Sie sind auf Seite 1von 18

Recursive sequences of the form

yn = anyn1 + yn3 with integer coefficients

Kenneth S. Berenhaut a, , Augustine B. OKeefe a ,


a Wake Forest University, Department of Mathematics, Winston-Salem, NC 27109

Abstract

This paper studies recursive sequences of the form yn = an yn1 + yn3 with positive
integral coefficients. Several properties of terms related to the coefficient sequence
are determined as well as some coefficients with maximal characteristics.

Key words: Recursive sequence, Integer coefficients, Third order, Partitions.


2000 MSC: 39A10, 11B37, 11P81, 05A17.

1 Introduction

This paper studies general linear recurrences of the form

yn = an yn1 + yn3 (n 1), (1)

with ai N for i 1, y2 = y1 = 0 and y0 = 1. These are third order


generalizations of the famous second order Fibonacci sequence, which has
been studied extensively for over eight centuries (cf. Dickson [9]).

In [7], the equation

zn = an zn1 + znm (n 1), (2)


1 The first author acknowledges financial support from a Sterge Faculty Fellowship.
Corresponding author.
Email addresses: berenhks@wfu.edu (Kenneth S. Berenhaut),
okeeab4@wfu.edu (Augustine B. OKeefe).

Preprint submitted to Indian Journal of Mathematics. 17 November 2006


with m = 2 was studied, and several properties of terms related to the coeffi-
cient sequence as well as some fixed sum results were proven. Here we prove
comparable results for the third order equation given in (1) (i.e. for m = 3
in (2)). Several new intricacies are introduced when considering m = 3 versus
m = 2 in (2).

For {yi } satisfying (1), and a given sequence {ai }, the first few yi = yi (a1 , . . . , ai )
are

y2 = 0
y1 = 0
y0 = 1
y1 = a 1
y2 = a1 a2
y3 = a1 a2 a3 + 1
y4 = a1 a2 a3 a4 + a4 + a1
y5 = a1 a2 a3 a4 a5 + a4 a5 + a1 a5 + a1 a2 . (3)

Note that the first term in yi is simply the product of the arguments a1 , a2 ,
. . . , ai .

Recurrences with varying or random coefficients have been studied by many


previous authors. For a partial survey of such literature see Viswanath [20] and
[21], Viswanath and Trefethen [22], Embree and Trefethen [10], Wright and
Trefethen [23], Mallik [14], Popenda [16], Kittapa [13], Odlyzko [15], Berenhaut
and Goedhart [4,5], Berenhaut and Morton [6], Berenhaut and Foley [3], and
Stevic [1719] (and the reference therein). For a comprehensive treatment of
difference equations and inequalities, c.f. Agarwal [1].

For x 1 and w = x + 1, by yk (wb1 , xb2 , wb3 ) we shall mean yk (a1 , a2 , . . . , ak )


where ai = x for b1 + 1 i b1 + b2 and ai = w otherwise. For example

y10 (w2 , x5 , w3 ) = y10 (w, w, x, x, x, x, x, w, w, w). (4)

Among our results, we prove the following fixed sum theorem.

Pk
Theorem 1 Suppose that a1 , a2 , . . . , ak N with i=1 ai = N . Then

yk (a1 , a2 , . . . , ak ) yk (wb1 , xb2 , wb3 ), (5)

where x = [N/k], w = x + 1 and

2




(0, k, 0), N 0 mod k





(1, k 1, 0), N 1 mod k



(b1 , b2 , b3 ) = (1, k 2, 1), N 2 mod k . (6)







(2, k 3, 1), N 3 mod k




(2, k v, v 2), N v mod k; 4 v k 1

Here the number of ws in (5) is precisely N kx, while the number of xs is


k(x + 1) N .

In the spirit of our Theorem 1 (and Theorem 1 in [7]), we phrase the following
conjecture, which is a natural extension. Recall that m is the shift parameter
in (2).

Conjecture 1 Suppose that N v mod k, 0 v k 1, m 2, {zi }


P
satisfies (2), and ki=1 ai = N . Then,

yk (a1 , a2 , . . . , ak ) yk (wb1 , xb2 , wb3 ), (7)

where x = [N/k], w = x + 1 and


v v
, k v, , if 0 v 2(m 1)
(b1 , b2 , b3 ) = 2 2 (8)
.

(m 1, k v, v (m 1)), if 2(m 1) + 1 v k 1

Here the number of ws in (7) is precisely N kx, while the number of xs is


k(x + 1) N .

The paper proceeds as follows. In Section 2 we introduce some preliminary


results and lemmas regarding properties of {(a1 , , ak )} while in Section 3
we prove Theorem 1, which highlights the extent to which the main product
Q
term aj dominates the value of yi .

2 Preliminary results and lemmas

This section contains several preliminary results and lemmas regarding prop-
P
erties of {yk (a1 , a2 , . . . , ak )} for fixed sum aj = N .

For ease of notation we will use the vector (a1 , a2 , . . . , ak ) in place of


yk (a1 , a2 , . . . , ak ).

3
Also, let us denote (m,k ) = (am , . . . , ak ) and (similar to (4)) (bk ) = (b, b, . . . , b)
when the number of bs is k. For convenience, in keeping with (3), we take
(i+1,i ) = () = 1 and (i+t,i ) = 0 for t > 1. Some of the most basic properties
of operations on k-tuples are given in the following fundamental lemma (anal-
ogous to Lemma 1 in [7]). The symmetry property in (iv) will be particularly
crucial to our investigations. It may also be noted that the property holds
for all recursive sequences satisfying (2) for any m 2. This fact as well as
variants of other components of Lemma 1 which hold true for general m (in
(2)) were indeed part of our motivation for the formulation of Conjecture 1.
Decomposition rules as in (ii) and (iii), for larger m, could be especially useful
in proving the conjecture.

Lemma 1 The following hold.

(i) For all b, c, d 1, (c) = c, (b, c) = bc and (b, c, d) = bcd + 1.


(ii) For all (m,k ), we have

(m,k ) = (m,k1 )(ak ) + (m,k3 ) (9)

(iii) For all (1,k ) and 1 m k 1,

(1,k ) = (1,m )(m+1,k ) + (1,m1 )(m+3,k ) (10)


+(1,m2 )(m+2,k ).

(iv) (Symmetry) For all (1,k ),

def
(1,k ) = (ak , ak1 , . . . , a2 , a1 ) = (1,k ). (11)

(v) (Monotonicity) The function

yk (x1 , x2 , . . . , xk ) = (x1 , x2 , . . . , xk ) (12)

is monotone in each variable xi .

Proof. The equalities in (i) and (ii) follow directly from (1), as does (v). To
see (iii), note that, for k = 2,

(1,k ) = (a1 , a2 ) = a1 a2 = (1,1 )(2,2 ) + (1,1 )(0,2 ) + (1,0 )(1,2 ). (13)

Hence, from (ii) and induction, we have

4
(1,k ) = (1,k1 )(ak ) + (1,k3 )
= (ak )[(1,m )(m+1,k1 ) + (1,m1 )(m+3,k1 ) + (1,m2 )(m+2,k1 )]
+ (1,m )(m+1,k3 ) + (1,m1 )(m+3,k3 ) + (1,m2 )(m+2,k3 )
= (1,m )[(m+1,k1 )(ak ) + (m+1,k3 )]
+ (1,m1 )[(m+3,k1 )(ak ) + (m+3,k3 )]
+ (1,m2 )[(m+2,k1 )(ak ) + (m+2,k3 )]
= (1,m )(m+1,k ) + (1,m1 )(m+3,k ) + (1,m2 )(m+2,k ). (14)

Similarly, for (iv), we have from (iii) and induction that

(1,k ) = (1,k1 )(ak ) + (1,k3 ) = (ak )(1,k1 ) + (1,k3 ) = (1,k ). (15)

Note that the final equality in (15) follows via an application of (iii). 2

Now, using the properties in Lemma 1, we prove several lemmas that will
help us to take steps towards maximizing arbitrary fixed sum k-tuples, and
eventually obtaining a proof of Theorem 1.

Lemma 2 For all positive integers k and b, c 1 satisfying b c 2

(1,k , b, c) (1,k , b + 1, c 1). (16)

Proof. Employing (iii), we have

(1,k , b + 1, c 1) (1,k , b, c) = (1,k )(b + 1, c 1) + (1,k1 ) + (1,k2 )(c 1)


(1,k )(b, c) (1,k1 ) (1,k2 )(c)
= (1,k )[bc b + c 1] + (1,k1 ) + (1,k2 )[c 1]
(1,k )[bc] (1,k1 ) (1,k2 )[c]
= (1,k )[b + c 1] (1,k2 ) 0, (17)

since (1,k ) (1,k2 ) and c b 2. 2

Lemma 3 For all positive integers k and b, c 1 satisfying b c + 1,

(1,k , b, c) (1,k , b 1, c + 1). (18)

Proof. As in the proof of Lemma 2, we have

5
(1,k , b 1, c + 1) (1,k , b, c) = (1,k )[bc + b c 1] + (1,k1 ) + (1,k2 )[c + 1]
(1,k )[bc] (1,k1 ) (1,k2 )[c]
= (1,k )[b c 1] + (1,k2 ) 0, (19)

since b c 1. 2

Lemma 4 For all positive integers 1 k m1 and b, c satisfying b c2,


we have

(1,k , b, c, k+1,m ) (1,k , b + 1, c 1, k+1,m ). (20)

Proof. Employing Lemma 1 (iii) and Lemma 2, we have

(1,k , b, c, k+1,m ) = (1,k , b, c)(k+1,m ) + (1,k , b)(k+3,m )


+(1,k )(k+2,m )
(1,k , b + 1, c 1)(k+1,m ) + (1,k , b + 1)(k+3,m )
+(1,k )(k+2,m )
= (1,k , b + 1, c 1, k+1,m ). (21)

Lemma 5 For all positive integers 1 k m and b, c 1 satisfying b c2,


we have

(c, 1,k , b) (c 1, 1,k , b + 1). (22)

Proof. We have

(c, 1,k , b) = (c, 1,k )(b) + (c, 1,k2 )


= [(c)(1,k ) + (3,k )](b) + [(c)(1,k2 ) + (3,k2 )] (23)

and

(c 1, 1,k , b + 1) = (c 1, 1,k )(b + 1) + (c 1, 1,k2 )


= [(c 1)(1,k ) + (3,k )](b + 1)
+ [(c 1)(1,k2 ) + (3,k2 )]
(24)

6
Comparing (23) and (24), gives

(c 1, 1,k , b + 1) (c, 1,k , b) = (c b 1)(1,k ) + (3,k ) (1,k2 )


(c b 1)(1,k ) (1,k2 ) 0. (25)

Lemma 6 For all positive integers b, k 1,

(1,k , b + 1, k+1,m ) = (1,k , b, k+1,m ) + (1,k )(k+1,m ). (26)

Proof. Repeated use of Lemma 1, Parts (iii) and (iii) and regrouping yields

(1,k , b + 1, k+1,m ) = (1,k , b + 1)(k+1,m ) + (1,k )(k+3,m )


+(1,k1 )(k+2,m )
= (1,k )(b + 1)(k+1,m ) + (1,k2 )(k+1,m )
+(1,k )(k+3,m ) + (1,k1 )(k+2,m )
= (1,k )(b)(k+1,m ) + (1,k )(k+1,m ) + (1,k2 )(k+1,m )
+(1,k )(k+3,m ) + (1,k1 )(k+2,m )
= (1,k , b)(k+1,m ) + (1,k )(k+3,m ) + (1,k1 )(k+2,m )
+(1,k )(k+1,m )
= (1,k , b, k+1,m ) + (1,k )(k+1,m ). (27)

Lemma 7 For all D, C 1 and x > C, we have

(1,k , x C + 1, xn , x + D 1, k+1,m ) (1,k , x C, xn , x + D, k+1,m )


(k+1,m ) [(1,k )[D + C 1](xn ) (1,k1 )(xn1 ) (1,k2 )(xn )].
(28)

In particular, if ak (D + C 1) 2, then

(1,k , x C + 1, xn , x + D 1, k+1,m ) (1,k , x C, xn , x + D, k+1,m ).


(29)

Proof. Splitting, we have

7
(1,k , x C, xn , x + D, k+1,m ) = (1,k )(x C, xn , x + D, k+1,m )
+ (1,k1 )(xn1 , x + D, k+1,m )
+ (1,k2 )(xn , x + D, k+1,m ). (30)

Meanwhile

(x C, xn , x + D, k+1,m ) = (x C, xn , x + D)(k+1,m )
+ (x C, xn )(k+3,m )
+ (x C, xn1 )(k+2,m ), (31)
(xn1 , x + D, k+1,m ) = (xn1 , x + D)(k+1,m )
+ (xn2 )(k+3,m )
+ (xn3 )(k+2,m ), and (32)
(xn , x + D, k+1,m ) = (xn , x + D)(k+1,m )
+ (xn1 )(k+3,m )
+ (xn2 )(k+2,m ). (33)

Now,

(x C, xn , x + D) = (x C)(xn , x + D) + (xn2 , x + D)
= (x C)(xn )(x + D) + (x C)(xn2 )
+ (xn2 )(x + D) + (xn4 ), (34)

and

(x C, xn ) = (x C)(xn ) + (xn2 ),
(x C, xn1 ) = (x C)(xn1 ) + (xn3 ),
(xn1 , x + D) = (xn1 )(x + D) + (xn3 ),
(xn , x + D) = (xn )(x + D) + (xn2 ). (35)

Combining the information in (30)(35) gives

8
(1,k , x C, xn , x + D, k+1,m )
= (1,k )(k+1,m )[(x C)(x + D)(xn ) + [(x + D) + (x C)](xn2 ) + (xn4 )]
+ (1,k )(k+3,m )[(x C)(xn ) + (xn2 )]
+ (1,k )(k+2,m )[(x C)(xn1 ) + (xn3 )]
+ (1,k1 )(k+1,m )[(xn1 )(x + D) + (xn3 )]
+ (1,k1 )(k+3,m )(xn2 )
+ (1,k1 )(k+2,m )(xn3 )
+ (1,k2 )(k+1,m )[(xn )(x + D) + (xn2 )]
+ (1,k2 )(k+3,m )[(xn1 )]
+ (1,k2 )(k+2,m )[(xn2 )]. (36)

Plugging in C 1 for C and D 1 for D in (36), and taking the difference


gives

(1,k , x C + 1, xn , x + D 1, k+1,m ) (1,k , x C, xn , x + D, k+1,m )


= (1,k )(k+1,m )(xn )[(x C + 1)(x + D 1) (x C)(x + D)]
+ (1,k )(k+3,m )(xn )
+ (1,k )(k+2,m )(xn1 )
(1,k1 )(k+1,m )(xn1 )
(1,k2 )(k+1,m )(xn ). (37)

The result in (28) then follows from (37), upon noting that (x C + 1)(x +
D 1) (x C)(x + D) = D + C 1. The inequality in (29) is proven upon
noting that (1,k ) ak (1,k1 ). 2

It will be useful to have available the following lemma, which follows directly
from Lemma 7 and symmetry.

Lemma 8 For all D, C 1 and x > C, we have

(1,k , x + D 1, xn , x C + 1, k+1,m ) (1,k , x + D, xn , x C, k+1,m )


(1,k ) [(k+1,m )[D + C 1](xn ) (k+2,m )(xn1 ) (k+3,m )(xn )].
(38)

Lemma 9 For all 1,k , x, c, d 1 and w = x + 1,

(xc , w, 1,k ) (w, xc , 1,k ) (39)

9
Proof. If c = 1, the result follows from Lemma 3 and symmetry. Otherwise,
we have

(w, xc , 1,k ) = (w, xc )(1,k ) + (w, xc1 )(3,k ) + (w, xc2 )(2,k ), (40)

and

(xc , w, 1,k ) = (xc , w)(1,k ) + (xc )(3,k ) + (xc1 )(2,k ). (41)

Taking the difference and applying monotonicity and symmetry, gives the
result in this case. 2

Lemma 10 If d, c 1 and (1,k ) = (wd , xc ), with w = x + 1 and x 1, then

(1,k1 ) (2,k2 ) (2,k ) (3,k1 ). (42)

Proof. Note that a1 = w and ak = x. We have

(1,k1 ) (2,k2 ) = a1 (2,k1 ) + (4,k1 ) (2,k2 )


= w[(2,k2 )ak1 + (2,k4 )] + (4,k1 ) (2,k2 )
= [wak1 1](2,k2 ) + w(2,k4 ) + (4,k1 ), (43)

while

(2,k ) (3,k1 ) = a2 (3,k ) + (5,k ) (3,k1 )


= a2 [(3,k1 )ak + (3,k3 )] + (5,k ) (3,k1 )
= [a2 x 1](3,k1 ) + a2 (3,k3 ) + (5,k ) (44)

The result follows upon noting that a2 w, ak1 x, and by monotonicity,


(2,k2 ) (3,k1 ), (2,k4 ) (3,k3 ) and (4,k1 ) (5,k ). 2

The following lemma is crucial in dealing with the final steps in the proof of
Theorem 1.

Lemma 11 For all (1,k ), (1,j ) and (1,l ), we have

(1,k , 1,j , 1,l ) (1,k , 1,j , 1,l )


= [(1,k )(2,l ) (1,k1 )(1,l )][(1,j2 ) (3,j )]
+ [(1,k )(3,l ) (1,k2 )(1,l )][(1,j1 ) (2,j )]
+ [(1,k1 )(3,l ) (1,k2 )(2,l )][(3,j1 ) (2,j2 )] (45)

10
Proof. Note that repeated splitting and symmetry gives


(1,j )(1,l ) (3,j )(1,l )


(1,k , 1,j , 1,l ) = (1,k )
+ ( 1,j1 )( 3,l ) + (1,k1 ) + (3,j1 )(3,l )


+ (1,j2 )(2,l ) + (3,j2 )(2,l )

(2,j )(1,l )


+(1,k2 )
+ ( 2,j1 )( )
3,l
, (46)

+ (2,j2 )(2,l )

and


(1,j )(1,l ) (1,j2 )(1,l )


(1,k , 1,j , 1,l ) = (1,k )
+ (2,j )(3,l )
+ (1,k1 ) + (2,j2 )(3,l )


+ (3,j )(2,l ) + (3,j2 )(2,l )

(1,j1 )(1,l )


+(1,k2 )
+ ( 2,j1 )( 3,l ) ,

(47)

+ (3,j1 )(2,l )

where we have used the fact that, by symmetry, (c,d ) = (c,d ).

Taking the difference in (47) and (46) and collecting terms gives that for
= (1,k , 1,j , 1,l ) (1,k , 1,j , 1,l ),

= (1,j2 )[(1,k )(2,l ) (1,k1 )(1,l )]


+(1,j1 )[(1,k )(3,l ) (1,k2 )(1,l )]
+(3,j )[(1,k1 )(1,l ) (1,k )(2,l )]
+(3,j1 )[(1,k1 )(3,l ) (1,k2 )(2,l )]
+(2,j )[(1,k2 )(1,l ) (1,k )(3,l )]
+(2,j2 )[(1,k2 )(2,l ) (1,k1 )(3,l )]. (48)

Finally, regrouping terms, gives the result. 2

The following two lemmas follow from Lemmas 10 and 11.

Lemma 12 Suppose (1,l ) = (b1 , b2 , . . . , bl ). If c, d 1, w = x+1, b1 = x 1,


and b2 {x, w}, then

11
(w2 , xc , wd , 1,l ) (wd+2 , xc , 1,l ) (49)

Proof. Setting (1,k ) = (w2 ) and (1,j ) = (wd , xc ), and applying Lemma 11
gives

(w2 , wd , xc , 1,l ) (w2 , xc , wd , 1,l )


= [(w2 )(2,l ) (w)(x, 2,l )] [(1,j2 ) (3,j )]
+ [(w2 )(3,l ) (x, 2,l )] [(1,j1 ) (2,j )]
+ [(w)(3,l ) (2,l )] [(3,j1 ) (2,j2 )] (50)

Now, note that, by monotonicity,

min{(1,j2 ) (3,j ), (1,j1 ) (2,j ), (2,j2 ) (3,j1 )} 0. (51)

As well,

(x, 2,l ) min{(w)(2,l ), [xw + x + 1](3,l )}


= min{(w)(2,l ), [w2 ](3,l )}. (52)

If (2,l ) w(3,l ), then each of the three summands in (50) are non-negative
and the result follows. Otherwise, suppose (2,l ) < w(3,l ). Then,

(w2 )(3,l ) (x, 2,l ) (w)(3,l ) + (2,l )


(w2 )(3,l ) (w)(2,l ) (w)(3,l ) + (2,l )
(w 1)[(w)(3,l ) 2,l )] 0. (53)

Now, combining (53), (51) and applying Lemma 10, gives the positivity of the
sum of the last two terms in (50). Positivity of the remaining term follows
from (52) and (51). 2

Similarly, we have the following lemma.

Lemma 13 If c, d 1, w = x + 1 and a1 = x 1, then

(w, xc , wd , 1,l ) (wd+1 , xc , 1,l ) (54)

Proof. Setting (1,k ) = (w) and (1,j ) = (wd , xc ), and applying Lemma 11
gives

12
(w, wd , xc , 1,l ) (w, xc , wd , 1,l )
= [(w)(2,l ) (x, 2,l )] [(1,j2 ) (3,j )]
+ (w)(3,l ) [(1,j1 ) (2,j )]
+ (3,l ) [(3,j1 ) (2,j2 )]. (55)

The result follows upon noting that (x, 2,l ) (w)(2,l ), (1,j2 ) (3,j ),
(w)(3,l ) > (3,l ), and applying Lemma 10 with (1,j ), in place of (1,k ). 2

We are now in a position to move onto the proof of our main results.

3 Main fixed sum results

In this section we prove our main results regarding maximal a1 , a2 , . . . , ak


under fixed sum constraints. First we prove a result similar to Theorem 2
in [7] which dealt with the case m = 2 for solutions to (2). For the current
scenario (i.e. m = 3), some additional complications arise essentially due to
the structure of the lemmas available.
P
Theorem 2 Suppose that aj 1 for 1 j k and kj=1 aj = N then
P
there exists b1 , b2 , . . . , bk with kj=1 bj = N and bj {[N/k], [N/k] + 1} for
j {1, 2, . . . , k}, such that

yk (a1 , a2 , . . . , ak ) yk (b1 , b2 , . . . , bk ). (56)

Proof. First note that by Lemmas 2 and 3 and Lemma 1 (iv), we have

yk (a1 , a2 , . . . , ak ) yk (b1,1 , b1,2 , a3 . . . , ak ), (57)

where b1,1 + b1,2 = a1 + a2 and |b1,1 b1,2 | 1.

Hence, suppose that

yk (a1 , a2 , . . . , ak ) yk (bt1,1 , bt1,2 , . . . , bt1,t , at+1 , . . . , ak ), (58)

with bt1,1 + bt1,2 + + bt1,t = a1 + a2 + + at and |bt1,i bt1,j | 1 for


1 i, j t.

Then, successively employing either Lemma 7 or Lemma 4, we obtain

13
yk (a1 , a2 , . . . , ak ) yk (bt1,1 , bt1,2 , . . . , bt1,t , at+1 , . . . , ak )
yk (bt,1 , bt,2 , . . . , bt,t , bt,t+1 , at+2 , . . . , ak ), (59)

where bt,1 + bt,2 + + bt,t+1 = a1 + a2 + + at+1 and |bt1,i bt1,j | 1


for 1 i, j t + 1. In proving (59) (via Lemmas 4 and 7), several cases
need to be considered. For instructive purposes, we include the details here,
as the arguments differ considerably from those in [7]. The method involves
transforming S = {bt1,1 , bt1,2 , . . . , bt1,t , at+1 }, while strictly lowering the
def
value of D = D(S) = max{|x y| : x, y S} at each step, and not decreasing
the value of yk .

First, note that by Lemma 4, we may reduce to the case that

max{|at+3 at+2 |, |at+2 at+1 |, |at+1 at |} 1. (60)

Now, set m = min{bt1,i } and M = max{bt1,i }. If at {m, M }, or m = M


and at+1 {m 1, m, m + 1} then (59) follows directly.

Case 1 (m = M ). Suppose at+1 m + 2. Then,

yk (bt1,1 , bt1,2 , . . . , bt1,t , at+1 , . . . , ak ) = yk (m, m, . . . , m, at+1 , . . . , ak )


yk (m, m, . . . , m + 1, at+1 1, . . . , ak ),

by Lemma 4. Similarly, if at+1 m 2, then decay of D follows by application


of Lemma 4 and symmetry.

Case 2 (M = m + 1, bt1,t = M , at+1 M + 1). Here decay of D follows by


application of Lemma 7. To see this, first let n = t max{i t 1 : bt1,i =
m} (the set is nonempty, since m 6= M . Now, apply Lemma 7 with C = 1,
x = M , D = at+1 M , n = t n and 1,k = (bt1,1 , bt1,2 , . . . , bt1,n 1 ). If
D 2 or m 2, then the decay of D follows directly by Lemma 7. Hence
assume m = 1, x = M = 2 and D = 1. Set Q = (1,k )(xn ) (1,k1 )(xn1 )
(1,k2 )(xn ). Then,

Q = [(1,k ) (1,k2 )](xn ) (1,k1 )(xn1 )


(xn1 ) (2[(1,k ) (1,k2 )] (1,k1 ))
= (xn1 )(2[(1,k1 ) + (1,k3 ) (1,k2 )] (1,k1 ))
= (xn1 )((1,k1 ) 2[(1,k2 ) (1,k3 )])
= (xn1 )(ak1 (1,k2 ) + (1,k4 ) 2[(1,k2 ) (1,k3 )])
= (xn1 )([ak1 2](1,k2 ) + 2(1,k3 ) + (1,k4 ))
= (xn1 )([ak1 2][ak2 (1,k3 ) + (1,k5 )] + 2(1,k3 ) + (1,k4 )).
(61)

14
For the inequality in the second line of (61), we employed the fact that (xn )
x(xn1 ) 2(xn1 ). Now, if ak1 2 then (61) gives that Q 0; hence suppose
ak1 = 1. Then, ak2 2 (indeed, this also follows from the fact that M = 2),
and (61) implies that

Q (xn1 )[[2(1,k3 ) + (1,k5 )] + 2(1,k3 ) + (1,k4 )]


= (xn1 )[(1,k4 ) (1,k5 )] 0. (62)

The decay then follows via Lemma 7.

Case 3 (M = m + 1, bt1,t = M , at+1 m 1). Here decay of D follows by


application of Lemma 4 and symmetry.

Case 4 (M = m + 1, bt1,t = m, at+1 M + 1). Here decay of D follows by


application of Lemma 4.

Case 5 (M = m + 1, bt1,t = m, at+1 m 1). Here decay of D follows by


application of Lemma 8 and symmetry. This case is very similar to Case 2.

Let n = t max{i t 1 : bt1,i = M }. Now, apply Lemma 8 with D = 1,


x = m, C = m at+1 , n = t n and (at+2 , at+3 , . . . , ak ) in place of (ak+1,m ).
If C 2 or at+2 2, then the decay of D follows directly by Lemma 8. Hence
assume C = 1 and at+1 = 1. We have x 2, and hence the analogue of (61)
follows as before.

As before, if ak1 = at+2 2 then the result holds, hence suppose at+2 = 1.
Then, by (60), at+3 2, and (61) implies that the analogue of (62) holds. The
decay then follows via Lemma 8.

Successive application of the results in Cases 15, gives a set of values S =


{bt,1 , bt,2 , . . . , bt,t , bt,t+1 }, such that D(S) 1 for which (59) holds. The result
then follows by induction, with bj = bk1,j for 1 j k. 2

We now turn to a proof of Theorem 1.

Proof of Theorem 1. First note that by Theorem 2, we may restrict our


attention to a1 , a2 , . . . , ak where aj {x, w} for j {1, 2, . . . , k}, where x =
[N/k] and w = x + 1.

Now, suppose ||{j : aj = x}|| = s and ||{j : aj = w}|| = v N mod k with


s + t = k, where for any set S, by ||S|| we denote its cardinality.

For k = 2 the theorem follows from Lemmas 2 and 3. Hence, suppose the
theorem holds for k K 1.

15
The case when v = 0 is immediate. Now, if v = 1 or v = 2, then application of
Lemma 9 (and symmetry) leads to the result. If v = 3 or v = 4, then successive
application of Lemmas 9 and 13 give the result. Hence, suppose v 5. Here,
applications of Lemmas 9, 13 and 12 reduce to the case a1 = a2 = a3 = w or
ak2 = ak1 = ak = w. Assume the former, i.e. a1 = a2 = a3 = w (the case of
the latter follows similarly, via symmetry). By induction (and symmetry), we
get

(a1 , a2 , . . . , aK ) = (w, w, w, a4 , . . . , aK )
= (w)(w, w, a4 , . . . , aK ) + (a4 , . . . , aK )
(w)(wv3 , xKv , w, w) + (wv5 , xKv , w, w)
= (wv2 , xKv , w2 ), (63)

and the theorem is proven. 2

Remark. Note that additional motivation for studying recursive sequences


of the form in (2) with integer coefficients was provided by a recent study of
recurrences on partially ordered sets (see [3] and [8]).

Acknowledgements

We are very thankful to a referee for comments and suggestions that improved
this manuscript.

References

[1] R. P. Agarwal, Difference Equations and Inequalities, Theory, Methods and


Applications, Second Edition (Revised and Expanded), Marcel Dekker, New
York, (2000).

[2] E. J. Barbeau, Pells Equation, Springer, New York (2003).

[3] K. S. Berenhaut and J. D. Foley Explicit Bounds for Multi-dimensional


Linear Recurrences with Restricted Coefficients, Journal of Mathematical
Analysis and Applications, Volume 322, Issue 2, (2006), 11591167.

[4] K. S. Berenhaut and E. G. Goedhart, Explicit bounds for second-


order difference equations and a solution to a question of Stevic, Journal of
Mathematical Analysis and Applications, 305, (2005), 110.

[5] K. S. Berenhaut and E. G. Goedhart, Second-Order Linear Recurrences


with Restricted Coefficients and the Constant (1/3)1/3 , Mathematical
Inequalities & Applications, Volume 9, (2006), 445452.

16
[6] K. S. Berenhaut and D. C. Morton, Second-order bounds for linear
recurrences with negative coefficients, Journal of Computational and Applied
Mathematics, 186 (2006), 504522.

[7] K. S. Berenhaut, A. B. OKeefe, and F. Saidak, Recursive sequences


of the form yn = an yn1 + yn2 with integer coefficients, Indian Journal of
Mathematics, 48, No. 1 (2006), 116.

[8] K. S. Berenhaut, A. B. OKeefe, and F. Saidak, Bounds for Linear


Recurrences on Partially Ordered Sets, In preparation, (2006).

[9] E. L. Dickson, History of the Theory of Numbers. Vol 1., Chap. XVII,
University of Chicago Press, Chicago (1919).

[10] M. Embree, and L. N. Trefethen. Growth and decay of random Fibonacci


sequences, The Royal Society of London Proceedings, Series A, Mathematical,
Physical and Engineering Sciences, 455 (1999), 2471-2485.

[11] G. H. Hardy and E. M. Wright, An Introduction to the Theory of Numbers.


Fifth edition. The Clarendon Press, Oxford University Press, New York (1979).

[12] A. Y. Khinchin, Continued Fractions. University of Chicago Press, Chicago


(1964).

[13] R. K. Kittappa, A representation of the solution of the nth order linear


difference equations with variable coefficients, Linear Algebra and Applications,
193 (1993), 211-222.

[14] R. K. Mallik, On the solution of a linear homogeneous difference equation


with varying coefficients, SIAM Journal on Mathematical Analysis, 31 (2000),
375-385.

[15] A. M. Odlyzko, Asymptotic enumeration methods, In:Handbook of


Combinatorics (R. Graham, M. Groetschel, and L. Lovasz, Editors), Volume
II, Elsevier, Amsterdam, (1995), 1063-1229.

[16] J. Popenda, One expression for the solutions of second order difference
equations, Proceedings of the American Mathematical Society, 100 (1987), 87-
93.

[17] S. Stevic , Growth theorems for homogeneous second-order difference


equations. ANZIAM J. 43 (2002), no. 4, 559566.

[18] S. Stevic, Asymptotic behavior of second-order difference equations. ANZIAM


J. 46 (2004), no. 1, 157170.

[19] S. Stevic, Growth estimates for solutions of nonlinear second-order difference


equations, ANZIAM J. 46 (3) (2005), 459-468

[20] D. Viswanath, Lyapunov exponents from random Fibonacci sequences to the


Lorenz equations, Ph.D. Thesis, Department of Computer Science, Cornell
University, (1998).

17
[21] D. Viswanath, Random Fibonacci sequences and the number 1.13198824...,
Mathematics of Computation, 69 (2000), 1131-1155.

[22] D. Viswanath and L. N. Trefethen. Condition numbers of random


triangular matrices, SIAM Journal on Matrix Analysis and Applications, 19
(1998), 564-581.

[23] T. G. Wright and L. N. Trefethen, Computing Lyapunov constants for


randomrecurrences with smooth coefficients, Journal of Computational and
Applied Mathematics 132 (2001), no. 2, 331-340.

18

Das könnte Ihnen auch gefallen