CONTINUOUS GRAPHS AND C*-ALGEBRAS

Transcription

1 CONTINUOUS GRAPHS AND C*-ALGEBRAS by VALENTIN DEACONU 1 Timisoara, June 1998 Abstract We present the continuous graph approach for some generalizations of the Cuntz- Krieger algebras. These C*-algebras are simple, nuclear, and purely infinite, with rich K-theory. They are tied with the dynamics of a shift on an infinite path space. We explain the connection with the Ruelle algebras of Putnam, the Pimsner algebras, and the polymorphisms of Arzumanian and Renault Mathematics Subject Classification: Primary 46L05; Secondary 46L55 INTRODUCTION Recent papers are dealing with different generalizations of the Cuntz-Krieger algebras O A (see [Pi], [P1], [D2], [AR], etc). The exact relationship between these approaches remains to be explored, but certainly there are overlaps. In [Pi], the author considers a Hilbert bimodule H over a C*-algebra, and creation operators on a corresponding Fock space. These operators generate the Toeplitz algebra T H and, taking a quotient by an ideal generated by finite rank operators, one obtains the algebra O H. If the Hilbert bimodule is projective and finitely generated over an abelian, finite dimensional C*-algebra, then one recovers the algebras O A. In [P1], the starting point is a Smale space (a compact metric space endowed with an expansive homeomorphism with canonical coordinates), on which one defines the stable 1 Research partially supported by NSF grant DMS

2 and unstable equivalence relations. The associated C*-algebras have natural shift automorphisms, and the crossed products are called Ruelle algebras. These are strongly Morita equivalent to particular Cuntz-Krieger algebras if the Smale space is a topological Markov shift. Our point of view is to start with a continuous oriented graph (E, V, o, t), to consider the space of one-sided infinite paths (obtained by concatenation of edges in E), and to associate a groupoid (à la Renault) using the unilateral shift on this path space. The C*-algebra of this groupoid plays the role of a continuous version of the Cuntz-Krieger algebras, since these could be obtained by the same construction from a finite graph defined by a 0-1 matrix. In many cases, this groupoid C*-algebra is simple, purely infinite, with computable K-theory. This approach offers more freedom for constructing easy, concrete examples, with prescribed K-theory. It should be mentioned that C*-algebras associated with discrete graphs were studied in [KPRR], [KPR], [KP]. See also the survey [K2]. The continuous graph approach is very similar to the point of view of polymorphisms or correspondences, introduced earlier in a measure theoretical context by Vershik and Arzumanian (see [AR] for a precise definition and references). The authors consider the C*-algebra of a groupoid on the vertex space rather than on the infinite path space, which in general is different from ours. We show that our groupoid algebras could be reobtained by using the Pimsner approach, with a right choice of the Hilbert bimodule. We feel that each point of view has certain advantages, ours beeing tied with the dynamics of a shift. Even in a case where this shift is not expansive, so the space of two-sided infinite paths has no obvious Smale space structure, we will prove that the corresponding algebra is simple and purely infinite. In the particular case when the vertex space is a disjoint union of tori, we call the corresponding space of paths a generalized solenoid, and we obtain results similar to those of Brenken (see [B]). It is interesting to notice how these fairly complicated dynamical systems appear in a natural way from embeddings of toral algebras. Acknowledgements. Thanks are due to several people who helped me while this paper 2

3 was growing, especially Paul Muhly, Alex Kumjian, Jean Renault, Jack Spielberg, Berndt Brenken, Ian Putnam. 1. CONTINUOUS GRAPHS AND DYNAMICAL SYSTEMS 1.1 DEFINITION. A continuous graph is V o E t V, where E (edges) and V (vertices) are topological spaces, and o (origin) and t (terminus) are onto local homeomorpisms. In the case when E and V are compact spaces, the set of edges could be identified with a closed subset of V {1, 2,..., n} V, since there are only a finite number of edges with prescribed origin and terminus. The maps o and t are the projection onto the first and the last coordinate. 1.2 EXAMPLES 1. Let V = {1, 2,..., N} and let E be defined by an N by N matrix A with positive integer entries. We have a ij edges with origin the vertex i and terminus the vertex j, for i, j = 1...N. 2. Let V = E = T the unit circle, with o(z) = z p, t(z) = z q for relatively prime integers p and q. 3. Let V = T and E = T 1 T 2 be the dijoint union of two circles, with o(z 1 ) = z 1, t(z 1 ) = z 2 1, o(z 2 ) = z 3 2, t(z 2 ) = z If f 1,..., f n : V V are onto local homeomorphisms, let n E = {(v, i, f i (v)) v V }. n=1 5. Let D be a discrete graph (the vertex space is discrete) and (D, A) a diagram of groups. By definition, for each vertex v of D we have a compact group A v, and for each edge e from o(e) to t(e) we have a covering morphism A e : A o(e) A t(e). In a natural way we can associate to (D, A) a continuous graph, by taking V = v A v, E = {(v, A e v) v A o(e) }, e 3

4 with obvious origin and terminus maps. 1.3 DEFINITION. Given a compact graph E V {1, 2,..., n} V, we define the dynamical systems (X +, σ + ), (X, σ) as follows: the space X + is the space of one-sided infinite paths, and σ + : X + X + is the unilateral shift, X + = {(x i, k i ) i=0 (x i, k i, x i+1 ) E}, (σ + (x i, k i )) p = (x p+1, k p+1 ). The space X is the space of two-sided infinite paths, X = {(x i, k i ) i Z (x i, k i, x i+1 ) E}, and σ is the bilateral shift. The dynamical system (X +, σ + ) unifies in a natural way the notion of a covering map T : V V, a (finitely-generated) semigroup or group of covering maps S : V V and the (unilateral) Markov shifts (see also [F]). It has the advantage that we get a single local homeomorphism. Example 1 will produce the unilateral and bilateral Markov shifts. In general, we may think that we have a continuous set of symbols. In fact, (X +, σ + ) is an extension of a Markov shift: let K + = {(k i ) 0 there are x i, x i+1 V such that (x i, k i, x i+1 ) E}, and let τ + : K + K + be the shift. Then the following diagram is commutative σ + X + X+ K + τ +, K+ where the vertical maps are the canonical projections. In the third example above, (X +, σ + ) is an extension of the Bernoulli shift ({1, 2} N, τ + ). Note that the fibers over sequences containing only a finite number of 2 s are circles, while the fibers over the other sequences are solenoids. 4

5 1.4 PROPOSITION. The dynamical system (X, σ) could be obtained from (X +, σ + ) by the usual inverse limit process by which one associates a homeomorphism to a continuous onto map. Proof. Indeed, let X = {(ξ n ) X + σ + (ξ n+1 ) = ξ n }. 1 We have π : X X+, π(ξ 1 ξ 2...) = ξ 1, and σ + : X X, σ+ (ξ 1 ξ 2...) = (σ + (ξ 1 )ξ 1 ξ 2...), such that σ + π = π σ +. Since X + = {(e n ) E t(e n ) = o(e n+1 )}, 1 X E could be identified with X, the space of two-sided infinite paths, and σ + with 1 1 the bilateral shift σ. 1.5 DEFINITION. For each continuous graph E we define its dual (or transpose) graph Ê by interchanging the origin and the terminus maps. This way we get dynamical systems ( ˆX +, ˆσ + ), ( ˆX, ˆσ), where ˆX +, ˆX are constructed from Ê, and ˆσ +, ˆσ are the unilateral and bilateral shift, respectively. It is easy to see that the systems (X, σ) and ( ˆX, ˆσ 1 ) are conjugated. But (X +, σ + ) and ( ˆX +, ˆσ + ) could be very different. 1.6 EXAMPLE. Take V = T the unit circle, and E the graph of the map z z 2, E = {(z, z 2 ) z T}. Then X + = T, σ + (z) = z 2, and ˆX + is a solenoid, ˆX + = {(z 1, z 2,..., ) z n T, zn+1 2 = z n, n 1}, ˆσ + (z 1, z 2,...) = (z 2, z 3,...). 1.7 REMARK. Note that if V has a group structure and E V V is a subgroup, then X + and X have also natural group structures, with componentwise multiplication. 5

6 2. THE C*-ALGEBRA OF A COMPACT GRAPH Assume that V is a compact metric space, and the two projections o, t : E V, o(x, k, y) = x and t(x, k, y) = y are onto local homeomorphisms. We can associate to the graph E a C*-algebra C (E), using the Renault groupoid of the dynamical system (X +, σ + ). The space X + is endowed with a metric defining the product topology. If δ denotes the metric on V, then one can take d((x i, k i ), (x i, k i)) = i 0 as a metric on X +. Similarly, we obtain a metric on X. δ(x i, x i)+ k i k i 2 i The unilateral shift σ + is a local homeomorphism, and we consider the following locally compact r-discrete groupoid: Γ = Γ(X +, σ + ) = {(x, n, y) X + Z X + k, l 0, n = k l, σ k +(x) = σ l +(y)}. The range map, the source map, and the operations are given as follows: r(x, n, y) = x, s(x, n, y) = y, (x, n, y)(y, p, z) = (x, n + p, z), (x, n, y) 1 = (y, n, x). The unit space of Γ is X +, if we identify (x, 0, x) with x. A basis of open sets for Γ is given by Z(U, V, k, l) = {(x, k l, (σ l + V ) 1 σ k +(x)), x U}, where U and V are open subsets of X +, and k, l are such that σ k + U and σ l + V are homeomorphisms with the same open range. 2.1 DEFINITION. Given a compact graph E with the maps o, t onto local homeomorphisms, we define its C*-algebra C (E) to be C (Γ), the C*-algebra of the Renault groupoid associated with the dynamical system (X +, σ + ). To understand the structure of C (E), consider the homomorphism c : Γ Z, c(x, n, y) = n, and let s denote by R the subgroupoid c 1 (0). If we denote by B the C -algebra of the 6

7 equivalence relation R, the local homeomorphism σ + induces a *-endomorphism α of B by the formula α(f)(x, y) = 1 p(σ + (x))p(σ + (y)) f(σ +(x), σ + (y)), f C c (R ), where for x X +, p(x) is the number of paths z such that σ + (z) = x. Moreover, assuming that σ + is not one-to-one, α is induced by a non unitary isometry v, in the sense that α(f) = vfv, where (p(σ + (x)) 1/2, if n = 1 and y = σ + (x) v(x, n, y) = 0, otherwise. Indeed, v v = 1, and p(σ vv + (x)) 1 if σ + (x) = σ + (y) (x, n, y) = 0, otherwise. Thus, α is a proper corner endomorphism of B, and C (E) is isomorphic to the crossed product B α N (see [R1]). In order to compute the K-theory of C (E), we can use the exact sequence K 0 (C (R )) id α 0 K 0 (C (R )) i 0 K 0 (C (E)) 1 0 K 1 (C (E)) i 1 K 1 (C (R )) id α 1 K 1 (C (R )) where i : C (R ) C (E) is the inclusion map. If on E we consider the equivalence relation R defined by t: two edges (x, k, y) and (x, k, y ) are equivalent iff y = y, then the C*-algebra C (R) is a continuous trace algebra with spectrum V, and there is a canonical embedding f(x), if x = x and k = k Φ : C(V ) C (R), Φ(f)((x, k, y), (x, k, y)) = 0, otherwise. Using the same method as in the Main Result of [D2], we get 7

8 2.2 THEOREM. If Φ 0 and Φ 1 are the maps induced on K-theory by the embedding Φ : C(V ) C (R), and if the K-theory groups K 0 (V ) and K 1 (V ) are free and finitely generated, then K 0 (C (E)) = ker(id Φ 1 ) coker(id Φ 0 ), K 1 (C (E)) = ker(id Φ 0 ) coker(id Φ 1 ). Using this theorem, we can get interesting examples of simple purely infinite C*-algebras with prescribed K-theory groups. In particular, in the next example, we construct C*- algebras A n with K 0 (A n ) = 0 and K 1 (A n ) = Z n. 2.3 EXAMPLE. Let V = V 1 V 2, where V i, i = 1, 2 are copies of the unit circle, and E = {(v, w) V 1 V 1 v = w 2 } {(v, w) V 1 V 2 v = w} {(v, w) V 2 V 1 v = w} {(v, k, w) V 2 {1, 2,..., n + 2} V 2 w = v n }. Then Φ : C(V 1 ) C(V 2 ) C(V 1 ) M 2 C(V 2 ) M n(n+2)+1, σ 1 f Φ(f g) = σ 2f 0 0 ˆσ n g σ 1 g ˆσ n g Here σ k f(z) = f(z k ), and ˆσ k is the k-times around embedding (the homomorphism compatible with the covering z z k ). There are n + 2 copies of ˆσ n g in the definition of Φ. Note that It follows that Φ 0 = n(n + 2), Φ 1 = n + 2 ker(id Φ 0 ) = 0, Z 2 /(id Φ 0 )Z 2 = 0, ker(id Φ 1 ) = 0, Z 2 /(id Φ 1 )Z 2 = Z n,. 8

9 therefore the corresponding C*-algebra C (E) has K 0 = 0, K 1 = Z n. One can check that every orbit with respect to the equivalence relation R is dense, therefore C (R ) and C (E) are simple. The latter algebra is purely infinite because it appears as a crossed product of an inductive limit of circle algebras by an endomorphism that does not preserve any trace (see Theorem 2.1 in [R2]). More general, we can state 2.4 PROPOSITION. If σ + is minimal (the orbits with respect to the equivalence relation R are dense), and there is no shift invariant trace on C (R ), then C (E) is simple and purely infinite. A solenoid is a compact connected abelian group of finite dimension. For example, if T is the unit circle, T(m) = {z T Z z m k = z k+1, k Z} is such a group, for any integer m > 1. The bilateral shift σ on T(m), σ(z) p = z p+1 is a homeomorphism, and in many respects it is an analogue of the Bernoulli shift. In [B], Brenken considered the dynamical system (G 0, σ) for G 0 the connected component of the identity of the group G = {z (T d ) Z F z k = Mz k+1, k Z}, where M, F are surjective endomorphisms of the d-torus, given by matrices M, F M d (Z) with nonzero determinant. (Note that the case d = 1, M = 1, F = m corresponds to the above example T(m)). The space G 0 has a natural local product structure, being a principal bundle over T d with fiber the Cantor set. Moreover, it has a Smale space structure (see [P1]), and the author identifies the C -algebras associated with the stable and unstable equivalence relations. 2.5 DEFINITION. By a generalized solenoid we mean the space X of two-sided infinite paths with edges in the graph E described bellow. Let V = T d 1... T d N be the disjoint union of N copies of the d-dimensional torus T d, and let L = (l(i, j)) i,j be an N N matrix 9

10 with positive integer entries (the incidence matrix of the graph). We require that in the matrix L each row and each column has at least a nonzero entry. For each pair (i, j) with l(i, j) 1, consider a family of closed, connected subgroups G ij 1, G ij 2,..., G ij l(i,j) of Td i T d j, not necessarily distinct, such that all the projections on T d i and T d j are surjective. For the pairs (i, j) with l(i, j) = 0, this family is empty by definition. We take E to be the disjoint union of all the groups G ij k, 1 i, j N, 1 k l(i, j), with obvious origin and terminus maps. It is known (see [KS]) that there are families of d d nonsingular matrices with integer entries, A ij = {A ij 1,..., A ij l(i,j) }, B ij = {B ij 1,..., B ij l(i,j) }. such that G ij l = {(z, w) T d i T d j A ij l z = Bij l w}. The matrices A ij k, Bij k are not necessarily distinct. Note that a generalized solenoid X has no longer a group structure, and the dynamical system (X, σ) is a continuous analogue of the Markov shift. 2.6 EXAMPLE. Let d = 2, N = 2, A 11 1 = A 22 1 = The space of edges is, B 11 1 = A 12 2 =, B 22 1 = , A 12 1 =, B 12 2 =, A 22 2 = , B 12 1 =,, B 22 2 = {((z, w), 1, (t, u)) T 2 1 {1} T 2 1 z 3 w = t 1, zw = t 2 u 3 } {((z, w), 1, (t, u)) T 2 1 {1} T 2 2 z = t 3, w 2 = tu} 10,.

11 {((z, w), 2, (t, u)) T 2 1 {2} T 2 2 z 2 w = t 1 u, w = t} {((z, w), 1, (t, u)) T 2 2 {1} T 2 2 z = t 2, w = u} {((z, w), 2, (t, u)) T 2 2 {2} T 2 2 z = t, w = u}. The corresponding embedding C(V ) C (R) of toral algebras is where and Φ 11 (f) = ˆσ B 11 1 Φ : C(T 2 ) C(T 2 ) C(T 2 ) M 3 C(T 2 ) M 7, Φ(f g) = Φ 11 (f) σ A 11 1 (f), Φ 12(f) = Φ 22 (g) = ˆσ B1 22 Φ 12(f) 0 0 Φ 22 (g) ˆσ B1 12 σ A 22 1 (g) 0 0 ˆσ B 22 2, σ A 12 1 (f) 0 0 ˆσ B 12 2 σ A 22 2 (g). σ A 12 2 (f) Here σ A : C(T 2 ) C(T 2 ) denotes the -homomorphism induced by the covering A, defined by (σ A f)(z) = f(az), and ˆσ A : C(T 2 ) C(T 2 ) M deta is the homomorphism compatible with A, in the sense that ˆσ A σ A (f) = f I deta., 3. RUELLE ALGEBRAS 3.1 DEFINITION. Recall that σ + : X + X + is (positive) expansive if there is a constant c > 0 such that x y implies d(σ+(x), n σ+(y)) n c for some integer n 0. An element x X + is eventually periodic if there are two integers p q with σ+(x) p = σ+(x). q 3.2 THEOREM (Delaroche) If σ + is expansive and the eventually periodic points form a dense set with empty interior, then C (Γ), and therefore C (E), is nuclear, purely infinite, and belongs to the bootstrap class N. Note that in the above hypotheses, the groupoid Γ = Γ(X +, σ + ) is essentially free, i.e. the set of points in the unit space with trivial isotropy is dense. 11

12 3.3 REMARK. When σ + is expansive, then σ is also expansive and (X, σ) has a Smale space structure. For the definition of a Smale space, see again [P1]. Renault and Arzumanian (see Theorem 4.5 in [AR] are proving the following result: Let R s the stable equivalence relation, R s = {(x, y) X X d(σ n (x), σ n (y)) 0 as n + }. Then C (R s ) is strongly Morita equivalent to C (R ), and its Ruelle algebra C (R s ) Z is strongly Morita equivalent to C (E). We show in the following example that even for non-expansive σ +, the C*-algebra C (E) is simple and purely infinite. compact graphs strictly contains the Ruelle algebras. Then This example arose in a discussion with Jack Spielberg. 3.4 EXAMPLE. Let V = T, the unit circle, and In particular, we believe that the class of C*-algebras of E = {(z, 1, z 2 ) z V } {(z 3, 2, z) z V }. X = {(z n, k n ) (V {1, 2}) Z k n = 1 z n+1 = z 2 n, k n = 2 z 3 n+1 = z n }. We will show that σ : X X, σ(z n, k n ) p = (z p+1, k p+1 ) is not expansive, therefore the space (X, σ) has not a Smale space structure. It suffices to show that for any ε > 0, we can find two distinct sequences (z n, k n ) and (w n, k n ) such that δ(z n, w n ) ε for all n Z. Fix z 0, w 0 V. The idea is that, taking in a certain order squares, cubes, square roots and cubic roots, the corresponding vertices remain close together. We can choose two sequences of integers (a n ) n 1, (b n ) n 1 such that 2 a a n lim n 3 b b n = 1. Consider the symmetric sequence (k n ) n Z described as }{{} }{{} }{{} }{{} }{{} }{{} }{{} }{{}..., b 2 a 2 b 1 a 1 a 1 b 1 a 2 b 2 12

13 where the bar indicates k 0. Given ε > 0, we can choose z 0 and w 0 sufficiently close together (but distinct), and z n and w n in a consistent way (when we take square or cubic roots) such that δ(z n, w n ) ε. It follows that d(σ p (z n, k n ), σ p (w n, k n )) ε p Z, and the shift is not expansive. Nevertheless, the orbits with respect to R are dense in X +, and there is no shift invariant trace, therefore the C*-algebra C (E) is simple and purely infinite. It is interesting to notice that C (E) and C (Ê) are both simple, purely infinite, with K-theory K 0 (C (E)) = K 1 (C (Ê)) = Z 2, K 1 (C (E)) = K 0 (C (Ê)) = Z 3. In [P1] it is proved that the Ruelle algebra associated to the graph of the map z z p on the unit circle is isomorphic to the one obtained from the dual graph. Whether this is true for more general graphs is an open question. 4. THE PIMSNER ALGEBRAS O H In this paragraph, we recall the Pimsner construction from [Pi], and we show how the C*-algebra of a compact graph could be thought as O H, for a particular Hilbert bimodule H. To a pair (H, A), where H is a (right) Hilbert module over a C -algebra A, and A acts to the left on H via a -homomorphism ϕ : A L(H), Pimsner constructs a C*-algebra O H, which generalizes both the crossed products by Z and the Cuntz-Krieger algebras. The algebra O H is a quotient of the generalized Toeplitz algebra T H, generated by the creation operators T ξ, ξ H on the Fock space H + = H n. Here H 0 = A, and for n 1, H n denotes the n-th tensor power of H, balanced via the map ϕ. By definition, T ξ a = ξa, for a A, and T ξ (ξ 1... ξ n ) = ξ ξ 1... ξ n, for ξ 1... ξ n H n. To give another description of the algebra O H, Pimsner considers a new pair (H, F H ), where F H is the C*-algebra generated by all the compact operators K(H n ), n 0 in n=0 13

14 lim L(H n ), and H = H F H. The advantage is that H becomes an F H F H bimodule, such that the adjoint H is also an F H F H bimodule. The C*-algebra O H is represented on the two-sided Fock space H = H n, n Z where for n < 0, H n means (H ) n. In fact, it is isomorphic to the C*-algebra generated by the multiplication operators M ξ L(H ), where ξ H, and M ξ ω = ξ ω. Given a compact graph E such that the origin and terminus maps E V are onto local homeomorphisms, let A = C(V ) as a C*-algebra, and let H = C(E) as a vector space, with the structure of Hilbert A-module given by (ξf)(e) = ξ(e)f(t(e)), ξ H, f A, e E, < ξ, η > (v) := ξ(e)η(e), v V, ξ, η H. t(e)=v In other words, the inner product is given by < ξ, η >= P ( ξη), where P : C(E) C(V ) is the conditional expectation (P ξ)(v) = ξ(e). t(e)=v The left module structure is given by ϕ : A L(H), (ϕ(f)ξ)(e) = f(o(e))ξ(e) f A, ξ H. Note that indeed ϕ(f) is in L(H), having the adjoint ϕ( f), f A. To prove that O H with this choice of A, H and ϕ is isomorphic to C (E), let s identify the C*-algebra F H in this case. Note that H ϕ H is a quotient of C(E) C(E), where we identify ξf η with ξ ϕ(f)η for any ξ, η H and any f A. Therefore H ϕ H could be identified as a vector space with the continuous functions on the set {(e 1, e 2 ) E E t(e 1 ) = o(e 2 )}. This set will be denoted by X 2, and is precisely the set of paths of length 2. In a similar way, H n is identified (as a vector space) with C(X n ), where X n is the set of paths of length 14

15 n. The Hilbert A-module structure on H n for n 2 is given by (ξf)(x) = ξ(x)f(t n (x)), x X n where t n : X n V, t n (e 1 e 2...e n ) = t(e n ), and by < ξ, η > n = P n ( ξη). Here P n is the conditional expectation P n : C(X n ) C(V ), P n (ξ)(v) = ξ(x). t n(x)=v 4.1 PROPOSITION. The C*-algebra K(H) is isomorphic with C (R), where R = {(e 1, e 2 ) E E t(e 1 ) = t(e 2 )} is the equivalence relation associated with the map t. The map ϕ : A L(H) could be identified with the embedding Φ : C(V ) C (R), defined before Theorem 2.2. Moreover, K(H n ) C (R n ), where R n = {(x, y) X n X n t n (x) = t n (y)} is the equivalence relation associated with t n. Proof. Taking into account the fact that o and t are local homeomorphisms, we have L(H) = K(H), since H is algebraically finitely generated. Now K(H) = H H, the tensor product balanced over A, where H is the adjoint of H. Since ξf η = ξ fη, it follows that, as a set, K(H) = C(R). The multiplication of compact operators turns out to be the convolution product on C(R), therefore, as C*- algebras, K(H) = C (R). 4.2 COROLLARY. We have F H = lim C (R n ). Therefore, F H is isomorphic to the algebra C (R ). 15

16 Proof. As above, using the fact that K(H n ) = (H n ) (H n ), we get K(H n ) = C (R n ). Note that for n 1, the inclusion φ n : C (R n ) C (R n+1 ), f(x 1...x n, y 1...y n ) if x n+1 = y n+1 (φ n )(f)(x 1...x n+1, y 1...y n+1 ) = 0, otherwise is just the map K(H n ) K(H n+1 ), T T I. Here R 1 =R. In order to establish an isomorphism between C (Γ) and O H, we show that they appear as the C -algebras associated to isomorphic Fell bundles over the group Z. This point of view was suggested by Abadie, Eilers and Exel in [AEE]. The definition of a Fell bundle and of the associated C -algebra is taken from [K1]. To the pair (H, F H ), we can associate the Fell bundle B, where B n := H n, n Z. The multiplication is given by the tensor product, identifying H H with F H and H H with the ideal F 1 H of F H, generated by K(H n ) with n 1. But F 1 H is equal to F H in our case. The involution is obvious. Then Since H is generated by F H L 2 (B) = H = H n. n Z operators M ξ is isomorphic to C (B). Hence, O H C (B). For the groupoid Γ and l Z, take and H, it follows that the C -algebra generated by the Γ l := {(x, k, y) Γ k = l} = {(x, y) X X x n = y n+l for large n}, and D l = C c (Γ l ) (closure in C (Γ)). This way, we obtain a Z-grading on C (Γ), and it is easy to see that this C*-algebra could be recovered as C (D). But D 0 = C (R ) F H = B 0, and D 1 = C c (Γ 1 ) H A F H = H = B 1. We get 16

17 4.3 THEOREM. With the above choice of A, H and ϕ, the C*-algebras C (E) and O H are isomorphic. 5. POLYMORPHISMS A continuous graph V r Σ s V with V, Σ locally compact is called a polymorphism by Arzumanian and Renault. They associate a groupoid and a C*-algebra to a commuting polymorphism in the following way (see [AR]). Let G be the pseudogroup on V generated by the partial homeomorphisms r U (s U ) 1 affiliated to Σ. Here U Σ is an open set such that both r U and s U are one-to-one. Let G be the groupoid of germs constructed from G, and let C (G). In the case that Σ = V, and the polymorphism is essentially free, they show that the above groupoid G is isomorphic to Notice that G and G G = {(r n s m (z), m n, s n r m (z)) : m, n N, z V }. are groupoids on V, rather than on a space of infinite paths. They show that the properties of C (G) could be conveniently expressed in terms of the pseudogroup G. For example, it is purely infinite if G is locally contracting. But the polymorphism T z2 T z T and its transpose will generate the same groupoid and the same C*-algebra. Our approach and the Pimsner approach do not have this symmetry. Hence, the C*-algebra of a polymorphism is in general different from the C*-algebra of a compact graph. REFERENCES [AR] V. Arzumanian, J. Renault, Examples of pseudogroups and their C*-algebras, Operator Algebras and Quantum Field Theory, Editors: S. Doplicher, R. Longo, J.E. Roberts and L. Zsido, International Press 1997, 17

18 [B] B. Brenken, The local product structure of expansive automorphisms of solenoids and their associated C*-algebras, Canad. J. Math. 48(1996), no. 4, , [D1] V. Deaconu, Groupoids associated with endomorphisms, Trans. AMS 347(1995), , [D2] V. Deaconu, Generalized Cuntz-Krieger algebras, Proc. AMS 124(1996), , [De] C. Anantharaman-Delaroche, Purely infinite C*-algebras arising from dynamical systems, Preprint, June 1996, [F] S. Friedland, Entropy of graphs, semigroups and groups, in Ergodic theory of Z d - actions, , Cambridge 1996, [K1] A. Kumjian, Fell bundles over groupoids, to appear in Proc. AMS, [K2] A. Kumjian, Notes on C*-algebras of graph, to appear, [KP] A. Kumjian, D. Pask, C*-algebras of directed graphs and group actions, to appear in Ergodic Theory Dyn. Sys., [KPR] A. Kumjian, D. Pask, I. Raeburn, Cuntz-Krieger algebras of directed graphs, to appear in Pacific J. Math, [KPRR] A. Kumjian, D. Pask, I. Raeburn, J. Renault, Graphs, groupoids and Cuntz- Krieger algebras, J. Funct. Anal. 144(1997), , [KPS] J. Kaminker, I. Putnam and J. Spielberg, Operator algebras and hyperbolic dynamics, Operator Algebras and Quantum Field Theory, Editors: S. Doplicher, R. Longo, J.E. Roberts and L. Zsido, International Press 1997, [KS] B. Kitchens, K. Schmidt, Automorphisms of compact groups, Ergod. Th & Dynam. Sys 9(1989), , [KW] T. Kajiwara, Y. Watatani, Hilbert C*-bimodules and Continuous Cuntz-Krieger Algebras considered by Deaconu, Preprint 1998, [Pi] M.V. Pimsner, A class of C*-algebras generalizing both Cuntz-Krieger algebras and crossed products by Z, in Free Probability Theory, edited by D.V. Voiculescu, Fields Institute Communications vol 12, 1996, , [P1] I. Putnam, C*-algebras from Smale spaces, Can. J. Math. 48(1996), , 18

19 [P2] I. Putnam, Hyperbolic Systems and Generalized Cuntz-Krieger Algebras, Lecture Notes, August 1996, [R1] M. Rørdam, Classification of Certain Infinite Simple C*-algebras, to appear in J. Funct. Anal., [R2] M. Rørdam, Classification of Certain Infinite Simple C*-algebras III, Operator Algebras and Their Applications, Fields Institute Communications No. 13, 1997, DEPARTMENT OF MATHEMATICS/084, UNIVERSITY OF NEVADA, RENO NV 89557, USA address: 19