Ferromagnetism

Ferromagnetism is the load-bearing wall of the electric age. Every transformer, generator, and motor core is soft iron alloy doing what this article priced: concentrating flux at $\mu_0M_s \approx 2$ T while forgetting its history 50–60 times a second with the narrowest achievable loop. At the hard end, Nd$_2$Fe$_{14}$B — coercivity a million A/m, single-domain grains pinned beyond persuasion — put permanent-magnet motors in electric vehicles and drones; its rare-earth supply chain is a standing geopolitical issue that is, at root, a hysteresis-loop specification. The memory reading became civilization's archive. Hard-disk platters store bits as the remanent magnetization of nanoscale grain clusters, coercivity tuned so that only the write head — not neighbors, not time — can flip them; heat-assisted recording (HAMR) briefly warms each bit toward its Curie point to write it cheaply, then cools it back to stubbornness: the $T_C$ phenomenon, weaponized per bit. Barkhausen noise is now a nondestructive-testing tool reading residual stress in rails, pipelines, and turbine shafts — listening to the domain walls stick. And the family name-drops at the article's edge grew into fields: Néel's antiferromagnets (neighbors locked antiparallel, net $M = 0$, invisible to a compass) are the quiet workhorses of spintronic memory, and ferrimagnets — including magnetite, the original lodestone, so the first magnet humanity met was technically not a ferromagnet — bridge the two.

Picture the compass-needle crowd of Paramagnetism once more, but grant the needles one new power: each pulls its immediate neighbors into line with itself. If that mutual pull beats the thermal shaking, the crowd no longer needs an external field — one needle aligns the next, and order propagates wall to wall. That locked collective state is spontaneous magnetization, and it changes every rule of the previous two articles: the response to $H$ is no longer linear, no longer small, and — because reorganizing a locked crowd leaves scars — no longer reversible. The pull between neighbors must be enormous by magnetic standards, and here the chapter owes its most important honesty. The obvious candidate — each moment sitting in the dipole field of its neighbor — fails by four orders of magnitude, as the first derivation computes. What actually locks the spins is the exchange interaction: the Pauli exclusion principle acting on electrostatic Coulomb energy. Electrons with parallel spins are forbidden from occupying the same place; forced apart, they repel each other less; in iron, cobalt, and nickel the geometry of the $3d$ orbitals makes that Coulomb saving largest when neighboring spins align. The energy at stake is electrostatic — tenths of electron-volts, thousands of kelvin — which is why iron's order survives to 1043 K when honest dipole magnetism would surrender below a tenth of a kelvin. How the Pauli-plus-Coulomb mechanism produces an effective spin–spin coupling is quantum mechanics proper, owned by Exchange Interaction and Ferromagnetic Ordering at the chapter's end; this article uses the interaction's existence and its measured scale. And the Bohr–van Leeuwen theorem, prosecuted in Diamagnetism, already warned us: no classical mechanism could have done this — spin, the Pauli principle, and exchange are all quantum hardware, and ferromagnetism is the most theatrical quantum effect visible on a refrigerator door. And the door brings up the name. Ferrum is Latin for iron — this subject, the magnetism of iron, is the etymology's home. When early twentieth-century physicists found crystals whose electric polarization showed the same spontaneous order, the same domains, the same looping memory, they named them ferroelectrics by pure analogy, iron-free though they are. Ferroelectrics and Spontaneous Polarization confessed the borrowed name and promised the original story; the loops, walls, and Curie points below are that story — the template the electric twin was traced from.

Why, then, isn't every iron nail born a permanent magnet? Because spontaneous magnetization has an external bill to pay. A uniformly magnetized block pours field into the surrounding space — think of the bar magnet's $\mathbf{B}$ portrait — and field costs energy, of order $\mu_0M_s^2V/6$ for a compact body: about 0.6 joules for a single cubic centimetre of iron, an enormous sum by atomic standards. The crystal cheats. It splits into domains — regions, microns across, each locked at full $M_s$ by exchange, but pointing in different crystallographically easy directions so their external fields cancel. A virgin nail is not disordered; it is a parliament of perfectly ordered constituencies that vote in all directions. Magnetizing the nail is redistricting, not persuasion atom by atom: domains aligned with the field grow at their neighbors' expense as the domain walls between them sweep sideways, each wall a thin ribbon in which the moments rotate from one domain's direction to the next. The wall's internal anatomy — width, energy per area, why it is not atomically sharp — is owned by Magnetic Domains and Domain Walls; here we need only its existence, its price tag, and its mobility. Because the walls move through a real crystal — grain boundaries, dislocations, inclusions — their motion sticks and jumps, and that friction writes the hysteresis loop we meet below. The macroscopic bookkeeping runs exactly as the chapter built it: $\mathbf{B} = \mu_0(\mathbf{H} + \mathbf{M})$, with $\mathbf{H}$ the free-current bookkeeper of The Auxiliary Field H and $M_s(\mathrm{iron}) = 1.7\times10^{6}$ A/m the saturation value owned by Magnetization — worth $\mu_0M_s = 2.1$ T of flux density from the material alone, which is why iron transforms every electromagnet it touches. What ferromagnetism breaks is the constitutive step: no single $\chi_m$ or $\mu$ exists — $\mathbf{M}$ depends on where you are on the loop, not just on $\mathbf{H}$ — a failure of linearity managed, by name, in Magnetic Permeability and Linear and Nonlinear Magnetic Media.

Now put the domains to work and watch the memory appear. Start from virgin (demagnetized) iron and raise $H$ slowly. The response unfolds in three acts. First, at small $H$, walls bow — favorable domains grow slightly, reversibly, like a stretching membrane; remove the field and they relax back. The initial slope is already spectacular ($\chi_{\rm initial}$ of $10^3$–$10^4$; reading such slopes as "permeability" is the business of Magnetic Permeability). Second, at moderate $H$, the walls tear free of the defects that pin them — grain boundaries, inclusions, strain fields — and jump discontinuously from one pinning configuration to the next. Each jump flips a micron-scale packet of moments at once and is irreversible: the energy released does not return when $H$ is lowered; it degrades to sound and heat. This is the steep, history-writing segment of the virgin curve. Third, near saturation, the walls are gone and only coherent rotation remains: the last moments, held along easy crystal axes by anisotropy, are turned into the field direction. Beyond that, $M = M_s$ and the material has nothing more to give. The wall jumps are literally audible. Wrap a coil around the iron, feed it to an amplifier, and a smooth ramp of $H$ produces a rushing crackle in the speaker — the Barkhausen effect (1919): each crackle a wall breaking free of a pin, an irreversible packet of flux change, and collectively the first direct evidence that magnetization proceeds by domain avalanches rather than smooth atomic alignment.

where: $W_{\rm hyst}$ — the energy per unit volume converted to heat in one full cycle; the integral runs once around the closed $B$–$H$ trajectory and equals the enclosed area. This is the loop's thermodynamic verdict, carried here at statement-and-reading level: on the way up you do work $H\,\mathrm{d}B$ against the material, on the way down it returns only part, and the shortfall — the area — is what the irreversible wall jumps dissipated. A transformer core cycled at 50 Hz pays this bill fifty times a second: silicon steel with a loop area of $\sim$40 J/m³ dissipates 2 kW per cubic metre — a few watts of standing heat in a 10 kg core, which is why loop area is a headline specification of core steel. The derivation of equation (1) from magnetic work, its split against eddy-current losses, and the engineering of narrow loops are owned by Hysteresis; the loop's wall-by-wall anatomy by Magnetic Domains and Domain Walls.

Trace the full loop and the memory vocabulary falls out. Drive to saturation and bring $H$ back to zero: the walls do not retrace their jumps — pinning holds much of the redistricted territory — and a remanence $M_r$ survives at $H = 0$. The iron is now a permanent magnet; nothing but its own history distinguishes it from the virgin bar. To erase it, apply a reverse field: $M$ reaches zero only at the coercivity $-H_c$, the field that un-pins enough walls to re-balance the domains. Push on to reverse saturation, swing back, and the trajectory closes into the loop: two branches, two possible magnetizations at every intermediate $H$, the choice recording which extreme you visited last. A ferromagnet is a physical bit — the loop-as-memory reading — and everything from tape archives to credit-card stripes is engineering built on choosing $H_c$ large enough that stray fields cannot rewrite the record. The choice of $H_c$ splits the materials catalogue in two:

Six orders of magnitude separate the softest keeper from the hardest magnet — same $M_s$, same exchange, the entire difference being how strongly the microstructure pins the walls. Soft materials are engineered pure (few pins: low loss, no memory — a transformer core must forget sixty times a second); hard materials are engineered obstructed, up to single-domain grains that cannot host a wall at all and must reverse by full rotation, the most expensive route. Fridge magnet and transformer lamination differ not in their atoms' opinions but in their defects' stubbornness. One more phenomenon completes the portrait: heat the magnet. At 770 °C (1043 K) — a dull red heat — iron's spontaneous magnetization vanishes. Not gradually into weakness: it collapses at a sharp critical temperature, the Curie point, above which exchange loses its race against thermal agitation and iron becomes an ordinary paramagnet, its susceptibility obeying a modified Curie law $\chi = C/(T - T_C)$ that diverges as cooling approaches the transition. Blacksmiths have always known the phenomenon: hot iron does not hold to the magnet. The quantitative story — the Curie–Weiss law, the mean-field theory that predicts $T_C$, the critical behavior at the transition — is deferred whole to Curie Law and Curie Temperature; this article's business is the phenomenon and its scale.

1 · Four orders of failure — the dipole bid for iron (confidence builder)

Evaluate the case for magnetic dipole–dipole coupling as the source of iron's order: compute the coupling for single-magneton moments at $a = 0.25$ nm, convert to an ordering-temperature scale, and compare with $T_C = 1043$ K; then extract the energy scale exchange must supply.

Step 1. Dipole coupling at lattice distance: $U_{dd} \sim \dfrac{\mu_0\mu_B^2}{4\pi a^3} = \dfrac{(1.257\times10^{-6})(9.274\times10^{-24})^2}{4\pi(2.5\times10^{-10})^3} = 5.5\times10^{-25}$ J.

Step 2. As an ordering temperature: $U_{dd}/k_B = \dfrac{5.5\times10^{-25}}{1.381\times10^{-23}} = 0.040$ K. With iron's measured atomic moment of $2.2\,\mu_B$ the estimate rises by $(2.2)^2 \approx 4.8$ to 0.19 K — still hopeless.

Step 3. The verdict ratio: $\dfrac{T_C}{U_{dd}/k_B} = \dfrac{1043}{0.040} \approx 2.6\times10^4$ — four orders of magnitude. A mechanism short by a factor of twenty-six thousand is not a candidate needing refinement; it is the wrong mechanism.

Step 4. What the real mechanism must pay: $k_BT_C = (1.381\times10^{-23})(1043) = 1.44\times10^{-20}$ J $= 0.09$ eV per atom — chemical-bond currency, which reorganized Coulomb energy (exchange) has to spend.

2 · Redistricting a cube of iron — the domain economy (exam level)

A 1.0 cm cube of iron ($M_s = 1.7\times10^6$ A/m, wall energy $\sigma_w = 1.5\ \mathrm{mJ/m^2}$) may stay uniformly magnetized or split into $N$ antiparallel slab domains. Find the uniform-state field energy, the optimal $N$, the domain width, and the energy saved.

Step 1. The uniform bill: $U_0 \approx \dfrac{\mu_0M_s^2}{6}V = \dfrac{(1.257\times10^{-6})(1.7\times10^6)^2}{6}(10^{-6}) = 0.61$ J. Per atom that is $\sim 4\times10^{-4}$ K — trivial against exchange, which is why exchange tolerates the splitting that removes it: walls cost exchange energy only inside their thin ribbons.

Step 2. The split budget: $U(N) \approx U_0/N + N\sigma_wL^2$ with $L = 10^{-2}$ m, so the wall term is $N\times(1.5\times10^{-3})(10^{-4}) = 1.5\times10^{-7}N$ J.

Step 3. Minimize: $N^* = \sqrt{\dfrac{U_0}{\sigma_wL^2}} = \sqrt{\dfrac{0.61}{1.5\times10^{-7}}} \approx 2.0\times10^3$ walls, giving domain width $d = L/N^* \approx$ 5 μm — squarely the scale Kerr microscopy reports in centimetre iron crystals.

Step 4. The saving: $U(N^*) = 2\sqrt{U_0\sigma_wL^2} = 2\sqrt{0.61\times1.5\times10^{-7}} = 6.1\times10^{-4}$ J — the 0.61 J field bill cut a thousandfold, at a wall cost the crystal barely notices. Domains are not a failure of ferromagnetic order; they are its cheapest possible packaging.

References. Griffiths, Introduction to Electrodynamics, 4th ed., §6.1.4 (Ferromagnetism) and §6.4 · Jackson, Classical Electrodynamics, 3rd ed., §5.8 (Ferromagnetism) · Purcell & Morin, Electricity and Magnetism, 3rd ed., Ch. 11 · Zangwill, Modern Electrodynamics, §13.5 · Kittel, Introduction to Solid State Physics, 8th ed., Ch. 12 · Ashcroft & Mermin, Solid State Physics, Ch. 33

Provenance. Equations and numbers cross-checked against Griffiths §6.4 (hysteresis, nonlinear media), Jackson §5.8, Purcell Ch. 11, and Kittel Ch. 12 / Ashcroft–Mermin Ch. 33 (Weiss theory, Curie–Weiss law). A verifier should: (1) reproduce the dipolar vs exchange energy estimate ($U_{dd}/k_B\sim1\,$K vs $T_C^{\mathrm{Fe}}=1043\,$K); (2) re-derive $T_C$ from the Brillouin small-$x$ expansion ($\mathcal{B}_J\approx(J+1)x/3J$); (3) confirm the Curie–Weiss susceptibility pole; (4) verify the hysteresis loss equals the loop area $\oint H\,\mathrm{d}B$. Standard constants: $\mu_0=4\pi\times10^{-7}\,\mathrm{T\,m/A}$, $\mu_B=9.274\times10^{-24}\,\mathrm{J/T}$, $M_s^{\mathrm{Fe}}\approx1.7\times10^{6}\,\mathrm{A/m}$, $T_C^{\mathrm{Fe}}=1043\,$K (verifiable in Kittel). Numbers recomputed for this edition: the dipolar estimate at face value ($m = \mu_B$, $a = 0.25$ nm) gives $U_{dd} = 5.5\times10^{-25}$ J $= 0.040$ K (0.19 K for $2.2\,\mu_B$), a $2.6\times10^4$-fold — four-order — shortfall against $T_C$, sharper than the "$\sim$1 K, three orders" often quoted; $k_BT_C = 1.44\times10^{-20}$ J $= 0.09$ eV; domain economy for a 1 cm cube: $U_0 = \mu_0M_s^2V/6 = 0.61$ J, $\sigma_w = 1.5$ mJ/m² $\Rightarrow N^* \approx 2000$, $d \approx 5\ \mathrm{\mu m}$, residual $6.1\times10^{-4}$ J; silicon-steel loop $\sim$40 J/m³ at 50 Hz $\Rightarrow$ 2 kW/m³.

Builds on: Magnetization · Magnetic Dipole Moment of Atoms · Paramagnetism · The Auxiliary Field H · Magnetic Susceptibility

Leads to: Magnetic Permeability · Linear and Nonlinear Magnetic Media · Magnetic Boundary Conditions · Bound Currents