When are crossed products classifiable in the Elliott program?

Classifiable means nuclear, simple, and Z-stable

The paper works inside the Elliott program, whose classification theorem covers simple, separable, unital, nuclear, Z-stable C*-algebras satisfying the Universal Coefficient Theorem (UCT). Such algebras are completely classified by their K-theories and traces, together known as the Elliott invariant. The survey calls a C*-algebra classifiable when it meets exactly these assumptions.

For the crossed products considered here, two of the hypotheses come for free. Unitality of C(X)⋊G is equivalent to compactness of X together with discreteness of G, and separability is equivalent to metrizability of X together with countability of G. So the real work is deciding nuclearity, simplicity, and Z-stability, and translating each into the language of the action G↷X.

Nuclearity is equivalent to amenability of the action

The first structural result the survey establishes is that nuclearity of C(X)⋊G is equivalent to amenability of the action G↷X, a property introduced by Anantharaman-Delaroche. This is stated as Theorem 3.2.8, with the definition of amenability of an action in Definition 3.2.1.

The equivalence matters because it converts an operator-algebraic regularity condition into an intrinsic dynamical one. Once the action is amenable, the crossed product also satisfies the UCT, so the UCT hypothesis of the classification theorem is not an extra obstacle in this setting. The survey notes that amenability of the action is therefore a necessary dynamical condition for classifiability, not an optional convenience.

Simplicity splits into minimality and topological freeness

Assuming the action is amenable, simplicity of C(X)⋊G is equivalent to the combination of two dynamical properties: minimality and topological freeness. This is Theorem 3.3.9 in the survey. Neither condition alone suffices; the equivalence is stated for the pair.

The survey also connects traces on the crossed product to invariant probability measures on X. Theorem 3.5.2 describes this relationship, and Proposition 3.5.7 shows that for an amenable dynamical system, every trace on C(X)⋊G has the form τ_μ for some G-invariant probability measure μ precisely when the action is essentially free. Essential freeness is defined by requiring that for every non-identity group element g and every invariant probability measure μ, the fixed-point set X^g has μ-measure zero. This is a measure-theoretic weakening of freeness, and it is the condition under which the trace space is canonically the space of invariant measures.

Z-stability is the only condition left to translate

With nuclearity and simplicity accounted for, the only remaining hypothesis of the classification theorem to express dynamically is tensorial absorption of the Jiang-Su algebra Z. The survey does not review the construction of Z, because the property that matters is absorbing Z tensorially, and in the simple nuclear case that absorption condition can be characterized without referring to Z at all, using only matrix algebras (Theorem 3.4.7). In practice the paper takes this criterion as the definition of Z-stability.

This is where the survey stops being a closed account and becomes a report on an active frontier. The paper states that relating Z-stability to dynamical properties of the system is not finished, and that research effort is increasingly directed at completing it. The organizing principle for the later chapters is two conjectures that are believed to capture the full picture.

Two conjectures split amenable from nonamenable groups

Conjecture 1.0.1 covers free, minimal actions of amenable groups. It asserts that C(X)⋊G is Z-stable if and only if G↷X has the small boundary property, meaning there is a basis for the topology of X such that every basis element has boundary of measure zero for every G-invariant probability measure on X (Definition 4.4.1).

Conjecture 1.0.2 covers nonamenable groups. It asserts that if G↷X is an amenable, topologically free, minimal action of a nonamenable group, then C(X)⋊G is Z-stable. The expectation here is that Z-stability is automatic for actions of nonamenable groups, which the survey describes as perhaps surprising.

The two conjectures are not symmetric. For amenable groups, Z-stability is conjecturally equivalent to a nontrivial dynamical property that can fail. For nonamenable groups, it is conjectured to hold with no additional dynamical hypothesis beyond amenability, topological freeness, and minimality of the action.

Z-stability can fail even for amenable groups

The amenable-group conjecture is not vacuous. Giol and Kerr produced the first examples of minimal, free actions of the integers whose crossed products are not Z-stable. Since the integers are amenable, these examples show that Z-stability can indeed fail when G is amenable, and that the small boundary property in Conjecture 1.0.1 is doing real work rather than being automatically satisfied.

This is the sharpest contrast in the survey. For amenable groups, the question is which actions have the small boundary property, and the answer is not all of them. For nonamenable groups, the conjecture predicts the opposite behavior, with Z-stability expected without any extra dynamical input. The survey collects evidence on both sides and isolates the remaining obstacles.

Kirchberg's dichotomy takes a dynamical form here

Kirchberg's dichotomy says that a classifiable C*-algebra is either stably finite or purely infinite, and the classification theorem splits into these two cases. In the crossed-product setting the dichotomy simplifies: a simple, nuclear crossed product C(X)⋊G is stably finite if and only if G is amenable (Proposition 3.4.15).

So amenability of the group, not just of the action, decides which half of the classification theorem applies. This is a separate use of amenability from the equivalence with nuclearity of the action, and the survey keeps the two distinct. The later chapters on paradoxical groups and purely infinite C*-algebras follow the purely infinite side of this split.

The open end is Z-stability, not nuclearity or simplicity

The survey's own framing is that amenability, minimality, and topological freeness are necessary dynamical conditions for classifiability, and that the remaining challenge is to determine when C(X)⋊G is Z-stable. That is the priority the paper sets: not re-deriving nuclearity or simplicity, but closing the gap on Z-stability.

For a reader deciding where to look, the decisive split is by group. If G is amenable, the question is whether the action has the small boundary property, and Giol and Kerr's integer examples show the answer can be no. If G is nonamenable, the conjecture predicts Z-stability whenever the action is amenable, topologically free, and minimal, with no further dynamical condition. The survey presents both conjectures as the believed full picture while noting that the machinery and techniques are still being assembled.

Sources used in this answer