Axion Dark Matter

A puzzle the Standard Model never asked for

Quantum chromodynamics is, by accident of its non-abelian gauge structure, allowed to violate CP. The Lagrangian admits a "topological" term

L_θ = θ_QCD · (g_s² / 32π²) · G_μν^a · G̃^μν_a

where G is the gluon field strength and θ_QCD is a dimensionless angle. After diagonalising the quark mass matrix, the physically observable parameter is θ̄ = θ_QCD + arg det M_q. Nothing in the Standard Model fixes this number: it could naturally have been of order unity. Yet a nonzero θ̄ would produce an electric dipole moment for the neutron of order d_n ≈ θ̄ × 10⁻¹⁶ e·cm, while precision experiments cap |d_n| at about 1.8 × 10⁻²⁶ e·cm. The bound is

|θ̄| ≲ 10⁻¹⁰

This is the strong CP problem. Anthropics will not save you: nothing about life would change if θ̄ were 0.1. Nor is θ̄ = 0 protected by any continuous symmetry of the gauge-coupled Standard Model. The smallness is just there, like a coin balanced on its edge.

Peccei, Quinn, and the dynamical relaxation idea

In 1977 Roberto Peccei and Helen Quinn published a clever workaround. Extend the Standard Model by a global U(1)_PQ symmetry, broken spontaneously at some high scale f_a. The associated pseudo-Nambu-Goldstone boson, the axion field a(x), shifts under the symmetry: a → a + const · f_a. Crucially, U(1)_PQ has a colour anomaly, so the axion couples to gluons exactly like the θ term:

L ⊃ (a/f_a) · (g_s² / 32π²) · G G̃

The effective θ is therefore promoted from a constant to a dynamical field θ̄_eff = θ̄ + a/f_a. QCD instantons induce a potential that, by general theorems (Vafa-Witten), is minimised at θ̄_eff = 0. The axion field automatically rolls to that minimum, neutralising any starting value of θ̄. CP is conserved in QCD not because of a tuning but because the universe found the bottom of a bowl.

Steven Weinberg and Frank Wilczek pointed out independently in 1978 that this implied a physical particle — the axion — with calculable properties. The mass and decay constant are linked through chiral perturbation theory:

m_a · f_a ≈ m_π · f_π · √(z) / (1 + z)   with z = m_u/m_d ≈ 0.48
m_a ≈ 5.7 μeV · (10¹² GeV / f_a)

So a high symmetry-breaking scale gives a feeble, light axion; a low scale gives a heavier, more strongly interacting one. Astrophysical and cosmological bounds cut deeply into both extremes, leaving the canonical "QCD axion window":

f_a ~ 10⁹ – 10¹² GeV    ↔    m_a ~ μeV – meV

Producing dark matter without thermal contact

The original axion proposal said nothing about dark matter. The connection came in 1983, in three near-simultaneous papers by Preskill-Wise-Wilczek, Abbott-Sikivie, and Dine-Fischler. The mechanism is the misalignment mechanism, and it is elegantly simple.

Before the QCD phase transition at T ≈ 150 MeV the instanton-induced axion potential is essentially flat: the axion has no preferred value. In each causally connected patch the field sits at some initial angle θ_i = a_i / f_a uniformly drawn from [-π, π]. As the universe cools and QCD's chiral condensate forms, the cosine potential

V(a) = m_a(T)² f_a² · [1 − cos(a/f_a)]

switches on. The mass grows from zero to its zero-temperature value. The field, initially over-damped by Hubble friction (3H > m_a), is left out of equilibrium and starts to oscillate about its minimum when 3H ∼ m_a, at a temperature T_osc ∼ 1 GeV. These coherent oscillations of a classical scalar field have zero pressure and redshift like cold matter:

ρ_a ∝ a(t)⁻³

where a(t) here is the cosmic scale factor. The relic density scales as

Ω_a h² ≈ 0.12 · (f_a / 9 × 10¹¹ GeV)^(7/6) · θ_i²

For a natural θ_i of order unity, all of dark matter is explained by a QCD axion with f_a ≈ 10¹² GeV, m_a ≈ 5 μeV. This is the "sweet spot" of the post-inflation scenario. The axion is never in thermal contact with the bath, never freezes out — it is produced cold by a coherent oscillation of a vacuum field, the opposite of the thermal-WIMP story.

The two-photon coupling — the experimentalist's handle

Once U(1)_PQ mixes with electroweak anomalies, the axion picks up a coupling to two photons:

L_aγγ = −(1/4) g_aγγ · a · F_μν · F̃^μν     (so a → γ γ)
g_aγγ = (α / 2π f_a) · (E/N − 1.92)

The model-dependent ratio E/N comes from the electromagnetic over colour anomaly coefficients. The two canonical "benchmark" models give different values:

• KSVZ (Kim 1979; Shifman-Vainshtein-Zakharov 1980). Adds a single heavy vector-like quark carrying PQ charge; SM fermions are PQ-neutral. E/N = 0.
• DFSZ (Dine-Fischler-Srednicki 1981; Zhitnitsky 1980). Two Higgs doublets carry PQ charge; SM fermions couple to the axion at tree level. E/N = 8/3.

Their predicted g_aγγ differ by about a factor of three at fixed m_a, and both define the "QCD axion band" that experimentalists chase. Outside this band live "axion-like particles" (ALPs) — particles with the same couplings but a free mass–coupling relation, predicted in string compactifications and many BSM extensions.

In an external magnetic field B, a → γ conversion proceeds through the Primakoff process. The conversion probability in a volume L is

P(a → γ) ≈ (g_aγγ · B · L / 2)²    (when m_a L / 2E ≪ 1)

That dependence on g², B², L² is the lever every haloscope and helioscope pulls.

Hunting the local galactic axion

If axions make up the dark matter halo, the local density is ρ_DM ≈ 0.4 GeV/cm³. For m_a ∼ μeV this is an enormous number density of bosons: n_a ∼ 10¹⁴/cm³. They behave classically as a coherent field oscillating at frequency ν = m_a c² / h:

m_a = 1 μeV   ↔   ν ≈ 242 MHz   ↔   λ_Compton ≈ 1.2 m

Pierre Sikivie's 1983 proposal: put a tunable microwave cavity inside a strong solenoidal field. When the cavity resonance ν matches m_a c² / h, the a → γ conversion is enhanced by the cavity quality factor Q ∼ 10⁵–10⁶, and the dark matter halo "sings" into the cavity at a power of order 10⁻²³ W. Scan the cavity frequency, look for a thermal-noise excess.

Experiment	Mass window	Technique	Status
ADMX (Seattle)	2.7 – 4.2 μeV (and climbing)	Resonant cavity, 8 T, dilution-fridge SQUID	KSVZ-sensitive; DFSZ-sensitive in low-mass band
HAYSTAC (Yale)	17 – 30 μeV	Cavity + squeezed-state readout (below quantum limit)	Probing above ADMX, scanning ongoing
ORGAN (Perth)	~26 μeV	Higher-frequency cavity	First exclusion bands published
MADMAX (DESY)	40 – 400 μeV	Dielectric haloscope (booster stacks)	Demonstrator phase
ABRACADABRA / DMRadio	10⁻¹² – 10⁻⁶ eV	Broadband toroidal magnet + SQUID magnetometer	Active R&D toward DMRadio-50L and beyond
CASPEr	10⁻²² – 10⁻⁷ eV (gradient channel)	NMR — axion DM modulates nuclear precession	Multiple frequency channels operating
BREAD	~meV	Reflector dish — coherent emission from a magnetised boundary	Prototype
IAXO (helioscope)	up to ~eV (solar)	X-ray telescope behind a magnet, pointing at the Sun	Successor to CAST; under construction

The reason there are eight rows is that the QCD axion mass window spans six orders of magnitude, and a single instrument cannot scan all of it. Each technology fits a different frequency band, the way radio receivers, optical telescopes, and X-ray detectors partition the photon spectrum.

Ultralight axions and fuzzy dark matter

Push f_a to the GUT scale and beyond, and the axion mass drops below 10⁻²⁰ eV. The de Broglie wavelength of a non-relativistic axion at typical galactic velocity v ∼ 200 km/s is

λ_dB = h / (m_a · v) ≈ 0.5 kpc · (10⁻²² eV / m_a) · (200 km/s / v)

For m_a ∼ 10⁻²² eV — the canonical "fuzzy" or "wave" dark matter target — this is a kiloparsec, comparable to the size of a dwarf galaxy. On those scales the dark matter behaves as a coherent quantum wave rather than as classical particles. The Schrödinger-Poisson equations replace the Vlasov-Poisson equations, and a quantum pressure term ℏ²/(2m_a²) ∇²(√ρ)/√ρ — sometimes called Madelung pressure — appears in the fluid equations.

The phenomenology:

• Cores instead of cusps. Halos develop a flat-density "soliton" at the centre, of radius ∼ λ_dB, set by the balance of quantum pressure and gravity. This evades the cuspy-halo prediction of cold dark matter.
• Suppressed small-scale power. Structures with mass below ∼10⁹ M☉ are washed out by the de Broglie scale. This eases the missing-satellites problem.
• Granules and interference. Inside larger halos the field is a superposition of de Broglie wavelets, producing time-varying density "granules" of size λ_dB that gravitationally perturb cold stellar streams and globular clusters.

The same physics provides constraints. Lyman-α forest power spectra at z ∼ 5 push m_a ≳ 2 × 10⁻²¹ eV. Ultra-faint dwarf galaxy kinematics, globular-cluster heating in Eridanus II, and stream perturbations narrow the window further. Pure-fuzzy m_a = 10⁻²² eV is now tightly constrained for a 100% DM fraction, but the upper end of the band remains a target for 21-cm cosmology and pulsar-timing-array searches for axion-induced oscillating pressure.

Black-hole superradiance: a no-target detector

A spinning Kerr black hole carries a reservoir of rotational energy outside its horizon, the ergoregion. Any bosonic field whose Compton wavelength is comparable to the gravitational radius can, in suitable angular-momentum eigenstates, extract that energy through a process called superradiance. The condition for amplification is

ω < m_z · Ω_H            (Penrose-Misner condition)
α ≡ G M m_a c / ℏ ≈ 0.1 – 1   (efficient regime)

An ultralight boson builds up an exponentially growing "gravitational atom" of bound states around the hole, drawing spin and mass-energy from the Kerr geometry. The hole spins down on timescales much shorter than its lifetime over a sweet spot in m_a. Because we observe astrophysical black holes that are rapidly spinning — both stellar-mass (X-ray binary spin measurements) and supermassive (continuum fitting of AGN disks) — entire bands of axion mass are excluded.

• Stellar-mass black holes (1 – 100 M☉) exclude roughly 10⁻¹³ ≲ m_a ≲ 10⁻¹¹ eV.
• Supermassive black holes (10⁶ – 10¹⁰ M☉) exclude roughly 10⁻²⁰ ≲ m_a ≲ 10⁻¹⁷ eV.

The result is a particle-physics constraint derived without any direct detection. The detectors are the black holes themselves; the signal is that they still spin.

Astrophysical bounds — stellar cooling and SN1987A

Even without a direct laboratory signal, axions show up wherever they would drain energy from hot environments. Three classes of stellar-cooling bound matter:

• Horizontal-branch stars. Helium-burning stars in globular clusters compare predicted lifetimes against observed numbers of stars on the HB versus red giant branch. Excess axion bremsstrahlung in the He core would shorten the HB lifetime, depleting the HB-to-RGB ratio. The bound is g_aγγ < 6.5 × 10⁻¹¹ GeV⁻¹.
• Red giants and white dwarfs. Axion-electron coupling g_aee accelerates RGB-tip cooling and WD luminosity-function evolution. Hints of "anomalous" extra cooling at the few-σ level have been claimed in WDs and HB stars, fuelling speculation that a real axion-electron interaction is nearby. The bound is currently g_aee < 1.5 × 10⁻¹³.
• SN1987A. The proto-neutron-star core sits at T ∼ 30 MeV and ρ ∼ 3 × 10¹⁴ g/cm³. Axion bremsstrahlung off nucleons (NN → NN a) would carry energy out faster than neutrinos, shortening the observed ν̄_e burst at Kamiokande (12 events) and IMB (8 events) below the measured ∼10 s. The classic Raffelt analysis excludes a swath of axion-nucleon couplings centred near f_a ∼ 10⁸ – 10⁹ GeV; modern reanalyses set f_a ≳ 4 × 10⁸ GeV.

This combined astrophysical floor — together with the upper bound from over-closing the universe — is what brackets the "classical QCD axion window" of f_a between roughly 10⁸ and 10¹² GeV.

Axion-photon conversion in galactic magnetic fields

If axions stream through cosmic magnetic fields, a small fraction convert into photons at the same energy. This opens cosmological lines of sight:

• Solar axion conversion. Axions produced in the solar core (Primakoff, ABC processes) reach Earth at keV energies. CAST at CERN — a magnet pointed at the Sun — set g_aγγ < 0.66 × 10⁻¹⁰ GeV⁻¹. IAXO (under construction) will improve by 1–2 orders of magnitude.
• Galactic-cluster magnetic fields. ALPs from distant AGN convert to X-rays in cluster magnetic fields, producing irregularities in observed X-ray spectra. Chandra and XMM-Newton constraints on M87, NGC 1275 and similar systems probe g_aγγ < 10⁻¹² GeV⁻¹ for m_a ≲ 10⁻¹¹ eV.
• Galactic magnetic field. Neutron-star magnetospheres act as resonant converters when the local plasma frequency crosses m_a; SKA, FAST and Effelsberg radio searches for a coherent emission line are now under way.

Post-inflation vs. pre-inflation cosmology

The relic abundance formula depends sensitively on whether U(1)_PQ breaking happens before or after inflation:

• Pre-inflation (single domain). The whole observable universe is in one PQ vacuum, so θ_i is a single random number. The relic abundance can be tuned by choosing θ_i small, allowing higher f_a and lighter axions. Isocurvature constraints from CMB tightly bound the inflation scale H_I.
• Post-inflation (multi-domain). Many causally disconnected PQ patches form. Topological defects — axion strings and domain walls — radiate axions until they decay. The relic density is the integrated yield of string radiation; this fixes m_a to a narrow band, roughly 25–500 μeV, that experiments are now scanning.

The post-inflation scenario therefore has a sharp target. Recent lattice simulations of cosmic-string networks predict m_a ≈ 40–100 μeV, putting MADMAX and HAYSTAC's planned reach right on top of theory.

Variants and extensions

• QCD axion (KSVZ, DFSZ). The "true" axion — couplings related to mass by m_a f_a ≈ m_π f_π. Two benchmark models, both in the laboratory's sights.
• Axion-like particles (ALPs). Pseudoscalars with axion-style couplings but no fixed mass-coupling relation. Generic in string compactifications, where dozens of moduli give "axiverses" populating every decade in m_a.
• Fuzzy / wave dark matter. m_a ∼ 10⁻²²–10⁻²⁰ eV. Quantum-pressure halos, granular interference, oscillating gravitational potentials detectable by pulsar-timing arrays.
• Dark photon (heavy gauge boson). Often searched for with the same haloscope hardware, since a → γ conversion via kinetic mixing yields a similar resonant signal. ADMX and DMRadio cover both targets.
• Composite axions / familons. Axion-like states from broken family symmetries; couple flavour-diagonally and show up in rare meson decays.

Common pitfalls

• Conflating "axion" with "ALP". The QCD axion has a specific mass-coupling band; ALPs do not. Most exclusion plots distinguish the two — the diagonal "DFSZ" and "KSVZ" lines are predictions only for the QCD axion.
• Forgetting that m_a depends on temperature. The instanton potential is suppressed at T > 1 GeV by Q_topological(T). The oscillation start, and therefore Ω_a, depend on m_a(T_osc), not the zero-temperature value. Misalignment formulae must use the temperature-dependent mass.
• Treating the misalignment angle as fixed. In the pre-inflation scenario, θ_i is a free parameter of order unity; one can choose it small to evade overclosure at high f_a. In the post-inflation scenario, θ_i is averaged ⟨θ_i²⟩ = π²/3 over patches, and isocurvature is gone — but defect emission complicates the abundance calculation.
• Quoting only g_aγγ. Astrophysical bounds also constrain g_aee, g_aNN. KSVZ has g_aee = 0 at tree level; DFSZ does not. Stellar-cooling bounds therefore constrain models differently.
• Ignoring the local DM velocity distribution. Haloscope signals depend on the local axion field's coherence time τ_coh = 2π / (m_a v² / 2) ∼ 10⁶ / ν oscillations. Mis-estimating v_local mis-estimates the bandwidth of the signal and the optimal integration time.
• Equating "no detection" with "ruled out". Decades of WIMP searches have taught the community to distinguish exclusion of parameter space from exclusion of a model. The QCD axion has six orders of magnitude in mass to scan; only a few decades have been touched.