AF Algebras, Bratteli Diagrams, and the Irrational Rotation Algebra

Intuition Beginner

Some operator algebras are too big to write down all at once, but you can build them as a tower. Start with a few small blocks of matrices. Then build a slightly bigger algebra in which each old block sits inside a new block, perhaps copied several times. Repeat forever. The algebra you reach in the limit is an approximately finite-dimensional algebra: every element can be approximated as closely as you like by something living at a finite stage of the tower. You never need an infinite computation to get close to any element.

The whole tower is recorded by a simple branching picture called a Bratteli diagram: rows of dots, one row per stage, with each dot a matrix block and the lines between rows saying how many copies of each old block get packed into each new block. The diagram is the blueprint, and the algebra is what you get by pouring matrices into it forever.

The surprise is that a single bookkeeping gadget — a group of "sizes" that you can add and compare — remembers the whole tower. Two towers that look different but produce the same group of sizes give the same algebra.

Visual Beginner

A Bratteli diagram is rows of dots joined by edges; the dots are matrix blocks, the edge counts say how many copies embed, and the running totals of block sizes track how the algebra grows.

The dictionary reads: each dot is a block of square matrices of the labelled size; each edge says "embed one copy of this old block into that new block"; a dot with several incoming edges packs several old blocks down its diagonal; and following the diagram downward forever, then filling in the gaps, produces the algebra $A$.

Worked example Beginner

Build the simplest infinite tower. At stage one take the two-by-two matrices, a single block of size $2$. To pass to stage two, embed each matrix block-diagonally into a four-by-four matrix as two side-by-side copies:

$$a⟼\left(\begin{array}{cc}a& 0\\ 0& a\end{array}\right).$$

So stage two is the four-by-four matrices, a single block of size $4$, and the embedding doubled the size. Do it again: stage three is the eight-by-eight matrices, size $8$. Each stage doubles, giving sizes $2,4,8,16,\dots$, the powers of two.

The Bratteli diagram is a single column of dots labelled $2,4,8,16,\dots$, each joined to the next by a double edge (two copies embed each time). The limit algebra is written $M_{2^{\infty }}$, the CAR algebra. Track the sizes as fractions of the whole: a single matrix unit at stage one is a fraction $1/2$ of the block; after one doubling it is $2/4=1/2$ again; the "sizes" you can reach are the fractions with a power of two in the denominator.

What this tells us: the reachable sizes are exactly the dyadic numbers $\{k/2^n\}$. That set of sizes, with addition and the order on the line, is the invariant that names this tower.

Check your understanding Beginner

Formal definition Intermediate+

A finite-dimensional C-algebra* is, up to $\ast$-isomorphism, a finite direct sum of matrix algebras $F=M_{n_1}(\mathbb{C})\oplus \cdots \oplus M_{n_k}(\mathbb{C})$; this is the C*-algebraic content of the Artin-Wedderburn theorem, and the data $(n_1,\dots ,n_k)$ is a complete invariant. A $\ast$-homomorphism $\varphi :F\to F^{\prime }$ between finite-dimensional C*-algebras is determined up to unitary equivalence by its multiplicity matrix $(m_{ij})$, where $m_{ij}$ counts how many copies of the $i$-th summand of $F$ embed block-diagonally into the $j$-th summand of $F^{\prime }$; compatibility requires ${\sum }_i{m}_{ij}{n}_i\le n_j^{\prime }$ (with equality when $\varphi$ is unital).

A C*-algebra $A$ is approximately finite-dimensional (AF) if it is the closure of an increasing union $\overline{{\bigcup }_n{F}_n}$ of finite-dimensional C*-subalgebras $F_1\subseteq F_2\subseteq \cdots$, equivalently the inductive limit $A=\underrightarrow{\lim }(F_n,{\varphi }_n)$ of a sequence of finite-dimensional C*-algebras with connecting $\ast$-homomorphisms ${\varphi }_n:F_n\to F_{n+1}$ ^{[Davidson Ch. III]}. The inductive limit is the completion of the algebraic direct limit in the unique C*-norm $\Vert x\Vert ={\lim }_n\Vert x{\Vert }_{F_n}$, which is well-defined because each ${\varphi }_n$ is norm-decreasing and isometric on its image.

A Bratteli diagram is the directed graph encoding this data: a vertex of weight $n_i$ for each matrix summand at each level, and $m_{ij}$ edges between consecutive levels recording the multiplicity matrix of ${\varphi }_n$. The diagram, together with the level vertex weights, determines the AF algebra up to $\ast$-isomorphism, though many diagrams give the same algebra.

For a unital C*-algebra $A$, two projections $p,q$ are Murray-von Neumann equivalent, $p\sim q$, if $p=v^{\ast }v$ and $q=v{v}^{\ast }$ for some partial isometry $v\in A$ (equivalently in $M_{\infty }(A)=\underrightarrow{\lim }{M}_k(A)$). The equivalence classes of projections in $M_{\infty }(A)$ form an abelian monoid $V(A)$ under the orthogonal-direct-sum operation $[p]+[q]=[p\oplus q]$. The $K_0$ group is its Grothendieck enveloping group $K_0(A)=G(V(A))$. The image of $V(A)$ generates the positive cone $K_0(A{)}_{+}$, making $(K_0(A),K_0(A{)}_{+})$ an ordered abelian group; the class $[1_A]$ of the unit determines the scale $\Sigma (A)=\{[p]:p\text{ a projection in }A\}$. For an AF algebra this ordered group with scale is called the dimension group.

Counterexamples to common slips

• Being a union of finite-dimensional subalgebras is not the same as being AF unless the subalgebras are nested. An arbitrary dense subalgebra built from many small matrix algebras with no inclusion chain need not give an AF algebra; the inductive system must be a directed chain $F_1\subseteq F_2\subseteq \cdots$.
• A Bratteli diagram does not determine the algebra through its shape alone — telescoping (replacing the chain by a subsequence and composing connecting maps) changes the diagram while preserving the algebra. Two diagrams give isomorphic algebras exactly when they have a common telescoping, equivalently when their dimension groups agree.
• The dimension group is an ordered group; the underlying abelian group $K_0(A)$ alone is not a complete invariant. The CAR algebra has $K_0\cong \mathbb{Z}[\frac{1}{2}]$ as a group, but it is the order embedding into $\mathbb{R}$ via the trace that distinguishes it from other UHF algebras such as $M_{3^{\infty }}$ with $K_0\cong \mathbb{Z}[\frac{1}{3}]$.

Key theorem with proof Intermediate+

Theorem (Elliott's classification of AF algebras). Let $A$ and $B$ be unital AF algebras. Then $A\cong B$ as C-algebras if and only if there is an isomorphism of ordered abelian groups $\alpha :K_0(A)\to K_0(B)$ preserving the scale, $\alpha (\Sigma (A))=\Sigma (B)$, equivalently $\alpha ([1_A])=[1_B]$. Moreover every such scaled ordered-group isomorphism is induced by a $\ast$-isomorphism, and $\ast$-isomorphisms are determined up to approximate unitary equivalence by their $K_0$-effect.* ^{[Elliott 1976; Effros]}

Proof. The forward direction is functoriality: a $\ast$-isomorphism $\Phi :A\to B$ carries projections to projections preserving Murray-von Neumann equivalence and orthogonal sums, hence induces a monoid isomorphism $V(A)\to V(B)$, which Grothendieck-completes to an ordered-group isomorphism $K_0(A)\to K_0(B)$ sending $[1_A]$ to $[1_B]$.

The substantive direction is the intertwining argument. Write $A=\underrightarrow{\lim }(F_n,{\varphi }_n)$ and $B=\underrightarrow{\lim }(G_m,{\psi }_m)$ with each $F_n,G_m$ finite-dimensional. For finite-dimensional C*-algebras the functor $K_0$ is faithful and full on unital embeddings: $K_0(M_{n_1}\oplus \cdots \oplus M_{n_k})={\mathbb{Z}}^k$ with positive cone ${\mathbb{Z}}_{\ge 0}^k$, and a unital $\ast$-homomorphism $F\to F^{\prime }$ corresponds exactly to its multiplicity matrix acting ${\mathbb{Z}}^k\to {\mathbb{Z}}^{k^{\prime }}$ as a positive map respecting the scale. Two unital $\ast$-homomorphisms between finite-dimensional algebras inducing the same map on $K_0$ are unitarily equivalent, because the multiplicity matrix is the complete unitary invariant.

Given the scaled ordered isomorphism $\alpha$, one builds a commuting ladder. Using that $K_0(A)=\underrightarrow{\lim }{K}_0(F_n)$ and $K_0(B)=\underrightarrow{\lim }{K}_0(G_m)$ are themselves inductive limits of the ${\mathbb{Z}}^{k_n}$, the map $\alpha$ restricts, after telescoping, to a sequence of positive scale-preserving maps $K_0(F_n)\to K_0(G_{m_n})$ and back, $K_0(G_{m_n})\to K_0(F_{n+1})$, that compose to the connecting maps in the limit. By the finite-dimensional lifting, each is realised by a unital $\ast$-homomorphism $F_n\to G_{m_n}$ and $G_{m_n}\to F_{n+1}$, unique up to unitary equivalence. A standard approximate-intertwining lemma (Elliott's two-sided intertwining) corrects the unitary discrepancies stage by stage, producing genuine commuting triangles in the limit and hence a $\ast$-isomorphism $\Phi :A\to B$ inducing $\alpha$ on $K_0$. The approximate-unitary uniqueness follows from the same lifting uniqueness pushed to the limit. $\square$

Bridge. Elliott's theorem builds toward the entire C*-classification program and appears again in 39.02.04 pending, where the Effros-Handelman-Shen theorem characterises which ordered groups arise as dimension groups (the unperforated Riesz-interpolation groups), closing the invariant. The foundational reason the argument works is exactly that $K_0$ is a complete invariant at finite stages — the multiplicity matrix is the unitary invariant of an embedding of matrix blocks — so the limit invariant is assembled by an intertwining that is dual to the inductive system itself. This is exactly the abstract face of the dyadic-fractions picture of the CAR algebra from the Beginner worked example, where the reachable sizes $\{k/2^n\}$ are the ordered group $\mathbb{Z}[\frac{1}{2}]$; the construction generalises the commutative Gelfand picture 39.01.02, where $C(X)$ is reconstructed from the algebra of functions, to a reconstruction of an AF algebra from its group of projection-sizes. Putting these together, the dimension group is the noncommutative analogue of a measured size, and the bridge is that classifying AF algebras becomes classifying ordered abelian groups.

Exercises Intermediate+

Lean formalization Intermediate+

lean_status: none — Mathlib has matrix C*-algebras and sequential colimits but no named C*-inductive limit, no AF class, and no $K$-theory functor for C*-algebras: there is no ordered $K_0(A)$, no positive cone, no scale, no dimension group, and so no statement of Elliott's theorem or of the Pimsner-Voiculescu computation $K_0(A_{\theta })\cong {\mathbb{Z}}^2$ with trace image $\mathbb{Z}+\theta \mathbb{Z}$.

The intended statement reads schematically:

import Mathlib.Analysis.CStarAlgebra.Basic
import Mathlib.Analysis.CStarAlgebra.Matrix

variable {A B : Type*} [CStarAlgebra A] [CStarAlgebra B]

/-- Elliott: a scaled ordered-K₀ isomorphism classifies unital AF algebras. -/
theorem elliott_af_classification
    (hA : IsAF A) (hB : IsAF B)
    (α : K0Ordered A ≃o K0Ordered B) (hscale : α (K0.unitClass A) = K0.unitClass B) :
    Nonempty (A ≃⋆ₐ[ℂ] B) :=
  sorry  -- finite-stage multiplicity-matrix lifting + approximate intertwining

/-- Pimsner-Voiculescu / Rieffel: K₀ of the irrational rotation algebra. -/
theorem K0_irrational_rotation (θ : ℝ) (hθ : Irrational θ) :
    (K0 (RotationAlgebra θ) ≃+ (ℤ × ℤ)) ∧
      (traceImage (RotationAlgebra θ) = AddSubgroup.closure {1, θ}) :=
  sorry  -- six-term exact sequence for the crossed product C(T) ⋊_θ ℤ

Advanced results Master

The CAR algebra and the UHF classification. The CAR (canonical anticommutation relations) algebra is the AF algebra $M_{2^{\infty }}=\underrightarrow{\lim }{M}_{2^n}$, generated by a sequence of operators $a_i$ satisfying $a_i{a}_j+a_j{a}_i=0$ and $a_i{a}_j^{\ast }+a_j^{\ast }{a}_i={\delta }_{ij}1$; it is the observable algebra of a Fermi gas and is the unique unital AF algebra with $K_0=\mathbb{Z}[\frac{1}{2}]$, scale $[0,1]$, $[1]=1$. More generally the UHF algebras $M_{\mathfrak{p}}$ are indexed by supernatural numbers $\mathfrak{p}={\prod }_p{p}^{a_p}$ with $a_p\in \{0,1,2,\dots ,\infty \}$, and $M_{\mathfrak{p}}\cong M_{\mathfrak{q}}$ iff $\mathfrak{p}=\mathfrak{q}$, with dimension group the subgroup of $\mathbb{Q}$ of fractions whose denominators divide $\mathfrak{p}$. This was Glimm's classification, the prototype Elliott generalised.

Commutative AF algebras and the Cantor set. A commutative C*-algebra $C(X)$ is AF iff $X$ is totally disconnected (zero-dimensional) compact metric — for instance the Cantor set. Its finite-dimensional subalgebras are the locally constant functions on finite clopen partitions, the Bratteli diagram has multiplicity-one edges, and the dimension group is $C(X,\mathbb{Z})$ with the pointwise order. Bratteli-Jorgensen and Herman-Putnam-Skau showed these diagrams classify Cantor minimal systems up to strong orbit equivalence, linking AF theory to topological dynamics.

The irrational rotation algebra. For $\theta \in \mathbb{R}\setminus \mathbb{Q}$, the noncommutative torus $A_{\theta }=C^{\ast }(u,v:vu=e^{2\pi i\theta }uv)$ is the crossed product $C(\mathbb{T}){⋊}_{\theta }\mathbb{Z}$ of rotation by $\theta$, a simple unital C*-algebra with a unique normalised trace $\tau$ determined by $\tau (u^m{v}^n)={\delta }_{(m,n),(0,0)}$. The Pimsner-Voiculescu six-term exact sequence for a crossed product by $\mathbb{Z}$ computes $K_0(A_{\theta })\cong K_1(A_{\theta })\cong {\mathbb{Z}}^2$, and Rieffel constructed an explicit projection $e_{\theta }\in A_{\theta }$ (a band-limited element built from a continuous function on $\mathbb{T}$) with $\tau (e_{\theta })=\theta$. Thus the trace realises ${\tau }_{\ast }(K_0(A_{\theta }))=\mathbb{Z}+\theta \mathbb{Z}$, and since $A_{\theta }$ has a unique trace this ordered-group invariant is complete enough to give $A_{\theta }\cong A_{{\theta }^{\prime }}⟺{\theta }^{\prime }\equiv \pm \theta \mathrm{mod}1$. Although $A_{\theta }$ itself is not AF (it has nontrivial $K_1$), it is an inductive limit of circle algebras (an AT algebra) and is covered by the later Elliott-Evans classification, which is why it sits in this chapter as the first noncommutative space whose isomorphism class is pinned by ordered K-theory.

Synthesis. The dimension group is the central insight that organises this chapter: it is the foundational reason a tower of matrix blocks is reconstructible, because $K_0$ is a complete invariant at each finite stage and the inductive limit of the invariants is the invariant of the limit. This is exactly the noncommutative form of measure: the trace embeds $K_0$ into $\mathbb{R}$ as a group of reachable sizes — dyadic fractions for the CAR algebra, the dense group $\mathbb{Z}+\theta \mathbb{Z}$ for the rotation algebra — and the order is the positivity detected by states 39.01.03. Elliott's classification generalises Glimm's UHF theorem and is dual to the Gelfand reconstruction 39.01.02: where Gelfand rebuilds a commutative algebra from its characters, Elliott rebuilds an AF algebra from its projection-sizes. Putting these together, the irrational rotation algebra shows the program reaching genuinely noncommutative spaces: $A_{\theta }$ is a torus that no longer has points, yet its isomorphism type is read off a single irrational number through the trace on $K_0$, and the bridge is that classifying operator algebras has become classifying ordered abelian groups with scale — the launch of the Elliott program that runs through the rest of the subject.

Full proof set Master

Proposition (the inductive limit carries a unique C-norm).* Let $(F_n,{\varphi }_n)$ be a sequence of finite-dimensional C*-algebras with $\ast$-homomorphism connecting maps. On the algebraic direct limit $F_{\infty }={\underrightarrow{\lim }}_{\mathrm{alg}}{F}_n$ define $\Vert x\Vert ={\lim }_n\Vert x_n{\Vert }_{F_n}$ for any representative $x_n$ eventually mapping to $x$. The limit exists and is independent of the representative because each ${\varphi }_n$, being a $\ast$-homomorphism of C*-algebras, is norm-decreasing, so $\Vert {\varphi }_n(x_n)\Vert \le \Vert x_n\Vert$, and the sequence $\Vert x_n\Vert$ is non-increasing and bounded below by $0$; when ${\varphi }_n$ is injective it is isometric, so the limit equals the eventual constant value. This $\Vert \cdot \Vert$ satisfies the C*-axiom because each $\Vert \cdot {\Vert }_{F_n}$ does, and the completion $A=\overline{F_{\infty }}$ is a C*-algebra in which ${\bigcup }_n{\varphi }_{\infty ,n}(F_n)$ is dense.

Proposition ($V(A)$ has cancellation for AF $A$). In an AF algebra projections satisfy cancellation: $[p]+[r]=[q]+[r]$ in $V(A)$ implies $[p]=[q]$. At each finite stage $F_n={⨁}_i{M}_{n_i}$ this is the elementary fact that Murray-von Neumann equivalence of projections in a finite direct sum of matrix algebras is determined by the tuple of ranks, where cancellation is cancellation of natural numbers. Projections in $A$ (and in $M_k(A)$) are, by a standard perturbation argument, equivalent to projections living in some $F_n$ — a projection within $\epsilon$ of a self-adjoint element of $F_n$ is unitarily equivalent, via the functional calculus 39.01.01, to a projection in $F_n$ — so cancellation passes to the limit. Hence $V(A)$ embeds into its Grothendieck group $K_0(A)$, and the scale $\Sigma (A)$ is exactly the order interval $[0,[1]]$.

Proposition ($K_0$ of a finite-dimensional C-algebra).* For $F=M_{n_1}(\mathbb{C})\oplus \cdots \oplus M_{n_k}(\mathbb{C})$ one has $(K_0(F),K_0(F{)}_{+},[1])=({\mathbb{Z}}^k,{\mathbb{Z}}_{\ge 0}^k,(n_1,\dots ,n_k))$. A projection in $M_m(F)$ is a tuple of projections in matrix algebras, classified up to equivalence by the rank in each summand, giving $V(F)={\mathbb{Z}}_{\ge 0}^k$ with coordinatewise addition; Grothendieck completion gives ${\mathbb{Z}}^k$, the cone is the image of $V(F)$, and the class of the unit is the rank tuple $(n_1,\dots ,n_k)$. A unital $\ast$-homomorphism $F\to F^{\prime }$ induces the multiplicity matrix as a positive map ${\mathbb{Z}}^k\to {\mathbb{Z}}^{k^{\prime }}$ sending $[1_F]\mapsto [1_{F^{\prime }}]$.

Proposition ($A_{\theta }$ is simple with a unique trace for irrational $\theta$). The dual action of ${\mathbb{T}}^2$ on $A_{\theta }$, $(z,w)\cdot (u^m{v}^n)=z^m{w}^n{u}^m{v}^n$, has a faithful conditional expectation $E$ onto the fixed-point algebra $\mathbb{C}1$ given by averaging, $E(x)={\int }_{{\mathbb{T}}^2}(z,w)\cdot xdzdw$, and $\tau =E$ is a trace because $E(xy)=E(yx)$ on the dense span of monomials. Irrationality forces simplicity: a nonzero ideal $I$ is invariant under the (ergodic, since $\theta \notin \mathbb{Q}$) translation, and the conditional expectation of a positive element of $I$ is a positive scalar, so $1\in I$. Uniqueness of the trace follows because any trace must vanish on $u^m{v}^n$ for $(m,n)\ne 0$: the commutation relation gives $u(u^m{v}^n)u^{-1}=e^{-2\pi in\theta }{u}^m{v}^n$, so a trace sees $\tau (u^m{v}^n)=e^{-2\pi in\theta }\tau (u^m{v}^n)$, and $e^{-2\pi in\theta }\ne 1$ for $n\ne 0$ (irrationality) forces $\tau (u^m{v}^n)=0$; the same with $v$-conjugation handles $m\ne 0$. Hence $\tau$ is the averaging trace, unique.

Proposition (the trace image is $\mathbb{Z}+\theta \mathbb{Z}$). The Pimsner-Voiculescu sequence for $A_{\theta }=C(\mathbb{T}){⋊}_{\theta }\mathbb{Z}$ reads $K_0(C(\mathbb{T}))\stackrel{1-{\theta }_{\ast }}{\to }{K}_0(C(\mathbb{T}))\to K_0(A_{\theta })\to K_1(C(\mathbb{T}))\stackrel{1-{\theta }_{\ast }}{\to }{K}_1(C(\mathbb{T}))$. Since rotation is homotopic to the identity, ${\theta }_{\ast }=\mathrm{id}$ on $K$-theory, so $1-{\theta }_{\ast }=0$, and the sequence splits to give $K_0(A_{\theta })\cong K_0(C(\mathbb{T}))\oplus K_1(C(\mathbb{T}))\cong \mathbb{Z}\oplus \mathbb{Z}$. The generator $[1]$ has ${\tau }_{\ast }[1]=1$; the second generator is the class of the Rieffel projection $e_{\theta }$, for which Rieffel's explicit computation gives $\tau (e_{\theta })=\theta$. Hence ${\tau }_{\ast }(K_0(A_{\theta }))=\mathbb{Z}\cdot 1+\mathbb{Z}\cdot \theta =\mathbb{Z}+\theta \mathbb{Z}$, dense in $\mathbb{R}$ because $\theta$ is irrational.

Connections Master

• C-algebras: axioms, spectrum, and the continuous functional calculus 39.01.01* — the inductive-limit norm exists because $\ast$-homomorphisms of C*-algebras are norm-decreasing, and projections are produced and compared using the functional calculus on self-adjoint elements; the finite-dimensional building blocks are the matrix algebras whose spectrum and positivity were established there.

• States, the GNS construction, and Gelfand-Naimark 39.01.03 — the trace on an AF or rotation algebra is a tracial state, its GNS representation builds the finite von Neumann algebra closure, and the positive cone of $K_0$ is exactly the positivity detected by states; the irrational rotation algebra's unique trace is the averaging state used throughout this unit.

• Commutative C-algebras and Gelfand duality 39.01.02* — the commutative AF algebras are the $C(X)$ for $X$ totally disconnected, so the Cantor-set examples are Gelfand-dual to zero-dimensional spaces, and Elliott classification is the noncommutative analogue of recovering $X$ from $C(X)$.

• Effros-Handelman-Shen and the axiomatisation of dimension groups 39.02.04 pending — the abstract characterisation of which ordered abelian groups arise as $K_0$ of an AF algebra (unperforated Riesz-interpolation groups) completes Elliott's theorem by describing the range of the invariant, closing the classification loop opened here.

• Von Neumann algebras and the bicommutant theorem 39.03.01 — the weak closure of the GNS representation of the trace on the CAR algebra is the hyperfinite II${}_1$ factor, the von Neumann-algebraic limit of the same matrix tower, so AF C*-classification sits beneath Murray-von Neumann's uniqueness of the hyperfinite factor.

Historical & philosophical context Master

The inductive-limit construction of operator algebras as towers of matrix algebras goes back to Murray and von Neumann's hyperfinite factor and to Glimm, who in 1960 classified the UHF algebras by their supernatural numbers ^{[Glimm 1960]}. Bratteli introduced the diagrams that bear his name in 1972, giving a combinatorial encoding of the full AF class and its embeddings ^{[Bratteli 1972]}. The decisive step was Elliott's 1976 theorem that the scaled ordered $K_0$ group — the dimension group — is a complete isomorphism invariant for AF algebras ^{[Elliott 1976]}, the first instance of what became the Elliott classification program for nuclear C*-algebras; Effros, Handelman, and Shen then characterised the dimension groups abstractly as the unperforated Riesz-interpolation groups.

The irrational rotation algebra entered through Rieffel's 1981 construction of projections of arbitrary trace value $\theta$ and the Pimsner-Voiculescu exact sequence of 1980 computing its K-theory ^{[Rieffel 1981; Pimsner-Voiculescu 1980]}. Connes had already proposed $A_{\theta }$ as the prototype of a noncommutative space — a torus whose points have dissolved into the noncommuting coordinates $u,v$ — and the result that $\theta$ is recovered up to $\pm$ from the trace on $K_0$ made it the worked example of noncommutative topology, where an isomorphism class of algebras is detected by an arithmetic invariant rather than by any underlying point set.

Bibliography Master

• Davidson, K. R., C-Algebras by Example*, Fields Institute Monographs 6, American Mathematical Society, 1996. Ch. III, VI.
• Effros, E. G., Dimensions and C-Algebras*, CBMS Regional Conference Series in Mathematics 46, American Mathematical Society, 1981.
• Elliott, G. A., "On the classification of inductive limits of sequences of semisimple finite-dimensional algebras", Journal of Algebra 38 (1976), 29–44.
• Bratteli, O., "Inductive limits of finite dimensional C*-algebras", Transactions of the American Mathematical Society 171 (1972), 195–234.
• Glimm, J., "On a certain class of operator algebras", Transactions of the American Mathematical Society 95 (1960), 318–340.
• Pimsner, M. and Voiculescu, D., "Exact sequences for K-groups and Ext-groups of certain cross-product C*-algebras", Journal of Operator Theory 4 (1980), 93–118.
• Rieffel, M. A., "C*-algebras associated with irrational rotations", Pacific Journal of Mathematics 93 (1981), 415–429.
• Effros, E. G., Handelman, D. E., and Shen, C.-L., "Dimension groups and their affine representations", American Journal of Mathematics 102 (1980), 385–407.

---

Operator-algebras spine, opening unit of the AF-algebras / K-theory chapter. Produced as the Elliott-classification anchor: AF algebras as inductive limits of matrix algebras, Bratteli diagrams, the dimension group $(K_0,K_0^{+},\Sigma )$, the UHF/CAR and commutative-Cantor examples, and the irrational rotation algebra as the first noncommutative space classified by ordered K-theory.