C*-Dynamical Systems and the KMS Condition

A thermal state of an infinite quantum field system is defined by analyticity relative to a specified time evolution, not by assuming that a global Gibbs density matrix exists. The Kubo–Martin–Schwinger condition gives that intrinsic definition. It recovers the Gibbs formula when a trace exists, remains meaningful in the thermodynamic limit, and fixes exactly which algebra, dynamics, and inverse temperature belong to an equilibrium claim.

Required background. Quasilocal C*-algebras and inductive limits supplies the observable algebra; states, GNS representations, and folia supplies positive functionals and their representations; and self-adjointness and unitary evolution supplies the generator theorem used after a dynamics is represented. Helpful background. Thermal density operators and the KMS condition develops finite-system calculations, while ground, KMS, and symmetry-selected states compares state-selection criteria on curved backgrounds.

C*-dynamics and analytic elements

A C*-dynamical system is a pair $(\mathcal A,\alpha)$ in which $\mathcal A$ is a unital C*-algebra and

$$\alpha:\mathbb R\longrightarrow\operatorname{Aut}(\mathcal A), \qquad \alpha_{t+s}=\alpha_t\alpha_s,$$

is strongly continuous: $\lVert\alpha_t(A)-A\rVert\to0$ as $t\to0$ for every $A\in\mathcal A$. The parameter $t$ is the named physical time only if the model identifies it with time translation. A rotating equilibrium or a grand-canonical dynamics can instead use a combination such as $H-\Omega J-\mu Q$; the algebraic definition does not silently choose among them.

An element $B$ is entire analytic when $t\mapsto\alpha_t(B)$ extends to an entire $\mathcal A$-valued function $z\mapsto\alpha_z(B)$. These elements form a norm-dense *-subalgebra. One concrete approximation is

$$B_n=\sqrt{\frac{n}{\pi}}\int_{\mathbb R}e^{-nt^2}\alpha_t(B)\,dt,$$

whose complex translate is obtained by shifting the Gaussian. This dense core lets one state the boundary identity algebraically, then extend it to arbitrary observables by continuity.

The KMS boundary condition

Let $\beta>0$. A state $\omega$ is $\beta$-KMS for $\alpha$ when, for every $A,B\in\mathcal A$, there is a function $F_{A,B}$ continuous and bounded on the closed strip $0\leq\operatorname{Im}z\leq\beta$, analytic inside it, with

$$F_{A,B}(t)=\omega\!\left(A\alpha_t(B)\right), \qquad F_{A,B}(t+i\beta)=\omega\!\left(\alpha_t(B)A\right).$$

Equivalently, for entire analytic $B$,

$$\omega\!\left(A\alpha_{i\beta}(B)\right)=\omega(BA).$$

The hypotheses include the algebra, the strongly continuous automorphism group, the sign convention for $\alpha_t$, and the positive number $\beta$. Changing any of them changes the claim. The original infinite-system formulation and its passage from local Gibbs ensembles to the boundary condition are given by Haag, Hugenholtz, and Winnink 1967, §§ 1–3, pp. 215–229.

Two consequences should be separated from the definition. First, every KMS state is invariant: $\omega\circ\alpha_t=\omega$. The proof applies the strip identity to analytic elements and uses boundedness plus Liouville-type continuation; density completes the argument. Second, in the GNS triple $(\mathcal H_\omega,\pi_\omega,\Omega_\omega)$, invariance defines

$$U_\omega(t)\pi_\omega(A)\Omega_\omega =\pi_\omega(\alpha_t(A))\Omega_\omega.$$

This is a strongly continuous unitary group with a self-adjoint generator $L_\omega$. It is a Liouvillean implementing the represented dynamics. It need not be positive, and in a thermal standard representation its spectrum is typically two-sided. KMS equilibrium therefore does not imply the vacuum spectrum condition.

Finite Gibbs systems as a special case

Suppose $\mathcal A=B(\mathcal H)$ with $\dim\mathcal H<\infty$, $H=H^*$, and $\alpha_t(B)=e^{itH}Be^{-itH}$. Then

$$\omega_\beta(A)=\frac{\operatorname{Tr}(e^{-\beta H}A)}{Z_\beta}, \qquad Z_\beta=\operatorname{Tr}(e^{-\beta H}),$$

satisfies the KMS identity. Indeed, for analytic $B$,

$$e^{-\beta H}\alpha_{i\beta}(B) =e^{-\beta H}e^{-\beta H}Be^{\beta H}$$

is not the useful arrangement; the safe step is instead

$$\operatorname{Tr}\!\left(e^{-\beta H}A\alpha_{i\beta}(B)\right) =\operatorname{Tr}\!\left(e^{-\beta H}BA\right),$$

obtained by substituting $\alpha_{i\beta}(B)=e^{-\beta H}Be^{\beta H}$ and using cyclicity twice. The same argument extends when all displayed products are trace class. This calculation also shows why the automorphism convention matters. If one uses $e^{-itH}Be^{itH}$, the sign of the strip changes.

The converse statement is not “every KMS state has this density matrix.” It is only that in this type-I, trace-class setting the KMS functional is the Gibbs functional for the stated dynamics, up to the familiar central qualifications. Infinite-volume representations generally are not type I and have no trace suitable for this formula.

Free-scalar Weyl application

For the massive free scalar, let $h=\sqrt{-\Delta+m^2}$ on the one-particle space, $m>0$, and let $W(f)$ denote Weyl generators with

$$W(f)W(g)=e^{-i\sigma(f,g)/2}W(f+g), \qquad \alpha_t(W(f))=W(e^{ith}f).$$

On a real test-function space stable under $e^{ith}$, define the quasifree covariance

$$\mu_\beta(f,g)=\operatorname{Re}\left\langle f, \coth\!\left(\frac{\beta h}{2}\right)g\right\rangle, \qquad \omega_\beta(W(f))=e^{-\mu_\beta(f,f)/4}.$$

For vectors analytic for $h$, the two-Weyl function

$$F_{f,g}(z)= e^{-i\sigma(f,e^{izh}g)/2} \exp\!\left[-\frac14\mu_\beta(f+e^{izh}g,f+e^{izh}g)\right]$$

has the required strip continuation. The scalar identity relating $\coth(\beta\lambda/2)$ to the Bose factors $(1-e^{-\beta\lambda})^{-1}$ exchanges the order of the two Weyl operators at $z=t+i\beta$. Density then yields the KMS condition on the Weyl algebra. This doubled thermal representation and its infinite-volume, non-type-I character originate in Araki and Woods 1963, §§ 2–5, pp. 640–657. The physical correlation-function calculation is developed at thermal density operators and the KMS condition.

Adversarial test: the nonexistent global Gibbs trace

On bosonic Fock space over $L^2(\mathbb R^3,d^3k)$, the formal partition function would be

$$\operatorname{Tr}_{\mathcal F}\!\left(e^{-\beta d\Gamma(h)}\right) =\det(1-e^{-\beta h})^{-1}.$$

But $e^{-\beta h}$ is a multiplication operator on a nonatomic infinite-volume one-particle space and is not trace class. The determinant and the Fock trace therefore do not exist. The strongest surviving statement is the quasifree KMS functional on the quasilocal Weyl algebra and its GNS representation—not a density operator in the vacuum Fock representation. A finite box repairs trace class but changes the object; its thermodynamic limit must be taken at the level of local observables or states.

Common pitfalls

Treating β as a property of a state alone. KMS is a relation among $(\mathcal A,\alpha,\omega,\beta)$. The same functional may be KMS for a rescaled dynamics at a rescaled inverse temperature.

Replacing analyticity by imaginary-time periodicity. Periodicity of one two-point function is not the all-observable boundary identity. State the analytic domain, boundary values, and class of observables before calling a correlation function KMS.

Exercises

1. Let $H$ have discrete spectrum and $e^{-\beta H}$ be trace class. Verify the KMS identity for matrix units $E_{mn}=|m\rangle\langle n|$.

Solution

Since $\alpha_t(E_{pq})=e^{it(E_p-E_q)}E_{pq}$, both boundary values vanish unless $n=p$ and $m=q$. In that case the lower boundary is $Z_\beta^{-1}e^{-\beta E_m}e^{it(E_n-E_m)}$. At $t+i\beta$ the exponential contributes $e^{-\beta(E_n-E_m)}$, giving $Z_\beta^{-1}e^{-\beta E_n}e^{it(E_n-E_m)}$, exactly the expectation of $\alpha_t(E_{nm})E_{mn}$.

2. Explain why a ground state is not a finite-$\beta$ KMS state merely because it is invariant.

Solution

Invariance gives only equality under real time translation. KMS also requires a bounded analytic strip and the order-reversing boundary relation at height $\beta$. A ground state instead has a positive-energy half-plane analyticity structure; without exceptional trivial dynamics it does not satisfy the finite-strip boundary identity.

References

• Araki, H., and Woods, E. J. “Representations of the Canonical Commutation Relations Describing a Nonrelativistic Infinite Free Bose Gas.” Journal of Mathematical Physics 4 (1963): 637–662. DOI.
• Haag, R., Hugenholtz, N. M., and Winnink, M. “On the Equilibrium States in Quantum Statistical Mechanics.” Communications in Mathematical Physics 5 (1967): 215–236. DOI.