12 1

THE FIBONACCI QUARTERLY
THE OFFICIAL JOURNAL OF

THE FIBONA CCI ASSOC IA TION
VOLUME 12
\ \
$P^
J J
NUMBEE 1
CONTENTS
Fibonacci Notes 1. Zero-One Sequences
and Fibonacci Numbers of Higher Order
L.
Linearly Eecursive Sequences of Integers
Carlitz
L. Raphael
11
H. E. Huntley
38
Brother
The Golden Ellipse

On a Problem of M. Ward
R. R.
Generalized Hidden Hexagon Squares
A. K. Gupta
Letter to the Editor

Diagonal Sums
of the Trinomial Triangle
Laxton
. V. E. Hoggatt,
Jr.,
41
45
Donald
E. Knuth
46
and Marjorie
Bicknell
47
Fibonacci Sequences Modulo m
Agnes
Andreassian
51
Phi: Another Hiding Place
H. E. Huntley
65
Some Geometrical Properties

of the Generalized Fibonacci Sequence
D. V. Jaiswal
67
Sets of Binomial Coefficients

with Equal Products
Calvin
T. Long and V. E. Hoggatt,
On Daykin's Algorithm
for Finding the G. C D
V. C. Harris
Stufe of a Finite Field
Sahib Singh
Iteration Algorithms
for Certain Sums of Squares
Edgar
Certain Congruence P r o p e r t i e s
(Modulo 100) of Fibonacci Numbers
G, E. BER.GUM
71
80
81
Karst
83
R.
Turner
87
Santosh
Kumar
92
Michael
Fibonaccian Pathological Curves

Irreducibility of Lucas and
Generalized Lucas Polynomials.
Jr.
and V. E. Hoggatt,
Jr.
95
Elementary Problems and Solutions
Edited
by A. P . Hillman
101
Advanced Problems and Solutions
Edited
by R. E.
107
FEBBUAEY
Whitney
1974
THE FIBONACCI QUARTERLY

THE OFFICIAL JOURNAL OF THE FIBONACCI ASSOCIATION
DEVOTED TO THE STUDY
OF INTEGERS WITH SPECIAL PROPERTIES
CO-EDITORS
V. E. Hoggatt, J r .
Marjorie Bicknell
EDITORIAL BOARD
H. L. Alder
John L. Brown, J r .
Brother A. Brousseau
Paul F. Byrd
L. Carlitz
H. W. Eves
H. W. Gould
A. P.
David
C. T.
M. N
D. E.
Hillman
A. Klarner
Long
S. Swamy
Thoro
WITH THE COOPERATION OF

T e r r y Brennan
Maxey Brooke
Calvin D. Crabill
T. A. Davis
John H. Hal ton
A. F. Horadam
Dov Jarden
Stephen Jerbic
L. H. Lange
D. A. Lind
J a m e s Maxwell
Sister M. DeSales McNabb
Gerald Preston
D. W. Robinson
Azriel Rosenfeld
Lloyd Walker
Charles H. Wall
The California Mathematics Council
All subscription correspondence should be addressed to Brother Alfred Brousseau, St. Mary's
College, California 94575. All checks ($8.00 per year) should be made out to the Fibonacci
Association or the Fibonacci Quarterly.
Two copies of manuscripts intended for publication
in the Quarterly should be sent to Verner E. Hoggatt, J r . , Mathematics Department, San

Jose State University, San J o s e , California 95192. All manuscripts should be typed, doublespaced.
Drawings should be made the same size as they will appear in the Quarterly, and
should be done in India ink on either vellum o r bond paper.
Authors should keep a copy of
the manuscript sent to the editors.

The Quarterly is entered as third-class mail at the St. Mary T s College Post Office, Calif. ,
as an official publication of the Fibonacci Association.
The Quarterly is published in February, April, October, and December, each year.
FIBONACCI NOTES
1. ZERO-ONE SEQUENCES AND FIBONACCI NUMBERS OF HIGHER ORDER
L. CARL1TZ
Dyke University, Durham, North Carolina 27706
1.
INTRODUCTION
It is well known (see, for example [ 1 , p. 14]) that the number of sequences of zeros and
ones of length n with two consecutive ones forbidden is equal to
For example for
n = 3, the allowable sequences are

(000),
(100)9
(010),
(001),
(101) .
As a first extension of this result, let f(n,k) denote the number of sequences of length
n.
(1.1)
(a ls
a2,
-, an)
(a. = 0 or 1)
such that
(1.2)
a . a . + 1 a. + k
= 0
(i = 1, 2, - , n - k) ;
that i s , a string of k + 1 consecutive ones is forbidden. Also, let f(n,k,r) denote the number of such sequences with exactly r ones and let f.(n,k,r)
quences with r ones and beginning with j ones,
It suffices to evaluate f 0 (n,k,r).
(1.3)
where c . ( n , r )
denote the number of such s e -
0 < j < r.
We shall show that
f 0 (n,k,r)
= cfc(Q - r , r) ,
is defined by
kn
(1.4)
(1 + x + . . . + x )
= ^
ck(n,r) x r .
r=0
Also if
f
0( n ^ k >
^,1)
r
is the number of allowable sequences beginning with at least one z e r o , we have
* Supported in part by NSF Grant GP-17031.
1
FIBONACCI NOTES
f 0 (n,k)x n =
~;
n=0
(1.5)
[Feb.
i
l - x - x - - . - x
In the next place let r , s be fixed positive integers and let f(m,n; r,s) denote the number of sequences of length m + n
(a1s a 2 ,
1
, a , )
' m+n
(a. = 0 or 1)
l
with exactly m zeros and n ones, at most r consecutive zeros and at most s consecutive
ones. Also, let f.(m,n; r , s ) denote the number of such sequences beginning with exactly j
zeros,
0 < j < r; let f. (m,n; r , s ) denote the number of such sequences beginning with ex-
actly k ones,
0 < k < s.
As in the previous problem, it suffices to evaluate f 0 (m,n; j , k ) and 7 0 (m,n; j , k ) . We

shall show that
(1.6)
fo(m,n; r , s ) = ] P
k
+ ^
k i
" k)cs_i^kj
c r _ 1 ( k + 1, m - k - l)c j g _ 1 (k, n - k) ,
k
(1.7)
f 0 (Hi,n; r , s ) = ^
r_i
(k
>
"
1)c
-l(k'
k)
"
+ Y^
k
(1.8)
r-l(k'
f(m,n; r , s ) = ] T ( c r _ l ( k '
k
s-l(k
9 n
~ k)cs_l(k'
"
"
k)c
"
'
X)
'
k)
+ c r _ 1 ( k + 1, m - k - l)c g _ 1 (k, n - k)
+ c
_,(k, m - k)c n (k + 1, n - k - 1)}

.
J
r-1
s-1
As a further extension one can consider sequences of 0 ? s, l ' s , and 2 ! s , .say with r e strictions on the number of allowable consecutive elements of each kind.
this for another occasion.
However, we leave
1974]
FIBONACCI NOTES
2.
FIRST PROBLEM
We now consider the first problem as defined above. It is clear from the definition that
k
(2.1)
f(n,k,r) = ] f (n.k.r) .
3=0
Also
k
f 0 (n 9 k,r) = J^ f j <n - L k> r ) = f < n - *> k> r>
(2.2)
5=0
and
(2.3)
fjfci.k.r)
Hence f 0 (n,k,r)
= f 0 (n - j , k, r - j)
(0<j<k) .
satisfies the mixed r e c u r r e n c e

k
(2.4)
f0(n,k,r)
= ] P $, (n - j - 1, k s r - j)
(n > k + 1) .
3=0
Now, for r < k,

f0(l,ksr)
= 50jr
fo(2,k,r) _= j\ 1j
(r = 0, 1)
> J;
(r = 0)
2
1
0
(r = 1)
(r = 2)
(r > 2)
'
Generally, for 1 < m < k + 1, we have
(2.5)
f 0 (m,k,r)
(
If we take n = k + l 5
"
(r > m)
r < k in (2.4) we get
f 0 (k + 1, k, r) = J^
3=0
By (2.5) this reduces to
(0 < r < m)
o( k " h k, r - j)
[Feb.
FIBONACCI NOTES
O J - t f ; ^ 1 ) * *
r - k)
Since
k-1
Z^\
j=0
k-1
- j
k-1
/ " ^ ^ k - r - l J ' ^ f k - r - l J ' V k - r j ' l r /

f
j=0
]=0 X
it follows that (2.4) holds for n ^. k + 1 > r provided we define

f 0 (0,k,t)
=0
(t < 0) .
Moreover (2.4) holds for r > k, n = k + 1, since both sides vanish.

Now put
(2.6)
^
fo(nkr)xI1yr
n,r=0
F(x,y) =
Then, by (2.4) and (2.5),

k
f0(n,k,r)xnyr + ^
F(x,y) = ] P ^
n=0 r=0
1 +
f0 (n,k,r) x n y r
n=k+l r=0
EE^^v
n=l r=0
oo
E 1212 fo(n - J - !fc'r - J )XI1 ^

n=k+l r=0 j=0
k
= 1 +
n=l r=0
OO
j+
2> V
j=0
k
00
Efo(n - J -1. k>r - j>* n ~ 3 ~V- j
n=k+l r=0
1 +
n=l r=0
k-j-1
k
F(x,y) =
3=0
J^
][]f(n,k,r)xnyr
n=0
r=0
19 7 4]
FIBONACCI NOTES
Since
k
j+
k-j-1
k - j - 1 *>
E* V E Ef(n.k,r)xv = E* V E Ef 11 ; 1 )* 11y r
j=0
n=0 r=0
5+
j=o
n=0 r=0 ^
= EExVE(v;: 2 )
n=l
n=l
it follows that
k
F(x :,y) == 1
*E x
y3 . F(x,y) .
Therefore
(2.7)
F(x,.y)
iyj
1 3=0
If we define the coefficient c. (n,r) by means of
kn
(2.8)
(1 + x + . . . +
)n = ^
\(n,k)xT
r=0
it is clear that
^ = 5> S [X>V
1-xJ
xV
s=0
\ j=0
3=0
00
ks
X X Z ck(s>r>xV
s=0
r=0
E E ck<n - r> r>*v
n=0 r=0
Therefore by (2.6) and (2.7) we have
[Feb.
FIBONACCI NOTES
f 0 (n,k,r)
(2.9)
= c, (n - r , r)
and, in view of (2.2),

(2.10)
f(n,k,r) = c k (n - r + 1, r) .
Moreover if we put
f 0 (n,k)
= ^2
fo (n,k,r) ,
it follows that
(2.11)
2J f0(n,k) xn =
k+T
n=0
5=1
In particular, for k = 1, it is clear from (2.11) that
(2.12)
f 0 (n,D
3.
= F,
n+1
SECOND PROBLEM
We turn now to the second problem defined in the Introduction. It is convenient to define
(f 0 (0,0; r , s) = f(0,0; r , s ) = 1
(3.1)
( ^ ( 0 , 0 ; r , s) = T k (0,0; r , s ) = 0
(j > 0, k > 0) .
The following relations follow from the definition.
f 0 (m,n; r , s ) = ^
f k (m,n; r , s)
k=l
(3.2)
f0(m,.n; r , s ) = ^ J f (m,n; r , s)
3=1
s
f (m,n; r , s ) = 2 ^ f k (ni - j , n; r , s )
(j ^ 0)
k=^l
(3.3)
r
f k (m,n; r , s) = J j V**1'
5=1
" k'
r,S
^ ~ ^
197
4]
FIBONACCI NOTES
where it is understood that

^ ( m , n ; r , s ) = f 0 (m,n; r , s ) = 0
if either m or n is negative.
We have also
/ ff.(m,n;
(m,n; rr9,sj ) = f0(m - j , n; r, s)
(j > 0)
( T k (m 3 n; r ,s)
= f 0 (m 9 n - k; r , s )
5j
(k > 0)
(3.4)
In particular
f. (j, n; r , s )
= f 0 (0, n; r , s) = 1
(0 < n < s)
(3.5)
1 (m,k; r , s ) = lQ(m90;
r, s) = 1
(0 < m < r)
It follows from (3.2) and (3.4) that

s
f 0 (m t n; r , s ) = ^
f 0 (m, n - k; r, s)
k=l
(3.6)
r
7 0 (m,n; r , s ) = ^
fo(m - j , n; r , s)
3=1
and therefore
r
f 0 (m,n; r, s) = ^
s
^
f0(m - j , n - k; r , s)
j=l k=l
(3.7)
r
7 0 (m,n; r s s ) = ^
s
X^
( m
- h
- k5 r s )
j=l k=l
Now put
F 0 = F 0 (x 9 y) =
f 0 (m s n; r , s ) x m y n
m,n=0
00
0 = "F0(x9y) =
^J
m,n=0
Then by (3.5) and (3.7),
"f 0 (m s n; r , s ) x m y n
FIBONACCI NOTES
s
F
2J y
k=Q
z2
Mm>n; r,s)xmyn
m,n=l
s
=
2
k=0
Zy
1C Yl 2 fo(m
m,n=l j=l k=l
[Feb.
k=0
ZZ
" J'
'
k; r
>
>xinyn
xJy
j=l k=l
S Mm>n; r,s)xmyn
m,n=0
= z> k+ x;i> j y k - F o.
k=0
j=l k=l
so that
Ey 1
(3.8)
F0 = -
~
r
s
. ,
1 -
xJyk
j=l k=l
Similarly
r
y
(3.9)
r
s
. ,
i - E E xJyk
3=1 k=l
Clearly
f(m s n; r , s ) = f 0 (m,n; r , s ) + 7 0 ( m , n ; r , s )
(m + n > 0) ,
so that
F(x,y) =
f(m,n; r , s ) x m y n
22
= - 1 + F 0 (x,y) + F"0(x,y)
m s n=0
Hence
(3.10)
r
s .
x3 + y
F(x,y) = J = 2
!S=
r
s
. .
i - E E xJyk
j=l k=l
To get explicit formulas we take
1974]
FIBONACCI NOTES
= Exkyk(1
i - E E ^ )
j=l k=l
--- + x r - 1 ) k d
...+ys-1)k
k=0
OO
OO
xkyk
OO
(k i)xl
X vi >
X vi(k*j)x3
k=0
i=0
j=0
OO
m s n=0
^ ^ X) cr-l(k' m " k)c s-l (ks
~ k)
Then
r
E
-,
-crJ
OO
,
, r - l-,v k+1^ _,
, ,, k,
E xk+1
y (1 + + x
)
(1 +
/n
x3 k
i - E E
k=0
s-lv
, k
+ y
)
j=l k=l
k=0
i,j
OO
xmyn^cr_1(k
2
m,n=0
r1 s
- E
+ 1, m - k - D c ^ ^ k , n - k ) ,
^y
^ ^ E cr-l(k' m " k ) Vl ( k
= E
m3n=0
1. - - k - 1)
j=l k=l
It follows that
f 0 (m,n; r , s ) = J ^ ( c r i < k j
(3.11)
+ c
"
k)c
_i(k'
"
k)
,(k + 1, m - k - l)c n (k, n - k)}

J
r-1
s-1
{ c r i^ k
7 0 (m,n; r , s ) = ^
(3.12)
"
k)c
s_i^
'
+ c _x(k, m - k)c g _ 1 (k + 1, n - k + 1)}

r-
10
FIBONACCI NOTES
Feb. c1974
f(m,n; r , s ) = ^ ) ( c r _ 1 ( k , m - k)c g _ 1 (k, n - k)

k
(3,13)
+ c
,(k, n - k)
n (k + 1, m - k - l)c
r-1
s-1
+ c r - : L (k, m - k)c g _ 1 (k + 1, n - k - 1)} .
When r o the restrictions in the definition of f, f. reduce to the single restriction
that the number of consecutive ones is at most
s.
The generating functions (3.8),
(3.9),
(3.10) reduce to
(1 " x) D yk
k=0
(3.80
i - x E yk
k=0
(3.9')
1 - x yk
k=0
(3.10T)
/
k=0
i - xz
yk
k=0
It can be verified that these results are in agreement with the results of Sec. 2 above.
REFERENCE
1. J. Riordan, An Introduction to Combinatorial Analysis, Wiley New York, 1958.
LINEARLY RECURSIVE SEQUENCES OF INTEGERS

BROTHER L RAPHAEL, FSC
St Mary's College,, Moraga, California 94575
PART 1.
INTRODUCTION
As h a r m l e s s as it may appear, the Fibonacci sequence has provoked a remarkable

amount of r e s e a r c h .
It seems that there is no end to the results that may be derived from
the basic definition

F ^ 0 = F _,_., + F
n+2
n+1
n
and
F0 = 0
u
and
F 4 = 1,
l
'
which Leonardo of Pisa found lurking in the simple rabbit problem.
For example, an exten-
sion of the definition yields the so-called Lucas numbers:

L
40
n+2
= L ,. + L
n+1
n
and
Lftu = 2
and
Li1 = 1.
Evidently, any two integers may be used "to start" the sequence.
However, it is well known
that there is an extraordinary relationship between the Fibonacci and Lucas sequences.
In
particular:
Fn
= F L
2n
n n
and
F _,, + F n = L .
n+1
n-1
n
P r e c i s e l y where does this peculiarity a r i s e ?

Then, again, many remarkable summation formulae are available.
n
partial sum of the Fibonacci sequence is expressed by F
erally for proving such formulae is induction on the index.
+2
- 1. The method used g e n -
This involves
1.
a guess provoked by the investigation of individual c a s e s ,
2.
an efficient formulation of the guess, and
3.
a proof by finite induction.
The drawbacks to this method are obvious.
In particular, the
F i r s t , it depends very heavily on insight and
cleverness, which qualities, while being desirable in any mathematician, do not lead to r e sults very quickly.
Second, this method is entirely inadequate for cases involving bulky for-
mulations, and, of course, many times a result suggested by individual observation does not
immediately come in convenient form.
alize results.
Finally, such a method is unable, to relate and gener-
Mathematics is incomplete until the specific, and perhaps surprising, facts
a r e brought back to a generalization from which they maybe deduced. Not only does this give
a foundation to the conclusions themselves, but it enables one to draw further,
unsuspected
conclusions, which a r e beyond inductive methods. Furthermore, as a result of a generalized

deduction, the formulation will be more elegant and notationally consistent.
11
12
[Feb.
What is required then, is a generalization of the Fibonacci sequence, discarding the

incidental.
At points this project will appear to be unnecessarily removed from the simplic-
ities of the original sequence, but attempts will be made to show the connections between the
more general case and the more familiar results.
DEFINITIONS
The Fibonacci sequence is based on an additive relationship between any t e r m and the
two preceding t e r m s . In our generalization, it is n e c e s s a r y to exploit two aspects of this r e lationship: we shall make it a linear dependence, and it will involve the preceding p t e r m s .
Here, and throughout,
f will note the general additive sequence:
(1)
f . = a-f __, - + af j . 0 + + a f
n+p
1 n+p-1
2 n+p-2
p n
It seems essential to the spirit of these sequences that they be integral.
(n = 0, 1, 2, ).
To insure this, we
must demand that the set

1
be integers.
This set will be called the spectrum.
(2)
But, returning to (1) and letting n = 0:
f = Y ^ a. f . ,
k=l
reveals that we must specify the first p t e r m s of the sequence in o r d e r that the others may
be obtained.
The set of integers

0
so specified will be called the initial set, or the initials.

It might be mentioned here that the Fibonacci sequence is obtained by letting p = 2 and
taking the spectrum
{l,l}
and the initials
{o,l}.
And the Lucas sequence has
p = 2,
spectrum { l , l} and initials {2, l} .

We wish now to extend the definition (1) so that negative values for the index a r e allowed.
Using the
back-up n approach, we obtain

f
., = a . f 0 + + a n fn + a f - ,
p-1
1 p-2
p-1 0
p -1
or
p-1
-1
1 p-l " / akfp-l-k

k=l
1974
13
Continuing, it can be seen that, for any n = 0, 1, 2, ,
(3)
f n = -II f
-n
-Va,
1 p-n
\
x^ k p-n-k
k=l
Clearly, in o r d e r to maintain an integral sequence for all values of the index, positive and
negative, it is n e c e s s a r y to take a
= 1. In any c a s e , we have that a2 = 1.
UNARY SEQUENCES (p = 1)
The number p of n e c e s s a r y initial values classifies the sequence as unary, binary,
tertiary, and so on.
The analysis of the unary sequences is r a t h e r trivial. The spectrum is
{aj} and the initial set {f 0 }. But since p = 1, we must have 2LX = 1, so that (1) comes
down to:
f n+1
,- = fn
o r , immediately:
f
n+1
= K( }l ) n + 1 f
0
In addition, it would seem altogether desirable to eliminate those sequences which can be
"reduced" by dividing each t e r m by a constant. That would leave only the primitive sequences
for which if d divides f.
for each value of k, then d = 1. In addition, we eliminate those
trivial sequences with each t e r m zero.
These conditions a r e met by demanding that neither
the spectrum nor the initial set be all z e r o , and that no constant be divisible into all the s p e c trum o r initial set.
With these r e s t r i c t i o n s , we see that the unary sequences become:

f,
= 1,
for all
fk
k,
or
= (-Dk .
This simply ends all discussion of unary sequences.

ALGEBRAIC GENERATORS
One of the most common manifestations of additive recursive sequences is the power
s e r i e s expansion of certain functions.
For example, a short calculation leads one to con-
clude that:
QO
" x2
k=o
The actual derivation of this result stems directly from the definition of the Fibonacci s e quence.
In what follows, we will use the same derivation in a generalized form.
What we
14
[Feb.
want to discover is an expression for:

00
(4)
ZX
*(X)'
k=0
where, by (1):
P
' J k n+p-k
k=l
n+p
Now, we multiply (1) by x
, and sum over the index n,

P
00
Z-J
n=0
n+
so that:
00
=V
a x k /V jf
/ * k
k=l
n+p-k
xn+p~k
n=0
But, taking into account (4), we may rearrange this expression, and:
/
W X
* ( "E V*
\
k=l
p-1
=f +
o E 1 k "Z a J f M
k=l
j=l
This singularly awkward expression can be made manageable by making the somewhat
a r b i t r a r y definition of a0 = - 1 . The introduction of a0 greatly simplifies the formulation of
the required function:
P-1
E
k
cD(x) = =
(5)
E= Q ai fk-i
J
k=0
We need hardly say that this is the required expression, which reduces to the familiar Fibonacci power s e r i e s when p = 2, aA = a2 = 1, and f0 = 0, ix - 1. But, further investigation of (5) leads to considerations which will be of crucial importance later. F i r s t , we r e m a r k that the denominator is a p -degree polynomial:
-a 0 + aAx + a 2 x 2 + . + a p x P ,
(a0 = -1) ,
which will be called the spectral polynomial.

Then, with regard to the numerator, the following definition will be made:
(6)
. = -Yâ.f ^ .
m,k
JLJ 3 m+k-j
j=0
for
0 <, k j
"*
1974]
In other words, h m, K. is a partial sum of t e r m s .

n
For example:
-aAf
m, 0
0 m
15
m
cf (1).
=0
m,p
- = a f
_, ,
m,p~l
p m-1
and, for convenience:

(7)
k =
o,k =
k"
ifk-i
kfo
The introduction of (7) into (5) yields the remarkably concise:
P"1
(5f)
*w = XX x k = -k=0
THE Q-SEQUENCE
In any f-sequence, it is possible to choose the initial set as any set of p
t e r m s , so that two "different" sequences may actually differ only in their indices.
then necessary to consider some sort of fundamental sequence.
sequences.
These sequences exist for all values of p,
Referring to (6), we will rename the partial sums h
It seems
This fundamental sequence
has the simplest non-trivial initial set; namely {o, 0, , 0, l } .

is a binary case.
consecutive
The Fibonacci sequence
and they will be called Q. :
k
(8)
m,k
= -\â.Q
for
0 ^
3=0
and, from (7):
k
k
but, from the definition of Q-sequences,

H k = 0,
for
0,k
/ ^ 3 k-j
3=0
Q, = 0 for 0 ^ k ^ p - 2, and Q
0 < k ^ p - 2,
H - = 1 .
p-1
Using these results in (5T), we have:
or
k = p ,
= 1:
16
(9)
^ Q
k X
- HI.fk_xx
!k
IFO
P
k=o
[Feb.
xP"1
P
y a. x
- y
The right-hand member of (9) may be treated as a geometric s e r i e s :
" JL a k x
k=0
X^0
and successive binomial expansions of the polynomial in parenthesis gives:
(EZ(v1)'
o o P
. \
k t =0 k=0
oo
ki
=qEE^-wr(v):
\ k i = 0 k2=0
'
k=2
and so on. After p steps, we may collect coefficients of x
^~
'
in (9) and equate them,
obtaining:
<*
< w - Z . U J U J U ; - l IT)&*&*' i r k p % p
s(SCXS)-(V)*
where the sum is taken over all {k.} such that

P
2>.
i=l
and m + p - 1 > 0.
Looking at the binary case (p = 2), we discover that
%+
EI
/ n - i \ a n-2i i
i=0 \
) i
**
where
(M
= 0
for
0 < n < k,
and
n + 1 > 0
1974]
17
so that, for SLt = a2 = 1:

F
C'-)
n+1
which, of course, is the well-known ''rising diagonal" result for Fibonacci numbers, derived
from P a s c a l ' s triangle.
(12)
And, for comparison, here is the ternary case:
sf-i-'Xir^
n+2 = E
Remark.
In this section, and throughout the r e s t , we choose to make the agreement
that I . J = 0 for all 0 < n < k.

plifies the summation notation.
This appears a bit a r b i t r a r y , but it is used since it s i m -
Notice that the upper index on the summation may be taken
as infinity, since by our agreement the binomial coefficients vanish for large enough k.
might be pointed out that the real upper index, for example, in (11) is | [ n / 2 | ,
greatest integer in n / 2 .
It
that i s , the
The bulky notation required for (10) in particular in this form w a r -
rants using our more simplified method.

THE Q-SEQUENCE AS BASIS
The fundamental nature of the Q-sequence is clearly shown in the following argument.
We return to (5 f ), and rearrange slightly:
P-1
- E \x
00
IV
k=0 R
Vp a, xk
k=0
- Ev k " p+1
P-1
xK
X
E
k=0
a xk
then, taking into account (9), we have:

00
p-1
k=0
k=0
\ i=0
00
HID+l
Comparing the coefficients of x
in this expression, we find that:

p-1
(13)
^ = Y ^ h. Q .
m-p+1
Z^J
k^m-k
k=0
"
18
[Feb.
This clearly shows the basic nature of the Q-sequence it forms a basis set for any other
sequence having the same spectrum.
But a more useful formulation may be derived by con-
sidering (7), and substituting into (13):

p-1
(14)
,- = -V^yVf,
.Q . ,
J
J
ni-p+1
'
'
3 k-j
k=0 j=0
and then, using (8):
m-k
k
~ Z-î j m-p+1+k-j
j=0
m-p+l,k
'
we have, from (14):

P-1
(15)
m-p+l
Y ^Hm-p+l.kfp-k-l
k=0
'
and finally, an obvious adjustment of index leaves us with:

P-1
(16)
Remark.
S Hm,k
k=0
where
p-l-k
For certain values of k,
m,k
m,k
is
îven *
p
a
(8)
'
an alternative form of (8) is desirable:
k
H
m+k - 2 - / j m + k - j
3=1
k
a Q
Z^ j m+k-j
3=1
" /,aiQm+k-i
3=1
p-k
(17)
H . = y ^ a . ^.Q
m,k
/ J k+j m - j
3=1
THE H-SEQUENCE
What appeared in (8) to be m e r e l y notational convenience can now show more positive
results.
For example, a linear combination of p consecutive H
the spirit of (1)):
. over the m index (in
1974]

P
Zâj
j=l
"^âjZ-iaiQni-j+k-i
j=l
i=0
m-],k ~
(18)
19
a Q
-E iE J m + k- H
i=0
j=l
X a i Q i m+k-i
m,k
i=0
shows that H
is itself an f-sequence for any choice of k. In fact, for
k = 0,
the H-
sequence reduces to the Q-sequence due to:

0
Hm,Qn = - y aj . jQ m - j. = -a AO^m
Q
= Qm
mm
3=0
But, for any choice of k, we must have in general, that H-sequences satisfy (16), since they
a r e f-sequences:
P-1
(19)
. = Y^H
m,k
Z^
.H
m,j
..
p-l-j,k
3=0
which is a remarkable formula suggestive of a whole s e r i e s of important results.
PART 2. MATRIX REPRESENTATIONS
A great many of the familiar Lucas and Fibonacci identities have been shown to be r e lated to the properties of m a t r i c e s .
The attempt to generalize these results for higher o r -
d e r s of sequence directly leads to various s o r t s of results depending largely on the aspect

taken for generalization.
(8)
and
(19)
as well as
(16)
But our previous work has led up to the following formulae:
H , = - /
a.Q M
m,k
x^ j ^ m + k - j
j=o
p-1
H
m,k
= y^H
Z/ m , j
for
0 ^ k ^ p
"*
^ ^
.H
+~h ,k
p-
j=o
p-1
f
f
= > H
m
/ J m , k p - 1-k
in
k=0
These three equations are strongly suggestive of matrix multiplication, particularly the last
(16). In fact, if the following definitions are made, a singularly simple formulation may be
given:
20

H1J
(20)
H
m+p-1,0""'
TT
H
m,0
Hm , p - l
[Feb.
m+p-l,p-l
Then:
H
p-1,0
H
p-l.p-1
H =1
H0,0
and
... w
I ,
the identity,
0,p-l
H1 = H
0
.0
A glance at (19) shows that the matrix II is really multiplicative; that i s :

H
Tm+n
mHn
since (19) is merely a statement of such a multiplication, row by column.
Here again, what
began in (8) as m e r e convenience, is seen to have something of a fundamental character with

regard to the recursive sequences.
Now, in addition, let us define:
F
m
(21)
Finally, then, it is evident that (16) may be written in the matrix form:
(22)
= HmFn
A particularly useful r e m a r k may be inserted h e r e :

d e t H m = ( d e t H ) m = (a ( - l ) p + 1 ) m .
(23)
This can be seen by considering definitions (20). However, in o r d e r to maintain a sequence

which has integers for all values of the index we need a
m
for any value of m , d e t l l
= 1.
Also, the h a r m l e s s observation that II
= II
= 1, as was seen in (3).
, when compared entry for entry
leads to the remarkable:

TT
(24)
k=l
TT
m+p-k,j-l
n+p-i,k-l
Hence,
TT
m+n+p-i,j-l
1974]
21
which is actually a generalization of (19).

In particular, by taking i = p and j = 1 in (24), and recalling that H
ft
= Q
in
(8), and then rearranging index in (24):

P-l
(25)
= Q.
JLJ n,l Q.
k=0
GENERAL REDUCTIONS
Rather than considering column matrices of
f. ,
we now extend the treatment to the
square matrix, having columns given by (21):
m+2p-l
f
(26)
F
m
m+p-1
f
m+2p-2
m+p-2
\ f

\ m+p-1
f
m
Then (22) becomes an expression involving p X p m a t r i c e s :
Imlo
where:
-m
f
F
2p-l
f
p-l
f
. "p-l
U0
Taking determinants, and simplifying, using (23):

det F
= det H
m
(27)
det F A
0
= (ap(-l)p"1)mdetF0
or:
detF0
det F
m
for any m .
Clearly, the number det F 0 is an extraordinary constant for any sequence which depends on
the initial set If.
P~ , and which will be called the characteristic.
IJo
The characteristic of the Fibonacci sequence is
while that of the Lucas sequence Is
1 ,
22

3
1
1
2
[Feb.
5 ,
as is well known.
Again, the simple r e m a r k that:
F + = Hm+nFn = HmF
m+n
0
n
plus, a comparison of e n t r i e s , gives:
(28)
P-l
y H
f
m+n
/ J m , k n+p-l-k
k=0
which is the general reduction.
It is a generalization of (16).
EXAMPLES
The binary c a s e , of course,, yields the most familiar results:
so that:
H-(?
i)
and
+ 1
- <-a2)m(f2fo - f?)
det IT = -a 2
F r o m (27) in the binary case:
Wm-fL
(30)
so that, for the Q-sequence:

Q 0Q
- Q2
=
^m+2^m
^m+1
Q2
(3D
- Q W ^ Q W -, =
(-a2)m(-l)
2/
'
(-*i)m~l
and the binary reduction becomes, referring to (28):

(32)
Q f ^ +a0Q
J
= f , .
^ m n+1
2^m-l n
m+n
The correspondence with the usual Fibonacci results may be worked out in detail directly
from these identities.
Now, turning our attention to the ternary case (p = 3), we discover several important
points.
F i r s t , the elegant formulations of the binary case do not hold up for
higher c a s e s .
p = 3,
or for
Also, symmetry of expression begins to fade with the higher sequences.
1974]
23
Clearly, most of the interesting properties of the Fibonacci sequence stem from its being a
binary sequence, while a few come from its being a sequence in general.
We will here give
the ternary results:

H111 =
(33)
'Qm+2
Sm+i
2Qm+l
3Qm
2^m
3^m-l
a0Q
., + a 0 Q
2^m-l
3^m-2
3Qm+l
3Qm-l
3Qm-2
and
/ai
0J
\ 0
0 /
=( 1
so that: det II = a3
a3\
hence:
m+4
(34)
f
m+3
m+3
m+2
m+2
m+l
m+2
f
= (a 3 )
m+1
m
For Q-sequences in the t e r n a r y case:
(35)
m+4
m+3
m+3
m+2
m+2
m+1
Qm+2
Qm+1
-(a 3 )
And the ternary reduction:

Q f , 0 + (Q _, - - an Q )f ^ + a 0 Q n f
^ m n+2
^m+1
1 rm n+1
3^m-l n
(36)
= f ,
m+n
NEGATIVE INDEX
Already, we have investigated the nature of the general sequence for negative values of
the index. A n e c e s s a r y and sufficient condition that a sequence be primitive and integral is
that a2p = 1. Now, using;
& more recent r e s u l t s , it is possible to look into the matter a bit
more deeply, and obtain expressions relating t e r m s of negative index with those of positive
index.
Were the matrix equation B[ m F 0 = F
o r , in particular:
(37)
to hold for negative value of the index:
F
= H"mFn
-m
0
(Hm)_1I0
H"m = (H m f
so that, after the indicated inversion, we may equate the entries in (37):
24
LINEAELY RECURSIVE SEQUENCES OF INTEGERS
(38)
j. . . n =
-m+p-1,3-1
d e
t H
[Feb.
minor H __, . . n .
m+p-j,i-l
Then, letting j = 1, and recalling (8):
. n = Q
. =
-m+p-i,0
-m+p-i
, , m
minor H
, - . .,
m+p-l,i-l
and, then, letting i = p, we have, after reference to (23):

(39)
= (a ( - l ) p + 1 ) ~ m
-m
p
minor H _, 1
m+p-l,p-l
Then, for the general c a s e , we need only note that from (16):
P-l
(i6?)
f
-ni
= y^H
/ J
k=0
,.,
.f
-m,k p-l-k
where:
k
(8f)
. =\â.Q
3=0
As a footnote, we add two identities coming from the equation II If
= I ,
where entries
a r e compared, after completing the multiplication on the left member:
(40)
\ ^ H _,...- H
J . . I 1
_j
m+p-k,j-l -m+p-i,k-l
k=l
H . . 1 = s . .
p-i,j-l
ij
for 0 ^ 1 ^ p, 0 < j ^ p and 0 ^ k < p, and where 6... is Kronecker's delta.

If i = p and j = 1 in (40):
P
J ^ HU n H p -,_k ,n0Hn - ...

m , k,_- l, =
" "H,
0,0
k=lwhich may be rewritten:
P-l
(41)
, , ,H
. = 0
JL^S ^ m + p - k - 1 - m , k
V Q
k=0
'
1974]
25
Applying (39) to the binary case yields the intriguing result:

Q_m = - ( _ a 2 r m Q m
(42)
or that:
JQ_m| = | Q m | ,
for
p = 2,
while in the ternary case:

Q_m = ( a 3 ) - m ( Q ^ + 1 - Q m Q m + 2 ) .
(43)
Clearly, the beauty of the expression for
for
p = 3 .
p = 2 does not c a r r y over to the situations for
g r e a t e r values of p.
MATRIX SEQUENCES
An obvious, but interesting result of (26) is the matrix expression (using entrywise
addition):
< 44 >
2>k^m-k = FIm
k=l
From (1), it is evident that the m a t r i c e s | F, i form an f-sequence with spectrum {a, I .
F u r t h e r m o r e , (44) m a y b e written, using the definition of F :
k=l
however,
F 0 ^ 0,
k=l
so that, dividing it out:

P
(45)
XA- m " k
'
k=0
in which case the powers of the matrix H form an f-sequence.
In fact, (45) is really a r e -
sult of the Cayley-Hamilton Theorem.
ROOTS OF THE SPECTRAL POLYNOMIAL

Returning to the e a r l i e r question of explicit determination of f
that (10) was obtained, which expressed Q
and Q ,
we recall
in t e r m s of a sum of binomial coefficients.
different approach now will yield the so-called Binet form, which may then be compared with
(10) for a s e r i e s of remarkable relationships.
But first we return to (5 f ):
26
[Feb.
Kx
oo
2L,
k
k=0
V^f
> f. X =
Z j k
p
k=0
- V
a,x
recalling that the spectral polynomial appears in the denominator.
Now consider that this
polynomial has been factored in the complex field:
(46)
->
Xx*
P
= n (1 - r.x) ,
* -1
k=0
where the roots a r e \ 1/r. y , a set of complex numbers, none of which are zero.
Now, let
us make the very strong assumption that the roots a r e distinct, so that:
P-l
Vuk=
(47)
Z ^ k
k=0
k=Q
=Y
p
n (1 - r.x)
i=l
*
T-i-
Z-rf 1 - r . x
i=l
where the right-hand member is a sum of partial fractions. What is needed is an expression
for each A..
Using a geometric s e r i e s and (47):
_00
k=0
00
i=l
j=0
k
and, equating coefficients of x * :
(48)
k=ZVf
3=1
Then, from (5') and (46):
00
f k x
k=0
/ P
(
n (1 - r.x)
N1 1
I
P-1
^ V
k=0
"
1974
27
and so, cancelling the t e r m (1 - r. x), after introducing (47):

P
p-1
A n (l-.x) = J ] V k k=i
lfK
k=o
Now, we substitute r Q (n = 1, 2, , p) for 1/x,
and recall that the {r. I a r e distinct:
i^-i)Yt:
AJ n ( 1 - ^ - 1 ) = > > k ( l / r n ) k
so that, finally:
P-1
(49)
i i
> h. r p
iS) k n
= SJL-
A
n
. /
ifn
which is exactly the expression for the A
(r
- r.)
demanded for (47), and (48).
In fact, now we
may introduce into (48) the value for A. derived from (49):
p p-1
- _ _ ^
^m+p-l-k
r^
<--^v^
m
i=l k=0
n (r. - r . )
-/
or:
p-1
(50)
/ p
rm+p-l-k
EME^.
k=0
\j=l
n (r. - r . )
l
#i J
However, comparing (50) with (13) and rearranging index shows that:
(51)
m
r.
3
Qm
i=l
r. - r.)
-j.-
where, of course, the r. are assumed to be distinct.

Binet-form for the Q-sequences.
This expression (51) is the general
28

Remark.
[Feb.
We note that, at the opposite extreme, the assumption might have been made
that r. = r for all i,
so that all the spectral polynomial roots are equal:
"X) a k x k
(52)
= (1
" rx)P '
k=0
in which case (5T) becomes:

p-^-1
Z\
ZV^
(53)
k=o
(i -
rx
>P
and a geometric expansion, and comparison of coefficients of x

pP --11 ,
gives:
f = yy^(( mm + + Pp "n
- i1- "MMh y. * " 3
C
(54)
L-d\
p - 1
1=0
and, again, comparing this expression with (13) gives that:
(55)
CD
"Mrm-P+1
=1
EXAMPLES
In the binary c a s e , many of the above results produce elegant formulae.
(51) the roots are 1/r^ f l / r 2
and 1 - a t x - a2x2 = (1 - r t x)(l - r 2 x),

m
*i
(56)
^m
vt - r 2
m
*i
Hence, if in
then:
m
r2
=
r 2 - rt
rt - r 2
where rt + r 2 = at and r ^ 2 = -a 2 .
In the case that r t = r 2 = r ,
we have 2r = SLt and
r 2 = -a 2 ,
cases: 1) r = +1, a 1 = 2 and a2 = - 1 , and 2) r = - 1 , SL^ = -2
so that there are two

and a2 = 1.
either case:
(57)
where:
Q
^m
m\
m-1
m-1
= mr
1Jr
Q ^ = 2rQ _L1 - Q
^m+2
^m+1
^m
Evidently, by factoring and dividing in (56), and then allowing rA to approach
r2:
And, in
1974]

m-1
_ ^
Q m - 2^
m-l-k
ri
r2
mr
m-1
lf
k=0
29
2'
hence, all three cases may be said to be derived from (56).

Now, foregoing the tedious calculation, we give the ternary results (p = 3):
m
i
Q
m ~ {rt - r 2 ) ( r 1 - r 3 )
r
(r 2
m
r2
- r 3 )(r 2 - r t )
(r 3
m
r3
- rt)(r3 - r2)
where
1 - a t x - a 2 x 2 - a3x3 = (1 - rjxMl - r 2 x)(l - r 3 x) .
If two roots are equal, then r 2 = r 3 say, and:
m
m
rt - r 2
Q
=
m
/
\o
(r-L - r2r
(59)
where
m-1
mr 2
,
rt - r 2
1 - SLtx - a2x2 - a 3 x 3 = (1 - rjxMl - r 2 x) 2
And, if r 1 = r 2 = r 3 = r ,
then:
Q
= - m(m - l ) r m-2
^m
2
(60)
v
where
Q 0 = 3rQ ^ - 3Q ^ + rQ
^m+3
^m+2
^m+1
^m
and
r = +1 o r - 1 .
Once again, although now there a r e an infinite number of cases depending on the nature of the
roots, it can foe seen that (59) and (60) can be derived from (58) directly, using in part the
identity:
m
G^ - r 2 ) 2
m
^
m-1
rt - r 2
In summary, then, we can, with minor adjustments in view of multiple roots of the spectral
polynomial, consider that the form (51) actually is the expression for
roots of the spectral polynomial. On the other hand, (10) expresses Q
in t e r m s of the
in t e r m s of the c o -
efficients of the spectral polynomial. That this is a source of a multitude of fascinating p r o b lems is left to the imagination of the reader, as well as to his leisure.
30
[Feb.

PART 3. SYMMETRIC FUNCTIONS
By attacking the entire problem from another point of view, it will be possible to derive
a generalization of the Lucas sequence, and thence derive a set of remarkable identities involved with this generalization similar to the usual Fibonacci-Luc as result that F L = F~ .
Consider, first, a set of complex numbers | r . 1 , and a defining relation:
k
n
(61)
i=1
3 x
The coefficients
_ ,
(x - r . ) = >
J
La
t
(-D's.x*-1
i
k=0
S. are clearly the elementary symmetric functions of { r . |.

S0 = 1,
In particular:
in any c a s e ,
SA = r t + r 2 + r 3 + + r R ,
(62)
S 2 = rt r 2 + rt r 3 + ,
S k = r 1 r 2 r 3 r fe ,
and
By substituting r
= 0,
for
n >k
into (61):
k
= 2>'-V;'
i=l
o r , after multiplying by r
n-k
i=l
Suppose that t
is any linear combination of the { r. }, so that from (63), it is clear that:

k
ttn
i 1
=y^(-l)
" S.t
. .
/ ^
I n-i
i=l
In which c a s e , if we further define a. = (-1) 1
S., we have
(letting k = p):
1974]
(64)
P
= V V t
* i n-1
'
31
i=l
Hence, from (1), a sequence of linear combinations of n

sequence.
powers of
r.
is actually an f-
Further, the Q-sequence defined by (51) is a specific case of linear combination.
It seems reasonable to investigate the properties of the simplest t-sequence, namely,

the simplest linear combination of n
powers of r.,
(65)
which will be called a T-sequence:
n=E r "
5=1
in which case:
T0 = P
Tj = aj
T 2 = 4 + 2a 2 ,
and, in general,
(66)
Remark.
k =
iTk-l
+ a
2Tk-2
+ a
k-lTl
+ ka
Since a. = (-1) " S,, then, in particular,
In addition, it must be remarked that the
a.
for
^ P
ao = - 1 , as was defined e a r l i e r .
defined just before (64) must be integers, in
keeping with the definitions made in the first part of this paper guaranteeing that
the
f-sequences be sequences of integers.

F r o m (7) and (66), it can be seen that h
= -a k (p - k) for k p for any T-sequence.
Immediately, (5 ) becomes
(67)
P-l
k
J2 a k (p - k)x
k=0
ZTkxk
k=0
a, x
and, if s(x) denotes the spectral polynomial

P
a. x
k
k=0
(68)
y^Tk
k=0
x k
P "
' s ! (x)/s(x) .
32
[Feb.
Already 9 from (9):

J^Qkxk
= xP"Vs(x)
k=0
so that clearly, we have:
(69)
^SI^V^^SV"
k=0
k=0
in the derivation of which, a bit of the tedious rearrangement has been passed over.
In addition, noting again that h
, = h, = -a, (p - k) and substituting into (13):

p-1
_,_-,= - V ^ a .k (p - k)Q .
m-p+l
' *
m-k
k=0
P
pF ) a. Q 1 + 7
a. kQ ,
/ J k m-k
/ ^ k in~k
k=0
k=0
but the first term on the right is exactly zero by (1); so:
^ = V^ka._Q
.
m-p+1
/ J
k^m-k
k=0
(70)
T
m
= V ^ k a . Q __, 1 .
/ /
k^m+p-1-k
k=l
Inspecting (70) and looking at various c a s e s leads to the r e m a r k that, in fact, (70) is exactly
generalizati of L
the &
generalization
Lucas identity.
= F
,-,+F
= F +2F
-,
m+1
m - 1n
m
m-1
which is a familiar Fibonacci-
EXAMPLES
What follows now is a rather long discussion of the binary case for T-sequences.
The
most fascinating results occur when p = 2, so that a presentation of this situation is rewarding.
F i r s t , in the binary case:

T
= rt + r 2 ,
where
r^ - a4 rj - a2 = 0, i = 1, 2 ;
1974]
33
or, if rt = r2 = r:
(71)
= 2r
In either case:
n
n
l - r2
\{T1 + r 2 )
T
i~- ~*i
r
Q T
n n
2n
2n
i
" r2
_ _ _
so that:
(72)
Q T
n n
Q 2n
Then, from (70):

ajQ
+ 2a2(
or:
(73)
Q m+1
j.i
2^Q m-1
i
The symmetry of (72) and (73) reveals the underlying charm of the Lucas sequence, which,
of course, carries over to any binary T-sequence.
Continuing, using (73) and (42):
= Q
_,, + aQQ
-m+1
2 -Hi-1
,
v-m+1^
/
v-m-l
-(-a 2 )
Qî-ajt-aj)
Qm+1
-m
(-a2)
(a,
2^m-l
m+r
(-a2)" Tr
or, as in (42):
(74)
Applying (73), we have:
(75)
(aj + 4a 2 )
, n + aQT
n
m+1
2 m-1
while the characteristic expression (27) is:
(76)
v
'
_,0T
- T2 _L1 = (-a 2 ) m (a^ + 4a 2 ) .
m+2 m
m+1
But the general reduction
(28) provides the most elegant formulae,
T-sequences:
Q
^m
(77)
o.i
n+1
+ a
^m^n+1
oQ
J
2^m-l
=
n
2^m-l^n
^
m+n
^m+n
and, taking (74) into account:

T
(78)
= (-a2)
Ll
m-n
Q
= (-a2)
L
^m-n
'
(Q
T - Q T ^ )
^m+l n
^m n+1
(Q M Q - Q Q , - )
^m+l^n
^m^n+1
both for
Q- and
34
[Feb.
so that, adding (77) and (78):

^ + (-a2)nT
=
L
m+n
m-n
Q ^ + (-a 2 ) n Q
=
L
^m+n
' ^m-n
T
(79)
o r , subtracting:
T
(80)
TT
n m
TQ
n^m
^ - (-a2)nT
= Q Q (a? + 4a 2 )
L
L
m+n
m-n ^n ^m 1
n
Q ^ - (-a 2 ) Q
= QT
L
^m+n
^m-n ^n m
cf. (75)
and rearranging index in (79) and (80):

<Wn-k-
<81)
(T2n-(-a2)n-kT2k)/(a?+4a2)
n + k T n - k = %n
n + kVk = %n -
n+kT-k = T2
^"Sk
^
^ )
n _ k T
k
2 k
and, finally:
(82)
Q 9 Q , . = Q*
2n
Remark.
2k
n+k
- Q* . = T n + k " T n - k
n
"k
(a21 + 4a 2 )
The ternary and higher c a s e s yield no such results; that i s , the symmetry
and conciseness do not c a r r y over for p > 2 .
Then, it is clear, the Lucas-Fibonacci r e l a -
tionship i s based almost entirely on the character of the two sequences as binary sequences.
PART 4.
FINITE SUMS
A number of Fibonacci identities are concerned with the formulation in t e r m s of the

Fibonacci sequence of the sum of a certain s e r i e s of t e r m s of the sequence.
For example,
the simplest case:

F
V2-1
k=0
We now seek to generalize this result.
(i)
f a. = y ^ u
m+p
(6)
Recalling e a r l i e r definitions and theorems:
/ ^ k m+p-k
k=l
h . = - 7 a.f M .
m,k
Z4 3 m+k-j
3=0
1974]
and we define: a0 = - 1 and h Q
= h, , so that:
p-1
(5.)
^ ^
1
K
k=0
k=0
But the initial set {f.}
J2 a.x
may be chosen a r b i t r a r i l y , so it is possible to choose for initials
the set { f m + i } where i = 0, 1, 2, , p = 1. In that c a s e , (5f) becomes:
p-^-1
X ) f -m+k
"-xk ~
(83)
k=
p
k
V* a, x
k=0
and rearranging the left member:

P-l
- V1 h
oo
E
o r , multiplying by x :
r^n
k-m _
k=0
P
kX
k=m
,
, x
m,k
J2 \
P"1
(84)
35
Efkxk=
k=
m kx
m+k
P
k
V a. x
k=m
&b
Then, by a simple substitution:
oo
2k=n>
kk
and, subtracting these two expressions:
- 1n
~E
k=0
x > *
n ki
a x
36
p-1
n 1
(85)
[Feb.
k-o
k xk
>
k
ai
i^x
=0 k
Z-u
k=m
Letting x = 1, and assuming that 2 a, f 0:
p^1
n-l
(86)
J\ (hn , k. - h m , .k )
~f^
k=0
> . f, =
k
k=m
Remark.
that 2 a, = 0,
J2 a,
Evidently, the sum in the left member of (86) is finite, so that in the case
the numerator on the right must be divisible by the denominator.
In the event that p = 2, we have the simpler expression:
li
(f - f ) + a2 (f - - f
-)
2
\ ^ f
n
m'
Vn-l
m-1
Z*t f k
- 1 + a 4 + a2
k=n
<*7\
(87)
and
n-l
(83)
^ Q
k=l
Q + a0Q n - 1
^n
2^n-l
a t + a2 - 1
which reduces to the Fibonacci expression when a1 = a2 = 1.
SUMMARY
At the outset, it was proposed to find a generalization from which all the familiar r e sults for Fibonacci-Lucas sequences might be deduced, in addition to which a consistent notation might be developed, and finally, that the sources of the peculiarity of the FibonacciLucas sequences might be found. It is hoped that such proposals are worked out in the course
of the paper.
All that remains to be said concerns the sources of peculiarity which is the
bulk of the charm surrounding the Fibonacci-Lucas sequences.
Of course, some of these
properties stem from the very nature of a recursive sequence of integers (such as (5) and
(27)); while other properties stem from the Q-sequence in particular (for example, (10) and
(51)); while others still come from those formulae which assume different forms when aA =
a2 = 1. Actually, it is quite extraordinary how many of the properties of the Fibonacci-Lucas
1974]
37
sequences a r e shared by a l a r g e r class of sequences.
REFERENCES
1. J a m e s A. J e s k e , "Linear Recurrence Relations, P a r t I , " Fibonacci Quarterly, Vol. 1,
No. 2 (April 1963), pp. 69-74; P a r t II, Vol. 1, No. 4 (Dec. 1963), pp. 35-40; P a r t III,
Vol. 2, No. 3 (Oct. 1964), pp. 197-202.
2. Ken Siler, " Fibonacci Summations," Fibonacci Quarterly, Vol. 1, No. 3 (Oct. 1963), pp.
67-70.
3.
E. P. Miles, J r . , "Generalized Fibonacci Numbers and Associated M a t r i c e s , " Amer.

Math. Monthly, 67 (1960), p. 745.
4.
R. A. Rosenbaum, "An Application of Matrices to Linear Recursion Relations, " Amer.

Math. Monthly, 66 (1959), p. 792 (note).
5. David Zeitlin, "On Summation Formulae for Fibonacci and Lucas N u m b e r s , " Fibonacci
Quarterly, Vol. 2, No. 2 (Apr. 1964), pp. 105-107.
6. David Zeitlin, "Power Identities for Sequences defined by W
= dW
- - cW , " Fib-
onacci Quarterly, Vol. 3, No. 4 (Dec. 1965), pp. 241-256.

7. V. C. H a r r i s and Carolyn C. Styles, "A Generalization of Fibonacci Numbers," Fibonacci Quarterly, Vol. 2, No. 4 (Dec. 1964), pp. 277-289.
8. I. I. Kolodner, "On a Generating Function Associated with Generalized Fibonacci Seq u e n c e s , " IlboaC_Qurter2y, Vol. 3, No. 4 (Dec. 1965), pp. 272-278.
9. M. E. Waddill and Louis Sacks, "Another Generalized Fibonacci Sequence," Fibonacci
Quarterly, Vol. 5, No. 3 (Oct. 1967), pp. 209-222.
10. C. H. King, "Conjugate Generalized Fibonacci Sequences," Fibonacci Quarterly, Vol. 6,
No. 1 (Feb. 1968), pp. 46-49.
11.
T. W. Cusick, "On a Certain Integer Associated with a Generalized Fibonacci Sequence,"

Fibonacci Quarterly, Vol. 6, No. 2 (Apr. 1968), pp. 117-126.
12,. Leon Bernstein and Helmut Hasse, "Explicit Determination of the P e r r o n Matrices in
Periodic Algorithms of the Perron-Jacobi Type with Applications to Generalized Fibonacci Numbers with Time Impulse," Fibonacci Quarterly, Vol. 7, No. 4 (Nov. 1969), pp.
394-436.
13., Hyman Gabai, "Generalized Fibonacci k-Sequences," Fibonacci Quarterly, Vol. 8, No.
1 (Feb. 1970), pp. 31-38.
14. V. E. Hoggatt, J r . , Fibonacci and Lucas Numbers, Houghton Mifflin, 1969.
THE GOLDEN ELLIPSE

H. E. HUNTLEY
Nethercombe Cottage, Canada Combe, Hutton, Weston - Super - Mare, England
The ellipse in which the ratio of the major to the minor axis (a/b) is the golden ratio
<p the "divine proportion" of Renaissance mathematicians [l]has interesting properties.
In the figure, let OA = a = cp units, OB = b = 1 unit. Clearly, if a rectangle be c i r c u m scribed about the ellipse, having its sides parallel to the axes, it will be what has been called
the golden rectangle, frequently realized in ancient Greek architecture.
V
The Golden Ellipse
a/b = cp
The modern symbol for "golden section," as it was called in the nineteenth century, is
the Greek letter phi: cp, and cp and cp1 are the solutions of the equation x2 - x - 1 = 0.
cp = (1 + V 5 ) / 2 = -1.6180
38
Feb. 1974
THE GOLDEN ELLIPSE

<P% = (1 - \[b)/2
39
= -0.6180
so that cp + cp1 = 1 and cpcf = - 1 .

Now, the eccentricity e of an ellipse is given by
b 2 = a 2 (l - e 2 )
so that the eccentricity of the golden ellipse is
(1)
e = 1/ \t<p = N^cp* .
Using the familiar notation of the figure (e. g. , S and S" are the foci), from the known
properties of the ellipse, we may write the following equations:

(2)
OS = ae = \ f ?
(3)
BS = ^(b 2 + a 2 e 2 ) = a = cp .
If L OBS = 6 ,
(4)
sec 6 = a/b = <P .
If ON is perpendicular to the directrix ND ,

ON = a/e = cpz/2
(5)
(6)
SN = ON - OS = a/e - ae = 1/ sfip .
A property of any ellipse may be stated thus: the minor axis is the geometric mean of
the major axis and the latus rectum; that i s , if
L is the length of the semi-latus rectum,
aL - b . Hence, for the golden ellipse,

L = b 2 /a = 1/cp = -cp1
(7)
Thus
a : b : c = cp : 1 : - cp1 = cp2 : cp : 1
From Eqs. (2), (5), and (6),
ON/OS = 1/e 2 = <p
and
OS/SN = a e / ( a / e - ae) = l/cp = -<?f .
Hence, the focus S divides ON in the Golden Ratio.

Again, if PP T is the latus rectum,
(8)
OP 2 = OS2 + SP 2 = a 2 e 2 + b 4 / a 4 = cp + cp'2 = 2 .
40
THE GOLDEN ELLIPSE

Using the cosine rule for A P O P ' ,
Feb. 1974
it may be deduced from this that the latus rectum
subtends at the center O an angle a given by cos a = l/cp , so that, from (4),
(9)
a = 6 .
Another property of the ellipse has it that a tangent at P passes through

cot J_ SPN = e.
Since cot 6 = b / a e ,
it follows in the case of the golden ellipse that
c o t / S P N = l/4<p
so that / S P N = 6.
N and that
and
cot 6 =
l/\lcp9
Thus, MPSB is a parallelogram, and
(10)
MP = BS = a = cp .
Moreover, since MP/PN and OS/SN = <p ,
(11)
PN = 1
and
MN = <p + 1 = cp2 ,
and P divides MN in the Golden Ratio.

It is easily shown that OM = <p so that M lies on the auxiliary circle of the ellipse
and APOM is isosceles.
2
b /ae
Moreover, if OP produced intersects the directrix in D,
= 1, BDNO is a rectangle,
ND =
and P divides OP in the Golden Ratio.
The interested reader may, by searching, discover for himself many other hiding
places where the Golden Ratio is lurking in this ellipse.
For example, superimpose on this
ellipse a second, similar ellipse, with center O but rotated through a right angle.
Draw a
common tangent cutting OY, OX in R, S respectively, to touch the ellipse in Tt
and T 2 .
Examine the ratios of the several segments of RS.
Reference
1. H. E. Huntley, "The Golden Cuboid," Fibonacci Quarterly, Vol. 2, No. 3 (Oct. 1964),
p. 184.
ON A PROBLEM OF M WARD
R. R. LAXTON
University of Nottingham, Nottingham, England
1.
INTRODUCTION
In [3] M. Ward showed that a general non-degenerate integral linear recurrence of o r d e r two has infinitely many distinct prime divisors.
He conjectured that the result was true
for linear r e c u r r e n c e s of higher order (again excluding certain degenerate ones) and, indeed,
confirmed this in [4] for the case of cubic r e c u r r e n c e s .
Here we prove Ward's conjecture;
the method is straightforward and uses the most elementary form of p-adic analysis. We end
by discussing the limitations of the method together with the problems it r a i s e s and posing
further questions concerning divisors of r e c u r r e n c e s (especially in connection with the work
ofK. Mahler).
2.
STATEMENT OF THE PROBLEM
2.1. Let the polynomial f(x) = x
-a
no root nor ratio of distinct roots a root of unity.
~ - - aAx - a0 G Z [x] , m > 1, have

Say
m
f(x) = IT (x - 9 ) ,
i=l
where the 6. are algebraic integers.
the rational field
Put K Q(0i, , 6
); it is a normal extension of
Q.
W = {w 0 , w l 5 , wn> } is a (integral) linear recurrence with companion polynomial f(x) if given w 0 , w l5 , w ^ ^ G Z , not all z e r o , we have
w,
= a
-, w ^
+ .. . + a n w , T + aAwM ,
n+m
m - 1 n+m-1
1 n+1
0 n
for all non-negative integers n.
2.2.
Thus all the t e r m s of
W are rational integers.
We can assume that a0 f 0 since otherwise we would have a linear recurrence of
degree less than m.

All the roots 9V , 0 m a r e distinct, so we may write
(2.1)
Dw
= A ^ f + .. . + A
n
: l
0n
mm
for all n, where the A, are algebraic integers in K and the rational integer D is the d i s criminant of f(x).
41
42
ON A PROBLEM OF M. WARD
[Feb.
If at most one of the A. of (2.1) is not z e r o , then we have a degenerate recurrence

and this we exclude.
Hence we may assume that AjA2 * A m f- 0 and m ^ 2 since other-
wise we would use (2.1) but with the zero A. f s deleted.

I
To make the exposition clear, we shall assume that f(x)

v '' ' > ^ m
are
splits in Z [ x ] ,
i. e. , that
rational integers. The following proof remains valid in general (with prime
ideals replacing rational p r i m e s , etc.) apart from one step and this we shall deal with at the
end of the present proof.
2.3. An integer n is a divisor of W if n divides some t e r m w
oi W. We shall
be concerned with prime divisors of W. It & prime p divides all the roots Qi9 ,0
then
p divides all the t e r m s w, of W with t ^ m; these divisors (which a r e called null-divisors)

are of no interest to us and we eliminate them.
Let u = g. c. d. (01? , 0 m ) and rewrite
(2.1) as
(2.2)
with 6. = 0. / u for all i = 1, , m. It follows that given any prime p, there is at least
one 8. which is not divisible by p. This fact we will need in the subsequent proof.
2.4.
F r o m now on we will assume that the recurrence W
finite number of prime divisors.
given by (2.1) has only a
It follows from (2.2) that there are only a finite number of
p r i m e s , say p 1? , p^., which a r e prime divisors of the integers U
= A18\ + . . . + A ^ S - ^
for all n = 0, 1, 2, . Essentially we prove that this assumption implies that the t e r m s
U
assume the same integer value for infinitely many distinct values of n.
3.
3.1.
THE ANALYSIS
Let p = p.. be one of the p r i m e s p l 5 , pt which divide some U .
construction in 2.3, we know that some
From the
5. is not divisible by p say 8 j , , 8^ (d =
d(i) ^ 1) are p-adic units and 6 , - , , 5
a r e divisible by p.
For the moment, we will assume that

(3.1)
A$
+ + A d 5 ^ = 0
for only finitely many n E Z ,
n ^ 0.
Let k = k(i) be such that A t 6^ + + A 8^ f 0. Say A ^ + + A 5^ = 0 (mod

k
k
s+1
p ) but A1S1 + - + A , 5 , ^ 0 (mod p
) for some integer s = s(i) ^ 0. For each j =
1, , d, there exists a positive integer b. such that
s
b.
8.J = 1
(mod p
).
Now put b = b(i) = b ^ tyj. Then for each r E Z such that v = k + rb > s, the rational integer
1974]
a
mm
x 1
43
dd dd
= A ^ + . . . + A,sJ
(mod p S + 1 )
5
d d
Thus for all v = k + rb
>
the t e r m s U
are exactly divisible by p S .
3.2. Now repeat the argument of 3.1 for each of the primes
pl5 , p .
It is clear
that provided the assumption (3.1) holds for each of these p r i m e s , the value selected for k =
k(i) can be chosen to be the same for all p 1? , p . Assuming then that (3.1) holds for each
p.., i = 1, , t, we have constructed a subsequence
U ,.x of f u ,
M
i
all r E Z with v(i) = k + rb(i)
>
v(i)
n = 0, 1, J}
s(i), all of whose t e r m s a r e exactly divisible by p
Therefore for all r G Z such that v = k + rb(l)b(2) b(t)

the infinite subsequence ( U v ) of ( u } takes on the form
>
for
( )
max (s(l), , s(t)),
N for some fixed integer N
(since the p r i m e s p l 5 , p, a r e the only prime divisors of t e r m s of this sequence {u }

t
n
of rational i n t e g e r s ) .
3.3. Now both the derivation in 3.2 and the denial of assumption (3.1) for some prime
among p l 5
, p
give rise to statements of the form: "AjS 1 + ... . + A f 5 f takes the same
value for infinitely many n E Z, n 0. n Here A. and 6.
are non-zero algebraic inte-
gers (actually we have assumed they are rational; see 2.2) and f 2 (f = m 2 for the
derivation in 3.2 and for (3.1) to be false we must have f = d(i) ^ 2).
By p-adic methods (see for example K. Mahler's article [l]) we can conclude from
this that some ratio 5, / 5 . , i ^ j , is a root of unity and hence 9. /9. = u5. / u 5 . is a root
of unity.
This contradicts our initial hypothesis and so the assumption that the recurrence
W has only finitely many prime divisors is false.

4.
REMARK ON THE GENERALIZATION OF THE PROOF
We need consider the case when f(x) does not split in Z[x] and so 6l9 , 9
not rational integers but only algebraic integers.
ideals p of the normal extension K
stead of p-adic analysis.
As we remarked in 2.2, we use prime
instead of rational primes p and p-adic analysis
in-
This part of the generalization causes us no trouble but there is a
slight difficulty in getting rid of the null-divisors of W in 2.3 and forming Eq. (2.2).
we put u = g. c.d.
are
(0 ls , 0 m )
and subsequently considered the sequence
There
= A1&1 +
+ A 6 n of rational integers
and the fact that the U a r e rational integers is impor&
m m
n
tant in Sec. 3.2 where we used the fact that the only units in the rationals are 1 (and t h e r e by deducing that we obtained an infinite subsequence { U v } of { U n } taking the values N for
some fixed N E Z ) .
To overcome this we let qu
rational primes dividing g. c. d.

extension K = Q(0 l 9 , 0 m )
(a 0 , - , a m _ 1 ) ,
, q s be the set (possibly empty) of all

the coefficients of f(x).
In the normal
let the ideal (q,) have prime ideal decomposition

a.
<V = <Ji(i)--- qi(r)) '

Now each prime ideal _q.,,. contains all 9V , 9m;
-2i(k) d
ivi(
in
* S ^ ^1 > ' " ' ' ^m-
e Z
'
i "
let jS.,,, > 1 be the highest power of
Since K is normal we have
44
"id)
say.
=/3
i ( 2 ) =
We do this for all i = 1, , s.
= V )
Put a = a^
=/3
. . -as
Feb. 1974
'
and then
TT(M-where u is a rational integer.
i,k\-3i(k)/
Now instead of considering all t e r m s
of W we consider only the subsequence
(w
} and (2.2) then becomes
L
cmJ
Dw
an
= A<u*
l 1
+ + A 0
mm
with 5. = 9, / u for all i = 1, , m.
= u \Afii
1 1
+ + A 5 ) ,
mm
Here { AJSJ + + A 5
is a sequence of r a -
tional integer t e r m s and for any prime at least one 5. is a p-adic integer.
The analysis
can now proceed as previously.

5.
5.1.
PROBLEMS CONCERNING FURTHER GENERALIZATIONS
One would suppose that the result established here can be generalized to a r b i t r a r y
linear r e c u r r e n c e s , not just those W
all of whose t e r m s are integers.
However,
our
method of proof breaks down in this general situation since in Sec. 3.2, we needed the fact
that there a r e only a finite number of units in Z
5.2.
In [2] , K. Mahler has shown (using the p-adic generalization of Roth's Theorem)
that in a non-de gene rate linear recurrence of order two (with c. p. x2 - Px + Q, 4p > Q2 and
Q ^ 2) every infinite subsequence has an infinite number of prime divisors. Again one would
suppose that this is true for linear r e c u r r e n c e s of higher o r d e r .
5.3.
Let f(x) = x2 - Px + Q e Z[x] , Q ^ 0, and
w 0 , wA E 3E be a linear recurrence satisfying w
= Pw
W = { , w 0 , w l5 , w n , }
- - Qw , for all n E Z
(we are
allowing the recurrence to go in both directions so that now not all t e r m s are integers). Then
one can establish that if every prime is a divisor of W and Q = 1, then some t e r m of W
is 0 and so it is essentially the Lucas sequence
, 0, 1, P ,
associated with
f(x).
Is this result true for a r b i t r a r y Q ?

REFERENCES
1.
K. Mahler, "Eine arithmetische Eigenshaft der Taylor koeffizienten rationaler Funktionen, M Proc. Amsterdam Acad. , 38 (1935), pp. 50-60.
2.
K. Mahler, "A Remark on Recursive Sequences," J. Indian Math. Soc. (Dehli) 1.
3.
M. Ward, " P r i m e Divisors of Second-Order Recurring Sequences," Duke Math. J. , 21
4.
M. Ward, "The Laws of Apparition and Repetition of P r i m e s in a Cubic Recurrence,"
(1954), pp. 607-614.

Trans. Amer. Math. Soc. , 79 (1955), pp. 72-90.
GENERALIZED HIDDEN HEXAGON SQUARES

A. K. GUPTA
The University of Michigan, Ann Arbor, Michigan 48104
The triangular a r r a y of binomial coefficients is well known.
Recently, Hoggatt and
Hansel [2] have obtained a very surprising result involving these numbers.
Cowan [3] and Gupta [ l ] have generalized this triangular a r r a y to a tableau.
Stanton and
In this paper,
we generalize the results due to Hoggatt and Hansel.

Let, for any positive integer m and any integer
n
<
n,
J =. 0 if either
n > m
or
0. Then we prove the following theorem.

Theorem.
The product of the six binomial coefficients spaced around (

r
m n
i \ /
- rA/
n
2 / \
/ \
m
n
\ /
r
"
2 / \
+ r2\ / m + r2\ /
m
n
i / \
/ \
J , viz. ,
\
r
i/
where rt and r 2 are positive integers, is a perfect integer square.

Proof.
The product of the six binomial coefficients is

(m - r j ) !
(m - r A )!
(n - r 2 )t(m - rt - n + r 2 ) !
(m)!
(n)!(m - rt - n)\ ' (m - r 2 ) ! (m - n + r 2 ) f .

i2
r
(m - r ^ ' . N K m + r 2 ) !
"j
L (n - r 2 )t(m - rt - n + r 2 )t(n)!(m - ri - n)!(m - n + r 2 )!(n + r ^ t j
Now, the product of binomial coefficients is an integer, since each binomial coefficient
is an integer.
And the square of a rational number is an integer if and only if the rational
number is an integer.
Hence the product is an integer square.
It is interesting to note that
m
n
"
r 2
\/m
/ \
i\/m
/ \
z\
+ F l
"
i\/m
r 2
/\
2\/
/ \
\
r i
'
which is what really happens to make the product of six numbers a perfect square.
Corollary l.
If
rt = r 2 ,
equally spaced around (
we get the product of six binomial coefficients which are
Corollary 2. If rt = r 2 = 1, we get the product of six binomial coefficients that s u r round I
J.
This is the result of Hoggatt and Hansel [ 2 ] .
special case of our general theorem.

45
Hence their result is a very
46
GENERALIZED HIDDEN HEXAGON SQUARES

By taking different values for rt and r 2 ,
Feb. 1974
we can obtain several configurations which
yield products of binomial coefficients which are squares.
In fact, one can build up a long
serpentine configuration, o r snowflake curves, as noted by Hoggatt and Hansel.

Note that the theorem holds for generalized binomial coefficients (and hence for qbinomials), and in particular for the Fibonomial coefficients.
REFERENCES
1. A. K. Gupta, "On a 'Square' Functional Equation," unpublished.
2. V. E. Hoggatt, J r . , and Walter Hansel, "The Hidden Hexagon Squares," Fibonacci Quarterly, Vol. 9, No. 2 (April, 1971), pp. 120 and 133.
3.
R. G. Stanton and D. D. Cowan, "Note on a ' S q u a r e ' Functional Equation," Siam Review,
Vol. 12 (1970), pp. 277-279.
LETTER TO THE EDITOR

Dear Editor:
Here are two related problems for the Fibonacci Quarterly, based on some remarkable
things discovered last week by Ellen Crawford (a student of mine).
Problem 1.
Prove that if m and n are any positive integers, there exists a solution
x to the congruence
F
Solution.
= m (modulo 3 ) .
Let m be fixed: we shall show that it is possible to solve the simultaneous
congruences
F
= m (modulo 3 )
(*)
,
F
+ F , - f 0 (modulo 3) .
x
x+1 '
This is clearly true for n = 1. It is also easy to prove by induction, using

Y
= F
m+n
(Continued on page 79.)
F
m-1
+ F
n
F
n+1
DIAGONAL SUMS OF THE TRINOMIAL TRIANGLE

V. E. HOGGATT, JR., and MARJORIE BICKNELL
San Jose State University, San Jose, California 95192
In an e a r l i e r paper [ l ] , a method was given for finding the sum of t e r m s along any
rising diagonals in any polynomial coefficient a r r a y , given by
(1 + x + x2 + - + x 1 "" 1 ) 11 ,
n = 0, 1, 2, ,
r > 2 ,
which sums generalized the numbers u(n; p,q) of H a r r i s and Styles [2] , [ 3 ] . In this paper,
an explicit solution of the general case for the trinomial triangle is derived.
If we write only the coefficients appearing in the expansions of the trinomial (1 + x +
2
x ) , we have
1
1
1 2
1 3
10
16
19
16
10
15
30
45
51
45
30
15
Call the top row the zero
row and the left-most column the zero
column.
Then, the col-
umn generating functions are
G0 =
(1)
_ x,
Gt = ,
(1 - x) 2
Gn+2 = j - ^
(Gn+1
G2 =
G n >.
: ,
(1 - x) 3
n > 0
We desire to find the sums u(n; p,q) which are the sums of those elements found by
beginning
ning in the zero
column and the n
row and taking

takin steps
right throughout the left-justified trinomial triangle.
(2)
G = ^x
n p
n q
Let
= ^ u ( n , p,q)x n
n=0
n=0
47
p units up and q
units
48

Our first problem is to find a recurrence for every q
[Feb.
column generator.
We need
two sequences,
(3)
n
nn
) = &^ _ -^ B_ ,
~Q ,n (v x_) = n +, / 50n
Both P (x) and Q (x) obey
V2(X)
l~I ( Vl ( x )
U (x)1
So let
then
P , 0 (x) = A(P , n (x) + P (x))
n+2
n+1
n
Q , 0 (x) = A(Q Ax) + Q (x))
^n+2 n + 2
^n+1
^n
a
= A(an+l + c n )
(4)
^n+2
A ( j J n+l
+ i3n)
Next, we list the first few members of P (x) and Q (x).

n
TLV
n
Q
^ n (x)
P (x)
,n
A2 + 2A
2
A3 + 3A2
A 3 + 2A2
4
+ A
3
A4 + 4A3 + 2A2
2
A 5 + 4A 4 + 3A3
+ 3A
+ A
A 5 + 5A4 + 5A3
A6 + 6A5 + 9A4 + 2A3
Note that the coefficients of Q (x) are simply the t e r m s appearing on rising diagonals
of the Lucas triangle [ 4 ] . The coefficients of P (x) and Q (x) are the same as those of the
Fibonacci and Lucas polynomials, and P (1) = F ,
Q (1) = L ,
Lucas number, respectively.

By mathematical induction, it is easy to show that
(5)
Q n (x) = P n + 1 (x) + AP n _ 1 (x)
Then, the general recurrence for the k
t e r m s is
the
Fibonacci and
1974]
(6)
u k ( n + 2 ) ( x ) = Q k (x)u k ( n + 1 ) (x)
Then, a recurrence relation for every q

(7)
(-l)
k+1
V(x)
49
.
column generator is
+ (-l)q+1AqG
G , MOAx) = Q (x)G t M,Ax)

q(n+2)
q
q(n+l)
(x) .
qn
In summing elements to find u(n; p,q) from the column generators, we need to multiply the column generators by powers of x so that the coefficients summed lie along the
chosen diagonals of the trinomial a r r a y .
(8)
G* ( n + 2 ) (x)
Then
= xP Q q (x)G* ( n + 1 ) (x)
G *(x)
x2P(-l)q+1AqG*n(x) ,
G*(x) = x P G q (x)
= _ L _ ,
Let
G
n
= >
mmd
G*
iq
i=0
and
lim G = G ,
n oo n
the generating function for the numbers u(n; p,q).
XX
We next sum Eq. (8),
XX
i=0
X )
xPQ
i=0
XX
(X)G
qU + l)
(x)+
x2P( 1)q+lAqG
i=0
Qi (X) '
which becomes, upon expansion,

G n - G*(x) - G*(x)
= xPQ q (x)G* n + 1 ) q (x)
G* n+1)q (x)
+
G* n+2)q (x)
x P Q q ( x ) G n - xPQ q (x)G*(x)
+ x2p(-l)q+1A(3Gn .
Collecting t e r m s , our sum simplifies to
G (1 - x P Q (x) - x 2 p ( - l ) q + 1 A q )
n
^q
where R
involves only t e r m s involving G^
= G*(x)(l - x P Q (x)) + G*(x) + R ,

0
q
q
n
^ (x) and G^
lim G*(x) = 0
n oo n
?\
(x). It can be shown that
50
for Ixl < 1/r,
r > 2, so that
lim
=0.
Feb. 1974
Then, taking the limit as n oo of our sum
and simplifying,
G*(x)(l - x P Q (x)) + G*(x)
G = -JJ
-2
1 - x P Q q (x) + x ^ P ( - A ) q
which becomes Eq. (9) from the identity given in Eq. (8):
G(x)(l - x p Q ( x ) ) + x P G n (x)
"
G = -2
2
JL_ = V
u(n; p,q)x n ,
1 - x P Q (x) + x 2 p ( - A ) q
j ^
(9)
where G (x) is defined by Eq. (1), A = =
and
[(k+l)_/2] r
(10)
REFERENCES
1. V. E Hoggatt, J r . , and Marjorie Bicknell, "Diagonal Sums of Generalized Pascal T r i a n g l e s , n 5Tbojiaci_Qua^
Vol. 7, No. 4, Nov. 1969, pp. 341-358.
2. V. C. H a r r i s and Carolyn C. Styles, n A Generalization of Fibonacci Numbers," Fibonacci

Quarterly, Vol. 2, No. 4, Dec. 1964, pp. 277-289.
3.
V. C. H a r r i s and Carolyn C. Styles, "Generalized Fibonacci Sequences Associated with a

Generalized Pascal T r i a n g l e , " Fibonacci Quarterly, Vol. 4, No. 3, Oct. 1966, pp. 2 4 1 248.
4. Verner E. Hoggatt, J r . , "An Application of the Lucas T r i a n g l e , " Fibonacci Quarterly,

Vol. 8, No. 4, Oct. 1970, pp. 360-364.
FIBONACCI SEQUENCES MODULO M

AGNES ANDREASSSAN
American University of Bierufr, Bserut-, Lebanon
Most of the questions concerning the length of the period of the r e c u r r i n g sequence obtained by reducing a general Fibonacci sequence by a modulus m have been answered by D.
D. Wall [ l ] .
The problem discussed in this paper is to determine the number of ordered
p a i r s (a,b) with 0 ^ a < m
and 0 ^ b < m
that produce these various possible lengths.
The results that have been used in this study a r e summarized below.
The proofs of
these theorems are omitted here except for ''Theorem 12" whose proof in [1] is incorrect.
The outline of a c o r r e c t proof of "Theorem 12 n was proposed by D. D. Wall in answer to a
letter sent to him asking for clarification.
SUMMARY OF KNOWN RESULTS
Using the notation in [l] , let f
f0 = a,
ij = b, and f
- = f +f
...
denote the n
term of the Fibonacci sequence where
Let h = h(a,b,m) denote the length of the period of
this sequence when it is reduced modulo m ,
taking least non-negative residues.
does not depend on a and b we may write h = h(m)
instead.
When h
The special Fibonacci s e -
quence which s t a r t s with the pair (0,1) will be denoted by {u } and its period when reduced
modulo m by k(m). The sequence which s t a r t s with (2,1) will be denoted by ( v }.
letter p will be used to denote a prime and e a positive integer.
The
In studying the possible
values of h ( a , b , m ) w e m a y a s s u m e , without any loss of generality, that (a,b,m) = 1.

1. If
e.
lip. 1
e. \
and hfa, b, p . 1 J = h.
then h ( a , b , m ) = LCM[h.] [ l , Theorem 2 ] .

2. If t is the largest integer such that k(p ) = k(p) then k(p e ) = p e ~ \ ( p ) for e ^ t
[ 1 , Theorem 5 ] .
Remark. The proof of this theorem as given in [l] is rather incomplete. It is possible
to give a complete proof by using induction on e as suggested, but a much neater proof for
the case when p is an odd prime is given by Robinson [2] , by the use of matrix algebra.
6+1
e
e
For p = 2, Robinson 1 s proof that k(p
) is either k(p ) or pk(p ) still holds, and
the proof that shows that if k ( p e + 1 ) = pk(p e ), then k ( p e + 2 ) = pk(p 6 + ) is Still applicable
for e > 1. The case p = 2 and e = 1 is verified by direct computations since we have
k(2) = 3, k(2 2 ) = 6 ,
and k(2 3 ) = 12.
In particular, if k(p 2 ) f k(p), we obtain
k(p ) = p
k(p).
In [3] Mamangakis has
shown that (1) if c and p are relatively prime and cp occurs in {u }, then k(p 2 ) f k(p),
51
52
and (2) if c and p a r e relatively p r i m e ,
e d,
[Feb.
and u. = cp
is the first multiple of p
to occur in ( u }, then k(p ) = k(p) if and only if u. ^ has the same o r d e r

e
mod p .
mod p
and
F o r all p up to 10,000 it has been shown that k(p ) f k(p). However, it has not
yet been proved that k(p 2 ) = k(p) is impossible.

3. If m > 2, then k(m) is even [ l , Theorem 4 ] .
4. If (b2 - ab - a 2 , p e ) = 1, then h(p e ) = k(p e )
5. If p E 3 (mod 10), then h(p e ) = k(p e )
6. h(2 e ) = k(2 e )
2
[ l , Corollary to Theorem 8 ] .
[ l , Theorem 8 ] .
[1, Theorem 9 ] .
7. If b - ab - a ^ 0 (mod 5), then h(5 e ) = k(5 e ); and if b 2 - ab - a2 = 0 (mod 5),

then h(5 e ) = (l/5)k(5 e )
[1, Theorem 9 ] .
e
e
8. If m = p , p > 2, and if there is a pair (a,b) which gives h ( a , b , p ) = 2t + 1,
then k(p e ) = 4t + 2 [ l , Theorem 10].
9. If m = p , p > 2, and if k(p ) = 4t + 2 then h ( a , b , p ) = 2t + 1 for some pair
(a,b)
[1, Theorem 11].

10. If m = p , p > 2, p f- 5, and h is even, then h(p ) = k(p ) [ l , Theorem 12].
Proof.
Since f, = u, . a + u . b , we have
h
h-1
h
(1)
(mod p e ) ;
fh - a = bu h + a(u h _ 1 - 1) = 0
(2)
fh+1
(mod p e ) .
- b = b ( u h + 1 - 1) + au h = 0
Since we are assuming that ( a , b , p ) = 1, considering a and b as the unknowns, the d e Hence u2 - (u, ,. - l)(u, _, - 1) = 0 (mod p ). But it is known
,
h
h+1
h-1
,
^
= (-1) " , and so u, . + u,
= 1 + (-1)
(mod p ). Since h is even
terminant must be zero.

that u2 - u , + 1 u ,
and u, - = u, + u, - ,
this gives
2u, _ 1 + u, = 2 (mod p ),
or
u, = 2(1 - u, ..) (mod
p ). It has been shown that if b2 - ab - a2 ^ 0 (mod p) we obtain the unique solution u, = 0

and
u,
= 1 (mod p ),
and so h(p ) = k(p ).
b - ab - a = 0 (mod p). Since u, = 2(1 - u, _ 1 )

2b(l - u h
) + a(u h
- 1) = 0 (mod p e ) ,
Next consider the cases for which
(mod p ), substituting in (1) we obtain
or
(2b - a)(l - i ^
) = 0 (mod p e ) .
We will show that (2b - a, p ) = 1. The condition b 2 - ab - a2 = 0 (mod p) can be written

in the equivalent form (2b - a) 2 = 5a2 (mod p). Now if p|(2b - a), then p|5a 2 ; but p ^ 5,
hence p|a.
sumption.
Therefore p|2b, and since p
>
2, p|b.
Thus
( a , b , p ) ^ 1 contrary to a s -
Hence p/f (2b - a), and so we may cancel 2b - a from the above congruence ob-
taining 1 - u,
=0
(mod p e ) , or u,
= 1 (mod p e ) . Since u, = 2(1 - u,
) (mod p e ), :
this implies that u, = 0 (mod p ), and so again h(p ) = k(p ) .

11. If h(a,b,p) = k(p), then h ( a , b , p e ) = k(p e )
12.
[ l , Corollary 2 to Theorem 12].
Let f(m) denote the smallest positive integer,
and let p be an odd prime.
If 2/f(p),
then k(p e ) = 4f(p e )
n,
for which
[4].
= 0 (mod m),
1974
53
THE PROBLEM
For any given modulus
these m
m,
m 2 possible ordered pairs in sequence.
there are
Of
ordered p a i r s we would like to determine the number of pairs corresponding to
each of the various possible lengths for that modulus.
For example, if m = 7 we obtain
0, 0,
length
(1 pair)
0, 1, 1, 2, 3, 55 1, 6, 0 S 6 9 6, 5, 4 , 2 , 6, 1, 0,
length 16 (16 pairs)
0, 2, 2, 4, 6, 3, 2, 59 0, 5, 5-, 3, 1, 4, 5, 2, 0, -
0, 3, 3, 6, 2, 1, 3, 4 5 0, 4, 4, 1, 5, 6, 4, 3, 0,
Hence 1 pair produces a sequence of length 1 and 48 pairs produce sequences of length 16.
Viewing these as infinite sequences extending to the right as well as to the left, some of
these sequences become indistinguishable.
Thus instead of number of p a i r s it is convenient
to talk about number of distinct sequences of a given length.
In the above example, there is
1 distinct sequence of length 1 and there are 3 distinct sequences of length 16.
Let n(h,m) denote the number of distinct sequences of length h when the sequence is
reduced mod m.
This will be abbreviated to n(h) when it is clear what modulus is used.
Thus the problem is to determine the values of n(h) corresponding to the various possible
values of h for any given modulus m.
e
( a , b , p ) = 1,
Since the results summarized from [1] hold when

what happens when
( a , b , p ) f 1.
When m = p,
we must consider
there is only one pair, namely
(0,0),
with (a,b,p) f 1 and it produces a sequence of length 1. When m = p 2 , then sequences for
which ( a , b , p ) = 1 are all the sequences for
mod p multiplied throughout by p.
When
m = p 3 , the sequences for which (a,b,p 3 ) f 1 a r e all the sequences for mod p 2 multiplied
throughout by p. Thus, in general when m = p we can t r a c e back all the sequences except
e
e1 e2
the one arising from (0,0) to pairs for mod p , p " , p " , - - - , p where the condition of
being relatively prime holds.
The pair (0,0) will always have length 1 no matter what the
modulus is.
We shall henceforth abbreviate k(p) as k.
Theorem 1. Let m = p e where p = 2 or
p = 3 (mod 10).
If k(p 2 ) ^ k(p)
then
n(l) = 1 and
. i. v
n(p k) =
2
p (p
F
1)
for i = 0, 1, , e - 1.
Proof.
By 5 and 6, if p = 2 or p = 3 (mod 10) and if ( a , b , p 6 ) = 1, then h ( a , b , p G )
= k(p ). If ( a , b , p ) f 1, then we still have h ( a , b , p ) |k(p ). Since

2
possible values of h ( a , b , p ) a r e 1, k, pk, p k, , p " k.
k(p ) = p " k,
the
We know that there is always
54
[Feb.
one sequence of length 1, namely when a = 0 and b = 0. Thus n(l) = 1. We will show that
e1
e
all of the n(p
k) sequences come from cases where ( a , b , p ) = 1. We know that the s e e
e1
quences for which (a 9 b,p ) f 1 a r e the same sequences as for mod p "
multiplied
throughout by p, and these sequences have the same lengths as the corresponding sequences
for mod p "~ . Since none of the sequences for mod p " has a length g r e a t e r than k(p ~ )
e2
e
e1
= p
k, no sequence for which (a, b , p ) ^ 1 can have a length of p
k. Moreover, all
the sequences for which (a, b, p ') = 1 have lengths of p " k and so a r e included in
ntp6"1^.
Since 2n(h.)-h. = m 2 where h. a r e the different possible lengths, we must have
e-1
i + J2
n(
p i k ) pik
p2e
i=0
and
e-2
i XT* / ii \
ii
2e-2
1 + 2_j nvP k) p k = p
i=0
Subtracting we obtain
n(p " k) p " k = p
(p2 - 1)
and so
n(p e-l k)
=P^V-D
e2
e3
o
Now since n(p " k), n(p
k), , n(p k) represent the numbers of the sequences
for which (a, b, p ) ^ 1, they correspond to the sequences for mod p " . But for mod
e1
e2
e~1
p
, the sequences that have lengths of p " k are those for which (a 1 , b ' , p
) = 1
e-2
where a = paf and b = pb ! . The number of these sequences gives n(p
k). Hence we may
use the formula derived above and obtain
e-2. ,
p e - 2 ( p 2 - 1)
n(p
k) =
't
Thus in general for mod p
we have
1974

n(p i k)
55
= P V -D
for i = 0, 1, , e - 1.
Since k(2) = 3 and k(2 2 ) f k(2) we have:
For mod 2 e , n(l) = 1 and n(3-2 1 ) = 2 1 for i = 0, 1, , e - 1.
Corollary.
Theorem 1 ? . Let m = p
where p = 3 (mod 10).
If t is the largest integer such
that k(pfc) = k(p) with t > 1, then (1) for e < t, n(l) = 1 and
n ( k)
2e
= P z - i
and (2) for e > t, n(l) = 1,
n(k) = B ^ r = - i - ,
and
nfp1"^
= P1+t~V - D
for i = t, , e - 1.
Proof.
(1)
For
e < t,
have length k.
k(p ) = k(p) and so all the sequences except the
Since 2h.n(h.) = p
(0,0)
sequence
this means
2e
e-t
(2) For e > t, the possible lengths a r e 1, k, pk, , p
of the sequences for mod p can be identified as the lengths for
k. Since all the lengths

mod p , p " , , p , p
where (a, b, m) = 1, we have:

For mod p, n(l) = 1.
For mod p , n(l) = 1 and n(k) = (p2 - l ) / k .
t
2t
For mod p , n(l) = 1 and n(k) = (p - l ) / k .
For mod p t + 1 , n(l) = 1, n(k) = ( p 2 t - l ) / k , and
2t 1
(p - (p2-l))/k.
For mod p t + 2 ,
n(pk)
n(l) = 1,
2t-l, 2
t\
P
~
(ELLJ>
k
n(pk) = ( p 2 t + 2 - p 2 t ) / p k =
n(k) = ( p 2 t - l ) / k ,
2t+4
and
n(p2k)
^P
p2k
e
2t
Therefore, for mod p , n(l) = 1, n(k) = (p - l ) / k ,
2t+2
2 t / 9 ,v
. , P fc - 1)
k
and
56

. , (1
i + t - 1 / 2 -.\
, l-t+1. v
P
(P^ - 1)
n(p
k) = -
'F
-
,
for
[Feb.
.
,
l = t, , e - 1 .
Theorem 2. If m = 5 e 5 then n(l) = 1, n(4) = 1, n(4-5 1 ) = 6-5 1 " 1 for
i = 1, ,
e - 1, and n(45 ) = 5 " .

Proof,
by 7 that if
We always have n(l) = X* With the assumption that (a, b, 5 ) = 1 we know

(b2 - ab - a25 5) = 1
then h(5 e ) = k ( 5 e ) ,
and if
(b2 - ab - a 2 , 5) f 5 then
h(5 e ) = ( l / 5 ) k ( 5 e ) .
e
It can be shown that the assumption (a, b, 5 ) = 1 is superfluous in the first case b e cause if (a, b, 5 ) f X, then 5|a and 5|b; hence 5|(b2 - ab - a 2 ) contradicting (b2 - ab a 2 , 5) = 1. Thus, if (b2 - ab - a 2 , 5) = 1, then (a, b, 5 e ) = 1.
In generalj we know that there a r e p - p
p a i r s (a,b) with (a,b,p ) = 1. We
2e
2e2
wish to determine how many of these 5 - 5
pairs give (b2 - ab - a 2 , 5) = 5. This is
equivalent to b 2 - ab - a2 = 0 (mod 5), or (2a + b) 2 = 5b2 (mod 5), or (2a + b) = 0 (mod 5)..
e
Hence b = -2a (mod 5), or b = 3a (mod 5). Thus If (a, b, 5 ) = 1 and
(b2 - ab - a 2 , 5) = 5,
e
eX
eX
a can take 5 - 5
different values and corresponding to each value of a, b can have 5
values. Therefore, there will be 5 " (5 - 5 " ) = 4.5 "
such pairs (a,b). Since the
total number of pairs (a,b) for which ( a , b , 5 ) = 1 is 5
- 5
and all the cases for
which (b2 - ab - a 2 , 5) = X a r i s e from these, the number of p a i r s (a,b) such that (b2 - ab a 2 , 5) = X is given by 5
- 5
- 4=5
"
= 4-5 e~~ . This is the number of pairs that
produce sequences of length k(5 '). Since k = k(5) = 20, k(5 e ) = 5 e ~ k = 4-5 e and so
4.526"1
e-X
u _e*
n(4-5
) = - = 5R
e
4-5
We have 4*5 e ~
pairs with (a, b, 5 ) = X producing sequences of length -|k(5 ) = 4-5 " .
There a r e also the cases for which (a, b, 5 ) f X. But there a r e the sequences for mod
eX
eX
eX
5
multiplied throughout by 5. Since k(5
) = 4-5
the number of p a i r s that produce
e-X
2e-2
2(e-X)-X
sequences of length 4-5
Is given by 4-5
+4-5
and so
,A C e - L
45
+ 4*5
n(4-5
) =
n _e-2
= 6-5
45e*"1
5
We have
-|k(5 1 + 1 ) = k(5X) = 4-5 1
for
n(45 ) =
i = X, 2, , e - X and so
+ 4.52(i+l)-3
r
=6-5
4.52(i+l)-2
4-5 1
1974]
57
for i = 1, 2, , e - 1. In addition to these there are the pairs that produce sequences of
1
length ik(5)
= 4. The number of such p a i r s is 4-5
, where e = 1.
Hence,
n(4) = (4.5)/4 = 1 .
Theorem 3. Let m = p e where p = 1 (mod 10). If k(p 2 ) f k(p) then
(1) If 4Ik, n(l) = 1 and
n(pik)
for i = 0, 1, , e - 1
(2) if 4|k 5
= P, V Z J>
and
n(l) = l ,
n(pik/2)
2fc>zJ>
and
n(pi k ) = (P - I H P ^ 1 + p 1 - 1)
for i = 0, 1, , e - 1.
Proof.
By 3, k(p ) is even, and so it is either of the form 4t or of the form 4t + 2.

e
e
e
(1) If k(p ) = 4t, by 8, h(p ) cannot be odd; and if h is even then by 10, h(p ) =
i-*
t*
iP
f1
k(p ). Thus on condition (a, b, p ) = 1, h(p ) = k(p ) = p

k, and so the proof of
Theorem 1 is applicable here. We also note that the condition 4|k(p ) is equivalent to that
__ p e -"1 k. and, 02|p'
I^ e-1
of 4 |k since k(p e .) =
(2) If k(p e ) = 4t + 2, by 9 h(p e ) = 2t + 1 for some (a,b).
By 4 if
(b2 - ab - a 2 , p e )
= 1, then h(p ) = k(p ). Now consider (b2 - ab - a 2 , p ) ^ 1; if h(a, b, p ) is even, by

10 h(a, b, p ) = k(p ); and if h(a, b, p ) is odd, by 8, h(a, b , p ) = 4k(p ).
Let us first consider the case for mod p.
for which
(mod p).
(b - ab - a , p) ^ 1,
consider
To determine the number of pairs
b - ab - a = 0 (mod p),
Since 5 is a quadratic residue of p r i m e s of this form,
or
(2b - a)
(a,b)
= 5a2
x = 5 (mod p) has two
solutions c. Thus the above condition is equivalent to 2b - a = ca (mod p),
or
(mod p)
o r bt = r a and b 2 = sa (mod p), where r = (1 + c)/2 and s = (1 - c)/2 (mod p). Note
that r s (mod p) for this would imply c = 0 and hence c 2 = 0 (mod p).
To have (a, b, p) = 1, we must have
(a, p) = 1 because if
(a,p) ^ 1,
then
pja;
but
-M
(mod p)
and so b s 0 (mod p) and p|b; hence (a9 b, p) ^ 1. Therefore for mod p there are p 1 possible values of a that will give
(a, b, p) = 1
and (b2 - ab - a 2 , p)
1;
and
58
corresponding to each value of a,
there are two values of b.
[Feb.
Hence, there a r e
2(p - 1)
p a i r s (a,b) with (a,b,p) = 1 and (b2 - ab - a 2 , p) ^ 1, We obtain:

If a = 1, b-L = r and b 2 = s (mod p) ;
If a = 2, ht = 2r and b 2 = 2s (mod p);
If a = p - 1, b t = (p - l ) r
and
b 2 = (p - l)s
(mod p).
It is clear that no matter what a i s , for mod p, the pairs (a, ar) will all produce sequences
of the same length as the pair ( l , r ) , and similarly the p a i r s
(a, as)
will all produce s e -
quences of the same length as the pair (1, s).

Now, we know that since k(p) = 4t + 2 there exist (a,b) such that h(a,b,p) = -^k(p) =
2t + 1. But if there is one pair (a,b) satisfying this, there a r e at l e a s t p - 1 p a i r s
(a,b)
with h(a,b,p) = 2t + 1. We will show that there a r e only p - 1 such p a i r s .

Without any loss of generality we may assume a to be
1.
We will show that either
( l , r ) or (l,s) but not both, will produce a sequence of length 2t + 1 when reduced mod p.
Now suppose that both ( l , r ) and ( l , s ) produce sequences of length 2t + 1. We have
1, r , 1 + r , 1 + 2r, 2 + 3r, , u .. + u r,
n-1
n
1, s, 1 + s, 1 + 2s, 2 + 3s, , u
+ us,
n-1
Therefore, we must have u , + u ,

= 0 (mod p).
u
By 12,
f(p)
(mod p) ;
^
(mod p) .
v
f/
r = 1 and u ? , + u ? , - s = 1 (modp). Hence u , ( r - s )
must be even for if f(p) is odd, then 4|k(p ). This gives
of+i ^ (mod p) for otherwise f (p) j(2t + 1) which is impossible.
(mod p), and we have shown that this is impossible.
Hence we have
r = s
Thus, the p a i r s ( l , r ) and ( l , s ) can-
not both produce sequences of length 2t + 1.

An alternative proof is the following.
2
r - r - 1 = 0 (mod p), or 1 + r = r
f
0 we may obtain r + r
nA
Thus the sequence
Since b 2 - ab - a 2 = 0 (mod p) we must have
(mod p). Using the recurrence relation f
= r ( l + r) = r
(mod p), r + r
1, r , 1 + r , 1 + 2r, 2 + 3r,
may be written as 1, r , r 2 , r 3 , r 4 , (mod p).
2s, 2 + 3 s ,
= r(r + r ) = r
= f _ +
(mod p), etc.
(mod p)
Similarly, the sequence

2
(modp) may be written as 1, s, s , s , s ,
1, s, 1 + s, 1 +
(modp).
Therefore, the assumption that these two sequences have periods of length 2t + 1 when
2t+l
2t+l
reduced mod p , implies that r
= 1 and s
= 1 (modp). Multiplying these two
2t+l
congruences we obtain (rs)
= 1 (mod p). But
rs
= " = _i
(mod p)
9 f_j-l
because c 2 = 5 (mod p), and so (-1)

= 1 (modp) which is impossible.
and ( l , s ) cannot both produce sequences of length ^k(p) = 2t + 1.
Hence
(l,r)
1974]
59
Therefore of the 2(p - 1) pairs (a,b) for which (b2 - ab - a 2 , p) ^ 1 and (a,b,p) = 1,
p - 1 pairs produce sequences of length |-k and the other p - 1 pairs produce sequences of
length k.
(a,b,p) = 1 is given by p 2 - 1,
Since the total number of pairs (a,b) with

now find the number of p a i r s (a,b) for which
2
we can
(b2 - ab - a 2 , p) = 1.
(a 9 b,p) = 1 and
obtain (p - 1) - 2(p - 1) = (p - l) . All of these produce sequences of length k.
We
Therefore
for mod p we have

n(l) = 1,
and
(^)=2(p_^)
0)
n(k) = (P - I <P - 1>2 = P ( P k -

We shall next consider the case for mod p . The condition
2
(b2 - ab - a 2 , p ) ^ 1 is
equivalent to (b - ab - a , p) f 1. Therefore we must again have
-W
(mod p)
e
e
e
eX
We know that (a, b, p ) = 1 if and only if (a, p ) = 1. Hence there a r e p - p
possible
e-1
values of a, and corresponding to each value of a there a r e 2p
values of b. Thus
there a r e 2p " (p - p " ) pairs (a,b) with ( a , b , p ) = 1 and (b2 - ab - a 2 , p ) ^ 1
e
e1
If a = 1, hi = r + jp and b 2 = s + jp (mod p ) where j = 0, 1, 2, , p
- 1. If
a = 2, bt = 2r + jp and b 2 = 2s + jp (mod p ), where j = 0, 1, 2, , p " - 1. These
a r e equivalent to bt = 2(r + jp) and b 2 = 2(s + jp) (mod p ), where j = 0, 1, 2, ,.
e-1 p
- 1.
Since for any a,
the sequences
(a, a(r + jp)) and
(a, a(s + jp)) will all have the

e-1
same length as (1, r + jp) and (1, s + jp), respectively, for j = 0, 1, , p
- 1, it
e-1
is sufficient to consider the sequences (1, r + jp) and (1, s + jp) for j = 0, 1, , p
- 1.
Since k(p e ) = 4 t + 2, we know that for at least one value of j , at least one of (1, r +
jp)
and (1, s + jp) produces
a sequence of length
2t + 1.
Suppose for some value of j ,
h = h ( l , r + jp, p ) = 2t + 1. We will show that then for any i where i is one of 0, 1, 2,

. .. , pe~
_ i,
h ( l , s + ip 3 p 6 ) ^ 2t + 1. Suppose for some i ?

h = h ( l , r + jp, p e ) = h ( l , s + ip, p e ) = J k = 2t + 1.
We have
and so
1, r + jp, , u n _ x + u n (r + jp),
(mod p 6 ) ;
1, s + ip, , u n _ 1 + u n (s + ip),
(mod p e ) ;
60

U
h-1
+ U
h^r
1 ~
h-1
h^ S
+ ip
[Feb.
^md
p6
o r u, ( r + jp) s u, (s + ip) (mod p ). Since by 12 u, ^ 0 (modp),
^ '
we may cancel
u,
and obtain r + jp = s + ip (mod p ), or r = s (mod p) which is impossible.

Hence if for some value of j s h(l + r + jp, p ) = 2t + 1 then for no value of i can
h ( l , s + ip, p ) be equal to 2t + 1. Similarly, if for some value of j ,
h ( l , s + jp, p e ) = 2t + 1,
then for no value of i can h ( l , r + ip, p ) be equal to 2t + 1.
Next, we will show that only one value of j gives a length of
2t + 1.
Suppose both
11, r + jp) and (1, r + ip) produce sequences of length h = 2t + 1, where i and j a r e two
e-1
different numbers from 0, 1, , p
- 1. Therefore
U
h-1
+ U
h^r
3P) = ! = \_i
+ u
(r
*P)
(mod
PG)>
or
u h (r + jp) = u^(r + ip)
(mod p e ) .
Since by 12, u, ^ 0 (modp), we have r + jp = r + ip (modp ), or jp = ip (modp ),

e-1
e-1
o r j = i (mod p
) which is impossible. Therefore of the 2p
values corresponding
to each value of a, only one can produce a sequence of length 2t + 1. But there a r e p p " possible values of a. Hence there are p - p
or p " (p - 1) pairs (a,b) that
.
Q
e1 e
e1
e
e1
produce sequences of length | k ( p ). The remaining 2p
(p - p
) - (p - p
) or
p " (p - l)(2p
- 1) pairs (a,b) that have (a, b, p ) = 1 and (b2 - ab - a 2 , p ) f 1
e
2e
2e2
produce sequences of length k(p ). Also since there are p
-p
p a i r s (a,b) for which
(a, b, p ) = 1, we have
, 2e
2e-2 x
e-1/ e
e-lv
(p
- p
) - 02p
(p - p
)
or p
(p - l) 2 pairs with (b2 - ab - a 2 , p ) = 1. All of these produce sequences of length
k(p ). In addition to these, there are the sequences for which

mod p
(a, b, p ) ^ 1.
Thus for
we have n(l) = 1,
n(pik/2)
= P > - 1) = 2 ( P ^ 1 )
k
p'k/2
and
n(pik) = P'<P - W
+ p21(
P - ' - <P - ^
^ - ^
(i = 0 , l , . . . , e - l ) .
P k
Theorem 3 T . Let m = p
where p = 1 (mod 10) and let t be the largest integer
such that k(p ) = k(p) with t > 1.
1974

(1) if 41 k9 then (a) for e < t, n(l) = 1 and
2e
nfc) =
P - ^
and (b) for e > t, n(l) = 1,

2t
m(k) = S^ZJ,
and
n(pi4+1k)
(2) if 4 | k ,
pl+t-1
[P2 -
1}
for i = t, . . . . e - 1.
then (a) for e t, n(l) = 1,
(P 6 - D
and
/. v
p (p
- 1)
NF
n(k) =
i
and (b) for e > t, n(l) = 1,
(f)
n(k) =
p (
\ -
i)
1}
n ^ W V s = 2pt"XLP - 1}
and
n(pi-t+ik)
P U 1 ( P - 1HP 1 + 1 + P 1 - i)
for i = t, , m - 1.
Proof. (1) Same as the proof for Theorem l f .
e
(a, b, p ) = 1,
(2) We have shown in Theorem 3 that if

e1
(p - 1) pairs (a,b) produce sequences of length |-k(p ) and

p
e-1,
, 2e-2,
W o e-1
(p - ll)(2p
- lX1) + p
(p -
or
p
e-1/
e-1
W e ,
(p - ll)(p
+ p
- 1X
1)
ix2
ir
for
e
mod p ,
62
p a i r s (a,b) produce sequences of length k(p ).
Thus we have:
For mod p, n(l) = 1.

For mod p, n(l) = 1,
,2/
_ 2(p - 1)
k
and
(k) _ P(P - 1)
k
For mod p 2 , n(l) = 1,
H + 2p(p - 1)
( * ) - * !k
2(p2 - 1)
and
n(k) -
p(p
1}
+ P(P - D(P 2 + P - 1) _ P 2 (p 2 - 1)
For mod p , n(l) = 1,
-1
*(t)-
i=0
'
t
(p - 1) _ 2(p I - 1)
~E
k
and
n(k) =
pi(p l)(pi+1 + pi 1]
- = pt(pt - *>
i=0
For mod p
, n(l) = 1;,
-(0n(k)
2(p t - 1)
k
P (p - 1)
k
t 1
n(pk/2) = 2p " (p^ - i)
and
t-1,
t
1X t+1 ,
n(pk) = P " (P - DP
+ P - 1)
[Feb.
1974]
63
t+2
For mod p , n(l) = 1 5
(1)=*^.
n(pk)
*>.&f.
(pk/2)
= P " < P - D(P k t+1 + P* - 1) ,
n(pV
2)
(P - 1)
k
2p"(p - 1)
and
t-1,
1W
t+2 ,
t+1
Thus for e ^ t , we have n(l) = 1,

e-1
( l ) = E ^ (p - 1)
k
2(p e - 1)
i=0
and
e-1
n(k) = Y
P1^ - D(P
+P 1 - 1) = P e (p 6 - 1)
i=0
and for e > t we have n(l) = 1,
(|)
= V _ ^
n ( k ) =
P!(P^)
n(pi-t+lk/2)
2p^(p - 1)
and
n(p i - t + 1 k) = P t - 1 ( P - 1HP 1+1 + P 1 - 1)
Theorem 4.
f o r
i =t, , e - 1 .
Let N(h,m) = h-n(h 5 m). If

n e.
m = n p.* ,
i=l
then
n
e.
n N(h., p . 1 )
n(h
>m>=
S
LCM[h.]=h
e.
where h. = h(a 5 b , p. ) .
izi
64
Proof.
[Feb. 1974]
n
Consider the equivalent problem for which the modulus is of the form
II m.
where the m. are pairwise relatively p r i m e .
Suppose first that m = m^m^ and (m 1 ,m 2 ) ~
1. By 1, if h ( a , b , m i ) = ht and h ( a , b , m 2 ) = h2 then h(a, b, m ^ )
is the least com-
mon multiple of hj and h 2 . Also, by the Chinese Remainder Theorem, we know that each
pair
(a,b) (mod m ^
(mod m ^ )
and each pair
such that
e = a,
(c,d)
(mod m 2 )
f = b (mod n ^ ) ,
gives r i s e to a unique pair (e,f),
and e = c,
f = d
(mod m 2 ) .
Byl,
h(e, f, m 1 m 2 ) is the least common multiple of h(e, f, m A ) and h(e, f, m 2 ) . But e = a and
f = b (mod m.1) imply that h(e, f, m ^ = h(a, b, m A ) = h l s and similarly h(e, f, m 2 ) = h2;,
and so h(e, f, niim 2 ) is the least common multiple of hA and h 2 .
Let [hi, h2 ] denote the least common multiple of hA and h 2 . We have seen that each
pair of p a i r s (a,b) (mod m t ) and (c,d) (mod m 2 ) gives a unique pair (e s f)
of length h = [h l 3 h 2 ] .
(e,f)
with
(mod Hiim 2 )
with
length h when reduced mod mi produces a sequence of length hi and when reduced
mod
length hi
(mod m^)
Therefore there are

and length h 2
N(h1? mi)-N(h 2 , m 2 )
(mod m 1 m 2 ),
(mod m 2 ) .
Now any pair
m 2 produces a sequence of length h 2 such that [hi, h 2 ] = h.

^
N(h,m 1 m 2 ) =
2-*
N(hi,mi)-N(h 2 ,m 2 ),
[h l 5 h 2 ]=h
such pairs
(e,f)
Hence
and so n ( h , m 1 m 2 ) =
N(h 1 ,m 1 ).N(h 2 ,m 2 )
JL/
K
'
[hi,h 2 ]=h
n
By induction, this result is now easily extended to the case m = n m.,
where n > 2 ,
and all the m.1 are pairwise relatively p r i m e .
Thus we obtain
n
n NOt^m.)
n(h,m) =
2^
^ L _
LCM[h.]=h
e.
In particular, if m. = p.
for i = 1, , n, we have
n(h,m) =
2^
LCM[h.]=h
n
e.
n N(h., p. 1 )
i=l
h
These four theorems cover all possible values of m.
Thus if k(p ) is known, the
values of h(a, b, m) as well as n(h 9 m) can be determined.

I would like to acknowledge the assistance Prof. D. Singmaster gave me with his c r i t icisms and suggestions in putting this paper in its final form.
REFERENCES
1. D. D. Wall, "Fibonacci Series Modulo m , " Amer. Math. Monthly, 67(1960), 525-532.
2. D. W. Robinson, "The Fibonacci Matrix Modulo m , n Fibonacci Quarterly, 1 (1963), pp.
29-35.
3. S. E. Mamangakis, "Remarks on the Fibonacci Series Modulo m , " Amer. Math. Monthly,
68 (1961), pp. 648-649.
4. J. Vinson, "The Relation of the Period Modulo m to the Rank of Apparition of m in the
Fibonacci Sequence," Fibonacci Quarterly, 1(1963), pp. 37-45.
PHI: ANOTHER HIDING PLACE

H. E. HUNTLEY
Nethercombe Cottage, Canada Combe, Hutron, Weston - Super - Mare, England
F r o m an a r e a A of any outline, regular o r i r r e g u l a r , there is cut an a r e a B, having

the same outline as that of A under the following conditions: (i) The peripheries of A and
B have one point O in common; (ii) B is oriented so that O and the centroids
C, of A and B a r e colinear.
lies in the straight line OC C,
and
a
It follows that C, the centroid of the remnant (A - B) also
produced.
Fig. 1
Let the ratio of the linear dimensions of A and B be
2
a:b,
their respective a r e a s
being Aa , Ab ; OC /OC, = a/b.

a
D
Taking moments about O,
Ab 2 .OC b + A(a2 - b 2 )-OC = Aa 2 -OC a ,
whence, multiplying by l/Ab 2 -OC, ,
OC
OC
b
b3
Since (a/b) - 1 / 0 ,
/a
Vb
y>'ocb~b2
_OC_
a*
a
b
+ 1 .
Phi, the Golden Section, is now uncovered by writing OC/OC, = 2,

^ - - ^ - 1
b
b2
65
= 0
giving
66
PHI: ANOTHER HIDING PLACE
[Feb. 1974]
whence a/b = (j) o r a/b = 1/0.

The result i s , of course, applicable to regular plane figures.
In the case of the circle
the centroid C of the remnant lune falls on the endpoint of the diameter of B through O.
Fig. 2
Any chord of circle A through O is cut by the circumference of
Section: PO/QO = 0 = (1 +
B in the Golden
\fE)/2.
THE FIBONACCI ASSOCIATION

PROGRAM OF SATURDAY, OCTOBER 20, 1973
St. Mary's College
SOME PROPERTIES OF TRIANGULAR NUMBERS
Marjorie Bicknell, A. C. Wilcox High School, Santa Clara, California
THE GOLDEN SECTION REVISITED
Edmundo Alvillar, San Francisco, California
OPERATORS ASSOCIATED WITH STIRLING NUMBERS
Elaine E. Alexander, California Polytechnic State University
ALGORITHMS FOR THIRD-ORDER RECURSION SEQUENCES
Brother Alfred Brousseau, S t Mary's College, California
ON THE DIOPHANTINE EQUATION 1 + x + + x a = y b
Hugh Edgar, San Jose State University, San J o s e , California
PASCAL, CATALAN, AND LAGRANGE WITH CONVOLUTIONS
Verner E. Hoggatt, J r . , San Jose State University, San J o s e , California.
SOME GEOMETRICAL PROPERTIES OF THE GENERALIZED FIBONACCI SEQUENCE

D. V. JAISWAL
Holkar Science College, in do re, India
1.
INTRODUCTION
In this paper, some geometrical properties of the generalized Fibonacci sequence { T }

have been discussed.
The sequence {T } being defined by

T
n+1
+ T
n -n1
b .
a,
On taking a = b = 1, the Fibonacci sequence {F } is obtained.

We shall make use of the following identities [1]
T F J_1 + T
, F .
m n+1
m-1 n
m+n
F F ,
- F
= (-l) n "" S F F M
n n+m
s s+m
n - s n+m+s
(-l)mF, F
D
T T
- T
T
k n-m
m n+k
m+k n
(1.1)
(1.2)
(1.3)
where D is the characteristic number of the sequence and is given by

T 2 - T n T _L1 = ( - l ) n D ;
n
n - 1 n+1
2.
2a < b .
THEOREM 1
A r e a of the triangle having vertices at the points designated by the rectangular cartesian
coordinates (T , T __, ), (T , , T __, ^ ), (T __, , T L ^ ) is independent
of n.
p
n
n+r
n+p
n+p+r
n+q
n+q+r
Proof. Twice the a r e a of the specified triangle is equal to the absolute value of the
determinant
n+r
n
n+p
T ,
"n+q
n+p+r
T
n+q+r
Using (1.1) for the second column the determinant can be written as
n-1
"r+1 "n+p
n+q
n+p
+ F
n+p
r ,
n+q
n+q
67
n+p-1
n+q-1
68
SOME GEOMETRICAL PROPERTIES
[Feb.
The first determinant is obviously zero; in the second on alternately subtracting the
second and first column from each other, the suffixes can be reduced and finally we get
T
l
p+1
T
T
IT
q+l
p+2
q+2
according as n is odd or even.

On expanding the determinant along the third column, we obtain
+F F(T
T
- T
T
) - (T T
- T T
)
u
" r LV p + l q + 2
p+2 q + r
lq+2
2 q+r
+ (T T
- T T
)1
u
l p+2
2 p+l;j
which on using (1.3) reduces to
F L[ F - F - (-1) P F
]D .
r
q
p
q-p
Thus the area of the specified triangle is independent of n.
P a r t i c u l a r Case.
On taking
r = h,
p = 2h,
a r e a of the triangle
whose vertices are
to
q = 4h,
(F , F ,, ),
n
n+h
a = b = 1,
(F , n, , F , OI ),
n+2h
n+3h
we find that the

(F , ., , F , _. ) is
n+4h
n+5h
equal to the value of

(2.1)
4Fh(F4h-2F2h>
Duncan [2] has proved that the a r e a of this triangle is
which on using (1.2) simplifies to the value given in (2.1).

3.
THEOREM 2
Lines drawn through

with the direction ratios T , T , , T , ,
& the origin
&
n
n+p
n+q
and q a r e a r b i t r a r y constants are airways coplanar for every value of n.
where
Proof.
T., T.+
Direction ratios of any three such lines are T., T., , T., ; T., T., , T., ;
J
i
i+p
i+q
J
3+P
J+q
, T. + . These will be coplanar if
T.
l+p
(3.1)
T.
3
T
-4+
3 P
k+p
-a.
i+q
3+q
= 0 .
T
x
k+q
On using the relation (1.1), the left-hand side of (3.1) can be written as the sum of four
determinants, each of which is zero.
Hence proved.
4.
THEOREM 3
Set of points designated by the cartesian coordinates (T , T
, T
) where p and
1974]
OF THE GENEEALIZED FIBONACCI SEQUENCE
69
q are a r b i t r a r y constants and n = 1, 2, 3, , are always coplanar.
This plane passes
through the origin, and its equation is independent of n.

Proof.
Equation to the plane passing through any three points of the set is
T.
(4.1)
i+p
T.
J
j+p
-i+q
^
= o ,
j+q
T, ,
T,
k+p
k+q
where i, j and k are particular values of n.
Here the coefficient of x is
[(TV
^ - T.^_
_, ) - (TV
_, - T.,_
__, )
L
j+p T.k+q
j+q T.k+p
i+p T.k+q
i+q T.k+p
+ (TV TV - T.^ T._, ) 1J
i+p j+q
i+q j+p
1) P F q - p u{ ( - l ^ F ,k - j. - ( - D ' F , k - i. + ( - D ' F .j - i.}D
J
The coefficient of y is obtained on putting p = 0 in the coefficient of x;
the coeffi-
cient of z is obtained from the coefficient of y on replacing q by p; the constant t e r m is

zero as is already proved in (3.1).
Thus the equation to the plane simplifies to
(-DPF
(4.2)
F Jy + F z = 0 .
q
p
q-p
This equation is independent of n. Also it does not depend on the initial values a and
b.
Q.E.D.
On taking a = 1, b = 3 we obtain the Lucas sequence {L } .
points (F., F . + 2 , F . + 5 ) ,
Tk+5),
k = 1 , 2 , 3,
i = 1, 25 3,
(L., L j + 2 , L . + 5 ) ,
The
j - 1, 2, 3,- - ; ( T ^ T R + 2 ,
all lie on the plane 2x - 5y + z = 0.

5. THEOREM 4
The set of planes

T x + T y J + T4_z+ T , = 0
n
n+p
n+q
n+r
all intersect in a given line
where p, q, r are a r b i t r a r y constants, and n = 1, 2, 3,

whose equation is independent of n.
Proof.
Let two such planes be

T . x + TV yJ + TV z + TV
i
i+p
i+q
i+r
T . x + TV yJ + T., z + TV
J
j+p
j+q
j+r
(5.1)
=0
=0
The equation to the line of intersection of the parallel planes through the origin is
T
- T
T
i+p j+q
i+q j+p
T.TV
l
j+q
- T._^ T.
i+q
T.TV
l
j+p
- TV
i+p
T.
SOME GEOMETRICAL PROPERTIES

OF THE GENERALIZED FIBONACCI SEQUENCE
,
*et)'
iy
On using (1.3) and proceeding as in (4.2) this simplifies to

_
-y
(-1) F
q-p
Similarly the line of intersection of the planes given by (5.1) meets the plane z = 0, at
the point given by
x
_ -y_ _ _J_
(-1) F
"
r "
r-p
p
*
'
Thus the equation to the line of intersection of the planes given by (5.1) becomes
(-1)PF x - F
F y + F
q-p
Hence proved.
The set of planes whose equations are
F . x + F i + 1 y + F . + 3 z + F . + 4 = 0,
i = 1, 2, 3 , - - - ;
L . x + L . + 1 y + L . + 3 z + L . + 4 = 0,
] = 1, 2. 3, ;
kX
k = l , 2 , 3, ;
+ T
k+iy
+ T
k+3Z
+ T
k + 4 = '
all intersect along the line

x + 2
1
y + 3
2
_z
1
I am grateful to Dr. V. M. Bhise, G. S. Tech. Institute, Indore, for his help and guidance in the preparation of this paper.
REFERENCES
1.
Brother Alfred Brousseau,
An Introduction to Fibonacci Discovery,
The Fibonacci A s -
2.
Dewey C. Duncan, n Chains of Equivalent Fibonacci-wise Triangles," Fibonacci Q u a r t e r -
sociation, San J o s e , California (1965).

terly, Vol. 5, No. 1 (February 1967), pp. 87-88.
'4
SETS OF BINOMIAL COEFFICIENTS WITH EÛAL PRODUCTS

CALVIN T. LONG
Washington State University, Pullman, Washington 99163
and University of British Columbia, Vancouver, B.C., Canada
and
VERNER E. HOGGATT, JR.
1.
INTRODUCTION
In [2] , Hoggatt and Hansell show that the product of the six binomial coefficients s u r rounding any particular entry in P a s c a l ' s triangle is an integral square.
They also observe
that the two products of the alternate triads of these six numbers a r e equal.
Quite r e m a r k -
ably, Gould conjectured and Hillman and Hoggatt [l] have now proved that the two greatest
common divisors of the numbers in the above-mentioned triads are also equal though their
l e a s t common multiples a r e , in general, not equal.
Hillman and Hoggatt also generalize the
greatest common divisor property to more general a r r a y s .

The integral square property was further investigated by Moore [4] , who showed that
the result is true for any regular hexagon of binomial coefficients if the number of entries
p e r side is even, and by the present author [ 3 ] , who generalized the e a r l i e r results to nonregular hexagons, octagons, and other a r r a y s of binomial coefficients whose products are
squares.
In the present paper, we generalize the equal product property of Hoggatt and Hansell
along the lines of [3] and also make some observations and conjectures regarding a generalized greatest common divisor property.
It will suit our purpose to represent P a s c a l ' s triangle (or, more precisely, a portion of
it) by a lattice of dots as in Fig. 1. We will have occasion to refer to various polygonal figu r e s and when we do, unless expressly stated to the contrary, we shall always mean a simple
closed polygonal curve whose vertices are lattice points. Occasionally, it will be convenient
to represent a small portion of P a s c a l ' s triangle by letter arranged in the proper position.
Fig. 1
Fig. 2
71
72
SETS OF BINOMIAL COEFFICIENTS WITH EQUAL PRODUCTS

2.
[Feb.
As in [3] , we begin by deriving a fundamental lemma which is basic to all of the other
results of this section.
Lemma 1. Consider two parallelograms of binomial coefficients oriented as in Fig. 2
and with corner coefficients
a, b, c, d and e , f, g, h a s indicated.
Then the products
acfh and bdeg are equal.

Proof. For suitable integers m, n, r , s , and t, the binomial coefficients in question
may be represented in the form
+ B
* - ( \
c
_/
m+r \
In + r + tJ'
) '
=( n ) '
m
f = (
\
yn + r + t y '
= (n+\r)>
* = (mn\\+r)'
=(
m+s
\
^n + s + r + ty'
/ m + s + r \
\n + a + r + t J '
Thus, the desired products a r e

a c fh
<m + s)'.
(m + r)t
n! (m - n + sjl (n + r)! (m - n)!
ml
_
(m + s + r)l
(n + r + t)! (m - n - r - t)l ' (n + s + r + t)! (m - n - t)!
and
, ,
DQeg
m!
(m + s + r)!
n! (m - n)! " W+ r)(m - n + s)!
(m + r)
(m + s)t
(n + r + t)! (m - n - t)t ' (n + r + s + t)! (m - n - r - t)!
and these a r e clearly equal as claimed.

As a first consequence of Lemma 1, we obtain the equal product result of Hoggatt and
Hansell.
Theorem 2. Let a, b, c , d, e , f, and g denote binomial coefficients as in the a r r a y
a
f
b
g
c .
d
Then aec = fbd.

Proof.
Therefore,
The parallelograms a, f, e , g and b, g, d, c are oriented as in Lemma 1.

fgbd = aegc and this implies the desired result.
By essentially the same argument, we obtain the following more general statement about
products of coefficients at the vertices of hexagons in P a s c a l ' s triangle.
Theorem 3. Let m > 1 and n > 1 be integers and let H be a convex hexagon whose
sides lie on the horizontal rows and main diagonals of P a s c a l ' s triangle.
Let the number of
1974]
coefficients on the respective sides of H be m, n, m , n, m,

a, b, c, d, e,
73
and n in that o r d e r , and let
and f be the coefficients in cyclic order at the vertices of H.
Then
ace =
bdf.
Proof.
Without loss in generality, we may take
to be the number of coefficients
along the bottom side of H. If we consider two m-by-n parallelograms with a common v e r tex and with corner coefficients
a, b, c, d, e, f,
and g as in Fig. 3, then, again by Lemma
1, fgbd = aegc and this implies the equality claimed.
Fig.
Now, as in [3] , let us call the hexagons of Hoggatt and Hansell fundamental hexagons
and say that a polygonal figure P on P a s c a l ' s triangle is tiled with fundamental hexagons if
P is "covered" by a set F of fundamental hexagons F in such a way that
i.
ii.
iii.
The vertices of each F in F a r e coefficients in P or in the interior of P.

Each boundary coefficient of P is a vertex of precisely one F in F,
and
Each interior coefficient of P Is interior to some F in F o r is a vertex shaped

by precisely two elements of
F.
We can then prove the following result.

Theorem 4.
Let P
be a polygonal figure on P a s c a l ' s triangle with boundary coef-
ficients a1? a 2 , , a n in order around P . If P
can be tiled by fundamental hexagons,
then n = 2s for some s ^- 3 and

s
"
s
E
Proof.
Suppose that P
2i-1 = "
1=1
can be tiled with
2i
1=1
fundamental hexagons.
The proof p r o -
ceeds by induction on r. Clearly the least value of r is 1 which occurs only in the Case Of
the fundamental hexagon itself.
In this c a s e , n = 6 and the result is true by Theorem 2.
Now suppose that the result is true for any polygon that can be tiled with fewer than k fundamental hexagons where k > 1 is fixed and let P
damental hexagons.
be a polygon that can be tiled with k fun-
Let H be one of the hexagons which tiles P
boundary point of P . We distinguish five cases.
and contains at least one
74

Case 1. H contains just one boundary point of P .
may let a
J
a'
[Feb.
Without loss in generality, we
be the boundary point of P which is in H. Let af , a' ,.., aT 1 0 , a' ini

J F
n
n
n
n+1
n+2
n+3
. denote the other five boundary points of H in o r d e r around H. Let P
and
be the poly-
gon obtained from PT by deleting H. Then the boundary points of P

a r e a 1} a 2 , ,
a .,, af , a'
, a l 0 , a' o 5 and a' , ,. Thus, m = n + 4. Also, since P
can be
n-1
n
n+1
n+2
n+3
n+4
m
tiled by k - 1 fundamental hexagons, n + 4 = m = 2t for some t and
a..1 a63 a - a? ,., a' 0 = za 2 4a 4 a 0 a? af , 0 a f , , .
n - 1 n+1 n+3
n-2 n n+2 n+4
But, since H is a fundamental hexagon, it follows that
nan+lan+3 "
nan+2an+4
and this clearly implies that

s
."
2i-i
1=1
"
2i
1=1
since n = 2t - 4 = 2s. This completes the proof for Case 1.

Cases 2-4. In these c a s e s , respectively,
H contains 2, 3, 4, or 5 boundary points of
P . We omit the proofs of these cases since they essentially duplicate the proof of Case 1,
This completes the proof.
With Theorem 4 and Lemma 1 as our principal tools we a r e now able to give several
quick results.
Theorem 5. Let H
be a convex hexagon with an even number of coefficients per side,
with sides oriented along the horizontal rows and main diagonals of P a s c a l ' s triangle, and with
boundary coefficients
a l 5 a 2 , , a n in order around H . Then n = 2s for some
s ^- 3
and
s
n
s
n
2i-i =
1=1
Proof.
2i
1=1
This is an immediate consequence of Theorem 4 since H
can be tiled by fun-
damental hexagons as shown in Theorem 5 of [3].

Theorem 6. Let K
be any convex octagon with sides oriented along the horizontal and
vertical rows and main diagonals of P a s c a l ' s triangle and with boundary coefficients
a l9 a 2 ,
, an in o r d e r around K . Let the number of coefficients on the various sides of K be

n
n
2r, 2s, t, 2u, 2v, t, and 2s as indicated in Fig. 4. Then n = 2h for some h > 4 and
s
"
1=1
2i-l =
s
"
1=1
2i
1974]
75
2v
2s
2s
2r
Fig. 5
Fig. 4
Proof.
The proof is the same as for Theorem 5 and will be omitted.
We observe that the convexity conditions in both Theorems 3 and 5 a r e n e c e s s a r y since

neither result is true for the hexagon of Fig. 5. Also, it is easy to find examples of convex
hexagons where the results of Theorems 3 and 5 do not hold if the conditions on the number of
elements per side are not met.
In fact, we conjecture that the conditions in these theorems
a r e both necessary and sufficient.
On the other hand, the convexity condition of Theorem 6
is not n e c e s s a r y since the result holds for the octagon of Fig. 6 which is clearly not convex.
We make no conjecture regarding n e c e s s a r y and sufficient conditions for the. result of Theor e m 6 to hold for octagons in general, or indeed, for hexagons whose sides may not lie along
the horizontal rows and main diagonals of P a s c a l ' s triangle.
We note that the octagon of Fig.
6 cannot be tiled by fundamental hexagons but can be tiled by pairs of properly oriented "fundamental parallelograms" as indicated by the shading in the figure.
Thus, the most general
theorem for these and other polygons will most likely have to be couched in t e r m s of tilings
by sets of pairs of fundamental parallelograms.
<$/ <>/ \</ \ /

,\ / ' \ A-\ / Z \ /\> /. >\
/ W
^?
T5:/
./
A\ i'\
'vh
Fig. 6
Fig. 7
76
[Feb.
3. ADDITIONAL COMMENTS ON EQUAL PRODUCTS

Theorem 3 gives an equal product result for the corner coefficients of hexagons and it
is natural to seek similar results for octagons.
It is easy to find octagons like those in Figs.
4 and 6 for which the equal product property on vertices does not hold.
Nevertheless, it is
possible to find classes of octagons for which the equal product property does hold for the
products of alternate corner coefficients.
Theorem 7.
Let K be a convex octagon formed as in Fig. 7 by adjoining parallelo-
g r a m s with r and s and r and t elements on a side to a parallelogram with r elements

on each side.
If the corner coefficients a r e a l5 a 2 , , a8 as shown, then

a1a3a5a7 = a2a4a6a8 .
Proof.
We have only to observe that a1? a 4 , a 5 , a 8 and a 2 , a 3 , a 6 , a 7 are vertices of
p a i r s of parallelograms oriented as in Lemma 1. The result is then immediate.

Again it is clear that the convexity condition of Theorem 7 is not necessary. The proof,
after all, r e s t s on the presence of the properly oriented p a i r s of parallelograms.
In p r e -
cisely the same way we show that a 1 a3 a 5 a 7 = a2 a4 a6 a 8 for each of the three octagons of
Fig. 8. Note that for K2 the two products a r e not products of alternate vertices around the
octagon.
Clearly the preceding methods can be used to obtain a wide variety of configurations
of binomial coefficients which divide into sets with equal products. As illustrations we give
several examples of polygons (sometimes not closed, simple, or connected) with this property.
a7
a4
a7
a6
Fig. 8
4.
THE GREATEST COMMON DIVISOR PROPERTY
As mentioned in Section 1, if the a r r a y

a
c
b
d
e
g
r e p r e s e n t s coefficients from P a s c a l ' s triangle, then afe = cbg and Hillman and Hoggatt
1974]
S E T S O F BINOMIAL C O E F F I C I E N T S WITH E Q U A L P R O D U C T S
12
12
a
"
2i-l
1=1
"
*
2i
1=1
2i-l
1=1
"
1=1
2i
II
ai
a2
at
*
a2
*
a17
*
a3
*
a18*
*
*
24 /
\l9
a5
a10
a23 4
A7
*
21
*
a
12
3ft
a9
ia 2 0 I a7
a
22
*
*
a4
*
io
a9
12
12
"
2i-l
1=1
= ."
.nî-l
2i
III
1=1
IV
a4
a5
i
a
s
*
l6
an '
*
a
a12
an
10
10
. n - a*02 i - l
. n1 a20 i - l = . n
" a02i
i=l
* 1a
11
i=l
i=l
VI
Fig.
20 1
a
l
11
.V2i
1=1
1=1
r as
" a2i
i=l
77
78
[Feb.
have shown that (a, f, e) = (c, b, g) where we use parentheses to indicate greatest common divisors.
In view of the preceding results on equal products, one wonders if the g r e a t -
est divisor property also holds in more general settings.

Unfortunately, it is easy to find examples of regular hexagons with sides oriented along
the main diagonals and horizontal rows of P a s c a l ' s triangle where the two alternate triads of
corner coefficients have different greatest common divisors in spite of the fact that theyhave
equal products by Theorem 3. We have such examples for hexagons with 3 , 4 , 5 and 6 c o efficients per side and conjecture that the property only holds in general for the fundamental
hexagons of Hoggatt and Hansell.
Also, we observe that, for the parallelograms of Fig. 2,
the products acfh and bdeg a r e equal but that
(a, c, f, h) is not necessarily equal to
(b, d, e, g). At the same time, we have been unable to find examples of hexagons of the type
of Theorem 5 where the greatest common divisor of the two sets of alternate boundary coefficients are not equal.
Of course, these greatest common divisors are usually equal to one,
but the three regular hexagons with four elements per side whose upper left-hand coefficients
a r e , respectively,;
\S)'
Ce)'
and
(")
have p a i r s of greatest common divisors equal to 13, 13, and
34,
respectively.
We con-
jecture that the greatest common divisors of the two sets of alternate boundary coefficients
for the hexagons of Theorem 5 a r e equal.
This is not t r u e , however, of the octagons of Theorem 6, since, in particular,
((:)
() ( 0 -
( ( ! )
( : )
( : )
(:))->
(
and these are alternate boundary coefficients of such an octagon.
Of c o u r s e , this makes it
clear that not all polygons that can be tiled with fundamental hexagons have the equal greatest
common divisor property.
At the same time, some figures that cannot be tiled with funda-
mental hexagons appear to have the equal greatest common divisor property.
For example,
this appears to be true of the octagon of Fig. 11 in [3] though we have no proof of this fact.
This leaves the question of the characterization of figures having the equal greatest common
divisor property quite open.
5.
GENERALIZATIONS AND EXTENSIONS
There are an infinitude of other Pascal-like a r r a y s in which the Hexagon Squares p r o p erty holds.
F o r example, the Fibonomial triangle and the generalized Fibonomial triangle.
If, indeed we replace F
by f (x), the property holds and thus for each x integral yields
an infinitude of such a r r a y s .
k
For example, if x = 2,
Pell Number Sequence works.
we get the Pell numbers, or every
1974]
79
REFERENCES
1. A. P. Hillman and V. E. Hoggatt, J r . , "On Gould f s Hexagon Conjecture, Fibonacci Quarterly, Vol. 10, No. 6 (Dec. 1972), pp. 561-564.
2.
V. E. Hoggatt, J r . , and W. Hansell, "The Hidden Hexagon Squares," FibonacciQuarterlXi Vol. 9, No. 2 (Apr. 1971), pp. 120, 133.
3.
Calvin T. Long, "Arrays of Binomial Coefficients whose Products are S q u a r e s , " Fibonacci Quarterly, Vol. 11, No. 5 (Dec. 1973), pp. 449-456.
4.
Carl L. Moore, "More Hidden Hexagon S q u a r e s , " Fibonacci Quarterly, Vol. 11, No. 5
(Dec. 1973), p. 525.
(Continued from page 46. )

that
F
F
3 n (modulo
8. 3 n - i
8- 3 n--
3n+1)
1 + 3 n (modulo 3 n + 1 ) .
-!
Therefore
F
8- 3n--i+x
ss
If x satisfies (*), then either

modulo 3
FF
H+
3n(F
x or
+ F _,,) (modulo 3 n + 1 ) .
X
8-3
X+l
+ x or
163 ~ + x
will be congruent to m
. Therefore (*) has solutions for a r b i t r a r i l y large n.
Problem 2.
The number N is said to have complete Fibonacci residues if there ex-
ists a solution to the congruence

F
= m (modulo N)
for all integers m. A computer search shows that the only values of N ^ 500 having complete Fibonacci residues are the divisors of
35,
2 2 .5 3 ,
2-3-5 3 , 5-3 4 , or 7-53 .
Determine all N which have complete Fibonacci residues.

Problem 3 is submitted by the undersigned and Leonard Carlitz, Duke University, Durham, North Carolina.
Problem 3. Show that if
(Continued on page 82.)
= em'n9
then
ON DAYKIN'S ALGORITHM FOR FINDING THE G.C.D.

V. C. HARRIS
San Diego State College, San Diego, California 92115
In a recent issue of the Fibonacci Quarterly, Daykin [l] has given an algorithm for
finding the greatest common divisor of two positive integers.
The process can be obtained
by changing the signs in Euclid T s algorithm (using subtraction in Euclid's algorithm instead
of addition, as possibly Euclid may have done [2]) and taking numbers modulo
10 ,
where
k is the number of digits in the l a r g e r of the two numbers whose g. c. d. is being obtained.
It a p p e a r s , then, that the number of additions required is the sum of the quotients inEuclid f s
method; also, that any modulus (larger than the numbers whose g. c. d. is being obtained) may
be used in place of 10 .
To illustrate this, we have Daykin's example and, on the right, the modification of
Euclid's as suggested above.
To find
(2847,1168):
2847 (+8832)
-2847 (+1168)
1679 (+8832)
-1679 (+1168)
511
-511
8832 (+511)
-1168 (+511)
9343 (+511)
-657 (+511)
9854
-146
511 (+9854)
-511 (+146)
365 (+9854)
-365 (+146)
219 (+9854)
-219 (+146)
73
-73
9854 (+73)
-146 (+73)
9927 (+73)
-73 (+73)
0
0
Hence (2847, 1168) = 73.
REFERENCES
1. D. E. Daykin, "An Addition Algorithm for Greatest Common Divisor," Fibonacci Quarterly, Vol. 8, No. 3 (Oct. , 1970), pp. 347-349.
2. Donald E. Knuth, The Art of Computer Programming, Addison-Wesley Pub. Co., Vol. 2,
pp. 293-337.
80
STUFE OF A FINITE FIELD

SAHIB SINGH
Clarion State College, Clarion, Pennsylvania 16214
INTRODUCTION
Stufe of a field is connected with the property of integer - 1 in that field.
to be the l e a s t integer s such that - 1 = a\ + a2 + + a^,
field.
where each a,
It is defined
belongs to the
In [2] Chowla and Chowla have determined the stufe of a cyclotomic field.
has shown in [3] that the stufe of a finite field is ^ 2 .

further.
Our aim is to elaborate this
Pfister
result
We do this in the following theorem.
Theorem.
Stufe of GF(p ), where p is prime and n ^ 1, is always one except for
the case when n is odd and p = 3 (mod 4), in which case its value is two.
Proof.
We know that the non-zero elements of GF(p ), denoted by GF*(p ), form a
cyclic multiplicative group. Also, it is well known that if G is a cyclic group of o r d e r

and m divides k,
then there exists a unique subgroup of o r d e r
divides (p - 1) for all n,
in G.
Since
(p - 1)
therefore it follows that the m e m b e r s of GF*(p) constitute the
unique subgroup of o r d e r (p - 1) in GF*(p ). Now we develop the proof by considering different c a s e s .

Case 1. Let p = 2. If A is a generator of GF*(2 ), then X
that A
- 1, which means
= A implying that A is a square which enables us to conclude that each element of
GF*(2 ) is a square and thus - 1 is a square.
In the subsequent c a s e s , p is understood to
be an odd prime.
Case 2.
n
of GF*(p ),
Let n be even. From the above analysis it is clear that if A is a generator

then
is a primitive root mod p. In view of the values of p and n we conclude that
( ^ )
is even, which again means that this primitive root mod p is a square implying that
a square.
Case 3. Let n be odd. In this case,
-1
81
-1
is
82
[Feb. 1974]
STUFE OF A FINITE FIELD
is odd.
Thus half the m e m b e r s of GF* (p) which are quadratic residues
squares and the remaining half a r e not.
If p = 1 (mod 4),
a quadratic residue mod p and hence is a square.
mod p would be
it is well known that
(-1)
is
If p = 3 (mod 4), then (-1) is a quad-
ratic non-residue mod p and therefore is not a square.
In this case - 1 is the sum of two
squares, which easily follows from (3) or (4).
REFERENCES
1. G H. Hardy and E. M. Wright, An Introduction to the Theory of Numbers, Oxford University P r e s s , London, 1965.
2.
P. C howl a and S. Chowla, "Determination of the Stufe of Certain Cyclotomic F i e l d s , "

Journal of Number Theory, Vol. 2 (1970), pp. 271-272.
3. Albert Pfister,
Z u r Darstellung Von I Als Summe Von Quadraten in einem K o r p e r , "
Journal of London Math. Society, Vol. 40 (1965), pp. 159-165.

4.
Sahib Singh, "Decomposition of Each Integer as Sum of Two Squares in a Finite Integral
Domain," to appear in the Indian Journal of P u r e and Applied Mathematics.
5. B. L. Van P e r Waerden, Algebra, Vol. 1, Frederick Ungar Pub. C o . , N. Y. , 1970.
(Continued from page 79.)
(i)
(ii)
, 2k-l + , +
-2H-1.
1 1 ( G )
+3+U)
)
k=l
tn/2]
2k
2k
=
n
(u)
+ 3 + u AK)
k=l
Donald E. Knuth
Professor
Stanford University
Stanford, California 94305
FIBONACCI CURIOSITY
The THIRTEENTH PERFECT NUMBER is built on the prime p = 521 = L i;
2 520 ( 2 521
1}
Brother Alfred Brousseau

*~o-#-o~*
ITERATION ALGORITHMS FOR CERTAIN SUMS OF SQUARES

EDGAR KARST
University of Arizona, Tuscon# Arizona 85721
The following three-step iteration algorithm to generate x simultaneously in 2x + 1 =
a
and 3x + 1 = b 2 was mentioned, but not proved, in [4, p. 211]:

92 = 81
1-10 - 1 = 9
9 1 0 - 1 = 89
89
(81 - l ) / 2 = 40 = xt
= 7921
(7921 - l ) / 2 = 3960 = x 2
8 9 - 1 0 - 9 = 881
881 2 = 776161
8 8 1 - 1 0 - 8 9 = 8721
8721 2 = 76055841
8 7 2 1 - 1 0 - 8 8 1 = 86329
863292 = 7452696241
Proof.
(776161 - l ) / 2 = 388080 = x 3
(76055841 - l ) / 2 = 38027920 = x 4
(7452696241 - l ) / 2 = 3726348120 = x 5 .
F r o m 2x + 1 = a2 and 3x + 1 = b2 comes
3a 2 - 2b2 = 1.
If a , b
n n
solution of this generalized
Pell equation,
then a . n = 5a + 4b , b , = 6a + 5b
&
M
n+1
n
n
n+1
n
n
next l a r g& e r one. From these, we obtain immediatelyJ a , - + a n = 10a , b , - +
n+1
n-1
n
n+1
10b , which is equivalent to the algorithm.
F o r the n
ple,
is any
J
is the
b - =
n-1
t e r m formula we use the usual approach by linear substitutions (for exam-
[ 1 , p. 181]) and obtain

x
= [(\/6 + 2)(5 + 2-s/6) n + (N/"6 - 2)(5 - 2 \ / 6 ) n ] / 4 8 - 1/2 .
This formula has three shortcomings: (1) it uses fractions, (2) it employs roots, and (3) it
has n in the exponent.
The algorithm above has none of them.
Similar arguments a r e valid for a four-step iteration algorithm [3] to generate
in
x 2 + (x + l) 2 = y2 .
Sometimes, the n
term formula may be simple, as for a2 + b 2 + (ab)2 = c 2 ,
and
b consecutive positive integers [2]. Here we have

(n - l) 2 + n2 + [ (n - l ) n ] 2 = (n2 - n + l) 2 ,
and hence we need no algorithm.
But for a = 1 an algorithm would be helpful.
find some clues to such an algorithm.
l2 +
o2
O2
l2 +
22 +
22
+ 12
Let us first
We have by hand and by a table of squares:
+ 12 2
l 2 + 70 2 + 70 2
= l 2 = (0 2 + D2
= 3 2 = (2 2 - D2
= 17 2 = (4 2 + D2
= 99 2 = (10 2 _ D2
83
84
I T E R A T I O N A L G O R I T H M S FOR C E R T A I N SUMS O F SQUARES
The alternating
+1 and
- 1 in the l a s t c o l u m n , w h i c h s h o w s a c o n s t a n t p a t t e r n , s u g g e s t s the
p o s s i b i l i t y of a n a l g o r i t h m .
all
c.
[Feb.
If w e c a n find a l l b ,
say, from
b 3 = 12 o n , we w i l l a l s o h a v e
A f t e r s o m e t r i a l s and e r r o r s , we o b t a i n
Iteration Algorithm 1
6-2 - 0 = 12
6-12-2
= 70
6-70 - 12 = 408
6-408 - 70
= 2378
6-2378 - 408
13860
6-13860 - 2378 =
80782
w h i c h y i e l d s e a s i l y the n e x t four r e s u l t s :
l2 +
l
4 0 8 2 +*
408 2 =
577 2 =
(24 2 + l ) 2
(58 2 - l ) 2
2378
2378
I 2 + 13860 2 + 13860 2 =
l
+ 80782
S i m i l a r l y , we a p p r o a c h the c a s e
a = 2,
22 +
2
+ 80782
22 +
19601 2 = (140 2 + l ) 2
= 114243 2 = (338 2 - l ) 2 .
We h a v e by h a n d and a t a b l e of s q u a r e s :
l2 +
2
3363
22 =
3 2 = ( l 2 + 2) 2
72
(32
)2
8 2 + 16 2 = 18 2 = (42 + 2) 2
2 2 + 2 1 2 + 42 2 = 47 2 = (7 2 - 2) 2 .
T h e a l t e r n a t i n g +2 and
- 2 in t h e l a s t c o l u m n , w h i c h s h o w s a c o n s t a n t p a t t e r n , s u g g e s t s the
p o s s i b i l i t y of a n a l g o r i t h m .
all
c.
If we c a n find a l l b ,
say, from
A f t e r s o m e t r i a l s a n d e r r o r s we o b t a i n :
3-3-1
= 8
3-8 - 3 = 2 1
3-21 - 8 = 55
3 55 - 2 1 = 144
3-144 - 55 = 377
3-377 - 144 = 987
w h i c h y i e l d s e a s i l y the n e x t f o u r r e s u l t s :
b3 = 8
o n , we w i l l a l s o h a v e
1974]
I T E R A T I O N A L G O R I T H M S FOR C E R T A I N SUMS O F SQUARES

22 +
2
55 2 +
110 2 =
123 2 = ( l l 2 + 2) 2
322 2 = (18 2 - 2) 2
754 2 =
843 2 = (29 2 + 2) 2
+ 144
2 2 + 377 2 +
2
+ 987
S l i g h t l y d i f f e r e n t b e h a v e s the c a s e
32 +
3
a = 3.
80
3 2 + 154 2 +
3
H e r e the d o u b l y a l t e r n a t i n g
the exceptional values
two a l g o r i t h m s .
57
+ 684
12 2 =
32
(0 2 + 3) 2
?2
(2 2 + 3) 2
13 2 =
(4 2 - 3) 2
54
57
240
253 2
= (16 2 - 3) 2
462 2 =
487 2
= (22 2 + 3) 2
+ 2052
+3 and
We h a v e b y h a n d a n d a t a b l e of s q u a r e s :
18 2 +
2
= 2207 2 = (47 2 - 2) 2 .
02 =
42 +
32 +
+ 1974
02 +
32 +
2
288
85
= 2163
- 3 in the l a s t c o l u m n would show a c o n s t a n t p a t t e r n ,
and 2 1 6 3
could be e l i m i n a t e d .
if
T h i s s u g g e s t s the p o s s i b i l i t y of
To obtain further r e s u l t s , we w r i t e an I n t e g e r - F O R T R A N p r o g r a m for the
I B M 1130 w h i c h y i e l d s
32 +
3038 2 +
9114 2 =
9607 2 =
(98 2 + 3) 2
32 +
5848 2 +
17544 2 =
18493 2 =
(136 2 - 3) 2
32 +
25974 2 +
77922 2 =
82137 2
32 +
3
115364
222070 2 +
986328
+ 4380794
346092
666210 2 =
(604 2 - 3)2
702247 2 =
(838 2 + 3)2
364813 2
2
2958984
+ 13142382
= 13853287 2 = (3722 2 + 3) 2
3119043
Now w e w a n t to find an a l g o r i t h m w h i c h s h o u l d g e n e r a t e the s e q u e n c e

115364, 2 2 2 0 7 0 , 4 3 8 0 7 9 4 , .
b0 = - 4
i s the left n e i g h b o r of
l o g i c a l e x t e n s i o n to the left.
L e t the t e r m s
bt
= 0,
since
bt = 0,
b2 = 2 ,
and
80, 154, 3 0 3 8 ,
b3 = 4
be given;
3 2 + (-4) 2 + (-12) 2 = 13 2 = (4 2 - 3) 2
With t h i s t r i c k and s o m e t r i a l s a n d e r r o r s , we o b t a i n
38-2 - (-4)
2-80-2-4+2
= 80
= 154
38-80 - 2 = 3038
2-3038 - 2-154 + 80 = 5848
38-3038 - 80 =
115364
2-115364 - 2-5848 + 3038 = 222070

38-115364 - 3038 = 4380794
5848,
then
i s the
86
ITERATION ALGORITHMS FOR CERTAIN SUMS OF SQUARES
[Feb. 1974]
Now there remains only to find an algorithm which should generate 25974, 986328, . Here
we have not far to go, since such an algorithm is already contained in the former one, and
we obtain easily
38-684 - 18 = 25974
38-25974 - 684 = 986328
Finally, one could ask: Does there exist a general formula for solving x2 + y2 + z 2 = w 2 ?
The answer is yes.
2
x +y +z = w
Let x = p 2 + q2 - r 2 ,
becomes 0 = 0.
y = 2pr,
z = 2qr,
and w = p 2 + q2 + r 2 ;
then
But this formula has two shortcomings: (1) it uses f r a c -
tions, and (2) it employs roots, since, for example, the solution of 3 2 + 22 + 62 = 72 requires
p =
N/2/2,
q = 3N/2"/2,
and
r =
\/2 .
REFERENCES
1. Irving Adler, "Three Diophantine Equations," Fibonacci Quarterly, Vol. 6, No. 3 (1968),
pp. 360-369; Vol. 7, No. 2 (1969), pp. 181-193.
2.
E. P. Starke, Elem. Prob. E2151, Amer. Math. Monthly, 76 (1969), p. 187.
3. Edgar K a r s t ,
T,
A Four-Step Iteration Algorithm to Generate x in x 2 + (x + l) 2 = y 2 , n
Fibonacci Quarterly, Vol. 7, No. 2 (1969), p. 180.

4.
Zentralblatt fur Mathematik, 151(1968), pp. 211-212.
CERTAIN CONGRUENCE PROPERTIES (MODULO 100) OF FIBONACCI NUMBERS

MICHAEL R. TURNER
Regis College, Denver, Colorado 80221
Remark.
F
= P
It was originally observed by the author that if p is a prime ^ 5 ,
then
(mod 100). Further study led to this theorem which characterizes those Fibonacci
numbers which terminate in the same last two digits as their indices. The original observation is proved as a corollary to the theorem.
Theorem.
n
Proof.
= n (mod 100) if and only if

1, 5, 25, 29, 4 1 , o r 49 (mod 60) o r n = 0 (mod 300) .
From [l] , we have the well known formula

m-1
where m = n if n is odd, and m = n - 1 if n is even.

Lemma 1. F 6 Q k E 20k (mod 100).
Proof.
Observe that (1) implies

2n-1Fn = n + 5
(2)
n(n
- f"
2)
(mod 25) .
From [l] , we have for n, m > 2, (n,m) = d implies that (F , F

implies
2 6
"
l F
6 0 k
E 10k (mod 25).
) = F d - Now (2)
= 60k + 50k (60k - 1) (60k - 2) (mod 25), which reduces to
Since
20
= 1 (mod 25), we get F 6 Q k = 20k (mod 25).
60k, it follows that F 6 divides
F6Qk.
Now F 6 = 8,
26
k~lF60k
Since 6 divides
so F g ( ) k = 0 (mod 4).
Combining
this with F 6 Q k = 20k (mod 25), we get F 6 Q k = 20k (mod 100), which proves Lemma 1.
We now prove one of the congruences in the theorem.
(3)
n = 1 (mod 60)
Proof.
Clearly n = 1 implies F
60k + 1 implies F
implies
= n (mod 100) .
= n (mod 100). Assume that for all k < N , n =
= n (mod 100). Now if n = 60N + 1 for even N,
then n = 120k + 1
for k = N/2 < N.

From [ 2 ] , we have the following identity, which will prove extremely useful in what
follows.
(4)
F
n+m+1
= F F + F
F
n m
n+1 m+1
87
88
CERTAIN CONGRUENCE PROPERTIES (MODULO 100)
[Feb.
In particular,
+ F2
60k
60k+l
Y2
-p
=
120k+l
Using Lemma 1 and induction hypotheses, we get

F
n =
L60k+l
i j.i
L60k
= (60k
If n = 60N + 1 for odd N5
I)2
(20k)2
= l20k + 1 = n
then n = 120 k + 60 + 1
for
(mod
100) .
k = (N - l ) / 2 . Then
F^ = FF 1 o n l r FF ^ n ++ FF 19nLr _ l _F1 F fi1 . Inspection of any large table such as [3] verifies that F 6 1
n
120k 60
120k+1 61*
61 (mod 100). Thus, by Lemma 1 and induction hypothesis, we have
F
= 40k-20 + (120k + 1).61 s 120k + 60 + 1 s n (mod 100) .
This proves the congruence.

Lemma 2.
F AI __, = 20k-F n + (60k

+ 1)-F (mod 100) .
v
60k+n
n-1
n
Proof. Lemma 2 follows from (3) and Lemma 1. The remainder of the proof is divided
into five c a s e s .
Case 1. n = 1 (mod 5).
Assume F
implies 2 ~ F
= n (mod 100). Then F

= n (mod 25),
Also,
(5,n) = 1 and F
= n (mod 4) and
= n (mod 25).
Now (2)
since
n(n
- gfr
2)
= 0 (mod 25) .
= n (mod 100), so we may cancel the n and F
to get 2 ~ = 1
(mod 25). Since 2 belongs to the exponent 20 (mod 25), it follows that n = 1 (mod 20).
Thus n = 1 (mod 4). But F = n = 1 (mod 4), so F must be odd. But F
is even if
n
n
n
and only if n = 0 (mod 3), so n = 1 or 2 (mod 3). Combining r e s u l t s ,
n = 1 (mod 3) )
Jn
n = 1 (mod 20)
1 (mod 60)
Now suppose that n = 41 (mod 60).

F
or
n = 2 (mod 3) )
s ( m o d 20) | n
41 (mod 60) .
Let n = 60k + 4 1 . By Lemma 2,
= 20kF40 + (60k + 1)F 41
(mod 100) .
By inspection of tables, we have F 40 = 55 (mod 100) and F 4 1 = 41 (mod 100).

we have
F
= (60k + 41) + 20k-55 = 60k + 41 = n (mod 100) .
This result, along with (3), completes the proof of Case 1.
Therefore,
1974]
OF FIBONACCI NUMBERS
89
Case 2. n = 2 (mod 5).

This case is impossible,
for as in Case 1, it follows that n = 1 (mod 20), a
contradiction.
Case 3. n = 3 (mod 5).
Let n = 3 + 5k. Then from (2),
,,2+5^
2
F
Assuming F
Q-j-^k
ing 3-2
=n
__,
5(2 + 5k) (1 + 5k)
= n + n i - p
'-
(mod 100), we may cancel the F

Rk+fi
= 6 + 5-2-1 (mod 25). Thus 2
.
, ^.
(mod 25)c .
and n f s ,
since (n,25) = 1,
obtain-
(mod 25). But this congruence implies
5k + 6 = 0 (mod 20), or 5k = 14 (mod 20). This congruence is not possible, so case 3 is

impossible.
Case 4. n = 4 (mod 5).
4+^k
Assume F = n (mod 100). Let n = 4 + 5k. Then 3-2
= 6 + 5-3-2 (mod 25), so
n
5k- 5
2 = 1
(mod 25), and 5k = 5 (mod 20). Thus n = 5k + 4 = 9 (mod 20). F
and n
a r e therefore odd, so n = 1 or 2 (mod 3). Combining r e s u l t s ,
n = 1 (mod 3) )
n = 9 (mod 20) >n = 49 (mod 60)
or
n = 2 (mod 3) )
_ g ( m o d 2 0 ) ?n = 29 (mod 60)
Now suppose that n = 29 (mod 60). Let n = 29 + 60k.

F
By inspection of tables,
60kF28
60k+1F29
(mod
By Lemma 2,
100)
'
F 2 8 = 11 (mod 100), and F 2 9 = 29 (mod 100). Thus by Lemma 1,
we have
F
= 20k-11 + (60k + l)-29 = 60k + 29 = n (mod 100) .
Suppose n = 49 (mod 60). Let n = 49 + 60k.

F
By similar reasoning,
= 20k-F 48 + (60k + 1)F 49 = 20k-76 + (60k + l)-49 = 60k + 49 = n (mod 100) .
This result completes the proof of Case 4*

Case 5. n = 0 (mod 5).
Let n = 5s-k, where s ^ i, and (5,k) = l. We shall consider in order the possibilities n = 0, 1, 2, and 3 (mod 4). Assume F = n (mod 100). If n = 0 (mod 4), and
s = 1, then n = 5k5 where (5,k) = 1. Thus we get 2
F
= n (mod 25) from (2). Now
= n = 5k (mod 25) implies 2 ~ .5 = 5 (mod 25), so n = 1 (mod 4). But in this c a s e ,
the l a s t result is impossible, so it follows that s ^ 2. Also, since
must be even, we
have n = 0 (mod 3). Finally, n = 0 (mod 5 S ) implies n = 0 (mod 25).

have
Combining,
we
90
CERTAIN CONGRUENCE PROPERTIES (MODULO 100)
[Feb.
n = 0 (mod 3) J
n = 0 (mod 4) ) n = 0 (mod 300) .
n = 0 (mod 25) J
Let us suppose that n = 0 (mod 4); we have F
odd, so there are two combinations:
n = 1 (mod 3) J
n = 2 (mod 3)
n s 1 (mod 4) > n = 25 (mod 60)
or
n = 1 (mod 4) } n = 5 (mod 60)
n = 0 (mod 5) 1
n = 0 (mod 5)
If n = 2 (mod 4), we have

n = 0 (mod 3)
n s 2 (mod 4) } n = 30 (mod 60) .
n = 0 (mod 5)
Let n = 30 + 60k.
By Lemmas 1 and 2,
But this reduces to F
60k + 30
20kF
29
(60k + 1)F
s 20k + 40 (mod 100). Now F
2 0 k + 4 0 = 60k + 30 (mod 100),
or
30
(mod
100)
"
= n = 30 + 60k (mod 100)
40k = 10 (mod 100),
which is impossible.
implies
If n = 3
(mod 4), we get two combinations:

n = 1 (mod 3) J
n = 2 (mod 3)
n s 3 (mod 4) In = 55 (mod 60)
or
n = 3 (mod 4) }n = 35 (mod 60)
n = 0 (mod 5) 1
n = 0 (mod 5)
The first congruence results in

F
and
n =
55 + 60k
= n = 55 + 60k implies
40k
+ 45
( m 0 d 100)
20k = 90 (mod 100),
which is impossible.
The second
congruence results in
F
and F
n =
35 + 60k
40k
+ 65
(mod
100)
'
= n = 35 + 60k implies 20k = 30 (mod 100), which is also impossible.

F
Then F
= F-
fift.
, so
= 20k-F 4 + (60k + 1)-F 5 = 60k + 5 = n (mod 100) .
1974]
OF FIBONACCI NUMBERS
91
Then
F n = F 2 5 + 60k,
so
F n = 20k-F 24 + (60k + 1).F 25 s 60k + 25 = n (mod 100) .
Finally, if n = 0 (mod 300), then 300 divides n,
F 300 = 0 (mod 100), and thus F
so F 300 divides F . By Lemma 1,
= 0 = n (mod 100).
This result completes the proof of the theorem.

Corollary.
Proof.
If p is a prime ^ 5 , then F
By the theorem,
= p 2 (mod 100).
F 5 = 5 (mod 100). If p is a prime > 5 ,
p = 1, 3, 7, 9, 11, 13, 17, o r 19
then
(mod 20) .
Thus p 2 = 1 o r 9 (mod 20). Since p 2 = 1 (mod 3), it follows that
p2 = 1 or
49 (mod
60).
REFERENCES
1.
G. H. Hardy and E. H. Wright, An Introduction to the Theory of Numbers,
2.
S. L. Basin and V. E. Hoggatt, J r . , "A P r i m e r on the Fibonacci Sequence, P a r t I, M

Fibonacci Quarterly, Vol. 1, No. 1 (Feb. 1963), pp. 65-72*
3.
Dov Jarden, Recurring Sequences, Rive on Lematematika, J e r u s a l e m , I s r a e l , 1958.
FIBONACCI AN PATHOLOGICAL CURVES

SANTOSH KUMAR
Armament Research and Development Establishment, Poona, India
There are many curves which possess peculiar properties not possessed by ordinary
curves.
These are the so-called "pathological curves 1 ' of mathematics.
In the present note
a few curves which a r e not normal and healthy and which possess idiosyncrasies have been
generated and analyzed.
It may be pointed out that these curves cannot be analyzed with the
help of ordinary calculus.

We generate Fibonaccian pathological curves as follows.
of length H , where H
Start with a square with side
is the generalized Fibonacci number obtained by the recurrence
relation
H
= H , + H 0
n-1
n-2
n > 2 ,
where H1 and H2 are any positive integers. Divide each side of the square into three p a r t s ,
two of length H Q and one p a r t of length H Q , a s shown in F i g . 1.
nA
nu
H
n-2
H
n-3
Y
T
n-2
Fig. 1
On each section with length H
figure and call this curve Sv
and H
erect a square outwards. E r a s e the basic side of this new

In S l s
shown in Fig. 2, we will have the sides of length H n _ 2
n~3*
92
Feb. 1974
FIBONACCIAN PATHOLOGICAL CURVES
93
T
fn-2
n-3
\n-d
Fig. 2
Again, each side is divided into three p a r t s . The sides of Si will be divided as follows:
Hn-2
0 = H / 1 + H I ,
/ I
n-4
n-5 + H n-4
H Q = H _ + H + H K.
n-3
n-5
n-6
n-5
Again we will form a square on the middle p a r t , forming curve S2 as shown in Fig. 3.
We continue dividing each side into lengths equal to two lower Fibonacci numbers and
constructing squares on the middle p a r t s until finally we get sides of length H2 and H1? and
have formed the curve S.
We called the curve S the Fibonaccian Pathological Curve.
In such a construction, it is of interest to find the total length L and the rate of inc r e a s e of a r e a A A of the curve S at the r
successive subdivision. It can be seen
r
r
that, for n > 3r,
Lr=
<
It is of interest to note that Et and H2 can be chosen as a r b i t r a r i l y small positive

numbers, and the curve after allowing all successive subdivisions will be a continuous curve
which is not differentiate anywhere.
It may be noted that an inwards curve can also be
94
FIBONACCIAN PATHOLOGICAL CURVES
Feb. 1974
generated on similar lines, but, due to lack of symmetry, the expression for obtaining the
total area after r successive subdivisions is difficult to obtain.
IRREDUCIBILITY OF LUCAS AND GENERALIZED LUCAS POLYNOMIALS

GERALD E. BERGUM
South Dakota State University, Brookings, South Dakota 57006
and
VERNER E. HOGGATT, JR.
1.
INTRODUCTION
In [5] , Webb and P a r b e r r y discuss several divisibility properties for the sequence
\ F (x)} of Fibonacci polynomials defined recursively by
(1)
F0(x) = 0,
Fife) = 1,
F n + 2 (x) = xF n + 1 (x) + F n (x),
n ^0 .
In particular, Webb and P a r b e r r y prove that F (x) is irreducible over the integral domain of
the integers if and only if p is a prime.
In [l] , Bergum and Kranzler develop many relationships which exist between the s e quence \ F (x) } of Fibonacci polynomials and the sequence \L (x)} of Lucas polynomials d e fined recursively by
(2)
L0(x) = 2,
Ljfe) = x 5
n+2
(x)
xL
n+l(x)
+ L
n(x)'
~ '
Specifically, Bergum and Kranzler show that

(3)
L (x) I L
(x)
iff
m = (2k - l)n,
k ^ 1.
With n = 1, we see that X | L (X) for all odd integers m so that the result of Webb
and P a r b e r r y does not hold for the sequence (L (x)}.
In [4] , Hoggatt and Long show that the result of Webb and P a r b e r r y does hold for the
sequence {u (x,y)} of generalized Fibonacci polynomials defined by the recursion
(4)
U 0 (x 5 y) = 0,
Ui(x,y) = 1,
n+
2(x'y)
xU
n+l(x'y)
+ yU
(x y)
'
'
- -
The purpose of this paper is to obtain n e c e s s a r y and sufficient conditions for the i r reducibility of the elements of the sequence {L (x)} as well as the elements of the sequence
{v fe,y)} of generalized Lucas polynomials defined by the recursion
(5)
V 0 fe,y) = 2,
Vi(x,y) = x,
V n + 2 fe,y)
= xV n + 1 fe,y) + yV n (x 5 y),
The first few t e r m s of the sequence { V fe,y)} are

95
n > 0.
96

n
[Feb.
V (x,y)
n J
X
2
x + 2y
x 3 + 3xy
x 4 + 4x 2 y + 2y2
x 5 + 5x3y + 5xy2
x 6 + 6x4y + 9 x V + 2V3
x 7 + 7x5y + 14x3y2 + 7XV3
7
8
x + 8x6y + 20x4y2 + 16x2^
+ 2V4
x 9 + 9x7y + 27x5y2 + SOxV + 9XV4
Observe that L (x) = V (x,l) so that with Jy = 1, we also have the first nine t e r m s
n
n
of the sequence { L (X)} .
2. IRREDUCIBILITY OF L (x)
nN
The basic fact that we shall use is found in [2, p. 77] and is
Theorem 2.1. (Eisenstein's irreducibility criterion.)
F(x) = a x
n
+a
.. x ~
n-1
For a given prime p,
let
+ + a<x
+ au0
2
be any polynomial with integral coefficients such that

a
nx
= a 9 = = a 0 = 0 (mod p ) ,
\\u
a
n
0 (mod p ) ,
a 0 0 (mod p 2 )
then F(x) is irreducible over the field of rationals.

To establish our first irreducibility theorem, we use the following.
Lemma 2.1. Every coefficient of L
2n
(x), except for the leading coefficient, is divisible
by 2 and 4 does not divide the constant term.

Proof. If n = 1 then L2(x) = x2 + 2 and the lemma is obviously true.
Assume the
lemma is true for n.

In [1] , we find
L 2 k (x) = L2k(x) - 2 ( - l ) k .
(6)
Hence,
(7)
L 2 n + 1 (x) = Lj n (x) - 2 .
By the induction hypothesis, it is obvious that L
of L
.. (x) is divisible by 2.
2n+l
,-,(x) is monic and every coefficient
F u r t h e r m o r e , since L
2^
(x) has constant t e r m
+2 we see
that L 2 (x) has constant t e r m +4, thus L ^(x) has constant t e r m +2. Therefore,
v
2n+1
2n
constant term of L
.. (x) is divisible by 2 but not by 4 and the lemma is proved.
An immediate result of Lemma 2.1 with the aid of Theorem 2.1 is
the
1974]
97
Theorem 2.2. The Lucas polynomial L . (x) is irreducible over the rationalsfor k ^ 1.
2K
Although L (x) is not irreducible if p is a p r i m e , we can show that
L (x)/x is i r -
reducible for every odd prime p.

F i r s t we note, as is pointed out in [l] , that
L n (x) = an + /3 n ,
(8)
where a = (x + ^x 2 + 4)/2 and |3 = (x - N/X2 + 4 ) / 2 .

L n (x) = (x + N;x2 + 4 ) n / 2 n + (x n
Hence, if n = 2m + 1 we have
N/xr+T)n/2n
n
* *
n k
k/2
x - (a x 2 + 4 )
= 2 I > . x i r i r v + 4 r - + > 11 K-D^X
(X2 + 4)
\k=o
k=o \
'
(9)
= 2
k=0 \
m
/
k
ti:(i)(^
(n 1
= 2- - >WL?W k U n - 2s 2 2s .
k=0 s=0
Therefore,
m
For each s,
(11)
\ /
22s,
k=0 S=0 X / V /
0 < s < m, we see that the coefficient of x
-(n-2s-l)>
, - 11-1
W0)
n = 2m + 1
is
= 2m + l .
k=s
When s = 0, we have the leading coefficient of L (x) which is 1 so that
When s = m in (11), we have the constant term of L (x) which is n. If we now let n be an
odd prime p and recall that p divides
U)
98
[Feb.
if p is a p r i m e , then p is a factor of (11) for each value of s,
Hence, by Eisenstein's criterion, the following is true.

Theorem 2.3. The polynomials L (x)/x are irreducible over the rationals if p is an
odd prime.
By (11) and the fact that the coefficients of L (x) are integers, we have
Corollary 2.1. If n = 2m + 1 then 2
divides
UW)
for any s such that 0 ^ s ^ m.
Using (3) together with Theorems 2.2 and 2.3, we have
Theorem 2.4.
(a) The ^Lucas polynomials L n (x), n ^ 1,

rationals if and only if n = 2 for some integer k 1.
are irredicuble over the
(b) The polynomials L (x)/x, n odd, are irreducible over the rationals
if and only if n is a prime.
3.
IRREDUCIBILITY OF V (x,y)
It is a well known fact that

(13)
U (x,y) =
n
0n
" P ,
a. -
n s> 0
and
V n (x,y) = an + / 3 n ,
(14)
n ^ 0 ,
where a = (x + \/x2 + 4 y ) / 2 and j3 = (x - ^x 2 + 4 y ) / 2 .

In [4] , we find
Lemma 3.1. (a) For n ^ 0,
TT
U (x y) =
n '
[(n-l)/2]
\ ^
/ n - k - 11 n-2k-l k
LJ I
k=0
)X
'
(b) For n > 0, m > 0,

(U (x,y),
U (x,y))
= U,
, (x,y)
.
J
J
m
n ,J
(m,n)
Using (13) and (14), a straightforward argument yields
1974]

Lemma 3.2.
(a) V ^ x . y ) = y U ^ x . y ) + U n + 1 ( x , y ) ,
(b)
(C)
U 2 n (x,y) = U n (x,y)V n (x,y),

U
2n(x'^V(2k+l)n+l(x>y>
n a 1;
n 2= 0 ;
^ k - D n ^
V
(2k+l)n(x'y)U2a+l(x'y)-
Using (a) of Lemma 3.1 and 3.2, we have, for n 1,
,, ,
V^
[(n-2)/2] .
.
XP
n - k - 2 )
2^ \
k=0
n-2k-2 k+1 ^
that
[n/2] ,
v
V""* / n - k \
XJ
'
I k )
k=0
[n/2]
99
n-2k k
'
[n/2]
(15)
k=l
k=0
'
[n/2]
/n - k - l\
1
k=l \
k-i
n
;
n-2k k ,
+ x
Hence,
Lemma 3.3.
(a) For n ^ 1, V (XjV2) is homogeneous of degree n.

(b) If n is odd then x is a factor of V (x,y 2 ) and V (x,y 2 )/x
is h o -
mogeneous of degree n - 1.
By (b) of Lemma 3.1, (U
(x,y), U
(x,y)) = 1. Using this fact together with (b) of
Lemma 3.2 and induction on k in (c) of Lemma 3.2, one obtains

Lemma 3.4. If k > 1 then V n (x,y) | v ( 2 k - i ) n ( x ' y ) '
In [3, p. 376, Problem 5 ] , we find
Lemma 3.5. A homogeneous polynomial f(x,y) over a field F is irreducible over F
if and only if the corresponding polynomial f(x, 1) is irreducible over F.
Using Lemmas 3.3 and 3.5 with Theorem 2.4, we have
2
Theorem
- 3.1. (a) The polynomials Vn (xjV ) are irreducible over the rationals if and
only if n = 2 for some integer k ^ 1.
(b) The polynomials V (x,y2)/x, n odd a r e irreducible over the r a tionals if and only if n is an odd prime.
Since f(x,y) is irreducible if f(x,y 2 ) is irreducible and x is a factor of V (x,y) for
n odd by (15), we apply Lemma 3.4 and Theorem 3.1 to obtain
Theorem 3.2.
only if n = 2
(a) The polynomials V (x,y) are irreducible over the rationals if and
for some integer k greater than o r equal to one.
100
IEREDUCIBILITY OF LUCAS AND GENERALIZED LUCAS POLYNOMIALS

(b) The polynomials V (x,y)/x, n odd,
Feb. 1974
a r e irreducible over the r a -
tio nals if and only if n is an odd prime.

Letting y = 1 and n = 2m + 1 in (15), we see that
/..^
(16)
/ v/
n(x)/x
\ " " s n - k - 11 n
2 J
k=l
kX
k-1
n-2k-l ,
+X
n-1
'
'
Comparing the coefficients of x ~
in (16), 1 ^ s ^ m ,
with the result obtained
in (11), we have
Corollary 3.1. If n = 2m + 1 and 1 ^ s ^ m then
,-w-2(iX:) - (n;: il) f

REFERENCES
1. G. B e r g u m a n d A . Kranzler, Linear Recurrences-Identities and Divisibility P r o p e r t i e s ,
unpublished.
2.
G. Birkhoff and S. MacLane, A Survey of Modern Algebra, The Mac Millan Company, New
3.
G. B i r k o f f a n d S . Mac Lane, Algebra, The Mac Millan Company, New York, N. Y., 1967.
4.
V. E. Hoggatt, J r . , a n d C . T. Long, "Divisibility P r o p e r t i e s of Generalized Fibonacci
York, N Y. , 1965.
Polynomials," Fibonacci Quarterly, to appear April, 1974.

5. W. A. Webb and E. A. P a r b e r r y , "Divisibility P r o p e r t i e s of Fibonacci Polynomials,"
Fibonacci Quarterly, Vol. 7 (Dec. 1969), pp. 457-463.
ELEMENTARY PROBLEMS AND SOLUTIONS

Edited by
A. P. HILLMAN
University of New Mexico, Albuquerque, New Mexico 87131
Send all communications regarding Elementary Problems and Solutions to Professor

A. P. Hillman, Dept. of Mathematics and Statistics, University of New Mexico, Albuquerque,
New Mexico 87131. Each problem or solution should be submitted in legible form,
prefer-
ably typed in double spacing, on a separate sheet or sheets, in the format used below. Solutions should be received within four months of the publication date,
DEFINITIONS
The Fibonacci numbers F
F ô = F ^ + F , Fu0
n+2
n+1
n'
and the Lucas numbers L

satisfy
J
n
n
= 0, Fi = 1 and L ^ 0 = L x 1 + L , L 0 = 2, L,
l = 1.
*
n+2
n+1
n
PROBLEMS PROPOSED IN THIS ISSUE

B-274
Proposed
by C. B. A. Peck,
State College,
Pennsylvania.
Approximate ( \IE - l ) / 2 to within 0.002 using at most three distinct familiar symbols.
(Each symbol may represent a number or an operation and may be repeated in the expression.)
B-275
Proposed
Show that
B-276
Proposed
by Warren
Cheves,
Littleton,
North
F mn = L m F m(n-l)
, -v + ( - l ) m
by Graham
Lord,
Temple
Carolina.
F m(n-2)
, Qv .
University,
Philadelphia,
Pennsylvania.
Find all the triples of positive integers m , n, and x such that

F, = x m
h
B-277
Proposed
where
by Paul S. Bruckman,
Prove that L 2 n ( 2 k + 1 ) = L 2 n
B-278
Proposedby
B-279
and
University
m > 1.
of Illinois,
Chicago
Circle,
III.
of Illinois,
Chicago
Circle,
III.
(mod F 2 n ) .
Paul S. Bruckman,
Prove that ^(2n+1}(4k+1)
h = 2
- ^2n+1
University
(mod F 2 n + 1 ) .
Proposed
by Richard M. Grassl, University
of New Mexico, Albuquerque,
New
Mexico.
Find a closed form for the coefficient of x in the Maclaurin s e r i e s expansion of
0
(x + 2x)/(l - x - x 2 )
101
102
[Feb.
SOLUTIONS
SEVEN DO'S FOR TWO SUSY'S
B-250
Proposed
by Guy A.R.
Guillotte,
Montreal,
Quebec,
Canada.
DO
YOU
LIKE
SUSY
In this alphametic, each letter stands for a particular but different digit, nine digits
being shown h e r e .
Solution
What do you make of the perfect square sum SUSY ?
by Raymond
E. Whitney,
Lock
There are four possible SUSYTs.
Haven
They a r e
State College,
Lock Haven,
2025, 3136, 6561, 8281.
Pa.
SUSY = 2025
leads to the six solutions shown below:
76
560
1389
2025
86
560
1379
2025
49
590
1386
2025
89
590
1346
2025
98
580
1347
2025
48
580
1397
2025
SUSY = 3136 leads to one solution:

57
671
2408
3136
The other two 4-digit numbers lead to no solutions. Thus the likelihood is that SUSY =
2025 and she is definitely 2025 or 3136.
Also solved by Richard Blazej, Donald Braffitt,
Paul S. Bruckman,
Juliana D. Chan,
Warren Cheves, Herta T. Freitag, Ralph Garfield,
Myron Hlynka, J. A. H.
Hunter,
John W. Milsom,
C. B. A. Peck, Jim Pope, Richard
W. Sielaff
Charles
W.
Trigg,
Lawrence
Williams,
David Zeitlin,
and the
Proposer.
FAIR GAME
B-251
Proposed
by Paul S. Bruckman,
San Rafael,
California.
A and B play a match consisting of a sequence of games in which there are no ties.
The odds in favor of A winning any one game is m.
The match is won by A if the number
of games won by A minus the number won by B equals 2n before it equals -n.
in t e r m s of n given that the match is a fair one, i. e. , the probability is
win the match.
1/2
Find
that A will
1974]
103
Solution by Dennis Staples, The American School in Japan, Tokyo, Japan,

The game specifies that A wins whenever A's wins - B f s wins reaches 2n before it
reaches -n.
Said another way, A wins whenever A f s wins - B f s wins + n reaches 3n be-
fore it reaches 0.
Recalling the notion of Markov chains, let u(i) be the probability that A reaches
(in other words, wins), given that i = A f s wins - B's wins + n.
3n
Using techniques designed
for solution of Markov chain problems, it can be found that
u(i)
( l - m V _ (l - m \
mj
=V m / ;\
-NO
3n
Since A and B begin competition when i = n, and since A's chances of winning are
to be
1/2,
1 - m I
1 - m 3n
m / " \
m /
3n
J = \
( i ^r-(Hr) , -( i irJ
3n
/ 1
TVI
+ 1 = 0
This is a r a t h e r familiar equation, and it can be easily shown that the roots, (1 - m ) / m ,
are equal to the following when n = 1 :
-1 +
_ _
_1,
-1 -
N/5
or
N/5
Of these values, only (-1 + \[E)/2 = 0.618- is acceptable in the case we are considering.
Thus, when n = 1,
~m
m
= 0.618
1 - m = (0.618 ) m
m
For the general c a s e ,
1.618-..
-618-"
'
104

,1/n
I 1.618 )
Also
solved
by Ralph
Garfield
and the
[Feb.
= (0.618 ) l / n
Proposer.
SOMEWHAT ALTERNATING SUM OF TRINOMIALS

B-252
Proposed
by Wray
Pennsylvania.
G. Brady,
Slippery
Rock
State College,
Slippery
Rock,
Prove that
(-Dk _ i
A--^
i! j ! k!
n!
New
York.
i+j+k=n
Solution
by Harvey
J. Hindin,
Dix Hills,
The multinomial theorem (G. Chrystal, Textbook of Algebra, P a r t 2, Dover Reprint,

New York, 1961, page 12), may be stated as:
(1)
(x
+ y
z)n=
X )
TrfWxVz
i+j+k=n
If we let x = y = 1, and z = - 1 , we have:
(1 + 1 - If - 1 = nt 2
(2)
i+j+k=n
or
(3)
X ) FJr^-3-
Q-E-D-
i+j+k=n
Problem 34> page 20 of Chrystal is similar.
Also solved by Paul S. Bruckman,
Michael Capobianco,
Timothy B. Carroll,
T. Freitag,
Ralph Garfield,
Lawrence
D. Gould, Myron Hlynka,
Graham
C. B. A. Peck, Raymond
E. Whitney,
David Zeitlin,
and the
Proposer.
Herta
Lord,
TRINOMIAL EXPANSION WITH F'S AND L f S

B-253
Proposed
by Wray
Pennsylvania.
Prove that
G. Brady,
Slippery
Rock
State
College,
Slippery
Rock,
1974]

( - l ) k L+^2 k
Y^
Z-J
i+j+k=n
Solution
by C. B. A. Peck,
i! j ! k!
Z^J
i+j+k=n
State College,
105
(-l)kF
j+2k
it j k!
Pennsylvania.
In the trinomial expansion
^"(u.k)
-4 4-1;- :r
i+j+k=n
(x + y + z)
'
where
(i.J.k) = '"*'*
z = -a2 (-j82).
with i + j + k = n, let x = 1, y = a(fi),

2
expressions are proportional to (1 + a - a )

Comment.
From the Binet formulas, the two
(1 + /3 - /32)
= 0 0n = 0 .
A number of solvers pointed out that the F T s or L's could be replaced by
generalized Fibonacci numbers.

Timothy B. Carroll,
Garfield,
Harvey J. Hindin,
Graham Lord, David Zeitlin,
Herta T. Freitag,
Ralph
and the
Proposer.
MORE OR LESS LUCAS

B-254
Proposed
Let A
by Clyde
= a +b +c
x - 2x - 1 and d, e,
A
A,
Bridger,
and B
Springfield,
= d +e
Illinois.
+ i' , where a, b, and c a r e the roots of

3
and f a r e the roots of x - 2x2 + 1.
Find recursion formulas for the
and for the B . Also express B in t e r m s of A .

n
^
n
n
Solution
by Graham
Lord,
Temple
University,
Philadelphia,
The roots of x3 - 2x - 1 a r e
Pennsylvania.
- 1 , a, (3 and of x 3 - 2x2 + 1 a r e 1, a, /3 where

th
have their usual (Fibonacci) meaning. Hence if L is the n
Lucas number, then
A = (-l)n + L
n
n
Consequently from the properties of L
A
B
and
l0
n+3
-LO
n+3
and
B = 1 + L .
n
n
for n ^ 0 :
= 2A ,.. + A ,
n+1
n
= 2B ^ 0 - B ,
n+2
n
a,p
106

4A
= <* A n ' 2
n
J A +
B3
Feb. 1974
n even
n odd
B n = AQ + 1 - ( - l ) n
Q.E.D.
Also solved by Richard Blazej,

Paul S. Bruckman,
Timothy B. Carroll,
Herta T.
Freitag, Ralph Garfield, Robert McGee and Juliana D. Chan, Raymond E.
Whitney,
Gregory Wulczyn,
David Zeitlin,
and the
Proposer.
FIBONACCI CONVOLUTION REVISITED
B-255
Proposed
by L. Carlitz
North
Carolina.
and Richard
Scoville,
Duke
University,
Durham,
Show that
k ) =E
7
2k<n
Solution
by C. B. A. Peck,
k V k = [tn - DF n+1 + (n + DF^l/5 .
k=0
State College,
Pennsylvania.
The l.h. result is proved by Carlitz in the Fibonacci Quarterly, Vol. VII, No. 3, pp. 285286 (proposed by Hoggatt as H-131 in Vol. VI, No. 2, p. 142), so we confine ourselves to the
r . h . result (stated by Wall in Vol. I, No. 4, p. 28). The result for n = 0 is just F 0 F 0 i 0 =
0-0 = 0 = [-1-1 + 1-1]/5 = [(0 - 1)F 0 + 1 + (0 + l ) F 0 . i ] / 5 . and for n = 1 is F 0 F 1 - 0 + F J F ^ J =
0-1 + 1-0 = 0 = [0-2 + 2-0]/5 = [(1 - 1)F 1 + 1 + (1 + D F ^ i l / 5 . Suppose the result true for all
n up to some m > 1.
E
k=0
Then
F, F _ , , . = >
F. F
. + \ ^ F. F
- . + F x1F ^ . ^ + F F
k m+l-k
Zd k m-k
/ J k m-l-k
m+1 m+l-(m+l)
m m-l-m
k=0
k=0
= L[(m - 1)F _,_-, + (m + 1)F

-, + (m - 2)F
+ mF
+ F 5J] / 5
m+1
m-1
m
m-2
m
= L[ m F
J0
m+2
+ (m + 2)F 1J / 5 ,
m '
so that by the Second Principle of Finite Induction, the right-hand result is true.
Timothy
Garfield,
Phil Tracy,
and the
Proposer.
B. Carroll,
Herta
T. Freitag,
Ralph
ADVANCED PROBLEMS AND SOLUTIONS

Edited by
RAYMOND E. WHITNEY
Lock Haven State College, Lock Haven, Pennsylvania 17745
Send all communications concerning Advanced Problems and Solutions to Raymond E.

Whitney, Mathematics Department, Lock Haven State College, Lock Haven, Pennsylvania
17745.
This department especially welcomes problems believed to be new or extending old
results.
P r o p o s e r s should submit solutions o r other information that will a s s i s t the editor.
To facilitate their consideration, solutions should be submitted on separate signed sheets

within two months after publication of the problems.
H-230
Proposed
by Gregory
Wulczyn,
Bucknell
University,
Lewisburg,
Pa.
(a) If 5 is a quadratic nonresidue of a prime p (p f- 5), then P | F . / -v, k a positive integer.

(b) If 5 is a quadratic residue of a prime p, then P | F , ,
H-231
Proposed
by L. Carlitz,
Duke
University,
Durham,
-v, k a positive integer.

North
Carolina.
1. Let A 0 = 0, At = 1,
A
2k+1
2k
A
A
2k+1 "
2k+2
2k+2
Find A .
n
2k-1
2k+l
2k
2k
2k+1 "
2k-1
B
2k
Find B .
H-232
Proposed
by R.
Garfield,
the College
Define a sequence of polynomials,
1 - (x + l)t
{ G , (x)},
- xt
of Insurance,
Q
New York,
New
York.
as follows:
= G k (x)t k
k=Q
1.
Find a recursion formula for G, (x).
2.
Find G, (1) in t e r m s of the Fibonacci numbers.
3.
Show that when x = 1, the sum of any 4 consecutive G numbers is a Lucas number.
107
108
H-233
Proposed
by A. G. Shannon, NSW Institute
The University
of New England,
Armidale,
[Feb.
of Technology,
Australia.
Broadway,
and
The notation of Carlitz* suggests the following generalization of Fibonacci numbers.

Define
(r)
nk+k
, nk+k w . k
,k
V
= t(a
- b
)/(a - b x ) ,
where k = r - 1, and a,b a r e the zeros of x 2 - x - 1, the auxiliary polynomial of the o r (2)
dinary Fibonacci numbers, f
Show that
oo
2 fn)xR
n=0
(a)
Let f, = (a
- b
>
^-
1/(1
~ ^
+ bk)x
)/(a - b), and prove that
(?)(n;m)CiC2Bf
(Note that when r = 2 (and so k = 1),
f. = f,
- = 1,
f,
= 0,
and (b) reduces to the
well known
<r - s (n-j)
SOLUTIONS
SUCCESS!
Editorial Note.
this is incorrect.
We previously listed H-61, H-73, and H-77 as unsolved.
H-61 is solved in Vol. 5, No. 1, pp. 72-73.
No. 3, pp. 255-256.
However,
H-73 is solved in Vol. 5,
H-77 is solved in Vol. 5, No. 3, pp. 256-258.

ANOTHER OLDIE
H-62
Proposed
by H. W. Gould,
West
Virginia
University,
Morgantown,
W.
Va.
Find all polynomials f(x) and g(x), of the form
f(x + 1) = 2^
-xJ '
a. an integer,
j=0
*L. Carlitz, "The Characteristic Polynomial of a Certain Matrix of Binomial Coefficients,"
Fibonacci Quarterly, Vol. 3 (1965), pp. 81-89.
1974]
109
g(x) = ^ b . x J ,
b. an integer,
j=0
such that
2 J x 2 f % + 1) - (x + D V W } + 3{x 2 f 2 (x + 1) - (x + l) 2 g 2 (x)}
+ 2(x + l ) | x f ( x + 1) - (x + l)g(x)j- = 0 .
H-87 Proposed by Monte Boisen, Jr.,
San Jose State University, San Jose, Calif.
Show that, if
u0 = u2 = u3 = = u n _ 1 = 1
and
U
k-1
+ U
k-2
''
Vn
"
then
1 - x2 - 2x3 - - (n - 2)x 11 " 1
i
l - X - X ^ - ' - ' - X
Eukxk
.
k=0
Solution by Clyde
Write
and
A.
Bridger,
Springfield,
Illinois.
g(x) = c 0 + Cjx + c 2 x 2 + . . . + c n _ 1 x I 1 ~
f (x) = a0 + ajx + a2x
+ + a x
g/f = q(x) = A0 + A l X + A2x2 + + A n x
where a0 and a
,
-.
+ A ^ x
are not z e r o , at least one c, is not z e r o , and
f(x) = 0 has no multiple
roots.
Then g/f generates a recurrence of length n,
as is at once apparent by either long
division or by equating coefficients of like powers of x in g(x) = f(x)-q(x).

depend entirely on the c ' s ,
as the following set of equations shows.

c0 = a 0 A 0
Ci = a 0 A 1 + a ^ o
c 2 = a 0 A 2 + Q,1A1 + a 2 A 0
n - 11
= , a 0 A n + a-. A n + + a n Au0
^ n-1
* n-2
n-1
0 = a0An + a1An_1 + . . . + anA0

0 = a0Ak + a1Ak_1 + . + akA0
(k ^ n)
The first n A f s
HO
Beginning with the (n + 1)

preceding A ' s .
st
equation, A
The general t e r m A,
equation that the fraction g/f
[Feb.
can be expressed directly in t e r m s of the
for any k ^ n
n - 1
provides the formula o r difference
generates.
To solve the given problem, one sets c 0 = 1 and c, = -(n - 1) for i = 1 to i = n 1, an = 1, a. = - 1 for i = 1 to n, and A. = u. .
^
FIT TO A " T "

H-197
Proposed
by Lawrence
Somer,
University
of Illinois,
Urbana,
Illinois.
Let
}.i
be the t-Fibonacci sequences with positive entries satisfying the recursion relationship:
t
,(t)
- V*
wt
u
= > u . .
n
A* n-i
i=l
Find
lim
t 5
n-*oo
Solution
by the
Proposer.
By analyzing the convergents of continued fractions, one can easily see that for any
fixed t, and any initial e n t r i e s ,
will be a constant.
u} , u2 , , u, ,
(t)
*.
n+l
lim rrr
n -* oo un(t)
Let
At)
At)
...
(t) _
For this choice we have

(t)
Let
Vi
t+l
= 2
-- 1
and
(t)
V2
= 2
0 t+2
"3
(t)
* = lim
n oo
J*
u
(t)
If one examines the convergents of the continued fractions, he finds
1974]

u(t)
u
for large enough n.
111
u(t)
Vi
Thus
2
< *!/ <
fc
2 _ J - < /> <

2l
t+1
- 1
2
2
t + 1
- l
and
Jim
f = 2 .
I oo
It thus follows that the desired limit is 2.
Also solved by P. Tracy and one unsigned solver.

PELL-MELL
H-198 Proposed by E. M, Cohn, National Aeronautics and Space

Washington, D.C.
Administration,
There is an infinite sequence of square values for triangular numbers *

k2 = m(m + l ) / 2 .
Find simple expressions for k and m in t e r m s of Pell numbers, P
(pn+2
= 2P
n+l
+ P
n'
where P 0 = 0 and Vt = 1).
Solution by the Proposer,

Since (m, m + 1) = 1, the odd factor must be a square, say (2s + l) 2 . Then the even
factor is (2s 2 + 2s) or (2s2 + 2s + 1).
Let k 2 /(2s + l) 2 = q 2 ,
(After division by 2.)
so that either
2s 2 + 2s - q2 = 0
or
2s 2 + 2s + 1 - q2 = 0 .
Solving for s and r e - a r r a n g i n g ,
(2s + l) 2 = 2q2 1
k = q^2q 2 " l
*A. W. Sylvester, Amer. Math. Monthly, 69 (1962), p. 168.
112
It has been shown* that q
= P
discriminant itself equals P

n
[Feb. 1974]
for Diophantine solutions of such discriminants, and that the
Jl - P .
n+1
n
Furthermore,
P (P , - - P ) = i - P n .
n n+1
n
2 2n
Thus
2 2n
For even n,
m
= 2P2 ,
n
and for odd n,

m
= 2P2 1 .
n
Since even Pell numbers a r e alternately congruent to 0 (mod 4) and 2 (mod 4),
pairs of
values of k are of different parity.
Also
solved
by P.
Bruckman,
who also solved
H-192,
H-193,
and
H-194.
*E. M. Cohn, submitted to the Fibonacci Quarterly.
HERE AT LAST!! - FIBONACCI NOTE SERVICE

The Fibonacci Quarterly is offering a service in which it will be possible for its r e a d e r s
to secure background notes for articles.
This will apply to the following:
(1) Short abstracts of extensive r e s u l t s , derivations, and numerical data,,

(2) Brief articles summarizing a large amount of r e s e a r c h .
(3) Articles of standard size for which additional background material maybe obtained.
Articles in the Quarterly for which this note service is available will indicate the fact,
together with the number of pages in question.
Requests for these notes should be made to
Brother Alfred Brousseau

St. Mary's College
Moraga, California 94574
The notes will be Xeroxed.
The price for this service is four cents a page (including postage, materials and labor).
This service will begin with the April, 1974 issue, coming soon.

12 1

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

12 1

Загружено:

Авторское право:

Доступные форматы

THE FIBONACCI QUARTERLY

THE OFFICIAL JOURNAL OF

Linearly Eecursive Sequences of Integers

The Golden Ellipse

Generalized Hidden Hexagon Squares

Letter to the Editor

Fibonacci Sequences Modulo m

Phi: Another Hiding Place

Some Geometrical Properties

Sets of Binomial Coefficients

T. Long and V. E. Hoggatt,

Stufe of a Finite Field

Fibonaccian Pathological Curves

Elementary Problems and Solutions

Advanced Problems and Solutions

THE FIBONACCI QUARTERLY

WITH THE COOPERATION OF

Two copies of manuscripts intended for publication

in the Quarterly should be sent to Verner E. Hoggatt, J r . , Mathematics Department, San

should be done in India ink on either vellum o r bond paper.

Authors should keep a copy of

the manuscript sent to the editors.

For example for

n = 3, the allowable sequences are

denote the number of such s e -

We shall show that

As in the previous problem, it suffices to evaluate f 0 (m,n; j , k ) and 7 0 (m,n; j , k ) . We

_,(k, m - k)c n (k + 1, n - k - 1)}

f 0 (n 9 k,r) = J^ f j <n - L k> r ) = f < n - *> k> r>

satisfies the mixed r e c u r r e n c e

Now, for r < k,

Generally, for 1 < m < k + 1, we have

r < k in (2.4) we get

/ " ^ ^ k - r - l J ' ^ f k - r - l J ' V k - r j ' l r /

it follows that (2.4) holds for n ^. k + 1 > r provided we define

Moreover (2.4) holds for r > k, n = k + 1, since both sides vanish.

Then, by (2.4) and (2.5),

E 1212 fo(n - J - !fc'r - J )XI1 ^

Efo(n - J -1. k>r - j>* n ~ 3 ~V- j

E E ck<n - r> r>*v

and, in view of (2.2),

The following relations follow from the definition.

where it is understood that

It follows from (3.2) and (3.4) that

Then by (3.5) and (3.7),

To get explicit formulas we take

^ ^ X) cr-l(k' m " k)c s-l (ks

,(k + 1, m - k - l)c n (k, n - k)}

+ c _x(k, m - k)c g _ 1 (k + 1, n - k + 1)}

f(m,n; r , s ) = ^ ) ( c r _ 1 ( k , m - k)c g _ 1 (k, n - k)

that the number of consecutive ones is at most

The generating functions (3.8),

LINEARLY RECURSIVE SEQUENCES OF INTEGERS

As h a r m l e s s as it may appear, the Fibonacci sequence has provoked a remarkable

the basic definition

which Leonardo of Pisa found lurking in the simple rabbit problem.

For example, an exten-

sion of the definition yields the so-called Lucas numbers:

However, it is well known

P r e c i s e l y where does this peculiarity a r i s e ?

partial sum of the Fibonacci sequence is expressed by F

erally for proving such formulae is induction on the index.

- 1. The method used g e n -