Bol. Soc. Mat. Mexicana (3) Vol.

11, 2005

THE LINEAR ALGEBRA OF THE PELL MATRIX

EMRAH KILIC AND DURSUN TASCI

Abstract. In this paper we consider the construction of the Pell and sym-
metric Pell matrices. Also we discuss the linear algebra of these matrices.
As applications, we derive some interesting relations which includes the
Pell numbers by using the properties of these Pell matrices.

1. Introduction
The Pell sequence {Pn } is defined recursively by the equation
(1.1) Pn+1 = 2Pn + Pn−1
for n ≥ 2, where P1 = 1, P2 = 2. The Pell sequence is
1, 2, 5, 12, 29, 70, 169, 408, . . ..
Matrix methods are major tools in solving many problems stemming from
linear recurrence relations. As is well-known (see, e.g., [1]) the numbers of this
sequence are also generated by the matrix
 
2 1
M= ,
1 0
since by taking successive positive powers of M one can easily establish that
 
Pn+1 Pn
Mn = .
Pn Pn−1
In [4] and [3], the authors gave several basic Pell identities as follows, for
arbitrary integers a and b,
n
(1.2) Pn+a Pn+b − Pn Pn+a+b = Pa Pb (−1) ,

(1.3) P2n+1 = Pn2 + Pn+1

2
,

[(n−1)/2]  
X n
(1.4) Pn = 2r .
r=0
2r + 1
These identities occurs as Problems B-136 [8], B-155 [11] and B-161 [5], respec-
tively.
Now we define a new matrix. The n × n Pell matrix Hn = [hij ] is defined as

Pi−j+1 , i − j + 1 ≥ 0,
Hn = [hij ] =
0, i − j + 1 < 0.

2000 Mathematics Subject Classification: 05A19, 11B37, 15A18, 15A42.

Keywords and phrases: Pell matrix, symmetric Pell matrix, majorization.

1
2 EMRAH KILIC AND DURSUN TASCI

For example,  
1 0 0 0 0 0

 2 1 0 0 0 0 

 5 2 1 0 0 0 
H6 =  ,

 12 5 2 1 0 0 

 29 12 5 2 1 0 
70 29 12 5 2 1
and the first column of H6 is the vector (1, 2, 5, 12, 29, 70)T . Thus, the matrix
Hn is useful to find the consecutive Pell numbers from the first to the nth Pell
number.
The set of all n−square matrices is denoted by An . Any matrix B ∈ An of the
form B = C t · C, C ∈ An , may be written as B = L · Lt , where L ∈ An is a lower
triangular matrix with nonnegative diagonal entries. This factorization is unique
if C is nonsingular. This is called the Cholesky factorization of B. In particular,
a matrix B is positive definite if and only if there exists a nonsingular lower
triangular matrix L ∈ An with positive diagonal entries such that B = L · L∗ . If
B is a real matrix, L may be taken to be real.
A matrix D ∈ An of the form
 
D11 0 ... 0
 0 D22 . . . 0 
 
D= . .. .. .. 
 .. . . . 
0 0 . . . Dkk
Pk
in which Dii ∈ Ani , i = 1, 2, . . . , k, and i=1 ni = n, is called a block diago-
nal. Notationally, such a matrix is often indicated as D = D11 ⊕ D22 ⊕ . . . ⊕
Pk
Dkk or more briefly, ⊕ i=1 Dii ; this is called the direct sum of the matrices
D11 , D22 , . . . , Dkk .

2. Pell Identities
In this section we give some identities of the Pell numbers. We start with the
following Lemma.
Lemma (2.1). If Pn is the nth Pell number, then
2 n
(2.2) 2Pn Pn−1 + Pn−1 − Pn2 = (−1) .
Proof. We will use the induction method. If n = 1, then we have
2P1 P0 + P02 − P12 = −1.
We suppose that the equation holds for n.
Now we show that the equation holds for n + 1. Thus
2
2Pn Pn−1 + Pn−1 − Pn2 = Pn−1 (2Pn + Pn−1 ) − Pn2
= (Pn+1 − 2Pn ) Pn+1 − Pn2
which, by definition of the Pell numbers, satisfy
2
2Pn Pn−1 + Pn−1 − Pn2 = −2Pn Pn+1 − Pn2 + Pn+1
2


= − 2Pn Pn+1 + Pn2 − Pn+1
2
 THE LINEAR ALGEBRA OF THE PELL MATRIX 3

which also, by induction hypothesis, satisfy

2Pn Pn+1 + Pn2 − Pn+1
2
= (−1) (−1)n = (−1)n+1 .
So the proof is complete.
Lemma (2.3). Let Pn be the Pell number. Then
2 2
2Pn−1 Pn = Pn+1 − Pn−1 − 2Pn Pn+1 .
Proof. By considering the proof of the previous Lemma, the proof is clear.
Lemma (2.4). If Pn is the nth Pell number, then
Pn Pn+1
(2.5) P12 + P22 + . . . + Pn2 = .
2
Pi Pi+1
Proof. Let we take ai = 2 , now since
Pi Pi+1 Pi Pi−1
ai − ai−1 = −
2 2
Pi (Pi+1 − Pi )
= ,
2
by definition of the Pell numbers, we have
Pi (2Pi )
ai − ai−1 = = Pi2 .
2
Now, using the idea of “creative telescoping” [13], we conclude
n
X n
X
Pi2 = (ai − ai−1 ) = an − a1
i=2 i=2

or equivalently (P1 = 1) ,
n
X Pn Pn+1
Pi2 = an − a1 + 1 = an = .
i=1
2
The proof is complete.
Lemma (2.6). If Pn is the nth Pell number, then
P2n+1 − 2Pn+1 Pn − 1
P1 P2 + P2 P3 + . . . + Pn−1 Pn =
(2.7) 2
P2n−1 + 2Pn Pn−1 − 1
= .
2
Proof. From Lemma (2.3) we write the following equations for 1, 2, . . . , n,
2P1 P2 = P32 − P12 − 2P2 P3
2P2 P3 = P42 − P22 − 2P3 P4
2P3 P4 = P52 − P32 − 2P4 P5
..
.
2Pn−2 Pn−1 = Pn2 − Pn−2
2
− 2Pn−1 Pn
2 2
2Pn−1 Pn = Pn+1 − Pn−1 − 2Pn Pn+1 .
4 EMRAH KILIC AND DURSUN TASCI

By addition, we obtain
2 2
2 (P1 P2 + P2 P3 + . . . + Pn−1 Pn ) = Pn+1 − Pn−1 − P12 − P22 − 2Pn+1 Pn
−2 (P1 P2 + P2 P3 + . . . + Pn−1 Pn − P1 P2 ) .

If we arrange this equation by P1 = 1, P2 = 5 and equation (1.3), then we have

P2n+1 − 2Pn+1 Pn − 1
P1 P2 + P2 P3 + . . . + Pn−1 Pn = .
2
The proof is complete.

In [2], the authors gave the Cholesky factorization of the Pascal matrix. Also
in [6], the authors consider the usual Fibonacci numbers and define the Fibonacci
and symmetric Fibonacci matrices. Furthermore, the authors give the factoriza-
tions and eigenvalues of Fibonacci and symmetric Fibonacci matrices. In [7],
the authors consider the generalized Fibobacci numbers and discuss the linear
algebra of the k-Fibonacci matrix and the symmetric k-Fibonacci matrix.

3. Factorizations
In this section we consider construction and factorization of our Pell ma-
trix of order n by using the (0, 1, 2) −matrix, where a matrix said to be a
(0, 1, 2) −matrix if each of its entries are 0, 1 or 2.
Let In be the identity matrix of order n. Further, we define the n×n matrices
Ln , Hn and Ak by
   
1 0 0 1 0 0
L0 =  2 1 0  , L−1 =  0 1 0  ,
1 0 1 0 2 1

and Lk = L0 ⊕ Ik , k = 1, 2, . . . , Hn = [1] ⊕ Hn−1 , A1 = In , A2 = In−3 ⊕

L−1 , and, for k ≥ 3, Ak = In−k ⊕ Lk−3 . Then we have the following Lemma.

Lemma (3.1). Hk · Lk−3 = Hk , k ≥ 3.

Proof. For k = 3, we have H3 · L0 = H3 . From the definition of the matrix

product and familiar Pell sequence, the conclusion follows.

Considering the previous work on Pascal functional matrices, we can rewrite

L0 , L−1 as follows:

L−1 = [1] ⊕ P1,1 [1] , L0 = CP2,0 [1] ([1] ⊕ P1,0 [−1])

in which Pn,k [x] and CPn,k [x] are Pascal k−eliminated functional matrices [12].
From the definition of Ak , we know that An = Ln−3 , A1 = In , and A2 =
In−3 ⊕L−1 . The following Theorem is an immediate consequence of Lemma (3.1).

Theorem (3.2). The Pell matrix Hn can be factored by the Ak ’s as follows:

Hn = A 1 A 2 . . . A n .
 THE LINEAR ALGEBRA OF THE PELL MATRIX 5

For example
H5 = A1 A2 A3 A4 A5 = I5 (I2 ⊕ L−1 ) (I2 ⊕ L0 ) ([1] ⊕ L1 ) L2
     
1 0 0 0 0 1 0 0 0 0 1 0 0 0 0
 0 1 0 0 0   0 1 0 0 0   0 1 0 0 0 
     
= 
 0 0 1 0 0 . 0 0 1 0 0 . 0 0 1
    0 0 

 0 0 0 1 0   0 0 0 1 0   0 0 2 1 0 
0 0 0 0 1 0 0 0 2 1 0 0 1 0 1
   
1 0 0 0 0 1 0 0 0 0
 0 1 0 0 0   2 1 0 0 0 
   
 0 2 1 0 0 . 1 0 1 0 0 
.   
 0 1 0 1 0   0 0 0 1 0 
0 0 0 0 1 0 0 0 0 1
 
1 0 0 0 0
 2 1 0 0 0 
 
 5 2 1 0 0 .
=  
 12 5 2 1 0 
29 12 5 2 1
We give another factorization of Hn . Let Tn = [tij ] be n × n matrix as
 
 P1 0 ... 0
 Pi , j = 1,  P2 1 ... 0 
 
tij = 1, i = j, , i.e., Tn =  . .. .. ..  .
  .. . . . 
0, otherwise
Pn 0 ... 1
The next Theorem follows by a simple calculation.

Theorem (3.3). For n ≥ 2, Hn = Tn (I1 ⊕ Tn−1 ) (I2 ⊕ Tn−2 ) . . . (In−2 ⊕ T2 ) .

We can readily find the inverse of the Pell matrix Hn . We know that
   
1 0 0 1 0 0
L−1
0 =  −2 1 0 , L−1
−1 =  0 1 0 , and L−1 −1
k = L0 ⊕ Ik .
−1 0 1 0 −2 1

Define Jk = A−1
k . Then
 
1 0
J1 = A−1
1 = I n , J2 = A−1
2 = In−3 ⊕ L−1
1 = In−2 ⊕ , and Jn = L−1
n−3 .
−2 1
Also, we know that
 
P1 0 0 . . . 0
 −P2 1 0 . . . 0 
 
Tn−1 =  −P3 0 1 . . . 0
 
 and (Ik ⊕ Tn−k )−1 = Ik ⊕ Tn−k
−1
.
 .. .. .. .. 
 . . . . 
−Pn 0 0 . . . 1
Thus the following Corollary holds.
6 EMRAH KILIC AND DURSUN TASCI

Corollary (3.4).
−1 −1 −1
Hn−1 = A−1
n An−1 . . . A2 A1 = Jn Jn−1 . . . J2 J1

= (In−2 ⊕ T2 )−1 . . . (I1 ⊕ Tn−1 )−1 Tn−1 .

From Corollary (3.4), we have
 
1 0 0 0 ... 0
 −2 1 0 0 ... 0 
 
 −1 −2 1 0 . .. 0 
 
 0 −1 −2 1 . . . 0 
(3.5) Hn−1 =  .. .. .. . . . ..
.

 . . . . .. . 
 
 .. .. .. 
 . . . 
0 . . . . . . −1 −2 1
We define a symmetric Pell matrix Qn = [qij ] as, for i, j = 1, 2, . . . , n,

i
 P
Pk2 , i = j,
qij = qji = k=1

qi,j−2 + 2qi,j−1 , i + 1 ≤ j,
in which q1,0 = 0. Then we have q1j = qj1 = Pj and q2j = qj2 = Pj+1 .
For example,
 
1 2 5 12 29 70 169
 2 5 12 29 70 169 408 
 
 5 12 30 72 174 420 1014 
 
Q7 =  12 29 72 174 420 1014 2448 .

 29 70 174 420 1015 2450 5915 
 
 70 169 420 1014 2450 5915 14280 
169 408 1014 2448 5915 14280 34476
From the definition of Qn , we arrive at the following Lemma.
 
Pj−2 P3
Lemma (3.6). For j ≥ 3, q3j = P4 Pj−3 + .
2
P3 P4
Proof. By Lemma (2.4), we have that q3,3 = P12 + P22 + P32 = ; hence
   2 
P3 P4 P1 P3 Pj−2 P3
q3,3 = = P 4 P0 + for P0 = 0. By induction, q3,j =P4 Pj−3 + .
2 2 2

We know that q3,1 = q1,3 = P3 and q3,2 = q2,3 = P4 . Also we have that
q4,1 = q1,4 , q4,2 = q2,4 and q4,3 = q3,4 . By similar argument, we have the
following Lemma.
 
Pj−3 P5
Lemma (3.7). For j ≥ 4, q4,j = P4 Pj−4 + Pj−4 P3 + .
2
From Lemmas (3.6) and (3.7), we obtain q5,1 , q5,2 , q5,3 and q5,4 . From these
results and the definition of Qn , we arrive at the following Lemma.
Pj−4 P5 P6
Lemma (3.8). For j ≥ 5, q5,j = Pj−5 P4 (1 + P3 + P5 ) + .
2
 THE LINEAR ALGEBRA OF THE PELL MATRIX 7

P5 P6
Proof. Since q5,5 = we have, by induction, q5j = Pj−5 P4 (1 + P3 + P5 )+
2
Pj−4 P5 P6
.
2
From the definition of Qn together with Lemmas (3.6), (3.7) and (3.8) we
have the following Lemma by induction on i.
Lemma (3.9). For j ≥ i ≥ 6,
Pj−i+1 Pi Pi+1
qij = Pj−i P4 (1 + P3 + P5 )+Pj−i P5 P6 +Pj−i P6 P7 +. . .+Pj−i Pi−1 Pi + .
2
Considering the above lemmas, we obtain the following result.
Theorem (3.10). For n ≥ 1 a positive integer, Jn Jn−1 . . . J2 J1 Qn = HnT and
the Cholesky factorization of Qn is given by Qn = Hn HnT .
Proof. By Corollary (3.4), Jn Jn−1 . . . J2 J1 = Hn−1 . So, if we have Hn−1 Qn =
HnT , then the proof is immediately seen.
Let V = [vij ] = Hn−1 Qn . Then, by (3.5), we have following:

 Pj , if i = 1,
vij = Pj−1 , if i = 2,

−qi−2,j − 2qi−1,j + qij , otherwise.
Now we consider the case i ≥ 3. Since Qn is a symmetric matrix, −qi−2,j −
2qi−1,j + qij = −qj,i−2 − 2qj,i−1 + qji . Hence, by the definition of Qn , vij = 0 for
j + 1 ≤ i. Thus, we will prove that −qi−2,j − 2qi−1,j + qij = Pj−i+1 for j ≥ i. In
the case in which i ≤ 5, we have vij = Pj−i+1 by Lemmas (3.6), (3.7) and (3.8).
Now we suppose that j ≥ i ≥ 6. Then by Lemma (3.9) we have
vij = −qi−2,j − 2qi−1,j + qij = (Pj−i − 2Pj−i+1 − Pj−i+2 ) P4 (1 + P3 + P5 ) +
(Pj−i − 2Pj−i+1 − Pj−i+2 ) P5 P6 + . . . + (Pj−i − 2Pj−i+1 − Pj−i+2 ) Pi−3 Pi−2
 
Pj−i+3
+ Pj−i − 2Pj−i+1 − Pi−2 Pi−1
2
Pi Pi+1
+ (Pj−i − Pj−i+2 ) Pi−1 Pi + Pj−i+1 .
2
Pj−i+3
Since Pj−i −2Pj−i+1 −Pj−i+2 = −4Pj−i+1 , Pj−i −2Pj−i+1 − = − 92 Pj−i+1
2
and Pj−i − Pj−i+2 = −2Pj−i+1 , we obtain
 
−4P4 − 4 (P3 P4 + P4 P5 + . . . + Pi−3 Pi−2 ) −
vij = Pj−i+1  
1 Pi Pi+1
2 Pi−2 Pi−1 − 2Pi−1 Pi + 2 .
Since P4 = 12, using Lemma (2.6) we get
   
P −2P P −1
−48 − 4 2(i−1)+1 4 i−1 i − 12−
 
vij = Pj−i+1  
Pi−2 Pi−1 Pi Pi+1
2 − 2P P
i−1 i + 2
 
Pi−2 Pi−1 Pi Pi+1
= Pj−i+1 −P2i−1 + 1 − + .
2 2
8 EMRAH KILIC AND DURSUN TASCI

Using equation (1.3) and the definition of the Pell numbers we obtain
 2

vij = Pj−i+1 −2Pi−1 − 2Pi2 + 2 − Pi−2 Pi−1 + Pi (2Pi + Pi−1 )
= Pj−i+1 .
Therefore, Hn−1 Qn = HnT , i.e., the Cholesky factorizaton of Qn is given by
Qn = Hn HnT . The proof is complete.


T −1

T
In particular, since Q−1
n = Hn Hn−1 = Hn−1 Hn−1 , we have
 
6 0 −1 0 ... ... 0
 .. 
 0 6 0 −1 . 
 
 .. 
 −1 0 6 0 . 
 
 0 −1 0 6 ... ... 0 
(3.11) Q−1
n = 
 .. .. .. 
.

 . . . 
 . ..


 6 0 −1  
 0 5 −2 
0 ... 0 . . . −1 −2 1
From Theorem (3.10), we have the following Corollary.
Corollary (3.12). If Pn is the nth Pell number and k is an odd number,
then


Pn Pn−(k−1)
− Pk /2, if n is odd,
Pn Pn−k + . . . + Pk+1 P1 =
Pn Pn−(k−1) /2, if n is even.
If k is an even number, then


Pn Pn−(k−1) /2,
if n is odd,
Pn Pn−k + . . . + Pk+1 P1 =
Pn Pn−(k−1) − Pk /2, if n is even.
For the case when we multiply the ith row of Hn and the ith column of HnT ,
we obtain the formula (2.5). Also, formula (2.5) is the case when k = 0 in
Corollary (3.12).

4. Eigenvalues of Qn
In this section we consider the eigenvalues of Qn .
Let B = {x = (x1 , x2 , . . . , xn ) ∈ Rn ; x1 ≥ x2 ≥ . . . ≥ xn } . For x, y ∈ B, x ≺ y
k
P Pk
if xi ≤ yi , k = 1, 2, . . . , n and if k = n, then equality holds. When x ≺ y,
i=1 i=1
x is said to be majorized by y, or y is said to be majorize x. The condition for
k
P k
P
majorization can be written as follows: for x, y ∈ B, x ≺ y if xn−i ≥ yn−i ,
i=0 i=0
k = 0, 1, . . . n − 2, and if k = n − 1, then equality holds.
The following is an interesting simple fact:
n
P
xi
i=1
(x, x, . . . , x) ≺ (x1 , x2 , . . . , xn ) , where x =
.
n
More interesting facts about majorizations can be found in [9] and [10].
 THE LINEAR ALGEBRA OF THE PELL MATRIX 9

An n × n matrix P = [pij ] is doubly stochastic if pij ≥ 0 for i, j = 1, 2, . . . , n,

n
P n
P
Pij = 1, j = 1, 2, . . . , n, and Pij = 1, i = 1, 2, . . . , n. In 1929, Hardy,
i=1 j=1
Littlewood and Polya proved that a necessary and sufficient condition that x ≺ y
is that there exist a doubly stochastic matrix P such that x = yP.
We know that both the eigenvalues and the main diagonal elements of real
symmetric matrix are real numbers. The precise relationship between the main
diagonal elements and the eigenvalues is given by the notion of majorization as
follows: the vector of eigenvalues of a symmetric matrix is majorized by the
diagonal elements of the matrix.
Note that det Hn = 1 and det Qn = 1. Let λ1 , λ2 , . . . , λn be the eigenvalues
Pk Pk+1 Pk
of Qn . Since Qn = Hn · HnT and Pi2 = , the eigenvalues of Qn are all
i=1 2
positive and
 
Pn+1 Pn Pn Pn−1 P2 P1
, ,..., ≺ (λ1 , λ2 , . . . , λn ) .
2 2 2
[(n−1)/2]
P

n
In [4], we find the combinatorial property, Pn = 2r+1 2r . Therefore
r=0
we have following Corollaries.
Corollary (4.1). Let λ1 , λ2 , . . . , λn be the eigenvalues of Qn . Then
  !2  

 [n/2]
P n+1
r

2r+1 2 − 1 /4 , if n is odd,

 


 r=0

λ1 + λ 2 + . . . + λ n = 

 !2  

 [n/2]
P n+1 r


2r+1 2

   /4 , if n is even.

 r=0

 
Pn+1 Pn Pn Pn−1
Proof. Since 2 , 2 , . . . , P22P1 ≺ (λ1 , λ2 , . . . , λn ) , and from Corol-
lary (3.12),
(
(Pn+1 )2 −P1
4 , if n is odd,
λ1 + λ 2 + . . . + λ n = 2
Pn+1
4 , if n is even.
By formula 1.4, the proof is immediately seen.

Corollary (4.2). If n is an odd number, then

 2
[n/2] 
X n+1
4nλn ≤  2r  − 1 ≤ 4nλ1 .
r=0
2r + 1

If n is an even number, then

 2
[n/2]  
X n + 1 r
4nλn ≤  2 ≤ 4nλ1 .
r=0
2r + 1
10 EMRAH KILIC AND DURSUN TASCI

Proof. Let Sn = λ1 + λ2 + . . . + λn . Since

 
Sn Sn Sn
, ,..., ≺ (λ1 , λ2 , . . . , λn ) ,
n n n
Sn
we have λn ≤ n ≤ λ1 . Therefore, the proof is readily seen.

From equation (3.11), we have

 
1 1 1
(4.3) (6, 6, . . . , 6, 5, 1) ≺ , ,..., .
λn λn−1 λ1
Thus there exists a doubly stochastic matrix G = [gij ] such that
 
g11 g12 ... g1n
  g g22 ... g2n 
1 1 1  21 
(6, 6, . . . , 6, 5, 1) = , ,...,  . .. .. .
λn λn−1 λ1  .. . . 
gn1 gn2 . . . gnn
1 1 1
That is, we obtain λn g1n + λn−1 g2n +. . .+ λ1 gnn = 1 and g1n +g2n +. . .+gnn = 1.
λi
Lemma (4.4). For each i = 1, 2, . . . , n, gn−(i−1),n ≤ .
n−1
λi
Proof. Suppose that gn−(i−1),n > . Then
n−1
λ1 λ2 λn
g1n + g2n + . . . + gnn > + + ...+
n−1 n−1 n−1
1
= (λ1 + λ2 + . . . + λn ) .
n−1
n
P
Since g1n + g2n + . . . + gnn = 1 and λi ≥ n, this yields a contradiction, so
i=1
λi
gn−(i−1),n ≤ .
n−1
From Lemma (4.4), we have 1−(n − 1) λ1i gn−(i−1),n ≥ 0. Let γ = Sn −(n − 1) .
Therefore, we have the following Theorem.
Theorem (4.5). For (γ, 1, 1, . . . , 1) ∈ B, (γ, 1, 1, . . . , 1) ≺ (λ1 , λ2 , . . . , λn ) .

Proof. A necessary and sufficient condition that (γ, 1, 1, . . . , 1) ≺ (λ1 , λ2 , . . . , λn )

is that there exist a doubly stochastic matrix C such that (γ, 1, 1, . . . , 1) =
(λ1 , λ2 , . . . , λn ) C.
We define an n × n matrix C = [cij ] as follows:
 
c11 c12 . . . c12
 c21 c22 . . . c22 
 
C = . .. ..  ,
 .. . . 
cn1 cn2 . . . cn2
where ci2 = λ1i gn−(i−1),n and ci1 = 1 − (n − 1) ci2 , i = 1, 2, . . . , n. Since G is
doubly stochastic and λi > 0 and ci2 ≥ 0, i = 1, 2, . . . , n. By Lemma (4.4),
 THE LINEAR ALGEBRA OF THE PELL MATRIX 11

ci1 ≥ 0, i = 1, 2, . . . , n. Then
gnn gn−1,n g1n
c12 + c22 + . . . + cn2 = + +...+ =1
λ1 λ2 λn
ci1 + (n − 1) ci2 = 1 − (n − 1) ci2 + (n − 1) ci2 = 1,
and
c11 + c21 + . . . + cn1 = 1 − (n − 1) c12 + 1 − (n − 1) c22 + . . . + 1 − (n − 1) cn2
= n − n (c12 + c22 + . . . + cn2 ) + c12 + c22 + . . . + cn2 = 1.
Thus, G is a doubly stochastic matrix. Furthermore,
gnn gn−1,n g1n
λ1 c12 + λ2 c22 + . . . + λn cn2 = λ1 + λ2 + . . . + λn
λ1 λ2 λn
= gnn + gn−1,n + . . . + g1n = 1
and
λ1 c11 + λ2 c21 + . . . + λn cn1 = λ1 (1 − (n − 1) c12 ) + . . . + λn (1 − (n − 1) cn2 )
= λ1 + λ2 + . . . + λ n −
(n − 1) (λ1 c12 + λ2 c22 + . . . + λn cn2 )
= λ1 + λ2 + . . . + λn − (n − 1) = γ.
Thus, (γ, 1, 1, . . . , 1) = (λ1 , λ2 , . . . , λn ) C, so (γ, 1, 1, . . . , 1) ≺ (λ1 , λ2 , . . . , λn ) .

From equation (4.3), we arrive at the following Lemma.

1
Lemma (4.6). For k = 2, 3, . . . , n, λk ≥ .
6 (k − 1)
Proof. From equation (4.3), for k ≥ 2,
1 1 1
+ +...+ ≤1 | + 5 + 6 {z
+ + . . . + 6} = 6 (k − 1) .
λ1 λ2 λk
k
Thus,  
1 1 1 1
≤ 6 (k − 1) − + +...+ ≤ 6 (k − 1) .
λk λ1 λ2 λk−1
1
Therefore, for k = 2, 3, . . . , n, λk ≥ . So the proof is complete.
6 (k − 1)
Acknowledgement
I wish to thank to an annonymous reviwer for his/her many valuable com-
ments, suggestions and constructive critism.
12 EMRAH KILIC AND DURSUN TASCI

Received November 19, 2004

Final version received July 18, 2005
Department of Mathematics
Gazi University
06500 Teknikokullar
Ankara, Turkey
[email protected] (corresponding author)
[email protected]

References
[1] N. Bicknell, A primer on the Pell sequence and related sequence, Fibonacci Quart. 13
(4), (1975), 345–349.
[2] R. Brawer & M. Pirovino, The linear algebra of the Pascal matrix, Linear Algebra
Appl. 174 (1992), 13–23.
[3] J. Ercolano, Matrix generator of Pell sequence, Fibonacci Quart. 17 (1), (1979), 71–77.
[4] A. F. Horadam, Pell identities, Fibonacci Quart. 9 (3), (1971), 245–252, 263.
[5] J. Ivie, Problem B-161, Fibonacci Quart. 8 (1), (1970), 107–108.
[6] G. Y. Lee, J. S. Kim, and S. G. Lee, Factorizations and eigenvalues of Fibonacci and
symmetric Fibonacci matrices, Fibonacci Quart. (40) (3), (2002), 203–211.
[7] G. Y. Lee and J. S. Kim, The linear algebra of the k-Fibonacci matrix, Linear Algebra
Appl. 373, (2003), 75–87.
[8] P. Mana, Problems B-136, B-137, Fibonacci Quart. 7 (1), (1969), 108–109.
[9] A. W. Marshall & I. Olkin, İnequalities: Theory of Majorization and Its Applications,
New York: Academic Press, 1979.
[10] D. S. Mitrinovic, Analytic Inequalities, Berlin-Heidelberg-New York, Springer-Verlag,
1970.
[11] M. N. S. Swamy and C. A. Vespe, Problem B-155, Fibonacci Quart. 7 (5), (1969), 547.
[12] H. Teimoori and M. Bayat, Pascal k-eliminated functional matrix and Eulerian num-
bers, Linear Multilinear Algebra 3, (2001).
[13] D. Zeilberger, The method of creative telescoping, J. Symbolic Comput. Vol. 11, (1991),
195–204.