39.05.10 · operator-algebras / nuclearity-exactness

Exact Groups, Amenable Actions, and Property A

shipped3 tiersLean: none

Anchor (Master): Brown-Ozawa *C*-Algebras and Finite-Dimensional Approximations* (AMS, 2008) Ch. 4-5, 15; Ozawa 2000 *C. R. Acad. Sci.* 330; Yu *The coarse Baum-Connes conjecture for spaces which admit a uniform embedding into Hilbert space* (Invent. Math. 139, 2000); Higson-Roe *Amenable group actions and the Novikov conjecture* (J. reine angew. Math. 519, 2000); Guentner-Higson-Weinberger *The Novikov conjecture for linear groups* (Publ. IHES 101, 2005)

Intuition Beginner

The previous units found that a group is amenable when it can average over itself: it can spread a fixed, fair weighting across its own elements that no symmetry of the group can disturb. Amenability is generous, and many interesting groups do not have it. The free group on two letters, which branches like a tree, cannot average over itself at all.

But there is a softer thing such a group might still manage. Even if a group cannot average over itself alone, it might be able to average once it is handed a little extra room to work in — a stage built at the edge of the group, far out at infinity, where its branches finally settle down. The group spreads its weight not over itself but over points of this edge, and now the spreading can be made fair in the limit. Averaging with help, rather than averaging alone.

A group that can average with this kind of help is called exact. Every averaging group is exact, because it never needed the help in the first place. The surprise is that almost every group you meet — free groups, the symmetry groups of tilings of curved space, groups of matrices — is exact even when it is nowhere near being an averaging group. Exactness is the wide, forgiving property; genuine averaging is the narrow, demanding one.

There is also a purely geometric way to spot an exact group, by looking at the group as a landscape of points and measuring how its faraway regions can be tidied up. That landscape test, and the averaging-with-help test, turn out to name the same groups.

Visual Beginner

The picture shows the same group answering two questions. On the left it tries to average over itself alone and fails; on the right it averages with the help of a stage built at its edge and succeeds.

The dictionary reads: averaging-with-help on the edge is an amenable action on a boundary; the landscape test on faraway patches is property A; and the operator-algebra statement is that the group's reduced algebra sits inside a tame one. The three panels are one property wearing three coats.

Worked example Beginner

We watch the free group on two letters $a$ and $b$ average with help, where averaging alone is impossible. The group is a tree: from a central point, four roads lead out (toward $a$ , toward $a^{- 1}$ , toward $b$ , toward $b^{- 1}$ ), and each new junction sprouts three fresh roads, forever. The edge at infinity is the set of endless roads — pick a direction and keep walking, never turning back.

A point of the group sits somewhere on this tree. To average with help, we hand each such point a recipe for spreading weight toward the edge: from where you stand, push all your weight a fixed distance $R$ further out along the unique road that heads to a chosen end. Two points that sit close together, a few junctions apart, push their weight toward almost the same far-out region, because their roads merge once you walk far enough.

Now measure the disagreement. Two neighbouring points produce weight-spreads that overlap almost completely; only the first few junctions, where their roads had not yet merged, differ. As we let the push-distance $R$ grow, that mismatch shrinks toward nothing. The spreading has become fair in the limit: nearby points average to nearly the same thing.

What this tells us: the free group, which can never average over itself, averages cleanly once it is allowed to throw its weight onto the edge of its tree. The merging of roads far out is exactly the help that makes the group exact, even though no fair self-average will ever exist.

Check your understanding Beginner

Exercise (easy, multiple choice).

What does it mean for a group to be exact, in the averaging picture?

A. It can average fairly over itself, with no outside help

B. It can average fairly once it is allowed to spread its weight onto a stage built at its edge at infinity

C. It has finitely many elements

D. It cannot average at all, under any conditions

Hint

Exactness is the softer, more forgiving cousin of genuine averaging: help is allowed.

Answer

B. Exactness is averaging-with-help: the group spreads its weight over points of an edge built at infinity, and the spreading can be made fair in the limit. Feedback-correct: this is an amenable action on a boundary, and every genuinely averaging group is exact because it never needed the help. Feedback-wrong: option A is the stronger property of amenability; exactness has nothing to do with the number of elements, and most non-averaging groups are still exact, so option D is wrong.

Exercise (easy, multiple choice).

In the worked example, what made the free group's averaging-with-help become fair in the limit?

A. The roads of nearby points merge once you walk far enough toward the edge

B. The tree has only finitely many junctions

C. The group has a fair self-average after all

D. The weight was erased instead of spread

Hint

Two neighbouring points push their weight toward almost the same far-out region.

Answer

A. Far out along the tree the roads of two nearby points merge, so their weight-spreads overlap almost completely, and the mismatch shrinks as the push-distance grows. Feedback-correct: the merging of roads at infinity is the help that makes the group exact. Feedback-wrong: the tree is infinite, no fair self-average exists, and the weight is spread, never erased.

Formal definition Intermediate+

Throughout, $Γ$ is a countable discrete group with a fixed finite generating set $S = S^{- 1}$ , $∣ \cdot ∣$ the associated word length, and $d (x, y) = ∣ x^{- 1} y ∣$ the left-invariant word metric. We write $Prob (Γ) \subseteq ℓ^{1} (Γ)$ for the probability measures (positive, $ℓ^{1}$ -norm one) and $ℓ^{1} (Γ)_{1, +}$ for the same set. The group acts on $ℓ^{1} (Γ)$ by $(g \cdot μ) (t) = μ (g^{- 1} t)$ .

Definition (exact group). $Γ$ is exact if its reduced C*-algebra $C_{r}^{*} (Γ)$ is an exact C*-algebra in the sense of 39.05.05: the functor $- \otimes_{m i n} C_{r}^{*} (Γ)$ preserves short exact sequences. By Kirchberg's theorem this is the same as $C_{r}^{*} (Γ)$ being nuclearly embeddable.

Definition (topologically amenable action). Let $Γ$ act on a compact Hausdorff space $X$ by homeomorphisms. The action is (topologically) amenable if there is a net of weak- $*$ continuous maps $$ m_n : X \longrightarrow \mathrm{Prob}(\Gamma), \qquad x \mapsto m_n^x, $$ such that for every $g \in Γ$ , $$ \sup_{x \in X}, \big| g\cdot m_n^{x} - m_n^{,g x} \big|_1 \xrightarrow[n]{} 0 . $$ Each $m_{n}^{x}$ is a probability measure on $Γ$ depending continuously on $x$ , and the displayed condition asks the assignment to be $Γ$ -equivariant in the limit, uniformly in $x$ . A group $Γ$ is boundary amenable (or exact in the dynamical sense) if it admits an amenable action on some compact Hausdorff space; the action on the Stone-Čech-type universal boundary may always be taken, and amenability of any such action implies amenability of the action on every compact $Γ$ -space generated suitably.

Definition (Yu's property A). $Γ$ has property A if for every $R > 0$ and $ε > 0$ there exist a map $$ \xi : \Gamma \longrightarrow \ell^1(\Gamma)_{1,+}, \qquad x \mapsto \xi_x, $$ and a constant $L \geq 0$ such that (i) $∥ ξ_{x} - ξ_{y} ∥_{1} \leq ε$ whenever $d (x, y) \leq R$ , and (ii) each $ξ_{x}$ is supported in the ball ${t : d (x, t) \leq L}$ of radius $L$ about $x$ . The uniform bound $L$ on support diameter, independent of $x$ , is the geometric content; condition (i) is a Følner-type slow variation of the family along the metric. Property A is a coarse invariant of the metric space $(Γ, d)$ and makes sense for any bounded-geometry metric space.

The uniform Roe algebra $C_{u}^{*} (Γ)$ is the norm closure in $B (ℓ^{2} (Γ))$ of the operators of finite propagation — bounded operators $T$ with $⟨ T δ_{y}, δ_{x} ⟩ = 0$ whenever $d (x, y) > R_{T}$ for some $R_{T}$ . It records the large-scale geometry of $Γ$ and is the coarse-geometric companion of $C_{r}^{*} (Γ)$ .

Definition (reduced crossed product). For a $Γ$ -action on compact $X$ , the reduced crossed product $C (X) ⋊_{r} Γ$ is the C*-algebra generated on $ℓ^{2} (Γ) \otimes L^{2} (X, μ)$ (any quasi-invariant $μ$ ) by the copy of $C (X)$ acting by the covariant representation and the unitaries $λ (g) \otimes u_{g}$ implementing the action through the regular representation $λ$ of 39.05.07. It contains $C_{r}^{*} (Γ)$ as the subalgebra generated by the unitaries.

Counterexamples to common slips Intermediate+

Amenable action is not amenability of $Γ$ . A non-amenable group can act amenably; the free group acts amenably on its Gromov boundary. The measures $m_{n}^{x}$ live on $Γ$ but their existence is purchased by the space $X$ , not by an invariant mean on $Γ$ .
Property A is not the Følner condition. Følner sets give exact equivariance for an amenable group; property A only asks for slow variation of a family of finitely-supported probability measures along the metric, with no invariance. Amenable groups have property A, but so do free and hyperbolic groups, which have no Følner sequence.
Exactness of $Γ$ is a property of the metric space, not of the chosen generating set. Changing the finite generating set changes $d$ by a bi-Lipschitz (hence coarse) equivalence, and property A is a coarse invariant, so exactness is well defined.
Not every group is exact. Gromov monster groups contain coarsely embedded expanders, violate property A, and have non-exact $C_{r}^{*} (Γ)$ — the obstruction recorded in 39.05.05. Exactness is a genuine restriction, even if a hard one to violate.

Key theorem with proof Intermediate+

Theorem (Ozawa; Higson-Roe; Anantharaman-Delaroche). For a countable discrete group $Γ$ the following are equivalent: (i) $Γ$ is exact, i.e. $C^_r(\Gamma) $i s an e x a c tC \* - a l g e b r a; (ii)$ \Gamma $a d mi t s a t o p o l o g i c a l l y am e nab l e a c t i o n o n so m eco m p a c t H a u s d or f f s p a ce$ X $; (iii)$ \Gamma$ has Yu's property A.* ^{[Ozawa 2000; Higson-Roe 2000; Brown-Ozawa Ch. 4-5]}

Proof. (ii) $\Rightarrow$ (i). Suppose $Γ ↷ X$ is amenable, with maps $m_{n} : X \to Prob (Γ)$ . The crossed product $C (X) ⋊_{r} Γ$ is nuclear: amenability supplies completely positive contractive maps factoring its identity through matrix algebras. Concretely the functions $x \mapsto (m_{n}^{x})^{1/2} \in ℓ^{2} (Γ)$ assemble into a sequence of unital completely positive maps $Φ_{n}, Ψ_{n}$ between $C (X) ⋊_{r} Γ$ and $M_{k_{n}} (C) \otimes C (X)$ with $Ψ_{n} Φ_{n} \to id$ in point-norm, the equivariance defect $sup_{x} ∥ g \cdot m_{n}^{x} - m_{n}^{g x} ∥_{1} \to 0$ controlling the error on each generator $λ (g) \otimes u_{g}$ ; hence $C (X) ⋊_{r} Γ$ has the completely positive approximation property and is nuclear 39.05.04. Now $C_{r}^{*} (Γ) \subseteq C (X) ⋊_{r} Γ$ as the subalgebra generated by the canonical unitaries. A C*-subalgebra of a nuclear algebra is nuclearly embeddable, hence exact 39.05.05. Therefore $C_{r}^{*} (Γ)$ is exact and $Γ$ is exact.

(i) $\Rightarrow$ (iii). Suppose $C_{r}^{*} (Γ)$ is exact. Exactness gives, on the uniform Roe algebra $C_{u}^{*} (Γ) = C_{r}^{*} (Γ) ⋊ \dots$ side, finite-rank completely positive contractions approximating the identity on finite-propagation operators. Translating the matrix coefficients of these maps into kernels on $Γ \times Γ$ and applying a positive-type symmetrisation produces, for each $R, ε$ , a positive-definite kernel $u : Γ \times Γ \to R$ of finite propagation with $∣1 - u (x, y) ∣ \leq ε$ for $d (x, y) \leq R$ . The kernel factors as $u (x, y) = ⟨ ξ_{x}, ξ_{y} ⟩$ through finitely supported $ξ_{x} \in ℓ^{1} (Γ)_{1, +}$ (after normalisation), uniformly bounded support diameter $L$ coming from the propagation bound, and $∥ ξ_{x} - ξ_{y} ∥_{1} \leq 2 ε$ for $d (x, y) \leq R$ . These are exactly property-A data.

(iii) $\Rightarrow$ (ii). Suppose $Γ$ has property A with data $ξ^{(R)}$ . Let $X = \partial_{β} Γ$ be the corona of the Stone-Čech compactification $β Γ$ , on which $Γ$ acts by homeomorphisms extending left translation. Each property-A family $ξ^{(n)} : Γ \to Prob (Γ)$ extends continuously to $β Γ$ by the universal property and restricts to a weak- $*$ continuous map $m_{n} : X \to Prob (Γ)$ . The slow-variation estimate $∥ ξ_{x} - ξ_{g x} ∥_{1}$ small for the controlled displacement by $g$ becomes, in the limit at the boundary, $sup_{x \in X} ∥ g \cdot m_{n}^{x} - m_{n}^{g x} ∥_{1} \to 0$ . Thus the boundary action is amenable. $□$

Bridge. This theorem is the foundational reason a single regularity property of $Γ$ wears three coats: a dynamical coat (an amenable action on a boundary), a coarse-geometric coat (property A on the metric space), and an operator-algebraic coat (exactness of $C_{r}^{*} (Γ)$ ). It builds toward the boundary-amenability examples — hyperbolic and linear groups — and it appears again in the master-tier route from property A to the coarse Baum-Connes and Novikov conjectures. The engine is the reduced crossed product: an amenable action makes $C (X) ⋊_{r} Γ$ nuclear, and embedding $C_{r}^{*} (Γ)$ into that nuclear algebra is exactly the nuclear embeddability that 39.05.05 identifies with exactness — this is exactly the mechanism that converts dynamics into the tensor-functor condition. The implication (i) $\Rightarrow$ (iii) generalises the Følner picture of 39.05.06: property A is the slow-variation weakening of the Følner condition that survives the loss of an invariant mean, and putting these together gives the equivalence. The bridge is between the averaging-with-help of a non-amenable group and the subalgebra-stable weakening of nuclearity, and it appears again where exactness of the reduced algebra is read as a coarse invariant of the group's geometry.

Exercises Intermediate+

Exercise 3 (medium, symbolic).

Show directly that an amenable group $Γ$ has property A, using a Følner sequence $(F_{n})$ .

Hint

Set $ξ_{x} = ∣ F_{n} ∣^{- 1} 1_{x F_{n}}$ and estimate $∥ ξ_{x} - ξ_{y} ∥_{1}$ by the Følner ratio.

Answer

Fix $R, ε$ . Choose a Følner set $F$ with $∣ s F △ F ∣/∣ F ∣ \leq ε / (2 R^{'})$ for all $∣ s ∣ \leq R$ , where $R^{'}$ bounds the number of generators traversed; concretely $∣ g F △ F ∣/∣ F ∣ \leq ε$ for $d (e, g) \leq R$ after refining. Define $ξ_{x} = ∣ F ∣^{- 1} 1_{x F} \in ℓ^{1} (Γ)_{1, +}$ , a probability measure supported in the ball of radius $L = diam (F)$ about $x$ , uniform in $x$ by left invariance. For $d (x, y) \leq R$ write $y = xg$ with $∣ g ∣ \leq R$ ; then $$ |\xi_x - \xi_y|1 = |F|^{-1}|\mathbf{1}{xF} - \mathbf{1}_{xgF}|_1 = \frac{|F \triangle gF|}{|F|} \le \varepsilon . $$ Both property-A conditions hold: support diameter $L$ is uniform and the variation is at most $ε$ . Hence amenable $\Rightarrow$ property A.

Exercise 4 (medium, symbolic).

Let $Γ ↷ X$ be amenable with maps $m_{n} : X \to Prob (Γ)$ . Define $Ψ_{n} : C_{c} (Γ) \to C (X) ⋊_{r} Γ$ and verify the equivariance defect controls $∥ Ψ_{n} Φ_{n} (λ (g)) - λ (g) ∥$ on a generator.

Hint

The matrix coefficient picks up the inner product $⟨(m_{n}^{x})^{1/2}, g \cdot (m_{n}^{x})^{1/2} ⟩$ ; compare numerator and shifted measure.

Answer

Writing $η_{n}^{x} = (m_{n}^{x})^{1/2} \in ℓ^{2} (Γ)$ , the standard amenable-action CPAP sets $Φ_{n} (T) (x) =$ the compression of $T$ by $η_{n}^{x}$ and $Ψ_{n}$ the dual averaging. On a generator $λ (g) \otimes u_{g}$ , $$ \Psi_n\Phi_n(\lambda(g)\otimes u_g) = \big(x \mapsto \langle \eta_n^{gx},, g\cdot \eta_n^{x}\rangle\big),(\lambda(g)\otimes u_g), $$ so the coefficient is $1 - \frac{1}{2} ∥ η_{n}^{g x} - g \cdot η_{n}^{x} ∥_{2}^{2}$ at worst. The Powers-Størmer inequality gives $∥ η_{n}^{g x} - g \cdot η_{n}^{x} ∥_{2}^{2} \leq ∥ m_{n}^{g x} - g \cdot m_{n}^{x} ∥_{1}$ , which is $\leq sup_{x} ∥ g \cdot m_{n}^{x} - m_{n}^{g x} ∥_{1} \to 0$ . Hence the coefficient tends to $1$ uniformly in $x$ , so $∥ Ψ_{n} Φ_{n} (λ (g) \otimes u_{g}) - λ (g) \otimes u_{g} ∥ \to 0$ , and point-norm convergence on generators extends to the identity. This is the CPAP, so $C (X) ⋊_{r} Γ$ is nuclear.

Exercise 6 (hard, short-answer).

Prove that property A passes to subgroups: if $Γ$ has property A and $H \leq Γ$ , then $H$ has property A.

Hint

The inclusion $H ↪ Γ$ is a coarse (isometric for the restricted metric) embedding; restrict the property-A family and push it back to $H$ along a section of the coset projection.

Answer

Fix $R, ε$ and let $ξ : Γ \to ℓ^{1} (Γ)_{1, +}$ realise property A for $Γ$ at scale $R$ with support radius $L$ and variation $\leq ε$ . The word metric of $H$ (for a finite generating set of $H$ ) is bi-Lipschitz-dominated by the restriction of $d_{Γ}$ , so it suffices to produce a family on $H$ coarsely controlled by $d_{Γ}$ . Choose a set-theoretic section $σ : Γ/ H \to Γ$ of the right-coset projection and the induced retraction $r : Γ \to H$ sending $t$ to the $H$ -part $t σ (\overset{ˉ}{t})^{- 1}$ . For $x \in H$ define $ζ_{x} = r_{*} ξ_{x} \in ℓ^{1} (H)_{1, +}$ , the pushforward, which is supported within bounded distance of $x$ because $r$ is a coarse map of bounded displacement on balls, and satisfies $∥ ζ_{x} - ζ_{y} ∥_{1} \leq ∥ ξ_{x} - ξ_{y} ∥_{1} \leq ε$ for $d_{H} (x, y) \leq R$ , since pushforward is an $ℓ^{1}$ -contraction. Adjusting $L$ to absorb the displacement of $r$ gives property-A data for $H$ . Rubric: full credit for restricting the family along the coarse embedding $H ↪ Γ$ and using that pushforward contracts $ℓ^{1}$ -distance.

Exercise 7 (hard, short-answer).

Show that the free group $F_{2}$ has property A by exhibiting the boundary-averaging family explicitly, and name the obstruction that makes a Gromov monster group fail it.

Hint

For $x \in F_{2}$ push the unit mass a distance $R$ along the geodesic from $e$ through $x$ toward infinity; for the monster, recall that expanders cannot be coarsely embedded into Hilbert space.

Answer

For $x \in F_{2}$ at distance $∣ x ∣$ from $e$ , let $[e, x]$ denote the geodesic in the tree and define $ξ_{x}^{R}$ to be the uniform probability measure on the segment of the geodesic ray through $x$ from depth $max (0, ∣ x ∣ - R)$ outward of length $R$ — equivalently, push $x$ 's unit mass $R$ steps toward the end determined by $x$ . The support has diameter $\leq 2 R$ , uniformly. If $d (x, y) \leq 1$ then $x$ and $y$ share a geodesic prefix differing in only the last step, so the two pushed measures differ on at most a $1/ R$ fraction of their mass: $∥ ξ_{x}^{R} - ξ_{y}^{R} ∥_{1} \leq 2/ R$ , which is $\leq ε$ once $R \geq 2/ ε$ , and iterating over a path of length $\leq R^{'}$ controls $d (x, y) \leq R^{'}$ . Thus $F_{2}$ has property A. A Gromov monster group contains a sequence of expander graphs coarsely embedded in its Cayley graph; expanders have no uniform embedding into Hilbert space, and Yu's property A would supply one, so the monster violates property A and $C_{r}^{*} (Γ)$ is non-exact 39.05.05. Rubric: full credit for the geodesic-push family with the $1/ R$ overlap estimate plus the expander obstruction for non-exactness.

Advanced results Master

Exactness of a discrete group sits exactly one regularity step below amenability, and its theory is organised by the three-faces equivalence, the supply of examples that are exact without being amenable, and the topological-conjecture payoff that property A delivers.

The three faces and permanence. For countable $Γ$ , exactness of $C_{r}^{*} (Γ)$ , topological amenability of an action on some compact $Γ$ -space, and property A coincide (Ozawa) ^{[Ozawa 2000]}. The class of exact groups is closed under the operations one expects of a coarse-geometric class: subgroups, extensions $1 \to N \to Γ \to Q \to 1$ with $N$ and $Q$ exact, directed unions, free products and amalgams over exact subgroups, and direct/restricted products. The contrast with amenability is sharp: amenability is not closed under containing a free subgroup, whereas a free subgroup of an exact group keeps the ambient group exact, and the free group is the first non-amenable exact example. Property A is the lever for permanence because it is a coarse invariant, so any quasi-isometry or coarse embedding into an exact group transports exactness.

Hyperbolic groups via the boundary action. A finitely generated word-hyperbolic group $Γ$ (in Gromov's sense — geodesic triangles in the Cayley graph are $δ$ -thin) acts on its Gromov boundary $\partial Γ$ , the space of equivalence classes of geodesic rays with the visual topology, a compact metrisable $Γ$ -space. Adams proved this action is topologically amenable ^{[Adams 1994]}: the harmonic-measure / Patterson-Sullivan family $m_{n}^{ξ}$ supported on the geodesic toward $ξ$ furnishes the equivariant-in-the-limit measures, exactly the tree picture of the worked example generalised to a $δ$ -hyperbolic geometry. Hence every hyperbolic group is exact. This includes free groups, surface groups, fundamental groups of negatively curved manifolds, and random groups in the Gromov density model below density $1/2$ .

Linear groups. Guentner, Higson and Weinberger proved that every countable subgroup of $G L_{n} (K)$ , $K$ any field, is exact ^{[Guentner-Higson-Weinberger 2005]}. The proof builds an amenable action on a compact space assembled from the Furstenberg boundaries (flag varieties) of the relevant algebraic groups and, in positive characteristic and over local fields, from the Bruhat-Tits buildings; boundary amenability propagates through the field-by-field reduction. So $S L_{n} (Z)$ , mapping class groups' linear quotients, and arithmetic groups — many of them far from amenable and even with Kazhdan's property (T) — are all exact. Exactness and property (T) are compatible, which separates exactness cleanly from the Haagerup property of 39.05.07.

Property A, coarse embedding, and the conjectures. Yu introduced property A as a checkable sufficient condition for a bounded-geometry metric space to admit a uniform (coarse) embedding into Hilbert space, and proved that any such space satisfies the coarse Baum-Connes conjecture ^{[Yu 2000]}. For a group $Γ$ with property A this descends, via the Higson-Roe descent principle ^{[Higson-Roe 2000]}, to the strong Novikov conjecture: the assembly map $μ : K_{*}^{Γ} (\underline{E} Γ) \to K_{*} (C_{r}^{*} (Γ))$ is rationally injective, so the higher signatures of a closed aspherical manifold with fundamental group $Γ$ are oriented-homotopy invariants. The implication runs property A $\Rightarrow$ coarse embedding $\Rightarrow$ coarse Baum-Connes $\Rightarrow$ Novikov; none of the converses is needed here. The uniform Roe algebra $C_{u}^{*} (Γ)$ is the carrier of the coarse index, and property A is exactly the condition making it nuclear, the coarse mirror of the amenable-action nuclearity of $C (X) ⋊_{r} Γ$ .

The non-exact frontier. Property A is not universal. Gromov's monster groups contain coarsely embedded expander graphs; expanders are the canonical obstruction to coarse Hilbert-space embedding, so monster groups lack property A and have non-exact reduced C*-algebras ^{[Gromov 2003 via 39.05.05]}. These groups still satisfy the Novikov conjecture by other means in some constructions, but they show that boundary amenability is a real restriction and that the coarse-geometry route via property A genuinely fails for them. Whether every group satisfies Novikov remains open; exactness is the widest cleanly-characterised class for which the boundary-amenability route closes it.

Synthesis. Boundary amenability is the central insight that fuses three theories: the dynamics of a group acting on a boundary, the coarse geometry of the group as a metric space, and the tensor-functor regularity of its reduced C*-algebra. The foundational reason these coincide is the reduced crossed product $C (X) ⋊_{r} Γ$ : an amenable action makes it nuclear by a completely positive approximation built from the equivariant-in-the-limit measures, and the embedding $C_{r}^{*} (Γ) ↪ C (X) ⋊_{r} Γ$ is exactly the nuclear embeddability that 39.05.05 identifies with exactness — so dynamics becomes the tensor condition. The central insight, read at several levels, is that property A is the slow-variation weakening of the Følner condition of 39.05.06 that survives the death of an invariant mean: it generalises amenability by spreading a family of finitely-supported measures along the metric rather than demanding one invariant mean, and it is dual to the boundary picture through the Stone-Čech extension that turns slow variation into limiting equivariance. Putting these together, hyperbolic groups (Adams) and linear groups (Guentner-Higson-Weinberger) are exact though wildly non-amenable, the free group of 39.05.07 is the first such example, and the payoff is Yu's route from property A through coarse embedding to the coarse Baum-Connes and Novikov conjectures. The bridge from the amenable, Følner-averaging world of 39.05.06 to the non-amenable but boundary-amenable world here is the entire content the nuclearity-exactness chapter measures: where amenability collapses the two completions, exactness only plants the reduced one inside a tame host, and the monster groups mark the edge where even that fails.

Full proof set Master

Proposition 1 (amenable action $\Rightarrow$ nuclear crossed product). If $Γ ↷ X$ is topologically amenable on a compact Hausdorff space $X$ , then $C (X) ⋊_{r} Γ$ is nuclear.

Proof. Let $m_{n} : X \to Prob (Γ)$ witness amenability and set $η_{n}^{x} = (m_{n}^{x})^{1/2} \in ℓ^{2} (Γ)$ , a unit vector depending weak- $*$ continuously on $x$ , supported on a finite set $E_{n} (x)$ for the approximating compactly-supported version (truncate $m_{n}$ and renormalise; the truncation error is uniformly small). Define $$ \Phi_n : C(X)\rtimes_r\Gamma \to \mathbb{M}{E_n}\otimes C(X), \qquad \Psi_n : \mathbb{M}{E_n}\otimes C(X)\to C(X)\rtimes_r\Gamma, $$ with $Φ_{n} (T) = [x \mapsto (⟨ T (δ_{s} \otimes \cdot), δ_{t} \otimes \cdot ⟩)_{s, t \in E_{n} (x)}]$ the compression read against the $η_{n}^{x}$ -frame, and $Ψ_{n}$ the dual averaging $Ψ_{n} (e_{s t} \otimes f) = \overline{η_{n} (s)} η_{n} (t) f (λ (s t^{- 1}) \otimes u_{s t^{- 1}})$ . Both are unital completely positive: $Φ_{n}$ is a compression of a $*$ -representation 39.05.02 and $Ψ_{n}$ is a state-weighted sum of the completely positive maps implemented by the covariant pairs. On a generator $a (λ (g) \otimes u_{g})$ with $a \in C (X)$ , $$ \Psi_n\Phi_n\big(a(\lambda(g)\otimes u_g)\big) = \big(x\mapsto a(x)\langle \eta_n^{gx}, g\cdot\eta_n^x\rangle\big)(\lambda(g)\otimes u_g), $$ and $1 - ⟨ η_{n}^{g x}, g \cdot η_{n}^{x} ⟩ \leq \frac{1}{2} ∥ η_{n}^{g x} - g \cdot η_{n}^{x} ∥_{2}^{2} \leq \frac{1}{2} ∥ m_{n}^{g x} - g \cdot m_{n}^{x} ∥_{1}$ by Powers-Størmer, which tends to $0$ uniformly in $x$ by amenability. Hence $Ψ_{n} Φ_{n} \to id$ point-norm on generators, and by uniform contractivity $∥ Ψ_{n} Φ_{n} ∥ \leq 1$ on all of $C (X) ⋊_{r} Γ$ . This is the completely positive approximation property, so $C (X) ⋊_{r} Γ$ is nuclear 39.05.04. $□$

Proposition 2 (amenable action $\Rightarrow$ $Γ$ exact). If $Γ$ admits a topologically amenable action on a compact Hausdorff space, then $C_{r}^{*} (Γ)$ is exact.

Proof. By Proposition 1 the crossed product $N := C (X) ⋊_{r} Γ$ is nuclear. The canonical unitaries $λ (g) \otimes u_{g}$ generate a copy of $C_{r}^{*} (Γ)$ inside $N$ : the conditional expectation $E : N \to C_{r}^{*} (Γ)$ obtained by integrating against the unit of $C (X)$ restricts the inclusion to a faithful $*$ -isomorphism onto its image, so $C_{r}^{*} (Γ) \subseteq N$ isometrically. A C*-subalgebra of a nuclear C*-algebra is nuclearly embeddable, hence exact by Kirchberg's theorem 39.05.05. Therefore $Γ$ is exact. $□$

Proposition 3 (amenable $\Rightarrow$ property A). Every amenable group has property A.

Proof. Given $R, ε$ , pick a Følner set $F$ with $∣ g F △ F ∣/∣ F ∣ \leq ε$ for all $g$ with $∣ g ∣ \leq R$ 39.05.06. Put $ξ_{x} = ∣ F ∣^{- 1} 1_{x F} \in ℓ^{1} (Γ)_{1, +}$ ; left invariance makes the support diameter $L = diam (F)$ uniform in $x$ . For $y = xg$ with $∣ g ∣ \leq R$ , $$ |\xi_x-\xi_y|1 = |F|^{-1},|\mathbf 1{xF}-\mathbf 1_{xgF}|_1 = \frac{|F\triangle gF|}{|F|}\le\varepsilon, $$ using $(1_{A} - 1_{B})$ summed in absolute value equals $∣ A △ B ∣$ and left invariance of counting. Both property-A conditions hold. $□$

Proposition 4 (property A is a coarse invariant). If $f : (Γ, d_{Γ}) \to (Λ, d_{Λ})$ is a coarse equivalence of bounded-geometry metric spaces, then $Γ$ has property A iff $Λ$ does.

Proof. Coarse equivalence gives a coarse inverse $\overset{ˉ}{f}$ and control functions $ρ_{\pm}$ with $ρ_{-} (d_{Γ} (x, y)) \leq d_{Λ} (f x, f y) \leq ρ_{+} (d_{Γ} (x, y))$ , $ρ_{\pm}$ proper. Suppose $Λ$ has property A with family $ζ^{(R)} : Λ \to ℓ^{1} (Λ)_{1, +}$ . For $x \in Γ$ set $ξ_{x} = (\overset{ˉ}{f})_{*} ζ_{f (x)} \in ℓ^{1} (Γ)_{1, +}$ , the pushforward by $\overset{ˉ}{f}$ . The support radius is bounded by $ρ_{+}^{- 1}$ of $Λ$ 's support radius plus the displacement of $\overset{ˉ}{f}$ , uniform because $\overset{ˉ}{f}$ is bornologous; and for $d_{Γ} (x, y) \leq R$ we have $d_{Λ} (f x, f y) \leq ρ_{+} (R)$ , so $∥ ξ_{x} - ξ_{y} ∥_{1} \leq ∥ ζ_{f x} - ζ_{f y} ∥_{1} \leq ε$ since pushforward contracts $ℓ^{1}$ -distance. Thus $Γ$ has property A. Symmetry of the argument gives the converse. In particular property A is independent of the finite generating set, since two choices give bi-Lipschitz-equivalent, hence coarsely equivalent, metrics. $□$

Proposition 5 (property A $\Rightarrow$ exactness, via positive-type kernels). If $Γ$ has property A then $C_{r}^{*} (Γ)$ is exact.

Proof. Property A produces, for each $R, ε$ , finitely supported $ξ_{x} \in ℓ^{1} (Γ)_{1, +}$ of uniform support radius $L$ with $∥ ξ_{x} - ξ_{y} ∥_{1} \leq ε$ for $d (x, y) \leq R$ . Replace $ξ_{x}$ by $η_{x} = ξ_{x}^{1/2} \in ℓ^{2} (Γ)$ ; then $u (x, y) := ⟨ η_{x}, η_{y} ⟩$ is a positive-definite kernel of finite propagation ( $u (x, y) = 0$ when $d (x, y) > 2 L$ ) with $∣1 - u (x, y) ∣ \leq ε$ for $d (x, y) \leq R$ , using $∥ η_{x} - η_{y} ∥_{2}^{2} \leq ∥ ξ_{x} - ξ_{y} ∥_{1}$ . Such a kernel defines a finite-propagation Schur multiplier $M_{u}$ on $B (ℓ^{2} Γ)$ that is unital completely positive and approximates the identity on finite-propagation operators. Composing the $M_{u}$ with the finite-rank cut-downs to balls yields completely positive contractions $Φ_{R}, Ψ_{R}$ factoring $id_{C_{r}^{*} (Γ)}$ through matrix algebras with the outgoing leg landing in $B (ℓ^{2} Γ)$ , $Ψ_{R} Φ_{R} \to id$ point-norm. This is nuclear embeddability of $C_{r}^{*} (Γ)$ , hence exactness 39.05.05. $□$

Proposition 6 (hyperbolic groups are exact). A word-hyperbolic group $Γ$ acts amenably on its Gromov boundary $\partial Γ$ , hence is exact.

Proof (with citation). By Adams' theorem the action $Γ ↷ \partial Γ$ on the Gromov boundary is topologically amenable ^{[Adams 1994]}: for a basepoint $o$ in the Cayley graph one builds the measures $m_{n}^{ξ}$ as the normalised counting (or Patterson-Sullivan harmonic) measures on the length- $n$ initial segment of a geodesic ray from $o$ toward the boundary point $ξ$ , and $δ$ -thinness of geodesic triangles bounds $sup_{ξ} ∥ g \cdot m_{n}^{ξ} - m_{n}^{g ξ} ∥_{1}$ by a term $O (∣ g ∣/ n) \to 0$ , because two geodesic rays toward the same end synchronise after bounded delay. By Proposition 2 the amenable boundary action makes $Γ$ exact. (The detailed thin-triangle estimate is Brown-Ozawa Ch. 5 ^{[Brown-Ozawa Ch. 4-5]}.) $□$

Connections Master

Exact C*-algebras and nuclear embeddability 39.05.05. This unit is the group-theoretic incarnation of that theory: $Γ$ is exact precisely when $C_{r}^{*} (Γ)$ is an exact C*-algebra, and the mechanism is that an amenable action embeds $C_{r}^{*} (Γ)$ into the nuclear crossed product $C (X) ⋊_{r} Γ$ , exactly the nuclear embeddability that defines exactness there. The Gromov monster groups that supply the non-exact reduced algebras of that unit are exactly the groups that fail the property A developed here, so the two units share their single counterexample to universality.
Group C*-algebras: amenability and nuclearity 39.05.07. Where Hulanicki's theorem makes amenability of $Γ$ equal to nuclearity of $C_{r}^{*} (Γ)$ and the collapse of the two completions, this unit weakens both sides at once: boundary amenability replaces amenability, exactness replaces nuclearity, and the reduced algebra is only planted inside a nuclear host rather than being nuclear itself. The free group, the first non-nuclear example there, is the first non-amenable exact example here, and its boundary action on $\partial F_{n}$ is the prototype of the amenable actions that drive exactness.
Amenable groups, Følner sequences, and invariant means 39.05.06. Property A is the slow-variation descendant of the Følner condition: it spreads a family of finitely-supported probability measures along the word metric and asks only that nearby points produce nearby measures, dropping the demand for a single invariant mean. Amenable groups have property A by the Følner-set construction, and the boundary-averaging of a non-amenable exact group is what survives once the invariant mean is gone.
Quasidiagonality and group approximation 39.05.08. Exactness sits in the same hierarchy of finite-dimensional approximation properties: the uniform Roe algebra and the completely positive approximations that witness property A are the coarse-geometric companions of the local approximations that drive quasidiagonality, and boundary amenability is one of the structural inputs that propagate through the approximation-property tower assembled there.

Historical & philosophical context Master

Exactness for groups crystallised at the intersection of three programmes in the late 1990s. On the operator-algebra side, the question was when the reduced C*-algebra of a discrete group is exact in Kirchberg's tensor-functor sense; on the coarse-geometry side, Guoliang Yu introduced property A in his 2000 Inventiones paper as a checkable condition guaranteeing a bounded-geometry metric space embeds uniformly into Hilbert space, from which he deduced the coarse Baum-Connes conjecture ^{[Yu 2000]}. The unification was Narutaka Ozawa's, who proved in a 2000 Comptes Rendus note that exactness of $C_{r}^{*} (Γ)$ , topological amenability of a $Γ$ -action on a compact space, and Yu's property A are equivalent for countable discrete groups ^{[Ozawa 2000]}. Nigel Higson and John Roe simultaneously developed the descent route translating boundary amenability into the Novikov conjecture ^{[Higson-Roe 2000]}, and Claire Anantharaman-Delaroche and Jean Renault's monograph on amenable groupoids supplied the crossed-product nuclearity engine ^{[Anantharaman-Delaroche-Renault 2000]}.

The supply of examples came from dynamics and from algebra. Scot Adams had shown already in 1994 that a word-hyperbolic group acts amenably on its Gromov boundary ^{[Adams 1994]}, so all hyperbolic groups are exact; this fed directly into the boundary-amenability characterisation once it was available. Erik Guentner, Nigel Higson and Shmuel Weinberger closed the linear case in 2005, proving every countable linear group exact and thereby establishing the Novikov conjecture for linear groups ^{[Guentner-Higson-Weinberger 2005]}. The non-exact frontier was opened by Mikhael Gromov's monster groups, whose Cayley graphs contain coarsely embedded expanders and which therefore violate property A. Brown and Ozawa's 2008 monograph organises exactness, amenable actions, and property A as one of the central chapters of the finite-dimensional approximation programme, placing the equivalence at the heart of the modern structure theory of group C*-algebras.

Bibliography Master

@article{Ozawa2000,
  author  = {Ozawa, Narutaka},
  title   = {Amenable actions and exactness for discrete groups},
  journal = {C. R. Acad. Sci. Paris S\'er. I Math.},
  volume  = {330},
  year    = {2000},
  pages   = {691--695}
}

@article{Yu2000,
  author  = {Yu, Guoliang},
  title   = {The coarse Baum-Connes conjecture for spaces which admit a uniform embedding into Hilbert space},
  journal = {Inventiones Mathematicae},
  volume  = {139},
  year    = {2000},
  pages   = {201--240}
}

@article{HigsonRoe2000,
  author  = {Higson, Nigel and Roe, John},
  title   = {Amenable group actions and the Novikov conjecture},
  journal = {Journal f\"ur die reine und angewandte Mathematik},
  volume  = {519},
  year    = {2000},
  pages   = {143--153}
}

@article{Adams1994,
  author  = {Adams, Scot},
  title   = {Boundary amenability for word hyperbolic groups and an application to smooth dynamics of simple groups},
  journal = {Topology},
  volume  = {33},
  year    = {1994},
  pages   = {765--783}
}

@article{GuentnerHigsonWeinberger2005,
  author  = {Guentner, Erik and Higson, Nigel and Weinberger, Shmuel},
  title   = {The Novikov conjecture for linear groups},
  journal = {Publications Math\'ematiques de l'IH\'ES},
  volume  = {101},
  year    = {2005},
  pages   = {243--268}
}

@book{AnantharamanDelarocheRenault2000,
  author    = {Anantharaman-Delaroche, Claire and Renault, Jean},
  title     = {Amenable Groupoids},
  series    = {Monographies de L'Enseignement Math\'ematique},
  volume    = {36},
  publisher = {L'Enseignement Math\'ematique, Geneva},
  year      = {2000}
}

@book{BrownOzawa2008c,
  author    = {Brown, Nathanial P. and Ozawa, Narutaka},
  title     = {C*-Algebras and Finite-Dimensional Approximations},
  series    = {Graduate Studies in Mathematics},
  volume    = {88},
  publisher = {American Mathematical Society},
  year      = {2008}
}

@article{Gromov2003,
  author  = {Gromov, Mikhael},
  title   = {Random walk in random groups},
  journal = {Geometric and Functional Analysis},
  volume  = {13},
  year    = {2003},
  pages   = {73--146}
}

Prerequisites

39.05.05
39.05.07

Tier anchors

beginner: the difference between a group that can average over itself and one that can only average once it is allowed to spread its mass over a boundary it builds at infinity — averaging with help versus averaging alone
intermediate: Brown-Ozawa *C*-Algebras and Finite-Dimensional Approximations* (AMS, 2008) Ch. 4-5; Ozawa *Amenable actions and exactness for discrete groups* (C. R. Acad. Sci. 330, 2000); Anantharaman-Delaroche-Renault *Amenable Groupoids* (Geneva, 2000) Ch. 3
master: Brown-Ozawa *C*-Algebras and Finite-Dimensional Approximations* (AMS, 2008) Ch. 4-5, 15; Ozawa 2000 *C. R. Acad. Sci.* 330; Yu *The coarse Baum-Connes conjecture for spaces which admit a uniform embedding into Hilbert space* (Invent. Math. 139, 2000); Higson-Roe *Amenable group actions and the Novikov conjecture* (J. reine angew. Math. 519, 2000); Guentner-Higson-Weinberger *The Novikov conjecture for linear groups* (Publ. IHES 101, 2005)

References

Brown, N. P. and Ozawa, N. — C*-Algebras and Finite-Dimensional Approximations · Graduate Studies in Mathematics 88, AMS (2008), Ch. 4-5, 15 — exact groups, amenable actions, property A, the boundary action of a hyperbolic group
Ozawa, N. — Amenable actions and exactness for discrete groups · C. R. Acad. Sci. Paris Ser. I Math. 330 (2000), 691-695 — exactness of C*_r(Γ) is equivalent to topological amenability of an action on a compact space and to property A
Yu, G. — The coarse Baum-Connes conjecture for spaces which admit a uniform embedding into Hilbert space · Inventiones Mathematicae 139 (2000), 201-240 — property A, uniform embedding into Hilbert space, and the coarse Baum-Connes / Novikov conjecture
Higson, N. and Roe, J. — Amenable group actions and the Novikov conjecture · J. reine angew. Math. 519 (2000), 143-153 — topological amenability of the boundary action and the descent route to the Novikov conjecture
Adams, S. — Boundary amenability for word hyperbolic groups and an application to smooth dynamics of simple groups · Topology 33 (1994), 765-783 — the action of a hyperbolic group on its Gromov boundary is topologically amenable
Guentner, E., Higson, N. and Weinberger, S. — The Novikov conjecture for linear groups · Publ. Math. IHES 101 (2005), 243-268 — every countable linear group is exact (boundary-amenable)
Anantharaman-Delaroche, C. and Renault, J. — Amenable Groupoids · Monographies de L'Enseignement Mathematique 36, Geneva (2000), Ch. 3 — topological amenability for group actions and groupoids, nuclearity of amenable crossed products

Estimated time

beginner: 18m
intermediate: 54m
master: 96m