Exchange Interaction and Ferromagnetic Ordering

This chapter has been writing IOUs against one name for thirteen articles, and this is where they are paid. Ferromagnetism computed the failure: dipole–dipole coupling between neighboring iron moments is worth 0.04 K against an ordering temperature of 1043 K — four orders of magnitude short — so whatever locks the spins must supply chemical-bond currency, a tenth of an electron-volt per atom. Curie Law and Curie Temperature sharpened the embarrassment into a number: Weiss's molecular field works perfectly if each moment feels an internal field near a thousand teslas, which nothing magnetic in the crystal can produce. And standing behind both, Diamagnetism's prosecution of the Bohr–van Leeuwen theorem closed every classical escape route: classical statistical mechanics yields exactly zero equilibrium magnetization, from any mechanism, at any temperature. The aligning agent must be quantum, and it must be electrostatic in size. It is both, and it is called exchange. The derivation below runs the honest two-electron calculation: antisymmetry of the total wavefunction ties the spin configuration to the spatial configuration, and the spatial configuration decides how much Coulomb repulsion the electrons suffer. Parallel spins are forced into an antisymmetric spatial state that keeps the electrons apart; the Coulomb saving — an electrostatic integral with no magnetic content whatsoever — becomes an effective spin–spin coupling. Helium's spectrum shows the effect raw: 0.80 eV between states whose only difference is relative spin orientation, in a Hamiltonian containing no spin term at all. From there the chapter closes its arc: Dirac's rewriting gives the Heisenberg model, mean-field decoupling derives the Weiss field CURIE postulated — with $T_C = 2zJS(S+1)/3k_B$ turning iron's Curie point into a measurement of $J$ — and the honest caveats (itinerant iron, the Stoner criterion, magnons) mark where this track hands the story to solid-state physics.

Key takeaways

Exchange is Coulomb repulsion filtered through the Pauli principle: antisymmetry of the total two-electron wavefunction ties spin alignment to spatial separation, so the electrostatic energy depends on the spin configuration.
The two-electron calculation gives a singlet–triplet splitting $E_S - E_T = 2J_{\rm ex}$ , with $J_{\rm ex}$ an overlap-weighted Coulomb integral — electron-volt currency, no magnetic content.
Helium's 1s2s levels are the smoking gun: ortho (triplet) lies 0.80 eV below para (singlet) although the Hamiltonian contains no spin term; a magnetic origin would need a ~7000 T internal field.
Dirac's identity $\mathbf{S}_1\cdot\mathbf{S}_2 = -\tfrac34$ (singlet), $+\tfrac14$ (triplet) repackages the split as $H_{\rm eff} = -2J\,\mathbf{S}_1\cdot\mathbf{S}_2$ ; summed over neighbors this is the Heisenberg model $H = -2J\sum_{\langle ij\rangle}\mathbf{S}_i\cdot\mathbf{S}_j$ .
$J > 0$ locks neighbors parallel (ferromagnet); $J < 0$ locks them antiparallel (antiferromagnet, Néel order, superexchange through oxygen in MnO); conduction-electron-mediated RKKY coupling oscillates in sign with distance.
Mean-field decoupling of the Heisenberg model is the Weiss molecular field: $T_C = 2zJS(S+1)/3k_B$ . Iron's 1043 K then reads $J \approx 1.35\times10^{-21}$ J $\approx 8$ meV and a Weiss field $B_W \approx 700$ T — paying Curie Law and Curie Temperature's IOU with a mechanism.
KEY IDEA: no classical mechanism of any kind could align the moments (Bohr–van Leeuwen, prosecuted at Diamagnetism); exchange is quantum electrostatics masquerading as a spin–spin force, strong enough to hold a refrigerator magnet together at a thousand kelvin.

The core insight

The exchange interaction is the Pauli-enforced dependence of two electrons' Coulomb energy on their relative spin orientation — an electrostatic-scale effective spin coupling $H = -2J\sum\mathbf{S}_i\cdot\mathbf{S}_j$ whose mean-field average is precisely Weiss's molecular field, making ferromagnetism a quantum effect visible on a kitchen door.

Where this lives today

Exchange is a design variable now. Antiferromagnetic spintronics runs memory on the $J<0$ order nobody can see: Néel-vector bits are immune to stray fields and switch at terahertz rather than gigahertz rates, precisely because exchange, not anisotropy, sets their dynamics. Synthetic antiferromagnets — two ferromagnetic films RKKY-coupled through a ruthenium spacer tuned to the antiparallel sign — are standard issue inside every magnetic-tunnel-junction read head and MRAM cell. Superexchange's descendants run wilder: the same virtual-hopping arithmetic, on frustrated lattices where no arrangement satisfies every bond, produces quantum spin liquids, and cold-atom laboratories now build Heisenberg and Hubbard lattices atom by atom to watch exchange order emerge tunably. Even quantum computing pays the interaction rent: two-qubit gates in silicon spin-qubit processors are literally pulsed exchange couplings — $J(t)$ switched by a gate voltage that squeezes the orbital overlap this article's integral is built from.

Origins · 1 minWerner Heisenberg and how this was discovered

Werner Heisenberg

1901–1976 · Munich, Copenhagen, Leipzig

Heisenberg solved ferromagnetism in 1928 as an encore. Two years after inventing matrix mechanics at twenty-four, and one year after the uncertainty principle, he turned to the oldest embarrassment in magnetism — Weiss's unexplainable thousand-tesla molecular field — and recognized it as the same quantum resonance he and Dirac had formalized for identical particles and Heitler and London had just used to bind the hydrogen molecule. Exchange, he argued, is electrostatic energy sorted by spin symmetry, and its mean value is the Weiss field: this article's central derivations are that 1928 paper in modern dress. The Leipzig institute he then built trained a generation (Bloch's wall and spin waves were his student's habilitation), while Heisenberg himself went on to quantum field theory, the neutron–proton isospin idea, and cosmic-ray showers. His later career carried the century's weight — head of Germany's wartime nuclear project, a role still argued over; postwar rebuilder of German science at Göttingen and CERN's founding table. In the ferromagnet's ledger he is the physicist who found where a thousand teslas had been hiding: in a Coulomb integral with a minus sign.

History

1907

— Weiss postulates the molecular field, a thousand-tesla mystery bookkept but unexplained

1925

— Uhlenbeck and Goudsmit give the electron its spin; Pauli's exclusion principle acquires its full wavefunction form

1926

— Heisenberg and Dirac work out the quantum mechanics of identical particles: permutation symmetry ties spin to space

1927

— Heitler and London compute the hydrogen molecule; the exchange integral enters physics as the chemical bond's engine

1928

— Heisenberg identifies exchange as the origin of the Weiss field: ferromagnetism becomes quantum electrostatics

1929

— Dirac repackages the splitting as an effective spin Hamiltonian, the Heisenberg model's operator form

1932

— Néel proposes antiferromagnetic order for negative exchange; neutron diffraction confirms it in MnO in 1949

1950

— Anderson formalizes superexchange, the oxygen-mediated coupling of insulating magnets

The physical picture

Start with what the interaction is not. Two electron moments a lattice constant apart interact magnetically at the dipole scale — the two-line computation Ferromagnetism ran and this article recalls verbatim: about $5.5\times10^{-25}$ J, worth 0.04 K, a factor of $2.6\times10^4$ below what iron's ordering demands. And no cleverness with classical currents can do better, because Diamagnetism's Bohr–van Leeuwen prosecution stands: the classical partition function does not depend on the magnetic field at all, so classical matter in equilibrium has zero magnetization — not weak alignment, none, from any classical mechanism whatsoever. The theorem is a roadblock across the entire classical landscape, and it is why the resolution had to wait for 1926. Now the actual mechanism, in words before mathematics. Electrons are identical fermions: the total wavefunction — spatial part times spin part — must change sign when the two electrons are swapped. That bookkeeping rule has geometric teeth. If the spins are parallel (symmetric spin state), the spatial part must be antisymmetric: it has a node wherever the electrons coincide, so parallel-spin electrons are automatically kept out of each other's neighborhood. If the spins are opposed (antisymmetric singlet), the spatial part is symmetric and the electrons pile into the same region freely. Enter the only strong force in the problem: Coulomb repulsion. Electrons that avoid each other repel less. So the parallel-spin pair can sit lower in electrostatic energy than the antiparallel pair — by an amount set by Coulomb integrals over the overlapping orbitals, which is to say by chemistry-scale energies, tenths of electron-volts. The spins never touched each other; they merely selected, through antisymmetry, which spatial arrangement the pair was allowed to occupy, and the electric bill did the rest. Whether parallel or antiparallel wins depends on the orbital geometry — in the 3d bands of iron, cobalt and nickel, parallel wins; in most other arrangements antiparallel does, which is the sign story told below.

Figure

Antisymmetry ties spin to space - parallel spins are kept apart and pay less Coulomb repulsion

The two-electron derivation

Derivation AHeitler-London - the singlet-triplet split from pure electrostatics

The goal is the chapter's foundational quantum result: for two electrons shared between two neighboring atoms, the energy difference between the antiparallel (singlet) and parallel (triplet) spin configurations is $E_S - E_T = 2J_{\rm ex}$ , where $J_{\rm ex}$ is an electrostatic integral — no magnetic interaction anywhere in the Hamiltonian. The strategy is the Heitler–London construction from their 1927 hydrogen-molecule calculation: build the two symmetry-allowed spatial wavefunctions from atomic orbitals, attach the spin states that antisymmetry forces on each, and evaluate the Coulomb expectation in both. Watch what carries the spin dependence: not a spin operator in $H$ , but the $\pm$ sign in the spatial combination.

Step 1. Set up. Two atoms $a$ , $b$ with orthogonal-enough orbitals $\varphi_a$ , $\varphi_b$ ; two electrons 1, 2; Hamiltonian $H = h(1) + h(2) + V_{12}$ where $h$ is the one-electron part and $V_{12} = e^2/4\pi\varepsilon_0 r_{12}$ plus the cross nuclear attractions — every term spin-blind.

Step 2. Build the only two spatial states available to one electron per atom:

\Psi_\pm(\mathbf{r}_1,\mathbf{r}_2) = \frac{1}{\sqrt{2}}\left[\varphi_a(\mathbf{r}_1)\varphi_b(\mathbf{r}_2) \pm \varphi_a(\mathbf{r}_2)\varphi_b(\mathbf{r}_1)\right]

Antisymmetry of the total wavefunction now dictates the pairing: symmetric $\Psi_+$ must carry the antisymmetric spin singlet ( $S = 0$ ); antisymmetric $\Psi_-$ must carry the symmetric spin triplet ( $S = 1$ ). Why there is no choice here: this is the Pauli principle in its full form — not "no two electrons in the same state" but "total wavefunction odd under swap." The spin state is the tag that decides which spatial state the pair is permitted, and that is the interaction's entire origin.

Step 3. Evaluate the energy in each state (neglecting the overlap integral's small corrections, flagged below):

E_\pm = \varepsilon_a + \varepsilon_b + C \pm J_{\rm ex}

with the direct (Coulomb) and exchange integrals

C = \int |\varphi_a(\mathbf{r}_1)|^2\,V_{12}\,|\varphi_b(\mathbf{r}_2)|^2\,d\tau_1 d\tau_2, \qquad J_{\rm ex} = \int \varphi_a^*(\mathbf{r}_1)\varphi_b^*(\mathbf{r}_2)\,V_{12}\,\varphi_a(\mathbf{r}_2)\varphi_b(\mathbf{r}_1)\,d\tau_1 d\tau_2

where: $C$ — the classical repulsion of the two charge clouds; $J_{\rm ex}$ — the exchange integral, the same interaction evaluated between swapped configurations, weighted by orbital overlap. It has no classical reading: it is the interference term of two indistinguishable ways to distribute the electrons.

Step 4. Read off the splitting:

E_S - E_T = E_+ - E_- = 2J_{\rm ex} \qquad (1)

The singlet (symmetric space, electrons overlapping) and triplet (antisymmetric space, node at coincidence) differ by twice the exchange integral — electrostatic in magnitude because $V_{12}$ is, nonzero only where orbitals overlap, and existing at all only because of antisymmetry. Verified: this is the Heitler–London 1927 structure as presented in Ashcroft & Mermin ch. 32 and Blundell ch. 4; the honest fine print is retained — with overlap S included the energies are $(C \pm J_{\rm ex})/(1 \pm S^2)$ , and for the hydrogen molecule the singlet wins ( $J_{\rm ex} < 0$ effectively there), which is why H₂ bonds rather than magnetizes; the sign of the effective J is a genuine competition, set by orbital geometry — the sign story below.

The smoking gun is helium. Excite one electron to 2s: the configurations 1s2s come in two flavors, parahelium (singlet) and orthohelium (triplet), identical in orbitals and differing only in relative spin. The triplet sits 0.80 eV below the singlet. The helium Hamiltonian contains no spin-dependent term at that accuracy — no dipole coupling, no spin–orbit worth 0.8 eV — yet flipping a spin costs nearly an electron-volt. If you insisted on reading it magnetically, the required internal field would be $B = \Delta E/2\mu_B \approx 6.9\times10^3$ T. There is no such field in a helium atom; there is only Coulomb repulsion and antisymmetry. Exchange in a single atom, measured spectroscopically, at exactly the scale ferromagnetism needs — half a century before anyone knew why iron sticks.

Dirac's repackaging: the spin Hamiltonian

The splitting is electrostatic, but its label is spin — so Dirac observed (1929) that one can forget the orbitals and write an effective operator acting on spins alone that reproduces the spectrum.

Go deeper · Derivation B · 1 minFrom the two-level spectrum to H equals minus 2J S1 dot S2

Derivation BFrom the two-level spectrum to H equals minus 2J S1 dot S2

The goal is the Dirac–Heisenberg effective Hamiltonian: an operator built from $\mathbf{S}_1\cdot\mathbf{S}_2$ whose eigenvalues reproduce the singlet and triplet energies, so that all downstream physics can treat exchange as a spin–spin coupling. The strategy is two lines of angular-momentum algebra plus matching. Watch the conventions at the end — the literature's factor-of-2 and double-counting traps are real, and this track fixes its choice once.

Step 1. Evaluate $\mathbf{S}_1\cdot\mathbf{S}_2$ on the two sectors. With $\mathbf{S} = \mathbf{S}_1 + \mathbf{S}_2$ and $S_1 = S_2 = \tfrac12$ (units of $\hbar$ ):

\mathbf{S}_1\cdot\mathbf{S}_2 = \tfrac12\left[S(S+1) - \tfrac34 - \tfrac34\right] = \begin{cases} -\tfrac34 & \text{singlet } (S=0) \\ +\tfrac14 & \text{triplet } (S=1) \end{cases}

Step 2. Match a linear form $H_{\rm eff} = \alpha + \beta\,\mathbf{S}_1\cdot\mathbf{S}_2$ to the spectrum $E_S$ , $E_T$ . The two conditions give $\beta = -(E_S - E_T) = -2J_{\rm ex}$ and $\alpha = \tfrac14 E_S + \tfrac34 E_T$ . Dropping the spin-independent constant,

H_{\rm eff} = -2J\,\mathbf{S}_1\cdot\mathbf{S}_2, \qquad 2J \equiv E_S - E_T \qquad (2)

where: $J > 0$ makes the triplet (parallel spins) the ground state — the ferromagnetic sign. The operator is exact on the two-level subspace: nothing was approximated except forgetting where the numbers came from.

Step 3. Promote to a lattice. With one net spin per magnetic site and exchange active only across the overlap of near neighbors,

H = -2J\sum_{\langle ij\rangle}\mathbf{S}_i\cdot\mathbf{S}_j \qquad (3)

where: $\langle ij\rangle$ — each nearest-neighbor bond counted once. This is the Heisenberg model, the drosophila of magnetism.

Step 4. Convention honesty, stated once for the whole track: some texts absorb the 2 into $J$ ( $H = -J\sum\mathbf{S}_i\cdot\mathbf{S}_j$ ), some sum over ordered pairs (double-counting bonds, halving $J$ again), and older literature quotes $J$ per electron pair with overlap corrections. A quoted " $J$ " is meaningless without its Hamiltonian; factor-of-2 discrepancies between tables are almost always this, not physics. This track's convention is equation (3): bonds once, explicit 2. Verified: the S1·S2 eigenvalues follow from the angular-momentum identity (recomputed above); the matching is unique for a two-level spectrum; equation (3) with J>0 reproduces the ferromagnetic ground state and matches the convention of Blundell ch. 5 and Kittel ch. 12.

The sign story

Nothing in the derivation forced $J > 0$ ; the exchange integral's sign is a real competition between the direct overlap gain and the kinetic-plus-nuclear terms folded into the effective coupling, and nature uses both signs freely.

$J > 0$ — ferromagnetism. Parallel alignment wins where the same-site Coulomb saving dominates: the 3d metals iron, cobalt, nickel (with the itinerant caveat below), plus gadolinium and a shelf of alloys and compounds. Everything the last five articles built — domains, walls, hysteresis, energy products — rides on this sign.

$J < 0$ — antiferromagnetism. Antiparallel alignment wins in most insulating magnetic compounds. The lattice splits into two interpenetrating sublattices, magnetized oppositely — Néel order — with zero net moment: a hidden magnetism invisible to a compass, betrayed by susceptibility kinks at the Néel temperature and laid bare by neutron diffraction (which is how Shull vindicated Néel in 1949, in MnO). The usual mechanism is superexchange (Kramers 1934, Anderson 1950): in MnO the manganese moments sit too far apart to overlap directly and talk through the intervening oxygen — virtual hopping of the oxygen p-electrons is Pauli-allowed only when the flanking Mn spins are antiparallel, so the kinetic-energy saving picks the antiferromagnetic sign. Exchange routed through a nonmagnetic middleman, one more electrostatic conspiracy wearing spin clothing.

Oscillating $J$ — RKKY. In metals with localized moments (rare earths dissolved in conduction seas), the coupling is carried by the conduction electrons and oscillates in sign with distance, $J(r) \propto \cos(2k_Fr)/r^3$ — so the same pair of species can order ferro- or antiferromagnetically depending on spacing. Named for Ruderman, Kittel, Kasuya and Yosida (1954–57), it is the coupling that giant-magnetoresistance multilayers dial by choosing spacer thickness — the read-head physics of the hard disks whose grains Magnetic Domains and Domain Walls sized.

Figure

The sign of J picks the order - ferromagnet, Neel antiferromagnet, and superexchange through oxygen

Paying the Weiss IOU

Curie Law and Curie Temperature built its entire quantitative apparatus — Curie–Weiss law, spontaneous magnetization, $T_C = \lambda C$ — on a postulated molecular field $\mathbf{H}_{\rm eff} = \mathbf{H} + \lambda\mathbf{M}$ , and confessed that the $\lambda M \sim 10^3$ T pseudo-field could not be magnetic. Here is the payment: the Weiss field is the mean-field shadow of equation (3).

Go deeper · Derivation C · 2 minMean-field decoupling - the Weiss field derived and the Curie temperature priced

Derivation CMean-field decoupling - the Weiss field derived and the Curie temperature priced

The goal is to derive Weiss's molecular field from the Heisenberg model and extract $T_C = 2zJS(S+1)/3k_B$ — closing the loop with Curie Law and Curie Temperature by supplying the microscopic $\lambda$ it postulated. The strategy is the mean-field decoupling: each spin sees its neighbors only through their thermal average, converting the interacting lattice into one spin in a self-consistent effective field. Watch what is being thrown away — correlations between neighboring fluctuations — because that omission is exactly why mean-field $T_C$ overestimates and why the critical exponents come out wrong, honesty CURIE already flagged.

Step 1. Isolate one site $i$ in equation (3): $H_i = -2J\,\mathbf{S}_i\cdot\sum_{j\in{\rm nn}(i)}\mathbf{S}_j$ . Replace the neighbors by their average, $\mathbf{S}_j \to \langle\mathbf{S}\rangle$ :

H_i^{\rm MF} = -2zJ\,\mathbf{S}_i\cdot\langle\mathbf{S}\rangle

with $z$ the coordination number (bcc iron: $z = 8$ ).

Step 2. Read it as a Zeeman term. A spin in a field $\mathbf{B}$ has energy $-g\mu_B\mathbf{S}\cdot\mathbf{B}$ (dictionary from Magnetic Dipole Moment of Atoms), so the neighbors act as an effective field

\mathbf{B}_W = \frac{2zJ}{g\mu_B}\,\langle\mathbf{S}\rangle

With magnetization $\mathbf{M} = n g\mu_B\langle\mathbf{S}\rangle$ ( $n$ = moments per volume), this is $\mathbf{B}_W = \left(2zJ/n g^2\mu_B^2\right)\mathbf{M}$ — proportional to M, exactly Weiss's ansatz, with the proportionality now a formula in $J$ rather than a postulate. Why this pays the IOU: Curie Law and Curie Temperature needed $\lambda M \sim 10^3$ T and could name no magnetic source. The source is not magnetic: $B_W$ is a bookkeeping translation of electrostatic exchange into field language. No flux, no dipoles — a pseudo-field that acts on spin orientation only.

Step 3. Extract $T_C$ . Feed $B_W$ into the paramagnetic response CURIE derived: for spins $S$ the small-field susceptibility gives $\langle S_z\rangle = S(S+1)g\mu_B B/3k_BT$ . Self-consistency of the ordered state at zero applied field requires $\langle S_z\rangle = S(S+1)\,2zJ\langle S_z\rangle/3k_BT$ to have a nonzero solution, which first happens at

T_C = \frac{2zJS(S+1)}{3k_B} \qquad (4)

where: all the microscopy sits in $z$ and $J$ . Above $T_C$ the same algebra returns the Curie–Weiss law $\chi = C/(T - T_C)$ — CURIE's machinery, now with its $\theta$ microscopically priced.

Step 4. Invert for iron. With $T_C = 1043$ K, $z = 8$ , effective $S = 1$ :

J = \frac{3k_BT_C}{2zS(S+1)} = \frac{3(1.381\times10^{-23})(1043)}{2\cdot8\cdot2} = 1.35\times10^{-21}\ \mathrm{J} \approx 8.4\ \mathrm{meV}

— millielectron-volt exchange per bond, four orders above the dipole scale, exactly the currency the failure analysis demanded. The Weiss field itself: $B_W \approx k_BT_C/\mu_{\rm atom} = (1.381\times10^{-23})(1043)/(2.2\times9.274\times10^{-24}) \approx 7.1\times10^2$ T — the seven-hundred-tesla pseudo-field CURIE's honesty required and no magnet could supply. Verified: equation (4) matches Blundell ch. 5 and Ashcroft & Mermin ch. 33; iron numbers recomputed for this edition (J = 1.35×10⁻²¹ J = 8.4 meV, B_W ≈ 706 T); caveats carried honestly — mean-field neglects correlations (real T_C sits below the mean-field prediction for given J, so this J is an effective, slightly underestimating extraction) and iron's S = 1 with g = 2 is itself a fiction the itinerant paragraph below confesses.

One more loan is repaid here. Magnetic Domains and Domain Walls borrowed an exchange stiffness $A \approx 2\times10^{-11}$ J/m to price its walls. Stiffness is $J$ made continuum: tilting neighboring spins by small angles costs $JS^2\theta^2$ per bond (the expansion in that article's own derivation), and summing bonds across a lattice of spacing $a$ gives $A \sim c\,zJS^2/a$ with a lattice-geometry constant $c$ of order unity divided by a few. Iron's numbers — $J = 1.35\times10^{-21}$ J, $z = 8$ , $a = 0.287$ nm — land $A$ in the $10^{-11}$ J/m decade, consistent with both the dimensional estimate $k_BT_C/2a = 2.5\times10^{-11}$ J/m that article quoted and the measured value. The wall width, the domain pattern, the single-domain limit, the superparamagnetic floor: all of it was this article's $J$ , coarse-grained.

The honesty section: iron is not actually a Heisenberg magnet

The Heisenberg model assumes moments of fixed length localized on sites — a picture that is quantitatively right for insulating magnets (EuO, the rare-earth salts CURIE's Brillouin functions fit so well) and for gadolinium's buried 4f shell. Iron's magnetic electrons are 3d band electrons: itinerant, shared, delocalized. The tell is in Magnetic Dipole Moment of Atoms' own data: iron's moment is $2.2\,\mu_B$ per atom — a non-integer impossible for any localized shell, but natural for a band whose spin-up and spin-down populations differ by a non-integer average. The itinerant theory (Stoner, 1938) reframes the question: exchange in the band lowers the energy of a spin-split Fermi sea when the density of states at the Fermi level is large enough — the Stoner criterion — and ferromagnetism becomes a property of band structure, named here and owned by solid-state physics rather than this track. The localized picture survives as the right effective theory at domain-wall and thermodynamic scales (which is why this chapter got away with it), with $J$ , $S$ and $g$ read as effective parameters rather than atomic bookkeeping. The model's other frontier is its excitations. The Heisenberg ground state with all spins parallel is exact for $J>0$ , but its low-lying excitations are not single flipped spins — a flip costs $2zJS$ at once — rather spin waves: long-wavelength ripples of tilt spread across the lattice, costing arbitrarily little, quantized as magnons. Bloch's $T^{3/2}$ law for the low-temperature fall of $M_s$ (the same Bloch whose wall Magnetic Domains and Domain Walls built) is magnon counting, and inelastic neutron scattering maps magnon dispersions the way X-rays map phonons. Both threads — Stoner bands and magnon gases — are where this chapter's story continues in a condensed-matter course.

Limiting cases

Case	Result	Meaning
orbital overlap → 0	$J_{\rm ex} \to 0$	exchange is short-ranged: no overlap, no interaction — why it acts between near neighbors only, and why moments centimetres apart interact only dipolarly
classical limit ( $\hbar \to 0$ )	no exchange; $M = 0$	antisymmetry is quantum bookkeeping; Bohr–van Leeuwen (prosecuted at Diamagnetism) guarantees classical matter cannot magnetize at all
$J > 0$ , $T \ll T_C$	saturated ferromagnet, magnon corrections $\propto T^{3/2}$	Bloch's law; excitations are spin waves, not single flips
$J > 0$ , $T > T_C$	Curie–Weiss $\chi = C/(T-\theta)$ , $\theta = T_C > 0$	CURIE's high-temperature machinery with $\theta$ microscopically priced by equation (4)
$J < 0$	Néel order below $T_N$ ; $\chi = C/(T+\lvert\theta\rvert)$ above	antiferromagnetism: negative Weiss constant, susceptibility peak at $T_N$ , invisible to a compass
moments in a conduction sea	RKKY: $J(r)$ oscillates in sign	spacer-thickness-tunable coupling; the GMR multilayer dial
band electrons (Fe, Co, Ni)	Stoner criterion replaces localized $J$	itinerant ferromagnetism; non-integer moments (2.2 μ_B) are the tell

Common trapExchange is not a magnetic force - and the Heisenberg Hamiltonian does not mean the spins attract like compass needles

The operator $-2J\,\mathbf{S}_1\cdot\mathbf{S}_2$ looks like a direct spin–spin interaction, and the single most durable misconception in magnetism is to read it as one — as if neighboring moments align by magnetic attraction, merely stronger than expected. Every quantitative fact in this chapter convicts that reading: the honest dipole computation (Ferromagnetism's 0.04 K against 1043 K), helium's 0.80 eV splitting in a spin-free Hamiltonian, and the Bohr–van Leeuwen theorem, which forbids any classical interaction — dipolar or otherwise — from producing equilibrium magnetization at all. The spin operators in equation (3) are labels: they index which spatial symmetry sector the electrons occupy, and the energy differences are Coulomb integrals. Corollary worth stating: exchange conserves total spin and couples to nothing magnetic — it cannot rotate the net moment, which is why pure exchange gives no anisotropy and Magnetic Domains and Domain Walls had to import $K$ from spin–orbit coupling separately.

Worked examples

1 · Helium's spin premium, priced magnetically (confidence builder)

The 1s2s configuration of helium splits by 0.80 eV between ortho (triplet, lower) and para (singlet). What internal magnetic field would be needed to produce this splitting magnetically, and what temperature does it correspond to?

Step 1. A spin flip in a field $B$ costs $\Delta E = 2\mu_BB$ (one Bohr magneton reversing). Setting $2\mu_BB = 0.80$ eV:

B = \frac{0.80\times1.602\times10^{-19}}{2\times9.274\times10^{-24}} = 6.9\times10^3\ \mathrm{T}

Step 2. Seven thousand tesla — two hundred times the strongest continuous laboratory field — inside a two-electron atom with no magnetic machinery beyond two tiny dipoles. The magnetic reading is absurd; the electrostatic one is exact.

Step 3. As a temperature: $T = \Delta E/k_B = (0.80\times1.602\times10^{-19})/(1.381\times10^{-23}) = 9.3\times10^3$ K. Exchange energies run at thousands of kelvin — precisely the currency needed to keep iron ordered at 1043 K, where dipolar coupling (0.04 K, Ferromagnetism's verdict, recalled) fails by $2.6\times10^4$ .

2 · Iron's exchange constant and the Weiss field (exam level)

From $T_C = 1043$ K, $z = 8$ , effective $S = 1$ , extract $J$ ; from $\mu_{\rm atom} = 2.2\,\mu_B$ , price the Weiss field; then check the stiffness loan to Magnetic Domains and Domain Walls.

Step 1. Invert equation (4): $J = 3k_BT_C/[2zS(S+1)] = 3(1.381\times10^{-23})(1043)/(2\times8\times1\times2) = 1.35\times10^{-21}$ J = 8.4 meV per bond. Sanity: per atom the exchange budget is $\sim zJ \approx 0.07$ eV, matching the $k_BT_C \approx 0.09$ eV scale Ferromagnetism read off.

Step 2. Weiss field: $B_W \approx k_BT_C/\mu_{\rm atom} = (1.44\times10^{-20})/(2.2\times9.274\times10^{-24}) = 7.1\times10^2$ T ≈ 700 T — the thousand-tesla-class pseudo-field Curie Law and Curie Temperature required, now sourced: not a field at all, but exchange wearing field units.

Step 3. The stiffness check: $A \sim zJS^2/(2a)$ -class estimates with $J = 1.35\times10^{-21}$ J and $a = 0.287$ nm give $A \approx 1.9\times10^{-11}$ J/m (lattice constant of order unity absorbed), against the $2\times10^{-11}$ J/m Magnetic Domains and Domain Walls borrowed and the dimensional $k_BT_C/2a = 2.5\times10^{-11}$ J/m. Consistent to the O(1) factors the convention warning promised — the wall widths, domain sizes and blocking diameters built on that loan stand.

Going further

References. Griffiths, Introduction to Electrodynamics, 4th ed., §6.4 · Jackson, Classical Electrodynamics, 3rd ed., §5.8 · Ashcroft & Mermin, Solid State Physics, ch. 32–33 (the exchange and magnetic-ordering chapters this article tracks) · Kittel, Introduction to Solid State Physics, 8th ed., ch. 12 · Blundell, Magnetism in Condensed Matter, ch. 4–5 (the cleanest modern treatment at this register) · Anderson, "Antiferromagnetism. Theory of Superexchange," Phys. Rev. 79, 350 (1950).

Provenance. Equations are standard Heitler–London/Heisenberg exchange and Weiss mean-field results; a verifier should recompute the dipolar energy ( $\sim10^{-4}$ eV $\approx$ 1 K) from $\mu_0\mu_B^2/(4\pi a^3)$ with $a=2\times10^{-10}$ m, check the singlet–triplet algebra $\mathbf S_1\cdot\mathbf S_2=+1/4,-3/4$ , verify $k_BT_C=2zJS(S+1)/3$ and the Curie constant $C=Ng^2\mu_B^2 J(J+1)/3k_B$ , and confirm $T_C^{\text{Fe}}=1043$ K ( $k_BT_C\approx0.090$ eV). Cross-checks: Griffiths §6.4, Jackson §5.8, Ashcroft & Mermin Ch. 32–33, Kittel Ch. 12, Blundell Ch. 4–5. Constants: $\mu_B=9.274\times10^{-24}$ J/T, $\mu_0=4\pi\times10^{-7}$ T·m/A, $k_B=1.381\times10^{-23}$ J/K. Addendum for this edition: the base's " $\sim1$ K" dipolar figure is the generous literature estimate; this track's honest single-magneton computation is owned by Ferromagnetism (0.040 K at $a = 0.25$ nm, $2.6\times10^4$ -fold short) and recalled verbatim, not re-derived. Iron extraction recomputed: $J = 1.35\times10^{-21}$ J $= 8.4$ meV ( $z=8$ , $S=1$ ), $B_W \approx 706$ T; helium's 0.80 eV priced at $6.9\times10^3$ T equivalent; stiffness cross-check $A \sim 2\times10^{-11}$ J/m repays Magnetic Domains and Domain Walls' loan. $\mu_0$ used at its 2019-rebased value per Magnetic Permeability.

Builds on: Paramagnetism · Ferromagnetism · Magnetic Dipole Moment of Atoms · Magnetization

Leads to: Field of a Magnetized Object · The Auxiliary Field H · Magnetic Susceptibility · Linear and Nonlinear Magnetic Media

If exchange is just electrostatics, why does it show up only in magnetism courses and not in electrostatics ones?

Because its output is spin order, even though its engine is Coulomb energy. Electrostatics proper (E1–E3 of this track) treats charges whose arrangement you specify; exchange is a quantum statistical effect deciding which arrangements identical electrons are allowed, and the decision is labeled by spin. The energy is electric; the order parameter is magnetic. It also simply cannot be written in classical field language — there is no $\rho$ , no $\mathbf{E}$ , no configuration of classical charges that reproduces a spin-dependent energy — which is why it enters physics through quantum mechanics courses and enters this electrodynamics track as an honest import, flagged as such since Diamagnetism first showed classical magnetism is empty.

The Heisenberg model says J only couples nearest neighbors. What actually happens between two spins ten sites apart?

Directly, almost nothing: $J_{\rm ex}$ is built from orbital overlap, which dies exponentially with separation — a few lattice constants kill it. Two spins ten sites apart interact through chains of nearest-neighbor exchange (which is how order propagates and what spin waves are), through conduction electrons if the material is a metal (RKKY, oscillating and $\propto 1/r^3$ ), and always through the honest dipole–dipole coupling — negligible per pair (0.04 K) but long-ranged, and therefore the ruler of large-scale arrangements: it is exactly the magnetostatic energy that broke uniform order into domains in Ferromagnetism. Short range: exchange decides alignment. Long range: dipoles decide geography. The chapter's whole architecture is that division of power.

Why does the sign of the exchange integral differ between systems - what physically decides ferro versus antiferro?

It is a competition of channels. Direct exchange between well-overlapping orthogonal orbitals tends to favor parallel spins (the Coulomb-saving argument of the main derivation — Hund's first rule is exactly this within one atom). But when electrons can hop between sites, second-order kinetic energy favors antiparallel spins, because hopping onto an occupied orbital is Pauli-blocked unless the spins oppose — that channel dominates in most insulators (superexchange), which is why antiferromagnets vastly outnumber ferromagnets among magnetic compounds. Geometry arbitrates: overlap angles, on-site repulsion, orbital degeneracy. The Goodenough–Kanamori rules of solid-state chemistry are the case law; this track needs only the verdict that both signs occur and are electrostatic in size.

Does exchange explain why the compass needle aligns with Earth's field? Where does exchange stop and ordinary magnetism take over?

Exchange builds the needle; ordinary magnetism steers it. Inside the steel, exchange locks $\sim10^{23}$ spins into domains (with anisotropy and magnetostatics arranging them, per the last four articles), creating a mesoscopic moment. That collective moment then couples to Earth's 50 μT field through the ordinary Zeeman interaction — utterly negligible per atom ( $\mu_BB \sim 3\times10^{-9}$ eV, hopeless against thermal agitation), but multiplied by $10^{23}$ correlated atoms it becomes a torque that swings the needle. Every practical magnetic device runs this two-stage architecture: quantum electrostatics to build a rigid moment, classical field-dipole coupling to use it.

What is the Stoner criterion, in one honest paragraph?

For band electrons, ask whether the Fermi sea gains by spin-splitting: move a sliver of down-spin electrons into the up-spin band. Kinetic energy rises (they must occupy higher states), by an amount inversely proportional to the density of states at the Fermi level, $g(E_F)$ ; exchange energy falls, proportional to an interaction parameter $I$ (the band cousin of this article's $J_{\rm ex}$ ). Net gain — spontaneous band ferromagnetism — requires $I\,g(E_F) > 1$ : strong exchange and many states available at the Fermi surface. Iron, cobalt and nickel pass because their narrow 3d bands pile up enormous $g(E_F)$ ; copper, one element over, fails and stays the diamagnet Diamagnetism priced. Non-integer moments like iron's 2.2 μ_B fall out naturally, since the split populations need not differ by whole electrons. The full machinery is solid-state property; this track files the criterion under "named honestly, owned elsewhere."

Ferromagnetism needs J positive, but the hydrogen molecule - the original Heitler-London system - binds as a singlet. Why the opposite sign?

Because H₂'s effective exchange is dominated by the bonding channel: putting both electrons, with opposed spins, into the symmetric orbital combination between the protons lowers kinetic and potential energy enormously — that is the covalent bond, and it forces the singlet ground state (equivalently, the effective $J < 0$ ). The ferromagnetic sign needs the opposite regime: orbitals orthogonal enough that double occupation is expensive, so the system prefers one electron per site with parallel spins collecting the exchange-integral saving. Chemistry lives mostly in the first regime (hence paired, spin-silent bonding electrons — the fact Paramagnetism used to explain why most matter has no net moment), magnetism in the second. Same integrals, different balance — and the balance is precisely what the sign story's rules adjudicate.

Are magnons real particles? Can you count them?

As real as phonons — quantized collective excitations with definite energy and crystal momentum, obeying Bose statistics. Each magnon reduces the total spin by exactly $\hbar$ and carries a dispersion $\omega(k)$ (quadratic at small $k$ for a ferromagnet, $\propto Jk^2a^2$ — stiffness again). They are counted routinely: inelastic neutron scattering measures $\omega(k)$ bond by bond (it is the standard experimental route to $J$ itself, sharper than the $T_C$ inversion this article used), Bloch's $T^{3/2}$ law is thermal magnon census arithmetic, and magnonics laboratories now pipe coherent magnon beams through waveguides as information carriers. The chapter's mean-field story is the magnon vacuum; everything thermal about a real ferromagnet below $T_C$ is its magnon gas.