, ,

Throughout this paper, let

G

be a countable directed graph and let

F^{+} (G)

be the free semigroupoid of

G,

in the sense of Kribs and Power. i.e., it is a collection of all vertices of the graph

G

as units and all admissible finite paths, under the admissibility. As a set, the free semigroupoid

F^{+} (G)

can be decomposed by

F^{+} (G) = V (G) \cup F P (G),

where

V (G)

is the vertex set of the graph

G

and

F P (G)

is the set of all admissible finite paths. Trivially the edge set

E (G)

of the graph

G

is properly contained in

F P (G),

since all edges of the graph can be regarded as finite paths with their length

1 .

We define a graph

W^{*}

-algebra of

G

W^{*} (G) = d e f {\bar{C [{L_{w}, L_{w}^{*} : w \in F^{+} (G)}]}}^{w},

where

L_{w}

and

L_{w}^{*}

are creation operators and annihilation operators on the generalized Fock space

H_{G} = l^{2} (F^{+} (G))

induced by the given graph

G,

respectively.

Notice that the creation operators induced by vertices are projections and the creation operators induced by finite paths are partial isometries. We can define the

W^{*}

-subalgebra

D_{G}

W^{*} (G),

which is called the diagonal subalgebra by

D_{G} = d e f {\bar{C [{L_{v} : v \in V (G)}]}}^{w} .

Then each element

a

in the graph

W^{*}

-algebra

W^{*} (G)

is expressed by

a = \sum w \in F^{+} (G : a), u_{w} \in {1, *} p_{w} L_{w}^{u_{w}},

for

p_{w} \in C,

where

F^{+} (G : a)

is a support of the element

a

, as a subset of the free semigroupoid

F^{+} (G) .

The above expression of the random variable

a

is said to be the Fourier expansion of

a .

Since

F^{+} (G)

is decomposed by the disjoint subsets

V (G)

and

F P (G),

the support

F^{+} (G : a)

a

is also decomposed by the following disjoint subsets,

V (G : a) = F^{+} (G : a) \cap V (G)

and

F P (G : a) = F^{+} (G : a) \cap F P (G) .

Thus the operator

a

can be re-expressed by

a = \sum v \in V (G : a) p_{v} L_{v} + \sum w \in F P (G : a), u_{w} \in {1, *} p_{w} L_{w}^{u_{w}} .

Notice that if

V (G : a) \neq \emptyset,

then

\sum v \in V (G : a) p_{v} L_{v}

is contained in the diagonal subalgebra

D_{G} .

Thus we have the canonical conditional expectation

E : W^{*} (G) \to D_{G},

defined by

E (a) = \sum v \in V (G : a) p_{v} L_{v},

for all

a = \sum w \in F^{+} (G : a), u_{w} \in {1, *} p_{w} L_{w}^{u_{w}}

W^{*} (G) .

Then the algebraic pair

(W^{*} (G), E)

is a

W^{*}

-probability space with amalgamation over

D_{G}

(See [16]).

It is easy to check that the conditional expectation

E

is faithful in the sense that if

E (a^{*} a) = 0_{D_{G}},

for

a \in W^{*} (G),

then

a = 0_{D_{G}} .

For the fixed operator

a \in W^{*} (G),

the support

F^{+} (G : a)

of the operator

a

is again decomposed by

F^{+} (G : a) = V (G : a) \cup F P_{*} (G : a) \cup F P_{*}^{c} (G : a),

with the decomposition of

F P (G : a),

F P (G : a) = F P_{*} (G : a) \cup F P_{*}^{c} (G : a),

where

F P_{*} (G : a) = {w \in F P (G : a) :

both

L_{w}

and

L_{w}^{*}

are summands of

a}

and

F P_{*} (G : a) = F P (G : a) \ F P_{*} (G : a) .

The above new expression plays a key role to find the

D_{G}

-valued moments of the random variable

a .

In fact, the summands

p_{v} L_{v}

's and

p_{w} L_{w} + p_{w^{t}} L_{w}^{*},

for

v \in V (G : a)

and

w \in F P_{*} (G : a)

act for the computation of

D_{G}

-valued moments of

a .

By using the above partition of the support of a random variable, we can compute the

D_{G}

-valued moments and

D_{G}

-valued cumulants of it via the lattice path model

L P_{n}

and the lattice path model

L P_{n}^{*}

satisfying the

*

-axis-property.

At a first glance, the computations of

D_{G}

-valued moments and cumulants look so abstract and hence it looks useless. However, these computations, in particular the computation of

D_{G}

-valued cumulants, provides us how to figure out the

D_{G}

-freeness of random variables by making us compute the mixed cumulants. As applications, in the final chapter, we can compute the moment and cumulant of the operator that is the sum of

N

-free semicircular elements with their covariance

2 .

Based on the

D_{G}

-cumulant computation, we can characterize the

D_{G}

-freeness of generators of

W^{*} (G),

by the so-called diagram-distinctness on the graph

G .

i.e., the random variables

L_{w_{1}}

and

L_{w_{2}}

are free over

D_{G}

if and only if

w_{1}

and

w_{2}

are diagram-distinct the sense that

w_{1}

and

w_{2}

have different diagrams on the graph

G .

Also, we could find the necessary condition for the

D_{G}

-freeness of two arbitrary random variables

a

and

b .

i.e., if the supports

F^{+} (G : a)

and

F^{+} (G : b)

are diagram-distinct, in the sense that

w_{1}

and

w_{2}

are diagram distinct for all pairs

(w_{1}, w_{2})

\in

F^{+} (G : a)

\times

F^{+} (G : b),

then the random variables

a

and

b

are free over

D_{G} .

In [17], we considered some special

D_{G}

-valued random variables in a graph

W^{*}

-probability space

(W^{*} (G), E) .

The those random variables are the basic objects to study Free Probability Theory. We can conclude that (i) if

l

is a loop, then

L_{l} + L_{l}^{*}

D_{G}

-semicircular.

(ii) if

w

is a finite path, then

L_{w} + L_{w}^{*}

D_{G}

-even.

(iii) if

w

is a finite path, then

L_{w}

and

L_{w}^{*}

are

D_{G}

-valued R-diagonal. In [18], we observed the diagonal compressed random variables in the graph

W^{*}

-probability space

(W^{*} (G), E) .

Let

v_{1}, . . ., v_{N} \in V (G)

and let

a

be a

D_{G}

-valued random variable in

(W^{*} (G), E) .

Define the diagonal compressed random variable of

a

V = {v_{1}, . . ., v_{N}}

C_{V} (a) = L_{v_{1}} a L_{v_{1}} + . . . + L_{v_{N}} a L_{v_{N}} .

Notice that if

v \in V (G),

then

L_{v} a L_{v}

is the compressed random variable by

L_{v}

and the compressed random variable has its support contained in

{v}

\cup

l o o p_{v} (G),

where

l o o p_{v} (G)

=

{l \in l o o p (G) :

l = v l v} .

The main purpose of this paper is to show that the compressed random variable

P_{V} a P_{V}

of a projection

P_{V} = \sum_{j = 1}^{N} L_{v_{j}},

where

V = {v_{1}, . . ., v_{N}},

has the same free probabilistic information with the diagonal compressed random variable

C_{V} (a) .

i.e., the compressed random variables

P_{V} a P_{V}

and

C_{V} (a)

have the same

D_{G}

-valued moments and cumulants.

1 Graph $W^{*}$ -Probability Theory

Let

G

be a countable directed graph and let

F^{+} (G)

be the free semigroupoid of

G .

i.e., the set

F^{+} (G)

is the collection of all vertices as units and all admissible finite paths of

G .

Let

w

be a finite path with its source

s (w) = x

and its range

r (w) = y,

where

x, y \in V (G) .

Then sometimes we will denote

w

w = x w y

to express the source and the range of

w .

We can define the graph Hilbert space

H_{G}

by the Hilbert space

l^{2} (F^{+} (G))

generated by the elements in the free semigroupoid

F^{+} (G) .

i.e., this Hilbert space has its Hilbert basis

ℬ = {ξ_{w} : w \in F^{+} (G)} .

Suppose that

w = e_{1} . . . e_{k} \in F P (G)

is a finite path with

e_{1}, . . ., e_{k} \in E (G) .

Then we can regard

ξ_{w}

ξ_{e_{1}} \otimes . . . \otimes ξ_{e_{k}} .

So, in [10], Kribs and Power called this graph Hilbert space the generalized Fock space. Throughout this paper, we will call

H_{G}

the graph Hilbert space to emphasize that this Hilbert space is induced by the graph.

Define the creation operator

L_{w},

for

w \in F^{+} (G),

by the multiplication operator by

ξ_{w}

H_{G} .

Then the creation operator

L

H_{G}

satisfies that (i)

L_{w} = L_{x w y} = L_{x} L_{w} L_{y},

for

w = x w y

with

x, y \in V (G) .

(ii)

L_{w_{1}} L_{w_{2}} = {\begin{matrix} L_{w_{1} w_{2}} & if w_{1} w_{2} \in F^{+} (G) \\ 0 & if w_{1} w_{2} / \in F^{+} (G), \end{matrix}

for all

w_{1}, w_{2} \in F^{+} (G) .

Now, define the annihilation operator

L_{w}^{*},

for

w \in F^{+} (G)

L_{w}^{*} ξ_{w^{'}} = d e f {\begin{matrix} ξ_{h} & if w^{'} = w h \in F^{+} (G) ξ \\ 0 & otherwise. \end{matrix}

The above definition is gotten by the following observation ;

\begin{matrix} < L_{w} ξ_{h}, ξ_{w h} > & = < ξ_{w h}, ξ_{w h} > \end{matrix}

\begin{matrix} = 1 = < ξ_{h}, ξ_{h} > \end{matrix}

\begin{matrix} = < ξ_{h}, L_{w}^{*} ξ_{w h} >, \end{matrix}

where

<, >

is the inner product on the graph Hilbert space

H_{G} .

Of course, in the above formula we need the admissibility of

w

and

h

F^{+} (G) .

However, even though

w

and

h

are not admissible (i.e.,

w h / \in F^{+} (G)

), by the definition of

L_{w}^{*},

we have that

\begin{matrix} < L_{w} ξ_{h}, ξ_{h} > & = < 0, ξ_{h} > \end{matrix}

\begin{matrix} = 0 = < ξ_{h}, 0 > \end{matrix}

\begin{matrix} = < ξ_{h}, L_{w}^{*} ξ_{h} > . \end{matrix}

Notice that the creation operator

L

and the annihilation operator

L^{*}

satisfy that (1.1)

L_{w}^{*} L_{w} = L_{y}

and

L_{w} L_{w}^{*} = L_{x},

for all

w = x w y \in F^{+} (G),

under the weak topology, where

x, y \in V (G) .

Remark that if we consider the von Neumann algebra

W^{*} ({L_{w}})

generated by

L_{w}

and

L_{w}^{*}

B (H_{G}),

then the projections

L_{y}

and

L_{x}

are Murray-von Neumann equivalent, because there exists a partial isometry

L_{w}

satisfying the relation (1.1). Indeed, if

w = x w y

F^{+} (G),

with

x, y \in V (G),

then under the weak topology we have that (1,2)

L_{w} L_{w}^{*} L_{w} = L_{w}

and

L_{w}^{*} L_{w} L_{w}^{*} = L_{w}^{*} .

So, the creation operator

L_{w}

is a partial isometry in

W^{*} ({L_{w}})

B (H_{G}) .

Assume now that

v \in V (G) .

Then we can regard

v

v = v v v .

So, (1.3)

L_{v}^{*} L_{v} = L_{v} = L_{v} L_{v}^{*} = L_{v}^{*} .

This relation shows that

L_{v}

is a projection in

B (H_{G})

for all

v \in V (G) .

Define the graph

W^{*}

-algebra

W^{*} (G)

W^{*} (G) = d e f {\bar{C [{L_{w}, L_{w}^{*} : w \in F^{+} (G)}]}}^{w} .

Then all generators are either partial isometries or projections, by (1.2) and (1.3).

So, this graph

W^{*}

-algebra contains a rich structure, as a von Neumann algebra.

(This construction can be the generalization of that of group von Neumann algebra.) Naturally, we can define a von Neumann subalgebra

D_{G} \subset W^{*} (G)

generated by all projections

L_{v},

v \in V (G) .

i.e.

D_{G} = d e f W^{*} ({L_{v} : v \in V (G)}) .

We call this subalgebra the diagonal subalgebra of

W^{*} (G) .

Notice that

D_{G} = Δ_{| G |} \subset M_{| G |} (C),

where

Δ_{| G |}

is the subalgebra of

M_{| G |} (C)

generated by all diagonal matrices. Also, notice that

1_{D_{G}} = \sum v \in V (G) L_{v} = 1_{W^{*} (G)} .

a \in W^{*} (G)

is an operator, then it has the following decomposition which is called the Fourier expansion of

a

; (1.4)

a = \sum w \in F^{+} (G : a), u_{w} \in {1, *} p_{w} L_{w}^{u_{w}},

where

p_{w} \in C

and

F^{+} (G : a)

is the support of

a

defined by

F^{+} (G : a) = {w \in F^{+} (G) : p_{w} \neq 0} .

Remark that the free semigroupoid

F^{+} (G)

has its partition

{V (G), F P (G)},

as a set. i.e.,

F^{+} (G) = V (G) \cup F P (G)

and

V (G) \cap F P (G) = \emptyset .

So, the support of

a

is also partitioned by

F^{+} (G : a) = V (G : a) \cup F P (G : a),

where

V (G : a) = d e f V (G) \cap F^{+} (G : a)

and

F P (G : a) = d e f F P (G) \cap F^{+} (G : a) .

So, the above Fourier expansion (1.4) of the random variable

a

can be re-expressed by (1.5)

a = \sum v \in V (G : a) p_{v} L_{v} + \sum w \in F P (G : a), u_{w} \in {1, *} p_{w} L_{w}^{u_{w}} .

We can easily see that if

V (G : a) \neq \emptyset,

then

\sum v \in V (G : a) p_{v} L_{v}

is contained in the diagonal subalgebra

D_{G} .

Also, if

V (G : a) = \emptyset,

then

\sum v \in V (G : a) p_{v} L_{v} = 0_{D_{G}} .

So, we can define the following canonical conditional expectation

E : W^{*} (G) \to D_{G}

by (1.6)

E (a) = E (\sum w \in F^{+} (G : a), u_{w} \in {1, *} p_{w} L_{w}^{u_{w}}) = d e f \sum v \in V (G : a) p_{v} L_{v},

for all

a \in W^{*} (G) .

Indeed,

E

is a well-determined conditional expectation.

Definition 1.1. Let

G

be a countable directed graph and let

W^{*} (G)

be the graph

W^{*}

-algebra induced by

G .

Let

E : W^{*} (G) \to D_{G}

be the conditional expectation defined above. Then we say that the algebraic pair

(W^{*} (G), E)

is the graph

W^{*}

-probability space over the diagonal subalgebra

D_{G}

. By the very definition, it is one of the

W^{*}

-probability space with amalgamation over

D_{G} .

All elements in

(W^{*} (G), E)

are called

D_{G}

-valued random variables.

We have a graph

W^{*}

-probability space

(W^{*} (G), E)

over its diagonal subalgebra

D_{G} .

We will define the following free probability data of

D_{G}

-valued random variables.

Definition 1.2. Let

W^{*} (G)

be the graph

W^{*}

-algebra induced by

G

and let

a \in W^{*} (G) .

Define the

n

-th (

D_{G}

-valued) moment of

a

E (d_{1} a d_{2} a . . . d_{n} a),

for all

n \in N

, where

d_{1}, . . ., d_{n} \in D_{G}

. Also, define the

n

-th (

D_{G}

-valued) cumulant of

a

k_{n} (d_{1} a, d_{2} a, . . ., d_{n} a) = C^{(n)} (d_{1} a \otimes d_{2} a \otimes . . . \otimes d_{n} a),

for all

n \in N,

and for

d_{1}, . . ., d_{n} \in D_{G},

where

\hat{C} = (C^{(n)})_{n = 1}^{\infty} \in I^{c} (W^{*} (G), D_{G})

is the cumulant multiplicative bimodule map induced by the conditional expectation

E,

in the sense of Speicher. We define the

n

-th trivial moment of

a

and the

n

-th trivial cumulant of

a

E (a^{n})

and

k_{n} (\underset{︸}{a, a, . . ., a} n - t i m e s) = C^{(n)} (a \otimes a \otimes . . . \otimes a),

respectively, for all

n \in N .

To compute the

D_{G}

-valued moments and cumulants of the

D_{G}

-valued random variable

a,

we need to introduce the following new definition ;

Definition 1.3. Let

(W^{*} (G), E)

be a graph

W^{*}

-probability space over

D_{G}

and let

a \in (W^{*} (G), E)

be a random variable. Define the subset

F P_{*} (G : a)

F P (G : a)

F P_{*} (G : a) = d e f {w \in F^{+} (G : a) :

both

L_{w}

and

L_{w}^{*}

are summands of

a} .

And let

F P_{*}^{c} (G : a) = d e f F P (G : a) \ F P_{*} (G : a) .

We already observed that if

a \in (W^{*} (G), E)

is a

D_{G}

-valued random variable, then

a

has its Fourier expansion

a_{d} + a_{0},

where

a_{d} = \sum v \in V (G : a) p_{v} L_{v}

and

a_{0} = \sum w \in F P (G : a), u_{w} \in {1, *} p_{w} L_{w}^{u_{w}} .

By the previous definition, the set

F P (G : a)

is partitioned by

F P (G : a) = F P_{*} (G : a) \cup F P_{*}^{c} (G : a),

for the fixed random variable

a

(W^{*} (G), E) .

So, the summand

a_{0},

in the Fourier expansion of

a = a_{d} + a_{0},

has the following decomposition ;

a_{0} = a_{(*)} + a_{(n o n - *)},

where

a_{(*)} = \sum l \in F P_{*} (G : a) (p_{l} L_{l} + p_{l^{t}} L_{l}^{*})

and

a_{(n o n - *)} = \sum w \in F P_{*}^{c} (G : a), u_{w} \in {1, *} p_{w} L_{w}^{u_{w}},

where

p_{l^{t}}

is the coefficient of

L_{l}^{*}

depending on

l \in F P_{*} (G : a) .

1.1 $D_{G}$ -Moments and $D_{G}$ -Cumulants of Random Variables

Throughout this chapter, let

G

be a countable directed graph and let

(W^{*} (G), E)

be the graph

W^{*}

-probability space over its diagonal subalgebra

D_{G} .

In this chapter, we will compute the

D_{G}

-valued moments and the

D_{G}

-valued cumulants of arbitrary random variable

a = \sum w \in F^{+} (G : a), u_{w} \in {1, *} p_{w} L_{w}^{u_{w}}

in the graph

W^{*}

-probability space

(W^{*} (G), E)

1.1.1 Lattice Path Model

Throughout this section, let

G

be a countable directed graph and let

(W^{*} (G), E)

be the graph

W^{*}

-probability space over its diagonal subalgebra

D_{G} .

Let

w_{1}, . . ., w_{n} \in F^{+} (G)

and let

L_{w_{1}}^{u_{w_{1}}} . . . L_{w_{n}}^{u_{w_{n}}} \in (W^{*} (G), E)

be a

D_{G}

-valued random variable. In this section, we will define a lattice path model for the random variable

L_{w_{1}}^{u_{w_{1}}} . . . L_{w_{n}}^{u_{w_{n}}} .

Recall that if

w = e_{1} . . . . e_{k} \in F P (G)

with

e_{1}, . . ., e_{k} \in E (G),

then we can define the length

| w |

w

k .

i.e.e, the length

| w |

w

is the cardinality

k

of the admissible edges

e_{1}, . . ., e_{k} .

Definition 1.4. Let

G

be a countable directed graph and

F^{+} (G),

the free semigroupoid. If

w \in F^{+} (G),

then

L_{w}

is the corresponding

D_{G}

-valued random variable in

(W^{*} (G), E) .

We define the lattice path

l_{w}

L_{w}

and the lattice path

l_{w}^{- 1}

L_{w}^{*}

by the lattice paths satisfying that ; (i) the lattice path

l_{w}

starts from

* = (0, 0)

on the

R^{2}

-plane.

(ii) if

w \in V (G),

then

l_{w}

has its end point

(0, 1) .

(iii) if

w \in E (G),

then

l_{w}

has its end point

(1, 1) .

(iv) if

w \in E (G),

then

l_{w}^{- 1}

has its end point

(- 1, - 1) .

(v) if

w \in F P (G)

with

| w | = k,

then

l_{w}

has its end point

(k, k) .

(vi) if

w \in F P (G)

with

| w | = k,

then

l_{w}^{- 1}

has its end point

(- k, - k) .

Assume that finite paths

w_{1}, . . ., w_{s}

F P (G)

satisfy that

w_{1} . . . w_{s} \in F P (G) .

Define the lattice path

l_{w_{1} . . . w_{s}}

by the connected lattice path of the lattice paths

l_{w_{1}},

...,

l_{w_{s}} .

i.e.e,

l_{w_{2}}

starts from

(k_{w_{1}}, k_{w_{1}}) \in R^{+}

and ends at

(k_{w_{1}} + k_{w_{2}}, k_{w_{1}} + k_{w_{2}}),

where

| w_{1} | = k_{w_{1}}

and

| w_{2} | = k_{w_{2}} .

Similarly, we can define the lattice path

l_{w_{1} . . . w_{s}}^{- 1}

as the connected path of

l_{w_{s}}^{- 1},

l_{w_{s - 1}}^{- 1},

...,

l_{w_{1}}^{- 1} .

Definition 1.5. Let

G

be a countable directed graph and assume that

L_{w_{1}}, . . ., L_{w_{n}}

are generators of

(W^{*} (G), E) .

Then we have the lattice paths

l_{w_{1}},

...,

l_{w_{n}}

L_{w_{1}},

...,

L_{w_{n}},

respectively in

R^{2} .

Suppose that

L_{w_{1}}^{u_{w_{1}}} . . . L_{w_{n}}^{u_{w_{n}}} \neq 0_{D_{G}}

(W^{*} (G), E),

where

u_{w_{1}}, . . ., u_{w_{n}} \in {1, *} .

Define the lattice path

l_{w_{1}, . . ., w_{n}}^{u_{w_{1}}, . . ., u_{w_{n}}}

of nonzero

L_{w_{1}}^{u_{w_{1}}} . . . L_{w_{n}}^{u_{w_{n}}}

by the connected lattice path of

l_{w_{1}}^{t_{w_{1}}},

...,

l_{w_{n}}^{t_{w_{n}}},

where

t_{w_{j}} = 1

u_{w_{j}} = 1

and

t_{w_{j}} = - 1

u_{w_{j}} = * .

Assume that

L_{w_{1}}^{u_{w_{1}}} . . . L_{w_{n}}^{u_{w_{n}}}

=

0_{D_{G}} .

Then the empty set

\emptyset

R^{2}

is the lattice path of it.

We call it the empty lattice path. By

L P_{n},

we will denote the set of all lattice paths of the

D_{G}

-valued random variables having their forms of

L_{w_{1}}^{u_{w_{1}}} . . . L_{w_{n}}^{u_{w_{n}}},

including empty lattice path.

Also, we will define the following important property on the set of all lattice paths ;

Definition 1.6. Let

l_{w_{1}, . . ., w_{n}}^{u_{w_{1}}, . . ., u_{w_{n}}} \neq \emptyset

be a lattice path of

L_{w_{1}}^{u_{w_{1}}} . . . L_{w_{n}}^{u_{w_{n}}} \neq 0_{D_{G}}

L P_{n} .

If the lattice path

l_{w_{1}, . . ., w_{n}}^{u_{w_{1}}, . . ., u_{w_{n}}}

starts from

*

and ends on the

*

-axis in

R^{+},

then we say that the lattice path

l_{w_{1}, . . ., w_{n}}^{u_{w_{1}}, . . ., u_{w_{n}}}

has the

*

-axis-property. By

L P_{n}^{*},

we will denote the set of all lattice paths having their forms of

l_{w_{1}, . . ., w_{n}}^{u_{w_{1}}, . . ., u_{w_{n}}}

which have the

*

-axis-property. By little abuse of notation, sometimes, we will say that the

D_{G}

-valued random variable

L_{w_{1}}^{u_{w_{1}}} . . . L_{w_{n}}^{u_{w_{n}}}

satisfies the

*

-axis-property if the lattice path

l_{w_{1}, . . ., w_{n}}^{u_{w_{1}}, . . ., u_{w_{n}}}

of it has the

*

-axis-property.

The following theorem shows that finding

E (L_{w_{1}}^{u_{w_{1}}} . . . L_{w_{n}}^{u_{w_{n}}})

is checking the

*

-axis-property of

L_{w_{1}}^{u_{w_{1}}} . . . L_{w_{n}}^{u_{w_{n}}} .

Theorem 1.1. (See [15]) Let

L_{w_{1}}^{u_{w_{1}}} . . . L_{w_{n}}^{u_{w_{n}}} \in (W^{*} (G), E)

be a

D_{G}

-valued random variable, where

u_{w_{1}}, . . ., u_{w_{n}} \in {1, *} .

Then

E (L_{w_{1}}^{u_{w_{1}}} . . . L_{w_{n}}^{u_{w_{n}}})

\neq

0_{D_{G}}

if and only if

L_{w_{1}}^{u_{w_{1}}} . . . L_{w_{n}}^{u_{w_{n}}}

has the

*

-axis-property (i.e., the corresponding lattice path

l_{w_{1}, . . ., w_{n}}^{u_{w_{1}}, . . ., u_{w_{n}}}

L_{w_{1}}^{u_{w_{1}}} . . . L_{w_{n}}^{u_{w_{n}}}

is contained in

L P_{n}^{*} .

Notice that

\emptyset / \in L P_{n}^{*} .

)

□

By the previous theorem, we can conclude that

E (L_{w_{1}}^{u_{w_{1}}} . . . L_{w_{n}}^{u_{w_{n}}}) = L_{v},

for some

v \in V (G)

if and only if the lattice path

l_{w_{1}, . . ., w_{n}}^{u_{w_{1}}, . . ., u_{w_{n}}}

has the

*

-axis-property (i.e.,

l_{w_{1}, . . ., w_{n}}^{u_{w_{1}}, . . ., u_{w_{n}}} \in L P_{n}^{*}

1.1.2 $D_{G}$ -Valued Moments and Cumulants of Random Variables

Let

w_{1}, . . ., w_{n} \in F^{+} (G)

u_{1}, . . ., u_{n} \in {1, *}

and let

L_{w_{1}}^{u_{1}} . . . L_{w_{n}}^{u_{n}} \in (W^{*} (G), E)

be a

D_{G}

-valued random variable. Recall that, in the previous section, we observed that the

D_{G}

-valued random variable

L_{w_{1}}^{u_{1}} . . . L_{w_{n}}^{u_{n}} = L_{v} \in (W^{*} (G), E)

with

v \in V (G)

if and only if the lattice path

l_{w_{1}, . . ., w_{n}}^{u_{1}, . . ., u_{n}}

L_{w_{1}}^{u_{1}} . . . L_{w_{n}}^{u_{n}}

has the

*

-axis-property (equivalently,

l_{w_{1}, . . ., w_{n}}^{u_{1}, . . ., u_{n}} \in L P_{n}^{*}

). Throughout this section, fix a

D_{G}

-valued random variable

a \in (W^{*} (G), E) .

Then the

D_{G}

-valued random variable

a

has the following Fourier expansion,

a = \sum v \in V (G : a) p_{v} L_{v} + \sum l \in F P_{*} (G : a) (p_{l} L_{l} + p_{l^{t}} L_{l}) + \sum w \in F P_{*}^{c} (G : a), u_{w} \in {1, *} p_{w} L_{w}^{u_{w}} .

Let's observe the new

D_{G}

-valued random variable

d_{1} a d_{2} a . . . d_{n} a \in (W^{*} (G), E),

where

d_{1}, . . ., d_{n} \in D_{G}

and

a \in W^{*} (G)

is given. Put

d_{j} = \sum v_{j} \in V (G : d_{j}) q_{v_{j}} L_{v_{j}} \in D_{G},

for

j = 1, . . ., n .

Notice that

V (G : d_{j}) = F^{+} (G : d_{j}),

since

d_{j} \in D_{G} ↪ W^{*} (G) .

Then

Proposition 1.2. (See [16]) Let

a \in (W^{*} (G), E)

be given as above. Then the

n

-th moment of

a

E (d_{1} a . . . d_{n} a) = \sum (v_{1}, . . ., v_{n}) \in Π_{j = 1}^{n} V (G : d_{j}) (Π_{j = 1}^{n} q_{v_{j}})

\sum (w_{1}, . . ., w_{n}) \in F^{+} (G : a)^{n}, u_{w_{j}} \in {1, *}, l_{w_{1}, . . ., w_{n}}^{u_{w_{1}}, . . ., u_{w_{n}}} \in L P_{n}^{*} (Π_{j = 1}^{n} p_{w_{j}})

(Π_{j = 1}^{n} δ_{(v_{j}, x_{j}, y_{j} : u_{w_{j}})}) E (L_{w_{1}}^{u_{w_{1}}} . . . L_{w_{n}}^{u_{w_{n}}}),

where

δ_{(v_{j}, x_{j}, y_{j})} = {\begin{matrix} δ_{v_{j}, x_{j}} & if u_{j} = 1 \\ δ_{v_{j}, y_{j}} & if u_{j} = * \end{matrix}

□

Let

w_{1}, . . ., w_{n} \in F P (G)

be finite paths and

u_{1}, . . ., u_{n} \in {1, *}

. Then, by the Möbius inversion, we have (1.13)

k_{n} (L_{w_{1}}^{u_{1}}, . . ., L_{w_{n}}^{u_{n}}) = \sum π \in N C (n) \hat{E} (π) (L_{w_{1}}^{u_{1}} \otimes . . . \otimes L_{w_{n}}^{u_{n}}) μ (π, 1_{n}),

where

\hat{E} = {(E^{(n)})}_{n = 1}^{\infty}

is the moment multiplicative bimodule map induced by the conditional expectation

E

(See [16]) and where

N C (n)

is the collection of all noncrossing partition over

{1, . . ., n} .

Definition 1.7. Let

N C (n)

be the set of all noncrossing partition over

{1, . . ., n}

and let

L_{w_{1}}^{u_{1}},

. . .,

L_{w_{n}}^{u_{n}} \in (W^{*} (G), E)

D_{G}

-valued random variables, where

u_{1}, . . ., u_{n} \in {1, *} .

We say that the

D_{G}

-valued random variable

L_{w_{1}}^{u_{1}} . . . L_{w_{n}}^{u_{n}}

π

-connected if the

π

-dependent

D_{G}

-moment of it is nonvanishing, for

π \in N C (n) .

In other words, the random variable

L_{w_{1}}^{u_{1}} . . . L_{w_{n}}^{u_{n}}

π

-connected, for

π \in N C (n),

\hat{E} (π) (L_{w_{1}}^{u_{1}} \otimes . . . \otimes L_{w_{n}}^{u_{n}}) \neq 0_{D_{G}} .

i.e., there exists a vertex

v \in V (G)

such that

\hat{E} (π) (L_{w_{1}}^{u_{1}} \otimes . . . \otimes L_{w_{n}}^{u_{n}}) = L_{v} .

For convenience, we will define the following subset of

N C (n)

;

Definition 1.8. Let

N C (n)

be the set of all noncrossing partitions over

{1, . . ., n}

and fix a

D_{G}

-valued random variable

L_{w_{1}}^{u_{1}} . . . L_{w_{n}}^{u_{n}}

(W^{*} (G), E),

where

u_{1},

...,

u_{n} \in {1, *} .

For the fixed

D_{G}

-valued random variable

L_{w_{1}}^{u_{1}} . . . L_{w_{n}}^{u_{n}},

define

C_{w_{1}, . . ., w_{n}}^{u_{1}, . . ., u_{n}} = d e f {π \in N C (n) : L_{w_{1}}^{u_{1}} . . . L_{w_{n}}^{u_{n}}

π

-connected

},

N C (n) .

Let

μ

be the Möbius function in the incidence algebra

I_{2} .

Define the number

μ_{w_{1}, . . ., w_{n}}^{u_{1}, . . ., u_{n}},

for the fixed

D_{G}

-valued random variable

L_{w_{1}}^{u_{1}} . . . L_{w_{n}}^{u_{n}},

μ_{w_{1}, . . ., w_{n}}^{u_{1}, . . ., u_{n}} = d e f \sum π \in C_{w_{1}, . . ., w_{n}}^{u_{1}, . . ., u_{n}} μ (π, 1_{n}) .

Assume that there exists

π \in N C (n)

such that

L_{w_{1}}^{u_{1}} . . . L_{w_{n}}^{u_{n}} = L_{v}

π

-connected.

Then

π \in C_{w_{1}, . . ., w_{n}}^{u_{1}, . . ., u_{n}}

and there exists the maximal partition

π_{0} \in C_{w_{1}, . . ., w_{n}}^{u_{1}, . . ., u_{n}}

such that

L_{w_{1}}^{u_{1}} . . . L_{w_{n}}^{u_{n}} = L_{v}

π_{0}

-connected. Notice that

1_{n} \in C_{w_{1}, . . ., w_{n}}^{u_{1}, . . ., u_{n}} .

Therefore, the maximal partition in

C_{w_{1}, . . ., w_{n}}^{u_{1}, . . ., u_{n}}

1_{n} .

Hence we have that ;

Lemma 1.3. (See [16]) Let

L_{w_{1}}^{u_{1}} . . . L_{w_{n}}^{u_{n}} \in (W^{*} (G), E)

be a

D_{G}

-valued random variable having the

*

-axis-property. Then

E (L_{w_{1}}^{u_{1}} . . . L_{w_{n}}^{u_{n}}) = \hat{E} (π) (L_{w_{1}}^{u_{1}} \otimes . . . \otimes L_{w_{n}}^{u_{n}}),

for all

π \in C_{w_{1}, . . ., w_{n}}^{u_{1}, . . ., u_{n}} .

□

By the previous lemmas, we have that

Theorem 1.4. (See [16]) Let

n \in 2 N

and let

L_{w_{1}}^{u_{1}}, . . ., L_{w_{n}}^{u_{n}} \in (W^{*} (G), E)

D_{G}

-valued random variables, where

w_{1}, . . ., w_{n} \in F P (G)

and

u_{j} \in {1, *},

j = 1, . . ., n .

Then

k_{n} (L_{w_{1}}^{u_{1}} . . . L_{w_{n}}^{u_{n}}) = μ_{w_{1}, . . ., w_{n}}^{u_{1}, . . ., u_{n}} \cdot E (L_{w_{1}}^{u_{1}}, . . ., L_{w_{n}}^{u_{n}}),

where

μ_{w_{1}, . . ., w_{n}}^{u_{1}, . . ., u_{n}} = \sum π \in C_{w_{1}, . . ., w_{n}}^{u_{1}, . . ., u_{n}} μ (π, 1_{n}) .

□

1.2 $D_{G}$ -Freeness on $(W^{*} (G), E)$

Now, we will introduce the diagram-distinctness of finite paths ;

Definition 1.9. (Diagram-Distinctness) We will say that the finite paths

w_{1}

and

w_{2}

are diagram-distinct if

w_{1}

and

w_{2}

have different diagrams in the graph

G .

Let

X_{1}

and

X_{2}

be subsets of

F P (G) .

The subsets

X_{1}

and

X_{2}

are said to be diagram-distinct if

x_{1}

and

x_{2}

are diagram-distinct for all pairs

(x_{1}, x_{2})

\in

X_{1} \times X_{2} .

In [16], we found the

D_{G}

-freeness characterization on the generator set of

W^{*} (G),

as follows ;

Theorem 1.5. (See [16]) Let

w_{1}, w_{2} \in F P (G)

be finite paths. The

D_{G}

-valued random variables

L_{w_{1}}

and

L_{w_{2}}

(W^{*} (G), E)

are free over

D_{G}

if and only if

w_{1}

and

w_{2}

are diagram-distinct.

□

Let

a

and

b

be the given

D_{G}

-valued random variables. We can get the necessary condition for the

D_{G}

-freeness of

a

and

b,

in terms of their supports. Recall that we say that the two subsets

X_{1}

and

X_{2}

F P (G)

are said to be diagram-distinct if

x_{1}

and

x_{2}

are diagram-distinct, for all pairs

(x_{1}, x_{2})

\in

X_{1}

\times

X_{2} .

Proposition 1.6. (See [16]) Let

a, b \in (W^{*} (G), E)

D_{G}

-valued random variables with their supports

F^{+} (G : a)

and

F^{+} (G : b) .

The

D_{G}

-valued random variables

a

and

b

are free over

D_{G}

(W^{*} (G), E)

F P (G : a_{1})

and

F P (G : a_{2})

are diagram-distinct.

□

2 Diagonal Compressed Random Variables in $(W^{*} (G), E)$

Let

G

be a countable directed graph and

(W^{*} (G), E),

the graph

W^{*}

-probability space over the diagonal subalgebra

D_{G} .

In [18], we observed the diagonal compressed

D_{G}

-valued free probability on

(W^{*} (G), E) .

Fix a finite subset

V = {v_{1}, . . ., v_{N}}

of the vertex set

V (G)

and define the diagonal compression

P_{V} : W^{*} (G) \to W^{*} (G)

P_{V} (a) = L_{v_{1}} a L_{v_{1}} + . . . + L_{v_{N}} a L_{v_{N}} \in W^{*} (G),

\forall a \in (W^{*} (G), E) .

If the given subset

V

is a singleton set, then we will call this diagonal compression the vertex-compression. Let

a \in (W^{*} (G), E)

be a

D_{G}

-valued random variable having its expression

a = \sum v \in V (G : a) p_{v} L_{v} + \sum l \in F P_{*} (G : a) (p_{l} L_{l} + p_{l^{t}} L_{l}^{*}) + \sum w \in F P_{*}^{c} (G : a), u_{w} \in {1, *} p_{w} L_{w}^{u_{w}} .

Then we have the following Fourier-like expression of the diagonal compressed random variable

P_{V} (a)

a

V

;

P_{V} (a) = \sum v \in V (G : P_{V} (a)) p_{v} L_{v} + \sum w \in F P_{*} (G : P_{V} (a)) (p_{w} L_{w} + p_{w^{t}} L_{w}^{*})

+ \sum w \in F P_{*}^{c} (G : P_{V} (a)), u_{w} \in {1, *} p_{w} L_{w}^{u_{w}},

with

V (G : P_{V} (a)) = V \cap V (G : a),

F P_{*} (G : P_{V} (a)) = (\cup_{j = 1}^{N} l o o p_{v_{j}} (G : a)) \cap F P_{*} (G : a)

and

F P_{*}^{c} (G : P_{V} (a)) = (\cup_{j = 1}^{N} l o o p_{v_{j}} (G : a)) \cap F P_{*}^{c} (G : a),

where

l o o p_{v_{j}} (G : a) = {l \in l o o p (G : a) : l = v_{j} l v_{j}},

for

j = 1, . . ., N .

Hence we can apply all amalgamated free probability information on this diagonal compressed case. In particular, we can compute the

D_{G}

-valued moments and

D_{G}

-valued cumulants of the diagonal compressed random variables like Section 1.3 and hence we can get

D_{G}

-valued moment series and

D_{G}

-valued R-transforms of the diagonal compressed random variables (See [18]). Also, by little modification of Section 1.5, we can get the following theorems ;

Theorem 2.1. (See [18]) Let

a, b \in (W^{*} (G), E)

be a

D_{G}

-valued random variable and let

V = {v_{1}, . . ., v_{N}}

be a finite subset of the vertex set

V (G) .

Let

P_{V} (a)

and

P_{V} (b)

be the diagonal compressed random variable of

a

and

b

V

(W^{*} (G), E),

respectively. Then (1) If

a

and

b

are free over

D_{G}

(W^{*} (G), E),

then

P_{V} (a)

and

P_{V} (b)

are free over

D_{G}

(W^{*} (G), E) .

(2) If

a

and

b

satisfy that

(V \cap V (G : a)) \cap (V \cap V (G : b)) = \emptyset

and

l o o p_{v_{i}} (G : a) \cap l o o p_{v_{j}} (G : b) = \emptyset,

for all choices

(i, j) \in {1, . . ., N}^{2},

then

P_{V} (a)

and

P_{V} (b)

are free over

D_{G}

(W^{*} (G), E) .

□

Theorem 2.2. (See [18]) Let

a \in (W^{*} (G), E)

be a

D_{G}

-valued random variable and let

V_{1} = {v_{1}^{(1)}, . . ., v_{N_{1}}^{(1)}}

and

V_{2} = {v_{1}^{(2)}, . . ., v_{N_{2}}^{(2)}}

be finite subsets of the vertex set

V (G) .

Suppose that

l o o p_{v_{i}^{(1)}} (G : a) \cap l o o p_{v_{j}^{(2)}} (G : a) = \emptyset,

for all choices

(i, j) \in {1, . . ., N_{1}} \times {1, . . ., N_{2}} .

Then the corresponding diagonal compressed random variables

P_{V_{1}} (a)

and

P_{V_{2}} (a)

a

V_{1}

and

V_{2}

are free over

D_{G}

(W^{*} (G), E) .

□

Therefore, we can again characterize the

D_{G}

-freeness of the diagonal compressed random variables in

(W^{*} (G), E)

by the subsets of the free semigroupoid

F^{+} (G)

of the graph

G .

Also, we can get the diagonal compressed R-transform calculus like in Section 1.6.

3 Off-Diagonal Random Variables in $(W^{*} (G), E)$

Throughout this chapter, let

G

be a countable directed graph and

F^{+} (G),

the free semigroupoid of

G

and let

(W^{*} (G), E)

be the graph

W^{*}

-probability space over the diagonal subalgebra

D_{G} .

In this chapter, we will consider the off-diagonal compressed random variables in

(W^{*} (G), E)

over

D_{G} .

Let's fix

v_{1} \neq v_{2}

V (G)

and let

L_{v_{1}}

and

L_{v_{2}}

be the corresponding projections in

(W^{*} (G), E) .

Define a subset

F P_{v_{1}}^{v_{2}} (G)

F P (G)

F P_{v_{1}}^{v_{2}} (G) = d e f {w \in F P (G) : w = v_{1} w v_{2}},

for

v_{1}, v_{2} \in V (G) .

Definition 3.1. Let

v_{1} \neq v_{2} \in V (G)

be given. For any

D_{G}

-valued random variable

a \in (W^{*} (G), E),

define the

(v_{1}, v_{2})

-off-diagonal compressed random variable of

a

(in short

(v_{1}, v_{2})

-compressed random variable of

a

)

_{v_{1}} a_{v_{2}}

_{v_{1}} a_{v_{2}} = d e f L_{v_{1}} a L_{v_{2}} \in (W^{*} (G), E) .

Let

a \in (W^{*} (G), E)

be an arbitrary

D_{G}

-valued random variable having the following Fourier expansion,

a = \sum v \in V (G : a) p_{v} L_{v} + \sum l \in F P_{*} (G : a) (p_{l} L_{l} + p_{l^{t}} L_{l}^{*}) + \sum w \in F P_{*}^{c} (G : a), u \in {1, *} p_{w} L_{w}^{u} .

We will denote

a_{d} = \sum v \in V (G : a) p_{v} L_{v},

a_{(*)} = \sum l \in F P_{*} (G : a) (p_{l} L_{l} + p_{l^{t}} L_{l}^{*})

and

a_{(n o n - *)} = \sum w \in F P_{*}^{c} (G : a), u \in {1, *} p_{w} L_{w}^{u},

for the given

D_{G}

-valued random variable

a .

Thus the

(v_{1}, v_{2})

-compressed random variable

_{v_{1}} a_{v_{2}}

is determined by

_{v_{1}} a_{v_{2}} = L_{v_{1}} a L_{v_{2}} = L_{v_{1}} a_{d} L_{v_{2}} + L_{v_{1}} a_{(*)} L_{v_{2}} + L_{v_{1}} a_{(n o n - *)} L_{v_{2}}

= 0_{D_{G}} + L_{v_{1}} a_{(*)} L_{v_{2}} + L_{v_{1}} a_{(n o n - *)} L_{v_{2}} .

By definition, we have the following partition of

F P (G : a),

for the given random variable

a \in

(W^{*} (G), E)

;

{l = v_{1} l v_{2} : l \in F P_{*} (G : a)} = F P_{*} (G : a) \cap F P_{v_{1}}^{v_{2}} (G)

and

{w = v_{1} w v_{2} : w \in F P_{*}^{c} (G : a)} = F P_{*}^{c} (G : a) \cap F P_{v_{1}}^{v_{2}} (G) .

So, we have that ;

Lemma 3.1. Let

a \in (W^{*} (G), E)

be a

D_{G}

-valued random variable and let

v_{1} \neq v_{2} \in V (G)

be the fixed vertices. Let

_{v_{2}} a_{v_{1}} \in (W^{*} (G), E)

be the

(v_{1}, v_{2})

-off-diagonal compressed random variable. Then

_{v_{1}} a_{v_{2}} = \sum w = v_{1} w v_{2} \in F P (G : a) p_{w} L_{w} + \sum w^{'} = v_{2} w^{'} v_{1} \in F P (G : a) p_{w^{'}} L_{w^{'}}^{*} .

Proof. By the relation that $L_{w} = L_{v w} = L_{v} L_{w},$ $L_{w} = L_{w v^{'}} = L_{w} L_{v^{'}}$ and $L_{w}^{*} = L_{w v}^{*} = L_{v} L_{w}^{*},$ $L_{w}^{*} = L_{v^{'} w}^{*} = L_{w}^{*} L_{v^{'}},$ under the weak topology ( $v, v^{'} \in V (G)$ ), we have that $L_{w} = L_{v w v^{'}} = L_{v} L_{w} L_{v^{'}}$ and $L_{w}^{*} = L_{v w v^{'}}^{*} = L_{v^{'}} L_{w}^{*} L_{v},$ whenever $w = v w v^{'}$ is a non-loop finite path, for $v, v^{'} \in V (G) .$ Thus, if $w \in F P (G : a),$ then we have that $L_{v_{1}} L_{w} L_{v_{2}} = {\begin{matrix} L_{w} & if w = v_{1} w v_{2} \\ 0_{D_{G}} & otherwise \end{matrix}$ and $L_{v_{1}} L_{w}^{*} L_{v_{2}} = {\begin{matrix} L_{w}^{*} & if w = v_{2} w v_{1} \\ 0_{D_{G}} & otherwise. \end{matrix}$ Therefore, the $(v_{1}, v_{2})$ -compressed random variable $_{v_{1}} a_{v_{2}}$ can be $_{v_{1}} a_{v_{2}} = L_{v_{1}} a L_{v_{2}} = L_{v_{1}} (\sum w \in F^{+} (G : a), u_{w} \in {1, *} p_{w} L_{w}^{u_{w}}) L_{v_{2}}$ $= L_{v_{1}} (\sum w \in F P (G : a), u_{w} \in {1, *} p_{w} L_{w}^{u_{w}}) L_{v_{2}}$ $= \sum w \in F P (G : a), u_{w} \in {1, *} p_{w} (L_{v_{1}} L_{w}^{u_{w}} L_{v_{2}})$ $= \sum w = v_{1} w v_{2} \in F P (G : a) p_{w} L_{w} + \sum w^{'} = v_{2} w^{'} v_{1} \in F P (G : a) p_{w^{'}} L_{w^{'}}^{*} .$

For the convenience, we introduce the following new notation ;

Notation Let $v_{1} \neq v_{2} \in V (G)$ be given as before and let $a \in (W^{*} (G), E)$ be a $D_{G}$ -valued random variable. Define

F P_{v_{1}}^{v_{2}} (G : a)_{*} = d e f F P_{*} (G : a) \cap F P_{v_{1}}^{v_{2}} (G)

F P_{v_{2}}^{v_{1}} (G : a)_{*}^{c} = d e f F P_{*}^{c} (G : a) \cap F P_{v_{2}}^{v_{1}} (G) .

$□$

3.1 Off-Diagonal Compressed Moments and Cumulants

Let

v_{1} \neq v_{2} \in V (G)

be the fixed vertices. In this section, we will consider the moments and cumulants of the

(v_{1}, v_{2})

-off-diagonal compressed random variables in the graph

W^{*}

-probability space

(W^{*} (G), E),

over the diagonal subalgebra

D_{G} .

We have that

_{v_{1}} a_{v_{2}} = \sum w = v_{1} w v_{2} \in F P_{v_{1}}^{v_{2}} (G : a) p_{w} L_{w} + \sum w^{'} = v_{2} w^{'} v_{1} \in F P_{v_{2}}^{v_{1}} (G : a) p_{w^{'}} L_{w^{'}}^{*} .

So, to compute the

D_{G}

-valued moments and the

D_{G}

-valued cumulants of

(v_{1}, v_{2})

-off-diagonal compressed random variables in

(W^{*} (G), E)

is to compute the

D_{G}

-valued moments and the

D_{G}

-valued cumulants of the

D_{G}

-valued random variables

x \in (W^{*} (G), E)

such that

x = \sum l_{1} = v_{1} l_{1} v_{2} \in F P (G : x) p_{l_{1}} L_{l_{1}} + \sum l_{2} = v_{2} l_{2} v_{1} \in F P (G : x) p_{l_{2}} L_{l_{2}}^{*}

(W^{*} (G), E) .

Suppose that

a \in (W^{*} (G), E)

is a

D_{G}

-valued random variable and assume that

F P_{*} (G : a) = {w_{1}, w_{2}, . . .} \subset F P (G) .

Then, in terms of

F P_{*} (G : a),

the

D_{G}

-valued random variable

a

has the following summands

p_{w_{1}} L_{w_{1}},

p_{w_{1}^{t}} L_{w_{1}}^{*},

p_{w_{2}} L_{w_{2}},

p_{w_{2}^{t}} L_{w_{2}}^{*}, . . .,

where

p_{w_{j}}, p_{w_{j}^{t}} \in C

. By the above observation, we have the following result ;

Theorem 3.2. Let

v_{1} \neq v_{2} \in V (G)

be the fixed vertices in the graph

G

and let

x \in (W^{*} (G), E)

be a

D_{G}

-valued random variable with its Fourier-like expression,

\sum l_{1} = v_{1} l_{1} v_{2} \in F P (G : x) p_{l_{1}} L_{l_{1}} + \sum l_{2} = v_{2} l_{2} v_{1} \in F P (G : x) p_{l_{2}} L_{l_{2}}^{*} .

Then the

n

-th moments and

n

-th cumulants of

x

vanish, for all

n \in N .

Proof. (1) Let $n = 1 .$ Then the first moments and the first cumulants of the $D_{G}$ -valued random variable $x$ vanish ; $E (x) = k_{1} (x) = 0_{D_{G}},$ since $V (G : x) = \emptyset .$ (2) Let $n > 1$ in $N .$ Then the $n$ -th $D_{G}$ -valued moments vanish ; By Section 1.4, we have that the $n$ -th moment of the $D_{G}$ -valued random variable $x$ is $E (d_{1} x . . . d_{n} x)$ $= \sum π \in N C (n) \sum (v^{(1)}, . . ., v^{(n)}) \in Π_{j = 1}^{n} V (G : d_{j}) (Π_{j = 1}^{n} q_{v^{(j)}})$ $\sum (w_{1}, . . ., w_{n}) \in F P (G : x)^{n}, w_{j} = x_{j} w_{j} y_{j}, u_{w_{j}} \in {1, *}, l_{w_{1}, . . ., w_{n}}^{u_{1}, . . ., u_{n}} \in L P_{n}^{*} (Π_{j = 1}^{n} p_{w_{j}})$ $(Π_{j = 1}^{n} δ_{(v^{(j)}, x_{j}, y_{j} : u_{w_{j}})}) E (L_{w_{1}}^{u_{1}} . . . L_{w_{n}}^{u_{n}}),$ where $x_{j}, y_{j} \in {v_{1}, v_{2}}$ and where $d_{j} = \sum v^{(j)} \in V (G : d_{j}) q_{v^{(j)}} L_{v^{(j)}} \in D_{G}$ are arbitrary, $j = 1, . . ., n,$ for all $n \in N,$ and where $L P_{n}^{*}$ is the lattice path model satisfying the $*$ -axis-property (See Section 1.2). But to get the nonvanishing $n$ -th cumulant of $x$ , we need to have at least one summand of $d_{1} a . . . d_{n} a,$ $L_{w_{1}}^{u_{1}} . . . L_{w_{n}}^{u_{n}} = L_{v},$ for some $v \in V (G) .$ Equivalently, the lattice path $l_{w_{1}, . . ., w_{n}}^{u_{1}, . . ., u_{n}}$ should have the $*$ -axis-property (i.e., $l_{w_{1}, . . ., w_{n}}^{u_{1}, . . ., u_{n}} \in L P_{n}^{*}$ ). To do that, at least, we need to have the nonempty $F P_{*} (G : x) .$ But $V (G) ⊈ F^{+} (G : a)$ and $F P_{*} (G : x) = \emptyset .$ Therefore, $k_{1} (d_{1} x) = 0_{D_{G}} = E (d_{1} a)$ and $k_{n} (a, . . ., a) = k_{n} (a_{(*)}, . . ., a_{(*)}) = 0_{D_{G}} .$ Indeed, we have that $F P (G : x) \subseteq F P_{v_{1}}^{v_{2}} (G) \cup F P_{v_{2}}^{v_{1}} (G) .$ Moreover, if $w = v_{1} w v_{2} \in F P (G : x) \cap F P_{v_{1}}^{v_{2}} (G),$ then the $L_{w}$ -term of $x$ exists but the $L_{w}^{*}$ -term does not exists in the Fourier expansion of $x .$ Similarly, if $w^{'} = v_{2} w v_{1} \in F P (G : x) \cap F P_{v_{2}}^{v_{1}} (G),$ then the $L_{w^{'}}^{*}$ -term of $x$ exists but the $L_{w^{'}}$ -term does not exists in the Fourier expansion of $x .$ Therefore, each lattice path of $L_{w_{1}}^{u_{1}} . . . L_{w_{n}}^{u_{n}}$ does not have the $*$ -axis-property. Since all $n$ -th $D_{G}$ -valued cumulants of $x$ vanish, all $k$ -th $D_{G}$ -valued moments of $x$ vanish, by the Möbius inversion.

Consider the

D_{G}

-valued random variable

a = L_{v} + L_{w_{1}} + L_{w_{1}}^{*} + L_{w_{2}}^{*},

where

v \in V (G)

and

w_{1} \neq w_{2} \in F P (G),

with

w_{1} = v_{1} w_{1} v_{2}

and

w_{2} = v_{2} w_{2} v_{1} .

Then the

(v_{1}, v_{2})

-off-diagonal compressed random variable of the

D_{G}

-valued random variable

a

_{v_{1}} a_{v_{2}} = L_{w_{1}} + L_{w_{2}}^{*} .

So, we have that

F P_{*} (G :_{v_{1}} a_{v_{2}}) = \emptyset

and

F P_{*}^{c} (G :_{v_{1}} a_{v_{2}}) = {w_{1}, w_{2}} .

Therefore, the

n

-th moments and the

n

-th cumulants of

_{v_{1}} a_{v_{2}}

vanish, for all

n \in N .

3.2 Off-Diagonal Compressed $D_{G}$ -Freeness

Let

G

be a countable directed graph and

F^{+} (G),

the free semigroupoid of

G

and let

(W^{*} (G), E)

be the corresponding graph

W^{*}

-probability space over the diagonal subalgebra

D_{G} .

In this section, we will consider the

D_{G}

-freeness of the off-diagonal compressed random variables. Throughout this section, let

v_{1}, v_{2}, v_{3}

and

v_{4}

be mutually distinct vertices in

V (G) .

Since

F P_{*} (G :_{v_{1}} a_{v_{2}}) = \emptyset,

it has vanishing

D_{G}

-valued

n

-th moments and

n

-th cumulants, for all

n \in N .

Therefore, automatically, the

D_{G}

-valued moments vanish, by [12].)

Proposition 3.3. Let

a

and

b

D_{G}

-valued random variables in the graph

W^{*}

-probability space

(W^{*} (G), E)

and let

v_{1} \neq v_{2}

be the given vertices in

V (G) .

a

and

b

have the diagram-distinct supports, then the

(v_{1}, v_{2})

-off-diagonal compressed random variables

_{v_{1}} a_{v_{2}} \equiv L_{v_{1}} a L_{v_{2}}

and

_{v_{1}} b_{v_{2}} \equiv L_{v_{1}} b L_{v_{2}}

are free over

D_{G}

(W^{*} (G), E) .

Proof. By the diagram-distinctness of $F P (G : a)$ and $F P (G : b),$ $l o o p^{c} (G : a)$ and $l o o p^{c} (G : b)$ are diagram-distinct, too. Note that $F P (G :_{v_{1}} a_{v_{2}}) \subset l o o p^{c} (G : a)$ and $F P (G :_{v_{1}} b_{v_{2}}) \subset l o o p^{c} (G : b) .$ Therefore, $_{v_{1}} a_{v_{2}}$ and $_{v_{1}} b_{v_{2}}$ are free over $D_{G} .$

Now, we will consider the other case ;

Proposition 3.4. Let

v_{1}, v_{2}, v_{3}

and

v_{4}

be the mutually distinct vertices in

V (G)

and let

a \in (W^{*} (G), E)

be a

D_{G}

-valued random variable. Then the

(v_{1}, v_{2})

-off-diagonal compressed random variable

_{v_{1}} a_{v_{2}}

and the

(v_{3}, v_{4})

-off-diagonal compressed random variable

_{v_{3}} a_{v_{4}}

are free over

D_{G}

(W^{*} (G), E) .

Proof. Since $v_{1}, . . ., v_{4}$ are mutually distinct, the $(v_{1}, v_{2})$ -off diagonal compressed random variable $_{v_{1}} a_{v_{2}}$ and the $(v_{3}, v_{4})$ -off-diagonal compressed random variable $_{v_{3}} a_{v_{4}}$ have the diagram-distinct supports. Therefore, they are free over $D_{G}$ in $(W^{*} (G), E) .$

Corollary 3.5. Let

v_{1}, v_{2}, v_{3}

and

v_{4}

be the given vertices in

V (G) .

Define two subsets

W^{*} (G)_{v_{1}}^{v_{2}} = L_{v_{1}} W^{*} (G) L_{v_{2}}

and

W^{*} (G)_{v_{3}}^{v_{4}} = L_{v_{3}} W^{*} (G) L_{v_{4}},

in the graph

W^{*}

-probability space

(W^{*} (G), E) .

Then these subsets are free over

D_{G}

(W^{*} (G), E) .

□

We can regard

W^{*} (G)_{v_{i}}^{v_{j}}

as an off-diagonal block of

W^{*} (G) .

4 Compressed Free Probability on $(W^{*} (G), E)$

Throughout this chapter, let

G

be a countable directed graph and

F^{+} (G),

the free semigroupoid of the graph

G

and let

(W^{*} (G), E)

be the graph

W^{*}

-probability space over the diagonal subalgebra

D_{G} .

In this chapter, we will consider the compressed random variable

P a P

of the

D_{G}

-valued random variable

a \in (W^{*} (G), E)

by the projection

P \in W^{*} (G) .

Let

v_{1}, . . ., v_{N}

be vertices in

V (G)

and define the projection

P = L_{v_{1}} + . . . + L_{v_{N}} \in W^{*} (G) .

Then it is indeed a projection in

W^{*} (G)

. From now, fix the finite vertices

v_{1}, . . ., v_{N} \in V (G)

and the corresponding projection

P

=

L_{v_{1}}

+

. . .

+

L_{v_{N}}

W^{*} (G) .

Let

a \in (W^{*} (G), E)

be a

D_{G}

-valued random variable. Then naturally, we can construct the compressed random variable

P a P

a

P .

Then this compressed random variable is again a

D_{G}

-valued random variable in

(W^{*} (G), E) .

Notice that the diagonal-compressed random variable

P_{V} (a)

by the diagonal compression

P_{V} : W^{*} (G) \to W^{*} (G),

for the fixed vertex-subset

V = {v_{1}, . . ., v_{N}}

(See Chapter 2 and [18]) and the compressed random variable

P a P

are totally different in

(W^{*} (G), E) .

For example, if

a

is a

D_{G}

-valued random variable in

(W^{*} (G), E),

then the diagonal compressed random variable is

P_{V} (a) = d e f L_{v_{1}} a L_{v_{1}} + . . . + L_{v_{N}} a L_{v_{N}}

but the compressed random variable by the projection

P

P a P = (\sum_{i = 1}^{N} L_{v_{j}}) a (\sum_{j = 1}^{N} L_{v_{j}}) = \sum (i, j) \in {1, . . ., N}^{2} L_{v_{i}} a L_{v_{j}} .

Therefore, we can say that the compressed random variable

P a P

a

P

satisfies that

P a P = P_{V} (a) + \sum (i, j) \in {1, . . ., N}^{2}, i \neq j L_{v_{i}} a L_{v_{j}} .

However, in this chapter, we will observe that

P_{V} (a)

and

P a P

have the same free probability information.

Again, remark that the compressed random variable

P a P

is the sum of diagonal compressed random variable

P_{V} (a)

and the (sum of

D_{G}

-free) off-diagonal compressed random variables

L_{v_{i}} a L_{v_{j}}

(

i \neq j

{1, . . ., N}

). Define

P_{V}^{c} (a) = \sum (i, j) \in {1, . . ., N}^{2}, i \neq j L_{v_{i}} a L_{v_{j}} .

Proposition 4.1. Let

a \in (W^{*} (G), E)

be a

D_{G}

-valued random variable and let

P = \sum_{j = 1}^{N} L_{v_{j}} \in W^{*} (G)

be a projection, where

V =

{v_{1}, . . ., v_{N}}

\subset

V (G)

is the finite subset of

V (G) .

Then the compressed random variable

P a P

a

P

P a P = P_{V} (a) + P_{V}^{c} (a),

where

P_{V} (a)

is the diagonal compressed random variable of

a

V

and

P_{V}^{c} (a) = \sum (i, j) \in {1, . . ., N}^{2}, i \neq j L_{v_{i}} a L_{v_{j}},

In particular,

P_{V} (a)

and

P_{V}^{c} (a)

are free over

D_{G}

(W^{*} (G), E) .

Proof. By the previous discussion, the compressed random variable of $a$ by the projection $P$ satisfies that $P a P = P_{V} (a) + P_{V}^{c} (a),$ where $P_{V} (a) = L_{v_{1}} a L_{v_{1}} + . . . + L_{v_{N}} a L_{v_{N}}$ and $P_{V}^{c} (a) = \sum (i, j) \in {1, . . ., N}^{2}, i \neq j L_{v_{i}} a L_{v_{j}} .$ Then, by [18] and by Chapter 3, $F^{+} (G : P_{V} (a)) \subset (V \cup l o o p (G))$ and $F^{+} (G : P_{V}^{c} (a)) \subset l o o p^{c} (G) .$ Therefore, the supports of $P_{V} (a)$ and $P_{V}^{c} (a)$ are diagram-distinct and hence they are free over $D_{G} .$

4.1 Amalgamated Moments and Cumulants of Compressed Random Variables

Remark that the compressed random variable

P a P

of a

D_{G}

-valued random variable

a

\in

(W^{*} (G), E)

by the projection

P =

\sum_{j = 1}^{N} L_{v_{j}}

\in

W^{*} (G)

has the form of

P a P = P_{V} (a) + P_{V}^{c} (a),

where

V = {v_{1}, . . ., v_{N}} \subset V (G)

is the finite subset. Futhermore, by the previous proposition, as

D_{G}

-valued random variables in

(W^{*} (G), E),

the diagonal compressed part

P_{V} (a)

a

and the off-diagonal compressed part

P_{V}^{c} (a)

are free over

D_{G}

(W^{*} (G), E) .

Therefore, we can get the following result ;

Theorem 4.2. Let

V = {v_{1}, . . ., v_{N}}

be the finite subset of the vertex set

V (G)

and let

P = \sum_{j = 1}^{N} L_{v_{j}} \in W^{*} (G)

be the corresponding projection. Let

a \in (W^{*} (G), E)

be a

D_{G}

-valued random variable and

P a P,

the compressed random variable of

a

P .

Then the

n

-th cumulants of

P a P

k_{1} (d_{1} P a P) = \sum v \in V \cap (V (G : d_{1}) \cap V (G : a)) (q_{v} p_{v}) L_{v}

and

k_{n} (\underset{︸}{d_{1} P a P, . . . ., d_{n} P a P} n - t i m e s)

= \sum (v^{(1)}, . . ., v^{(n)}) \in Π_{k = 1}^{n} V (G : d_{k}) (Π_{j = 1}^{n} q_{v^{(j)}})

\sum (w_{1}, . . ., w_{n}) \in {((\cup_{k = 1}^{n} l o o p_{v_{k}} (G : a)) \cup V \cap V (G : a) 3)}^{n}, w_{j} = x_{j} w_{j} x_{j}, l_{w_{1}, . . ., w_{n}}^{u_{w_{1}}, . . ., u_{w_{n}}} \in L P_{n}^{*}

(Π_{j = 1}^{n} p_{w_{j}}) (Π_{k = 1}^{n} δ_{v^{(k)}, x_{k}}) μ_{w_{1}, . . ., w_{n}}^{u_{w_{1}}, . . ., u_{w_{n}}} E (L_{w_{1}}^{u_{w_{1}}} . . . L_{w_{n}}^{u_{w_{n}}}),

for all

n > 1

N,

where

d_{k} = \sum v^{(k)} \in V (G : d_{k}) q_{v^{(k)}} L_{v^{(k)}} \in D_{G}

are arbitrary for

k = 1, . . ., n . .

Proof. Suppose we have the compressed random variable $P a P$ of the $D_{G}$ -valued random variable $a .$ Then $P a P = P_{V} (a) + P_{V}^{c} (a),$ where $P_{V}$ is the diagonal compression and $P_{V}^{c}$ is the off-diagonal compression by $V \subset V (G) .$ Clearly, we have the above first cumulant of $P a P .$ Also, we have that $k_{n} (d_{1} P a P, . . ., d_{n} P a P)$ $= k_{n} (d_{1} (P_{V} (a) + P_{V}^{c} (a)) + . . . + d_{n} (P_{V} (a) + P_{V}^{c} (a)))$ $= k_{n} (d_{1} P_{V} (a), . . ., d_{n} P_{V} (a))$ $+ k_{n} (d_{1} P_{V}^{c} (a), . . ., d_{n} P_{V}^{c} (a))$ by the $D_{G}$ -freeness of $P_{V} (a)$ and $P_{V}^{c} (a)$ $= k_{n} (d_{1} P_{V} (a), . . ., d_{n} P_{V} (a))$ $+ \sum (i, j) \in {1, . . ., N}^{2}, i \neq j k_{n} (d_{1} (L_{v_{i}} a L_{v_{j}}), . . ., d_{n} (L_{v_{i}} a L_{v_{j}}))$ $= k_{n} (d_{1} P_{V} (a), . . ., d_{n} P_{V} (a)) + 0_{D_{G}}$ by Section 3.1 $= k_{n} (d_{1} P_{V} (a), . . ., d_{n} P_{V} (a)),$ for all $n \in 2 N .$ Therefore, by [18], we can get the above result.

Remark 4.1. The above theorem simply shows that

k_{n} (d_{1} (P a P), . . ., d_{n} (P a P)) = k_{n} (d_{1} P_{V} (a), . . ., d_{n} P_{V} (a)),

for all

n \in N

and for any arbitrary

d_{1}, . . ., d_{n} \in D_{G} .

This says that the off-diagonal compressed part

P_{V}^{c} (a)

does not affect to compute the

D_{G}

-valued cumulants of the compressed random variable

P a P .

We can conclude that the compressed random variable

P a P

of the

D_{G}

-valued random variable by

P = \sum_{j = 1}^{N} L_{v_{j}}

and the diagonal compressed random variable

P_{V} (a)

of the random variable

a

V = {v_{1}, . . ., v_{N}}

have the same distributions and hence they have the same

D_{G}

-valued R-transforms.

4.2 $D_{G}$ -Freeness of Compressed Random Variables

In this section, we will consider the

D_{G}

-freeness of compressed random variables.

In this section, we will consider the various conditions for the

D_{G}

-freeness of compressed random variables.

Theorem 4.3. Let

a, b \in (W^{*} (G), E)

be a

D_{G}

-valued random variable and let

V = {v_{1}, . . ., v_{N}}

be a finite subset of the vertex set

V (G)

and

P = \sum_{j = 1}^{N} L_{v_{j}},

the corresponding projection in

W^{*} (G) .

Let

P a P

and

P b P

be the compressed random variable of

a

and

b

P

(W^{*} (G), E),

respectively.

a

and

b

satisfy that

(V \cap V (G : a)) \cap (V \cap V (G : b)) = \emptyset,

l o o p^{c} (G : a) \cap l o o p^{c} (G : b) = \emptyset,

and

l o o p_{v_{i}} (G : a) \cap l o o p_{v_{j}} (G : b) = \emptyset,

for all choices

(i, j) \in {1, . . ., N}^{2},

then

P a P

and

P b P

are free over

D_{G}

(W^{*} (G), E) .

Proof. By the previous section, we have that $P a P = P_{V} (a) + P_{V}^{c} (a)$ and $P b P = P_{V} (b) + P_{V}^{c} (b),$ where $P_{V}, P_{V}^{c} : W^{*} (G) \to W^{*} (G)$ are the diagonal compression and off-diagonal compression by $V,$ respectively. Moreover, $P_{V} (a)$ (resp. $P_{V} (b)$ ) and $P_{V}^{c} (a)$ (resp.
$P_{V}^{c} (b)$ ) are free over $D_{G}$ in $(W^{*} (G), E) .$ By the assumption and by [18], $P_{V} (a)$ and $P_{V} (b)$ are free over $D_{G}$ in $(W^{*} (G), E) .$ By the second condition $P_{V}^{c} (a)$ and $P_{V}^{c} (b)$ are free over $D_{G}$ in $(W^{*} (G), E) .$ This shows that ${P_{V} (a), P_{V}^{c} (a)}$ and ${P_{V} (b), P_{V}^{c} (b)}$ are free over $D_{G}$ in $(W^{*} (G), E) .$ So, $P a P$ and $P b P$ are free over $D_{G} .$

Theorem 4.4. Let

a \in (W^{*} (G), E)

be a

D_{G}

-valued random variable and let

V_{1} = {v_{1}^{(1)}, . . ., v_{N_{1}}^{(1)}}

and

V_{2} = {v_{1}^{(2)}, . . ., v_{N_{2}}^{(2)}}

be finite subsets of the vertex set

V (G) .

Let

P

and

Q

be the corresponding projections of

V_{1}

and

V_{2}

W^{*} (G),

respectively. Suppose that

(V_{1} \cap V (G : a)) \cap (V_{2} \cap V (G : a)) = \emptyset,

and

l o o p_{v_{i}^{(1)}} (G : a) \cap l o o p_{v_{j}^{(2)}} (G : a) = \emptyset,

for all choices

(i, j) \in {1, . . ., N_{1}} \times {1, . . ., N_{2}} .

Then the corresponding diagonal compressed random variables

P a P

and

Q a Q

a

P

and

Q

are free over

D_{G}

(W^{*} (G), E) .

Proof. By hypothesis and by [18], $P_{V_{1}} (a)$ and $Q_{V_{2}} (a)$ are free over $D_{G}$ in $(W^{*} (G), E),$ where $P_{V_{1}}, Q_{V_{2}} : W^{*} (G) \to W^{*} (G)$ are diagonal compressions by $V_{1}$ and $V_{2},$ respectively.

Reference [1] A. Nica, R-transform in Free Probability, IHP course note, available at www.math.uwaterloo.ca/ãnica.
[2] A. Nica and R. Speicher, R-diagonal Pair-A Common Approach to Haar Unitaries and Circular Elements, (1995), www .mast.queensu.ca/s͂peicher.
[3] B. Solel, You can see the arrows in a Quiver Operator Algebras, (2000), preprint [4] A. Nica, D. Shlyakhtenko and R. Speicher, R-cyclic Families of Matrices in Free Probability, J. of Funct Anal, 188 (2002), 227-271.
[5] D. Shlyakhtenko, Some Applications of Freeness with Amalgamation, J. Reine Angew. Math, 500 (1998), 191-212.
[6] D.Voiculescu, K. Dykemma and A. Nica, Free Random Variables, CRM Monograph Series Vol 1 (1992).
[7] D. Voiculescu, Operations on Certain Non-commuting Operator-Valued Random Variables, Astérisque, 232 (1995), 243-275.
[8] D. Shlyakhtenko, A-Valued Semicircular Systems, J. of Funct Anal, 166 (1999), 1-47.
[9] D.W. Kribs and M.T. Jury, Ideal Structure in Free Semigroupoid Algebras from Directed Graphs, preprint [10] D.W. Kribs and S.C. Power, Free Semigroupoid Algebras, preprint [11] I. Cho, Amalgamated Boxed Convolution and Amalgamated R-transform Theory, (2002), preprint.
[12] I. Cho, The Tower of Amalgamated Noncommutative Probability Spaces, (2002), Preprint. [13] I. Cho, Compatibility of a Noncommutative Probability Space and a Noncommutative Probability Space with Amalgamation, (2003), Preprint [14] I. Cho, An Example of Scalar-Valued Moments, Under Compatibility, (2003), Preprint. [15] I. Cho, Free Semigroupoid Probability Theory, (2004), Preprint. [16] I. Cho, Graph $W^{*}$ -Probability Theory, (2004), Preprint. [17] I. Cho, Random Variables in Graph $W^{*}$ -Probability Spaces, (2004), Preprint. [18] I. Cho, Diagonal Compressed Random Variables in the Graph $W^{*}$ -Probability Space, (2004), Preprint. [19] I. Cho, Free Product of Two Graph $W^{*}$ -Probability Spaces, (2004), Preprint. [20] P.Śniady and R.Speicher, Continous Family of Invariant Subspaces for R-diagonal Operators, Invent Math, 146, (2001) 329-363. [21] R. Speicher, Combinatorial Theory of the Free Product with Amalgamation and Operator-Valued Free Probability Theory, AMS Mem, Vol 132 , Num 627 , (1998).
[22] R. Speicher, Combinatorics of Free Probability Theory IHP course note, available at www.mast.queensu.ca/s͂peicher.

Univ. of Iowa, Dep. of Math, Iowa City, IA, U. S. A., E-mail address : ilcho@math.uiowa.edu

Ilwoo Cho