Magnons: Quantized Spin Waves in Magnetic Solids

From a tilted spin to a travelling wave

The physics starts with the Heisenberg exchange Hamiltonian, ℋ = −J Σ⟨ij⟩ 𝐒ᵢ·𝐒ⱼ, where J is the exchange coupling (energy, in joules or eV) between neighbouring spins 𝐒ᵢ and 𝐒ⱼ. For a ferromagnet J > 0, so the energy is minimised when every spin points the same way — the fully aligned state |↑↑↑…⟩. That ground state is exact.

Now flip one spin. You might expect the cheapest excitation to be a single localized flip, but the exchange term does not conserve which site is flipped — the operators 𝐒ᵢ⁺𝐒ⱼ⁻ hop the deficit from site to site. Diagonalising the Hamiltonian therefore produces not a stuck flip but a delocalized wave of small tilts, in which every spin precesses on a cone about the magnetization axis, each lagging its neighbour by a fixed phase.

• Each spin dips by the same small polar angle θ, so the deviation is spread thinly over the whole lattice — much cheaper than one fully reversed spin.
• The phase advances by 𝐤·𝐚 from one site to the next (𝐚 = lattice vector), defining a wavevector 𝐤 and wavelength λ = 2π/|𝐤|.
• The whole configuration precesses coherently at angular frequency ω — this is the spin wave, first described by Felix Bloch in 1930 and refined by Holstein and Primakoff in 1940.

Crucially, one spin wave carries a total spin deviation of exactly ℏ regardless of how many sites participate. Quantize that unit of collective precession and you have one magnon.

The dispersion relation: why long-wavelength magnons are nearly free

For a simple cubic ferromagnet with spin S, nearest-neighbour coupling J and lattice constant a, the exact linear-spin-wave dispersion is

• ℏω(𝐤) = gμ_B B₀ + 2JS Σ_δ (1 − cos 𝐤·δ)

where the sum runs over nearest-neighbour vectors δ, g ≈ 2 is the g-factor, μ_B = 9.274 × 10⁻²⁴ J/T is the Bohr magneton, and B₀ is an applied field. The first term is a rigid Zeeman gap Δ = gμ_B B₀ — for B₀ = 0.1 T it is about 11.6 μeV, or 2.8 GHz. The second term is the exchange cost of bending the spin texture.

Expand for small 𝐤 (long wavelength, cos x ≈ 1 − x²/2):

• ℏω ≈ Δ + Dk², with the spin-wave stiffness D = 2JSa².

Two lessons fall straight out of this:

• The dispersion is quadratic, like a free particle E = ℏ²k²/2m*, not linear like sound. A magnon behaves like a massive boson with effective mass m* = ℏ²/2D.
• As 𝐤 → 0 the energy → Δ, which itself → 0 when B₀ → 0. These are the Goldstone modes of broken rotational symmetry: rotating all spins together costs nothing, so infinitely long spin waves are gapless. This gaplessness is exactly what makes a ferromagnet's order so fragile to heat.

In iron, D ≈ 280 meV·Å² (4.5 × 10⁻⁴⁰ J·m²); in the workhorse insulator yttrium iron garnet (YIG), D ≈ 530 meV·Å². Antiferromagnets flip the story: staggered sublattices make ω ∝ |𝐤| at small k, so AFM magnons are linear, sound-like, and reach THz frequencies with group velocities up to tens of km/s.

Bloch's T^(3/2) law: counting magnons melts the magnet

Magnons are bosons, so at temperature T their number follows the Bose–Einstein distribution n(𝐤) = 1/[exp(ℏω/k_BT) − 1]. Every magnon excited flips a spin's worth of moment, so the magnetization drops by ΔM ∝ Σ_𝐤 n(𝐤). Convert the sum to an integral over the quadratic dispersion ℏω = Dk²:

• ΔM/M(0) ∝ ∫ d³k / [exp(Dk²/k_BT) − 1]
• Substitute x = Dk²/k_BT; the k² density of states and the Bose factor give an integral that scales as (k_BT/D)^(3/2).

The result is Bloch's law: M(T) = M(0)·[1 − (T/T_c)^(3/2)]. That characteristic 3/2 power is a direct fingerprint of the quadratic magnon dispersion in three dimensions — measure the exponent and you have measured the shape of the spin-wave spectrum. It works beautifully at low T (say below ~0.5 T_c); near T_c magnon–magnon interactions and critical fluctuations take over and the simple picture breaks down.

The same counting explains why a one- or two-dimensional Heisenberg ferromagnet cannot order at any finite temperature (the Mermin–Wagner theorem): in low dimensions the integral over gapless magnons diverges, so infinitely many long-wavelength spin waves are thermally excited and destroy the order. It takes 3D — or an anisotropy gap — to keep a magnet magnetized.

Dipolar magnons, geometry, and the two speeds of a spin wave

At the very long wavelengths used in devices (λ from ~100 nm up to millimetres, k far below the exchange regime), the dominant restoring force is not exchange but the magnetic dipole–dipole interaction. Here the sample's shape and the direction of 𝐤 relative to the static magnetization matter enormously — the dispersion becomes anisotropic. Three canonical dipolar modes appear in thin films:

• Magnetostatic surface waves (Damon–Eshbach): 𝐤 perpendicular to M in the film plane; unusually, they are non-reciprocal — they travel one way on the top surface and the opposite way on the bottom.
• Backward-volume waves: 𝐤 parallel to M; their group velocity points opposite to phase velocity (negative dispersion), a genuine backward wave.
• Forward-volume waves: M perpendicular to the film plane; isotropic in-plane.

Governing all of them is the Landau–Lifshitz–Gilbert (LLG) equation, d𝐌/dt = −γ 𝐌 × 𝐁_eff + (α/M_s) 𝐌 × d𝐌/dt, where γ = gμ_B/ℏ ≈ 1.76 × 10¹¹ rad·s⁻¹·T⁻¹ is the gyromagnetic ratio and α is the dimensionless Gilbert damping. The first term is precession; the second is dissipation that steadily relaxes spins back toward 𝐁_eff — the term that sets how far a magnon can travel before it dies.

YIG, magnonics, and moving spin without moving charge

The reason magnons became technology rather than curiosity is yttrium iron garnet (Y₃Fe₅O₁₂, YIG), an electrical insulator with the lowest known Gilbert damping of any magnetic material: α ≈ 3 × 10⁻⁵. Because YIG carries no conduction electrons, a magnon propagating through it moves spin angular momentum with zero Ohmic dissipation — the resistive I²R loss that limits every charge-based interconnect simply does not exist. Magnon decay lengths in YIG reach tens of micrometres, and lifetimes hit hundreds of nanoseconds.

What that enables:

• Magnonic logic and interconnects: GHz spin waves (typically 1–20 GHz, wavelengths of hundreds of nm) are steered through etched YIG waveguides. Because they are waves, interference does the computing — a spin-wave majority gate or Mach–Zehnder logic element needs no transistors.
• Frequency filters and delay lines: YIG-based magnetostatic-wave devices have been standard microwave components since the 1970s, tunable simply by changing the bias field.
• Spin-current conversion: a magnon current in YIG injects spin into an adjacent platinum film, where the inverse spin Hall effect turns it into a measurable charge voltage — the basis of the spin Seebeck effect, where a thermal gradient across a magnetic insulator pumps magnons and generates a transverse voltage.
• Magnon Bose–Einstein condensation: pump YIG hard with microwaves at room temperature and the magnon gas can accumulate in its lowest energy state, forming a condensate — one of the very few BECs observable without cryogenics (Demokritov et al., 2006).

How we see them, and where the intuition goes wrong

Magnons are measured across roughly five orders of magnitude in energy:

• Inelastic neutron scattering is the gold standard: a neutron carries a magnetic moment and can flip a spin, losing energy ℏω and momentum ℏ𝐤 that map out the full dispersion up to ~100 meV. This is how D was measured in Fe, Ni, and YIG.
• Brillouin light scattering (BLS): photons scatter off thermal magnons, shifting by the GHz magnon frequency; micro-focused BLS images propagating spin waves in a waveguide directly.
• Ferromagnetic resonance (FMR) probes the k ≈ 0 magnon (uniform precession), and its linewidth measures the Gilbert damping α.

Common misconceptions worth clearing up:

• "A magnon is a flipped spin hopping around." No — it is a collective precession spread over the whole crystal; every spin tilts by the same tiny angle. Only the total deviation equals one ℏ.
• "Magnons carry charge or heat like electrons." They carry spin and energy, not charge. That is the whole point: spin transport without Joule heating.
• "Magnons and phonons are the same kind of thing." Both are bosonic lattice excitations, but phonons come from atomic displacements (ω ∝ k, sound-like) while ferromagnetic magnons come from spin precession (ω ∝ k², massive). Where their dispersions cross they can hybridize into magnon-polarons.
• "Magnetization drops linearly with T." It drops as T^(3/2) at low temperature — the exponent is a measured consequence of the magnon dispersion, not an arbitrary fit.

Ferromagnetic vs. antiferromagnetic magnons — the sign of the exchange coupling changes everything.

Property	Ferromagnet (FM)	Antiferromagnet (AFM)
Ground state	All spins parallel (↑↑↑↑)	Alternating sublattices (↑↓↑↓)
Long-wave dispersion	ℏω ∝ k² (quadratic)	ℏω ∝ |k| (linear, sound-like)
Group velocity as k→0	→ 0	→ finite (up to ~40 km/s)
Spin carried per magnon	−1 ℏ	±1 ℏ (two chiralities)
Typical frequency	1–25 GHz (GHz band)	0.1–10 THz (THz band)
Example material	YIG, Fe, Ni, permalloy	MnF₂, NiO, hematite