Antiferromagnetism

Antiferromagnetism is magnetic order whose leading order parameter varies within the chemical unit cell or at a nonzero ordering wavevector. In the simplest collinear Néel state, two equivalent sublattices carry equal and opposite moments:

$$\mathbf M_0, \qquad -\mathbf M_0.$$

The ideal bulk uniform magnetization therefore vanishes even though each sublattice is ordered. This compensation is why ordinary magnetometry can miss an antiferromagnet and why diffraction, local probes, and symmetry-resolved measurements are central.

Antiferromagnetism is broader than the two-sublattice picture. Magnetic structures can be collinear or noncollinear, commensurate or incommensurate, localized-moment or itinerant, and single-$\mathbf Q$ or multi-$\mathbf Q$. Some antiferromagnets cant and acquire a small uniform moment; some remain metallic; some have strong antiferromagnetic correlations but no ordered phase. “Antiparallel neighboring spins” is therefore a useful motif, not a complete definition.

The magnetic pattern, rather than the net moment alone, identifies the phase. This page connects that pattern to diffraction, field-induced canting, spin waves, and local-moment or itinerant descriptions. Order Parameters develops phase selection, and Magnons derives the spin-wave formalism.

Required background. Exchange Interactions supplies material coupling signs, the Heisenberg Model supplies the lattice-spin convention, and Order Parameters supplies finite-wavevector source selection.

Helpful background. Ferromagnetism provides the uniform-order and domain contrast used throughout.

Néel order and its lattice pattern

Let a bipartite lattice be divided into sublattices $A$ and $B$, and define

$$\eta_i = \begin{cases} +1, &i\in A, \\[4pt] -1, &i\in B. \end{cases}$$

An ideal collinear Néel state has

$$\left\langle \mathbf s_i \right\rangle = \eta_i m_s\mathbf n,$$

where $m_s$ is the ordered spin per site and $\mathbf n$ is a unit vector specifying the Néel-axis orientation. The uniform spin average is zero when the two sublattices are equivalent, while the staggered average is nonzero:

$$\frac{1}{N} \sum_i \left\langle \mathbf s_i \right\rangle = \mathbf0, \qquad \frac{1}{N} \sum_i \eta_i \left\langle \mathbf s_i \right\rangle = m_s\mathbf n.$$

If an ordering wavevector $\mathbf Q_{\mathrm{AF}}$ satisfies

$$e^{i\mathbf Q_{\mathrm{AF}}\cdot\mathbf r_i} = \eta_i,$$

then the same order is the spin Fourier component at $\mathbf Q_{\mathrm{AF}}$. For a one-dimensional chain,

$$Q_{\mathrm{AF}} = \frac{\pi}{a},$$

and for the nearest-neighbor square lattice,

$$\mathbf Q_{\mathrm{AF}} = \left( \frac{\pi}{a}, \frac{\pi}{a} \right).$$

Wavevectors are defined modulo a chemical reciprocal-lattice vector. Crystallographic papers often quote them in reciprocal-lattice units, so the numerical coordinates depend on the chosen conventional cell.

A nonzero propagation vector is not compulsory. If a chemical cell already contains both opposing sublattices, the same pattern can repeat in every cell with $\mathbf Q=0$. For example, honeycomb Néel order has opposite $A$ and $B$ moments in its two-site primitive basis. The order is the antisymmetric basis combination, not a uniform magnetization. In general write $\langle\mathbf s_{\mathbf R\alpha}\rangle$ with both a cell coordinate $\mathbf R$ and a basis label $\alpha$; the propagation vector describes variation between cells, while basis amplitudes describe variation inside one.

Uniform and staggered variables

For equal populations, let $\mathbf M_A$ and $\mathbf M_B$ be each sublattice’s mean moment per site multiplied by the total site density. Then the physical bulk magnetization and the staggered component are

$$\mathbf M = \frac{ \mathbf M_A+\mathbf M_B }{2}, \qquad \mathbf L = \frac{ \mathbf M_A-\mathbf M_B }{2}.$$

Here $\mathbf M$ is the physical uniform component and $\mathbf L$ the Néel component. If instead each sublattice density is quoted per total sample volume, those two contributions add directly to the bulk magnetization. The factors of two depend on this normalization, not on different physics.

The inverse relations are

$$\mathbf M_A = \mathbf M+\mathbf L, \qquad \mathbf M_B = \mathbf M-\mathbf L.$$

If the two sublattice moments have equal fixed magnitude $M_0$, then

$$\mathbf M\cdot\mathbf L = 0, \qquad M^2+L^2 = M_0^2.$$

The ideal compensated state has $\mathbf M=\mathbf0$ and $\mathbf L=M_0\mathbf n$. A field-induced canting moment changes $\mathbf M$ without erasing $\mathbf L$.

The selecting source is staggered

A uniform magnetic field couples directly to $\mathbf M$, not to $\mathbf L$. The source conjugate to Néel order alternates between sublattices:

$$\Delta H_s = - \sum_i \eta_i \mathbf h_s\cdot\mathbf s_i.$$

The source-selected order parameter is

$$\mathbf m_s = \lim_{\mathbf h_s\to\mathbf0^+} \lim_{V\to\infty} \frac{1}{N} \sum_i \eta_i \left\langle \mathbf s_i \right\rangle_{\mathbf h_s}.$$

The thermodynamic limit comes first. A finite isotropic antiferromagnet can have a total-spin-singlet ground state and hence zero one-point staggered magnetization, while its correlations and low-lying spectrum already contain the approach to Néel order.

Correlation and Diffraction Diagnostics

The equal-time spin structure factor is

$$S(\mathbf q) = \frac{1}{N} \sum_{i,j} e^{i\mathbf q\cdot(\mathbf r_i-\mathbf r_j)} \left\langle \mathbf s_i\cdot\mathbf s_j \right\rangle.$$

For the simple phase-factor pattern above, true long-range order gives

$$S(\mathbf Q_{\mathrm{AF}}) \sim N m_s^2$$

as $N\to\infty$. A large but subextensive peak can instead represent a long finite correlation length. The finite-size test is the scaled quantity

$$\frac{ S(\mathbf Q_{\mathrm{AF}}) }{N},$$

not merely whether the peak is visually prominent. With several sites per chemical cell, use the structure-factor matrix in the ordered basis channel; a compensated $\mathbf Q=0$ pattern can cancel in the uniform sum even while its staggered channel is extensive.

Elastic magnetic neutron diffraction measures a transverse magnetic structure factor. Schematically,

$$\mathbf F_M(\mathbf K) = \sum_j f_j(\mathbf K) e^{i\mathbf K\cdot\mathbf r_j} \left[ \mathbf M_j - \widehat{\mathbf K} \left( \widehat{\mathbf K}\cdot\mathbf M_j \right) \right],$$

and

$$I_M(\mathbf K) \propto \left| \mathbf F_M(\mathbf K) \right|^2.$$

The magnetic form factor $f_j$ reflects the spatial distribution of the magnetic electrons. The transverse projection means moments parallel to $\mathbf K$ do not contribute to that reflection.

For a commensurate antiferromagnet, new magnetic Bragg peaks can appear at

$$\mathbf K = \mathbf G \pm \mathbf Q_{\mathrm{AF}},$$

where $\mathbf G$ is a chemical reciprocal-lattice vector. If $\mathbf Q_{\mathrm{AF}}$ is itself equivalent to a structural reciprocal vector in the chosen cell, magnetic intensity may overlap nuclear peaks; polarization analysis, temperature dependence, and symmetry refinement are then needed.

Sublattices and the Magnetic Unit Cell

A graph is bipartite if its sites can be colored $A$ and $B$ so every relevant bond joins opposite colors. Nearest-neighbor square, honeycomb, and simple-cubic lattices are bipartite. A triangle is not: any two-coloring leaves one bond connecting equal colors.

On a bipartite crystal, a primitive chemical translation can interchange $A$ and $B$. The Néel state then has a larger magnetic unit cell than the chemical cell. A one-site translation changes

$$\mathbf n \longrightarrow -\mathbf n.$$

Time reversal also changes $\mathbf n\to-\mathbf n$. Each symmetry can be broken while their product remains a symmetry of a collinear antiferromagnetic state. Spin–orbit coupling and the actual crystallographic operation determine the full magnetic space group.

The orientations $\mathbf n$ and $-\mathbf n$ should not be identified casually. When a chemical translation swaps the sublattices, they are distinct antiphase domains related by that broken translation. More generally they are time-reversed magnetic states, even when no chemical translation swaps the sublattices. Many bulk observables are nevertheless even in $\mathbf n$. Whether a coarse-grained theory can treat the axis as director-like depends on which lattice and domain information has been integrated out.

Beyond two sublattices

The defining structure may require:

• three sublattices for $120^\circ$ triangular order;
• four or more sublattices for complex commensurate structures;
• an incommensurate phase $e^{i\mathbf Q\cdot\mathbf r}$ for spirals or spin-density waves;
• several symmetry-related wavevectors for multi-$\mathbf Q$ order;
• coupled spin, orbital, charge, or structural order parameters.

“Néel order” is most precise for a simple staggered collinear pattern. “Antiferromagnetic order” includes the wider family whose leading uniform moment is absent or secondary.

Exchange Selects the Ordering Wavevector

Use the exchange convention

$$H_{\mathrm{ex}} = \sum_{i<j} J_{ij} \mathbf s_i\cdot\mathbf s_j,$$

so $J_{ij}>0$ favors antiparallel bond correlations. For a Bravais lattice with one spin per chemical cell, define

$$J(\mathbf q) = \sum_j J_{ij} e^{i\mathbf q\cdot(\mathbf r_j-\mathbf r_i)}.$$

At the classical or molecular-field level, the preferred wavevector minimizes $J(\mathbf q)$. With a multi-site basis, replace this scalar by an exchange matrix and examine its eigenvectors as well as eigenvalues. Fixed spin lengths can impose further constraints beyond minimizing one eigenvalue. Competing couplings can move the minimum or create several equally favored modes.

A positive nearest-neighbor $J$ therefore establishes an antiferromagnetic bond tendency, not a theorem that the material has Néel order. Dimensionality, frustration, quantum fluctuations, anisotropy, disorder, itinerancy, and competing phases still decide the thermodynamic state.

Mean-field Néel temperature

For equal spins $S$, let $\lambda_{\min}$ be the lowest eigenvalue of the exchange matrix over wavevector (or the minimum of scalar $J(\mathbf q)$ for a one-site basis). Molecular-field theory gives

$$k_{\mathrm B} T_N^{\mathrm{MF}} = - \frac{ S(S+1) }{3} \lambda_{\min}.$$

For nearest-neighbor exchange $J>0$ on a regular bipartite lattice with coordination $z$, the uniform and staggered eigenvalues are $+zJ$ and $-zJ$. For a one-site chemical Bravais basis these are

$$J(\mathbf0) = zJ, \qquad J(\mathbf Q_{\mathrm{AF}}) = -zJ,$$

and hence

$$k_{\mathrm B} T_N^{\mathrm{MF}} = \frac{ zJS(S+1) }{3}.$$

The high-temperature uniform susceptibility has Curie–Weiss form,

$$\chi = \frac{C}{ T-\Theta_{\mathrm{CW}} },$$

with

$$k_{\mathrm B} \Theta_{\mathrm{CW}} = - \frac{ S(S+1) }{3} J(\mathbf0).$$

Thus the simple nearest-neighbor mean-field model gives

$$\Theta_{\mathrm{CW}} = -T_N^{\mathrm{MF}}.$$

For honeycomb Néel order, the two-site exchange matrix at $\mathbf q=0$ is

$$\mathsf J(0)=3J\begin{pmatrix}0&1\\1&0\end{pmatrix}.$$

Its uniform eigenvector $(1,1)$ has eigenvalue $+3J$, while its staggered vector $(1,-1)$ has eigenvalue $-3J$. Thus the same zero propagation vector contains both channels; the basis eigenvector distinguishes them.

In a real material, $T_N$ and $|\Theta_{\mathrm{CW}}|$ need not agree. Short-range correlations, reduced dimensionality, frustration, itinerant fluctuations, disorder, and multiple exchanges can suppress or reshape the transition. A negative Weiss temperature is evidence for dominant antiferromagnetic correlations, not proof of a particular ordered structure.

Field Response, Canting, and Spin Flop

Compensation does not mean that an antiferromagnet cannot respond to a uniform field. The field cants the sublattice moments and produces a small $\mathbf M$ while the much larger staggered component remains.

A useful long-wavelength free-energy density is

$$f[\mathbf n,\mathbf M] ={} \frac{M^2}{ 2\chi_\perp } + \frac{\rho_s}{2} \left| \boldsymbol\nabla\mathbf n \right|^2 + \frac{K}{2} \left( 1-n_z^2 \right) - \mathbf h\cdot\mathbf M,$$

subject to

$$\mathbf n^2 = 1, \qquad \mathbf n\cdot\mathbf M = 0.$$

Here $\mathbf M$ is the physical bulk magnetization in $\mathrm{A/m}$ and $\mathbf h$ is its conjugate source in tesla, so $-\mathbf h\cdot\mathbf M$ has units of energy density. The coefficient $\chi_\perp=\partial M_\perp/\partial h_\perp$ has units $\mathrm{A/(m\,T)}$; the dimensionless SI susceptibility with respect to internal $H$ is $\chi_{H,\perp}=\mu_0\chi_\perp$. The stiffness $\rho_s$ has units $\mathrm{J/m}$, and $K>0$ is an easy-axis anisotropy density in $\mathrm{J/m^3}$. This expansion assumes small canting, $M\ll M_0$, and neglects longitudinal moment fluctuations.

Minimizing over $\mathbf M$ gives

$$\mathbf M = \chi_\perp \left[ \mathbf h - \mathbf n \left( \mathbf n\cdot\mathbf h \right) \right].$$

The field gains energy most efficiently when $\mathbf n\perp\mathbf h$. For $\mathbf h=h\widehat{\mathbf z}$, compare:

$$f_{\parallel} = 0$$

for $\mathbf n\parallel\widehat{\mathbf z}$, and

$$f_{\perp} = \frac{K}{2} - \frac{ \chi_\perp h^2 }{2}$$

for $\mathbf n\perp\widehat{\mathbf z}$. In this minimal model, the ideal spin-flop scale is

$$h_{\mathrm{sf}} = \sqrt{ \frac{K}{ \chi_\perp } }.$$

The corresponding internal field is $H_{\mathrm{sf}}=h_{\mathrm{sf}}/\mu_0$. Exchange-field notation often writes the same scale as $H_{\mathrm{sf}}\simeq\sqrt{2H_EH_A}$ when the anisotropy field $H_A$ is much smaller than the exchange field $H_E$. Factors depend on how those fields, sublattice moments, and SI units are defined.

The minimal formula omits longitudinal susceptibility, higher anisotropies, domains, magnetoelastic coupling, and the eventual saturation transition. Real spin-flop transitions can be rounded, split, hysteretic, or replaced by continuous rotation.

The same alternating motif has different propagation-vector descriptions when the chemical primitive cell contains one or two sites. Dashed boxes mark chemical cells; the magnetic period doubles in the first case but not the second. The bottom panel shows the ideal transverse-canting spin flop: the Néel vector turns perpendicular to the field while both sublattices acquire the same small field-parallel component.

Weak ferromagnetism from antisymmetric exchange

The Dzyaloshinskii–Moriya interaction

$$H_{\mathrm{DM}} = \sum_{i<j} \mathbf D_{ij}\cdot \left( \mathbf s_i\times\mathbf s_j \right)$$

can cant otherwise antiparallel moments. At the schematic two-sublattice level,

$$\mathbf M \propto \frac{ \mathbf D\times\mathbf L }{J}.$$

The proportionality and sign depend on bond geometry and normalization. Such a canted antiferromagnet has primary staggered order and a secondary weak uniform moment. It is not a ferrimagnet, whose net moment comes from unequal opposing sublattice moments. Magnetic Anisotropy explains how this coupling competes with crystal, shape, and texture energies.

Antiferromagnetic Domains

An antiferromagnet can have domains even when its external stray field is tiny. Distinct domain labels include:

• orientation domains, with different easy-axis or easy-plane directions;
• antiphase domains, related by $\mathbf n\to-\mathbf n$;
• wavevector domains, choosing different symmetry-related $\mathbf Q$ vectors;
• chirality domains in noncollinear structures;
• coupled structural, ferroelectric, or magnetoelastic domains.

Because a compensated bulk antiferromagnet has little magnetostatic energy, domain selection is often governed by crystalline anisotropy, magnetoelastic strain, defects, surfaces, exchange bias, and cooling history rather than the flux-closure balance familiar from ferromagnets.

Surfaces and interfaces can be uncompensated even when the bulk is compensated. A measured interfacial moment, exchange-bias shift, or local stray field therefore does not automatically represent the bulk uniform magnetization.

Spin Waves

Antiferromagnetic spin waves are collective oscillations of the coupled sublattices. For the nearest-neighbor bipartite Heisenberg model,

$$H = J \sum_{\langle i,j\rangle} \mathbf s_i\cdot\mathbf s_j, \qquad J>0,$$

linear spin-wave theory gives

$$\hbar\omega_{\mathbf k} = zJS \sqrt{ 1- \left| \gamma_{\mathbf k} \right|^2 },$$

where

$$\gamma_{\mathbf k} = \frac{1}{z} \sum_{\boldsymbol\delta} e^{i\mathbf k\cdot\boldsymbol\delta}$$

and $\boldsymbol\delta$ runs over nearest-neighbor vectors.

For the square lattice,

$$\gamma_{\mathbf k} = \frac12 \left[ \cos(k_xa) + \cos(k_ya) \right].$$

Near a gapless point, the dispersion is linear:

$$\hbar\omega \simeq \hbar c |\mathbf q|.$$

At the linear-spin-wave level for the square lattice,

$$c = \frac{ 2\sqrt2\,JSa }{\hbar}.$$

Momentum can be described in either the chemical Brillouin zone or the reduced magnetic Brillouin zone. Apparent duplication of gapless points must be interpreted with that folding convention.

Two type-A modes

An isotropic collinear antiferromagnet breaks

$$SO(3) \longrightarrow SO(2).$$

There are two broken spin generators. Unlike a ferromagnet, their commutator has no uniform magnetization expectation value in the compensated state. The two transverse coordinates therefore produce two type-A Goldstone polarizations with linear dispersion. Anisotropy, field, dipolar coupling, spin–orbit coupling, and intersublattice asymmetry can split or gap them.

Zero-point reduction

The classical Néel product state is not an eigenstate of an isotropic antiferromagnetic bond because

$$\mathbf s_i\cdot\mathbf s_j = s_i^zs_j^z + \frac12 \left( s_i^+s_j^- + s_i^-s_j^+ \right).$$

The transverse terms create correlated spin deviations. After the sublattice rotation and Bogoliubov transformation, even the magnon vacuum contains zero-point fluctuations in the original local-spin variables.

For the square-lattice nearest-neighbor model, use $N$ sites and sum over all $N$ wavevectors in the chemical Brillouin zone. Linear spin-wave theory gives

$$\delta S = \frac{1}{2N} \sum_{\mathbf k} \left[ \frac{1}{ \sqrt{ 1- |\gamma_{\mathbf k}|^2 } } - 1 \right],$$

so

$$m_s \simeq S-\delta S.$$

The exact Goldstone points represent collective rotations and require separate finite-size treatment; the quoted reduction is the thermodynamic momentum integral, not a finite sum evaluated at a zero denominator.

For the square-lattice spin-$1/2$ model, $\delta S\simeq0.197$ in linear spin-wave theory, giving $m_s\simeq0.303$. Sandvik’s quantum Monte Carlo calculation gives an ordered spin of about $0.307$ per site. The order survives at zero temperature, but it is strongly reduced from the classical value $1/2$.

Magnons develops the Holstein–Primakoff construction, Bogoliubov modes, neutron matrix elements, interactions, and breakdown tests in detail.

Frustration

Antiferromagnetic interactions become frustrated when their preferred pairwise correlations cannot all be satisfied simultaneously. The elementary geometric example is a triangle with equal $J>0$:

$$E_\triangle = J \left( \mathbf S_1\cdot\mathbf S_2 + \mathbf S_2\cdot\mathbf S_3 + \mathbf S_3\cdot\mathbf S_1 \right).$$

For dimensionless classical vectors of equal length $S$,

$$E_\triangle = \frac{J}{2} \left[ \left( \mathbf S_1+\mathbf S_2+\mathbf S_3 \right)^2 - 3S^2 \right].$$

The minimum satisfies

$$\mathbf S_1+\mathbf S_2+\mathbf S_3 = \mathbf0,$$

which permits a coplanar $120^\circ$ structure rather than a two-sublattice Néel pattern.

Frustration can also be competitive rather than geometric. In the square-lattice $J_1$–$J_2$ model, antiferromagnetic nearest- and next-nearest-neighbor exchanges favor incompatible ordering wavevectors.

Frustration can:

• reduce $T_N$ and the ordered moment;
• create many nearly degenerate classical states;
• select noncollinear or incommensurate order;
• make fluctuations choose among classically degenerate states through order by disorder;
• stabilize valence-bond order, multipolar order, or quantum-disordered phases in suitable regimes.

It does not automatically destroy magnetic order. The triangular-lattice antiferromagnet is frustrated yet can develop $120^\circ$ order.

The empirical ratio

$$f = \frac{ |\Theta_{\mathrm{CW}}| }{T_N}$$

is sometimes called a frustration parameter. A large value signals that ordering occurs far below the dominant interaction scale, but reduced dimensionality, disorder, weak interlayer coupling, and competing nonmagnetic phases can produce the same separation. There is no universal threshold that proves a spin liquid.

Quantum Fluctuations and Dimensionality

Quantum fluctuations are enhanced by:

• smaller spin $S$;
• lower spatial dimension;
• lower coordination number;
• geometric or exchange frustration;
• proximity to a competing singlet or itinerant phase.

Their consequences depend on the model.

One dimension

The spin-$1/2$ isotropic antiferromagnetic Heisenberg chain has no Néel long-range order, even at zero temperature. Its ground state is a singlet with algebraically decaying staggered correlations, and its elementary continuum is described by spinons rather than conventional sharp magnons.

Integer-spin chains can instead have a Haldane gap and exponentially decaying correlations. Thus “antiferromagnetic $J$” does not imply Néel order in one dimension.

Two dimensions

The spin-$1/2$ nearest-neighbor square-lattice Heisenberg antiferromagnet has Néel order at $T=0$, with a reduced ordered moment. At every $T>0$, however, the ideal isotropic short-range model lacks true long-range order by the Mermin–Wagner theorem.

Its low-temperature correlation length can become exponentially large. Weak interlayer exchange, easy-axis anisotropy, or other symmetry-breaking interactions can then convert a nearly two-dimensional correlated regime into a finite-$T_N$ transition in a real crystal.

Three dimensions and coupled layers

Three-dimensional short-range Heisenberg antiferromagnets can order at nonzero temperature. Quasi-two-dimensional compounds often have an in-plane exchange scale far above $T_N$; the small interlayer or anisotropy scale decides when three-dimensional coherence finally appears.

Strong antiferromagnetic coupling does not always favor order. In a bilayer, sufficiently strong interlayer exchange binds spins into singlets and destroys Néel order through a zero-temperature quantum phase transition. The relevant question is which correlations the coupling strengthens.

Finite-size singlets and the tower of states

An even finite lattice with exact spin-rotation symmetry often has a singlet ground state. In an ordered phase, a sequence of collective low-spin levels collapses toward the ground state as system size grows. This “tower of states,” extensive structure-factor scaling, and a finite spin stiffness distinguish incipient symmetry breaking from an isolated singlet.

Localized and Itinerant Antiferromagnets

Localized-moment limit

In a localized description, magnetic moments persist on atomic or molecular sites while exchange organizes their orientations:

$$H_{\mathrm{spin}} = \sum_{i<j} J_{ij} \mathbf s_i\cdot\mathbf s_j + H_{\mathrm{aniso}} + H_{\mathrm{DM}}.$$

Useful evidence includes local spectral weight above $T_N$, effective moments associated with atomic multiplets, magnetic entropy over a local-state manifold, and exchange-scale collective modes. A localized-moment antiferromagnet can be insulating or metallic.

Itinerant spin-density-wave limit

In an itinerant antiferromagnet, the ordered state mixes electronic states separated by $\mathbf Q$. A collinear spin-density-wave mean field can be represented schematically by

$$\boldsymbol\Delta_{\mathbf Q} \propto \frac{I}{N} \sum_{\mathbf k} \left\langle c^\dagger_{\mathbf k+\mathbf Q,\alpha} \boldsymbol\sigma_{\alpha\beta} c_{\mathbf k,\beta} \right\rangle.$$

For one spin sector of a commensurate one-orbital model with $2\mathbf Q$ a reciprocal-lattice vector, the folded Hamiltonian is

$$\mathcal H_{\mathbf k} = \begin{pmatrix} \varepsilon_{\mathbf k} & \Delta_{\mathbf Q} \\ \Delta_{\mathbf Q}^* & \varepsilon_{\mathbf k+\mathbf Q} \end{pmatrix},$$

with eigenvalues

$$E_\pm(\mathbf k) ={} \frac{ \varepsilon_{\mathbf k} + \varepsilon_{\mathbf k+\mathbf Q} }{2} \pm \sqrt{ \left[ \frac{ \varepsilon_{\mathbf k} - \varepsilon_{\mathbf k+\mathbf Q} }{2} \right]^2 + |\Delta_{\mathbf Q}|^2 }.$$

This two-state block opens avoided crossings where the original bands meet. For a generic incommensurate modulation, repeated shifts by $\mathbf Q$ generate a larger set of coupled states; a two-state block then needs an explicit nested-patch or other truncation approximation. A full insulating gap requires suitable filling and reconstructed band geometry. Antiferromagnetic metals are therefore ordinary possibilities.

Localized and itinerant language marks limiting descriptions, not a universal binary. Correlated metals can retain sizable local moments above $T_N$ while itinerant quasiparticles reconstruct at the transition. A trustworthy model must reproduce static moments, charge response, spectra, and temperature scales together.

The Néel Transition

The Néel temperature $T_N$ is the temperature below which thermodynamic antiferromagnetic long-range order appears. Depending on symmetry and coupling, the transition can be continuous, first order, split into several transitions, or obscured by structural order.

The uniform susceptibility need not diverge at $T_N$. It often shows a cusp, change of slope, or broad maximum because the critical field is staggered rather than uniform. The staggered susceptibility

$$\chi_s(\mathbf Q_{\mathrm{AF}}) = \left. \frac{ \partial m_s }{ \partial h_s } \right|_{h_s=0}$$

is the directly conjugate critical response.

Critical behavior depends on:

• dimensionality;
• whether the order parameter is Ising-like, XY-like, or Heisenberg-like;
• interaction range;
• coupling to strain, electrons, and other order parameters;
• quenched disorder;
• whether the transition is actually continuous.

Mean-field square-root behavior is a baseline, not an assumed material law. Finite-Temperature Phase Transitions develops the thermodynamic and scaling tests.

Experimental Identification

No single measurement establishes every aspect of antiferromagnetism. Reliable identification combines structure, thermodynamics, local fields, and dynamics.

Diffraction determines propagation vectors and basis patterns. Magnetic form factors, polarization selection, domain populations, and overlap with nuclear peaks determine which reflections are visible. Element-selective resonant x-ray scattering can help in thin films, but structural and orbital resonances must be separated from magnetic contributions.

Heat capacity and local-field probes provide complementary tests. A static local field need not identify the propagation vector; a susceptibility cusp alone can also arise from freezing, a low-dimensional crossover, or impurities. Magnetic Susceptibility develops those distinctions for bulk measurements.

Inelastic scattering tests mode energies, polarizations, and spectral weight. Linearly dispersing branches near the ordering channel support a conventional ordered reference state; continua or strongly damped modes may require a more complete description. Antiferromagnetic resonance gaps depend on anisotropy, field, and selection rules, so a terahertz frequency is not universal.

Magnetic imaging and transport probe particular symmetries. A zero net moment does not restore time reversal, and some antiferromagnets permit an anomalous Hall effect. Neither an image nor a resistance change determines the full magnetic structure without identifying its coupling to the order.

Antiferromagnetic Spintronics

Small stray fields and fast exchange-driven dynamics make antiferromagnets interesting for control, but writing and reading depend on symmetry, interfaces, damping, and domains. A particularly useful source has opposite sign on the sublattices:

$$\mathbf h_A^{\mathrm{so}}=-\mathbf h_B^{\mathrm{so}}.$$

It couples directly to $\mathbf L$. Železný and collaborators derive such current-induced Néel fields for appropriate local inversion asymmetry; they are not present in every antiferromagnet. Spin torque, strain, magnetoelectric coupling, and optical excitation provide other possibilities. Spintronics develops their mechanisms and limitations.

Reading the state

A simple anisotropic magnetoresistance form is

$$\rho(\mathbf n) = \rho_\perp + \Delta\rho \left( \mathbf n\cdot\widehat{\mathbf j} \right)^2.$$

It is even under $\mathbf n\to-\mathbf n$, so it distinguishes axes or domain populations but not the two antiphase states by itself. Other electrical or optical readouts can be odd in selected order-parameter components only when symmetry allows.

Current pulses also heat, strain, and modify a device. A resistance change is therefore convincing as a magnetic switch only when its symmetry, thermal behavior, and independent magnetic readout support that interpretation. Fast writing in a selected experiment does not by itself establish a memory clock rate or scalable readout.

Distinguishing Nearby States

State	Leading order	Uniform moment	Key distinction
Collinear Néel antiferromagnet	Staggered basis pattern, possibly at $\mathbf Q=0$	Zero ideally	Equivalent sublattices oppose
Canted antiferromagnet	Staggered $\mathbf L$	Small secondary $\mathbf M$	Primary order remains antiferromagnetic
Ferrimagnet	Opposing unequal sublattices	Usually nonzero	Net moment is intrinsic sublattice imbalance
Spin-density wave	Itinerant finite-$\mathbf Q$ polarization	Often zero	Band folding and amplitude may be itinerant
Spin glass	Random frozen moments	History dependent	No periodic magnetic Bragg order
Quantum paramagnet	No static order	Zero	Singlets or another symmetric state persist at $T=0$
Correlated paramagnet	Short-range AF correlations	Field induced	$S(\mathbf Q)$ is not extensive

Zero net magnetization is common to several rows. The ordering wavevector, scaling, dynamics, and symmetry decide the phase.

Exercises

1. Extensive antiferromagnetic structure factor

For a one-dimensional ideal Néel pattern

$$\left\langle \mathbf s_j \right\rangle = m_s(-1)^j\mathbf n$$

on an even periodic chain, evaluate the ordered contribution to $S(q)$ and show that it is extensive at $q=\pi/a$.

Solution

Factorizing the long-range ordered contribution gives

$$\begin{aligned} S_{\mathrm{ord}}(q) &= \frac{m_s^2}{N} \sum_{j,\ell} e^{iq a(j-\ell)} (-1)^{j-\ell} \\ &= \frac{m_s^2}{N} \left| \sum_j e^{i(q a+\pi)j} \right|^2. \end{aligned}$$

At $q=\pi/a$ modulo $2\pi/a$, every term has the same phase, so the sum has magnitude $N$. Therefore

$$S_{\mathrm{ord}} \left( \frac{\pi}{a} \right) = Nm_s^2.$$

At other allowed finite-size momenta, the perfect-pattern contribution vanishes. Fluctuations broaden or reduce the peak but do not change the extensive scaling criterion for long-range order.

2. Mean-field Néel and Weiss temperatures

A nearest-neighbor simple-cubic spin-$1/2$ antiferromagnet has coordination $z=6$ and $J=4\,\mathrm{meV}$ in the convention

$$H = J \sum_{\langle i,j\rangle} \mathbf s_i\cdot\mathbf s_j.$$

Estimate $T_N^{\mathrm{MF}}$ and $\Theta_{\mathrm{CW}}$.

Solution

For a bipartite nearest-neighbor model,

$$J(\mathbf Q_{\mathrm{AF}}) = -zJ = -24\,\mathrm{meV},$$

while $J(\mathbf0)=24\,\mathrm{meV}$. Since $S(S+1)=3/4$,

$$\begin{aligned} k_{\mathrm B}T_N^{\mathrm{MF}} &= - \frac{3/4}{3} \left( -24\,\mathrm{meV} \right) \\ &= 6\,\mathrm{meV}. \end{aligned}$$

Using $k_{\mathrm B}=0.08617\,\mathrm{meV/K}$,

$$T_N^{\mathrm{MF}} \simeq 69.6\,\mathrm K.$$

The uniform Curie–Weiss temperature is

$$\Theta_{\mathrm{CW}} = -69.6\,\mathrm K.$$

Fluctuations can reduce the measured $T_N$ without changing the high-temperature exchange sum in the same way.

3. Spin-flop scale

Starting from

$$f = \frac{M^2}{2\chi_\perp} + \frac{K}{2}(1-n_z^2) - hM_z$$

with $\mathbf M\cdot\mathbf n=0$, compare an easy-axis state $\mathbf n\parallel\widehat{\mathbf z}$ with a flopped state $\mathbf n\perp\widehat{\mathbf z}$.

Solution

For $\mathbf n\parallel\widehat{\mathbf z}$, the constraint forbids $M_z$ in this fixed-length transverse model. The minimum has $\mathbf M=\mathbf0$ and

$$f_\parallel = 0.$$

For $\mathbf n\perp\widehat{\mathbf z}$, $M_z$ is allowed. Minimization gives

$$M_z = \chi_\perp h.$$

Substitution yields

$$f_\perp = \frac{K}{2} - \frac{ \chi_\perp h^2 }{2}.$$

The two are equal at

$$h_{\mathrm{sf}} = \sqrt{ \frac{K}{\chi_\perp} }.$$

The derivation also shows the approximations: fixed sublattice lengths, one easy-axis constant, no domains, and no longitudinal susceptibility.

4. When an itinerant antiferromagnet is insulating

Assume perfect nesting at half filling,

$$\varepsilon_{\mathbf k+\mathbf Q} = -\varepsilon_{\mathbf k},$$

and a real spin-density-wave amplitude $\Delta$. Diagonalize the folded $2\times2$ Hamiltonian and determine the gap.

Solution

The Hamiltonian is

$$\mathcal H_{\mathbf k} = \begin{pmatrix} \varepsilon_{\mathbf k} & \Delta \\ \Delta & -\varepsilon_{\mathbf k} \end{pmatrix}.$$

Its characteristic equation is

$$E^2 - \varepsilon_{\mathbf k}^2 - \Delta^2 = 0,$$

so

$$E_\pm(\mathbf k) = \pm \sqrt{ \varepsilon_{\mathbf k}^2+\Delta^2 }.$$

At half filling the occupied and empty bands are separated by a direct gap

$$E_g = 2|\Delta|.$$

Without perfect nesting or suitable filling, reconstructed pockets can survive. Finite-$\mathbf Q$ magnetic order alone does not guarantee an insulator.

Ferrimagnetism treats unequal opposing sublattices. Quantum Spin Liquids and Glasses and Spin Glasses explain why strong antiferromagnetic correlations need not produce Néel order. Itinerant Magnetism develops collective modes and reconstruction in conducting systems.

References

• A. Auerbach, Interacting Electrons and Quantum Magnetism (Springer, 1994), doi:10.1007/978-1-4612-0869-3.

• S. Chakravarty, B. I. Halperin, and D. R. Nelson, “Two-Dimensional Quantum Heisenberg Antiferromagnet at Low Temperatures,” Physical Review B 39, 2344–2371 (1989), doi:10.1103/PhysRevB.39.2344.

• F. D. M. Haldane, “Nonlinear Field Theory of Large-Spin Heisenberg Antiferromagnets,” Physical Review Letters 50, 1153–1156 (1983), doi:10.1103/PhysRevLett.50.1153.

• R. Kubo, “The Spin-Wave Theory of Antiferromagnetics,” Physical Review 87, 568–580 (1952), doi:10.1103/PhysRev.87.568.

• N. D. Mermin and H. Wagner, “Absence of Ferromagnetism or Antiferromagnetism in One- or Two-Dimensional Isotropic Heisenberg Models,” Physical Review Letters 17, 1133–1136 (1966), doi:10.1103/PhysRevLett.17.1133.

• A. W. Sandvik, “Finite-Size Scaling of the Ground-State Parameters of the Two-Dimensional Heisenberg Model,” Physical Review B 56, 11678–11690 (1997), doi:10.1103/PhysRevB.56.11678.

• C. G. Shull and J. S. Smart, “Detection of Antiferromagnetism by Neutron Diffraction,” Physical Review 76, 1256–1257 (1949), doi:10.1103/PhysRev.76.1256.2.

• J. Železný et al., “Relativistic Néel-Order Fields Induced by Electrical Current in Antiferromagnets,” Physical Review Letters 113, 157201 (2014), doi:10.1103/PhysRevLett.113.157201.