16 6

J&e Fibonacci Quarterly
THE OFFICIAL JOURNAL OF

THE FIBONACCI ASSOCIATION
VOLUME 16
laBr
NUMBER 6
CONTENTS
On the Density of the Image Sets of Certain
Arithmetic FunctionsIII. . . . .
.
.Rosalind
Guaraldo 481
Evaluation of Sums of Convolved Powers Using
Stirling and Eulerian Numbers.
H. W. Gould 488
b-kdlc Numbers in Pascalfs Triangle Modulo b. ......Heiko Harborth 497
Divisibility Properties of Polynomials in Pascal's
Triangle*..... V. E. Hoggatt3 Jr.3 and Marjorie Bieknell-Johnson
501
The Golden Section in the Earliest Notated
Western Music
.
Paul Larson 513
On Fibonacci Numbers Which Are Powers. .
Neville
Bobbins 515
Primes, Powers, and Partitions.
B. de La Rosa 518
On Odd Perfect Numbers.
G. L. Cohen 523
A Simple Continued Fraction Represents a Mediant
Nest of Intervals.
Irving Adler 527
Folded Sequences and Bode!s Problem
....17. E. Greig 530
Fibonacci Numbers in Coin Tossing
Sequences.
.Mark Finkelstein
and Robert Whitley 539
Strong Divisibility Sequences with Nonzero
Initial Term
Clark Kimberling
541
Minimum Periods Modulo n for Bernoulli Numbers.....
W. Herget 544
The Rank-Vector of a Partition. .
Hansraj Gupta 548
The Andrews Formula for Fibonacci Numbers.
.....Hansraj
Gupta 552
Operational Formulas for Unusual Fibonacci Series.....H. W. Gould 555
A Figurate Number Curiosity: Every Integer is a
Quadratic Function of a Figurate Number...
Harvey J. Hindin 561
Elementary Problems and Solutions.........Edited by A. P. Hillman 562
Advanced Problems and Solutions
Edited by Raymond E. Whitney 566
DECEMBER
1978
Volume Index
570
^ e Fibonacci Quarterly
THE OFFICIAL JOURNAL OF THE FIBONACCI ASSOCIATION
DEVOTED TO THE STUDY
OF INTEGERS WITH SPECIAL
PROPERTIES
EDITOR
Verner E. Eoggatt3
Jr.
CO-EDITOR
Gerald E. Bergum
EDITORIAL BOARD
H. L. Alder
Marjorie Bicknell-Johnson
Paul F. Byrd
L. Carlitz
H. W. Gould
A. P. Hillman
David A. Klarner
Leonard Klosinski
Donald E. Knuth
C. T. Long
M. N. S. Swamy
D. E. Thoro
WITH THE COOPERATION OF
Maxey Brooke
Bro. A. Brousseau
Calvin D. Crabill
T. A. Davis
A. F. Horadam
Dov Jarden
L. H. Lange
James Maxwell
Sister M. DeSales
McNabb
John Mitchem
D. W. Robinson
Lloyd Walker
Charles H. Wall
The California Mathematics Council
All subscription correspondence should be addressed to Professor Leonard

Klosinski, Mathematics Department, University of Santa Clara, Santa Clara,
California 95053. All checks ($15.00 per year) should be made out to The
Fibonacci Association or The Fibonacci Quarterly. Two copies of manuscripts intended for publication in the Quarterly should be sent to Verner
E. Hoggatt, Jr., Mathematics Department, San Jose State University, San Jose,
California 95192. All manuscripts should be typed3 double-spaced.
Drawings should be made the same size as they will appear in the Quarterly3
and
should be drawn in India ink on either vellum or bond paper. Authors should
keep a copy of the manuscripts sent to the
editors.
The Quarterly is entered as 3rd-class mail at the University of Santa Clara
Post Office, California, as an official publication of The Fibonacci Assn.
The Fibonacci Quarterly is published in February, April, October, and
December each year.
ON THE DENSITY OF THE IMAGE SETS OF CERTAIN

ARITHMETIC FUNCTIONS111
St.
ROSALIND GUARALDO
College,
Brooklyn,
Francis
1.
NY 11201
INTRODUCTION
Let n be a fixed but arbitrary nonnegative integer. It is known (see [1],

for example) that n may be uniquely represented in the form n = d 11 +d 2! +
+ dkkl , 0 <_ dj <. J. Suppose that fid,
j) is a nonnegative integer-valued
function of j for each "digit" d, 0 d <_ j , j = 1, 2, . .., and define
k
S(w) = ^2f(dd9
j),
j-i
^(n) = n +
Sin),
fl(fc, 3?) = {^(ar) |fe x <_ p ) ,

(&, P ) = |fi(fc, P ) |
ft(p) = fi(0, P )
(p) = D(0,
p)
^P = {a:|a: = Tin)
for some n}, and
C= {x\x
for any
T{n)
n).
Our objective here is to prove some results concerning the asymptotic

density of the sets Q and C analogous to those which we proved when we considered the representation of n as an integer in base b (see [2] and [3]).
2.
Th&OKm 2.7:
(a)
(b)
EXISTENCE AND COMPUTABILITY OF THE DENSITY

Let fid,
j) , 0 <_ d <_ J be as described above.
/ ( 0 , j ) = 0, j = 1 , 2 , . . .
fid,
j ) = O(Q\) u n i f o r m l y i n j , i . e . ,
sup {fid,
j), 0 < d < j } = o ( j ! )
t h e n t h e d e n s i t y of <R e x i s t s .
VKOOfa: We f i r s t show t h a t
(2.2)
Didkl,
dk\
+ P ) = ( * ) , 0 <_r k\
- 1.
To prove 2.2, let us suppose that

k-l
x = dk\ + Y^ djjl
J
Clearly, T(x)
= T(y)
and
k-l
y = ^! + ]C ^P'!
=i
j= i
if and only if
481
If
482

ARITHMETIC FUNCTIONSI I I
Suppose that dk_
= dk_
= = dk_
=. 0 (or that dk_1
= dk_
[Dec.
= = dk_
= 0).
Since /(0, j) = 0, it must be the case that
We have therefore exhibited a one-one correspondence between the elements of

tt(dkl , dkl + p) and ^(r), 0 p kl - 1, and hence 2.2 follows. In particular, if v - kl - 1, we obtain
(2.3)
(d + l)kl
D(dkl9
- 1) = D(k\
- 1).
Our next result will enable us to find a relationship between

D{(k
fe_+i
and ^ Z?(dfe! - 1 ) .
+ 1 ) ! - 1)
c? = 0
Lojnmci 2.4:
There exists an integer k0 such that for all k >_ k0 the sets
ft(0, k\ - 1) ,fi(/c!, 2k\ - 1) , . . . ,fi(fefe!, (k + 1) ! - 1) are pairwise disjoint,
except possibly for adjacent pairs.
Vsioofi: The maximum value in Q,(dkl , (d + l)kl
- 1) is at most (d + )k\ 1 + kMk, where Mk- max {f(d9 j) , 1 . J ,fc},and the minimum value in Q((d +
2)kl , (6? + 3)fc! - 1) is at least (d + 2)/c!. By assumption (b) , there exists
krQ such that f(d9 j) < j!/2, for all j >L ?c0f, and there exists kQ _> kr0 such
that f(d> j) < jl/2
- k0'Mfc(r , for all j >. k0 , where Mfer = max {f(d9
j) | 1 <_ J <.
/CQ}. Therefore, if k >_k09 we have
E^>
^ = E/^> ^ + E fw** o) + E /WJ o) < KMK

i!/2 -fco^;(^"fco>1 E
+ E
J
= k
+ 1
j=k
J !/2 <
'
^!-
+l
In p a r t i c u l a r , kMk < kl i f fc >. k0 . Hence, we c e r t a i n l y have (d + l)/c! - 1 +

kMk < (d + 2)&! i f /c .> ?c0 , so t h e r e s u l t i s p r o v e d .
Now l e t Xdyk = \Q(dk\ , (d + Dkl - l ) fl fi((d + l)kl
d <_ k - 1. Using 2 . 3 and 2.4 and t h e f a c t t h a t
, (d + 2)fc! - l ) | , 0 <_
D((k + 1)! - l) = J^D(dk\9
W + Dkl - 1) - e,
J=o
where ^ depends on the number of elements that the sets 7(0, Zc! - 1 ) , Q(kl ,
2fc! - 1 ) , ..., ft(/c&!, (A: + 1) ! - l) have in common, we obtain
(2.5)
Z?((fe + 1 ) ! - 1) = (k + DD(kl
fc-i
- 1) - ] T X d . j k .
i =o
fc-i
Let 4 k = (&! - l ) / f c ! and Ek = ] T A d > k /(fe + 1) ! , k >_kQ.
d=o
Therefore,
Then 2 . 5 becomes
1978]

A
k+l
Ak
~
A
k-1
483
~ek
~
=
~ek-l
- Av = - e k
k
so Ak + 1 - Ak^ = -J2eJ>
obtain
k-1
i
- e * ' ^k + i
= A
k0 ~ J2
j = k0
J'
R e p l a c i n g k + 1 by k, we
j=k0
^ =i, o - > , .
(2.6)
J =
' "
fc-i
Clearly, 1/&! 4fe 1 and ]T E^ = Ak^ - Ak <_ Ak <_ 1.
Thus, ]P e^ is a
series of nonnegative terms bounded by Ak , hence is convergent. Let

(2.7)
L = AK - ^ e j .
Note that we have just shown that 0 <_ L <_ 1.

(2.8)
Then, 2.6 yields
Ak = L + ]T j.,fc^ k Q .
Since V ^ e^ = 0(1) as & -> , we have

J =k
^=L+o(l).
Multiplying both sides of this equation by kl and using the definition of the
Aki we obtain
(2.9)
D(kl - 1) = Lk! + o(fc!).
Using 2.3, 2.4, 2.9, and the definition of the A!s and the e's, we have
D dkl
d-l
- 1) = X ^ ^ ! , ^ ^
kl
d-l
" D " XA ^
(3=0
O= 0
d-l
= ^
(Lfc! + o{k\))
+ o((fc +
l)lek)
e=0
= dfc!L + o((k + 1)!) + o((k + 1)!),

i.e. ,
(2.10)
D(dk\ - 1) = dklL
+ o((fc + 1)!).
Now let n = ^ djjl
be any nonnegative integer. Then D(n) = D(dkkl
- 1) +
j-o
D(dkkl,
dkkl +dk_i(k
- 1) I + ) - , wherefiis the number of elements that
the sets Q(0, dkkl -1) and Q,(dkkl , dkkl + ^ ^ ( k - l ) ! + ) have in common.
Hence, if n is sufficiently large, then, by using 2.2, 2.10, and the definition of the A's, we obtain
D(n) = dkk\L
+ D{dk_1(k-l)l
+ ..-) + o((k + l)l)
= dkk\L + (aVi(k-D! + ...) +
+ o{(k + l)l)
o((k+l)\).
k8k

[Dec.
Applying the same type of reasoning yields

D(dk_(k
- 1)! + ...) = dk_(k

= dk_(k
- 1)! +
o(kl)
- 1)!L + o{(k
+ 1)!).
Continuing in this manner, we obtain
D(n) = Ll n - d ^ ! j + M Z dtf\ J + (k - k0),

errors of size o((/c + 1) !) ,
Therefore,
!
o(k\)9
= L - L o(l) + 0(1) + 0(1),
D(n)/n
which implies that the density of 9 is L, so the proof is complete.

Our next result is an immediate consequence of Theorem 2.1.
Con,oJULcUivj 2.11:
If f(d9 j) = f(d) depends only on d, where f(0) = 0 and
f(d) = o(j\)
uniformly in j for all other "digits" d9 then the density of Q
is L, where L is defined as in equation 2.7.
CoKottaJty
one-one.
2.12:
We have L < 1 if and only if the function T(n)
is not
" ^ e j = Ak ~ Z ec < f o r a 1 1 ^ ^ o w n e r e ^ o i s
Ja
fe
J = ô
defined as in Lemma 2.4. Therefore, L <_ Ak if k >_ k . if TOr) = T{y) 9 x + y,
and & is such that k >_ k0 and x <_ kl - 1, y <_ kl - 1; then, since
4* = D(kl - l)/fc!,
Vtioofa:
We have L = Ak^
it follows that L <_ Ak <_ 1. If T is one-one, then it follows from the definition of the A1 s and the fs that Ak = 1 and e^ = 0 for all ?C, so L = 1.
It seems to be true, although possibly difficult to prove, that L < 1 if
each f(d, j) = /(<i) depends only on d and / satisfies the hypotheses of Theoren 2.1. It also seems to be the case that we should always have L > 0 under these hypotheses; this result again will be left to conjecture.
3.
EXISTENCE OF THE DENSITY WHEN f(d,
j)
= 0(j!/j2 log2j)
The main drawback to Theorem 2.1 is the condition f(09 j) = 0. If we assume that f(d9 j) = 0(j!) uniformly in j for all "digits" d9 it seems to be
difficult to find a workable relationship between the quantities Ak9 but on
the other hand, it also seems to be difficult to find an example of an image
set <R which does not have density under this assumption. However, we do have
the following result.
JhzoJiQJtn 3.1:
of Q exists.
If f(d9
j) = 0(j!/j 2 log2j) uniformly in j , then the density

k
Vnoo{:
Let D and fi be as before. If n = Y] d j \
log 2 j) = 0(k\/k2
log2k).
j-i
k
9
then S(n) = ^ 0 ( j ! / j 2
j-i
1978]

Suppose that r <_ s <_ t(r
< t) and s < (k + 1) ! ; then,
D(r, t) = D(r, s) + D(s + 1, t) - |Q(r, s) H Q(s + 1, t) | .

Since S(n) = 0(/c!/k2 log2k), we have
(3.2)
D(r, t) = D(r9 s) + D(s + 1, t) + 0(k!/k2 log2k).
In particular, if r = 0, s = (/c - 1) ! - 1, and = k\ - 1, we obtain

D(k\
- 1) = 0(0, (fc - 1)! - 1) + D((k - 1 ) ! , k\ - 1))
+ 0((fc - l)!/(fc - D 2 log2(k - 1)).

Applying the same reasoning to compute the quantities D(09 jl - 1 ) , 2
<_ k - 1, we see that
(&! - 1) = 0(0) + 0(1!, 2! - 1) + (2!, 3! - 1) +
+ >((& - 1)! , k\ - 1)
+ 0((fc - l)!/(fc - D
log 2 (^ - 1))
+ 0((fc - 2)!/(fe - 2 ) 2 log2(/c - 2)) + '

so we finally obtain
(3.3)
fc-i
Z?(fe! - 1) = D(0) + ^Z?(q!, (q + 1)! - l) + 0(kl/k2
log2k).
Now, by 3.2, we have

D{dk\9
(d + l)k\
- 1) = D(dkl,
dk\) + D(dk\ + 1, (dfc+ 1) ! + 0(kl/k2
log2k)
and by repeated application of 3.2, we obtain

D{dk\,
(d + l)kl
- 1) = D(dkl,
dk\) + D(dkl + 1, dk\ +
+ - + D(dkl + (fc - 1 ) ! ,
2
e r r o r s of s i z e 0(kl/k
k_
i.e.,
(3.4)
!-
(d+l)k\-l
2
log k),
+
Z?(d/c!, W + l)k\ - 1 ) = (dfc!, dfc!) + ^D{dk\
4! > ^ !
+ (q + 1)! - 1 ) % 0(fc!/fc l o g 2 k ) .
Since all integers x which satisfy dk\ + q\ x <_ dk\ + (q + 1) S - 1

the same number of leading zeros, we have
D(dk\
+ q\, dk\ + (q + 1) - 1) = D(q\,
(cf. the argument used to prove 2.2).
(q + 1)! - 1 ) ,
1 1 q <. k - 1
Using this fact, 3.4 becomes

(3.5)
fc-i
0(dfe! , (d + 1) ! - 1) = 0(0) + X (<7! , (? + 1)! - 1)
+ 0(fc!/fc log2k)
and 3.3 and 3.5 imply that

(3.6)
D(dkl9
(d + l)fe! - 1) = D(k\ - 1) + 0(fc!/fc log2fe).
486

[Dec.
Now, using 3.6, we obtain

(k + 1)1 - 1 + 0(kl/k2
D((k + 1)! - l) = D(k\ - 1) + D(k\,

= D(kl - 1) + D(k\9
2k\ - 1) + D(2kl,
2
+ 0(k\/k
= 2D(k\ - 1) +D{2k\,
(k + 1)! - 1)
+ 0(k\/k
log k)
log2k)
log2k)
(k + 1)! - 1) + 0(k\/k
log2k).
By repeated application of 3.6, we finally obtain

D((k + 1)1 - 1) = (k + l)D(kl
- 1) + k + 1,
errors of size 0(kl/k
log k);
thus,
D((k + 1)1 - 1) = (k + l)(fc! - 1) + 0(ft + l)!/fe log 2 k).
(3.7)
Define Ak = Z?(fe! - l)/kl.

(fc + l)\Ak
+1
Then 3.7 becomes
- (k + l)M fc = 0((fc + l)!/fc log 2 k);
and by telescoping, we see that

k
A
k +i = o + E
^ '
lo
g^')k
It is not difficult to verify that ^ 0 ( l / j log2j) = <9(l/log2A:) . Therefore,

j-i
using the above equation, we may conclude that there exists a constant L such
that
(3.8)
Ak = L + 0(l/log fc).
Now l e t n = 2, dk.k\.
j
be any n o n n e g a t i v e i n t e g e r , where each dk. i2 0 .
Then
= 1
D(n) = D(dkmk\m - 1) + D(dkmk\
+A - i ^ _ i
) + 0(fc!/*4 l o g 2 ^ ) .
By the same type of reasoning employed to get 3.4 and 3.7, we see that
D{dkmk\m - 1) = dKD(kl
Since dk
log2km)
+ D{dKk\m,
dknk\m+
f 0 for any j , we have

/
Therefore,
- 1) + 0(kl/k
Z?() = ^ ^ ( L
/m-1
"1
'
\J=X
+ 0 ( l / l o g km))
'
'
+ 0{k\Jkm
log 2 /c m ) + D( Y.dk.kl)
Continuing in this manner yields

DM
= nL + 0(fc!/fcm l o g 2 ^ ) + o ( j ! / l o g
J =I
Hence,
>(n) = nL + Of/cl/log kj
so
Z?(n)/n = L + 0 ( l / l o g fcm) = L + 0 ( 1 ) ,
which proves that the density of ^ is L.
j).
1978]

487
RomoJtk 1: Theorem 3.1 has the drawback that the computability of the
density has been lost.
ROMCUik 2: If we assume that f(d9 j) = o(j\)
uniformly in j , then there
exists an image set < which does not have density. For example, let f(d9 j)
= 0 when j is even and f(d, j) = j! when j is odd. Then,
/
\zi
k-l
fe.i
if /c is odd, and
kl +
k-l
= kl +
X^ ! J
/
J=I
Z 6 ^ ' 1 + (fe - D ! + (fe - 3)! + --- + 1!

j-i
if fc is even. Therefore, the number of integers between k\ and 2kl that belong to ^ if fe is odd is at most 1 + (k - 2)! + (k - 4)! + + 1, and the
number of integers between k\ and 2k\ that belong to 9 if k is even is at
k\ - (k - 1)! - (k - 3)! - - 1!. Hence, if we let 6 and A denote the lower
and upper density of <, respectively, we see that
6 <. 0 + 0(1) and A >. 1 + 0(1),
so 6 = 0 and A = 1.
It is also interesting to note that, if we let f(d9 j) = o(jl)
uniformly
in j , there do exist image sets ^ of density 0. For example, if f(d, j) = 0
when d f 1 or j = 1 and f(d9 j) = 2j! if d =. 1 and j > 1, then no member of
(except 1) has the "digit" 1 anywhere in its factorial representation, and
the set
(3.8)
| n\n = J^ djJl>
dd + 19 1 <3
<k\
is easily seen to be the set of density 0.
Our next result is an immediate corollary of Theorem 3.1.
CotiolloAbj
3.9:
fid)
I f f(d9
= 0(j!/j
j)
= f(d)
depends only on d and
log j)
uniformly in j , then the density of ^P exists.

Finally, just as in [2] and [3], we wish to consider the special case that
arises when we assume that f(d, j) = f(d) =d for all "digits" d [so that Tin)
is the function n + the sum of the "digits" of n] . Clearly, f(d) satisfies
the assumptions of Corollary 2.11, so we know that the density of <R is L9
where L is defined as in 2.7. In this case, it is easy to verify that kQ = 0
and that the value of Xdyk does not depend on d, .Let us therefore set \d,k =
Afc, 0 d <_ k. In the following table, we give the values of Xk and ek to the
nearest 6 decimal places; it appears to be difficult to develop an algorithm
to calculate the \k in general.
Using this table together with Taylor's formula and Lagrange's form for
the remainder, we obtain the following result.
TknoKom 3.10:
When T(n) is the f unction n + the sum of the "digits" of
n9 the density of. 9 is 0.879888. The error made using this figure is less
than e/2 "9!. Therefore, Q has positive density in this case.
488
EVALUATION OF SUMS OF CONVOLVED POWERS

USING STIRLING AND EULERIAN NUMBERS
[Dec.
The Values of Xk and ek , 1 <. k <_ 10
1
2
3
4
5
6
7
8
9
10
0
0
0
2
6
8
14
17
26
39
fc
0
0
0
0.066667
0.041667
0.008929
0.002401
0.000375
0.000064
0.000009
REFERENCES
1.
2.
3.
G. Faber, "Uber die Adzahlberkei der Rationale)! Zahlen," Math, Ann., Vol.
60 (1905), pp. 196-203.
Rosalind Guaraldo, "On the Density of the Image Sets of Certain Arithmetic FunctionsI," The Fibonacci
Quarterly,
Vol. 16, No. 4 (Aug. 1978),
pp. 319-326.
Rosalind Guaraldo, "On the Density of the Image Sets of Certain Arithmetic FunctionsII," The Fibonacci
Quarterly,
Vol. 16, No. 5 (Oct. 1978),
pp. 428-434.
*****

West Virginia
H. W. GOULD
University,
Morgantown,
W. Va.
26506
ABSTRACT
It is shown here how the method of generating functions leads quickly to

compact formulas for sums of the type
0kn
using Stirling numbers of the second kind and also using Eulerian numbers.
The formulas are, for the most part, much simpler than corresponding results
using Bernoulli numbers.
1.
INTRODUCTION
Neuman and Schonbach [9] have obtained a formula for the series of convolved powers
n
(1.1)
S(i,j;n)
= fcf(n - k)s'
fc = o
1978]

489
using Bernoulli numbers. Although the formula expresses S(i,j;n)

as a polynomial of degree i + j + 1 in n, and this mode of expression is useful, still
the formula is rather clumsy and hard to recall. Below we shall show how the
method of generating functions can be used to obtain elegant closed forms for
(1.1) very quickly. The first of these uses the Stirling numbers of the second kind, and the second uses the Eulerian numbers. Both results give (1.1)
as series of binomial coefficients in n, rather than directly as polynomials
expressed explicitly in powers of n. For many purposes of computation and
number theoretic study, such expressions are desirable. The significant results below are formulas (3.6), (3.8), (5.3), and (7.3).
Glaisher [4] and [5] was the first to sum (1.1) using Bernoulli numbers.
Carlitz [3] has shown some extensions of [9] and connections with Eulerian
numbers. Our results overlap some of those of Carlitz, but were obtained in
August 1974 before [3] was written.
2.
(2.1)
G(t;i,j)
Y,
A GENERATING FUNCTION
tnS(i,j;n).
n=0
Then
G(t;i,j)
k =0
n=*k
k=0
n=0
so that we have at once the elegant generating function
(2.2)
= kHk ^ n -
G(t;i,j)
k=0
n=0
The generalized power series
2>v<
k
may be summed in a variety of ways. We shall use the methods of (i) Stirling
numbers of the second kind and (ii) Eulerian numbers. Our (2.2) is (3.4) in
Carlitz [3].
3.
METHOD OF STIRLING NUMBERS OF THE SECOND KIND
It is an old fact that

P
(3.1)
(tDff(t) = X S(p9k)tkDkf(t),
k=0
where D = d/dt and S(p,k)

citly,
(3.2)
k!S(p,k)
is a Stirling number of the second kind.
Expli-
= A V = E<-D*- J "0)fc P .
j-o
The formula dates back more than 150 years, but, for a recent example, see
Riordan [10, p. 45, ex. 18]. Riordan gives a full account of the properties
of Stirling numbers of both first and second kinds. Other historical remarks
and variant notations are discussed in [6], Applying the formula is easy because (tD)ptk
= kptk,
whence we have
490
(3.3)

^ V = V k\S(p,k)
[Dec.
^ .
This, too, is a very old formula. It converges for \t\ < 1, but we treat it
as a formal power series. Carlitz [2] gives a good discussion of formal power
series techniques.
Using (3.3) in (2.2), we find
i+j
(3.4)
, +2 Z ^ ! ( p -
=E "
G(t;iJ)
(1 - t)
r=0
k)lS(i,k)S(j,r-k).
k=0
Throughout the rest of the paper, we shall write, for brevity,

r
(3.5)
5,(i,j) =2/c!(r-/c)!5(i,/c)5(j,2-/c).
k=o
Applying the binomial theorem, we find next
i> = 0
n = 0
=Ei(i)-.'''
v = 0 n=r
n=0
r=0
In the next-to-last step here, the upper limit v = i, + j might as well have
been v = because of zero terms involved, since S(p,k)
= 0 when k > p. This
makes manipulation easier. Equating coefficients of tn and dropping some zero
terms, we find finally then our desired formula
(3.6)
S(iJ;n)
=J2[Hl)sr(i,j).
r =0
This simple expression may be compared with the bulky form of expression given
in [9] using Bernoulli numbers.
Having found our desired formula, we can next offer a much quicker proof.
Recall [10, p. 33] that
n
(3.7)
x" -
(X\v\S(n,r).
V =0
This gives at once

k
ky =Y(
?o
rlS(i,r)i2slS(j,s)(l)(n-k),
ZTo
\i /\ s /
whence, using formula (3.3) in [8], a modified Vandermonde addition formula,

we get on summing from k = 0 to k = n,
(3.8)
S(i,j;n)
^riSfi.rl^lsy.s)!^!
r=Q
s=0
1978]

491
By simply putting s - v for s and interchanging the summation order, we see

that this is nothing other than our former result (3.6).
h.
EXAMPLES OF THE STIRLING NUMBER METHOD
For the sake of completeness, we recall [10, p. 48] some of the values of
S(n,k):
0
0
1
2
3
4
5
6
7
1
1
1
1
1
1
1
1
3
7
15
31
63
4_
3_
1
15
140
1
21
7 -U
k_
1
1
6
25
90
301
1
10
65
350
n
Here, S(n,k)
= 0 when k > n and S(n,0)
= 0 for n >. 1.
For j = 0, formula (3.6) becomes the well known
1
(4.1)
S(i
,0;n) = (" + J)r!S(i,i>), w >. 0, i > 0 .

v> - n
>
'
Incidentally, in some places in the vast literature r\S(i-,r)

has been called
a Stirling number, and both arrays turn up very often in odd places with new
notations. There are at least 50 notations for Stirling numbers. Here are a
few examples of (4.1):
S(l,0;n)
(n+21),
5(2,0;) = ( n 2 1 )
+ 2
("31)'
5(3.0;,-(J 1 ) + 6(^)+ 6(; 1 ).

S(,,0;n)
- (n+21)
14(^)
Se^1)
2A(^1).
For j = 1 we shall obtain substantially the same coefficients, the difference

being that the lower indices are each increased by 1. Thus:
5(2,1;) = {
) + 2^
J=
12
>
3?r
5nd + 2n
60
^^^^crj^ftVKn )-^; )
2n6 - 5 ^ + 3n^
60
where we have indicated, for comparison, the values obtained in [9].
492
[Dec.

For j - 1, the following is a brief table of the coefficients in the array:

1
2
1 6
6
= 4
1
14 36 24
= 5
1
30 150 240 120
= 6
1
62 450 1560 1800 720
3, we find the following formulas:
i =2
i =3
i
i
i
For j =
sa,3;n)-(n+31)+6(nl1)
6(n+51)t
5(2,3;.) = (" +1 ) + a f j 1 ) + i s f ^ 1 ) + 12(^; 1 ),

S(3,3;n> - (" +1 ) + n f ^ 1 ) + A S ^ 1 ) + 7 2 ( ^ 1 ) + S G ^ 1 ) ,
and so forth.
5. METHOD OF EULERIAN NUMBERS
The Eulerian numbers [1], [10, pp. 39, 215] are given by
(5.D
Afi
^ ( - D T ^ C J
k)K
k=o
These must not be confused with Euler numbers appearing in the power series
expansion of the secant function. The Eulerian numbers satisfy
A n>3 = An,n-j
and
An,
= jAn.1}j
+ i9
row symmetry, n >_ 1 ,
+ (n - j + D ^ - i , j - i ,
n
J = l
Again, for completeness, here is a brief table of A n ^

0
0
1
2
3
4
5
6
7
1_
3_
4_
5_
1
120
J_
1
1
1
1
1
1
1
1
1
4
1
1
11
11
26
66
26
1
57 302 302
57
120 1191 2416 1191
These numbers are frequently rediscovered, for example, recently by Voelker

[11] and [12], where no mention is made of the vast literature dealing with
1978]

493
these numbers and tracing back to Euler. For our purposes, we need the wellknown expansion
(5.2)
x> K ^= a -
tr-lyEtkAntk.
k=0
k=0
This expansion is known to be valid for \t\ < 1, but again we treat all series
here as formal power series since we do not use the sums of any infinite series. We never assign t a value, but equate coefficients only.
Applying this to (2.2), we find
.
= (1 - t)" ~ XX^ j r X]t%. ) e
G(t;i,j)
s=0
r=0
i + 3
E( i+J 'r +1 )*5>x;

fe=0
r=0N
n=0
A .
r=0
s=o
"
's=0
and by comparison of coefficients of tn
A .
we have our desired formula
i + j
.3)
w^-zfi^nE
A-
A
j,
r-8
Here we have again dropped some of the terms that are zero by noting that
An = 0 whenever j > n.
Formula (5.3) is (3.6) in Carlitz [3].
As with our previous Stirling number argument, we could obtain (5.3) by
another method. We recall that in fact
^n=i2(x+i~i)^,i
(5.4)
J=0
and form the product kz (n - k)J and sum from k = 0 to k = n to obtain a formula for (5.3) analogous to (3.8). We omit the details.
6.
EXAMPLES OF THE EULERIAN NUMBER METHOD
When j = 0, formula (5.3) becomes, of course, the familiar relation
(6.1)
S(i,0;n)
= (" + l)Ai, r . n >0, i
>l.
r =l
To see that this is so, we proceed as follows.

î,
\
r= 0
i+1
)Ai,r,
By (5.3),
s ^0, v - s
s i n c e A0jr_s
= 0 f or r f
s,

k3k
iL \
)Ai,r
"+i
> since Ait0
[Dec.
= 0 for i >_ 1,
r= 1
= Y^ Q + iV^-^+i' by P u t t i n S ^ - ^ + 1 for p,
r= 1
t
( '+i)Ai,r9
by the symmetry relation.
I = 1
For J = 0, then, we have the following formulas:
5(1,0;n) = j ^ 1 ) ,
S(2,0;n)
=("^)
(^
),
W . O ^ - C ^ + l l ^ + ^ + ll^^J + f'J4),
etc.
For j = 1, we find
5(
2,i;)-(M;1)
s(3,i;n).(B;i)
and so on.
7.
+ (n:2)
*f,;2) + (n;3)
These again are a different way of saying what was found in [9],
ALTERNATIVE EXPRESSION OF THE STIRLING NUMBER EXPANSION
( n +1\
Formula (3.6) uses the values of I
-. ). We wish to show now that we can
transform this result easily into a formula using just f . i)> i-e? directly as a series of binomial coefficients in n rather than n + 1. We will need
to recall, see [10], the recurrence relation for Stirling numbers of the second kind
(7.1)
S(m,k) = kS(m-l,k)
S(m-l,k-l).
In this, set m = j + 1 and replace k by v - k.

(7.2)
S(j+l,r-k)
= (r-k)S(j9r-k)
We get
+ S(j ,r - k - 1).
Now, by (3.6) and the usual recurrence for binomial coefficients, we have
1978]

1 +J
i+
'
^95
D(")5,(i,;) + U
(Z)Sr-iV>3)
V=0
r=0
However, Sr(i,j)
Sr_1(iij)
V - l
= J^ kl (r -k)\S(i,k)S(j,r
k=o
-k)
k\ (r - 1 - k) \S(i,k)S(j,r
- 1 - k)
=o
fc
v- 1
]T 7<! (p -fc- 1) \S(i9k)
(r - k)S(j,r
- k) + ^ fc! (p - 1 -fc)\S(i9k)SU,r
k =0
-1-k)
k =0
= X
feKr-l-WllW^W+l^-fe)
-5(i,/c)5'(j,3?-fe-l)},
by (7.2)
k=0
r- 1
+ J^kl(Y>-l
-k)\S(i,k)S(j,r-l
-k)
fe=0
v-1
= f c ! (r-fe-l)!5(t,W5(j+l,r-W + r\S(i,r)S(j
,0).
k=0
The extra term here may be dropped when we consider j >_ 1. . Therefore, we have
the new result that
i+j
(7.3)
= ]T ( J )fc! (r-fc) !S(i,fc)S(j+l,r+ 1-fc), j i l , i i O .
S(i,j;n)
r=0
fe
Let J = 1 again.
ExampteA:
=0
We find
5(0, l;n) = (J) + (2),

5(l,l;n)-(;)
(5),
5(2,l;) -(?)+ 3(5) + 2 ( ) ,

For j = 1, the general pattern of these coefficients begins as follows:
0
1
2
3
4
5
6
1
1
1
1
1
1
1
2_
1
3
7
15
31
63
3_
4_
5_
6_
2
12
6
50
60
24
180 390 360 120
602 2100 3360 3520
7 -
720
It is interesting to note that these coefficients appear in another old

formula:

k36
(7.4)
S(i,0;n)
<-!)*(*. + ) E
fc = 0
[Dec.
( )) W + D* ,
j" = 0
valid for J> 1, n >_ 1.

Ex.ampZ&>:
S<1,0;) - (J) + (^),
*"> ( I ) + 'G) + " ( 3 ) + 'fi)and so forth.

There is yet another old formula involving Stirling numbers of the second
kind which we should mention. It is
i
(7.5)
S(i,0;n)
=E
(-1)*" "("+ iVS'(i,r) ,
n >. 0, i >. 1.
r= 0
This occurs, for example, as the solution to a problem [13] in the .American
Mathemat-ieal
Monthly.
Examples:
S(l,0;n)
= j^1),
5(2,0;n)-.("J1) + 2 ( " + 2 ) ,
^<3,0;) - ("J1) - 6( + 2 ) +
^(4,0;n) = -(-; 1 )+ "(W32) " ^
and so forth.
8.
D+ 24 ( W D'
FINAL REMARKS
It is interesting to note that the original sum (1.1) is a type of convolution. So also formulas (3.6), (5.3), and (7.3) involve convolutions of
the Stirling and Eulerian numbers. The formula found in [9] is not of this
type. This is so because of the way in which the binomial theorem was first
used. It would evidently be possible to obtain convolutions of the Bernoulli
numbers. To get such a formula using Bernoulli polynomials is easy. Let us
recall that
(8.D
~=Y,h~B(x)>
1*1 < 27T >
defines the Bernoulli polynomial Bn(x).

Then Bn(0) = Bn are the Bernoulli
numbers. It is also a well-known old formula that then for all real x>
n
(8.2)
x" = ; r ^ T E r 1 H f e ) , n >
k-o
1978
fr-ADIC NUMBERS IN PASCAL'S TRIANGLE MODULO b
h31
Form the product k1'(n - kY by using this formula to expand kl and (n - k)J'.
Sum both sides and we get
(8-3)
S(iJ;n} - ^t^^^jt^^tBAmAn
r=0
'
s=0
- k),
k = 0
which brings in a convolution of Bernoulli polynomials. Since the Bernoulli

polynomials may be expressed in terms of Bernoulli numbers by the further
formula
n
(8.4)
{l)xn""B^
BAx) = X)
m = Q
it would be possible to secure a convolution of the Bernoulli numbers. However, the author has not reduced this to any interesting or useful formula
that appears to offer any advantages over those wev have derived here or those
in [9]. We leave this as a project for the reader.
It is also possible to obtain a mixed formula by proceeding first as in
[9] to get
7(i,j;n) = (-1)' (^J'~l>i+r>

v =0
'
k=0
apply one of our Stirling number expansions to the inner sum and get, e.g.,
J
i+r
S(i,j;n) = ('V ( ^""'"'E (l X 1}klS(i
(8.5)
+r k)
> >
fc=0
V=0
but the writer sees no remarkable advantages to be gained.

REFERENCES
1.
L. Carlitz, "Eulerian Numbers and Polynomials," Math. Magazine>

(1959), pp. 247-260.
, ,_.
v
'
vv
(continues
Vol. 32
^n\
on page 560)
b-ADIC NUMBERS IN PASCAL'S TRIANGLE MODULO b

HEIKO HARBORTH
Technische Universitat
Braunschweig, West Germany
For the binomial coefficients in Pascal's triangle we write their smallest nonnegative residues modulo a base b. Then blocks of consecutive integers
within the rows may be interpreted as Z?-adic numbers. What Z?-adic numbers
can occur in the Pascal triangle modulo bl In this article we will give the
density of such numbers and determine the smallest positive integer h(b) ,
such that its b-adc representation does not occur (see [3] for b = 2).
We use the notation
m
t
Y1 a
i=0
for positive integers t .

which do not occur.
(amam-l
'"
v \
>
' am $ >
First we will prove the existence of b-adc
numbers
498
&-ADIC NUMBERS IN PASCAL'S TRIANGLE MODULO b
L<imma 1:
gle modulo 2.
(1011) 2 is not to be found within any row of the Pascal trian-
[Dec.
PXOO&i We assume that there are integers n and k with
(*H*H*I S ) SI - (*Ii)s < -

These congruences substituted in
<* + ! + *>(* + ? + * ) - < * - * - * > ( * ; * )
a)
for i = 0, 1, 2, gives n E k (mod 2 ) , & E 0 (mod 2 ) , and n E 1 (mod 2 ) , respectively, which is a contradiction.
Lomma 2:
(111)^ is not to be found within any row of Pascalfs triangle
modulo b with b > 2.
Psioofi:
We assume that
Together with (1), for i = 0 and = 1, we conclude that n E 2& + 1 (mod 2?),
and w E 2fc + 3 (mod b), respectively. However, both congruences are possible
only if b = 2.
We are now able to determine the density.
ThdOJiQjn 1: Almost all b-adc
Pascal's triangle modulo b.
numbers cannot occur within the rows of
VKOO^X As noted in [4], it is well known that the density of those -adic
integers not containing a given sequence of digits is 0 (see [2], p. 120).
Thus, the proof is given by Lemmas 1 and 2.
Jk2.0h.Qyn 2: Let h(b) be the smallest b-adc number not being found within any row of Pascal's triangle modulo b. Then, h(b) - b2 + b + 1 = (111)^
for b > 2, and 7z(2) = 11 = (1011) 2 .
We first prove two lemmas.
Lemma 3: Let b = bYb2 with (bl9 b2) = 1. Then (am ... a0)b occurs in the
Pascal triangle modulo b if and only if (aim ... a ^ ) ^ for i = 1, 2 occur in
the triangles modulo bi with a^j E a3- (mod b^) , j = 0, 1, . . . , m.
VJ100&: One direction of the proof is trivial.
, ,
In the following, we use the result of [1] and [6], that (^j (mod b) is
periodic for fixed k with the minimal period N being the product of all prime
powers p a + $ with pa from the canonical factorization of b and 3 from p$
^k
< p+ 1. Thus, N depends only on the prime factors of b and on k (see [5] for
further references). By reasons of symmetry, a corresponding periodicity of
I n _j_ \
length L holds for I 7, , p ) with fixed n and k.
From this and by the assumption, we are able to find n^ and k^ such that
for i = 1, 2,
with minimal periods L^ each being the lowest common multiple of m + 1 minimal
1978]
&-ADIC NUMBERS IN PASCAL'S TRIANGLE MODULO b
499
periods. From (Z?l5&2) = 1 we have (L l 5 L 2 ) = 1. Thus, the diophantine equation,

K. 1
"t" X TJ I
has solutions o^ , # 2 .
N with
lni
fVo
'
vC lyLl p 5
For fixed values xl9
+xiLi+yiNi\
we then have minimal periods
x2J
_
(mod bi),
j = 0, 1,
. . . , m.
Finally, (/1/1, /l/2) = 1 guarantees solutions z/15 z/2 of

n 2 + a^Lj + ylNl
= n2 + x2L2
y2N2,
which completes the proof.

Lzmma 4: In Pascal's triangle modulo p a , p being a prime, there are arbitrarily large partial triangles with
0 ) = *(.*) ( dP),n>0, fc>0.

for every r from 1 to p a .
We first show
VK.00^',
(2)
(^/T)
( m d pa) f r pa3
" P a 3 " a + 1 < ^ < P a B + p a 3 ' a + 1,
fc ^ p a 3 .
Let y be the exponent of p in the canonical factorization of the binomial

coefficient in (2). Then, by a theorem of Legendre ([7], p. 13), we have,
rp a g - k
ppQ
y-Z
pi
i2-l
T. -
V%
t = a$ - a +1
where [#] means the greatest integer not exceeding x.

We further show by induction on a that (
a
a complete system of residues modulo p .
(3)
Pj (r) -
TT
(r
+l
"y
ag
1, for v = 1, 2,
Let
i =1
Then for a = 1 we can write

3
K)= ^TT^-^)
\ ^
Ev
< m o d p)
j -1
In general, with p = z;pa_1 + p, l ^ . p ^ p a _ 1 , 0 <. z; <L p - 1, we get

/
pp
a$\
a - 1
( ae ) = TJ =Tl j ( )
a _ 1
^TT
j( )
J = l
TT J(P)
J-.l
^p
a_1+
~1
P T<7T=P 1J ( P )
<mod p a >-
500
i-ADIC NUMBERS IN PASCAL'S TRIANGLE MODULO b
Dec. 1978
If we assume pirP. (p) to take all residues modulo pa~x,

then the induction is
complete.
As (3 may be chosen arbitrarily large, Lemma 4 follows with nr = rpa and
kT = p a e .
VKOOh ofi Tk&Ofl&n 2: Lemmas 1 and 2 yield h(b) <_ . . . . Because of Lemma
3, we need to consider only prime powers as moduli. Trivially,
teo)p-:x , 1 1 a0 < pa,
occur as [-L ) in the Pascal triangle modulo p a (let n = a0 and k = 1) , and so
do (la 0 ) pX
(let n - a0 and k = 0,1), with 1 _ a0 j p a .
We then multiply the
digits of (la0)pa by r, 1 < p < p , and obtain all numbers (a1a0)pa

ing those with (a1? p a ) > (a0, p a ) .
metry of binomial coefficients.
, includ-
This is because of Lemma 4 and the sym-
Further, (100)pa occurs if n = 2p a , k = 0,
1, 2, and (110)pa if n = 2pa + 1, k = 0, 1, 2.

Now
so that f^j E 0 (mod p a ) , if n = rp a - 2 and k = p a - l .
Using (3), and with
y being an integer, we have

(4)
(5)
(7aa_"22) " ^ - r T T ? - (r) = 1 + vp (mod p),

\P
* I
rpa - 1 j = i J
(,-- 2).
, . i , ' - - ^ ; - ^ : , - ! ) . * - !>(->>;-) <, P . , .
As (1 + vp, p a ) = 1 , we can find an integer x such that multiplying (4) and

(5) by x yields the residues 1 and r - 1. Because of Lemma 4, corresponding
binomial coefficients occur in the Pascal triangle, so that the existence of
all numbers (10(r - l))pa , 2 <_ r <_ p a , is proved.
Thus, we have shown h(b) >_ (lll)z? for b _> 2. The remaining binary numbers
(111)2, (1000)2, (1001) 2j and (1010)2 are to be found within the rows 3, 4,
5, and 6, respectively.
REFERENCES
1.
2.
3.
4.
5.
6.
7.
R. D. Fray, "Congruence Properties of Ordinary and ^-Binomial Coefficients," Duke Math, Journal,
Vol. 34 (1967), pp. 467-480.
G. H. Hardy & E. M. Wright, An Introduction
to the Theory of Numbers, 4th
ed. (Oxford: Oxford University Press, 1962).
H. Harborth, "Aufgabe P 424, Dualzahlen im Pascal-Dreieck," Praxis
der
Mathematik,
Vol. 13 (1971), pp. 76-77.
D. Singmaster, written communication to the author.
D. Singmaster, "Divisibility of Binomial and Multinomial Coefficients by
Primes and Prime Powers" (to appear).
W. F. Trench, "On Periodicities of Certain Sequences of Residues," Amer.
Math. Monthly,
Vol. 67 (I960), pp. 652-656.
E. Landau, Vorlesungen
uber Zahlentheorie
(New York: Chelsea, 1950).
###*#
D I V I S I B I L I T Y PROPERTIES OF POLYNOMIALS IN PASCAL'S TRIANGLE

V. E. HOGGATT, J R . ,
San Jose
State
and MARJORIE BICKNELL-JOHNSON
University,
San Jose,
California
95192
D i v i s i b i l i t y p r o p e r t i e s of t h e F i b o n a c c i sequence {Fn}
i n c l u d i n g t h e p r o p e r t y of g r e a t e s t common d i v i s o r s ,
\^m s ^n '
a r e w e l l known,
~ ^ (m, n) '
Here the derivation of the greatest common divisor of a sequence pair is extended to the Fibonacci polynomials, the Morgan-Voyce polynomials, the Chebyshev polynomials, and more general polynomials from a problem of Schechter [1] .
Moreover, all of these polynomials have coefficients which lie along rising
diagonals of Pascal's triangle, and all of these polynomials satisfy
(um(x),
un(x))
= U(m n x(x) with suitable adjustment of subscripts.
1.
INTRODUCTION
The Morgan-Voyce polynomials in [2], [3], and [4] are defined by

BQ(x)
= 1, Bx(x)
Bn(x)
= bn.(x)
bn(x)
= xBn_1{x)
Bn(x)
= Bn.(x)
= x + 2; bQ(x)
= 1, b1(x)
= x + 1,
and
(1.1)
+ (1 +
+
+
x)Bn-!(x),
bn_(x),
bn(x).
It is easy to show that B_x(x)

= 0, and b_1(x) = 1. These mixed recurrences
could be solved for pure recurrences as each separately satisfies
(1.2)
= (x + 2)un+1(x)
un+2(x)
un(x),
with uQ = 1 and ul = x + 2, and uQ = 1 and u1 = x + 1, respectively.

If one lists these polynomials,
b0(x)
B0(x)
b1(x)
Bl(x)
b2(x)
Bz(x)
b3(x)
BAx)
=
=
=
=
=
=
=
=
1
1
x + 1
x + 2
x2 + 3x
x1 + kx
x3 + 5x2
x3 + 6x2
+ 1
+ 3
+ 6x + 1
+ 10^ 4- 4
Clearly, we see that the coefficients of this double sequence lie along the
rising diagonals of Pascal's triangle.
The Fibonacci polynomials are
(1.3)
f0(x)
= 0, f.ix)
= 1, fn+2{x)
= xfn+1(x)
and we list the first few of these polynomials:

fx(x)
fz(x)
f3(x)
fn(x)
= 1
= x
= x2 + 1
= x3 + 2x
501
fn(x),
DIVISIBILITY PROPERTIES OF POLYNOMIALS IN PASCAL'S TRIANGLE
502
= xh
= x5
= xe
= x7
f5(x)
fe(x)
f7(x)
f8(x)
3x2
kx2
5xh
6x5
+
+
+
+
[Dec.
+1
+ 3x
+ 6x2 + 1
+ lO^r3 + kx
Once again, we see that the coefficients lie along the rising diagonals of
Pascal's triangle.
It can be shown that [3], [4]
(1.4)
bn(x2)
= f2n
+1
xBn(x2)
= f2n
+2
(x)
(x),
and the fact that coefficients lie on the rising diagonals of Pascal's triangle follows from that property for the Fibonacci polynomials. The Fibonacci
polynomials obey
/+(*) = (x2 + 2)fn
(1.5)
+ 2(x)
fn(x),
which agrees with (1.2) when x is replaced by x throughout.

Next, we are interested in finding the greatest common divisor of a pair
of Fibonacci polynomials.
lh<lOK<m 1.1:
For F i b o n a c c i p o l y n o m i a l s ,
(fm(x)>
Vtioofa:
fnW)
Rewrite the recursion (1.3) for the Fibonacci polynomials,

fm + 1(x)
and set (fm(x),

_ (x)
- xfm(x)
= d(x) .
fm + 1(x))
must have d(x) \fm_ (x).

d(x)\f
(#)-
= f(m,n)
fm_(x)9
Then, since d(x)\fm(x)
In turn, fm(x)
and, continuing, finally d(x)\f1(x)
and d(x) \fm + 1 (x) , we

= fm_2(x)
- xfm_1(x)
implies that
= 1. Therefore, d(x)
= 1,
and Theorem 1.1 holds for n = m + 1, or,

(1-6)
(/(*>> / n + i ^ = 1From [5], we also have
(1.7)
/p +r(a0 = fP-i(x)fr(x)
+ / p (a0/ r+ 1 ( * ) ,
and
(1.8)
/ m ( # ) ! / (#)
i f
an
d only i f 777 | n .
Next, let c = (m, ri) , and let c?(^) = (fm(x),

by (1.8), / e (a;) |/m(ar) and fa{x)\fn(x)
Since c\m and c\n,
fn(x)).
implies that fa(x)\d(x).
Since c = (rn,
ri) , by the Euclidean algorithm, there exist integers a and 2? such that c =
3??7 + bn. Since c <_ m, m9 n > 0, a 0 or b 0. Suppose a 0 and let & = -a.
Then bn = c + km applied to (1.7) gives
fhn (*) = fa
By ( 1 . 8 ) , fn(x)\fbn(x)
we have d(x) \fc (x)fkm
+ km(x)
= fc.1(x)fkm(x)
and fm (x)\fkm
+ 1(x)
+ fo(x)fkm
1(x).
(x), and s i n c e d(ff) | / n (x) and <2(#) | / m (#) ,
. But (/fc (a?) , / few + 1 ( a 0 ) = 1 by ( 1 . 6 ) , w h i c h
implies
1978]
that (d(x) , fkm
= 1, and d(x)\fa(x).
+ 1(x))
or (fm (x) , fn(x))
fa(x),
Also, since fc(x)\d(x),
d(x)
= / ( m > n ) (x) , concluding the proof, which is similar
to that by Michael [6] for Fibonacci numbers.

2.
503
Also see [7] and [8].
POLYNOMIALS FROM A PROBLEM BY SCHECHTER
Next, we consider some polynomials arising from a problem by Schechter

[1] and their relationships to the Fibonacci polynomials and the Morgan-Voyce
polynomials. Consider the sequence defined by Sl = 1, S2 = m, and
Sk = mSk_1
+ Sk_2,
k even,
Sk = nSk_x + Sk_2, k odd.

!
We now list the first few polynomials in m and ft, and compare to the MorganVoyce polynomials.
Sl(m,
n)
S2(m,
ri)
S3(m,
= bQ(mn)
= m
= mB0(mn)
n) = mn + 1
Sh(jn9 ri) = m(mn + 2)

S5(m,
S6(m,
ri)
= (mn)
ri)
= mBl(mn)
+ 3mn + 1
= m[(mn)
b1(mn)
b2(mn)
+ kmn + 3] = mB2(mn)
Thus, it appears that

S
s
2k + 2 (jn, n) = mBk(mn) ,
2k + i (m>
bk(mn).
Now, from (1.4), we have mnBk(m2n2)

(2.3)
S2k^2(m2,
n2)
= f2k+2(mn)
= m2Bk(m2n2)
= % f2k+2
For example, Sh(m2,

n2) = m2 (m2n2 + 2), Bl (m2n2)
2777ft, and we see that
Sk(m2,
n2)
= m2B1(m2n2)
thus,
(mn).
= m2n2 + 2, and fh (mn) = (mn) 3. +
= ^(mn) (m2n2
+ 2)
Next, we state and prove a matrix theorem in order to derive further results for the polynomials Sk(m, ri).
Th2.0A.rn 2.1:
(AB)k
Let A = (j
/ bk(xy)
(l
o) *
Then
'
xBk_1(xy)\
=
\yBk_1(xy)
where bk(x)
o ) '5
and Bk(x)
bk_(xy)
are the Morgan-Voyce polynomials.
504
{AB)
-\o
l
(AB) =/^
Assume that (AB)k
xybk(xy)
i
XyB.^xy)
, i
b_x(xy)
/ b Axy)
+ l *\ = f ^
+ bk(xy)
,
xBn(xy)\
^ M
has the form of the theorem.
+ xyBk_1(xy)
ybk(xy)
1/
[Dec,
x[(xy
+ yBk_1(xy)
Then,
+ l^.iGrz/)
xyBk_x(xy)
fcfc_i0c7/)r
+ fck-1(a:z/) ,
fcfc-iO*^) xBk{xy)
yBk(xy)
bk(xy)
by applying the mixed recurrences of (1.1), completing a proof by induction.
Now, returning to the matrices of Theorem 2.1, since the determinant of
AB is 1, it follows that
(2.4)
- xyB2k_1
bk(xy)bk_1(xy)
= 1.
Returning to the polynomials Sk(m, ri) , we have also that

fc
045)* =
71
mS2k
$2k + l
"
777
^zk -1 .
so that, taking determinants,

(2.5)
2k-iS2k
+ i " w^2k
1B
The polynomials 5^ (m, n) sire related to the Morgan-Voyce polynomials by

$2k + i (m9 n) = bk(mn) ,
(2.6)
(^2fc( w
S2k(m,
= nBk_1(mn)
n) -
mBk_1(mn).
Since the polynomials Sk(m, n), the Morgan-Voyce polynomials, and the
Fibonacci polynomials are interrelated by (1.4) and (2.3), which can be rewritten as
Sik + i(m> n) = f2k + 1Umn),
(2.7)
;
171
S2k(m9
n) = -7=/ 9 . (vmn)
1978]
505
and since the coefficients of the Fibonacci polynomials lie along the rising
diagonals of Pascal's triangle, we can write the following theorem.
ThdOtim 2.2: The coefficients of fk(x),
bk(x) , Bk (x) , and Sk(m, ri) are
all coefficients which lie along the rising diagonals of PascalTs triangle.
3.
Using the relationships of 2, we can expand upon Theorem 1.1 to write a

greatest common divisor property for Morgan-Voyce polynomials.
Th&Ofiem 3.1:
(i)
For t h e Morgan-Voyce p o l y n o m i a l s bn(x)
(Bm(x),
Bn(x))
D
(ii)
and Bn(x)
(m + 1, n + l ) - l
(bm(x)9
bn(x))
= b{(Zm
(iii)
(Bm(x)9
bn(x))
= b ((2m + 2j
Vnoo{\
(i)
x(Bm(x2),
Bn(x2))
+ i, 2 n + l ) - l ) / 2 (x) ,
2n + 1)
_l)/2
= (f2m+2(x)9
= CCB(m +
f2n
(x).
+ 2(x))
n+l)-l(%Z)
lt
= f2(ffl + 1 > n + 1 ) (*)
by a p p l y i n g ( 1 . 4 ) , Theorem 1 . 1 , and r e t u r n i n g t o ( 1 . 4 ) . For x ^ 0 , ( i ) i s

immediate by r e p l a c i n g x2 w i t h x a f t e r d i v i d i n g b o t h s i d e s by x.
If x = 0,
Bn = n + 1 , making ( i ) become (777 + 1 , n + 1) = (m + 1 , n + 1) - 1 + 1.
Applying ( 1 . 4 ) and Theorem 1.1 t o ( i i ) ,
(> m (x 2 ), bn(x2))
= (f2m
+ 1(ar),
/2n
+ 1(^))
f(2m + l, Zn + 1) (X) = Afc + l ^

since the greatest common divisor of 2m + 1 and 2n + 1 is odd.
Thus,
(M* ), M**)) = bk(x )

by (1.4), where 2k + 1 = ( 2 ^ + 1 , 2n + 1 ) , so that
fc = ((2m + 1, 2n + 1) - l)/2.
Replacing x 2 by x yields (ii).
Finally, we observe that bn(0)
and Theorem 1.1:
(Bm(x2),
Next, s e t
(2m + 2 ,
bn(x2))
= (xBm(x2)9
and again use (1.4)
bn(x2))
2n + 1) = 2/c + 1 , s i n c e i t must be odd, and

2
(S m (x ), M * 2 ) ) = f2k
where
= 1, so that x\bn (x),
+ 1(x)
= bk(x2)
fe = ((2m + 2 , 2n + 1) - l ) / 2 .
2
Replacing x by # establishes (iii), finishing the proof of Theorem 3.1.

Returning to the polynomials Sk(m, ri), and using (2.7) with Theorem 1.1,
gives us
IkdOKQjm 3.2:
(Si(m9 ri), Sj(m,
ri))
= (i,j) (m9 ri).
VfiOOî If i and j are both odd, (2,7) and Theorem 1.1 give the above result immediately. If i, and j are both even,
506

(Si(m9
n ) , Sj(m,
n))
= (S2k(m,n),
S2h (m,
lK
\vmn
n))
Vmn
= Vmn
-p=(f2k2K (i/?)> f2h
(Sb)
zn
= Si{k,h)
1 so t h a t Vmn)(S2k
{SiQn,
n),
+1
Qn9 n),
S3-(m9 n))
and s i n c e f2k
= (S2k
= (S2k
(m, n) = S(2k,2h)
I f i i s odd and j i s e v e n , s i n c e S2k

+ 1(m,
+ 1(m,
+ 1(m,
+1
()
(m, n) = ( i , j ) ( m ,
n).
a l s o ends i n 1 ,
S2h (m,
n),
/mnS2h(m,
m
~F=f2{kZKK>n) h^ff)
Vmn
n) always ends i n t h e c o n s t a n t
n),
= (/2fe + i ^ v / ^ " ) '
[Dec.
n))
f2yl^m<^))
n))
=
(f2k
+ 1(i/mn)
>
f2h(Jrnn))
/ ( 2 H i , 2h) ^^ = s(2k + i, 2/2) (m9 n) = SU}j)(m,

n) ,
where we can again use (2.7) because (2/c + l, 2/z) is odd, concluding the proof
of Theorem 3.2.
We quickly have divisibility properties for the polynomials Sk(m, n) .
Ikdonm
3.3:
Vnoo^\
If i\j,
Sîm,
n)\Sj(m,
n) if and only if
i\j.
then (i, j) = i, and 6^0??, n) l^-(777, n) by Theorem 3.2. If
S^(m, n)\Sj(m,
n) with i )(j, then /^(^)|/. (#) where i^j,
a contradiction of
J
(1.8).
From all of this, we can also write divisibility properties for MorganVoyce polynomials.
TkzotiQM 3.4:
VKOO^:
For the Morgan-Voyce polynomials,
Bm(x)\Bn(x)
if and only if (jn + 1) | (n + 1) ;
bm(x)\bn(x)
if and only if (2m + 1)|(2n + 1 ) ;
bm(x)\Bn(x)
if and only if (2m + 1)|(n + 1).
Bm(x)\Bn(x)
if and only if (Bm(x),
(Bm(x),
Bn(x))
Bn(x))
= Bm(x),
but
= B(
m+15rc+
1) - 1
by Theorem 3.1. Setting the subscripts equal, m = (m + 1, n + 1) - 1, or,

777 + 1 = (TT? + 1, n + 1 ) , which forces (m + 1) | (n + 1) . The case for bm(x) and
bn(x)
is entirely similar.
In the case of bm(x) and Bn(x),
Bn(x)
cannot divide bm(x) for n > 0 because bm(x) always ends in the constant 1, while the constant for Bn(x) is
greater than 1, n > 0. Since bm(x)\Bn(x)
if and only if
(bm(x), n(a;)) = bm(x),
and since
(ZÔ^), Bn(x))
= ^((2n+2, 2m + l)-l)/2^)
by carefully rearranging (iii) in Theorem 3.1, equating the subscripts leads
t0
777 = ((2n + 2 , 2777 + 1) - l ) / 2 ,
or
2777 + 1 = (2777 + 1 , In + 2 ) .
1978]
507
Thus, (2m + 1)I(2n + 2 ) , but since (2m + 1) is odd, we must have

(2m + 1) | (n + 1 ) ,
concluding the proof.
Returning to the greatest common divisor property of the Fibonacci polynomials, (fm(x) , fn(x))
= f(mtn)(x)9
we make some observations from Theorem
3.1(i) regarding the Morgan-Voyce polynomials Bn(x).
From
(Bn(x),
= B ( n + 1, m + 1) -
Bm(x))
it would follow that if B*(x)

(3.1)
(B*(x),
which sequence {B*(x)}

(3.2)
and B*(x)
= Bn^1(x)
= Bm_1(x),
then
= Bfn>a)
B*(x))
= { 0 , 1, x -f 2, ...} obeys
B*(*) = (x + 2)*_x(x) - B*_2(ar)
and is in fact the Fibonacci polynomial, so to speak, for the auxiliary polynomial A2 - (x + 2)X + 1 = 0, since
where Aj and A 2 are the roots.

U n = XU n _ ] _
But (3.2) can also be expressed as
Un _ 2
where x is replaced by (x + 2). Thus one set of polynomials with coefficients

on diagonals of Pascal's triangle transforms into another set with the same
property.
This property of transforming one set of polynomials whose coefficients
are on diagonals of Pascal's triangle to another set of polynomials with coefficients also on diagonals of Pascal's triangle is shared by the Chebyshev
polynomials {Tn(x)}
[9] of the first kind, defined by TQ(x) = 1, Tx(x)
= x,
and
(3.3)
since
(3.4)
Tn + 1(x)
= 2xTn(x)
= Tm(Tn(x))
Tn{Tm(x))
Tn_(x),
=
Tmn(x).
The property (3.4) is easy to prove from the Binet form associated with the
auxiliary polynomial
(3.5)
A2 - 2x\ + 1 = 0,
with roots Ax and A2.

The Chebyshev polynomials {Un (x)}
Ul(x) = 2x,
(3.6)
Un + 1(x)
= 2xUn(x)
of the second kind are UQ(x)
= 1, and
Un_(x).
First, to establish (3.4), we prove by induction that

(3.7)
Ax = Tn(x)
+ /(x2
xl
= Tn(x)
/(x2
- lJUnîix)
- !)/_! ( a ) .
We prove only one part, since the second part is entirely similar.
/-i(aO = 0, and TQ(x)
= 1, A^ = Tn(x)
+ /(x
- l)Un_1(x)
for n = 0.
Since,
Assume
508
[Dec.

/ O r 2 - )Uk_1(x)
t h a t Ai = Tk(x) +
Then, by ( 3 . 5 ) ,
Ai + 2 = 2xXk1+1 - x\ = (2xTk + 1(x)

= Tk+2(x)
and
x\
- Tk(x))
/ ( x 2 - l)Uk
+1
= Tk + 1(x)
/(x2
+
/ \ x 2 - 1) U
- 1) (2xUk + 1(x)
(x).
Uk(x))
1(x),
using (3.4) and (3.6), establishing the form of Xl in (3.7) by mathematical

induction.
Notice that, since A X A 2 = 1, by multiplying the forms of A^ and A2 from
(3.7), we can derive
(3.8)
T2(x)
- 1 = (ar2 - l)^_i(x).
Also, by adding in (3.7), we can establish

(3.9)
Now, Xl(x)
Tn(x)
= (Ar+
= x + /x^ - 1.
Xn2)/2.
Replace x by Tm(x)
and the root becomes
A^Gc)) = Tm(x) + A2(^) - 1,

satisfying the auxiliary polynomial (3.5), so that
X2(Tm(x))
That is,
But X1XZ
+ 1 = 0.
- 2Tm(x)X1{Tm(x))
A2 (Tm (x) ) + 1
Tm(x) =
= 1, so
Tm(x)
2X1(Tm(x))
[A
i^-(x))
+1AI(^W)]/2.
= [ A ^ f r ) ) + A2(Tm(x))]/2.
Referring back to (3.9), we write

Ax = Xml{Tm{x))
and
A* = A 2 ( T m ( * ) ) .
Now,
^ n W
= U 7 n + A 2 n ] / 2 = [(A?) n + ( A * ) n ] / 2 = [Xi(T w (a?)) +
Xn2(Tm(x))]/2,
so that Tmn(x) = Tn(Tm(x))

and similarly, Tmn(x) = T m (T n (x)), finishing the
proof of (3.4).
Returning to divisibility properties, observe that the Chebyshev polynomials of the second kind are the polynomials with the Fibonacci-like property
xl-
x\
where Ax and A2 are the roots of A2 - 2xX + 1 = 0. We now list the first few
polynomials and let
U*ix) = / n - i ( x )
U.l{x)
= 0
= U*{x)
U0(x)
= 1
= U*(x)
Ux(x)
= 2x
U2(x)
= hx
= U*(x)
- 1
U3(x)
= 8x
Uk(x)
= 16a: * - 12a;:2 + 1
- 4x = 4 x ( 2 x
= U*(x)
2
-- 1) = U*(x)
= U*(x)
1978]

U5(x)
= 32;c5 - 32^ 3 + 6x = 2x(Sxh
UAx)
= 64*6 - 80* 4 + 24* 2 - 1
509
- 8x2 + 3) = U%(x)
= U*(x)
It would appear that

(3.10)
U*(x)9
U*(x)
U*mtn)(x).
That this Is indeed the case can be established very simply.

satisfies
U*+1(x)
= 2xU*(x)
{[/ n (x)} i s a s p e c i a l c a s e of
Hoggatt and Long [7] as
(3.11)
Un+2(x,
(3.12)
- C^.!^),
t h e p o l y n o m i a l sequence {Un (x,
y) = xUn + 1(x9
where U0(x9 y) = 0 and U1{x9

x = 2x and y = -1. Since
Since U*(x)
y) + yUn(x9
z/) = 1.
d e f i n e d by
y),
Note that {U*(x)}
( M * , 2/), M * . 2/)) = U(min)(x,
y)}
is the special case
y)9
we see that (3.10) is immediate.

We summarize as
Tk&OSl&n 3.4:
By suitable shifting of subscripts in the original definitions, the Fibonacci Polynomials, the Morgan-Voyce polynomials Bn(x),
the
Chebyshev polynomials Un (x) , and the polynomials S]<(m9 n) all satisfy
(um9
= U(m> n ) .
un)
A MORE GENERAL POLYNOMIAL SEQUENCE
k.
Define Sk(a9
b9 c9 d) by taking Sl
^ = aSk_1
+ bSk_l9
= 1, S2 = a,
k even,
(4.1)
fc = c5fc_! + dSk_29
k odd.
Let S* = 1, S* = c, and define 5*(a, 2?, c, d) by taking

S* = <?* _ + dS* _ 2 , /c even,
(4.2)
S* = aS*
+ ^ * _ 2 , fe odd.
Let K0 = 0 , K1 = 1 , Z n = (ae + 2) + d)Kn_1

T
_ ^
Let
/#
&\
/<?
^\
( ac + b
li o j ' d o ) i
Now, {Kn}
^2fc + i
^2*
S*2k
dS2k.J
c
\
is the "Fibonacci sequence,"

Xx - X2
7/
Xx - A 2 ?
ad\
_,
dp
then
bdKn_2.
/ %-k + i ~ d%k
\oKk
daKk
d(Kk -
bKk_x)
510
[Dec.
for the quadratic X2 - (ac + b + d)x + bd = 0, with roots X19 X2.

Applying
results [7] for {Un(x9 y)} from (3.11) and (3.12) to {Kn}, we have immediately that
To continue, we write the first few terms of {Sk(a9
b,
c9
d)}.
S, = 1
52 = a
53 = ac + d
Sh = a2c + ad + ab
55 = a2c2
+ 2acd + abo + d2
56 = a3c2
+ 2a2cd + 2a2bo
7 = a3o3
+ 3a 2 c 2 J + 2a2bo2
+ a^2 + abd + a 2
+ 3aad2 + 2a&c^ + ab2o + ^ 3
We consider some special cases. If a - 0, then S2k + z = 0> and S2k + i ~ dk9
k >. 0. If fc= 0, ^2^ + 2= a(ac + d ) k and S2k + 1 = (ac + d)k, & >, 0. If c = 0,
then 5 2 k _! = c^-i and 2 k =a[(dk - bk)/ (d - b)]9 k > 1. If d = 0, then S2k =
a(ae + b)k~1 and /S^k + i = ao(aa + b)*"19
k >_1. The expansions of *(a, b9
c9 d) are not very interesting, since they are the same as those of Sk(a9
b9
c9 d) with the roles of a and c exchanged.
The special case of Sk(a9 b9 c9 d) where b = d proves fruitful. We list
the first few terms of {Sk(a9
b9 c)} below:
51 = 1
52 = a
53 = ac + b
S^ = a2c + lab
55 = a2c2
+ 3abc + b2
56 = a3c2
+ ha2bc
+ 3ab2 = a(ac
+ b) (ac + 32?) = S2S3(ao
+ 3b)
We are interested in the case b = d, or, taking Sk(a9 b, c) and S^(a,

b,
c) , so that 3 will divide S6.
It is not difficult to prove by induction
that
(4.3)
(4.4)
It
(4.5)
2k + j
^2^ +i +j
j + lS2k
= $*k+i
We now prove Sj\Sjm
bSjS2k-l>
Sj + iS 2k + i
i s not hard to see

$2k+i
+
+
bS^S2k'
that
a n d
aS
2k
cS
*k-
for j odd and m odd, or, jm = 2k + 1.
Sj(m + 1)
Sj + lSjm
bS^S
2k
= Sj
From (4.4),
iSjm + Z?^- 5 27c J
since 5j = S^ for j odd. So, if Sj \Sj and 6^ l^-^, then Sj \Sj(m + i) for j odd.
Thus, for j and m both odd, we see that Sj\Sjm
for all odd m.
Next, suppose that j is odd and m is even; then, from (4.3),
S
2m'j
- Sf+1S2mlj
+ bS-S^.j^,
m =
2m'.
1978]
Now, i f SjlSj
and Sj\S2mlj9
Next, l e t j be e v e n ;
S
= s
2k+2j
t h e n Sj\S(2ml+1)j
*j' + iszk
+ bS2j'S2k-i
511
Sj(m+1).
and
Ik =
2j'm.
Since S-\S2-,
and Sj\S2{jtm
= S2k , we have Sj \S 2j m + zj' ~ -Sj'o + i) T h i s comp l e t e s t h e proof t h a t i f i\j9
t h e n S^Sj,
S i n c e , a l g e b r a i c a l l y , {Si} a r e of
i n c r e a s i n g d e g r e e i n t h e two v a r i a b l e s a and c c o l l e c t i v e l y , SJ-)(Si f o r i < j .
L a s t , u s i n g ( 4 . 3 ) and ( 4 . 4 ) , i t i s now s t r a i g h t f o r w a r d t o show
ThzoKQjm 4 A:
Sj(a9
b> o)\Si(a9
b9 c) i f and o n l y i f j \ i .
We can a l s o now prove

Jh(LOK.m 4.2:
(Si(a9
b9 c),
Sj(a9
b9 c))
= S^t3-)(a9
b9
c).
VK.00^: Let P(x) be a monic polynomial of degree v + s with integral coefficients with two factors Q(x) and R(x) of degree v and s9 respectively.
Then,
br
+3
P(x/b)
P*(x9
brQ(x/b)bsR(x/b)
b) = Q*(x9 b)R*(x9
b).
In p a r t i c u l a r , i f P(x) i s of d e g r e e p , T(x)
(P(x)9
T(x)) = J/(ar), t h e n
{bpP(x/b)9
btrP{xlb))
of degree t , W(x) of d e g r e e w, and
bwW(x/b).
For a p p l i c a t i o n t o Theorem 4 . 2 :
(4.6)
c2S2m{a2,
(4-7).
S 2 r a + 1 ( a 2 , &2, c 2 ) =
Co6 7:
b2,
a2)
= aci2"-Y2BI(ac/Z>) ;
Both s u b s c r i p t s
2
(o SZm(a ,
fc2ffl/2m+1(ac/&).
even.
2
b,
o2S2n(a2,
a ),
b2,
= {acb2m-xfZm(,aclb),
a2))
aob2n-lfln{ae/b))
(2m, 2n) - 1 .
acb^m>^^f{2my2n)iac/b)
=
2
= c S(2mi
(a2, 2,
2n)
c2).
Therefore,
(S2m(a2,
Ca6 2 :
b2,
a2),
S2n(a2,
Both s u b s c r i p t s
2
b2,
S2n
2
+ 1(a ,
= (b2mf2m+1(ac/b),
Ca6 3 :
= S(2n>
b2,
a2))
odd.
2
( S 2 m + 1 ( a , Z? , a ),
c2))
b2nf2n+l(aa/b))
$ (2m + l, 2n + l) ($
s D , <3 ) .
One s u b s c r i p t odd, one s u b s c r i p t

2
(c S2m(a
c )9
S2n+1(a
h(2m,
2n + l)-1
(2m, Zn+1)
n
i.CC
b , c2))
b2nf2n
= (acb^-^Jac/b),
(rr^/h^
b
O )
even.
1(ao/b))
2B)
(a2, b2, e2).
512
since (ao9 b) = 1. Also, since (o2, S2n+1)

2
(c S2m(a ,
b,
c )9
= (S2m(a2,
b2,
c2)9
[Dec.
= 1,
S2n+1(a ,
b,
o2))
S2n+1(a2,
b2,
o2))
= ^(2/n, 2n+l) (& , & 2 , C2) ,
finishing the proof of Theorem 4.2 by replacing a2 with a, Z?2 with b9 and <?
with e.
Let f*(x) be a modified Fibonacci polynomial, with
fn to) = / to), n odd,
f2(x)
, n even.
Listing the first few values,

/*(*) =.1
f^W = 1
f*s(x)
= x2 + 1
f*(x)
= x2 + 2
/*(#) = ^ + 3jr2 + 1
f%(x)
+ kx2 + 3
= ^
/ * t o ) = ^ 6 + 5X4 + 6x 2 + 1
/*(a?) = x 6 + 6 ^ + 10;c2 + 4 .
Here,
=
fn + 2 <*>
f*+l
^)
fn^
'
even
'
/* +2 to) = * /* + 1 (x) + / ; t o ) , n odd.

This is {Sk(a,
b, o, d)} with a = b = d = 1, o - x2.
Thus, by Theorem 4.2,
(/*(*), /*(x)) = / f m > n ) ( x ) .

Let Vk(x)
be a modified Morgan-Voyce polynomial defined by

v2n+2(x)
= Bn(x),
v2n+1(x)
The first few values for {vk(x)}
bn(x).
are
vi to) = l
y 2 to) = l
z;3to) = x + 1
b0(x)
B0(x)
bx(x)
7^ to) = a; + 3
= Bito)
v5(x)
= x2 + 3x +
y6to) = x
2
3
y 7 t o ) = a:
vQ(x)
= x
= 2? 2 to)
+ 4x + 3
+ 5x
+ 6x
= S2to)
+ 6^ + 1
b3(x)
+ lOx + 4 = to + 2) to + 4^ + 2) =
B3(x)
1978]
THE GOLDEN SECTION IN THE EARLIEST NOTATED WESTERN MUSIC
Since Vk(x)
this is {Sk(a,
513
satisfies
( vn(x)
= vn_1{x)
( vn(x)
= xvn_(x)
n even,
+ vn_2(x),
+ vn_2(x),
n odd,
b, c, d)} with a = b = d = 1 and c = x*

(vn (x),
vm(x))
Then, by Theorem 4.2,
= v
(m,n)
REFERENCES
1.
2.
3.
4.
5.
6.
7.
8.
9.
Martin Schechter, Problem H-305, The Fibonacci, Quarterly

(to appear).
A. M. Morgan-Voyce, "Ladder Network Analysis Using Fibonacci Numbers,"
Proceedings
of the IRE3 IRE Transactions
on Circuit
Theory,
Sept., 1959,
pp. 321-322.
Richard A. Hayes, "Fibonacci and Lucas Polynomials" (Unpublished Master * s
Thesis, San Jose State University, January 1965).
V. E. Hoggatt, Jr., & Marjorie Bicknell, "A Primer for the Fibonacci NumbersPart XIV: The Morgan-Voyce Polynomials," The Fibonacci
Quarterly,
Vol. 12, No. 2 (April 1974), pp. 147-156.
Marjorie Bicknell, "A Primer for the Fibonacci NumbersPart VII: An Introduction to Fibonacci Polynomials and Their Divisibility Properties,"
The Fibonacci
Quarterly,
Vol. 8, No. 4 (October 1970), pp. 407-420.
Glenn Michael, "A New Proof for an Old Property," The Fibonacci
Quarterly,
Vol. 2, No. 1 (February 1964), pp. 57-58.
Verner E. Hoggatt, Jr. , & Calvin T. Long, "Divisibility Properties of Generalized Fibonacci Polynomials," The Fibonacci
Quarterly,
Vol. 12, No. 2
(April 1974), pp. 113-120.
W. A. Webb & E. A. Parberry, "Divisibility Properties of Fibonacci Polynomials," The Fibonacci
Quarterly,
Vol. 7, No. 5 (December 1969), pp. 457463.
Cornelius Lanczos, "Tables of Chebyshev Polynomials," U.S. Department of
Commerce, National Bureau of Standards, AMS-9, December 19, 1952.
*****
THE GOLDEN SECTION IN THE EARLIEST NOTATED WESTERN MUSIC

PAUL LARSON
Temple
University,
Philadelphia,
PA 19122
The persistent use of the golden section as a proportion in Western Art

is well recognized. Architecture, the visual arts, sculpture, drama, and poetry provide examples of its use from ancient Greece to the present day. No
similar persistence has been established in music. One possible reason is that
what ancient Greek music has survived is of such a fragmentary nature that it
is not possible to make reliable musical deductions from it. However, beginning with the early Middle Ages a large body of music has survived in manuscripts that from ca. 10th century can be read and the music can be performed.
This body of music is known as Roman liturgical chant or, more commonly, as
Gregorian chant. These chants have not previously been analyzed from the
standpoint of the golden section. Acknowledging the probability of the pres-
514
THE GOLDEN SECTION
IN THE EARLIEST NOTATED WESTERN MUSIC
[Dec.
ence of a number of structural designs and proportions in these chants, it is

the author's intention to establish the musical use of the golden section as
an organizing principle in them.
The official collection of Roman liturgical chant is the Libev
Usualis.
The chants selected for the present study are "Kyrie" chants of which there
are 30 in the collection. The chants span at least 600 years, having been
written beginning with the 10th century.
The basic structure of a "Kyrie" is determined by the text, as shown in
Diagram 1:
Diagram 1
un 11
Kyrie
Kyrie
Kyrie
un 11
Christe
Christe
Christe
un 11
Kyrie
Kyrie
Kyrie
eleison.~\
e'leison.
eleison.
sect ion
eleison.
eleison.
eleison.
total chant
eleison.
eleison.
eleison.
Each chant falls into nine separate sections. The three repetitions of the
sections form three larger units which, in turn, make up the complete chant.
While there is considerable variety in the melodic treatment of the text, the
text itself had remained constant in the above form since ca. 900.
The actual nature of the rhythm of these chants is still open to question.
Because music is a time art, any analysis that does not account for the proportional movement of the pitches in time cannot pretend to be a statement
about the total nature of the music. In this sense, the following findings,
though factual, remain theoretical to the degree that while pitches in succession imply time, exact temporal proportions are not deducible from that
succession alone. In addition, the reader should be advised that there are
more than 200 "Kyrie" melodies known to exist.
In this light, the chants
analyzed for this study represent a sampling of the repertory.
METHOD OF ANALYSIS
Because different treatments of the same text are usually set to different pitches, 146 distinct musical sections are present in the 30 chants, the
remaining being exact repetitions of other sections. The pitches in each section were totaled, and <f> was determined for each section. A section was examined to determine if any significant musical event occurred at either the
major or minor mean. A significant event was defined as the beginning or ending of a musical phrase. The three statements of the "Kyrie," the "Christe,"
and the "Kyrie" tend to form larger units; these were analyzed according to
the same procedure. Finally, the pitches in the complete chant were totaled,
i.e., nine separate sections of text.
THE FINDINGS
Applying the analytical method described above revealed the presence of
the golden section in 105 of the 146 individual sections of the "Kyries" in
the Liber
Usualis.
These 105 sections make up .72 of the cases. The major
1978]
ON FIBONACCI NUMBERS WHICH ARE POWERS
515
mean precedes the minor mean twice as often as the minor mean precedes the
major mean. Example 1 is a section of chant conforming to the M:m proportion,
Ky-n - e
|e_j_son
Example 1*
Example 2 shows the proportion in reverse.
l^y-* -* ' *-* *,* " *j " a , + ^

C h r i s - te
e-
le-i-son
Example 2*
Twenty-one sections have phrase divisions occurring at the arithmetic mean.

The same method was applied to the next larger formal unit, i.e., the
three repetitions of each exclamation. In 30 chants there are 90 such units.
<t> is found in 53 (.59) of these units. Where the musical phrase either falls
short of the exact mean or extends beyond it, a tolerance of .02 of the total number of pitches was maintained in defining the unit as a golden section.
A performance of an entire chant includes nine sections as shown in Diagram 1. An analysis of the 30 chants revealed that 20 (.66) exhibit the
golden section proportion. In more than half of the cases, the mean occurs
at the end of the first or at the beginning of the second "Christe eleison."
CONCLUSION
At this stage, these findings tend to establish
den section in one of the earliest notated forms of
"Kyrie" chants. To establish the presence of the
other than the "Kyrie," requires further analysis
Gregorian chant.
the presence of the golWestern music, i.e., the

golden section in chants
of the general body of
0N FIBONACCI NUMBERS WHICH ARE POWERS

NEVILLE R0BBSNS
of San Francisco,
San Francisco,
University
CA 94117
INTRODUCTION
Let F(n), L(n) denote the nth Fibonacci and Lucas numbers, respectively.
(This slightly unconventional notation is used to avoid the need for secondorder subscripts.) Consider the equation
(0)
F(m)
"Source:
Liber
GP,
Usualis
(Desclee & Co., Tournai [Belb.], 1953), p. 25.
516
[Dec.
where p is prime and m > 2, so that c > 1. (The restriction on m eliminates

from consideration the trivial solutions which arise because cp=c
if o = 1,
o = 0, or c - -1 and p is odd.)
The complete solution of (0) was given for p = 2 by J. H. E. Cohn [1] and
by 0. Wyler [4], and for p == 3 by H. London and R. Finkelstein [3]. In this
article, we consider (0) for p >_ 5. It follows from Theorem 1 that if a nontrivial solution exists, then one exists such that m is odd. In Theorem 2,
we give some necessary conditions for the existence of such a solution.
PRELIMINARIES
We will need the following definitions and formulas; r, s denote odd integers such that (p, s) = 1.
q,
V^ZyiObLon 7: If q is a prime, then z(q) is the Fibonacci entry point of

i.e., z(q) = min{m: q\F(jn)} .
V^l^wUtLoyi
2: If q is a prime, then y (q)
is the least prime divisor of
z(q).
If (x, y) = 1 and xy = zn, then a? = un and y = vn, where (u, v) = 1 and
wi> = z.
(1)
(2) F(2n) =
F(n)L(n).
/o\ /n/ \ r/\\ ( 2 if n = 0 (mod 3)
(3) (*(n), L(n)) = j x
^0 (mod3)
(4) F(n) = 2r -* n = 3 (mod 6) ++ L(n) = 4s.

(5) If (a:, y) = 1 < x, and xmy = zn, then n|/7?.
(6) F(n) = 2kr9
k > 1 -> k >_ 3, 3^2fc"2|n *-> L(n) = 2s.
(7)
2|F(w) -* 3|n.
(8)
3|F(n) <- 4|n.
(9)
(F(n), F(kn)/F(n))\k.
(10)
t odd -* (F(t), F(3t)/F(t))
(11)
> 0 -* F() < F(6t).
= 1.
(12) ^|F(w) -> s(q) Im.

(13) F(2n + 1) = F(n) 2 + F(n + I ) 2 .
(14)
c, n odd + cn = c (mod 8).
Re.maA.kA: (1) through (8) and (11) through (14) are elementary and/or wellknown; for proof of (9), see [2], Lemma 16; (10) follows from (8) and (9).
THE MAIN THEOREMS
For a given prime, p, let m = m(p) >2 be the least integer such that, by
assumption, (0) has a nontrivial solution. By inspection,
777(2) = 1 2
a n d 7??(3) = 6.
Tk&Ofim 7: If m = 2n > 2 is the least integer such that F(m) = cv, where
p is prime, then either (i) m = 6, p = 3, or (ii) m = 12, p = 2.
Cctt>& J I f n 0 (mod 3), then by hypothesis, (1), (2), and (3), we have
1978]
517
F(n) = bp.
If b > 1, we have a contradiction, since n < m. If b = 1, then
hypothesis ->n = 2->m = 4-> F(m) = 3, a contradiction.
C<UZ 2 I f ft = 3 (mod 6 ) , then (4) -> F(n) = 2p, L(n) = 4s, with rs odd.
Now hypothesis and (2) -> F(/77) = 8PS = cp, so that (5) - p 13 -> p = 3. By [3] ,
we must have
c=2,n=3,m=6.
Co6e 3-If n = 0 (mod 6 ) , let n = n0 = 2J3fet, where j , fc > 1 and (6, ) =
1. Let m = 2~Vz0 for each i such that 1 <_ i <_ j . Let fr0 = rij = 3 k t, and let
hi = 3~^/z0 for each i such that 1 <_ i <_ k, so that t = hk.
By (6), we have
F(n) = 22+^r, L(n) = 2s, where rs is odd and (p, s) = 1. Now hypothesis, (1),
and (2) imply r = r%9 s = SQ, with r0s0 odd and (P 0 , s 0 ) = 1. Therefore, F(n)
= F(n0)
= 22+JTQ,
L{n)
= L(nQ)
= 2s p ,
ÔSO
'
Since ni
= 2n^ + i, we may re-
peat our reasoning to obtain F(m) = F(rii + 1)L(ni + 1) = 2 Z + J'"2'p?, L(ft^) = 2s|
for i = 0, 1, 2, ..., j - 1. By (4) we have F(hQ) = F(rij) = 2r?, L(n3-) = 4sJ;
moreover, r^s^ = 2^ _ x is odd and (r^ , s^) = 1 for i = 1, 2, 3, ..., J. Now,
let rj = Mo, so that F(7z0) = 2up. We have F(hi-i)
= F{hi)^F{hi.{)/F{hi)
for
i = 1, 2, 3, . . . , k.
By (7), (10), and (1), if i < k, we have F(hi) = 2u\,
F{hi-i)lFQii)
= v\\ if i = k, we have F(t) = F(hk) = up, F(hk_x) /F(hk)
= 2^;
moreover, (ui, Vi) = 1 and uÔi = u^_i is odd for i = 1, 2, 3, ..., fc.
But (11) + F(t) < F(6t) <. F(n) < F(m) = c p - * u k = l - * t = l .
If k >_ 2,
then F(hk-1)lF(hk_1)
= F(9)/F(3)
= 17 = v_ -> p = 1, a contradiction. Hence,
k = 1, ft0 = nJ = 3- I f C ^ 2> t h e n Liô-2)
= (12) = 322 = 2s|_ 2 -* sf_ 2 =
161 -> p = 1, a contradiction. Therefore, k = j = 1, ft = 6, m = 12, p = 2.
Cofio&Lafiy: If (0) has a nontrivial solution for p _> 5, then it has a nontrivial solution such that m is odd.
VfLOO&i The proof follows directly from Theorem 1.
Tk&Oltm 2: If F(m) = cv > 1, where the prime p ^ 5, and m is odd, then
either (i) m = 1 (mod 12) and c = 1 (mod 8) or (ii) m = 5 (mod 12) and <? =
5 (mod 8; furthermore, if q is any prime factor of c, then 2/(q) >. 5, so that
q e {5, 13, 37, 73, 89, 97, 113, 149, 157, . . . } .
Vnooi'. If 2\c9 then 2 P | op -> 2 P |F(m) , so that by (6), 3*2p~2\m,
contradicting hypothesis. Now c is odd, so that F (m) is odd, and by (7), 3)(m. Therefore, m E 1 or 5 (mod 12). If q is any prime factor of o, then
(12) + #(4) |g(q) |TTZ.
Since ( 6 , m) = 1, we must have y (q) >_ 5 .
Co6 ?If w = 1 2 t 1, then (13) -> F(w?) = F ( 6 t ) 2
F(6t) = 0 (mod 8) and F(6t 1) i s odd, so F(6t l ) 2
cp E 0 + 1 El (mod 8 ) , and (14) i m p l i e s c = 1 (mod
Co6e 2 I f m = 1 2 t 5 , t h e n (13) > F(TT?) = F ( 6 t
Now, F ( 6 t 3) = 2 (mod 8) and F(6t 2) i s odd, so
T h e r e f o r e , c p E 4 + l E 5 (mod 8 ) , and (14) i m p l i e s
+ F ( 6 t l ^ 2 = c p . Now,
E l (mod 8 ) . T h e r e f o r e ,
8).
3 ) 2 + F(6t 2 ) 2 = cp.
F(6t 2 ) 2 = 1 (mod 8 ) .
c = 5 (mod 8 ) .
REFERENCES
1.
2.
3.
4.
J. H. E. Cohn, "Square Fibonacci Numbers, Etc.," The Fibonacci

Quarterly,
Vol. 2, No. 2 (1964), pp. 109-113.
J. H. Halton, "On the Divisibility Properties of Fibonacci Numbers," The
Fibonacci
Quarterly,
Vol. 4, No. 3 (1966), pp. 217-240.
H. London & R. Finkelstein, "On Fibonacci and Lucas Numbers Which Are
Perfect Powers," The Fibonacci
Quarterly,
Vol. 7 (1969), pp. 476-481.
0. Wyler, American Math. Monthly,
Vol. 71 (1964), pp. 221-222.
PRIMES, POWERS, AND PARTITIONS

B. DE LA ROSA
University
of The Orange Free
State
Bloemfontein
9300, Republic
of South
Africa
Elementary arguments are employed in this paper to give a characterization of the set of primes and to extend this set to a larger one whose elements are defined by a single property: we show that a positive integer is
either a prime or a power of 2 if and only if such an integer cannot be expressed as a sum of at least three consecutive positive integers. This fact
provides an easy sieve to isolate the primes (and if one prefers, the immediately recognizable powers of 2) less than or equal to any preassigned positive
integer. We also describe the possible ways in which a given composite number may be expressed as a sum of at least three consecutive positive integers;
such representations for the sake of brevity shall be termed a-partitions of
the given integers. Furthermore, "number" shall mean "positive integer" and
the set of all these numbers will, as usual, be denoted by IN.
LzmmCL 1: An odd number m admits a a-partition if and only if m is a composite number.
VKOO{ (=): Let m = n + (n + 1) + + (n + ft), n e IN, ft >_ 2.
m = ^(2n
Then,
^).
If ft + 1 is even, then (ft + l)/2 is an odd number _> 3, since ft >_ 3 in this
case; obviously, thereforey In 4-ftis an odd number >_ 5. Hence, m is a composite number. If ft + 1 is odd, then ft is even, and since ft >. 2, one must
have that In + ft is an even number >_ 4. Since m is an odd number, it follows
that {In + ft)/2 is an odd number > 2. The fact that ft + 1 >. 3 now shows that
m is a composite number.
VKOO^ (<=): Consider an arbitrary factorization m = kg,
(3 <_ft<_ g).
m = kg = |(2<7) = ffifg - ^ - ^ ) + ft - l] = |(2a + ft - 1)

where
a = g
Hence, we have
E:
ê
m = / \ , (q + p - 1) ,
p = i
which i s a a - p a r t i t i o n ,
s i n c e a e IN and ft J>. 3 .
CoKolLoJtij 1 : An odd number i s prime i f and o n l y i f i t a d m i t s no a - p a r t i tion.

L&nma 1\ An even number m a d m i t s a a - p a r t i t i o n i f and only i f m i s n o t a
power of 2 .
(Cf. [ 1 ] , p . 1 7 . )
Vfioo^ (=) :
Let m = n + (n + 1) + + (n + ft); n e IN, ft >. 2 .
Then,
ft + l
, N
w = - (/ 20 n + 7ft).
If m = 2s , then, since ft >. 2, we must have that ft + 1 = 2t, _> 2, and In + ft
= 2", u >. 2. This is a contradiction, since ft 4- 1 = 2* would imply that ft is
518
Dec. 1978
odd, so that In + k would also be odd.

of 2.
519
Hence,, we have that m is not a power
VKOO^ (<=): Suppose that m is not a power of 2. Set m = 2 v,

odd number >. 3.
y < 2n.
CcU>& U) :
W r i t e fc = i; and g = 2 n .
m = kg = ~^{2g) = where
2(g - ^~Y^)
n >_ 1, y an
Then,
+k
- A= l(2a
+k
- 1} >
^ - 1
a
= g _
i$jm
Thus we have
k
]T (a + r - 1)
m =
r=1
which is clearly a a-partition.

C<U ill) : y > 2 n . Write Zc = 2 n and g = V.
g <_ 2k - 1 ^ k ^^^^
Now,
k > ^-^k
<L
-^
^>
and we have
m = gk = f (2/0 = f [ 2 ^ - 2 - ^ A ) + ^ _ ij = | ( 2 a + ^ - 1)
where
Hence,
a = k - ^"2"^.
9
m = y^ (a + p - 1 ) ,
r =1
a a-partition.
On the other hand,
g > 2k - 1=> S 1 ^ - >fe=^ 4 - ^ - k e IN,

and now
Ik
2k
-Y9 = -y
where
a = ^ ^
( S L X^ - k) + 2k - l\ = *, ^ + r - 1)
- k e IN, and clearly 2k >. 4.
This completes the proof.

CoswLlcUiy 2:
a-partition.
An even number is a power of 2 if and only if it admits no
We now have a natural extension of the sequence of primes, defined by a

single property, in the following direct consequence of our two corollaries.
Tk2.Oh.1im 1 : A number m is either a prime or a power of 2 if and only if m
admits no a-partition.
This theorem provides an easy sieve to isolate the set of primes (and immediately recognizable powers of 2) less than or equal to any preassigned
number x of moderate size: one simply writes down the segment 1, 2, 3, . . . , x
and crosses out the a-partitions less than or equal to x, starting with those
with leading term a = 1, then those with a = 2, etc. The least upper bound
[Dec.
520
x 3l
< x 3
, for clearly a = - if and

only if 3a + 3 f #, which is equivalent to a + (a + 1) + (a + 2) = x.
A simple example of "sieving" out the primes, for instance with x = 15,
shows that a given composite number may admit more than one a-partition. Our
next problem is to give an account of the different possible a-partitions of
a given composite number 77?. First, we deal with the case where 77? is an odd
number. Consider an arbitrary factorization m = kg (3 j< k <. g) . Corresponding with this factorization, one always has the a-partition
^)
L [(^ " ^T"^) + r ~ X ] of m'
If g < 2k + 1, then k - ^^
Z
that
(2)
t[(*
IN, and once again direct computation shows
~^)+r-l]
is a a-partition of 77?. Clearly, (2) coincides with the fixed partition (1)
if and only if k = g9 i.e. , the case where m is a square and is factored as
such,
If g _> 2k + 1, then clearly ^z
k E IN, and we obtain the a-partition
2k
(3)
17
E [(^HP - *) + * - ! ] f m-
The partitions (1) and (3), having different lengths, can never coincide.
Conversely, the indicated possible a-partitions corresponding to the particular type of factorization of m are the only possible ones m can have.
For, consider an arbitrary a-partition
(a + P - 1) = ~(2a + n - 1), n > 3 .
r=1
If n is even, then n/2 is an odd divisor of m and so is 2a + n - 1. Clearly,

2 a + n - l > n > n/2 so that we may write k = n/2 and g = 2a + n - 1, k < g.
In this notation, we have that g = 2a + 2k - 1 >_ 2k + 1, since 2a >_ 2, and a
= (g + l)/2 - /c. Hence, the given partition is of the form (3). If n is an
odd number, we have that 2a + n - 1 is even and that W = (2a + n - l)/2 is an
odd divisor of m. If n < w, we put k = n and # = U. Then, 2^ = 2a + k - 1,
so that a = g - (fe - l)/2, and the given partition has the form (1). If n >
W9 we write k = w and g = n. Then, 2k = 2a + g - 1, so that g = 2/c + 1 - 2a
< 2k + 1, since a >_ 1, and a = k - (g - l)/2.
This shows that the given
partition has the form (2).
Summarizing these observations, we obtain the following characterization
of the a-partitions of a given composite odd number.
Tk&OSl&m 2: The a-partitions of a composite odd number m are precisely
those determined by the factorizations of the form 777 = kg (3 <. k <_ g), namely
V = 1
and exactly one of
(2)
[(fc " ^T^)+r ~ x ]
(lf g< 2/c + 1}
1978]
521
and
<}
Lvîr1 ~ k) + r ~ X J
(if
a > 2k + i)
If g < 2k + 1, the two valid partitions (J) are (2) are different, except in
the case where g = k9 and if g >_ 2k + 1, the valid partitions (1) and (3) are
always different.
Now consider any two different factorizations (if they exist) of 777 into
two factors: m = kg (3 k <_ g) and m = kTgr (3 <_kr <_ gr) . Then, comparing
the lengths of the resulting a-partitions, one sees that every possible partition corresponding with m = kg differs from every possible one corresponding with 777 = kTgr.
Hence, if 777 admits t different factorizations of the form
m = kg (3 <_ k g) , then 7?? admits It different a-partitions, except in the
case where m is a square in which case the number is 2t - 1.
Finally, we consider the nature and number of a-partitions of even numbers other than powers of 2.
Tfieoêm 3: Let m be an even number other than a power of 2. Then, there
exists at least one factorization of the form m = kg (k < g) , where one of
the factors is an even number and the other an odd number J> 3. For each such
factorization, exactly one of the following three conditions holds:
(1) k is even, g is odd and g < 2k + 1;
(2) k is even, g is odd and g >_ 2k + 1;
(3) k is odd, g is even;
and only t h a t sum i n t h e
list
(V)
2k
r/
-1
(2<)
'
which corresponds with the valid condition is a a-partition of m.

these are the only possible types of a-partitions of m.
Finally,
PflOO&i First we observe that m may be written in the form m = 2nu, where
n >_ 1 and u is an odd number >_ 3. Now, consider an arbitrary factorization,
m = kg (k < g), into an even and an odd factor. If k is even, then the possibility (3) is ruled out and the ordering axiom ensures that exactly one of
(1) and/or (2) holds. If k is odd, then the first two possibilities are excluded and (3) obviously holds.
Concerning the next part of Theorem 3, we note that each of the indicated
sums is of the form
E',(a + v - 1),
V = 1
and that n >_ 3 and a e fft, providing that the condition to which the particular
sum corresponds holds. Moreover, in each of these cases the indicated summation results in the product kg. On the other hand, the validity of any given
condition (i) clearly results in a f IN in the sums (j f) 5 i 4 j . This concludes
this part of the proof.
Dec. 1978
522
F i n a l l y , we consider an a r b i t r a r y cr-partition
n
m = ^2 (a + v -" !)
j(2a
+ n - 1).
r=1
If n is even,
one must have
write k = n/2
Moreover, g >
then 2a + n - 1 is an odd divisor of w, and since m is even,

that n contains a factor 4. Since 2 a + n - l > n > n/2, we may
and a = 2a H- n - 1. Then m = kg, k < g, k Is even and ^ is odd.
2k, so g >_ 2k + 1. Finally, from g = 2a + 2k - 1, we obtain
a + 1 7
a = z^
k.
Hence, the given partition has the form (2f).

If n is odd, then n is a divisor of 7?? and 2a + n - 1 Is an even number. Since 777 is even, we must have that
(2a + n - l)/2 is an even divisor of m and we may write (2a + n = l)/2 = 2w.
Considering first the case where n > 2w and using the notation g = n and k =
2w, one easily checks that k and g satisfy the requirements of condition (1)
and that the given partition has the form (1 f). A similar straightforward
analysis of the case n < 2w shows that k = n and g = 2w satisfy condition (3)
and that the given partition has the form (3f).
This completes the proof.
In conclusion, we want to determine the number of different a-partitions
of an even number 7?7 other than a power of 2. We once again consider two different factorizations as specified in the theorem (if they exist):
(a) 777 = kg,
(3) m =
krgr.
Let condition (i) in the theorem be satisfied in (a), and let condition (j)
be satisfied in (3). We consider two possibilities:
i, = j : Here k and kf are both even or both odd. Since k ^ kr and g ^ gr
we must have that the a-partition (i ') relative to (a) is different from the
a-partition (j') = (ir)
relative to (3).
i ^ j : Suppose that k and kr are both even. Then g and gr are both odd
and of the resulting a-partitions one is of the form (1') and the other of the
form{2') Noting that one of the lengths here is odd and the other one even,
we conclude that the two a-partitions are different. Suppose now that one of
the factors k and kr is even and the other one odd. Without loss of generality we may assume that k is even. Then j = 3 and the factorization (3) yields
the a-partition (3r) of odd length. If i = 1, then the equality of (1 ') relative to (a) and (3f) relative to (3) would imply that g = kr, so that k = gr.
This, however, would imply that k > g, a contradiction. Hence, we have that
the two resulting a-partitions are different in this case as well. If -i = 2,
then the partition (2r) relative to (a) has even length, while (3r) relative
to (3) has odd length, so they do not coincide. Therefore, we may conclude
that if 777 admits t factorizations m = kg where one of the factors is an even
number and the other one an odd number >_ 3, then m admits t different a-partitions.
REFERENCE
1.
I. Niven & H. S. Zuckerman, An Introduction

to
3rd ed. (New York: John Wiley & Sons, 1972).
the
Theory
of
Numbers,
ON ODD PERFECT NUMBERS

The New South
G. L. COHEN
Wales Institute
of Technology,
Sydney,
Australia
If O(n) denotes the sum of the positive divisors of a natural number n,

and o(n) = 2n, then n is said to be perfect. Elementary textbooks give a
necessary and sufficient condition for an even number to be perfect, and to
date 24 such numbers, 6, 28, 496, ..., have been found. (The 24th is
219 9 36 /ol 99 37 _ -, x
discovered by Bryant Tuckerman in 1971 and reported in the Gu-iness Book of

Records
[3]. The three preceding ones were given by Gillies [2].)
It is not known whether there are any odd perfect numbers, though many
necessary conditions for their existence have been established. The most interesting of recent conditions are that such a number must have at least
eight distinct prime factors (Hagis [4]) and must exceed 100 2 0 0 (Buxton and
Elmore [1]).
Suppose p , . . . , p are the distinct prime factors of an odd perfect number. In this note we will give a new and simple proof that
(1)
- -<log2,
i = 1 r"z-
a result due to Suryanarayana [5], who also gave upper and lower bounds for
t
when either or both of 3 and 5 are included in (p , ...,p }.

Most of these bounds were improved in a subsequent paper with Hagis [6],
but no improvement was given for the upper bound in the case when both 3 and
5 are factors. We will prove here that in that case
t
Z ^
< .673634,
the upper bound in [5] being .6 73770. We will also give a further improvement
in the upper bound when 5 is a factor and 3 is not; namely,
V- 1
> < .677637,
the upper bound in [6] being .678036. (These are six-decimal-place approximations to the bounds obtained.)
We assume henceforth that n is an odd perfect number.
An old result, due to Euler, states that we may write
i =l
where p , . . . , p are distinct primes and pk E ak = 1 (mod 4) for just one k

in {l, . .., t) and a{ = 0 (mod 2) when i ^ k. We will assume further that
p < ... < p , and later will commonly write a(r) for a^ when pi = r.
The
subscript k will always have the significance just given and IIr and Z' will
denote that i, = k is to be excluded from the product or sum.
523
524

We w i l l need t h e well-known
(2)
\{pk
[Dec.
result
l)\n,
which is easily proved (see [6]). It follows that
(3)
PX \ipk
+ 1).
We a l s o use t h e i n e q u a l i t y
1 + x + x2
(4)
> expLr + -rx2} ,
0 < x <_ -k
To prove t h i s , n o t e t h a t
exp(x
+ ^x2^)
- (1 + x + x2)
= 1 + x + ~~ + - y L r + -)
_, x3 ^ xh , 1 /
+ - (1 + x + x2)
,x2V
so we wish to prove that

x , x2
x2\3
x2Y
Now,
f + f U n + Jiiir- 09
and
i / , * 2 V . i I , *2X*
a; + -7+
r uc + -7+
4
3!x2\
W
4!a; 2 \
<
^(*
1 /l3\
~ 18 \ 1 2 /
T )
3
36
23
<
T) + (*
T)2
'2'
Hence (4) is true. Other and better inequalities of this type can be established but the above is sufficient for our present purposes.
Now we prove (1). Since n is perfect,
t
In = o(n) = Y\ (1 + p. + p?
+ . . + p*1)
' - f t K - * )
By E u l e r ' s r e s u l t , ak >_ 1 and a^ >_ 2 (i- ^ k) , so
P k /i = i \
by ( 4 ) .
Pi
P?/
Hence,
log2>1 8 1+
( t) + 5'(t + ^)
Pfc/i.'i
VPi
4P;
1978]

t ,
t ,
525
- - -5- + y A. + I y'J_ > y _!_ + J^ _ _i_
>
~ti?i
ttx Pi
(Pk + D
4&P?
4Pl2
M P ,
2p 2
2p , 4 l P i
using ( 3 ) .
We end w i t h t h e
Th2.OA.2m: ( i )
I f 15 I n ,
then
x - l ^ l ^ l ^ l ^ l , ,
AFT
(ii)
I f 5 | n and 3 | n ,
l 3+ 6 l
+ l o g
= lr^
<
= a s a y
'
'
then
v - 1 ^ 1 ^ 1 ^ 1 ^ ,
Z-^7
2950753
2815321
31 + 61 +
l o g
293105
190861
= 6
'
Say
'
Vh.00^1 The proofs consist of considering a number of cases which are

mutually exclusive and exhaustive.
(i) We are given that p, = 3 and p = 5. Suppose first that ax = 2 and
a2 = 1 (so that we are assuming, until the last paragraph of this proof, that
k = 2). Since cr(32) = 13, we have 13 \n.
Suppose 01(13) = 2, so that, since a(132) = 183 = 38 61, 6l\n. Since also
0(5) = 6 = 2 3, we cannot have a ( 6 1 ) = 2, for a(612) = 3783 = 3 13 97 and
we would have 33|n (i.e., a,1 > 2). Hence, oi(61) iL 4. Then, using a simple
consequence of (4),
t
61 61 2
P;* 13,61
so, taking logarithms and rearranging,

t
V^ 1 <, -,
o , 13 , 6
^ - < l o g 2 - l o g T - l o g - -
183
14076605
log ^
- log 1 3 8 4 5 8 4 1
i =i
+ 1 x 1 ,1 , 1
3
If ot/13N .> 4 , t h e n we s i m i l a r l y
13
61
'
obtain
^7 < log 2 - log(l + i + i ) - log(l + i )

- l o g ( l + - r \ + ~ \ + - ^ + - ^ - ) + ~ + i + ~~<
13
V
13 1 3 2 1 3 3 1 3 V 3
5 13
Suppose now t h a t OL1 >. 4 and a 2 = 1 . Then,
ZjT
< ^g 2 - log(l + f + ^ + ^ + i ) - log(l + i ) + i + \ < a.
526
Next, suppose that a2 >_ 5.
[Dec.
Then,
X ) ~ < lg 2 - logfl +f + ^
Log
(1+I
Finally, suppose k > 2, so a 2 > 2,

above,
A+J_
< a
Since ak >_ 1, we obtain, proceeding as
log 2 > log(l + j-) + log(l + I + ^ ) + log(l + I + -L) + g'-L
j. i
a. i
!
+
log T +
log 25 " !3 " j
" - !r
But p _>_ 13 (though we can easily demonstrate that in fact p >_ 17), so,
t
l ^
,
13
3 1 1 1 1 ^
< l o g 02 - l o g - nl o g + - + - + 333 < a.
i = l î
This completes the proof of (i).

(ii) We are given that pl = 5. The details in the following are similar
to those above. Suppose, until the last paragraph of this proof, that ax = 2.
Since a(52) = 31, we have 31\n. Now, a(312) = 993 = 3 - 3 3 1 and 3Jn, so we
must have a( 31 ) >_ 4. It follows from (2) and from the fact that 3/fft, that
if p < 73, then pk must be either 13, 37, or 61 (so we cannot have a1 = 1).
Suppose first that p = 6 1 . Then a( 6 1 \ >_ 1 and
t^<iog2-iog(i
i)-io g (i
- i T + -i ?
-i,;
- i o g ( l + ^ + i + ^- + ^-= b.
If pk = 1 3 , t h e n , by ( 2 ) , p 2 = 7 .
A l s o , ot , j 3 .> 1 , so
L ^
a ( 7 2 ) = 57 = 3 - 1 9 , so a 2 >. 4 , s i n c e 3/fn.
< log 2 - iog(i + j + jr) - iog(i + y + -yr + yj + y^

-log(l + ^)-lo g (l + i + ^
If p k = 37, then, by (2), 19 \n.
a(192) = 381= 3-127, s o a ( 1 9 ) >_ 4.
V < log 2 - logfl + 5i + - 2M - logfl + -~r + 2 + -^3 +

f^Pi
\
5 /
*\
19 19
19 19L
logfl +-~
+~
31
+ ^j 3+ ~4
31 z
31 31
Since
1978]
If p
A SIMPLE CONTINUED FRACTION REPRESENTS

A MEDIANT NEST OF INTERVALS
527
>: 73, then, as in the last paragraph of the proof of (i), we h

ave
t
Finally, suppose 04 >. 4.
Then pk
>_ 13 and, as in the preceding paragraph,
S i < l o g 2 - l o g ( 1 + i + ^ + ^ + ^ ) + i + rri? < & This completes the proof of (ii).

I am grateful to Professor H. Halberstam for suggesting a simplification
of this work through more explicit use of the inequality (4).
REFERENCES
1.
2.
3.
4.
5.
6.
M. Buxton & S. Elmore, "An Extension of Lower Bounds for Odd Perfect Numbers," Notices
Amer. Math. Soc., Vol. 23 (1976), p. A-55.
D. B. Gillies, "Three New Mersenne Primes and a Statistical Theory,"
Math. Comp. , Vol. 18 (1964), pp. 93-97.
Guiness Book of Records, 22nd ed., 1975, p. 81.
P. Hagis, Jr., "Every Odd Perfect Number Has at Least Eight Prime Factors,"
Notices
Amer. Math. Soc,
Vol. 22 (1975), p. A-60.
D. Suryanarayana, "On Odd Perfect Numbers II," Proc. Amer. Math.
Soc,
Vol. 14 (1963), pp. 896-904.
D. Suryanarayana & P. Hagis, Jr., "A Theorem Concerning Odd Perfect Numbers," The Fibonacci
Quarterly,
Vol. 8, No. 3 (1970), pp. 337-346, 374.

North
IRVING ADLER
Bennington,
VT 05257
1.
While working on some mathematical aspects of the botanical problem
of phyllotaxis, I came upon a property of simple continued fractions that is
simple, pretty, useful, and easy to prove, but seems to have been overlooked
in the literature. I present it here in the hope that it will be of interest
to people who have occasion to teach continued fractions. The property is
stated below as a theorem after some necessary terms are defined.
2. TeAmZnoZogy:
For any positive integer n, let n/0 represent . Let
us designate as a "fraction" any positive rational number, or 0, or , in the
form alb,
where a and b are nonnegative integers, and either a or b is not
zero. We say the fraction is in lowest terms if (a, b) = 1. Thus, 0 in lowest terms is 0/1, and in lowest terms is 1/0.,
If inequality of fractions is defined in the usual way, that is
alb
< eld
if ad < be,
it follows that x < for x = 0 or any positive rational number.
528

[Dec.
3. Tfie M&cLiant: If a/b and c/d are fractions in lowest terms, and a/b
< c/d, the mediant between a/b and c/d is defined as (a + c) / (b + d) . Note
that a/b < {a + c)/(b + d) < c/d.
ExamptoJsThe mediant between 1/2 and 1/3 is 2/5. If n is a nonnegative
integer, the mediant between n and is n + 1. If n is a nonnegative integer
and m is a positive integer, the mediant between n and n + l/m is n +1/(777 + 1).
4. A Mediant N&>t: A mediant nest is a nest of closed intervals
..., T n , ... defined inductively as follows:
I0,Il9
ô = [0, " J .
For n >_ 0, if In = [r, s] , then .Tn+ 1 = either [r, m] or [m, s] , where 777 is the
mediant between v and s.
It is easily shown that: if at least one In for n >_ 1 has for form
[r,m],
then the length of Jn approaches 0 as n -* , so that such a mediant nest is
truly a nest of intervals, and it determines a unique number x that is contained in every interval of the nest. For the case where every In for n >_ 1
has the form [m, s] , let us say that the nest determines and "contains" the
number . Mediant nests are obviously related to Farey sequences.
5. Long Notation
ÔH. a Mediant N&>t: A mediant nest and the number it
determines can be represented by a sequence of bits b1b2b3 ...b^
... , where,
for > 0, if Iiî
= [P, s] and 77? is the mediant between r and s, bi = 0 if
Ii = [P, m] , and bi = 1 if _Z\- = [777, s] .
ExampZ&> 0 = 0; 1 = ; 10 = T, the golden section; where each of these
three examples is periodic, and the recurrent bits are indicated by the dots
above them.
6. kbbfizvtat&d
Notation
ÔK. a Mediant N&6t: The sequence of bits representing a mediant nest is a sequence of clusters of ones and zeros,
b1b2b3
. . . bi
. . . = 1 ... 10 ... 01 ... 1 .. .
where the a^ indicate the number of bits in each cluster; 0 ^ a 1 ^ . o o ; 0 < a

<_ for n > 1; and the sequence (at) terminates with an if ccn = . As an abbreviated notation for a mediant nest and the number x that it determines we
shall write x = (a19a29
. . . ) Then a 1 < _ x < a 1 + l. The sequence (a^) terminates if and only if x is rational or . Every positive rational number is
represented by exactly two terminating sequences
(at).
Example()
= i = 00; (o, 00) = 6 = 0; (0, 2, >) = 00I = % ; (0, 1, 1, <) =
010 = \ .
In general, if x = (a2 , . . . , an_ l9 an, ) where an > I, then x = (a\,
. . . , a n _i, an - 1, 1, ) , and vice versa.
7. ThdOKom: If x = (a1? a 2 , . . . , a n , . . .) , then x = a1 + /a2 + +l/an
+ and conversely. If x = (a1 , . . . , an , 00) 9 then x = aY + l/a2 + + l/an
and conversely.
P/LOO^ 0^ the.
Tko.on.zmi
I. The nonterminating case, # = (a1, a 2 , ..., a^, . . . ) . Thus, x is irrational. Let p^/q^9
for >_ 1, be the principal convergents of a 2 + l/a2 +
... . Then a straightforward proof by induction establishes that for all
even _> 2,
1978]

529
and for all odd _> 1,

J
x + --'+^ = [Pi/Rt>
Pi-i/Vi-i]Consequently, the nest determined by successive pairs of consecutive principal convergents of al + l/a2 + + l/an + defines the same number as the
mediant nest (ax, a2, . . . , a n , ... ).
II.
The terminating case, x = (ax, ...,a n , ) . It follows from I that

Ia1 + -..+an+1
where
Pn + 1/qn
Since
+1
= [pn/qn,
Pn + 1/qn
= (pn_! + an + lPn)/(qn_1
l]
o r
[ P n + l / ^ n + 1* P / ? ] >
+ an + 1qn).
it follows that
x =
^lim^ J ai + ... +an+1
= p n /q n = a1 + l/a2 + + l/an.
III. The "conversely" in the theorem follows from the fact that the mapping of the set of mediant nests into the set of simple continued fractions
established in I and II is one-to-one and onto.
ExamptzThe mediant nest (0, 2, 3, ) and the continued fraction 0 + 1/2
+ 1/3 represent the same number. Verification:
a.
(0,2,3, ) is the abbreviated notation for the sequence of bits

001110.
The intervals In defined by this sequence of bits are:

it
Interval
Mediant between Endpoints of
Interval
lo = [0/1, 1/0]
Ii = [0/1, 1/1]
(0 + 1)/(1 + 0) = 1/1
Ii
= [0/1, 1/2]
(0 + 1)/(1 + 2) = 1/3
I,
= [1/3, 1/2]
(1 + l)/(3 + 2) = 2/5
I,
= [2/5, 1/2]
(2 + l)/(5 + 2) = 3/7
I5
= [3/7, 1/2]
(3 + l)/(7 + 2) = 4/9
I,
= [3/7, 4/9]
(3 + 4)/(7 + 9) = 7/16
In = [3/7, mn-i]
(0 + 1)/(1 + 1) = 1/2
n >. 6, mn.\ = the mediant between the

endpoints of Xn_i-
Since lim rnn_1 = 3/7, the number defined by this mediant nest is 3/7.
n --
b.
The continued fraction

0 +
^ - = 3/7.
2+
FOLDED SEQUENCES AND BODE'S PROBLEM

W. E . G R E I G
West Virginia
University,
Morgantown,
WV 26506
Readers of this journal have long been interested in Bode's Rule, see,
e.g., [26] and [15]. Indeed attempts to solve it have ranged from those devoid of science but fairly accurate to those based on some physical principle (s) but rather inaccurate, such as Berlage's and 0. Schmidt's theories.
The problem of planetary motions was first tackled by Eudoxus, who proposed
rotating tilted concentric spheres, rather like a gyroscope, to explain each
planet. When Kepler solved the problem of their motions with the concept of
areal velocity, the area swept out in an invariable plane divided by the time
is a constant, the problem of a law for their spacing remained. Indeed, I
think the unit of angular momentun should be named after Kepler for his contribution of area as a vector. Bode's problem is of great value to the history and especially the philosophy of science. The qualities that distinguish
pure mathematics are succinctness, elegance, fertility, and relevance to the
unsolved problems. But to a scientist the first criterion is reproducibility.
The multifarious, variegated, and at times loquacious and mellifluous monographs on this aspect of cosmogony attest to man's persistent and insistent
attempts, at times based on specious assumptions, to find order in a theatre
of nature that may have no reason to be other than nearly random. Such is
one view. But while science corrects its mistakes (so far) it must be remembered that Boltzmann committed suicide because his contemporaries would not
accept his counting of molecules, Wagoner's 1911 theory of continental drift
was not believed until the 'sixties, and Newton's theory was not believed on
the continent until Clairaut's prediction of the return of Halley's comet in
1759 (P = 76.75 1.5 yr) came true. Such is the lag between prediction and
proof. The final answer to Bode's problem will be known within twenty years
when sophisticated computer simulations are finished. My work will probably
remain the most accurate, namely 1-percent with a few exceptions either way.
In any case, my work has led to some interesting mathematics, especially the
Self-Lucas property (see Section 2). May it be that Urania and Euterpe have
recessed a part of Nirvana to sequester all who have slaved over this vexing
problem. The impetus for this paper comes from Kowal's [17] recent discovery
of an object between Saturn and Uranus that prima faciae,
see Table 1, fits
my rule [6, 7] and does not fit any other rule published! Sequentially, I
present an overview of the history of Bode's Rule, Kowal's discovery and then
generalized folded sequences.
1.
BODE'S PROBLEM AND KOWAL'S DISCOVERY
Historically, the first offered solution to Bode's problem was Kepler's

[1] perfect solids, which model, in fact, antedates Bode by two centuries.
Gingerich [11] has discussed the accuracy of Kepler's youthful proposal. At
first, the Titius-Bode mnemonic was successful with the Asteroids and Uranus
but it fails badly for Neptune and Mercury. Attempting to save it, Miss Blagg
[3] introduced two more parameters into it. Nieto, see [9] of paper [7], supports her work. The literature is full of algebraic rules of the form, distance bn and indeed of more complicated rules. It is interesting to look
at the range of b values. A partial list is:
530
Dec. 1978
Dermott
von Weizsacker
Greig
Quadranacci
Dermott
Cale
Pierucci & Dermott
Gaussin
Blagg
Dermott
Belot
von Weizsacker
Greig
Quadranacci
Titius
b
b
b
b
b
b
b
b
b
b
b
b
b
b
b
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
531
2
= 1.2599 (Saturnian moons)
1.370889 (10 eddies, inner planets),
1.378241 (Saturn's inner moons)
1.380 (see below)
3
= 1.4422 (Uranian moons)
0/J3 = 1.5115 (see [15])
41/3 = 1.5874 (Jovian moons)
1.72
1.73
61/3 = 1.817121
1.886
1.894427
1.8995476
1.905 (see below)
2.00
See Gould [15] for references I have not cited. Dermott (see [2] and [10] of
paper [7]) was forced to take Earth and Venus together to retain his period
factor of /6~. DermottTs arbitrary period factor of /2~ for Saturn's moons
misses both Rhea and Janus.
Indeed, I have myself happened upon some rather well-fitting arbitrary
rules. One such is a bisection of the Quadranacci recurrence Qn+i = Qn+Qn_3,
namely: 1, 1, 1, 1, 2, tf=3, 4, /=5, 7,4 = 10, 14, J = 19, . . . , = 95,
..., which is very good at representing reciprocal distances. Another one
for the distances begins: 4, 7, 10, 15, ... . The distance factor, b, is
(1.380278)2 = 1.905166. The most complex rule of which I am aware is Rothman's (see [9]), d = n(5. 5 + F2)/9 (1 +Fn ). It has seven parameters: n, 5.5,
2, 9, and 1, and two to determine the Fibonacci sequence, and since only 9
planets are fitted then only 2 degrees of freedom are left which is unscientific. All of the above rules are arbitrary, except von Weizsacker's and my
own. My own view is that any rule with more than two parameters violates
Occkham's razor, "Essentia
non sunt multvpticanda
yvaetev
necessitatem."
Dermott {ibid.) proposed different period factors for each satellite system, whereas my theory is simpler since the same limiting ratio, 0 , applies
to all. He also ignored the outer Jovian and Saturnian moons. I chose to
emphasize them. This suggests the principle of Contrary Ignorability: whatever earlier researchers ignorethat is the path to pursue. My work indicates that outer Jovian moons should cluster well within 10 percent of 97,
257, 730, and 608 days (Table 2 of [6]). The announcement of Jupiter's XIII
moon [21] at 239 day and i = 27 [25] came after my initial work [4], [22]
and satisfies the above sequence. I also studied the relevance, if any, of
rotation periods and grazing periods of parent bodies in [4], using:
(35)
PgPM = 3TT/C7 where
G = 498 day"2 (g/cc) "x.
The period of a satellite just grazing the surface of Saturn, the Sun, Jupiter, and Uranus would be 0.167, 0.116, 0.12, and 0.11 day. The criterion
for coalescence against tidal forces may be written
(36)
pm > /M/distance3
or
Pm > PgJ (4TrfpM /3pm)
where m, M are the satellite and parent masses, p m and pM their densities, Pm
the period of a satellite at this Roche limit, and / a factor between 2 and
10 [24, p. 18]. When (36) is not satisfied "rings" result. In [4] an inner
Uranian moon was suggested which would have a period of 1/1.3292 = 0.752 day
532
[Dec.
according to (17) of [6]. It has even been proposed that the separation, a,
between binary stars satisfies Bode's rule [27], but I am very skeptical of
that.
My own interest in Fibonacci numbers dates back at least to 1966 when I
obtained a copy of Vorobyev's book. I tried these numbers on the planets
with what seemed good accuracy and communicated this to Gould [13] , pointing
out the relevance of bisected Fibonacci sequences. After long arduous efforts, I thought I had put the problem to rest when news of Kowal's discovery
[17] of a planetoid between Saturn and Uranus, too big to be a comet nucleus
and too small to be a large planet, was announced. He calls the object Chiron
after one of the Greek half-man/half-horse animals. Is this discovery to
prognosticate that this Chinese year 4676 (see [18]), beginning 7 Feb. 1978,
should not be the year of Earth-Horse, but rather the year of the Centaur!?
One can see in Table 1 that Chiron fits very neatly into the bisected halfinteger sequence. I could have predicted this object three years ago [5, 6]
from the folded sequences I had discovered but it would have been considered
wildly delusionary at the time. The major body that should occur before Neptune in my sequence given in [5] and Table 2 of [6] is easily calculated to
have a reciprocal period corresponding to -1974 - 4558 = -6532. Its period
should then be 317816/6532 = 48.66 yrs. The agreement with ChironTs period
of 47 to 51 yrs is quite good. The ellipsis (...) in [5] and [6] clearly indicated that the sequence continued in both directions so that a body at Chiron's position was implied. In an earlier work [4] I had stated, " . . . one
should really ask why don't Jupiter, Uranus and Neptune have a plane of particulate matter [rings] inside Roche's limit since that is natural considering the pervasiveness of grains and cometesimals . . ." (p. 16). As we now
know, rings have been found around Uranus [16], [28]. The way to test my theory would be a computer simulation using reciprocal periods given by (23a) or
(23b) or [7] (or, perhaps, using part of a bisected odd-N folded sequence in
[6]) to see if the broad maxima in [10] can be sharpened.
The agreement of Chiron's period with Folded Fibonacci sequences is
reassuring but not perfect. I have been able to represent Neptune and outer
satellites in general more accurately than any other rule simply because I
worked with recursive sequences rather than naive power laws. Also, mine is
the only work to represent the several comet groups (see Table 1). So that,
in terms of completeness and goodness of fit, my hypothesis is the best. It
remains for a computer simulation to test whether my proposal gives maximum
stability. Such a simulation may solve the following question: Why are some
period ratios nearly but not quite small integers? Saturn:Jupiter is not 5:2
but is 6551:2638 to seven, significant digits. Accurate to only five digits
is 149:30. Neptune:Uranus is not 2:1 but is 51:26 to five digits. And Uranus : Saturn is 77:27 to nearly five digits. Similarly, Earth:Venus is not F7:
F6 but 1172:721 to eight-digit accuracy. These ratios suggest that low-order
commensurabilities (LOG) are avoided, except for the ratio 2 among the Galilean satellites. The Kirkwood gaps indicate that LOC are unstable if the
ratio >i 2, such as 11/5, 9/4, 7/3, 5/2, 8/3, and 3/1.
The problem is ancient. The Pythagoreans believed in orbits in arithmetic progression and added a Central Fire and a Counter-Earth [23] to obscure
that Fire so that the total number of moving bodies be the "magic" number
1 + 2 + 3 + 4 = 10. Yet Aristarchus placed the Sun in the center for reasons
of simplicity 18 centuries before Copernicus. Later, Ptolemy and others confounded the picture with equants and epicycles, until Kepler discovered that
blemished curvethe ellipse.
1978]
533
Further back in time, the concepts become anthropocentric and folklorish

as in the mural of Ra and Noot found in the tomb of Rameses VI.
There is still the possibility that Bode's problem has no solution or that
the distribution of planets is random on a logarithmic axis, save that they
cannot be too close to each other. But my work has led to some interesting
sequences that I will discuss in the future.
TABLE 1
The Correspondence between the Half-Integer Sequence and the Planets
Reciprocal
Period
Period
(yrs)
-550 - 233
340 + 144
-210 - 89
0.999969
1.618
2.6183
130 + 55
-80 - 34
4.235
6.859
50 + 21
-30 - 13
20 + 8
-0 - 5
0+ 3
11.07
18.035
28.66
48.66
69.73
- 2
0+ 1
0- 1
20
30
50
80
130
210
340
550
890
1440
2330
3770
6100
9870
1
1
2
3
5
8
13
21
34
55
89
144
233
Solar System
161.00
123.0
521.0
99.5
83.55
45.42
29.42
17.85
11.11
6.849
4.237
2.6177
1.618
1.0
0.618
0.382
0.236
2.
null
Hungaria #434 (991 da = 2.71 yrs), etc. (Average
= 2.75 yrs)
null
Faye (7.35 yrs); Brooks II (6.72 yrs); d'Arrest
(6.67 yrs), Finlay (6.90 yrs), etc.
null
Neujmin (17.97 yrs)
null
Chiron (47 to 51 yrs), other Centaurs
Olbers (69.6 yrs); Brorson-Metcalf (69.1 yrs);
Pons-Brooks (71 yrs); Halley
Neptune (164.79 yrs); N + P (168.4 yrs)
Swift-Tuttle (119.6 yrs); Barnard II (128.3 yrs)
Planet X (464.? yrs)
null
Uranus (84.01 yrs)
null
Saturn (29.46 yrs)
null
(Jupiter 11.86 yrs) not meant to fit Jupiter.
null
Astrea (4.13 yrs); asteroids (0.23 yr )
null
(Mars)
(Earth)
(Venus, 0.615 yr)
null
(Mercury, 0.241 yr)
GENERALIZED FOLDED SEQUENCES
The obvious generalization of the definition of folded sequences, (4) of

[6], is
(37)
{%,}*= Pj.k + (~lf + 1Pj,k-N
with
0<k<Fl-l
where a script
letter denotes a folded sequence, {Pj} is the jth coprime sequence as in (1) of [20], and {^-.jy} is finite if N is finite. This latter
534
[Dec.
point differs from (1) of [6] wherein folded sequences were made infinite by
repeating the cycle ad infinitum.
It will help to display the Folded Pell
array {q>2} in Table .2:
TABLE 2
Folded Pell Sequences for N Odd
Sum
N
1
3
5
29
7
9
985
-1
-11
13
41
169
-69
31
-7
17
27
71
-407
171
-65
41
17
75
167
239
1393
409
11
5741
-2377
987
-403
181
-41
99
157
413
983
2379
13
-13859
5743
-2373
997
-379
239
99
437
973
2383
5739
-13855
5753
-2349
1055
-239
577
915
2407
5729
13865
5811
-2209
1393
577
2547
5671
13889
6+5/8
-2 -2/8
2+/8~
2 +0
6 +/8~
15
17
6 +/8 -2-0
-5/2
-7/2
-3/2
8119
13861
47321
2 +/8~ 2 +2/8" 6+5/8
-h
/2
3/z
5/2
7/2
The last row is the half-integer subscript v defined by 2k = 2v + N. Both

infinite folded sequences (which I often call half-integer sequences) are
given, namely for N= 1 (mod 4) and N = 3 (mod 4). Subscripts within braces
are part of the name of the array/sequence, whereas those outside the braces
indicate the value of the row/element.
In order to find the jth Half-Integer sequence the theorem in [6] is generalized.
Th<LOH.Qjmi {%-,s)k
+1
/{cPj,N)k
approaches a limit as N+ for each k depen-
dent only upon the value of N modulo 4.

Pl00&:
Define 0 = N modulo 4.
k = (N - l)/2 =
We will need P_kP_k_i/(Pk+1Pk)

Pk/Pk
+1
Since r = (k - N/2)
then r = -h gives
[N/2].
= -1, from which one finds
= -z =
-Pk/P-k..
Thus z -> 3 as N-+, while 0 = 3 or 1, respectively.
= i^â
Then,
- g ) / a + z)
which -> ((a - 1) / (a + 1)) x as il7->, while /V = 1 or 3 (mod 4), respectively.

Hence, where the limit of 9 is S,
(38)
{S.-U/isA
= (a + D/(a - 1) = -{S$}/{$}-,,.
In the following the first subscript of 5j\r and Pj,n

A number of relations follow from the elegant
(39)
r= P*+h + dPr_h
and S* = P*_ h + d P r + h ;
is suppressed.
1978]
535
these are (40), (41), (42), (43), and (45).

(40)
jP% = (Qm + h~ *_h)
(41)
(2 + d)P*+h
and Pn = (*n+h-
= 2*+ 1 + r
Qn.h)/jd,
and d(d + 2)Pr_h
= ( p + 5 ? ^ ) ,
(42)
(5 r + 5*) = 2 Cot* and ($ r - S$ ) = 2 ^ 3 ,
and the Binet-like formulas

(43)
where
(44)
Q r = (ar + i3 r )
and *
= (ah + ^3_/z) = (a
n+1
and analogous to Fn
(-l)
(45)
(5.
^ > Y>
'
^ 2
+ oT") =
F.n.
-i($}
) , where i,
-1,
we have
r+h
*PH
Now the initial values are determined by (37) or (38) and are
(46)
(2 + d) =
j,-fe
and
for all j .
J.?J
The bisections of the general Half-Integer sequence for N = 1 (mod 4) appear in (47) and (48), where t instead of j is used from now on for the parameter. Compare with the penultimate rows of Table 2. Also (49) is the
h
/5.
subscript r.
In Table 3 note that (2 + d) = Z,' = (a + a~n)/L and V
(47)
(th+dth+it2+3dt2+d+2)
(t2+dt2+d+2)
-(t3+dt3+3t+2dt)
(48)
2+d
(t2+2+d)
(t3+3t+dt)
-(t+dt)
(49)
TABLE 3
Parameters of Half-Integer -Fib Sequences
I-h/Ih
(ah + a-h)
(ah-a'h)
td
Jv - 2
/y + 2
1.272020
8.352410
2.014490
0.241187
1.000
1//I
3//2
1.414213
5.828427
2.030104
0.348311
1.500
/5
1.618034
4.236068
2.058171
0.485868
2.236
1.5
5/2
2.0
3.0
2.121320
0.707107
3.750
/8
2.414213
2.414213
2.197368
0.910180
5.657
/5
2.618034
2.236068
2.236068
1.0
6.708
8/3
10/3
3.0
2.0
2.309401
1.154700
8.888
/n
3.302776
1.868517
2.367605
1.267103
10.817
3.732050
1.732050
2.449490
1.414213
13.856
/20
4.236068
1.618034
2.544039
1.572303
17.888
/32"
5.828427
1.414214
2.828427
2.000000
33.941
2/3
536
[Dec.
Table 3 shows us that the Half-Integer Pell sequence is the only one in
which the ratio of the central pair, S_-h/Sh equals the characteristic root.
This is clear from (a + l)/(a - 1) = a whose solution is 1 + /l.
Sequences
given by (a^ + a~h) = a have a very simple Binet-like formula but the larger
root a = 2.1478990 is the solution to a cubic. Are there Half-Integer sequences with integer terms? Consider t-Fib sequences,
^t,
n+1
tPt
t n
+ Pt, n-1 >
for which d is an integer, d = t + 4. Consider the Root-Five sequences,

t = /F, which have the PL-types: 1, 0, 1, v, 6, 7v, 41, 48y, 281, (7 47)i>,
(18 107), (55_- 41) , ... and 2, v, 7, Sv 9 47, 55y, (23 14), 377y, 2207,
... where v = /5. In the corresponding Half-Integer sequence, from v = -llh
to +ll/z,
(50)
..., -199v, 170, -29y, 25, -4i>, 5, v,
10, lly, 65, 76y, 445, ...
we see that both bisections are integers (after dividing by common factors).
Furthermore, one bisection consists of every fourth Fibonacci number including F5 and the other consists of every fourth Lucas number including L5. Can
this be generalized? As a little algebra shows, yes,
(51)
di
= d\ - 2
or
d = t\
+ 2
or
= thdh
ti
where t 4 , dh refer to the sequence from which every fourth term is extracted
and di, ti refer to the chosen sequence. To illustrate this, the Root-32
Half-Integer sequence, see Table 3, has from (46)
Sh = t = k/2
and
S_h = 2 + d = 8.
The bisections of this reduced by common factors are:

(52)
..., 33461,
985, 29, 1,
5,
169,
5741, ...
(53)
..., -8119, -239, -7, 1, 41, 1393, 47321, ...
which are every fourth Pell number as expected beginning with P5 = 29 and
P*/2 = 41. Proofs of statements above follow easily from (39) or (43). So
(23a, b) are every fourth term of the t = /(/5 - 2) sequences. The "F" sequence to 3 decimals is ..., 1.236, -0.486, 1, 0, 1, 0.485868, 1.236, 1.086,
1.764, 1.943, ... and clearly the ratios
1.236:1:1.764 = 3 + v:2 + v:3 + 2v = 20:0 + 1 : 0 + 3
show that every fourth term of "F" gives (23a) of [7] and equivalently a bisection of Table 1. The general recurrence of these bisections is
Bt,n
where P*
rence
+1
P't,>+Bj,n
P>t,n-1
= 34 for (52) and (53). A bisected t-Fib sequence has the recurPn+2
= (t2 + 2)P - P_ 2
or
&2Pn =
t2Pn.
Indeed the recurrence's middle term for ^-sectioning has a coefficient given
by the 777th rising diagonal of the Lucas triangle. The bi-bisection case is
Pn+1+ = (ah + ka2b + 2b2)Pn - bhPn.hi
and so on.
We come now to what I regard as the most important property of these sequences. The Self-Lucas property, (14) of [6.], remains unchanged in this
generalization, namely
(54)
(2 r + 1 + gr.j)/d
= (-l)r'\.P
and
(.i* +
1+S*.1)/d=(-lf
+h
S*1.,
1978]
537
where d = (a - 3) and (a + g) = j = t .
Now (54) may be proven from (43).
Note that terms with subscripts (r + 1) , (r - 1) and -r all belong to the
same bisection of t or . This is obvious since (r + 1) - (r - 1) = 2 and
(r - 1) - (-p) = 2k - (N + 1) which is also even. Taking ratios of (54) one
may form the triplet rule, for both t an<3 ?,
(55)
( r + Q>v + 2)/(Q>r
+ r _ 2 ) = S-r-i/S-r+i.
Now this is readily illustrated when J is an integer so consider (50). Obviously (10 + 65)/3 = 25 and (lit; + 76y)/3 = 29i> and both are members of (50).
Again from (52) one has (985 + 29)/6 = 169. From Table 1, I illustrate (55)
by ((80 - 2) + (30 - l))/((30 - 1) + (0 - 1)) = (0 + 5)/2 by using the trick
(0 + 2) = 0v.
But this last ratio, (0 + 5):2 is Chiron:Neptune.
I introduce a new operator capital lambda, A:
(56)
A E I + E,
where E and I are the forward shift and identity operators and, therefore,
(A - V) E (E + E~l).
Then the Self-Lucas property may be written
(57)
(A - V)$t,r = d(-l)"*St,-r,
where 2" is another notation for and *, respectively. We may also write
KBn - dB-n or -dB-n depending upon which bisection of Qt or * is being considered. The Self-Lucas property does not hold for F-like sequences: P t ) 1 =
1 = P t ) _ 1 ? or L-like sequences: P%t 1 = 1 = -P%t_1.
In this sense, the HalfInteger sequences are more important.
How are (54) and (57) to be interpreted geometrically?
Let alternate
terms of the bisection be made negative, then the recurrence is
(58)
6zNn = ~(t2
+ 4)tfn.
Further let the terms of Nn be reciprocal periods of planets and let a minus
sign mean retrograde (backward) motion. Then the Self-Lucas property may be
written
(59)
ANn =
-dN-n,
which in words says that the set of synodic (apparent) frequencies of a collection of alternately pro- and retrograde planets are simply proportional to
the negative of the sidereal (real) frequencies in reverse order.
3.
COMMENTS ON THE RECIPROCAL PERIOD RULE
Why are the planetary frequencies not Folded or Half-Integer Pell sequences? The limiting distance ratio would be 3.2386766. One solution to this is
point (x) of [7], namely to bring the planets closer to each other, thereby
minimizing their potential energy which is negative; that is,
(60)
m^YJ{GMJmi/d2av),
i *J
where Mj is Jupiter's mass, m^ the mass of any planet except Jupiter, and dav
is a time-weighted distance from Jupiter. The Pell and, indeed, all t -Fib
sequences satisfy point (ix), the avoidance of low-order commensurabilities,
since gcd(Pt,n+i, Pt,n)
= 1 for all integers t and n. This can also be seen
by noting that the continued fraction of the roots of any t-Fib recurrence
consists of repeated (l/t) T s, so no one convergent is a great deal better
than another. The sequence, 11, 12, 16, 24, 38, ..., is an example who---
538
[Dec.
gcd(12, 16) ^ 1. Given that the total number of planets is a constant, then
minimization of the cumulative perturbation frequencies (synodic) occurs as
t becomes small (point xii). Of course, as t becomes small the average distance becomes smaller and the average perturbation force becomes very large.
k.
FINALE
The logic [7] of my rule suggests that other civilizations may be signaling us in binary code with 1/02 = 0.01100,00111,00100,01000,01100, .... But
let me assure you that my getting into Bode's Rule was not a matter of choice.
Its rewards, though, have been a large number of empyreal highs, some over
ideas I later rejected; but now I am glad to be through with this whirlpool.
Finally, we all know that the idea of the "music of the spheres" which dates
back to Eudoxus is poetic license, nonetheless I could not help noting that
though most of the "notes" in my scale are cacophpnous, the first note, 2+v,
corresponds to C-sharp two octaves high, since (2+v)
= 225 .
REFERENCES
1. Johannes Kepler, Mysterium Cosmographicum
(Tubingen, 1597).
2. Johann Elert Bode, Vom dem neuen3
zwischen
Mars und Jupiter,,
entdeckten
achten Haupt-Planeten
des Sonnensystems
(Berlin, 1802).
3. M. A. Blagg, Mon. Not. Roy. Soc. Astr.,
Vol. 73 (1913), pp. 414-422.
4. W. E. Greig, The Distance
Rule in Planetary
and Satellite
Systems
(1974),
55 pp. (private circulation).
5. W. E. Greig, "The Reciprocal Period Rule," Bull. Amer. Astron.
Soc. , Vol.
7 (1975), p. 449.
6. W. E. Greig, "Bodefs Rule and Folded Sequences," The Fibonacci
Quarterly,
Vol. 14 (1976), p. 129.
7. W. E. Greig, "The Reciprocal Period Law," The Fibonacci
Quarterly,
Vol.
15 (1977), p. 17.
8. W. E. Greig, "Abstract of Reference," Zentralblatt
fur Matematik,
1978 (in
press).
9. I. J. Good, "A Subjective Evaluation," J. Amer. Statistical
Assn. , Vol. 64
(1969), pp. 23-49.
10. J. G. Hills, Nature,
Vol. 225 (1970), p. 840.
11. Owen Gingerich, Icarus,
Vol. 11 (1969), pp. 111-113.
12. J.L.Brady, Astron.
Soc. Pacific,
Vol. 84 (1972), pp. 314-322.
13. W. E. Greig, letter to H. W. Gould, dated 30 Aug. 1973.
14. M. G. Monzingo, "Extending the Fibonacci Numbers," The Fibonacci
Quarterly,
Vol. 12 (1974), p. 292.
15. H.W.Gould, "Cale's Rule," Proc. W. Va. Acad. Sci. , Vol. 37 (1965), pp.
243-257.
16. Anon., "Rings of Uranus," Science News,Vol.
Ill (1976), pp. 180, 245,52.
17. C. T. Kowal, "Chiron the Centaur," Science News, Vol. 112 (1977), pp. 311,
388, 409, 419.
18. J. L. Friend, Sky and Telescope,
Vol. 32 (1963), p. 329.
19. D. E. Thomsen, "Toro! Toro!," Science
News, Vol. 103 (1972), pp. 186, 173.
20. W. E. Greig, "On Sums of Fibonacci-Type Reciprocals," The Fibonacci
Quarterly,
Vol. 15 (1977), pp. 356-358.
21. Anon., "Jupiter!s 13th Moon," Science
News, Vol. 109 (1974), pp. 195 and
367.
22. W. E. Greig, Bull. Amer. Astron.
Soc,
Vol. 7 (1974) , p. 337.
23. N. A. Pananides, Introductory
Astronomy
(Reading: Addison-Wesley, 1973).
1978]
24.
25.
26.
27.
28.
FIBONACCI NUMBERS IN COIN TOSSING SEQUENCES
539
0. Struve, The Universe

(Cambridge, Mass.: MIT Press, 1962).
Brian Marsden, letter to the author dated 1976.
B. A. Read, The Fibonacci
Quarterly,
Vol. 8 (1970), pp. 428-438.
F.X.Byrne, Bull. Amer. Astron.
Soc., Vol. 6 (1974), pp. 426-427.
L. H. Wasserman, et al. , Bull. Amer. Astron.
Soc,
Vol. 9 (1977), p. 498.

MARK FINKELSTEIN and ROBERT WHITLEY
University
of California
at Irvine,
Irvine,
CA 92717
The Fibonacci numbers and their generating function appear in a natural

way in the problem of computing the expected number [2] of tosses of a fair
coin until two consecutive heads appear. The problem of finding the expected
number of tosses of a p-coin until k consecutive heads appear leads to classical generalizations of the Fibonacci numbers.
First consider tossing a fair coin and waiting for two consecutive heads.
Let 0n be the set of all sequences of H and T of length n which terminate in
BE and have no other occurrence of two consecutive heads. Let Sn be the number of sequences in 0n. Any sequence in 0n either begins with T, followed by
a sequence in 0n-i,
or begins with ET followed by a sequence in C n _ 2 . Thus,
(1)
Sn
= Sn-1
+ Sn-2,
Si
= 0, S2
= 1.
Consequently, Sn-i = Fny the nth Fibonacci number.

termination in n trials is Sn/2n.
Letting
^)
The probability of
Z2 "n^n>
and using the generating function (1 - x - x2) ~l for the Fibonacci numbers,
yields g{x) = x2/(1 - x - x2).
Hence, the expected number of trials is
J2nSn/2n
= (1/2)^(1/2) = 6.
n=l
We generalize this result to the following

Tk<lOKQjn'. Consider tossing a p-coin, Pr(E) = p, repeatedly until k consecutive heads appear. If Pn is the probability of terminating in exactly n
trials (tosses), then the generating function
(2)
G(x)
=Y,Fnxn
k
is given by G(x)
r
The expected number of trials, G (l)
(3)
1/p + 1/p + ... + l/p
(px)
1 - x + -^
U-px)
^-(px)k
+1
is
=r
'
P\
VtiOO^: Let 0n be the set of all sequences of H and T of length n which

terminate in k heads and have no other occurrence of k consecutive heads.
Let Sn be the number of sequences in 0n and Pn = Pr(On)
be the probability
of the event 0n.
One possibility is that a sequence in 0n begins wi h a T,
followed by a sequence in 0n_i\ the probability of this is
5^0

Pr(T)Pr(On^)
[Dec.
q = 1 - p.
= qPn.l9
The next possibility to consider is that a sequence in On begins with HT>

followed by a sequence in 0n-2\
this has probability
=
Pr(HT)Pr(On_2)
qpPn.2.
Continuing in this way, the last possibility to be considered is that a sequence in 0n begins with k - 1 #Ts followed by a T and then by a sequence in
(9n_k, the probability of which is qpk~1Pn_^.
Hence, the recursion:
(4)
Pn = qPn.
-- + ^ p ^ - 1 ? ^ ,
+ qpPn.2+
(Note that the probability of achieving k heads with k tosses is pk, while
with less than k tosses it is impossible.) The technique to find the generating function for the Fibonacci numbers applies to finding
00
G(X) = J^ Pn*"'
Consider
H{x) = ) Pn + 1xn;
n = k
then
xH(x) = Pn + 1xn
+1
n
= ^P]T
nx
Hence,
- Pkxk
(px)h
= G(x) -
H(x) = [G(x) -
(px) ]/x.
On the o ther hand,
HM = p n
k
+ 1tf
= ] T (qPn + qpPn_^ + + qpk~ xPn _k + Jx"

k
and r e c a l l i n g t h a t Pj = 0 f o r j < k,
=
^XX*" + <lPX12PnXn + + ^(px)^ 1 ^]?^"

fe
'
= ^[1 + P* + ... + (p*)*-M = ^ p x r ^ r ] Solving for (9 yields (2).

In the case p = 1/2, the combinatorial numbers Sn = 2 Pn satisfy the recursion Sn = _ i + 5 n _ 2 + ... + Sn_ic.
For these numbers, the generating
function (1 - x - x2 - - xk)~l
was found by V. Schlegel in 1894. See [1,
Chap. XVII] for this and other classical references.
An alternate solution to the problem can be obtained as follows. Consider
a sequence of experiments: Toss a p-coin Xl times, until a sequence of k - 1
heads occurs. Then toss the p-coin once more and if it comes up heads, set
Y = 1. If not, toss the p-coin X2 times until a sequence of k - 1 heads occurs again, and then toss the p-coin once more and if it comes up heads, set
Y = 2. If not, continue on in this fashion until finally the value of Y is
set. At this time, we have observed a sequence of k heads in a row for the
first time, and we have tossed the coin Y + X + X2 + + XY times. The X^
are independent, identically distributed random variables and J is independent
1978]
STRONG DIVISIBILITY SEQUENCES WITH NONZERO INITIAL TERM
5^1
of all of the Xi.

Let Ek = the expected number of tosses to observe k heads
in a row. Let Z = Z x + + XY. Then,
Ek = E(Y + Z) = E(Y)
+ E(Z)
= E(Y)
+ tf(Z|y = l)Pr(Y
= E(Y)
+ J2E(Z\Y
= 1) + #(Z|j = 2)Pr(J = 2) + . . .
n)Pr(Y
= n) = E(Y)
+ Y^rLE{Xx)Pr{Y
n=l
= E(Y)
= n)
n=l
E{XX)E{Y).
But E(Y) = the expected number of tosses to observe a head = 1/p, and E(X{)
E7;,.!. Thus S7?, = 1/p + (l/p)Ek_,
which yields (3).
REFERENCE
1.
2.
L. E. Dickson, History
of the Theory of Numbers, Vol. I (1919; Chelsea reprint 1966).
W. Feller, Introduction
to Probability
Theory and Its Applications,
Vol.
I (New York: John Wiley & Sons, 1968).
STRONG DIVISIBILITY SEQUENCES WITH NONZERO

CLARK KIMBERLING
of Evansville,
Evansville,
University
IN
INITIAL TERM
47702
In 1936, Marshall Hall [1] introduced the notion of a kth order

divisibility
sequence
as a sequence of rational integers uQ, u19 ,.,,un9
satisfying a linear recurrence relation
(1)
un
+k
= alun+k_l
+ .- - +
linear
...
akun9
where a19 a2, . . . , ak are rational integers and um\un whenever m\n.
Some divisibility sequences satisfy a stronger divisibility property, expressible in
terms of greatest common divisors as follows:
(um,
un)
= U(m^ n )
for all positive integers m and n.

bility
sequence.
An example is the
It is well known that for any
sequence {un} is periodic modulo m.
M depending on m and al9 a2, . . . 9 ak
(2)
un+M
E un
We call such a sequence a strong

divisiFibonacci sequence 0, 1, 1, 2, 3, 5, 8,... .
positive integer m9 a linear recurrence
That is, there exists a positive integer
such that
(mod m)
for all n >. n0[m9 a19 a2, . . . , ak] ; in particular, nQ = 0 if (afe, w) = 1.

Hall [1] proved that a linear divisibility sequence {un} with u Q ^ 0 is
degenerate
in the sense that the totality of primes dividing the terms of
{un} is finite. One should expect a stronger conclusion for a linear strong
divisibility sequence having u0 4- 0. The purpose of this note is to prove
that such a sequence must be, in the strictest sense, periodic. That is,
there must exist a positive integer M depending on al9 al9 ..., ak such that
^n + M ~ ^n>
n \J 9 I, ...
5^2
[Dec.
Suppose {un} is a kth order linear strong divisibility sequence. In terms

of a generating function for {un}, we write
U
U ;
+ U
ir+
U V
K(t)
(1 - Xit)a
- x2t)
-. (1 -
where H(t) and K(t) are polynomials with integer coefficients.

. . . xk (=ak) . We assume that q ^ 0.
um\qmu0
Lmma 1:
Vh.00^:
xkt)'
Let q = x x
for w = 1, 2, ... .
The 0th ^-multisection of (3) (e.g., Riordan [2]) gives
where the Mi are integers.
Since um\uQm
for <3 = 1, 2, ..., we have
and this finishes the proof, because Mk = qm.

Another proof of Lemma 1, depending on the periodicities (2), may be found
in Hall [1].
Henceforth, we assume uQ ^ 0. Let p 1 , p 2 , . . . , pv be all the prime divisors of quQ9 so that we may write
q= p ^ p s z '
Then, since um\qmuQ
... pys and uQ = pjl- p*'-' ... p*- .
for m = 0, 1, 2, ..., we can write
"m = Pi1" Pi'"
?""" , m = 0, 1, 2, .... .
Consider the set o = { t l5 ij 2 , . . . } , = 1, 2, ..., i?. Let |a | be the

number of elements in C , with j 0" ^ j = if G is an infinite set. Define
az(j)
for j = 1, 2, ... inductively as follows:

a(l) = 1
(l if K | = 1
a(2) =
least w such that iliW
\a%U
a
(j)
, if | cf | > 1
~ 1) if \z\ l i - 1
~ J least w such that i ) W Ui,ai{r)
:1 <. r < j -
l\
(if K l > j - 1.
Thus, either the sequence a (l), a (2), a (3), ... is strictly increasing and
unbounded, or else it is strictly increasing up to some point and constant
thereafter, or else it is the constant sequence 1, 1, ... .
Lemma 2:
Suppose I <_ <v.
Then a(j)|a(j + 1) for j = 1, 2, ... .
VK00^\ T O simplify notation, let a = a (j), b = &iU + 1) > a n d c = (a, b) .

Without loss we assume a ^ b. Clearly c _<_ a. Suppose 1 <_ e < a. Then i,c =
ti,as(r)
for some r < j , so that i , c ^ ^, a and i%t0 i i%,b- From u c = (ua,
=
Ub) follows iii0
niin{tj()a, ii,b}>
This contradiction shows that c = a, as
required.
LmmOL 3:
Suppose 1 <. I < V and j > 1.
If 1 <_ w <_ a(j) = a, then
1978]
VKooi'. I f 1 <. W <. a, then il}W = iZiai(r).

for some v < j .
by Lemma 2, we have ua^r)
\ua3 so that ii>ai(i>)
<_iziaSuppose 1 <_ I <_ v and j >_ 1.
Lemma, 4:
Vtiooi:
(uW9 ua) = U(Wta)9
so mn{iZtW,
by Lemma 3, so iZi(<Wya)
= î,w
Lmma. 5: Suppose 1 <_ <_'z; and j >_ 2.
positive integer. Then
(ti,b
a +
i,
t [ , k + 2j
Since az(r)
If 1 <. w <. aîj)
ig,a}
Suppose a = az(j)
Â.fca + a-l)
(^Z,ly
\a9
= a, then
Now
= ii,(w,a)
5^3
iitWiita,
>_ 2 and H s a
^ 1, 2 > s ^ , a - l )
VhOO&i Suppose- 1 w ^ a - 1. Then (w^ a+w , u a ) = U(ba + W,a) = "(w,a) so

min{^,foa+ w , ii,a) = ^,(w,a) = ^,w by Lemma 4. Since i > w < i jCr by definition of a, we conclude in h ,
Lemma 6: Suppose 1 <_ <_ v and 2 <_ |a | < <. Let L = a ( | a ^ | ) , and let
b be a positive integer. Then
(^,W
Vh.00^1
+ l ) ^l,bL
+ 2 J
^ , 2 L - l ) = (^Jl, 1 ^ , 2 J J
^Jt,2>L-l)
By Lemma 5, we already know
(^,2L+1 s J ^ , 3L-1 )
(^.(fc-DL + l ,
---J ^,L-l)5
so it remains only to see that -#,, = ii,2L

= = ii,(b-i)L'
2> - 1, we have (w c L , w L ) = u L , so that m i n { i > c L , i j L } = *,,L".
< i i i L , we conclude i^ 5oL = ilyL*
LQMnCL 7: There exists a positive
1 , 2 , ..., fc.
For
1 <_ <2 <.

Since i ^ c L
integer M such that w MHJ

+,
P/LOO^: For 1 <. I <^ v, if \oz\ = , choose j so large that a (j ) > k9

and if |G | < , let az(jz)
= a e ( | a | ) . Let M be the least common multiple
of the numbers a1(j1)
, a2(j2^
j^y(j y ) 2^- (We include 2fc to ensure that
M >fcin case \oz\ < for all .)
Now, by Lemma 5, for each I with | G | = , we have
This same equation holds, by Lemma 6, for each with 2 _< [G^l < , and clearly
holds also for d = 1. Therefore, for 1 _< j _< k, we have ii,M + j = ^,7- f r
1 _< <. V, so that u M+J- = Uj for 1 <_ j _<_ /c.
Tfieo/Lem: Suppose { u n } , n = 0, 1, . . . , is a kth. order strong divisibility
sequence with uQ 0. Then the sequence { w n } is periodic and has a generating function of the form H(t)/(1
- tp), where p is the fundamental period of
{un).
If H(t) has no linear factor of the form 1 - rt, where rp = 1, then p
is the least possible recurrence order of {un}. If
P
Pl
'p2
...
pt'
if the prime factorization of p, then

Up = UUê, Upe?
. . . UQet
5hk
[Dec.
MINIMUM PERIODS MODULO n FOR BERNOULLI NUMBERS
for some nonzero integer U.
Finally, u0 = up,
and un\u0
for n = 0, 1, ... .
VK.00^'. By Lemma 7 and the fact that {un} is a kth order recurrent sequence, the sequence {un) is periodic with period M. Letting p be the fundamental period, we now show that the denominator of the generating function
H(t)/K(t)
must be of the form 1 - tp:
Y^r
+ + Up.^"1
= u0 + uxt
= u0(l
+ t
= (u0 + uYt
+ t
2p
+ uQtp
+ u1tQ
p
+ ...) + uxt(X
+ t
p 1
+ + Mp_it " )(l + t
= (w0 + w x t + ... + Up^t9'1)-
+1
+ t
+ t
2p
2p
+
+ ...) +
+ )
.
A. "0
If H(t) has no linear factors 1 - rt with rp = 1, then H(t)

has no linear
factors in common with K(t).
This means that no recurrence order for {un\
can be less than p.
We see that pf<: |p and (pfS p Sj ) = 1 for 1 _< i < j <_ t, so that
Up
= tfW 8l U
fll
. . . U et
for some integer U. For n 2>_ 1> we have w n p = up and w n |^ n p , so that

That wo = Up, so that un\u0
for all n, follows from
un\up.
aû0 = uk - <^2w/c-i - - afcu2

= Mp+fc- ^2 U P+k-l ~ "" ~ akuP + l
akup.
REFERENCES
1.
2.
Marshall Hall, "Divisibility Sequences of Third Order," Amer. J.

Math.,
Vol. 58 (1936), pp. 577-584.
John Riordan, Combinatorial
Identities
(New York: John Wiley & Sons,
1968).

Technische
W. HERGET
Braunschweig,
Universitat,
Fed.
Rep.
Germany
The Bernoulli numbers Bm may be defined by
(1)
*-d^2( n :>
(m>o)
i =0
By the Rummer congruence, we have [2, p. 78 (3.3)],
(2)
(-iy
(r.)î^
= oiodp",
with u: = pe~1{p
- 1), where r >_, e >_ 1, m > re, p prime such that p - 1^/7
With r = 1 we get, in particular
1978]
(3)
mod p e ,
=
m+y6-1^
5^5
- 1)
where m > e, p - \m.

Therefore, the sequence of the Bernoulli numbers is periodic after being
reduced modulo n (where n is any integer) in the following sense. A rational a/b with a, b e Z, gcd(a, b) = 1, may be interpreted as an element of Z n ,
the ring of integers modulo n, if,and only if the congruence relation yb = a
mod n has a unique solution y e {0, 1, 2, . .., n - l} , i.e., if and only if
gcd(b, ri) = 1. In this case, a/b is said to be
n-integral.
By the famous von Staudt-Clausen theorem we have for integer i and prime
p (cf. [1] and [2]),
B2i
p - i n t e g r a l =>p -
l\li.
Since B0 = 1, B1 = - 1 / 2 and B 2 i + 1 = 0 f o r i e N, we g e t
(4)
Bm p - i n t e g r a l <=^>p - l)(mVm = OVw ( 3 , 5 , 7, . . . } .
Now let Lin)

erty:
be the smallest integer greater than 1 with the following prop3 77?0 V/c, m >_ m0:
(5)
fSj, n-integral A /c = 7?7 mod L(n)
^ Bm ?z-integral f\Bk
E 5 m mod n ) .
L(n) is called the period-length

of the sequence {Bk mod n}.
The smallest possible integer mQ in (5) is then called the pvepeviod
\Bk mod n} and will be denoted by Vin) .
If n = n 1 n 2 , where ni, n 2 are coprime, then clearly
L(n) = lcm(L(ni), L(n2))
and
7(n) = max(7(n1), F(n 2 )).
Hence, it suffices to discuss the case n = pe , p a prime.

Thzo/im
J:
of
We will prove
(a) L(2 ) = L(3 ) = 2

(b)
F(2e) = F(3e) = 2
(c)
L(pe) = pe(p
(d)
V(p )
- 1), where p > 3
<.e + 1.
VK.00^'. If l\n or 3|n, none of the B2i is n-integral by (4); since B2

= 0, this proves (a) and 7(2e) , V(3e) <_ 2. But 7(2e) = 1 and V(3e) = 1, respectively, is impossible because Bl = -1/2 is not 2-integral and Si ^ 0 mod
3 e . So we get (b) too.
Now let p > 3. From (3) we have, for m > e, p - l)fm, t >_ 0,
(6)
+ p e - 1 ( p - 1)
ni
1
1 j _ moc | pe. hence,
m + tpe (p - 1)
k = m + spe(p - l)Ap --\mhm
> e =^Bk ~ Bm mod p e .
Consequently, L(pe)|pe(p - 1). On the other hand, we first prove p - 1 \ L ( p e ) :

suppose p - l)(L(pe);
we may choose 7?7 >. V(pe) + L(pe) such that p - 1|777 (and
therefore m f 0 and 7?? (3, 5, 7, ...}), hence by (4) Bm is not p-integral.
For k: = m - L(pe) , we have k = m mod p e , k >_ V(pe) and p - l\k, hence by (4)
Bk is p-integral. But this is a contradiction to (5). So L(pe) = pl (p - 1)
where t- e {0, . . . , e).
It remains to show % = e.
For this, we choose q e N
such that s: = (qp(p - 1) + 2)p^ > V(pe).
Because pe|s and p - l\s, we have
546
Bs - 0 mod pe [2, p. 78, Theorem 5].

mod pe if k = s mod p.(p - 1). Take
Now suppose i < e.
[Dec.
Then, Bk = Bs l
k: = s + (p - l ) p * = (2 + (<7p2 + 3 ) ( p - D ) p * = 2 + t(p - 1) ,
where
t : = 2 r _ ^ + (qp* + 3 ) p i e ff;
P
t h e n by (3) w i t h e = 1 and m = 2 ,
52 _
B2 + ( p - l )
2 + (p - T )
_ Bk
_
=
" = x
m o d p
'
where Bk E 0 mod p . But, p |s and p \(p - l)pl gives p 6 ^ and, therefore,

B2/2 E 0 mod p, contradictory to 5 2 = 1/6. Hence, i = e holds, and thus
e
L(pe)
= pe(p - 1)
and F(pe) <_ e + 1
by (6).
Now we may improve this last inequality as follows:
Th&QJim 2:
1. V(p) = 2 for p prime.
2. Let p be a prime, p > 3 and e e (2, 4, 6, . . . } . Then,
(a) Be f 0 mod p A p - l f e =>V(pe) = e + 1.
(b) fc maximal such t h a t
V0<.i<.k:
(5e.2i E0 modp2i +1 V p - l | e - 2t)

^V(pe)
3.
= e - 1 - 2k.
Let p b e a p r i m e , p > 3 and e e { 3 , 5 , 7 , . . . } . Then,

(a) B e -i ? 0 mod p 2 A p - l\e - 1 => V(pe) = e.
(b) & maximal such that
E 0 mod p2i + 2\/p - l\e - 1 - 2)
V O i <_ k: {Be_l_2i
=>F(pe) = e - 2 - 2k.
VKOOl: By Theorem 1(d), we have V(p) <. 2. But V(p) < 2 is impossible
since B1 - -1/2 ^ 0 mod p and S 1+ L ( P ) = 0, thus V(p) = 2.
For the proof of the other assertions we note that [4, p. 321, Cor.]:
Y,(-iy(l)Bm + iv(l
- p -i + <vj = O B o d p ' C " ) - 1 ,
i =0
where p prime, p ^ 2, p - l|v, and p w is the highest power of p contained in

V.
Setting r: = 1 and v: =fc- #?, we get
5^(1 - p m _ 1 ) - Bk(l - pk-x) E 0 mod pe,
where pe(p - 1) \k - m and k >_ m >_ 1. Because
k - 1 >_ m + pe(p - 1) - 1 >_ pe(p - 1) ^ 3 e 2 >_ e,
we h a v e , f o r k > m >_ 1 , p - l/fw:
(7)
fc
E m mod p e (p - 1) =>Bk - Bm E p " 7 - 1 ^ mod p e .
Now it is easy to verify the assertions.

It is not very difficult to derive the following corollary, which gives
the value of V(pe) "explicitly" for regular p (a prime p is said to be regular if and only if Bk f 0 mod p for each k e {2, 4, ..., p - 3}.
1978]
5^7
Cosiollasiy 7: Let p be r e g u l a r , p > 3 and e > 0.

(a) I f 2\e t h e n
F(pe) = e + l < ^ > p | e A p e
7 ( p ) <. e - l < ^ = > p | e V p e
F(p ) e
- 3 <=> (p\eAp
^ e
l|e
l|e
- l\e
= 2 p mod p(p
- 2 ) V ( p - l\eAp3\e
- 1) V e = 2 - 2p
- 2)
3
mod p 3 ( p - 1)
F ( p e ) = e - 5 =>p = 5 A e E 252 mod 500

V(pe)
(b)
If 2 | e
>. e - 5 .
then
F ( p ) = e <=^p2\e
- l A p - \\e
- 1
F ( p ) e - 2 < = p | e - l V p - l\e
e
- 1
7 ( p ) <. e - 4 < ^ > ( p | e - l A p - l\e

2
<=> e = 2p
h
= -2p
- 3) V (p - l\e
2
+ 1 mod p ( p -
- hph\e
- 3)
l)Ve
+ 3 mod p ^ ( p - 1)
V(pe)
= e - 6=^p = 5 A e E 1253 mod 2500
V(pe)
>_ e -
6.
For the proof, note that l\V{pe)

p and p - \\li,, we have
holds for e > 1 and that in case of regular
S 2 i E 0 mod p e =^>pe|2i.
The assertions of Corollary 1 with "<==" are also valid for any irregular
prime p.
By Corollary 1, you may see that only for greater integers pe, the value
e
V(p ) differs from e and e + 1, respectively. We get
CoKolLcUty 2: For prime p , p > 3 , l e t e
2 p 2 + l , e5 = 252, e6 = 1253. Then we have
(a)
F(peO < e
- i ,
= p - l ,
e2 = p ,
e3 = 2p,
e^ =
{l, . . . , 4}.
If p is regular, then V(pei)

= ei - i, i e {l, . . . , 4 ) , and there is
no smaller power of p such that V(pe) = e - i.
(b) 7(5&t ) = e{ - t, i e {5, 6}, and there is no smaller power of 5 such
that V(5e) = e - i .
(c) If p is regular and p > 5, then V(pe) >_ e - 4.
For irregular primes, it is naturally somewhat more difficult to derive
similar results about the smallest power of p such that V(pe) = e - i , where
i >_ 1. By Theorem 2, we get
Be E 0 mod pA2|e => 7(pe) <. e - 1;
hence, for each irregular prime p, we have V(pe) <L e - 1 for at least one e
such that e e = p - 1.
Considering the table of irregular primes in [1] we may compute that n =
691 2 is the smallest power of an irregular prime such that V(pe) = e - 1.
There are still some open questions:
1. Are there powers n - pe of some (necessarily irregular) prime p such
548
THE RANK-VECTOR OF A PARTITION
2.
[Dec.
that e < et and V(pe) < . - , where i e {-2, 3, 4}? (By the computational results in [5] we may conclude that this does not happen when
p < 30,000.)
Is there a power n = pe of some irregular prime such that
V(pe)
<e-5?
VAJIGJL RojncUik: Professor L. Carlitz and Jack Levine in [3] asked similar
questions about Euler numbers and polynomials. Analogous results about the
periodicity of the sequence of the Bernoulli polynomials reduced modulo n and
the polynomial functions over Z generated by the Bernoulli polynomials will
be derived in a later paper.
REFERENCES
1.
2.
3.
4.
5.
Z. I. Borevic & I. R. Safarevic, Number Theory,

"Nauka" (Moscow, 1964; English trans, in Pure and Applied Mathematics,
Vol. 20 [New York: Academic
Press, 1966]).
L. Carlitz, "Bernoulli Numbers," The Fibonacci
Quarterly,
Vol. 6, No. 3
(1968), pp. 71-85.
L. Carlitz & J. Levine, "Some Problems Concerning Rummer's Congruences
for the Euler Numbers and Polynomials," Trans. Amer. Math. Soc. , Vol. 96
(1960), pp. 23-37.
J. Fresnel, "Nombres de Bernoulli et fonctions L p-adiques," Ann.
Inst.
Fourier,
Grenoble,
Vol. 17, No. 2 (1967), pp. 281-333.
W. Johnson, "Irregular Prime Divisors of the Bernoulli Numbers," Mathematics of Computation,
Vol. 28, No. 126 (1974), pp. 652-657.

Pan jab
HANSRAJ GUPTA
University,
Chandigarh,
1.
India
INTRODUCTION
The Ferrars graph of a partition may be regarded as a set of nested right

angles of nodes. The depth of a graph is the number of right angles it has.
For example, the graph
T
T
e
o
)
)
c)
<)
1978]
5^9
is four deep or is of depth four. It is clear that a graph of depth k cannot have less than k nodes.
Denote by x the number of nodes on the horizontal, and by yi the number
of those on the vertical section of the ith. right angle, starting with the
outermost right angle as the first. Then, the partition can be very conveniently represented by the 2 x k matrix:
^2
^3
ys
hi
or simply by
.yi
Evidently, we must have
(1.1)
xi
>_ xi
+1,
+1
_> yi
+1
+ 1,
i <_ k - 1.
It must be remembered that x s and y s are positive integers. The Atkin-ranks

of the graph [1] are given by
(1.2)
ykl
y^
yi<.
= I** -
yih>
which we shall call the rank-vector both of the graph and of the partition it
represents.
The number of nodes in the graph is given by
^
(1.3)
+ yi
(xi
- 1).
In our graph, the matrix

"7
2"
.9
1.
represents a partition of 30 and its rank-vector is

[-2
1].
Obviously, if Rk is the rank-vector of a partition, then the rank-vector of

its conjugate partition is -i?/<. Hence, the rank-vector of a self-conjugate
partition of depth k must be [0]k.
_ _.
Again, if [rAk
is the rank-vector of the partition given by /
, then
we have
(1.4)
yi
2.
SOME CONSEQUENCES OF (1.1]
Since yi >_ y^ + 1 + 1, we must have x^ - v >_ xi

each i <_ k - 1,
(2.1)
Since yk
(2.2)
Xi >_ max(xi+ 1 + 1, xi
+ ri
+1
+1
+ 1).
is a positive integer, we conclude that

xk
>_ max(rk
+ 1 , 1).
From (1.3) and (1.4), we further have

(2.3)
- ri
Y^Xi = Un + k
i=i
+Y,rA.
i=i
+1
+ 1.
Hence, for
550
[Dec.
Hence a partition of n with a given rank-vector [ 2 ^ ] ^ can exist only if n has

the same parity as
k
+ E rt
i= 1
In what follows, w e assume that our n T s satisfy this condition.

shall invariably run over the integers from 1 to L
3.
Moreover, i
THE BASIS OF A GIVEN RANK-VECTOR
There are an infinite number of Ferrars graphs which have the same rankvector. All such graphs have the same depth but not the same number of nodes
necessarily.
ThJL0h&m\
Among the graphs with the same rank-vector,
with the least number of nodes.
there is just one
VJWOfa: Using the equality sign in place of the sign _>_ in (2.2) and ( 2 . 1 ) ,
we obtain the least value of each of the x 's, i k.
(1.3) and (1.4) then
give n0 that is the least n for which a graph with the given rank-vector e x ists. This proves the theorem.
Incidentally, w e also get the unique partition with the given rank-vector
and the least number of nodes. W e call this unique partition the basis of
the given rank-vector.
Example:
Let us find the basis of the rank-vector [-2 3 0 1 ] . With
the equality sign in place of the of the inequality sign, (2.2) gives xh = 2.
With the equality sign in place of _> , (2.1) n o w gives, in succession,
= 3, x2 - 7, and x1 = 8.
x3
From (4) of Section 1, we now have

= 1, y3 = 3, y2 = 4, and yl
yh
= 10.
Hence, the required basis is

"8
2"
.10
1_
This represents a partition of 34.

We leave the reader to verify the following two trivial-looking but very
useful observations:
(a)
If
is the basis of [r3^] and h is an integer, then the basis of
the vector [v^ + h] is given by
+ K
y<i
lyi
according as h is positive or negative.
(b)
If hi >_ h2 _> ... >_ hk >_ 0 are integers, then the graphs of
XJS
+ 7z,-l
and
.Mi
Mi
have the same rank-vector.
7*1
1978]

PARTITIONS OF n WITH A GIVEN RANK-VECTOR
k.
Let
551
M be the basis of the given rank-vector and n 0 the number of nodes
in the basis. For our n to have any partitions with the given rank-vector,
it is necessary that n has the same parity as n0 and n>_nQ. Assume that this
Write
m = j(n
- n0).
List all the partitions of m into at most k parts. Let
m = hl + h2 + + hk,
with hi >_ h2 >_
L b>k O J t> e a
y s u c n p a r t i t i o n of m.
Then the matrix
+ hi
(4.1)
2/t
+ /z,
provides a partition of n with the given rank-vector.

The one-one correspondence between the partitions of m and the matrices
(4.1) establishes the following
Tfieo/LCm: The number of partitions of n with the given rank-vector is the
same as the number of partitions of m into at most k parts where m is as defined above.
Example: Let the given rank-vector be [-3
the basis of the vector is readily seen to be
-1] and n = 43. Then
" 7 6 4 1 "
_10
2.
so that n 0 = 33 and m = 5.
The partitions of 5 into at most 4 parts are:
5; 4 + 1 , 3 + 2 ; 3 + 1 + 1 ,
2+2+1;
2+1+1+1.
Therefore, the r equired partitions o E 41

)a
[12
Ll5
l2 > 1"11
J-4
"10
-13
2_
10
2,
13
l"
2.
12
J-2
We leave it to the reader to see how the graphs of partitions of n can be

constructed directly from that of the basis. As an exercise, he/she might
also find a formula for the number of self-conjugate partitions of n.
As a corollary to the theorem of this section, we have
CofiolZa/iy:
The number of partitions of n + hk, h > 0, with rank-vector
[v. + h] is the same as the number of partitions of n with rank-vector \vi ].
This follows immediately from observation (a) in the preceding section.
5.
THE BOUNDS FOR THE ATKIN-RANKS
What can be said concerning the Atkin-ranks of partitions of n for which
THE ANDREWS FORMULA FOR FIBONACCI NUMBERS
552
[Dec.
We show that these ranks are bounded both above and below. Since x1 <. a,
the number of rows a partition of n can occupy is not less than u, where
u - 1 < n/a <_ u.
Hence, none of the ranks can exceed (a - u).
Similarly, none of the ranks can fall short of (v - b), where
v - 1 < n/b <_ v.
Of course, for n to have a partition of said type, it is necessary to have
n <_ ab,
REFERENCE
1.
A. 0. L. Atkin, "A Note on Ranks and Conjugacy of Partitions," Quart.

Math., Vol. 17, No. 2 (1966), pp. 335-338.
J.

HANSRAJ GUPTA
Panjab
University,
Candigarh,
India
7. In what follows: small letters denote integers; n > 0; p denotes an

odd prime other than 5; [ ] is the greatest integer function; and for convenience, we write
(n;p) for [J .
The two relations
(1.1)
(n;p) = (n;n - r), and
(1.2)
(n;r - 1) + (n;r) = (n + l;r)
are freely used, and we take, as usual,

(t;0)
= 1 for all integers t, and
(n;p) = 0 if v > n, and also when v is negative.

We further define
(1.3)
S(n,r) = Y,(n;j),
J
where j runs over all nonnegative integers which are E p (mod 5 ) .
As a consequence of this definition and the relations (1.1) and (1.2) we
have
(1.4)
S(n,r)
(1.5)
S(n9r
2.
=S(n,n
- r), and
- 1) + S(n,r)
= S(n + l,r).
The Fibonacci numbers Fn are defined by the relations
(2.1)
Fl = 1 = F29 and
(2.2)
Fn + Fn + 1 = Fn + 2 for each n >_ 1.
1978]
553
G. E. Andrews [1] has given the following formulas for F :

<2'3)
n =H
(-l)'(n
- 1 - 5j)/2]);
- l;[(n
(2.4)
(-DJ'(n;[(n
Fn = Y,
- 1 - 5j)/2]);
where j runs over the set of integers.

The object of this note is to provide a simple proof of these formulas
and to obtain some congruence properties of Fn . Let
[(n - l)/2] = m (mod 5)
so that
n - 1 = 2m
or
2m + 1 (mod 10)
according as n is odd or even.

(2.5)
Fn = S(n - l9m)
(2.6)
Fn = S(n9m)
Then (2.3) and (2.4) can be written as:
- S(n - l,m
- 2);
- 1).
- S(n9m
We first assert that (2.5) and (2.6) are equivalent and prove the assertion as follows:
For any integer j , we have
(n;m + 5j) - (n - ;m + 5j) = (n - l;m + 5j - 1).
Also
(n;m - 1 + 5j) - (n - l;m - 2 + 5j)
= (n;n
- m + 1 - 5j) - (n - l;n
- m + 1 -
5j)
- m - 5j)
= (n - l;n
= (n - ;m + 5j - 1).
Hence, letting J vary suitably, we get
S(n,m)
- Sin
- l,m)
- 1) - S(n - l9m - 2 ) ,
= S(n,m
and our assertion follows immediately.

3. Proof of (2.5) is by induction. It is easy to verify that (2.5) and
(2.6) hold for n = 1 and n = 2. Assume that they hold for each n <_ t + 1.
Then, from (2.6), we have
(3.1)
with m = [(t
(3.2)
Ft = S(t,m)
- l)/2] (mod 5).

Ft+1
(3.3)
- 1)
- S(t9m
= S(t,m)
= S(t9m
For the same TT?, (2.5) gives
- S(t9m
- 2) for t odd,
+ 1) - S(t9m
- 1) f o r t e v e n .
I f t i s odd, l e t t = 10fc + 2777 + 1 ; t h e n

(3.4)
S(t9m)
= S(t9t
- m) = S(t910k
+ m + 1) = S(t9m
+ 1).
I f t i s e v e n , l e t t = 10/c + 2TT7 + 2 ; t h e n
(3.5)
S(t9m
- 1) = S(t9t
-777 + 1) = S(t910k
+ m + 3) = S(t9m
- 2);
so t h a t
(3.6)
Ft + 1 = S(t9m
+ 1) - S(t9m
- 2) f o r t odd as w e l l as t e v e n .
554
THE ANDREWS FORMULA FOR FI BO ACCI NUMBERS
[Dec.
From ( 3 . 1 ) and ( 3 . 6 ) , we g e t
Ft + Ft + 1 = {S(t,m)
= S(t
+ S(t9m
+ 1)} - {S(t,m
+ 1,777 + 1) - S(t
- 1) + S(t,m
2)}
+ 1,777 - 1 ) .
Thus,
F
t + 2 = S(t
+ 1,777 + 1) - S(t
+ 1,777 - 1 ) .
Inductive reasoning now proves (2.5) for all n > 0.

4. From (2.5) and (2.6), we can derive not only the well-known congruences modulo p for Fp , Fp + i, and Fp-i (in the manner of Andrews) , but also
some congruences modulo p2.
We first give the expressions for Fp2 , Fp2 + i , and F p2 _i.
(i)
If p is a prime of the form 10k 1, then we have

[(p2 - l)/2] E 0 (mod 5 ) ,
and so also
[p2/2] = 0 (mod 5 ) .
Hence,
= S(p 2 , 0) - S(p 2 , 4 ) ,
Fp2
V +i
and t h e r e f o r e ,
5(
P2'
0)
5(
"
P2'
3);
F p , . ! = 5 ( p 2 , 4) - 5 ( p 2 , 3 ) .
(ii)
I f p i s a prime of t h e form 10k 3 , t h e n

[ ( p 2 - l ) / 2 ] E 4 (mod 5 ) ,
and so a l s o
is
[ p 2 / 2 ] E 4 (mod 5 ) .
Hence
F p 2 = 5 ( p 2 , 4) - 5 ( p 2 , 3 ) ,
*Wi
and
= (p 2 > 4) - ^ ( p 2 ,
);
therefore,
= 5(
^2'
3)
Lmmcii
For l ^ p
"
5(
2)
P2'
All that we need now for our purpose is the
*V-i
-l,
(p ; h) E ( - l ^ ' V A (mod p 2 ) .
P/LOOfJ:
We have
2.
*-P '
(p
fe)
n>
j . P2 - 1 . P2 - 2
h
1
2
Since f o r 1 <. r ft - 1 ,
^-f^t h e lemma f o l l o w s
= -1 (mod p 2 )
immediately.
p 2 - ft + 1
h - 1
1978]
OPERATIONAL FORMULAS FOR UNUSUAL FIBONACCI SERIES
555
Evidently, if pj[h9 then

(p2;h)
(4.1)
E 0 (mod p 2 ) ;
otherwise,
(p2;h)/p
(4.2)
5 (-D^p/fc (mod p ) .
We have, of course,
(p2;p2).
(p2;0) = 1 -
(4.3)
As an application of the lemma, we have, for example:

(i) when 1 _ m j 4,
(4.4)
S(p\m)
= Y, (P2;^ + 5j) (mod p 2 ).

j'>.o
On the right of the sigma in (4.4), we need consider only those nonnegative
values of j for which
m + 5j <_ p 2 and 77? + 5j E 0 (mod p) ;
(ii)
when 77? = 0 , we h a v e ,
S(p2,0)
(4.5)
so t h a t
(4
*(p2,0) -
.6)
(p2;5j)
- 1 =
(mod p 2 ) ,
~1
E^
(-l^-VSj
(mod p ) ,
where 1 <_ j < p / 5 .
Thus
^121 ~ 1
1
1
1
1
11
^ - ^ + t - i E
(mod 1 1 ) .
Therefore,
Fl21
E 89 (mod 1 2 1 ) .
REFERENCE
1.
G. E . A n d r e w s , The Fibonacci
Quarterly,
Vol. 7, No. 2 (1969), p p . 113-130.

H.
West Virginia
University,
W. GOULD
Morgantown,
W.Va.
26506
Operational formulas can play a fascinating role in finding transformations and sums of series. For instance, by using the differential operator
D (=d/dx) we can transform
(1)
2 , * *
k = 0
into
YJkxk~1
ri
(1 - x)2
556
[Dec.
The operator 0 = xD is even more interesting. It has the basic property that
0pxk = kvxk, so that (1) can be transformed into
<2>
S* v - 6 '{5-M-
and since it can also be shown (and is well known) that

P
(3)
epf(x) = Sip, k)xkDkf(x),
k =0
where Sip, k) are Stirling numbers of the second kind, explicitly
(4)
k\S(p,
k) = A*0* = (-D""J'(5)jp5
X J /
j-o
then series (2) can be found in closed form for it is trivial to find the
higher derivatives needed in (3). The result is a very old and well-known
formula. In [7] is given an extension of (3) applied to generalized Hermite
polynomials. There are numerous similar generalized expansions involving the
D operator. Here we propose to examine some rather unusual variations that
are not too well known, and which have applications to Fibonacci numbers
among other things.
We shall need several, other well-known operational formulas whose proofs
involve some calculus and/or mathematical induction, and we tabulate these
below:
(5)
9 = Dz, where x = ez 9
(6)
(7)
e" = D?,
xD-nl(Dz),
where the binomial coefficient is defined as usual by ( J = x(x - 1) ...

V
(x - n + l)/n!, with (g) = 1.
'
(8)
e D = 1 + A = E,
where
A/(a:) = fix + 1) - f{x)
and Ef(x) = f(x + 1) .
More generally
(9)
etD* = f(x + t) =
Efix).
x ,h
The ^-operator
(10)
fiqx)
= Qf(x),
where Q = q* .
This was used, e.g., in [10], and is very convenient when working with basic
hypergeometric series.
In the references at the end are several papers, viz. [1], [2], [4], [5],
from the older literature where properties of a great number of familiar and
unfamiliar operators were developed. The master calculator was almost certainly George Boole. The English literature for the period from about 1830
to 1890 is especially rich in papers on unusual operators.
In [1], Boole gave the pair of very remarkable operational expansions
(11)
f(x + 0riD))uix)
e^D)fix)e-*{D)uix),
f(D + 0rix))uix)
e-4(x)f(D)e0(xy>uix),
and
(12)
1978]
557
which hold for arbitrary functions /, 0, and u.

The formulas are certainlytrue for polynomials, and in order to avoid matters of convergence of any
series we shall explain that we interpret these as statements about formal
power series. In that context there is no difficulty and we use formal power
series definitions of all operators. Thus, if L is a linear operator, we
should like to define eL by
*L-t,WLK
(13)
k=o
'
Boole's formulas (11)-(12) have a bearing on expansions in [7]. They are

representative of some of the most unusual operational formulas.
But stranger still, we shall consider the operator LL, which we define as
follows:
(14)
LLf(x) = {(L - 1) + l}Lf(x) = E(^)( - D"fW

n=0
= ZEÊ(-i)- (^ k /(x)
n =0 j =0
k =0
where Cj are Stirling numbers of the first kind, i.e., coefficients in the
expansion of a binomial coefficient:
j =0
In the familiar notation of Riordan, nlCj = s (n, j).

For a particular choice of L we may be able to give a more compact definition. Thus, with / =
f(x),
DDf={(D-l)+l}Df=Y,(Dn)(D-l)nf
(16)
n =0
=Z
f r ^ A c - 1)"/, by (7), with z = e*,
= ESt(-D"-k(?)o^).
n =0
' k =0
For an example of this expansion, let f(x)

whence
n=0
n=0
" k=0
k=0
k =0
= eax.
Then
558
OPERATIONAL FORMULAS FOR UNUSUAL FIBONACCI
= zaJ2{n)(a " 1)n
= zCCaa =
SERIES
[Dec.
aae<ZX
>
n=0
so that w e have the attractive formula
DDeax
(17)
aaeax,
It is instructive to compare this with Qpx = kpx , and to recall a little

terminology from vector analysis. A characteristic vector for a linear transformation L is a non-zero vector / such that Lf - of for some scalar c. With
each operator w e like to find a natural function or characteristic function.
For the operator 0 it is xk9 for D'ît
is eax, etc.
Formula (17) allows us to write down symbolic sums for various peculiar
series. Thus
(18)
"
n- l
Y,kkek*tk
n-l
= ^tkDDekx
n-l
= DD ^(te*)k
= DD {*"*"* " H .
In particular,
(19)
y>*** = W ^ \\
All that would be necessary to sum (19) would be to find a different method
of attaching a meaning to the right-hand member.
For a Fibonacci-Lucas application, recall the general Lucas function
Ln = Ln(a,
Then
(20)
b) = an + bn.
>**%* = W g p ,+ V V
fc=o
* ap - 1
and for the generalized Fibonacci function
Fn = Fn(a,
bpex
b) = (an - bn) / (a - b),
then
(21)
- 1
aP"gn* _
(a - b)^kkek*Fpk
= W -
p x
ae
- 1
2)Pn5nX - 1
bpex
Following the methods outlined in [ 3 ] , [ 6 ] , [ 8 ] , [ 9 ] , or [11], we could set

down complicated symbolic formulas for the general series
(22)
J2kkek*u%lLk,
k=0
but we shall not take the space to exhibit the result.

For another application, let us rewrite (17) as aa = e~xDDeax,
have an obvious application in the two forms
(23)
LLnn = e-xDDeLnX
F*" = e-xDDeF"x
which allow us to introduce Fibonacci powers of Fibonacci

of Lucas) numbers into known series. In particular,
(24)
!>*"" =
rc = 0
so that we
(and Lucas powers
e-'DD^ttneL-x,
n 0
and a similar formula with F in place of L .

In principle then w e could sum such series if w e could sum the series
(25)
S(t,
u) = ] tnuLn ,
1978]
559
and
(26)
T(t9 u) = ]T tnuFn .
n= 0
These a r e o f f e r e d a s r e s e a r c h p r o j e c t s ; t h e a u t h o r would b e i n t e r e s t e d i n
h e a r i n g of any s u c c e s s by o t h e r s . DUS\U=1 and DUT\U=1 a r e known.
The o p e r a t o r 0 e may be c o n s i d e r e d f i n a l l y .
We f i n d
' / = {(e - i ) + i } f l / = E ()(e - iff

n= 0
= E f r W " DV, by (7),

n=0
"
-tfi*- ( - i ) B ~*(*) e ^
n=0
k=0
so we have
(27)
=Y,^D^(-l)n-k(n)&Kf(x).
Bf(x)
n
=0
"
k=0
^'
Let / ( x ) = x , t h e n
)exp
= xP E(n) ( P " 1)K= P Px? '

or therefore, another formula analogous to (17),
(28)
QBxp =
ppxp.
As an application we can get a different version of formula (19) as
(29)
**** = eeJ>* = e 9 ! ^ } .
We wish to remark that even stranger formulas have been published. Cayley [4], [5] expressed the Lagrange series inversion formula in the most
curious operational form
(30)
F(x) =
(DJ^'^F'ê^^)},
where x = u 4- hf(x) and F(x) is an arbitrary function.

he expressed the second form of this expansion as
By differentiation,
Cayley says these are well known, and goes on to write similar formulas for
functions of several variables.
Bronwin [2] writes
(32)
f(a + x) =
D!{f(a)ex)
as a symbolic form of Taylor's expansion. This is, of course, a special case

of the Lagrange expansion.
In conclusion, we wish to emphasize that the formulas presented here are
offered more for further research than as final answers to any of the questions
560
[Dec.
raised. It certainly is possible to introduce unusual terms into generating

functions by the use of unusual operators.
REFERENCES
1. George Boole, "On the Theory of Developments, I," Cambridge Math.
Journal,
Vol. 4 (1845), pp. 214-223.
2. Brice Bronwin, "On Certain Symbolical Representations of Functions," Cambridge and Dublin Math. Journal,
N.S., 2(6) (1847), 134-140.
3. L. Carlitz, "Generating Functions for Powers of Certain Sequences of
Numbers," Duke Math. Journal,
Vol. 29 (1962), pp. 521-537.
4. Arthur Cayley, "On Lagrange's Theorem," Cambridge Math. Journal,
Vol. 3
(1843), pp. 283-286.
5. Arthur Cayley, "On Lagrange's Theorem," Cambridge and Dublin Math. Journal, N.S., 6(10) (1851), 37-45.
6. H. W. Gould, "Generating Functions for Products of Powers of Fibonacci
Numbers," The Fibonacci
Quarterly,
Vol. 1, No. 2 (1963), pp. 1-16.
7. H. W. Gould & A. T. Hopper, "Operational Formulas Connected with Two Generalizations of Hermite Polynomials," Duke Math. Journal, Vol. 29 (1962),
pp. 51-63.
8. I. I. Kolodner, "On a Generating Function Associated with Generalized Fibonacci Sequences," The Fibonacci
Quarterly,
Vol. 3, No. 4 (1965), pp.
272-278.
9. John Riordan, "Generating Functions for Powers of Fibonacci Numbers,"
Duke Math. Journal,
Vol. 29 (1962), pp. 5-12.
10. L. J. Slater & A. Lakin, "Two Proofs of the 6 6 Summation Theorem," Proc.
Edinburgh Math. Soc. , (2) 9 (1956), 116-121.
11. David Zeitlin, "Generating Functions for Products of Recursive Sequences,"
Trans. Amer. Math. Soc,
Vol. 116 (1965), pp. 300-315.
(continued
from page
497)
2. L. Carlitz, "Generating Functions," The Fibonacci Quarterly,

Vol. 7, No. 4
(1969), pp. 359-393.
3. L. Carlitz, "Note on Convolved Power Sums," SIAM J. Math. Anal.,
Vol. 8
(1977), pp. 701-709.
4. J. W. L. Glaisher, "On a Class of Relations Connecting Any
Consecutive
Bernoullian Functions," Quart. J. Pure and Appl. Math.,
Vol. 42 (1911),
pp. 86-157.
5. J.W. L. Glaisher, "On ln(x-l)'7? + 2n (x - 2)m + + (x - l)nlm and Other
Similar Series," Quart. J. Pure and Appl. Math., Vol. 43 (1912), pp. 101122.
6. H. W. Gould, "Noch einmal die Stirlingschen Zahlen," Jber. Deutsch. Matin. Verein.,
Vol. 73 (1971), pp. 149-152.
7. H. W. Gould, "Explicit Formulas for Bernoulli Numbers," American
Math.
Monthly,
Vol. 79 (1972), pp. 44-51.
8. H. W. Gould, Combinatorial
Identities
(rev. ed.; Morgantown, W. Va. : By
the author, 1972).
9. C. P. Neuman & D. I. Schonbach, "Evaluation of Sums of Convolved Powers Using Bernoulli Numbers," SIAM Review, Vol. 19 (1977), pp. 90-99.
10. John Riordan, An Introduction
to Combinatorial
Analysis
(New York: John
Wiley & Sons, 1958).
1978]
A FIGURATE NUMBER CURIOSITY
561
11. D. H. Voelker, "On a Class of Polynomials," Notices

of Amer. Math. Soc. ,
Vol. 18 (1971), p. 800. Abstract 71T-A162.
12. D. H. Voelker, "On a Class of Polynomials," Rev. Un. Mat. Argentina,
Vol.
26 (1972), pp. 115-124.
13. Problem 1125, American Math. Monthly,
Vol. 61 (1954), p. 423; Solution to
Problem 1125, American Math. Monthly,
Vol. 62 (1955), pp. 125-126. Posed
by Walter James; solution by A. R. Hyde.
A FIGURATE NUMBER CURIOSITY:
EVERY INTEGER IS A
QUADRATIC FUNCTION OF A FIGURATE NUMBER

Empire
State
HARVEY J. HIND IN
College,
Stony Brook,
NY 11790
In this note we prove the following: Every positive integer n can be expressed in an infinite number of ways as a quadratic function for each of the
infinite number of figurate number types.
The nth figurate r-sided number pr is given by
(1)
p* = n((r
- 2)n - r + 4)/2,
where n = 1, 2, 3, ... and r = 3, 4, 5, ... . Therefore, the snth figurate

number is given by
(2)
pvsn
= sn((r
- 2)sn
- r + 4)/2.
However, (2) is a quadratic in n. Solving for n and taking the positive root
yields
(p - 4) + /(r - 4 ) 2 + 8(r - 2)psrn
which allows us to express n as stated above. A special case of (3) for pentagonal numbers (r = 5) was obtained by Hansen [1].
REFERENCE
1.
R. T.Hansen, "Arithmetic of Pentagonal Numbers." The Fibonacci

Vol. 8, No. 1 (Feb. 1970), pp. 83-87.
Quarterly,
ELEMENTARY PROBLEMS AND SOLUTIONS

Edited by
A. P. HILLMAN
of New Mexico, Albuquerque,
University
New Mexico
87131
Send all communication* hegaJtdlng ELEMENTARY PROBLEMS AND SOLUTIONS to

Professor A. P. Hillman, 709 Solano Dr., S.E., Albuquerque, New Mexico 87108.
Each solution
on phoblem should be on a Aepa/iate Aheet [oh Aheet*).
Vhe{oJtence wiUL be given to tho*e typed with, double spacing in the ôhmat u*ed below.
Solution* should be heceived within 4 month* ofi the publication
date.
DEFINITIONS
The Fibonacci numbers Fn and Lucas numbers Ln satisfy Fn + 2 = Fn + 1 + Fn ,
FQ = 0, Fx = 1 and Ln + 2 = Pn + i + ^n? L0 = 2, L x = 1. Also a and b designate
the roots (1 + /5)/2 and (1 - /5~)/2, respectively, of x2 - x - 1 = 0.
PROBLEMS PROPOSED IN THIS ISSUE
B-388
Proposed
by Herta
T. Freitag,
Roanoke,
VA.
Let Tn be the triangular number n(n + l)/2.

T
+ T
+ i
Show that
l + 3 + 5 2 + + (2n - l ) 2
s + '" + 2n-i
and express these equal sums as a binomial coefficient.

B-389
Proposed
by Gregory
Wulczyn,
Bucknell
University,
Lewisburg,
PA.
Find the complete solution, with two arbitrary constants, of the difference equation
(n2 + 3n + 3)Un + 2 - 2(n2 + n + l)Z7w+ i + (n2 - n + l)Un = 0.
B-390
Proposed
by V. E. Hoggatt,
Jr., San Jose
State
University,
San Jose,
CA
Find, as a rational function of x9 the generating function
B~391
Proposed
by M. Wachtel,
Zurich,
Switzerland.
Some of the solutions of 5x2 + 1 = y2 in positive integers x and y are

(x9y)
= (4,9), (72,161), (1292,2889), (23184,51841), and (416020,930249).
Find a recurrence formula for the xn and yn of a sequence of solutions (xn,
yn) and find m(xn + 1/xn) in terms of a = (1 + /5")/2.
n -y 00
B-392
Proposed
by Phil
Mana, Albuquerque,
Let Yn = (2 + 3n)Fn + (4 + 5n)Ln.

Yn+2 ~ Yn+l ~ Yn
B-393
Proposed
by V. E. Hoggatt,
Let Tn = ( n 2 X ) '
Find c o n s t a n t s h and k such t h a t
hFn + kLn.
Jr., San Jose
o = X> Pn = TiT2
"Tn
f o r i n t e g e r s k and n w i t h 0 . k n .
fnl
lk\
NM.
State
^r
Show t h a t
1
(n\(n + \
n - k + l U / U + 1 /
562
University,
San
n > 0 , and [ j ] =
Jose,CA.
Pn/PkPn.k
Dec. 1978
563
SOLUTIONS
1NCONTIGUOUS ZERO DIGITS
B-364
Proposed
by George Berzsenyi,
Lamar University,
Beaumont,
TX.
Find and prove a formula for the number R(n) of positive integers less
than 2 whose base 2 representations contain no consecutive O's. (Here n is
a positive integer.)
Solution
by C. B.A.
Peck,
State
College,
PA.
Let Sn be the number of integers 777 with 2n~ <_ m < 2n and having a binary
representation 5(777) with no consecutive pair of 0Ts. Clearly Sn = Rn - i?n_i
for n > 1 and S1 = R. Also,
Sn = n_i + 5 n _ 2 for w > 2,
since Sn_1 counts the desired 77? for which B(m) starts with 11 and Sn_2 counts
the desired 77? for which 5(777) starts with 101. It follows inductively that
S
= Fn+1>
and
then
Rn
= S,
+ S2
+ + Sn
= F2 + F3 + . . + Fn + 1 = Fn
Also solved by Michael Brozinsky,

Pauls.
A. G. Shannon, Sahib Singh, Rolf Sonntag,
+3
Bruckman, Graham Lord, Bob

Gregory Wulczyn, and the
- 2.
Prielipp,
proposer.
CONGRUENT TO A G.P.
B-365
Proposed
by Phil
Mana, Albuquerque,
NM
Show that there is a unique integer m > 1 for which integers a and r exist with Ln E avn (mod 777) for all integers n >_ 0. Also, show that no such 777
exists for the Fibonacci numbers.
Solution
by Graham Lord,
Universite
Laval,
Quebec.
2 5
Since 7 = LiiLl = a r
E LZL3 = 12 (mod 777), then 77? divides 5, hence 777 = 5.
Furthermore, a = ar E L0 = 2 (mod 5). And finally, or2 =L2 = L1 + LQ E av +
a (mod 5) together with a E 2 (mod 5) implies v1 E v + 1 (mod 5), i.e., v E 3
(mod 5). In all, 777 = 5, and a and v can be taken equal to 2 and 3, respectively. Note for any n >_ 1, Ln + 1 = Ln + Ln_1 E ap n + a p n _ 1 E ar n+ 1 (mod 5).
For the Fibonacci numbers, if 777 were to exist, then
3 = FxFh
E a 2 p 5 E F2F3 = 2 (mod 777),
i.e., 1 E 0 (mod 777), which is impossible if 777 > 1.

Also solved by George Berzsenyi,
Paul S. Bruckman,
Sahib Singh, Gregory Wulczyn, and the
proposer.
Bob Prielipp,
A. G.
Shannon,
LUCAS CONGRUENCE
B-366
Proposed
Slippery
by Wray G. Brady, University

Rock State College,
Slippery
Prove that LiLj
E LhLk
of Tennessee,
Rock, PA.
Knoxville,
TN and
(mod 5) when i + j - h + k.
Solution
by Paul S. Bruckman, Concord,
lege,
Clarion,
PA
(independently)
Using the result of B-365,
CA and Sahib Singh,
Clarion
State
Col-
564
L.L-
= 2 3i
- LhLk
+j
[Dec.
- 2 3h + k = 0 (mod 5 ) ,
since i + j = h + k.
Also solved by George Berzsenyi, Herta T. Freitag,
Graham Lord, T.
Bob Priellpp,
A. G. Shannon, Gregory Wulczyn, and the proposer.
Ponnudurai,
ROUNDING DOWN
B-367
Proposed by Gerald E. Bergum, Sr. ,

SD.
'
Dakota State
University,
Let [x] be the greatest integer in x, a = (1 + /5)/2

that
(a)
F2n = [aF 2n _ L ]
and
(b)
F2n+1 = [a^^.J.
Solution
T
In view of Binet s
Lamar University,
_ 2)2n~1
7
a - b
Prove
Beaumont, TX.
2n
a2n-i
l
and n >. 1.
formula,
ccFr>n_ T F 0y. = a
zn
Brookings,
Zn
b2n
7
a - b
-
Similarly,
2p
^ '
Since - 1 < b =
follow.
j?
"
+1
<
~2&
~ O
a - b
"
t h a t 0 < -b2n~x
implies
a - b
< 1,
the
" ~b
2n-l
desired
Also solved by J. L. Brown, Jr., Paul S. Bruckman, Graham Lord, Bob

A. G. Shannon, Sahib Singh, and the
proposer.
results
Priellpp,
CONVOLUTING FOR CONGRUENCES

B-368
Proposed by Herta T. Freitag,
Obtain functions g(n)
and h(n)
Roanoke, VA.
such that
^2^FiLn-i=
g(n)Fn + h(n)Ln
i-l
and use the results to obtain congruences modulo 5 and 10.

Solution
by Sahib Singh,
Clarion
State
College,
Clarion,
PA.
Let An = 2_\î^n-i'
Then the generating function A1 + A2x + A3x2
i=i
is a rational function with (1 - x - x ) as the denominator. It follows that

g(n) and lain) are quadratic functions of n. Then, solving simultaneous equations for the coefficients of these quadratics leads to
g(n)
= (5n2 + lOn + 4)/10
and
h(n)
= n/10
so t h a t
(5n2 + l\)Fn + nLn = 0 (mod 10).
This also gives us nLn - Fn (mod 5 ) .
Also solved
poser.
by Paul S. Bruckman,
Graham Lord, Gregory Wulczyn,
and the
pro-
1978]
565

NO LONGER UNSOLVED
B-369
Proposed
Lamar University,
Beaumont,
TX.
For all integers n >_ 0, prove that the set

>n
-^2n+3J
\^Zn+l9
-^2n+5/
has the property that if x> y Sn and x ^ y then xy + 5 is a perfect square.

Forft= 0, verify that there is no integer z that is not in Sn and for which
(ss ^2n+i' ^2n+3s -^n + s) n a s this property. (For n > 0, the problem is unsolved. )
Solution
by Graham Lord,
Universite
Laval,
Quebec.
That Sn has the property follows from the identities:

^2n+1^2n+3
2
+ 5 = L2 n + 2 5
and
L
2n + l L 2n+5 +
2r7 + 3 '
In the second part of this solution use is made of the results:
lL6/c + i a n d 2 lL6fe+5
4= L L
3 l 6fc+3
4|L2k
Of these, (l) is somewhat well known and the latter three are consequences of the results in "A Note on Fibonacci Numbers," The Fibonacci
Quarterly,
Vol. 2, No. 1 (February 1964), pp. 15-28, by L. Carlitz.
By (T) and \2) there is exactly one even number, L6fc+3 > in the set Sn,
n >. 0. So if {2} U 5 n has the desired property, then zLsk+3 + 5 will be an
odd square and thus congruent to 1 modulo 8. This implies that 2, if it exists, is odd.
Now the other two members of Sn are either:
(a)
L6k_i,
(b) L
Lsk+1;
s k + 5
L6k+7;
o r (c) L6k+l9
LBk+5-
Each of these is odd by (l), and hence the sum of 5 and any one of them
multiplied by z will equal an even square. Thus, in case (a) [and similarly
in case (b)]:
zLsk_1
+ 5 E 0 (mod 4 ) , and zL6k+1
+ 5 = 0
(mod 4 ) ;
i. e. ,
zL&k
= z(L6k
+1
- L 6k _i) = 0 (mod 4).
But this is impossible by (5) and the fact that z is odd.

And in case (c),
z 5Fek+3
= (zLsk+5
+ 5) - (zL6k+1
+ 5) = 0 (mod 4 ) ,
which is also impossible by (4).

Consequently, no z exists such that the set {z} U Sn has the desired
property. Note that it was not assumed that n = 0.
Also solved by Paul S. Bruckman, Herta T. Freitag,
A. G. Shannon, Sahib Singh, and the
proposer.
T. Ponnudurai,
Bob
Prielipp,
ADVANCED PROBLEMS AND SOLUTIONS

Edited by
RAYMOND E. WHITNEY
Lock Haven State College,
Lock Haven,
PA 17745
Send alt commuyiLccutlonA concerning

ADVANCED PROBLEMS AND SOLUTIONS to
Raymond E. Whitney, Mathematics Department, Lock Haven State College, Lock
Haven, Pennsylvanis 17745. ThjJ> department eApecioLty welcomes problems believed to be new on, extending old fie&ultA. ?siopo&&i& should submit AolmtLonA
on, other information
that wilt aA&lbt the ecUXor. To facilitate
their
consideration,
solutions
should be submitted on separate Signed sheets within 2
months a^ter publication
o^ the problems.
H-290
Proposed
by Gregory
Wulczyn,
Bucknell
University,
Lewisburg,
PA.
Show that:
(a)
(b)
H-291
Fk* k + 6r +3 ~ ^k+kr + 2 ~ ( " 1 /
FkFLsr
Proposed
~ ?U,r
^2r +l^k+8r
(-Dk + 1F2Zr(Fk
+ 8p
+h ~ ^k+kr+2'
>
+ 2F k + l f r ) .
Lamar University,
Beaumont,
TX
Prove that there are infinitely many squares which are differences of
consecutive cubes.
H-292
Proposed by F. S. Cater
land, OR.
and J. Daily,
Portland
State
University,
Port-
Find all real numbers v (0,1) for which there exists a one-to-one function fr mapping (0,1) onto (0,1) such that
(1) fr and f~ are infinitely many times differentiable on (0,1), and
(2) the sequence of functions fr, frofP9
frofrofr,
fv^fvfvfv
> '
converges pointwise to v on (0,1).
H-293
Proposed
by Leonard
Carlitz,
Duke University,
It is known that the Hermite polynomials <Hn(x)\

n=0
Durham, NC.
defined by
'
satisfy the relation
I>+ *(*>S = e2XZ-*2Hk(x

n=0
- z) (k = 0, 1, 2, . . . ) .
'
Show that conversely if a set of polynomials \fn(x)\
<x>
Z-4+fc^Jr = E ^ ^ J r 4 ( x - z) (k = 0, 1, 2, . . . ) ,
n =0
where f0(x)
satisfy
n=0
= 1, fx(x) = 2x, then

fn(x)
= Hn(x)
(n = 0, 1, 2, . . . ) .
566
Dec. 1978
H-294
Proposed
by Gregory
Wulczyn,
Bucknell
University,
567
Lewisburg,
PA.
Evaluate
^2r+l
^6r+3
-^10r+5
^lhr+7
^ I 8r +9
* hr +2 ^ 1 2 r + 6
^ 2 OP + 1 0 ^ 2 8 r + 1 h^ Z Sv + 1 8
^6r+3
^ 3 6 r + 1 5 ^ 4 2 ^ + 2 1^5 4 r + 2 7
^18r+9
^ 7 8 r + L f - ^ 7 2 i + P + 1 2 ^ 7 i + 0 P + 2 0 ^ 5 6r + 2 8 ^ 7 2 r + 3 6
^ 1 Or + 5 ^ 2 Or + 1 5 ^ 5 Or +2 5 ^ 7 Or + 3 6 ^ 5 Or+1+5
SOLUTIONS
SYMMETRIC SUM
H-272 ( C o r r e c t e d )
Proposed
by Leonard
Carlitz,
Duke University,
Durham,
NC.
Show t h a t
EmcAX^rvÂ)
is symmetric in p, q,
Solution
C m (p, (7, r)
v.
by Paul Bruckman,
Concord,
CA.
Define
a)
Clearly,
therefore
/?? - j in
volving p
j =0
Cm(p, q, v) = Cm(q, p, r). A moment's reflection reveals that it

suffices to show that Cm(p, q, r) = Cm(q, p, p ) . Replacing j by
(1) and applying Vandermonde? s convolution theorem on the term inand q yields:
".<p-"-')-i;(.^)(5)(j)/(;)fc^))
k=0
j-o
m
m -J
E0t(A)UKi)(.?oc;')/(j)-
j=0 k=0
7 = 0fe= 0
Replacing k by m - k in the
last expression yields
te
last
expression yields:
m
m
'.<*'>-E(.-*)(5)(.?*)(0(*)/(;)
J
= 0fc= j
-EU)(.!,)E(.-*)0)(5)/(")However, it is easy to verify that
(-,)/(?)-O/rTO-O^'/fT1)Therefore,
777
*.<*.. w C)E(?)(.f*)?:w)'(J)(5)/(-r1)-
568
Now, formula (7.1) in Combinatorial

is as follows:
<
[Dec.
(H. W. Gould, Morgantown,1972),
Identities
D-u'UK^/cr)/-
fc-0
Letting k = J, n = k, % = q, y = m - r - 1 In (4), we therefore simplify (3)

as follows:
*..*> = Q|:axA)r i T~ i )/CT 1 )

fr\SLy(P\t
\m)ls\k)\m-k)\
1 \/q+r-m
k
+ k\ I IT -m + k\
) /\
k
)
k=0
once again and replacing k by j y i e l d s :

m
<p.,.rt=E(?)(^-)L-J-)r'r+j')/()
3
= Cm(q, p, p ) .
Also solved
by the
Q.E.D.
proposer.
A RAY OF LUCAS
H-273 Proposed by W. G. Brady, Slippery
Rock State
College,
Slippery
Rock, PA.
Consider, after Hoggatt and H-257, the array D, indicated below, In which
^2n+i (n = 0, 1, 2, ...) is written in staggered columns:
i.
ii.
iii.
1
4
1
11
4
1
29
11 4
1
76
29 11 4 1
Show that the row sums are L2n + 2 ~ 2;
Show that the rising diagonal sums are F2n+3 ~ 1 where L2n+1
is the
largest element in the sum.
Show that if the columns are multiplied by 1, 2, 3, ... sequentially to
the right then the row sums are L2n+3
- {In + 3).
Solution
by A. G. Shannon,
The N.S.W. Institute
of Technology,
Australia.
In effect we are asked to prove:

n
i*
2Ll ^2n-2j +1
^2n+2
~ 2;
J-0
[n/2]
i1,
2Ls ^2n-kj + i ~ F2n+3
i;
J-0
n
iii.
(j + l)L2n_2j+1
= L2n + 3 - {In + 3 ) .
j-o
n
n
(^
Z^L2^-2J + 1
J=0
I-** \
j - 0
J2n+2 " ^ O J a
2n-2(j-
1)
s
"
2n-2j)
required.
/ ^ L2n - 2 (j - 1)
j=0
n+1
"
,L2n-2(,i-l)
j=l
1978]

[/2]
(ii)
2 ^
[n/2]
L
2n-Hj
[n/2]
+ 1 = 2-J {F2n-Hj
J-o
+3 "
2n - 4 j - 1 ) =
j=o
~ î^(25 ) " F.1o(29
if n\m9
if n\m.
[n/2] + l
F
2n-kj
2-j\^2n-2i+
=0
2n-kj+3
n + 1) = F 2 n + 3 - 1
j-i
(iii) (J + l)L2n_2j,+ 1 - S L 2 n - 2 i + l = Z)(L2n-2i+ 2 i =0j =1

'
*-0
n
rc + 1
+3 ~ X ,
^.Q
'
= F2n+3
in which
569
*Yj 2in+i)-n+i - Ll
)
[from (i)]
3 ~ -^2n-2i+l ~ ^j
- (L2n + 2 - 2) - 2(n + 1)
i =Q
= L2n+,
- 2 - 1 - L 2 n + 2 + 2 - 2(n + 1)
= L2n+3
- (In + 3 ) , as required.
Also solved by P. Bruckman, G. Wulczyn,

and the
proposer.
Late Acknowledgments:
H. Freitag,
B. Prielipp,
Dinh
The'Hung,
F. T. Howard solved H-268 and M. Klamkin solved H-270.

A CORRECTED OLDIE
H-225
Proposed
by G. A. R. Guillotte,
Quebec,
Canada.
Let p denote an odd prime and xp + yp = zp for positive integers, x, y,

and z. Show that
(A) p < xj(z - x) + y/(z - y), and
(B)
z/2(z - x) < p < y/(z - y).
VOLUME INDEX
ADLER, IRVING. "A Simple Continued Fraction Represents a Mediant Nest of Intervals," Vol. 16, No. 6, pp. 527-529.
ANDERSON, 0. D. "Some More Patterns from PascalTs Triangle," Vol. 16, No. 4,
pp. 296-301.
ANDERSON, PETER G. "On the Formula IT = 2Z arcot /2fc+i>"
118.
Vo1
16
>
No
> P-
AUSTIN, RICHARD. "Binary Sequences without Isolated Ones," Vol. 16, No. 1,
pp. 84-86 (co-author, Richard Guy).
BEITER, MARION. "Coefficients of the Cyclotomic Polynomials F3qr(x)9"
16, No. 4, pp. 302-306.
Vol.
BERGUM, G. E. "A Combinatorial Problem Involving Recursive Sequences and Tridiagonal Matrices," Vol. 16, No. 2, pp. 113-118 (co-author, V. E. Hoggatt,
Jr.); "A Family of Tridiagonal Matrices," Vol. 16, No. 3, pp. 285-288 (coauthor, V. E. Hoggatt, Jr). Problem proposed: B-367, Vol. 16, No. 6, p.
564. Problem solved: B-367, Vol. 16, No. 6, p. 564BERNSTEIN, LEON. "An Invariant for Combinatorial Identities," Vol. 16, No. 4,
pp. 354-369.
BERZSENYI, GEORGE. Problems proposed: B-351, Vol. 16, No. 1, p. 91; H-266,
Vol. 16, No. 1, p. 94; B-378, Vol. 16, No. 2, p. 184; B-387, Vol.16, No. 5,
p. 473, H-269, Vol. 16, No. 5, p. 478; B-364, B-369, Vol. 16, No. 6, pp.
563, 565, H-291, Vol. 16, No. 6, p. 566. Problems solved: B-346, B-347,
B-248, B-349, B-350, B-351, Vol. 16, No. 1, pp. 89-91, H-266, Vol. 16, No.
1, p. 94; B-352, B-356, Vol. 16, No. 2, pp. 185, 186; B-358, B-359, B-362,
Vol. 16, No. 5, pp. 474, 475, H-269, Vol. 16, No. 5, p. 478; B-364, B-365,
B-366, B-367, B-369, Vol. 16, No. 6, pp. 563-565.
BEVERAGE, D. Problems Solved: H-264, H-265, Vol. 16, No. 1, p. 93.
Proposed: H-283, Vol. 16, No. 2, p. 188.
Problem
BICKNELL-JOHNSON, MARJORIE. "A Golden Double Crostic," Vol. 16, No. 1, pp.
67-69; "A Golden Double Crostic Solution," Vol. 16, No. 1, p. 83; "A Primer
for the Fibonacci Numbers XVII: Generalized Fibonacci Numbers Satisfying
un+iUn-i
- Un = 1," Vol. 16, No. 2, pp. 130-137 (co-author, V. E. Hoggatt,
Jr.); "Properties of Generating Functions of a Convolution Array," Vol. 16,
No. 4, pp. 289-295 (co-author, V. E. Hoggatt, Jr.); "Convolution Arryas for
Jacobsthal and Fibonacci Polynomials," Vol. 16, No. 5, pp. 385-402 (coauthor, V. E. Hoggatt, Jr.); "Fibonacci Chromotology or How To Paint Your
Rabbit," Vol. 16, No. 5, pp. 426-428; "Divisibility Properties of Polynomials in Pascal's Triangle," Vol. 16, No. 6, pp. 501-513 (co-author, V. E.
Hoggatt, Jr.).
BRADY, WRAY G. "More on Benford's Law," Vol. 16, No. 1, pp. 51-52. Problems
proposed: B-366, Vol. 16, No. 6, p. 563, H-273, Vol. 16, No. 6, p. 568.
Problems solved: B-347, B-348, B-349, Vol. 16, No. 1, pp. 89-91; B-352, B354, B-355, Vol. 16, No. 2, pp. 185-186; H-270, Vol. 16, No. 5, p. 479; B366, Vol. 16, No. 6, p. 563, H-273, Vol. 16, No. 6, p. 568.
BRIDGER, CLYDE A. Problem solved:
B-347, Vol. 16, No. 1, p. 90.
570
Dec. 1978
VOLUME INDEX
571
BROUSSEAU, BROTHER ALFRED. "Formula Development through Finite Differences,"

Vol. 16, No. 1, pp. 53-67.
BROWN, J. L., JR. "Some Sequence-to-Sequence Transformations which Preserve
Completeness," Vol. 16, No. 1, pp. 19-22; "A^ Primer for the Fibonacci Numbers, Part XVI, The Central Column Sequence," Vol. 16, No. 1, pp. 41-46
(co-author, V. E. Hoggatt, Jr.); "Simplified Proof of a Greatest Integer
Function Theorem," Vol. 16, No. 4, pp. 307-309. Problem Solved: B-367,
Vol. 16, No. 6, p. 564.
BROZINSKY, MICHAEL. Problems Solved: B-351, Vol. 16, No. 1, p. 91; B-359,
Vol. 16, No. 5, p. 474; B-364, Vol. 16, No. 6, p. 563.
BRUCKMAN, PAUL S. Problems proposed: H-280, Vol. 16, No. 1, p. 92; B-377,
Vol. 16, No. 2, p. 184; H-268, Vol. 16, No. 5, p. 477. Problems Solved: B346, B-347, B-348, B-349, B-350, B-351, Vol. 16, No. 1, pp. 89-91, H-264,
H-265, H-266, Vol. 16, No. 1, pp. 94-95; B-352, B-353, B-354, B-355, B-356,
B-357, Vol. 16, No. 2, pp. 185-186, H-267, H-268, Vol. 16, No. 2, pp. 191192; B-358, B-359, B-360, B-361, B-362, B-363, Vol. 16, No. 5, pp. 474476, H-269, H-270, Vol. 16, No. 5, pp. 478-479; B-364, B-365, B-366, B-367,
B-368, B-369, Vol. 16, No. 6, pp. 563-565.
BUNDER, M. W. "More Fibonacci Functions," Vol. 16, No. 2, pp. 94-98.
BUTCHER, J. C. "On a Conjecture Concerning a Set of Sequences Satisfying the
Fibonacci Difference Equation," Vol. 16, No. 1, pp. 80-83.
CARLITZ, LEONARD. "Recurrence of the Third-Order and Related Combinatorial
Identities," Vol. 16, No. 1, pp. 11-18; "Generalized Eulerian Numbers and
Polynomials," Vol. 16, No. 2, pp. 138-146, 151 (co-author, V. E. Hoggatt,
Jr.); "Some Polynomials Related to Fibonacci and Eulerian Numbers," Vol.
16, No. 3, pp. 216-226; "A Recurrence Suggested by a Combinatorial Problem," Vol. 16, No. 3, pp. 227-242; "Some Remarks on a Combinatorial Identity," Vol. 16, No. 3, pp. 243-248; "Enumeration of Certain Weighted Sequences," Vol. 16, No. 3, pp. 249-254; "The Number of Derangements of a
Sequence with Given Specification," Vol. 16, No. 3, pp. 255-258; "Enumeration of Permutations by Sequence," Vol. 16, No. 3, pp. 259-268; "Some
Classes of Fibonacci Sums," Vol. 16, No. 5, pp. 411-426. Problems Solved:
H-264, H-265, Vol. 16, No. 1, pp. 93-94; H-268, Vol. 16, No. 2, p. 191; B361, Vol. 16, No. 5, p. 475, H-270, Vol. 16, No. 5, p. 479; H-272, Vol. 16,
No. 6, p. 567. Problems Proposed: H-264, Vol. 16, No. 1, p. 92; H-268,
Vol. 16, No. 2, p. 191; B-361, Vol. 16, No. 5, p. 475; H-289, H-270, Vol.
16, No. 5, pp. 477, 479; H-293, Vol. 16, No. 6, p. 566.
CATER, F. S. Problem Proposed:
J. Daily).
H-292, Vol. 16, No. 6, p. 566
(co-proposer,
COHEN, G. L. "On Odd Perfect Numbers," Vol. 16, No. 6, pp. 523-527.
CULL, PAUL. "Knight's Tour Revisited," Vol. 16, No. 3, pp. 276-284 (co-author,
Jeffery De Curtins).
DAILY, J. Problem Proposed:
Cater).
H-292, Vol. 16, No. 6, p. 566 (co-proposer, F.S.
DE CURTINS, JEFFERY. "Knight's Tour Revisited," Vol. 16, No, 3, pp. 276-284
(co-author, Paul Cull).
572
VOLUME INDEX
[Dec.
DESMOND, JAMES E. "On the Existence of the Rank of Apparition of m in the

Lucas Sequence," Vol. 16, No. 1, pp. 7-10; "on the Equality of Periods of
Different Moduli in the Fibonacci Sequence." Vol. 16, No. 1, pp. 86-87.
DE VITA, JOSEPH. "Fibonacci, Insects, and Flowers," Vol. 16, No. 4, pp. 315-317.
DRESSLER, ROBERT E. "Interpolation of Fourier Transforms on Sums of Fibonacci
Numbers," Vol. 16, No. 3, pp. 193-194 (co-author, Louis Pigno); "Topological, Measure Theoretic and Analytic Properties of the Fibonacci Numbers,"
Vol. 16, No. 3, pp. 195-197 (co-author, Louis Pigno).
ENGLE, ROGER. Problems Solved: B-358, B-360, B-361, B-362, B-363, Vol. 16,
No. 5, pp. 474-476 (co-solver, Sahib Singh); B-359, Vol. 16, No. 5, p. 474
(co-solvers, Benjamin Freed & Sahib Singh).
ESWARATHASAN, A.
310-314.
"On Square Pseudo-Fibonacci Numbers," Vol. 16, No. 4, pp.
FERGUSON, HELAMAN ROLFE PRATT. "The Fibonacci Pseudogroup, Characteristic

Polynomials and Eigenvalues of Tridiagonal Matrices, Periodic Linear Recurrence Systems and Application to Quantum Mechanics," Vol. 16, No. 5, pp.
435-447.
FIELD, R. S. Problem Proposed: B-359, Vol. 16, No. 5, p. 474. Problem solved:
B-359, Vol. 16, No. 5, p. 474.
FLANIGAN, JIM. "Generalized Two-Pile Fibonacci Nim," Vol. 16, No. 5, pp. 459469.
FREED, BENJAMIN. Problem Solved: B-359, Vol. 16, No. 5, p. 474 (co-solvers,
Roger Engle & Sahib Singh).
FREITAG, HERTA T. Problems Proposed: B-371, B-372, Vol. 16, No. 1, p. 88; B379, B-356, Vol. 16, No. 2, pp. 184, 186; B-385, B-362, B-363, Vol. 16, No.
5, pp. 473, 475-476; B-388, B-368, Vol. 16, No. 6, pp. 562,564. Problems
Solved: B-346, B-348, B-350, B-351, Vol. 16, No. 1, pp. 89-91; B-352, B353, B-355, B-356, Vol. 16, No. 2, pp. 185-186; B-358, B-362, B-363, Vol.
16, No. 5, pp. 474-476; B-366, B-368, B-369, Vol. 16, No. 6, pp. 563-565,
H-273, Vol. 16, No. 6,. p. 568.
FULLER, LEONARD E. "Vectors Whose Elements Belong to a Generalized Fibonacci
Sequence," Vol. 16, No. 5, pp. 447-450.
GARFIELD, RALPH. Problems Solved: B-346, B-348, B-349, Vol. 16, No. 1, pp.
89-91; B-352, B-353, B-354, B-355, Vol. 16, No. 2, pp. 185-186.
GERDES, WALTER. "Convergent Sequences and Generalized Fibonacci Numbers,"
Vol. 16, No. 3, pp. 269-275.
GIULI, ROBERT M.
Problem Solved:
B-361, Vol. 16, No. 5, p. 475.
GLADWIN, A. S. "Expansion of the Fibonacci Numbers Fnm in nth Powers of Fibonacci or Lucas Numbers," Vol. 16, No. 3, pp. 213-215.
GOULD, H. W. "Remark on Problem H-123," Vol. 16, No. 2, p. 189; "Evaluation
of Sums of Convolved Powers Using Stirling and Eulerian Numbers," Vol. 16,
No. 6, pp. 488-497, 560-561; "Operational Formulas for Unusual Fibonacci
Series," Vol. 16, No. 6, pp. 555-560. Problem Proposed: H-282, Vol. 16,
No. 2, p. 188 (co-proposer, W. E. Greig).
GRASSL, RICHARD M. Problems Proposed: B-349, B-350, Vol. 16, No. 1, pp. 9091. Problems Solved: B-349-B-350, Vol. 16, No. 1, pp. 90-91.
1978]
VOLUME INDEX
573
GREGORY, M. B. "Fibonacci Sine Sequences," Vol. 16, No. 2, pp. 119-120 (coauthor, J. M. Metzger).
GREIG, W. E. "On Generalized G^k Numbers," Vol. 16, No. 2, pp. 166-170;
"Folded Sequences and Bode's Problem," Vol. 16, No. 6, pp. 530-539. Problem Proposed: H-282, Vol. 16, No. 2, p. 188 (co-proposer, H. W. Gould).
GUARALDO, ROSALIND. "On the Density of the Image Sets of Certain Arithmetic
FunctionsI," Vol. 16, No. 4, pp. 318-326; "On the Density of the Image
Sets of Certain Arithmetic FunctionsII," Vol. 16, No. 5, pp. 428-434; "On
the Density of the Image Sets of Certain Arithmetic FunctionsIII," Vol.
16, No. 6, pp. 481-488.
GUERIN, E. E. "Matrices and Convolutions of Arithmetic Functions," Vol. 16,
No. 4, pp. 327-334.
GUILLOTTE, G. A. R. Problem Proposed: H-225, Vol. 16, No. 6, p. 569 . Problems Solved: H-272(corrected), H-273, Vol. 16, No. 6, pp. 567-568.
GUPTA, HANSRAJ. "The Rank-Vector of a Partition," Vol. 16, No. 6, pp. 548552; "The Andrews Formula for Fibonacci Numbers," Vol. 16, No. 6, pp. 552555.
GUY, RICHARD. "Binary Sequences without Isolated Ones," Vol. 16, No. 1, pp.
84-86 (co-author, Richard Austin).
HANSEN, RODNEY T. "General Identities for Linear Fibonacci and Lucas Summations," Vol. 16, No. 2, pp. 121-128.
HARBORTH, HEIKO. "&-Adic Numbers in Pascal's Triangle Modulo &," Vol. 16, No.
6, pp. 497-500.
HEED, JOSEPH L. "Entry Points of the Fibonacci Sequence and the Euler 0
Function," Vol. 16, No. 1, pp. 47-50 (co-author, Lucille Kelly).
HENDY, M. D. "Stolarsky's Distribution of Positive Integers," Vol. 16, No. 1,
pp. 70-80.
HENSLEY, DOUGLAS. "Fibonacci Tiling and Hyperbolas," Vol. 16, No. 1, pp. 3740.
HERGET, W. "Minimum Periods Modulo n for Bernoulli Numbers," Vol. 16, No. 6,
pp. 544-548.
HICKERS0N, DEAN R. "Identities Relating the Number of Partitions into an Even
and Odd Number of Parts," Vol. 16, No. 1, pp. 5-6; "An Identity Relating
Compositions and Partitions," Vol. 16, No. 1, pp. 23-26.
HIGGINS, FRANK. Problem Proposed: B-357, Vol. 16, No. 2, p. 186.
Solved: B-357, Vol. 16, No. 2, p. 186.
Problem
HILLMAN, A. P. "A Property of Wythoff Pairs," Vol. 16, No. 5, p. 472 (coauthor, V. E. Hoggatt, Jr.). Editor: "Elementary Problems and Solutions,"
Vol. 16, No. 1, pp. 88-91, 96; Vol. 16, No. 2, pp. 184-187; Vol. 16, No. 5,
pp. 473-476; Vol. 16, No. 6, pp. 562-565.
HOGGATT, V. E., JR. "A Primer for the Fibonacci Numbers, Part XVI, The Central Column Sequence," Vol. 16, No. 1, pp. 41-46 (co-author, J. L. Brown,
Jr.); "A Combinatorial Problem Involving Recursive Sequences and Tridiagonal Matrices," Vol. 16, No. 2, pp. 113-118 (co-author, G. E. Bergum); "A
Primer for the Fibonacci Numbers XVII: Generalized Fibonacci Numbers
574
VOLUME INDEX
[Dec.
Satisfying un+1un_1
- u\
= 1," Vol. 16, No. 2, pp. 130-137 (co-author,
Marjorie Bicknell-Johnson); "Generalized Eulerian Numbers and Polynomials,"
Vol. 16, No. 2, pp. 138-146, 151 (co-author Leonard Carlitz); "A Family
of Tridiagonal Matrices," Vol. 16, No. 3, pp. 285-288 (co-author, G. E. Bergum); "Properties of Generating Functions of a Convolution Array," Vol. 16,
No. 4, pp. 289-295 (co-author Marjorie Bicknell-Johnson); "Convolution Arrays for Jaeobsthal and Fibonacci Polynomials," Vol. 16, No. 5, pp. 385402 (co-author, Marjorie Bicknell-Johnson); "A Property of Wythoff Pairs,"
Vol. 16, No. 5, p. 472 (co-author, A. P. Hillman); "Divisibility Properties
of Polynomials in Pascal!s Triangle," Vol. 16, No. 6, pp. 501-513 (co-author,
Marjorie Bicknell-Johnson). Problems Proposed: B-375, Vol. 16, No. 1, p.
88, B-346, B-347, Vol. 16, No. 1, p. 89; H-278, H-265, Vol. 16, No. 1, pp.
91, 94; B-381, B-352, B-353, Vol. 16, No. 2, pp. 184-185, H-281, Vol. 16,
No. 2, p. 188, H-267(corrected), Vol. 16, No. 2, p. 190; H-285, Vol. 16,
No. 5, p. 477; B-390, B-393, Vol. 16, No. 6, p. 562. Problems Solved: B346, Vol. 16, No. 1, p. 89; B-352, B-353, Vol. 16, No. 2, p. 185; H-267,
Vol. 16, No. 2, p. 190.
H0RADAM, A. F. "Diagonal Functions," Vol. 16, No. 1, pp. 33-36; "Wythoff
Pairs," Vol. 16, No. 2, pp. 147-151.
H0RADAM, E. M. "Solved, Semi-Solved, and Unsolved Problems in Generalized
Integers: A Survey," Vol. 16, No. 4, pp. 370-381.
HOWARD, F.
Problems Solved: H-268, Vol. 16, No. 2, pp. 191-192.
HUNG, DINH THE'. Problems Solved: B-348, B-351, Vol. 16, No. 1, pp. 90, 91;
B-352, B-353, B-354, Vol. 16, No. 2, pp. 185-186; H-273, Vol. 16, No. 6,
p. 568.
HUNTER, J. A. H.
407-411.
IV'IE, JOHN.
"Congruent Primes of Form (8r + 1 ) , "
Problems Solved:
Vol. 16, No. 5, pp.
B-352, B-353, Vol. 16, No. 2, p. 185.
JONES, BURTON W. "A Second Variation on a Problem of Diophantus and Davenport," Vol. 16, No. 2, pp. 155-165.
KELLY, LUCILLE. "Entry Points of the Fibonacci Sequence and the Euler 0 Function," Vol. 16, No. 1, pp. 47-50 (co-author, Joseph J. Heed).
KIMBERLING, CLARK. "Strong Divisibility Sequences with Nonzero Initial Term,"
Vol. 16, No. 6, pp. 541-544.
KLAMIN, M.
Problem Solved:
H-270, Vol. 16, No. 5, p. 479.
KNUTH, DONALD E. "Identities from Partition Involutions," Vol. 16, No. 3,

pp. 198-212 (co-author, Michael S. Paterson).
K0CHER, FRANK. Problem Proposed: B-376, Vol. 16, No. 2, p. 184 (co-proposer,
Gary L. Mullen).
KRAVITZ, SIDNEY. Problem Proposed: B-348, Vol. 16, No. 1, p. 90.
Solved: B-348, Vol. 16, No. 1, p. 90.
LA ROSA, B. DE.
522.
Problem
"Primes, Powers, and Partitions," Vol. 16, No. 6, pp. 518-
1978]
VOLUME INDEX
575
LARSON, PAUL. "The Golden Section in the Earliest Notated Western Music,"
Vol. 16, No. 6, pp. 513-515.
LORD, GRAHAM. Problems Solved: B-346, B-347, B-349, B-350, B-351, Vol. 16,
No. 1, pp. 89-91; B-352, B-353, B-354, B-355, B-356, Vol. 16, No. 2, pp.
185-186; B-358, B-359, B-361, B-362, B-363, Vol. 16, No. 5, pp. 474-476;
B-365, B-366, B-367, B-368, B-369, Vol. 16, No. 6, pp. 563-565.
MANA, P. L. Problems Proposed: B-373, Vol. 16, No. 1, p. 88; B-354, Vol. 16,
No. 2, p. 185; B-358, Vol. 16, No. 5, p. 474; B-392, B-365, Vol. 16, No. 6,
pp. 562, 563.
METZGER, J. M. "Fibonacci Sine Sequences," Vol. 16, No. 2, pp. 119-120 (coauthor, M. B. Gregory) .
MILSOM, JOHN W.
pp. 185-186.
Problems Solved: B-352, B-353, B-354, B-355, Vol. 16, No. 2,
MOHANTY, S. P.
384.
"The Number of Primes Is Infinite," Vol. 16, No." 4, pp. 381-
MULLEN, GARY L. Problem Proposed: B-376, Vol. 16, No. 2, p. 184 (co-proposer,
Frank Kocher).
MULLIN, A.
Problem Proposed:
H-287, Vol. 16, No. 5, p. 477.
0'CALLAHAN, T. Problem Proposed: B-360, Vol. 16, No. 5, pp. 474-475. Problem Solved: B-360, Vol. 16, No. 5, pp. 474-475.
PATERSON, MUCHAEL S. "Identities from Partition Involutions," Vol. 16, No. 3,
pp. 198-212 (co-author, Donald E. Knuth).
PECK, C. B. A. Problems Solved: B-346, B-348, Vol. 16, No. 1, pp. 89-90; B352, B-353, B-354, B-355, Vol. 16, No. 2, pp. 185-186; B-364, Vol. 16, No.
6, p. 563.
PIGNO, LOUIS. "Interpolation of Fourier Transforms on Sums of Fibonacci Numbers," Vol. 16, No. 3, pp. 193-194 (co-author, Robert E. Dressier); "Topological, Measure Theoretic and Analytic Properties of the Fibonacci Numbers," Vol. 16, No. 3, pp. 195-197 (co-author, Robert E. Dressier).
PONNUDURAI, T. Problems Solved:
B-366, B-369, Vol. 16, No. 6, pp. 563,565.
PR I EL IPP, A. G. Problems Solved: B-346, B-347, B-351, Vol. 16, No. 1, pp.
89-91; B-352, B-353, B-354, B-355, B-356, Vol. 16, No. 2, pp. 185-186;
B-364, B-365, B-366, B-367, B-369, Vol. 16, No. 6, pp. 563-565. H-273,
Vol. 16, No. 6, p. 568.
ROBBINS, NEVILLE.
pp. 515-517.
"On Fibonacci Numbers Which Are Powers," Vol. 16, No. 6,
SHANNON, A. G. "Fibonacci and Lucas Numbers and the Complexity of a Graph,"

Vol. 16, No. 1, pp. 1-4; "On the Multiplication of Recursive Sequences,"
Vol. 16, No. 1, pp. 27-32; "Pellian Diophantine Sequences," Vol. 16, No. 2,
pp. 99-102. Problem Proposed: B-382, Vol. 16, No. 5, p. 473. Problems
Solved: B-346, B-347, Vol. 16, No. 1, pp. 89-90; B-352, B-353, B-354, B355, B-356, B-357, Vol. 16, No. 2, pp. 185-187; H-269, Vol. 16, No. 5, p.
478; B-364, B-365, B-366, B-367, B-369, Vol. 16, No. 6, pp. 563-565.
SHIELDS, CHARLES.
Problem Solved: .B-359, Vol. 16, No. 5, p. 474.
576
VOLUME INDEX
Dec. 1978
SINGH, SAHIB. Problems Solved: B-358, B-360, B-361, B-362, B-363, Vol. 16,
No. 5, pp. 474-476 (co-solver, Roger Engle); B-359, Vol. 16, No. 5, p. 474
(co-solvers, Benjamin Freed & Roger Engle).
SINNAMON, GORDON.
Problems Solved:
SMITH, PAMELA GRAVES.
B-351, Vol. 16, No. 1, p. 91.
"Expansion," Vol. 16, No. 2, p. 112.
SOMER, LAWRENCE. Problem Proposed: B-386, Vol. 16, No. 5, p. 473.

Solved: B-351, Vol. 16, No. 1, p. 91.
SONNTAG, ROLF.
Problem Solved:
Problem
B-364, Vol. 16, No. 6, p.
STEINER, RAY. "On Nth Powers in the Lucas and Fibonacci Series," Vol. 16,
No. 5, pp. 451-469; "On fcth Numerical Centers," Vol. 16, No. 5, pp. 470471.
STERN, FREDRICK.
Problem Proposed:
B-374, Vol. 16, No. 1, p. 88.
SUN, HUGO S. "Embedding a Group in the pth Powers," Vol. 16, No. 1, p. 4.
SUTTENFIELD, JAMES M., JR.
"A New Series," Vol. 16, No. 4, pp. 335-343.
T0SCAN0, L. "Some Results for Generalized Bernoulli, Euler, Stirling Numbers,"

Vol. 16, No. 2, pp. 103-112.
TRIGG, CHARLES W.
VINCE, ANDREW.
407.
WACHTEL, M.
Problem Solved:
B-348, Vol. 16, No. 1, p. 90.
"The Fibonacci Sequence Modulo /I/," Vol. 16, No. 5, pp. 403-
Problem Proposed:
B-391, Vol. 16, No. 6, p. 562.
WADDILL, MARCELLUS E. "Some Properties of a Generalized Fibonacci Sequence

Modulo m," Vol. 16, No. 4, pp. 344-353.
WHITLEY, ROBERT. "Fibonacci Numbers in Coin Tossing Sequences," Vol. 16, No.
6, pp. 539-541 (co-author, Mark Finkelstein).
WHITNEY, RAYMOND E. "Geometric Sequences and the Initial Digit Problem," Vol.
16, No. 2, pp. 152-154. Problem Proposed: H-271(corrected), Vol. 16, No.
5, p. 480. Editor: "Advanced Problems and Solutions," Vol. 16, No. 1, pp.
92-96; Vol. 16, No. 2, pp. 188-192; Vol. 16, No. 5, pp. 477-480; Vol. 16,
No. 6, pp. 566-569.
WULCZYN, GREGORY. Problems Proposed: B-370, Vol. 16, No. 1, p. 88, H-279,
Vol. 16, No. 1, p. 92; B-353, Vol. 16, No. 2, p. 186, H-284, Vol. 16, No.
2, p. 188; B-383, B-384, Vol. 16, No. 5, p. 473, H-288, Vol. 16, No. 5, p.
477; B-389, Vol. 16, No. 6, p.
, H-290, H-294, Vol. 16, No. 6, pp. 566567. Problems Solved: B-346, B-347, Vol. 16, No. 1, pp. 89-90; B-352, B354, B-355, B-356, Vol. 16, No. 2, pp. 185-186; B-360, B-362, B-363, Vol.
16, No. 5, pp. 474-476; B-364, B-365, B-366, B-368, Vol. 16, No. 6, pp.
563-565, H-273, Vol. 16, No. 6, p. 568.
ZEITLIN, DAVID. "An Inequality for a Class of Polynomials," Vol. 16, No. 2,
pp. 128-129, 146. Problems Solved: B-237, B-349, Vol. 16, No. 1, pp. 90,
91; B-352, B-353, B-356, Vol. 16, No. 2, pp. 185, 186.
ZENZ, F. A. "The Fluid Mechanics of Bubbling Beds," Vol. 16, No. 2, pp. 171183.
ZWILLINGER, DAN.
Problem Proposed:
B-380, Vol. 16, No. 2, p. 184.
SUSTAINING MEMBERS
*H. L. Alder
*J. Arkin
Do A* Baker
Murray Berg
Gerald Bergum
Je Berkeley
George Berzsenyi
C. A. Bridger
John L. Brown, Jr.
Paul Bruckman
Paul F. Byrd
Co R0 Burton
Lo Carlitz
Go Do Chakerian
P. J, Cocuzza
Mo Jo DeLeon
Harvey Diehl
Jo Lo Ercolano
^Charter Members
Do R* Farmer
*James Maxwell
Harvey Fox
R. Ko McConnell, Jr.
Eo To Frankel
*Sister M. DeSales McNabb
Ro Mo Giuli
Lo P. Meissner
*Ho Wo Gould
Leslie Miller
Nicholas Grant
F. J. Ossiander
William Greig
Fo Go Rothwell
Vo Co Harris
Co Eo Serkland
Ao Po Hillman
A. G o Shannon
*A F9 Horadam
Jo A. Sehumaker
*Verner E Hoggatt, Jr 8 D. Singmaster
Virginia Kelemen
Co Co Styles
Ro Po Kelisky
L. Taylor
Co Ho Kimberling
Ho Lo Umansky
Jo Krabacker
*Lo Ao Walker
George Ledin, Jr e
Marcellus Waddill
*C. To Long
Paul Willis
Jo Ro Ledbetter
Co Fo Winans
Do Po Mamuscia
Eo L o Yang
ACADEMIC OR INSTITUTIONAL MEMBERS

DUKE UNIVERSITY
Durham^ North Carolina
ST. MARY f S COLLEGE

St. Mary's College5 California
SACRAMENTO STATE COLLEGE

Sacramento, California
UNIVERSITY OF SANTA CLARA

Santa Clara, California
SAN JOSE STATE UNIVERSITY

San Jose, California
WASHINGTON STATE UNIVERSITY

Pullman -9 Washington
THE BAKER STORE EQUIPMENT COMPANY
Typed by
~*J0 ANN VTSEh
Campbell, California

16 6

Uploaded by

Copyright:

Available Formats

16 6

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

16 6

Uploaded by

Copyright:

Available Formats

J&e Fibonacci Quarterly

THE OFFICIAL JOURNAL OF

WITH THE COOPERATION OF

All subscription correspondence should be addressed to Professor Leonard

ON THE DENSITY OF THE IMAGE SETS OF CERTAIN

Let n be a fixed but arbitrary nonnegative integer. It is known (see [1],

fl(fc, 3?) = {^(ar) |fe x <_ p ) ,

for some n}, and

Our objective here is to prove some results concerning the asymptotic

EXISTENCE AND COMPUTABILITY OF THE DENSITY

j) , 0 <_ d <_ J be as described above.

To prove 2.2, let us suppose that

ON THE DENSITY OF THE IMAGE SETS OF CERTAIN

Suppose that dk_

=. 0 (or that dk_1

Since /(0, j) = 0, it must be the case that

We have therefore exhibited a one-one correspondence between the elements of

Our next result will enable us to find a relationship between

^ = E/^> ^ + E fw** o) + E /WJ o) < KMK

In p a r t i c u l a r , kMk < kl i f fc >. k0 . Hence, we c e r t a i n l y have (d + l)/c! - 1 +

D((k + 1)! - l) = J^D(dk\9

ON THE DENSITY OF THE IMAGE SETS OF CERTAIN

series of nonnegative terms bounded by Ak , hence is convergent. Let

Note that we have just shown that 0 <_ L <_ 1.

Then, 2.6 yields

Since V ^ e^ = 0(1) as & -> , we have

D(kl - 1) = Lk! + o(fc!).

= dfc!L + o((k + 1)!) + o((k + 1)!),

Now let n = ^ djjl

be any nonnegative integer. Then D(n) = D(dkkl

+ ..-) + o((k + l)l)

= dkk\L + (aVi(k-D! + ...) +

ON THE DENSITY OF THE IMAGE SETS OF CERTAIN

Applying the same type of reasoning yields

- 1)! + ...) = dk_(k

Continuing in this manner, we obtain

D(n) = Ll n - d ^ ! j + M Z dtf\ J + (k - k0),

= L - L o(l) + 0(1) + 0(1),

which implies that the density of 9 is L, so the proof is complete.

We have L < 1 if and only if the function T(n)

EXISTENCE OF THE DENSITY WHEN f(d,

j) = 0(j!/j 2 log2j) uniformly in j , then the density

Let D and fi be as before. If n = Y] d j \

ON THE DENSITY OF THE IMAGE SETS OF CERTAIN

Suppose that r <_ s <_ t(r

< t) and s < (k + 1) ! ; then,

D(r, t) = D(r, s) + D(s + 1, t) - |Q(r, s) H Q(s + 1, t) | .

D(r, t) = D(r9 s) + D(s + 1, t) + 0(k!/k2 log2k).

In particular, if r = 0, s = (/c - 1) ! - 1, and = k\ - 1, we obtain

- 1) = 0(0, (fc - 1)! - 1) + D((k - 1 ) ! , k\ - 1))

+ 0((fc - l)!/(fc - D 2 log2(k - 1)).

+ 0((fc - 2)!/(fe - 2 ) 2 log2(/c - 2)) + '

Now, by 3.2, we have

dk\) + D(dk\ + 1, (dfc+ 1) ! + 0(kl/k2

and by repeated application of 3.2, we obtain

dk\) + D(dkl + 1, dk\ +

Since all integers x which satisfy dk\ + q\ x <_ dk\ + (q + 1) S - 1

+ q\, dk\ + (q + 1) - 1) = D(q\,

(cf. the argument used to prove 2.2).