Original text: CC0 1.0. Prerequisite proofs and component terms.

Kernels, local fixed parts, and freeness

CC0 1.0.

The largest identity-part projection and intertwining-multiplier theorem supplies the projection used in the covariance and abelian-action proofs below.

An action can be ergodic even when some nonidentity group elements do nothing. The right first step is to identify the kernel. A second issue is local: an automorphism can move part of an algebra while acting identically on a nonzero summand. We will detect that summand by a projection, before choosing any measure-space realization.

OA-FLOW.KERNEL.FOUNDATIONS — Setting and exact prerequisites

Let MM be a nonzero commutative von Neumann algebra with identity 11. All automorphisms are unital and normal. Initially GG is an arbitrary group with identity eGe_G, and

α:G⟶Aut⁡(M),αgh=αgαh \alpha:G\longrightarrow\operatorname{Aut}(M),\qquad \alpha_{gh}=\alpha_g\alpha_h

is a group homomorphism. No topology, countability, separable predual, faithful normal state, or measure-space presentation is required for the algebraic results.

We use the elementary realization of MM as a strongly closed algebra on a Hilbert space, bounded functional calculus, and the fact that a von Neumann algebra's normal functionals separate its elements. The required projection and support arguments are proved in the largest identity-part projection theorem. These basic operator-algebra foundations are prerequisites, not new proofs of the representation theorem for abstract von Neumann algebras. The point-space statements additionally use countable separating families for standard Borel spaces and elementary measure theory; the periodic-flow example uses Fubini's theorem for finite Lebesgue measure. No modular theory, operator-valued weight, disintegration theorem, or crossed-product theorem is imported.

For the periodic-flow continuity argument, the precise additional measure-theory import is L∞(R/Z,m)∗=L1(R/Z,m)L^\infty(\mathbb R/\mathbb Z,m)_*=L^1(\mathbb R/\mathbb Z,m) under integration, where mm is normalized Lebesgue measure. Thus testing against every L1L^1 density tests every normal functional. This foundational identification is not proved in this unit.

Write

Mα={x∈M:αg(x)=x for every g∈G},K=ker⁡α={g∈G:αg=id⁡M}. M^\alpha=\{x\in M:\alpha_g(x)=x\text{ for every }g\in G\}, \qquad K=\ker\alpha=\{g\in G:\alpha_g=\operatorname{id}_M\}.

The action is ergodic when Mα=C1M^\alpha=\mathbb C1 and faithful when K={eG}K=\{e_G\}. A nonzero projection pp is an identity part for an automorphism β\beta when β(p)=p\beta(p)=p and the restriction of β\beta to MpMp is the identity. We call β\beta free when it has no identity part. The action is free when αg\alpha_g is free for every g≠eGg\ne e_G.

This is a definition on the algebra. A pointwise action and a common conull set of free points are separate objects, treated below. Also, MβM^\beta records globally fixed elements; it does not itself identify the region where every element is fixed.

OA-FLOW.KERNEL.COVARIANCE — Transport and commuting symmetries

Proposition. For β,γ∈Aut⁡(M)\beta,\gamma\in\operatorname{Aut}(M),

pγβγ−1=γ(pβ).(4) p_{\gamma\beta\gamma^{-1}}=\gamma(p_\beta). \tag{4}

If a family of automorphisms commutes with β\beta and its common fixed algebra is C1\mathbb C1, then β\beta is either the identity or free.

Proof. Apply γ\gamma to β(x)pβ=xpβ\beta(x)p_\beta=xp_\beta and replace γ(x)\gamma(x) by an arbitrary y∈My\in M. The result says that γ(pβ)\gamma(p_\beta) satisfies the defining identity for γβγ−1\gamma\beta\gamma^{-1}. Thus γ(pβ)≤pγβγ−1\gamma(p_\beta)\leq p_{\gamma\beta\gamma^{-1}}. Applying the same reasoning with γ−1\gamma^{-1} gives the reverse inequality. If γ\gamma commutes with β\beta, (4) gives γ(pβ)=pβ\gamma(p_\beta)=p_\beta. Under the stated common fixed-algebra hypothesis, pβp_\beta is a scalar projection and therefore is 00 or 11. Equation (1) says that pβ=1p_\beta=1 is equivalent to β=id⁡\beta=\operatorname{id}; the preceding proposition says that pβ=0p_\beta=0 is equivalent to freeness. □\square

For completeness, ergodicity can equivalently be stated by saying that the only projections fixed by every automorphism in the family are 00 and 11. One implication is immediate. For the other, any common fixed self-adjoint element has all its spectral projections fixed, by normality and spectral functional calculus. If it were nonscalar, a spectral cut strictly between two points of its spectrum would be a nonzero proper fixed projection. Thus every common fixed self-adjoint element is scalar. Taking real and imaginary parts proves that the common fixed algebra is C1\mathbb C1.

OA-FLOW.KERNEL.ABELIAN — The ergodic dichotomy and effective quotient

Theorem. Let GG be an arbitrary abelian group acting ergodically on MM. Then

pαg={1,g∈K,0,g∉K.(5) p_{\alpha_g}= \begin{cases} 1,&g\in K,\\ 0,&g\notin K. \end{cases} \tag{5}

Consequently, the action is free if and only if it is faithful. For an arbitrary kernel KK, the action

α‾:G/K⟶Aut⁡(M),α‾gK=αg(6) \overline\alpha:G/K\longrightarrow\operatorname{Aut}(M), \qquad \overline\alpha_{gK}=\alpha_g \tag{6}

is well defined, faithful, ergodic, and free.

Proof. Every αh\alpha_h commutes with αg\alpha_g. The preceding proposition applies to the ergodic family α(G)\alpha(G), so each pαgp_{\alpha_g} is either 00 or 11. Its value is 11 exactly when αg\alpha_g is the identity, that is, when g∈Kg\in K. This proves (5).

If the action is faithful, no g≠eGg\ne e_G lies in KK, and (5) proves freeness. Conversely, a nonidentity element of KK fixes the nonzero projection 11 pointwise, contradicting freeness. This converse does not require ergodicity or commutativity of GG.

As the kernel of a homomorphism, KK is a normal subgroup. If gK=hKgK=hK, then h−1g∈Kh^{-1}g\in K, so αg=αh\alpha_g=\alpha_h; hence (6) is well defined and is a homomorphism. If α‾gK=id⁡\overline\alpha_{gK}=\operatorname{id}, then g∈Kg\in K, proving faithfulness. The two actions have the same set of automorphisms and therefore the same fixed algebra. Finally G/KG/K is abelian, so the already proved faithful case applies. □\square

The proof does not use a countable union of projections or null sets. In particular, it applies when GG is uncountable or MM has nonseparable predual. If M=CM=\mathbb C, every action is ergodic and K=GK=G; the effective quotient is the trivial group, whose freeness is vacuous. If GG is not abelian, (4) still gives

αh(pαg)=pαhgh−1.(7) \alpha_h(p_{\alpha_g})=p_{\alpha_{hgh^{-1}}}. \tag{7}

Thus central group elements satisfy the same dichotomy under ergodicity. General elements need not do so.

Editable source · Proof dependencies and component terms