Symmetry Classification of Order Parameters

Ask what actually changes when water freezes, a magnet cools below its Curie point, or a metal turns superconducting, and the honest answer is not "the density" or "the magnetization" but the symmetry. Landau's great insight was that a phase of matter is labelled by which symmetries it keeps and which it loses. A microscopic Hamiltonian is invariant under some group $G$ — spatial rotations, translations, internal spin rotations, or a gauge transformation. At high temperature the equilibrium state samples all of $G$ and looks symmetric; below a transition it settles into a state invariant only under a subgroup $H\subset G$ . The transition is the reduction $G\to H$ . To make this quantitative we need a variable that can see the difference. That variable is the order parameter $\phi$ : zero in the symmetric phase, nonzero in the ordered one. Crucially $\phi$ cannot be a scalar invariant — an invariant takes the same value on both sides and so is blind to the transition. Instead $\phi$ must transform under a nontrivial representation $\rho$ of $G$ . The triple $(G,H,\rho)$ is the symmetry data of the phase, and it does an astonishing amount of work: $\rho$ fixes the number of components of $\phi$ , which terms are allowed in the Landau free energy, and therefore whether the transition can even be continuous; the coset space $\mathcal{M}=G/H$ counts the gapless Goldstone modes and, through its topology, catalogues the domain walls, vortices, and monopoles the ordered phase can support. This is why the classification is the skeleton of the whole subject. Two systems as unalike as a uniaxial ferromagnet and the liquid–gas critical point share the $\mathbb{Z}_2$ (Ising) data, so the renormalization group hands them identical critical exponents. Microscopic detail is an irrelevant operator; the symmetry data survive the flow to the fixed point. Learn to read $(G,H,\rho)$ off a phase and you can predict its Goldstone spectrum, its defect zoo, whether it orders continuously, and which universality class it belongs to — before writing down a single microscopic coupling.

Key takeaways

A phase is classified not by appearance but by the symmetry group $G$ of the disordered phase and the subgroup $H$ left unbroken below the transition, encoded in an order parameter $\phi$ that transforms under a specific representation $\rho$ of $G$ .
The order parameter lives on the coset manifold $\mathcal{M}=G/H$ , whose dimension counts Goldstone modes and whose homotopy groups $\pi_k(G/H)$ classify topological defects (walls, vortices, monopoles).
The representation $\rho$ dictates which invariants may appear in the Landau free energy; the Landau criterion says a continuous transition is possible only when $\rho$ admits no cubic invariant.
A one-component (scalar) order parameter realizes $\mathbb{Z}_2$ (Ising); a complex scalar realizes $U(1)\cong O(2)$ (XY); an $n$ -component vector realizes $O(n)$ (Heisenberg for $n=3$ ); a headless director realizes the nematic $\mathbb{RP}^2$ .
Discrete broken symmetries give no Goldstone modes but allow domain walls; continuous broken symmetries give $\dim(G/H)$ Goldstone bosons plus defects from $\pi_k(G/H)$ .
Key idea: a phase carries the data $(G,H,\rho)$ — the disordered group $G$ , the unbroken subgroup $H$ , and the representation $\rho$ under which $\phi$ transforms; $\mathcal{M}=G/H$ fixes Goldstone modes and defects, while $\rho$ fixes the allowed Landau invariants and hence the universality class.

The core insight

A phase is labelled by the data $(G,H,\rho)$ : the symmetry group $G$ of the disordered phase, the subgroup $H$ left unbroken by the ordered phase, and the irreducible representation $\rho$ of $G$ under which the order parameter transforms. The manifold $\mathcal{M}=G/H$ and its topology determine the Goldstone modes and topological defects, while $\rho$ fixes the symmetry-allowed Landau invariants — and therefore whether the transition can be continuous and which universality class it joins.

Where this lives today

Symmetry classification is the backbone of modern condensed matter, and it has grown rather than shrunk. The renormalization-group verdict that critical exponents depend only on $(d,n,\rho)$ is now checked to spectacular precision: the conformal bootstrap pins the 3D Ising exponents ( $\nu\approx0.6300$ , $\gamma\approx1.2372$ ) to many digits, and the same numbers are measured at the liquid–gas critical point of simple fluids and at binary-fluid demixing — different microscopics, identical symmetry data, identical exponents. Beyond conventional order, the Landau paradigm has been extended and, in places, transcended: topological phases (the quantum Hall states, topological insulators, spin liquids) are precisely the phases not labelled by a local symmetry-breaking order parameter, and classifying them required new tools (symmetry-protected topological order, entanglement invariants). Even there, the symmetry group $G$ remains the organizing input — now acting on the entanglement structure rather than on a local order parameter. Reading $(G,H,\rho)$ off a system is still the first move a condensed-matter theorist makes.

The physical picture

Start with a hot paramagnet. Its spins point in every direction and the macroscopic state has the full rotational symmetry $O(3)$ of the Hamiltonian. Cool below $T_c$ and the spins line up along some axis $\hat n$ . Nature has not lost the symmetry — the Hamiltonian is still perfectly rotationally invariant — but the realized state has picked one $\hat n$ , breaking $O(3)$ down to $O(2)$ , the rotations about that axis. The distinguishing variable is the magnetization vector $\mathbf M$ , transforming as the vector (spin-1) representation of $O(3)$ . The leftover freedom — which $\hat n$ was chosen — is a point on the sphere $S^2=O(3)/O(2)$ . That sphere is the order-parameter manifold, and almost everything follows from its geometry. Walking slowly across the sphere costs vanishing energy in the long-wavelength limit: those are the two spin-wave (Goldstone) modes of a Heisenberg magnet. The sphere also supports point defects — hedgehogs — precisely because $\pi_2(S^2)=\mathbb{Z}$ assigns each configuration an integer winding. Change the representation and you change the manifold and hence the whole story.

Figure

Symmetry breaking G → H: the order parameter picks a point on the coset manifold M = G/H

Contrast a uniaxial (Ising) magnet, where only $\uparrow$ or $\downarrow$ is allowed. Now $G=\mathbb{Z}_2$ , $H=\{e\}$ , and $\mathcal{M}=\mathbb{Z}_2$ is two discrete points. There is no continuous direction to rotate, hence no Goldstone mode; but the discrete set supports domain walls separating $\uparrow$ from $\downarrow$ regions. The entire difference between a Heisenberg and an Ising magnet is which representation $\phi$ carries — a vector versus a scalar. That single choice reorganizes the spectrum (gapless vs gapped), the defects (points vs walls), and the universality class. The deepest payoff is predictive economy. To know how a system behaves near criticality you do not solve its microscopic dynamics; you read off its symmetry data $(G,H,\rho)$ and its spatial dimension $d$ . Universality is the statement that the messy microscopics are irrelevant operators in the renormalization-group sense, and only the symmetry-protected structure survives. The classification below is therefore less a taxonomy than a lookup table for physics.

Figure

Four symmetry classes: order-parameter representation fixes M = G/H, Goldstone count, defects, and Landau order

The core equations

The order parameter is defined by what it does at the transition: it must vanish above and be nonzero below, so that its value alone certifies which phase you are in.

\langle \phi \rangle = 0 \;\;(T>T_c), \qquad \langle \phi \rangle \neq 0 \;\;(T<T_c) \qquad (1)

where: $\phi$ is the order parameter; $\langle\cdot\rangle$ the thermal average; $T_c$ the transition temperature. Physically, $\langle\phi\rangle$ is the "flag" that reports symmetry breaking — but for it to distinguish two phases it must itself change under the symmetry, which is the content of the next equation. For that flag to work, $\phi$ must carry a nontrivial representation $\rho$ of $G$ : the group acts on its components by a matrix $\rho(g)$ , not by leaving it fixed.

\rho:\;G \longrightarrow GL(V),\qquad \phi \in V,\qquad \phi \mapsto \rho(g)\,\phi \qquad (2)

where: $\rho$ is the representation; $V$ the vector space its $n$ components span; $g\in G$ a symmetry element. A trivial $\rho$ would make $\phi$ invariant and blind to the transition; a generic continuous transition uses an irreducible $\rho$ so that a single quadratic invariant $|\phi|^2$ tunes it. The symmetry that survives is exactly the set of transformations that fix the chosen ordered value — the stabilizer of $\langle\phi\rangle$ .

H=\{\,g\in G:\;\rho(g)\,\langle\phi\rangle=\langle\phi\rangle\,\} \qquad (3)

where: $H\subseteq G$ is the unbroken subgroup (the residual symmetry of the ordered state). Reading it physically: everything in $G$ that moves $\langle\phi\rangle$ is broken; everything that fixes it is kept. The physically distinct ordered states are the images of one state under all of $G$ , i.e. the coset space, whose dimension is the number of independent flat directions.

\mathcal{M}=G/H,\qquad \dim \mathcal{M}=\dim G-\dim H \qquad (4)

where: $\mathcal{M}$ is the order-parameter (vacuum) manifold. For discrete $G$ it is a set of points; for continuous $G$ it is a smooth manifold whose dimension counts Goldstone modes and whose homotopy classifies defects. For a broken continuous symmetry, each flat direction of $\mathcal{M}$ is a gapless mode, so the Goldstone count is simply the dimension of the coset.

N_{\text{Goldstone}} = \dim G - \dim H = \dim(G/H)\quad(\text{type-A}) \qquad (5)

where: $N_{\text{Goldstone}}$ is the number of gapless bosons for Lorentz-invariant or "type-A" systems. Ferromagnets are the famous exception: their broken generators pair under a nonzero conserved charge into "type-B" modes with quadratic dispersion, halving the count (two broken generators, one magnon branch). The Landau free energy may contain only $G$ -invariant combinations of $\phi$ ; symmetry, not convenience, dictates its form.

F[\phi]=\int \mathrm{d}^dx\Big[\tfrac{1}{2}(\nabla\phi)^2 + r\,|\phi|^2 + u\,(|\phi|^2)^2 + \cdots\Big] \qquad (6)

where: $r\propto(T-T_c)$ is the quadratic coefficient that changes sign at the transition; $u>0$ stabilizes the ordered state; $d$ is the spatial dimension. The gradient term penalizes spatial variation of $\phi$ ; the polynomial terms are the only ones allowed because every term must be invariant under $\rho(g)$ . The single most consequential invariant is a possible cubic term — if the representation admits one, the transition is forced to be first order.

f(\phi)=r\,\phi^2 - w\,\phi^3 + u\,\phi^4,\qquad r_c=\frac{w^2}{4u},\quad \phi_*=\frac{w}{2u} \qquad (7)

where: $w$ is the coefficient of a symmetry-allowed cubic invariant, $r_c>0$ the temperature at which the ordered minimum becomes degenerate with $\phi=0$ , and $\phi_*=w/2u$ the value it jumps to. A nonzero $w$ means the order parameter appears discontinuously: a first-order transition with latent heat. (A common textbook shorthand quotes the midpoint $3w/8u$ of the two extrema; the coexisting minimum is at $\phi_*=w/2u$ — see the cheat sheet.) Finally, the topology of $\mathcal{M}$ classifies the defects the ordered phase can host, via its homotopy groups.

\pi_k(\mathcal{M})=\pi_k(G/H)\;\Longrightarrow\;\text{stable defect of spatial dimension } D-k-1 \qquad (8)

where: $\pi_k$ is the $k$ -th homotopy group; $D$ the spatial dimension of the system. $\pi_0$ nontrivial gives domain walls, $\pi_1$ gives line defects (vortices/disclinations), $\pi_2$ gives point defects (monopoles/hedgehogs).

Derivation AFrom symmetry breaking to Goldstone modes: the coset construction

The goal is to prove rigorously that breaking a continuous symmetry $G\to H$ produces exactly $\dim(G/H)$ gapless modes, one for each broken generator, and to see that these modes parameterize motion along the coset manifold. The strategy is to differentiate the $G$ -invariance of the effective potential twice at the ordered minimum: the first derivative encodes invariance, the second produces the mass-squared (Hessian) matrix, and the broken generators are shown to be its null eigenvectors. This is the cleanest route because it never needs the explicit dynamics — invariance plus the location of the minimum is enough, and it makes transparent why the number of massless modes equals the number of broken generators rather than something representation-dependent.

Step 1. Take a $G$ -invariant effective potential $V(\phi)$ with $V(\rho(g)\phi)=V(\phi)$ for all $g\in G$ , minimized on the orbit of $\langle\phi\rangle$ . Why: the coarse-grained free energy inherits the full symmetry of the microscopic Hamiltonian; only the chosen state, not the potential, is less symmetric.

Step 2. Expand $g=1+\epsilon_a T^a$ with generators $T^a$ and demand invariance to first order: $\dfrac{\partial V}{\partial \phi_i}\,(T^a)_{ij}\,\phi_j = 0$ for every $a$ . Why: an infinitesimal symmetry moves $\phi$ but cannot change $V$ , so the gradient of $V$ is orthogonal to every symmetry direction.

Step 3. Differentiate again with respect to $\phi_k$ and evaluate at the minimum, where $\partial V/\partial\phi=0$ . The surviving identity is $M_{ik}\,(T^a\langle\phi\rangle)_k=0$ , with $M_{ik}=\partial^2 V/\partial\phi_i\partial\phi_k$ the Hessian (mass-squared matrix). Why: the term with $\partial V/\partial\phi$ drops at the minimum, leaving the Hessian acting on the symmetry direction.

Step 4. Read off the modes. If $T^a\langle\phi\rangle\neq0$ (a broken generator) then $T^a\langle\phi\rangle$ is a null eigenvector of $M$ : a massless mode. If $T^a\langle\phi\rangle=0$ (an unbroken generator, i.e. an element of $H$ ) it contributes nothing. Why: a null direction of the Hessian is a flat direction of the potential — precisely a Goldstone mode.

Step 5. Count: the broken generators number $\dim G-\dim H=\dim(G/H)$ , so there are exactly that many Goldstone bosons, each with $\omega\to0$ as $k\to0$ . For a Heisenberg ferromagnet $O(3)\to O(2)$ gives $\dim(S^2)=2$ broken generators; because they are conjugate under the magnetization density they combine into a single type-B magnon with $\omega\propto k^2$ . Verified: verify_aq-sm-symmetry-classification-order-parameters.py checks $\dim(G/H)$ for $O(3)/O(2)=S^2$ (2), $U(1)=S^1$ (1), and $SU(2)/U(1)$ (2), and confirms the broken-generator/null-eigenvector count on an explicit $O(3)$ vector potential.

Go deeper · Derivation B · 2 minThe Landau criterion: which symmetries permit a continuous transition

Derivation BThe Landau criterion: which symmetries permit a continuous transition

The goal is to derive the condition that a continuous (second-order) transition is possible only when the order-parameter representation admits no cubic invariant, and to apply it to concrete cases (Ising, three-state Potts, crystallization). The strategy is to expand the $G$ -invariant free energy in powers of $\phi$ , note that a cubic invariant exists exactly when the symmetric cube of the representation contains the trivial representation, and then analyze the shape of $f=r\phi^2-w\phi^3+u\phi^4$ to show that any nonzero cubic coefficient forces a discontinuous jump. This matters because it converts a hard dynamical question — is the transition continuous? — into a finite group-theory check on the representation, computable once and for all.

Step 1. Write $f=\sum_n a_n I_n(\phi)$ where the $I_n$ are the independent $G$ -invariant polynomials of degree $n$ . For an irreducible $\rho$ there is a unique quadratic invariant $|\phi|^2$ , whose coefficient $r\propto(T-T_c)$ tunes the transition. Why: only invariants can appear in a symmetric free energy, and the quadratic invariant is the one that softens at $T_c$ .

Step 2. A cubic invariant exists iff $[\mathrm{Sym}^3\rho]\supset\mathbf{1}$ — the symmetric cube of $\rho$ contains the trivial representation. For a $\mathbb{Z}_2$ scalar, $\phi\to-\phi$ sends $\phi^3\to-\phi^3$ , so no cubic invariant exists; for a three-state order parameter (complex $\psi$ with $\psi\to\omega\psi$ , $\omega=e^{2\pi i/3}$ ) the combination $\mathrm{Re}\,\psi^3$ is invariant. Why: the cube of an odd object is odd and cannot be invariant, whereas a $\mathbb{Z}_3$ phase rotates $\psi^3$ by $\omega^3=1$ , leaving it fixed.

Step 3. Suppose $w\neq0$ and minimize $f=r\phi^2-w\phi^3+u\phi^4$ . The nonzero extrema solve $4u\phi^2-3w\phi+2r=0$ ; at $r_c=w^2/4u$ the discriminant is $w^2$ , giving $\phi=w/4u$ (a barrier maximum) and $\phi=w/2u$ (a minimum). Why: a cubic term tilts the quartic well asymmetrically, creating a second minimum separated from $\phi=0$ by a barrier.

Step 4. At $r_c=w^2/4u$ the minimum at $\phi_*=w/2u$ becomes degenerate with $\phi=0$ and then takes over: as $r$ decreases through $r_c$ the equilibrium $\phi$ jumps from $0$ to $\phi_*$ . That discontinuity is a first-order transition with latent heat. Why: two coexisting minima of equal depth mark phase coexistence; below $r_c$ the ordered one wins abruptly.

Step 5. Apply to crystallization: the order parameters are density-wave amplitudes $\rho_{\mathbf G}$ , and any three reciprocal vectors forming a closed triangle $\mathbf G_1+\mathbf G_2+\mathbf G_3=0$ (present for BCC/FCC) allow the cubic invariant $\rho_{\mathbf G_1}\rho_{\mathbf G_2}\rho_{\mathbf G_3}$ . Hence freezing is generically first order, while Ising and $O(n)$ magnets (no cubic invariant) can order continuously. Verified: verify_aq-sm-symmetry-classification-order-parameters.py confirms $\phi^3$ is odd under $\mathbb{Z}_2$ (no cubic invariant), that $\mathrm{Re}\,\psi^3=p_1(p_1^2-3p_2^2)$ is invariant under the $\mathbb{Z}_3$ rotation (cubic invariant present), and that $f=r\phi^2-w\phi^3+u\phi^4$ has coexistence at $r_c=w^2/4u$ , $\phi_*=w/2u$ .

Limiting cases

Each phase corresponds to specific group-theoretic data; the classification is concrete, not abstract.

Case	Result	Meaning
Ising ferromagnet / liquid–gas	$G=\mathbb{Z}_2,\ H=\{e\},\ \mathcal{M}=\{2\ \text{pts}\},\ n=1$	Scalar order parameter (magnetization component or density difference). Discrete: no Goldstone modes, but domain walls from $\pi_0=\mathbb{Z}_2$ . Continuous transition allowed (no cubic invariant).
XY / superfluid $^4$ He	$G=U(1)\cong O(2),\ H=\{e\},\ \mathcal{M}=S^1,\ n=2$	Complex order parameter $\psi=\|\psi\|e^{i\theta}$ . One Goldstone mode (phase / second sound). Vortices from $\pi_1(S^1)=\mathbb{Z}$ .
Heisenberg ferromagnet	$G=O(3),\ H=O(2),\ \mathcal{M}=S^2,\ n=3$	Vector order parameter $\mathbf M$ . Two broken generators; type-B magnon. Hedgehogs from $\pi_2(S^2)=\mathbb{Z}$ ; no stable line defects ( $\pi_1(S^2)=0$ ).
Nematic liquid crystal	$G=O(3),\ H=D_{\infty h},\ \mathcal{M}=\mathbb{RP}^2$	Headless director: tensor order parameter $Q_{ij}$ . Half-integer disclination lines from $\pi_1(\mathbb{RP}^2)=\mathbb{Z}_2$ ; a cubic invariant $\mathrm{Tr}\,Q^3$ makes the isotropic–nematic transition weakly first order.
3-state Potts	$G=S_3,\ \mathcal{M}=\{3\ \text{pts}\}$	Discrete three-fold order parameter with a cubic invariant $\mathrm{Re}\,\psi^3$ : generically first order (continuous only in special low $d$ ).

Worked examples

Example 1 — counting Goldstone modes and defects from the coset, for real magnets. For (a) an XY magnet, (b) a Heisenberg magnet, and (c) a nematic, read off the number of Goldstone modes and the leading topological defect purely from the symmetry data.

Step 1. XY: $G=O(2)$ ( $\dim 1$ ), $H=\{e\}$ ( $\dim 0$ ), so $N_{\text{Goldstone}}=1-0=\mathbf{1}$ and $\mathcal{M}=S^1$ . Since $\pi_1(S^1)=\mathbb{Z}$ , the defects are vortices with integer winding. (One phase mode; quantized vortices — exactly the physics of superfluid films.)

Step 2. Heisenberg: $G=O(3)$ ( $\dim 3$ ), $H=O(2)$ ( $\dim 1$ ), so the broken-generator count is $3-1=\mathbf{2}$ and $\mathcal{M}=S^2$ . Because $\pi_2(S^2)=\mathbb{Z}$ the point defects are hedgehogs; $\pi_1(S^2)=0$ means no stable line defects. (For a ferromagnet the two broken generators fuse into one type-B magnon branch.)

Step 3. Nematic: $G=O(3)$ , $H=D_{\infty h}$ , $\mathcal{M}=\mathbb{RP}^2$ . The broken-generator count is again $3-1=2$ orientational modes, but $\pi_1(\mathbb{RP}^2)=\mathbb{Z}_2$ permits half-integer disclination lines impossible for a vector — the Schlieren textures seen under crossed polarizers. The manifold, fixed entirely by $(G,H)$ , dictates the whole defect zoo.

Example 2 — is the transition continuous? The cubic-invariant test on a three-state clock model. A three-state order parameter is the complex field $\psi=p_1+ip_2$ with the $\mathbb{Z}_3$ symmetry $\psi\to\omega\psi$ , $\omega=e^{2\pi i/3}$ . Decide whether Landau theory permits a continuous transition, and contrast with Ising.

Step 1. Build the candidate cubic invariant. Under $\psi\to\omega\psi$ , $\psi^3\to\omega^3\psi^3=\psi^3$ since $\omega^3=1$ . So $\psi^3$ — and its real part $\mathrm{Re}\,\psi^3=p_1(p_1^2-3p_2^2)$ — is invariant: a genuine cubic invariant exists.

Step 2. Contrast Ising. A $\mathbb{Z}_2$ scalar obeys $\phi\to-\phi$ , so $\phi^3\to-\phi^3\neq\phi^3$ : no cubic invariant, and the leading terms are $r\phi^2+u\phi^4$ . (Even powers only — the hallmark of an Ising order parameter.)

Step 3. Conclude. With a cubic invariant present, the three-state free energy $f=r|\psi|^2 - w\,\mathrm{Re}\,\psi^3 + u|\psi|^4$ has $f=r\phi^2-w\phi^3+u\phi^4$ along the ordering direction, which jumps at $r_c=w^2/4u$ to $\phi_*=w/2u$ : the three-state transition is generically first order, while the Ising transition can be continuous. The whole distinction is a three-line group-theory check, with no microscopic input.

Origins · 1 minsymmetry group and how this was discovered

symmetry group

organizing concept, formalized 1930s–1960s · A symmetry group G is the set of transformations — rotations, translations, internal spin or gauge rotations — that leave a system's Hamiltonian invariant, together with the composition law that makes them a group. Emmy Noether (1918) tied continuous symmetries to conservation laws; Eugene Wigner brought group representation theory into quantum physics in the 1930s, showing that states organize into irreducible representations. Lev Landau made the symmetry group the centrepiece of phase-transition theory in 1937, arguing that a transition is a change of symmetry captured by an order parameter transforming under a representation of G. Jeffrey Goldstone (1961) connected broken continuous symmetries to gapless modes, and the coset construction G/H by Callan, Coleman, Wess and Zumino (1969) made the order-parameter manifold a computational tool. The modern classification of phases — from magnets to liquid crystals to topological matter — is written in the language of these groups and their representations.

History

1918

— Noether links continuous symmetries to conservation laws

1937

— Landau formulates phase transitions as symmetry breaking with an order parameter

1950

— Ginzburg and Landau write the order-parameter free energy for superconductors

1961

— Goldstone identifies gapless modes of a broken continuous symmetry

1966

— Mermin and Wagner forbid continuous-symmetry breaking in two dimensions

1969

— Callan, Coleman, Wess, Zumino build the coset G/H nonlinear realization

1979

— Mermin's review systematizes the homotopy classification of ordered-media defects

Going further

References. Landau & Lifshitz, Statistical Physics Part 1 (3rd ed.), §142–147 · Kardar, Statistical Physics of Fields, Ch. 2–3 · Pathria & Beale, Statistical Mechanics (3rd ed.), Ch. 12 · Sethna, Statistical Mechanics: Entropy, Order Parameters, and Complexity, Ch. 9 · Chaikin & Lubensky, Principles of Condensed Matter Physics, Ch. 2–4, 9 · N. D. Mermin, Rev. Mod. Phys. 51, 591 (1979).

Provenance. The symmetry data and homotopy classifications are standard: $O(3)/O(2)=S^2$ , $U(1)\to S^1$ , nematic $\mathbb{RP}^2$ with $\pi_1=\mathbb{Z}_2$ (Mermin 1979; Chaikin–Lubensky Ch. 9). Goldstone counting and the Landau cubic-invariant criterion follow Landau & Lifshitz §142–147 and Kardar Ch. 2–3. The cubic-term analysis of $f=r\phi^2-w\phi^3+u\phi^4$ gives coexistence at $r_c=w^2/4u$ with the ordered minimum at $\phi_*=w/2u$ (the two nonzero extrema are $w/4u$ and $w/2u$ ; the frequently quoted $3w/8u$ is their midpoint, not the minimum — corrected here). Quoted 3D Ising exponents ( $\nu\approx0.6300$ , $\gamma\approx1.2372$ ) are conformal-bootstrap/RG values (Pelissetto–Vicari, Phys. Rep. 368, 549 (2002)).

Builds on: Order Parameter · Landau Theory of Phase Transitions · Spontaneous Symmetry Breaking · Continuous Phase Transitions. Leads to: Goldstone Modes · Universality and Universality Classes · Critical Exponents · Correlation Function.

What is the difference between the symmetry group G and the unbroken subgroup H?

$G$ is the symmetry group of the disordered (high-temperature) phase — the full set of transformations leaving the Hamiltonian invariant. $H$ is the subgroup that still leaves the ordered state invariant: the stabilizer of the chosen $\langle\phi\rangle$ . A magnet's Hamiltonian is $O(3)$ -symmetric ( $G=O(3)$ ), but once the spins align along $\hat n$ only rotations about $\hat n$ preserve that state, so $H=O(2)$ . The transition is precisely the reduction $G\to H$ , and the pair $(G,H)$ — together with how $\phi$ transforms — is what labels the phase.

How does the order-parameter manifold M = G/H predict Goldstone modes?

The coset $\mathcal{M}=G/H$ is the set of all physically distinct ordered states — the ways the system could have broken the symmetry. Moving along $\mathcal{M}$ takes one ground state to another, and because the free energy is $G$ -invariant this motion costs no potential energy: these are flat directions. Making such a flat direction vary slowly in space costs only gradient energy $\sim(\nabla\theta)^2$ , giving modes whose frequency vanishes as the wavevector goes to zero — Goldstone bosons. Their number equals $\dim(G/H)$ for type-A systems; ferromagnets are the type-B exception, where two broken generators fuse into one quadratically dispersing magnon.

Why does a cubic term in the Landau free energy force a first-order transition?

A cubic term $-w\phi^3$ breaks the symmetry between $\phi$ and $-\phi$ in the free-energy landscape, tilting the quartic well. This creates a second minimum at finite $\phi$ separated from $\phi=0$ by a barrier while the disordered state is still locally stable. As temperature drops, that second minimum deepens until, at $r_c=w^2/4u$ , it becomes degenerate with $\phi=0$ ; just below, it becomes the global minimum and the system jumps to $\phi_*=w/2u$ . Because $\phi$ changes discontinuously and latent heat is released, the transition is first order. Only representations with no cubic invariant (Ising, $O(n)$ ) escape this and can order continuously.

Why do a uniaxial ferromagnet and the liquid–gas critical point share critical exponents?

Both have a scalar order parameter with $\mathbb{Z}_2$ symmetry: the magnetization component flips sign under $\uparrow\leftrightarrow\downarrow$ , and the density difference $\rho-\rho_c$ flips under the particle–hole symmetry at the critical point. Since the renormalization group says critical behaviour depends only on the spatial dimension $d$ , the number of order-parameter components $n$ , and the symmetry $\rho$ — not on microscopic detail — the two systems flow to the same fixed point and inherit identical exponents ( $\beta$ , $\gamma$ , $\nu$ , …). This is the sharp, quantitative content of universality: symmetry data, not chemistry, sets the exponents.

Is symmetry breaking possible in two dimensions?

It depends on whether the broken symmetry is discrete or continuous. Discrete symmetries (like Ising $\mathbb{Z}_2$ ) can break in $d=2$ — the 2D Ising model orders below $T_c$ . But a continuous symmetry with $\dim(G/H)\ge1$ cannot break at $T>0$ in $d\le2$ : the Goldstone fluctuations diverge logarithmically and destroy long-range order (the Mermin–Wagner theorem). The 2D $XY$ model is the famous loophole — it has no true long-range order but still undergoes a topological Kosterlitz–Thouless transition driven by the very vortices that $\pi_1(S^1)=\mathbb{Z}$ guarantees. So the same symmetry data that predicts Goldstone modes also predicts their dimensional fragility.

Why is crystallization first order but ferromagnetism can be continuous?

Crystallization's order parameters are density-wave amplitudes $\rho_{\mathbf G}$ at the reciprocal-lattice vectors. For common lattices (BCC, FCC) three reciprocal vectors form a closed triangle, $\mathbf G_1+\mathbf G_2+\mathbf G_3=0$ , which allows a cubic Landau invariant $\rho_{\mathbf G_1}\rho_{\mathbf G_2}\rho_{\mathbf G_3}$ . By the Landau criterion a nonzero cubic term forces a discontinuous jump, so freezing is generically first order (and indeed water→ice has latent heat). A ferromagnet's $\mathbb{Z}_2$ or $O(3)$ order parameter admits no cubic invariant — its free energy has only even powers — so its transition can be continuous, with diverging correlation length and power-law singularities.

Do gauge symmetries fit this classification?

Not straightforwardly, and it is a genuine subtlety. A global symmetry can be spontaneously broken with a nonzero local order parameter, giving Goldstone modes and defects as described. A gauge (local) symmetry cannot be broken in that sense — Elitzur's theorem shows the local order parameter averages to zero. In a superconductor the "broken $U(1)$ " is gauged: the would-be Goldstone phase mode is absorbed into the gauge field via the Anderson–Higgs mechanism, giving the photon a mass (the Meissner effect and finite penetration depth) rather than a gapless boson. So while symmetry still organizes the physics, one must distinguish global from gauge symmetry before applying the Goldstone and coset machinery.