Thermal Nuclearity, Return to Equilibrium, and Mixing

Thermal existence and thermal relaxation answer different questions. Phase-space nuclearity can produce infinite-volume KMS states without proving uniqueness or dynamics toward them. Return to equilibrium is a spectral statement about a specified Liouvillean and folium; mixing is stronger than ergodicity, and neither follows from KMS analyticity alone.

The free-field perturbation test below is handed back to infinite-volume KMS states, passivity, and phase multiplicity for its physical interpretation.

Required background. Phase-space nuclearity and compactness maps supplies the local degree-of-freedom bounds; C*-dynamical systems and the KMS condition fixes equilibrium; modular dynamics and equilibrium representations supplies the standard Liouvillean. Helpful background. Equilibration, thermalization, and dephasing compares finite-system and many-body notions.

From phase-space bounds to KMS existence

In a vacuum representation, consider the energy-damped localization map

$$\Theta_{\gamma,O}(A)=e^{-\gamma H}A\Omega, \qquad A\in\mathcal A(O)_1.$$

A quantitative nuclearity estimate controls the number of locally preparable states at energy resolution $\gamma$. To construct a thermal state, place the system in increasing bounded regions, form normalized local Gibbs-type functionals, and use nuclearity to obtain compactness of their restrictions to each fixed local algebra. A diagonal subnet gives a locally normal state on the quasilocal algebra. Boundary corrections vanish in the limit, and the limiting analytic functions satisfy the $\beta$-KMS boundary identity.

The conclusion is existence at the specified inverse temperature, not a trace-class global Gibbs operator. The required estimates must be uniform in the approximating volume and strong enough at the relevant energy scale. Buchholz and Junglas prove this route for local QFT under their nuclearity hypothesis in Buchholz and Junglas 1989, §§ 2–4, pp. 258–270. A mass gap alone does not control species growth, and nuclearity alone does not choose one extremal phase.

Ergodicity, mixing, and return

Let $\omega_\beta$ be invariant under $\alpha_t$. Three conclusions must be kept distinct:

$$\begin{aligned} \text{ergodic: }& \lim_{T\to\infty}\frac1T\int_0^T \omega_\beta(A\alpha_t(B))\,dt =\omega_\beta(A)\omega_\beta(B),\\ \text{mixing: }& \lim_{t\to\infty}\omega_\beta(A\alpha_t(B)) =\omega_\beta(A)\omega_\beta(B),\\ \text{return: }& \lim_{t\to\infty}\varphi(\alpha_t(A)) =\omega_\beta(A) \quad\text{for every stated normal }\varphi. \end{aligned}$$

The last line requires the initial state class to be named—usually the normal folium of $\omega_\beta$, sometimes with an energy or regularity restriction. For vector states $\varphi_C(A)=\langle C\Omega_\beta,AC\Omega_\beta\rangle$, mixing extends to return on a dense class; norm bounds and density then cover the declared folium.

Let $L$ be the standard Liouvillean, $e^{itL}\Omega_\beta=\Omega_\beta$. The mean ergodic theorem shows that a simple zero eigenspace gives ergodicity. Pointwise mixing needs decay of nonzero spectral matrix elements. A clean sufficient condition is

$$\ker L=\mathbb C\Omega_\beta, \qquad L\big|_{\Omega_\beta^\perp} \text{ purely absolutely continuous}.$$

Then the Riemann–Lebesgue lemma sends the off-vacuum Fourier transforms to zero. A simple eigenvalue at zero does not exclude singular continuous spectrum and therefore does not, by itself, prove mixing or a rate. A resonance estimate or Mourre bound can provide absolute continuity and quantitative decay in a concrete model.

A locally perturbed Araki–Woods test

Start with the massive free-scalar Araki–Woods representation and its Liouvillean

$$L_0=d\Gamma(h\oplus(-\overline h)).$$

Let $V=V^*$ be a bounded operator affiliated with a bounded-region algebra, and perturb with small coupling $\lambda$. In standard form the formal interacting Liouvillean has the balanced structure

$$L_\lambda=L_0+\lambda\bigl(V-JVJ\bigr),$$

after the corresponding Araki perturbation fixes the interacting KMS vector $\Omega_{\beta,\lambda}$. The commutant term is essential: $V$ alone would not implement the perturbed automorphism while preserving the standard cone.

The return-to-equilibrium test is now explicit:

1. construct the perturbed KMS vector and identify the self-adjoint domain of $L_\lambda$;
2. prove that $0$ is a simple eigenvalue with vector $\Omega_{\beta,\lambda}$;
3. exclude other eigenvalues and singular spectrum on its orthogonal complement;
4. derive the claimed decay only in the coupling, temperature, and form-factor regime where the commutator estimates close.

For atom–radiation and spin–boson models, this program uses the Araki–Woods representation, a virial theorem, and a Mourre positive-commutator estimate; the small-coupling and infrared hypotheses are explicit in Fröhlich and Merkli 2004, Theorems 2.1–2.3, pp. 241–260. For the scalar local perturbation above, the spectral conditions license return if proved; writing $V$ does not prove them. The physical interpretation remains with infinite-volume KMS states, passivity, and phase multiplicity.

Adversarial test: a persistent Liouvillean mode

Suppose a conserved charge $Q$ produces a normal vector $\Psi_Q\perp\Omega_\beta$ with $L\Psi_Q=0$. Then

$$\dim\ker L\geq2.$$

The Cesàro limit retains the projection onto both zero modes, and observables distinguishing their sectors cannot converge to one equilibrium expectation. KMS and even thermal nuclearity may still hold. The strongest conclusion is equilibrium-state existence, perhaps phasewise ergodicity after fixing $Q$; the missing hypothesis for global return is simplicity of the stationary subspace.

Common pitfalls

Using nuclearity as a relaxation rate. Nuclearity controls phase-space size and compactness. Long-time decay is governed by Liouvillean spectral or resonance estimates.

Calling Cesàro convergence mixing. Time averaging removes oscillations that need not decay pointwise. Always state the mode of convergence.

Exercises

1. Show that pure absolute continuity away from a simple zero eigenvalue implies mixing for vectors whose spectral densities are $L^1$.

Solution

Split each vector into its multiple of $\Omega_\beta$ and an orthogonal part. The invariant components give the product of expectations. The remaining matrix coefficient is the Fourier transform of an $L^1$ spectral density, so it tends to zero by the Riemann–Lebesgue lemma.

2. Give a finite-dimensional reason why nontrivial closed systems generally fail to mix.

Solution

Their Liouvilleans have discrete spectra. A correlation function is a finite or countable sum of persistent phases $e^{it(E_m-E_n)}$; time averages may remove nonzero frequencies, but pointwise limits generally do not exist. A continuous reservoir spectrum is one mechanism that permits decay.

References

• Buchholz, Detlev, and Peter Junglas. “On the Existence of Equilibrium States in Local Quantum Field Theory.” Communications in Mathematical Physics 121 (1989): 255–270. DOI.
• Fröhlich, Jürg, and Marco Merkli. “Another Return of ‘Return to Equilibrium’.” Communications in Mathematical Physics 251 (2004): 235–262. DOI. Open PDF.