Exact C*-Algebras and Nuclear Embeddability

Intuition Beginner

The previous unit found a special class of operator algebras, the nuclear ones, that can be rebuilt from finite matrix stacks. Nuclearity is a strong, demanding property, and it has one inconvenient flaw: a piece carved out of a nuclear algebra need not stay nuclear. You can sit inside a tame algebra and yet fail to be tame yourself. So there is room for a gentler property — one that only asks you to live nicely inside a tame algebra, not to be tame on your own.

That gentler property is called exactness, and there are two ways to picture it. The first is about cutting and gluing. Take a big algebra, throw away a chunk to form a smaller quotient, and glue a fixed partner algebra onto all three pieces. Exactness of the partner says the gluing respects the cut: the glued chunk plus the glued quotient still reassemble into the glued whole, with nothing torn and nothing lost.

The second picture is the surprise of the unit. An algebra is exact in that cut-and-glue sense exactly when it can be planted as a piece inside some nuclear algebra. Living inside a tame host and respecting cuts are the same property. For algebras you can list out one element at a time, the host can even be taken to be a single famous algebra, the Cuntz algebra, so every reasonable algebra of this kind fits inside one universal tame room.

Visual Beginner

Exactness has a cut-and-glue face and an embedding face, and the theorem of the unit is that they are the same face.

The dictionary reads: on the left, cutting the big algebra into a chunk and a quotient and gluing a partner onto each is the short-exact-sequence picture; on the right, planting the algebra as a piece of a tame host is the embedding picture. The dashed connector is the theorem: these are one property, exactness.

Worked example Beginner

We test the cut-and-glue picture on the smallest honest example, where we can count everything. Let the big algebra be pairs of numbers $(x_1,x_2)$ with slotwise multiplication — picture diagonal two-by-two matrices. The chunk we throw away is the pairs whose second slot is zero, written $(x_1,0)$. The quotient, what is left after collapsing the chunk to zero, is the single remaining slot: just the number $x_2$.

Now glue a fixed partner $A$ onto each of the three pieces. Gluing the partner onto the big algebra gives two independent copies of $A$, one per slot. Gluing onto the chunk gives one copy of $A$, sitting in the first slot. Gluing onto the quotient gives one copy of $A$, the second slot.

Check that the pieces reassemble. The glued chunk is the first-slot copy of $A$. The glued quotient is the second-slot copy of $A$. Stacking them back gives exactly the two-slot glued big algebra: first-slot copy plus second-slot copy, nothing torn, nothing extra. The cut survived the gluing.

What this tells us: for a commutative big algebra the reassembly always works, against every partner, so the partner side never causes trouble here. The interesting failures come from large noncommutative partners, where gluing can lose a piece. Exactness names the partners for which the reassembly never fails.

Check your understanding Beginner

Formal definition Intermediate+

Let $A$ be a C*-algebra. Recall from 39.05.03 that for any C*-algebra $B$ the minimal tensor product $B{\otimes }_{\min }A$ is the spatial completion of $B\odot A$, and that the minimal norm is injective: a C*-subalgebra inclusion $B_0\subseteq B$ induces an isometric inclusion $B_0{\otimes }_{\min }A\subseteq B{\otimes }_{\min }A$. Quotients are the problematic direction.

A C*-algebra $A$ is exact if the functor $-{\otimes }_{\min }A$ preserves short exact sequences: for every C*-algebra $B$ and every closed two-sided ideal $J⊴B$, the sequence $$ 0 \longrightarrow J \otimes_{\min} A \longrightarrow B \otimes_{\min} A \longrightarrow (B/J) \otimes_{\min} A \longrightarrow 0 $$ is exact ^{[Wassermann 1994; Brown-Ozawa Ch. 3-4]}. By injectivity of $\min$ the left map is always an isometric inclusion, and the right map is always a surjection with kernel containing $J{\otimes }_{\min }A$; the single non-automatic point is that this kernel is exactly $J{\otimes }_{\min }A$ rather than something larger. Equivalently, $A$ is exact iff for every $(B,J)$ the natural map $$ (B \otimes_{\min} A),/,(J \otimes_{\min} A) \longrightarrow (B/J) \otimes_{\min} A $$ is a $\ast$-isomorphism. The maximal tensor product is always exact in this sense, so exactness is a defect measured by $\min$ alone.

A C*-algebra $A$ is nuclearly embeddable if there is a nuclear C*-algebra $N$ and an injective $\ast$-homomorphism $\iota :A\hookrightarrow N$ — equivalently, if the inclusion $A\hookrightarrow A$ factors approximately through finite-dimensional matrix algebras by completely positive contractive (CPC) maps in the point-norm topology, where now the second leg need only land back in some C*-algebra containing $A$ rather than in $A$ itself. The latter is the nuclear embeddability property: there are CPC maps ${\phi }_{\lambda }:A\to M_{n_{\lambda }}(\mathbb{C})$ and ${\psi }_{\lambda }:M_{n_{\lambda }}(\mathbb{C})\to B(H)$ (for a faithful $A\subseteq B(H)$) with ${\psi }_{\lambda }{\phi }_{\lambda }\to {\mathrm{id}}_A$ point-norm. This relaxes the CPAP of 39.05.04 by dropping the requirement that the ${\psi }_{\lambda }$ land inside $A$.

A C*-algebra $A$ is locally reflexive if for every finite-dimensional operator system $E\subseteq A^{\ast \ast }$ the inclusion $E\hookrightarrow A^{\ast \ast }$ is a point-weak-$\ast$ limit of CPC maps $E\to A$. Local reflexivity is the operator-space lifting property that controls when nuclear embeddability transfers cleanly through quotients.

Counterexamples to common slips

• Exactness is a property of $A$ against all $(B,J)$, witnessed by one universal test. It is not enough that the sequence be exact for some convenient ideal; failure is detected by tensoring against a single bad pair. The reduced C*-algebra of a Gromov monster group fails the test.
• Nuclear embeddability is weaker than the CPAP. The CPAP requires the maps ${\psi }_{\lambda }$ to land in $A$; nuclear embeddability lets them land in a larger host. Dropping that requirement is exactly the gap between nuclear and merely exact.
• Exact does not imply nuclear. The reduced free-group algebra $C_r^{\ast }(F_n)$ is exact (a subalgebra of a nuclear algebra) and not nuclear. The implication runs one way only.
• Local reflexivity is automatic but not vacuous. Every C*-algebra is locally reflexive by Effros-Haagerup, yet the property does real work: it is what lets the operator-space approximation of $A$ inside its bidual be pulled back into $A$ itself.

Key theorem with proof Intermediate+

Theorem (Kirchberg: exactness equals nuclear embeddability). A C-algebra $A$ is exact if and only if it is nuclearly embeddable, that is, isomorphic to a C*-subalgebra of a nuclear C*-algebra.* ^{[Kirchberg 1995; Kirchberg 1994; Brown-Ozawa Ch. 3-4]}

Proof. ($\Leftarrow$) Suppose $A\subseteq N$ with $N$ nuclear. Fix a short exact sequence $0\to J\to B\to B/J\to 0$; we must show $0\to J{\otimes }_{\min }A\to B{\otimes }_{\min }A\to (B/J){\otimes }_{\min }A\to 0$ is exact. Because $N$ is nuclear, $-{\otimes }_{\min }N=-{\otimes }_{\max }N$, and the maximal tensor product preserves short exact sequences 39.05.04; hence $-{\otimes }_{\min }N$ is exact, so $$ 0 \to J \otimes_{\min} N \to B \otimes_{\min} N \to (B/J) \otimes_{\min} N \to 0 $$ is exact. Now restrict each term along $A\subseteq N$. By injectivity of the minimal norm 39.05.03, $X{\otimes }_{\min }A\subseteq X{\otimes }_{\min }N$ isometrically for $X\in \{J,B,B/J\}$. The kernel of $B{\otimes }_{\min }A\to (B/J){\otimes }_{\min }A$ is the intersection of $J{\otimes }_{\min }N$ (the kernel in the nuclear sequence) with the subalgebra $B{\otimes }_{\min }A$. That intersection is $J{\otimes }_{\min }A$: an element of $B{\otimes }_{\min }A$ killed in $(B/J){\otimes }_{\min }A$ lies in $J{\otimes }_{\min }N\cap B{\otimes }_{\min }A$, and the slice-map characterisation of $J{\otimes }_{\min }A$ inside $B{\otimes }_{\min }A$ — an element lies in $J{\otimes }_{\min }A$ iff every slice $(\mathrm{id}\otimes \phi )$ by a functional $\phi$ on $A$ lands in $J$ — pins it there, since the same slices computed in $N$ already land in $J$. Thus the kernel is exactly $J{\otimes }_{\min }A$ and $A$ is exact.

($\Rightarrow$) Suppose $A$ is exact and separable. Realise $A\subseteq B(H)$ faithfully. Exactness gives, for each finite $F\subseteq A$ and $\epsilon >0$, a CPC factorisation of ${\mathrm{id}}_A$ on $F$ through a finite-dimensional algebra with the outgoing leg landing in $B(H)$: concretely, exactness against the Toeplitz/cone sequence $0\to SA\to CA\to A\to 0$ (the suspension-cone sequence) forces the identity of $A$ to lift to a CPC map factoring through matrices up to $\epsilon$ on $F$, by the Effros-Haagerup local reflexivity of $A$ ^{[Effros-Haagerup 1985]} applied to finite-dimensional operator subsystems of $A^{\ast \ast }$. Assemble the resulting CPC maps ${\phi }_{\lambda }:A\to M_{n_{\lambda }}(\mathbb{C})$, ${\psi }_{\lambda }:M_{n_{\lambda }}(\mathbb{C})\to B(H)$ with ${\psi }_{\lambda }{\phi }_{\lambda }\to {\mathrm{id}}_A$ point-norm. This is precisely nuclear embeddability: the maps exhibit $A$ inside the nuclear algebra generated by the images, and one packages the host as $N=$ the inductive limit of the finite-dimensional ranges under the connecting CPC data, a nuclear C*-algebra containing a copy of $A$. Hence $A$ embeds in a nuclear algebra. $\square$

Bridge. The exactness-equals-embeddability theorem builds toward the classification of purely infinite simple nuclear algebras, and it appears again in the Kirchberg-Phillips embedding of every separable exact algebra into the Cuntz algebra ${\mathcal{O}}_2$. The foundational reason the two pictures coincide is exactly that nuclearity of the host makes $-{\otimes }_{\min }$ behave like $-{\otimes }_{\max }$ 39.05.04, so the host's good short-exact behaviour descends to its subalgebra through the injectivity of $\min$ 39.05.03; this is exactly the slice-map machinery read one tensor leg at a time. The two halves of the proof are dual to each other — the embeddability-implies-exact direction pushes the host's exactness down to the piece, while the exact-implies-embeddable direction pulls finite CPC factorisations out of the preserved sequences via local reflexivity. The central insight is that exactness generalises nuclearity by asking only to live inside a finite-dimensionally approximable algebra rather than be one, and putting these together the bridge is that this is the largest subalgebra-stable weakening of the CPAP, which is the central insight organising the chapter's passage from nuclearity 39.05.04 to the group C*-algebras of 39.05.07.

Exercises Intermediate+

Lean formalization Intermediate+

lean_status: none — Mathlib carries the C*-algebra layer (CStarAlgebra, positivity through cfc, the spectrum) and the algebraic tensor product of algebras (Algebra.TensorProduct), but neither the minimal C*-tensor norm and its functoriality in $B$ (the prerequisite layer of 39.05.03), nor the short-exact-sequence definition of exactness, nor nuclear embeddability, nor Kirchberg's equivalence, nor operator-space local reflexivity. The intended statement reads schematically:

import Mathlib.Analysis.CStarAlgebra.Basic
import Mathlib.RingTheory.TensorProduct.Basic

variable {A : Type*} [CStarAlgebra A]

/-- A C*-algebra is exact iff it is nuclearly embeddable: it is isomorphic to a
C*-subalgebra of a nuclear C*-algebra. Exactness on the left means - ⊗_min A
preserves short exact sequences. -/
theorem exact_iff_nuclearly_embeddable :
    IsExact A ↔
      ∃ (N : Type*) (_ : CStarAlgebra N) (_ : IsNuclear N)
        (ι : A →⋆ₐ[ℂ] N), Function.Injective ι :=
  sorry  -- (⇐) nuclear host makes -⊗_min behave like -⊗_max, which is exact;
         -- restrict along A ⊆ N using injectivity of the minimal norm.
         -- (⇒) local reflexivity pulls finite CPC factorisations of id_A out of
         -- the preserved short exact sequences; package the ranges into N.

Advanced results Master

Exactness sits one notch below nuclearity in the regularity hierarchy, and its theory is organised by the embedding characterisation, the permanence package, and the boundary examples that separate it from both nuclearity above and from the wild algebras below.

The hierarchy and permanence. Nuclear $\Rightarrow$ exact, and the implication is strict. The class of exact C*-algebras is closed under the operations nuclearity is closed under, and one more: passage to C*-subalgebras. Exactness passes to subalgebras, quotients (of separable exact algebras), extensions, inductive limits, and minimal tensor products; a separable C*-algebra is exact iff each of its separable subalgebras is. The contrast with nuclearity is precise. Nuclearity fails for subalgebras — a non-exact algebra is a quotient of a nuclear algebra, and every separable C*-algebra is a quotient of a nuclear one — so nuclearity is not subalgebra-stable, while exactness is engineered to be exactly the subalgebra-stable shadow of nuclearity. This is why $C_r^{\ast }(F_n)$, a subalgebra of the nuclear crossed product $C(\partial F_n){⋊}_r{F}_n$ (the boundary action is amenable), is exact though not nuclear (Haagerup-de Cannière) 39.05.07.

Kirchberg's slice and embedding theorems. Kirchberg's slice-map property says $A$ is exact iff for every $B$ and every closed subspace $X\subseteq B{\otimes }_{\min }A$, an element $x\in B{\otimes }_{\min }A$ lies in $X$ as soon as every left slice $(\phi \otimes \mathrm{id})(x)$ by a functional $\phi$ on $B$ lies in the corresponding slice of $X$; the slice maps detect membership. From this Kirchberg deduced the structural high point: every separable exact C*-algebra embeds into the Cuntz algebra ${\mathcal{O}}_2$ (Kirchberg-Phillips) ^{[Kirchberg-Phillips 2000]}. Since ${\mathcal{O}}_2$ is nuclear, simple, and purely infinite, and ${\mathcal{O}}_2\otimes {\mathcal{O}}_2\cong {\mathcal{O}}_2$, this realises a single universal tame host absorbing all of separable exact C*-theory, and it is the embedding side of the Kirchberg-Phillips classification of separable, simple, nuclear, purely infinite (Kirchberg) algebras by $KK$-theory.

Local reflexivity and property C. Every C*-algebra is locally reflexive: finite-dimensional operator systems of $A^{\ast \ast }$ are point-weak-$\ast$ approximated by CPC maps into $A$ (Effros-Haagerup) ^{[Effros-Haagerup 1985]}. Kirchberg's property C for $A$ asks that the canonical map $A^{\ast \ast }{\otimes }_{\min }B\to (A{\otimes }_{\min }B{)}^{\ast \ast }$ be injective for all $B$; property C lies between exactness and local reflexivity and is the operator-space lever that converts exactness of $A$ into the lifting needed for the ${\mathcal{O}}_2$-embedding. The hierarchy reads: nuclear $\Rightarrow$ property C $\Rightarrow$ exact $\Rightarrow$ locally reflexive, with each implication strict and each arrow detectable by a tensor test against a suitable $B$.

Non-exact algebras: monster groups. Non-exact C*-algebras are not pathological curiosities concocted by hand; they arise from geometric group theory. Gromov constructed monster groups — finitely generated groups containing expander graphs coarsely embedded in their Cayley graphs — and for such a group $\Gamma$ the reduced group C*-algebra $C_r^{\ast }(\Gamma )$ is not exact ^{[Gromov 2003]}. The expander geometry obstructs the coarse (Yu) embedding into Hilbert space, and the failure of exactness is the operator-algebraic trace of that obstruction. These were the first explicit non-exact reduced group algebras, settling that exactness is a genuine restriction on discrete groups: a group is exact (boundary-amenable, i.e. it admits an amenable action on a compact space) iff $C_r^{\ast }(\Gamma )$ is exact, and monster groups are non-exact in this sense while remaining far from any amenability.

Synthesis. The exactness-equals-nuclear-embeddability theorem is the foundational reason exactness is the right subalgebra-stable weakening of nuclearity: it presents an exact algebra as a piece of a finite-dimensionally approximable host, and this is exactly the statement that the $\min$-functor inherits short-exact behaviour from a nuclear ambient algebra through the injectivity of $\min$ 39.05.03. The central insight is that the same property, read at several levels, is the short-exact preservation by $-{\otimes }_{\min }A$, the embeddability of $A$ into a nuclear algebra 39.05.04, the slice-map detection of membership, and boundary-amenability of the underlying group 39.05.07 — these generalise the single fact that nuclear algebras tensor exactly, and the two faces are dual through the bidual via property C and local reflexivity, the $W^{\ast }$-side of the embedding. Putting these together, the Kirchberg-Phillips ${\mathcal{O}}_2$-embedding refines the qualitative property into a universal host, and the bridge to classification is that exactness plus nuclearity plus pure infiniteness is exactly the regularity making the $KK$-theoretic invariant complete — this is exactly where the approximation theory of the chapter feeds the Kirchberg-Phillips classification of purely infinite simple algebras, the central insight that organises the boundary between the classifiable and the non-exact wild.

Full proof set Master

Proposition (nuclearly embeddable $\Rightarrow$ exact). If $A\subseteq N$ with $N$ nuclear, then $A$ is exact. Proof: for a short exact sequence $0\to J\to B\to B/J\to 0$, nuclearity of $N$ gives $-{\otimes }_{\min }N=-{\otimes }_{\max }N$ 39.05.04, and $-{\otimes }_{\max }N$ preserves short exact sequences (the maximal norm respects quotients and ideals 39.05.03); hence $0\to J{\otimes }_{\min }N\to B{\otimes }_{\min }N\to (B/J){\otimes }_{\min }N\to 0$ is exact. Injectivity of $\min$ 39.05.03 embeds $X{\otimes }_{\min }A\hookrightarrow X{\otimes }_{\min }N$ isometrically for each term. The kernel of $B{\otimes }_{\min }A\to (B/J){\otimes }_{\min }A$ is $\left(J{\otimes }_{\min }N\right)\cap \left(B{\otimes }_{\min }A\right)$; the slice-map characterisation — $z\in B{\otimes }_{\min }A$ lies in $J{\otimes }_{\min }A$ iff $({\mathrm{id}}_B\otimes \phi )(z)\in J$ for every state $\phi$ on $A$ — identifies this intersection with $J{\otimes }_{\min }A$, because the same slices taken in $N$ already certify membership in $J$. Hence the minimal sequence is exact and $A$ is exact.

Proposition (exact $\Rightarrow$ nuclearly embeddable, separable case). If $A$ is separable and exact then $A$ embeds in a nuclear C*-algebra. Proof: by Effros-Haagerup local reflexivity ^{[Effros-Haagerup 1985]} every finite-dimensional operator system $E\subseteq A$ admits CPC maps $A\to M_n(\mathbb{C})\to A^{\ast \ast }$ approximating ${\mathrm{id}}_E$ point-weak-$\ast$. Exactness upgrades the approximation to the point-norm topology with the outgoing leg landing in $B(H)$ for a faithful $A\subseteq B(H)$: tensoring against the cone-suspension sequence $0\to SA\to CA\to A\to 0$ and using preservation of $\min$-exactness forces the lifted maps to converge in norm on finite sets. Indexing by $(F,\epsilon )$ yields CPC maps ${\phi }_{\lambda }:A\to M_{n_{\lambda }}(\mathbb{C})$, ${\psi }_{\lambda }:M_{n_{\lambda }}(\mathbb{C})\to B(H)$ with ${\psi }_{\lambda }{\phi }_{\lambda }\to {\mathrm{id}}_A$ point-norm. Form $N$ as the C*-algebra generated by the inductive system of finite-dimensional ranges under the connecting CPC data; $N$ is an inductive limit of nuclear (finite-dimensional) pieces, hence nuclear 39.05.04, and the maps embed $A$ into $N$.

Proposition (exactness passes to subalgebras). If $A$ is exact and $A_0\subseteq A$ is a C*-subalgebra, then $A_0$ is exact. Proof: by the equivalence theorem $A\subseteq N$ for some nuclear $N$, so $A_0\subseteq N$ is a C*-subalgebra of a nuclear algebra, hence exact by the embeddability-implies-exact proposition. (Tensor-side: injectivity of $\min$ restricts the exact sequence for $A$ to the subalgebra $A_0$, and the slice-map membership criterion is inherited.)

Proposition (exactness passes to quotients). If $A$ is separable and exact and $I⊴A$ is a closed ideal, then $A/I$ is exact. Proof: by the embedding theorem $A\hookrightarrow {\mathcal{O}}_2$ (Kirchberg-Phillips) ^{[Kirchberg-Phillips 2000]}. Quotients of exact algebras are exact because exactness is detected by the slice-map property, which is stable under the quotient map: a CPC local lifting $A/I\to A$ on finite-dimensional operator systems exists by Choi-Effros/Arveson 39.05.02, and composing the embeddability data of $A$ with the lifting and the quotient produces nuclear-embeddability data for $A/I$. Hence $A/I$ is exact. (This is the one permanence property where separability and local reflexivity are genuinely used.)

Proposition (minimal tensor products of exact algebras are exact). If $A_1,A_2$ are exact then $A_1{\otimes }_{\min }{A}_2$ is exact. Proof: embed $A_i\subseteq N_i$ with $N_i$ nuclear. Then $A_1{\otimes }_{\min }{A}_2\subseteq N_1{\otimes }_{\min }{N}_2$ by injectivity of $\min$ 39.05.03, and $N_1{\otimes }_{\min }{N}_2$ is nuclear because the minimal tensor product of nuclear algebras is nuclear 39.05.04. So $A_1{\otimes }_{\min }{A}_2$ is a subalgebra of a nuclear algebra, hence exact.

Proposition (a non-exact reduced group algebra exists). There is a finitely generated group $\Gamma$ with $C_r^{\ast }(\Gamma )$ not exact. Proof (stated with citation): Gromov's monster groups contain coarsely embedded expander graphs in their Cayley graphs ^{[Gromov 2003]}. A group $\Gamma$ is exact (its reduced algebra is an exact C*-algebra) iff $\Gamma$ is boundary-amenable, equivalently admits an amenable action on a compact Hausdorff space, equivalently coarsely embeds into Hilbert space via Yu's property A. The expander geometry of a monster group violates property A — expanders are the canonical obstruction to coarse Hilbert-space embedding — so $\Gamma$ is not boundary-amenable and $C_r^{\ast }(\Gamma )$ is not exact. This is the first explicit family of non-exact reduced group C*-algebras and shows the class of exact algebras is proper. (The full construction of monster groups via random-walk methods is beyond the scope of this proof set; see the cited primary source.)

Connections Master

• Nuclear C-algebras and the CPAP 39.05.04* — exactness is the subalgebra-stable weakening of nuclearity: nuclear embeddability drops the requirement that the CPAP's outgoing maps land in $A$, asking only that $A$ live inside a nuclear host, so nuclear $\Rightarrow$ exact while $C_r^{\ast }(F_n)$ is exact and not nuclear; this unit is the immediate sequel that names the property a piece of a nuclear algebra inherits.

• Tensor products of C-algebras: the minimal and maximal norms 39.05.03* — exactness is a defect of the minimal tensor functor alone, measured by whether $-{\otimes }_{\min }A$ preserves short exact sequences; the injectivity of $\min$ and the slice-map machinery of that unit are exactly the tools that transport exactness from a nuclear host down to a subalgebra, and Takesaki's $\min \ne \max$ obstruction is what keeps the exact $C_r^{\ast }(F_n)$ from being nuclear.

• Operator systems, Arveson's extension theorem, and the Choi-Effros theorem 39.05.02 — the CPC local liftings that pull nuclear-embeddability data through quotients, and the operator-space approximations underlying local reflexivity and property C, are produced by Arveson-extending compressions off finite-dimensional operator systems; the entire lifting theory of exactness runs on that unit's extension machinery.

• Group C-algebras: full and reduced 39.05.07* — a discrete group is exact (boundary-amenable) iff its reduced C*-algebra $C_r^{\ast }(\Gamma )$ is an exact C*-algebra; the free groups give exact-but-not-nuclear examples via the Haagerup-de Cannière approximation, while Gromov monster groups give the non-exact examples, so the group-theoretic face of this unit's hierarchy is developed there.

• Von Neumann algebras and the bicommutant theorem 39.03.01 — property C and local reflexivity are conditions on the canonical map $A^{\ast \ast }{\otimes }_{\min }B\to (A{\otimes }_{\min }B{)}^{\ast \ast }$ into the enveloping von Neumann algebra, so the operator-space lifting theory that separates exactness from mere local reflexivity is read through the $W^{\ast }$-structure of the bidual.

• Density matrix, pure and mixed states 12.17.01 — the finite-dimensional shadow of nuclear embeddability is the routing of a quantum channel on a subsystem through a finite-dimensional ancilla inside a larger tame system; there every algebra is a matrix algebra, hence nuclear and exact, which is why finite quantum systems never see the exactness obstruction the infinite theory measures.

Historical & philosophical context Master

Exactness emerged from the tensor-product theory of the 1970s but was isolated as a structural notion by Eberhard Kirchberg across a sequence of papers in the 1990s. Simon Wassermann's 1994 Seoul lecture notes consolidated the short-exact-sequence definition — a C*-algebra $A$ is exact when $-{\otimes }_{\min }A$ preserves short exact sequences — and recorded the early permanence results, including the exactness of the reduced free-group algebras ^{[Wassermann 1994]}. Kirchberg's 1995 Journal of Functional Analysis paper "On subalgebras of the CAR-algebra" proved the characterisation that gives the subject its conceptual unity: exactness is equivalent to nuclear embeddability, so an algebra is exact precisely when it is a C*-subalgebra of a nuclear algebra ^{[Kirchberg 1995]}. His 1994 ICM address set the slice-map property, property C, and the classification programme for purely infinite algebras in their final framing ^{[Kirchberg 1994]}.

The embedding side reached its sharpest form in the Kirchberg-Phillips theorem: every separable exact C*-algebra embeds into the Cuntz algebra ${\mathcal{O}}_2$, published by Kirchberg and N. Christopher Phillips in the 2000 Journal für die reine und angewandte Mathematik ^{[Kirchberg-Phillips 2000]}. The operator-space underpinning is Edward Effros and Uffe Haagerup's 1985 Duke Mathematical Journal proof that every C*-algebra is locally reflexive ^{[Effros-Haagerup 1985]}. The question of whether exactness is a genuine restriction on groups was settled by Mikhael Gromov's 2003 Geometric and Functional Analysis construction of monster groups containing coarsely embedded expanders, whose reduced C*-algebras are not exact ^{[Gromov 2003]}. Brown and Ozawa's 2008 monograph organises exactness, alongside nuclearity and quasidiagonality, as one of the finite-dimensional-approximation properties, with boundary amenability supplying the group-theoretic translation.

Bibliography Master

• Wassermann, S., Exact C-Algebras and Related Topics*, Lecture Notes Series 19, Seoul National University, 1994.
• Kirchberg, E., "On subalgebras of the CAR-algebra", Journal of Functional Analysis 129 (1995), 35-63.
• Kirchberg, E., "Exact C*-algebras, tensor products, and the classification of purely infinite algebras", in Proceedings of the International Congress of Mathematicians (Zurich, 1994), Birkhauser (1995), 943-954.
• Kirchberg, E. and Phillips, N. C., "Embedding of exact C*-algebras in the Cuntz algebra ${\mathcal{O}}_2$", Journal fur die reine und angewandte Mathematik 525 (2000), 17-53.
• Effros, E. G. and Haagerup, U., "Lifting problems and local reflexivity for C*-algebras", Duke Mathematical Journal 52 (1985), 103-128.
• Gromov, M., "Random walk in random groups", Geometric and Functional Analysis 13 (2003), 73-146.
• Brown, N. P. and Ozawa, N., C-Algebras and Finite-Dimensional Approximations*, Graduate Studies in Mathematics 88, American Mathematical Society, 2008. Ch. 3-4, 9.

---

Operator-algebras spine, fifth structural unit of the nuclearity-exactness chapter. Produced as the exactness anchor: exact C-algebras as the algebras for which $-{\otimes }_{\min }A$ preserves short exact sequences, Kirchberg's equivalence with nuclear embeddability (subalgebra of a nuclear algebra, and into ${\mathcal{O}}_2$ for separable algebras by Kirchberg-Phillips), local reflexivity and property C, the hierarchy nuclear $\Rightarrow$ exact with $C_r^{\ast }(F_n)$ exact-not-nuclear (Haagerup-de Cannière) and Gromov monster groups non-exact, and the permanence package. Builds on the nuclearity/CPAP of 39.05.04 and the min/max tensor norms of 39.05.03; the group C*-algebra translation is developed in the co-produced 39.05.07.*