galaxy cluster magnetic field.
We can also bound gaγγ using observations of SN 1987A, a supernova observed in 1987 in the nearby
(astronomically speaking) Large Magellanic Cloud. A supernova is an explosion at the end of a star’s life,
which produces vast quantities of neutrinos and photons. If ALPs exist, a supernova would also produce them
copiously via nuclear interactions between ALPs and the supernova constituents. Some of the ALPs produced
by SN 1987A would have been converted into γ - ray photons in the Milky Way’s magnetic field. No such
extra γ - ray photons have been observed, which allows us to set a bound gaγγ < 5.3 × 10−12 GeV−1 for ALP
masses ma . 4.4 × 10−10 eV (e.g.[46]). As we will see below, SN 1987A can also be used to constrain the
interactions of higher-mass axions.
The constraints outlined above apply only to rather low mass ALPs. Photons travelling through the plas-
mas of galaxies and galaxy clusters acquire a low effective mass, but axions or ALPs with masses higher than
this effective mass cannot mix efficiently with the photon. For this reason, astrophysical axion-photon mixing
constrains the ALP parameter space, but not the standard QCD axion for which gaγγ ∝ ma .
The coupling between axions and photons would also allow axions to be copiously produced in stars.
This would create an additional cooling mechanism for the star and would therefore alter the course of stellar
evolution. This effect can also be used to constrain gaγγ , requiring gaγγ < 6.6 × 10−11 GeV−1 over a very
wide range of masses [47], thus constraining both the QCD axion and a more general ALP. Our nearest star,
the Sun, would also copiously produce axions, which could be observed experimentally via controlled axion-
photon conversion in the laboratory. This is the aim of the CERN Axion Solar Telescope (CAST) experiment
[48], and other so-called “helioscopes”, discussed in the companion experimental review.

3 Axion Dark Matter Waves

The advent of quantum mechanics in the early 20th century taught us that for every particle there is a wave,
and for every wave there is a particle. The wavelength, λ , and frequency, f , of the wave are related to the
particle mass m and velocity v (in the non-relativistic limit) by:

h 1
λ= , h f = E = mc2 + mv2 , (8)
mv 2
where h is Planck’s constant. When the particle is relatively heavy compared to the axion, as in the case of
the electron, the wavelength is small, and so in our day-to-day lives we do not notice the particle behaving
as a wave. Electrons are also fermions (particles of half integer spin). This means that they must obey
the ‘exclusion principle’ - more than one electron cannot occupy the same state, and so collective wavelike
behaviour does not occur (except when they bind to form pairs with integer spin, as in superconductors).
On the other hand, for light and massless particles, the wavelength of the matter waves can be large,
and their frequency low. If these particles are also bosons (particles of zero or integer spin, see Fig, 1),
then many particles can occupy the same state. When the occupation number is macroscopic, the bosons
can be described using classical, rather than quantum, field theory. This gives rise to macroscopic wavelike

9
behaviour. Historically, we were first aware of photons in their guise as the classical electromagnetic radio
waves of Hertz. Our technology to observe electromagnetic waves begins in this realm of wavelengths larger
than 1 mm or so, and frequencies lower than around 1 THz.
The axion field, a, is just one massive bosonic field among many possibilities, as also is its parent Peccei-
Quinn field, Φ. The energy density of dark matter in the Universe is one of the most macroscopic quantities
imaginable, which in Einstein’s theory of General Relativity affects the very geometry of spacetime and the
expansion of the Universe. The theory of axion DM is the theory of classical waves in the fields Φ, a, and
their dynamics under the influence of gravity. We begin this section by considering the origin of the energy
density in these fields in the very early Universe.

What is axion DM and how did it get here?

• Axion DM consists of the energy stored in spatial and temporal gradients, and po-
tential energy, of the axion field, a(t, x).

• Initial fluctuations of the axion field arose from a phase transition involving sponta-
neous symmetry breaking that occurred in the early Universe.

3.1 Two Roads to Axion Dark Matter

Observations of the CMB tell us that to a very good first approximation the early Universe was homogeneous,
isotropic, flat, and hot. Under these approximations, Einstein’s equations of General Relativity reduce to a
single ordinary differential equation, the Friedmann equation (see Supplementary Material). This equation
relates the temperature of the plasma in the early Universe to its expansion rate, expressed in terms of the
Hubble parameter H(t).
The isotropy of the CMB on the largest scales also provides evidence that this hot, dense state of the early
Universe must have been preceded by a far more mysterious epoch that sets the large-scale initial conditions
of the Universe. That such an epoch must have existed is the only relevant point at present: which theory
describes it (for example inflation or a cyclic Universe) has no bearing on our story (although in the case
of inflation, the axion could be used to test key predictions of the theory). After the initial conditions of
the Universe were set, the well understood “hot Big Bang” phase began, at some finite time t hot when the
temperature of the Universe was at its maximum, T hot .When cosmologists refer to the “age of the Universe”
or “time after the Big Bang”, this is in reference to the time since t hot , and not dependent on how much time
there was, or was not, before this (for example, the Universe could be infinitely old in its past). Thus, in the
box “What is known about dark matter?”, the “Creation of DM” refers to the amount of time ∆t = t − t hot .
The key distinction for the origin of axion dark matter is whether the Peccei-Quinn phase transition occurs
for T PQ > T hot or T PQ < T hot . This is summarised in Fig. 4. The fact that the oscillations of the axion field
caused by spontaneous symmetry breaking leads to a viable DM candidate was first realised in 1983 by three
separate groups working independently [51–53]. This happy accident, a DM candidate “for free” from a
particle postulated only to solve the strong-CP problem, is now a main virtue of the theory. There are two
scenarios for spontaneous symmetry breaking that we must consider, shown schematically in Fig. 4.
In Scenario A, (T PQ > T hot ), the Peccei-Quinn phase transition occurs in the epoch during which the
initial conditions were set, before the hot Big Bang. This is the simplest case to describe mathematically, as
this epoch serves to smooth the Universe on the largest scales, in accordance with observation. This removes
all “memory” of the details of the phase transition (e.g., all the spatial gradients in the axion field), and
replaces all that complexity with just two quantities: a single random number, the initial value of the axion
field, θ (t hot ), and the axion mass, ma . With these two inputs, the evolution of the classical axion field, and
the subsequent DM density, is determined by a simple wave equation in which the expansion of the Universe
acts as a friction term damping the waves (see Supplemental Material).
In Scenario B (T PQ < T hot ), the Peccei-Quinn phase transition occurred during the hot Big Bang phase,
after the initial conditions epoch ended. Thus the axion field is not smoothed, leaving a messy field con-
figuration. The physics involves the Kibble mechanism [54] for the formation and subsequent disposal of
“topological defects” in the Peccei-Quinn field (the very same mechanism occurs in the superfluid phase tran-
sition of Helium [55]). Topological defects are axion field configurations that occur between regions where
the axion field takes different values, which may subsequently decay into axion DM. There is also more model
dependence in this case, as different models of the axion lead to different kinds of topological defect. The
dynamics in Scenario B is highly non-linear. However, in Scenario B the random number, θ (t hot ), of Scenario

10
 A: TPQ>Thot B: TPQ<Thot
Initial
conditions

Thot thot
Temperature

Time

At most one year

Harmonic motion Topological defects
thot
Axion Wave
Production

tdyn
Non-linear dynamics
Time

300000 years

tcold
✓i 0
Axion Field, ✓

The
CMB
Figure 4: Production of axion DM. The initial conditions epoch prior to the hot phase serves to smooth the
Universe on the largest scales. The axion dynamics and resulting DM density depends on when the Peccei-
Quinn (PQ) phase transition at the temperature TPQ takes place relative to the onset of the hot phase of the
evolution of the Universe at Thot , leading to two scenarios, A and B. Axion production occurs during the period
between the two times tdyn and tcold , and is complete in around 1 year at most. After tcold , axions are described
by the Cold Dark Matter model on length scales larger than their de Broglie wavelength. The axions then
“free stream” up to the time of formation of the Cosmic Microwave Background (CMB), and beyond into the
epoch of galaxy formation and the present day. In Scenario A, the axion mass is not fixed by the observed
dark matter density, since there is an additional free parameter, θ (thot ). In Scenario B there is no additional
free parameter and the mass can in principle be predicted. (Some images adapted from Refs. [49, 50])

11
A is averaged over in the initial distribution and is no longer a free parameter. In Scenario B large density
variations in the initial state of the axion field lead to the formation of compact DM objects known as “mini-
clusters” [56]. These objects offer new opportunities to discover the axion by astrophysical means, but also
lead to an increased theoretical uncertainty in the local axion density for DM detection in the laboratory.
In both Scenarios A and B an important change happens to the axion field when the expansion rate of the
Universe (which is falling during cosmic time) drops below the natural frequency of axion oscillations, which
is determined by its mass. At this time, tdyn , the axion field becomes highly dynamical (the wave equation
becomes underdamped) and begins to oscillate. Prior to tdyn , there were no non-relativistic axions: on the
length and time scales of the expansion of the Universe, the axion field was effectively static. This is because
the expansion of the universe is responsible for damping the axion field oscillations.
Thus it is at tdyn that we think of axion DM as being produced. In Scenario A, the axion field undergoes
damped simple harmonic motion, while in Scenario B, we think of axion field oscillations being emitted from
the decaying topological defects. This epoch of particle production lasts only for a relatively short period, as
the large scale energy density in the axion field, stored up from the phase transition, is converted into field
oscillations, which in quantum theory are equivalent to axion particles. From tdyn up to some later time tcold
the number of axion particles grows. Then, from tcold up to the present day the number of axion particles in
any sufficiently large “comoving” volume (a volume that increases in proportion with the expansion of the
Universe) in the Universe is conserved , i.e., the axion number density is simply diluted by the expansion of
the Universe just like any other particle would be. Axion cold dark matter is born, and starts to form cosmic
structures such as galaxies.

3.2 Cosmic Structure Formation

Because of the very large number of axions needed to supply the observed DM density, there are a large
number of axions occupying each quantum state. This means that axion DM is described to a very good
approximation by the classical scalar field a(t, x). Averaged over cosmic scales (distances of more than 10
Megaparsecs), the axion field is very uniform, and its energy density is stored in harmonic oscillations: ρ =
1 2 3 2 2 3 3 2 2 2 3 3
2 [ȧ /(}c ) + ma a /(} c )], with a = a0 cos ma c t/}, thus ρ = ma a0 /(2} c ). The value of a0 is determined
by the early evolution of the axion field as described above. Crucially, for the axion to behave as cold DM
with weak self-interactions, we require a0  fa , which permits the harmonic motion approximation for small
displacements. The frequency of oscillation of the field is determined by the particle mass and is given by the
Compton frequency.
On smaller scales, gravity causes the axion field to develop inhomogeneities, and it clusters into so-called
“DM halos”. On larger scales, the halos are linked by filaments and sheets in the so-called “cosmic web”.
The process of structure formation begins very early in the history of the Universe: the gravitational potential
wells of the DM halos began to form at very high temperatures, when ordinary matter was still completely
ionized (i.e., before the CMB formed). The development of inhomogeneities in the axion field is governed
by gravity, and the (small) axion self-interactions. The equations describing this evolution are known as the
Gross-Pitaevski-Poisson equations, a form of non-linear Schrödinger equation for the mean field occupation,
where the potential is determined by the density of the field itself via the Poisson equation (see Supplemental
Material).
It is here that axion DM begins to take on some unique characteristics compared to heavier particle DM
such as WIMPs. These features are:
• The De Broglie wavelength. Gradient energy dominates over gravitational energy on small scales. The
axion field is uniform on small scales and there is a minimum mass for a DM halo [57].

• Axion Stars. At very high density, the axion forms a kind of soliton (a stable wavepacket-like field
configuration) supported by an equilibrium between gravity and gradients [58, 59].
• Wave turbulence and interference. In the structures of the cosmic web, axion waves have dynamic
velocities. Where there are coherent flows, this leads to interference patterns in filaments. In ther-
malised/virialised environments (halos) the velocities are Maxwell-Boltzmann distributed, and are de-
scribed by wave turbulence [60, 61].
A numerical simulation of axion structure formation is shown in Fig. 5, where the above mentioned effects
can be seen. Similar phenomena showing the formation of solitons and “incoherent” turbulent solitons occur
in certain non-linear optics systems [62].

12
 “axion star”/
soliton

“incoherent
soliton”/
interference
turbulence
fringes

Figure 5: Zoom in to part of the Cosmic Web formed by gravitational interactions and the interference of
axion waves. The dense, almost spherical nodes of the web are dark matter halos, which host galaxies like our
own. The inset shows a volume rendering of such a halo. Notice the interference of coherent waves outside
of dark matter halos forming interference fringes. Inside halos, there are self bound solitonic objects at high
density (“axion stars”), and turbulent waves in the outer regions (“incoherent solitons”). In order to make the
unique wave effects visible on cosmic scales (kiloparsecs, kpc), a very low particle mass, ma ≈ 10−22 eV,
is used in this simulation. The same physics operates and the same effects occur, however, for all types of
bosonic dark matter, with the length scale set by the particle de Broglie wavelength. Figure adapted from
work presented in Refs. [63, 64].

Inside the Milky Way, at the location of the Earth, the axion field is in the turbulent regime. It is given as:
1  −ima c2 t/} 2

a= Ψe + Ψ∗ e+ima c t/} (9)
2
The amplitude field Ψ is Rayleigh distributed (up to small corrections) with a coherence length and time
governed by the de Broglie wavelength at the local galactic orbital velocity, and an amplitude fixed by the local
DM density. The overall frequency f is determined by Eq. (8), with the velocity drawn from the Maxwell-
Boltzmann distribution, leading to a frequency dispersion ∆ f / f ≈ 10−6 . The frequency dispersion leads a
natural linewidth for axion DM.

4 What is the Axion Mass?

The QCD axion has a single free parameter in the low-energy theory, the decay constant fa , which in turn
sets the axion particle mass (at zero temperature), ma , according to Eq. (3). In order to try to detect the
axion in the laboratory, one would like to know the approximate mass scale, so that experiments can be
appropriately targeted to the mean oscillation frequency, f = ma c2 /h, which in turn determines the frequency
of any radiation or other associated phenomenon (as discussed elsewhere in this review and in the experimental
companion review). Currently, astrophysical considerations provide the best clues to the range of allowed
values for the axion mass, summarised in Fig. 6, which we now elaborate on.
Beginning at the lightest end of the scale are bounds on so-called “fuzzy” DM. When the particle mass is
very low, the de Broglie wavelength is cosmologically large, and affects the formation and internal dynamics
of galaxies. Galaxies cannot form below the de Broglie wavelength, which suppresses galaxy formation in
the early Universe (where low-mass galaxies form first). Conservative constraints on the axion mass in this
scenario are m & 10−22 eV, although the bounds can be tightened by up to three orders of magnitude including
more data, and better knowledge of the structure formation process and galactic dynamics [65, 66]. This limit
also applies to other forms of bosonic DM (not just axions, see Fig, 1), and is the fundamental lower limit
on DM particle mass. If DM is to be lighter than 10−22 eV, observationally it can be only a sub-dominant
fraction of the total DM density, with current bounds around the 2% level [67].
The lack of a significant observed length scale in cosmic structure formation is one of the key facts known
about DM, as described in the Box “What is known about DM?”. Similarly, there is no evidence of a pre-

13
 22 11 2
10 10 10
Sub-dominant Exotic heavy
Exotic light ALPs
dark matter The QCD Axion ALPs
Axion
Mass [eV]
dark matter

Supernova
“Fuzzy”

1987A
Black hole spins

Figure 6: Constraints on the axion mass, measured in electronvolts (eV). Red arrows and the shaded region
indicate exclusions. The region outside that marked “The QCD Axion” corresponds to more exotic axion-like
particles (ALPs). “Fuzzy dark matter” shows a simulation of cosmic structure formation with very low particle
mass DM (adapted with permission from Ref. [68]), at the lower limit of what is acceptable observationally
for the dominant DM component. “Black hole spins” shows the ergoregion and horizons of spacetime around
a spinning black hole, the geometry of which allows very light bosonic particles to extract angular momentum
from the black hole. These last two constraints apply under the assumption that axion self-interactions are
negligible. The SN1987A constraint applies only to the QCD axion, where the axion-nuclear coupling is
proportional to the particle mass, and not to ALPs (image is a composite [69]).

ferred time scale related to the formation of DM, and only a bound on when this must have been complete.
Demanding that axions or other ultralight bosons (see next section) were formed early enough sets tcold & 1
year, and similarly bounds m & 10−22 eV.
The next important constraint on the particle mass results from a property known as “black hole super-
radiance”. When the Compton wavelength of a boson is resonant with the radius of the “ergoregion” of a
spinning (Kerr) black hole, the so-called “Penrose process” [70] operates, and vacuum fluctuations of the bo-
son extract angular momentum from the black hole, reducing its rate of spin. (The ergoregion is a region close
to a spinning black hole in which objects are necessarily forced to co-rotate with the black hole.) The radius
of the ergoregion is fixed by the black hole mass and spin, while the spin of the black hole can be inferred,
for example, from the Doppler shift in the emission of X-rays from its accretion disk. Rapidly spinning black
holes would quickly spin down if a resonant boson existed, and so the observation of spinning and old black
holes places exclusions on boson masses. The most robust, and also most important, range of these exclusions
comes from stellar mass black holes with spins measured from X-ray images, and for spin-0 bosons excludes
the range 1 × 10−13 eV ≤ m ≤ 2 × 10−11 eV [71, 72], with a similar range excluded for massive spin-1 (vec-
tor) bosons. The importance of this bound is that it overlaps the range of the masses allowed for the QCD
axion with spontaneous symmetry breaking scale below the Planck scale, 1019 GeV. Since it is widely be-
lieved that symmetry breaking above the Planck scale is forbidden in quantum gravity and string theory [73],
superradiance leads to a solid lower bound to the mass range of the QCD axion.
An upper bound on the QCD axion mass can be derived using the relationship between the mass and the
coupling strength to nuclear matter, gaNN , and is derived from observations of the supernova SN1987A [74].
The axion-nucleon coupling leads to emission of axions from neutrons and protons inside a collapsing super-
nova core in the Bremstrahhlung process, N + N → N + N + a. The produced axions interact only very weakly
with the surrounding nuclear matter, and so can escape from the supernova core, leading to excess cooling
compared to the case without axions. Neutrinos are also produced inside supernovae, and provide a known
cooling channel. If axions are also produced, then the supernova cools more quickly and thus has less time
to emit neutrinos, leading to a lower neutrino flux. The neutrinos produced in SN1987A were observed in
neutrino detectors on Earth, and the duration and timing of the neutrino burst is known, and is consistent with
predictions of standard supernova models. Thus, the axion production rate must be low, leading to an upper
bound on gaNN , and thus an upper bound on ma (see Eq. 7), which is ma . 2 × 10−2 eV.
The superradiance and SN1987A bounds thus limit the QCD axion natural frequency, of critical impor-

14
tance for direct searches, to the range:

10 kHz . f . 10 THz . (10)

The QCD axion mass can be further narrowed down by appeal to the dark matter density, ρa , in the two
cosmological scenarios discussed in the previous section. Observationally, we have the restriction that the
axion density cannot be larger than the observed dark matter density (although it could be smaller, if the
axions are not all of the dark matter).
The important difference between Scenarios A and B is one of predictability. In both cases, the dark matter
density depends inversely on the axion mass. In Scenario A, there are two continuous free parameters θi and
ma . Thus it is possible to reproduce the correct cosmic density of axions for (almost) any value of the mass ma ,
with smaller values of the mass having θ (t hot )  1. Without a prediction for the axion mass, the experimental
task of searching for the axion could be long and arduous. This scenario is probed by experiments operating
in the 10 kHz to 10 GHz regime.
In Scenario B, on the other hand, there is only one continuous parameter, ma , and the discrete parameter N.
In principle, the observed cosmic DM density then predicts the value of ma . The problem with trying to make
such a prediction is the computational complexity: the non-linear dynamics of the axion field in Scenario B
cannot be solved exactly. Unfortunately, computational limitations even with modern supercomputers prevent
a complete, end-to-end, ab initio simulation, and the resulting predictions for the mass vary considerably
depending on the theoretical methods employed.
Demanding that the axion is all the DM, the predicted value of the mass is in the range 0.025 meV .
ma . 0.5 meV when N = 1 (where the range covers approximately the predicted regions from the simulations
of Refs. [75, 76], which span those also of other groups). The case N > 1 predicts a slightly higher range
0.1 meV . ma . 20 meV (the upper bound comes from SN1987A) [77]. Lower values of the axion mass are
excluded in Scenario B. The theoretical uncertainty is large, but in general such predictions provide motivation
to search for axion dark matter using experiments in the 10 GHz to 10 THz frequency range. This promise to
predict the axion mass uniquely in this scenario is driving many theoretical investigations, and is happening
now due to advances in computer simulation techniques.

5 ALPs Everywhere
It has been realized since the proposal of the axion that light spin-zero particles with properties resembling
those of the axion may appear in a number of different theoretical contexts. Such axion-like particles (ALPs)
may either be pseudoscalar bosons, whose couplings are odd under parity transformations (reversing spatial
directions), or scalar bosons, whose couplings are even under parity. The latter are sometimes referred to
loosely as dilatons or moduli, and may be thought of as lower-mass siblings of the Higgs boson of the Standard
Model. Generically, both classes of ALPs possess, like axions, couplings with fermion-antifermion pairs, and
also with pairs of photons, gluons, and the other vector bosons via their field strengths, of the forms E · B for
pseudoscalars and |E|2 + |B|2 for scalars (where E and B are generalised for the other forces of the Standard
Model). ALPs are distinguished from the QCD axion of our previous discussion, since ALPs do not solve the
strong-CP problem. ALPs do, however, still make excellent dark matter candidates.
There is a single scalar Higgs boson in the Standard Model associated with the generation of particle
masses via breaking of the underlying electroweak symmetry. However, including another symmetry-breaking
Higgs field requires the appearance of an additional scalar particle as well as a pseudoscalar partner particle.
The original axion model is an example of this feature, but it can be generalized by including more Higgs
fields of different types. This scenario occurs in many theories, including supersymmetry, and grand unified
theories.
Another theoretical framework where ALPs appear is provided by string theory [78]. Generic four-
dimensional particle physics models obtained from string theory, by the compactification of extra dimensions
of space on small manifolds, contain many additional pseudoscalar and scalar fields, which may have (very)
small masses. A useful and famous analogy for compact dimensions of space is to imagine our world as a
tightrope. From the perspective of large humans, the rope is one-dimensional, and one can only move along
its length. However, a small creature such as an ant can circle the “extra dimension” of the rope’s circum-
ference. In quantum mechanics, length is inversely proportional to energy, and thus the small length scales
of extra dimensions of space tend to be associated to large energy scales, possibly even the Planck scale of
quantum gravity. The sizes and shapes of the compact spaces are unspecified a priori, since General Relativity
and Quantum Mechanics tell us that the compact dimensions must vary in size and shape at different places.

15
These changes in size and shape can be described, from a four-dimensional perspective, by a collection of
new fields, analogous to the Newtonian gravitational field, which in General Relativity describes the curvature
of spacetime. Futhermore, there must be some dynamical energy principle that fixes the average size and
shape of the compact dimensions; as the potential in Fig. 3 fixes the axion angle and also gives it a mass, so
these dynamics give masses to the ALPs. This new dynamics is by nature extremely speculative, and (due to
the vast array of possibilities for the topology of the compact space) even the most predictive approaches to
this problem currently make only statistical statements about the masses of the resulting ALPs. However, the
existence of extra dimensions of space is one of the central predictions of string theory and, since it leads to
the existence of ALPs, they too can be considered a prediction of the theory.
Although the forms of ALP couplings are similar to those of the original axion, their strengths are quite
model-dependent, and even less is known about them than about the ALP masses. For this reason, there is
great interest in designing experimental searches for axions that are sufficiently general to also be sensitive to
wide ranges of the parameter spaces of ALP couplings to both fermions and gauge bosons.
We also mention the growing interest in the possible existence of light vector (spin 1) bosons. In the
Standard Model there is, on the one hand, the photon whose mass is thought to be exactly zero and is very
tightly constrained by experimental observations, and on the other hand the massive vector bosons responsible
for the weak interactions, with no vector bosons of intermediate mass. However, it is possible that there might
be additional light vector bosons that are very weakly coupled to the particles of the Standard Model, so-called
‘dark’ or ‘hidden’ vector bosons, as also appear in some string theory models, for example. Their couplings
to Standard Model particles would take different forms from those of ALPs. For example, couplings to pairs
of photons or gluons would be forbidden, and their couplings to fermion-antifermion pairs would be vector-
or axial-vector-like. For this reason, different types of experimental strategies are required to search for them,
and we do not discuss them further in this review.
Finally, we discuss the possibility of axion-like quasiparticles appearing in certain exotic materials in solid
state physics. Quasiparticles such as phonons (quanta of vibrational modes of solids) and magnons (quanta of
magnetization) have been staples of condensed matter theory since the earliest days of quantum mechanics.
For a quasiparticle to behave as an axion, it must be a pseudoscalar, which requires the parent solid to have
certain specific symmetry properties under parity and time-reversal. The would-be axion quasiparicle should
also have the trademark axion-like couplings, in particular to E · B, giving rise to axion electrodyanmics. The
possibility of axion electrodynamics in solids was identified already by Wilczek, one of the fathers of the
axion theory, in the 1980s [79], but only recently has the quest to realise this idea in the laboratory picked up
pace.
Mixing of parallel electric and magnetic fields occurs in general in magnetoelectric materials, and in par-
ticular in topological insulators. The existence of the axion quasiparticle requires the breaking of time-reversal
symmetry, and is provided by magnetic ions and antiferromagnetic order. Candidate materials to host the ax-
ion quasiparticle include Bi(Fe)2 Se3 and Mn2 Bi2 Te5 [80, 81]. The axion quasiparticle has yet to be observed
in the lab, although various experiments have been designed to hunt for it, and there is growing interest in
these candidate materials with possible applications to high-speed computer memory [82]. Furthermore, as
discussed in the experimental companion to this review, axion quasiparticles could also prove useful in the
detection of axion dark matter.

6 Summary, Conclusions, and Musings

The dark matter zeitgeist is in flux. For many years experimental and theoretical efforts were directed at
WIMPs and similar particle candidates. The technology to detect WIMPs was readily available in the form
of the LHC, and the experimental programme of direct WIMP searches also developed rapidly [15, 16].
Theories attempting to explain the origin of the Higgs mass seemed to predict naturally the existence of
WIMP dark matter, with mass comparable to that of the Higgs itself, providing a very tantalising prospect. The
experimental searches for WIMPs performed excellently, yet to date have found nothing. While there are still
important parts of the parameter space left to explore, initial optimism has been dampened by experimental
cold water.
Almost in parallel to the WIMP story, there was, somewhat behind the scenes, the ever-present and de-
veloping story of the axion. Also proposed in the 1970s as a sticking plaster for problems in the Standard
Model [5–7], and with a dark matter candidate for free [51–53], that is where the shared story of axions and
WIMPs ends. The axion is far lighter and more weakly coupled than the WIMP, and initially experimental
searches for the axion made slow and painful progress towards the theoretically defined goals. All of this is

16
changing at present and the axion’s star is in ascendance. Theorists are working hard trying to hem in the
predictions of the axion model, while new technological advances offer prospects for detection in the near
future (see the companion experimental review).
Axion dark matter is an ultralight bosonic field, which manifests itself on Earth primarily as exotic oscil-
lating phenomena violating CP symmetry. The primary and defining oscillation is in electric dipole moments,
the neutron EDM being of particular interest. The axion also acts as an oscillating source in Maxwell’s equa-
tions, and can be detected as an anomalous electric or magnetic field. The oscillations in the axion field were
set up in the early Universe by a process of symmetry breaking and phase transition, and the discovery of the
axion could shed light on those extremely high energy processes in the most distant cosmic past.
The parameter space of axion models is relatively well defined. Astrophysical considerations hem in the
oscillation frequency (and thus the particle mass) to a range between kHz and THz, tightening to GHz to THz
given some assumptions about the phase transition (see Fig. 4). Intense theoretical efforts are being made at
present to try and tighten this range still further, limited by computational resources. Experimental searches
for the axion often rely on resonant effects [29], and it is the task of tuning these experimental “radios” over
the wide range of available frequencies that makes axion searches so long and laborious. If the theorists
could predict the axion mass accurately, which might be within reach, the experimental search could be tuned,
and accomplished within a very short period. However, we should temper our optimism. The theoretical
predictions rely on a large number of assumptions, which could be violated in Nature.
Axion theorists are also intensely engaged in thinking about the axion couplings, g, to ordinary matter [22].
Astrophysical constraints, for example from stellar evolution, often provide a good guide for how large these
couplings can be in general, without theoretical bias. The axion theory predicts g ∝ ma , but the constant of
proportionality varies for different models. A given experiment or observation constrains the value of the
constant of proportionality, and it is for theorists to interpret which, if any, of the available axion models has
been excluded. Again, efforts to pin down the range of allowed couplings are starting to reach some degree of
consensus, just in time for the data onslaught that is coming our way from new experiments.
As described in the companion experimental review, this onslaught of data is already beginning, and
should reach a conclusion in the next decade or so. Experiments with resonant circuits and nuclear magnetic
resonance operate in the kHz range and below, microwave cavities in the GHz range, and the use of dielectrics
and novel materials can push into near THz and above. If these experiments succeed, the next task will be to
identify the axion precisely. One would like to measure the axion via all its interactions, Eq. (5). Fortunately,
once one experiment succeeds and finds the right axion “radio station” (i.e. the value of the axion mass), all
the others will quickly be able to tune in and measure the different couplings of the axion. For example, with
the axion photon and electron couplings in hand, it should be possible to determine whether the axion speaks
to the Standard Model via new heavy quarks [25, 26], or via the Higgs boson [23, 24] (see Supplemental
Material).
An axion dark matter experiment in reality measures a combination of the axion coupling and the dark
matter density, so another key task will be to use overlapping experiments that do not rely on the dark matter
density to break this degeneracy. Fortunately, ideas are available using nuclear magnetic resonance, electric
dipole measurements, searches for new forces, and combinations of searches for axions from the Sun and other
astrophysical sources (see the experimental companion to this review). Axion searches also offer a wealth of
information about the frequency structure of the dark matter, which would provide a detailed view of dark
matter in our galactic neighbourhood, and open a new field of dark matter astronomy [83]. This combination
of astrophysics and particle physics could eventually determine not just if the axion is indeed the dark matter,
but exactly what fraction of the dark matter it is and how it is distributed. If the axion is just one part of the
dark matter, a new series of more challenging problems arise, since all dark matter searches are harder when
aimed at sub-components.
Fortunately, theorists are prepared for such an eventuality. Many theories possessing the axion also possess
a zoo of related ALPs, all or none of which could also contribute to the dark matter, and some of which could
even be related to hidden dimensions of spacetime [78]. If the axion is discovered, it will open up the doors
to experimentally testing these far reaching theories. The interest in these related particles, and experimental
searches for them, is also reaching new heights, but that is a story too varied for this review.
The story of the axion, like much of modern particle physics, began in the 1970s, rooted in ideas of
symmetry. The Standard Model is a symmetry success story, but other theories based on symmetry have
failed. Does Nature really play by these rules? The popularity of the axion has risen as a prime candidate to
explain dark matter, the cosmic glue that in many ways is responsible for our own existence. What composes
this mysterious substance, and how did it get here? The discovery of the axion seems tantalisingly close,
promising answers to these deepest of questions.

17
Acknowledgements
We were saddened to hear during the completion of this paper of the deaths of Toshihide Maskawa and Steven
Weinberg, whose deep theoretical insights underlie the topics reviewed here.

We are indebted to Marco Gorghetto for providing an image used in Fig. 4, to Matthew Stott for provid-
ing an image used in Fig. 6, and to Philip Mocz who provided the images used in Fig. 5, and to all authors
who have allowed us to reproduce images from their published work. The Chandra composite image in Fig. 6
has credit to the following sources, X-ray: NASA/CXC/SAO/PSU [84]; Optical: NASA/STScI [85]; Mil-
limeter: ESO/NAOJ/NRAO/ALMA [86]. We thank Yannis Smertzidis (CAPP), Ian Williams (University of
Surrey), and SungWoo Youn (CAPP) for comments on the draft. FC-D is supported by the UK STFC grant
ST/P001246/1 and Stephen Hawking Fellowship EP/T01668X/1. The work of JE was supported in part by the
UK STFC via grant ST/T000759/1, and in part by the Estonian Research Council via grant MOBTT5. DJEM
was previously supported by the Alexander von Humboldt Foundation and the German Federal Ministry of
Education and Research, and is now supported by a UK STFC Ernest Rutherford Fellowship.

Data and Materials Availability

All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary
Materials.

Author Contributions
All authors contributed equally to the writing and editing of the manuscript.

Competing Interests
The authors declare no competing interests.

References
[1] Aad, G. et al. Observation of a new particle in the search for the Standard Model Higgs boson with the
ATLAS detector at the LHC. Phys. Lett. B 716, 1–29 (2012). 1207.7214.
[2] Chatrchyan, S. et al. Observation of a New Boson at a Mass of 125 GeV with the CMS Experiment at
the LHC. Phys. Lett. B 716, 30–61 (2012). 1207.7235.
[3] Higgs, P. W. Broken Symmetries and the Masses of Gauge Bosons. Phys. Rev. Lett. 13, 508–509 (1964).

[4] Englert, F. & Brout, R. Broken Symmetry and the Mass of Gauge Vector Mesons. Phys. Rev. Lett. 13,
321–323 (1964).
[5] Peccei, R. D. & Quinn, H. R. CP conservation in the presence of pseudoparticles. Phys. Rev. Lett. 38,
1440–1443 (1977).

[6] Weinberg, S. A new light boson? Phys. Rev. Lett. 40, 223–226 (1978).
[7] Wilczek, F. Problem of strong P and T invariance in the presence of instantons. Phys. Rev. Lett. 40,
279–282 (1978).
[8] Dirac, P. A. M. The quantum theory of the electron. Proc. Roy. Soc. Lond. A 117, 610–624 (1928).

[9] Sakharov, A. D. Violation of CP Invariance, C asymmetry, and baryon asymmetry of the universe. Pisma
Zh. Eksp. Teor. Fiz. 5, 32–35 (1967).
[10] Christenson, J. H., Cronin, J. W., Fitch, V. L. & Turlay, R. Evidence for the 2π Decay of the K20 Meson.
Phys. Rev. Lett. 13, 138–140 (1964).
[11] Kobayashi, M. & Maskawa, T. CP Violation in the Renormalizable Theory of Weak Interaction. Prog.
Theor. Phys. 49, 652–657 (1973).

18
[12] Rubin, V. C. & Ford, W. K., Jr. Rotation of the Andromeda Nebula from a Spectroscopic Survey of
Emission Regions. Astrophys. J. 159, 379–403 (1970).
[13] Aghanim, N. et al. Planck 2018 results. VI. Cosmological parameters. Astron. Astrophys. 641, A6
(2020). 1807.06209.
[14] Peebles, P. J. E. Large scale background temperature and mass fluctuations due to scale invariant
primeval perturbations. Astrophys. J. Lett. 263, L1–L5 (1982).
[15] Jungman, G., Kamionkowski, M. & Griest, K. Supersymmetric dark matter. Phys. Rept. 267, 195–373
(1996). hep-ph/9506380.
[16] Aprile, E. et al. Dark Matter Search Results from a One Ton-Year Exposure of XENON1T. Phys. Rev.
Lett. 121, 111302 (2018). 1805.12562.
[17] Clowe, D., Gonzalez, A. & Markevitch, M. Weak lensing mass reconstruction of the interacting cluster
1E0657-558: Direct evidence for the existence of dark matter. Astrophys. J. 604, 596–603 (2004).
astro-ph/0312273.
[18] Abel, C. et al. Measurement of the permanent electric dipole moment of the neutron. Phys. Rev. Lett.
124, 081803 (2020). 2001.11966.
[19] Nambu, Y. Axial vector current conservation in weak interactions. Phys. Rev. Lett. 4, 380–382 (1960).
[20] Goldstone, J. Field Theories with Superconductor Solutions. Nuovo Cim. 19, 154–164 (1961).
[21] Grilli di Cortona, G., Hardy, E., Pardo Vega, J. & Villadoro, G. The QCD axion, precisely. JHEP 01,
034 (2016). 1511.02867.
[22] Di Luzio, L., Giannotti, M., Nardi, E. & Visinelli, L. The landscape of QCD axion models. Phys. Rept.
870, 1–117 (2020). 2003.01100.
[23] Dine, M., Fischler, W. & Srednicki, M. A Simple Solution to the Strong CP Problem with a Harmless
Axion. Phys. Lett. B 104, 199–202 (1981).
[24] Zhitnitsky, A. R. On Possible Suppression of the Axion Hadron Interactions. (In Russian). Sov. J. Nucl.
Phys. 31, 260 (1980).
[25] Kim, J. E. Weak-interaction singlet and strong CP invariance. Phys. Rev. Lett. 43, 103–107 (1979). URL
https://link.aps.org/doi/10.1103/PhysRevLett.43.103.
[26] Shifman, M., Vainshtein, A. & Zakharov, V. Can confinement ensure natural cp invariance of strong
interactions? Nuclear Physics B 166, 493–506 (1980). URL https://www.sciencedirect.com/
science/article/pii/0550321380902096.
[27] Moody, J. & Wilczek, F. NEW MACROSCOPIC FORCES? Phys. Rev. D 30, 130 (1984).
[28] Arvanitaki, A. & Geraci, A. A. Resonantly Detecting Axion-Mediated Forces with Nuclear Magnetic
Resonance. Phys. Rev. Lett. 113, 161801 (2014). 1403.1290.
[29] Sikivie, P. Experimental tests of the ’invisible’ axion. Phys. Rev. Lett. 51, 1415–1417 (1983).
[30] Barbieri, R., Cerdonio, M., Fiorentini, G. & Vitale, S. AXION TO MAGNON CONVERSION: A
SCHEME FOR THE DETECTION OF GALACTIC AXIONS. Phys. Lett. B 226, 357–360 (1989).
[31] Budker, D., Graham, P. W., Ledbetter, M., Rajendran, S. & Sushkov, A. O. Proposal for a Cosmic Axion
Spin Precession Experiment (CASPEr). Phys. Rev. X 4, 021030 (2014). 1306.6089.
[32] Weinberg, S. Nonlinear realizations of chiral symmetry. Phys. Rev. 166, 1568–1577 (1968).
[33] Asztalos, S. J. et al. SQUID-Based Microwave Cavity Search for Dark-Matter Axions. Phys. Rev. Lett.
104, 041301 (2010). 0910.5914.
[34] Braine, T. et al. Extended Search for the Invisible Axion with the Axion Dark Matter Experiment. Phys.
Rev. Lett. 124, 101303 (2020). 1910.08638.

19
[35] Semertzidis, Y. K. et al. Axion Dark Matter Research with IBS/CAPP (2019). 1910.11591.
[36] Graham, P. W. & Rajendran, S. New observables for direct detection of axion dark matter. Phys. Rev. D
88, 035023 (2013). 1306.6088.
[37] Crescini, N. et al. Operation of a ferromagnetic axion haloscope at m a=58 µ eV. European Physical
Journal C 78, 703 (2018). 1806.00310.
[38] Marsh, D. J. E., Fong, K. C., Lentz, E. W., Šmejkal, L. & Ali, M. N. Proposal to Detect Dark Matter
using Axionic Topological Antiferromagnets. Phys. Rev. Lett. 123, 121601 (2019). 1807.08810.
[39] Aybas, D. et al. Search for Axionlike Dark Matter Using Solid-State Nuclear Magnetic Resonance.
Phys. Rev. Lett. 126, 141802 (2021). 2101.01241.
[40] Horns, D. et al. Searching for WISPy cold dark matter with a dish antenna. Journal of Cosmology and
Astroparticle Physics 2013, 016 (2013). 1212.2970.
[41] Caldwell, A. et al. Dielectric Haloscopes: A New Way to Detect Axion Dark Matter. Phys. Rev. Lett.
118, 091801 (2017). 1611.05865.
[42] Egge, J., Knirck, S., Majorovits, B., Moore, C. & Reimann, O. A first proof of principle booster setup
for the MADMAX dielectric haloscope. Eur. Phys. J. C 80, 392 (2020). 2001.04363.
[43] Kahn, Y., Safdi, B. R. & Thaler, J. Broadband and Resonant Approaches to Axion Dark Matter Detec-
tion. Phys. Rev. Lett. 117, 141801 (2016). 1602.01086.
[44] Salemi, C. P. et al. Search for Low-Mass Axion Dark Matter with ABRACADABRA-10 cm. Phys. Rev.
Lett. 127, 081801 (2021). 2102.06722.
[45] Reynolds, C. S. et al. Astrophysical limits on very light axion-like particles from Chandra grating
spectroscopy of NGC 1275 (2019). 1907.05475.
[46] Brockway, J. W., Carlson, E. D. & Raffelt, G. G. SN1987A gamma-ray limits on the conversion of
pseudoscalars. Phys. Lett. B 383, 439–443 (1996). astro-ph/9605197.
[47] Ayala, A., Domı́nguez, I., Giannotti, M., Mirizzi, A. & Straniero, O. Revisiting the bound on axion-
photon coupling from Globular Clusters. Phys. Rev. Lett. 113, 191302 (2014). 1406.6053.
[48] Anastassopoulos, V. et al. New CAST Limit on the Axion-Photon Interaction. Nature Phys. 13, 584–590
(2017). 1705.02290.
[49] Vaquero, A., Redondo, J. & Stadler, J. Early seeds of axion miniclusters. JCAP 04, 012 (2019). 1809.
09241.
[50] Armengaud, E. et al. Physics potential of the International Axion Observatory (IAXO). JCAP 2019,
047 (2019). 1904.09155.
[51] Preskill, J., Wise, M. B. & Wilczek, F. Cosmology of the invisible axion. Phys. Lett. B 120, 127–132
(1983).
[52] Abbott, L. F. & Sikivie, P. A cosmological bound on the invisible axion. Phys. Lett. B 120, 133–136
(1983).
[53] Dine, M. & Fischler, W. The not-so-harmless axion. Phys. Lett. B 120, 137–141 (1983).
[54] Kibble, T. W. B. Topology of Cosmic Domains and Strings. J. Phys. A 9, 1387–1398 (1976).
[55] Zurek, W. H. Cosmological Experiments in Superfluid Helium? Nature 317, 505–508 (1985).
[56] Hogan, C. J. & Rees, M. J. Axion miniclusters. Phys. Lett. B 205, 228–230 (1988).
[57] Khlopov, M., Malomed, B. & Zeldovich, I. Gravitational instability of scalar fields and formation of
primordial black holes. Mon. Not. Roy. Astron. Soc. 215, 575–589 (1985).
[58] Seidel, E. & Suen, W. M. Oscillating soliton stars. Phys. Rev. Lett. 66, 1659–1662 (1991).

20
[59] Guzman, F. S. & Urena-Lopez, L. A. Gravitational cooling of self-gravitating Bose-Condensates. As-
trophys. J. 645, 814–819 (2006). astro-ph/0603613.
[60] Schive, H.-Y., Chiueh, T. & Broadhurst, T. Cosmic structure as the quantum interference of a coherent
dark wave. Nature Physics 10, 496–499 (2014). 1406.6586.

[61] Hui, L., Joyce, A., Landry, M. J. & Li, X. Vortices and waves in light dark matter. JCAP 01, 011 (2021).
2004.01188.
[62] Picozzi, A. & Garnier, J. Incoherent soliton turbulence in nonlocal nonlinear media. Phys. Rev. Lett.
107, 233901 (2011). URL https://link.aps.org/doi/10.1103/PhysRevLett.107.233901.

[63] Mocz, P. et al. Galaxy formation with BECDM – I. Turbulence and relaxation of idealized haloes. Mon.
Not. Roy. Astron. Soc. 471, 4559–4570 (2017). 1705.05845.
[64] Mocz, P. et al. First star-forming structures in fuzzy cosmic filaments. Phys. Rev. Lett. 123, 141301
(2019). 1910.01653.
[65] Marsh, D. J. E. & Niemeyer, J. C. Strong Constraints on Fuzzy Dark Matter from Ultrafaint Dwarf
Galaxy Eridanus II. Phys. Rev. Lett. 123, 051103 (2019). 1810.08543.
[66] Rogers, K. K. & Peiris, H. V. Strong bound on canonical ultra-light axion dark matter from the Lyman-
alpha forest. Phys. Rev. Lett. 126, 071302 (2021). 2007.12705.
[67] Hlozek, R., Marsh, D. J. E. & Grin, D. Using the Full Power of the Cosmic Microwave Background to
Probe Axion Dark Matter. Mon. Not. Roy. Astron. Soc. 476, 3063–3085 (2018). 1708.05681.
[68] Veltmaat, J., Niemeyer, J. C. & Schwabe, B. Formation and structure of ultralight bosonic dark matter
halos. Phys. Rev. D 98, 043509 (2018). 1804.09647.
[69] URL https://chandra.harvard.edu/photo/2017/sn1987a/.

[70] Penrose, R. & Floyd, R. Extraction of rotational energy from a black hole. Nature 229, 177–179 (1971).
[71] Arvanitaki, A., Dimopoulos, S., Dubovsky, S., Kaloper, N. & March-Russell, J. String axiverse. Phys.
Rev. D 81, 123530 (2010). 0905.4720.
[72] Stott, M. J. & Marsh, D. J. E. Black hole spin constraints on the mass spectrum and number of axionlike
fields. Phys. Rev. D 98, 083006 (2018). 1805.02016.

[73] Arkani-Hamed, N., Motl, L., Nicolis, A. & Vafa, C. The String landscape, black holes and gravity as the
weakest force. JHEP 06, 060 (2007). hep-th/0601001.
[74] Raffelt, G. G. Astrophysical axion bounds. Lect. Notes Phys. 741, 51–71 (2008). hep-ph/0611350.
[75] Klaer, V. B. & Moore, G. D. The dark-matter axion mass. JCAP 11, 049 (2017). 1708.07521.

[76] Gorghetto, M., Hardy, E. & Villadoro, G. More Axions from Strings. SciPost Phys. 10, 050 (2021).
2007.04990.
[77] Hiramatsu, T., Kawasaki, M., Saikawa, K. & Sekiguchi, T. Axion cosmology with long-lived domain
walls. JCAP 01, 001 (2013). 1207.3166.

[78] Svrcek, P. & Witten, E. Axions In String Theory. JHEP 06, 051 (2006). hep-th/0605206.
[79] Wilczek, F. Two applications of axion electrodynamics. Phys. Rev. Lett. 58, 1799–1802 (1987). URL
https://link.aps.org/doi/10.1103/PhysRevLett.58.1799.
[80] Li, R., Wang, J., Qi, X.-L. & Zhang, S.-C. Dynamical axion field in topological magnetic insulators.
Nature Physics 6, 284–288 (2010). 0908.1537.
[81] Zhang, J. et al. Large Dynamical Axion Field in Topological Antiferromagnetic Insulator Mn2 Bi2 Te5 .
Chinese Physics Letters 37, 077304 (2020). URL https://doi.org/10.1088%2F0256-307x%2F37%
2F7%2F077304.

21
[82] Cheng, Y., Peng, B., Hu, Z., Zhou, Z. & Liu, M. Recent development and status of magnetoelectric
materials and devices. Physics Letters A 382, 3018–3025 (2018).
[83] O’Hare, C. A. J. & Green, A. M. Axion astronomy with microwave cavity experiments. Phys. Rev. D95,
063017 (2017). 1701.03118.

[84] Frank, K. A. et al. Chandra Observes the end of an era in SN 1987a. Astrophys. J. 829, 40 (2016).
1608.02160.
[85] Orlando, S., Miceli, M., Pumo, M. L. & Bocchino, F. Supernova 1987A: a Template to Link Supernovae
to their Remnants. Astrophys. J. 810, 168 (2015). 1508.02275.

[86] Indebetouw, R. et al. Dust Production and Particle Acceleration in Supernova 1987A Revealed with
ALMA. Astrophys. J. Lett. 782, L2 (2014). 1312.4086.
[87] Dematteis, G., Grafke, T., Onorato, M. & Vanden-Eijnden, E. Experimental evidence of hydrodynamic
instantons: The universal route to rogue waves. Phys. Rev. X 9, 041057 (2019). URL https://link.
aps.org/doi/10.1103/PhysRevX.9.041057.

[88] Coleman, S. Aspects of Symmetry (Cambridge University Press, 1988).

[89] Vafa, C. & Witten, E. Parity Conservation in QCD. Phys. Rev. Lett. 53, 535 (1984).
[90] Georgi, H. & Randall, L. Flavor conserving cp violation in invisible axion models. Nuclear
Physics B 276, 241–252 (1986). URL https://www.sciencedirect.com/science/article/pii/
0550321386900222.

22
A Supplemental: Theoretical Methods
A.1 The Axion and Quantum Chromodynamics
We look here in more detail at the strong CP problem and the axion. When we write down the Lagrangian
density for Quantum Chromodynamics (QCD) we must include all terms allowed by the symmetries of the
theory. This means that the Lagrangian density should include the CP-violating term In the supplemental
material, we use natural units with c = } = 1.

g23
L ⊃ θ̄ Gµν · G̃µν , (11)
32π 2
where g3 is the coupling constant of the strong force, Gµν is the gluon field strength tensor, G̃µν = εµνρσ Gρσ
is its dual, and θ̄ includes contributions from the quark masses and from the structure of the vacuum itself:

θ̄ = θ + arg det(M), (12)

where θ is the QCD vacuum angle and M is the quark mass matrix. We seek an explanation for why θ̄ is so
small.

To this end, we introduce the global Peccei-Quinn symmetry U(1)PQ . The spontaneous breaking of this
symmetry yields a massless Nambu-Goldstone boson, the axion. This means that under a U(1)PQ transforma-
tion with parameter ε the axion field transforms like:

a(x) →
− a(x) + ε fa . (13)
When the U(1)PQ symmetry is exact, the axion must be exactly massless, so that adding ε fa to the axion field
results in the same energy density.
The axion interacts with gluons via the Lagrangian density

a g23
L ⊃ Gµν · G̃µν . (14)
fa 32π 2
Notice that this has exactly the same form as the problematic strong CP-violating term (11), but the coefficient
of G · G̃ is now a dynamical field. This is what allows the total coefficient of G · G̃ to relax to zero in the ground
state of the theory. How this term is generated is discussed in more detail in the next subsection. We note in
g
passing that the axion’s coupling to photons takes the same form L ⊃ aγγ 4 F · F̃ = gaγγ E · B, where F is the
electromagnetic field strength tensor.

The axion’s coupling to gluons also ensures that the U(1)PQ symmetry is anomalous. An anomalous
symmetry is a symmetry that would be conserved in a classical theory but is broken by quantum effects. Put
another way, the theory’s Lagrangian is invariant under the symmetry, but the path inegral measure is not.
This means that the axion is no longer massless, but is a pseudo-Nambu-Goldstone boson. The potential
energy, and therefore the mass, of the axion field is generated by QCD instantons. Instantons are non-trivial
solutions to the classical equations of motion, whose effects are not captured by perturbative quantum field
theory methods. Instantons have been observed experimentally in the context of fluid dynamics [87].
The effects of instantons in QCD can be accounted for by a θ -dependent change to the vacuum energy,
leading to an effective potential for the axion. In the dilute instanton gas approximation (see e.g. Ref. [88])
applicable to high-temperature QCD, the potential is:
 
a(x)
V (a(x)) ∝ −cos θ̄ + . (15)
fa

Therefore, minimising the potential energy of the axion field sets the total coefficient of G · G̃ to zero. The
mass of the axion is found from a Taylor expansion of V (a(x)). The constant of proportionality is the QCD
topological susceptibility, which is temperature-dependent and can be computed by instanton methods, in
effective field theory, or using non-perturbative lattice QCD.
How do we know that the minimum of the axion potential is at precisely the value that leads to no CP
violation in strong interactions? One way to see this is via the Vafa-Witten theorem, which tells us that in
vector-like theories, such as QCD, dynamical parity-violating terms are zero in the ground state of the theory

23
 Q Aµ
@µ a
Q
Q
Aµ
Figure S1: The colour anomaly of the Peccei-Quinn symmetry. The current of the PQ symmetry is JµPQ = ∂µ a.
Due to the chiral interaction between the axion and a quark, Q (under which the left- and right-handed Weyl
spinors carry opposite charge), this current is not conserved in the quantum theory. The loop diagram mediates
a non-vanishing interaction between axions and gluons, Aµ .

[89]. The neutron EDM violates P as well as CP, and the axion field make the total neutron EDM a dynamical
variable. Therefore, in the absence of weak interactions, the Vafa-Witten theorem guarantees that the axion
field potential minimum will be at zero neutron EDM. However, the theory of quarks is not quite vector-like
because the quarks are also charged under the chiral weak interaction. This means that the minimum of the
axion potential is slightly shifted from the CP conserving value. The magnitude of this effect can be calculated
using effective field theory, and is found to lead to an effective CP violating angle θ̄ ' O(10−17 ) [90], far
below current experimental bounds.

A.2 Invisible axion models

So far, we have described in only general terms how the axion solves the strong CP problem. Now, we
would like to be more specific. The key property of the axion is its coupling to gluons described above. In
order to solve the strong CP problem, this coupling should be generated by the chiral anomaly mediated by
the coupling between the axion and fermions charged under the strong interactions, leading to the triangle
Feynman diagram in Fig. S1. In the absence of the chiral anomaly, this diagram would have zero amplitude.
In the Standard Model, the fermions charged undeer the strong interactions are the quarks, u, d, c, s,t, b.
In the original axion model [6, 7], the Peccei-Quinn field Φ is a second Higgs field, with one field giving
mass to up-type quarks, and another to down-type quarks. The axion field is then the complex phase of the
combined Higgs field. However, given the known vacuum expectation value of the Higgs field, this model is
ruled out by experimental data from colliders. This led to the development of “invisible” axion models. In
this section we will describe the first two invisible axion models A review of these and more general models
can be found in Ref. [22].. In these models, the Peccei-Quinn field is not related to the Higgs field, and so
its vacuum expectation value vPQ is a free parameter. If vPQ is high enough, the axion’s interactions are weak
enough to have evaded experiments to date The axion decay constant is given by fa = vPQ /N, where N is the
(model dependent) color anomaly of the Peccei-Quinn symmetry.

We describe first the Kim [25] Shifman-Vainshtein-Zakharov [26] (KSVZ) model. The KSVZ model
extends the Standard Model by the introduction of two new fields. The first is the Peccei-Quinn field Φ,
which is uncharged under all the Standard Model interactions. The second is a new heavy vector fermion Q,
which is charged under the strong interactions, i.e., a new exotic quark.The historical KSVZ model also has
an electrical charge for the new quark, with a value inspired by grand unified theories. For simplicity, we do
not discuss further this option. The Lagrangian of this sector is

λ ¯ µ − ig3 Aµ )γ µ Q + (yQ Q¯L QR Φ + y† Q¯R QL Φ† ) .

L = (∂µ Φ)(∂ µ Φ) − (|Φ|2 − v2PQ )2 + iQ(∂ Q (16)
4
The first term is the kinetic term of the PQ field. The second term is the potential of the PQ field, with the
characteristic symmetry breaking form, with λ the PQ self-interaction strength. In the vacuum, the field takes
the form Φ = (vPQ + χ)eia/vPQ , with χ a massive field and a the massless Nambu-Goldstone boson arising
from breaking of the rotational symmetry in the complex plane of Φ, i.e., U(1) symmetry. The third term is

24
the quark kinetic term, with the SU(3) covariant derivative: Aµ is the matrix gluon field, an element of SU(3)
(which appears in the right-hand side of the Feynman diagram). Note that the new quark does not couple to
the Higgs boson, and so has no mass term in the Lagrangian.
The fourth term is where the real action of the PQ mechanism resides: the coupling between the PQ field
and the new quark, with a Yukawa coupling constant yQ . This term gives a mass to the new quark when Φ
gains a vacuum expectation value: Φ acts like the Higgs for the new quark. If the product yQ vPQ is large, then
the new quark is too heavy to have been observed in any collider experiment so far. This interaction term also
leads to a coupling between the axion and the new heavy quark, leading to the required chiral anomaly (this
gives the left-hand side of the Feynman diagram).

The Dine-Fischler-Srednicki [23] Zhitnitsky [24] (DFSZ) model is slightly more subtle to describe, since
it involves changes to the Higgs sector of the Standard Model, splitting the Higgs into up- and down-type
doublets. This additional subtlety avoids the introduction of the new quark, Q, of the KSVZ model, and
instead mediates the coupling between axions and gluons using the quarks of the Standard Model. We do
not discuss this model in detail, since it requires detailed background on the Higgs and the Standard Model.
From the perspective of low-energy phenomenology, the DFSZ model differs from the KSVZ model in that
it possesses a direct coupling between the axion and the electron, gaee , whereas the KSVZ model acquires
this coupling only at loop order, and so it is suppressed in magnitude. Since the electron and the quarks of
the Standard Model also carry electric charge, the DFSZ axion acquires its coupling to electromagnetism via
loop diagrams mediated by the Standard Model fermions, in addition to its mixing with the neutral pion, thus
leading to a different strength of coupling, gaγγ than that in the KSVZ model (see below). These two differ-
ences, in the values of gaee and gaγγ , will allow (in the event that the axion is discovered in the future) for the
KSVZ and DFSZ models (and other axion variants) to be distinguished by measurements as described in the
experimental companion to this review.

Many axion search strategies depend on the axion’s mass and coupling to electromagnetism. These are
related by [21]
αem E E
   
ma
gaγγ = − 1.92(4) = 0.203(3) − 0.39(1) , (17)
2π fa N N GeV2
where the numbers in brackets denote theoretical uncertainties, αem is the electromagnetic fine structure con-
stant, E is the electromagnetic anomaly of the PQ symmetry and N the colour anomaly. The model dependent
coupling coefficient Cγ = EN − 1.92(4) is often used. The first term in Eq. (17) is generated by the mixing
between the axion and the neutral pion, and is model-independent (so long as the axion has a coupling to glu-
ons and solves the strong CP problem: the coupling can be absent for an ALP), while E and N in the second
term are model-dependent. For example, in the DFSZ model EN = 83 (generated by the fermions charged under
both electromagnetism and PQ), whereas for the KSVZ model described here EN = 0. In principle, EN could be
large, leading to a large value for |gaγγ | (there is some maximum set by unitarity). Alternatively, EN could by
chance be such that the two terms in Eq. (17) approximately cancel, leading to a very small value for |gaγγ |.
However, such a model would need to be deliberately constructed so as to hide the axion, and in general we
take the DFSZ model to give the smallest canonical value of |gaγγ |. Precisely outlining the allowed range of
gaγγ in viable models is an active area of research.

A.3 Equations of Axion Dark Matter

The Friedmann equation:
2
8π 3 GN

1 dR
≡ H(t)2 = g? (T )T 4 , (18)
R(t)2 dt 90
where t is time “since the Big Bang”, T is temperature, GN is the Newton constant, g? (T ) counts the num-
ber of relativistic particles at temperature T (with different factors for bosons and fermions), and we have
defined the Hubble rate H(t) from the cosmic scale factor R(t). The evolution of temperature with time, and
also g? (T ), can be found from elementary thermodynamic considerations. At the beginning of the hot Big
Bang, the temperture was at its maximum value, Thot , the time was thot , and the value of R(thot ) is chosen by
normalisation.

25
 In the simple case described by Scenario A, the axion field evovles according to the Klein-Gordon equa-
tion:
d2a da
+ 3H(t) + ma (T )2 a = 0 , (19)
dt 2 dt
with initial condition θ (t hot ) = θi , θ̇ (t hot ) = 0, and the axion number density is
"   #
1 1 da 2 1
na = + ma (T )2 a2 . (20)
ma (T ) 2 dt 2

The temperature dependence of the axion mass, ma (T ), can be computed from the topological susceptibility as
described briefly above. At temperatures above the QCD phase transition the mass grows with temperature,
roughly described by a power law, ma ∝ T 4 , while at low temperature it has the asymptotic value given in
Eq. (3). The Hubble parameter acts as a friction term in the Klein-Gordon equation, which falls over time
as the Universe expands and cools. The axion field begins damped harmonic motion at the epoch when
H(T ) ≈ ma (T ) (labelled tdyn in Fig. 4). At some slightly later time, tcold , the oscillation period of the axion
field becomes large compared to the Hubble rate, and the Klein-Gordon equation enters an attractor solution
in which the number density is conserved modulo the expansion of the Universe, i.e., na (t)/R(t)3 = const..
The axion dark matter density is then given by ρa = ma na , and is a function of the zero-temperature axion
mass and the initial field value.
At late times the axion is described by the Gross-Piatevski-Poisson equations. These equations arise as
the non-relativistic limit of the Klein-Gordon equation for the axion
p field in a weakly perturbed Friedmann-
Robertson-Walker spacetime, under a phase approximation, a = 2/ma (ψeima t/} + ψ ∗ e−ima t/} ). The “wave
function” ψ obeys:

∂ψ }2 2 λa
i} + ∇ ψ− |ψ|2 ψ − mVN ψ = 0 , (21)
∂t 2ma 2ma
∇2VN = 4πGN (ma |ψ|2 + ρ) , (22)

where λa is the dimensionless axion self-interaction coupling constant. In the dilute instanton gas approxi-
mation, λa = −m2a / fa2 . The mass density of axions to leading order is ma |ψ|2 . The Newtonian gravitational
potential is VN , and ρ represents the density of all other gravitating matter.

26