, ,

We proceed in analogy to results concerning the disk algebra, the fundamental commutative graph operator algebra.

For commutative function algebras, the notion of a point derivation can be connected intimately to analytic structure. There are examples where this connection breaks down, but for the disk algebra it is well known, see [2,Theorem1] and [3,Section1.6] , that point derivations occur only at interior points of the maximal ideal space. In fact, a point of the maximal ideal space of the disk algebra is a peak point if and only if there are no nonzero point derivations at the point, see [3,Corollary1.6.7] .

In this paper we will extend these results to the graph algebras coming from

n

-cycles. Besides the interest in peak points, the analysis below is also useful in studying the homology groups associated to certain graph operator algebras. We will see that the first homology group of the graph operator algebras we discuss is trivial, in analogy with the same result for the disk algebra.

We begin by defining the two classes of algebras we will study in this paper. For the first class of algebras we will look at

M_{n} \otimes A (D)

. This algebra can be viewed as

n \times n

matrices with entries from

A (D)

. This is an operator algebra when it inherits the matricial norms from

A (D)

The second class of algebras arise in the context of directed graph operator algebras, see [10,Example6.5] . These algebras have also been studied as semicrossed product operator algebras, see in particular [1] and [6] . From directed graphs, they arise as the left regular representation of the directed graphs with

n

vertices and

n

edges connecting each successive vertex in turn, to form a single loop, or

n

-cycle.

We will use the notation

A (C_{n})

for these algebras, where

n

is the length of the cycle in the algebra.

We can view

A (C_{n})

as a matrix function algebra of the form

[\begin{matrix} f_{1, 1} (z^{n}) & z f_{1, 2} (z^{n}) & z^{2} f_{1, 3} (z^{n}) & \cdot \cdot \cdot & z^{n - 1} f_{1, n} (z^{n}) \\ z^{n - 1} f_{2, 1} (z^{n}) & f_{2, 2} (z^{n}) & z f_{2, 3} (z^{n}) & \cdot \cdot \cdot & z^{n - 2} f_{2, n} (z^{n}) \\ z^{n - 2} f_{3, 1} (z^{n}) & z^{n - 1} f_{3, 2} (z^{n}) & f_{3, 3} (z^{n}) & \cdot \cdot \cdot & z^{n - 3} f_{3, n} (z^{n}) \\ . . . & . . . & . . . & . . . & . . . \\ z f_{n, 1} (z^{n}) & z^{2} f_{n, 2} (z^{n}) & z^{3} f_{n, 3} (z^{n}) & \cdot \cdot \cdot & f_{n, n} (z^{n}) \end{matrix}]

where

f_{i, j} \in A (D)

for all

1 \leq i, j \leq n

These algebras will inherit a matricial norm from the matricial norm on

A (D)

A (C_{n})

can be viewed as sitting inside

M_{n} \otimes A (D)

. There is some evidence that the algebra

A (C_{n})

is a natural analogue of

A (D)

to the noncommutative setting.

For example, the automorphism group of

A (C_{n})

has connections with that of

A (D)

, [1] . Further, as a graph algebra there are many analogues of classical properties of

A (D)

to the quiver algebras, see for example [5] and [9] . In this paper we discuss another such generalization to a particular class of graph algebras.

In the first section we discuss the notion of noncommutative point derivations and discuss some basic properties. We then establish semisimplicity of one class of algebra in the next section. The third section is devoted to the noncommutative point derivations of the first class of operator algebras. The next section then discusses the point derivations of the graph operator algebras we are interested in. We then formulate results on the homology groups for the graph operator algebras. In the last section we discuss amenability for non-selfadjoint graph operator algebras.

We now fix some notation:

(x_{i j})

will denote a matrix, where the

i

j

entry is

x_{i j}

, and

[x_{i j}]

will denote the matrix with all zero entries except the

i

j

entry which will be

x_{i j}

1 Noncommutative point derivations

In analogy with the function algebra theory we define a noncommutative point derivation. Here, the noncommutative “point”, is given by a representation of the algebra into

n \times n

matrices for some

n

. In this section we will assume that

A

is always an operator algebra.

Definition 1. We let

A

be an operator algebra, and we let

π : A \to B (ℋ)

be a unital representation. We say that a bounded, linear map

D : A \to B (ℋ)

is a point derivation at

π

D (a b) = D (a) π (b) + π (a) D (b)

for all

a, b \in A

Notice that if

ℋ = C

then we are in the usual situation of a point derivation.

Certainly continuity is not necessary, as the study of discontinuous derivations has an interesting theory, but we are not concerned with that theory here. Now, if

D (a) = 0

for all

a

we say that

D

is the trivial derivation. In the context of this paper we will mostly be dealing with finite dimensional representations (i.e. where

B (ℋ) = M_{n}

for some

n \geq 1

We now record some useful facts which will allow us to build point derivations.

We refer the reader to [3] for a survey of classical results on point derivations for function algebras. In contrast to the classical case, we will need to look at a notion of inner point derivations.

Definition 2. Let

π : A \to B (ℋ)

be a unital representation, and

X \in B (ℋ)

Let

δ_{X} : A \to B (ℋ)

be given by

π (a) X - X π (a)

for all

a \in A

Lemma 1. Let

π : A \to B (ℋ)

be a unital representation, and

X \in B (ℋ)

, then

δ_{X}

is a point derivation at

π

Proof. Clearly

δ_{X}

is linear and bounded. Now let

a, b \in A

and notice that

\begin{matrix} δ_{X} (a b) & = π (a b) X - X π (a b) \end{matrix}

\begin{matrix} = π (a) π (b) X - X π (a) π (b) \end{matrix}

\begin{matrix} = π (a) π (b) X - π (a) X π (b) + π (a) X π (b) - X π (a) π (b) \end{matrix}

\begin{matrix} = π (a) δ_{X} (b) + δ_{X} (a) π (b) \end{matrix}

\begin{matrix}  \end{matrix}

and hence

δ_{X}

is a derivation at

π

■

In the case that

ℋ = C

the range of the representation

π

C

and hence any inner derivation yields the trivial derivation. In the noncommutative setting, however, inner derivations do arise and we must take them into account. We first show that if the range of

π

is not commutative there always exist nontrivial inner point derivations.

Proposition 1. Let

π : A \to B (ℋ)

be such that

r a n π

is not isomorphic to

C

, then there exists a nontrivial inner derivation at

π

Proof. Let $(B (ℋ))^{'}$ denote the commutant of $B (ℋ)$ . It is well known that $(B (ℋ))^{'} = {λ 1_{ℋ} : λ \in C}$ and since $r a n π$ is not isomomorphic to $C$ there is $a \in A$ such that $π (a) \notin (B (ℋ))^{'}$ . Since $π (a) \notin (B (ℋ))^{'}$ there exists $X \in B (ℋ)$ such that $π (a) X \neq X π (a)$ . Notice that $δ_{X}$ is then a nontrivial inner derivation. $■$

We know look at a method of describing the inner derivations at

π

for a given onto representation

π : A \to M_{n}

, where

n \geq 1

Proposition 2. Let

π : A \to B (ℋ)

be a unital representation such

r a n π

is amenable and

ker π

is a complemented subspace of

A

. A continuous derivation

D

π

is inner if and only if

D |_{ker π} = 0

Proof. Clearly if $D$ is a derivation at $π$ such that $D (a) = π (a) X - X π (a)$ then $D |_{| k e r π} = 0$ .
For the reverse direction, let $D$ be a derivation with $ker π \subseteq ker D$ . Now define $\hat{D} : r a n π \to r a n π$ by $\hat{D} (x) = D (a)$ where $π (a) = x$ . First notice that if $π (a) = π (b) = x$ , then $π (a - b) = 0$ and hence $D (a - b) = 0$ . It follows by linearity of $D$ that $\hat{D} (x)$ is well defined. Let ${x_{λ}}$ is a convergent net in $r a n π$ with limit point $x$ . Now since $ker π$ is a complemented subspace of $A$ , then as subspaces the complement of $ker π$ and $r a n π$ are isomorphic. In other words, for each $λ$ there is a unique $a_{λ}$ in the subspace complement of $ker p i$ such that $π (a_{λ}) = x_{λ}$ and since ${x_{λ}}$ is convergent so is ${a_{λ}}$ . Further $π (lim (a_{λ}) = x$ . Now by continuity of $D$ it follows that $lim \hat{D} (x_{λ}) = lim D (a_{λ}) = D (lim a_{λ}) = \hat{D} (x)$ and hence $\hat{D}$ is a continuous $B (ℋ)$ valued derivation on $r a n π$ .
But by hypothesis $r a n π$ is amenable and hence every continuous $B (ℋ)$ valued derivation on $r a n π$ is inner. The result now follows. $■$

Notice that if

r a n π

is finite dimensional and

r a n π

is amenable then the previous result will apply. In particular, if

r a n π

M_{n}

for some finite

n

then any derivation

D

, at

π

sending

ker π

to zero is inner. The same result holds also in the context of

r a n π

a finite direct sum of copies of

C

. We know look at information concerning

ker π

that will allow us to identify the previous situation in practice (i.e. when is

D |_{ker π}

identically zero.) For an ideal

M

, we let

M^{2} = s p a n {f g : f, g \in M}

Clearly

M^{2}

is also an ideal, although not necessarily closed.

Lemma 2. Let

π : A \to B (ℋ)

be a representation and let

M

be the kernel of

π

. Then if

M = \bar{M_{2}}

then for any derivation

D

π

D |_{ker π}

is identically zero

Proof. Let $f, g \in M$ and notice that $D (f g) = D (f) π (g) + π (f) D (g) = D (f) \cdot 0 + 0 \cdot π (g) = 0$ . It follows by linearity that $D |_{M^{2}}$ is identically zero. Now since $\bar{M^{2}} = M$ , then for every $m \in M$ there exists a net ${m_{λ}} \in M^{2}$ such that $m_{λ} \to m$ . Now $D (m_{λ} = 0$ for all $λ$ and hence by continuity of $D$ , $D (λ) = 0$ . As $m$ was an arbitrary element of $M$ the result now follows. $■$

We say that an ideal

M

has a bounded approximate left identity if there exists a bounded net

{f_{λ}}_{λ \in I}

M

such that

lim f_{λ} g = g

for all

g \in M

. Alternatively we say that

M

contains a bounded approximate left identity if there exists

K > 0

such that for all

ɛ > 0

and

f_{1}, f_{2}, \dots, f_{n} \in M

there exists

f

with

∥ f ∥ < K

and

∥ f f_{i} - f_{i} ∥ < ɛ

for

1 \leq i \leq n

Proposition 3. Let

M

be the kernel of

π

, then any nonzero point derivation at

π

is inner if

M

has a bounded left approximate identity.

Proof. Clearly if $M$ has a bounded approximate identity ${e_{λ}}$ then for each $f \in M$ we know that $e_{λ} f \in (ker π)^{2}$ for each $λ$ . It follows that $f = lim e_{λ} f \in \bar{M^{2}}$ . As $f$ was arbitrary the previous result applies and we are done. $■$

It is clear that the same result holds in the presence of a bounded approximate right identity.

2 Representations for $M_{n} \otimes A (D)$

Proposition 4. There is a one-to-one correspondence between maximal ideals of

M_{n} \otimes A (D)

and kernels of representations of the form

(f_{i, j} (z)) \mapsto (f_{i, j} (λ))

where

λ \in \bar{D}

Proof. Certainly each representation of the form described will give rise to a maximal ideal since the range of the representation is the simple algebra $M_{n}$ . The reverse direction is more complicated.
Let $M$ be a maximal ideal in $M_{n} \otimes A (D)$ , and let $x$ be a nonzero element of $M$ . Assume, without loss of generality that $x_{i, j} = f (z) \neq 0$ . Now, if $e_{l, m}$ denotes the elementary matrix with a $1$ in the $l$ - $m$ position and zeroes elsewhere, then $e_{l, m} \in M_{n} \otimes A (D)$ for all pairs $m$ and $n$ . Thus, $e_{i, i} x e_{j, j} \in M$ and $e_{l, i} e_{i, i} x e_{j, j} e_{j, l} \in M$ .
This matrix is the matrix with $f (z)$ in the $l$ - $l$ position and zeroes elsewhere.
Now if $g (z) \in A (D)$ then if $G$ is the matrix with $g (z)$ along the diagonal then, $G e_{l, i} e_{i, i} x e_{j, j} e_{j, l} \in M$ and hence the matrix with $f g$ in the $l$ - $l$ position is in $M$ . It follows that the subalgebra $M_{l, l}$ of the form ${e_{l, i} e_{i, i} M e_{j, j} e_{j, l}}$ forms an ideal in the commutative subalgebra $A (D)_{l, l}$ given by ${e_{l, i} e_{i, i} A (D) e_{j, j} e_{j, l}}$ .
Further, $M$ is the set of all matrices of the form $e_{i, l} M_{l, l} e_{l, j}$ for all $i$ and $j$ . In other words, componentwise the maximal ideal is given by the same sets of analytic functions. It follows that $M_{l, l}$ is maximal in $A (D)_{l, l}$ and hence $M_{l, l}$ is the kernel of the evaluation map of $A (D)$ . The result now follows since each componentwise subspace is the same. $■$

We will denote the representations of the preceding theorem by

φ_{λ}

where

λ

is a complex scalar of modulus less than or equal to one.

Corollary 1. The algebra

M_{n} \otimes A (D)

is semisimple.

Proof. Let $f$ be an $A (D)$ valued matrix such that $φ_{λ} (f) = 0$ for all $φ$ . In particular, $f_{i, j} (λ) = 0$ for all $λ$ and hence $f_{i, j} = 0$ for all $i, j$ . It follows that $f$ is the zero matrix and the algebra is semisimple. $■$

3 Point derivations and homology of $M_{N} \otimes A (D)$

We are now in a position to describe the point derivations for

M_{n} \otimes A (D)

Theorem 1. Let

n \geq 0

then there exists a nontrivial point derivation at

φ_{λ}

if and only if

| λ | < 1

Proof. If

| λ | < 1

define a continuous linear map by

D_{λ} ([\begin{matrix} f_{1, 1} (z) & f_{1, 2} (z) & \cdot \cdot \cdot & f_{1, n} (z) \\ f_{2, 1} (z) & f_{2, 2} (z) & \cdot \cdot \cdot & f_{2, n} (z) \\ . . . & . . . & \cdot \cdot \cdot & . . . \\ f_{n, 1} (z) & f_{n, 2} (z) & \cdot \cdot \cdot & f_{n, n} (z) \end{matrix}]) = [\begin{matrix} f_{1, 1}^{'} (λ) & f_{1, 2}^{'} (λ) & \cdot \cdot \cdot & f_{1, n}^{'} (λ) \\ f_{2, 1}^{'} (λ) & f_{2, 2}^{'} (λ) & \cdot \cdot \cdot & f_{2, n}^{'} (λ) \\ . . . & . . . & \cdot \cdot \cdot & . . . \\ f_{n, 1}^{'} (λ) & f_{n, 2}^{'} (λ) & \cdot \cdot \cdot & f_{n, n}^{'} (λ) \end{matrix}] .

This map is bounded since it is the matrix version of the completely bounded map sending

f (z) \mapsto f^{'} (λ)

acting on

A (D)

We know verify that

D_{λ}

is indeed a derivation. Choose two elements

f, g \in M_{n} \otimes A (D)

. Now notice that

\begin{matrix} D_{λ} (f g) & = D_{λ} (\sum_{j = 1}^{n} f_{i j} g_{j k}) \end{matrix}

\begin{matrix} = (\sum_{j = 1}^{n} (f_{i, j}^{'} (λ) g_{j k} (λ) + f_{i j} (λ) g_{j k}^{'} (λ))) \end{matrix}

\begin{matrix} = (\sum_{j = 1}^{n} (f_{i, j}^{'} (λ) g_{j k} (λ)) + (\sum_{j = 1}^{n} f_{i j} (λ) g_{j k}^{'} (λ))) \end{matrix}

\begin{matrix} = (f_{i j}^{'} (λ)) (g_{i j} (λ)) + (f_{i j} (λ)) (g_{i j}^{'} (λ)) \end{matrix}

\begin{matrix} = D_{λ} (f) φ_{λ} (g) + φ_{λ} (f) D_{λ} (g) . \end{matrix}

\begin{matrix}  \end{matrix}

Notice that this derivation is not inner since

D_{λ} |_{ker (φ_{λ})} \neq 0

. In particular, let

f

be an analytic function with nonzero derivative such that

f (λ) = 0

. Then the matrix with

f

in each entry is in the kernel of

φ_{λ}

but is not in the kernel of the derivation

D_{λ}

We now show that if

| λ | = 1

then there are only inner derivations. We do this by showing that

ker (φ_{λ}) = \bar{(ker (φ_{λ}))^{2}}

. By [3,Corollary1.6.4] there exists a bounded net

f_{ι} \in A (D)

such that

f_{ι} (λ) = 0

and if

f

is analytic with

f (λ) = 0

then

f_{ι} f \mapsto f

Let

F_{ι}

be the matrix with diagonal entries given by

f_{ι}

and off-diagonal entries all zero. Then

{F_{ι}}

is a bounded net in

ker φ_{λ}

. Further

F_{ι} G

converges to

G

for all

G \in ker φ_{λ}

and hence

{F_{ι}}

is a bounded approximate identity in the ideal

ker (φ_{λ})

It follows by Proposition 3 that for

| λ | = 1

any derivation at

φ_{λ}

is inner.

■

Let

M_{n}

be a module with module action given by

a X = φ_{λ} (a) X

for

a \in M_{n} \otimes A (D)

and

X \in M_{n}

. Denoting this module by

(M_{n})_{λ}

, then in terms of Hochschild cohomology, the previous result tells us that

H^{1} (M_{n} \otimes A (D), (M_{n})_{λ}) \neq 0

if and only if

| λ | < 1

. It is immediate that

M_{n} \otimes A (D)

is not amenable, for all

n \geq 1

4 Point derivations on $A (C_{n})$

A characterization of maximal ideals of

A (C_{n})

is found in [1,Corollary6] . We present the results here without proof.

Proposition 5. The maximal ideals of

A (C_{n})

are of one of the following forms:

(1) the kernel of a representation of the form $(f_{i, j} (z)) \mapsto (f_{i, j} (λ))$ where $λ \in \bar{D} \ {0}$ .
(2) the kernel of a representation that sends $f_{i j} (z)$ to $0 if i \neq m or j \neq m$ where $m$ is a fixed integer and $f_{m m} (z) \mapsto f_{m m} (0)$ .

So we have a maximal ideal of codimension

n^{2}

for each

λ \neq 0

and we have

n

maximal ideals of codimension 1, given by evaluating a fixed diagonal entry at 0.

For this algebra, we will denote the maximal ideals corresponding to evaluation of a diagonal entry at zero, by

φ_{0, m}

where the

m

corresponds with the diagonal entry which is evaluated at zero.

Corollary 2. The algebra

A (C_{n})

is semisimple.

Proof. Assume that $φ_{λ} (a) = 0$ for all $0 < | λ | < 1$ . Then in particular. $f_{i, j} (λ^{n}) = 0$ for all $0 < | λ | < 1$ . But since $f_{i, j}$ is analytic in $D$ , if $f_{i, j} = 0$ on a set containing a limit point then $f_{i, j} \equiv 0$ . The result now follows. $■$

Another approach, and a more general approach, to the semisimplicity result for graph algebras is given in [4] , [9] and alternatively for semicrossed products in [12] .

Proposition 6. Let

n \geq 2

, then there does not exist a nontrivial point derivation of

A (C_{n})

φ_{0, j}

for all

1 \leq j \leq n

Proof. Notice that since the range of $φ_{0, j}$ is $C$ we need not worry about inner derivations. Now let $D : A (C_{n}) \to C$ be a point derivation for $φ_{0, j}$ . Further, since the range of $φ_{0, j}$ is $C$ we know that $A (C_{n}) \sim = C \oplus ker φ_{0, j}$ as vector spaces.
Without loss of generality we will assume that $j = 1$ . Then $ker (φ_{0, j}) = {[\begin{matrix} z^{n} f_{1, 1} (z^{n}) & z f_{1, 2} (z^{n}) & z^{2} f_{1, 3} (z^{n}) & \cdot \cdot \cdot & z^{n - 1} f_{1, n} (z^{n}) \\ z^{n - 1} f_{2, 1} (z^{n}) & f_{2, 2} (z^{n}) & z f_{2, 3} (z^{n}) & \cdot \cdot \cdot & z^{n - 2} f_{2, n} (z^{n}) \\ z^{n - 2} f_{3, 1} (z^{n}) & z^{n - 1} f_{3, 2} (z^{n}) & f_{3, 3} (z^{n}) & \cdot \cdot \cdot & z^{n - 3} f_{3, n} (z^{n}) \\ . . . & . . . & . . . & . . . & . . . \\ z f_{n, 1} (z^{n}) & z^{2} f_{n, 2} (z^{n}) & z^{3} f_{n, 3} (z^{n}) & \cdot \cdot \cdot & f_{n, n} (z^{n}) \end{matrix}]}$ where $f_{i, j} \in A (D)$ for all $i, j$ . A short calculation shows that $(ker (φ_{0, 1}))^{2} = ker (φ_{0, 1})$ and hence any derivation must send $ker (φ_{0, 1})$ identically to zero. The result now follows, since any derivation must also send the copy of $C$ to zero.
A similar argument gives the result for arbitrary $j$ . $■$

Proposition 7. For

1 < m \leq n

let

π : A (C_{n}) \to M_{m}

be given by

π (x) = d i a g (φ_{0, j_{1}} (x), φ_{0, j_{2}} (x), \dots, φ_{0, j_{m}} (x))

where

j_{1} < j_{2} < \dots < j_{m}

. Then there exist nontrivial point derivations at

π

Proof. Notice that the derivation on $M_{n} \otimes A (D)$ given by $(f_{i, j}) \mapsto (f_{i, j}^{'} (0))$ restricts to a derivation $D : A (C_{n}) \to M_{n}$ . Further, by examining the derivation applied to the matrix $A = [\begin{matrix} 0 & z & 0 & 0 & \cdot \cdot \cdot & 0 \\ 0 & 0 & z & 0 & \cdot \cdot \cdot & 0 \\ 0 & 0 & 0 & z & \cdot \cdot \cdot & 0 \\ . . . & . . . & . . . & . . . & . . . & . . . \\ 0 & 0 & 0 & \cdot \cdot \cdot & 0 & z \\ z & 0 & 0 & \cdot \cdot \cdot & 0 & 0 \end{matrix}],$ we see that $D (A) \neq 0$ but $φ_{0} (A) = 0$ . It follows that $D$ is not inner.
Now if we truncate the matrix $D (A)$ by only keeping those rows and columns which correspond to one of the $j_{i}$ , then we have a noncommutative point derivation which is not inner at the representation $π$ . $■$

We denote the representation of the preceding result by

(φ_{j_{i}, 0})_{1 < i < m}

with the special case

{j_{1}, j_{2}, \dots, j_{n}} = {1, 2, \dots, n}

denoted by

φ_{0}

Here the set

M \ \bar{M^{2}}

m - 1

dimensional spanned by the

m - 1

matrices with a single

z

in the super diagonal, corresponding to the truncated matrix, and all other entries

0

Proposition 8. For

n \geq 1

then

H^{1} (A (C_{n}), φ_{0}) = C^{n},

Proof. Let

D : A (C_{n}) \to M_{n}

be a point derivation at

φ_{0}

. Notice that

r a n φ_{0}

is finite dimensional and hence

ker φ_{0}

has a Banach space complement in

A (C_{n})

which we will denote by

(ker φ_{0})^{c}

. Further, every

a \in A (C_{n})

can be written as

x_{a} + y_{a}

where

x_{a} \in (ker φ_{0})^{c}

and

y_{a} \in ker φ_{0}

. Further there exist

λ_{i}

such that

\sum_{i = 1}^{n} λ_{i} e_{i i}

where

e_{i i}

is the matrix with

1

in the

i

i

position and zero everywhere else.

We claim that if

a, b \in A (C_{n})

then, with respect to the decomposition above,

λ_{0} (x_{a} x_{b}) = 0

if and only if

x_{a} x_{b} = 0

. Writing

x_{a} = \sum_{i = 1}^{n} λ_{i} e_{i i}

and

x_{b} = \sum_{i = 1}^{n} μ_{i} e_{i i}

then,

x_{a} x_{b} = \sum_{i = 1}^{n} λ_{i} μ_{i} e_{i i}

and the claim follows.

Define the map

D^{'} : A (C_{n}) \to M_{n}

by letting

D^{'} (x_{a} + y_{a}) = D (y_{a})

with respect to the above decomposition. We will use the claim in the previous paragraph to show that

D^{'}

is a derivation at

φ_{0}

. Linearity, and continuity are clear. We need only establish the derivation property. Now

\begin{matrix} D^{'} (a b) & = D^{'} ((x_{a} + y_{a}) (x_{b} + y_{b})) \end{matrix}

\begin{matrix} = D^{'} (x_{a} x_{b} + y_{a} x_{b} + x_{a} y_{b} + y_{a} y_{b}) \end{matrix}

\begin{matrix} = D (y_{a} x_{b}) + D (x_{a} y_{b}) \end{matrix}

\begin{matrix} = D (y_{a}) φ_{0} (x_{b}) + φ_{0} (y_{a}) D (x_{b}) + D (x_{a}) φ_{0} (y_{b}) + φ_{0} (x_{a}) D (y_{b}) \end{matrix}

\begin{matrix} = D (y_{a}) φ_{0} (x_{b}) + φ_{0} (x_{a}) D (y_{b}) \end{matrix}

\begin{matrix} = D (y_{a}) φ_{0} (x_{b} + y_{b}) + φ_{0} (x_{a}) D^{'} (x_{b} + y_{b}) \end{matrix}

\begin{matrix} = D^{'} (x_{a} + y_{a}) φ_{0} (x_{b} + y_{b}) + φ_{0} (x_{a} + y_{a}) D^{'} (x_{b} + y_{b}) \end{matrix}

\begin{matrix} = D^{'} (a) φ_{0} (b) + φ_{0} (a) D^{'} (b) \end{matrix}

\begin{matrix}  \end{matrix}

and hence

D^{'}

is a derivation at

φ_{0}

It follows that every point derivation at

φ_{0}

can be written as an inner derivation and a derivation which sends

(ker φ_{0})^{c}

to zero. Since

D - D^{'}

will be zero on

φ_{0}

and hence inner. It follows that

H^{1} (A (C_{n}), φ_{0})

is isomorphic to the group generated by derivations which send

(ker φ_{0})^{c}

to zero.

Notice that each such derivation is uniquely determined by it's value on

ker φ_{0} \ \bar{(ker φ_{0})^{2}}

. A technical calculation shows us that the set

ker φ_{0} \ \bar{(ker φ_{0})^{2}}

is given by

{[\begin{matrix} 0 & λ_{1} z & 0 & 0 & \cdot \cdot \cdot & 0 \\ 0 & 0 & λ_{2} z & 0 & \cdot \cdot \cdot & 0 \\ 0 & 0 & 0 & λ_{3} z & \cdot \cdot \cdot & 0 \\ . . . & . . . & . . . & . . . & . . . & . . . \\ 0 & 0 & 0 & \cdot \cdot \cdot & 0 & λ_{n - 1} z \\ λ_{n} z & 0 & 0 & \cdot \cdot \cdot & 0 & 0 \end{matrix}] : λ_{i} \in C} .

For each

i

define the continuous map from

D_{i} : A (C_{n}) \to M_{n}

by sending

(ker φ_{0})^{c}

to zero,

\bar{(ker φ_{0})^{2}}

to zero, and

a

from the set

ker φ_{0} \ \bar{(ker φ_{0})^{2}}

as described above to

\frac{1}{z} e_{i i} a

D_{i}

will be a derivation sending

(ker φ_{0})^{c}

to zero and linear combinations of the

D_{i}

will span all possible point derivations at

φ_{0}

. The result now follows.

■

A similar result is true for the representations

(φ_{j_{i}, 0})_{1 < i < m}

and the proof follows in a method similar to the previous.

We now deal with the case of

0 < | λ | < 1

Proposition 9. There exists a nontrivial point derivation at

φ_{λ}

if and only if

0 < | λ | < 1

Proof. The existence of a derivation for $| λ | < 1$ follows from Proposition 3 since $A (C_{n})$ is a subalgebra of $M_{n} \otimes A (D)$ , and the representations coincide. We need only show that $| λ | = 1$ implies that there are no nonzero point derivations. If we investigate the ideal $M = ker (φ_{λ})$ we will see that $M^{2}$ is given by matrices in the set ${[\begin{matrix} (M_{λ})^{2} & z (M_{λ})^{2} & z^{2} (M_{λ})^{2} & \cdot \cdot \cdot & z^{n} - 1 (M_{λ})^{2} \\ z^{n - 1} (M_{λ})^{2} & (M_{λ})^{2} & z (M_{λ})^{2} & \cdot \cdot \cdot & z^{n - 1} (M_{λ})^{2} \\ z^{n - 2} (M_{λ})^{2} & z^{n - 1} (M_{λ})^{2} & (M_{λ})^{2} & \cdot \cdot \cdot & z^{n - 2} (M_{λ})^{2} \\ . . . & . . . & . . . & . . . & . . . \\ z (M_{λ})^{2} & z^{2} (M_{λ})^{2} & z^{3} (M_{λ})^{2} & \cdot \cdot \cdot & (M_{λ})^{2} \end{matrix}]},$ where $M_{λ}$ is the ideal in $A (z^{n})$ given by the kernel of the map $f (z^{n}) \mapsto f (λ^{n})$ . It suffices to show, then, that $M_{λ} = \bar{M_{λ}^{2}}$ . As the algebra $A (z^{n})$ is not a uniform algebra we can not use [3,Corollary1.6.4] directly as before.
We let $π_{n} : A (D) \to A (z^{n})$ be the completely contractive isomorphism induced by sending $z \mapsto z^{n}$ . (We are not making any claims about contractivity of the reverse map). Notice that $π_{n} ({f \in A (D) : f (λ) = 0}) \subseteq {f \in A (z^{n}) : f (λ^{n}) = 0} .$ Further, since $| λ | = 1$ we know that there is a uniformly bounded net ${f_{ι}} \subseteq {f \in A (D) : f (λ) = 0}$ such that $f_{ι} g \to g$ for all $g \in {f \in A (D) : f (λ) = 0} .$ Notice that $∥ π_{n} (f_{ι}) ∥ \leq ∥ f_{ι} ∥$ and hence ${π_{n} (f_{ι})}$ is a bounded net in ${f \in A (z^{n}) : f (λ^{n}) = 0}$ . Now if $g (λ^{n}) = 0$ then $h = π_{n}^{- 1} (g) \in {f \in A (D) : f (λ) = 0} .$ It follows that $f_{ι} h \to h$ . Now $π_{n} (f_{ι} g) \to g$ and hence the ideal ${f \in A (z^{n}) : f (λ^{n}) = 0}$ has a bounded approximate identity. The result now follows in a manner similar to the proof for $M_{n} \otimes A (D)$ . $■$

A uniqueness result for point derivations at

φ_{λ}

, where

0 < λ < 1

is not obvious.

We suspect that a result similar to 8 will hold but a different proof is necessary, since

a, b \in (φ_{λ})^{c}

does not imply that

a b \in (φ_{λ})^{c}

for

λ \neq 0

5 Homology for $A (C_{n})$

Now

M_{n}

is a right

A (C_{n})

module with module multiplication given by

a (X) = φ_{λ} (a) X

for all

X \in M_{n}

. To denote this module we will write

(M_{n})_{λ}

. Similarly, for

k \leq n

M_{k}

is a module over

A (C_{n})

with module multiplication given by

a X = (φ_{j_{i}, 0})_{1 \leq i \leq m} (a) X

We will denote this module by

M_{m, j_{1}, j_{2}, \dots, j_{m}}

Summarizing the results of the previous section in the language of Hochschild cohomology groups we have the following theorem.

Theorem 2. Let

A (C_{n})

be the quiver algebra associated to the

n

cycle graph. Then

H^{1} (A (C_{n}), (M_{n})_{λ}) \neq 0

if and only if

| λ | < 1

. Similarly,

H^{1} (A (C_{n}), M_{m, j_{1}, j_{2}, \dots, j_{m}}) = C^{m - 1},

where

C^{0} = {0}

In this section, in analogy with the classical theory of uniform algebras, we will show that the algebra

A (C_{n})

has no nontrivial derivations (i.e the homology group

H^{1} (A (C_{n}, A (C_{n}))

is trivial).

We begin with a definition.

Definition 3. Let

A

be a Banach algebra and

D : A \to A

be a continuous derivation. For a representation

π : A \to B (ℋ)

we say that

D

is locally inner at

π

π \circ D : A \to B (ℋ)

is inner.

Notice that in the context of

A (C_{n})

, a derivation

D

will be locally inner at

φ_{λ}

D |_{ker φ_{λ}}

is identically zero.

Theorem 3. Let

D : A (C_{n}) \to A (C_{n})

be a derivation such that

D

is locally inner at

φ_{λ}

for all

λ \in \bar{D}

. Then

D

is an inner derivation.

Proof. Let $D_{λ} : = φ_{λ} \circ D$ . Then, by hypothesis, $D_{λ} (a)$ can be written as $M_{λ} φ_{λ} (a) - φ_{λ} (a) M_{λ}$ for all $a \in A (C_{n})$ .
Now define a matrix function $M (z) : \bar{D} \to M_{n}$ by $M_{i j} (z) = M_{z}$ . We will show that there exists $(f_{i j}) \in \in A (C_{n})$ such that $D (a) = (f_{i j}) a - a (f_{i j})$ for all $a \in A (C_{n})$ .
Notice that $D (e_{i i}) \in A (C_{n})$ and further $φ_{λ} \circ D (e_{i i}) = M_{λ} e_{i i} - e_{i i} M_{λ}$ . As we let $λ$ vary we see that $M_{i j}$ is a uniquely determined element of $z^{| i - j |} A (z^{n})$ for $i \neq j$ . So we define $f_{i j} (z) = M_{i j} (z)$ for $i \neq j$ . We need only determine the entries of $(f_{i j})$ along the diagonal.
Notice that $M_{i i} (λ)$ is uniformly bounded. Now let $A$ be the norm closed algebra generated by ${M_{i i} (z) : 1 \leq i \leq n}$ and $A (z^{n})$ . Notice that $A (z^{n})$ is a closed subalgebra of $A$ , and further as a Banach space $A / A (z^{n})$ is finite dimensional and hence $A (z^{n})$ is complemented. Now let $f_{i i} (z) = M_{i i} (z) P_{A (z^{n})}$ , where $P_{A (z^{n})}$ is the projection of $A$ onto the complemented subspace given by $A (z^{n})$ . Notice that $(f_{i j}) \in A (C_{n})$ . Since elements of $A (C_{n})$ are uniquely determined by the value of $φ_{λ}$ as we let $λ$ vary, $(f_{i j})$ induces the derivation $D$ and the result follows. $■$

We will show that the homology group

H^{1} (A (C_{n}), A (C_{n}))

is trivial by showing that every derivation on

A (C_{n})

is locally inner.

Theorem 4. Every derivation on

A (C_{n})

is inner, hence homology group

H^{1} (A (C_{n}), A (C_{n}))

is trivial for all

n \geq 1

Proof. The case of $n = 1$ is well known, see [7,Proposition9.1] . We will show the result for $n \geq 2$ . Fix $D$ , a derivation on $A (C_{n})$ . Now since $A (C_{n})$ is semisimple, every derivation is continuous, see [8] . Thus $φ_{λ} \circ D$ is a noncommutative point derivation for all $| λ | \leq 1$ .
In particular, since every point derivation at $φ_{λ}$ is inner for all $| λ | = 1$ , it follows that $φ_{λ} \circ D$ is inner for all $| λ | = 1$ .
Now $φ_{0} \circ D$ is also a point derivation at $φ_{0}$ . By the description of point derivations in Proposition 8 , and since $r a n φ_{0}$ is contained in the diagonal matrices, it follows that $(φ_{0} \circ D)$ is an inner point derivation at $φ_{0}$ . In fact, simple calculations with the matrices $e_{i i}$ tells us that the matrix $X$ such that $φ_{0} \circ D (a) = φ_{0} (a) X - X φ_{0} (a)$ is a diagonal matrix and hence in the range of $φ_{0}$ .
We need only show that $φ_{λ} \circ D$ is locally inner at $φ_{λ}$ for all $0 < | λ | < 1$ .
Let $(f) \in ker φ_{λ}$ , then notice that $(e_{i i}) (f) (e_{j j}) \in ker φ_{λ}$ , for all $i, j$ . Further notice that $(e_{i i}) (f) (e_{j j})$ is the matrix which is the same as $(f)$ in the $i$ - $j$ entry and zero everywhere else. Now $φ_{λ} \circ D ((e_{i i}) (f) (e_{j j})) = (e_{i i}) (φ_{λ} \circ D (f)) (e_{j j})$ which is a matrix which is zero in all components except perhaps the $i$ - $j$ entry.
Further notice that for $| μ | = 1$ $e_{i i} [φ_{μ} \circ D ((e_{i i}) (f) (e_{j j}))] e_{j j} = e_{i i} [(e_{i i} (f) (e_{j j}) X - X (e_{i i}) (f) (e_{j j})] e_{j j}$ which is zero. In particular, the $i$ - $j$ entry of $φ_{μ} \circ D$ will be zero for all matrices $g$ with the only nonzero entry being the $i$ - $j$ entry. Notice that $μ$ was an arbitrary element of $T$ .
Now suppose $f \in ker φ_{λ}$ with $φ \circ D \neq 0$ . Then there exists some $i$ and $j$ such that $φ \circ D ((e_{i i}) f (e_{j j})) \neq 0$ , and in particular the $i$ - $j$ entry of $φ \circ D ((e_{i i}) f (e_{j j}))$ is nonzero. But since $D$ is a derivation it follows that $D ((e_{i i} f (e_{j j})$ yields a matrix of analytic functions. But since the $i$ - $j$ entry of $φ_{μ} \circ D$ is zero for all $μ \in T$ it follows that the $i$ - $j$ entry of $φ_{λ} \circ D$ must be zero, by the maximum modulus theorem. This is a contradiction and hence every point derivation induced by $D$ must be inner.
The result now follows. $■$

An interesting question is for which

n

H^{n} (A (C_{n}, A (C_{n})) = 0

. It is shown in [7] that

H^{2} (A (D), A (D)) \neq 0

. We are pursuing a similar result in this setting.

6 Amenability for graph algebras

We can put together the results of this paper, and the paper [13] to present a proof of the perhaps well known result that the quiver algebras associated to a directed graph with at least one edge are never amenable. This extends the result of [13,Corollary4.3] . Clearly if a graph has no edges then the associated quiver algebra is a direct sum of copies of

C

and hence is amenable.

Theorem 5. Let

A

be the quiver algebra associated to the directed graph

Q

. If

Q

has at least one edge then

A

is not amenable.

Proof. We first show the result for the semisimple quiver algebras. Let

Q

be a graph such that

A (Q)

is a semisimple quiver algebra. By Theorem 5.1 in [10] each vertex is contained in a cycle. Choose a vertex

v

and let

e_{1}, e_{2}, \dots, e_{n}

be a minimal cycle of length

n

beginning and ending at

v

. Here by minimal, we mean that

s (e_{i}) \neq s (e_{j})

and

r (e_{i}) \neq e (e_{j})

for

i \neq j

Fixing a consecutive ordering on the edges

{f_{i}}

C_{n}

define a representation

π : A (Q) \to A (C_{n})

π (P_{v}) = {\begin{matrix} P_{v_{i}} & if v = s (e_{i}) \\ 0 & else \end{matrix}

and

π (L_{e}) = {\begin{matrix} L_{f_{i}} & if e = e_{i} \\ 0 & else . \end{matrix}

Since amenability is preserved by homomorphism, and

A (C_{n})

is not amenable for all

n \geq 1

, the result follows.

For graphs with no cycles we use a different representation. Let

e

be an edge in

Q

with

r (e) \neq s (e)

. Define

π : A (Q) \to M_{2}

by extending the map with

\begin{matrix} π (P_{s (e)}) & = [\begin{matrix} 0 & 0 \\ 0 & 1 \end{matrix}], \end{matrix}

\begin{matrix} π (P_{r (e)}) & = [\begin{matrix} 1 & 0 \\ 0 & 0 \end{matrix}], \end{matrix}

\begin{matrix} π (T_{e}) & = [\begin{matrix} 0 & 1 \\ 0 & 0 \end{matrix}], \end{matrix}

\begin{matrix}  \end{matrix}

π (P_{v}) = 0 for v \notin {s (e), r (e)}

, and

π (T_{f}) = 0 for f \neq e

. Notice that

r a n π = T_{2}

which is not amenable and hence the result follows.

■

References

M. Alaimia, Automorphisms of some Banach algebras of analytic functions, Linear Algebra Appl. 298 (1999) 87-97.
A. Browder, Point derivations on function algebras, J. Funct. Anal. 1 (1967) 22-27.
A. Browder, “Introduction to Function Algebras,” W. A. Benjamin, New York, 1969
K. Davidson and E. Katsoulis, Nest representations of directed graph algebras, preprint.
K. Davidson and D. Pitts, The algebraic structure of non-commutative analytic Toeplitz algebras, Math. Ann. 311 (2) (1998), 275-303.
L. DeAlba and J. Peters, Classification of semicrossed products of finite-dimensional $C^{*}$ -algebras, Proc. Amer. Math. Soc. 95(4) (1985), 557-564.
B. Johnson “Cohomology in Banach Algebras” American Mathematical Society, Providence, 1972.
B. Johnson and A. Sinclair, Continuity of derivations and a problem of Kaplansky, Amer. J. Math. 90 (1968), 1067-1073.
M. Jury and D. Kribs, Ideal structure in free semigroupoid algebras from directed graphs, J. Operator Theory to appear.
D. Kribs and S. Power, Free semigroupoid algebras, J. Ramanujan Math. Soc. 19 (2004), 75-114.
V. Paulsen, “Completely Bounded Maps and Operator Algebras” Cambridge University Press, Cambridge, 2002.
J. Peters, The ideal structure of certain nonselfadjoint operator algebras, Trans. Amer. Math. Soc. 305(1) (1988), 333-352.
G. Popescu, Noncommutative joint dilations and free product operator algebras, Pacific J. Math. 186(1) (1998), 111-140.

Department of Mathematics, 300 Minard Hall, North Dakota State University, Fargo, ND 58105-5075 E-mail address : benton.duncan@ndsu.edu

Benton L. Duncan