Symmetries and Dynamical Automorphisms

Closed quantum dynamics does more than move vectors in Hilbert space. In the Heisenberg viewpoint it moves observables into observables while preserving the algebraic relations among them. For the abstract C*-dynamical-system treatment, including equilibrium states and Hilbert-space representations, see Dynamics as Automorphisms.

The map

$$\alpha_t(A) = U^\dagger(t)AU(t)$$

is a dynamical automorphism: it preserves sums, products, adjoints, identity, commutators, spectra, and positivity. This page explains that structure-preserving language and why it is useful before the full algebraic formulation is introduced.

The guiding point is simple. Schrödinger evolution is unitary on states; Heisenberg evolution is automorphic on observables. These are dual descriptions of the same closed-system physics.

Observables as an Algebra

For a finite-dimensional system, the observables and their complex linear combinations form a matrix algebra such as

$$\mathcal A=M_n(\mathbb C).$$

The relevant operations are

$$A+B, \qquad \lambda A, \qquad AB, \qquad A^\dagger.$$

Self-adjoint elements $A=A^\dagger$ represent observables in the narrow measurement sense, but the algebra must also include non-self-adjoint elements. Products, ladder operators, projectors, unitaries, and complex linear combinations all belong to the same structure.

The noncommuting product is not decorative. It carries the compatibility and dynamics information:

$$[A,B]=AB-BA.$$

Reversible closed-system evolution preserves products and adjoints. Reduced noisy evolution need not preserve products, while still defining consistent measurement probabilities. For example, if $A$ and $B$ multiply to a projector, or if $A^\dagger A$ is positive, the evolved quantities should retain the corresponding algebraic relation.

Automorphisms

An automorphism of an observable algebra $\mathcal A$ is an invertible map

$$\alpha:\mathcal A\to\mathcal A$$

that preserves the algebraic operations:

$$\alpha(A+B) = \alpha(A)+\alpha(B), \qquad \alpha(\lambda A) = \lambda\alpha(A),$$ $$\alpha(AB) = \alpha(A)\alpha(B), \qquad \alpha(A^\dagger) = \alpha(A)^\dagger, \qquad \alpha(I)=I.$$

The product rule is the most restrictive part. A general linear transformation on operators is not an automorphism. Complete positivity and trace preservation do not make a quantum channel an automorphism: product preservation is an additional requirement. A noisy channel can even be invertible as a linear map while its inverse fails to be a physical channel. The depolarizing example below separates these properties; see also Nayak and Sen on invertible quantum operations.

Several useful consequences follow immediately. Since commutators are built from products,

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

Since adjoints and products are preserved, positivity is preserved:

$$A=C^\dagger C \quad\Longrightarrow\quad \alpha(A) = \alpha(C)^\dagger\alpha(C).$$

The spectrum of every element of a unital algebra is preserved. This follows from invertibility:

$$\lambda I-A \text{ is invertible} \quad\Longleftrightarrow\quad \lambda I-\alpha(A) = \alpha(\lambda I-A) \text{ is invertible}.$$

Thus an automorphism is not merely a convenient rewriting of time evolution. It identifies the features of an observable algebra that closed dynamics must preserve.

Unitary Evolution as an Automorphism

Let $U(t,t_0)$ be the closed-system time-evolution operator. The Heisenberg evolution of an observable is

$$\alpha_{t,t_0}(A) = U^\dagger(t,t_0)AU(t,t_0).$$

This map is an automorphism. Product preservation follows directly:

$$\begin{aligned} \alpha_{t,t_0}(AB) &= U^\dagger ABU\\ &= U^\dagger A(UU^\dagger)BU\\ &= (U^\dagger AU)(U^\dagger BU)\\ &= \alpha_{t,t_0}(A)\alpha_{t,t_0}(B). \end{aligned}$$

Adjoints are preserved because

$$\alpha_{t,t_0}(A^\dagger) = U^\dagger A^\dagger U = (U^\dagger AU)^\dagger.$$

The inverse is conjugation by the opposite evolution:

$$\alpha_{t,t_0}^{-1}(A) = U(t,t_0)AU^\dagger(t,t_0).$$

In a finite-dimensional Hilbert-space representation, every automorphism of the full matrix algebra is implemented by a unitary up to an overall phase. In infinite-dimensional and algebraic settings, one must be more careful: an automorphism of the abstract algebra need not be implemented by a unitary operator in every representation.

Schrödinger and Heisenberg Duality

The same physical expectation value can be computed in two dual ways. If the state evolves by

$$\rho(t) = U(t,t_0)\rho(t_0)U^\dagger(t,t_0),$$

then

$$\operatorname{Tr}(\rho(t)A) = \operatorname{Tr}\!\left( \rho(t_0)\alpha_{t,t_0}(A) \right).$$

The Schrödinger map sends states forward. The Heisenberg automorphism sends observables in the opposite dual action. This is why product order can feel reversed when switching pictures: the observable map is defined so that expectation values stay unchanged.

In algebraic notation, a state is an expectation-value functional $\omega$. Its dual evolution can be written

$$\omega_t(A) = \omega_{t_0}(\alpha_{t,t_0}(A)).$$

This formula says that the state at time $t$ assigns to $A$ the same value that the initial state assigns to the Heisenberg-evolved observable.

Time-Translation Groups

For a time-independent Hamiltonian,

$$U(t)=e^{-iHt/\hbar},$$

and the automorphisms form a one-parameter group:

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

For this one-parameter group, $\alpha_s$ and $\alpha_t$ commute, so either order gives $\alpha_{t+s}$.

When the dependence on time is continuous in the appropriate sense, the family has an infinitesimal generator. Define

$$\delta(A) = \left. \frac{d}{dt}\alpha_t(A) \right\vert_{t=0}.$$

For Hamiltonian evolution,

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

This generator is a derivation, meaning it obeys the Leibniz rule

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

and is compatible with the adjoint:

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

The derivation property is the infinitesimal form of product preservation. It is the algebraic core of Commutator Dynamics.

For explicitly time-dependent Hamiltonians, one usually has a two-time family

$$\alpha_{t,s}(A) = U^\dagger(t,s)AU(t,s),$$

with the specific composition law

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

This reversed composition follows from $U(t,r)=U(t,s)U(s,r)$. Each map remains an automorphism, but there may be no single time-independent generator.

Constants of Motion as Fixed Points

An observable $A$ is conserved under a time-independent dynamics when it is a fixed point of every automorphism:

$$\alpha_t(A)=A \qquad \text{for all }t.$$

Equivalently,

$$\delta(A)=0.$$

For Hamiltonian evolution this becomes

$$[H,A]=0.$$

This fixed-point language is useful because it does not require choosing a basis of eigenvectors. Conserved projectors, conserved spectral subspaces, and conserved functions of $A$ are all part of the fixed-point algebra:

$$\mathcal A^\alpha = \{A\in\mathcal A:\alpha_t(A)=A \text{ for all }t\}.$$

In finite-dimensional Hamiltonian mechanics, this fixed-point algebra is the commutant of $H$:

$$\mathcal A^\alpha = \{A\in\mathcal A:[H,A]=0\}.$$

The dynamics page Constants of Motion gives the direct conservation criteria. The symmetry volume explains how such fixed points are often produced by continuous symmetries and Noether-type reasoning.

Example: Spin Precession

For a spin-half system in a static magnetic field along $z$, take

$$H=\frac{\hbar\omega}{2}\sigma_z.$$

The ladder operators

$$\sigma_\pm = \frac{1}{2}(\sigma_x\pm i\sigma_y)$$

satisfy

$$[H,\sigma_\pm] = \pm\hbar\omega\sigma_\pm.$$

Therefore

$$\alpha_t(\sigma_\pm) = e^{\pm i\omega t}\sigma_\pm.$$

Equivalently,

$$\alpha_t(\sigma_x) = \cos(\omega t)\sigma_x - \sin(\omega t)\sigma_y,$$ $$\alpha_t(\sigma_y) = \sin(\omega t)\sigma_x + \cos(\omega t)\sigma_y,$$

while

$$\alpha_t(\sigma_z)=\sigma_z.$$

The automorphism rotates the noncommuting observable algebra while leaving the field-axis component fixed. This is the Heisenberg-picture version of Larmor precession.

Example: Harmonic Oscillator

For

$$H=\hbar\omega\left(a^\dagger a+\frac{1}{2}\right),$$

the canonical operators obey

$$[H,a]=-\hbar\omega a, \qquad [H,a^\dagger]=\hbar\omega a^\dagger.$$

Thus

$$\alpha_t(a)=e^{-i\omega t}a, \qquad \alpha_t(a^\dagger)=e^{i\omega t}a^\dagger.$$

The number operator $N=a^\dagger a$ is fixed:

$$\alpha_t(N)=N.$$

This shows two typical features at once: ladder operators acquire phases, while observables commuting with the Hamiltonian lie in the fixed-point algebra.

Discrete-Time Automorphisms

A single unitary step $U$ also defines an automorphism:

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

Repeated steps give

$$\alpha^n(A) = (U^\dagger)^nAU^n.$$

This is the algebraic form of a closed quantum map. Floquet operators, kicked systems, product-formula steps, and ideal circuit layers all fit this pattern when the system is closed and the step is unitary.

The important distinction is that a discrete-time automorphism need not have a unique Hamiltonian logarithm. If $U=e^{-iH_{\mathrm{eff}}T/\hbar}$, then $H_{\mathrm{eff}}$ is branch-dependent. The automorphism $\alpha(A)=U^\dagger AU$ is unambiguous even when the effective Hamiltonian is not.

Automorphisms Versus Channels

Open-system evolution usually acts on observables by a unital completely positive map or on states by a trace-preserving completely positive map. Such maps preserve positivity and probabilities, but they do not usually preserve products:

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

For example, noise can shrink Pauli operators while leaving the identity fixed. That cannot be an automorphism of the Pauli algebra, because product relations and spectra would not be preserved.

This is why automorphism is the correct word for reversible closed dynamics, while channel is the correct word for general irreversible or noisy dynamics. A unitary channel is the special case that is both a channel on states and, dually, an automorphism on observables.

Algebraic Formulation Preview

The abstract continuation is Dynamics as Automorphisms. The preview here records why the concrete Heisenberg construction extends naturally to that setting.

The algebraic formulation begins from $\mathcal A$ and its states rather than from a preferred Hilbert space. Dynamics is then a family of automorphisms

$$t\longmapsto\alpha_t$$

of the observable algebra.

This language becomes essential in infinite systems. The Hamiltonian may be unbounded or may not belong to the algebra of bounded observables. A dynamics can still be defined as an automorphic flow on the algebra, with a generator understood as a densely defined derivation. In a chosen representation, the same automorphism may or may not be implemented by a unitary operator.

Those subtleties are not needed for ordinary finite-dimensional calculations, but the vocabulary is worth learning early. It explains why the Heisenberg equation, Liouvillian superoperators, symmetries, constants of motion, and algebraic quantum mechanics are not separate topics. They are different ways of organizing the same structure-preserving evolution.

Common Mistakes

• Calling any linear operator map an automorphism. It must preserve products, adjoints, identity, and be invertible.
• Confusing the Schrödinger state map with the Heisenberg observable automorphism; they are dual, not identical.
• Assuming every automorphism is implemented by a unitary in every representation. This is safe for full finite-dimensional matrix algebras, not as a general algebraic statement.
• Treating a noisy quantum channel as an automorphism. Most channels preserve positivity but not products or spectra.
• Forgetting that a time-dependent Hamiltonian gives a two-time family of automorphisms rather than a single one-parameter group.
• Assuming a discrete-time automorphism has a unique effective Hamiltonian.

References

• O. Bratteli and D. W. Robinson, Operator Algebras and Quantum Statistical Mechanics 1, 2nd ed., Springer, 1987.
• 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, Vol. I, Academic Press, 1983.
• A. Nayak and P. Sen, “Invertible Quantum Operations and Perfect Encryption of Quantum States”, arXiv:quant-ph/0605041, 2006.
• M. Reed and B. Simon, Methods of Modern Mathematical Physics I: Functional Analysis, revised and enlarged ed., Academic Press, 1980.
• J. J. Sakurai and J. Napolitano, Modern Quantum Mechanics, 3rd ed., Cambridge University Press, 2020.
• R. Shankar, Principles of Quantum Mechanics, 2nd ed., Springer, 1994.

Exercises

1. Verify directly that unitary Heisenberg evolution preserves products and adjoints.

Solution

Let $\alpha(A)=U^\dagger AU$ with $U^\dagger U=UU^\dagger=I$. Then

$$\begin{aligned} \alpha(AB) &= U^\dagger ABU\\ &= U^\dagger A(UU^\dagger)BU\\ &= (U^\dagger AU)(U^\dagger BU)\\ &= \alpha(A)\alpha(B). \end{aligned}$$

Also,

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

So unitary conjugation is a $*$-automorphism of the operator algebra.

2. Show that the infinitesimal generator $\delta(A)=\frac{i}{\hbar}[H,A]$ is a derivation.

Solution

Use the commutator identity

$$[H,AB] = [H,A]B+A[H,B].$$

Multiplying by $i/\hbar$ gives

$$\begin{aligned} \delta(AB) &= \frac{i}{\hbar}[H,AB]\\ &= \frac{i}{\hbar}[H,A]B \;+ A\frac{i}{\hbar}[H,B]\\ &= \delta(A)B+A\delta(B). \end{aligned}$$

This is the Leibniz rule, the infinitesimal version of product preservation.

3. Suppose $H$ is time independent and $[H,A]=0$. Prove that $\alpha_t(f(A))=f(A)$ for any polynomial $f$.

Solution

Since $[H,A]=0$, the observable itself is fixed:

$$\alpha_t(A) = U^\dagger(t)AU(t) = A.$$

Because $\alpha_t$ preserves products and sums,

$$\alpha_t(A^n) = \alpha_t(A)^n = A^n.$$

For a polynomial

$$f(A)=\sum_n c_nA^n,$$

linearity gives

$$\alpha_t(f(A)) = \sum_n c_n\alpha_t(A^n) = \sum_n c_nA^n = f(A).$$

The spectral projections of $A$ are the limiting functional-calculus version of the same idea.

4. Let $\Phi$ be the qubit depolarizing Heisenberg map

$$\Phi(I)=I, \qquad \Phi(\sigma_j)=\lambda\sigma_j, \qquad 0\lt\lambda\lt1.$$

Show that $\Phi$ is not an automorphism. Is it nevertheless invertible as a linear map? Test whether its inverse preserves positivity.

Solution

An automorphism must preserve products. But

$$\sigma_x\sigma_y=i\sigma_z.$$

Therefore

$$\Phi(\sigma_x\sigma_y) = i\lambda\sigma_z,$$

while

$$\Phi(\sigma_x)\Phi(\sigma_y) = (\lambda\sigma_x)(\lambda\sigma_y) = i\lambda^2\sigma_z.$$

For $0\lt\lambda\lt1$, these are not equal. This is a valid noisy Heisenberg-picture channel, but it is not a product-preserving automorphism.

Its linear inverse does exist: it fixes $I$ and sends $\sigma_j\mapsto\sigma_j/\lambda$. Applying it to the positive projector $P_0=(I+\sigma_z)/2$ gives

$$\Phi^{-1}(P_0)=\frac12\left(I+\frac{\sigma_z}{\lambda}\right), \qquad \operatorname{spec}\Phi^{-1}(P_0) =\left\{\frac{1+1/\lambda}{2},\frac{1-1/\lambda}{2}\right\}.$$

The second eigenvalue is negative. Thus the linear inverse is not positive, and cannot undo the noise as a channel on arbitrary inputs. For example, $\lambda=1/2$ gives eigenvalues $3/2$ and $-1/2$. The full depolarizing-channel description gives its physical parameter range and state-space contraction.

5. For $H=\hbar\omega\sigma_z/2$, compute $\alpha_t(\sigma_x)$ using commutators.

Solution

The Heisenberg equation gives

$$\frac{d}{dt}\alpha_t(\sigma_x) = \frac{i}{\hbar}\alpha_t([H,\sigma_x]).$$

Since

$$[\sigma_z,\sigma_x]=2i\sigma_y,$$

one has

$$[H,\sigma_x] = i\hbar\omega\sigma_y.$$

Thus

$$\frac{d}{dt}\alpha_t(\sigma_x) = -\omega\alpha_t(\sigma_y).$$

Similarly,

$$\frac{d}{dt}\alpha_t(\sigma_y) = \omega\alpha_t(\sigma_x).$$

With initial conditions $\alpha_0(\sigma_x)=\sigma_x$ and $\alpha_0(\sigma_y)=\sigma_y$, the solution is

$$\alpha_t(\sigma_x) = \cos(\omega t)\sigma_x - \sin(\omega t)\sigma_y.$$