Skip to content
LumenaryAcademy
Functional 75 min

Magnetic materials

Where magnetism comes from, why iron is magnetic and copper is not, domains and hysteresis simulated, soft and hard magnets, the rare-earth supply question, and how data is stored in grains a few nanometres across.

Code tested · 6AI-draftedAwaiting review by a materials scientist

Your guides: Chloe · The Materials Scientist, Margaret · The Tester, Gideon · The Failure Analyst

MotivationH

The materials that turn the world

Almost all the electricity in the world is generated by spinning a magnet past a coil, or a coil past a magnet, and much of it is consumed the same way, by motors. Between generator and motor it passes through transformers whose cores are tonnes of iron alloy, magnetised and demagnetised fifty or sixty times a second. A wind turbine's generator may hold more than a tonne of permanent magnets; an electric car's traction motor a kilogram or two; a phone, a few grams in its speaker, its vibration motor and its camera's focusing system. And until solid-state drives took over laptops, most of the world's data sat in magnetic grains a few nanometres across; most of it in data centres still does.

Yet magnetism in solids is strange. Every atom of copper has electrons spinning and orbiting, and copper is not magnetic; iron, its neighbour in the periodic table but one, is strongly magnetic, while stainless steel, mostly iron, is often not. This lesson explains where magnetic moments come from, why in a few materials they line up spontaneously, why those materials break into domains, and how engineers make magnets that are easy to remagnetise (for transformers) or almost impossible to (for motors).

IntuitionH

Where moments come from, and what lines them up

An electron carries a magnetic moment for two reasons: its spin, an intrinsic angular momentum, and its orbital motion around the nucleus. The natural unit is the Bohr magneton, μB=eℏ/2me=9.274×10−24\mu_B = e\hbar/2m_e = 9.274 \times 10^{-24} J/T, roughly the moment of one electron spin. In a filled shell, spins pair up and orbits cancel, so atoms with only filled shells have no net moment. The interesting atoms are those with partly filled inner shells: the 3d transition metals (iron, cobalt, nickel, manganese, chromium) and the 4f rare earths (neodymium, samarium, dysprosium, gadolinium). By Hund's first rule, electrons in a partly filled shell keep their spins parallel as long as they can, so an isolated iron atom has a moment of about 4 μB\mu_B, and a neodymium or dysprosium ion a large spin and orbital moment from its 4f shell.

In a metal the picture changes: the 3d electrons form bands, and iron's moment falls to about 2.2 μB\mu_B per atom, a non-integer number that only band theory explains. Still, moments are common. What is rare is for them to line up spontaneously, and here the obvious explanation fails: the magnetic attraction between neighbouring moments is far too weak.

python
import numpy as np

mu0 = 4e-7 * np.pi
muB = 9.2740100783e-24    # Bohr magneton, J/T
kB = 1.380649e-23

# 1. Could magnetic dipoles alone line up iron's moments? Field of one moment at its neighbour's distance:
m = 2.2 * muB            # moment per iron atom in the metal
r = 0.248e-9             # nearest-neighbour distance in iron, m
B_dip = mu0 / (4 * np.pi) * 2 * m / r**3     # on the dipole's axis
print(f"dipolar field from one neighbour: {B_dip:.2f} T; energy {m * B_dip / kB:.2f} K x kB")
B_weiss = kB * 1043 / m  # field that would align the moments against heat up to iron's Curie point
print(f"field needed to order iron's moments up to 1043 K: about {B_weiss:.0f} T")
print(f"ratio: {B_weiss / B_dip:.0f} -- the ordering force is exchange, not magnetism\n")

# 2. Weiss mean-field theory for spin-1/2 moments in reduced units: m = tanh((m + h) / t),
#    with m = M / Ms, t = T / Tc and h the applied field in units of kB Tc / moment.
def solve_m(t, h=0.0):
    """Largest self-consistent magnetisation, by bisection on m - tanh((m + h)/t) = 0."""
    lo, hi = 1e-12, 1.0
    if lo - np.tanh((lo + h) / t) > 0:      # no ordered solution: only m = 0 (when h = 0)
        return 0.0
    for _ in range(100):
        mid = 0.5 * (lo + hi)
        if mid - np.tanh((mid + h) / t) < 0:
            lo = mid
        else:
            hi = mid
    return lo

print(" T/Tc    M/Ms (mean field)")
for t in [0.1, 0.3, 0.5, 0.7, 0.8, 0.9, 0.95, 0.99, 1.05]:
    mm = solve_m(t)
    print(f" {t:4.2f}     {mm:.4f}  " + "#" * int(round(40 * mm)))

# Near Tc the mean-field magnetisation vanishes as sqrt(3 (1 - T/Tc))
for t in [0.99, 0.999]:
    print(f"T/Tc = {t}: M/Ms = {solve_m(t):.4f}, sqrt(3(1 - t)) = {np.sqrt(3 * (1 - t)):.4f}")

# Above Tc a small field h induces m = chi h: compare with the Curie-Weiss law chi = 1/(t - 1)
print("\n T/Tc   chi from the model   Curie-Weiss 1/(T/Tc - 1)")
h = 1e-6
for t in [1.1, 1.5, 2.0, 3.0, 5.0]:
    print(f" {t:4.1f}       {solve_m(t, h) / h:8.3f}            {1 / (t - 1):8.3f}")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

A neighbouring iron moment produces a field of about 0.27 T, enough to hold the moments in line only below about half a kelvin. Iron stays ferromagnetic up to 1043 K, which would need an aligning field of about 700 T, more than two thousand times larger. In 1907 Pierre Weiss simply postulated such a "molecular field". Two decades later Werner Heisenberg showed where it comes from: the exchange interaction, a consequence of electrostatic repulsion combined with Pauli's principle. Two electrons with parallel spins must have an antisymmetric spatial wavefunction, which keeps them apart and lowers their repulsion; so, depending on the details of the bonding, parallel or antiparallel neighbouring spins can be favoured by energies of tens of millielectronvolts. Magnetic order is, at heart, electrostatics plus quantum mechanics.

The second part of the program solves Weiss's model, in which each moment feels a field proportional to the average magnetisation. Below the Curie temperature TCT_C a spontaneous magnetisation appears, growing from zero as 3(1−T/TC)\sqrt{3(1 - T/T_C)} and approaching saturation at low temperature. Above TCT_C the material is paramagnetic, with a susceptibility that follows the Curie–Weiss law χ=C/(T−TC)\chi = C/(T - T_C) exactly as the model reproduces.

DefinitionsH

Five kinds of magnetic order

Definition— magnetisation, susceptibility, permeability

The magnetisation M\mathbf M is the magnetic moment per unit volume, in A/m. In a field H\mathbf H (A/m), a linear material has M=χH\mathbf M = \chi\mathbf H with dimensionless volume susceptibility χ\chi, and the flux density is B=μ0(H+M)=μ0μrH\mathbf B = \mu_0(\mathbf H + \mathbf M) = \mu_0\mu_r\mathbf H, with relative permeability μr=1+χ\mu_r = 1 + \chi.

Definition— diamagnetism, paramagnetism

In a diamagnet an applied field induces moments that oppose it, so χ\chi is small and negative (about −10−5-10^{-5}). Every material has this response; it is all there is in materials without net atomic moments. In a paramagnet atoms carry moments that point at random until a field partly aligns them, so χ\chi is small and positive, typically 10−510^{-5} to 10−310^{-3}, and falls with temperature as C/TC/T (the Curie law).

Definition— ferromagnetism, antiferromagnetism, ferrimagnetism

In a ferromagnet exchange aligns neighbouring moments parallel below the Curie temperature, giving a spontaneous magnetisation. In an antiferromagnet it aligns them antiparallel on two interpenetrating sublattices, so the net magnetisation is zero; order is lost above the Néel temperature. In a ferrimagnet the sublattice moments are antiparallel but unequal, so a net magnetisation survives.

diamagnetparamagnetferromagnetantiferromagnetferrimagnet
Arrangements of atomic moments in the five kinds of magnetic behaviour, drawn as arrows on a row of atoms. A diamagnet has no permanent moments (only the small opposing moments induced by a field). A paramagnet's moments point at random. A ferromagnet's are all parallel; an antiferromagnet's alternate and cancel; a ferrimagnet's alternate but are unequal, leaving a net moment. Examples: copper, bismuth and water (diamagnetic); aluminium and oxygen (paramagnetic); iron, cobalt, nickel and Nd₂Fe₁₄B (ferromagnetic); MnO and chromium (antiferromagnetic); magnetite and the ferrites (ferrimagnetic).

Some numbers make the classes concrete. Water has χ≈−9×10−6\chi \approx -9 \times 10^{-6} and bismuth, the most diamagnetic element, about −1.7×10−4-1.7 \times 10^{-4}; aluminium is paramagnetic with χ≈+2×10−5\chi \approx +2 \times 10^{-5}. These are tiny, but not negligible: in a 16 T laboratory magnet, the diamagnetism of water was enough to levitate a living frog, in a famous 1997 demonstration at Nijmegen. Ferromagnets are in another league, with effective susceptibilities in the thousands to hundreds of thousands. Iron's Curie temperature is 1043 K, cobalt's 1388 K, nickel's 627 K, and gadolinium's only 293 K, so a piece of gadolinium is ferromagnetic in a cold room and not in a warm one.

Ferrimagnets matter more than their exotic name suggests. Magnetite, Fe₃O₄, the lodestone of ancient compasses, is one. So are the ferrites, oxides such as manganese–zinc and nickel–zinc ferrite and the hexagonal barium and strontium ferrites. Being oxides, they are electrical insulators or poor conductors, which makes them the core material of choice at high frequencies, where currents induced in a metal core would waste energy.

ComputationC

Domains and domain walls

If iron's moments line up spontaneously, why is an ordinary iron nail not a magnet? Because a ferromagnet divides itself into domains, regions each magnetised to saturation along an easy direction, but pointing different ways so that the total nearly cancels. Pierre Weiss proposed them in 1907; Francis Bitter first made them visible in 1931 by sprinkling fine magnetic particles on a polished crystal.

The reason is energy. A uniformly magnetised block produces a large field outside itself, the magnetostatic energy of its own magnetic poles. Splitting into domains, and especially adding triangular closure domains at the ends, keeps the flux inside the material and removes that energy. But each boundary between domains, a domain wall, costs energy too. Inside a wall the moments turn gradually from one direction to the other. Exchange wants the turn spread over many atoms, so that neighbours differ only slightly; magnetocrystalline anisotropy, the preference of moments for particular crystal directions, wants the turn abrupt, so that few moments point in hard directions. The balance gives a wall width δ≈πA/K\delta \approx \pi\sqrt{A/K} and an energy per unit area γ≈4AK\gamma \approx 4\sqrt{AK}, where AA is the exchange stiffness and KK the anisotropy constant.

one domaintwo domainsclosure domainsdomain wall, width δ
Domains in a small square crystal of a ferromagnet. Left: a single domain sends its flux out into the surroundings, which costs magnetostatic energy. Middle: two antiparallel domains roughly halve that energy. Right: two main domains with triangular closure domains at the ends keep the flux almost entirely inside the crystal, at the cost of four 90° walls. Inset: inside a 180° wall the moments rotate gradually over a width δ.

The program computes the wall widths and energies for iron and two permanent-magnet compounds, along with two further quantities. Below a critical size, a particle cannot afford a wall at all and stays a single domain. And the anisotropy field HK=2K/(μ0Ms)H_K = 2K/(\mu_0M_s) is the field needed to rotate the magnetisation of a perfect single-domain particle against its anisotropy: an upper limit on coercivity. The last column is the upper limit on the energy product of a permanent magnet, Js2/4μ0J_s^2/4\mu_0, which we will meet below.

python
import numpy as np

mu0 = 4e-7 * np.pi

# Exchange stiffness A (J/m), first anisotropy constant K1 (J/m^3) and saturation magnetisation Ms (A/m),
# rounded room-temperature literature values
mats = {"iron": (2.1e-11, 4.8e4, 1.71e6),
        "Nd2Fe14B": (7.7e-12, 4.9e6, 1.28e6), "SmCo5": (1.2e-11, 1.7e7, 0.86e6)}

print("material    wall width/nm  wall energy/(mJ/m^2)  single-domain size/nm  mu0*H_K/T  (BH)max limit/(kJ/m^3)")
for name, (A, K, Ms) in mats.items():
    K = abs(K)
    width = np.pi * np.sqrt(A / K)                 # a 180-degree wall: exchange (wants it wide) vs anisotropy (narrow)
    gamma = 4 * np.sqrt(A * K)                     # its energy per unit area
    d_sd = 72 * np.sqrt(A * K) / (mu0 * Ms**2)     # sphere below which one domain beats two (uniaxial estimate)
    HK = 2 * K / (mu0 * Ms)                        # anisotropy field: the coercivity of a perfect, uniform particle
    bh = mu0 * Ms**2 / 4                           # (BH)max can never exceed Js^2 / (4 mu0)
    print(f"{name:10} {width * 1e9:9.1f}      {gamma * 1e3:9.2f}          {d_sd * 1e9:10.0f}        {mu0 * HK:7.2f}"
          f"        {bh / 1e3:7.0f}")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

Iron's walls are about 66 nm wide, a few hundred atoms; those of the neodymium magnet compound Nd₂Fe₁₄B, whose anisotropy is a hundred times stronger, are only 4 nm. A sphere of Nd₂Fe₁₄B smaller than about 200 nm should be a single domain, and SmCo₅'s limit is about a micrometre. The anisotropy fields tell the same story: rotating iron's magnetisation against its anisotropy takes only 0.06 T, which is why iron is magnetically soft, while Nd₂Fe₁₄B needs 7.7 T and SmCo₅ about 40 T. Iron's theoretical energy product limit is the highest of the three, but it is useless as a permanent magnet: with so little anisotropy, nothing holds its magnetisation in place.

How a ferromagnet magnetises. Apply a field to a demagnetised piece of soft iron and the domains already pointing roughly along the field grow, by their walls moving, at the expense of the others. In a perfect crystal the walls would glide freely; in a real one they are snagged by inclusions, grain boundaries, dislocations and precipitates, and jump from one pinning site to the next. (Those jumps make tiny voltage pulses in a coil around the sample, heard through a loudspeaker as a crackle: the Barkhausen effect, discovered in 1919.) At higher fields the last domains rotate their magnetisation towards the field until the sample is saturated. Remove the field, and the walls do not all return: the material keeps a remanent magnetisation. That irreversibility is hysteresis.

ComputationC

Hysteresis loops, simulated

The simplest model of hysteresis has no walls at all. Edmund Stoner and Erich Wohlfarth, at Leeds in 1948, considered a particle small enough to be a single domain, with one easy axis. Its magnetisation can only rotate as a whole, and its energy, relative to its anisotropy energy KVKV, is

e(θ)=sin⁡2(θ−ψ)−2hcos⁡θ,e(\theta) = \sin^2(\theta - \psi) - 2h\cos\theta,

where θ\theta is the angle between the magnetisation and the applied field, ψ\psi the angle between the easy axis and the field, and h=H/HKh = H/H_K the reduced field. For small fields there are two minima, magnetisation roughly along or against the field; the particle sits in whichever it was in. As the field is reversed, the occupied minimum becomes shallower and finally disappears, and the magnetisation jumps irreversibly to the other. The program sweeps the field down and up for single particles at several angles, and for an assembly of particles with random easy axes, as in a powder or an isotropic sintered magnet.

python
import numpy as np

# Stoner-Wohlfarth: a single-domain particle with a uniaxial easy axis at angle psi to the applied field.
# Reduced energy e(theta) = sin^2(theta - psi) - 2 h cos(theta), theta = direction of M from the field,
# h = H / H_K. The moment sits in a local minimum and stays there until that minimum disappears.
def sweep(psi, h_values, theta0):
    theta = np.array(theta0, dtype=float)
    m = []
    for h in h_values:
        for _ in range(15):                        # relax into the nearest local minimum
            g = np.sin(2 * (theta - psi)) + 2 * h * np.sin(theta)        # de/dtheta (halved)
            gp = 2 * np.cos(2 * (theta - psi)) + 2 * h * np.cos(theta)   # its derivative
            newton = np.clip(g / np.maximum(gp, 1e-3), -0.05, 0.05)       # small Newton steps inside a well
            theta -= np.where(gp > 1e-3, newton, 0.3 * np.sign(g))        # outside a well, roll downhill
        m.append(np.cos(theta))                    # component of M along the field
    return np.array(m)

h_down = np.linspace(2, -2, 801)
h_up = np.linspace(-2, 2, 801)

def loop(psi, weights):
    m_down = sweep(psi, h_down, np.zeros_like(psi)) @ weights        # start saturated along +h
    m_up = sweep(psi, h_up, np.full_like(psi, np.pi)) @ weights       # start saturated along -h
    return m_down, m_up

def remanence_coercivity(m_down, m_up):
    Mr = m_down[np.argmin(abs(h_down))]                                # M on the descending branch at h = 0
    i = np.flatnonzero(m_down <= 0)[0]                                 # first field where M has reversed
    Hc = h_down[i - 1] - m_down[i - 1] * (h_down[i] - h_down[i - 1]) / (m_down[i] - m_down[i - 1])
    return Mr, -Hc                                                     # linear interpolation to M = 0

for deg in [0, 30, 45, 60]:
    psi = np.array([np.radians(deg)]) + 1e-6      # a tiny offset avoids balancing exactly on a maximum
    m_down, m_up = loop(psi, np.ones(1))
    Mr, Hc = remanence_coercivity(m_down, m_up)
    print(f"easy axis at {deg:2} deg to the field: Mr/Ms = {Mr:.2f}, Hc/H_K = {Hc:.2f}")

# Randomly oriented particles in 3D: cos(psi) uniformly distributed
u = (np.arange(400) + 0.5) / 400
psi = np.arccos(u)
w = np.full(psi.size, 1 / psi.size)
m_down, m_up = loop(psi, w)
Mr, Hc = remanence_coercivity(m_down, m_up)
print(f"\nrandom orientations: Mr/Ms = {Mr:.2f}, Hc/H_K = {Hc:.2f}")

# ASCII loop of the random assembly
rows, cols = 15, 41
grid = [[" "] * cols for _ in range(rows)]
for hs, ms in [(h_down, m_down), (h_up, m_up)]:
    for h, mm in zip(hs, ms):
        c = int(round((h + 2) / 4 * (cols - 1)))
        r = int(round((1 - mm) / 2 * (rows - 1)))
        grid[r][c] = "*"
for r in range(rows):
    if grid[r][cols // 2] == " ":
        grid[r][cols // 2] = "|"
print("\n".join("".join(row) for row in grid))
print("h = -2" + " " * 29 + "h = +2")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

With the easy axis along the field, the loop is a perfect square: the magnetisation stays fully along +h+h until the reversed field reaches exactly HKH_K, then flips. At 30° and 45° the particle switches at about 0.52HK0.52H_K and 0.50HK0.50H_K, and at 60° the magnetisation rotates through zero (at 0.43HK0.43H_K) before it jumps. Averaged over random orientations, the remanence is exactly half the saturation and the coercivity about 0.48HK0.48H_K, reproducing the two famous numbers of Stoner and Wohlfarth's paper, 0.5 and 0.479. Aligning the easy axes of the grains, which manufacturers of sintered magnets do by pressing the powder in a magnetic field, is worth up to a factor of two in remanence.

Mr = 0.5 Ms−Hc ≈ −0.48 H_K+Hcpermanent magnetswork here−2−1+1+2 = H/H_K+1−1M/Ms
The hysteresis loop of randomly oriented Stoner–Wohlfarth particles, computed above: magnetisation (as a fraction of saturation) against applied field (in units of the anisotropy field). The loop crosses the vertical axis at the remanence, ±0.5, and the horizontal axis at the coercive field, about ±0.48. For a magnet, the useful part is the second quadrant, where the magnetisation stays positive against a reversed field.
ExplanationH

Soft and hard magnets

The loop sorts magnetic materials into two families with opposite goals.

Soft magnets should magnetise and demagnetise easily, with a narrow loop (coercivity below about 1000 A/m, sometimes below 1 A/m), high permeability and high saturation. Every cycle around the loop dissipates its area as heat, and a transformer core goes round fifty or sixty times a second for decades. The engineering rule for softness is: low anisotropy, low magnetostriction, and nothing to pin the walls, so few inclusions, large grains, and no internal stress. A second loss is just as important. A changing flux induces eddy currents in a conducting core, so cores are made of thin insulated sheets (laminations), and the loss falls with the square of their thickness.

python
import numpy as np

# 1. Eddy-current loss in a laminated core. For sinusoidal flux density of peak B in sheets of thickness d,
#    classical theory gives a loss per volume P = pi^2 d^2 f^2 B^2 / (6 rho).
rho, density = 4.7e-7, 7650.0       # resistivity (ohm m) and density (kg/m^3) of iron with about 3 % silicon
B = 1.5                             # T
print("sheet thickness   eddy loss at 50 Hz   at 400 Hz   (W/kg, classical estimate)")
for d in [2.0e-3, 0.5e-3, 0.35e-3, 0.2e-3, 0.1e-3]:
    loss = [np.pi**2 * d**2 * f**2 * B**2 / (6 * rho) / density for f in (50, 400)]
    print(f"   {d * 1e3:4.2f} mm           {loss[0]:8.3f}           {loss[1]:8.2f}")

# 2. Magnetic recording: a grain keeps its bit only if its energy barrier K V is large compared with kB T.
#    Neel-Arrhenius: mean time before a thermal flip, tau = tau0 exp(K V / kB T), with tau0 about 1 ns.
kB, T, tau0 = 1.380649e-23, 330.0, 1e-9      # a warm drive, 330 K
year = 3.156e7
print("\nK V / kB T     mean time before a flip")
for ratio in [20, 30, 40, 50, 60]:
    t = tau0 * np.exp(ratio)
    print(f"    {ratio:3}         {t:10.2e} s = {t / year:10.2e} years")

# A bit is spread over many grains, and every grain must survive, so drives aim for K V / kB T of about 60.
# Smallest grain (a column 10 nm tall) that meets that, for two media materials:
for name, K, Ms in [("CoCrPt-oxide (conventional media)", 0.4e6, 0.5e6), ("FePt, L1_0 ordered (HAMR media)", 6.6e6, 1.1e6)]:
    V = 60 * kB * T / K
    diameter = np.sqrt(4 * V / (np.pi * 10e-9))
    print(f"{name:34}: V >= {V * 1e27:6.0f} nm^3, diameter >= {diameter * 1e9:4.1f} nm, "
          f"write field needed ~ 2K/Ms = {2 * K / Ms:4.1f} T")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

A solid core 2 mm thick would lose 10 W/kg to eddy currents alone at 50 Hz; laminations 0.35 mm thick cut that to 0.3 W/kg. At 400 Hz, typical of a fast-spinning electric-vehicle motor or an aircraft generator, the loss is 64 times larger, which is why such machines use sheets of 0.2 mm or thinner. (Measured losses are higher than this classical estimate, because domain walls moving through the sheet concentrate the eddy currents; manufacturers quote total losses measured by standard methods.) The main soft materials are:

  • Electrical steel: iron with about 3 per cent silicon, which raises resistivity fourfold and reduces anisotropy and magnetostriction. Grain-oriented steel, invented by Norman Goss in the 1930s, is rolled and annealed so that almost every grain has an easy ⟨100⟩ axis along the rolling direction: the material of large transformers. Non-oriented grades are used in rotating machines, where the flux turns. Electrical steel is the largest-tonnage magnetic material by far.
  • Amorphous and nanocrystalline ribbons: metallic glasses of iron, boron and silicon, quenched from the melt into ribbons about 25 µm thick, have no grains to pin walls and very low losses, and are used in efficient distribution transformers. Annealing a related alloy with small additions of copper and niobium gives 10–20 nm iron–silicon grains in an amorphous matrix ("Finemet", developed at Hitachi Metals in 1988), whose random grain anisotropies average out to almost nothing.
  • Soft ferrites (manganese–zinc and nickel–zinc): ceramic, insulating, used from kilohertz to hundreds of megahertz in switch-mode power supplies, chargers and interference filters, where metal cores would be eaten by eddy currents (see Lumenary EE's transformers and machines and power electronics).
  • Nickel–iron alloys (Permalloy, around 80 per cent nickel): extraordinarily high permeability, for sensors and magnetic shielding.

Hard (permanent) magnets are the opposite: they should keep their magnetisation against reversing fields and temperature, with high remanence BrB_r, high coercivity, and a large maximum energy product (BH)max⁡(BH)_{\max}, the largest value of ∣BH∣|BH| in the second quadrant of the BB–HH loop, which measures how much field a given volume of magnet can supply to a gap. Its upper limit, if the loop were perfectly square, is Js2/4μ0J_s^2/4\mu_0, where Js=μ0MsJ_s = \mu_0M_s; the program above gave 515 kJ/m³ for Nd₂Fe₁₄B.

Magnet familyTypical (BH)max⁡(BH)_{\max} (kJ/m³)Curie temperatureNotes
Hard ferrite (Sr or Ba hexaferrite)25–40about 450 °Ccheap, oxide, no critical elements; loudspeakers, small motors
Alnico (Al–Ni–Co–Fe)10–80about 800 °Cstable at high temperature; easily demagnetised
Samarium–cobalt (SmCo₅, Sm₂Co₁₇)120–260700–850 °Chigh temperature, corrosion resistant; aerospace
Neodymium–iron–boron (Nd₂Fe₁₄B)200–440about 310 °Cstrongest at room temperature; motors, generators, drives

Nd–Fe–B magnets were found in 1982–84, independently, by Masato Sagawa's group at Sumitomo Special Metals in Japan (by powder sintering) and by John Croat's group at General Motors in the United States (by rapid solidification). They made compact, powerful motors possible, and today an electric car's motor or a large direct-drive wind turbine's generator typically uses them. Their weakness is temperature: the coercivity falls as they warm, and their Curie temperature is only about 310 °C. To keep coercivity at the 150–200 °C inside a motor, makers add the heavy rare earths dysprosium or terbium, increasingly placed only at grain boundaries by diffusion, where reversal would nucleate, to use as little as possible.

ExplanationH

Rare-earth magnets and supply

Despite their name, rare-earth elements are not especially rare: cerium is about as abundant in the Earth's crust as copper, and neodymium more abundant than lead. What is rare is ore rich enough to mine economically, and above all the capacity to separate the elements, which are chemically almost identical. Separation uses solvent extraction through hundreds of mixer–settler stages, and produces chemical wastes and, in many ores, radioactive thorium-bearing residues that must be managed carefully. These are the reasons production became concentrated.

According to the International Energy Agency's 2025 analysis, the three largest producing countries accounted for about 86 per cent of rare-earth mining and 97 per cent of refining in 2024, with China by far the largest in both; China also makes the large majority of sintered Nd–Fe–B magnets. That concentration has repeatedly become visible. A 2010 diplomatic dispute was followed by restricted exports and a price spike, after which Japan in particular diversified its supply, including investment in the Australian producer Lynas, which separates rare earths in Malaysia. In April 2025 China introduced export licensing for several medium and heavy rare earths, including dysprosium and terbium, and for magnets containing them; shortages briefly halted some vehicle production lines in other countries, and the rules have been adjusted more than once since. As of 2026, new mining, separation and magnet capacity is being built in the United States, Australia, Europe, Japan and elsewhere, though it will take years to change the overall picture.

Materials science offers several responses, all of them active work as of 2026:

  • Use less of the heavy rare earths, through grain-boundary diffusion and finer grains, or none, in motors designed to run cooler.
  • Recycle: magnets from hard drives, motors and wind turbines can be recovered, and processes that break sintered magnets apart with hydrogen and remake them directly are being scaled up; at present only a small fraction of end-of-life magnets is recycled.
  • Avoid permanent magnets: induction motors and wound-rotor synchronous motors, used by several car makers, need no magnets at all, at some cost in efficiency or size; ferrite-magnet motors trade size for supply security.
  • New magnet compounds without rare earths, such as iron nitride (Fe₁₆N₂), manganese-based compounds and tetrataenite (ordered FeNi, found in meteorites), are being researched; none yet matches Nd–Fe–B. Materials discovery for this problem is one of the case studies of AI-assisted materials discovery, and the wider question of critical materials returns in Sustainable materials.
ExplanationH

Magnetic storage

A hard disk stores each bit in a patch of a thin film made of magnetic grains, magnetised up or down (since about 2005, perpendicular to the disc). To store more bits per square centimetre the patches must shrink, and each must still contain enough grains for a clean signal, so the grains must shrink too. But a grain's magnetisation is held in place by an energy barrier KVKV, and if that barrier is small compared with kBTk_BT, heat flips it at random: the grain becomes superparamagnetic and forgets. The last program shows how steeply the retention time depends on the ratio, following the Néel–Arrhenius law τ=τ0eKV/kBT\tau = \tau_0e^{KV/k_BT}: at KV/kBT=40KV/k_BT = 40 a grain lasts about seven years on average, at 60 billions of years. Because a drive has billions of grains and every one must survive, designs aim for about 60.

This sets the recording trilemma. Smaller grains need higher anisotropy KK to stay stable; higher KK needs a larger field to write; and write heads are limited by the best soft magnetic alloy available, iron–cobalt, which saturates at about 2.4 T. The computation shows the way out. Conventional cobalt–chromium–platinum media with K≈0.4K \approx 0.4 MJ/m³ cannot go much below 9 nm grains. Chemically ordered iron–platinum, with sixteen times the anisotropy, could be stable at grains near 2–3 nm, but would need a 12 T field to write. Heat-assisted magnetic recording (HAMR) resolves this by heating each spot with a laser, through a tiny plasmonic antenna, to near FePt's Curie temperature for about a nanosecond, writing it while its anisotropy is low, and letting it cool. After two decades of development, HAMR drives began shipping in volume in 2024, alongside microwave-assisted and other approaches from different manufacturers.

Reading uses a different effect. In 1988 Albert Fert's group in Orsay and Peter Grünberg's in Jülich found that layers of iron separated by chromium change their resistance by tens of per cent when their magnetisations are switched between parallel and antiparallel: giant magnetoresistance (GMR), recognised with the 2007 Nobel Prize in Physics. GMR read heads entered drives in 1997, and tunnelling magnetoresistance through crystalline magnesium oxide barriers, much larger again, replaced them from the mid-2000s. The same magnetic tunnel junctions are the bits of MRAM, non-volatile memory now offered as embedded memory by several chip foundries. Magnetic tape, using barium ferrite particles, remains the cheapest medium for archives, and holds much of the world's long-term data.

Common misconceptionsH

Common misconceptions

Misconception

"Iron is magnetic because its atoms attract each other magnetically." The magnetic interaction between neighbouring moments is thousands of times too weak to explain iron's Curie temperature. The ordering comes from exchange, an electrostatic and quantum-mechanical effect.

Misconception

"All steels are magnetic." Common austenitic stainless steels (with nickel, such as the familiar "18/8" grades) have a face-centred cubic structure that is not ferromagnetic, though cold working can make them slightly magnetic by forming martensite. Ferritic and martensitic steels are ferromagnetic.

Misconception

"An unmagnetised piece of iron has no magnetic order." Every domain inside it is magnetised to saturation; the domains point in different directions and cancel. Magnetising it rearranges domains rather than creating order.

Misconception

"Rare-earth elements are rare." Several are as abundant in the crust as common metals. The bottlenecks are concentrated deposits, separation capacity and magnet manufacturing.

Misconception

"A higher coercivity is always better." For a permanent magnet, yes, up to a point; for a transformer core it would be a disaster, because every cycle around a wide loop wastes energy. The right loop depends on the job.

Exercises

Exercises

1. Magnetic materials

Warm-upHuman

Which of these have a spontaneous net magnetisation (within each domain) at room temperature? Select all that apply.

Choose all that apply

2. Magnetic materials

CoreHuman

A magnetic material has exchange stiffness A=1.0×10−11A = 1.0\times10^{-11} J/m and uniaxial anisotropy constant K=1.0×106K = 1.0\times10^{6} J/m³. Estimate the width of a 180° domain wall, δ=πA/K\delta = \pi\sqrt{A/K}, in nanometres (nearest whole number).

3. Magnetic materials

CoreHuman

With the Néel–Arrhenius law τ=τ0eKV/kBT\tau = \tau_0e^{KV/k_BT} and τ0=1\tau_0 = 1 ns, what is the smallest whole-number value of KV/kBTKV/k_BT for which a single grain's mean time before a thermal flip is at least 10 years? (Take a year as 3.156×1073.156\times10^7 s.)

4. Magnetic materials

CoreHuman

Why are transformer and motor cores built from thin, insulated sheets of iron–silicon alloy rather than solid iron?

Choose one

5. Magnetic materials

CoreComputational

In Weiss's mean-field model for spin-½ moments, the reduced magnetisation m=M/Msm = M/M_s at reduced temperature t=T/TCt = T/T_C (with no applied field) satisfies

m=tanh⁡(m/t).m = \tanh(m/t).

Write magnetisation(t) returning the positive solution for 0<t<10 < t < 1 and 0.0 for t≥1t \ge 1. Use any root-finding method (bisection is safest), accurate to 10−810^{-8}.

Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

6. Magnetic materials

StretchComputational

A Stoner–Wohlfarth particle has reduced energy e(θ)=sin⁡2(θ−ψ)−2hcos⁡θe(\theta) = \sin^2(\theta - \psi) - 2h\cos\theta. Starting magnetised along +h+h, a reversed field h=−hswh = -h_{\text{sw}} makes its occupied minimum disappear and the magnetisation jump.

Write switching_field(psi_deg) returning hswh_{\text{sw}} (a positive number, in units of HKH_K) for an easy axis at psi_deg degrees to the field, 0≤ψ≤90°0 \le \psi \le 90°. You may derive a formula or find it numerically; the tests need three-decimal accuracy.

Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

7. Magnetic materials

Warm-upHuman

Name the effect discovered in 1988 by Albert Fert's and Peter Grünberg's groups, in which the resistance of magnetic multilayers changes strongly when their layers' magnetisations switch between parallel and antiparallel.

Further reading and resources