Nilpotent Adjacency Matrices and Random

Graphs
Rene Schott

, George Stacey Staples

October 16, 2006

Abstract
While powers of the adjacency matrix of a nite graph reveal infor-
mation about walks on the graph, they fail to distinguish closed walks
from cycles. Using elements of an appropriate commutative, nilpotent-
generated algebra, a new adjacency matrix can be associated with a
random graph on n vertices and |E| edges of nonzero probability. Let-
ting X
k
denote the number of k-cycles occurring in a random graph, this
algebra together with a probability mapping allow E(X
k
) to be recov-
ered in terms of tr A
k
. Higher moments of X
k
can also be computed,
and conditions are given for the existence of higher moments in growing
sequences of random graphs by considering innite-dimensional algebras.
The algebras used can be embedded in algebras of fermion creation and
annihilation operators, establishing connections with quantum computing
and quantum probability theory. In the framework of quantum probabil-
ity, the nilpotent adjacency matrix of a nite graph is a quantum random
variable whose m
th
moment corresponds to the m-cycles contained in the
graph.
AMS subject classication: 05C38, 05C80, 60B99, 81P68
Key words: cycles, Hamiltonian, enumeration, random graphs, quantum
computing
1 Introduction
The reader is referred to [8] for essential graph theory terminology and notation.
All graphs in this work are assumed to contain no multiple edges and no loops.
Graphs may be directed or undirected.
When working with a nite graph G on n vertices, one often utilizes the
adjacency matrix A associated with G. If the vertices are labeled 1, . . . , n,

IECN and LORIA, Universite Henri Poincare-Nancy 1, 54506 Vandoeuvre-l`es-Nancy,

France, email: schott@loria.fr

Department of Mathematics and Statistics, Southern Illinois University at Edwardsville,

Edwardsville, IL 62026-1653, email: sstaple@siue.edu
1
one denes A by
A
ij
=
_
1 if v
i
, v
j
are adjacent
0 otherwise.
(1.1)
A simple but useful result of this denition, which can also be generalized to
directed graphs, is given here without proof.
Proposition 1.1. Let G be a graph on n vertices with associated adjacency
matrix /
G
. Then for any positive integer k, the (i, j)
th
entry of /
G
k
is the
number of k-walks i j. In particular, the entries along the main diagonal of
/
G
k
are the numbers of closed k-walks in G.
What the adjacency matrix fails to provide, however, is a method of counting
self-avoiding walks and cycles in G. This problem is overcome by constructing
a nilpotent adjacency matrix.
The methods employed here are original with the authors. The technique
involves mapping combinatorial structures into algebras where self-intersections
are sieved out by multiplication. Then the remaining structures, representing
cycles and paths, are recovered by projection.
Other algebraic-probabilistic approaches to graph theory include the works
of Hashimoto, Hora, and Obata [2] and Obata [4]. Overlaps between quantum
probability and graph theory have also been discussed by Lehner [3].
1.1 Algebraic preliminaries
Let (
n
nil
denote the abelian algebra generated by the collection
i
(1 i n)
along with the scalar 1 =
0
subject to the following multiplication rules:

j
=
j

i
for i ,= j, and (1.2)

i
2
= 0 for 1 i n. (1.3)
A general element (
n
nil
can be expanded as
=

iP([n])

i
, (1.4)
where i T([n]) is an element of the power set of [n] = 1, 2, . . . , n used as a
multi-index,
i
R, and
i
=

.
Let (
n
idem
denote the abelian algebra generated by the collection
i
(1
i n) along with the scalar 1 =
0
subject to the following multiplication rules:

j
=
j

i
for i ,= j, and (1.5)

i
2
=
i
for 1 i n. (1.6)
It is evident that a general element (
n
idem
can also be expanded as in
(1.4).
2
The inner-product is dened by
u, v) =
_

iP([n])
u
i

i
,

jP([n])
v
j

j
_
=

iP([n])
u
i
v
i
. (1.7)
Hence, arbitrary u (
n
nil
has the canonical decomposition
u =

iP([n])

u,
i
_

i
. (1.8)
Finally, dene the double angle bracket to mean the sum of all scalar coef-
cients. That is, for u (
n
nil
,
u)) =

iP([n])
u
i
. (1.9)
1.2 Nilpotent Adjacency Matrices
Denition 1.2. Dene the nilpotent adjacency matrix associated with G by
A
ij
=
_

j
, if (v
i
, v
j
) E(G)
0, otherwise.
(1.10)
Observe that A Mat((
n
nil
, n), the algebra of n n matrices with entries in
the abelian nilpotent-generated algebra (
n
nil
.
Proposition 1.3. Let A be the nilpotent adjacency matrix of a graph G on n
vertices. For any m > 1 and i ,= j, summing the coecients of (A
m
)
ii
yields
the number of m-cycles based at v
i
occurring in G.
Proof. Proof is by induction on m. When m = 2,
_
/
2
_
ii
= (//)
ii
=
n

=1
/
i
/
i
. (1.11)
By construction of the nilpotent adjacency matrix,
/
i
1-paths v
i
v

, and (1.12)
/
i
1-path v

v
i
. (1.13)
Hence, the product of these terms corresponds to 2-cycles v
i
v
i
.
Now assuming the proposition holds for m and considering the case m+ 1,
_
/
m+1
_
ii
= (/
m
/)
ii
=
n

=1
(/
m
)
i
/
i
. (1.14)
3
Considering a general term of the sum,
(/
m
)
i
=

m-paths w
m
:v
i
v

w
m
, and (1.15)
/
i
=

1-paths w
1
:v

v
i
w
1
. (1.16)
It should then be clear that terms of the product
(/
m
)
i
/
i
(1.17)
are nonzero if and only if they correspond to m+1-paths v
i
v

v
i
. Summing
over all vertices v

gives the sum of all m+ 1-cycles based at v

i
.
Because A has entries in (
n
nil
, A
k
is identically the zero matrix for all k > n.
As a result, (I tA)
1
exists as the nite sum
n

k=0
t
k
A
k
for real parameter t,
and tr A
k
is recovered as the (
n
nil
-valued coecient of t
k
in the power series
expansion of tr(I tA)
1
.
Example 1.4. The 5-cycles contained in the randomly generated graph in Fig-
ure 1.1 are recovered by examining the trace of /
5
. Dividing by ve compensates
for the ve choices of base point and dividing by two compensates for possible
orientations.
A nilpotent adjacency matrix for random graphs is dened by attaching
edge existence probabilities to the nilpotent generators of (
n
nil
. Using this
approach, E(X
k
) is recovered from the trace of A
k
[7].
In the number of algebra multiplications required, cycle enumeration is re-
duced to matrix multiplication. Hence, the time complexity of enumerating a
graphs k-cycles requires no more than O(kn
3
) algebra multiplications. Several
NP-complete problems are moved into class P in this context[6].
However, computing higher moments of X
k
requires computing probabilities
P(X
k
= ) for 0, and the abelian nilpotent-generated algebra (
n
nil
is not
sucient for this purpose. In order to compute higher moments, it is necessary
to dene a nilpotent adjacency matrix with entries in (
n
nil
(
|E|
idem
, where
n denotes the number of vertices and [E[ denotes the number of edges in the
associated graph.
2 Cycles in random graphs
Consider a random graph G
n
= (V
n
, E
n
) on n vertices, V
n
= v
1
, . . . , v
n

and [E
n
[ edges, E
n
= (v
i
1
, v
j
1
), . . . , (v
i
|E
n
|
, v
j
|E
n
|
). Let 2 k n, and
let 1, 2 be dened by
=
_
1 if G
n
is directed or k = 2
2 otherwise.
(2.1)
4
In[50]:= NilpotentLabeledPlotGraphA

7
In[58]:= NilpotentAdjacencyMatrixA MatrixForm
Out[58]//MatrixForm=

0
2

3
0
5
0
7

1
0
3

4

5

6

7

1

2
0 0 0 0 0
0
2
0 0 0 0
7

1

2
0 0 0
6
0
0
2
0 0
5
0 0

1

2
0
4
0 0 0

In[55]:= SimplifyTrClMatrixPowerM, 525

Out[55]=
1,2,3,4,7

1,2,3,5,6

1,2,4,5,7

1,2,5,6,7
In[56]:= ScalarSum%
Out[56]= 4
Figure 1.1: A randomly generated graph on 7 vertices.
5
For each ordered pair (v
i
, v
j
) V (G
n
) V (G
n
), dene the probability of
existence of edge (v
i
, v
j
) in the graph G
n
by
p
ij
= P(v
i
, v
j
) E(G
n
). (2.2)
Let : [n] [n]
_
2

_
n
2
_
be an enumeration of ordered pairs of vertices,
excluding the diagonal. Because G
n
is assumed to contain no loops, (i, i) is
dened to be zero for all 1 i n.
Dening the random variable X
k
as the number of k-cycles occurring in the
graph, the goal is to compute E(X
k
) as well as the variance and the higher
moments.
Denition 2.1. Labeling the vertices with nilpotents and edges with idempo-
tents, the nilpotent adjacency matrix of G
n
is dened by
A
i j
=
(i,j)

j
Mat
_
(
|E
n
|
idem
(
n
nil
, n
_
(2.3)
for 1 i, j n. Here Mat
_
(
|E
n
|
idem
(
n
nil
, n
_
denotes the algebra of n n
matrices with entries in (
|E
n
|
idem
(
n
nil
.
Denition 2.2. Let u (
|E
n
|
idem
(
n
nil
for some n, [E
n
[ > 0, and dene

n
=

iP([|E
n
|])
_
_

i
p

_
_

i
(
|E
n
|
idem
. (2.4)
The
n
-evaluation of u is then dened as the nonnegative linear functional
)

n
: (
|E
n
|
idem
(
n
nil
R,
u)

n
=

iP([|E
n
|])
jP([n])
u
i j

i

j
,
n
) =

iP([n])
jP([n])
[u
i j

ni
[ , (2.5)
where
ni
denotes the product

i
p

.
If u = u
i j

i

j
for some i T([[E
n
[]) , j T([n]) where [i[ = k, and [j[ = ,
then u is referred to as a k -vector.
When k 3, tr A
k
will give k copies of each k-cycle in G
n
. In the particular
case k = 2, only two copies will be obtained because only one orientation is
possible. Let
(k, n) =
1
k
tr A
n
k
. (2.6)
Because the graph contains no multiple edges and no loops, (1, n) = 0,
and all values of k are hereby assumed to be greater than or equal to 2. Then
(k, n) represents a collection of k k-vectors associated with the edges and
vertices belonging to the k-cycles of nonzero probability in G
n
. Because the
6
edge probabilities are independent, the
n
-evaluation of each k k-vector is the
probability of existence of a k-cycle in G
n
. Further,
E(X
k
(n)) =

k-cycles
P(U
i
) = (k, n))

n
, (2.7)
where U
i
denotes the event that the i
th
k-cycle exists, X
k
(n) is the number of
k-cycles in G
n
, and (k, n))

n
denotes the
n
-evaluation of (k, n).
Now dene the map

1
: (
|E
n
|
idem
(
n
nil
(
|E
n
|
idem
(
2
|E
n
|
nil
by linear extension of

1
_

j
_
=

b
1
()
, (2.8)
where T([[E
n
[]) is a xed multi-index, j T([n]) is an arbitrary multi-index,
and b
1
: T([[E
n
[]) [2
|E
n
|
] is a one-to-one mapping from the power set to the
integers 1, 2, . . . , 2
|E
n
|
.
Dene the map

2
: (
|E
n
|
idem
(
2
|E
n
|
nil
(
|E
n
|
idem
(
4
|E
n
|
nil
by linear extension of

2
_

j
_
=

b
2
(j)
, (2.9)
where T([[E
n
[]) is a xed multi-index, j 2
[2
|E
n
|
]
is an arbitrary multi-
index, and b
2
: T
_
[2
|E
n
|
]
_
[4
|E
n
|
] is a one-to-one mapping from the power
set to the integers 1, 2, . . . , 4
|E
n
|
.
An easy realization of maps b
1
and b
2
is to think of multi-indices as binary
representations of integers. Images of multi-indices under maps b
1
and b
2
are
then the integers themselves.
It is worth noting that a graph on n vertices contains at most
_
n
k
_
(k 1)!

k-cycles. The maximum number of -tuples of k-cycles is then

__
n
k
_
(k1)!

_
, and
the maximum number of j-tuples of -tuples of k-cycles is given by
__
(
n
k
)
(k1)!

_
j
_
.
Where these quantities appear, they should be regarded in this context.
7
Proposition 2.3.
PX
k
(n) = =
_
1
( + 1)!
exp
_
_
_
_

2
_

1
_
trA
n
k
_
+1
_
( k)
+1
_
_
_
_
_

_
1
!
exp
_
_
_
_

2
_

1
_
trA
n
k
_

_
( k)

_
_
_
_
_

n
. (2.10)
Proof. Utilizing idempotency of the edges and nilpotency of the vertices, assum-
ing lexicographical ordering of multi-indices, and expanding (k, n) in terms of
the k-cycles it represents, (k, n) =
(
n
k
)
(k1)!

i=1
(k, n)
i
, one can see that

i<j
(k, n)
i
(k, n)
j
gives the collection of 2k 2k-vectors associated with edge- and vertex-sets of
pairs of k-cycles. It is further evident that

i
1
<i
2
<<i
m
P(U
i
1
U
i
2
U
i
m
) =
_

i
1
<<i
m
(k, n)
i
1
(k, n)
i
m
_

n
.
(2.11)
So the probability that G
n
contains one or more k-cycles is
P(U
1
U
(
n
k
)
(k1)!

) =

i
P(U
i
)

i
1
<i
2
P(U
i
1
U
i
2
)
+

i
1
<i
2
<i
3
P(U
i
1
U
i
2
U
i
3
) + (1)
[(
n
k
)
(k1)!

1]
U
i
1
U
i
(
n
k
)
(k1)!

=
_

i
(k, n)
i
_

i<j
(k, n)
i
(k, n)
j
_

n
+
_

i<j<
(k, n)
i
(k, n)
j
(k, n)

n

+ (1)
[(
n
k
)
(k1)!

1]
_
(k, n)
1
(k, n)
(
n
k
)
(k1)!

n
. (2.12)
Similarly, the probability that G
n
contains two or more k-cycles is computed by
dening U
i
as the event the i
th
pair of k-cycles exists.
Let (k, n)
()
i
denote the multivector representation of the edge-set associated
with the i
th
-tuple of k-cycles occurring in G
n
, and assume lexicographical
8
ordering of the multivector indices. In other words,
(k, n)
()
=

1i
1
<i
2
<<i

(
n
k
)
(k1)!

(k, n)
i
1
(k, n)
i

. (2.13)
It is now evident that if X
k
(n) denotes the number of k-cycles appearing
in G
n
, the probability that G
n
contains or more k-cycles is equal to the
probability one or more -tuples of k-cycles exist in G
n
. In other words,
PX
k
(n) =
(
(
n
k
)
(k1)!

i=1
_
(k, n)
()
i
_

i<j
(k, n)
()
i
(k, n)
()
j
_

n
+
+ (1)

(
(
n
k
)
(k1)!

)1

_
(k, n)
()
1
(k, n)
()
(
(
n
k
)
(k1)!

)
_

n
. (2.14)
Using nilpotency of multivectors associated with vertices in G
n
, one nds
(k, n)
()
=

i
1
<i
2
<i

(k, n)
i
1
(k, n)
i

=
1
!

1
((k, n))

=
1
!( k)

1
(tr A
n
k
)

. (2.15)
Similarly,

i
1
<i
2
<i
j
(k, n)
()
i
1
(k, n)
()
i
j
=
1
!j!

2
(
1
((k, n))

)
j
=
1
!j!( k)
j

2
(
1
(tr A
n
k
)

)
j
. (2.16)
Therefore,
PX
k
(n) = = PX
k
(n) PX
k
(n) + 1 , (2.17)
where
PX
k
(n) =
_

2
_

1
_
trA
n
k
_

_
!(k)

2
_

1
_
trA
n
k
_

_
2
2!(k)
2
_

n
+
+
(1)
(
(
n
k
)
(k1)!

)
!
_
(
n
k
)(k1)!

_
!(k)
(
(
n
k
)
(k1)!

)
_

2
_

1
_
trA
n
k
_

_
(
(
n
k
)
(k1)!

)
_

n
, (2.18)
9
and
PX
k
(n) +1 =
_

2
_

1
_
trA
n
k
_
+1
_
( + 1)!(k)
+1
_

2
_

1
_
trA
n
k
_
+1
_
2
2( + 1)!(k)
2(+1)
_

n
+
+
(1)
(
(
n
k
)
(k1)!
+1
)
( + 1)!
_
(
n
k
)(k1)!
+1
_
!(k)
(+1)(
(
n
k
)
(k1)!
+1
)
_

2
_

1
_
trA
n
k
_
+1
_
(
(
n
k
)
(k1)!
+1
)
_

n
.
(2.19)
Hence,
PX
k
(n) = =
(
(
n
k
)
(k1)!

j=1
(1)
j1
_

2
_

1
_
trA
n
k
_

_
j
!j!(k)
j
_

(
(
n
k
)
(k1)!
+1
)

j=1
(1)
j1
_

2
_

1
_
trA
n
k
_
+1
_
j
( + 1)!j!(k)
j(+1)
_

n
. (2.20)
Observing that
2
_

1
_
trA
n
k
_

_
j
= 0 whenever k > n, >
_
n
k
_
(k 1)! or
j >
_
(
n
k
)(k1)!

_
, one has
PX
k
(n) = =

j=0
(1)
j1
!j!
_
_
_
_
_
_

2
_

1
_
trA
n
k
_

_
j
( k)
j
_

1
+ 1
_

2
_

1
_
trA
n
k
_
+1
_
j
(k)
j(+1)
_

n
_
_
_
_
_
.
(2.21)
10
Rewriting the innite series gives
PX
k
(n) = =

j=0
(1)
j1
!j!
_

2
_

1
_
trA
n
k
_

_
j
( k)
j
_

j=0
(1)
j1
( + 1)!j!
_

2
_

1
_
trA
n
k
_
+1
_
j
(k)
j(+1)
_

n
=

j=0
(1)
j
!j!
_

2
_

1
_
trA
n
k
_

_
j
( k)
j
_

n
+

j=0
(1)
j
( + 1)!j!
_

2
_

1
_
trA
n
k
_
+1
_
j
(k)
j(+1)
_

n
=
_

1
!
exp
_
_
_
_

2
_

1
_
trA
n
k
_

_
( k)

_
_
_
_
_

n
+
_
1
( + 1)!
exp
_
_
_
_

2
_

1
_
trA
n
k
_
+1
_
( k)
+1
_
_
_
_
_

n
=
_
1
( + 1)!
exp
_
_
_
_

2
_

1
_
trA
n
k
_
+1
_
( k)
+1
_
_
_
_
_

_
1
!
exp
_
_
_
_

2
_

1
_
trA
n
k
_

_
( k)

_
_
_
_
_

n
.
(2.22)
Corollary 2.4. Let n, m > 0 be xed and let G
n
be a random graph on n-
vertices with associated nilpotent adjacency matrix A
n
. Then for k n,
E(X
k
(n)
m
) =

=1

m
( + 1)!
_
exp
_
_
_
_

2
_

1
_
trA
n
k
_
+1
_
( k)
+1
_
_
_
_
_

=1

m
!
_
exp
_
_
_
_

2
_

1
_
trA
n
k
_

_
( k)

_
_
_
_
_

n
. (2.23)
11
The variance of X
k
(n) is then given by
varX
k
(n) = E(X
k
(n)
2
) E(X
k
(n))
2
=

=1

2
( + 1)!
_
exp
_
_
_
_

2
_

1
_
trA
n
k
_
+1
_
( k)
+1
_
_
_
_
_

=1

2
!
_
exp
_
_
_
_

2
_

1
_
trA
n
k
_

_
( k)

_
_
_
_
_

_
_
1
k
trA
n
k
_

n
_
2
. (2.24)
3 Convergence of Moments
Let ( = G
n
denote a family of random graphs. For each n > 0, let G
n
denote a random graph on n vertices having [E
n
[ =
2

_
n
2
_
edges of probability
p
1
, . . . , p
2(
n
2
)
. Further assume that each G
n
is a subgraph of G
n+1
. In other
words,
v
i
V
n
v
i
V
n+1
, and
(v
i
, v
j
) V
n
V
n
(v
i
, v
j
) V
n+1
V
n+1
.
It is apparent that V
n+1
contains vertex v
n+1
, and that V
n+1
V
n+1
contains a
collection of edges (v
i
, v
n+1
) and (v
n+1
, v
i
) where 1 i n.
For each n > 0, the adjacency matrix of G
n
has entries in (
|E
n
|
idem
(
n
nil
.
Each algebra is therefore canonically embedded in the innite-dimensional al-
gebra (
idem
(
nil
, dened by
(
idem
(
nil
=

n=1
_
( 2

(
n
2
)
idem
(
n
nil
_
. (3.1)
For each n, let
n
(
|E
n
|
idem
be dened as in (2.4). Because G
n
is a subgraph
of G
n+1
for all n,
u
m
)

m
= u
n
)

n
(3.2)
holds for all n > m whenever u
m
(
E
m
idem
.
It is required that
= lim
n

n
(
idem
(
nil
12
exists. A necessary and sucient condition for existence of is
||
2
= lim
n
|
n
|
2
= lim
n

iP([|E
n
|])
[
ni
[
2
= lim
n

iP([|E
n
|])
_
_

i
p
2

_
_
< .
(3.3)
Note the use of an inner-product norm for the idempotent-generated algebra.
The denition of this norm should be clear from the canonical expansion

n
=

iP([|E
n
|])

ni

i
, (3.4)
where
ni
R are real scalar coecients.
Theorem 3.1. Let G
n
be an increasing sequence of random graphs such that
(3.2) and (3.3) are satised. Let k 2 and m 1 be xed. For each n N, let
[V (G
n
)[ = n and let A
n
denote the nilpotent adjacency matrix for G
n
. Let B
m
denote the m
th
Bell number. Suppose that > 0, N

N such that j, 0
and n
1
, n
2
> N

the following inequality is satised:

2
_

1
_
tr A
n
1
k
_

_
j
_

2
_

1
_
tr A
n
2
k
_

_
j
_

eB
m
. (3.5)
Then lim
n
E(X
k
(n)
m
) exists.
Proof. For xed ,

j
(1)
j1

m
j!!(k)
j
=

m
!
e

1
(k)

. (3.6)
Now e

1
(k)

< 1 , k, , and by Dobinskis Formula [1],

=0

m
!
= eB
m
. (3.7)
Thus,

j,
(1)
j1

m
j!!(k)
j

j,

m
j!!(k)
j
=

m
!
e

1
(k)

eB
m
. (3.8)
Now let > 0 be arbitrary and suppose N

N such that n
1
, n
2
> N

implies

2
_

1
_
tr A
n
1
k
_

_
j
_

2
_

1
_
tr A
n
2
k
_

_
j
_

eB
m
(3.9)
13
for all j, 0. Then

,j
(1)
j1

m
j!!(k)
j
_
_
_

2
_

1
_
tr A
n
1
k
_

_
j
_

2
_

1
_
tr A
n
2
k
_

_
j
_

_
_

,j

(1)
j1

m
j!!(k)
j
_
_
_

2
_

1
_
tr A
n
1
k
_

_
j
_

2
_

1
_
tr A
n
2
k
_

_
j
_

_
_

eB
m

j,

(1)
j1

m
j!!(k)
j

eB
m
eB
m
= . (3.10)
Thus, E(X
k
(n)
m
) is a Cauchy sequence, and lim
n
E(X
k
(n)
m
) exists.
If the m
th
moment of X
k
exists,
E(X
k
m
) =
lim
n

,j
(1)
j1

m
j!!
_
_
_
_
_
_

2
_

1
_
tr A
n
k
_

_
j
(k)
j
_

2
_

1
_
tr A
n
k
_
+1
_
j
( + 1)(k)
j(+1)
_

n
_
_
_
_
_
.
(3.11)
Moreover,
E(X
k
) = lim
n

1
k
trA
n
k
)

n
, (3.12)
provided the limit exists. In light of these equations, one has
varX
k
= E(X
k
2
) E(X
k
)
2
, (3.13)
provided the limits (3.12) and (3.11) exist for m = 2.
Remark 3.2. Because m is xed, the quantity eB
m
could be absorbed into the
of inequality (3.5). As stated, the theorem reveals a connection between the
m
th
Bell number and the existence of the m
th
moment.
3.1 Characterizing the Moments
Proposition 3.3. Let G = (V, E) be a random graph on n vertices. Fix k 2,
and dene the following quantities:
c

= PX
k
= (3.14)
= max
c

=0
. (3.15)
Then
lim
m
E(X
k
m
)

m
= c

. (3.16)
14
Proof. By denition of the m
th
moment of X
k
,
E(X
k
m
) =

m
PX
k
= =

m
c

. (3.17)
Let denote the maximum value of such that c

,= 0. Then
E(X
k
m
)

m
=

m
c

= c

m
c

. (3.18)
Observing that < and 0 c

1, the proof is complete.

4 Links to Quantum Computing
The algebras (
n
nil
and (
n
idem
can be constructed within fermion creator /
annihilator algebras of appropriate dimension.
The algebra (
n
nil
can be constructed within the 2n-particle fermion algebra
by writing

i
= f
i
+
f
+
n+i
. (4.1)
Here, f
i
+
denotes the i
th
fermion creation operator.
The algebra (
n
idem
can also be constructed within the 2n-particle fermion
algebra. Fix n > 0 and consider elements of the form

i
=
1
2
_
1 +
_
f
i
+f
i
+
2
_
_
f
n+i
+f
+
n+i
2
__
. (4.2)
Here, f
i
+
denotes the i
th
fermion creation operator, and f
i
denotes the i
th
fermion annihilation operator.
Direct calculation shows

i
2
=
_
_
_
_
1 +
_
f
i
+f
i
+
2
_
_
f
n+i
+f
+
n+i
2
_
2
_
_
_
_
2
=
1
4
+
1
2
_
f
i
+f
i
+
2
_
_
f
n+i
+f
+
n+i
2
_
+
1
2
_
f
i
+f
i
+
2
_
_
f
n+i
+f
+
n+i
2
_
_
f
i
+f
i
+
2
_
_
f
n+i
+f
+
n+i
2
_
=
2
_
f
i
+f
i
+
2
_
_
f
n+i
+f
+
n+i
2
_
+ 2
4
=
i
. (4.3)
Because each
i
is written using a pair (i, n+i) of fermion creation/annihilation
operator pairs and because these pairs are disjoint for i ,= j, direct calculation
also shows that
i

j
=
j

i
for i ,= j.
15
Letting T denote the innite-dimensional fermion algebra,
(
idem
(
nil
T T.
The nilpotent adjacency matrix associated with a nite graph can itself
be considered a quantum random variable whose m
th
moment corresponds to
the number of m-cycles occurring in the graph [5]. Considering sequences of
such quantum random variables associated with ascending sequences of random
graphs is a topic for further research.
References
[1] G. Dobinski, Summierung der reihe f ur m = 1, 2, 3, 4, 5, .... Grunert Archiv
(Arch. Math. Phys.) 61 (1877), 333-336.
[2] Y. Hashimoto, A. Hora, N. Obata, Central limit theorems for large graphs:
method of quantum decomposition, J. Math. Phys., 44 (2003), 71-88.
[3] F. Lehner, Cumulants in noncommutative probability theory III, Creation
and annihilation operators on Fock spaces, Inn. Dimens. Anal. Quantum
Probab. Relat. Top. 8 (2005), 407-437.
[4] N. Obata, Quantum probabilistic approach to spectral analysis of star
graphs, Interdisciplinary Information Sciences, 10 (2004), 41-52.
[5] R. Schott, G.S. Staples, Cliord algebras, random graphs, and quantum
random variables, presented at 27
th
International Conference on Quantum
Probability, Nottingham, England, July, 2006.
[6] R. Schott, G.S. Staples, How can NP problems be moved into P?, Preprint,
2005. http://www.siue.edu/sstaple/index les/NPtoP.pdf.
[7] G.S. Staples, A new adjacency matrix for nite graphs, submitted to Pro-
ceedings of the 7
th
International Conference on Cliord Algebras and Their
Applications, Toulouose, France, May, 2005.
[8] D. West, Introduction to Graph Theory, Second Ed., Prentice Hall, Upper
Saddle River, 2001.
16