Dynamics as Automorphisms

Once observables are organized into an algebra $\mathcal A$, closed-system time evolution should preserve that algebra’s physical structure. It should carry products to products, adjoints to adjoints, positive elements to positive elements, and the identity to itself. The algebraic formulation expresses this by a continuous family of *-automorphisms

$$\alpha_t:\mathcal A\to\mathcal A.$$

The resulting object is a C-dynamical system*. It describes dynamics without first selecting a Hilbert-space representation or requiring a Hamiltonian to be a bounded observable in $\mathcal A$.

The concrete operator-dynamics bridge, including spin and oscillator examples, fixed-point algebras, discrete maps, and the contrast with channels, is developed in Symmetries and Dynamical Automorphisms. This page owns the abstract continuous-time formulation, its action on states, and its equilibrium and QFT extensions.

Automorphisms of the Observable Algebra

A -automorphism of a unital C-algebra $\mathcal A$ is a bijective complex-linear map $\alpha:\mathcal A\to\mathcal A$ satisfying

$$\begin{aligned} \alpha(AB) &= \alpha(A)\alpha(B),\\ \alpha(A^*) &= \alpha(A)^*,\\ \alpha(I) &=I. \end{aligned}$$

Linearity and multiplicativity imply preservation of polynomial relations. In particular,

$$\alpha([A,B]) = [\alpha(A),\alpha(B)].$$

If $X=A^*A$ is positive, then

$$\alpha(X) = \alpha(A)^*\alpha(A) \geq0.$$

The inverse automorphism preserves positivity as well, so order is preserved in both directions. A C*-algebra automorphism is also isometric:

$$\lVert\alpha(A)\rVert = \lVert A\rVert.$$

It consequently preserves spectra:

$$\operatorname{spec}(\alpha(A)) = \operatorname{spec}(A).$$

These properties distinguish reversible closed dynamics from a general linear operator map. A noisy Heisenberg-picture channel may preserve the identity and positivity while shrinking spectra and failing

$$\Phi(AB)=\Phi(A)\Phi(B).$$

Such a map is not an automorphism.

An automorphism is inner if there is a unitary $U\in\mathcal A$ such that

$$\alpha(A)=U^*AU.$$

An automorphism that is not inner is outer. This is an algebra-relative distinction: an outer automorphism of $\mathcal A$ may still be implemented by a unitary after $\mathcal A$ is represented on a larger Hilbert space.

Time-Translation Groups

Autonomous time evolution is represented by a homomorphism from the additive group of real times into the automorphism group:

$$\alpha:\mathbb R\to\operatorname{Aut}(\mathcal A), \qquad t\longmapsto\alpha_t.$$

The group laws are

$$\alpha_0=\operatorname{id}, \qquad \alpha_{t+s} = \alpha_t\circ\alpha_s, \qquad \alpha_t^{-1}=\alpha_{-t}.$$

For a C*-dynamical system, the standard continuity requirement is point-norm continuity:

$$\lim_{t\to0} \lVert\alpha_t(A)-A\rVert =0 \qquad \text{for each }A\in\mathcal A.$$

This is often called strong continuity in the C*-algebra literature. It should not be confused with the stronger operator-norm condition

$$\lVert\alpha_t-\operatorname{id}\rVert\to0.$$

The latter would force a bounded generator and is too restrictive for many physical systems.

If a Hamiltonian depends explicitly on time, the natural object is instead a two-parameter propagator of automorphisms,

$$\alpha_{t,s}:\mathcal A\to\mathcal A,$$

with

$$\alpha_{t,r}\circ\alpha_{r,s} = \alpha_{t,s}.$$

There is then no reason for $\alpha_{t,s}$ to depend only on $t-s$, so a one-parameter time-translation group need not exist.

Infinitesimal Generators

The generator of a point-norm continuous automorphism group is defined on the elements for which the norm limit exists:

$$D(\delta) = \left\{ A\in\mathcal A: \lim_{t\to0} \frac{\alpha_t(A)-A}{t} \text{ exists} \right\},$$

and

$$\delta(A) = \lim_{t\to0} \frac{\alpha_t(A)-A}{t}.$$

The domain $D(\delta)$ is a dense *-subalgebra. The generator is generally an unbounded, closed linear operator on the Banach space $\mathcal A$. Differentiating the product and adjoint identities gives

$$\delta(AB) = \delta(A)B+A\delta(B),$$

and

$$\delta(A^*) = \delta(A)^*.$$

Thus $\delta$ is a *-derivation. The Leibniz rule is the infinitesimal form of product preservation.

For sufficiently regular $A),

$$\frac{d}{dt}\alpha_t(A) = \alpha_t(\delta(A)) = \delta(\alpha_t(A)).$$

The group can be denoted formally by

$$\alpha_t=e^{t\delta},$$

but this exponential notation should not hide the domain theory of an unbounded generator.

Hamiltonian Generation

If a bounded self-adjoint $H$ belongs to a represented unital algebra, then

$$\alpha_t(A) = e^{iHt/\hbar} A e^{-iHt/\hbar}$$

is an inner automorphism group, with bounded derivation

$$\delta(A) = \frac{i}{\hbar}[H,A].$$

This is the algebraic form of the Heisenberg equation. The canonical operator derivation is treated in Commutator Dynamics, while the unitary-group theorem and its domain assumptions are treated in One-Parameter Unitary Groups.

Physical Hamiltonians are often unbounded and therefore do not belong to the C*-algebra of bounded observables. Even then, their unitaries may implement an automorphism in a Hilbert-space representation:

$$\pi(\alpha_t(A)) = e^{iHt/\hbar} \pi(A) e^{-iHt/\hbar}.$$

The abstract dynamics can be perfectly well defined even if no element $H\in\mathcal A$ generates it by a commutator. The automorphism group may be outer on $\mathcal A$, and its derivation may be unbounded.

This distinction becomes concrete in an infinite quantum lattice. Finite regions $\Lambda$ have Hamiltonians $H_\Lambda$, but the total energy generally grows without bound as $\Lambda$ fills the lattice. Under suitable locality and interaction assumptions, a local observable $A$ can nevertheless have a thermodynamic-limit evolution

$$\alpha_t(A) = \lim_{\Lambda\nearrow\Gamma} e^{iH_\Lambda t/\hbar} A e^{-iH_\Lambda t/\hbar}.$$

The limiting automorphism acts on the quasi-local observable algebra even though an infinite-volume Hamiltonian is not an element of that algebra. This is a principal reason to regard $\alpha_t$, rather than a formal total $H$, as the primary dynamical object.

Relation to the Heisenberg Picture

Choose the convention

$$U_t=e^{-iHt/\hbar}, \qquad \alpha_t(A)=U_t^*AU_t.$$

If a density operator evolves in the Schrödinger picture as

$$\rho_t=U_t\rho_0U_t^*,$$

then cyclicity of the trace gives

$$\operatorname{Tr}(\rho_tA) = \operatorname{Tr}(\rho_0\alpha_t(A)).$$

The same duality can be stated without density matrices. If $\omega_0$ is the initial state functional, define

$$\omega_t = \omega_0\circ\alpha_t.$$

Then

$$\omega_t(A) = \omega_0(\alpha_t(A)).$$

Because $\alpha_t$ preserves products, adjoints, and the identity, $\omega_t$ remains a state:

$$\begin{aligned} \omega_t(A^*A) &= \omega_0 \!\left( \alpha_t(A)^*\alpha_t(A) \right) \geq0,\\ \omega_t(I)&=1. \end{aligned}$$

On the generator domain,

$$\frac{d}{dt}\omega_t(A) = \omega_t(\delta(A)).$$

For a finite system with $\delta(A)=i[H,A]/\hbar$, the dual density-operator equation is

$$\frac{d\rho_t}{dt} = -\frac{i}{\hbar}[H,\rho_t].$$

This is the Liouville–von Neumann Equation. The sign difference reflects which side of the expectation pairing carries the time dependence.

A state is invariant or stationary if

$$\omega\circ\alpha_t=\omega \qquad \text{for all }t.$$

An observable is a fixed point if $\alpha_t(A)=A$ for all $t$. Stationary states and fixed observables are dual notions, but neither condition by itself characterizes thermal equilibrium.

Kubo–Martin–Schwinger Equilibrium Preview

In a finite system with a trace-class Gibbs state, thermal equilibrium at inverse temperature $\beta$ is represented by

$$\rho_\beta = \frac{e^{-\beta H}} {\operatorname{Tr}(e^{-\beta H})}.$$

For infinite systems, the total Gibbs operator and partition function may not exist. The Kubo–Martin–Schwinger condition characterizes equilibrium directly from the state $\omega$, the algebra $\mathcal A$, and the dynamics $\alpha_t$.

Set $\hbar=1$ in this subsection, so the imaginary-time strip has height $\beta$. An element $A$ is analytic for the dynamics if $t\mapsto\alpha_t(A)$ extends to an entire algebra-valued function $z\mapsto\alpha_z(A)$. A state is a KMS state at inverse temperature $\beta$ if, for analytic $A$ and $B$,

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

Equivalently, for arbitrary $A,B\in\mathcal A$, there is a function $F_{A,B}(z)$ that is analytic in the open strip

$$0\lt\operatorname{Im}z\lt\beta,$$

continuous on its closure, and has boundary values

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

If physical time is retained rather than setting $\hbar=1$, the strip height is $\hbar\beta$.

For finite-dimensional Hamiltonian dynamics,

$$\alpha_{i\beta}(A) = e^{-\beta H}Ae^{\beta H}.$$

The Gibbs state satisfies the KMS identity by cyclicity of the trace. The condition is therefore a twisted trace relation, with the twist supplied by imaginary-time evolution.

Every KMS state at nonzero inverse temperature is invariant under the dynamics, but the converse is false. Commuting with $H$ is enough for stationarity in a finite system; it does not force the particular Boltzmann weights required at temperature $\beta$.

The KMS condition survives thermodynamic and continuum limits in which no global Gibbs density matrix exists. It also connects equilibrium correlation functions, imaginary-time analyticity, passivity, and modular theory. KMS Condition Preview owns the finite Gibbs derivation, thermal-strip figure, detailed-balance checks, and the fuller physics-level bridge; this section retains the defining C*-dynamical context.

GNS Implementation of Invariant Dynamics

Let

$$(\mathcal H_\omega,\pi_\omega,\Omega_\omega)$$

be the GNS representation of a state $\omega$. If $\omega$ is invariant, define on the dense set of cyclic vectors

$$V_t\pi_\omega(A)\Omega_\omega = \pi_\omega(\alpha_t(A))\Omega_\omega.$$

Invariance makes this map isometric:

$$\begin{aligned} \big\lVert V_t\pi_\omega(A)\Omega_\omega \big\rVert^2 &= \omega \!\left( \alpha_t(A^*A) \right)\\ &= \omega(A^*A). \end{aligned}$$

It therefore extends to a unitary group satisfying

$$V_t\Omega_\omega=\Omega_\omega$$

and

$$V_t\pi_\omega(A)V_t^* = \pi_\omega(\alpha_t(A)).$$

Under the appropriate continuity assumptions, Stone’s theorem gives a self-adjoint generator $L_\omega$:

$$V_t=e^{itL_\omega}.$$

This operator implements time translations in the GNS representation. It need not be an observable belonging to $\pi_\omega(\mathcal A)$, and in thermal representations it is often called a Liouvillian rather than the Hamiltonian of a single vector-state system.

This result sharpens an important distinction. An automorphism can be outer as a map of the abstract algebra yet spatially implemented by unitaries in the representation selected by an invariant state.

Algebraic QFT Bridge

Algebraic QFT assigns an observable algebra to each spacetime region:

$$\mathcal O \longmapsto \mathcal A(\mathcal O).$$

A time translation acts covariantly on the net:

$$\alpha_t \!\left( \mathcal A(\mathcal O) \right) = \mathcal A(\mathcal O+t e_0),$$

where $e_0$ denotes the chosen time direction. The automorphism moves both observables and their localization while preserving algebraic relations and causal commutativity.

A vacuum state is normally invariant under spacetime translations and yields a unitary implementation in its GNS representation. Thermal states are characterized by the KMS condition relative to a chosen time flow. Different phases or thermal sectors can lead to inequivalent GNS representations even though the same abstract quasi-local algebra and dynamics are used.

This framework avoids assuming that an interacting relativistic field theory has one globally preferred Schrödinger Hilbert space or that the generator of time translations is a bounded local observable. The algebra, its automorphisms, and the selected state jointly determine the represented dynamics.

Common Mistakes

• Calling every positive or unital operator map an automorphism. Multiplication, adjoints, identity, and invertibility must all be preserved.
• Requiring $\lVert\alpha_t-\operatorname{id}\rVert\to0$. A C*-dynamical system normally requires point-norm continuity for each algebra element.
• Assuming the generator is bounded or defined on all of $\mathcal A$.
• Assuming every derivation has the form $i[H,\cdot]/\hbar$ for an $H$ inside the observable algebra.
• Equating innerness with unitary implementability in every representation.
• Evolving states and observables in the same direction without checking the expectation-value pairing and sign convention.
• Treating every invariant state as a thermal state. The KMS condition is stronger than stationarity.
• Writing a global Gibbs density operator for an infinite system without checking that the trace and thermodynamic limit exist.
• Forgetting that reversing the convention for $\alpha_t$ reverses the imaginary-time sign in the KMS formula.
• Treating the GNS implementation generator as automatically identical to a local Hamiltonian observable.

References

• O. Bratteli and D. W. Robinson, Operator Algebras and Quantum Statistical Mechanics 1, 2nd ed., Springer, 1987.
• O. Bratteli and D. W. Robinson, Operator Algebras and Quantum Statistical Mechanics 2: Equilibrium States, Models in Quantum Statistical Mechanics, 2nd ed., Springer, 1997.
• R. Haag, N. M. Hugenholtz, and M. Winnink, “On the equilibrium states in quantum statistical mechanics,” Communications in Mathematical Physics 5, 215–236, 1967.
• R. Haag, Local Quantum Physics: Fields, Particles, Algebras, 2nd ed., Springer, 1996.
• R. V. Kadison and J. R. Ringrose, Fundamentals of the Theory of Operator Algebras, Volume II, AMS, 1997.
• H. Araki, Mathematical Theory of Quantum Fields, Oxford University Press, 1999.

Exercises

1. Prove that pulling a state back along an automorphism gives another state.

Solution

Let

$$\omega_t=\omega\circ\alpha_t.$$

Linearity follows by composition of linear maps. For positivity,

$$\begin{aligned} \omega_t(A^*A) &= \omega(\alpha_t(A^*A))\\ &= \omega \!\left( \alpha_t(A)^*\alpha_t(A) \right) \geq0. \end{aligned}$$

Normalization follows from identity preservation:

$$\omega_t(I) = \omega(\alpha_t(I)) = \omega(I) =1.$$

Thus $\omega_t$ is a normalized positive linear functional.

2. Derive the Leibniz and adjoint identities for the generator $\delta$.

Solution

For $A,B\in D(\delta)$ such that $AB\in D(\delta)$,

$$\begin{aligned} \delta(AB) &= \lim_{t\to0} \frac{ \alpha_t(A)\alpha_t(B)-AB }{t}\\ &= \lim_{t\to0} \left[ \frac{\alpha_t(A)-A}{t}\alpha_t(B) + A\frac{\alpha_t(B)-B}{t} \right]\\ &= \delta(A)B+A\delta(B). \end{aligned}$$

Similarly,

$$\begin{aligned} \delta(A^*) &= \lim_{t\to0} \frac{\alpha_t(A)^*-A^*}{t}\\ &= \left( \lim_{t\to0} \frac{\alpha_t(A)-A}{t} \right)^*\\ &= \delta(A)^*. \end{aligned}$$

Norm continuity justifies passing the limits through multiplication and adjoint.

3. Recover the Liouville–von Neumann equation from the duality between states and observables.

Solution

For

$$\alpha_t(A) = e^{iHt/\hbar}Ae^{-iHt/\hbar},$$

the generator is

$$\delta(A) = \frac{i}{\hbar}[H,A].$$

Let $\omega_t(A)=\operatorname{Tr}(\rho_tA)$. Duality requires

$$\frac{d}{dt}\omega_t(A) = \omega_t(\delta(A)).$$

Therefore

$$\operatorname{Tr} \!\left( \frac{d\rho_t}{dt}A \right) = \frac{i}{\hbar} \operatorname{Tr} \!\left( \rho_t[H,A] \right).$$

Cyclicity gives

$$\operatorname{Tr} \!\left( \rho_t[H,A] \right) = \operatorname{Tr} \!\left( [\,\rho_t,H\,]A \right).$$

Because this holds for every $A$,

$$\frac{d\rho_t}{dt} = -\frac{i}{\hbar}[H,\rho_t].$$

4. Verify the KMS identity for a finite-dimensional Gibbs state.

Solution

Set $\hbar=1$ and let

$$\omega_\beta(X) = \frac{\operatorname{Tr}(e^{-\beta H}X)} {Z}, \qquad Z=\operatorname{Tr}(e^{-\beta H}).$$

For the dynamics

$$\alpha_t(A)=e^{itH}Ae^{-itH},$$

analytic continuation gives

$$\alpha_{i\beta}(A) = e^{-\beta H}Ae^{\beta H}.$$

Then

$$\begin{aligned} \omega_\beta \!\left( B\alpha_{i\beta}(A) \right) &= \frac{1}{Z} \operatorname{Tr} \!\left( e^{-\beta H} B e^{-\beta H} A e^{\beta H} \right)\\ &= \frac{1}{Z} \operatorname{Tr} \!\left( B e^{-\beta H}A \right)\\ &= \frac{1}{Z} \operatorname{Tr} \!\left( e^{-\beta H}AB \right)\\ &= \omega_\beta(AB). \end{aligned}$$

The second and third lines use cyclicity of the finite-dimensional trace.

5. Show that invariant dynamics is unitarily implemented in the GNS representation.

Solution

On the dense cyclic subspace, define

$$V_t\pi_\omega(A)\Omega_\omega = \pi_\omega(\alpha_t(A))\Omega_\omega.$$

For $A$ and $B$,

$$\begin{aligned} \big\langle V_t\pi_\omega(A)\Omega_\omega \big\vert V_t\pi_\omega(B)\Omega_\omega \big\rangle &= \omega \!\left( \alpha_t(A)^* \alpha_t(B) \right)\\ &= \omega \!\left( \alpha_t(A^*B) \right)\\ &= \omega(A^*B), \end{aligned}$$

where the last equality is invariance. Hence $V_t$ is isometric and extends to a unitary; $V_{-t}$ supplies its inverse.

Because $\alpha_t(I)=I$,

$$V_t\Omega_\omega=\Omega_\omega.$$

Acting on the dense cyclic subspace also gives

$$V_t\pi_\omega(A)V_t^* = \pi_\omega(\alpha_t(A)).$$

Thus the abstract automorphism becomes spatially implemented in the invariant state’s GNS representation.