AN INTRODUCTION TO GRAPH C ∗ -ALGEBRAS

DAVID PASK

Abstract. These are notes from the lectures given at the CIMPA workshop Leavitt
path algebrs and graph C ∗ -algebras held at the Nesin Mathematics Village in Turkey, 29
June – July 12 2015. They are intended for wide circulation in due course. If you find
any errors in these notes please do not hesistate to contact me at the address given at
the end of these notes.

1. A bit of history
1977-1981 Seminal papers by Cuntz and Krieger, [10, 12, 11].
1981-1982 Papers by Watatani and Enomoto giving graphical iterpretation, [14].
1993-1995 Papers by Mann, Pask, Raeburn and Sutherland on Doplicher–Roberts algebras
and connection with Cuntz-Krieger algebras, [18, 21].
1996-2000 Papers by Bates, Kumjian, Pask, Raeburn, Renault, Szymánski developing theory
of infinite Cutz-Krieger algebras and then graph C ∗ -algebras associated to row-
finite directed graphs, [20, 16, 15, 6].
2000-2015 Explosion of interest in graph C ∗ -algebras, generalisation of earlier results and
applications to nonabelian duality, discrete topology.

2. Directed graph basics

A directed graph E = (E 0 , E 1 , r, s) consists of a countable set E 0 of vertices, a countable
set E 1 of edges, and maps r, s : E 1 → E 0 giving the direction of each edge.
e
s(e) r(e)

A path in E of length n ≥ 1 is a sequence α = α1 · · · αn of edges such that r(αi ) =

s(αi+1 ) for i = 1, . . . , n − 1.
s(αi ) αi r(αi )
... ...
s(αi+1 ) αi+1 r(αi+1 )

Such α is said to have length n, the set of paths in E of length n is denoted E n . We set
s(α) = s(e1 ) and r(α) = r(en ).
Note: In this note paths run from left to right in accordance with the usual conventions
of graph theory. In 2004 the direction was changed to right to left in order to mimic
the order of composition of operators and comply more with the conventions of category
theory (see [22]) – this convention is often termed “the Australian convention”. In order
not to cause confusion we will use the old convention as the majority of Leavitt path
algebra literature uses it.
Date: July 24, 2015.
1
2 DAVID PASK

If v ∈ E 0 then we define r(v) = s(v) = v. v ∈ E 0 is a sink if it emits no edges (s−1 (v) = ∅).
v ∈ E 0 is a source if it receives no edges (r−1 (v) = ∅). E is row-finite if each vertex has
finite out-valency, i.e. |s−1 (v)| < ∞ for all v ∈ E 0 . Let E ∗ = ∪n≥0 E n denote the set of
finite paths in E. E ∗ can be given the structure of a category.
A path α ∈ E n where n ≥ 1 is a cycle based at v = s(α) if r(α) = s(α) and r(α1 ), . . . r(αn )
are distinct. A infinite path is an infinite sequence x = (xi )i≥1 of edges such that r(xi ) =
s(xi+1 ) for all i ≥ 1. We set s(x) = s(x1 ). The set of infinite paths in E is denoted E ∞ .
A boundary path in E is either an infinite path x ∈ E ∞ or α ∈ E ∗ with r(α) a sink. We
denote the set of boundary paths in E by ∂E. If E has no sinks then ∂E = E ∞ .
Example 2.1.
f
u x

E := e h

w v
g

The graph E is row-finite, with sink at w and source at x. The boundary paths are
∂E = {f e, hg, e, g, w}

3. Cuntz-Krieger families
Definition 3.1. Let E be a row-finite graph, then a Cuntz-Krieger E-family {S, P } on
a Hilbert space H consists of mutually orthogonal projections {Pv : v ∈ E 0 } and partial
isometries {Se : e ∈ E 1 } satisfying
(CK1) for all e ∈ E 1 we have Se∗ Se = δe,f Pr(e) ;
(CK2) for all v ∈ E 0 which is not a sink we have Pv = {e:s(e)=v} Se Se∗ .
P

This definition was later ammended to deal with non row-finite graphs (see [5]).
Example 3.2. For H = `2 (∂E)1we may define operators Pv , Se 6= 0 which satisfy (i)-(iv)
above, so there is always a non-degenerate Cuntz-Krieger E-family (that is a Cuntz-
Krieger E family for which Pv 6= 0 for all v ∈ E 0 ):
(
δx if v = s(x),
Pv δx =
0 otherwise.

We may define {Se : e ∈ E 1 } on H by


δex if x is not a sink and r(e) = s(x),

S e δx = δe if x is a sink and r(e) = s(x),

0 otherwise.

1The Hilbert space completion of the finitely supported complex functions on ∂E.
 AN INTRODUCTION TO GRAPH C ∗ -ALGEBRAS 3

Hence 
δy
 if x is not a sink and x = ey,
∗
Se δx = δr(e) if x = e and r(e) is a sink ,

0 otherwise.
Since {δx : x ∈ ∂E} is a basis for H these formulas determine bounded linear maps
Pv , Se : H → H which are mutually orthogonal projections and partial isometries. It
is then straightforward to check that these operators satisfy conditions (CK1)–(CK2) of
Definition 3.1. Note that `2 (∂X) is not necessarily separable. With a bit more work we
can modify this construction to one based on a seperable Hilbert space.
Lemma 3.3 ([12, 1, 22]). Let {S, P } be a non-degenerate Cuntz-Krieger E-family then:
(a) For any path α = α1 · · · αn in E n , then Sα := Sα1 · · · Sαn is a nonzero partial isometry.
Similarly Sα∗ = Sα∗ n · · · Sα∗ 1 is a nonzero partial isometry.
(b) For any paths α, β in E ∗ with r(α) = r(β), then Sα Sβ∗ is a nonzero partial isometry.
(c) Any finite product of {Pv , Se : v ∈ E 0 , e ∈ E 1 } can be written as a finite sum of
elements of the form Sα Sβ∗ where α, β ∈ E ∗ with r(α) = r(β).
(d) C ∗ ({S, P }) = span{Sα Sβ∗ : r(α) = r(β)}, in particular C ∗ ({S, P }) is separable as
{α, β ∈ E ∗ : r(α) = r(β)} is countable.
Note LC (E) ∼ = span{Sα S ∗ : r(α) = r(β)}.
β

f
u x
Examples 3.4. (i) Let E := e h as in Example 2.1. Then if we define {P, S} on
w v
g
`2 (∂E) as in Example 3.2 then since `2 (∂E) ∼ = C5 as a vector space then we may
associate the generators of C ∗ ({P, S}) with the matrix units in M5 (C) as shown
∗
Sf e Sf e Sg∗ Sf e Shg
 
Pw Se
 Se∗ Pu Sf Se Sg∗ Se Shg ∗ 
 Sf∗e ∗ ∗ ∗
 
S f P x S h S hg

∗ ∗ ∗
 
 Sg Sf e Sg Se Sh Pv Sg 
Shg Sf∗e Shg Se∗ Shg Sg Pw
then we may see that C ∗ ({P, S}) ∼ = M5 (C).
(ii) Let F be the directed graph v . e Then ∂F = F ∞ = {eee . . .}, so `2 (∂F ) = C
as a vector space. Then we have Se = Se∗ = Pv , so C ∗ ({S, P }) ∼= C.
(iii) If X is any closed subset of T then C(X) the space of continuous functions on X
acts by multiplication on L2 (X) the Hilbert space of square integrable functions
on X. If we define (Qv f )(z) = f (z) and (Te f )(z) = zf (z) for f ∈ L2 (X) then
Te Te∗ = Qv = Te∗ Te so {T, Q} is a Cuntz-Krieger F –family and C ∗ ({T, Q}) ∼ = C(X)
by the Stone-Weierstrass Theorem.
(iv) The F family in (ii) corresponds to taking X = {1} ⊂ T.
The Cuntz-Krieger relations for F in Examples 3.4 (ii) imply that Se is a unitary – so there
are lots of Cuntz-Krieger F -families, generating different C ∗ -algebras. To get around this
problem, we use a universal property to define the C ∗ -algebra associated to a directed
graph.
4 DAVID PASK

Theorem 3.5 (Existence of Universal C ∗ -algebra – see [1], [16], [6]). Let E be a row-
finite directed graph. There is a C ∗ -algebra C ∗ (E) generated by a Cuntz-Krieger E-family
{p, s} such that if {Q, T } is a Cuntz-Krieger E-family in a C ∗ -algebra B then there is a
∗-homomorphism πQ,T : C ∗ (E) → B such that πQ,T (pv ) = Qv and πQ,T (se ) = Te .
For the graph F in Example 3.4 (ii), we have C ∗ (F ) ∼
= C(T). By convention the generators
of the universal algebra are written using lower case letters and concrete Cuntz-Krieger
families in a C ∗ -algebra are written using upper case letters.

4. Cuntz-Krieger algebras
In [12] Cuntz and Krieger study a C ∗ -algebra A associated to a 0-1 matrix A = (aij )ni,j=1
with no zero rows or columns. They then go on to prove that A has a universal propery
provided that the matrix A satisfies a condition called (I). Following [1] we now define the
Cuntz-Krieger algebra OA be the universal C ∗ -algebra generated by a family {s1 , . . . , sn }
of partial isometries satisfying

n
X
(4.1) s∗i si = aij sj s∗j .
j=1

The relation (4.1) looks a lot like relation (CK2) from Definition 3.1. This is because there
is a close relationship between Cuntz-Krieger algebras and the C ∗ -algebras associated to
finite graphs with no sinks or sources. In fact [1, Theorem 2.1] defines a Leavitt path
algebra associated to A, and later in [15, Theorem 1.2] for row-finite directed graphs.
Furthermore OA and C ∗ (E) are defined to be the completion of this Leavitt path algebra
with respect to a certain universal norm k · k0 .
For a 0-1 matrix A = (aij )ni,j=1 with no zero rows or columns define a directed graph
EA by EA0 = {1, . . . , n}, EA1 = {ij : aij = 1} and r(ij) = j, s(ij) = i. Since A has no zero
rows or columns the graph EA has no sinks or sources or multiple edges.
Let {s1 , . . . , sn } be partial isometries satisfying (4.1) for some A then we may define
Qi = si s∗i and Tij = si sj s∗j where aij = 1, i, j = 1, . . . , n.
One checks that {Q, T } is a Cuntz-Krieger EA family in OA , and so by the universal
property of C ∗ (EA ) there is a ∗-homomorphism πQ,T : C ∗ (EA ) → C ∗ ({Q, T }) ⊆ OA .
This then raises the question: Is C ∗ (EA ) isomorphic to OA ? This graphical approach
Cuntz-Krieger algebras was spotted in the 1980’s by Enomoto and Watatani in [14], and
we will come back to it again soon.
Proposition 4.1 (see [1], [6]). Let E be a row-finite directed graph. Then there is a
strongly continuous action γ of T on C ∗ (E) such that γz (se ) = zse and γz (pv ) = pv .
Sketch proof. Let {p, s} be a Cuntz-Krieger E-family generating C ∗ (E). Then for each z ∈
T the elements {p, zs} are also a Cuntt-Krieger E-family and so by the universal property
there is a ∗-homomorphism γz := πp,zs : C ∗ (E) → C ∗ (E) which is a ∗-automorphism
since γz γz = γz γz is the identity map on C ∗ (E). One checks that γz is multiplicative and
strongly continuous using a standard /3 argument. 
The existence and importance of the gauge action on C ∗ (E) can be traced back to the
original paper of Cuntz and Krieger (see [12]).
 AN INTRODUCTION TO GRAPH C ∗ -ALGEBRAS 5

Theorem 4.2 (Gauge-Invariant Uniqueness Theorem – see [1],[6]). Let E be a row-finite

directed graph and suppose that {Q, T } is a Cuntz-Krieger E-family in a C ∗ -algebra B
with Qv 6= 0 for all v ∈ E 0 . If there is a continuous action β : T → Aut(C ∗ ({Q, T }) such
that
(4.2) βz (Te ) = zTe and βz (Qv ) = Qv , where z ∈ T,
then πT,Q : C (E) → C ∗ ({Q, T }) is an isomorphism.
∗

We nw give an application of the use of the gauge–invariant uniqueness theorem: Let

E = (E 0 , E 1 , r, s) be a row-finite directed graph with no sources. Define E(1, 2) =
(E 1 , E 2 , r0 , s0 ) where
r0 (ef ) = f and s0 (ef ) = e.
It is easy to see that E(1, 2) is also row-finite with no sources.
Corollary 4.3 (see [6]). Let E be a row-finite graph with no sinks, then C ∗ (E) ∼ =
C ∗ (E(1, 2)).
Sketch proof. Let {p, s} be the universal Cuntz-Krieger family generating C ∗ (E). Define
Qe = se s∗e , Tef = se sf s∗f , then {Q, T } is a Cuntz-Krieger E(1, 2)-family. Since the se are
nonzero partial isometries the Qe are nonzero projections. One checks that the map πQ,T
intertwines the gauge actions on C ∗ (E) and C ∗ (E(1, 2)), so Theorem 4.2 implies that πQ,T
is injective. Since E has no sinks we can show that se , pv lie in the range of πQ,T , and so
it is surjective. 
An argument similar to Corollary 4.3 shows that OA ∼ = C ∗ (EA ).
1 1
For a graph E, the E × E edge matrix BE is defined by
(
1 if r(e) = s(f ),
BE (e, f ) =
0 otherwise.
Suppose that E has no sinks and sources and E 0 , E 1 are finite, then BE is a 0-1 matrix
with no zero rows or columns.
Observe that EBE ∼ = E(1, 2), then we have
C ∗ (E) ∼
= C ∗ (E(1, 2)) ∼
= C ∗ (EB ) ∼
E = OB . E

Hence every Cuntz-Krieger algebra is a graph C -algebra, and every graph C ∗ -algebra of
∗

a finite graph with no sinks and sources is a Cuntz-Krieger algebra. So for finite graphs
with no sinks or sources, there is a close relationship between Cuntz-Krieger algebras and
graph C ∗ -algebras.
Example 4.4. Recall the graph F from Example 3.4 (ii).
v. e

It was shown that there is a Cuntz-Krieger F -family {Q, T } with C ∗ ({Q, T }) ∼

= C whereas
C ∗ (F ) ∼
= C(T). Certainly Qv 6= 0, however since the spectrum of Te is not all of T
there is no action β of T on C ∗ ({Q, T }) ∼
= C which satisfies (4.2). Hence Theorem 4.2
cannot be used to deduce that πT,Q is an isomorphism unless Te has spectrum T, that is
C ∗ ({Q, T }) ∼
= C(T).
Definition 4.5. Let E be a directed graph. A cycle α ∈ E n , n ≥ 1 has an exit if there
is 1 ≤ j ≤ n and e 6= αj ∈ E 1 such that s(e) = s(αj ).
6 DAVID PASK

αj
αj−1 e

Th property that every cycle has an exit is often known as condition (L) (see [15]).
Theorem 4.6 (Cuntz-Krieger Uniqueness Theorem – see [6, 5, 22]). Let E be a row-
finite directed graph in which every cycle has an exit. Let {Q, T } be a Cuntz-Krieger
E-family in a C ∗ -algebra B such that Qv 6= 0 for every v ∈ E 0 . Then the homomorphism
πQ,T : C ∗ (E) → B is an isomorphism of C ∗ (E) onto C ∗ ({Q, T }).
Theorem 4.6 is named after a similar result in [12] for Cuntz-Krieger algebras which were
not defined by a universal property. The version of Theorem 4.6 in [12] says that if A
satisfies condition (I) then all C ∗ -algebras generated by partial isometries satisfying (4.1)
are isomorphic. If A satisfies condition (I) then every cycle in EA has an exit and the
Cuntz-Krieger Uniqueness Theorem gives us an alternative proof that C ∗ (EA ) ∼ = OA ∼ = A.
Example 4.7. Recall the graph F from Example 3.4 (ii).

v. e

In Example 3.4 (iv) we showed that there is a Cuntz-Krieger F -family {Q, T } with
C ({Q, T }) ∼
∗
= C whereas C ∗ (F ) ∼= C(T). Certainly Qv 6= 0, however the loop e does not
have an exit and so the Cuntz-Krieger uniqueness theorem does not apply, and gives us
an alternative proof that C ∗ ({Q, T }) is not isomorphic to C(T).

5. Simplicity
Definition 5.1 (see [22]). Let E be a row-finite directed graph. Then E is cofinal if for
every x ∈ ∂E and v ∈ E 0 there is i ≥ 1 and α ∈ E ∗ with s(α) = v and r(α) = s(xi ).
v α
x1 x2 x3
...
s(x)

Theorem 5.2 (see [22]). Let E be a row-finite graph then C ∗ (E) is simple if and only if
every cycle has an exit and E is cofinal.

Examples 5.3. (i) The graph is cofinal but has a cycle without an exit and so its
∗ ∼
C (E) = C(T) which is not simple.
(ii) The graph is cofinal and every cycle has an exit and so is simple; in fact
C ∗ (E) ∼
= O2 .
(iii) The graph is cofinal and there is a cycle without an exit, and is hence not
simple. In fact C (E) ∼
∗
= M2 (C(T)).
 AN INTRODUCTION TO GRAPH C ∗ -ALGEBRAS 7

(iv) The graph is not cofinal even though it has a cycle which has an exit and
so its C -algebra is not simple; in fact its C ∗ -algebra is isomorphic to the Toeplitz
∗

algebra, T .
(v) The graph . . . is cofinal and has no cycles, so its C ∗ -algebra,
the compact operators K, is simple.
(vi) The graph . . . is cofinal and has no cycles, so its C ∗ -algebra,
the compact operators K, is simple.
Definition 5.4. A graph E is strongly connected or transitive if for every u, v ∈ E 0 there
is α ∈ E ∗ with s(α) = u and r(α) = v.
The strongly connected condition is usually only used for finite graphs.
Corollary 5.5 (see [12]). Let E be a finite graph which is not itself a cycle. Then C ∗ (E)
is simple if and only if E is strongly connected.
Example 5.6. The following finite graph is strongly connected and not a cycle.

RP := • •

Hence C ∗ (RP ) is simple.

6. Purely infinite C ∗ -algebras

Let p, q be projections in a C ∗ -algebra A.
(a) p ≤ q if and only if pq = p,
(b) p ∼ q if there is a partial isometry u ∈ A such that uu∗ = q and u∗ u = p,
(c) p is infinite if there is a projection q 6= p such that q ≤ p and p ∼ q.
Let a be an element of a C ∗ -algebra A then a is positive if and only if there is b ∈ A
such that a = b∗ b, and we write a ≥ 0. A C ∗ -subalgebra B of A is hereditary if whenever
0 ≤ a ≤ b where b ∈ B and a ∈ A we have a ∈ B.
Definition 6.1. A simple C ∗ -algebra A is purely infinite if every hereditary C ∗ -subalgebra
contains an infinite projection.
In the nonsimple case alternative (and inequivalent) definitions of purely infinite are often
used (see [23, Proposition 4.1.1]).
Lemma 6.2 (see [6]). Let E be a row-finite directed graph, and α a cycle with an exit.
Then pr(α) ∈ C ∗ (E) is an infinite projection.
Proof. Without loss of generality let e 6= α1 be such that s(e) = s(α1 ). Let {p, s} be a
Cuntz-Krieger E-family generating C ∗ (E), then
ps(α) = s∗α sα ∼ sα s∗α ≤ sα1 s∗α1 < sα1 s∗α1 + se s∗e ≤ ps(α) .
Hence ps(α) ≥ sα s∗α and ps(α) ∼ sα s∗α and so is an infinite projection. 
Theorem 6.3 (see [15]). Let E be a row-finite graph which is cofinal and every cycle has
an exit. Then C ∗ (E) is purely infinite if and only if for every vertex v ∈ E 0 there is a
cycle α with v ≤ s(α).
8 DAVID PASK

Examples 6.4. (i) The C ∗ -algebra of the graph RP := • • is simple

and purely infinite since it is cofinal and every vertex lies on a cycle.
In fact the graph is isomorphic to R2 (1, 2) where R2 is the graph Hence
∗
its C -algebra is isomorphic to the Cuntz-algebra O2 .
(ii) The C ∗ -algebra of the graph . . . is not purely infinite since it
has no cycles at all and so the second condition in Theorem 6.3 cannot be satisfied.
As we saw earlier, its C ∗ -algebra is isomorphic to the compact operators K, which
is an AF algebra. If we write K = ∪∞ n=1 An where An = Mn (C), then this graph is
the Bratteli diagram for this approximating sequence.
Definition 6.5. A C ∗ -algebra A is approximately finite dimensional (AF) if there is a
sequence An of finite dimensional C ∗ -algebras such that An ⊆ An+1 and A = ∪∞
n=1 An .

Theorem 6.6 (see [15]). Let E be a row-finite directed graph. Then C ∗ (E) is AF if and
only if E has no cycles.
If C ∗ (E) is simple, we have the following dichotomy.
Theorem 6.7 (see [15]). Let E be a row-finite directed graph which is cofinal and every
cycle has an exit. Then C ∗ (E) is either purely infinite or AF.
Proof. Suppose E has no cycles, then C ∗ (E) is AF by Theorem 6.6.
Suppose E has a cycle α, then by cofinality for every v ∈ E 0 we have s(α) ≥ v and
hence C ∗ (E) is purely infinite by Theorem 6.3. 

7. Shift spaces
For an excellent treatment of shift spaces see the book by Lind and Marcus [17]. Let A
be a set, usually finite, of symbols, called the alphabet. Let A∗ denote the Kleene Star of
A, which consists of words or finite strings of symbols drawn from A. The full A-shift is
the set AZ together with the shift map σ : AZ → AZ defined by
(σx)n = xn+1 n ∈ Z.
A shift space is a closed, shift invariant subset X of AZ . x ∈ X has period p if p is the
least positive number such that σ p (x) = x.
A shift space X ⊂ AZ can be described by giving a list F ⊂ A∗ of forbidden words that
is
X = XF := {y ∈ AZ : no subword of y contains an element of F}.
For n ≥ 1, the subwords of length n appearing in x ∈ X are denoted Bn (X), then
B(X) = ∪n≥1 Bn (X) denotes the collection of all words which occur in some x ∈ X.
The set B(X) is usually called the language of X. The languages of shift spaces are
characterised by the following two properties:
Prolongable for all w ∈ B(X) there are a, b ∈ A such that awb ∈ B(X).
Factorial for all w ∈ B(X) every subword of w belongs to B(X).
The language of a shift space X completely determines X; that is is X and Y are shift
spaces with the same language, then X = Y .
 AN INTRODUCTION TO GRAPH C ∗ -ALGEBRAS 9

The entropy h(X) of a shift space X is a measure of the complexity of X in terms of

the growth rate of the size of B(X). Hence we have
1
h(X) = lim log |Bn (X)|,
n→∞ n

where the limit in the definition can be shown to exist (see [17]). A shift space X is called
irreducible if for all u, v ∈ B(X) there is w ∈ B(X) such that uwv ∈ B(X).
The shift spaces (X1 , σ1 ), (X2 , σ2 ) are conjugate if there is a homeomorphism h : X1 →
X2 such that h ◦ σ1 = σ2 ◦ h.
There is a weaker notion of equivalence between shift spaces called flow-equivalence:
Let (X, σ) be a shift space, then the suspension flow is defined to be SX = (X × R)/ ∼
where (x, t + 1) ∼ (σ(x), t) with dynamic φt ([x, s]) = [x, s + t)]. An flow equivalence
between the flows (U, {φt }) and (V, {ψt }) is a homeomorphism π : U → V which maps
orbits of {φt } to orbits of {ψt } in an orientation-preserving way; that is for all x ∈ U we
have π(ψ(t)x) = ψfx (t) (π(x)) for some fx : R → R which is monotonically increasing.

8. Shifts of finite type

Definition 8.1. A shift space X over a finite alphabet A is of finite type if it may be
described using a finite list of forbidden words.
Example 8.2. Let E = (E 0 , E 1 , r, s) be a finite directed graph with no sinks or sources.
Let
XE = {x ∈ (E 1 )Z : s(xi+1 ) = r(xi ), i ≥ 1}
then XE is a shift of finite type with F = {ef : r(e) 6= s(f )}. The shift space XE is called
the edge shift associated to E. Note that B(XE ) = E ∗ \E 0 . Furthermore X is irreducible
if and only if E is strongly connected.
Theorem 8.3 (Folklore). Let X be a shift of finite type, then there is a directed graph E
such that X is conjugate to XE .
One of the most basic examples of a conjugacy is to take shift space and create a new
shift using an alphabet which consists of words of a fixed length in the original shift, this
process is called a higher-block code, which can easily be seen for shifts of finite type.
Example 8.4. Let E = (E 0 , E 1 , r, s) be a directed graph with no sinks; for m ≥ 0 let
E(m, m + 1) = (E m , E m+1 , r, s) where
s(α1 . . . αm+1 ) = α1 . . . αm and r(α1 . . . αm+1 ) = α2 . . . αm+1 ,
so that E(0, 1) = E.

Then XE is conjugate to XE(m,m+1) . For instance if R2 is the graph a v b , then

ab

R2 (1, 2) is the graph aa a b bb . Hence XR2 is conjugate to XR2 (1,2) . Notice

ba
that R2 (1, 2) ∼
= RP .
More generally there are graphical moves called in/out-splitting and in/out-amalgamation
which induce conjugacies amongst shifts of finite type:
10 DAVID PASK

The idea behind an outsplitting of a graph E is as follows: Fix v ∈ E 0 and partition

s−1 (v) into non-empty sets P1 , . . . , Pn . Remove v from the graph and replace it by n
copies {v1 , . . . , vn } and distribute the edges in s−1 (v) such that vi emits the edges in Pi
for i = 1, . . . , n. Remove r−1 (v) from the graph and replace them by n copies {ei : 1 ≤ i ≤
n, e ∈ r−1 (v)}. For each 1 ≤ i ≤ n place the edges {ei : e ∈ r−1 (v)} such that s(ei ) = s(e)
and r(ei ) = vi .
For an insplitting, we proceed in a similar way, but work with edges with range v:
Fix v ∈ E 0 and partition P −1 (v) into non-empty sets P1 , . . . , Pn . Remove v from the
graph and replace it by n copies {v1 , . . . , vn } and distribute the edges in r−1 (v) such that
vi receives the edges in Pi for i = 1, . . . , n. Remove s−1 (v) from the graph and replace
them by n copies {ei : 1 ≤ i ≤ n, e ∈ r−1 (v)}. For each 1 ≤ i ≤ n place the edges
{ei : e ∈ r−1 (v)} such that s(ei ) = vi and r(ei ) = r(e).
For more information about in- or out-splittings see [17].
Example 8.5. The graphs E, F shown below are related via an out-splitting procedure,
where vertex v splits into v1 , which emits e, f and v2 which emits g; and so XE is conjugate
to XF .
f2
f e e 1

v1 e2 v1
E := v F := g1

g f1 g2

Moving from F to E is termed an in–amalgamation. For more details on in/out–splittings

and in/out–amalgamations see [17].
The process of inserting (resp. deleting) a vertex in an edge emitted by a vertex v, is a
graphical move called an delay.
... ... ⇔ ... ...

Example 8.6. The graph F is obtained from E by inserting a vertex in the edge f . So XE
and XF are flow equivalent.

e f e1 f (1)
E := v F := v1 v2

Theorem 8.7 ([25],[21]). Let (XE , σE ), (XF , σF ) be shifts of finite type. Then they are
conjugate if and only if there are a sequence E = E1 , . . . , En = F of directed graphs such
that Ei+1 is either an in/out-splitting or in/out-amalgamation of Ei for i = 1, . . . , n − 1.
If, in addition, Ei+1 is a delay of Ei then (E, σE ) is flow equivalent to (XF , σF ).
We now investigate how the graph moves involved in conjugacy and flow-equivalence for
shifts of finite type affect their associated graph C ∗ -algebras. The first hint that something
nice happens is given by:
Theorem 8.8 (see [8]). Let E be a row-finite graph with no sinks and m ≥ 0. Let
{p, s} be a Cuntz-Krieger E-family generating C ∗ (E). For α ∈ E m and β ∈ E m+1 set
 AN INTRODUCTION TO GRAPH C ∗ -ALGEBRAS 11

Tβ := sβ1 ...βm s∗β , Qα = sα s∗α . Then {Q, T } is a Cuntz-Krieger E(m, m + 1) family in

C ∗ (E) and the map πT,Q : C ∗ (E(m, m + 1)) → C ∗ (E) is an isomorphism.

Example 8.9. If R2 is the graph a v b , then C ∗ (R2 ) ∼

= C ∗ (R2 (1, 2)) where R2 (1, 2)

ab
is the graph aa a b bb
.
ba

In [3] (see also [13, 9]) it was shown how in/out-splittings and in/out-amalgamations
and delays affect the corresponding graph C ∗ -algebras of a graph with no sinks (for more
general moves in [24]):
Theorem 8.10 ([3]). Let E be a graph with no sinks. Then
(i) if F is formed from an outsplitting of E then C ∗ (E) ∼= C ∗ (F ),
(ii) if F is formed from an insplitting of E then C ∗ (E) and C ∗ (F ) are morita equivalent,
(iii) if F is formed from a delay of E then C ∗ (E) and C ∗ (F ) are morita equivalent.
More general results may be found in [24].
Examples 8.11. For the graphs E, F below, C ∗ (E) and C ∗ (F ) are Morita equivalent since
the graph F is obtained from E by a delay.

e f e1 f (1)
E := v F := v1 v2

For the graphs E, F below, C ∗ (E) and C ∗ (F ) are isomorphic since the graph F is obtained
from E by an outsplitting.
f2
f e e 1

v1 e2 v1
E := v F := g1

g f1 g2

Let E be a row-finite graph with no sinks and r vertices. Define the r × r vertex matrix
AE by
AE (v, w) = #{e ∈ E 1 : s(e) = v, r(e) = w}.
Then the Bowen-Franks group associated to the shift of finite type (XE , σE ) is
BF (XE ) = coker(1 − AtE : Zr → Zr ) := Zr / Im(1 − AtE )Zr .
In [7] Bowen and Franks showed that this abelian group is an invariant for conjugacy
and flow equivalence of shifts of finite type.
For a C ∗ -algebra A there are two abelian groups K0 (A), K1 (A) which are invariants for
Morita equiavlence. Roughly speaking, K0 (A) measures the dimension of projections in
A and K1 (A) measures the connectivity of the unitaries in A.
12 DAVID PASK

Theorem 8.12 (see [11], [6]). Let E be a row-finite directed graph with no sinks and r
vertices. Then
K0 (C ∗ (E)) ∼
= coker(1 − At : Zr → Zr ), E

K1 (C (E)) ∼
∗
= ker(1 − AtE : Zr → Zr ).
In particular we note that K0 (C ∗ (E)) is the Bowen-Franks group for the shift of finite
type XE .
9. Back to graphs
Let E be a finite, row-finite directed graph, then DE := C ∗ ({sµ s∗µ : µ ∈ E ∗ }) is a
commutative subalgebra (masa) of C ∗ (E) called the diagonal of C ∗ (E). It can be shown
that DE is a maximum abelian subalgebra of C ∗ (E), and that DE ∼ = C(E ∞ ) via the
isomorphism sµ s∗µ 7→ 1Z(µ) where Z(µ) =P{x ∈ E ∞ : x = µy, for some y ∈ E ∞ }.
Define φ : C ∗ (E) → C ∗ (E) by φ(x) = e∈E 1 s∗e xse . The map φ is not a ∗-homomorphism
but is completely positive. For such maps there is a notion of entropy, and it can be shown
that h(φ) = h(XE ).
References
[1] A. An Huef and I. Raeburn, The ideal structure of Cuntz–Krieger algebras, Ergod. Thy. &
Dynam. Sys. 17 (1997), 611-624.
[2] T. Bates, Applications of the gauge-invariant uniqueness theorem, Bull. Austral. Math. Soc., 66
(2002), 57–67.
[3] T. Bates and D. Pask, Flow equivalence of graph algebras, Ergodic Th. & Dynam. Sys. 24 (2004),
367–382.
[4] T. Bates and D. Pask, C ∗ -algebras of labelled graphs, J. Operator Theory 57 (2007), 207–226.
[5] T. Bates, J. Hong, I. Raeburn and W. Szymański, The ideal structure of the C ∗ -algebras of
infinite graphs, Illinois J. Math., 46 (2002), 1159–1176.
[6] T. Bates, D. Pask, I. Raeburn and W. Szymański, The C ∗ -algebras of row-finite graphs, New
York J. Math. 6 (2000), 307–324.
[7] R. Bowen and J. Franks, Homology for zero-dimensional nonwandering sets, Ann. Math. 106
(1977), 73–92.
[8] O. Bratteli, Inductive limits of finite-dimensional C ∗ -algebras, Trans. Amer. Math. Soc. 171
(1972), 195–234.
[9] T. Crisp and D. Gow, Contractible subgraphs and Morita equivalence of graph C ∗ -algebras, Proc.
Amer. Math. Soc. 135 (2006), 2003-2013.
[10] J. Cuntz, Simple C ∗ -algebras generated by isometries, Comm. Math. Phys. 57 (1977), 173-185.
[11] J. Cuntz, A class of C ∗ -algebras and topological Markov chains II: Reducible chains and the
Ext-functor for C ∗ -algebras, Invent. Math., 63 (1981) 25–40.
[12] J. Cuntz and W. Krieger, A class of C ∗ -algebras and topological Markov chains, Invent. Math.
56 (1980), 251–268.
[13] D. Drinen and N. Sieben, C ∗ -equivalences of graphs, J. Operator Theory, 45 (2001), 209–229.
[14] M. Enomoto and Y. Watatani, A graph theory for C ∗ -algebras, Math. Japon 25 (1980), 435–442.
[15] A. Kumjian, D. Pask and I. Raeburn, Cuntz–Krieger algebras of directed graphs, Pacific. J. Math.
184 (1998), 161–174.
[16] A. Kumjian, D. Pask, I. Raeburn, and J. Renault, Graphs, groupoids and Cuntz–Krieger algebras,
J. Funct. Anal. 144 (1997), 505–541.
[17] D. Lind and B. Marcus, An introduction to symbolic dynamics and coding. Cambridge University
Press., 1995.
[18] S. Mann, I. Raeburn and C. Sutherland,
[19] W. Parry, W. and D. Sullivan, A topological invariant of flows on 1-dimensional spaces, Topology
14 (1975), 297–299.
 AN INTRODUCTION TO GRAPH C ∗ -ALGEBRAS 13

[20] D. Pask and I. Raeburn,

[21] D. Pask and C. Sutherland,
[22] I. Raeburn, Graph Algebras, CBMS Regional Conference Series in Mathematics, 103, Amer.
Math. Soc., (2005).
[23] M. Rørdam, Classification of nuclear C ∗ -algebras,...
[24] A. Sørensen, Geometric classification of simple graph algebras, arXiv:1111.1592).
[25] R.F. Williams, Classification of subshifts of finite type, Ann. of Math. 98 (1973) 120–153.

School of Mathematics and Applied Statistics, University of Wollongong, NSW 2522,

AUSTRALIA
E-mail address: dpask@uow.edu.au