The Fibonacci Quarterly

THE FIBONACCI QUARTERLY
THE OFFICIAL JOURNAL OF

THE FIBONACCI ASSOCIATION
VOLUME 8
NUMBER 2
SPECIAL
ISSUE
CON-TENTS
Fibonacci Representations II
L. Carlitz 113
A Limited Arithmetic
on Simple Continued Fractions n . . .
Convolution Triangles for
Generalized Fibonacci Numbers . . . . .
Some Properties of Stirling Numbers
of the Second Kind
. . . . . . . .
A Generalized Fibonacci Sequence
Over an Arbitrary Ring
A Remarkable Lattice
Generated by Fibonacci Numbers . *
An Extension of Fibonacci Numbers II .
Fibonacci Sequence Modulo a Prime
p = 3 (mod 4)
Summation of Powers of Roots
of Special Equations . . . . . . .
MARCH
,C. T. Long and J. H. Jordan 135

. .
.
Verner E. Hoggatt, Jr. 158

.
Rosedith Sitgreaves 172

D. J. DeCarli 182
S. K. Zaremba 185
Joseph Arkin and 199
Verner
Ho att
Jr
^ >
.Gottfried Bruckner 217
N. A Draim and Marjorie Bicknell
2 21
1970
THE FIBONACCI QUARTERLY

THE OFFICIAL JOURNAL OF THE FIBONACCI ASSOCIATION
DEVOTED TO THE STUDY
OF INTEGERS WITH SPECIAL PROPERTIES
EDITORIAL BOARD
Donald E. Knuth
George Ledin, J r .
D. A. Lind
C. T. Long
Leo Moser
I. D. Ruggles
M.N. S Swamy
D. E. Thoro
H. L. Alder
Marjorie Bicknell
John L. Brown, J r .
Brother A. Brousseau
L. Carlitz
H. W. Eves
H. W. Gould
A P. Hillman
V. E. Hoggatt, Jr
WITH THE COOPERATION OF
P. M. Anselone
T e r r y Brennan
Maxey Brooke
Paul F. Byrd
Calvin D. Crabill
John H. Halton
Richard A. Hayes
A. F. Horadam
Dov Jarden
Stephen Jerbic
R. P. Kelisky
Charles H. King
L. H. Lange
James Maxwell
Sister M. DeSales McNabb
C. D. Olds
D. W. Robinson
Azriel Rosenfeld
John E. Vinson
Lloyd Walker
Charles H. Wall
The California Mathematics Council
All subscription correspondence should be addressed toBro. A. Brousseau, St.

Mary f s College, Calif. All checks ($6.00 per year) should be made out to the
Fibonacci Association or the Fibonacci Quarterly. Manuscripts intended for
publication in the Quarterly should be sent to Verner E. Hoggatt, J r . , Mathematics Department, San Jose State College, San Jose, Calif. All manuscripts!
should be typed, double-spaced. Drawings should be made the same size as
they will appear in the Quarterly, and should be done in India ink on either
vellum or 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 f s College Post
Office, California, as an official publication of the Fibonacci Association..
FIBONACCI REPRESENTATIONS
II
L. CARLITZ*
Duke University, Durham, North Carolina
1.
Let R(N) denote the number of representations of
(1.1)
N = F, + F, + . . . + F. ,
kt
k2
kt
where
(1.2)
k t > k2 > > kj. j>
The integer t is allowed to vary. We call (1.1) a Fibonacci representation of

N provided (1.2) is satisfied.
(1.3)
If in (1.1), we have
k - k.+1 > 2
(j = 1,- , t - 1);
> 2 ,
then the representation (1.1) is unique and is called the canonical representation of N.
In a previous paper [1], the writer discussed the function R(N).
The
paper makes considerable use of the canonical representation and a function

e(N) defined by
(1.4)
e(N) = V l
Vl
'" + Vl
'
It is shown that e(N) is independent of the particular representation.
The
first main result of [1] is a reduction formula which theoretically enables one
to evaluate R(N) for arbitrary N. Unfortunately, the general case is very
complicated.
However, if all the kt in the canonical representation have the
same parity, the situation is much more favorable and much simpler results
are obtained.
* Supported in part by NSF grant GP-7855.
113
114
FIBONACCI REPRESENTATIONS - II
[March
In the present paper, we consider the function R(t,N) which is defined

as the number of representations (1.1) subject to (1.2) where now t is fixed.
Again we find a reduction formula which theoretically enables one to evaluate
R(t,N) but again leads to very complicated results. However, if all the kj in
the canonical representation have the same parity, the results simplify considerably.
In particular, if
N = F2k + ... +F2k
j
J
(kA > k2 > > k r 2 1) ,
= k
s
- k ^
(1 < s < r); j = k ,
J
s
s+1
r
r '
f r (t) = f(t; jt, , j r ) = R(t,N) ,
00
F r (x) = F(x; ji, , j r ) = ] ^ f ( t ; j
l f
, j r ) x ,
t=l
G r (x) = F(x; j ^ , j r _ l f j
+ 1) ,
then we have
(1.5)
j
x(l - x r + 1 )
r-l+2
X
r X
G r (x) ' G r + 1 (x) - x
G r _ 2 (x) = 0
1^
x
XU
(r > 2) ,
where
G0(x) = 1,
Gi(x) =
X(1
Ji+1
X
]
"_
1
X
In particular, if ji = *' = j > then

oo
r=0
G (x)z r = { l - [J + l ] x z + x j + 2 z 2 }
r
_i
from which an explicit formula for G (x) is easily obtained.
Also the case
1970]
Ji =
= Jr-1 = h
115
= k
leads to simple results.

In the final section of the paper some further problems are stated.
2. Put
o
F F <J>(a,x,y) = II (1 + ax n y n + 1 ) .
n=l
(2.1)
Then
F
î
\
o /i ,
n+1 F n + 2 ,
,- __, n F n F n + L
n
4>(a, x, xy) = II (1 + ay
x
) = 11 (1 + ay x
) ,
n=l
n=2
so that
(1 + axyXtKa, x, xy) = 4>(a, y, x) .
Now put
(2.2)
<J)(a, x, y) =
k m n
A(k, in, n) a x y .
k,mn=0
Comparison of coefficients gives
(2.3)
A(k, m, n) = A(k, n - m , m) + A(k - 1, n - m , m - 1) ,
where it is understood that A(k,m,n) = 0 when any of the arguments is

negative.
In the next place, it is evident from the definition of e(N) and R(k,N)
that
116
FIBONACCI REPRESENTATIONS - H
(2.4)
(l + a x F n y F n + 1 ) =
a=1
[March
R(k, N) a k x e ( N ) y N .
N=0
Comparing (2.4) with (2.1) and (2.2), we get

(2.5)
R(k, N) = A(k, e(N), N) .

In particular, for fixed k, n,
(2.6)
A(k, m, n) = 0
(m ^ e(n) ) .
It should be observed that A(k,e(n),n) may vanish for certain values of k

and n. However, since
A k
< > e<n)> n ) >
R(n) = R(k, n) =
k=0
k=0
it follows that, for fixed n, there is at least one value of k such that
A(k, e(n), n) f 0 .
If we take m = e(n) in (2.3), we get
(2.7)
R(t,N) = A(t, N - e(N), e(N) ) + A(t - 1, N - e(N), e(N) - 1) .
Now let N have the canonical representation

(2.8)
with k
N = Fkj
...
. . .
k r
odd. Then
e(N) = F
M +
N-e(N) = F
+..-
k r l ( I
+Fkr_2
1970]
Since k
11?
> 3, it follows that
(2.9)
N - e(N) = e(e(N)) 8
On the other hand, exactly as in [ 1 ] , we find that

e(e(N) - 1) = N - e(N) - 1 .
It follows that
A(t, N - e(N), e(N) - 1) = 0 ,
and (2.7) reduces to
R(t, N) = A(t, e(e(N)) .
We have, therefore,
(2.10)
R(t, N) = R(t, e(N) )

Now let k
(k r odd) .
in the canonical representation of N be even.
We shall
show that
S
k -1
(2.11)
R(t, N) = R(t - 1, e
k -2
(Nj)) +
R(t - j , e
J=2
where k
(2.12)
= 2s,
Ni = F k i + . . . + F k r _ i
and
(2.13)
e k (N) = e(e k "' 1 (N)) s
e(N) = N .
(N^ ) ,
118
[March
Assume first that s > 1. Then as above

(2.14)
N - e(N) =e(e(N) ) ,
and
(2.15)
e(e(N) - 1) = e(e(N) ) .
Thus (2.7) becomes

(2.16)
R(t, N) = R(t, e(N) ) + R(t - 1, e(N) - 1)

When k
(k > 2) .
= 2, we have, as in [1],
N - e(N) = F k _ 2 + . . . + F k r - 1 - 2 = ((Ni) ) ,
e(N) - 1 = F ^ ^ + . . . + F k r - 1 _ ! = e(Ni) ,
e(e(N) ) = N - e(N) - 1.
It follows that
(2.17)
R(t, N) = R(t - 1, e(Ni) )
(k r = 2) .
Returning to (2.16), since

e(N) - 1 = F
+ ... + F
+ (F 2 + F 4 + . . . +
F ^ )
K.~~ X
= e(N t ) + (F2 + F 4 + . . . + F 2 t _ 2 ) ,
it follows from (2.17) and (2.10) that
R(t, e(N) - 1) = R(t - 1, e 2 (Ni) + F 3 + . . . + F 2 t _ 3 )
= R(t - 1 , e 3 (Ni) + F 2 + . . . + F 2 t _ 4 ) .
Repeating this process, we get
1970]
FIBONACCI REPRESENTATIONS H
119
R(t, e(N) - 1) = R(t - s, e 2 s " 2 (N 1 ) ) ,

so that (2.16) becomes
(2.18)
R(t, N) = R(t, e2(N)) + R(t - s, e 2 8 " 2 ^ ) )

If k
(k
= 2s > 2) .
= 4, Eq. (2,18) reduces, by (2.17) and (2.10), to

R(t, N) = R(t - 1, e4(NA) ) + R(t - 2, e 2 ^ ) ) ,
since
(2.19)
R(t, N) = R(t, e(N t ) )
(k r = 2) .
For k4 = 2s > 4, Eq. (2,18) gives

R(t, N) = R(t, e4(N) + R(t - s + 1, e 2s "" 2 (Ni) ) + R(t - s, e 2s "~ 2 (Ni) )
= R(t, e6(N)) + R ( t - s + 2 , e 2 s " 2 ( N i ) ) + R(t - s + 1, e 2 s " 2 ( N i ) )
+ R ( t - s, e 2 s " 2 ( N i ) ) .
Continuing in this way, we ultimately get
s
2s 2
R(t, N) = R(t, e " (N) ) + J ] R(t - | , e 2 s " 2 (N!) ) .
(2.20)
3=2
By (2.17),
R(t, e 2 s ~ 2 (N) ) = R(t - 1, e 2 s " 1 (N 1 ) ) ,
so that (2.20) reduces to (2.11).
This proves (2.11) when k
is identical with (2.17).
We may now state
> 2; for k
= 2, it is evident that (2.11)
120
[March
Theorem 1. Let N have the canonical representation

=
...
where
k. - k j + 1 >
(j = 1, . . . , r - 1);
kr > 2
Then, for r > 1, t > 1,

S
k -1
(2.21)
R(t, N) = R(t - 1, e
k -2
(N^ ) +
R(t - j , e
(N*) ) ,
j=2
where s = [k / 2 ], Nj = F,
r
+ + F,
r-1
3. For N = F , r > 2, Eq. (2.7) reduces to

(3.1)
R(t, F r ) = Aft, F r _ 2 , F r - 1 ) + A(t - 1, F r _ 2 , F ^

= R(t, F r _ 1 ) + Aft - 1, F r - 2 , F ^
- 1)
- 1) .
Also,
(3.2)
Rft, F r - 1) = Aft, F r - 1 ~ e ( F r - 1), e ( F r - 1) )

+ Aft - 1), F r - 1 - e ( F r - 1), e ( F r - 1) - 1)
Since
e
<F2s+l "
1}
"
2s
<F2S "
1J
2s-1 -
we have
Aft - 1, F 2 s _ 2 , F ^
- 1) - Eft - 1, P a B _ 2 - 1),
Aft...- 1, F 2 g - 1) = 0 .
1970]
121
Thus (3.1) becomes
(32)
|R(t, F 2 s )
R(t
W, F^))
= Rft, F 2 s )
'
2s-1>
+ R(t
"
lj
2s-1 "
In the next place, Eq. (3.2) gives

R(t, F 2 s - 1) = Aft. F 2 g _ 2 , F 2 g _ 1 - 1) + Aft - 1, F 2 s _ 2 , F , ^ - 2)
= Eft. F ^
R
<t> F 2s+1 "
1} = A ( t
- 1)
' F 2 s - 1 " X'
2s>
+ A(t
"
2 s - 1 " X' F 2 s "
1}
= Rft - 1, F 2 s - 1),
that i s ,
(3.3)
Rft, F r - 1) = Rft - X, F r _ 1 - 1)
(r > 2) ,
where
X =
(r even)
^1
(r odd) .
It follows from (3.3) that

Rft, F 2 s - 1) = Rft - s + 1,0),
Rft, F 2 g + 1 - 1) = Rft - s + 1,1)
which gives
(3.4)
| Rft, F 2 g - 1 ) = 8
I
(Rft, F 2 g + 1 - 1) = 8 t j S
Combining (3.2) with (3.4), we get
Rft, F 2 g ) = Rft, F 2 s + 1 ) = Rft, F2a_)
+ tjS ,
122
[March
so that
R ( t 5 F 2 g ) = R(t, F 2 s _ 2 ) + 8 t > s
It follows that
11
0
R(t, F 2 g ) =
(1 ^ t s)
(t > s)
We may now state

Theorem 2. We have, for s _> 1, t > 1,
X)
= St,s
2s+2 "
1 (1 < t < s)
R(t, F 2 g ) - R(t, F 2 g + 1 ) - >Q ( t - g )
R(t
(3.5)
>
2s+1 "
X)
R(t
'
Let m(N) denote the minimum number of summands in a Fibonacci

representation of N and let M(N) denote the maximum number of summands.
It follows at once from (2.21) that
m(N) = r ,
(3.6)
where r is the number of summands in the canonical representation of N.

Moreover, it is easily proved by induction that
R(r, N) = 1.
(3.7)
As for M(N), it follows from (2.21) that

(3.8)
M(N) < M ( F k i _ k 2 + 2
-.
r-1
where
N
is the canonical representation.
kl
+
*
---
+ F
k
r
Now, by Theorem 2,
1970]
123
M(Fk) = [Ik] .
Hence by (3.8),
M(F
kt
+ F
k2) ^ ^ ( k * " k2)] + [|k2] + 1 .
Again, applying (3.8), we get

M(F
kj
+ F
k2
+ F
k8) ^ ^ ( k l "
k2>]
"
3^
ti^l
+ 2
It is clear that in general we have

(3.9)
M(N) < [l(ki - k 2 )] + . . . + [i(k r _ >1 - k r ) ] + [ | k r ] + r - 1 s
so that
(3.10)
M(N) < [-Jki] + r - 1 .

We note also that (2.21) implies
(3.11)
R(M(N), N) = 1 .
We may state
Theorem 3. Let
(3.12)
N = Fk + ... + Fk
1
be the canonical representation of N. Let m(N) denote the minimum number

of summands in any Fibonacci representation of N and let M(N) denote the
maximum number of summands.
Then m(N) = r and M(N) satisfies (3.9).
Moreover,
(3.13)
R(m(N), N) = R(M(N), N) = 1 .
It can be shown by examples that (3.9) need not be an equality when r > 1.
124
[March
4. While Theorem 1 theoretically enables one to compute R(t,N) for

arbitrary t,N, the results are usually very complicated. Simpler results can
be obtained when the k. in the canonical representation
(4.1)
N = Fki + .-. + F k r
have the same parity.
In the first place, if all the k. are odd, then, by (2.10),
R(t, F ^ + . . . + F k r ) = R(t, F ^ ^ + . . . + F k r _ ! ) .
There is therefore no loss in generality in assuming that all the k. are even
It will be convenient to use the following notation. Let N have the canonical representation
(4.2)
N = F2kj
...
F2kr
where
(4.3)
kt>
k2 > > k r 2 1
Then, by (2.21) and (2.10),

(4.4)
R(t,N) = R(t - 1, F
k
+
2 V 2 k r
. . . +Fak r _ 1 -2k p >
E * " h F 2 k i _ 2 k r + 2 + + F 2 k r _ 1 _ 2 k r f 2 > 3=2
Put
(4.5)
jJ
=k
-k
,
s-1
(s = 1, , r - 1);
Jj
and
(4.6)
f r (t) = f(t; J!, - - - . J r ) = R(t, N) .
=k
1970]
FIBONACCI REPRESENTATIONS -
125
Then (4.4) b e c o m e s
(4.7)
f(t; j
l f
. . . , j r ) = f(t - 1; Ji, . . . j r _ i )
j
+
r
f(t
u=2
u;
h> '"
> Jr-2 Jr-1
1)
By (2.18), we have
R(t
'
2k1-2kr+2
R(t
2kr_i-2kr+2>
'F2kr2kr
-"
+F
2kr_r2kr)
R(t-kr_1+kr-l;F2kr2kr_i+2
+ F
...
2kr_2-2kr_1+2) '
s o that
(4.8)
f(t; j j , , j r _ 2 , j r _ i + 1)
= f(t; j i , ,j r _ 2 > j r _ i > + f ( t - ' j r _ i - 1; Ji- j r _3j r _2
D-
If we put
00
(4.9)
t
f(t; J l t , j r ) x ,
F r (x) = F(x; j , j r ) =
t=l
it follows from (4.7) that (for r > 1) ,

(4.10)
F(x; j j , , j r ) = xF(x; j t , , j ^ )
j
Similarly, by (4.8),
X(X
" X
1 - x
F(x; Ji, . . , j r _ 2 , j r _ i + D-
126
(4.11)
FIBONACCI REPRESENTATIONS F(x; Ji, . . . , j
2,
[March
j T m m l + 1)
3
= F(x; j i , - , j r _ 2 3 r - l ) + x
rr
-l+1
Ffcji,fjr_3ljr_2 + l).
which yields
(4.12)
F(x; j l f , j r _ 2 J r - 1
1)
3
= F(x; j i , - - . , j r _ 2 J r - l )
K l+V-2+2
1
+X "
+ x
r-l+1
F(x; J ! , - - . , j r - 3 > 3 r _ 2 )
K i+---+32+r-l
F(x; J i , - - - , j r _ 3 ) + . . . + x x ~
F(x;jt).
F o r b r e v i t y , put
(4.13)
G r (x) = F(x; ]t, , j
r - 1
, j
+ 1) ,
so that (4.10) b e c o m e s
3r
(4.14)
Fr(x) - x F r _ 1 ( x )
= *V-
} G
r-l(x) '
while (4.11) b e c o m e s
(4.15)
G^x)
j -+1
= Fr_1(x) + x r " 1
G r _ 2 (x) .
Combining (4.14) with (4.15), we get
(4.16)
v
G (x) r\
(i
3r+1
iTl^
1 - x
' G
-W + x
r-1
r _ i
i+2
(x) = 0 .
r-2
T h u s G (x) s a t i s f i e s a r e c u r r e n c e of the second o r d e r .

r
Gi(x) = F(x; jt'.+ 1) =
R(t, F 2 j
+2)
xt
t=l
- 4- xJ - x(l - x j l +1 )
"
L,
fe=l
-=
l -
Note that
1970]
G2(x) = F(x;
Jlf j2 +
1) = R(t, F 2 J i + 2 J 2 + 2
127
F ^ )
t=2
Now, by (2.21),
J2+1
R(t
'
2J 1+ 2j 2+ 2
+ F
2 j 2 + 2 ) = * " * F 2 j l + 1>
E R ( t " "> F 2 J l + 2>

u=2
so that
Ji
J2+1
Ji+1
G2(x) = x Z x* + 2 xuu ^J ) x t
t=l
u=2
= x 2 (l - x j l )
1-x
t=l
x 2 (l - x J2 ) x(l - x 3 l + 1 )
1-x
1-x
Hence, if we take G0(x) = 1, Eq. (4.16) holds for all r 2 2.

We may state
Theorem 5. With the notation (4.2), (4.6), (4.9), (4.12), f (t) = R(t,N)
is determined by means of the recurrence (4.16) with
G0(x) = 1,
Gift) =
X{
_XX
'
and
F r (x) = G r (x) - x
Jr +1
G^x)
It is easy to show that G (x) is equal to the determinant
[March
128
-x"
x[Jl + l ]
-1
D (x)
r
(4.17)
X[j2 + 1]
...
Jl
...
x[js+l]
...
* [ J r + l]
where
[j] = (1 - x j ) / ( l - x)
(4.18)
Indeed,
Di(x) = x[Jt + 1] = GiW ,

D2(x) = x\jt
+ l ] [ j 2 + 1] - x j i + 2 = x 3 [ji.+ l ] [ j 2 ] + x*[}t] = G2(x) ,
and
J
r-1
D r (x) = x [ j r + l J D ^ W - x ^
(4.19)
D r _ 2 (x)
Since the recurrence (4.16) and (4.19) are the same, it follows that
G r (x) = D r (x) .
5. When
(5.1)
Ji
32
Jr = 3
we can obtain an explicit formula for G (x). The recurrence (4.16) reduces to
(5.2)
Then
G r (x) - x[j - l]G r _ 1 (x) + x j + 2 G r _ 2 (x) = 0
( r 2)
1970]
oo
129
oo
G r (x)z
r=0
= 1 + [j + l]xz + ^ G, r (x)z r
r=2
+2
n r _ fcrrt
- xJJ+ ^G
2 (x)}
= 1 + [J + l]xz + ( x [ j + l j G ^ W
r=2
00
= 1 + ([j + l]xz + x
j+2
z ) C
G r (x)
zr ,
rv
r=0
so that
oo
_i
G r (x) z
= (1 - [ j + 1] xz + x
j+2
z )
r=0
= E *s *s (D + 1 ] -
xJ z
>
S=0
= E x S z S E (-^(^[j + i r ^ ^ z *
s=0
t=0
* '
Hence
(5.3)
Gr(x) = X ( - D ' ^ j l J +
lf*^
2t<r
Finally, we compute F (x) by using
(5.4)
F r (x) = G r (x) - x j + 1 G ^ 1 ( x )
When j = 1, we have
130
G
0
(x)r = 1
(1 - xz)(l - x2z)
r=0
= -_J_/_i
1 - x \ 1 - xz
[March
L_\
1 - x2z/
which gives
(5.5)
G r (x) =
X 2[r]
^ J *
F r (x) = x r
(5.6)
(j = 1; r > 1)
0 = 1).
In this case, we evidently have

N = F
+ F
+-..+F
=F
- 1 .
x
*2r
*2r-2
*2
*2r+l
'
iN
so that (5.6) is in agreement with (3.4).

For certain applications, it is of interest to take
(5.7)
% =
Then G^x), G2(x), , G
(5.8)
Gg(x) =
= j
^ = j;
= k .
Ax) are determined by
( - W S " ^ [j - i] s " 2 t x s+jt
2t<s
while
(5.9)
G^x) = x[k - l]G r _ 1 (x) - x j + 2 G r _ 2 (x)
where
G J = G r (x; j , , j , k)
Also,
(i <
< r),
1970]
(5.10)
FJ.(x) = F r (x; j , . - .
J, k) - x[k]G r _ 1 (x) - ^+\_2(x)
We shall now make some applications of these results.

L
2j+lF2k
131
2k+2j
+ F
2k+2j-2
"''
.
Since
2k-2j
it follows from (5.10) that
(5.M)
(^L2j+iF2k)xt
= x2J+1 2
[ J ] t k " J] " x 2 J + 2 [ 2 j - 1]
0 < k) .
(Note that formula (6.17) of [l] should read

R(L
2 j + l F 2 k ) = 2J(k - ~
(2
J "
1}
in agreement with (5.11).) If we rewrite (5.11) as

R(t, L 2 j + 1 F 2 k ) x t = x 2 j + 1 { l + X + . . . +
t
, k-i-1 , /
+x
, 2j-lw
+(x + . ' + x J
, k-j-lo
)(x + - - - + x
we can easily evaluate R(t, L 2 . + 1 F 2 , ). In particular, we note that

(5.12)
R(t, L 2 j + 1 F 2 k ) > 0
(j < k)
if and only if
2j + 1 ^ t 3j + k - 1 .
Note that, for k = 3j,
R ( t , L 2 . + 1 F 6 . ) x t = x2i+1{l
+ x + . . . + x 2 ] V l + ( x + x 2 + - +x 2 j '"' 1 ) } .
)}
132
[March
This example shows that the function R(t,N) takes on arbitrarily large values.
When j = k, we have
L
2k+l F 2k
4k+1 ~ * '
so that, by (3.4),
LIt.L2fcflF2k)xt
(5.13)
=x2k
Next, since
L
2j+l F 2k "" F 2j+2k
+ F
2j+2k-2
+ F
'''
2j-2k-2
(j
>
k)
we get
(5.14)
R ( t , L 2 ^ F ^ J x * = x 2 k [ j - k - 1] [2k - 1] - x 2 k + 1 [ 2 k - 2]
t
(j > k > 1) .
Corresponding to (5.15), we now have

(5.15)
R(t, L 2 j H K L F 2 k ) > 0
(j > k > 1) ,
if and only if
2k < t < j + 3k - 2
The case k = 1 is not included in (5.14), because (5.5) does not hold
when r = 0. For the excluded case, since
L
we get, by Theorem 1,
2j+1
2j+2
+ F
2j
1970]
(5.16)
R(t, L 2 j + 1 ) x t = X2 + (x2 + x 3) ^ z A
133
(j > 1) .
t
For t = 1, Eq (5.16) reduces to the known result:
R(L2j+1) = 2 j - 1
In [ 1] a number of formulas of the type
R
< F 2n + l " X> = F 2 n + 1
<n > >
< F L > = F 2n
<n >
were obtained. They depend on the identities

F4
F8
F2+
Fi
+
+
.-.
...
4 n
= F|Q+1-
F 4 n + 2 = F2n
We now apply (5.10) to these identities.
(5.17)
G r (x) =
1,
.
Then G (x) is determined by
( - ^ ( ^ " ^ [ S r-2t
f ^ x 1r+2t
i-i; | t
j L^J
"
2t<r
Thus (5.10) yields
(5.18)
R(t, F | n + 1 - l)x t = x(l + x)G n _ 1 (x) - x 4 G^ 2 (x) ,

4n+l
(5.19)
R(t
>
2n)xt
xG
with GnJ...(x), Gn4

M 9 (x) given by (5.17)
n-l(x) "
x4
n-2(x) '
134
Mar. 1970
It may be of interest to note that
Gr(l) =
(-l)Yrt" M3r"2t = F2r+2
2t<r
6. The following problems may be of some interest.
A. Evaluate M(N) in terms of the canonical representation of N.
B. Determine whether R(t,N)_> 1 for all t in m(N) < t < M(N).
C. Does R(t9N) have the unimodal property? That i s , for given N,
does there exist an integer /x(N) such that
R(t, N) < R(t + 1, N)
(m(N) < t < /x(N) ),
R(t, N) > R(t + 1, N)
(M(N)
< t < MCN) ) ?
D. Is R(t, N) logarithmically concave? That i s , does it satisfy

R 2 (t,N) > R(t - l,N)R(t + l,N)
(m(N) < t < M(N) )?
E. Find the general solution of the equation

R(t, N) = 1 .
REFERENCES
1. L. Carlitz, "Fibonacci Representations," Fibonacci Quarterly, Vol. 6
(1968), pp. 193-220.
2. D. A. Klarner, "Partitions of N into Distinct Fibonacci Numbers," Fibonacci Quarterly, Vol. 6 (1968), pp. 235-244.
A LIMITED ARITHMETIC ON SIMPLE CONTINUED FRACTIONS - II

C. T. LONG and J. H. JORDAN*
Washington State University, Pullman, Washington
1. INTRODUCTION
In the first paper [2 ] in this s e r i e s , we developed certain properties of
the simple continued fraction expansions of integral multiples of quadratic
surds with expansions of the form
[a,b] or
[a,b,c] where the notation is
that of Hardy and Wright [1, Chapter 10]. For easy reference, we restate the
principle results here.
Theorem 1. Let = [ a , b ] , let n be a positive integer, let p, /q.
denote the k
convergent to and let t, = q, - + q, - for k > 0 where
we take q_x = 0. Then n = [ r / s ] if and only if n = <l2in_2>
and s = t 02m-20 for some m > 1.
Theorem 2. Let , n, p, /q,
and t,
2r * ^
2r-l
hv-l
' *
[ p 2r 9 fc2r> c t 2 r
/b]
X l
2m-2'
(c
'
+ 4c/b)q
and vw = t 2 m _ 1 - 2
th
k
convergent to ,
1. Then, for every
'
t 2r-l " ' > S r - i "
[ S 2r-1> q 2 r - l '
= p
be as in Theorem 1. Then n =
[ u , v , w ] if and only if vn = q 2 m _ l 9 vu = P 2 m _ 1 - If
for some integer m ^ 1.
* 8
Theorem3. Let = [ a , b , c ] , let p, /q, be the
let t^ = q k _ 1 + q k + 1 and s k = p k _^ + p k + 1 for k >
integer r > 1, we have
2]
2r-l J
and
t
2 r
' ( =
[ s 2 r - 1, 1, q 2 r - 2, 1, (be + 4)q 2 r - 2 ] .
Of course, for a = b = c = 1, the preceding theorems give results involving the golden ratio, (1 + V*5)/2, and the Fibonacci and Lucas numbers
since, in that case,
*The first author was supported by NSF Grant GP-7114.
135
136
A LIMITED ARITHMETIC
[March
{ = (1 + VB)/2. p k = F k + 2 , q k = F k + 1 , ^ = L k + 1 , and ^ =
where F
and L
denote respectively the n
Fibonacci and Lucas numbers.
In the present paper, we devote our attention primarily to the study of

the simple continued fraction expansions of positive rational multiples of quadratic surds with expansions of the form
[ a ] . Again, we note that, for a = 1,
the theorems specialize to results about the golden ratio and Fibonacci and
Lucas numbers.
2. PRELIMINARY CONSIDERATIONS
Let the integral sequences {f }
n
(1)
f0 = 0,
and {g }
be defined as follows:
n
ri>0
n>0
ft = 1, fn = a ^ ! * ^ '
and
(2)
go = 2, gt = a, g n = ag n _ 1 + g ^ ,
n > 0 ,
where a is any positive integer. These difference equations are easily solved
to give
(3)
fn = *
"
n=> 0 ,
7a + 4
and
(4)
gn = t
+ ?n , n > 0 ,
where
= (a + Va 2 + T ) / 2
and
J = (a - */a2 + 4 ) / 2
are the two irrational roots of the equation
(5)
x 2 - ax - 1 = 0 .
1970]
ON SIMPLE CONTINUED FRACTIONS - H
137
Incidentally, if fl is a quadratic surd, we will always denote the conjugate surd

by p. The following formulas, of interest in themselves, generalize results
for the Fibonacci and Lucas numbers and are easily proved by induction,
2i+1
f 2 n == >Z: ^(2!
-M
+ li)a
i=0 ^
(6)
n > 0 ,
fo
2n+l
i=0
(7)
%n = f n - l + W
(8)
(9)
^rnn-n+l = f m^n
(10)
m + n + l = fmfn
Wn + 1>
+
Wn+1'
^>
'
* '
f f -f
A ^ = (-l)m~h
JJt9
mn
m - 1 n+1
m-n+1'
* >
^ >
l < m < n .
Also, we obtain in the usual way from (8) the following lemma.
Lemma 4.
For the integral sequence (f } ^
we have that f If
if
and only if m| n, where m and n are positive integers and m > 2 if a = 1.

3, PRINCIPAL RESULTS
Our first theorem, together with the results of the first paper in this
couplet, yields a series of results concerning the simple continued fraction
expansion of multiples of =: [a] by the reciprocals of positive integers* The
theorem is also of some interest in its own right.
Theorem 5.
Let = (a + b*/c)/d with a, b , c,
not a perfect square, and c and d positive.
and d integers,
Let r be a positive rational
number such that 2ar/d is an integer. Let a2 + d2 = b2c and let 1 < < r .
Then
r = [a 0 , a l8 a 2 , ] ,
138
[March
if and only if
2
L
ar
1
a
r
| ' " ~ c P ai> a2' ' " 1 *
Proof. We note first
st that a^ - 2ar/d is ppositive. This is so since
a0 -
[ 3
(
ar + r l W c | ^ r a + rbVc"
_^
"!J ^
cg.
so that
2ar ^ - r a + rbN/c"
d
- b2c
(*-
51
>
b's/c"
Urd
1
a + h\l~c
=
r
{
- 1 =>
o ,
by hypothesis. Now let /x = [a^ a 2 , a 3 , ] so that

r = a0 + Then
2ar
0, a0 - ,
"1
a l s a 2 ,
1
=
2ar
2ar
r i
+ bVc
2a\
~ dJ
r ( - a + b's/c)
-d(a + b ^ F )
r(a 2 - b 2 c)
a + b^c"
~ . dr
=
_
r
1970]
139
and the proof is complete.

Corollary 6. Let a and n be positive integers. Let = [a] and let
n >f.
Then
n = [a 0 , a l9 a 2 , ]
if and only if
|
= [0, a 0 , - an, a l9 a 2 , " ' ]
Proof. Since
*
a + \/a2 + 4
* >
f = [a] =
we may use the preceding theorem with a = a, b = 1, c = a2 + 4, d = 2,

and r = n. The result then follows immediately since
2ar _ 2an
d ~ 2
is an integer and
+ d2 = a2 + 4 = b 2 c ,
as required.
Now for
f = [a] =
The convergents p, /q.
p 0 = a,
Pl
are given by the equations

= a2 + 1, p n = a p n _ 1 + p n _ 2 ,
(11)
n ^ 2 ,
q0 = 1, qt = a, q n = a q ^ + qn__2 ,
and it is clear that p

f
and q
= f
= f
and q
f
- for n ^ 0, where p /q
= f
- for n ^ 0.
is the n
Also,
p' = f
convergent to l/f.
The
140
[March
following results could all be stated in terms of the sequences ( p } and {q };

instead, we use the sequences ( f } and ( g }.
Corollary 7.
Let r , s , and n be the positive integers with n > =
[ a ] . Then f / n = [0, r , s ] , if and only if, n = f 2 m - l ' r = f 2m' a n d S =

qn
- for some m ^ 2.
^2m-l
Proof. This i s an immediate consequence of Theorem 1 with a = b ,
and Corollary 6.
Corollary 8.
[a].
Let u , v., w, and n be positive integers with n > =
Then f/n = [0, u, v, w ] , if and only if, vn = f2m>
and vw = g 2
vu = ^ m + l " 1 '
- 2 for some integer m ^ 2.
Proof. This is an immediate consequence of Corollary 6 and Theorem 2

with u = v = w = a.
The next corollary results from Theorem 3 and Corollary 6 by taking
a = b = c. However, since, in this special case, parts (a) and (b) of Theorem 2 yield results already obtained, we concern ourselves only with parts
(c) and (d).
Corollary 9.
Let n be a positive integer greater than . Then for
r >1,
4=
[> < W f2r' <a2+'4>f2r]
and
= [0, g 2 r - 1. 1, f2r+1 - 2, 1, (a*
4)f 2 r + 1 - 2 ] .
The next theorem shows that the periodic part of the simple continued
fraction expansion of n for siny positive integer n > f = [a] is almost symmetric. Of course, by Corollary 6, the same thing is true of f / n .
Theorem 10.
Let a and n be positive integers with n > = [a] .
Then n = [a 0 , a t , , a ] and the vector (a l9 a2, , a _-) is symmetric

if r ^ 2.
1970]
ON SIMPLE CONTINUED FRACTIONS - II
141
Proof. Since a0 = [ n ] , we have that

0 < nf - fy < 1
and
f! = 1
1
- i ,
n( - a0
where f j is the first complete quotient in the expansion of n .

fl
n - a0
^ + a0
-1 <|J" <0
Moreover,
so that
since ao + n/f
is clearly greater than one, Thus, fj is a reduced quadratic
surd and by the general theory (see, for example [3, Chapter 4]) has a purely
periodic simple continued fraction expansion, say
fl = [ait a2, , a r ] .
Additionally, we also have that [a , a
- , ' , a-]
negative reciprocal of the conjugate of lo

r.
Lar ,*
oi
is the expansion of the
Thus,
I
v
ar_]_
., j a ' , a1 JJ = - j = 7- + a*
so that
(12)
J = [0, a r - a Q , a r - l f a r _ 2 , - , a ^ a r ]
142
[March
But, from above,

n = [a 0 , &i, , a r ]
and by Corollary 6,
(13)
- = [0, a0 - an, a lf a2, , a r ] .
Thus, comparing (12) and (13), we have that the vector (a ls a2, , a _-) is
symmetric.
We now turn our attention to the consideration of more general positive
rational multiples of = [a] .
Theorem 11. Let r be rational with 0 < r < 1. If the simple continued fraction expansion of r is not purely periodic, then
r = [0, a1? a 2 , ,
aj
and
I _
r ~
r
La
.
V l '
n " V
n-2' "
2'
. ,
nJ
for some n ^ 2.
Proof. If r had a purely periodic simple continued fraction expansion,
then r would have to be a reduced quadratic surd so that r > 1 and -1 ^
rT
<
(14)
0. But the first of these inequalities implies that
< r < 1 ,
and, since = - l / f , the second implies that

(15)
>r > 0
1970]
143
which is already implied by (14). Therefore, since r is not purely periodic,

we have
(16)
0 < r < i
so that 0 = [ r ] = a0. Now consider
f1 = ^
> 1 .
and set aj = [i] ^ 1. Again,

1
f i - aj = h > 1
1 _
and
(22 =
"T
T~~
- H
7 + at
since ?" = - 1 . Therefore, - 1 < J2 < 0 and 2
is
reduced.
Thus, 2
a purely periodic simple continued fraction expansion,

6
= [a2i a3, , a n ] ,
and
r = [0, a l5 a 2 , , a j
as claimed.
Also,
+ ai = - - =
r
*
T*2
a , a - , , a 0
n ' n-1
' 2J
,
'
nas
144
[March
so that
"~" ~~ I a a., a .., a ~

L
r
n
1 n-1
r-2
a^, a i
2
nJ
and the proof is complete.

Theorem 12. Let r be rational with 0 < r < 1. If the simple continued fraction expansion of rf is purely periodic, then
= [ v ai " *J
and
- = fa , a - , ,
L
r
n
n-1
aA1
0J
for some n > 0.

Proof. Since the simple continued fraction expansion of rf is purely
periodic, it is reduced and we have by the preceding proof that
< r < 1
Since we also have

f i r .2. = =
a , a
-i
. , , a~ J ,
n-1
the proof is complete.

In passing, we note that the periodic part of the expansions of rf need
not exhibit any symmetry or even near symmetry.
a = [1] =
we have that
For example, for
1970]
145
| o r = [0, 2, l f 3, 1, 1, 3, 9]
and
| a = [ 1 , 4, 1, 2, 6, 2] .
Also, it is easy to find rational numbers r with 0 < r < 1 such that the
surds va and a/v
are not equivalent where we recall that two real numbers
JJL and v are said to be equivalent if and only if there exist integers a, b , c,
and d with | ad - be j = 1 and such that
u, = S L 4 .
^
cv + d
However, as the following theorems show, there exist interesting examples,

where near symmetry of the periodic part of the expansions of r f and / r
and equivalence of r and r / f
both hold. We will indicate that r f and
f/r
are equivalent by the notation
Theorem 13.
Let a be a positive integer, let f = [ a ] ,
and let the
sequences (f }
and {g }
be as defined above. Then, for n > 1,
n
nn
nÔ
nn>0
-0
f
2n+l
2n+2
= [a Q , a x , ,
ar]
and
2n+2
7
f
2n-KL
r.
.o1
a ,
r
, ? " ,
anJ
where the vector (a2, a 3 , *, a r , ao) is symmetric, a 0 = a2 = 1, and aj =

f
- 1
lo
*4n+3
146
[March
Proof. We first demonstrate the purely periodic nature of the expansions in question. From the definition, it i s clear that f
ing for n ^ 2. Also, f /f - i s the n
(17)
i s strictly increas-
convergent to l/f. Therefore,
^ L . < i <^1< 1 .
i
2n+l
2n+2
and it follows from the proof of Theorem 11 that f f2

periodic expansion.
- /fL
Also, from Theorem 12, f2
/f2
has a purely
- has a purely
periodic expansion whose period is the reverse of that for f 2 n+l ^ 2 +2*
Additionally, from (17), it follows that
f
<
2n+l _ i
2n+2
<
1 / f 2n+l __ f 2n \
2 f
\ 2n+2
_
2f
2n+l/
1_
f
2n+l
2n+l ! 2n+2
so that
- . f 2n+l
> _
g
, ,
1 < ^ _ . <
_
+ 1
x
*2n+2
2n+l 2n+2
<
2f
tan
y
+ 1 < 2
2n+l 2n+2
Thus, a 0 = [ff 2 n + 1 / f 2 n + 2 ] = 1. Now
2n+l .
2n+2
. !
and we claim that
W
* - K
4n+3
^
so that a4 = f4
is the n
^1
! < f
4n+3
- 1. To see that this i s so, we note mat, since f
convergent to ,
/f -
1970]
ON SIMPLE CONTINUED FRACTIONS

f
2n+2 ^ f 4n+4
<
f
f
"
2n+l
4n+3
<
.
^
and
1
<
2n+l f 4n+4
^
^ f2n+l
<
2n+2 4n+3
2n+2
But this gives, using (10),

2n-fl
2n+2
* f
2n+l 4n+4
2n+2 4n+3
f
f
2n+2 4n+3
>
2n+2
2n+2 4n+3
4n+3
or
(18)
as desired.
< f
4n+3
A l s o , we have that
f
2n+2
f
2n+l
f
4n+3
* "^ f
4n+2
so t h a t , again by (10),
o <^SJi.f
-i
2n+2
*2n+l f 4n+3 ~ f 2n+2 f 4n+2
2n+2 4n+2
f
f
2n+l
2n+2
4n+2
2n+l
2n+l 4n+3
2n+l
147
148
[March
and
Thus, aj = [f i] = f.
- 1 as claimed.
Finally, to show the symmetry of
the vector (a2, &$, , a , a Q ), it suffices to show that
(19)
-=..
2n+l
a2
at
f
2n+l
a
f7~7
2n+2 " * " 0
Making use of the determined values of a0 and aj and setting a2 = 1,
this
means that we must show that
(20)
= **** f
2n+l
" <f4n+3 -
f
2n+l
f
x
2n+2
( ' 1
which will also, of course, confirm the fact that a2 = 1. Now (20) is true if
and only if
1970]

1
- _
149
2n+l
2n+2
(f
4n + 3 "
2n+2
which i s t r u e if and only if

1
f
2n+l
(f
^2n+2
+ ff
x
2n+l
* 2n+2
1) '
4n+3 "
- 1
2n+2
which, in t u r n , i s t r u e if and only if

f
H
s
2n+2
2n+2
^ f 2 n + l - f2n+2 " f 2 n + l + ^ 2 n + 2 "
to+3
"
T o s e e that t h i s l a s t equation i s t r u e we m a k e u s e of (8), (10), and the fact that

2
2n+2
i 0 obtain
^ f 2n+2
f 2n+l " f 2 n + 2 " f 2n-KL
2n+2 f 2n+l
^ 2 Q + 2 " |f|n+1
2
ff
2n+l
* f 2n+2 " ^ f 2n+l f 2n+2
V < - l f 2n + 2 - W
^2n+2 "
+
aff
"
^ f 2n+2
^
^ f 2n+l f 2n+2
2n+lf2n+2 "
ff
2n+2
(f
+ f )
2n+2 u 2n+2
2n;
2
f
f
- f
1
2 n + l 2n+3
2n+2
f2
2n+2
4-f
f + f
2n+2 2n
f
-1
I
4n+3
2n+l 2n+3
-f2
- 1
2n+2
150
[March
T h i s c o m p l e t e s the proof.
B e c a u s e of the s i m i l a r i t y of m e t h o d , the following t h e o r e m s a r e stated
without proof.
The notation i s a s before.
T h e o r e m 14.
F o r n > 2,
2n
*.
r
7 = La0>
2n+l
i>
2 "
a i
rJ
and
f
2n+l
^
r
-i
-f- = La 3 - 1, a 4 , a 5 , , a r ,
2n
w h e r e the v e c t o r
f
(a 3 , a 4 , , a r )
is symmetric,
4 n + l " l f a n d a 3 = f 3 + !
T h e o r e m 15. L e t n >: 2 be an i n t e g e r .
n+2
i
a 2 , a3J
a 0 = 0,
Then
r.
f = [a 0 , a t , , a r J ,
11
>
?"""
n+2
* '
- ~
~" L a r '
. . . "
r-l'
' a o Js
= [b0, bj, , b s ] ,
and
T^-t
T h e o r e m 16.
[ V V l ' ' fc o]
L e t n be a positive i n t e g e r .
Then
2n+l
.
. ,
t
T
f = [ a 0 , aA, a 2 , , a r J ,
8
2n+2
,
a 4 = 1,
a2
1970]
151
and
g
2n+2
= [ 29 a4? a 5 , . . , a .
r
2n+l
a 2 , a3 ]
and the vector (a3, a 4 , , a r ) is symmetric with a0 = 0,

f
4n+3 " l s a n d a 3 = 3e
Theorem 17, Let n be a positive integer.
g
2n
=2n+l
= [a 0 , *U ,
a,t = 1,
Then
ar]
and
g
2n+i
g
= [ a r , a r _ l S , a if a 0 ] ,
2n
where the vector (a29 a3, , a r s a 0 ) is symmetric with a^ - a2 = l f
and
l = f 4n+l " X
In view of the preceding results 9 one would expect an interesting theorem
concerning the simple continued fraction expansion of
i.
s
and
.
n
but we were not able to make a general assertion value for all a. To illustrate
the difficulty, note that s when a = 2 and = 1 + *J"29 we have
f4
= [ 0 , 1 , 5, 1, 3, 5, 1, 7] ,
&4
f
g5
and
f = [0, 1, 5, 1, 5, 3, 1, 4, 1, 7] ,
152
[March
f
1- = [ 0 , 1, 5, 1, 4, 1, 3, 5, 1, 4, 1, 7] .
&6
However, for
i; = = [1]
= ^-
we obtain the following rather elegant result:

Theorem 18. Let a = (1 + *v/5)/2 and
an let F
Fibonacci and Lucas numbers, respectively.
(21)
and L
denote the n
Then, for n > 4,
F
_ . a = [ 0 , 1, 2, 1, , 1, 3, 1, , 1, 4 ]
L
n
and
(22)
L
-I- a = [ 3, 1, , 1, 3, 1, - , 1, 2, 4 ] ,
n
where, in (21), there are n - 4 ones in the first group and n - 3 ones in the
second group and just the reverse in (22).
Proof. Set
x n = [ 2 , 1, - . , 1, 3, 1, , 1, 4]
= [ 2 , 1, , 1, 3, 1, , 1, 4, x n ] .
Then it is easy to see by direct computation as on computes convergents, that

ax
\ JT7T
n n
where
+ b
n
1970]

a
n "
=
4(L
4F
n =
n "
4(L
n-lFn-l
+ F
nFn-l
n-lFn-l
+ F
n-lFn-3
4F
n-l
+ F
+ F
n-2> 2
(Ln-2Fn-l
n-3 F n-2>
<-*'
n-2 =
+ F
153
n-2 F n-4>
< L n-2 F n-3
VsW
nFn-3
and
d = L
F
+ F
F
=F2
n
V l n - 3
n-2 n-4
n-1
Moreover, from (23),
(an - dn 7) + Jy( a n - dn )2 + 4bn c n

2c n
and
yn = [o, i. x j
X
x n+ 1
n
(a - d ) + J ( a - d )2 + 4b c
n
n
y n
n
n n
(an - d n +2c n )+
(an - d n') g +4b n c n
(a - d - 2 b ) + J ( a - d )2 + 4b c
=
2(a
Now
Jn
- b + c - d )
n
n
n
154
a
- d
[March
- 2b = 4 F 2 + F F , + ( - l ) n - F 2
- 2F2
n
n
n n-1
n-1
n
= 2 F 2 + F F
, - F F
n
n n-1
n n - 20
(25)
= F (2F + F n - F Q )
n
n
n-1
n-2'
F
n^Fn+2 "
n-2*
= F L ,
n n
a
- b
n
+c
n
- d
n
= (a
n
- d
n
- 2b ) + b + c
n'
n
n
= F L + F2 + 4F2
+ F F 0
n n
n
n-1
n n-3
(26)
F L + 2F -L
n n
n-1 n
= L2
n
and
<an " V
+ 4
V n
= <4Fn
+
(27)
4F^(4F^_1
= KFl+3
=
nFn-l
n<Fn+3
+ 4F
"
n-l
+ F F
n a-3>
4F
nFn-3>
= 5F2L2
n n
T h u s , using (25), (26), and (27), in (24), we obtain
Yn
=
"
Ll>'
FnFn_3)
^4FLl
16F
F L + F L \/5
F
- J /=
n n
n n
__ _ji
1 + V5
=
2
2L2
'
L
n
n
1970]
ON SIMPLE CONTINUED FRACTIONS H
as claimed.
155
The other part of the proof is an immediate consequence of
Theorem 11.
Finally, we comment on the question of the equivalence of r f and
If r = g ^ /fn
or r = g m /g^9
f/r.
where m and n are nonnegative integers,
it frequently turns out to be the case that rf ~ f/r.
However, this is not
necessarily the case and hence, & fortiori, it is.not necessarily the case for
more general r.
For example, for a = (l+\[5)f2
= [1] ,
J . a = [0, 1, 2, 3, 1, 4]
and
1 a = [ 3 , 1, 3, 2, 4 ]
where 3 = f4 = g2 and 7 = g 4 ; and other examples are easily found.

ever, if r = f
and s = f
How-
for nonnegative integers m and n then we
always have
as the following theorem shows.

Theorem 18. If m and n are nonnegative integers, then
f
m
n
f
*.
n
m
156
[March
Proof. Without loss of generality, we may assume that 0 < m < n and
that (m,n) = 1. We let
f f
=
3m+2
J'
n
b = c = f
2qm+l'
D
c
^qm+l'
^
f f
d = -HJ
a
f
where q is chosen so that

2q + 2 = 0 (mod n) ,
as may easily be done since (m,n) = 1. With this choice for q it follows
from Lemma 4 that fnlf2Gim+2 ^
integers. Also, by (10),
ml f 2am
so tnat
k c
and
are a
^-
ad - be = J*L*E12 . V p q m _ f2
f
f
2qm+l
n
m
^
= f
f
- f2
2qm+2 2qm
2qm+l
= -1 .
Finally, we show that
f
^
(28)
f
//
F"
\
+ b
+c
\m
for this choice of a, b , c,
and d.
Making the indicated substitutions, we
have that (28) holds if and only if

f
f
m " *t
T~
n '*
m f 2qm+2 / f n
J\
+ f
2qm+l
f
IF" ' * /
n 7 \Tm
/
\
TT0
// f n
f
t. \ I' j . nf: 2 q m
1970]
157
and this is true if and only if
^ f 2qm+l
^ f 2qm
^ f 2qm+2
2qm+l
But this is clearlyJ true since af2 = a + 1 and af0

, - + fn
= f0
,
*
2qm+l
2qm
2qm+20
and the proof is complete,
Finally 8 we note that the list of stated theorems is not exhaustive.
One
could no doubt prove theorems concerning
f
n
? "s
n+2
f
n
f
n
f
n+4
n+5
and so on. However ? we were not able to arrive at general formulations of the
expansions of
f
t
r'f'
n
T" ^
&
^
or
T" *
&
for arbitrary positive integers m and n. The results stated seem to be the
most interesting.
REFERENCES
1. G. H. Hardy and E. M. Wright, An Introduction to the Theory of Numbers,
Oxford University P r e s s , London, 1954.
2. C. T. Long and J, H, Jordan 9 "A Limited Arithmetic on Simple Continued
F r a c t i o n s , " Fibonacci Quarterly, 5 (1967), pp. 113-128.
3. C. D. Olds, Continued Fractions, Random House, New York, 1963.
CONVOLUTION TRIANGLES FOR GENERALIZED FIBONACCI NUMBERS

VERNER E. HOGGATT, JR.
San Jose State College, San Jose, California
DEDICATED TO THE MEMORY OF R.J. WEINSHENK

1. INTRODUCTION
The sequence of integers Fj = 1, F 2 = 1, and F
= F
+F
are
called the Fibonacci numbers. The numbers Fj and F 2 are called the starting pair and F
= F
n+i
+ F
n
2
i s ca
êc^
tne
recurrence relation.
The long
division problem 1/(1 - x - x ) yields
= F t + F2x + F3x2 + . . . + F n + 1 x n +
1 - X - X'
This expression is called a generating function for the Fibonacci numbers.

The generating function yielding
(1 - x - x 2 ) K + 1
= F1(k)+F,(k)x
is the generating function for the k
...
( k
n+1
+ -"
convolution of the Fibonacci numbers.
For k = 0, we get just the Fibonacci numbers.
We now show two different
ways to get the convolved Fibonacci numbers.

2. CONVOLUTION OF SEQUENCES
If a j , a 2 , a 3 , 9 a , and b l 9 b 2 , b 3 , , b n , are two s e quences, then the convolution of the two sequences is another sequence c j , c 2 ,
c 3 , , c n , whose terms are calculated as shown:
cj
ajbj
c2
= ajb 2 + a 2 bi
c 3 = ajb3 + a2b2 + a 3 bj
c n = a^b
, , . , + + an b-1
I n + a 20 b n-1- + a n3b n-2. + . . . + a.k bn-k+1
158
CONVOLUTION TRIANGLES
FOR GENERALIZED FIBONACCI NUMBERS
M a r . 1970
159
T h i s l a s t e x p r e s s i o n m a y also be w r i t t e n
n
c
= E akbn-k+l
k=l
ii
L e t u s convolve the Fibonacci n u m b e r sequence with itself. T h e s e n u m b e r s we

call the F i r s t Fibonacci Convolution Sequence:
Fi ( 1 )
F2
(1)
F!1)
FJFJ
= 1
= FiF2 + F2Fi
= 1 + 1
= F4F3 + F2F2 + F3F4
=2
+ 1 + 2
F J P = F j F 4 + F 2 F 3 + F 3 F 2 + F 4 F t = 3 + 2 + 2 + 3 = 10
*P
= k E F k F 5 _ k+1
= 20
F{1} = E FkF6_k+1
k=l
= 38
(i)
SiFkVk+i
= 71
'
Now l e t u s "convolve" the f i r s t Fibonacci convolution sequence with the

Fibonacci sequence to get the Second Fibonacci Convolution Sequence:
F< 2 ) = F i r i ( 1 )
F2(2) = F 2 F l ( 1 )
FiFU
F3(2) = F 3 F l ( 1 )
F2F2(1>
F ^
'
F 4 ( 2 ) = F 4 Fi ( 1 ) + F a F 2 ( 1 ) + F 2 F 3 ( 1 ) + F 1 F 4 ( 1 ) = 3 . 1 + 2 - 2 + 1-5 + 1- 10 = 22
F|2) = s ^Kk+i
=4l
Fi2) = E F f F 6 _ k + 1
=111
k=l
k=l
160
[March
T h e Fibonacci sequence i s obtained from
1
1 -
X -
= F j + F2x + F3x2 + . . .
+ FL n+1
-x
T h e f i r s t Fibonacci convolution sequence i s obtained from
(1 - x - x 2 )
,(D.' x + Fo(1).
= F,(D
i
+ Fo
V
F ( 1 > x
n+1
T h e second Fibonacci convolution sequence i s obtained from

,(2)
,(2).x + F i(2).
+ Fo
'xi
x2)
(1
F(2>xn
n+1
..
T h e s e could all have been obtained by long division and continued to find a s
(k)
a s d e s i r e d o r one could have found the convoluted sequence by the
many F
method of this section. In the next section we shall s e e yet a n o t h e r way to find
the convolved Fibonacci s e q u e n c e s .
3.
THE FIBONACCI CONVOLUTION TRIANGLE
Suppose one w r i t e s down a column of z e r o s .

down place a one.
To the right and one s p a c e
To get the e l e m e n t s below the one we add the e l e m e n t s one
above and the one d i r e c t l y left.

1
Pascal s arithmetic triangle.
T h i s i s , of course., the r u l e of formation

Such a r u l e g e n e r a t e s a convolution t r i a n g l e .
Next suppose i n s t e a d we add the one above and then diagonally left.
the row s u m s a r e the Fibonacci n u m b e r s .
0
0
0
0
0
0
0
0
0
0
0
Column:
for
1
1
1
1
1 "N*
1
3 \
1 s\ 5
1
1
0
^ l
1 3
We i l l u s t r a t e :
\ 6
6 \ s 1<N
7 1 15
1
2
1
4
1 310
1
4
Now
1970]
161
However, if we add the two elements above and diagonally left, we generate
the Fibonacci convolution triangle as follows.
Please note these are the same
numbers we got in Section 20 The zero-th column are the Fibonacci numbers,
F ; the first column are the first convolution Fibonacci numbers, F (1) , etc.
0
0
0
0
0
13
21
34
55
~~\1
2V
U-^
ten
20
1
3
38
71
21
Column:
4. COLUMN GENERATORS OF CONVOLUTION TRIANGLES

It is easily established that the column generating functions for Pascal 1 s
triangle are
gk(x) =
(1 - x)k+1
n=0
when the triangle is generated normally as the expansion of (1 + x) , n = 0 , 1 ,

2, and as we said to do in the first part of Section 3. The column generators become
2k
g k (x) =
(1 - x)ETi
k = 0, 1, 2,
162
[March
if we follow the second set of instructions. These column generators are such
that the elements across the rows each are multiplied by the same power of x.
We make the column move up or down by changing the power of x inVê.
erator of the column generating function.
k=0
k=0
"
X)
YXQWLY-
If we now sum
/ k=0 X
1 - X
1
>
2
X^
2
1 - X - XT
" 1 - x
Thus the row sums across the specially positioned (Position 2) Pascal triangle
are Fibonacci numbers. These a r e , of course, the numbers in the zero-th
column of the Fibonacci convolution triangle. If we multiply the column generators of Pascal* s triangle by a special set of coefficients, we may obtain
other columns of the Fibonacci convolution triangle.
Recall that the k
gk
column generator of PascaPs triangle is
(x) =
(kV
n=0 \ /
(1 -\k-ri
x)*
Replace x by
1 - x
in the above to obtain
t *> )
(l-A)'
1970]
163
while
2k
\ 1 - x/ g k ^ l - x)
Z^ \ k / g k
Thus the row s u m s a r e the k
th
x2
\k+l
( 1 - x - x 22>k+l
)
convolution of the Fibonacci n u m b e r s s i n c e
that i s the column g e n e r a t o r we have obtained.

( F i r s t column of P a s c a l )
Multipliers: 0
1 2
3
4
We i l l u s t r a t e :
0)
o
o
Row S u m s :
&
o
a
1 o
:tf
GO
3-1 + 1-2
5 o
4-1 + 3-2
= 10
5-1 + 6-2 + 1-3
= 20
10
6*1 + 10-2 + 4 - 3
= 38
15
10
21
20
F4
iH
rH
7-1 + 15-2 + 10-3 + 1-4 = 71 m

u
8-1 + 21-2 + 20-3 + 5-4 =130 r-4
(Second column of
Multipliers: 0
10
Row S u m s :
0)
1
1
10
22
15
10
51
21
20
111
28
35
15
233
o
o
o
a
o
a$
a
o
ci
o
o
0)
CO
164
Thus if we use the numbers in the k
[March
column of Pascal* s arithmetic triangle
(Position 1) as a set of multipliers with the columns of Pascal's triangle

(Position 2), we get row sums which form the k
Fibonacci sequence.
5. EXTENSION TO GENERALIZED FIBONACCI NUMBERS CONVOLUTION

TRIANGLES
The Fibonacci numbers are the sums of the rising diagonals of Pascal's
triangle which is generated by expanding (1 + x) . The generalized Fibonacci
numbers a r e defined as the sums of the diagonals of generalized Pascal's t r i angles which are generated by expanding
(1 + x + x 2 + . . . + x 1 "" 1 ) 1 1 .
i-2
The sequences can be shown to satisfy u4 = 1, u. = 2J
for j = 2 , 3 , 8 , r ,
and
u,
= 7 u , .,
n+r
/ JI n+r-j
n > 1 ,
and the generating functions are
XA+IXI1
r-1
n=0
The simplest instance is the Tribonacci sequence, where Tj = 1, T 2 = 1,

T 3 = 2, and T + 3 = T + 2 + T - + T , and these sums are the rising diagonal sums of the expansions of (1 + x + x 2 )
for n = 0, 1, 2, 3, . The
first few terms are 1, 1, 2, 4, 7, 13, 24, 44, 81.
If we return to our Fibonacci convolution triangle at the end of Section 3,
we note the row sums are the Tribonacci numbers. The column generators of
the Fibonacci convolution are
3k
g
(x)
"
2^+1
(1 - X - X 4 )
'
1970]
165
where the numbers on each row in the triangle all multiply the same powers of
x in the column generators. Adding, we get
fv . (_*_)/_*_ .
^ ^
\l -X- W^V.l - X - W
1-X-X2-X3
which is the Tribonacci sequence generating function.

multipliers [ , J as before, we get
If we use the special
b%P>UJ (=)and this is the k
3k
k+
(1 - X - X22 - vX33 )> l
Tribonacci convolution sequence generator and the coef-
ficients appear in the k
column of the Tribonacci convolution triangle. Thus
we can obtain all the columns of the Tribonacci convolution triangle from the
Fibonacci convolution triangle in the same way we obtained the Fibonacci convolution triangle from Pascal 1 s arithmetic triangle.
We can thus generate a sequence of convolution triangles whose zero-th
columns are the rising diagonal sums taken from generalized Pascal triangles
induced from expansions of (1 + x + x2 + - + x
for the r
) . The column generators
case
rk
>k
( T - x - x 2 - ...
~Xr-l)k+1
St
can easily be seen to generate the column generators for the (r+1)
x (r+l)k
gk(x) =
^
(1 - X - X^ - - X )
using the preceding methods.
case
166
[March
Referring back to the Fibonacci convolution triangle of section three,

each number in the triangle is the sum of the one number above and the number diagonally left.
Because the column generators must obey that law and
multiplying by powers of x so that the proper coefficients will be added, we

could write a recurrence relation for the column generators of the Pascal
convolution triangle as follows:
Gk(x) = xGk(x) + x 2 G k _ 1 (x)
or
Gk =^ % ^ )
By similar reasoning, each number of the Fibonacci convolution triangle is

the sum of the two terms above it and one diagonally left. Proceeding to column
generators, then,
Gk(x) = xGk(x) + x2Gk(x) + x 3 G k _ 1 (x)
or
k(x) = ,
k-i(x)
1 - X - X^
6. THE REVERSE PROCESS

One can retrieve the Fibonacci convolution triangle from the Tribonacci
convolution triangle quite simply. First recall
(k)
xn=
a-Dk+1
Replace x by - x ; then it becomes
k/ . < - i ) V = /,
(1
n=0 ' '
xk+1
A
+ x)
'
1970]
167
or
<n=0 "
-vn+k n
x
1)
x = (
1 + x)
With these multipliers,
a (-D n + k .
we can return from Tribonacci to Fibonacci.
Let the column generators of the Tribonacci case be
3n
gnvx) =
11
zrrr
(1 - X -
X2
X3)n+1
and multiplying through by
n \ , -vn+k
and summing, yields

C<0
n=0
n=0
+1
\
/-,
2\k+1
(1 - X - X^)
1 - X - X2 - X 3 /
which are the column generators of the Fibonacci convolution triangle.

st
same thing applies, in general, to return from the (r+1)
to the r
convolution triangle.
The
convolution triangle
168
[March
7. SPECIAL PROBLEMS
1. Assuming Pascal's triangle in Position 1 and the column generators
are
k
Sk
(x)
,k-KL
(1 - x)
'
then show the row sums of Pascal's triangle are the powers of 2. Hint:
00
inns
V* 0n
Zn=0
^
2. Assuming the Fibonacci convolution triangle has its columns positioned

so that
k
gtto =
(1 - X - X 4 )
then show the row sums are the Pell numbers P 4 = 1, P 2 = 2, P

+ P . Hint:
l _ 2 x -
=*-H
VL/
n=0
n+1
+2
= 2P
n+i
3. Show that the convolution triangle for the sequence 1, 3, 3 2 , , 3 ,

can be obtained from the convolution triangle for the sequence 1, 2, 2 2 ,
2 3 , , 2 , using the techniques discussed in this paper.
4. By using the coefficients in
J () <-* *",
1970]
169
as multipliers, show how to get the convolution triangle for the alternate Fibonacci numbers from the convolution triangle for the powers of three.
Hint:
00
!_3i+x2
ZF2n+lXn
n=0
5. By using the multipliers from
n=0
n=0
LW
on the Fibonacci convolution triangle with column generators

k
gk(x)
1 -
obtain the convolr"
A.o
X -
X4
;riangle for every third Fibonacci number sequence,
Hint:
oo
2
1 - 4x + x-
LJ
n=Q
F ,- xn
" 3n+l
8. OTHER CONVOLUTION TRIANGLES

Let,
k^r-I-"**--
be the k,th convolution of the sequence u(n; p , q ) ,
where the sequences
u(n; p,q) are the generalized Fibonacci numbers of Harris and Styles [1].
(Also see [2]..)
170
[March
Let
Jp+q)n
(1 - x) n ( * + 1
(1 - x ) ^ 1
n=0 V 1
n=0
(1 - x) q - x P + q
> /
But,
U &-^
Thus,
oo /
v n
d - A ' .'M-1
(1 - x ) q - x P ^
(1 - x) q -
XP
(1 - x ) ^
(1 - x) q
J.
+(
k+l
^
( ),
,
^r = > u k '(n; p,q)x n
- XP^
K=0
(k)
and the g (x) are the corresponding column generators in the Pascal's triangle with the first k columns trimmed off.
1970]
171
9. REVERSING THE PROCESS, AGAIN

If we consider the convolution triangles whose column generators are
(x
gn(x) -
>
((l-x^-x^J
and if we sum these with alternating signs,
(-l) n g (x)
n-0
^
& " ^
xP+q
*P+q
(1 - x) q - x p + q
(1 - x) q
while
H k S(^ k ) ( """ s w = (i->^
.(1 - x ) 4 - xJ
Thus, we can recover the columns of Pascal 1 s triangle from the above convolution triangle. This may be extended in many ways. Thus, we can obtain
the convolution triangles for all the sequences u(n; p,q) by using multipliers
from Pascal 1 s triangle on the column generators of Pascal 1 s triangle and taking
row sums.
REFERENCES
1.
V. C. Harris and Carolyn C. Styles, "A Generalization of Fibonacci Numbers , " F i b o n a c c i ^ ^
2.
V. E. Hoggatt, J r . ,
Vol. 2, No. 4, D e c , 1964, pp. 277-289.

M
A New Angle on PascaPs Triangle," Fibonacci
Quarterly, Vol. 6, No. 4, Oct. , 1968, pp. 221-234.

3.
H. T. Leonard, J r . ,
'Fibonacci and Lucas Identities and Generating
Function," San Jose State College* Master's Thesis, January? 1969.

4.
David Zeitlin, "On Convoluted Numbers and Sums," American Mathematical

Monthly, March, 1967, pp. 235-246.
SOME PROPERTIES OF STIRLING NUMBERS OF THE SECOND KIND

ROSEDITH SITGREAVES
Logistics Research Project, George Washington University, Washington, D. C.
INTRODUCTION
In attempting to predict the number of demands that will occur during a
given period of time, for supplies in military inventory systems, it becomes
necessary to formulate suitable probability models for the distribution of demands of individual items of supply. One such model, described in [ 1 ] , involves two parameters, to be estimated from available data.
For example,
in the case of predicting demands for items installed on Polaris submarines,

the data might consist of items demanded in a series of patrols.
In studying the properties of estimation procedures for parameters of
any model, one is led to a consideration of the sampling distributions of the
estimates. For the model described in [ 1 ] , the sampling distributions of some
proposed estimates were found to involve Stirling numbers of the second kind,
and in the derivation of these distributions from the initial probability assumptions, some properties of these numbers become of interest.
A Stirling number of the second kind, J l ,
, is the number of ways of
partitioning a set of T elements into m non-empty subsets.
Thus, if we
have the set of elements (1, 2, 3) with T = 3 and m = 2, we have

sible partitions
(1,2), (3)
(1,3), (2)
(2,3), (1)
with
,i3(2) = 3 .
If the order of the partitions is taken into account, that i s ,
(1,2) L (3) ,
172
Mar. 1970
SOME PROPERTIES
OF STIRLING NUMBERS OF THE SECOND KIND
173
and
(3),
(1,2).
are considered to be two partitions, the number of ordered partitions is
For example, suppose that a given item installed in aPolaris submarine

is demanded in each of m partols, with a total quantity demanded of T units,
(T ;> m) The number of different ways of partitioning T into m demands is
0T
; the number of ways in which a particular partition can be assigned to
the m patrols is ml ; thus the number of possible assignments of the total

quantity demanded to the m patrols is m!*L;
PROPERTIES OF STIRLING NUMBERS OF THE SECOND KIND

The generating function of Stirling numbers of the second kind is
T
x = J2 i i ^ x(x - 1) fe - m + 1) .
T
m=0
In closed form,
>*r = ^ra<- m - k *
k=0
Various properties of these numbers are known (e.g. , see [2]). Thus,
J
= 0 for
T< m
J ^
= 1 for
T = 0, 1, 2,
J^0)
= 0 for
T = 1, 2,
Jjp
= 1 for
T = 1, 2,
174
SOME PROPERTIES
[March
We have the recurrence relations
(2)
4rt
jta> + ^toi-i)
or T
m 21
If r = 1, Eq. (3) becomes
(4)
mJW
(J)JM
- 2
=m-l
k=m-
x f
The following results appear to be less well known.

Lemma 1. For any integers r and k, with k = 0, 1, ,
and r ;
k + 1, k + 2, ,
r
(5)
3 =0
='
'
Proof, We prove the lemma by induction on k. In the proof, we use

the recurrences
(r + k\
(r + k - l\
/ v
IT
+ k - l\
/ Ij-w
and
*r-j+k
vr
'
r-j+k-1
r-j+k-1
1970]
For k = 0 and r = 19 29 * * 9
E(5) < - 1)ii S j) = i : (;)(-!)' = a - r = o

For k = 1 and r = 2, 3 9 ,
f-j)
s ( r ; 1 ) - i ) y ^ i " 2 ( j ) < - u U -j+i

^
i=o '
j=o
r
f-j)
E^l)^!:
-j+l
j=0
The last two terms cancel each other while the first one becomes
rr -- 1i
j=0
We suppose the result holds for k = m - 1, and r = m, m +

and show that it holds for k = m, r = m + l , m + 2,
89
We have
|$6
SOME PROPERTIES
g( H
[March
f(r-j)
r-j+m
j=o
'
r-l
2( r + T-)<-^Si
r-l
y- /r
TT \
m - l\
/
(^^(r-j-D
r-j+m-l
Again the last two t e r m s cancel each other. The first term becomes
r
r
1=0
'
-(r-.wf (*-, ' j ^ ^ V ,

3=0
Since r > m + 1 and (r - 1) > m , both of these sums are zero, giving the
desired result.
Lemma 2. For any integers m and T such that T >: 2m
1970]

T-2
.,
177
m-1
E mÊftV-^r-"
k=2(m-l)
'
j=0
X /
)-0 * I
Proof.
T-2
k=2(m-l)
m-1
'
i=0 \ /
m-1
= Cm - 1)!
EI<-
j=0
T-j-2
'
E. I:i,>S-,-I)
(k-j)=2(m-l)-j
From Eq. (4),

T-j-l
E1. (l.-jV^ - *-<*
(k-j)=m-j-l
It follows that
T-j-2
(k-j)=2(m-l)-j
T-j-l
(lij)^- '=
x
'
T
-j-D
.
lk-i/^k-i
(k-j)=m-j-l
E ( ,:])^-
/ T - j W m i-j-D
-j-1
^T - j - 1/^T-j-l
2m-J-3
E (uHr *
(k-j)=m-j-l
= (m - j)JH> - ( T - j ) ^ : ^
m-j-3
2m-j-3
E
r=m-j-l
;'VrH)
178
SOME P R O P E R T I E S
[March
T h e r i g h t - h a n d side of Eq. (7) b e c o m e s

m-1
7
j)
- (m - 1)!
j=o V /
TI (-1) 3
(m-J)
(j - 1)! (T - J)! * T - J
Li=i
m-2
TI (-D 3
JKT- 1-j)|
j(m-l-j)
*T-1-J
j=0
m-1
. v
2m-3-j
'
- (m - 1)
j=0
'
r=m-l-j
The two s u m s in b r a c k e t s cancel each o t h e r since

m-1
3=1
Ti (-1) J
o(m-j)
(j - DKT^jTT^T-j
m-2
V
Z-.
j=0
T!(-1)J
j ! ( T - 1-j)!
In the final t e r m , we s e t k = (m - 1 - j) s o that

(m-2).
jf(m-l-j)
^T-l-j
k r a n g e s from z e r o to
Interchanging the o r d e r of s u m m a t i o n and r e w r i t i n g the e x p r e s s i o n ,
we have
m-1
.v
2m-3-j
(m - 1)!
j=0
'
r=m-l-j
m-2
k=0
'
.THt)fE("rk)jti
' L j=o v
From Lemma 1, each of the inner sums is zero, so that Eq. (8) becomes
j+k|
19701

m
and the lemma follows.

These properties are useful in proving the following theorems.
Theorem 1, Let t i i ^ " - , ^
t. > 1
be m integers such that
i = 1, 2, , m
and
m
m
i=l
Then
T!
-* m
(9)
tm n t.s
i=lx
k t2
= m!
p(m)
Proof. We write
ti + t2 = T 2
ti + t2
... + t m = Tm = T
The summation in (9) can be rewritten as
S xL - Eft E ?I
rm-1
\mi/L
LT2=2
'LTÎ^
179
180
SOME PROPERTIES
But
T,-l
Tf*l
'
From Eq. (4),
T 2 =2 ^
and, in general,
T -1
<T-IM
E ( T T r W" 1 } = r l 4 r )
T
T r ^=(r-l)
iATr-V
r-1
and the theorem follows*

Theorem 2. Let t j , t 2 , , t m be m integers such that
t. > 2 i = 1, 2, , m
and
S fci = T ^ 2m
i=l
then
[March
1970]

m
(10)
2 2 1 - mf V( T \f_i>J.!< m -J>
m
tj t 2
tm n 1.1
x
181
i=i
Proof, Again, let
t 2 = Tj
ti + t2 = T 2
ti1 + toL + + tm
= Tm = T
The desired summation can now be written as
T
T
r T3 " 2 / \
T 2 -2
\ ) L(?J
.=2(m-l)
m-1
T
=2(m-l)\ m - V
2~2 /
To=4
T2
2
Ti=2
TA=2 \
'
T2
- 2 - 2
'
3=0
Repeated application of Lemma 2 leads to the desired result.

REFERENCES
1. R. Sitgreaves, "Estimation Procedures for a Generalized Poisson," (in
preparation).
2. Milton Abramowitz and Irene A. Stegun (ed.), Handbook of Mathematical
Functions, U. S. Department of Commerce, National Bureau of Standards,
Applied Mathematics (1964), 55, pp, 825-826.
A GENERALIZED FIBONACCI SEQUENCE OVER AN ARBITRARY RING

D. J. DeCARLI
Rosary Hill College, Buffalo, New York
L e t S be a r i n g with identity I.
m e n t s of S,
C o n s i d e r the sequence {M } of e l e -
r e c u r s i v e l y defined by:
(1)
M ^Q = A , M ^ + AAM
n+2
1 n+1
On
w h e r e M 0 , Mi9 A 0 ,
for
n 2
0 ,
'
and A 4 a r e a r b i t r a r y e l e m e n t s of S.
Special c a s e s of (1) have been c o n s i d e r e d by B u s c h m a n [ 1 ] , H o r a d a m

[ 2 ] , and Vorobyov [ 3 ] w h e r e S w a s taken to be the s e t of i n t e g e r s .
Wyler
[ 4 ] a l s o worked with such a sequence o v e r a p a r t i c u l a r c o m m u t a t i v e r i n g with

identity. In this n o t e , we e s t a b l i s h s e v e r a l r e s u l t s for such s e q u e n c e s o v e r S
(not n e c e s s a r i l y commutative) which a r e analogues of r e s u l t s d e r i v e d for s i m i l a r l y defined s e q u e n c e s of i n t e g e r s .
We begin by c o n s i d e r i n g a special c a s e of (1), denoted by { F } and d e fined by:
(2)
F ^ = A - F ^ + AAF
n+2
1 n+1
On
for
n > 0 ,
'
w h e r e F 0 = 0,
F j = I and A 0 , A* a r e a r b i t r a r y e l e m e n t s of S.
The fact that S need not be c o m m u t a t i v e c a u s e s difficulty in t r y i n g to

d e r i v e r e s u l t s for the
{ F } sequence,,
However, we note that the t e r m s of
t h i s sequence p o s s e s s an i n t e r n a l s y m m e t r y which e n a b l e s u s to m a k e a s t a r t
at d e r i v i n g i d e n t i t i e s .
T h e o r e m 1.
<3>
If F , = A - F ,- + A A F ,
n+2
1 n+1
0 n
F
n+2 =
n+1A1
then
nA0
Proof: The proof i s s t r a i g h t f o r w a r d by induction.

C o r o l l a r y 1:
(i)
x
'
F ^ - F - - F2 = F nAAF - - F AnF 0 ,
n+1 n - 1
n
n-1 0 n-1
n 0 n-2
182
A GENERALIZED FIBONACCI SEQUENCE

OVER AN ARBITRARY RING
Mar. 1970
(li)
n - l F n + l " Fl = F n - l A 0 F n - l " F n - 2 A 0 F n
183
n
Proof of (i): From (3), we have
Wn-l
" *1 = KA1
=
+ F
n - l V F n - l " F n (A l F n-l
n-lA0Fn-l " FnA0Fn-2
The second result can be obtained in a similar manner.
We note that the r e -
sults of Corollary 1 are analogues of Equation (11) of Horadam ? s paper [2].

The { M } sequence does not, in general, possess the symmetry of the
{ F } sequence and consequently it is even more difficult to work with. There
i s , however, a relation between the ( M } sequence and the ( F } sequence.
Theorem 2:
M ^ = F AAM - + F _,,M ,
n+r
r 0 n-1
r+1 n '
n > 1, r > 0 .
*
Proof: The result i s easily established by induction.

Corollary 2;
M
n = FnMl
+ F
n-1A0M0'
Proof: Interchange r and n, replace n by n - 1 and set r = 1 in

Theorem 2.
We note that the result of Theorem 2 is identical with Equation (12) of
Buschman's paper [1] which was derived for a similarly defined sequence of
integers.
For the { F } sequence. Theorem 2 becomes
(4)
w
F
= F AAF , + F _ F ,
n+r
r 0 n-1
r+1 n*
n > 1.
If we replace n by n + 1 and r by n in (4), then we have

<5>
n+1
+ F
nA0Fn = F2n+l '
The commutator of the { F } sequence i s characterized by
A GENERALIZED FIBONACCI SEQUENCE

OVER AN ARBITRARY RING
184
Mar. 1970
Theorem 3:
F F
- F F
n n+r
n+r n
= F F A F - - F -A.F F ,
n r o n-1
n-1 0 r n
n > 1, r > 1 .
Proof: If we replace n by r + 1 and r by n - 1 in (4), we have

(6)
F ^ = F -A A F + F F ^ .
n+r
n-1 O r
n r+1
From (4), (6), and the fact that S satisfies the associative law for multiplication, we have:
F
n(Fr+lV
= <FnFr+l)Fn
'VWn-^Wn-Vo^l'
= (F -A^F + F F _,, - F ..A.F )F .
n-1 O r
n r+1
n-1 0 r n
' F n ( F n + r " F r A 0 F n - l ) = <Fn+r " F n - l A 0 F r ) F n
'FnFn+r " Fn+rFn = FnFrA0Fn-l " Fn-lA0FrFn
The {M } sequence appears to be very difficult to work with directly.
Investigations indicate that the best that can be done is to concentrate effort on
the ( F } sequence and use Theorem 2 and Corollary 2 to derive analogous r e sults for the {M } sequence.
As a final remark, we note that the sequence obtained from (l)by setting
MQ = R, M t = P + Q ,
P , Q arbitrary elements of S, and A0 = Ai = I,
yields a nice set of identities which are analogues of those derived byHoradam
[2] for a similarly defined sequence of integers.
REFERENCES
1. R. G. Buschman, "Fibonacci Numbers, Chebyschev Polynomials, Generalizations and Difference Equations," Fibonacci Quarterly, Vol. 1, No. 4 ,
(1963), pp. 1-7.
[Continued on p. 198. ]
A REMARKABLE LATTICE GENERATED BY FIBONACCI NUMBERS

S. K. ZAREMBA
University College of Swansea, Wales
Functions which can be represented in the s~dimensional unit interval by

rapidly convergent Fourier series of unit period in each coordinate can be integrated numerically over this interval with great efficiency by averaging their
values over all the points obtained by reducing modulo 1 the coordinates of the
multiples of a suitable vector "g = < g i / p , 8 * , g ,/p> , where g l9 r g ,
s
and p are integers. The crucial property of this vector can be described as
follows: For any vector h" = <h if , h > put
s
R(h) = max(l, h 4 ) max(l, h )
s
and denote by p(g) the minimum of R(h) for all the vectors having integral
coordinates not all zero, and satisfying
g h = 0 (mod 1) ,
where the dot denotes, as usual, the scalar product* Hlawka [5] describes
pg as a good lattice point modulo p if
(1)
p ( g ) > pCSlogp) 1 " 8
because upper bounds for the e r r o r of integration can be expressed as rapidly

decreasing functions of
( g ) , and he proves the existence of good lattice points
modulo any prime for any number of dimensions.
The requirement that p
should be a prime was introduced only in order to facilitate the proof. Understandably, however, one assumes in any event (gi,
M ,
, g . i p ) = l so that
s
g generates exactly p different multiples modulo 1. Of course, here and in
f From September, 1969 at the Centre deMecherchesMatheâtiques, Universite
de Montreal, Montreal, Canada.
JAs a result of a misprint, the exponent of 8 log p appears to be - s in
Hlawka's paper, but his proof applies to lattice points satisfying (1). Thus
his results are sharper than those of Korobov ( [ 7 ] , [8]).
185
186
A REMARKABLE LATTICE
[March
what follows, by a multiple modulo 1 of any vector, we understand the result

of reducing modulo 1 each coordinate of the multiple of the given vector.
In the case of more than two dimensions no recipe other than trial and
e r r o r is known for finding good lattice points, and indeed such a recipe seems
unlikely to exist. However, in two dimensions, the best lattice points in the
sense of maximizing the ratio
(g):p are obtained by putting
P = Fn,
gi = 1,
g2 = Fn__1 ,
where <F > are the Fibonacci numbers [9]. One finds, then, p(g) = F
j
0,
n &
which is of a better order of magnitude than (1).

The case when the integrand has not the required properties of periodicity can be reduced to the periodic case. In the case of two dimensions, denoting the coordinates by x and y, we add to the integrand a suitable polynomial in x with coefficients depending on y, and a polynomial in y with
coefficients depending on x.
The precise upper bounds for the e r r o r ( [ 9 ] ,
[ 12]) are too complicated to be discussed here in detail. Let it suffice to say
that if the integrand f has partial derivatives up to
a 2r f
dx
dyr
of bounded variation in the sense of Hardy and Krause (for a precise definition
see, for instance [5] or [9]), and if we add to it suitable polynomials of degree
r , this allows us to obtain the value of the integral with an e r r o r of the order
F
log F by averaging f over the F points defined above. Trial
computations carried out by this method [12] gave a very high degree of
accuracy.
For instance, taking r. = 3, the value of the integral over the unit
square of exp (-x2 - 2y 2 ) (true value 0.446708379 to nine decimals) was obtained with eight correct decimals from n = 7 onwards, i . e . , using 13 or
more points.
The sets of points corresponding to n = 5, 6, and 7 are shown in Figs.
1 , 2 , and 3. It will be seen that they define regular grids, and indeed square
grids when n is odd. The importance of these grids lies not so much in the
fact that they may be thought to be picturesque, as rather in conclusions of a
far-reaching nature which can be drawn from their existence.
1970]
GENERATED BY FIBONACCI NUMBERS
Fig. 1: n = 5
Fig.2 :n = 6
Fig.3:n = 7
187
188
[March
We begin, however, with a description of the grids themselves.

easily seen that the sets of points in question form lattices.
It is
The lattice gen-
erated modulo 1 by the vector
v = <F" 1 ,
n '
-F"^
n-1 n
will be denoted by L . It obviously has a base formed by the vectors V and

"ej = <0, 1>. The more detailed nature of L
depends on the parity of n.
In
its investigation, we shall repeatedly use the identities

(2)
m+lFn+l
+ F
mFn "
m-*H-1
(see, for instance, I2e in [6]), and

F_n = (-l)n+1Fn.
(3)
When n = 2[x + 1, an alternative basis of L

V7 =
<FMFI1^,
2jLt+l'
-F
-iFl1- , >
-/i-1
and
T2 = <F ^ F l *
L
2jLX+l
is formed by the vectors
jLl+1
, -F
2jLt+l'
F l * ->.
-jll
2/X+l
Indeed, from (3) and from (2) with n = -2fx - 1 and with m = jit - 1 and
m = jLi, respectively, we deduce
F F0
= -F
[X 2jLt
(mod F 9
-/Ll-1
, -)
2jLt+l
and
jll+1
2j[i
-jU
-so that
F V = VA (mod 1) ,
and
2jLt+l
1970]
Thus L 2
189
- contains the lattice generated by "Vj, and "V^.
To prove that, conversely, L 2
- is contained in the lattice generated
by Vj and "V^f we note that by (3)s

-F
- F
=0,
while by (2) and (3),

F2
+ F2
= F
Hence
-F
-V- - F V 0 = e 0
-jLl-1 1
-]Lt 2
2
On the other hand s by (3) and by (2) with m = -\x and with n = \x and n
-\x - 1, respectively! we find
-MPM
WM+1
and
-F
F
- = F, F
= F9
,
-JX -jtl-1
1-jU -jtl
2jLt
so that
F
V- + F- VQ = V .
-jU 1
1-jLt 2
Thus Vi and V^ generate the same lattice as V and e^9 that is the lattice
V+l'
Since Vj and V2 are orthogonal and of equal length 9 L 2 + - forms a

grid of squares with sides inclined to the axes of coordinates. It will be seen
that this grid is invariant with respect to rotations preserving the unit square.
190
[March
Since clearly such rotations transform the grid into a parallel one, it suffices
to show that a rotation by a right angle about the centre of the square transforms at least one lattice point into a lattice point. Now the point V of LJ 0
2]Ll+l
is transformed by such a rotation into the point
<VlF2U'F21M-KL> " % - l ^
(m dl)
'
since, by (2) with m = 1 - 2ju, n = 2JU,
*VlF2
* <m0d * W
Further rotations transform the point in question into
These points form a square with vertices close to the sides of the unit square,
but it does not follow that the sides of this new square are contained in the. grid
formed by LQU+I* ^ i s s o ^ ' anc^ o n ^ ^* ^ i s e v e n anc ^ ^ s
seen as follows.
By (I25) in [6] with n = ju, p = 1, or by Qio), we have
<4)
^-
1-
M-1
can
êst ê
2M
and by (Ii9) in the same book, with n = \x - 1, k = 2,

F2
= F
+ (-1)^ + 1
M-3)
Hence
F
H+1 (F M+1 "
2f*
when jx is even. It follows that in this case, the abscissa of the point
1970]
191
IS
^F2M1
+ F
Since no pair of points L~
2JU
1)F
2jU+l
~ F2M+1
- in the unit square can have the same abscissa,
this is necessarily the point

<1 - F " 1
. F
F"1
>
which was mentioned above as another vertex of the square in question. Let it
be noted in passing that there are 4(F
--F
^) points of L 2 - on the
perimeter of this square. In Fig* 1, this square is shown by thicker lines*

When JU is odd* the ordinate of V2 is negative*
Consequently, it is
along V\ that we should attempt to move from
<%-!%+!'
%+!>
to
-1
K1
"
-1
2JUH-1* F 2jU-l F 2iLi-H > e
But b y (2) with m = JU, n = JU - 1 ,
Wi-V-i* = *V
Hence if w e add (F , - + F - )V- to t h e s t a r t i n g point, we obtain a point of
jLt +1
]LI - I 1
abscissa
<%-l + V F 2iU
= 1
'
192
[March
which shows that, for \x > 1, adding multiples of Vj to our starting point
cannot produce a point of abscissa 1 - F~
+-.
Thus indeed the square in
question is not formed by the grid; this is illustrated in Fig. 3, where this
square is marked in dotted lines.
When n is even, say n = 2JU, the vectors
v f = <F F l 1 , F
jit
2pt
F;1>
and
v\
-jLA 2jLt
.-Fl1, F,
= <F
*
y+l
2JLT
Fl1 >
1-jLt 2jtl
form a basis of L 2 . Indeed, writing V as
<F
2M'
1-2 M F 2]U >
'
we find, by easy applications of (2),

(5)
F V = v j (mod 1)
and
+1V
s Vj (mod 1)
On the other hand, by (2) and (3), we find

(6)
V + F-jLA
V == V
and
-F
1-jU 2
-V' - F
jLtX 1
V' = e 0
-jLl 2
Now (5) and (6) show that the lattice generated by V\ and V^ is nothing else
but
L2]L|.
-r
-T
However, VA and Vj are not orthogonal, their scalar product being

2
(F F
fJL
+ F
F
-jit
pi+1
)F
1-fJL
= F F
/Lt
2fJL
,
2pt
since
F F _ + F
[X
jLl+1
F-]Li l - ] L t
= F (F
jit
Ll
jLt+1
- F
- ) = F2
jtl-1
jLl
When n = 2jL4, Ln does not form a square grid; The determinant of Lu\x
Q|I
being equal to F~* , this would indeed require a pair of orthogonal vectors of
1970]
193
lengths F~"^ each. The cases of M = 2 and M = 3 being trivial, assume

JU > 3. In our search for the required vectors, we can dismiss those which
have a coordinate equal to, or bigger than, F X F~ in absolute value, since
by (4), their length exceeds F~'* . All linear combinations o>V\ + (N\ in
which p ^ 0 are thereby excluded because of their abscissa if j S Z o and
because of their ordinate if afi < 0. There remain the multiples of "vf. But
((78) in [1])
(7)
FQ
= F 2 + 2F F ,, ;
2jLt
jLt
fX j t l + 1
'
This identity can also be deduced from (4) noting that

F2
- F2
= F 2 + 2F F
It follows from (7) that when JU > 3, we have F Q
> 2F 2 , so that Vj
is
jLl
Ci fX
too short for our purposes, while 2V[ has an abscissa exceeding F
-,
so
that it is too long.

The figures representing the lattices with F 5 = 5, F 6 = 8, and F 7 =
13 points show that in each case there is a relatively large number of lattice
points on a straight line passing through the origin. In order to evaluate this
number in general, we must again distinguish two cases according to the parity
of n.
If n = 2(JL + 1, one of the vectors Vj" and "V2
has
both
its
coordinates
positive, one being equal to

F
M F 2 M +1
and the other to

F
The origin being a lattice point, it follows that the line passing through it and
parallel to the vector in question contains
194
rF2M+iF"Ai] +
lattice points, where, as usual,
[March
[x] denotes the biggest integer not exceeding
x. This number is easily determined as follows: By (2) and by (I13) in [6],

we have
xp2
jp2
jU+1
jU
2JLH-1
and
H+l
F2
/Lt X
= (-if
[i
Hence
Vl(Vl
V l > " F2M+1 + ('lf
'
and consequently the number of lattice points on the line in question is
When n = 2jit, one of the vectors V[ and Vj has its coordinates equal
to F F~
and F
-F"
in either order.
JU+1 F JU
%FM-1
But by (2),
fy'
and
F F-
+ F ^-F0
jU, 1-fJL
jU+1 2-jU
= F0 ,
2
or
F F
- F
= (-1)^
1970]
195
Hence
which shows that the line through the origin parallel to the vector mentioned
above contains
F +
, V 2 + i ( 1 + (-1)P)
points of L .
Thus, in either case* there is a line, say 1, which contains a number of
points of L n in the unit square which is of the order v/TT". The importance
of this fact is a consequence of the following considerations.
Let S be any finite set of, say p points of the unit square
0 < x < 1;
and denote by
P(X9J)
0 < y < 1 ,
the number of points of S with coordinates smaller than,
or equal to, x and y, respectively. The function

g(x,y) = p " i/(x,y) - xy
can be regarded as describing the equidistribution of S oyer Q 2 . If a single
number is required to characterize this equidistribution, it can be obtained by
taking any of the plausible norms of g.
In particular, it has been proposed
([1], [11]) to call
D(S) =
sup Jg(x,y)|
<x,yxEQ2
the extreme discrepancy of S in order to distinguish it from other possible

norms of g; the previously used term is simply discrepancy.
If f is any
function of bounded variation in the sense of Hardy and Krause over the closure
of Q 2 , then its integral over Q2 is approximately equal to the average value
196
[March
of f over the points of S, the absolute value of the e r r o r not exceeding VD(S),
where V is the sum of the two-dimensional variation of f over Q2 in the
sense of Vitali and of the (one-dimensional) variations of f(x,l) and f(l,y)
over [0,1]
([3]; for a slight sharpening of this result, see [9]).
Thus sets of points with low extreme discrepancies provide us with a

method of numerical integration over Q2 even when the integrand cannot be
expanded into a uniformly convergent Fourier series. In the case of the set of
points determined by the multiples modulo 1 of V, the extreme discrepancy
has been shown to be smaller than (7/6)F log (15 F ) [9].
However,
the
integrals we may want evaluated numerically do not necessarily lend themselves to a reduction to integrals over Q2. It might seem that, if the domain
of integration, say D, is contained in Q 2 , we could replace the integrand by
a function equal to it in D, and to O outside, integrating this new function
over the whole of Q 2 ; this is what would be likely to be done if the Monte
Carlo method were applied. The difficulty lies in the fact that in general the
new integrand will not be of bounded variation in the sense of Hardy and Krause
over Q 2 , however regular the initially given function might be, and indeed
even if it is a constant, consequently Hlawka f s theorem cannot be applied to this
situation.
The sets of points which have been described above show that even when
the integrand is a constant, say 1, and the domain of integration i s , for instance, convex, the integration e r r o r can be of the order of VF instead of
n
-1
that of F log F . To see this, it suffices to consider two lines, say 1 4 and
1 2 , on opposite sides of 1, parallel to it, and arbitrarily near to each other.
Let D. be the part of Q2 above 1. (i = 1,2). Then the integrals over Dj
and D2 will differ arbitrarily little, while the numerical values found for them
will differ by the number of points of L on 1 divided by F , so that for
one at least of the two integrals the e r r o r of the computation will indeed be of
2
the order of F n-4
.
When n = 2 jbt+ 1, JU, being even, by slightly expanding, or contracting
the previously discussed square formed by the grid, it is possible to obtain a

similar example of an integration domain leading to e r r o r s of the order of F~"2
when applying the method in question; of course, many variations on this theme
are possible.
1970]
197
All these considerations extend to an arbitrary number of dimensions,

and the phenomenon illustrated by the lattice L
in two dimensions becomes
even more accentuated as the number of dimensions increases.
The present
author, impressed by the pattern of L n , proved [10] that if a set of p points

of the s-dimensional unit interval Q s is generated by a good lattice point,
then there exists an s-1-dimensional linear variety (or hyperplane, to use a
rather old-fashioned terminology) which forms with Q s an intersection con1-s 1-l/s
taining more than (4s)
p
'
points of the set. This leads again in an
obvious way to an example of convex domains (actually simple polyhedral domains) having the property that by integrating over them, by Hlawka ! s method,
arbitrarily regular functions, which could even reduce to constants, we are
always liable to commit e r r o r s of the order of p - 1 / s .
This contrasts sharply with the e r r o r committed when integrating over
s
a function of bounded variation in the sense of Hardy and
the whole of Q
Krause and using the same set of points; the discrepancy is then O (Log p) /p)
4 , and the integration e r r o r is of the same order of magnitude. With s > 2,
in the former case the bound obtained for the e r r o r (and this is a sharp bound!)
is much less favorable than that which is practically claimed by the Monte
Carlo method, namely (3(p~2).
The irrelevance of some traditional tests applied to so-called random
numbers, or to pseudo-random numbers with a view to applications to MonteCarlo integration was discussed in detail by the present author
11 ; the con-
siderations adduced here show that when the domain of integration is not r e duced to a multidimensional unit interval, even discrepancy tests in the
appropriate number of dimensions do not guarantee the success of Monte Carlo,
although, naturally, nobody can be denied the right of hoping for the best.
REFERENCES
1. J. H. Hal ton and S. K. Zaremba, n The Extreme and L2 Discrepancies of
Some Plane Sets, n to appear in Monatsh. Math.
2. J. H. Halton,
On a General Fibonacci Identity," Fibonacci Quarterly,
Vol. 3 (1965), No. 1, pp. 31-43.

3. E. Hlawka, "Funktionen von beschrankter Variation in der Theorie der
Gleichverteilung," Ann. Mat. Pura AppL (4) 54 (1961), pp. 325-334.
198
4. E.
Hlawka,
ff
Zur
angerfaherten
Berechnung mehrfacher
Mar. 1970
Integrale, "
Monatsh. Math. 66 (1962), pp. 140-151.

5. E. Hlawka, "Uniform distribution modulo 1 and numerical analysis,"
Comp. Math. 16 (1964), pp. 92-105.
6. Verner E ^ H o g g a t t , J r . ,
Fibonacci and Lucas Numbers,
Houghton
Mifflin Co. , Boston, 1969.

7. N. M. Korobov,
fT
The Approximate Computation of Multiple Integrals,"
(Russian), Doid. Akad. Nauk SSSR 124 (1959), pp. 1207-1210.

8. N. M. Korobov, "Teoretikocislovye Metody v
riblizhennom analize"
(Numbertheoretical methods in numerical analysis), Fizmatgiz, Moscow,

1963.
9. S. K. Zaremba, "Good Lattice Points, Discrepancy and Numerical Integration," ^ ^ ^ ^ ^ 1 ^ :
(4) 73 (1966), pp. 293-318.
10. S. K. Zaremba, "Good Lattice Points in the Sense of Klawka and MonteCarlo Integration," Mojia^sl^âth. 72 (1968), pp. 264-269.
11. S. K. Zaremba, "The Mathematical Basis of Monte-Carlo and Quasi-MonteCarlo Methods," SIAM Rev. 10 (1968), pp. 303-314.
[Continued from p. 184. ]
2. A. F. Horadam, "A Generalized Fibonacci Sequence," American Mathematical Monthly, 68 (1961), pp. 445-459.
3. N. No Vorobyov, The Fibonacci Numbers, translated from the Russian by
Normal D. Whaland, J r . , and Olga A. Tittlebaum, D. C. Heath and C o . ,
Boston, 1963.
4. O. Wyler, "On Second Order Recurrences," American Mathematical
Monthly, 72 (1965), pp. 500-506.
A N EXTENSION OF THE FIBONACCI NUMBERS (PART II)

JOSEPH ARKIN
Nanuet, New York
and
VERNER E. HOGGATT, JR.
San Jose State College, San Jose, California
In this section we consider the generalized Fibonacci and Tribonacci

numbers.
We write the generalized Fibonacci numbers as
OQ
(1 - a lX - a 2 x2)~ k =
(1)
F^ k ) x V
(F y = F ^ ) ,
v=0
where
F
= a
*Fn-l
+ a
2Fn-2'
k = 1, 2, 3,
ls
Fl
and
2 = al
n = 0, 1, 2,
+ a
2>
The generalized Tribonacci numbers we write as
(1 - a lX - a2x2 - a 3 x 3 ) = ] "T
(2)
v<W x v
v=0
where
Ty = T ^ ,
T
= a4T
T 0 = 1,
Tj = a1?
- + T 9 a 2 + a3T .,
nx
n&
n o
T2 = 4 + a 2 ,
k = 1,2,3,"-
F 3 = a? + 2aja2 + a 3 ,
and
n = 0, 1, 2,-
Note: Throughout this section we consider a i9 a 2 , and a 3 , as rational integers only.

199
200
AN EXTENSION OF THE FIBONACCI NUMBERS
[March
CONVOLUTED SUM FORMULAS

FOR THE GENERALIZED FIBONACCI AND TRIBONACCI NUMBERS
By elementary means, it i s easy to prove, if
(1 - y r k = f ) b y v
(3)
v=0
then
( " ^ - l
) - ^
where
V k - 1 1)
( n + k - l ) l / n l ( k - 1)1, b 0 (k) = 1, k = 1 , 2 , 3 , - - . and

n = 0,1,2,---.
Now, in (1), we replace a*x + a2x2 with y so that combining (1) with (3)
we may then write
v=0
v=0
It is easy to prove with induction that
EbS( n J " , )-^J-'n W .

5=0
and combining this result with
/n + k - l\
V k- 1 /
1970]
201
leads to the following generalized Fibonacci convoluted sum formula:
*'"(n+k-i",)(iJ)a?"2i4
(n = 0 , 1 , 2 , - - - , k = l , 2 , 3 , - - - ) .
ajx + a2x2 + %x3
Now in (2), we replace
W ith
y so that combining (2)
with (3), we may then write
t *?*" t *?-r
v=0
v=0
and by comparing coefficients, it is easy to prove with induction, that
*r - E t k^t-i) (vk4r4l2r~2)^
r=0 j=0
-j
(n - 2r - l \ / 2 r + 1 - j \ n-4r-2+j 2r+l-2j j
(k)
Pn^r-l^r-l +J ^
a3
and combining this result with
tf-f;-; )
leads to the following generalized Tribonacci convoluted sum formula:
^ = E E [ ( k + Y - V *)X ( ^ - ^ ( ^ f j)a?-4r+ja22r-2jai
r =0
j=0 L^
'
, /k+n-2r-2 \ /n-2r-l\ /2r+l-j\ n-4r-2+j 2r+l-2j

+
k-i
J\2T+I-})\]
202
AN EXTENSION O F THE FIBONACCI NUMBERS
[March
w h e r e n = 0, 1, 2 , and k = 1, 2, 3 , .
THE GENERALIZED FIBONACCI NUMBER
EXPRESSED E X P L I C I T L Y AS A DETERMINANT
We shall now p r o v e the following five s t a t e m e n t s :
n FP ^r) '
A\r
J= Oaj(k
+ nv, - 1D\ T ? WF ^
iO\r
_L
Q\pW
+ ao 2 (2k
+
n - 2)F
2
-L
where
F^k)
F } k ) = ajk,
= 1,
nF^
*
H.
- p j + q,
Jb ^
n-1
where
%!+!
2(
a^
P2 + P3
p . = aA(k + n - j)
=
n = 2, 3 , ,
k = 2, 3,
q ^
+
"
qn
P n - 1 + Pn
(j = 1, 2 , 3 , , n ) ,
- m)(2k + n - m - 1 ) . (m = 1 , 2 , 3 , - , n - 1),
(n -
2, 3, )
(k -
2, 3, )
F j k ) = ajk .
(a2t + 4 a 2 ) k F ^ 1 = a i n F ^ k ) + a 2 (4k + 2n - 2 ) F * ) 1 ,
m.
where
(k)
Fp
= 1
F ^
= a,k
n = 1, 2 ,
k = 1, 2 , 3 , .
iv. y > ( v ) =
~
!
*'
((a 1 + l + ( ( a 1 + l ) 2 + 4 a 2 ) i 4 a 2 ) ^ ) / 2 ) n - ((a 1 +l-((a 1 +l)2+4a 2 )^)/2) n
"n"-vv
where n = 1, 2, 3,
((a 1+ l)2 + 4 a 2 ) 2
1970]
,(k)
V. F
= K(p l9 q 2 , q n sP n )/nS, (p n >q n are identical to those in (ii) withqj

= 0).
where n,k = l f 2 f 3 , - below in (6).
(6)
203
K(pl9q29"'
and K(p ls q 2 , , q n + p n ) is the determinant given
Pi
Q2
-1
P2
Q3
0 -1
P3
,qn,Pn)
Q4
0 -1
P4
45
...
..
...
...
...
. * o
-a V i
0
-l
%
Pn
The table below of the generalized Fibonacci Numbers (in the table, we
have replaced a l9 a2 in (1) with aA = a and a2 = b)
a2 + b
a3 + 2ab
a4 + 3a2b + b 2
2a
3a 2 + 2b
4a3 + 6ab
5a4 + 12a2b + 3b2
3a
6a 2 + 3b
10a3 + 12ab
15a4 + 30a2b + 6b2
(7)
ka
may be constructed as follows:

(8)
To get the k
multiplied by the k
element in the n
column, we add the product of a
element in the (n-1)
column and the product of b
204
multiplied by the k
element in the n
element in the (n-2)
[March
column together with the (k-1)
column,
We write the k
element in the n
column as F
, so that a restaten '
ment of (8) reads

F<k> = a ^
n
1 n-1
(9)
a-F^
2 n-2
F * ^ ,
n
'
where
F, - 1
F< k) = a l k
0 = F< 0) = F<> = F< 0)
= -
n = 2, 3, '
k = 1,2,3,-'- .
PROOF OF I, H, III, AND IV
We use (9) to get
(10)
F x* = a,
n=2
F< x*
n=2
F<k_>2 x
a2
n=2
F ^
x*
n=2
00
at
4x
F n W x n + a2x2
n=l
FnW xn +
n=0
F * " 1 * x 11 ,
n=2
for k = 1, 2, . Then
F xn - F - F<k)x = a
n=0
F f xn - a r f * x
n=0
l X
a2x* F ^ x11
n=0
F ^ 1 ' xn - FF" 1 ' - Ff- i; x ,

n=0
and therefore
1970]
205
00
(1 - a t x - a2x2) F f x n = F< k) - F
^ + (F^ - alF<k) - F ^ x
1 1
+ -*
V F
.
n^ x
n=0
Now
n(k)
_ ^(k-1)
j 1 - 0 = 1 if k = 1
1 - 1 = 0 if k 2
and
(k)
F! -
aiF0
F (k-1)
- F4
(
" |
i - a i - =
ifk= l |
2| " '
a i k - aj - aj(k - 1) = 0 if k
for k = 1, 2, 3, , and
00
--*
n=0
Therefore
F x11 = (1 - a l X - a 2 x2)- 1 (
n=0
F*"1* x n |
\n=0
and
F^x 1 1 = (1 - alX - a^)" 1

n=0
From this, we have at once
(k = 2, 3, . ) ,
206
[March
oo
(11)
(1 - a lX - a 2 x 2 )~ k = ^
F^ k ) x11
(k = 1,2,3,-) .
n=0
Differentiation of (11) leads to
k(2a2x +
00
00
n=0
n=l
and comparing the coefficients we conclude that

(12)
M^F^
+ 2a2F^1))
= nF^
(k = 1 , 2 , 3 , - - - , n = 2 , 3 , - - - ) .
Combining (12) with (9), we get

nF^ k ) = aj(k + n - D F ^ + a2(2k + n - 2)F^ k) 2
(13)
for k = 2 , 3 , * ' , n = 2 , 3 / , F^ k ) = 1, and F ^ = 54k.

the proof for L
When we divide (13) by F (k)- , we have
J p
Fn X
-
= aj(k + n - 1) + a2(2k + n - 2)(n - 1)

(n F
which in turn along with FQ
(n=2,3,- ,k=2,3,- )
D F ^
n-2
= 1 and F J
= a^k, implies II.
The identity
aA + 4a2 = 4a 2 (l - a4x - a 2 x 2 ) + (aj + 2a 2 x) 2
may be written as
This completes
1970]

A.
+ 4a
(a t + 2a 2 x) 2
4a 2
+
(1 - a l X - a 2 x 2 ) k
(1 -
- a2x2 ) k *
aix
207
(1 - a l X - a 2 x 2 ) k
(M)
(k = 1 , 2 , 3 , - - - )
Differentiation l e a d s to
(a\ + 4a 2 )kx
^r
(1 - &tx - a 2 x 2 )
4a 2 kx
T,
(1 -
/ oo
7F
ajx - a 2 x 2 )
+ (ai + 2a2X)
nF
,n=l
n x"
Now, by c o m p a r i n g coefficients, we conclude that
(a* + 4 a 2 ) k F ^ 1 )
(15)
when FQ
= 1 ,
Fj
= ajk,
= a^F^
+ a 2 (4k + 2n - 2 ) F ( k ) x
n = 1,2,3,98-,
and k = 1 , 2 , 3 , ' " ,
which
p r o v e s III.
We o b s e r v e that Equations (n) and (HI) i m m e d i a t e l y give an e x p r e s s i o n
for
(a21
4a2)F^11)
in the f o r m of a continued fraction, for n = 2 , 3 , " ' , and k = 2 , 3 ,

(Proof of IV).
In (9), we have
F< k ) = a l1F < k >
n
n-1
a2LF<k>
n-2
F^
n
so that
n
(16)
%
V
>
L-4
v=l
n-1
(V)
n-v
= a 2i -**
y>
v=l
n-2
(v)
n - v - 11 +
a L2 L^d
VF
v=l
n
(v)
+^y F <nV- v- 1 )
n-v-2
v=2
(n = 2 >3) -
208
We see that
n-1
v=l
W
n-v-1
n
y > (v-1)
Z ^ n-v
v=2
and we write (16) as
v=l
\v=l
v=l
We let
n
n
A)
n-v
x-f
v=l
then
Vi-E'SU-
V.-S-J1,.
v=l
v=l
so that (17) becomes

(18)
u n = 04 + l )
V l
+ a2un_2 .
Replacing n with n + 2 in (18), we have

< 19)
V2
(a
+ 1)u
n+l
Vn
where
Ui = F 0 = 1, Fj
+ Fi = 1 + a4 = u 2 ,
and
n = 1, 2, 3,
[March
1970]
209
We now solve (19) for u^ by continued fractions (see [ l ] ) , and get

n
un = a i H
s )
. ( a
1 +
l -
s) n )/2 n s = ^
V v
v=l
where
1
s = ((aj + l) 2 + 4a 2 ) r , n = 1, 2, 3,
which completes the proof of IV.
PROOF OF V
Combining Euler f s expression for a continuant as a determinant (see [2])
with (n) and (6), leads to
nF^ k )
(20)
"^^
n-1
K ( p i , q 2 f - - - f q ,p )
n n
K(p 2 ,q 3 ,--- QnPn)
'
for n,k = 2 , 3 , 4 , ' .

Note: For convenience we let
*f/F-U k <>.
Now, using the values of p. and q
(21)
- in (II), we write
nUk(n) = K(aj(k + n - 1),a 2 (n - 1)(2k + n - 2), ,a 2 (2k),a 4 k)

K(a t (k + n -
2)9SL1(JI
- 2)(2k + n - 3 ) , . . .,a 2 (2k),aik)
(n - l)U k (n - 1) =K(ai(k + n - 2),a 2 (n - 2)(2k + n - 3),- ,a 2 (2k),aik)

k(aj(k + n - 3), a2(n - 3)(2k + n - 4), ,a 2 (2k), a ^ )
3Uk(3) = K(a1(k + 2),a 2 (2)(2k + l ) , - , a 2 (2k), aAk)
aj(k + 1)
affikf
-1
a4k
'
210

2U k (2)
aj(k + 1)
[March
a 2 (2k)
-1
ajk
ajk
We now multiply all the equations in (21) from top to bottom to get
(22)
n!
n
j
= n!F^k)/Fi(k)
U k (j)
= K(Pl,q2,. ,qn,pn)/a1k
- 2
for n , k = 2 , 3 , 4 , .
(k)
Now combining ( n , with F*
= ajk) with (22) c o m p l e t e s the proof of V.
We r e s o l v e for k = l
(n = 0, 1, 2 , ) by the u s e of continued f r a c -
tions (see [ l ] ) , and we have
Fn =
( ( a ^ V ^ - f a ^ v r
) ^
i>
where
V = (a? + 4 a 2 ) 2 ,
and
lFn-l
a2F
n-2
(F
0 =
>
i =
FORMULAS
F o r F,(t)
n
Let
(t
N
= 2 , 3 , and 4) a s a function of F - and F .

' '
n-1
n
A = &l + 4a 2 , B(k,n) = 4k + 2n - 2 ,
where
1970J

F ^ k ) = 1,
F ^ k ) = aik,
n,k =
211
1,2,3/',
and
F
= a xt F
- + aL? F 0 ,
n-1
n-2
(where aj and a 2 a r e r a t i o n a l i n t e g e r s ) ; then from (III), we have
^^l}
(23)
a nF
nk)
In (23), we have the following: when k = 1,
(24)
when k = 2 ,
A F ^
2AF^
ai
2B(k,n)F n ( k ) 1
then
nFn + a2B(l,n)Fn-1 ,
then
ainF^
2)
+ a ^ . n j F ^
so t h a t multiplying by 1:A, we get
2 ^ ^ ^
= a j n A F ^ + a2B(2,n)AF^)1
and combining this with (24), we w r i t e (using the identity

a
= a- F - +
1 n-1
2Fn-2>
21A2F\
= a2B(2,n)(ainFn + a2B(l,n)Fn
-)
(25)
+ a 1 n(a 1 (n + l ) F n + 1 + a 2 B ( l , n + l ) F n >
and r e p l a c i n g F ... (in (25)) with F ,- = a - F

n+j.
n-ri
i n
+ a 0 F - l e a d s to
^ n-1
212
A2

,(3) =_ a a n
n-1
[( i 2 B(l*n + 1) + aja 2 nB(2,n) + afn(n + 1))F
[March
(26)
+ (a|B(l f n)B(2,n) + aia2n(n + D ) F n _ 1 ]

when k = 3, then in the exact way we found (26), we prove that
3!A 3 F ( 4 ) =- M + N ,
n-1
(27)
where
ajainBCUn + l)B(3,n) + a ^ n E ^ ^ E ^ n )
+ aja2n(n + l)B(3,n) + a ^ n B d , ! ! + l)B(2,n + 1)
M =
+ aâ2n(n + l)(n + 2) + aia2n(n + l)B(l,n + 2)
F n',
+ aja2n(n + l)B(2,n + 1) + ajn(a + l)(n + 2)
and
a 2 B(l,n)B(2,n)B(3,n) + afa2n(n + l)B(3,n)

N =
+ aia|n(n + l)B(l,n + 2) + aja2n(n + l)B(2,m + 1)
n-1
+ aia2n(n + l)(n + 2)
REMARKS
(k)
The above method may be used to evaluate formulas of the FN
for
values of k = 5 and higher.
1970]
213
THE GENERALIZED FIBONACCI NUMBER EXPRESSED AS A LIMIT

We now prove that
VI
lim
( F ( k + 1 ) /(n + l ) k F ) = (1 + aâj + 4 a 2 ) ^ ) k / 2 k k !
when
lim ((4k - 2)/n) = 0
n oo
(k,n = 1 , 2 , 3 , " )
Let
(28)
A =
SL\
+ 4a 2 ,
V = AJ,
+ a
2Fn-2J
H = {(aj + V) ,
where
F
a F
i n-l
lf
and
l> a 2
a r e rational integers.
It is easy to prove by use of continued fractions (see [1]) that
F n = ((a* + V ) n + 1 - ( a i - V ) n + 1 ) / 2 n + 1 V
(n = 0 , 1 , 2 , . . - ) ,
and then by elementary means we show that

(29)
lim (F / F 1 ) = i(aj + V) = H .
n cj oo nn nni
Now, combining (28) with (III), we have
(30)
AkF
n^l
where n,k = 1, 2, 3,
1)
80e
a nF(k) + a
2< 4k
+ 2n
"
2)F 1
n ?i
'
214
[March
In (30), we have the following: when k = 1, then

A F ( 2 ' = a<nF + a2(2n + 2)F
n-1
* n
*
n-1
and dividing this equation by nF _-, we have
i&
nF
n-1
- Si -(^r 1 )
F
n-1
where combining this result with (29), we write

(31)
A(v lim F . / n F - ) = lim (a txF / F - + a2(2n + 2)/n) = aiH

l + 2aL2 :
n oc n - 1 ' n-1
n->oo
n ' n-1
^
'
-
when k = 2 (in (30)), then
(32)
2AF (3) - = a<nF (2) (F / F ) + a2(2n + 6)F ( 2 ) ,

l
n-1
* n v n' n
n-1
Multiplying both sides of (32) by A/n 2 F
- , we now write
(3)
2
(2)
2A 2 (Fn-1
1X
1 /n F
'
n-1- ) = ai(AF
n '/ F n )(l/n)(F
'
n '/ F n-1- )
(33)
+ a5 \
Ix
n-1 '
Then combining (33) with (31) leads to

, ..
2A2(v lim F ( 3 ) , /n 2 F - ) = aJajH

a14(s J + 2aL2)H + 2a2(aiH + 2a 2 )
n->oo n - 1 '
n-1
* *
*
= (ajH + 2 a 2 ) 2 ;
when k = 3 (in (30)L then

(35)
v
'
3AF (4) - = ainF ( 3 ) (F / F ) + a2(2n + 10)F (3) - ,

1
dX
n-1
n v n
n
n-1
multiplying both sides of (35) by 2A 2 /n 3 F
- , we now write
n-1
1970]

SIA^F^/n'F^)
215
ai(2A2Ff/n2Fn)(Fn/Fn_1)
-<^Hf>X-l>
where combining (36) with (34) leads to

3!A3(nlimooF^)1/n3Fn_1)
(37)
= a ^ H + 2a 2 ) 2 H + 2a2(aiH + 2a 2 ) 2
= (ajH + 2 a 2 ) 3 .
Then, step-by-step, and with induction, we prove that

k!A k lim
(38)
(F< k + 1 ) /(n + l ) k F n ) = ( a ^ + 2a 2 ) k ,
ii > oo
where replacing the A and H in (38) with their respective values in (28), we
complete the proof of VI.
REMARK. It may be interesting to note that if a2 + 4a2 is replaced by
a + 4a2 = (ajk)2 in the right side of VI, then of course
2
(2 k k!F ( k + 1 ) /(n + l ) k F ) = e
lim
n oo
(e = 2.71828-) .
k > oo
AN EXPLICIT FORMULA FOR THE TRIBONACCI NUMBERS

Let
t
" 2 V^l
r=l
where the a
\-l
/
= 1+
oo
2 c(n,t)xn ,
n=l
are rational integers.
In a recent paper (see [3]), it was proved that it is always possible to

express the c(n,t) by an explicit formula when t = 1, 2, 3, 4, and 5.
Then, using the methods in [3] we find the following Tribonacci formula
(T n = c(n,3):
216

, n+2
Xj(x 3
(39)
Tn =
n+2 * ,
- X2
, n+2
) + X 2 (Xi
- Xg
n+2x ,
n+2
) + X3(X2
Mar. 1970
n+2 x
- X4
X^Xg - X 2 ) + X 2 (Xj - Xg) + X 3 (xf - Xj)
where
xj = zj + 4/9z t + 1/3 ,
x 4 = z 2 + 4/9z 2 + 1/3 ,
x 3 = z 3 + 4/9z 3 + 1/3 ,
with
Zi = ( l / 3 ) ( 3 / 3 3 + 1 9 ) 1 / 3 ,
z 2 = -(zj/2)(l - i / 3 )
(.
z 3 = -(zi/2)(l + i / 3 )
/(-D)
and
n = 0,1,2,-" .
REFERENCES
1. G. H. Hardy and E. M. Wright, Theory of Numbers, 4th Ed. (reprinted
with corrections), Oxford University P r e s s , 1962, pp. 146-147.
2. G. Chrystal, Textbook of Algebra, Vol. ii (1961), 502, Dover Publications,
Inc. , New York.
3. J. Arkin, "Convergence of the Coefficients in a Recurring Power Series, TT
The Fibonacci Quarterly, Vol. 7, No. 1, February 1969, pp. 41-55.
4. J. Arkin, "An Extension of the Fibonacci Numbers," American Math.
Monthly, Vol. 72, No. 3, March 1965, pp. 275-279.
5. David Zeitlin, "On Convoluted Numbers and Sums," American Math.
Monthly, March, 1967, pp. 235-246.
FIBONACCI SEQUENCE M O D U L O a p r i m e p = 3 ( m o d 4)
GOTTFRIED BRUCKNER
DAW, Institut fur Reine Mathematik, Berlin-Adlershof, Germany
Shah [ l ] proved: For a prime q > 7 the Fibonacci sequence might contain a complete residue system mod q only if q = 3 o r 7 (mod 20).
Here
we show the
Theorem. Let p be a prime, p > 7, p s 3 (mod 4), then in the Fibonacci sequence, a complete residue system mod p doesn't exist.
It follows from this and Shah f s result: The only primes for which the
Fibonacci sequence possesses a complete residue system are 29 39 5, and 7.
Let p be a prime, p > 7, p = 3 (mod 4). In the following all residues
and congruences are meant mod p . For the Fibonacci sequences
uj = 0f u 0 = 1, Ui = 1*
u2 = 2,
is true:
(1)
n = u aVa + u a-lVa-l'
(2) u k + b = u k _ b ,
= '
"'
n;
n =
*1**"
b=0,---,k,
where g = 2k + 1 is the minimal index so that plu
(for p = 3 (mod 4) g i s
uneven).
(3)
x(g+l)+y
s U
y = * e * > S;
y>
x = 0, l f 29
(You verify these known facts by easy calculations.)

Lemma. The residues
u
s V l
'
s = 1, - - - . g ,
a r e all different.
Proof. From
217
218
FIBONACCI SEQUENCE
U
we define (putting u
aVl
= u
"bVl'
- +u
[March
^
and u, = u, - + u,
Vlub-2
Vl
a-2'
continuing this way, we get

u
lVa
Va+lu0
'
this means
V a
Va+1
'
hence u,
- = 0, hence b = a.
b-a-1
Corollary 1. g ^ p.
Corollary 2. The residues
VVe'
are all different,
Proof,
s = e, , g + e - 1,
e being a given number 1 <1 e < . g.
From
u
aVe
bua-e '
we conclude with
u a = u eu a-e + u e-1-u a-e-1
and
u
(from (1))
eVe
Ve-1
.
2)
1970];
MODULO a PRIME p = 3 (mod 4)
Ve-lVe
219
Ve-lVe
and by the Lemma, a - e = b - e, a = b.

(The Lemma and Corollaries hold, of course, for all primes,)
Proof of the Theorem,
u
From (2) and (3), it is clear that
= 0 or
u * 1 < c < k
c
holds for all n. Therefore the question is whether

{0, u c , 1<: c k}
forms a complete residue system or not.
takes its maximum (p - l ) / 2 .
This might be the case only if k
Hence to prove the Theorem, it suffices to
prove: Is g = p then there is a congruence

<*) u a -
i%
for at least one pair (a,b), 1 < a < b < (p - l ) / 2 .

Putting e = 5, Corollary 2 givess The p residues
u u" r9 s = 5, ,9 p + 4,
s s-5'
*
are all different.
Hence there is a t ,
5<t<^p+4,
u U
t t-5 =
satisfying
From this,
u
= u, ,
for one t,
5<Lt<Lp + 4 .
We differ 4 cases:
a)
b)
t > p,
p > t > t - 5 ( p -
l)/2,
FIBONACCI SEQUENCE
MODULO a PRIME p = 3 (mod 4)
220
c),
t > ( p - l ) / 2 > t - 5,
d)
Case a) is impossible:
U
p4
Mar. 1970
^3'
(from (2) and (3)).
p3 "
(p - l ) / 2 > t .
2'
p2 " *"!'
V l
"
0'
p '
(Check the cases t = p , , p + 4 one after the other
and take into account p > 5.) While Case (d) is a congruence (*) itself, we
easily get such a congruence in Case (b) by utilizing (2).
In the remaining
Case (c), we put

t = (p - l ) / 2 + r ,
1 r<^4 .
We have
u
(p-l)/2+r " % - l ) / 2 - ( 5 - r ) '
From (2), we conclude

u
(p-l)/2+r ' *u(p-l)/2-r
hence
{
**> u (p-l)/2-(5-r)
(p-l)/2-r
p > 7 implies (p - l ) / 2 > 4, therefore in (**), both indices are > 1 , r
and
5 - r always being different (**) is a congruence (*). This finishes the proof
of the Theorem.
REFERENCE
<L. A. P . Shah, "Fibonacci Sequence Modulo m , , ! Fibonacci Quarterly, Vol.
6 (1968), pp. 139-141.
SUMMATION OF POWERS OF ROOTS OF SPECIAL EQUATIONS

FM. A. DRA1M
Ventura, California
and
MARJORIE BICKNELL
A. C. Wilcox High School, Santa Clara, California
It is well known that the sums of the n
powers of the roots of x2 - x -
1 = 0 give rise to the famous Lucas sequence 1, 3, 4, 7, 11, ,

in which each term is the sum of the preceding two t e r m s .
...,
In this case, one
can find the roots

a = (1 - V 5 ) / 2 ,
and
p = (1 + V 5 ) / 2 f
and easily calculate L
= a +p
for n = 1, 2, 3, .
But what of the sums of the n
powers of roots of other equations of the
form
n
x
n-1
-x
n-2
-x
.
- - x - l
A 0
= 0?
Soon the roots cannot be found directly but an interesting pattern of sequences
of integers emerges.
The problem can be solvedusing symmetric functions derived in elementary theory of equations , but we prefer matrix theory.
We use the following
matrix properties. The trace (sum of elements on mail diagonal) of a square

matrix of order n is the coefficient of x
in its characteristic equation,
when the characteristic equation is computed by subtracting x from each main

diagonal element and then taking the determinant. If a matrix is raised to the
k
power, then its new characteristic equation has as its roots the k
powers of the roots of the original equation. So, raising a matrix to successive
powers and summing its main diagonal is a way of calculating sums of powers
of the roots of an equation.
221
222
SUMMATION OF POWERS OF ROOTS OF SPECIAL EQUATIONS [Mar.

To sum the n
powers of the roots of x2 - x - 1 = 0,
which is the
characteristic equation of
ill].
simply calculate the trace for successive powers of the matrix.
And for x 3
x2 - x - 1 = 0, we use
0
I
Lo
oj
For
k-1
0,
write the square matrix of order k having each element in the first row equal
to one, each element in the k
column except the first equal to zero, and
bordering an identity matrix of order k - 1 , as

1
In the table of values occurring that follows, let r

the n
signify the sum of
powers of the roots, and

f(x)i = x - 1
2
f(x)2 = x - x
n
n-1
n-2
ct v
f(x)
= x - x
- x
n
= 0
= 0
- x - 1 = 0
1970] SUMMATION OF POWERS O F ROOTS O F SPECIAL EQUATIONS

I
rH
lO
rH
rH
CO
CO
CD
LO
rH
rH
00
CD
II
I
I
i
J4
J*
II
CO
H
O
&
rH
I
o
rW
II
rH
CO
o
o
14
CO
bQ
d
^
rH
CD
rH
0)
5H
CD
d
o .
II
OS
CO
CO
CO
CQ
||
CD
-d
14
1
3$
00
rH
I-J
J -*!
14
t4
H
ca
co
us
co
J4
^J
wwwwwwwwww
223
224
SUMMATION OF POWERS OF ROOTS OF SPECIAL EQUATIONS
Mar. 1970
Matric theory explains the general form given in the right column of the
table. It can be proved by mathematical induction that if the nxn matrix M =
(a..) is defined as
lj
i = j +1
fl
= J i . i = 1 or
a
ij
) 0 , i ^ 1 and i j + 1
then, if k ^ n, M = (b..) has the following form:
by = 2 k " 1
for
b.. = 0
j = 1,2,- ,n + i - k,
if
k< i
Thus, the trace of M is

2 k - l + 2 k-2 + . . .
and
+ 2
i = l,2,---,n;
i > i .
+ 1 = 2
- i ,
k<n.
Since M satisfies its own characteristic equation,

Mn
= M 11 " 1 + M n " 2 + + M + I
M n + 1 = M n + M n _ 1 + + M2 + M ,
and the trace of M
equals the sum of the traces of the matrices on the
right, giving trace of

M n+1 = (2 n _
1}
= 2
+ (2n-l
1} +
. . . + (22 _ i) + (21-
1)
n+1
- n - 2.
k
Since finding the trace of M for k ;> n involves summing the n preceding
traces already obtained, we can form our table inductively without actually
raising the matrices to powers.
For example, to get the series for n = 5,
the sum of the five roots
raised to the zero power is 5. The sum of the five roots to the first power is
one, the coefficient of -x 4 in
x 5 - x 4 - x 3 - x2 - x - 1 = 0.
The sum of the second, third, and fourth powers are given by
22 - 1, 23 - 1, 24 - 1.
The sum of the fifth powers is either 2 5 - 1 or the sum
5 + 1 + 3 + 7 + 15.
The sum of the sixth powers is the sum of the preceding five entries,
1 + 3 + 7 + 15 + 31 .
SUSTAINING MEMBERS
*H. L. Alder
V. V. A l d e r m a n
G. L. A l e x a n d e r s o n
R. H. Anglin
* Joseph Arkin
L. L. Badii
Don B a k e r
Col. R. S. B e a r d
Murray Berg
Leon B e r n s t e i n
* M a r j o r i e Bicknell
J. H. Biggs
J . H. Bohnert
M. B. B o i s e n , J r .
C. A. B r i d g e r
*Bro A. B r o u s s e a u
* J . L. Brown, J r .
C. R. Burton
N. S. C a m e r o n
L. C a r l i t z
P . V. C h a r l a n d
P. J. Cocussa
L e e Corbin
Nannette Cox
A. B. C u m m i n g s
D. E. Daykin
F r a n c i s DeKoven
J. E. Desmond
A. W. Dickinson
M. H. D i e m
N. A. D r a i m
D. C. Duncan
M. H. E a s t m a n
* Charter Members
Merritt Elmore
R. S. E r l i e n
H. W. E v e s
F . A. F a i r b a i r n
A. J . F a u l c o n b r i d g e
* H . H. F e r n s
D. Co F i e l d e r
E. T. F r a n k e l
C. L. G a r d n e r
G. H. Glabe
*H. W. Gould
Nicholas G r a n t
G. B. G r e e n e
G. A. Guillotte
B. H. Gundlach
V. C. H a r r i s
W. R. H a r r i s , J r .
* A . P . Hillman
M r . and M r s . B. H. H o e l t e r
*V. E. Hoggatt, J r .
*A. F . H o r a d a m
J. A. H. Hunter
D. F . Howells
F . R. J a c h i m
A. S. J a c k s o n
*Dov J a r d e n
J . H. J o r d a n
R. P . K e l i s k y
Kenneth K l o s s
D. E. Knuth
Sidney K r a v i t z
George L e d i n , J r .
Eugene Levine
*D. A. Lind
* C . T. Long
A. F . Lopez
F . W. Ludecke
J. S. Madachy
* J . A. Maxwell
* S i s t e r M a r y d e S a l e s NcNabb
W. A. M o r i n
L u c i l e Morton
Stephen Nytch
R. D. 0 T Connell
P . B. Onderdonk
F . J. O s s i a n d e r
Ann P a p e
R. J . P e g i s
J . W. Phillips
Alvin P o r t n e y
M. M. Risueno
* D . W. Robinson
F . G. Rothwell
B. B. Sharpe
L. R. Shenton
J . A. S h u m a k e r
D. C. Stone
A. S y l w e s t e r
* D . E. T h o r o
H. L. U m a n s k y
M. E . Waddill
* L . A. W a l k e r
R. J . Weinshenk
R, A. White
R. E . Whitney
P . A. Willis
C h a r l e s Ziegenfus
ACADEMIC OR INSTITUTIONAL MEMBERS

DUKE UNIVERSITY
D u r h a m , No. C a r o l i n a
ST. MARY f S COLLEGE

St. Mary T s C o l l e g e , California
NORWICK UNIVERSITY
Northfield, V e r m o n t
VALLEJO UNIFIED SCHOOL DISTRICT

V a l l e j o , California
SACRAMENTO STATE COLLEGE

S a c r a m e n t o , California
WASHINGTON STATE UNIVERSITY

P u l l m a n , Washington
SAN JOSE STATE COLLEGE

San J o s e , California
THE CALIFORNIA
MATHEMATICS COUNCIL
BINDERS NOW AVAI LABILE

The F i b o n a c c i A s s o c i a t i o n is making available a binder which
can be used to take c a r e of one volume of the p u b l i c a t i o n at a t i m e .
This binder is d e s c r i b e d as follows by the company p r o d u c i n g it:
11
. . . . The binder is m a d e of heavy weight v i r g i n vinyl,

e l e c t r o n i c a l l y s e a l e d over r i g i d b o a r d equipped with
a c l e a r label holder extending 2 - 3 / 4 " high from the
bottom of the backbone, round c o r n e r e d , fitted w i t h
a 1 1/2 " multiple m e c h a n i s m and 4 heavy w i r e s . "
The n a m e , FIBONACCI QUARTERLY, is p r i n t e d in gold on the
front of the binder and the s p i n e . The color of the binder is d a r k
green.
T h e r e is a s m a l l pocket on the spine for holding a tab
giving y e a r and v o l u m e ; T h e s e l a t t e r w i l l be supplied with e a c h
o r d e r if the volume or v o l u m e s to be bound a r e i n d i c a t e d .
The p r i c e p e r binder is $ 3 . 5 0 which includes postage ( r a n g i n g
f r o m 50 to 80 for one b i n d e r ) . The t a b s will be s e n t w i t h the
r e c e i p t or i n v o i c e .
All o r d e r s should be sent t o : B r o t h e r Alfred B r o u s s e a u ,
Managing E d i t o r , St. Mary 1 s C o l l e g e , Calif. 94575

The Fibonacci Quarterly

Uploaded by

Copyright:

Available Formats

The Fibonacci Quarterly

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Fibonacci Quarterly

Uploaded by

Copyright:

Available Formats

THE FIBONACCI QUARTERLY

THE OFFICIAL JOURNAL OF

,C. T. Long and J. H. Jordan 135

Verner E. Hoggatt, Jr. 158

Rosedith Sitgreaves 172

N. A Draim and Marjorie Bicknell

THE FIBONACCI QUARTERLY

All subscription correspondence should be addressed toBro. A. Brousseau, St.

Let R(N) denote the number of representations of

k t > k2 > > kj. j>

The integer t is allowed to vary. We call (1.1) a Fibonacci representation of

paper makes considerable use of the canonical representation and a function

It is shown that e(N) is independent of the particular representation.

However, if all the kt in the canonical representation have the

In the present paper, we consider the function R(t,N) which is defined

(kA > k2 > > k r 2 1) ,

F r (x) = F(x; ji, , j r ) = ] ^ f ( t ; j

In particular, if ji = *' = j > then

from which an explicit formula for G (x) is easily obtained.

Also the case

leads to simple results.

(1 + axyXtKa, x, xy) = 4>(a, y, x) .

A(k, m, n) = A(k, n - m , m) + A(k - 1, n - m , m - 1) ,

where it is understood that A(k,m,n) = 0 when any of the arguments is

Comparing (2.4) with (2.1) and (2.2), we get

R(k, N) = A(k, e(N), N) .

It should be observed that A(k,e(n),n) may vanish for certain values of k

< > e<n)> n ) >

R(t,N) = A(t, N - e(N), e(N) ) + A(t - 1, N - e(N), e(N) - 1) .

Now let N have the canonical representation

> 3, it follows that

On the other hand, exactly as in [ 1 ] , we find that

R(t, N) = R(t, e(N) )

in the canonical representation of N be even.

e k (N) = e(e k "' 1 (N)) s

Assume first that s > 1. Then as above

Thus (2.7) becomes

R(t, N) = R(t, e(N) ) + R(t - 1, e(N) - 1)

R(t, N) = R(t - 1, e(Ni) )

Returning to (2.16), since

R(t, e(N) - 1) = R(t - s, e 2 s " 2 (N 1 ) ) ,

R(t, N) = R(t, e2(N)) + R(t - s, e 2 8 " 2 ^ ) )

= 4, Eq. (2,18) reduces, by (2.17) and (2.10), to

R(t, N) = R(t, e(N t ) )

For k4 = 2s > 4, Eq. (2,18) gives

R(t, N) = R(t, e " (N) ) + J ] R(t - | , e 2 s " 2 (N!) ) .

= 2, it is evident that (2.11)

Theorem 1. Let N have the canonical representation

Then, for r > 1, t > 1,

3. For N = F , r > 2, Eq. (2.7) reduces to

R(t, F r ) = Aft, F r _ 2 , F r - 1 ) + A(t - 1, F r _ 2 , F ^

Rft, F r - 1) = Aft, F r - 1 ~ e ( F r - 1), e ( F r - 1) )

Thus (3.1) becomes

In the next place, Eq. (3.2) gives

<t> F 2s+1 "

' F 2 s - 1 " X'

2 s - 1 " X' F 2 s "

It follows from (3.3) that

Rft, F 2 g + 1 - 1) = Rft - s + 1,1)