Factorial KMS States, Phases, and Symmetry Breaking

At fixed dynamics and inverse temperature, the KMS states form a convex set. Its extreme points are the pure thermodynamic phases: their GNS von Neumann algebras are factors, so macroscopic observables have sharp central values. Nonextremal KMS states are mixtures of such phases. This representation-theoretic statement is an infinite-system theorem; finite-volume metastable peaks are not additional algebraic phases.

Required background. Local normality, quasiequivalence, and folia distinguishes local agreement from global equivalence; representation types, factors, and local-algebra structure supplies factoriality; C*-dynamical systems and the KMS condition fixes the equilibrium simplex. Helpful background. Infinite-volume KMS states, passivity, and phase multiplicity develops physical phase selection, and convexity, gauge dependence, and physical phase criteria separates phase coexistence from effective-potential artifacts.

The KMS simplex and factoriality

Fix $(\mathcal A,\alpha)$ and $\beta>0$, and denote the set of $\beta$-KMS states by $K_\beta$. It is convex: if $\omega_1,\omega_2\in K_\beta$ and $0\leq\lambda\leq1$, then

$$\omega=\lambda\omega_1+(1-\lambda)\omega_2$$

has the same analytic strip and boundary identity. Under the usual weak-* topology it is compact when nonempty. A member is extremal if it has no nontrivial convex decomposition inside $K_\beta$.

Let $(\mathcal H_\omega,\pi_\omega,\Omega_\omega)$ be its GNS representation and $\mathcal M_\omega=\pi_\omega(\mathcal A)''$. The central theorem is

$$\omega\text{ extremal in }K_\beta \quad\Longleftrightarrow\quad \mathcal Z(\mathcal M_\omega)=\mathbb C1.$$

Thus an extremal KMS state is primary or factorial. In one direction, a nontrivial central projection $Z$ yields two normalized KMS components by conditioning with $Z$ and $1-Z$. In the other, a convex decomposition gives a nontrivial Radon–Nikodym operator in the center of the supported standard representation. The decomposition and its relation to pure thermodynamic phases are developed in Emch and Knops 1970, §§ 2–4, pp. 3010–3018; the equilibrium representation input goes back to Haag, Hugenholtz, and Winnink 1967, §§ 3–4, pp. 225–236.

Factorial does not mean irreducible. At nonzero temperature, $\mathcal M_\omega$ is often type III, so its commutant is large even when its center is trivial. The correct phase condition is factoriality, not purity of the C*-state in the zero-temperature sense.

Central decomposition and disjoint phases

Under standard separability hypotheses, a general KMS representation decomposes over its center:

$$\mathcal H_\omega=\int_X^\oplus\mathcal H_x\,d\mu(x), \qquad \pi_\omega=\int_X^\oplus\pi_x\,d\mu(x), \qquad \omega(A)=\int_X\omega_x(A)\,d\mu(x),$$

where $\omega_x$ is extremal KMS for almost every $x$. A bounded central observable is multiplication by a function on $X$; it records a classical random phase label. Uniqueness statements concern this measure only up to the appropriate measure class and null sets.

Distinct factorial phases selected by different macroscopic boundary conditions are normally disjoint: no nonzero subrepresentation is shared. Disjointness is stronger than inequivalence and compatible with local normality on every bounded region. A phase transition therefore appears globally even when all finite local experiments are described by mutually normal restrictions.

Symmetry breaking as an orbit of phases

Let a group $G$ act by automorphisms $\gamma_g$ commuting with $\alpha_t$. It maps $K_\beta$ to itself. An extremal phase $\omega_+$ breaks the symmetry when

$$\omega_+\circ\gamma_g\neq\omega_+$$

for some $g$, and the transformed state belongs to a distinct phase. The symmetry can remain unbroken in their mixture. The key distinction is

$$\text{symmetric state} \not\Rightarrow \text{symmetric factorial component}.$$

For a $\mathbb Z_2$ scalar order parameter, suppose thermodynamic limits with plus and minus boundary conditions produce extremal KMS states $\omega_\pm$ with

$$\omega_+(\phi)=v, \qquad \omega_-(\phi)=-v, \qquad v>0.$$

The symmetric equilibrium state

$$\omega_{\mathrm{sym}}=\frac12(\omega_++\omega_-)$$

is KMS but not factorial. Its GNS representation is the direct sum of the two disjoint phase representations, and the projection onto the plus summand lies in the center. Spatial averages

$$M_R=\frac{1}{|B_R|}\int_{B_R}\phi(0,\mathbf x)\,d^3\mathbf x$$

converge weakly to $+v1$ in the plus phase and $-v1$ in the minus phase when the appropriate clustering theorem holds. Their distinct central limits prove disjointness. This is the exact phase decomposition used by infinite-volume KMS states, passivity, and phase multiplicity; existence of $\omega_\pm$ and convergence of $M_R$ remain model-specific hypotheses, not consequences of the abstract simplex theorem.

Adversarial test: a finite-volume double peak

In a finite box with a $\mathbb Z_2$-invariant Hamiltonian, the Gibbs density matrix is typically unique and faithful. Its representation of the full matrix algebra is factorial. A bimodal probability distribution for a coarse magnetization may signal long-lived metastability, but it does not create central projections in that algebra. Only after specifying a thermodynamic limit and proving distinct limiting KMS states can one claim phase multiplicity. The surviving finite-volume statement is a distribution with two peaks; the missing objects are disjoint infinite-volume representations.

Common pitfalls

Calling an extremal KMS state pure. It is pure as a thermodynamic phase, meaning factorial within the KMS simplex. It need not be a pure state of the C*-algebra.

Using a local order parameter to prove global equivalence. Different phases can be locally quasiequivalent. Disjointness is detected by macroscopic limits, the center, or asymptotic observables.

Exercises

1. Show that a nontrivial convex combination of two disjoint factorial KMS states is not factorial.

Solution

The GNS representation of $\lambda\omega_1+(1-\lambda)\omega_2$ is the direct sum of the disjoint GNS representations. The projection $P_1$ onto the first summand commutes with both represented algebras and their commutants, so $P_1$ is a nontrivial central projection. Hence the represented von Neumann algebra is not a factor.

2. Why can $\omega_{\mathrm{sym}}(\phi)=0$ coexist with symmetry breaking?

Solution

The expectation cancels between the two components: $(v+(-v))/2=0$. Symmetry breaking is a property of the extremal components and the action that permutes them. The symmetric mixture retains a classical central uncertainty over the two broken phases.

References

• Emch, Gérard G., and Hubert J. F. Knops. “Pure Thermodynamical Phases as Extremal KMS States.” Journal of Mathematical Physics 11 (1970): 3008–3018. DOI.
• Haag, Rudolf, Nicolaas M. Hugenholtz, and Marinus Winnink. “On the Equilibrium States in Quantum Statistical Mechanics.” Communications in Mathematical Physics 5 (1967): 215–236. DOI.