INTEGERS: ELECTRONIC JOURNAL OF COMBINATORIAL NUMBER THEORY 6 (2006), #A12

LINEAR RECURRENT SEQUENCES AND POWERS OF A SQUARE

MATRIX
Hac`ene Belbachir
1
USTHB, Department of Mathematics, P.O.Box 32 El Alia, 16111, Bab Ezzouar, Algiers, Algeria
hbelbachir@usthb.dz and hacenebelbachir@gmail.com
Farid Bencherif
2
USTHB, Department of Mathematics, P.O.Box 32 El Alia, 16111, Bab Ezzouar, Algiers, Algeria
fbencherif@usthb.dz and fbencherif@yahoo.fr
Received: 8/10/05, Revised: 3/20/06, Accepted: 4/4/06, Published: 4/24/06
Abstract
In this paper, we establish a formula expressing explicitly the general term of a linear recur-
rent sequence, allowing us to generalize the original result of J. McLaughlin [7] concerning
powers of a matrix of size 2, to the case of a square matrix of size m 2. Identities concerning
Fibonacci and Stirling numbers and various combinatorial relations are derived.
1. Introduction
The main theorem of J. McLaughlin [7] states the following:
Theorem 1. Let A =
_
a b
c d
_
be a square matrix of order two, let T = a +d be its trace,
and let D = ad bc be its determinant. Let
y
n
=
n/2

i=0
_
n i
i
_
T
n2i
(D)
i
. (1)
Then, for n 1,
A
n
=
_
y
n
dy
n1
by
n1
cy
n1
y
n
ay
n1
_
. (2)
1
Partially supported by the laboratory LAID3.
2
Partially supported by the laboratory LATN.
INTEGERS: ELECTRONIC JOURNAL OF COMBINATORIAL NUMBER THEORY 6 (2006), #A12 2
We remark that in this theorem, (y
n
)
n1
is a linear recurrent sequence that satises
_
_
_
y
1
= 0,
y
0
= 1,
y
n
= Ty
n1
Dy
n2
for all integer n 1.
(3)
By setting A
0
=
_
1 0
0 1
_
and A
1
=
_
d b
c a
_
, relation (2) of Theorem 1 may be
written as follows:
A
n
= y
n
A
0
+y
n1
A
1
for all integers n 0, (4)
with A
0
= I
2
and A
1
= A TI
2
(where I
2
is the unit matrix).
In Section 3, we extend this result (relation (4)) to any matrix A M
m
(A) of order
m 2, A being a unitary commutative ring.
We prove the following result:
Let A M
m
(A) and let P (X) = X
m
a
1
X
m1
a
m1
X a
m
be the characteristic
polynomial of A. Let A
0
, A
1
, . . . , A
m1
be matrices of M
m
(A) dened by
A
k
=
k

i=0
a
i
A
ki
, for 0 k m1, with a
0
= 1,
and let (y
n
)
n>m
be the sequence of elements of A satisfying
y
n
=

k
1
+2k
2
++mkm=n
_
k
1
+k
2
+ +k
m
k
1
, k
2
, . . . , k
m
_
a
k
1
1
a
k
2
2
. . . a
km
m
, for n > m.
Then, for all integers n 0, A
n
= y
n
A
0
+y
n1
A
1
+ +y
nm+1
A
m1
.
The proof of this result is based on Theorem 2 given in Section 2.
In this section, we generalize Theorem 2, which permits us to express the general term
u
n
of a recurrent linear sequence satisfying the relation
u
n
= a
1
u
n1
+a
2
u
n2
+ +a
m
u
nm
for all n 1,
in terms of the coecients a
1
, a
2
, . . . , a
m
and u
0
, u
1
, u
2
, . . . , u
(m1)
. Applications to
Fibonacci, generalized Fibonacci and multibonacci sequences are also given.
Finally, in Section 4, further combinatorial identities are derived, including identities
concerning the Stirling numbers of the rst and second kind.
As an illustration, we give a nice duality between the two following relations (Corollaries
5 and 7):

k
1
+2k
2
++mkm=n
_
k
1
+ +k
m
k
1
, . . . , k
m
_
(1)
n(

m
i=1
k
i)
_
m
m1
_
k
1
. . .
_
m
0
_
km
=
_
n +m1
m1
_
,
INTEGERS: ELECTRONIC JOURNAL OF COMBINATORIAL NUMBER THEORY 6 (2006), #A12 3

k
1
+2k
2
++mkm=n
_
k
1
+ +k
m
k
1
, . . . , k
m
_
(1)
n(

m
i=1
k
i)
_
m
m1
_
k
1
. . .
_
m
0
_
km
=
_
n +m1
m1
_
,
where
_
k
1
+ +k
m
k
1
, . . . , k
m
_
is the multinomial coecient (Section 2), and
_
n
k
_
and
_
n
k
_
are, respectively, the Stirling numbers of the rst and second kind as dened in [5].
2. Explicit Expression of the General Term of a Recurrent Linear Sequence
In this section, we let m 2 be an integer, A a unitary commutative ring, a
1
, a
2
, . . . , a
m
,

0
,
1
, . . . ,
m1
elements of A, and (u
n
)
n>m
a sequence of elements of A dened by
_
u
j
=
j
for 0 j m1,
u
n
= a
1
u
n1
+a
2
u
n2
+ +a
m
u
nm
for n 1.
(5)
The aim of this section is to give an explicit expression of u
n
in terms of n, a
1
, a
2
, . . . , a
m
,

0
,
1
, . . . ,
m1
(Theorem 3).
Let us dene the sequence (y
n
)
nZ
of elements of A, with the convention that an empty
sum is zero, by
y
n
=

k
1
+2k
2
++mkm=n
_
k
1
+k
2
+ +k
m
k
1
, k
2
, . . . , k
m
_
a
k
1
1
a
k
2
2
. . . a
km
m
, for n Z, (6)
the summation being taken over all m-tuples (k
1
, k
2
, . . . , k
m
) of integers k
j
0 satisfying
k
1
+ 2k
2
+ + mk
m
= n. With the previous convention, we have y
n
= 0 for n < 0. The
multinomial coecient that appears in the summation is dened for integers k
1
, k
2
, . . . ,
k
m
0, by
_
k
1
+k
2
+ +k
m
k
1
, k
2
, . . . , k
m
_
=
(k
1
+k
2
+ +k
m
)!
k
1
!k
2
! k
m
!
,
and can always be written as a product of binomial coecients
_
k
1
+k
2
+ +k
m
k
1
, k
2
, . . . , k
m
_
=
_
k
1
+k
2
+ +k
m
k
1
+k
2
+ +k
m1
_

_
k
1
+k
2
+k
3
k
1
+k
2
__
k
1
+k
2
k
1
_
.
Let us adopt the following convention. For k
1
+k
2
+ +k
m
1, we put
_
k
1
+k
2
+ + (k
j
1) + +k
m
k
1
, k
2
, . . . , k
j
1, . . . , k
m
_
= 0 when k
j
= 0,
for any j {1, 2, . . . , m} . We can now state the following lemma [p. 80 (Vol. 1), 4].
Lemma 1. Let k
j
0 be integers for j {1, 2, . . . , m} , such that k
1
+ +k
m
1. Then
_
k
1
+k
2
+ +k
m
k
1
, k
2
, . . . , k
m
_
=
m

j=1
_
k
1
+ + (k
j
1) + +k
m
k
1
, . . . , k
j
1, . . . , k
m
_
.
INTEGERS: ELECTRONIC JOURNAL OF COMBINATORIAL NUMBER THEORY 6 (2006), #A12 4
This lemma permits us to easily prove the following result.
Lemma 2. The sequence (y
n
)
nZ
, dened by relation (6) , satises the recurrence relation
y
n
= a
1
y
n1
+a
2
y
n2
+ +a
m
y
nm
for n 1
with y
0
= 1 and y
n
= 0 for n < 0.
Proof. First notice that, for n 1, for all j {1, 2, . . . , m} we have
a
j
y
nj
=

k
1
+2k
2
++mkm=n
_
k
1
+ + (k
j
1) + +k
m
k
1
, . . . , k
j
1, . . . , k
m
_
a
k
1
1
. . . a
k
j
j
. . . a
km
m
.
Applying Lemma 1, we obtain
m

j=1
a
j
y
nj
= y
n
. The relations y
0
= 1 and y
n
= 0 for n < 0
follow immediately.
We can now state the following result.
Theorem 2. Let (u
n
)
n>m
the sequence of elements of A dened by
_
_
_
u
j
= 0 for 1 j m1,
u
0
= 1,
u
n
= a
1
u
n1
+a
2
u
n2
+ +a
m
u
nm
for n 1.
Then for all integers n > m,
u
n
=

k
1
+2k
2
++mkm=n
_
k
1
+k
2
+ +k
m
k
1
, k
2
, . . . , k
m
_
a
k
1
1
a
k
2
2
. . . a
km
m
.
Corollary 1. Let q 1 be an integer, a, b A, and let (v
n
)
nq
be the sequence of elements
of A dened by
_
_
_
v
j
= 0 for 1 j q,
v
0
= 1,
v
n+1
= av
n
+bv
nq
for n 0.
(7)
Then, for all n q,
v
n
=

n
q+1

k=0
_
n kq
k
_
a
nk(q+1)
b
k
, (8)
and, for all n 0,
v
n+1
+bv
nq
= 2v
n+1
av
n
=

n+1
q+1

k=0
n + 1 k (q 1)
n + 1 kq
_
n + 1 kq
k
_
a
n+1k(q+1)
b
k
. (9)
INTEGERS: ELECTRONIC JOURNAL OF COMBINATORIAL NUMBER THEORY 6 (2006), #A12 5
Proof. We deduce relation (8) directly from Theorem 2, with m = q +1, a
1
= a, a
m
= b and
a
k
= 0 for 1 < k < m. From (8) , we deduce (9) as follows:
v
n+1
+bv
nq
=

n+1
q+1

k=0
_
n + 1 kq
k
_
a
n+1k(q+1)
b
k
+

nq
q+1

k=0
_
n (k + 1) q
k
_
a
nq(q+1)k
b
k+1
=

n+1
q+1

k=0
_
n + 1 kq
k
_
a
n+1k(q+1)
b
k
+

n+1
q+1

k=1
_
n kq
k 1
_
a
n+1k(q+1)
b
k
=

n+1
q+1

k=0
_
n + 1 kq
k
_
a
n+1k(q+1)
b
k
+

n+1
q+1

k=0
k
n+1kq
_
n + 1 kq
k
_
a
n+1k(q+1)
b
k
=

n+1
q+1

k=0
n + 1 k (q 1)
n + 1 kq
_
n + 1 kq
k
_
a
n+1k(q+1)
b
k
.

We now give some applications of the above corollary.

Application 1. Let (F
n
)
n0
be the Fibonacci sequence
_
_
_
F
0
= 0,
F
1
= 1,
F
n+1
= F
n
+F
n1
for n 1.
Then, by setting
n
= F
n+1
for n 1, we see that (
n
)
n1
is also dened by
_
_
_

1
= 0,

0
= 1,

n
=
n1
+
n2
for n 1.
The application of Corollary 1 gives us that

n
= F
n+1
=
n/2

k=0
_
n k
k
_
, for n 1,
the relation given in [pp. 18-20, 12], and announced in [9] . Also
F
n
+F
n+2
=

n+1
2

k=0
n + 1
n + 1 k
_
n + 1 k
k
_
, for n 0,
and we nd the relations given in Problem 6.98 of [10] , which state that
F
2n1
+F
2n+1
=
n

k=0
2n
2n k
_
2n k
k
_
INTEGERS: ELECTRONIC JOURNAL OF COMBINATORIAL NUMBER THEORY 6 (2006), #A12 6
and
F
2n
+F
2n+2
=
n

k=0
2n + 1
2n + 1 k
_
2n + 1 k
k
_
.
Application 2. For q 1, let
_
G
(q)
n
_
n0
be the generalized Fibonacci sequence as cited in
[8] , and let
_
H
(q)
n
_
n0
be a sequence of numbers dened as follows:
_
G
(q)
0
= G
(q)
1
= = G
(q)
q
= 1,
G
(q)
n+1
= G
(q)
n
+G
(q)
nq
for n q,
and
_
H
(q)
0
= H
(q)
1
= = H
(q)
q
= 1,
H
(q)
n+1
= H
(q)
n
H
(q)
nq
for n q.
We can extend easily the above sequences to (G
n
)
nq
and (H
n
)
nq
by
_

_
G
(q)
j
= 0 for 1 j q,
G
(q)
0
= 1,
G
(q)
n+1
= G
(q)
n
+G
(q)
nq
for n 0,
and
_

_
H
(q)
j
= 0 for 1 j q,
H
(q)
0
= 1,
H
(q)
n+1
= H
(q)
n
H
(q)
nq
for n 0.
The application of Corollary 1 gives us, for n q, the relations
G
(q)
n
=

n
q+1

k=0
_
n kq
k
_
, and H
(q)
n
=

n
q+1

k=0
(1)
k
_
n kq
k
_
.
Notice that G
(1)
n
= F
n+1
=
n
, and H
(2)
n
is the integer function studied by L. Bernstein
[1] , who showed that the only zeros of H
(2)
n
are at n = 3 and n = 12. This result was treated
also by L. Carlitz [2, 3] and recently by J. McLaughlin and B. Sury [8].
The following theorem gives us an explicit formulation for u
n
in terms of n, a
1
, a
2
, . . . ,
a
m
,
0
,
1
, . . . ,
m1
, and thus generalizes Theorem 2.
Theorem 3. Let (u
n
)
n>m
be a sequence of elements of A dened by
_
u
j
=
j
for 0 j m1,
u
n
= a
1
u
n1
+a
2
u
n2
+ +a
m
u
nm
for n 1.
(10)
Let (
j
)
0jm1
and (y
n
)
n>m
be the sequences of elements of A dened by

j
=
m1

k=j
a
kj

k
, for 0 j m1, with a
0
= 1, (11)
and
y
n
=

k
1
+2k
2
++mkm=n
_
k
1
+k
2
+ +k
m
k
1
, k
2
, . . . , k
m
_
a
k
1
1
a
k
2
2
. . . a
km
m
, for n > m.
INTEGERS: ELECTRONIC JOURNAL OF COMBINATORIAL NUMBER THEORY 6 (2006), #A12 7
Then for all integer n > m, we have u
n
=
0
y
n
+
1
y
n+1
+ +
m1
y
n+m1
.
Remark. Note that (11) is equivalent to
_
_
_
_
_
_
_
_
_

2
.
.
.

m2

m1
_
_
_
_
_
_
_
_
_
=
_
_
_
_
_
_
_
_
_
1 a
1
a
2
a
m2
a
m1
0 1 a
1
a
2
a
m2
0 0 1 a
1
a
m3
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0 0 0 1 a
1
0 0 0 0 1
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_

2
.
.
.

m2

m1
_
_
_
_
_
_
_
_
_
.
or
[
0
,
1
, . . . ,
m1
]
t
= C [
0
,
1
, . . . ,
m1
]
t
, (12)
where
C = (c
ij
)
1i,jm
, with c
ij
=
_
_
_
0 if i > j,
1 if i = j,
a
ji
if i < j.
We deduce also from relations (10) and (12) the matrix equality
_
_
_
_
_
_
_
u
n
u
n+1
.
.
.
u
n+m2
u
n+m1
_
_
_
_
_
_
_
=
_
_
_
_
_
_
_
y
n
y
n+1
y
n+m2
y
n+m1
y
n+1
y
n+2
y
n+m1
y
n+m
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
y
n+m2
y
n+m1
y
n+2m4
y
n+2m3
y
n+m1
y
n+m
y
n+2m3
y
n+2m2
_
_
_
_
_
_
_

_
_
_
_
_
_
_
1 a
1
a
2
a
m1
0 1 a
1
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
. a
2
0 0 1 a
1
0 0 0 1
_
_
_
_
_
_
_
_
_
_
_
_
_
_

1
.
.
.

m2

m1
_
_
_
_
_
_
_
Proof. Let S be the A-module of sequences (v
n
)
n>m
satisfying the recurrence relation
v
n
a
1
v
n1
a
2
v
n2
a
m
v
nm
= 0 for all n 1.
Let us consider the family of sequences
_
v
(k)
n
_
n>m
, for 0 k m1, dened by
_

_
v
(0)
n
= y
n
,
v
(1)
n
= y
n+1
a
1
y
n
,
v
(2)
n
= y
n+2
a
1
y
n+1
a
2
y
n
,
.
.
.
v
(m1)
n
= y
n+m1
a
1
y
n+m2
a
2
y
n+m3
a
m1
y
n
,
INTEGERS: ELECTRONIC JOURNAL OF COMBINATORIAL NUMBER THEORY 6 (2006), #A12 8
i.e.,
v
(k)
n
= y
n+k
(a
1
y
n+k1
+a
2
y
n+k2
+ +a
k
y
n
) (13)
=
k

i=0
a
i
y
n+ki
, with a
0
= 1.
By Theorem 2, we have (y
n
)
n>m
S. Consequently, (y
n+q
)
n>m
S for q 0, and nally
we deduce that
_
v
(k)
n
_
n>m
S, for 0 k m1. (14)
It is easy to observe that we also have, for j, k {0, 1, 2, . . . , m1} ,
v
(k)
j
=
kj
=
_
1 if k = j
0 if k = j
(15)
In fact, with a
0
= 1, we can write, for j, k {0, 1, . . . , m1} ,v
(k)
j
=

k
i=0
a
i
y
j+ki
.
If k < j, then j + k i < 0 for 1 i k and v
(k)
j
= 0 (because y
q
= 0 for q < 0). If
k = j, then v
(k)
j
= v
(j)
j
=
j

i=0
a
i
y
i
= a
0
y
0
= 1. If k > j, then for r = k j < 0, we have
r 1 and
v
(k)
j
= y
r
(a
1
y
r1
+a
2
y
r2
+ +a
k
y
rk
) , and by using r k = j 0,
= y
r
(a
1
y
r1
+a
2
y
r2
+ +a
k
y
rk
+ +a
m
y
rm
)
= 0 because (y
n
)
n>m
S and r 1.
Relations (14) and (15) give easily, that for all n > m,u
n
=
m1

k=0

k
v
(k)
n
and, with (13) ,
u
n
=
m1

k=0

k
_
k

i=0
a
i
y
n+ki
_
=
m1

k=0
k

j=0
a
kj

k
y
n+j
=
m1

j=0
_
m1

k=j
a
kj

k
_
y
n+j
=
m1

j=0

j
y
n+j
,
where
j
is as dened in (11) .
INTEGERS: ELECTRONIC JOURNAL OF COMBINATORIAL NUMBER THEORY 6 (2006), #A12 9
Theorem 3 allows us to nd various formulae for the Fibonacci numbers.
Corollary 2. For all integers n 1,
F
n
=
1
2
n+2
n

k=1
n + 1 k
k
_
k
n + 1 2k
_
8
k
.
Remark. In this summation, we may, in fact, restrict the sum to those integers i ranging
between
n+1
3
and
n+1
2
, the binomial coecient of the formula being zero for the other integers.
Proof. Note that F
n
=2F
n2
+F
n3
for n 3. Denoting by (y
n
)
n2
the sequence dened by
_
_
_
y
2
= y
1
= 0,
y
0
= 1,
y
n
= 2y
n2
+y
n3
for n 1,
we see that, for n 1,F
n
= y
n1
+y
n2
. Theorem 3 allows us to state that, for n 2,
y
n
=

2k+3l=n
_
k +l
l
_
2
k
=

0tn
_
t
n 2t
_
2
3tn
,
and Corollary 2 follows by simple calculations.
The following result can also be easily deduced from Theorem 3.
Corollary 3. For all integers m 1, F
2m1
=
m/3

k=0
m+k
mk
_
mk
2k
_
2
m13k
.
Now, let us consider for q > 1, the multibonacci sequence
_

(q)
n
_
n>q
dened by
_

(q)
(q1)
= =
(q)
2
=
(q)
1
= 0,

(q)
0
= 1,

(q)
n
=
(q)
n1
+
(q)
n2
+ +
(q)
nq
for n 1,
(16)
where
(2)
n
=
n
= F
n+1
= G
(1)
n
. Theorem 3 also implies that, for all n 0,

(q)
n
=

k
1
+2k
2
++qkq=n
_
k
1
+k
2
+ +k
q
k
1
, k
2
, . . . , k
q
_
.
Thus, for q = 3, we obtain

(3)
n
=

i+2j+3k=n
_
i +j +k
i, j, k
_
=

2i+3jn
_
n i 2j
i +j
__
i +j
i
_
.
INTEGERS: ELECTRONIC JOURNAL OF COMBINATORIAL NUMBER THEORY 6 (2006), #A12 10
We deduce from (16) that
(q)
n+1
= 2
(q)
n

(q)
nq
, for n 1. Let us consider (
n
)
nq
,
the sequence dened by
n
:=
(q)
n
, for n > q and
q
= 1, which satises the recurrence
relation
n+1
= 2
n

nq
, for n 0. After applying Theorem 3, we nd that, for n 0,

n
_
=
(q)
n
_
=
0
z
n
+
1
z
n+1
+ +
q
z
n+q
= z
n
2z
n+q1
+z
n+q
,
with z
n
=

n
q+1

k=0
_
n kq
k
_
2
nk(q+1)
(1)
k
, for n q. We know, via Theorem 3 and
Corollary 1, that the sequence (z
n
)
nq
satises the recurrence relation z
n+1
2z
n
+z
nq
=
0, for n 0. This gives, for n 0,

(q)
n
= z
n
z
n1
+ (z
n+q
2z
n+q1
+z
n1
)
= z
n
z
n1
.
Applying relation (9) in Corollary 1, we can write, for n 1,

(q)
n
=

n
q+1

k=0
n k (q 1)
n kq
_
n kq
k
_
2
n1k(q+1)
(1)
k
.
Thus,

k
1
+2k
2
++qkq=n
_
k
1
+ +k
q
k
1
, . . . , k
q
_
=

n
q+1

k=0
n k (q 1)
n kq
_
n kq
k
_
2
n1k(q+1)
(1)
k
. (17)
For q = 3, we obtain

i+2j+3k=n
_
i +j +k
i, j, k
_
=

2i+3jn
_
n i 2j
i +j
__
i +j
i
_
=

n
4

k=0
n 2k
n 3k
(1)
k
_
n 3k
k
_
2
n14k
.
3. Powers of a Square Matrix of Order m
We start this section with the main result of this paper.
Theorem 4. Let A M
m
(A) and let P (X) = X
m
a
1
X
m1
a
m1
X a
m
be the
characteristic polynomial of A. Let A
0
, A
1
, . . . , A
m1
be matrices of M
m
(A) dened by
A
k
=
k

i=0
a
i
A
ki
, for 0 k m1, with a
0
= 1, (18)
INTEGERS: ELECTRONIC JOURNAL OF COMBINATORIAL NUMBER THEORY 6 (2006), #A12 11
and let (y
n
)
n>m
be the sequence of elements of A satisfying
y
n
=

k
1
+2k
2
++mkm=n
_
k
1
+k
2
+ +k
m
k
1
, k
2
, . . . , k
m
_
a
k
1
1
a
k
2
2
. . . a
km
m
, for n > m.
Then, for all integers n 0,
A
n
= y
n
A
0
+y
n1
A
1
+ +y
nm+1
A
m1
(19)
Proof. Dene, for 0 k m,
P
k
(X) =
k

i=0
a
i
X
ki
, with a
0
= 1. (20)
For 0 k m1, we have
XP
k
(X) = P
k+1
(X) +a
k+1
. (21)
Relation (20) shows that the degree of P
k
is k for 0 k m. It implies that (P
0
, P
1
, . . . , P
m1
)
is a basis of the free A-module A
m1
[X] consisting of polynomials of A[X] of degree m1.
For all n 0, the remainder R
n
of the euclidean division of X
n
by P
m
is written as
a linear combination of polynomials P
0
, P
1
, . . . , P
m1
. Then, for all n 0, there exists a
unique family (
n,k
)
0km1
such that
R
n
=
m1

k=0

n,k
P
k
. (22)
For 0 n m1, we have R
n
(X) = X
n
, where X
n
is a linear combination of P
0
, . . . , P
m1
,
and
_

0,0
= 1

n,k
= 0 for n < k m1.
(23)
Relations (21) and (22) imply that
XR
n
(X) =
m1

k=0

n,k
(P
k+1
(X) +a
k+1
)
=
m1

k=0
a
k+1

n,k
+
m1

k=1

n,k1
P
k
(X) +
n,m1
P
m
(X) .
As a consequence, the polynomial R
n+1
(X) XR
n
(X)
n,m1
P
m
(X) , of degree m1,
is divisible by P
m
(X), which is of degree m. This polynomial is thus the zero polynomial as
INTEGERS: ELECTRONIC JOURNAL OF COMBINATORIAL NUMBER THEORY 6 (2006), #A12 12
well as its components in the basis (P
0
, P
1
, . . . , P
m1
). This provides us with the following
relations:

n+1,0
=
m1

k=0
a
k+1

n,k
, for n 0, (24)

n+1,k
=
n,k1
for 1 k m1. (25)
Let us set, for all integers n Z,
z
n
=
_

n,0
for n 0,
0 for n < 0.
(26)
One checks easily that for all integers n 0 and for 0 k m1,

n,k
= z
nk
. (27)
Indeed, if n k this relation follows from (25) and (26) , and if 0 n < k m1 it follows
from (23) and (26). From (25) , (26) and (27) , we nd that
_
_
_
z
n
= 0 for n < 0,
z
0
= 1,
z
n
= a
1
z
n1
+a
2
z
n2
+ +a
m1
z
nm1
+a
m
z
nm
for n 1.
Theorem 2 implies that z
n
= y
n
, for all n Z, where
y
n
=

k
1
+2k
2
++mkm=n
_
k
1
+k
2
+ +k
m
k
1
, k
2
, . . . , k
m
_
a
k
1
1
a
k
2
2
. . . a
km
m
.
This last fact, together with the Cayley-Hamilton Theorem, (22) , (20) , (18) and (27) now
give that
A
n
= R
n
(A) =
m1

i=0

n,i
P
i
(A) =
m1

i=0
z
ni
A
i
=
m1

i=0
y
ni
A
i
.
which completes the proof of (19) .
4. Further Combinatorial Identities
Some nice combinatorial identities can be derived from Theorem 4 by considering various
particular matrices with simple forms.
Corollary 4. Let n be a positive integer. Then

k
1
+2k
2
++mkm=n
_
k
1
+ +k
m
k
1
, . . . , k
m
_
(1)
n(k
1
++km)
_
m
m1
_
k
1
. . .
_
m
0
_
km
=
_
n +m1
m1
_
,
INTEGERS: ELECTRONIC JOURNAL OF COMBINATORIAL NUMBER THEORY 6 (2006), #A12 13
where
_
n +m1
m1
_
is the number of m- combinations with repetition of nite set {1, . . . , n} ,
and the summation is taken over all m-tuples (k
1
, k
2
, . . . , k
m
) of integers k
j
0 satisfying
the relation k
1
+ 2k
2
+ +mk
m
= n.
Proof. Let J
m
be the mm Jordan matrix,
J
m
=
_
_
_
_
_
_
_
_
1 1 0 0
0 1 1
.
.
.
.
.
.
0 0
.
.
.
.
.
. 0
.
.
.
.
.
.
.
.
. 1 1
0 0 0 1
_
_
_
_
_
_
_
_
.
The characteristic polynomial of J
m
is (X 1)
m
. We also have J
n
m
=
__
n
j i
__
1i,jm
.
Applying Theorem 4 with A = J
m
, and considering the (1, m) entries of both sides of
(19), we obtain the relation
_
n
m1
_
= y
n(m1)
, which leads to the result.
From Corollary 4, we obtain the following combinatorial identities, for m = 2, 3, 4.

2in
(1)
i
_
n i
i
_
2
n2i
= n + 1,

2i+3jn
(1)
i
_
n i 2j
i +j
__
i +j
i
_
3
ni3j
=
_
n + 1
2
_

2i+3j+4kn
(1)
i+k
_
n i 2j 3k
i +j +k
__
i +j +k
i +j
__
i +j
i
_
2
2n3i4j8k
3
i
=
_
n + 3
3
_
Like J. Mc Laughlin and B. Sury [8, Corollary 6], we can also derive Corollary 4 from the
following result.
Corollary 5. ([8, Theorem 1]) Let x
1
, x
2
, . . . , x
m
be elements of the unitary commutative
ring A with s
k
=

1i
1
<i
2
<<i
k
m
x
i
1
x
i
2
x
i
k
, for 1 k m. Then, for each positive integer n,

k
1
+k
2
++km=n
x
k
1
1
. . . x
km
m
=

k
1
+2k
2
++mkm=n
_
k
1
+ +k
m
k
1
, . . . , k
m
_
(1)
nk
1
km
s
k
1
1
. . . s
km
m
,
with the summations being taken over all m-tuples (k
1
, k
2
, . . . , k
m
) of integers k
j
0 satis-
fying the relations k
1
+k
2
+ +k
m
= n for the left-hand side and k
1
+2k
2
+ +mk
m
= n
for the right-hand side.
INTEGERS: ELECTRONIC JOURNAL OF COMBINATORIAL NUMBER THEORY 6 (2006), #A12 14
Proof. Let us give a proof of this result by using Theorem 2. For n > m and 1 l m,
consider q
(l)
n
:=

k
1
+k
2
++k
l
=n
x
k
1
1
x
k
2
2
. . . x
k
l
l
, and q
n
:= q
(m)
n
. Let E :
n

n+1
be the shift
operator which acts on any sequence (
n
)
n
, and, for 1 l m, let Q
l
be the operator given
by Q
l
:= (E x
1
) (E x
2
) (E x
l
) . Notice that Q
m
= E
m
s
1
E
m1
+ + (1)
m
s
m
.
Then, for n 1 and 2 l m, we have (E x
l
) q
(l)
nl
= q
(l1)
n(l1)
. Therefore,
Q
m
_
q
(m)
nm
_
= Q
m1
_
q
(m1)
n(m1)
_
= = Q
1
_
q
(1)
n1
_
= 0.
Thus,
q
n
= s
1
q
n1
s
2
q
n2
+ + (1)
m1
s
m
q
nm
, for n 1. (28)
By the denition of q
n
, we also have
q
n
= 0 for n < 0 and q
0
= 1. (29)
Applying Theorem 2 to the sequence (q
n
)
n>m
, we obtain the result.
Corollary 4 follows immediately by setting x
i
= 1 for all i in the previous corollary.
The following theorem is an extension of a result of J. Mc Laughlin and B. Sury [8,
Theorem 3].
Theorem 5. Let K be a eld of characteristic zero, and let x
1
, x
2
, . . . , x
m
be independent
variables. Then, in K (x
1
, x
2
, . . . , x
m
) ,

k
1
++km=n
x
k
1
1
x
k
2
2
. . . x
km
m
=
m

i=1
x
n+m1
i

j=i
(x
i
x
j
)
.
Proof. For
1
,
2
, . . . ,
m
K (x
1
, x
2
, . . . , x
m
) , let V (
1
,
2
, . . . ,
m
) denote the Vander-
monde determinant dened by V (
1
,
2
, . . . ,
m
) = det
_

j1
i
_
1i,jm
. It is well known that
V (
1
,
2
, . . . ,
m
) =

1i<jm
(
j

i
) . By (28) , the sequence (q
n
)
n>m
dened by
q
n
=

k
1
+k
2
++km=n
x
k
1
1
x
k
2
2
. . . x
km
m
is a recurrent sequence with characteristic polynomial
X
m
s
1
X
m1
+ + (1)
m
s
m
= (X x
1
) (X x
2
) (X x
m
) .
This polynomial has m distincts roots x
1
, x
2
, . . . , x
m
. We deduce that there exist m elements
A
i
= A
i
(x
1
, x
2
, . . . , x
m
) K (x
1
, x
2
, . . . , x
m
) , 1 i m, such that
q
n
=
m

i=1
A
i
x
n
i
for n m.
INTEGERS: ELECTRONIC JOURNAL OF COMBINATORIAL NUMBER THEORY 6 (2006), #A12 15
The initial conditions given by (28) lead to the Cramer system
m

i=1
A
i
x
j
i
=
j,0
for 0 j
m1. The resolution of this system gives
A
i
=
V
_
1
x
1
, . . . ,
1
x
i1
, 0,
1
x
i+1
, . . . ,
1
xm
_
V
_
1
x
1
, . . . ,
1
x
i1
,
1
x
i
,
1
x
i+1
, . . . ,
1
xm
_ =
x
m1
i

j=i
(x
i
x
j
)
, for 1 i m.
This completes the proof.
Let us now give an application to Stirling numbers. For n 0, with the notations of [5] ,
the Stirling numbers of the rst kind
_
n
k
_
, and the Stirling numbers of the second kind
_
n
k
_
, can be dened by the equations
X
n
=
n

k=0
_
n
k
_
X
k
, (30)
with X
n
:=
_
1 if n = 0,
X (X + 1) (X + 2) (X +n 1) if n 1,
and
X
n
=
n

k=0
_
n
k
_
X
k
, (31)
with X
k
:=
_
1 if k = 0,
X (X 1) (X 2) (X k + 1) if k 1.
It is well known [p. 38 (Vol. 2), 4] that
_
n
k
_
=
1
k!
k

i=0
(1)
ki
_
k
i
_
i
n
. From Theorem 5
and Corollary 5, we deduce

k
1
+2k
2
++mkm=n
_
k
1
+ +k
m
k
1
, . . . , k
m
_
(1)
nk
1
km
s
k
1
1
. . . s
km
m
=
m

i=1
x
n+m1
i

j=i
(x
i
x
j
)
. (32)
If we take x
k
= (k 1) for 1 k m, we have s
k
= (1)
k
_
m
mk
_
, for 1 k m,
and relation (32) gives the following result
Corollary 6. For all positive integers m and n,

k
1
+2k
2
++mkm=n
_
k
1
+ +k
m
k
1
, . . . , k
m
_
(1)
n(k
1
++km)
_
m
m1
_
k
1
. . .
_
m
0
_
km
=
_
n +m1
m1
_
.
INTEGERS: ELECTRONIC JOURNAL OF COMBINATORIAL NUMBER THEORY 6 (2006), #A12 16
Note that Theorem 5 gives the following relation, which is stated in [p. 42, (Vol. 2), 4].
For all positive integers m and n,
_
n +m1
m1
_
=

k
1
++k
m1
=n
1
k
1
2
k
2
. . . (m1)
k
m1
.
For m = 3, 4 and 5, we obtain

2in
(1)
i
_
n i
i
_
2
i
3
n2i
= 2
n+1
1,

2i+3jn
(1)
i
_
n i 2j
i +j
__
i +j
i
_
6
n2i2j
11
i
=
3
n+2
2
n+3
+ 1
2
,

2i+3j+4kn
(1)
i+k
_
n i 2j 3k
i +j +k
__
i +j +k
i +j
__
i +j
i
_
2
n2i2jk
3
k
5
nij4k
7
i
=
4
n+3
3
n+4
+ 3 2
n+3
1
6
.
Corollary 7. Let n be a positive integer and let x and y be indeterminates. Then

2i+3j+4kn
(1)
i+k
_
n i 2j 3k
i +j +k
__
i +j +k
i +j
__
i +j
i
_
(2x + 2y)
n2i2j4k

(xy)
j+2k
_
x
2
+ 4xy +y
2
_
i
=
(n + 1) (x
n+3
y
n+3
) (n + 3) xy (x
n+1
y
n+1
)
(x y)
3
.
Corollary 8. Let n a positive integer and x, y be indeterminates. Then

2i+3j+4kn
(1)
i+k
_
n i 2j 3k
i +j +k
__
i +j +k
i +j
__
i +j
i
_

(3x +y)
n2i3j4k
x
i+2j+3k
y
k
(3x + 3y)
i
(x + 3y)
j
=
(n + 1) (n + 2) (x
n+3
y
n+3
) (n + 3) y [n(x
n+2
y
n+2
) + (n + 2) x
n+1
(x y)]
2 (x y)
3
.
Proof of Corollaries 7 and 8. It suces to put, in Corollary 4, x
1
= x
2
= x and x
3
= x
4
= y
to obtain Corollary 7; x
1
= x
2
= x
3
= x and x
4
= y to obtain Corollary 8.
Acknowledgments. The authors are grateful to the referee and would like to thank him/her
for comments which improved the quality of this paper.
INTEGERS: ELECTRONIC JOURNAL OF COMBINATORIAL NUMBER THEORY 6 (2006), #A12 17
References
[1] L. Bernstein: Zeros of the function f (n) =

i
(1)
i
_
n 2i
i
_
. J. Number Theory, 6 (1974),
264-270.
[2] L. Carlitz: Some combinatorial identities of Bernstein. SIAM J. Math. Anal., 9 (1978), no.
1, 6575.
[3] L. Carlitz: Recurrences of the third order and related combinatorial identities. Fibonacci
Quarterly, 16 (1978), no. 1, 1118.
[4] L. Comtet: Analyse combinatoire. Puf, Coll. Sup. Paris, (1970), Vol. 1 & Vol. 2.
[5] R. L. Graham, D. E. Knuth, O. Patashnik: Concrete Mathematics. Addison Wesley
Publishing Company, Inc., (1994).
[6] R. C. Johnson: Matrix methods for Fibonacci and related sequences. www.dur.ac.uk/ bob.johnson/
bonacci/, (2004).
[7] J. Mc Laughlin: Combinatorial identities deriving from the n-th power of 22 matrix. Elec-
tronic Journal of Combinatorial Number Theory, A19, 4 (2004).
[8] J. Mc Laughlin, B. Sury: Powers of matrix and combinatorial identities. Electronic Journal
of Combinatorial Number Theory, A13, 5 (2005).
[9] Rajesh Ram: Fibonacci number formulae. http://users.tellurian.net/hsejar/maths/ bonacci/
(2003).
[10] Z. F. Starc: Solution of the Problem 6.98. Univ. Beograd. Publikac. Elektrotehn. Fak. Ser.
Mat. 13 (2002), 103-105.
[11] S. Tanny, M. Zuker: Analytic methods applied to a sequence of binomial coeents. Discrete
Math., 24 (1978), 299-310.
[12] N. N. Vorobev: Fibonacci numbers. Translated from the russian by Halina Moss, Translation
editor Ian N. Sneddon Blaisdell Publishing Co., New York-London, (1961).