You are on page 1of 100

^h ^Fibonacci

Quarterly

THE OFFICIAL JOURNAL OF
THE FIBONACCI
ASSOCIATION

VOLUME 15

NUMBER 1

[l^WST

CONTENTS
Residues of Generalized Fibonacci Sequences
C. C. Yalavigi
Composites and Primes Among Powers of Fibonacci
V. E. Hoggatt, Jr., and
Numbers, Increased or Decreased by One
Marjorie Bicknell-Johnson
Divisibility by Fibonacci
and Lucas Squares
V. E. Hoggatt, Jr., and Marjorie Bicknell-Johnson
Letter to the Editor
. . . . . . . . . John W. Jameson
An Elementary Proof of Kronecker's Theorem
J. Spencer
Fibonacci Numbers in the Count of Spanning Trees
Peter J. Slater
The Diophantine Equation
(x1 + x2 + - +xnf
= x3 + x32 + - + x3
. . . . W. R. Utz
An Application of W. Schmidt's Theorem
Transcendental Numbers and Golden Number
Maurice Mignotte
The Reciprocal Law
W. E. Greig
Binet's Formula Generalized
A. K. Whitford
On the Multinomial Theorem
David Lee Hilliker
A Fibonacci Formula of Lucas and its
Subsequent Manifestations and Rediscoveries . . . . . . . . . . . . . . H. W. Gould
Numerator Polynomial Coefficient Arrays for Catalan
V. E. Hoggatt, Jr., and
and Related Sequence Convolution Triangles. . . . Marjorie Bicknell-Johnson
Fibonacci-Like Groups and
Periods of Fibonacci-Like Sequences
.Lawrence Somer
Solution of a Certain Recurrence Relation
.Douglas A. Fults
On Tribonacci Numbers
and Related Functions
Krishnaswami Alladi and V. E. Hoggatt, Jr.
Sums of Fibonacci Reciprocals
W. E. Greig
Fibonacci Notes — 5. Zero-One Sequences Again
L. Carlitz
On the A/"-Canonical Fibonacci Representations of Order N . . . . . Robert Silber
A Rearrangement of Series Based on a
Partition of the Natural Numbers
H. W. Gould

1
2
3
8
9
11
14
15
17
21
22
25
30
35
41
42
46
49
57
67

n

A Formula for £ ' Fk (x)yn"k

and its Generalization. . . . . . . M. N. S. Swamy

73

7

Some Sums Containing the Greatest Integer Function
Wythoff's Nim and Fibonacci Representations

Advanced Problems and Solutions
Elementary Problems and Solutions

FEBRUARY

L Carlitz 18
Robert

Silber

85

Edited by Raymond E. Whitney
Edited by A. P. Hillman

89
93

1977

tfie Fibonacci
THE OFFICIAL JOURNAL

Quarterly

OF THE FIBONACCI

ASSOCIATION

DEVOTED TO THE STUDY
OF INTEGERS WITH SPECIAL
PROPERTIES
EDITOR
V. E. Hoggatt, Jr.

EDITORIAL BOARD
H. L. Alder
Gerald E. Bergum
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
Rro. A. Brousseau
Calvin D. Crabill
T. A. Davis
Franklyn Fuller
A. F. Horadam
Dov Jarden

L. H. Lange
James Maxwell
Sister M. DeSales
McNabb
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 9 5 0 5 3 .
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 typed,
double-spaced. Drawings should be made the same size as they will appear in the
Quarterly, and should be done in India ink on either vellum or bond paper. Authors
should keep a copy of the manuscript sent to the editors.
The Quarterly is entered as third-class mail at the University of Santa Clara Post
Office, California, as an official publication of the Fibonacci Association.
The Quarterly is published in February, April, October, and December, each year.

Typeset by
H I G H L A N D S COMPOSITION SERVICE
P. 0 . Box 760
Clearlake Highlands, Calif. 95422

RESIDUES OF GENERALIZED FIBONACCI SEQUENCES

C. C. YALAVIGI
Government College, Mercara, India

Consider a sequence of GF numbers, [wn(b,c; P.Q)}^.
ing theorem.

For b =c= 1, L Taylor [1] has proved the follow-

Theorem.
The only sequences which possess the property that upon division by a (non-zero) member of
that sequence, the members of the sequence leave least +ve, or -ve residues which are either zero or equal in
absolute value to a member of the original sequence are the Fibonacci and Lucas sequences.
Our objective is to consider the extension of this theorem to GF sequences by a different approach, and show
that a class of sequences can be constructed to satisfy the property of this theorem in a restricted sense, i.e., for
a particular member only. For convenience, wn(b,l; 0,1), wn(b,1; 2,b), wn(b,1; P,Q) shall be designated by
u
n> vn, Hnr respectively.
Let Hk+r = (-l)r~
Hk.r (mod Hk). Assume without loss of generality, k to be +ve. We distinguish 2 cases:
(A)0<r<k,
and (B) r>k.
(A) Evidently, the members leave least residues which are either zero or equal in absolute value to a member
of the original sequence.
(B) Allow | / / _ 5 + ; | < \Hk\ < \H.S\. Let
H2k+S - (-Vk+S'1H.S

(D

(mod Hk), H2k+S+1

Clearly, the property of above-cited theorem holds for
_

(_1)k+s-lH_s

HQ ( m o d Hj<)i

= (-Vk+sH.s.j

(-Dk+sH-s-i

for some c such that -s + 1 < e < 2k. Denote the period of

Hk).

iff

{Hp}^,

and

(mod

= HQ + 1 (mod

Hk),

{ Hn (mod Hk)} Q by k(Hk). Rewrite the given

sequence as { Hp'}™^, where H'n' = Hn. Set k' = k + t, s' = s - t, and 2'=£ + t. Then, it is easy to show that
k(Hk)

= 2k + s-z,

k(H'k>) = 2k'+ s'-

c',

and

k(Hk)

=

k(H'k-).

We assert that ^ / / ^ ' ^ is even, for f = ^ - z)/2 obtains s'= si', k(H'k') = 2k', and the substitution ois-s. = 2t + 1
leads to s' - si' = 1, k(H'k') = 2k'+ 1, which is a contradiction. Hence, it is sufficient to examine the following
system of congruences, viz.,
(2)

H'2k- = H'0 (mod H'k-\

H'2k+1 = H', (mod H'k>).

These congruences imply
(3)

H2k+t

= Ht = (-Vk+t~1H^(mo6Hk)
=

= (-l)k~1

{Ht-

(2Q - bP)ut } (mod Hk )

k 1

(-1) ~ {Pvt-Ht}(m^Hk).

Therefore, (i) P = 0, Q = 1, and (\\) P = 2, Q = b, furnish readily the desired sequences, and they are the only
sequences for which the property of L. Taylor's theorem holds. For the restripted case, by using the well known
formula Hn = Pun-i + Qun, it is possible to express H.s = HQ (mod Hk), and //_ s _; =H%+1 (mod Hk) as two
simultaneous equations in P, Q, and obtain their solution for given s, si, and k. In particular, the latter case may
be handled by using k(H'k') = k(uk'), where H'k' is selected arbitrarily to satisfy k'= k(uk')/2 and
Hk- = Puk-„i + Quk,
determines/3 and Q.
Example:

H'g = 19,

k(H'9) =18,

P = 9,
1

Q =

-5.

F™ .Bn always has (A . since Fn± 1 = Fn-2Fn+2. This leaves only the case F™k + /. since A 3 ± B is always factorable. Now. for k= 1. 105. It is easily shown that Fn ± 1 is also composite for any n. we raise the question of when F™ ± 1 is composite. "On a Theorem of L Taylor. 2 Fn + 1 is composite for each n > 4. 1967). if k ^ 3 (mod 6).B) as a factor. prime for / = 0. Yalavigi. C. + / = 102'-t+1. 5. given by F-i = 1 = F2. ******* COMPOSITES AND PRIMES AMONG POWERS OF FIBONACCI NUMBERS. then Fk is odd. We note in passing that.2 RESIDUES OF GENERALIZED FIBONACCI SEQUENCES FEB. every even m except powers of 2 can be written in the form (2j + 1)2' = m.B)= 1.1m is composite except when (A. Fn + / = Fn+1Fn^1 . California 95192 It is well known that. We return to F™k + I For/77 odd. F™ is odd. 2. 298-304. C.. = 6 (mod 10).Ar=0(mod 5 ) . ''Residues of Fibonacci-Like Sequences. / ^ . INCREASED OR DECREASED BY ONE V. Since F3k=Q (mod 10). suppose we deal with F™k ± I Since An . JR. Fn+1 * Fn + Fn-i. that is. 3 (Oct. Here. while Fn . m =f 2'. Edn. from the known factors of A + B . and F™± 7 is even and hence composite. and 8m ± 1 = (23 ) ^ ± 1 = (2m ) 3 ± 7 is always composite. 1. San Jose." The Fibonacci Quarterly. F2' + 1 has the form 10t + 7.F6=8 = 23./ i s composite for/7 > 7. HOGGATT.?2 + /. F™k + 1 is composite. Vol. We can sum- 2. 1977 REFERENCES 1. First. E. It is worthwhile to note the actual factors in at least one case. 2. where m = 2'. No. k^ marize these remarks as Theorem (mod 5). pp. p. It is an unsolved problem whether or not 22' + 1 has other prime values. 4 (1970). L Taylor. if k£0 (mod 3).. we have the Fermat primes. . and MARJORIE BICKNELL\IOHNSON San Jose State University. we see that F™k . 4 but composite for / = 5. when k = 2. It is th ought that Fg + 1 is a prime. k ± 3. i> 2. Theorem 1. among the Fibonacci numbers Fn. m odd. . 0 (mod 5). If m is even. 6. so that F™k + 1m has the factor (F^k + 11 and hence is composite. Thus. When k= 1. The answer is yes. but if k = 3 (mod 6). so that F3k+1m = (F23k)2i+1 + (12)2i+1. Since Fk+2Fk-2-F2k F k+lFk-l~ (~Vk+1 = Fk = (-V moving Fk to the right-hand side and then multiplying yields F k-2Fk-lFk+lFk+2 ~ Fk-1. then only 30 can be guaranteed as a divisor. m m which. 3."Math. must have (Fjk + 1) as a factor. then Am + Bm is known to have the factor (A + B). and hence.1 is composite. k ?3s. F™ + 1 is composite. We now note that Fk-Fk = Fk-2Fk-iFkFk+iFk+2 which causes one to ask if this is divisible by 5!. F™k + /is composite. k4ft Since F?/.

.3) for k > 3. S\nce(Fm/Fm.p and that am = aFm + Fm_1f (1. E. We let a= (1 +\/5)/2. Since Fm m divides all terms of (1. DIVISIBILITY BY OTHER FIBONACCI POWERS The proof of (1. In [ 2 ] . and since Fj = F2= 1. Then it is well known that the Fibonacci numbers Fn are given by (1. Fmr *(rj) FmFr~l1F1 ^ rFmFr~l1 (mod F2m) . JR. Matijasevic's Lemma 17 was proved by Hoggatt. Since (Fm. . Lemma 17 states that F2m divides Fmr if and only if Fm divides r. we extend Lemma 17 to its counterpart in Lucas numbers and generalized Fibonacci numbers and explore divisibility by higher powers. % . (r . we repeat it here.yJs)/2. Phillips and Leonard using an identity for Fmr.2) with the binomial theorem expansion of amr and fimr gives „mr r n/nr .. and Fmr = rFmFrr^21(Fm. Fm-f) = 7.DIVISIBILITY BY FIBONACCI AND LUCAS SQUARES V. and MARJORIE B5CKNELL JOHNSON San Jose State University.4) if and only if Fm\r. it follows easily that F m\Fmr (1. (1-3) k=0 Since FQ= 0 and Fm divides all terms for k > 2.2) (3m = ^Fm + Fm. INTRODUCTION In Matijasevic's paper [1] on Hilbert's Tenth Problem..1+kFm) (mod Fm) . 2. Here. proceeding in a manner similar to that of Section 1 F = rF Fr~1 + (iLzJl F2 Fr~2 (mnri F3 ) When r is odd.V/2 = k is an integer. (5= (1 .1)=l Fm^(Fm.4) can easily be extended to give results for divisibility by higher powers. Combining (1.1) Fn = 5 L ^ | _ a. and Fm \ F^.1 . / _/r nk \ k=0 so that Fmr-J:(rk)FkmFr-11Fk. California 95192 1. San Jose. HOGGATT.1) and (1.1+kFm) so that F I Fmr if and only if F \r. Since that proof is the basis for our extended results.

then Fm\Fmr\i \UFm. we have proved Theorem 2.6.1)/2 = j and (r- F2 Fr~2 + r(r ~ 1 As before. Wheneverr=£*± 7.3).6F2m_1)= . If r is even. since F-j = F2.1 ) = 1.1 ) k If k\Msk'1 i r ~ 1 then „n-k\ If U) • tr-1\ \k. 6k±L ^Fm Fm-7 + 2(r . Fsm\Fmr\\\ F^1\r. Theorem 2. L FmFr^1(2Fm. s = 1.3. Thus.7 and Fj = 2. again considering the first terms of (1. m f 4q. then \UFm. Lemma. Whenever/^ is odd.2. F m\Fmr ™ FS~1 \r.1)!± 1.2.6. r = + 3 (mod F4m). Note that (Fm. Whenever r= k(s. whenever r = k(s . then r = Msn~1 for some integerM. In the general case. F4m\Fmr\\\ F%\r.6)= (mod F%) .rF Fr~1 + r(r ~ 1) since Fm divides every term for k > 4. F m\Fmr iff F%1 \rt s = 1. so that Fmr^rFmFr^1(F2m. Cases'/r = 7 and £ = /-are trivial. Proof.1i-jFmFm. 4q.M k U . then snk\(kr). If s11'1 \r. Thus. If sn~1\r. m ? 4q.3. \in<r.1)(r .3) we can write F . \ir=6k± 1. F6m\Fmr and also iff F%\r. 1 if /?? t 5 ^ /?? ^ The casesr = 6k±2 and /* = 6k ± J are similar.5.\r. F^\Fmr\ii If r = 5Ar. Fm.4.3. but the derivations are quite long. Whenever r = 6k± 1. Next. or m ? 3q. f ^ | / W i f f F^|r. then /r </7 < r. we make use of a Lemma to prove a final theorem for the general case. and tr\ \k) = L C. we can state that. F^\Fmr Sinilarly.4. F m\Fmr iff F%1 \r. from (1.1.1 I ' . We summarize these cases in the theorem below.1 W >(r~2> iff FJ*. since f f m . we have proved Theorem 2.1)1 ± 1. by carefully considering the common denominatorof the fractions generated from the binomial coefficients.V = Ms "~1 k 1 [I " 1 \ = Msk~1sn~k k. then/V. + iF2m) F^\r. Whenever rh odd. Continuing in a similar fashion and considering the first terms generated in the expansion of Fmn we could prove that whenever r = 6k± 1. Wheneverm ^ 3 ? .5. F5m\Fmr iff F4m\r. then F4m \Fmr iff F * | £ .4 DIVISIBILITY BY FIBONACCI AND LUCAS SQUARES [FEB. F3m\Fmr\\\ F^\r.2Fm-1)= + (r- l)Fm) (mod F3m) and only if F^\r.2)F2J ^~ 1.1 2 1.jFm F m-1 (®Fm-1 F3 Fr~3 (mnrl F4 ) 1)(r~ 2)/3 = / f o r integers/ and I.4. Whenever m 13q. Case s = 1 is trivial. k=1.n.-.

then Fbm divides each term. (Ln^z^/SF. we will derive and extend the counterpart of (1. s. When k= 7. then then F^ F ^ divides divic ( £ ) for k (Fm< Fm-1> = h so that if F^" *'r. DIVISIBILITY BY LUCAS SQUARES Next. then IfIf FF8^^1 divides/. pv /.1 > rf and where M is an inte- Fsm\Fmr. a contradiction. •••. 3. then 1 = r. It is well known that. Proof.1) Ln = a" +f and /nni _ „m _ Lm — \J5 Fm m _ Lm + \J5 Fm Combining (3. Theorem. or 6.).2) with the binomial theorem expansion of amr and Lmr = amr + $mr =(^m±^Im_) "+ fimr. r f mr = J2 (k) FmFm-1Fk k=0 - If k>s. nk-1 I r -1 \ Ms" n-k\ It is impossible for n > r. \\r=2t.1 > r. s 1 then FFssmmdivides dividesrF rFmmFFr^lr^[ divides/. since (^\ is an integer. .3) are divisible by L^ except the last term. / = [r/2]. k = pqNf wherep is some prime.1) and (3. —. But /r <pq for/7 > 2 and q > k > 0. ar and Fsm divides each successive term for k= 7. That is.»"" ^%rk) . (2t\ \2tl L0 Fr2trt rn m ° = t 2t 35 rF m - the last term is . 4.m 2 J=o When/ is odd. the entry point of F in the Fibonacci sequence \smF~ for k = 1. 5. 2. The entry point of a numbers in the Fibonacci sequence snee is the subscript of the first Fibonacci Fibonaci number of which n is ai divisor. since in the kth term we always have a factor Fm while F^ appears as a factor of ( f ) These theorems allow us to predict the entry point of Fm in the Fibonacci sequence in limited circumstances. n. analogous to (1. Si nee FQ= 0. . the Lucas numbers Ln obey (3. 3. then sn~1 \r implies Ms11 ger. the term i s r F ^ F ^ i .1977] DIVISIBILITY BY FIBONACCI AND LUCAS SQUARES 5 then k\Msq {rkZ_]) k < q < „.4) for the Lucas numbers. all terms are zero. If F%1\r. When m / 3j or 4j. \r. where n. Fbm divides the term k = 0. All terms on the right of (3.1). We let/ = 21 and simplify to write (3-3) [r/2] Lmr2r'1 = £ i=0 (^ Ltf'F**1 . so that k must divide a: 7 . B u t s " " 7 >rtars> 2. If n < r. s by the Lemma.

divides all terms of (3.6 DIVISIBILITY BY FIBONACCI AND LUCAS SQUARES [FEB.m > 7. Whenever/"is even. Z. If r = 2t./. then Lm \Lmr for any r. and Lm)((2t+1). m> U Return to (3. m > 7. then (Lm. To determine when Lm \ Fmr. 3. However. return to (3. Lm )(Lmr. 2) = 1. Lm divides all terms of (3. then Lm is even. If m = 3q.7. (Recall that sj~5 =a-$. Whenever/. Lm J(2r~1 Lmr. we obtain [r/2] £ i=0 j5Fmr=M L£2i~1F%+1-(j5)2i+1-2 ( n ^ ) [r/2] 2r-1Fmr= £ (3-4) (2/. m > 7. all terms are zero. m > 1.[r/2] = t. It is trivial that Fm\Lmr for m / 1. A * | Fmr iff A ^ | r. an m m%Fmrt°r Y > 1However. W h e n / . if r = 2t + 7. when r is even. 2. Lm I f 2 m f iff Lm \ t or.4) for / < [r/2] . 4. Setting/ = -?/ + 7 and rewriting. so that L3m\Lmr iff L2m\(2t+1). The same result holds for Lm even. We summarize these results as Theorem 3. If m / 3q. and use (1.1) and the binomial expansion of a W r a n d j8A77/" to write an expression for Fmr in terms of Lm.1) and (3.1.4) f o r / < //-/?/. Since 5 X Lm for any m and Lm >[ Fm for any m > 1.^ l ^ . f > 7. when r = 2t+ 7.3) and notice that.^.is odd. We can also determine criteria for divisibility of Lmr by Fm and Fmr by Lm. and Lm \ 5t for any t > 0. then the terms i= t . m > 1. if and only if Lm\(2t+ 1). L2m\Lmr iff and Lm\r.n for other values of m. so that Lm\22tL(2t+i)m. which case depends upon the fact that 4 is the largest powerdf 2that . whenr = -2f + 7. Lm\Lmr iff Lm\r. That is. If/77 = 3q > 1. since F ^ >^ Z. if r is odd.0 F2t+1rt \2t+ 1 I m m = ctF2t+1 m which is not divisible by Lm. the last term is r-2t [2t2t1) and 22tL(2t+Dm and L^\Lm(2t+i) LmF*5<=(2t+V5tLmF* is divisible by Lm if and only if Lm\(2t + 11 m > 1. Thus. since Lm j{ Fm.7 and i = t give L Now. and Lm^1. ^ ^ /w > 7. the last term is (2t+1\ .^ m f c t j t f a n d only if Lm \2t If A m is odd.) lr _ Rmr = (Lm + \J5 Fm\ _ lim-\l5 Fn 1 j=0 Here. all terms except the last are divisible by Lm.7)C2/-7^/^- i=0 Notice that.2). whenever/ is even.

E (rk) dwcV^w. Pm \Pmr'^pm +Pn-1. \r. Theorem 4. (4.1) also holds for Pell numbers. Similarly.Fm\Fmr then /7 |r and the smallest such r\sp itself. Pn+1 = 2Pn are given b y / ^ = fn(2). m = 3q. L m I F2mt If /-is odd. Fmp < Fp^+p Are there other primes than Fibonacci primes for which p2\Fn. . if/? .3) is also true if we replace Fn by fn(x) (see [2]). eventually.1. then Lm \ Fmr. DIVISIBILITY BY FIBONACCI PRIMES From [ 3 ] . Ln+1(x) = xLn(x) + fn-rfx).. If Lm is even. or. for Z. and 13 \Fgj = Fy.D k=0 Notice that Fm = fm(1) and Lm = Lm(1).1. Thus. Ln-j(x). If r is odd. The Pell numbers 1.13. so that p2\Fmp. For the Pell numbers/>„. Thus. Pn. From P\ Fp±i one can easily show that/? \Fp2±p. 70. 4.. GENERALIZED FIBONACCI NUMBERS The Fibonacci polynomials fn(x) are defined by f0(x) = 0. For example. n<p(p ± 1)? . If/? is a Fibonacci prime. where of course. and L3m\Fmr and L3m | F2mt L2m\r. 13. Lm'J(Fmr. but. then Ln(2) = Rn.. We could write similar theorems for other generalized Fibonacci numbers arising from the Fibonacci polynomials. since Fy= 13.2. since (3. we can write ww . Notice that. L3m\F2mt iff L2m\t. Lm contains exactly one factor of 2. which leads us to Theorem 4. If m is even. a multiple of F 12= 144 = 24-32. and the Lucas polynomials Ln(x) by L0(x) = 2. it can be shown in the same manner that L3m\Fmr iff L2m\r. while F mr = F(3q)(2t) = F6t contains at least three factors of 2. If/-is even. L m\Fmr iff Lm \r.e. If r is even.^ even. since F$ = 23 is a factor of % r . making Fmr = F^qs. 2. 13 < F131+13. We summarize these results as follows. Theorem 3. if the Lucas-analogue Rn of the Pell numbers is given by Rn = Pn+-j +Pn-i.2. and we can write. Since (1. 13 \Fi4. if Lm\2r'1 Fmr. note that the prime 13 enters the Fibonacci sequence earlier than that. and Lm \ 2t\\\ t = 2s for some integers. but squares of primes which are also Fibonacci numbers divide the sequence earlier than that. frfx) = I fn + lM = xfnM + Lj(x) = x. \f p = Fm. then m <p ± 1 since Fp±1 >/? for/? > 5. then Lm contains exactly two factors of 2. Rm \ Pmr iff Rm | r.3) and (3. [4] we know that a prime p\Fp-j or p\Fp+i depending upon if/? = 5k±1 or/7 = 5k ±2. Rm \ Rmr iff Rm \r. 'ff Lm \ * iff L2m 11 m > 1. (4.4) for /-even a n d / < [r/2] . 5. Fy . Thus. iff Further. . If m = 3q is odd. 5. since also Lm divides all terms of (3. 29.DIVISIBILITY BY FIBONACCI AND LUCAS SQUARES 1977] 7 divides the Lucas sequence. Fy. i.. 12.4) hold for Lucas and Fibonacci polynomials.

Graduate School of Science.40545. 2." The Fibonacci Quarterly. 1 (Feb. 13). Vol. Hoggatt. 5 ..8 DIVISIBILITY BY FIBONACCI AND LUCAS SQUARES FEB. "Twenty-Four Master Identities. 210-214.3361-23408. of the Academy of Sciences of the USSR. the smaller angles approximating 45 degrees as the sides increase: (1) 3 . it is possible by this means to obtain a sequence of Pythagorean triangles in which the ratio of the legs. AAAAAAA LETTER TO THE EDITOR March 20. New York. 4th Ed„. n = 2. 4 . . If #7 = l 92 = Z m = 1. known for centuries. 1 6 9 . V. However. 1971). Leonard. 1974 Dear Sir: I would like to contribute a note. as is well known. If the Fibonacci sequence itself is used (m = n = 7). Jr. 120. T. 9. Now let/77 and n be two integers. and Marjorie Bicknell. 9. It is easy to obtain m and n and good starting valuesg-j and#2 given/?/#..46817-564719. unless I missed it in a hasty reading. in his monograph Pythagorean Triangles (Scripta Mathematica Studies No. 2. "Enumerable Sets are Diophantine/'/Voc.E. John W. and 9k+9k-i are the sides of a Pythagorean triangle. m = -1. He shows also how to obtain Pythagorean triangles having one angle arbitrarily close to any given angle in the first quadrant (Chap. Following is a sequence of right triangles with integer sides. 47. In general.^ >gk-i • Then. n = 4. 3. Yeshiva University. and H./? > q). Oxford University Press. G.] . Yu V. 2 9 . 5. Following is another sequence of such "Pythagorean" triangles. "Some Congruences of the Fibonacci Numbers Modulo a Primep. approximates any given positive rational number/?/<7 (p and q positive non-zero integers. No. 2. Vol." Mathematics Magazine. but I shall leave all that for another communication. the smallest angle approximating 30 degrees as the sides increase: (2) 15. Phillips. E.. 3. No.326040. Let#£ and#£_/ be any two positive integers. Jr. He obtains sequence (1) above by a different method (Chap. 1-17. Vol. or paper to your publication expanding the topic presented below. For all I know.2 1 . Waclaw Sierpinski. 1 7 . this may be an old story.241-2911. Jr. 4). Hoggatt.119.652081 ••• The scheme for generating these sequences resembles that for generating the Fibonacci sequence 1.Matijasevic. M. letter. and so on. substitution in (4) and (3) gives the triangle sequence in (1) above. [Continued on page 10. the method differs from the one I have outlined. Wright. 1977 REFERENCES 1.H. 2 0 . 1680. 4. pp. Hardy and E. and let W) gk+1 = ngk + mgk-i to create a sequence of g's.8. 1962). An Introduction to the Theory of Numbers. No. but again.. the resulting triangle sequence is (2) above. pp. or of the hypotenuse to one leg. does not give this method of obtaining such triangle sequences. V. 1960. non-zero. If 91 = h 92 = 4. and there is more to the topic besides. 120. a triangle sequence results in which the ratio between the short sides approximates 2:1. 4 (September-October 1974).2 0 9 .. 11 (1970). (3) df-Sk-p ?9k9k-l.

California 90406 Kronecker's Theorem. Let PM = xn-J2 six""1 . This result was first proven by Kronecker using symmetric polynomials. The condition that/? is monic is necessary since p(x) = 5x .1. (Us. By (*) this implies (**) Setting K = (0 < / 7 . Us+j = Ut+j t-s. irreducible over the integers. for some 0 <s < t < (2N + 1)n + 1.6x + 5 has roots (3 ±4i)/5. It is also necessary that all roots a have \a\ = 1. •••. For let/? be the minimal polynomial of a = x + isji . Then \a\ = 1 butp((3) = 0 where (3 = y + i^T'-~yTf y = ->J2 .E ii<j+k 9 <o < i) . Us+(n-u) = (Ut./. Since the Us are integers there are < (2N + 1) possible Us and hence < (2N + 1)n possible sequences/7/5.x2 where x = ^/2 .Then all roots a are roots of unity.AN ELEMENTARY PROOF OF KRONECKER'S THEOREM JOEL SPENCER Santa Monica. such that all roots a of p have \a\= 1.-} defined by Uj = 0 [0 < i < n-2] Un-1 = 1 n Us = Y^ a/Us-i for s > n (*) Then us = E *. Us+j. Uf+f. Us+(n-l)).Therefore. Then l ^ i < £ IS/IN5 <£. Pro of of Theo rem.-<*/ > i=1 where a. an are the roots of/?. Ut+fn-u). Us+1. In this note we prove Kronecker's Theorem using Linear Recursive Sequences.1 and |j3 | > 7.. That is Us+j = Ut+j (0 < / < n. Let p(x) be a monic polynomial with integral coefficients. . i=i i=i independent of s. i£/i</v. Consider the sequence {u. .. £ sri* .1). •••.

10

AN ELEMENTARY PROOF OF KRONECKER'S THEOREM

E nM-iu*r

i=1

FEB. 1977

(o <j).

Setting x,- = %j(af - 1)
af+JXi = 0

£

(0 < j < n-

1).

i=1

The coefficient determinant
s
'n
s+1

a

1

77

det

=

^s+n-1

(a-j •••an)s

? o,

det

^s+n-1

^n-1

^n-1

since this is the Vandermonde matrix and the a,- are distinct since/? is irreducible. Hence the n linear forms are
independent, so
Xj = 0
(1 < / < n).
Some %j f 0 since Un-j i= 0. For that/, ay = 1. Since the a's are roots of an irreducible polynomial, by Galois
theory a!- = 1 for 1 <j<n.
Q.E.D.
Corollary.

Kronecker's Theorem holds even if p is not irreducible.

Proof.
We factor p(x) = Up,(x), where the p; are irreducible. All roots a of/7/ are roots of p so \a\ = 1
so all roots are roots of unity. But all roots of p are roots of some/7/ and hence roots of unity.
David Cantor has noted that the proof after (**) can be shortened using generating functions. For

2>,v=—xi=0

n-1

7 - £ a,*'

A(x)
xk-1

i=1

Hence
xnP(x~1)

= 7 - ] £ aixi\xk-

7

i=1

p(a) = 0 impliesp(a

1

) = 0 implies a

- 7 = 0, a~

= 1,zvak

= 1.

[Continued from page 8.]

I must tell you that I am short of proofs and most of the propositions would have to be presented as observations or conjectures. Co-authors with proofs are welcome.
Thank you for your attention to this letter. Please write and let me know whether the subject is of interest,
You are free, of course, to publish this letter or any part of it.
Sincerely,
John W. Jameson,
P. 0. Box 205
Edgewood, Maryland 21040

FIBONACCI NUMBERS IN THE COUNT OF SPANNING TREES

PETER J. SLATER*
Applied Mathematics Division, National Bureau of Standards, Washington, D.C. 20234

Hilton [3] and Fielder [1] have presented formulas for the number of spanning trees of a labelled wheel or
fan in terms of Fibonacci and Lucas numbers. Each of them has also counted thejiumber of spanning trees in
one of these graphs which contain a specified edge. The purpose of this note is to generalize some of their results. The graph theory terminology used will be consistent with that in [ 2 ] , Fk denotes the kth Fibonacci
number, and Lk denotes the kf Lucas number. All graphs will be connected, and ST(G) will denote the number of spanning trees of labelled graph, or multigraph, G.
A fan on k vertices, denoted N^, is the graph obtained from path Pk-i = 2, 3, •••, k by making vertex 1 adja—
cent to every vertex oiPk--/. The wheel on k vertices, denoted W^, is obtained by adding edge (2,k) t o / l / ^ .
That is, Wk= Nk + (2,k). Aplanar qraph G is one that can be drawn in the plane so that no two edges intersect;
G is outerplanar if it can be drawn in the plane so that no two edges intersect, and all its vertices lie on the same
face; and a maximal outerplanar graph G is an outerplanar graph for which G + (u,v) is not outerplanar for any
pair^/,1/ of vertices of G such that edge (u,v) is not already in G. For example, each fan is a maximal outerplanar
graph because, as will be used in the proof of Proposition 1, an outerplanar graph on k vertices is maximal outerplanar if and only if it has 2k - 3 edges.

Figure 1 Three Graphs on Six Vertices
As shown in Hilton [3] ,ST(Nf<) = F2k-2 a n d ST(Wk)= L.2k-2- 2- Let OPJk denote the set of maximal outerplanar graphs with k vertices, of which exactly/ are of degree two. Note that/I/^ e QP% for k > 4, and, with Gi
as in Figure \,G1-(3,6)
e OP§ .
*This work was done while the author was a National Academy of Sciences-National Research Council Postdoctoral Research Associate at the National Bureau of Standards, Washington, D.C. 20234.
* * A u t h o r is currently at Sandia Laboratories, Applied Mathematics Division-5121, Albuquerque, N.M. 87115.

11

12

Proposition

FIBONACCI NUMBERS IN THE COUNT OF SPANNING TREES

[FEB.

1. If// e OPJ;, then ST(H) = F2k.2.

Proof.
If k equals 4 or 5, then OPk = {/l/^}, and ST(Nk) = F2k~2 * o r anY k. The proposition will be
proved by induction on k. Suppose it is true f o r 4 < / r < / ? — 7 with n > 6, and suppose H e OPn. Assume the
vertices of H are labelled so that 1, 2, ••-, n is a cycle bounding the outside face and vertex n is one of the two
vertices of degree two, written deg (n) = 2. Now H is maximally outerplanar implies that edge (1, n - 1)\s in H.
Also, either (1, n - 2) or (n - 7, <?J is in H, and, by symmetry, one can assume (1, n - 2) is in //. (See Fig. 2.)
1

^\
\
!

J
n - 1

/? - 2

Figure 2 Structure in a Graph H e OP*
Since any spanning tree T contains at least one edge incident with vertex n, either T is a spanning tree of
H - (In) or H - (n - 7, /?A or else Tcontains both edges (In) and (n - 1, n). Now deg (n - 1) = 2'm H - n
i m p l i e s / / - / ? ^OP^v
Hence, ST(H - (1,n» = ST(H - (n - 1, n)) = ST(H - n) = F2n-4. Also, deg (1) > 3
and deg (n - 2) > 3 in H, but exactly one of these two vertices will have degree two in H - [n, n - 7}, that is,
H - [n, n - 7} e 0Pn_2. Now the number of spanning trees of H using both (l,n) and (n - 7, n) equals the
number of spanning trees of H - n using (1, n - 1). This is obtained by subtracting the number of spanning
trees of H - n that contain (n - 1, n -~ 2) but not (1, n - 1) from the total number of spanning trees of H - n,
and one obtains
F2n-4 ~ ST(H -n-(ln-D)
= F2n„4 - ST(H -{n,n-l})
= F2n.4 - F2n.6 = F2n.5.
Consequently,
ST(H) = ST(H- (In)) + ST(H - (n - 7, n» + F2n.„5 = 2F2n.4 + F2n,5 = F2n.2,
and the proposition is proved.
For OPI w i t h / > 3, no result like Proposition 1 is possible. Indeed, let// / = N7 + 8 + (8,4) + (8,5), and let
H2 = N7 + 8 + (8,3) + (8,4). Then H-, e OP%. H2 e OP%. ST(H7) = 368 and ST(H2) = 369.
Allowing there to be several edges connecting each pair of vertices, let G be any multigraph. Several observations can be helpful.
Observation
1. Suppose v is a cutpoint of (connected) multigraph G, and G - v has components
C i, C2l —, Ct. If/?,- is the subgraph of G induced by C-, and v (1 < / < t), then
t

ST(G) =

l\ST(Bj).
i=l

For example, vertex 1 is a cutpoint of % - (3,4), and ST(N6-

(3,4)) = ST(N3)*ST(N4)

= 3-8 = 24.

Observation
2. Suppose (u,v) is an edge of multigraph G and G' is obtained from G by identifying u
and v and deleting (u,v). (Note that even if G is a graph then G' may have multiple edges. Also, if (u,v) is one of
several edges between u and v, then G' will have loops, but no spanning tree contains a loop.) Then ST(G') is

1 k+j Figure 3 A "Biwheel" on k+j + 2 Vertices with 7 < / < k and k > 2 Observation 3.1 + 1. denoted Wkjr is as in Fig. v) in H = WkJ-{(k. 3 with deg (u) = k+ 7 and deg (v)= j+ 1. 5}. One can also find formulas for classes of graphs.v) or ft k +]) (or (1. The number of such spanning trees that contain a fixed spanning tree of V can be found. and let G' be as in Observation 2. Consider the spanning trees of Wkj that contain (k. LetU (respectively. 1} and { 3.v).6). 1. {3. Nj) containing u (respectively. As this procedure can be iterated. 2.+ ST(N2)). Proposition 2. {3. 1}.k+ 11 (u. 5}.1977] FIBONACCI NUMBERS IN THE COUNT OF SPANNING TREES 13 the number of spanning trees of G that contain edge (u.v)) is . Clearly there areST(U)-ST(V) labelled spanning trees of G containing (u. 4. This equals 2(ST(Nk) + ST(Nk. then ST(G) = ST(G .v).v) that contains */ (respectively. The number of spanning trees of Wkj which contain (u. one containing vertex 1 and the other containing vertex k. 2.2.v). V) be the fan % (respectively.+ ST(N2)) such spanning trees containing a fixed spanning tree of U. k + 1) or (u.v)." where the biwheel on k + j +2 vertices. Any such spanning tree of Wkj contains a spanning tree of U or V. Let G be any multigraph. 2. such as the "biwheels.(u. (I k +j)} . For example. For example. 1. V) by the subgraph of G induced by the set of vertices in the component of T. v). This presents another way to count the labelled spanning trees of G containing (u.v).v) is in spanning tree 7" of G.v)) +ST(G'). but not both. ST(G) is given by evaluating the number of spanning trees in two multigraphs. 4. each of which is a tree.v) is ST(Wk+j+i )• The number of spanning trees containing Ik. Suppose edge (u. one can compute ST(G) in tfeis manner tot any multigraph G. That is.3 + 3. k+ 7^ and (1. there are 2(ST(Nj) + ST(NH) + . {3. 2. each one with fewer edges and one with one fewer vertices. Similarly.2F2k . Thus one obtains 1. in graph G j of Fig.1) + .1 + 1-1 + 8-1 + 21-1 = 75 spanning trees containing (3. Let U (respectively. 4}.8 + 3. ST(WkJ) = L2k+2j + 2F2k+2j . Proof. The possibilities for the vertex set of U are { 3 } . let u = 3 and v = 6.4}. k +j) but not (k. 2. k + 1) but not (u. by enumerating the number of spanning subgraphs of U that have two components. 4. 5 } . 3) which contain the edge (u. { 3 . 4. k+j) but not (u. ST(Wk+j+1) is the number of spanning trees of "biwheel" Wkj (as in Fig.21 + 1-3 + 1. {3.v)that produce these same two subgraphs. {3. if / > 2. using a slight variation of Observation 3.2F2j .2}.(u. {3.

ST(WkJ) = L2k+2j- 2 + 2F2kF2j + 2F2j(F2+ F4 + ••+ F2h2>.Columbia. 12 (1974). Oxford. x2f •••. Clearly if (x j. then one has knowledge of a solution (1) for/7 . Simple Fibonacci identities reduce these equations to the desired formula. x2.. A. REFERENCES 1. Theorem. xn) is a solution of (1) wherein some entry is zero. we see that for a fixed positive integer. = i as well as permutations of this n-tuple since n n £ / . Vol. C. x. Thus. Fans and Baskets. Mo. UTZ University of Missouri . F. for any n. 1977 ST(Nk+1hST(Nj+1). R.2l F4 + •••+ F2k-2) + 2F2k(F2+ and. 355-359. Thus. <n . v n for all / = /. (1) has an infinite number of non-trivial solutions in positive integers. 1972." The Fibonacci Quarterly. Xf + x2 + •'• + xn < na. Harary. Reading. pp. we exclude all solutions with a zero coordinate hereafter. ******* T H E D I O P H A N T I N E E Q U A T I O N (x7 + x2 + •» +xn)2 = xf + x32+>~ + x3n W. Mass." Combinatorics. "The Number of Spanning Trees of Labelled Wheels. Hilton. The Institute of Mathematics and its Applications. n. ••-. is a solution of (1). xn). Equation (1) has only a finite number of solutions in positive integers for a fixed positive integer /7 but as n -> °° the number of solutions is unbounded. Addison-Wesley Publishing Company.n(n + 1)/2 and £ j3 = n2(n + l)2/4. equation (1) has only a finite number of solutions in positive integers and we have proved the following theorem.FIBONACCI NUMBERS IN THE COUNT OF SPANNING TREES 14 FEB. W. if/ > 2. For the same solution x3+x2+- + x3 > a3 2 and so a < n . except for/7 = /. 1969. i=l i=1 Also. let a be the largest coordinate in a solution (xj. •••.7 and so. Thus ST(Wkt1) = L2k+2-2+2F2k + 2(F2 + F4 + -+F2k. 2. The Diophantine equation (1) (x1+x2 + -+xn)2 = x3 + x32 + ~+x3n has the non-trivial solution X. Graph Theory. J.] . To see this. On the other hand if one assumes x\ > Of then for each / one has x. A7. D. 2. 3. Then. Fielder. "Fibonacci Numbers in Tree Counts for Sector and Related Graphs. [Continued on page 16.

SCHMIDT'S THEOREM TRANSCENDENTAL NUMBERS AND GOLDEN NUMBER MAURICE MIGNOTTE Universite Louis Pasteur. linearly independent over 0. (3) From (2) and (3). W. Now. . aj. this theorem can be used to prove that certain numbers are transcendental. France INTRODUCTION Recently. and let e > 0. Schmidt proved the following theorem. Then the number a2 . Of course. we get \qNa2l . Schmidt's Theorem. Let 1. Proof. (2) \a»aA d "Hn 1" ~ ' _ \/5Qn Since un\U2n.± *±J=Ui is transcendental. It is well known and easily proved that "n+l\ 3 / . We shall take a / equal to the golden number. a2 are ^-linearly independent We shall give only one example of such a number #2 but the proof shows clearly that there are many other possible choices of 32THE RESULT Proposition. It remains only to take a number a2 such that \\qa2II is small for these values of q and such that 1. (ItUf^JJL QN 15 . it is easily proved that I \B2- 2 + (-l)N+1 (2 + (-1)N+1)^ QN Thus. There are only finitely many integers q such that (1) Wqajl II<732i < c 7 q ~ 1 ~ € .AN APPLICATION OF W. -••) be the sequence of Fibonacci numbers. The integers q will be chosen in the sequence of Fibonacci numbers.112. where c-j is a positive constant and where II II denotes the distance from the nearest integer..- y/5U* Thus. N y „% 2 + (-1)n _Ph Qn Qfi ' wherep n is an integer. a1f a2 be algebraic real numbers. Let (u 1.Qn divides qn+v H ance. Strasbourg. Put qn = Un.

Dickson. (2. The case n = 2 is a special case of a well known theorem [1. We have verified that (1) holds with e= 1. In the complementary case (that is. 2) together. k. always an infinite number of solutions. Thus.37. Forfl =4. 0) which come from the case n = 1.. p. For n = 3 these solutions are exactly those revealed in the general case. the previous relation gives MW/I = ±k2\\qNa2\ • This contradicts (2) and (3).32 are 0-linearly independent Now Schmidt's theorem shows that a2 is not algebraic. 1.3) and permutations of (1.3. an identified problem [1. {a.. Vol. That is. etc. for each n > 1. p. We can now limit ourselves to the case of kt e Z For large /I/. -a) satisfies (1). (I 2). 1). Carnegie Institution of Washington. satisfies (1) in case n =3. 0). (3.C.3). 1). (0. 566].2. ******* . The assertion is proved. It remains only to show that 1.is a quadratic Pisot number and y its conjugate. The proposition remains true if we put xn -yn x . E. with (0. D. For/? =3 the solution will be a subset of the solutions of x] + x32+x33 = u2. some coordinate is negative). In case n = 2 the reader will have no difficulty in showing that all solutions are (a. 412 etseq.e 0. 1920.l.]. L. ******* [Continued from page 14. REFERENCE 1. For example. there are. SCHMIDT'S THEOREM TRANSCENDENTAL NUMBERS AND GOLDEN NUMBER FEB. Suppose that we can find a non-trivial relation ko + k1a1 + k2a2 = 0. (a.-a)f for any integers.16 AN APPLICATION OF W. -a).1977 \\qNai\\ |<7/v*2 II < "~" .y where A.] For small integers n the positive solutions of (Drnay be found with a machine because of the upper bound of n2 on the coordinates. a2 are linearly independent over 0.-a. VN where c is a positive constant. (2. History of the Theorey of Numbers. II. of course. REMARK. a. 3y. (1.

12. Intuitive considerations of stability require the minimization of the number of mutual perturbations between adjacent satellite orbits. Some think there is a reason for this [2. 2 where J.THE RECIPROCAL PERIOD LAW V\LE. This will obtain if adjacent periods are coprime. Alone this restricts us to even order difference equations. 8 . Von Weizsacker proposed a system of eddies of ellipticity = Vi lubricated by smaller eddies.c are constants and 5 = A . vi. 1. 11]. must turn around near the outermost body and be asymptotic to the inner bodies upon both leaving and returning to the primary. ••• is given b y / = %. have at most two boundary conditions (B. then the zeroth law should be the principle of Occkham's razor-the paring away of unnecessary assumptions. The idea that the asteroids were once a planet has been disproven [ 4 ] . The data on near-commensurabilities are not significant [15] if the peculiar ratios of 2 and 4 are omitted. thus a = ±3. namely a difference equation. are discrete and therefore discrete algebra should be used. Dermott's rules [12]) by almost all scientists because they have too many independent parameters and lack any logical basis.GREIG West Virginia University. Almost all theories proposing specific distance rules are discounted (e. as is proven later. Fig. while others such as Lecar [3] think that the distances are random subject to the restraint of not being too near to each other. Assuming that the Sun-Jupiter distance is fixed then a better way to state point (x) is that it is the minimization of the 17 .. cannot have B. Thus (20) 8*Zm=jZm. More than two parameters is too many.^ organtown. 7 . the relevant variable is the frequency of nearest approach. Now we can determine the value o f / x.73 rule [ 9 ] . The computer work of Hills [ 1 . by the Commonality Principle must all satisfy the same spacing law. Blagg's 1. 2] proves that an average period ratio exists and lies in the range 9/4 <P < 3. Y. West Virginia 26506 The opinion of scientists on Bode's rule falls into several camps. and -t6 gives alternate Pell numbers. If the first law of scientific reasoning is that equations should balance dimensionally. It forces the roots of (18) to be reciprocal pairs and by the theory of equations c = 0. ix. 16. 13]. The forces are attractive so the largest root of (20) should be as small as allowed.. may equally likely have pro or retrograde outer satellites since retrograde bodies are not irregular. where (19) with Ym+h = Zm+1-Zm h= 1 /2 viii. 9 . I begin this essay by a study of first principles. are stable due to weak (gravitational. This. determines/to be an integer. vii. 10.V = A\7 is the central difference operator. Zm.C. consist of one primary. the synodic frequency.g. To elucidate. v. the author's [13. and ±3 gives finite exponential (En) or alternate FL numbers. values of a = j + 2 equal to ±2 give arithmetic progressions. ii. iii. 24. tidal or gyroscopic) non-dissipative forces and hence.) and therefore 82Zm = jZm+cZm_1 (18) . where N is the number of eddies in a ring. in empty space and hence they must reside in the primary which means that the reciprocal period sequence.C. a pair of secondaries which we ignore and the rest tertiary masses. The only models not discounted in this way are von Weizsacker's [see 5 ] . iv. PRINCIPLES I insist that satellite and planetary systems: i. The mathematical theory in [16] is now given a logical derivation. The interested readership may consult any of the several summaries of physical theories of the origin of the solar system [6. 14] and perhaps Schmidt's [see 7 ] . One can derive the equation: distance factor = tan2 [TT (N + 1)/4N]. The sequence ••• 11.

\ This immediately gives by taking ratios (29) (Sf+2 + Si)/(Sj + Sj-2) = S-f.S2-i) • Now if satellites are alternately pro. not rigor. CONCEPTS A FL sequence. For convenience. where the " . Writing v = y/5 for brevity we have (23a) 5v + 7 (23b) 2v+3 -3v-4 (24) (25) / = -7/7 v+2 -v-1 v+3 0+1 2v + 7 v+4 Nept X Uran -3h h 5h 5v+18 3v+ll 13v+47 8v+29 Satur 9h 34v+123 21v + 76 (Jup) Astrea 13h 17h 89v +322 55v+199 Mars where either sequence may be regarded as the first-order differences (synodic frequencies) of the other.C. A system of satellites (planets) much lighter than their primary tries to act as if their synodic frequencies correspond to real bodies with their synodic frequencies in turn being the reciprocal periods of the original bodies. Sequence (23) gives an earth value of 521. Now a = 0 gives two constant sequences and so is trivial and a . (1) GQ= GN led via the theorem in [16] to the beautiful closure relation (14) £1"S. When M is infinite. Now in point (xi) we could not have written ZQ = AZQ since that leads to monotonically increasing sequences. define a sequence./ = (S-i-2 . sequence (24) has been placed parallel to (23). satisfies (27) (fr-l)En = 0.-. Therefore (21) Z0 = AZM = B.C. En. cannot be expressed as a function of 5 2 and / alone since (26) (A+ \f . or VZ0 = ZM = B. E'n. Hence a = ±3. I assert that only one physical B. If Zn represents the real reciprocal periods of satellites or planets this can be written as a minimum principle^ (28) £ (8*-l)Zm - 0.C. Hn.S-i)/(S-i .e. But a finite exponential (bisected FL) sequence. Thus (28) gives a closed system having a finite number of sidereal (true) and synodic frequencies.C.and retro-grade then we may interpret the first pair of (29) to mean that the ratio of adjacent synodic frequencies (since Sj is now negative) equals the ratio of the sidereal frequencies of two other members of the bisection of S aside from a (-1). . = (-1)f~hS. We may state this in words. which differ only in notation. 2. Arithmetic progressions. This is true even if all the bodies do not revolve in the same sense. This is equivalent to GQ = GJ\J in [16]. Further. I first used bisected FL (Fibonacci Lucas) sequences in a letter [17]. j = +1 o r . 1 and Table 1 of [16]). Then Zn satisfies (27). Hence (29) is an excellent approximation to the finite cases.and retro-grade we can use (31). (21) and (22a) give sequenced of [16]. xi. Real satellite systems have a finite number of bodies but the difference in the ratios given by [S] and {G 3 3 } for example is less than \ti~6. And from point (x) we have (22a. The index h = %. a = ±2.5 " case corresponds to outer satellites that are alternately prograde and retrograde.THE RECIPROCAL PERIOD LAW 18 [FEB. i.±1 gives cyclical sequences of periodicities 6 and 3 and so is also trivial.b) 82Zm = or 8*Zm = Zm -5Zm. namely (21) which is the same as Eq. Now let Zn be a bisection of Gn (Eq. are also trivial.l)Hn = 0 . Now this point. exists which must equal both matematical B. Eqs.5 . potential energy of the tertiaries with respect to their secondary. If they are alternately pro. such that (27a) (82 +5l)E'n = 0. Thus satellites try to get as close to their secondaries as other conditions (ix) will allow./ JS /..

. E'n. Hence a Saturnian satellite(s) of (mean) period 207. PREDICTIONS Dermott [12] ignored the outer Jovian and Saturnian satellites. I prefer (c). At the moment.. the Galilean quadruplet. Hence in place of (28) we may write (31) (o2-Dz'm - £ 0. . Similarly a stable orbit in the Jovian system is likely at 97 day with much less likelihood of another at 37 day because of its proximity to the Galilean quadruplet (secondaries). 5/11 of Jupiter's period as well as at 1/3. or (e) otherwise. In fact the gap at 3/8 is only 2 percent from the predicted asteroidal planet See [18. These then captured enough material to form the other planets and comets by coalesence. p.. 4/9. namely its mass. Now a system of alternately pro. The theory says little about the secondaries. spin.. 4) = 3.. It is then easy to show that (32) o2 = 41 + b2. If it is ever found.1 percent of the numbers but I regard 1 percent as acceptible.. We infer that the proto-solar system soon formed two clouds of cold dust and gas. (b) be a mean of the spins of the primary and secondary. The Simplicity Principle. (point XI) may: (a) lie in the mean angular velocity of the prinary.1977] THE RECIPROCAL PERIOD LAW 19 It is easy to show that the ratio of adjacent terms in (23b).84 < P < 208. or to common sense.. The reciprocal period law is the only theory to make very narrow predictions. I have chosen to give them an important place in this paper.and retro-grade satellites satisfies an £ primed sequence. reached a formula which according.. Z'is a bisection of S or G but with alternate terms multiplied by (-1). The physical B. where o is the central sum operator defined by of n = fn+h + fn-h * where fn is any sequence whatsoever. For the Sun. Snh+l/Snh-1 = (Ln + v)/Ln-2 = Ln+2/(Ln-v). (c) lie in the ter'tiaries as a whole in which case they constitute a selfenclosed system.. where mod (n. Jupiter. 2/5. it would constitute proof positive of the theory herein. Gaps exist at 3/8. The theory herein has been predicated upon: The Commonality Principle.C. The larger became the Sun and the smaller became Jupiter and Saturn. . the reciprocal period rule would begin to operate. Saturn their secondaries are Jupiter. Similarly the ratio of adjacent terms in (23a) is S nh + l/Snh-1 = (Ln-vJ/Ln-2 = Ln+2/(Ln + v). and the assumption that the physical reason for the stability of tertiary orbits is the avoidance of low-order commensurabilities (ALOC). C. 97-98] ". I propose to call it Aurelia. where mod (n. 4) = 1 and where (25) is the index. There is a blank midway between Saturn's Phoebe and lapetus in Table 2 of [ 16].03 day is predicted. Dn . 3/7 and 1/2. A Z' sequence satisfies (27a). respectively. During the late phases when dissipative forces were no longer important. For completeness we may define the double bisection of an FL sequence. But the synodic frequencies are no longer differences (since every other term is negative) but sums. pp. (d) be the period of a hypothetical satellite that skims the primary's surface." 3. J. Maxwell approached the motion of molecules in air in a similar vein of which James Jeans wrote [19. Application of the summing of adjacent terms twice is equivalent to the operator(82 +41). by (30) (82 ~5l)Dn = 0. 97]. The allowed range is 0. Titan+Hyperion.by a train of argument which seems to bear no relation at all to molecules or to their dynamics. Hence the distance between the primaries and secondaries and their mass ratio must be determined by the properties of the proto-solar system cloud. The Kirkwood gaps have prevented the coalesence of asteroids into a planet.was shown to be exactly right. to all the rules of scientific philosophy ought to have been hopelessly wrong. moment of inertia and magnetic field.

Astrophysics and Space Science. Lecar. 3. 56. But to align two prolate bodies one of whose axes is. 8 (1970). Consider Pn+1 = bPn + cPn. Astron. . F. p. Qtrly. These frequences 152/79. 18 (1946).8995476269 •••= 1 + r 1+ L + ^ 8+ + ^ 1+ + ^ 21+ + ^ + r4+ + 1+ T. 8. Of all two-point recurrences only those with the middle coefficient b an integer and c = ±1 have coprime adjacent terms given that an initial pair. 4. 94. Cousins. Vol.X (1967). Greig. Vol.W. Vol. 225(1970). PQ and Pf say. 267. Astr. Alven and G. Nieto. 30]. E. 449. p. p. are coprime. Tibor Herczeg.21 (1972). of a superior (exterior) body of period z > 1 is given by (33) 1/z+1/y = 1. 4. Vol.0 away from the line joining them requires a sin 6/d* force which is very weak. 1972. S. J. Mon. 7 (1975). F. ThenbPn is divisible by d and so also is Pn_1 = Pn+1 . Theorem. M. yet over long time periods must be sufficient. Physics. Let c = 1. 142 (1969).. 1.+ ' 1+ 1+ 1+ whose convergent is 1 + 25253/28073. 184. The first useful convergent is 416/219. I use ratios for clarity. As an example viz. McD. Vistas in Astronomy. 10. COPRIME SEQUENCES If the recurrence Pn+i = (integer)^ ±Pn-i holds we have a Coprime sequence because it satisfies the following theorem which is a generalization of one in [20.bPn . 5 (1967). But then Pn-2 = Pn-bPn^ is divisible by d and likewise all earlier terms by induction. Soc. Proof. 243/395 to be added in pairs.+ 7+ ^ + ^ + r . 11. 6. and 499. Arrhenus. S. New York City. Vol. Vol. p. 1. The proof obtains by postulating the contrariwise proposition. REFERENCES 1. P.20 THE RECIPROCAL PERIOD LAW [FEB. 18. Vol. p. D. LetPn+i and Pn be divisible by some integer d. p. Rev. 12. Amer. 224. Vol. The Solar System. W. Thinking of ourselves as Venusians.. 40. Pica Press. p. . p. p.701 day. 1972. Dodd. 11. Chandrasekhar. The latter two give 395/608 for the time between successive Earth transits. We have three frequencies: 1. 7. 242(1973). Oxford. Vol. Napier and R.-. J. Roy. Vol. Vol. VENUS The synodic period. p. 9. 9 (1970). Astr. Vol. I. W. then Venus is fixed and the Sun and Earth appear to revolve around us. 41-71. Cremin. Nature. 51] are interesting. Not. The first and third using (33) gives 395/152 for the earth's synodic period (584 da).HWte. Hence the theorem must be true. Soc. 0. p. 73/79. p. 3. The following relations [18. of Mod. p. 338. 4. Uermott. 11-19. The Titius-Bode Law of Planetary Distances. Hence both PQ and Pj are divisible by d which contradicts the assumption which says that at most one of PQ and Pi are divisible by any number. p. 9 (1968). to be the unit of time. H. 2. 242 (1973). and Astrophys. Vol. Ann. 5. p. The first pair gives 79/152 for one Venusian solar day. Astron. Choosing c = -1 changes no essential part of the argument. 840. In passing we give the continued fraction expansion of the distance factor derived from Kepler's III law. Nature. Bull. Venus must be aspherical so that torque forces can cause this. 244 (1973). F. 141 (1968). Ter Haar. M. 1. 13. Soc. J. p. 143. G. Tidal forces tend to pull a body apart and are inverse cube. Nature. Hence during every 584 day the same spot on Venus faces the Sun 5 times and the Earth 4 times. 363. Demon. Pergammon.. Nature. 608/395. 780. Williams and A. M. Revs. 4. 5. 15. y. S. Then to better than 5 significant figures the earth's period is 395/243 (13/8 is less accurate) and the rotation period of Venus is -79/73 (clockwise). Roy. Vol. 318. 250. W. Choose the Venu&ian sidereal (true) year. 152/395 are in the exact ratios 5.

where 2 . Not Roy. so that (3) implies GI = CL-J-= VP [Continued on page»14. 296. Z. Fibonacci Numbers. The Solar System. 8. of Chisla fibonachchi. J. England.WHITFORD Torrens College of Advanced Education. 21. 1951).. E. Mon. letter to the author dated 17 Nov. Blaisdell Publ. The Fiboancci Quarterly. W. N. E. N. James Clerk Maxwell. by starting with Binet's formula \ ~~ f~n n ii /F \ n ~" <" > 75 1> instead of (1). Thomson.$ = \fp and a +0 = /. Vol. Ovenden. A. p. Gn+2 = Gn+1 + {P-f1) G1 = G2 = I Gn (n > Clearly a . K. (n Thus [Gn] = {Fn} in the case/7 = 5. New York. Torrensville. 232. 15. Kopal. 20. we seek such a generalization indirectly. p. P 2 are roots of the equation (4) x2_x_^jy0 Corresponding to (1). Gould dated 30 Aug.. W. 2. Roy and M. London. South Australia Any generalization of the Fibonacci sequence {Fn} = 1. p. Vol. 114 (1954). 1961. Cambridge University Press. 115 (1955). we now have the equations (5) Proof. Vorob'ev. however. 17.] i G2 = (SLzm+1) = 7. 1972.1977] 14. VP V. We can also write (3) Gn = ^ X - (n > V. E. 129. for any positive integer^. 1975. 1973. .. 5031. p. J. W. Greig. THE RECIPROCAL PERIOD LAW 21 L. Cambridge. Soc. Oxford University Press. Suppose we define. BINET'S FORMULA GENERALIZED A. letter to H. Greig. 1. 16. 3. Astr. 30 (trans. 13. 5. Motz (Columbia Univ. 14 (1976). W. A Commemoration Volume. the sequence Gn by (2) GJnn = A__£___/ 1=_1___^___^_ (n > 1). 18. ••• necessarily involves a change in one or both of the defining equations (D F1 = F2= 1.). Vol. 1931. Fn+2 = Fn+1 + Fn (n > 11 Here. 19.

and various Multinomial Expansions (Hilliker [ 2 ] . = 1 implies that A7 . n2. dL As before (Hilliker [6]) we begin with a triple summation expansion: n-k y\a-. [ 4 ] .2 ••:"/ -" nr=n where n /. nr are nonnegative integers and where the multinomial coefficients are given by ( n )= .. We single out the terms for which n . for example. In this note we shall continue our sequence of expositions of the Binomial Theorem. =j=1E E m » /WE E <* k=1 (2) \ 2=7 Here. aversion of the Multinomial Expansion may be derived by an iterative argument which makes no reference to the Multinomial Theorem for a nonnegative integral exponent. [6]. so that the range in the summation with n . ) .X\ i=1 I 2 ni+n2 + + ( . for nonzero terms. —. let us make a change of notation and write Formula (2) as (3) ( £ a. the Multinomial Theorem.. c. Ohio 44115 The Multinomial Expansion for the case of a nonnegative integral exponent n can be derived by an argument which involves the combinatorial significance of the multinomial coefficients. [ 5 ] .V.22 = 0 and write (3) as A7-£0 «> (L «) E*/ .£I>0 m * \ £ */ C=/ . After/77 iterations of Formula (3). [ 3 ] . [7]) by making the observation that this iterative argument can be modified to cover the nonnegative integral case: (1) I T. ". See. .x and 22 with 1 < ex < r and 1 < c2 < n.1. Co +n-Q£=0 * e .c2 > 0 is 2 < e t </-and 1 < c 2 <n . m > 0 and not too large. 2 Note that. We now apply Formula (3) to the summation under 2 on the right side of (4). Here the derivation may be carried out by employment of the Binomial Theorem for an arbitrary exponent coupled with the Multinomial Theorem for a nonnegative integral exponent. Chrystal [1] for these details. We next assert that the Multinomial Theorem (1) is covered by the Formula (2).". In the case of an arbitrary exponent n these combinatorial techniques break down. \ i=l where the double summation on the right is taken under 9.fi2 = 0. We have observed (Hilliker [6]) that in the case where n isnotequal to a nonnegative integer.ON THE MULTINOMIAL THEOREM DAVID LEE H1LLIKER The Cleveland State University. we are using the convention that the empty sum is zero and that 0° = 1. Cleveland. . we obtain 22 . This iterative process may be continued. To see this.

.-Q2m>0 > n-SL? ^2m 0 '^2m+1 t*2m+1-1 ^2m+2 ( \^ I *2 £4 S V + Y / C k=1 n-%2- x „_ \ n-SL2 1=1 n / n \f *2m+2- r*2\ . David Lee Hilliker.. the Multinomial Theorem (1) now follows. 1977 23 n (5) [ 2 ^ \ «)"-„.ON THE MULTINOMIAL THEOREM FEB.The coefficients may be written as I n \(n-Q2 \ C2 M \. 1 (1972).(n -SL2 ) n! = %2lSL4i- SL2!Z4l—SL2r! •9-2r! It now follows from (8) that r .1) . then. so that the first two inequalities in (6) are possible and as long as (7) m < n.. so that the last inequality in (6) is possible. 1964.~ *2k-2 \ £* M ^2k %2k=0 If the range of £2/.fi2 \ i n ./ . the subscripts are uniquely determined: 2/ = r. the restriction (7) may be lifted. by (6) we have z2r-i %2r = = 1 / f ° r otherwise. Consequently.1. REFERENCES G. New York. n-5. " A Study in the History of Analysis up to the Time of Leibniz and Newton in Regard to Newton's Discovery of the Binomial Theorem.Y.3= r.Zo. the indices are subject to the restrictions f / < e." The Mathematics Student Vol. 28-34. Formula (5) now takes the form E <*> (x>)"-~ \ 1=1 n-SL2-'-^2r=0 I *E ("2)(^. the summation under k reduces to a single term. k = r. < r (6) < 1 < e 2 />/ < c 2 / .-. f>-*2 *2k-2\ %2-U *2k = 2A 0 ' Here.. 1. We \etm = rwould have £7 > r. N. we 0.2 %2r=0 ^2U ar ar-t ^2f -a7 With a change of notation.2)-("-fi2«2r^-2)^-42. No./ / I / < c 2/ >2 < fl-22 c for 7 < i < m. . U ) ("u2) E k=1 n-SL2 (""* 2 -* 2k -* 2k ~ 2 )#&-41% 9.2 I Then. is extended to include 0. 0.( U I n-Z2 *2r-2 \ *2r \ %2r + V n(n .. Chelsea. Formula (5) is meaningful as long as m <r. XL. for 1 < / < r. 1961. Also reprinted by Dover.. 2. i-1 L (4)("-S)-(—i^-*-) n-Z2-. by (6). Vols.n n! 2>) i=1 / n •9.. 2 / ' torO < / < m. First part. This is a reprint of works of 1886 and 1889.2k=0 r E ^p k=1 n-SL2 ( n \ I n . Textbook of Algebra. and. for nonzero terms. Contributions of Pascal. Chrystal. '"' z2r-i = 1.. I and II.

lp-1\nn Gn+2 = on-f-1 ( p__^J_ \ on ^jf— = ^ Thanks to (5) it is now a simple matter (despite the complicated appearance of (2)) to generate terms of the sequence {Gn}.] ^ 1 = 1 and Gn = F„ . 3. Contributions of Cavalieri. Vol. "On the Infinite Multinomial Expansion. 17. Vol. David Lee Hilliker. 15. Contributions of Newton." The Mathematics Student Vol. David Lee Hilliker. Fourth part. 1 (1974). pp. Contributions of Archimedes. Second part. " A Study in the History of Analysis up to the Time of Liebniz and Newton in Regard to Newton's Discovery of the Binomial Theorem. Third part. 203-205. XLII. 13. Generalizations of some of the better-known properties of [Fn} are listed below. 4. pp. namely p = 1." 77?e Mathematics Student Vol. 397-404. Of course. 5. using (3).ON THE MULTINOMIAL THEOREM 24 FEB. XLII.9. —. pp." The Fibonacci Quarterly. XLII. Therefore. lim Gn-Gn+2-Gt+1 = (-1)n+1 [£fl)n G Jl±L=i±JlL In > 1) . especially if our knowledge of {Fn} is taken into account. David Lee Hilliker. No. 7. 195-200. I I . in each case. 6. No.4 (1974). (i) (ii) [Continued on page 29. No. Thus the first five such sequences start as follows: p 4 Gi 1 5 9 13 17 0 1 2 3 4 1 1 1 1 1 G2 G3 G4 1 2 3 4 5 1 3 5 7 9 G5 G6 Gy G8 G9 1 5 11 19 29 1 8 21 40 65 1 13 43 97 181 1 21 85 217 441 1 34 171 508 1165 G10 1 55 341 1159 2929 We can use the above table to guess at various properties of the generalized Fibonacci sequence {Gn}. No. for any choice of/?. Assuming that we are interested only in integer-valued sequences.5. (5) tells us to take p of the form 4k + 1. pp. 15. "On the Infinite Multinomial Expansion. 392-394. 2 (1974). 107-110. David Lee Hilliker. 1977 3.] Also. David Lee Hilliker. 5. since a and |3 satisfy (4). the original result may be found by taking p = 5. it follows that nn+1. " A Study in the History of Analysis up to the Time of Liebniz and Newton in regard to Newton's Discovery of the Binomial Theorem." The Mathematics Student Vol. " The Fibonacci Quarterly. pp. " A Study in the History of Analysis up to the Time of Liebniz and Newton in Regard to Newton's Discovery of the Binomial Theorem. No. we have the equations 0P+2 = an + 1 + ^ J j anf pn+2 = ^n+1 + (*L^7) 0" (n > 1). ******* [Continued from page 21.

IT] (5) F f = r J2 (n~kk-1)Lnr-1-2k.7 6 ] .W. k=1) ' 2\r. M. West Virginia 26506 Almost everyone who works with Fibonacci numbers knows that diagonal sums in the Pascal triangle give rise to the formula r [?±] (D F„. k=0 but on the other hand for odd positive integers r we get the same terms but with all positive signs . £ ("-*"'). a>1. Formulas (2) and (3) generalize in a curious manner. Morgan town. with LQ = 2. F2n = £ k=0 or that \ri\ (3) F3n = 2 £ {"'I' 1 )4"-1-2k. As far as I can determine.a . On the one hand we have for even positive integers r [T] F (4) f1 = r H (~Vk {n~kk'1)L"r'1'2k. Formulas (4) and (5) may be written as a single formula in a clever way as noted by Hoggatt and Lind [4] who would write rn-U (6) T =11 r (-Vk(r'1> {"-kk-1)L"r-1-2k k=0 25 . 3 3 .GOULD West Virginia University. I was reminded of the formula recently when Leon Bernstein [1] found the formula again and asked me about it. He used a new technique involving algebraic number fields. . N. formula (2) first appeared in our Fibonacci Quarterly as a problem posed by Lurline Squire [10] when she was studying number theory at West Virginia University. where Lr is the usual Lucas number defined by Ln+1 = Ln + Ln-f. k=0 and that these are special cases of a very general formula given in 1878 by Edouard Lucas [5. Eqs. 2\r.A FIBONACCI F O R M U L A OF LUCAS A N D ITS SUBSEQUENT M A N I F E S T A T I O N S A N D R E D I S C O V E R I E S H. Lj = 7.3 4 ] . S. [6. pp. this of course in contrast with Fn+i = Fn + Fn--j and FQ = Of F-J = 1. Swamy's solution invoked the use of Chebyshev polynomials. k=0 but not many realize that (2) f-/i*("-*-')r'. 7 4 .

[FEB.

A FIBONACCI FORMULA OF LUCAS AND

26

valid now for any positive integers nj > 1.
Formula (6) of Hoggatt and Lind was posed as a problem by James E. Desmond [11] and solved by him using
a result of Joseph A. Raab [ 7 ] . The precise same problem was posed again by David Englund [12] and Douglas
Lind pointed out that it was just the same formula.
Formulas (4) and (5) were obtained by Hoggatt and Lind [4] by calculations using compositions and generating functions. Although they cite Lucas [5] for a number of items they were evidently unaware that the formulas appear in Lucas in a far more general form. Since Lr= F2r/Fr, formulas (4)-(5) can be written entirely
in terms of f ' s .
Lucas introduced the general functions U, V defined by
Un = a—=-f>
a—b

(7)

an+bn,

Vn =

where a and b are the roots of the quadratic equation
x2-Px

(8)

+ Q= 0,

2

so that a + b = P and ab = Q. When we have x - x - 7 - ft we gets and/? as (1 + ^Js)/2 and (1 -yJs)/2
then Un = Fn, Vn = Ln.
One of the general formulas Lucas gave is [6, pp. 33-34, note misprint in formula]

(9)

l"f]

^ =

£

r

(~Vk

(n~kk-1)

Vf-1-2*!]*

and

,

k=0

which unifies (4) and (5) and is more general than (6). Curiously, as we have intimated, Hoggatt and Lind do
not cite this general formula.
Now of course, there are many other such formulas in Lucas' work. Two special cases should be paraded here
for comparison. These are

(10)

Lrn

(-1)k

= £

--V--

(n-k)

Ln~2k

former,

and

f]

L„. £ - 2 - ("-*) cr-2k « ' -

mi

k=0

These can be united in the same manner as (4)-(5) in (6). Thus
(11.1)

Lrn

[I]

= £

(-7)k(r~1)

-JLj

{"-k)

Lnr~2k .

k=0

There is nothing really mysterious about why such formulas exist There are perfectly good formulas for the
sums of powers of roots of algebraic equations tracing back to Lagrange and earlier. The two types of formulas
we are discussing arise because of

[§]
(12)

£
k=0

formula (1.60) in [ 3 ] , and

(- 1)k (" ~ *) (xy)k(x +Yr2k

=

X

-^X-—
X

V

,

1977]

ITS SUBSEQUENT MANIFESTATIONS AND REDISCOVERIES

27

fL -2 1J
(13)

{n-k)(xy)k(x+yr2k

^ ( - j ) * "
n
k=0*

= xn+yn

,

formula (1.64) in [ 3 ] , familiar formulas that say the same thing Lucas was saying. The reason it is not mysterious that (2) holds true, e.g., is that Z^/? satisfies the second-order recurrence relation
F

=

2n+2

3F2n ~ F2n-2

with which we associate the characteristic quadratic equation
x2

= 3x - /

so that a formula like (2) must be true. For formula (4) with r = 4 we note that F4n+4 = 7F4n - F4n_4.
In general in fact,
(14)

Frn+r

for even r, or Fm+r + Fm-r=

= LrFrn-Fm.r

LrFm,

and
(15)

Fm+r

= LrFrn + Frn-r

ioroddr,

or Fm+r-

Fm.r=

LrFm .

Regularly spaced terms in a recurrent sequence of order two themselves satisfy such a recurrence. Set un = Frn
to see this for then we have
(16)

un+i

with

= Lrun±un-i,

z2 = Lrz± 7,

so we expect a priori that un must satisfy a formula rather like (1). Formulas like (12)— (13) give the sums of
powers of the roots of the characteristic equation, whence the general formulas.
Formula (12) corresponds to (B.1) and (13) corresponds to (A.1) in Draim's paper [2] which the reader may
also consult.
Another interesting fact is that these formulas are related to the Fibonacci polynomials introduced in a problem [9] and discussed at length by Hoggatt and others in later issues of the Quarterly. These are defined by
fn(x)

= xfn-f(x)

+ fn-2(x),

n > 2,

with fi(x)= 1 and f2(x) = x.
In general

m
(17)

fn(x)=

£

{"-kk-1)x"-2k-1

.

k=0
whence for odd r we have by (5) that

f„{Lr) = F-P .

(18)

' r

Many other such relations can be deduced.
Finally we want to note two sets of inverse pairs given by Riordan [8] which he classifies as Chebyshev inverse pairs:

i]
(19)

«W = £

( 1>k

-

^k

(nkk)9(n~2k)

if and only if

[I]
(20)

g(n) = £

k=0

("k)f(n-2k);

28

A FIBONACCI FORMULA OF LUCAS AND

[FEB.

and

tfi

(211

-2kl

IM- X. TlfH » ' "
k=0

if and only if

[i]

(22)

g(n) = £

(-if

(n~K)

f(n-2k).

k=0

Applying (19)—(20) to (10) we get the particularly nice formula

ui]

Lnr = £

(23)

("k) Lr(n-2k),

reven.

k=0

Using (21)—(22) on (4) we get the slightly more complicated formula

,„__£

(24)

r^itn
n-k+1

^ f r o ^ a
\kl

k^O

Fr
r

I do not recall seeing (23) or (24) in any accessible location in our Quarterly.
If we let /--> 0 in (4) we can obtain the formula (1.72) in [3] of Lucas, which is also part of Desmond's problem [11] who does not cite Lucas,

Ir'l
(25)

n=

£

(~1)k

(n-kk-1)2"-2k-1,

n>1.

k=0

It is abundantly clear that the techniques we have discussed apply to many of the generalized sequences that
have been introduced, e.g., Horadam's generalized Fibonacci sequence, but we shall not take the space to develop
the obvious formulas. It is hoped that we have shed a little more light on a set of rather interesting formulas all
due to Lucas.

REFERENCES
1.

2.
3.
4.
5.
6.
7.
8.

L. Bernstein, "Units in Algebraic Number Fields and Combinatorial Identities," Invited paper, Special Session on Combinatorial Identities, Amer. Math. Soc. Meeting, Aug. 1976, Toronto, Notices of Amer. Math.
Soc, 23 (1976). p. A-408, Abstract No. 737-05-6.
N. A. Draim, "Sums of nf Powers of Roots of a Given Quadratic Equation," The Fibonacci Quarterly, Vol.
4, No. 2 (April, 1966), pp. 170-178.
H. W. Gould, "Combinatorial Identities," Revised Edition, Published by the author, Morgantown, W. Va.,
1972.
V. E. Hoggatt, Jr., and D. A. Lind, '"Compositions and Fibonacci Numbers," The Fibonacci Quarterly, Vol.
7, No. 3 (Oct., 1969), pp. 253-266.
E. Lucas,"The'oriedes FonctionsNume'riquesSimplementPe'riodiques/'/l/77er. J. Math., 1 (1878), pp. 184—
240;289-321.
E. Lucas, "The Theory of Simply Periodic Numerical Functions," The Fibonacci Association, 1969. Translated by Sidney Kravitz and Edited by Douglas Lind.
J. A. Raab, "A Generalization of the Connection between the Fibonacci Sequence and Pascal's Triangle,"
The Fibonacci Quarterly, Vol. 1, No. 3 (Oct. 1963), pp. 21-31.
J. Riordan, Combinatorial Identities, John Wiley and Sons, New York, 1968.

pp. 11. Solved by the Proposer. Swamy. 1968). 1 (Feb. No. No. Carnegie Institution (Washington 1919). Vol. Dickson. N. Vol. we can apply the Binomial Theorem to (2) and obtain the general vertical sequence in the form ~~n~l 2 2 {nr)(4k-3)(r-1J/2} (k > 1). The Fibonacci Quarterly. 1 (Feb. 12. 6. it follows from (v) that the sequence under Gn (n > 1) is given by (6) is'c-. 4 (Dec. 192. Vol. No. A. No. Squire. Problem H-135. S. No. ibid. °r= r=1 (vi) G n+2-1 (n > V. Problem H-83. instead of using (6). Vol. 2 (April. Problem B-285. J so that for example the sequences under G4 and G5 are {2k . turning to the vertical sequences in the table given earlier. are fairly straightforward and left to the reader.7 ) G2 + G2H = G2n+1 (n > 1) (iv) G2n+2- (n > 1) (P-flf'G2 = G2n+2 r=0 {P-J-L)T. 3. 4.. 1966). W. Alternatively. Vol. History of the Theory of Numbers. pp. No. Problem H-172. 9. Posed by Mrs. which rely essentially on equations (2).. L.. (ii) was proved in a slightly more general form by E. J r=1 rodd REFERENCE 1. p. ibid. Swamy. respectively. The proofs of the above results. 1 (Feb. No...r )'*-"'}J \ X [n'l-r)(k-Ur\ 1t r=0 r=0 <k> 1). Posed by Barry Wolk. E. No. 383. B. 1965). 3 (Oct. ibid. 236.. Posed by J. 7. p. Solved by D. The Fibonacci Quarterly. For example. Vol. Of course. 8. 1970). pp. (3) and (5).7} and {k + k. No. 94-96. The Fibonacci Quarterly. 12. . Lucas as early as 1876 (see [1] page 396). p. 221.. Desmond. Vol./M/„ Vol. The Fibonacci Quarterly. 518-519. 1971)./ } . Vol.= 1 and a(S =-(&-—) . 4. Finally. results such as these are not new. p. ibid. Solved by M. 5 (Dec. E. 1968). N. 1966). Vol. 10. p. Peck. Vol. pp. S. 13. Solved by C. 54-55. 1. Zeitlin. a+p. 143144. 57. Problem B-74. 1969).1977] AND ITS SUBSEQUENT MANIFESTATIONS AND REDISCOVERIES 29 9. ******* [Continued from page 24. 13. Solved by Douglas Lind. 2 (April 1975). Posed by David England. together with a-j3 = Vp. 2 (April 1974). No..] (iii) ( ^ . The Fibonacci Quarterly.. 519. 6. Posed by M. p. 5 (Dec.

the general form is Gn(x) = Nn(x)/(1-x)n+1 . By the law of formation of the array.. HOGGATT. JR. N3(x) = 2-x. We compute the first few numerators as Nf(x) = 7. and higher arrays again interrelated with the Pascal array.» 9 ••• 54 •• 273 » 1260 •• Let Gn(x) be the generating function for the nth row. Gn-rfx) = xGn(x)-x2Gn(x) + Cn-1 . 1. THE CATALAN CONVOLUTION ARRAY The Catalan convolution array. Catalan. We now unfold an amazing panorama of Pascal.+ (n+1)x + - we see that by the law of formation that the denominators for Gn(x) continue to be powers of (1 .NUMERATOR POLYNOMIAL COEFFICIENT ARRAYS FOR CATALAN AND RELATED SEQUENCE CONVOLUTION TRIANGLES V. San Jose. we discuss numerator polynomial coefficient arrays for the row generating functions of the convolution arrays of the Catalan sequence and of the related sequences S. written in rectangular form. E.6x + 2x2. Since G0(x) = 1/(1-x) = 1+x+x2 G7(x) = 1/(1-x) 2 + x3 + -+xn 3 = 1+2x+3x +n + .f and MARJORIE BICKNELL-JOHNSON San Jose State University. 7. N5(x) = 14-28x + 20x2-5. n = O. . Thus.x).[ 1 ] . where Cn-j is a Catalan number. N4(x) = 5. In three different ways we can show that those rows are arithmetic progressions of order/. N2(x) = 7. [ 2 ] . ••• and record our results by writing the triangle of coefficients for these polynomials: Numerator Polynomial Nn(x) Coefficients Related to £7 1 1 2 5 14 42 132 429 -1 -6 2 -28 20 -5 -120 135 -70 14 -495 770 -616 252 -42 -2002 4004 -4368 2730 -924 132 30 . is Convolution Array f o r S / 1 1 2 5 14 42 1 1 1 1 1 1 1 4 2 3 7 5 6 8 5 9 14 27 35 44 20 14 28 48 75 110 154 208 42 90 165 275 429 637 910 132 297 572 1001 1638 1 . California 95192 In this paper. 2. •••.

First. 4. 8. 616/56. 1. 14/1. The first case could have divisors from the zero th column of Pascal's triangle and is£ 7 . the convolution array for$2 ' s Convolution Array forS2 1 1 1 1 1 1 3 4 6 2 1 5 33 3 7 12 18 25 12 30 55 88 130 182 55 143 273 455 700 1020 1 7 42 245 1 8 52 320 1428 1938 The numerator polynomial coefficient array is Numerator Polynomial Coefficients Related to S2 1 1 3 12 55 273 1428 -2 -18 -132 -910 -6120 7 108 -30 1155 -660 143 1 -• 9 62 •• 408 •• 2565 - . 84. 165. The coefficients for each row also can be used as a convolution with successive terms in rows of Pascal's triangle to write the terms in the rows for the convolution triangle. 5. 135/15. 27. The zeroth or left-most column is already the Catalan sequence Sp We look at successive columns: n = 0 1(1/1. = sf - sf The divisors are consecutive elements from column 1. -.= S 1 2(1/2.35-6-20 +2-0 +2-0 +2-1 + 24 + 2. 2.6 .20-6-10 5.) = 1. For example. appears in the first column. the diagonal whose elements are given by [2"\ The rising diagonals. 770/21. The next falling diagonal parallel to the Catalan numbers is the central diagonal of Pascal's triangle. 2. •••. 4368/84. or. 2n~1.1 5-10-6-4 5.NUMERATOR POLYNOMIAL COEFFICIENT ARRAYS FOR CATALAN AND RELATED SEQUENCE CONVOLUTION TRIANGLES 31 Notice that the Catalan numbers. 8. or the sequence £7. 1. and the / column of the convolution array for 5 .) = n = 3 4(5/20.16. Thus.4.6 . 14. the third row has coefficients 5. and again as the bordering falling diagonal of the array. - = s. taken with alternating signs and deleting the first one. 70/35. 32. are closely interrelated.are all one. 5 14 28 48 75 = = = = = 5-1 .10 We can take columns in the array of numerator polynomial coefficients to obtain columns in the Catalan convolution array. 120/5. 42/1. and 2.. 2/1. 495/6. 10. 2. 6. the / column of the Pascal array. 28/4. 56. column 2. ) = n = n = 2 3(2/6.6 . • 1. 20. 572. •• ) = 1. . •••. . 5/1. 44. and we can obtain the third row of the convolution array for^7 thus. 35. THE CONVOLUTION ARRAY FORS 2 Next we write the numerator polynomial coefficient array for the generating functions for the rows of the convolution array for the sequence^. 42. 6/3. have sums 1. . ••• . The row sums. 110. and column 3 of Pascal's triangle.0 5-4 . 208. 4. taken with the signs given. The third row of Pascal's rectangular array is 1. 48. the/ t / 7 column of the numerator coefficient triangle for the £7 array. 20/10.

32 NUMERATOR POLYNOMIAL COEFFICIENT ARRAYS FOR CATALAN [FEB. have sums which are half of the sums of the rising diagonals. 36. Each column above is divided by alternate entries of column 1. column 2. columns of the convolution array for £2 a r ' s e f r o m ficient array. 108/36. 182. of the numerator polynomial coefficient array related to Sp Again. The Convolution Array for S3 For the next higher sequence S3.) = 1. 55. Also notice that when the elements in the ith row of the numerator array are convolved with / successive elements of the iT row of Pascal's triangle written in rectangular form.) = 1. n = 0. •• = S122. 12/1. •••. 132/8. 126. a convolution of the numerator coefficients in the/ row with / elements taken from the ith row of Pascal's triangle produces the ith row of the convolution triangle for S3. ••• . The rising diagonal sums taken with signs. 63. For example. as follows: tne columns of the numerator polynomial coef- n = 0 1(1/1. •••. while the falling diagonal bordering the array is S3. ••) = 1. or. 33. 18/6. •••. S2^n+1^ is obtained by multiplying the/7 f/7 column of the numerator polynomial coefficient array by n and by dividing by every second term of the (n . 2. i = 1. . which is found in Pascal's triangle by starting in the third row of Pascal's triangle and counting right one and down two.1)st column of Pascal's triangle. which is found by starting in the first row and counting down one and right one in Pascal's rectangular array. 9. 2. •••. for/ = 3. The next falling diagonal is three times the diagonal 1. s. the first column is S3. 1155/55. •••.S% n = 3 4(30/120. 35. 6. we obtain the third row of the convolution array for S3 as . 12. 560. 1. S3. are related to the rising diagonal sums taken without signs. 3. and the falling diagonal bordering the array at the top \sS2. 55/1. we can write the/ row of the convolution triangle forS^3. . 9282/364. 660/220. taken without signs. 6. 1. 102. the row sums are one. taken with signs. and could be obtained by multiplying the column by one and dividing successively by 1.-) = 1. Again. 1. 220. or. 3/1. the convolution array is Convolution Array for S3 1 1 1 1 1 1 1 1 1 2 3 1 4 5 6 7 8 9 4 9 30 72 15 22 39 49 60 22 52 91 140 200 272 357 456 570 40 340 612 969 1425 1995 2695 3542 4554 and the array of coefficients for the numerator polynomials for the generating functions for the rows is Numerator Polynomial Coefficients Related to S3 1 1 4 22 140 -3 -36 -360 15 312 -91 ••• Again. 3. 910/10. 10. .) Again. 1. Again. 6120/12. The rising diagonals. ••• = s f n = 2 3(7/21.i. column 3 of Pascal's triangle. (The diagonal in the corresponding position in the array related to £7 is twice the diagonal 1. Note that the zeroth column could also be expressed as S2. the zeroth column is S2.-. •••.9. rjf of the numerator array related to S2 by the curious formula /-/ = 4s. . 78. 12. and the falling diagonal adjacent to that is four times the diagonal found in Pascal's triangle by beginning in the fifth row and counting right one and down three throughout the array.= S32 n = 1 2(2/4.

-. we write S" = \-k— !(<+1)m+k-lU m =0. one can be much more explicit here.-.3 6 . column 2. 1 1 4 .. 1. n(i+1)(n + 1) _ J 1 (i + 1)(n + 1) I mi + (i+1)(n m = O.- which gives successive terms of the (k . column 1. 3/ +2. Let an+m/n be the element in the numerator polynomial triangle for 5/. i = O. do not fail to notice the beautiful sequences which arise from the first terms used for divisors in each column division for the columns of the numerator polynomial coefficients of this section. 60.Dst column of the convolution array forS/. 2.-) = 1. We write expressions for each element in each array in what follows.. Now. we obtain ( d + Dim +n) +i \ ( iin + m) + n + i - L+J.. using the form of the m th element of S f given in [1]. . 2.36/9. 2. 8. ((i+1)(m+n) + 1) l / . — .-. —. Actually.360/12. the n column of the array of coefficients for the numerator polynomiials for the generating functions of the rows of the S. and from Pascal's triangle to the convolution array. And.1 +15-0 = 2 2 . 12.-) = 1. beginning with the [in + 1)i. to obtain the successive elements in the (in + i .n ~j = {(n + 1hn+mn J (ih+ml+n^i-l after equating the two expressions for the /??f/? term of 5/ = i l l I (' + V(m+n) m \ m-1 . —. Thus.1.1jst convolution of the sequence S. 1. the nf column of Pascal's triangle. 1. 1 4 0 / 1 . 41 + 3.-.22.. we obtain the columns/^ 2i + 1. s(i+1}(n+1) so that. . 1. 1 4 0 . 4. in the nth column and (n + m)th row. . n = O. For the Catalan .1)\ where we are working with the sequence S. the divisors are every third element taken from column 0.convolution array is multiplied by (n + 1) and divided by every / f successive element in the nt row of Pascal's rectangular array. n = O.2.312/78.. or the sequence S'Jn+1K That is.22/1.4. 2. we can go from the convolution array to the numerator polynomial array. and the mth term in the sequence of divisors. of the convolution array forS.. —.1 0 .= $ | n =1 2(3/6. for 5/. Then.O. Now. we can write the elements of the numerator polynomial coefficient array for the row generating function of the convolution array for the sequence S. 1. ••• of Pascal's triangle. 2. and from Pascal's triangle directly to the numerator polynomial array.= S§ n =2 3(15/45. upon solving for an+m/n = am+n. •••. . m = 1.) = 1.0 +15-0 = 22-4 . when k= (i + l)(n + 1). ( m + n ) +n + j + 1 "n+ { m 1 V n + i )( (i + Dn + (i .3 6 . 2.36-10+ 15-4 We obtain columns of the convolution array for S3 from columns of the numerator polynomial coefficient array as follows: n =0 1(1/1.l]st term.3 6 . The actual divisiors in the division process are / i(m +n) + in. 1. . 4 +15-1 = 22-20 . 2. 2. 1560/120. m = O. .1977] AND RELATED SEQUENCE CONVOLUTION TRIANGLES 22 52 91 140 33 = 22-1 .4/1. First. Then. 1.1) + mi ) 1x n ' Now. 3. •••.= S132 Here. the topmost element in the/7 t/? column is given by an/n. + i\\ ™ " ' n = O. THE GENERAL RESULTS FOR THE SEQUENCES Sf These results continue.

i)a 0 F +h N. "Fibonacci Representations of Higher Order. and Marjorie Bicknell. — ." The Fibonacci Quarterly. Vol. N . 70.4.Scoville and V. 1. 395-404. Vol.i 1=0 = FN. 2.. L. pp. L. No. •••. No. 5. Jr.i+1a k=0 REFERENCES 1. pp. the central column of Pascal's triangle which gave rise to the Catalan numbers originally.NUMERATOR POLYNOMIAL COEFFICIENT ARRAYS FOR CATALAN AND RELATED SEQUENCE CONVOLUTION TRIANGLES 34 FEB.i+1+k~^L F N. pp. "Fibonacci Representations. 1977 sequence S7. •••}. 10 (1972). E.i = F Nti+k+1~ k-1 Zl FN. and Marjorie Bicknell. •••} upon successive division by 1. "Catalan and Related Sequences Arising from Inverses of Pascal's Triangle Matrices. 2.i+1+j = F N. V. k = 0. Jr. 15. — . 3. Hoggatt.i+j+FN. 43-70. Vol. E." The Fibonacci Quarterly. Hoggatt. upon successive division by 1. 10(1972). 14. Hoggatt.. 91. a N-1 N+i+1 _ sp hFk anN-k+1 ~ 2^ N. Carlitz.3.2. 7. Scoville and V. 2. E." The Fibonacci Quarterly. R. 2. 60. For .45. " The Fibonacci Quarterly.i +F N. Jr. 6. —. If the theorem is true for some/ > 1. 3. pp. 20. . 14. 120. 6. REFERENCES 1.Sj. V.-". Jr. the first divisors are 1. which diagonal of Pascal's triangle yields S2 upon successive division by (3j+ 1). 2. Carlitz.. and S2 = {1. which p r o d u c e ^ = (/. 1. 10 (1972). j = 0. so the above equation reduces to N-1 N+i+1 _ T . 2." The Fibonacci Quarterly.i- Now k F N. then Proof. L. [Continued from page 66. Scoville and V. Hoggatt.i k=0 N-2 _ ^ Fhk+1 ~ ZJ N. —. E.i+1 • j=0 Also FI\IJ = FN~j+1. 71-80. the first divisors of successive columns were 1. 4. and General Sequence Convolution Arrays in a Matrix..i N-2 _ y * (F{tk+1 ~ 2^ N. pp. Vol. they are 1.rk N-k a ~ 2-J hN. 2. Then N-1 N+i ^ rk • _ „N-k 1 n n k=0 The case / = 7 amounts to F1^ 7= 1.i k=0 F0 )nN-k +hN. Carlitz. R. Hoggatt.. Vol. For S2.4.i k=0 a N-k 0 N+1 + +h F a n N.] ON THE N CANONICAL FIBONACCI REPRESENTATIONS OF ORDER N N-1 v^ N i i=0 for some N > 2. 136-142. •••. "Fibonacci Representations of Higher Order I I . 21.1. Jr. R. E. Catalan. 3. "Pascal. 1-28.

respectively.S. Definition 5. b. A primary Fibonacci-like sequence {Jn}. we will need a few definitions. p) = s will be called the multiplier of the P. b.L. North Merrick.S.S. J? = 7. where (p/q) is the Legendre symbol.F. It is the first positive integer/7 such that J^ =0 (mod p) andJn+i =Jj = 7 (mod/?). y/'FT4b)/2) The specific generalized Fibonacci sequence beginning with (1. and for which JQ = Q. p) is the period modulo p. Then s(a.L. modulo/7. will be designated by [Hn] and {H'n). of a P. hereafter called a P. Definition 1. beginning with either (KQ = 7. in which Jn+l = aJn +bJn-i. Generalized Fibonacci sequences satisfy the same recursion relation as the Fibonacci sequence. &(a. pi The next fact that we will need is that if (a2 + 4b/p) = 0 or /.S. modulop. b. then the period of the G.FIBONACCI-LIKE GROUPS AND PERIODS OF FIBONACCI-LIKE SEQUENCES LAWRENCE SOMER 1266 Parkwood Drive. p) is the exponent of s(a. Definition 2. p an odd prime. hereafter called G. is one which satisfies the recursion relation: Jn+i = aJn + bJn-j for some non-negative integers.p).S.L.b. To prove that these form multiplicative groups modulo /?. b. The G. It is equal to pi la. b. F.F. To continue.S.S. One of our main results will be to prove that the exponent to which any non-zero residue r of an odd prime p belongs is equal to either the period or one-half the period modulo/7 of a Fibonacci-like sequence. note that the congruence: be + acx = ex2 (mod/?) leads to the conqruence: 35 . b.(1-sj5)/2)t reduced modulo/7.{1+y/5)/2). K1 = (a - ^+Tb)/2).. is a Fibonacci-like sequence {Kn) in which KQ and Kj are arbitrary non-negative integers. L S. We will give a proof of this theorem and draw some consequences. and (1. is the least positive integerm Definition such that Jm = SJQ = 0 and Jm+1 ^ sJ-j = s (mod/?) for some residue s. in fact an infinite number of them. with a maximal period modulo/7. Mew York 11566 The purpose of this paper is to investigate Fibonacci-like groups and use them to show that for any odd prime /?.L. called the restricted period of a P. \sia.F. At the conclusion of this paper. a. A generalized Fibonacci-like sequence. 4.. reduced modulo/?.F. b. modulo p.F. beginning with (I (a + ^2~T~4b)/2) will be designated by {Mn} ar| d tne G.L. there are Fibonacci-like sequences.. we will present a program to show how one might apply Fibonacci-like groups to problems concerning primitive roots modulo an odd prime. forms a group under multiplication (mod/7). a(a. Definition 3. except when both/7 = 1 (mod 4) and r = ±^~l (mod/?).F. p)/a(a.L.F. L. beginning with (I (a by {M'n}. p) (mod p). K1 = (a + s/FT4b)/2) or (K0 = 1. antiJ2 = a.

. th If Fn = 0 (mod p). pf 5. 4. This group is {1. 16. If p = 17.. Now. Then. by assumption. is ((1 + ^5)/2)n-((1-^5)/2)n ^-1n Let us now assume either Hn = 1 or H'n EE 7 (mod/7) but that Fn£0 (mod/?). 2} and has a period of 4. and there is only one Fibonacci group. [Hn] and {Hn}. Then a Fibonacci-like group exists iff 0 (a2+4b/p) = (13/p) = 0 or 7. This has the solutions x = V2aiL1/2^T4b (mod/?). either. Then (mod/7).sj5)/2 = Mt = 3 (mod 5). Using the quadratic reciprocity formula. 5 = 0 (mod/7) and/7 must equal 5. Let us assume that both Hn and Hn are EE 1 (mod/7). let a = 1 and b = 3. Our method of proof of the main theorem will be based on the length of the periods of special types of Fibonacci-like groups. .s]~l3)/2 (mod 17) = (1±8)/2 is A. 4). To demonstrate my method of proof. Letting c = 7. (modp). Thus. If/7 = 5.36 FIBONACCI-LIKE GROUPS AND PERIODS OF FIBONACCI-LIKE SEQUENCES bcxn~1+acxn ^cxn+1 [FEB. For an example of a Fibonacci-like group.= 13 (mod 17). all the terms of the two Fibonacci groups {Hn} and [H'n} which are EE 1 (mod/7) can be expressed as: Hn =E ((1 + ^5)/2)n EE ((-1 + sf5)/2)Fn Hn = ((1-j5)/2)n EE (~1-sj5)/2)Fn + Fn+1 s 7 (mod/7). 13. (1) Hn EE ((~1+^5)/2)Fn + Fn+1 //. These sequences will be called Fibonacci-like groups modulop and the sequences {Hn} and [Hn] will be called Fibonacci-groups modulo p. Then Hn =±1 (mod/7) if Hn = 1 Hn EE ((-1+sj5)/2)Fn + Fn+1 EE 7 (mod/7). GJ =d. Note that the product of the/? terms of the two Fibonacci groups modulop. we will investigate the periods modulo p of the Fibonacci groups. can be generated from the Fibonacci sequence lFn\ by the formula: Gn = (d . EE ((-1-j5)/2)Fn ^ 7 (mod/7) {mod p) + Fn+J EE-7(mod/7) + Fn+1 = 7 (mod/7). namely those for which b = 1. (mod/7). suppose p i 5 and Fn ^0 (mod p). Note that these sequences have both the additive structure of a Fibonacci-like sequence and the multiplicative structure of a cyclic group. Hn-Hn Since Fn^O EE 5Fn EE 0 (mod/7). 3. then (1 + ^5}/2 = (1. then Fn+] must be EE 1 (mod p) and the n term of both the sequences [Hn } and [H'n) will be EE 1 (mod/7). or: + Fn+1 EE 7 (mod/?). Any generalized Fibonacci sequence {Gn} beginning with GQ = C. we see immediately that we obtain the group generated by the powers of x. or Hn EE±7 (mod/7) if H'n = 1 (mod/7). = ((-1-y/5)/2)Fn-f-Fn+1=-1 or (2) / / „ EE ((-1+^5)/2)Fn //. and Thus. H'n = ((-1-y/5)/2)Fn + Fn+1 = 1 (mod/7).c)Fn + cFn+ 7. and this gives rise to the Fibonacci-like group (1. we can see that Fibonacci groups exist modulo p only when/7 = 5 orp =±1 (mod 10). then a solution of x = (1±.

L.7. Thus. p) if/? ^ 5. 7. Note that F 2n = FnFn-i + FnFn+1 = Fn(Fn-7 + Fn+j).p). we observe that Fn =-2/«J~5 and Fn+j = ./ = Fn+1Fn = -Fn + 1 (mod/?). Then the product of Hn and Hn SEE . 7. p) = 4. were = -1n (mod /?) as before.. But the first solution for Hn or H'n = 7 generated by an Fn ^0 (mod/?) can only be n = 1/2a(7. Now.^T~4~b)/2)n EE (-b)n (mod/?).7.p)+1 = F^+F2n+1 = 7 (mod /?). then this sequence can be generated from the corresponding P. then a(7.i.S. we see that Fn = £?A/5 and / 7 7 + r . Mn = ((a + -jl>2~T4}/2)n and = ((-a+. as we have seen before.1977] FIBONACCI-LIKE GROUPS AND PERIODS OF FIBONACCI-LIKE SEQUENCES 37 In both (1) and (2). = ±1/\/5 (mod/?). 7fp) = 2a(7. /. n = 1/2a(7. by adding Hn and H'n we see that/ 7 .p). if Fn =2Fn+i. Fn+i = 7 (mod/?). n= 1/2a(7.7. look at the two congruences: F2n = Fa(i. Then a(7. This is true since /? must equal 1/2k>a/7.p)= 7. Thus. Fn^O (modp) can have no solutions since the zeros of Fn (mod/7) can only occur at multiples of a(7. n This product will be =-1 if b = 7. It is known that the only possibilities for fi(7. In (1). 7. From now on.. If (5/p) = 0 or 7. (See [2]. In (2).7. the period of the unique Fibonacci group is 4. not .F.p) = FnFn-i + FnFn+i'= 0 (mod/?) and F2n+1 = Fa(i. To generalize this result to other Fibonacci-like groups. in agreement with earlier results. We have now proved our first lemma. and.7fp). if |3f 7. L.the product of / / „ and Hn =-7n = 7 (mod /?). Lemma 1. in discussing Fibonacci-like groups [Mn] and [M'n] modulo /?. = 2Fn+i (mod p). p) = 2. p). The product of the nt terms of the two Fibonacci-like groups is: ((a + ^+Tb)/2)n• ((a .pl But since a(7. if ]3/7. p) = 0 (mod 4) [ 2 ] . Solving for Fn and />. 7.p) and H^^^p) are = 7 (mod/?). then a/7. 7. 7. if b = 7.p) = 2. F. p) = 0 (mod 4). the order of the two Fibonacci groups modulo/7 must both be oj(7f 7. p an odd prime.Hence.7. Hence. k= 7 and n = 1/2a(7.i. 7. But both Hfji(i. If j3f 7. 7.7.p)f while the period 1 of the other group is /2\±(7. we obtain Fn = 2/sJ~5 and thus Fn+1 = 7/^/5 (mod p). Thus. {Mn} and {M'n}. If j3/7. K1 = d are the first terms of a G. 7.p) = 4.p)and the period of the other is a(7./ A / 5 (mod/?).p) = 2 or 4 and p f 5. then ^ .p) = \ (mod 2) is the first place where either Hn or Hn can be = 1 and Fn ^ 0 (mod /?). Thus. 7.f + Fn+1 = 0 a n d F2n = 0(mod/?). if such a solution exists. or 4. Fn =2Fn+1 (mod/?) is not solvable if |3/7. the order of both Fibonacci groups must be \i(7. b will equal 1 and (a2 +4/p) will equal 0 or 1. Thus. Thus.7. 7. /. by the formula: Kn = (d . If |3/7. /. it would be helpful if the product of the nf terms of these sequences. the period of one Fibonacci group is1/2<i(7. p) are 1. Hence. the period of one Fibonacci group modulo p is a(7.1 .p) \\$(7.pi for some odd integer Ar. if (3(7. If/? = 5.7 " = .i. /?> is an odd number. 2. p).1 (mod/?).7. a contradiction. 7.p) . if Fn =2Fn+1/ Fn^O (mod/7). /?J = 4.#/>.7. then Fn. then the periods of the two Fibonacci groups [Hn) and {Hn} modulo/? are both \i(7.p) = Q (mod 2). 7. If Kg = c. Now.p).S. then F2n =0. The last case occurs if $(7.* 7. 7.jli2~^~4)/2)Jn+Jn+l (mod/?) .p) = \x(7./? divides li(7.ac)Jn + cJn+i.7.p) = 2 (mod 4) [ 2 ] . all solutions of Hn = 7 or Hn = 7 (mod/?) must be generated by Fn =Q.7fp) = 7.) But. by subtracting Hn from Hn. Thus.

l.S.p) = 2. \iJn^O (mod/?) and bothM^ and M'n are = 1 (mod/?). •/„ ^ t f anda 2 ^ 4 ^ 0 (mod/?). If (a2 +4/p) = 0.p) =4 (mod 8).p) while the period of the other group is p(a. To prove this let n = a(a. then no such instance can occur since afe 7. $(a.p) = I Thus. we see that.J^+1 = / and $6?.p) must be either 1. we solve for Jn andJn+j by the formulas: T ^ = 0 (mod/?).p) = 2. If $(a./ (mod 4) since (-1/p) = -1. then (a2 +4/p)= 7. Lemma 2.p) = 2 (mod 4) if j3fo 1.l n (mod /?).a 2 // = (a/2)2 = .1./ (mod/7). This also shows that1/2d(a. or 4. (S(a.S. we have now proved our second lemma.But Jn = 0 (mod /?) and 1-Jn-1 = Jn+i . we obtain exactly the same formulas as for the Fibonacci sequence. Thus.aJn = Jn+i (mod/?).1Jn+1 . then one otMrifMnm 1 and the other is = .L.1.L.p) = 0 (mod 4). 1.p)= 7.F. 1. The method for finding the periods of Fibonacci-like groups with b = 7 is along the same lines as before. then the period of the single Fibonacci-like group is.1. If 0/a.p)= 2 or 4 and (a2 +4/p) = 7. with 0 <a <p. 1.p) = 0 (mod 8). l. 1. Thus. As before \ijn=0 (mod/?). \ijn+1=-1 (mod/?)./. 1.1. a/? belongs to the exponent4 modulo/? if (a2 +4/p) = 0.S.p) = 1 (mod 2) and a(a. 1.J„ = .*/ = 1/2aJn and that 7„ = ±2/^]a~2~~+~4. as before. 1. Buta2+ 4 = 0 (mod /?).1. But then there is only one Fibonacci-like group [Mn) and Mn = (a/2)n (mod/?). Then Jn. [Jnj eralizations of some familiar Fibonacci identities: (a) J2n = (c) (mod/?). Thus.p) if $(a. Also.p)= 1 (mod 2).1.pl The remainder of this paper will be devoted to finding for a given odd primep all the P. with a and b unspecified. This shows that 1/2a(af 1. Solving. if eitherM n orM n = 7.1 (mod/?). Note that this further shows that if fi(a. l. if a2 + 4^0 (mod/?). since a(a. and the period of such a Fibonacci-like group (mod/?) is 4.)\\ Jn+1 = I then ^ ± 7 . Also. one finds that the product of Mn and Mn is = 1 (mod/?) in contradiction to what we have determined before.. then one can solve for Jn an6Jn+j by the congruences: J2n = 0 (motip).38 FIBONACCI-LIKE GROUPS AND PERIODS OF FIBONACCI-LIKE SEQUENCES M'n E= ((a-sja2 +4)/2)n = ((-a-s/a2 We wili next need a few formulas for P.p) = 1. we determine that their product = —1 (mod/?). J%+1 =-1n (mod p). then Jn+1 must be = 1 (mod/?) and bothM„ mdMn = 7 (mod/7).p) = 4. The periods of the Fibonacci-like groups {Mn} and [Mn] modulo p are both \±(a. (This also shows that no term J2n+1 of a P. if it exists. J2n+1 = 1 (mod/?).J^+1 =-1n (mod p)'.$(a. If n is o6d. If (a2 + 4/pf = 7 and fi(a. and studying the Fibonacci-like groups that they generate. Substituting back. and (a2 +4/p) = 0 or 1.F. lfjn+1 = 1..p) = 1.p).F. 1.J2n + i = ~1 ( m °d p).L.p) = 4. then id(a. and \x(a. If (3(a. bJnJn-i+JnJn+i J2n+1 = bJn +Jn+1 • These formulas can easily be proven by induction. Jn+1 = ftaJn = ±a/sja2 +4 (mod/?). we find that Jn = ±2/^Ja2 +4. is /7 = %a(a.4. in accordance with our previous results.p) = 0 (mod 2). with b = /can be divisible by a prime p = . then we have: /Va 2 +4)Jn =0 (mod/?) and a2 * 4 = 0 (mod/?). Thus. then the period of one Fibonacci-like group is'Apt(a. 1 'Jn-1 =Jn+l — aJn = 1/2aJn — aJn = —1/2aJn = —Jp+1 (mod /7). Jn+1 = ^ a ^ = ta/ja^+i . These formulas are simply genh1)nbn'1 Jn-lJn + l-Jn= (b) +4)/2)Jn+Jn+1 [FEB. . Solving for Jn a n d ^ 7 .1.p).p) = 7.l. if we substitute back to determine Mn and M'n. Hence. If (3fa. 1. a(a. If b = 7. the first/7 > #such t h a t / I ^ orM'n = 1 (mod/?) is generated by aJn £ 0 (mod /?).1. Let us now look at the terms of {Mn} and {/Wn} which are = 1 (mod/?). b = 7./?^ is odd.p) = 4. 1. 2.

then the number of P. However.a) = 4 (mod/?). If d / 4 and^/ = 0 (mod 4). Unless stated otherwise. the first terms of which are equal to each of the/? .L.p) for some P.. then e = \±(a.p) for some P.p)= I (3) We/4.a) = 4/k. If/? is an odd prime equal to either 4n + 1 or4n +3.L. and is not equal to 4.p) = 4. and (a2 +4/p) = 0 or 7. In each of the different cases.L.1 and b = 1. {Ja. There are p — / sets of solutions f o r * and a. 1.1.1./?. where 0 < a </? and (a2 + 4/p) = 0 or /.F.a2)2 = 0 (mod/?). ax = a 2 (mod/?).S.F. (a ± *Ja2 + 4)/2 gives rise to two distinct residues. then e = /ife.F. The exponent e to which a non-zero residue r belongs modulo p is equal to the period of the Fibonacci-like group of which it is the first term.a2 = (x + a)(x . If 0 < a < /?. then the number of P.i} with 0 < a < p . (1) If e = 1 (mod 2). \s(p(e).pj = d is (p(d).p) = d \%1/2(t)(d). (x+a) s -4/k. square both sides and collect terms. (x+a) = -k.a) = -k (mod p). Combining our previous results. ± ^f+~4)/2 imply t h a t ^ =a2 (mod/?). 7. divides/? ./?^ for some P.F. 4 sets of solutions lead to the same x2 and a2: (x + a) = k. {Ja.S.L.p) = 4.p) = 2.1. if we put the square roots on the same side of the congruence. (4) If 0 = 0 (mod 8).L. = 1 (mod 4) iff both ±4 are quadratic residues modulo /?. In general.a) = k.i} witha</? andp(a.S. such that (a2 + 4/p) = 0 or 7.F.7.L. I next claim that the set of numbers of the form ((a±yja2 + 4)/2)./for some P.L. (x-a)=4/k (mod/?).7. with recursion relation: Jn+i = aJn + bJn-f will be denoted by {Ja.F. and $(a. then the period of any P.S. (x .1. an odd prime.F. Also.7.S.S. then there are 2n + 1 P. The next corollary is very important.i}. Since k^Ofk^-k and 4/k £ -4/k (mod/?).i}. 4/k = k iff k = ±2 (mod /?). is even.8ala2 +4a\ = 4(ax . p / 5. thene = V2\i(a/l. (x + a) = 4/k. Thus. {Ja.1. In general. 7. The P. (5) If e = 4. we obtain the congruence: 4a\ .p) = p. a<pr with \x(a. then there exist 0(4) = 2 P. 1' < k < p .S.L. This theorem leads to a number of interesting corollaries. one finds that the number of solutions ofx 2 = a2 +4 (mod/?) is n + 7. {Ja.F.L. Combining all these conditions with the fact that a2 +4 = 0 (mod/?) is solvable only if/? = 1 (mod 4).F. except in the case where a = 0 (mod/?).F.7 Fibonacci-like groups. e = 4 (mod 8)./?. generated by (x+a) = -k. with y. all one needs to do is find all solutions of the congruence: x2 .L. if/? = either4/7 + 7 or 4n + 3.S.S. { J a f 1 ) with a <p dx\&fi(a. /? is an odd prime. there exist an infinite number of P.b}Theorem 1. b = 1. {Ja. If d divides/? . 1. = -4/k.L. gives rise to all the non-zero residues of /?.p)= 7. a and —a. l.S.7 non-zero residues modulo/?. we are now ready to state our main theorem. and b= 1. 3 </?.1977] FIBONACCI-LIKE GROUPS AND PERIODS OF FIBONACCI-LIKE SEQUENCES 39 To find all 0 <a <p such that (a2 +4/p) = 0 or 7.S. then the number of P. we see that all the non-zero residues are obtained if the congruences: (ax ± >Ja\T4)/2 = (a. Corollary 2. then e = id(a. Corollary 1. with the maximum possible period modulo/?. {Jaj}.i} with a maximal period of/? .7 is . {Ja.F. This follows from Theorem 1 and the fact that the number of residues belonging to a particular exponents modulo/?. It states that for any odd prime. Each P. If 0 < a <p .i} witha</? and $(a. a(a. Proof.l} with a <p ar\df5(a.L.1. {Ja.S. These generate/? . (x .L. Combining these facts with the fact that/?.F. generates a Fibonacci-like group with a period of 4. wheree dividesp . -4/k = * iff * =±^J-4 (mod /?).7 and d = 2 (mod 4). (x-a) (x .F.S. (2) If e = 2 (mod 4). {j3/i} with a < p.(a.1.

JU(1.S. {ja.1. Proof. 1.b.i} with with ($(a.F.L. If (a2 + 4/p) = -1. then so does 1/r. In addition.S. If (a2 + 4/p) = 1. If a=d (mod/?).1. the summation of all the a's of P.LS.F. I will use a result of Robert Backstrom [1].1. He stated that if s is a prime and/? = 2s+ 7 is prime such that(-b/p) = -1 and (a2 +4b/p) = +1.S.40 FIBONACCI-LIKE GROUPS AND PERIODS OF FIBONACCI-LIKE SEQUENCES [FEB. and we obtain the result.and ^ -^ 7 are primes. then/? = 3 (mod 4). One of my purposes in writing this paper was to see if I could get any general results on the relation between residues and the primes of which they were primitive roots.1.F.I Hence. and (5/13) = .13) = 28. {]n} and { . Unfortunately.S.lf a = 0 (mod p).F. then 1/r = (-a ±^J(-a)2 + 4)/2 (mod/?). I will prove. but we can still find isolated cases of P.p) = I This follows from the fact that/? .LS.F.13) = 4. {Jai} with \i(a. { ^ 7 } has (l(a. Examples are: (5/7) = .L.L. then every P. 1.1A\b= 1. then the P. has(3(a. then the number of P.p)= 1.5) = (3(4. — ) with fi(a. This gives rise to the 2 trivial Fibonacci-like groups.L. then Corollary 2 does not apply.1.F.F.F. {ja/1} (mod 2). 1.F. then there are an infinite number of P.S.7) =2 and JU(1. then the number of P. then (a2 +4/p) must = 1 as we have shown earlier.p).1. {Jaj} has (3(a.S. If p .1.F.L.L.1.p) is at most/? .L.7.7. 3. Corollary 4.7) = 16. every P. If 0 <a </? and/? = 3 (mod 4). then the P.5) = 4. %\x\z%(-b/p) = . then all quadratic non-residues are primitive roots modulo 2s + 1. 0(1.S.S.1 / 7 } . Thus. If (a2 + 4) = ~1 and fi(a.1)\\ p . The rest follows from Corollary 1.7 = 0 (mod 4).S. {Ja. \i(a. (a2 + 4/p)= 1.L.p) = p . Note that if $(a.1.7 is <j>(p ~ 1).L.1 .p) = 1. {j3f1} with fi(a. using my method. If 0 < a < /? and e is an even number dividing/? .S. If p = 5. $(a. {Ja/j} and { 7 ^ 7} have the same period modulo/?. s = 3 (mod 4). then one can generate a Fibonacci-like group whose period is at most/? .L. {Ja.F. then no P. Combine this with the fact that if r = (a ±sja2 +4)/2 (mod/?).S.i}.p) = 2 or 4.i} with \x(a. (0.7 = 2 (mod 4). excluding — 1 .) If a now ranges over the non-negative integers. 0(1. then \i(a. (These periods are maximal since T/2 +4/5) = (42 +4/5) = 0 and (}(1. and Fibonacci-like groups to obtain results about primitive roots.1. {Ja/j} with a maximal period modulo/?.S.p) = 2 is c/jp-r d=0 (mod 8) The number of P. 1 /2(p(p .L.S./?. then the summation of all the a's of P.p) = 4\s 1 /2 £ <t>(d) . I was unable to obtain any new results. the well-known result that if 5. 1. 0 < a < p. then a(a.1. {jaf1} with a maximal period modulo p of p .L. to prove this. If 7 < a </?.1.7. Corollary 5.1.1. {jjj} = [Fn] and {J4fi\each have periods of 20. and the maximal period modulo/? is/? .1-1 (mod 4).p) can be at most 2(p + 1).F. {Ja.L.i} with maximal periods. Corollary Proof.1. Thus.p)=* 1 by . d\p-1 d^4 (mod 8) Corollary 6.1.F.1.F.p)= 7 is If 0 < a < p and/?.p) = e is = 0 (mod /?).7 =2 (mod 4).1 .F. Proof One can prove this by using the fact that if r belongs to the exponents modulo/?.7 and which equals \x(a. 1.p) dividinge is = 0 (mod/?).0.1"=2kY\pj/ p/=] ' The number of P. then one gets the trivial P. But I will close this paper with an indication of how one might use P.p) = 7 iff/? is a Fermat prime = 22 + 1.

217-240. the general solution of (1) can be given in the form (2) Pr= C1Rr1+C2Rr2 + C3R. /?2 and R3 are the roots of the equation R3-2R2- (3) R+1 = 0." The Fibonacci Quarterly. { ^ . as texts on theory of equations seem to say. 354. and the others correspond to the quadratic non-residues.2469796. F . The roots of (3) are r Rl = 3 [1 + ^ cos< ^ R2 = I [2 .-. that represents the number of paths for/.3f where the quantities R?. 1966).1) correspond to the quadratic residues of p excluding 1.1) and one with a period o f / ? .] .4.reflections in three glass plates (with initial valuesPj = 7. Backstrom.S. This cubic. 4. r = 3. (Approximations of them are Rf = 2.L.sfi cos 0 + V ^ T sin 07 (4) R3 = 1. "On the Determination of the Zeros of the Fibonacci Sequence. excluding -1. L S .) The constants in the solution (2) are then found by solving the linear system [Continued on page 45. ?2 = 3 and P3 = 6). have a period o f / 7 .V27" sin 07 where (5) 0 = |. No. FULTS Student.7. Vol.7. 4. 2. Based on the usual theory for such relationships. No.(0ct. 3. } can have is 2 or/? . Vol. 3 (Oct.5549581." The Fibonacci Quarterly. whose discriminant is equal to 49. Robert P.F.8019377. 345-348.[2 . It is easily seen t h a t ^ / / ? . 4 (Dec. and the constants C-j. "The Fibonacci Group and a New Proof That Fp-(5/p) = Q (mod/?). 4.arc cos ( — L _ Such roots can be represented exactly only if they are left in this form. Halton.7. REFERENCES 1. each giving rise to one Fibonaccilike group with a period of 1Mp . 1966). Those with periods of 1Mp . I submit here an explicit expression forP r / and also obtain its generating function. Saratoga High School. Somer. Marjorie Bicknell-Johnson gave the recurrence relation (1) Prf1-2Pr-Pr-f+Pr-2 = 0. Saratoga. pp. has three real roots. and R3 = -0. The only periods that a P . California At the recent research conference of the Fibonacci Association." The Fibonacci Quarterly. Lawrence E. SOLUTION OF A CERTAIN RECURRENCE RELATION DOUGLAS A. the only even numbers dividing p . No. pp. 313-322. "On the Divisibility Properties of Fibonacci Numbers. 10. and they can best be expressed in trigonometric form.V 7 cos 0 . C2 and C3 must be determined to fit the specified conditions. John H.3) of these P. pp. R2 = 0. Vol.7.1977] FIBONACCI-LIKE GROUPS AND PERIODS OF FIBONACCI-LIKE SEQUENCES 41 Corollary 3. 1972).

Bicknell and Hoggatt [4] have shown another method of getting Tribonacci Numbers. E. >*__. on the arrays to make every row a homogeneous function i n * andy. 4 4 . The numbers g(n. California 95192 Stanton and Cowan [1] have discussed the two-dimensional analogue of Fibonacci Numbers.r) and discuss properties of functions related to Tribonacci Numbers.r) = 1 Carlitz [2] has discussed in detail a more general form of g(n. San Jose State University. T1 = 7.ym T0 = 0. 1. HOGGATTJR. Analogous identities have been established by Alladi [3] for Fibonacci Numbers.. 8 1 . ^ % 2l2^ 1 1 * "N. which is the Tribonacci sequence Tn = Tn-i + Tn-2+Tn„3.4. In this paper we get the Tribonacci Numbers from g(n. India 600004 and V.2. -s» ^_ 1 13 ~1 The descending diagonals are denoted by dotted lines. r) + g(n. San Jose. r) rfn > O integers. 42 . ^Currently at UCLA.r). g(n. 41 Array 1 25 ^ r-"7"'^ " l 3 k^5^. ^^3 k.. T2 = 1. 2 4 . 1 3 . 1 " 63 -^ s ^ 25 ^ 4 1 7 .ONTRIBONACC! NUMBERS AND RELATED FUNCTIONS KRISHNASWAMI ALLADl* Vivekananda College.7. They dealt with numbers g(n + I r + 1) = g(n + 1. Madras. "5 ^.r) can be represented on a lattice as follows: k 11 ^ * • » k 9-. We now add variables (suitably )xn.0) = g(0. . r+1) + g(n. The above figure is transformed into a Pascal-shaped triangle by changing the descending diagonals into rows 1 1 1 1 1 1 3 5 7 1 5 13 Array 2 Tribonacci Triangle 1 7 It is interesting to note that the sequence of diagonal sums in the Pascal-shaped triangle is 1.^^ 9 >* ^_ ^ ^11^ ^ ^_1 1 V.

y) y ™-2(x.y)-~ + y7^1(xfy) tn(x. Assume that statement holds far. From the recurrence relation \wTn(x. n = Of 7.yj = xTn-^xy) -yrn-2(x.y)+y k=0 r m-1 y \ H Tklx.y).y) = = tn(x.y) "" T<n(x>y)+X -Tx~ +y dTm-i(x.2(x/y)\ J -| Tk(xfy)Tm^k-3(x. Proofs.y) = J2 Tk(x.y) —^— .y).k.y).Y) + xyTn-3(x.2(x.k(x. + T we see that zTm-2(x.y) to g(x. do there exist relations between these functions? To get symmetric results we denote Tn(x. —.y)\+g(x.1977 ON TRIBONACCI NUMBERS AND RELATED FUNCTIONS 7 x x2 x3 x3 y 3xy 5x2y 7x3y Y2 5xy2 13x2y2 y3 7xy3 43 y4 The rising diagonal sums give a sequence of functions Tn with the following rule of formation: Tn(x.y) are the remainder terms from the first and second summations in each square bracket Now using the recurrence we may simplify g(x.y) = yTm.y)Tm-k-i(x.y)Tn.y)Tm-k(x. m.y) k=0 "I m-2 £ k=0 m-3 + y YJ k=0 m-4 Tk(x/y)Tm.y)+h(x. t*+1(x.y). Let us denote the partial derivatives and convolutions by the following n Tn(*.y) + Y J2 k=0 [ + m X) Tk(x.y)Tm-k-4(x.y) k=0 k=0 J applying recurrence for Tn(x.y) + h(x.y) dTm+1(x.y) which makes the right-hand side to be equal to the partial derivative *Tm+1(x) ax .y).y) by T*+1(x.y)Tm-k-1(x.y) + yTm-i(x.FEB.y)Tm-k-2(xfy) +y £ Tk(x.y) and h(x.y) . Theorem 3 follows immediately from Theorems 1 and 2.y) = ]C m-1 T k(x. 2.y) Lk=0 [ m-2 T k(x.Y) + yTn-2(x. whereg(x.y) + yTn-2(x.y) .y).y) = Tm(x. To prove Theorem 1 we would essentially have to show (D Tn(x.y)Tm-k+i(x. Theorem 1. Since Theorem 2 is similar to Theorem 1 we prove only Theorem 1.y)\ J *"| 2 Tk(x. ^n+1(x/y) Theorem 2.y)+yx —^— ' Now m+1 Tm+i(x. h(x. 3.y) = xTn-l(x. t*+1(x. 7n(x/y)+x7n-1(xfy) Theorem J. k=0 As in the case of Fibonacci Polynomials.y) —£T- dTm(x.y) = tn(x.

]P r=0 n r. 1 3 1 1 5 5 \1 7 13 Now denote by nT? the (n + Vth row(r+ -1 1 = 1 / 2 . Pi = /.44 ON TRIBONACCI NUMBERS AND RELATED FUNCTIONS [FEB. We conclude by considering the inverse of the following matrix.1. This means (1) holds for/7 = m + 1 and can be verified to hold for/7 = Of 7. In fact if we attached Xm Yr to the (m + l)th ior(r+ Vth column element of the Tribonacci triangle we get the generating function of the r/7 -^ />/f/7 colu mn as Gn(XfY) = X?YHll±A)^ (1-x)n+1 so that <3> E 1 Gn(X. v'. Po= 0. then the generating function of the (n + Vth column of this array would be (1-X)n+1 so that (2) 1-x-x2-x3 n=0 Now (2) indicates that the rising diagonal sums of array (2) are Tribonacci Numbers.'. We shall now discuss some more properties of the Tribonacci Triangle.1+Pn-2.Y) = n=o 1-X-XY-XJY which is the two-variable generating function of array (2). If on the other hand we shifted the (n + 1)*' column n steps downwards and in the new array added the term Xm to every member of the (m + Vth row. = o n > 1 (= Horn = 0) . If we attach the term xm to every member of the (m + Vth row then the generating function of the (n + Vth column is = xiii±xr Gnix) <1-x)n+1 so that £ (1) GnM= ____! 1-2X-X2 n=0 Now (T) clearly indicates that the row sums of the Tribonacci Triangle are Pell-Numbers Pn = 2Pn-. \ /:.3 7 1/ \-6 10 Vth column element of 1 1 2 6 1 -3 10 \ 1 -5 1 / 1 -5 Two interesting properties of nT*\stand out n p. By mathematical induction it holds for all positive integral values of n.

12. to appear. Stanton and D. using Cramer's rule.] R7R3 + R2R21 + R3R§ = 4 R1RI+R2R23 + R3R21 = -I. Such a generalization to higher dimensions may be possible but it is very complicated as it is exceedingly difficult to picture these numbers." The Fibonacci Quarterly. (12) [Continued on page 56. C1=L(3-R3). E.R3)(R2- R3) = s/W = 7. 4. pp. Alladi. r=0 REMARKS. L. We wish to draw attention to the fact that we obtained Tribonacci Numbers from Stanton and Cowan's Diagram. C3=1j(3-R2) when many discovered relations between the three roots are taken into account." The Fibonacci Quarterly.R3HR2 -R3) = 7 . pp. Math. 3. R1R2R3 while from the discriminant we have (11) (Ri- R2)(R1 . R1R2 + R1R3+R2R3 = -1. However there are other ways of obtaining these numbers as for example Tribonacci numbers from the expansion of (1 +x +x )nW. 11.. p.ON TRIBONACCI NUMBERS AND RELATED FUNCTIONS 1977] {n + \nT?\ E P. Carlitz. These involve the following. G. Jr. V. . 457. 277— 279. D. K. U 45 (= 1 for n = 0). R.] C1R1 + C2R2 + C3R3 = 1 C1R^ + C2R2 + C3R3 = 3 (6) C1R31+C2R2 + C3R3 = 6 . No. one obtains constants as C1 = f ^2^3(^3 (8) ~ R2>16 . 433446. and Marjorie Bicknell. "Note on a Square Functional Equation. 2. "Some q Analogues of Certain Combinatorial Numbers."SIAM Rev. Relations between the roots and the coefficient of the cubic gives (10) R7 + R2 + R3 = 2. Gowan. Hoggatt. Vol. Thus. 1970.3(R2 + Fl3) + R2R3] C2 = j RiR3<Fd-Fl3)[6-3(R1 [ C3 = 1 7 RjR2(R2- + R3) + R1R3] R1)[6-3(R1 + R2) + R1R2] which reduce simply to the fixed numbers (9) c2=17(3-R1). Use of these and the relation R1 + R2 + R3 = 6 furnish. after some manipulation. 4. 2." SIAM.. "Generalized Fibonacci Polynomials. "On Fibonacci Polynomials and Their Generalization. whose determinant is (7) D Rl R2 R3 01 ^2 p3 h2 ^3 n3 n3 n3 M1 = (R1 . [Continued from page 41. Anal.R2)(Ri . 1973. 5(1973). REFERENCES 1.

k. 658. the first ten values of Ck are: 3. Assuming it holds for n. 10. As k and m take on all integer values such that k > O and m > 0. we have a corollary oo (4) •£ -L = Ck . ••. 46. Since FJ/FJ-J -+ (1 + ^5)/2 as/ -> °°. 1674/2584. Morgan town.1 with a = l^fi- . then for /7 -^ / we find that we have to show that 7 F _ FkB-i F 2kB F2kB-l F kB 2kB ' and this comes by setting/ = kB in the formula (-D'Fj - F2JFH-F2HFJ . For k=\. FkB ~* where Ck is independent of /7 and in fact (3) r k f t k_j)/F =\ (1' l + rF k-V' ~k k \(1 + Fk_1)/Fk+2/F2k for even even K. (We use this notation to achieve clarity in printing. 3.1) Fj. 46/55. 7/11. 263.E. 1. 1674. or can be seen as a special case of the well known formula F m + 1 Fj + F F m j-1 = F m +j J when m = -2j and using F_j = (. which may be proved directly by the Binet formula.) A generalization may be given as follows: (2) F JL J~ = Ckm=0 . 6. 4305. 108. 7. then (2k + 1)2m generates each natural number once. 2.SUMS OF FIBONACCI RECIPROCALS W. 2. Hence for absolutely convergent series we have the general transformation formula 46 . GREIG West Virginia University. n > where b = 2m and B = 2n. • run "** kB-1 n. •••. Formula (2) is easily proved by induction. 21. Taking n = 1 we may determine Ck from 1/Fk+ VF2k = Ck-(F2k-l/F2kl It is from this that we have found (3). 10/8. 11146. tor for odd k. 3/4. 2/3. 263/377.. West Virginia 26506 Good [1] has shown that (1) V / = 3-^1 . 28980. This shows that Formula (2) holds with Ck independent of n. Our formula has an interesting application to sums of reciprocals of Fibonacci numbers in another way.k > 7. . If we write Ck with denominator as F2k t n e n the numerators form the interesting sequence 3.

. we get k odd k odd /c odd = 1. this becomes t n=1 n k=1 kodd fk k=1 kodd = 1.632 -+0..(-1/a)n)/\/5. By some simple manipulations with the Binet formula Fn = (an .389082--.^ 2k .030 ••• as given by Brousseau [2. using (9). 45].8245 -+1.+ 2 .1977 (5) SUMS OF FIBONACCI RECIPROCALS f((2k+ 1)2m) ]T f(n) = £ E n=1 ""XL k=0 m=0 tf*^- k=1 m=0 Arodd Applying this to the Fibonacci numbers we have (6) f i .218 = 3. we note that in view of (3).1462 -. using (8). + a4k-1 _L + _ ^ _ (9) ' = y/5 F2kg4k_1 g2k_1 To use these.i t n m=Q h n=1 k=1 k odd jr = i l-kb {Ck-\)> by W3J k=1 k odd = 2. series (6) becomes y F -J.080 -+0.0 . 7 5 7 1 . Next.8245 .5353 •-. k n=1 k A: odd K k=i k odd k=1 k odd ZK k=1 a k odd .8245 -+1.35988 - -+0.+ 0. Finally.382 . p. 2 9 2 4 .FEB.= y {-L +-2- no) ^ n A Fn ^ +^ F2k A Ift k=1 F7k k /rodd 1 - Fk aI so that by (7) we get (ID Z ^ L IU ^ F F? . a<*«_ 7 +1 -) . formula (6) m formed into some variant forms that we believe are of interest It is easy to verify the following: Fk-1 Fk (7) _J=_ a _L _ (8) F2k a2k+1 = _*£_ N/5 a2k •.+ 0.1.

2. Vol. 4. H. 1976).48 SUMS OF FIBONACCI RECIPROCALS FEB. V. Good later. 14. pp. No. k odd 0.083313 — = V* E k-i l { kT * a -1 2ak 4k a4k-1 . Gould for suggestions leading to the presentation of the ideas in the present form.5025865492-)V5 = 3. I. See [5] for an earlier treatment. " A Reciprocal Series of Fibonacci Numbers with Subscripts 2nk. E. A generalization of formula (5) will appear in Gould [ 4 ] . 346. 67-72. Hoggatt." The Fibonacci Quarterly. 15. 1977. California. No. p." The Fibonacci Quarterly. Gould and I. . 4 (Dec. "A Rearrangement of Series Based on a Partition of the Natural Numbers.. Jr. By bisection it is easy to see that E .E. J.580727 — )V5" = (1. pp. Gould. 1977). 12. Using the Binet formula it is also possible to rewrite (12) as (16) E in .68663 •••.E n-1 la3k+*2k+ak+1-2ak) . 453-455. San Jose. 5." The Fibonacci Vol. 1977 This last form of our result is most interesting because it is not at all what we get if we transform the reciprocals by simple bisection. W. No. and Marjorie Bicknell. 1972. 3. Vol. 5 (Dec. whereas it takes 12 terms of the other series to get this. 1 (Feb.-a'2n a2n 4n a -1 Comparing this with (13) •we find the interesting equivalence oo 0 oo (15) > a= l±f A: odd The series on the right seems to converge twice as fast as that on the left. J." The Fibonacci Association. A generalization of the main results here will appear in another paper [ 3 ] . REFERENCES 1. The author is also indebted to H.1 k odd = (2. Greig. "A Reciprocal Series of Fibonacci Numbers.1 = y -1 +E -1 = E f Fk F n=1 whence » h i M k odd j even k=l k odd E (14) h n n=1 k=1 i * „?. A preliminary form of this paper was written in October 1975 and communicated to H. and six terms give the sum as 0. t~k n=1 A: odd J5 a2". W. "On Sums of Fibonacci-Type Reciprocals. "Fibonacci and Related Number Theoretic Tables." to appear. Good. Brother Alfred Brousseau.^ - k~l a . W. 1974).359885665 ••• . W.

n./7 ones.n.0) = fi(0. n > 1 and it is understood that (j = O or V fj(m.s) it is convenient to define fj(m. r = 2.5) Fj = Fj(x./.r. Worth Carolina 27706 1.FIBONACCI NOTES 5.s^ the number of sequences (1.s is negative.nj. n = 2.777=4 n=2. We shall now consider the following problem.2.r.0) f0(lO.s=0 ^Supported in part by NSF Grant GP-37924X. Now put (1. r occurrences of (00).s) = fj(0.2. r= /.s) + frfm.7. s) fo(m. Examples. Let f(m. s = 1 ( 0 0 1 1 0 0 ) ( 0 0 0 1 1 0 ) f(4.0.s) = fo(m.1.s) = fgim . r(1 2) fi(m.s- where m > 1.r. n = 2.1) = 0 .r.s occurrences of (11) and with 57 = j. ( 1 0 0 1 0) f(3. 49 1). .3) and (1.v) = fj(m. In order to evaluate f(m. n ones.s) = O (r + s > O) (j = Oor 1) for all r. We also take (1.1) with m zeros.-.1.u. r = 1.n.s> 0.r.s) satisfies the following recurrences.r.n.r..r. 1. r occurrences of (00) and s occurrences of (11).2.s) denote the number of zero-one sequences of length m + n: (1.n.s)xmynurvs £ m.l. Durham.0) = 4 ( 0 1 0 0 1 ) \\.1) (ai. The point of view of the present paper is somewhat different from that in [ 1 ] . s= 1 /(4. r. where/= 0 or 1./?. n- 1.r.1.r.r.0.CARL1TZ* Duke University.4) fo(lO.r.= 0 or 1) am+n) with m zeros. n.n. n.a2.y.s) = 1 = 0 fi(OfO. (a.1) = 3 ( 0 1 1 0 0 0 ) 111.s) = O if any of the parameters /??. ZERO-ONE SEQUENCES AGAIN L. I. s) + f-jfrn . n- 1.n. It follows immediately from the definition that fj(m.2. m = 3. r. m = 4. s = 0 ( 0 0 1 0 0 ) ( 1 0 1 0 0) .n.r.

50 FIBONACCI NOTES and (1.E%M * k r z (rr)(sDww % min(m.k -2{m~1) ("-1)xmynum-k-1vn-k-1} Since F = f(m.r.n=0 (?) (V) »m-kv"-k £ k=0 It then follows from (1.k ( m.9) 11 .1 \ ( n .s=0 it follows from (1.r.xy In the next place.3).8) f1 = i x y +y(i -xu) (1 .y.XU)FQ- (1.u.7) -yF0 xFi = x + (1-yv)F1 =y .7 ) = i xmy"um-kvn~k-7 {\mk 1) [lZ\)xnVum-k-1vn-k+{mkZ]) (n-l^xmynum-k-1vn-k^ (n~k1) *mynum-kv"-k-] m.y.yv) +xy (1-xu)(1-yv)-xy 0 (1. (1.6).\/) = Fo(xfy.i/) Fi (x. we have o 7 <1-xu)(1-yV)-xy _ fo (xy r v- 7 (1_xujk+l(1_yvjk+1 °° .1 x . m. or more compactly (1 .xu)(1 -yv) .U vmx/nnm-k-1 + z9 \ k ) \ k ) x y u (n~k1) n-k-1 n 1 m nltn„m-k {m . Solving this system of equations we get p x(l .n.n.v) [FEB. (1.u.{„m 'k )x y um-kv t.10) _ ^ .-r-n-s) (m-r = n-s±1) (otherwise). It follows from (1.n.n.yv) . by (1.u.v) +xFi(x.y.11) f(m.v) + yvF i(x.1\ in .s) = < Cr~1){ns~1) 0 (.2).n-k-1 k v (1.10) that f2(--')("7?) (1.u.n.v) + Fi(xfy.s)xmyW £ .u. (1.ufv).y.5) that Fo(x.9) that F= E { C V ) (nk)xmynum'k-1vn-k+[mk) m. .yfufv) = y +y Fo(x.y.r.u.xy Therefore.y.xu)(1 .uey) = x +xuFo(x.4) and (1. .n) m n • Z * v m.6) F = F(x.

It is clear from (1.r.3) 1or-1). if j +k = n j . forx = y. First. For example. 2.4) we must have (2.1) F(x. (2. To evaluate f(n. in agreement with the worked examples. If m (orn) = ^7 clearly f(m. The enumerant tfVfl./: in (2.n.v) = 2x^~*-kU:vL (1 -xu)(1-xv)-x2 Put (2.1.1 .k.11) gives f(3.u.s) = £ f(j.s)\Ne make use of (1.2.r. We now examine several special cases.2. if (2.s) . (2.n > 0.11). The simplicity of this result suggests that one may be able to find a direct combinatorial proof.2.D.r.10) we must have f j +k = n [ j-k = r-s. for example.s + 11 and we get (2.9) becomes (2.2. Similarly. .7) J j+k = n I j-k = r-s + 1 we must have j n = r +s + 1 (2. Finally.2) that we need consider are those satisfying f j+k=n J / .r.r+s. if (2.1) = (32) []) f(4.11).5) ) n ^r + s (mod 2) n > \r-s\ .5) is satisfied it follows that (2. (1.s) is the number of zero-one sequences of length n with r occurrences of (00) and s occurrences of (11).1} provided at least one numerant is non-negative. exceptm = n = 0. Thus.k = r-s./ ) ) provided at least one of the numerators is non-negative.Q) = 2 (27) ( M f(4.11) that the only values of/./?.2) f(n.s) = ^ M n + r - s + 1 ) - 1 ) {%<n .s) = 0 unless r (ors) = 0.6) fifties) = 2 { * ( n + r - s ) - 1 ^ ( n .k = r-s + (0. We may now state the following Theorem 1.1977] FIBONACCI NOTES 51 This holds for all m.r./.l. j+k=n so that f(n.r. If (2.r + s).^ is evaluated by (1.8) { (mod 2) n > | r .9) Hn. (1.1) = 4 = 3 = 0.x.

/ A We get the following results: provided f m + + nn = t (mod 2) t > \m. We may state Theorem 2. (2.7) provided m m+ + n = t + 1 (mod 2) \ tn > \m .13) f(n.12) and (2.u) = 2x±2x^2££ (1 —xu) .4).x. 3. ~ xn T ^ uf.s) . 3.u.t) = f(m.t) is the number of zero-one sequences with m zeros.r. 1 o r . (1.n. The enumeranttf/wJ defined by (2.xy Define (3.x Thus F(x.9) and (3.r.n.10) We may state Theorem g(m. The enumerant^m^fJt defined by (3.u.1) F(x.2) ]T g(m. Forx = y.1\ (3. As in the previous case we need only consider (33) ( [6 6} r+s I m-n ' = r-s = t + (0. (I .t) . FIBONACCI NOTES /o 11 \ [ n = r + s + 1 (mod 2) U 'ni and we get { (2.13).n + 1\ (3.n.u) = ?-±y-±2*l-Jxy_ujL . (3.n.n.A(m X+\+ V ) (y. 4.) provided m + n = t+ 1 (mod 2) t > \m . n ones and t occurrences of either (00) or (11). u = v.7). (2.xu)(i — yu) .u. In all other cases (2.x.y.(-mn+-n lt-1.9) reduces to (4.9) in all other cases (3. r+s=t so that g(m. (3.n .s) = 0. We next take u = v in (19) so that (3.1 -1%<n+r-s-1)-1){%(n-r + *+1)-1) # provided at least one numerant is non-negative. ( " 7 ') v f ' .52 [FEB.9).( .u) = --2fiL^?M^m (1-x(u+1))(1-x(u-1)) = f* J-x(u+V = 2 V ~ ^ + J)"-1 = 2 T. is evaluated by (3.1) F(x.6).T2) f(nfrtS} = n > r-s .10).t) = 0.n\ (3.2) is evaluated by (2.2).5) 1 <3-6> S(m.

-.2) h(n.a2. si = 2.2.t: (4.uto) = ^±y-t2^---mi The right-hand side of (5.1) F(x.t) = 2 [n~ (4. where the summation is over non-negative /-#.4) (a1. We may state Theorem 4. Thus.2) is evaluated by (4. rj= 0.1.t) = 2^.an-t). we understand (0000) or (1111).k \ _ ( t . •••.x(y +u) .4) in the followingway.1977] FIBONACCI NOTES 53 Hence if we put (4. where each a.3) 1 (0 < t < n). 7. for the sequence.5) h(n. This result can be proved by a combinatorial argument in the following way.t) can be described as the number of zero-one sequences of length n with t occurrences of either (00) or (11).+ rk+s1 <4 6) " •• st + '-' + Sk = t + -+sk = n-k (k = O. j+k=n r+s=t it follows that h(n.1\ \ Si > 0 J " ^ 1 Sj > 0 1~ \ k ) ' w it follows from (4. 1 or*.S3= 7. For t = O there is nothing to prove so we assume t > 0. Since {ro+ri+"7ron-k-t} rg + r-j + •-• + /> = n .is equal to 0. Consecutive zeros and ones are ruled out. (1.t) defined by (4. where thex's denote doublets of the same kind.1) (5. also if 0 is followed byx.3). Assume it begins with 0 or (00). ) The enumerant h(n. t = 4.. then* stands for (00).-t . The enumeranth(n. By the subsequence (xxx)f for example.s). while if 1 is followed by x.1) is equal to 7 .7. r2 = 0. r-\ = 2.+ sk = t .k .)(V)-^(^7-/)^(. Then we have a subsequence (0101 •••) of length rg. Sk such that ( r0 + rj + .+ sk = t\ _ jsj + . (010(111)01(00X11)) we have ro = 3.) k=1 ' k=0 5.6) that ^-^£C*t)(tr)-^E(B-^_.5) and (4. For 1/ = Q.y. /> and positive57.t # and = c-kn n { s7 + .t) = £ f(JXr. thenx stands for (11). this is followed by a subsequence of length r-j which is either of the type (0101 •••) or (1010 •••) depending on the*. followed by a subsequence (xx •••) of length s-j. Hence (4. S2= 1.9) becomes 5. •••. Thus we are enumerating sequences of length n . and so on. Let the symbol x denote any doublet -either (00) or (11).. r/. Thus we can describe the sequence (4.

0> = .r y A: m. ( ? ) / " " V = E C" 7 ' ) x V ^ V ^ E ( ? )x x r=0 m.Q) = } . Similarly E E f(^.r.r.^ J ^ j 1-x-x = £ n=1 F n+2*n > where Fn+2 is a Fibonacci number in the usual notation.n ™-" Hence /r? £ tf^/^fl . If we take /y = / in (5. 1 ) xmym~rur.1)xmym-rur-Y. we get (5.) (m-n = r) (m-n = r±1).u.0) . m.r m.0) r=0 is the number of zero-one sequences with m zeros.r .J ] f(m.r.4) f( J^ £ '>*>r><» = Fn+2 • j+k=n r=0 Clearly m £ f(m.r u „mym-r+1ur +2Y. (5.o> j+k=n r=0 is the number of zero-one sequences of length n with (11) forbidden.54 FIBONACCI NOTES (x+y+2xy-xyu) £ xm £ AH^O [FEB.1.n. Thus (5. _ n M y .n.2) F(x.0)xmynur .A- A77.(m.2) reduces to = (2+x) £ ^ n=1 F(x. m=0 A?=0 " m.3) and (5._.')A". forx=y.n.1). 6.y.x.3) Finally. Put .( m ^ / ) - (5.2) is equal to oo m fr^^E^E(I)A-E((ra.yflO)=^±^-.4) are familiar results.r it follows that .'H.)^:/))A"-Er.^-1i)*n m. 1-x(y+l) The RHS of (5. It follows from (5.E ("71)xmym'r'1ur+ E (m7 ' ^ v ^ V ^ E C". n ones and doublets (11) forbidden. (2(m~ ) f(mfn.r Since F(x. /?7.3) that J (5.

7) For u = v = 1.n=o = = (1 -xu-yv-(1- uv)xy) F m n m^(u/v)x y £ . we get (6.v) m n = Y.v) = m. Comparing coefficients.11) we can evaluate Fmn(u.uv)xy uvjxv It follows that x+y + (2-u-v)xy Fmn(u. namely m — 1 \ (n — 1 \ . {mnZ1s){n-1)um~n+s+1vs (m + n > 2).y m. (6.XU Hence .v) + vFmtn-i(u..u. ' (n > 0).VFIQ(U.V) it follows that (6.v)xmYn .v) .3) ^±y-±!2_=u-Vlxy__ 1 -.V) Fh1(u.u).y.n(u.s s=0 n-1 (6g) +j.n (u.V) . s=0 a polynomial in u and v. Thus (1.(1 (1 -. taking y = 0.n.v) -• = 2-u-v.s)urvs . Since Fii(u.m-n+s.1977] FIBONACCI NOTES (6-1) F m.r.n=0 k m+n>Q SO that (6.1(ufv) It is evident from (6. E r=0 f(m.v) [n <m+n > °> explicitly. . = 2. £ F(x.3) becomes .8) FmjidD = (m+n) By mans of (1.v) = uFm-itn(u.xu xu -yv -yv -.— 1 .9) becomes (6.3) reduces to = 7 . .. J Fmt0(u.3) that (6.V)xmyr ' £ " _ m.4) Fmfn(u. s=0 s=0 For example.uv)Fm-1rn.UFQI(U.v) = Fnrm(v.5) Fmn(u. .v) m \ /cc ) E F m o(w)xm *-* m=0 = um~1 (m > 0) = vn~1 FQ^U. Also. for/7 = 7.v) + (1 . (6.n=0 so that (6-2) Fm.

The more general recurrences (6. Cowan. 277-279. Vol.] If the relations (10). 3.v) l)vn (n > 1)." The Fibonacci Quarterly.r) = A(n .r- l)+qnA(n. r) subject to the initial conditions g(n. + If we now make use of the summation of a geometric series. pp. Vol.l(u. pp. (10)."SIAM Review.Stanton and D. (11) and (12) are used.r- have been treated in [ 2 ] .FIBONACCI NOTES 56 = 2um'1 Fm. ') [m ~ '){?-. For m = n we get m—1 (6. upon employing the relations (9)."SIAM J.V) /T7-7 2 ( m . 12 (1972). (11) and (12). 1977 1)um (m > 1).12) A(n. D. 1. then (14) G = JLi— + _ £ * _ + -J*— 1-xRj = 1-xR2 1-xR3 Cl(1 xR2>(1 xRs>+ 2(1 ~ " ° ""xRl)(1 ~xR3>+ °3(1 ~xRl)(1""xR2> 1 -x(R1+R2 + R3) +x2(R1R2 + R?R3 + R2R3) x3R1R2R3 which. r=0 In connection with the recurrence (6.1. 433-446.r) and (6. so that = 2vn~1 + (n- Fhn(u.m(u. r > O). L. Carlitz. The generating function for the sequence Pr is defined by (13) fCJ(xR1)r J>7>r« £ r=0 r=0 G= C2(xR2)r+C3(xR3)r]. Vol. "Fibonacci Notes. it may be of interest to point out that Stanton and Cowan [3] have discussed the recurrence (6. it can be shown that the much simpler expressions for the constants in the explicit solution (2) are indeed given by equations (9).4). Zero-one Sequences and Fibonacci Numbers of Higher Order.10) = 2 £ Fm. 2. "Some ^-analogs of Certain Combinatorial Numbers. r) Afar) l)+pnA(n. 1-10.11) g(n + 1.r) = 1 (n > O.r- 1) + qrA(n- 1.r+ 1) = gfa r+ 1) + g(n + 1. AAAAAAA [Continued from page 45. 4 (1973). (uv)' + (u + v) £ r=0 ') ^ / . L Carlitz. "Note on a'Square' Functional Equation. 12 (1974). G. pp. on Math. r) + g(n.2x -x2 ******* +x3 - .0) = g(0.l.13) = A(n .r- 1) + qrA(n- 1. REFERENCES 1. Analysis. finally reduces to the simple equation (15) G = j—^^-r-Y 1 .v) + (m- FEB. R.

= 1 and N-1 (iii) n k for a '+J = ° " ' 1=0 *This enumeration of the Fibonacci numbers is shifted by one from that in [ 1 . j=0 Given a Fibonacci sequence {G. 3 ] . 1.2.ON THE N CANONICAL FIBONACCI REPRESENTATIONS OF ORDER AT ROBERTSILBER North CaroSinaState University.1. N .-} of order N and a positive integer/?. For each of these canonical representations there is determined a table WJ{i analogous to the tables studied in [ 1 .3. North Carolina 27607 SUMMARY Carlitz./} of order /V determined by the initial eonditionsf/v.i+k~^ N. Scoville and Hoggatt [ 1 . The first canonical representation parallels the usual Zeckendorff representation and the Nth canonical representation parallels what the aforementioned authors have called the second canonical representation. a canonical representation ofn by the sequence {G/ } is a su m in which (i) the summation extends over all positive indices / and all but a finite number of the ks are zero. 2] have investigated Fibonacci representations of higher order. . In the latter part of the paper we give a generating function for the canonical sequences (those which generate the canonical representations) and also give the extension of the elegant procedure in [1.* For each integer /r ^ 1. (ii) kjt 0=> k. N is called the sequence of Fibonacci numbers of order N. In this paper we introduce for each N > 2 a series of N distinct canonical Fibonacci representations of order N for each positive integer n which we call the first canonical through the Nth canonical representations.i = F F N..1 we define a Fibonacci sequence {FN •} of order /I/ by k-1 F N. It is interesting to note that no such table exists with analogous properties in terms of ordinary differences even for/1/ = 3. / = /. j=0 The particular Fibonacci sequence {F/y/} = {F/v./^ 7 " .i+j. As a result we obtain a remarkable theorem which states that for every 0 < k < N the table of Fibonacci differences of order k of the columns of W^ inherits the following characteristics (and more) from the table wfi: (1) Every entry of the table is a positive integer and every positive integer occurs exactly once as an entry in the table and. THE N CANONICAL REPRESENTATION OF ORDER N A sequence {Gj}T=i shall be called a Fibonacci sequence of order N (N > 2) iff ft N-1 2 J Gj+j = GJ+N for every / = 1. which is the table generated by the first canonical representation. 3] .-. (2) Every row and every column of the table is increasing. 2. ••-. 2. this shifting seems to be indicated by Theorem 1. ••-. 2. For 0 < k < N the tables WN are shown to be tables of Fibonacci differences of order k of the columns of Wjy. Raleigh. i = U. 3] for generating the tables W« and W°. 57 .

From 2G-. Let N be a fixed integer greater than one and let k be any integer between 0 and N . and so by the same reasoning Gm-N must be present unless it happens that m .2. Continue inductively. Proof. 3] called second canonical. Because G-j = 1 and because {G.-} is strictly increasing except possibly for 67 = ^ 2 a r | d (ii) if £ ^ / ^ / is any canonical representation by {Gj} and if the upper degree of representation is/?. M ' and this sum will be a canonical representation of/? by {Gj} . ik+u "> l'k+N-i De that set having first index /^ of smallest subscript k. then ^kjGj < Gp+i. let/^. This. . for all i.-} must be unbounded.i = FN'j1+1 for / = 1f2. At the other extreme. . The principal result of this section is the following theorem Theorem 1.ip.2. is well defined. Then every positive integer/? has one and only one canonical representation by {F^ • } of lower degree congruent to one of the integers {1. each term of the sequence//. Note that for k = 0 the theorem gives uniqueness of canonical representations by {FNJ} without restricting the lower degree of the representation. Let Gj be the term of {Gj} of largest index satisfying G.G. Of all sums determined by canonical representations by {Gj} of upper degree m let/7 be the largest. each of the numbers Gm. Then k+N-1 YJ G ij = G ik + 1 j=k which contradicts the choice o f / £ . If Gj £ n let Gj be the term of {Gj} of largest index satisfying G. N. a canonical representation of/? by {Gj} can be obtained from the following algorithm.k) modulo N.3. inclusive.1 and Gj+i < 2G. 12. > G.7.-} be a positive term Fibonacci sequence of order N having the property that k £ ' 0 / < Gk+1 for k = 1.+ i we must have / / > 12 > —>ip since the equality of any adjacent pair of these indices would contradict the choice of the one with smaller subscript. Gm-N+2 m u s t be present in the representation since otherwise its sum could be increased without altering its canonical properties or its upper degree. ••-..1 shall be called the k canonical Fibonacci representation of order N. Let {Gj} be a Fibonacci sequence of order N which is non-decreasing and satisfies Gj . canonical representations with respect to {FN~-) are required to have lower degree congruent to 1 modulo N. in which case 67 must be present if G2 is not and can be replaced by G2 without altering the sum. N . It then follows that . / ! / . Proof. The validity of (i) is clear as is that of (ii) for 1 </?</!/. The number Gm-^+i cannot be present. Lemma 1.} guaranteed by Theorem 1. 2.1 is accomplished with the aid of four lemmas. The proof of Theorem 1. If/? is represented canonically by {Gj}. Suppose (ii) holds for all/? < m for some m > N. < n. as well as/? itself. If there exist among /'// '2* '"* 'p sets of N consecutive integers.1. < n .58 ON THE N CANONICAL FIBONACCI REPRESENTATIONS OF ORDER N [FEB.2.1.N = 2 and G2= G1. after finitely many steps an index ip will be found such that P n = £ £/-.1. /=/ Then (i) {G. —. —. Let {G. the unique representation by {FN~t. Then for every positive integer /?. Gm-i. combined with the observation that F N. which we shall call exhaustion. For each k = 1. Lemma 1. The largest index / such that Ay f 0 is called the upper degree of the representation and the smallest index / such that k\f 0 is called the lower degree of the representation. —. ••• explains the connection of these representations by [FN~j } with the representations in [ 1 .

-} of order N. . so that none of these terms can be redundant. N. a term Gj shall be called redundant if G. •••. By definition F/v. Proof.2. and must contain all of the numbers Gm-f.N. Lemma 1. Otherwise by Lemma 1. Let N be an integer greater than one and let k be a nonnegative integer less than N. subtract Gm from both and apply the inductive hypothesis. The case p = 1 is vacuous.4. upon removal of the terms Gm-i. F^ N. 2.-} of upper degree m . We shall make use of the observation that a positive term Fibonacci sequence of order N can contain no redundant terms beyond the first N. •-.k. Gm-/\/+j since otherwise its value could be increased beyond that of Gm in contradiction to Lemma 1.N . 1=1 We note that { ^ / v / } n a s n o redundant terms.1 by Lemma 1.-. Gm-2.3.N. Proof for 2 < / < N follows by induction.i -11 ./ terms of the latter. Gm-2.l\l). Suppose some positive integer/7 has two distinct canonical representations by { £ / } . Since it is canonical it cannot contain the number Gm-/y. —. Then {G. Again by direct calculation one finds that N-k rk _ 0N-k 7 _ V * ck h z l N.7 which i = I 2./ = 2'~1 for / = 7. in which case the same is true of the lower degree of the other representation n = Gm . Suppose the lemma holds for all p < m and let/7 = m.2. FNN_k+2. in which case thesame is true of the canonical representation from which it was derived by the removal of the last/1/ .N-k+1 ~ ~ ~ JU hN.By direct calculation one now finds that F^JJ = 21'1 f o r / = 7.N is congruent to r modulo N.i i=1 . can be expressed as a sum of fewer than N terms of distinctsubscriptsfrom among {Gj.. / .-\ has a redundant term Grfor which one of the two canonical representations of n has lower degree congruent to r modulo N. so m n < Z Gi = Gm+1- i=m-N+1 Given a Fibonacci sequence {G. 2. If m < N then Gm is redundant and r = m. using the relation Fi\i[\i+j = 2FNfN+hl ~ FNJ-1.N+'I = 2I~2(2N+1 . which by the inductive hypothesis cannot sum to more than Gm-/\i+i. Gj+j.2 the representation of smaller upper degree can sum to at most Gm so the representation having upper degree m must consist of the single term Gm. Lemma 1. Proof is by induction on the maximum p of the upper degrees of the two representations of n. Then the redundant terms of {FNj} are precisely FNfN_k+1.2. and in fact /-/ FkNfi=12 FkN. By summation we obtain F/\/f/\/+j =2 proves for/ = 7 the formula FN. Therefore.1977] ON THE N CANONICAL FIBONACCI REPRESENTATIONS OF ORDER N 59 m n- Z * i=m-N+2 has a canonical representation by {G. where/ = max {/. . G.N. If both representations have upper degree m. If m > N the other representation must have upper degree m . Let {G/\ be a Fibonacci sequence of order N which satisfies the hypothesis of Lemma 1. N./ ^-/V-/-7 from the representation there results a canonical representation for n m-1 ~i=m-N+ 127 Gi = Gm~N with upper degree less than m . or else m . By the inductive hypothesis either the lower degree of this representation for 6/77-A/ is congruent to r modulo N.--i}. >~. ••-. Proof.i> > = N-k + 1.

3.2 and hence also of Lemma 1./+k-1-J2 F N. Then f o r / = / ! / .p p=1 and the induction is complete. f A / .i+1 < 2F N. By Lemma 1.. as above.i+p p=0 N.i Yl N.N-k+1 = 2F N.1 each positive integer has by exhaustion a canonical representation by [F£ .< F N.i+k-1~ Yl FN. For this case we describe a method for obtaining a canonical representation of the desired form which we shall call reduction.1. This representation fails to satisfy the condition imposed by the theorem on the lower degree only if it ha's lower degree of the form /77/I/ + p.k +j for some/such that 7 <j < k.2 < . Since/ < N we have E FNiP - p=i-(N-k+1) 2»-1 '£ = 2i-2_2HN-k+2. N F N. N .i+k-1-N P=1 = X ) FN.ON THE N CANONICAL FIBONACCI REPRESENTATIONS OF ORDER N 60 [FEB. Replace . Suppose the lemma is true for / " < / ! / .i+1 For k = 0 we know that FJ^N+1 = /-/ F YL N.k + 1. By the information contained in the statement and proof of Lemma 1.-.1.i-1 i+k-2 = FN.i+p-1 + F F N. which verifies the statement of the lemma for / = N .i i=1 = 2F N.i+k~ 2F F N. p=i-(N-k + 1) This gives 1-2 i-1 12 FNfp + FN.i+k-1-N P=1 i-2 = H FN.i-1 N. A / .k +jwe have/l/^.N-k.}. .. But k-1 F N.P p=i-1 N.N-k~ and that 1 F N./ < / + k <2N so that k-1 F N.i ^ ' = 1 > 2> '~>N+ *• It now follows by induction that { ^ / y / } satisfies the hypothesis of Lemma 1.i+k-1-N~ k-1 F = FN.i-1 ~ N.7 we have. _. Proof see that of Theorem 1. ? _ f / v ^/.i-1 ~ FN.k < p < /!/.i+p = p=0 N.N- Thus for each {FN •} we have 1 = FN.i < / < N. N .j = 2F N/r 1=1 / and for Ar = 1.j 1=1 = 2F^N- = F + F N.p p=i-(N-k+1) 7 > N.p~ = F S N.1 < F N.i+1 = for 2F N.N and F N.i+k-1-N by the inductive hypothesis. that F N.i-1 ~ N.i+j-1 = j=0 since i + k- i+k-2 2 p=i+k-N-1 F i-2 Yl FN.2.p p=1 + F F N. N-k+1 the latter following from / F N.4 we = 21'1 Ffrj for / - 1.2.i = F k-1 22 FN.N+1 = 12FN. and it is clear that {FN •} satisfies the hypothesis of Lemma 1.

Theorem \FN. until arriving at N+p-1 E F ki- i=p According to Lemma 1. YlkiGi differ and differ in such a way that the largest/for which k. For each N > 2 and for each nonnegative k < N any system of unique representations by is lexicographic. 1..2.4 we can now replace i=1 and the end result of all these replacements is seen to be a k + 1st canonical representation by j f / y f-} of lower degree one.mN+p by JT FJ^j i=(m-1)N+p and then replace (m-1)N+p-1 b F N.2. then m < n if and only if the representations and HkiG.(m-l)N+p y ^ i=(m-2)N+p F N.1977] ON THE N CANONICAL FIBONACCI REPRESENTATIONS OF ORDER N 61 mN+p.} of order /I/ and a system of canonical representations by { £ / } ..i i=1 .i differ and that kp = 0. This theorem is an extension of Lemma 1.i' and so on.1 F N. k'p= 1 and k.= k] for all / > p. Then remove E*/4. we shall say that the system is lexicographic if whenever 171 = and Ylki^i n = 2^/^/ are two canonical representations in the system.//=/ /=/ Since the upper degree of p J2k>FN.i) Proof.i T>k'iFN.4.-has/f/ = 0. k]= I Clearly this property implies uniqueness within the system (although it does not imply existence within the system or uniqueness outside of the system). so that it is sufficient to show that i>4./<E^.3 and 1. Given a Fibonacci sequence {£. Suppose that and HkiFN.-= E ^ / i>P i>p from both representations. The uniqueness of this representation comes immediately from Lemmas 1.-^ /:.

Suppose that m and n are positive integers having canonical representations within the system and that/7? < n. and therefore all canonical representations by \FN J) can be found by first applying exhaustion* and then (if the result is not k+ 1st canonical) reduction**.2 £ kiFN. Let N > 2 and / < k < N.62 ON THE N CANONICAL FIBONACCI REPRESENTATIONS OF ORDER N [FEB. k'jFN.2 the sum cannot exceed F < £ *.3.P Thus we have kF £ i NJ < £ i=1 k F 'i N. Then no positive integer has more than two distinct canonical representations by {?&./ • /=/ N. kF i N. kp = 7.r Kp By Lemma 1.7.i i<p-1 = 2. **Defined in the proof of Theorem 1. It suffices to prove that if a positive integer n has two distinct canonical representations by {Fj^ -\f then the one which is lexicographically inferior is k + 1st canonical and the other is given by exhaustion.i = F N. Proof. By the uniqueness of canonical representations within the system. kt = k]for all / >p.i > i=1 since if the two sums were equal one could replace £ kF = i N.>#. so that £ kiFN. by Lemma 1. k'p = 0 and k.i i>p and contradict the uniqueness assumption. Let kp = 0.1.p • . the representation £ kF i N. Theorem 1..i i<p-1 and thus also the representation £ *Defined in the statement of Lemma 1.i £ i>p kF i N.i i<p-1 = £ k'iFN.= /r7-for all / >p which gives/7? > n by the first half of the theorem.. the only way the theorem can fail is for these two representations to differ with kp = 1.1.}• A number has two distinct canonical representations by { ^ / v / } if and only if the representation given by exhaustion* is not k+ 1st canonical. is less than p. Let the canonical representations be ™=£kiFkNfi and n =^k)FkNi .i i<p If k < N . Let canonically with the first representation lexicographically inferior.

Corollary.1977] ON THE N CANONICAL FIBONACCI REPRESENTATIONS OF ORDER N 63 must be k + 1st canonical since otherwise F N. 2. If it were not. THE TABLES W$ AND FIBONACCI DIFFERENCES We now fix N > 2 and fix k such that 0 < k < N and consider the set of k + 1st canonical representations. It remains only to show that is given by exhaustion.1 the same remarks apply unless /r.P+i-V i = 2.1 + E i<p-1 kF i N. (2) Every row and every column of WJ^ is increasing.k and for any p. For k = N . and denote the (N . since if it did we would have p-1 N1 \ N. it would be lexicographically inferior to that representation of 2^ k F 'j N.3. 3 ] .1.i k+ 1st canonical. .F° ~. which cannot happen.Arrowed infinite matrix ((afc)) by W^.. If the k+ 1st canonical Fibonacci representation of order N of a 1 • is then ior afj it is Y. verified in the proof of Theorem 1.k let [af /}1LJ be the sequence generated by listing in increasing order those positive integers having k+ 1st canonical representations with lower degree congruent t o / modulo N. 2.N-k. and the first canonical representation for af-. Statement (4) makes use of the property F^ j+1 < 2FN jf i = 1. Each matrix wfc has the following properties: (1) Every entry of WJ^ is a positive integer and every positive integer occurs exactly once as an entry of WN./. Theorem 2.i k is also canonical and exceeds FNp.-.1. 2. in contradiction to Lemma 1. •-.tN f hF° N. Kk-F Z i^Nfi' i=2 p-2 V Kk'+1t. Scoville and Hoggatt [ 1 . have identical coefficient sequences [kp\ . 2. and (4) af+1J < 2a«jfor/ = 1. a?n <a!n if and only \\ain < ain.2^ Nfi i=1 FNp~1 ~~. . N-k1. Another useful corollary is the following.2.kpFkN. •••. For each / = 7.p-1 contradicting the uniqueness of the first canonical representation. (3) For k= 1. for any ij < N . = tf and k2 = k3 = •>•= k/\/= 1.q./!/. which by what has already been proven would make 2^ k F 'i N. The k + 1st canonical Fibonacci representation of order N for ar. •••. N .i which was given by exhaustion. The following theorem is an immediate consequence of the lexicographic property of the k+ 1st canonical representation. 2. (4^ and W°have been discussed by Carlitz.

. By Theorem 2.1.p+i+k-1~ 23 r=0 which gives F N. k = 0.7.N-1.. Proof. For each N > 2 and for each k= 1. by 3/\/) by 32. every column of W^ is the kth Fibonacci difference of its corresponding column in W^.1.b2.-. a2.p and that of afn is 23 kP FN.2. By Theorem 2. / 3N) k (ai.64 Corollary. ~ have identical coefficient sets. we define an (N . Proof. we see that a1 ~3 = n for every positive integer/?.1.2.2. But by statements (1) and (2) of the preceding corollary and the fact that WN~ has just one row.p+i-1 Given an N-tuple (aj. Let/7 be a positive integer. But k-1 Fr N. Then if the N th canonical representation of /7 is thek + 1stcanonical representation of af. the first canonical representations of a® n and the Nth canonical representation of . Thus if the Nth canonical representation of n is z2kPFN.. 2.p > the first canonical representation of a®n is 22kPFN..1 the k + 1st canonical representation of a7 n is Lv? N.p+i- " F N.p+i-1 > where is the k+ 1st canonical representation of a7 •. •••. -a2. ' = 1 > 2' '"' N ~ k- Theorem 2.1 we have = 3/J jLtkpFN.N-k.. ON THE N CANONICAL FIBONACCI REPRESENTATIONS OF ORDER N [FEB.p+i-1 N. . —. Thus the tables of kth Fibonacci differences of the columns of 1/l/f\l enjoy all of the properties listed in the first corollary to the preceding theorem. 3/y) and given an integer A ^ 1.n is Y*kPFN. a .P+i-P i = 1. N . 3/\/) = (bj. a2. b/y-k) with k-1 bj = dj+k ~ 2 3 a'+j' 1=0 Then we can prove the following theorem. N . —.AV-tuple called the kx Fibonacci difference of r(a-j..p and so by Theorem 2.p+i+r-1 .. ••-.

for each N > 2 and each k = 0.j 1 J.p+i+k-l p k F 2 r=0 p N. 41 46 1 3 4 6 8 I 9 11 12 14 2 5 7 10 15 18 20 23 f ^3) (29) 34 13 5 J 6 m ~T~! 15 & 52 8I 9 Fig.1 we show a portion of W% with its accompanying tables of Fibonacci differences.2 for Ordinary Differences (N = 3) Our next theorem gives the generalization of the procedure used in [1.S P N. I = 1.2. wherein d u p l i cations occur in the third and fifth.J Z^ •. . 2. 2. N .a?.p+i+r-1 p Again using Theorem 2. 3~ 8 6 12 23 5~ T~ 10 14 19 27 37 52 1 2 4 3 6 11 5 9 18 1 2 3 4 6 5 8 11 1 ~ 2 4 7 13 9~ 11 "~l3 I~l5~~ 16 18 31 17 21 25 29 35 33 41 48 56 60 I 68 131 64 79 93 108 116 14 8 16 31 10 20 12 23 38 45 27 52 7 14 9 17 10 20 2 | 3 | 4 | 5 6 7 8 25 20 39 75 145 15 29 56 17 33 63 12 13 15 23 25 * 28 31 10 11 9 19 36 70 16 I Fig. 2.7.1977] ON THE N CANONICAL FIBONACCI REPRESENTATIONS OF ORDER N 65 k-1 a k U ~ iL F . 1+a k 1.j+r r=Q which is the /'.2 suffice to determine the table of wft if it is also required that the rows of W^j be increasing sequences forming a disjoint partition of the positive integers. fourth and seventh and fifth and ninth columns (as far as the table goes). a i+1.1 we obtain k-1 i. 3] to define M^and Wj_.N-k- 7. If one tries the same thing for ordinary differences for N = 3.1 A Portion of fly^ and Accompanying Fibonacci Difference Tables 1 3 5 7 8 10 12 14 2 6 9 13 19 e> 26 4 11 fl6) (?..n Figure 2.j+k "I.2 Counter-Example to Theorem 2.2.3.-. the result is shown in Fig.-Z'l d i.-. One can see that the properties of the Fibonacci differences given in Theorem 2./' entry of the table of kr Fibonacci differences of the columns of W i 7vo .2. Theorem 2.

— . By Theorem 1.k and therefore / F 12 N. We note that the information in this theorem is sufficient for the construction of WN in the sense of [ 1 . 7.] . unless W^ has already been constructed.pl.4.p be mN + i + 1.qN+r *-*» P N.1 as follows. Theorem 2.3the/l/ f/7 canonical representation now follows by reduction. however.p = F N.j. Our last theorem provides a generating function for thesequences [F/y 7 -}._ dues 0.p 2-J 2-J N.J N+i F N. 2lkPFN.1.f V + + Kk h F ~ 2u N. Let aj+1 H k P FN.p-1Since F^J+1 = F%p\ oral I p. is not Nth canonical because it has lower degree congruent t o / + 7 modulo /I/. Then by the nature of the reduction process we know that the N canonical representation of a/\is given by / m-1 N+i N 1 N1 N 1 V V hF ' Y hF ~ +.p is a canonical representation for af'• by {FN~-} which.kPFN.1 we have that the k + 1st canonical representation of a 0 is 1a / 12 p=1 m-1 ' i. Thus if 1 is added to the k + 1st canonical representation of ak 0 the terms produced by the reduction process exactly recombine to yield the expression ta i.p '" which therefore has lower degree congruent t o / + 7 modulo N.p- p and hence a U/ = 1+a 1aP. Then by Theorem 2.p • p=1 q=0 r=i+2 p>mN+i+l By the second corollary to Theorem 2. Let the lower degree of 12 kP FN.we must havei <N . but not for the construction of W^.qN+r+ q=0 r=i+2 12 k F P N. hi have k + 1st canonical representation Proof.p+12 12 FN.j 12 k r-k P'FN.ON THE N CANONICAL FIBONACCI REPRESENTATIONS OF ORDER N 66 [FEB./r.1 the first canonical representation for s/y is _ 0 L. Let a be a (positive) root of [Continued on page 34.i+1 ~ 1 P=1 by what has been shown in the proof of Theorem 1.p- p>mN+i+1 Now since 2^kPFN. / ! / . A representation for a*" # can be obtained through the second corollary to Theorem 2.p st is a k+ 1 canonical representation of lower degree congruent to / + 7 modulo N and with 7 + 7 among the resi. 3 ] .

.= 3 . as well as show other uses. Our first observation is that the set {(2k + 1)2n\k > 0.W. West Virginia 26506 Good [3] showed that (1) £ m=0 -±. Our object here will be to discuss this partition and generalize it.1)/Fk+2/F2k The numbers Ck arose in his proof that (1) generalizes to 67 for even k. Pictorially.GOULD West Virginia University. n > #}is identical to the set of all natural numbers. Holding either k or n fixed and letting the other variable assume all non-negative integers. of course. 2m n> 1.~ I I f((2k+l)2n) (2> n=1 k=0 n=0 for an arbitrary function /provided only that the series on the left converges absolutely so that it can be rearranged at will. but its use in forming interesting series rearrangements does not seem to be widely known or appreciated. Our main results are some series rearrangement formulas that are related to multi-sections but differ and do not seem to appear in the literature. The theorem is used by Greig [4] to obtain the transformation (3> £J n=1 £ {ck-1-). we find that the natural numbers are generated as the union of countably many disjoint subsets of the naturals. The rearrangement theorem is as follows: E f<n> . For a convergent series of positive terms. The bibliography at the end of this paper gives an idea of what has been done with such series and their extensions. Morgantown. 2n The problem of summing this for/? -> °° was posed by Millin [8].^ . every natural number appears once and only once in the array: 1 2 4 8 16 32 3 5 7 9 11 13 6 10 14 18 22 26 12 20 28 36 24 40 48 96 15 17 19 ••• This seems to be common knowledge in the mathematical community. the formula always holds. A common thread may be found among many of these studies: explicit or implicit use is made of an interesting partition of the natural numbers. k=1 /rodd a = 2 where (4) r (1 + Fk_1)/Fk Ck =\ ^(1 + Fk. for odd k.A REARRANGEMENT OF SERIES BASED ON A PARTITION OF THE NATURAL NUMBERS H.

so that the result follows since everything is then known by simple geometric sums.z .bn)/(a . On the other hand.bn. The two sets are easily put down in the arrays . according to Hoggatt and Bicknell [5. Going first to mod 3. 275. For instance. n > 0} and {(3k + 2)3n\k > 0..E n=0 X 1 -• x y(2k+1)2 L X n=0 n=0 x E (x k=0 2" > Z and this is substantially the way that many related results can be found.E E (-*)k -(2k+D2 n l g 2n n=1 n=1 k=0 but the presence of the factor (-1) prevents us from going further as (2) cannot be applied then. Perhaps some other result can be found using (6). X 3romwiich [ 1 . If we apply the same argument to the Lucas numbers.b the roots o f z 2 . recalling that Ln= an' -/. The formula 1 K2" E (7) ^ T n=0 1- X = --x—. To make this as clear as possible. n=0 provided only that the series on the left converges absolutely. then the Binet formula is Fn = (an . we find that <6> k E r~ .68 A REARRANGEMENT OFSERIES BASED ON k9n 1 (5) £ k2m m=0 1 = Ck. jl— [FEB.E E E = xn k=0 n=1 . Method X ] . and so a-b = Er ~b c n=1 = (a--b) 2 E E n=l E £* n=0 -1)2n k=0 -(2k+ 1)2" (a-b) a-(2k< -2k-1 k=0 k=0 and the double series can be summed by using (2). The two disjoint sets {(3k+1)3n\k > 0. either using (7) or going back to (2) again.b = y/5. 1 . 24] to Augustus De Morgan f X 1.b). so that ab = -1. the numbers Ck being introduced in the course of an inductive proof. we have: (8) £ f(n) = 2 n=1 k=0 E f((3k+l)3n)+J£ n=0 k=0 £ f((3k+2)3n).n > I k2n but he did not make explicit use of (2) in determining (5).x . we rephrase the argument as follows: With a.1 = 0. and a. p.7 . n > O) form an interesting partition of the natural numbers. k. We come now to the generalization of (2). Carlitz used what is essentially (2) to sum (1) when/7 -* «>. we may set down the hyperbolic trigonometric analogue of (1) which is done for/? -» °° in (22) below.^ .x' 2n+1 \x\ < 7. p.

. m-1 °° r mk+ihn (mk+i)m n E E z^ / = / k=0 n=0 °° \^ °° \^ . with f(n) = xn. provided that the series converge absolutely.* < __ xm j which proves the finite series result in [2]. We pause now to exhibit a neat application of (10) to derive a general formula found by Bruckman and Good [2] whose argument is tantamount to formula (10) but it was not explicitly stated. For the Riemann Zeta function we find m-1 oo Us> . n=1 r > E f(n>.A PARTITION OF THE NATURAL NUMBERS 1977] 1 3 4 12 9 27 81 36 108 324 1 21 63 10 30 90 13 39 16 48 8 24 72 11 14 33 42 99 17 51 69 and 2 6 5 15 18 54 162 45 135 405 The general case.E -1 . We have. This formula. V -/1-7W -_ E im m n=0 i=1 „r m-1 (11) k k=0 n r n+1 -1 E E ^ ^ n=0 xmll+1 1 vm' (m-1) > = E . mod m.L *"-£ vn n=0 m r+1 n n=0 SO that r 7 1-x m-1 n V Z V x y im " T. This allows us often to write a more elegant formula. Tutte [9] has given an interesting new theory of formal power series. n=l ns i=i k=o (mk+i)s n=0 msn *>>.E E — J — E ~L. of course. provided the series on the left converges absolutely. We should remark that when f(n) is replaced by f(ri)xn we may use (9) and its special cases as a theorem on formal power series and matters of convergence may be ignored when we use such a formula to equate coefficients in proving combinatorial formulae. > 2> r > O. Notice that the series on the right may be written in an alternative manner when f(0) is defined as then the first terms cancel out.£ n=1 m > 2. is the extension to values other than m =2 of De Morgan's formula (7) and in a finite setting. It is not difficult to see that multiples of powers of m may be removed from the set of natural numbers and we obtain the following nice result: (10) E E 1=1 k=0 f((mk+i)mn) E = £ n=0 = f(mr+1nj. is: « m-1 (9) £ fin) = n=1 £ £ £— «<mki)m + nl i=1 h=0 n==0 -o m > 2. f(n) .E f(mr+1nl n=0 n=0 m O. Formula (9) may be further generalized usefully. We pause to exhibit a non-Fibonacci application of (10). — ^ i=1 n=0 (1_xm )(1 n -.

1-i 1 where brackets denote the usual greatest integer function. It is the same result found by using ordinary multisection modulo m.i/m). n-\x —ff-. n > 1. n > 0.^ — . . Finite multisection in the form (15) has always been a favorite of the author. 1) = $(s). k=0 It is well known that there is an analogy between the formulas for Fibonacci-Lucas numbers and trigonometric functions. and it has two interesting further special cases worth setting down for reference: mn-1 (16) £ m-1 n-1 ffk) = J2 k=0 f mk Yl i=0 ( + i >' m > 1. a arbitrary. k=0 and mn (17) £ m n-1 E f(k) = J^ k=0 i=1 f(mk + i)t m > 1. n i2 = The same may be said for the hyperbolic functions: (20) sinh x = e~=-e-- . Ordinary multisection means the following formula: oo (14) oo m £ E f(n) = T. which is defined by f (s. so that £(s. We know that this is true because of the similarities between the Binet formulas (18) Ln = an + bn Fn=^LiAlf a— b and the Euler formulas „ix (19) -ix sinx = e—-=f—. To every formula involving Fibonacci and Lucas numbers there is a corresponding formula involving sines and cosines. or m-1 (ms- (13) DUs) = £ i=1 l(s. Since we are speaking of multisection. in terms of Hurwitz' generalized Zeta function. — * — .70 A REARRANGEMENT OF SERIES BASED ON [FEB. cosh x = ^ . £ k_ f(mk+i)t n-a+1 > m > 1 I' a+m. which simplifies to - m-1 (12) (/- -i) ms/ £ . -1. But formula (12) or (13) is not new. (mk+if k=0 Us) = Z i=i *> /.a) = ^2 / k=0 (k + a)s s > 1. it may be worthwhile to set down the formula corresponding to (14) for a finite series: (15) n m-1 f(k) = £ k=a i=0 £ Lm . Qix cosx = e . n=1 i=1 f(mk + i)t m>U k=0 the result again being valid for absolutely convergent series on the left.

pn~1 (p = any prime).d.c. A more complicated result follows with f= (p in (9) or (10). 1 Although (7) and its congeners are often listed in compendia of series. + L ^s which I have not seen stated elsewhere. .— 7 n=o sinni" * ex- x > 0. For instance. f ^ = FnLn to remind of the analogy. 1tsIf k=0 (2k If /n. Since we can also use ordinary multisection of series we have besides (26) £ ^i = £ aim + f mui . e. sinh 2x = 2s\n\\x coshx. (9). we find from (2). e n=o s i n h ^ ~1 When n -* °° in (5) the special case of Greig's formula is (23) E -rn=0 k2 = Ck-^j-1. we may apply (2). n k >I Z Ck being given by (4). (25) T ^ n=1 n = T s n-0 T ^i2k + 1)())<2n> =T s ns k=0 (2k+1) 2 T (t)(2k + 1)(t)<2n > + T $(2k + 1) $ ns n=1 k=0 oo s k=o(2k+1) (2k+1) 2 oo _i__ V URttJl + T MZLU1 n=l n+l -n+1 2nS~ =0 k~ s > 2. (mk + i. and which I do not believe is immediately obvious. 1 T — = --?— = 1163953414 . and similar formulas to (25) and (27) may be found for other multiplicative functions.^~1 /n(2k . not even (22). The case/7 -»°° of (1) was (21) X ! TT~ = —.1 k=1 a A: odd .1. Since g. (9). upon comparing (25) and (26) we get the unusual formula (27) f —L. „=1 s n n=i (2nf k=o s >2 (2k+lf whence. mn) = 1 for all / < / < m . I am not aware of any ready listing for them written in the hyperbolic form (24). (8)... Possibilities exist for application to number theoretic functions. may be applied with success to the many generalizations of the Fibonacci-Lucas sequence that have been studied. and the hyperbolic analogue is 7 £ (24) —.381966012 . Besides these applications it is clear that the general formulas we have given.1 which was found in Greig's paper [4] by an entirely analogous procedure. (10). (2). (8).. It is hoped . relations like sin 2x =2sinx cosx.1977] A PARTITION OF THE NATURAL NUMBERS 71 and we merely cite. + .±M2k±lL Bj:mit n-1 2ns~n+1 k=0 (2k+1)s n=l (2n)s s>2.r^ n n=0 2 and the hyperbolic sine analogue is (22) = 2.= — .g. We should note that (27) is exactly analogous to the formula n=1 a . (10) to multiplicative number theoretic functions as well as completely multiplicative functions. To get these results we used <j)(pn) = pn . It is natural then to set down trigonometric analogues of formulas we have discussed above. using Euler's 0-function.

"Additions to the Summation of Reciprocal Fibonacci and Lucas Series.. 7. 12. and Marjorie Bicknell. No. V. Good. 14 (1976). T. "Sums of Fibonacci Reciprocals.//?/'^. Good. B. we may take a very large but convenient r and expect the remaining infinite series to converge very rapidly. "On Elementary Calculus and the Good Formula. using Greig's formulas. 402-404. 6. 8. pp. 1977 that our remarks may shed some light on the nature of the formula (1) and its analogues and why Others fail to exist. 97-137. 453-455. or 100 we could sum the first part and the remaining infinite series needs only a few terms to get a good approximation. 2nd Ed. p. 287." The Fibonacci Quarterly. Vol. " A Generalization of a Series of De Morgan.A REARRANGEMENT OF SERIES BASED ON A PARTITION OF THE NATURAL NUMBERS 72 FEB. "Corrigendum. No. W.. Bromwich. 1976)." The Fibonacci Quarterly. D. 46-48. 1971). we get „% F n n% [ F 2nH F4n+2 a I £ F^ For r = 10. 4 (Oct. pp. Vol. Shannon. £ k=0 n=0 n=1 and when we can sum the double series. V. "A Reciprocal Series of Fibonacci Numbers. Austral. 14 (1976). 309. T. Vol. Vol. W. No. The method must have been used before." ibid. what can be said about (22) with sine instead of hyperbolic sine? A final observation is that our formulas sometimes give transformed series that are very rapidly convergent. Jr.I'A. E. Brady. 5 (Dec. G. 346. No.. Thus. 1 (Feb.9. REFERENCES 1. Thus (10) gives oo (29) oo £ r-i f(n) = £ n=1 ff(2k+1)2n)+^f(2rn). 4." The Fibonacci Quarterly." The Fibonacci Quarterly. 15." Part XV: Variations on Summing a Series of Reciprocals of Fibonacci Numbers. An Introduction to the Theory of Infinite Series. Vol. 1977). 1974). Hoggatt.."/M/„ 19 (1975). 13 (1975). 18 (1975). pp. Tutte. Vol. 10. 20. Jr. 3 (Oct. No. Math. Greig. pp. "On the Transcendency of a Special Class of Functional Equations. E. 272-276. 1926. 4 (Dec. Theory. J. 5. P. " A Reciprocal Series of Fibonacci Numbers with Subscripts2"/:/' The Fibonacci Quarterly. Hoggatt." The Fibonacci Quarterly. for the Fibonacci case. Ser. 14." J. 1976). No. I suppose this is an old trick but I am not able to cite a reference. p. I.. pp. For example. 3. 2. with Applications of Fibonacci Type. 1976). Problem H-237.. ******* . Macmillan. 389-410. Vol.. 12.J. "Corrigendum. J. p. 3 (Oct. 1974). Comb. 186-187. "A Primer for the Fibonacci Numbers. 2. pp. Solution by A. 412. A. pp. Wray G. 3 (Oct." Bull. 14. Kurt Mahler. No. S. 9. and Marjorie Bicknell. 193-196. London. E. The Fibonacci Quarterly. pp. Bruckman and I. 14. Soc. 477-478. No. Millin.

y'-1+xyn-2 + xy-1 £ Fky"-k+y'1 2 y"-1+xy-1 Hence. Ff(x) = 1. FIBONACCI AND LUCAS POLYNOMIALS AS COEFFICIENTS The Fibonacci polynomials Fn(x) are defined by [3] (5) Fn(x) =xFn-1(x) + Fn-2M with FQ(X) = 0.F„_ 7 My . respectively. Lucas and generalized Fibonacci Polynomials.. Montreal. and then finally to extend these results to r-bonacci polynomials. The purpose of this article is to generalize these results to sums of the form XFk(x)yn . 1 lxFk-i<x) + 3 .Hk(x)yn~ . Fk„2(x)lyn-k (y2-xy- {S-FnM} " f Fkyn~k 1 + y'2 {S.SWAMY Sir George Williams University.Tny - y y2-y- o Tn^ i and (4) £ Tk2n~k-1 = T2(2")-Tn+2* 0 where the generalized Fibonacci numbers Tn are defined by Tn = Tn-i + Tn. Lk(x) and Hk(x) are. T2 = b.O.2t Tj = a. Now consider the sum S= E Fk(x)yn'k = yn-1+xyn-2+Y. Canada 1.Fn(x)} 1)S = yn+1-yFn+1(x)73 Fn(x). .A FORMULA FOR £ Fk (x)yn"k AND ITS GENERALIZATION 1 TOr-BONACCI POLYNOMIALS M. T. 2. Fibonacci. HLk(x)yn~ .S. where Fk(xj. INTRODUCTION Some years ago.N. King [2] generalized these results to obT yn-i<-l = (T0y + T. P. Recently.1)yn . Carlitz [1] had asked the readers to show that (D E Fk2n~k~1 = 2n-Fn+2 0 and n-1 (2) E Lk2n'k~1 =3(2n)-Ln+2f • 0 where Fn and Ln are the/7 tain the expressions: (3) Fibonacci and Lucas numbers.

Hi(x)= 7. ^ By letting x = 1.F1(x)yn . = "£.^±^^"^1 _ . y = 2 \n results in (7) and (9). GENERALIZED FIBONACCI POLYNOMIALS AS COEFFICIENTS Let us define the generalized Fibonacci polynomials Hn(x) as (13) Hn(x) = xHn. It is obvious that the polynomials Fn(x) are obtained by lettingHQ(X) = 0.Fn+3 = 2". The Lucas polynomials /.^ - ^ ^ ^ ^ . 1 Lk(x)yn-« = £ 1 Fk+1(x)yn-k+Y.A FORMULA FOR £ ' Fk(x)yn~k 74 AND [FEB. Further. using (7) = ^ *^ ~ ^ ( ^ 2 ^ * FnM) . . Fk(x)yn+1'k 2 Fk-lW" ^ Fk(x)y n-1-k 0 . 1 + Ln^M + Fn.^ W are defined by [3] (8) Ln(x) = xLn-jM with LQ(X) = 2. It may be shown by induction or otherwise that Ln(x) = Fn+1(x) Hence. while the Lucas polynomials Ln(x) are obtained by letting HQ(X) = 2 and Hf(x)= 1.y) + Fn-1M} Therefore (9) £ ^fr^^ .F3- Lk2n~k = 2n+2-Ln+3 Fn+3 1 £ = 2"-L3-Ln+3 1 which are the results of Carlitz [ 1 ] . = f ^ . by letting A. we obtain (10) £ and (11) Fk2n'k = 2n+1 . Y.{Fn+lM Gi(x.= y = 2 in (7) we get n (12) £ Pk2n'k = Pn+2-2n+1 = Pn+2-2n. <?. it can be established that Hn(x) is related to Fn(x) by the relation Hence.1(x) + Hn.y) = yn+1 S - £ f ^ " " * -yFn+1(x)-Fn(x).1(x).Z Fk(x)yn~1'k . 3. 1 where Pn is the nth Pell number. L-j(x) = x.W * 0 . Hn(x) = H1 (x)Fn (x) + H0 (x)Fn-1 (x).£ W K " ^ .j .2M with Hg(x) and Hj(x) arbitrary.P2 .».fr. In fact. / Letting (6) we may write S as (7) Gn(x.

M = 0.» + F(nr)(x) . F<2r>(x) = xr~1. 4. From (14) it is seen that for these generalized Fibonacci numbers (15) E Hky^ - ^HlfylzJ^y^n 1 y 2 .lHn+1(x) {Hn+1(x)-Hn(x)} + Hn(x)-H1(x)-HQ(x)l.1977] ITS GENERALIZATION TO r-BONACCI POLYNOMIALS Hk(x)yn'k Z = H7(x) Fk(x)yn~k Z 1 + H0(x) G . Let us now consider / .»1M 1 0^ +H0M Z FkV"-l~k.(x) = Rk (18) we have.£ F^Mvn~k • Denoting for the sake of convenience F(kr. + {H7(x)-H0(x)}] . r-BONACCI POLYNOMIALS AS COEFFICIENTS The /--bonacci polynomials Fp(x) r Fi !r-2)M have been defined by Hoggatt and Bicknell [5] as = ••• = Fl%) = F(0r. 1 It should be noted that by letting x = 7. we generate the generalized Fibonacci numbers Hn defined earlier by Horadam [ 4 ] .xn'k = Hn+2(x)-xnH2(x).y . using (7) 1 (x.Hn+3 = n 2 -H3-Hn+3 1 which are the results obtained by King [ 2 ] . F<r>(x) = 1. / = Riyn~1 +£ (xr~7Rk^ +xr-1R1yn-2 r 2 +x ~ Rk-2 + txl"1R2 + -+Rk„r)yn-k + xr'2Fli)yn'3+-+(x'"1Rr-i+x'"2Rr_2 = R1Y"-1 +xr-1f1 [Riyn~1 + xr-2y-2[R1yn-1 +.+ xR1)yn-r Rr-iyn~r+11 Rr-2yn-"2l .y) + H0(x)yn~yHn+1(x)-Hn(x) 2 1 y -xy .i and < ? n Z (16) n k Hk2 ~ = (2b+a)2 n .1 Some special cases of interest obtainable from (14) are. Z 1 1 n Z i=k-l(x)yn~k Z 1 75 (-Dk+1Hk(x) = t f(-Vn+1 H kM = 1. HQ(X) = a and H j(x) = b in (13). and F(nrlr(x) (17) = xr~1 F(nrir^(x) +xr-2F(nrlr_2(x) +.Gn <X>V) + H0MGnGi(x.y) The right-hand side may be simplified to show that = Z (14) H Hk(x)yn~k = Hl(x)y"+]^ y (x. n n Z Hk(x).+ +R2Vn~2 + -+ + R2yn-2 + .

+xr-2y-2 E r *&"* r-1 + .(x) - ylxF^M + yr~2 . We have from (20) that E 4r)-2"-k .(x ~ Rn -•••-y(xRn .+ (xy) + till + (xy) " Rkyn'k-- £ R kY°-k E 1 1 2 -(xyriRn-(Xyr2 E - n-r+2 Rkyn~k JT Rkyn'k • n-r+1 Thus. + ^ j +2^3[F(r) + .76 Fk(x)yn~k AND A FORMULA FOR Z / + xy-('-1>[Riy°-1] +xr-\-1 Y.... lyr = Riyn+r~1 + (xyr1 Rky"-k £ + (xyr2 1 1 n-r+1 n-r Rkvn~k + i: Rkyn'k E + -+xv 1 Riyn+r-1 = + Hxyf- 1 .(1.^ .. Let us now see if we can obtain for the r-bonacci numbers [5]. (x^F^M F<nrl1 -+Rn. + fjrj^j + .yr~2(xr-2Rn r 3 +xr~3Rn-i -+Rn-r+2) +x ~ Rn^ + -+Rn-r+3) + Rn-iS-Rn • r 4 -y ..r+1) + r 3 Denoting now G<nr)(x.2) / - Now we have from (19).y) = yn+r-1 + ^ ^ F ^ M + ••• +F(nr]r+2(x)J + .+ F<nrlr+3(x)] (x)J - Flnr)(x) we have / = £ F<r>(x)-yn-k ILLJM} (20) G\r)(x.. l\yr-E (xyA . + .^ (2D G(jr. Rky"-k [FEB.. a result corresponding to (10) for Fibonacci numbers.F^My'-1 (19) .yr-3[xr~3F(r>(x)+xr-4F<fl..y) 1 ' The above result for r-bonacci polynomials may be considered as a generalization of the result (7) for Fibonacci polynomials.R1yn+r-1-yr-1(xr-1Rn+xr-2Rn„1 .(xy) + . + xy-(r-1) "'£' Rkyn-k+y~rj: 2 Rkyn'k . 2n+r-1 _ G(r)(w = 2^-2^ + 2IFM + 2r-2[F<r) + F^1+FM + = . it may be noted that the r-bonacci numbers F'nr' we obtained by lettingx= 7 in (17). 1 Hence.

1. p. I968. Vol. 5. + FJll^l F^]+F^ + 2[F<„'> + F^J + F? . -1-+ tFn+1j 1 = rFn+r+1 Hence. for the r-bonacci numbers. Math. M. Continuing the process. F. No. No. Vol.f- 2 F^ 3 . 68. 11. 1973. 5. King. and M.] + 2^ [F<n% + + - + 2[F<nr> Fjfj. ******* . E. pp." The Fibonacci Quarterly.." The Fibonacci Quarterly. Hoggatt. 90. y +2^1F<« + - + FlfJ^J + + + 77 F&J + F^ 2^[F<n'> +. Problem B-I35. 3." Amer." The Fibonacci Quarterly. " A Polynomial with Generalized Fibonacci Coefficients. 6. 2. No. Vol..W. "Generalized Fibonacci Polynomials. + F<r> + ••• + F^+3] -+2[F^ + .1-F (22) n+r+1 Also r-1 G\r)(l2) = 2r- 2k -E 0 (23) = = 2'2'-1-2 Therefore from (21). Jr. Vol. 11.. 1961. Bicknell and V. pp. 455-459. Monthly. 3. Vol. Jr. (22) and (23) we get n Z r(r) 9n-k k 'z r _ " 9n+r-1 z p(r) ~ tn+r+1 _ ~ 9n p(r) z mt r+1 r r(r) n+r+1 • 1 The above result may be considered as a generalization. 11. "Roots of Fibonacci Polynomials. G(nr>(1. Bicknell. V. pp.f~ lF^+2 +2r-2fF<%] + f-3[F^ + •••+ F<nrlr+3] 3 + F<?>. L Carlitz. 457-465. " A Generalized Fibonacci Sequence. E. 1973.2-3[F<n% + F<n% + f W . 527-532. Horadam. 1973. pp.zj?) = z?r-rfF Lrn+r-i-+. the above may be reduced as n+r 1 2 z ' (r) (r) (r) _ u rJnrUla. 271-274.19771 ITS GENERALIZATION TO /--BONACCI POLYNOMIALS ... of the result of Carlitz [1] for the Fibonacci numbers.2) =nri+r. REFERENCES 1. 4. Hoggatt. A.. No. 5. The Fibonacci Quarterly. B.

by (1. 97] k-1 (1-3) £ [ x+'L]= [kx] .5). .HM + 1Mk .SOME SUMS CONTAINING THE GREATEST INTEGER FUNCTION L.1): k-1 (1.1) I = [kx] +1/2(h- ^Supported in part by NSF Grant GP-37924.1) . that k-1 Z r=n k-1 [x+h{)=H . For a later purpose we shall require the following extension of (1.2). North Carolina 27706 1. [f r=1 and (1. Indeed r i r=i r=1 \ s ' k-1 = (h . put (1. (1. Durham. r=0 To prove (1. k-1 ] T 0 ( * + j-^j = kx +1/2(k.3). Then clearly (1. 78 1)(k- V.[kx] (1.1)(k .V . p.4) <t>(x) = x .1) follows immediately.1). r=1 where (h.[x] . the fractional part of x. using (1. CARLJTZ* Duke University.6) . It is well known (see for example [4. \X+T-^{X+T)) r=n = * kx + y2h(k-1)-<t>(kx)-y2(k.k) = 1.5) <j)(x + 1) = (j)(x) and. p. r=0 It follows.2) reduces to the familiar result [4. 97] that an i: [f] - M-tM-t).2) £ [x + Jjf]= [kx] + V2(h .v(k -v-Y. r=0 For /7 = 7. Let [x] denote the greatest integer less than or equal to the real number*.

k)+s(k.1) Sn(h.2). To prove (2.k .h . v2 r=0 r=0 A=0 it follows that k-1 r 6kS2(h.l)2(k- where of course (h.5). Incidentally. (1.« «-Jji(Hr]-?) 2.k) satisfies (1.k) + 4h(h .l)(8hk . r=0 r=0 r= 1 Since.k) = 1. by (1.v- r hr" ^ .8) is equivalent to the reciprocity theorem for Dedekind sums [3.1)S3(k.1)(k ..& A . (h2 [j}~ ~ 1)(k 1)(k ~ ~ 1>(2k ~ 1>- It is known [3.7). It is not difficult to show that a similar result holds for S3(h.k) = 12h Y.h) = (h .h) = (h . 5=7 where (h.'f977 SOME SUMS CONTAINING THE GREATEST INTEGER FUNCTION 79 The writer has recently proved [2] the following result: h k-1 (1.(h2 . This formula can be proved rapidly in the following way.7) follows at once.k) + 6hS2(k.k) = 1. take 1)2(2hk-h-k+1). 9] that k-1 (1.S l f ] -L[f]' We have E *2 (f) .k) = Y [ j ] n (" = °> 1<2<~)' r=1 Thus S-j(h. p.'*».1) while S2(h.1)(k . Put o k-1 hr r k-1 .h) = -1-+ -± ( £ + + + k ' (1.[ * ] )2 . s=1 Thus 6kS2(h. Define k-1 (2.«-? S ^ ? ] • * * « .8) h-1 s 12h £ r\j]+1MH r=1 '" [f] = (h ~ D(8hk-h-k-l). p.1) .1)(2k .2) 4k(k- W3(h. 4 ] : s(h.FEB.k).7) 6k ^ E r=1 r h n 2 ~1 r ks -+6h * *H[j] ^ [j] " " " = (h-Wh-1)(k-1)(2k-1).1) -(k2- 1)(h-1)(2h-l) and (1. We shall prove that (2.k) is evaluated by (1.9) 4 where 12 \ k hk h ) k-1 <i. .£ ( * .

4) 2S2n+1(h.(2hk.1)(k .k) + 3(h . (-Vj ( 2n + 1 ) (h .1)2(2hk .k). (2.1)+ 10(h .I^Sjfrk) + 6(h . we get w.1)S3(h.k) + (h.k) = 3(h . by (1.--E 7 (/>-/-[f.1)S2(h. This proves (2. (~DJ ( F2 2n W ) ft " 1)2n~HSi(h.k) = (h. j=0 which reduces to 2n-1 (2. (2. k).10(h .k) .Z[»-?]3r=1 [FEB. for n = 2.4) becomes 2S5{h.k).5) becomes .k).k) + 5S4(h.1)2(2k -1)-(h- 1)2(k .k)+S4(h.h)} .V2S2(h.1)(2k.h . forA7 =3.S3(h. we have 2/7-/-7 S2n+i(h.7.| (h .k) .1)3(k . ( h.k) = -Mn- 1)(h-l)2n+1(k- D + Y.1)5(k .l)2(2h -1)(k .1)2(2hk .£ [ h ^ ] 3 .3) Thus 4k(k- . £ ("-'-[if r=1 ' r=1 = (h . 2n S2n(h.1)3(k -1) =0 in agreement with (2.k)-(h so that.2).1)(h .1)(k.[tjffj" (-tJi('?)(h-triSi(h. by (1.1) .l)S2(h.1)2k.1).3(h .k) + 6hS2(k.l)kS2(h.l)3S2(h.k).1)4(k -V- 4(h .k) = .1)2n-JSj(h. = £ In particular.k) .k) .h) = (h .l)2S3(h.3). For example.l)S2(h.1)2n-1(k -D+Y.k)= E (-1)' r~o (^/^(h-D^-^SjdiM so that 2n (2.I^S^h.(h .1)2n~i+1 Sj(h. = 0.k)\ = (h.k) = (h .k) = t .4(h - l)S3(h.h .1) .k) + 4h(h .6(h .1)2(k~ 1)2{(2h.l)3(k .k) .k) = £ (-Vn ( f ) (h .l)S3(k.1Mh .1){6kS2(h. so that.r ]) 4 = (h.k) . Generally.5) -(n .k). S„(k. Similarly. for arbitrary positive n. which reduces to 4S3(h. If we apply the same method to S4(h.h .1).7). (2. 4k(k .1)2(k . 2S3(h.1)3(k. while.SOME SUMS CONTAINING THE GREATEST INTEGER FUNCTION 80 w .1)2(k1)2(2hk-h-k+1). VS3(htk) = 6(h .k) .

r) [ ^ J 2 %k(k-D<h-U -2(h-1)Y./(f) •£(¥-&})'•%(? r=0 r=0 ' r=1 r=1 r=1 k-1 E r=1 r=1 Now put k-1 (3-D j hm.k) .TlS12(h.k).3) £ 1)(2k .1)(k- 1)(8hk-h-k-1).k) .V - Similarly .20(h + 15(h .r=1 E '•''[x] .k>. sj = s a / .1)3S2(h. k-1 4k .k) + 12kSh 1 (k.k).1977] SOME SUMS CONTAINING THE GREATEST INTEGER FUNCTION -2(h . satisfies a relation similar to (1.i (h.1)Sh 7 (h. (k~r.DS4(h.k) + kS2(h.k(k .k) = 1/2kS2(hfk) + (h .k).D<k~ 1>k • .k)-6S5(h. 3.7 (h .k)- (3.1)3(k- which is the same as (2.1)S2(h. \ k-1 S2.k) - Sh2(h. However we are k f->z (*)'[*] **£'* m2 i.2) for every n > 2.1)(h . St seems plausible \\\z\$n(h.-[j\) k-1 +_ £ k-1 2 (k r) h 1 2 (k .k) -kSt1 (h.1)5(k -1)+ 15(h . Consider the sum 1)-3(h .[ .h) = (h .= / i ~ 1){h j f o . By (1.k) = 0.1)2 r=1 + 2(h .4) Sh2(h.k) = E (k-r)2lh-1-[!£]\=-lk(k-1)(2k-1)(h-1)r=1 = 1 \ -k(k- 6 so that (3.k) unable to prove this.k) + j? 9 1 k(k 1)<2k ~ ^ " ^ [ T ] . k-1 Si.l(h.k) 81 1)2S3(h.8) 12hSh7(h.k) = 0.7) and (2. k2 In the next place.3).1)2 .l)S1f1(h.1 )(h - 1)2. Combining these two formulas we get 1 Mh .%k(k .1)(h .2) k Jj-(k .k)=Y.1)(k -l)k2 2 + 2kS 1 7 )h. Since k-1 k-1 r=0 r=0 r 3 k-1 ±?(%)'±y( r)'±(i) 'if'-'* r=0 we get S3(h.k) . ' 1 S2.2(h.S2 7 (h.l)2(h3 .1) = 0.1)(h .[^] . = Mfk .k) + 2S3(h.1) .k)+3\s2i(h.. = . so that (3.

i(k.1 (k.V(2k. (4. .1)(8hk-h2 + k(k.1)- 1 1)(2k- 1)+ -2<2h1h2-h1-h2)(k- 1 -hih2(k- 1)2+ \h22(k- l)(2k- V2.1)(k.5).k) similar to (3.h2k(k- 1)+ J. However.J . we have 1 .s=0 so that (4. in view of (1. .1 (h.1)(2k- l)h (h- k. comparing (3.4). (4.h22k(k- 1)(2k. r 2 [ k-1 = k) r. .jjrffk- l)(2k- 1) = tfi2(k- V(2k.1>(k~ 1>} = j*>ik(k- D(2k.5). On the other hand.s=0 ris=0 R3 = 1)(2k- R(h1fh2. k) = £ [ hll^i ] ((h lh2f k)= 1)m r.4) R2 = 1. 1 ' [ / T ] =\h2k(k- 1)(2k.V+ \ (h2.1).1) 1)k2(k- 1). it does not seem likely that S]2(h.(k- .1)2.1)k(k. r.2) l W ^ s ) . (4.8).1)2.6h2k2(h 1) + h(h.i(h.1)m.(h.1)(k.1).1). In the next place.x(±!^-\hL^]pJ. we get 12h2S2.2).h) = hk(h . 1)+ lh7h2k2(k- 1)2+ ^h22k2(k- 1)(2k.1)(k. . Simplifying.(2h 1h2-h1- 1)(2k- h2)k(k- I)2 + 1. by (1.h) = hk(h .1).4) with (1.k).2). We consider next the double sum k— 1 (4.k) (3.s=0 r.s=0 ' + /i3r k where k-1 Clearly (4.3) R1 = I 1 -h2k2(k- (hir + h2s)2 Rl = L r.7) and (1.s=0 We have (4. r.V + 1) R(hhh2.1)(6hk-2h -2k.k) + 12k2S2. by (4.Rl_lfi2 } r..3).1) satisfies any relation 2 R(h1f h2.^ f f t .v.k).£V(f-') = * £V(£) = *z (L)2= t(k.1)+ J.82 SOME SUMS CONTAINING THE GREATEST INTEGER FUNCTION Thus (3.s=0 t=0 t=*0 Hence. 4.^ ) .1)(2h- [FEB. r=0 Similarly k-1 2 .s=0 r. k-1 . <«> £ V ( .3) yields 12h2S2.5).5) + 12k2S2.1)k(k.s=0 £ r{hir + 1 <h 2 2.

]Th7 )) +R(hhh2.h2. t=0 Then. fj = R2 = r.s.. ((hlh2h3rk)= res.2%.1)(2k.s.. t=0 2/h7r+h2s+h3t k k-1 k-1 k-1 2 )> .\ lh2+h3- 1)k2(k- V2.r[hlr + h2s+l(h3-V(k-1)} r.1)(2k.h3.k)= l(h* + h%)(k. exactly as above k-1 y $2 \hir + h2s + h3t^ 1 Y" ( h1r + h2S + h3t = \h1r + h2s + h3tVY r..h3.1)(2ko 1 1)+ -(h1h2-h1 1)2+ 1-(k.1. put (4. \k(k-1)2Y.1)(2k- .k). t=0 r=0 r=0 . ^{2j2h7h2-J^h1 Thus £ ^{hilJ^lM.lk(k-1)(2k- VJ2h2+ jk(k- D^h^+ V2 ( < ? £ / ? 7h2.k(k. t=0 R3 = rh^r +h2s +h3t i I J' ~k R(h1fh2.s.k).6) R(hhh2. t=0 R 4 I 1~ TR2+R3 where ] T (hjr + h2s+h3t)2. r.10) [1.) = i k ( k _ 1)(2k_ vZh2+ i M-. t=0 -2 (4.h3.s.7) R(h1.s. t=0 J^ (h 1r + h2s + h3t) r. where the sums on the right denote symmetric functions.hih2+1- k(k- k) D2J^hi R(h1fh2. It follows that R2 = \k2(k- (4.. Clearly Ri = \k3(k- (4. On the other hand k-1 r..1)... By (1.1977] SOME SUMS CONTAIWIWG THE GREATEST INTEGER FUNCTION 83 Simplifying.h2)(k- z Next.s=0 = Ujk2(k- V(2k- D.8) D^h2+1-k3(k- 1)(2k- 1)2Y* hi/12 .9) 1)(2k- 1)J^h2+ )k2(k-V2. we get (4.2).E ^ \( " k^ ) ) =k2 *-< E r^ K(£) L^ ^ k J -^E' r =6 \ k(k.s.k)= Z [h-lr-+Jl2LtM] V.t=0 \ r.s. t=0 k T.s. E = r\^^f^) r.h3. o . r.

1) 6k £ h 1 [h-f] r=1 2 ~ 2 +6h J^ [jf] k-1 (5.p .11) R(h1.s=0 ( 5 . For the convenience of the reader.4) 3 [kj] +4h(h-1) £ r=1 (5..3) 1)(2k-l). 521-530.h2. 5) £ [lllLtAlLt!l^]2^m." Revue Roumaine de Mathematiques Pures et Appliquees.a) = ] £ rs [££*££] r." The Mathematical Association of America.b) = I The proof of (4. k-1 (5. s=1 2 =1-E(h21+ h2)(k-1H2k-1)+1-(h1h2-h1-h2)(k-1)2+l-(k-1)(2k-1) .l)(k- 1)(6hk. (5. 5.]m.])2{j:h^ 2A/)_ r.Vj^h^ . pp.lhLk(k. in (5.14) S(1.2b .k) = 1. Vol. L Carlitz.a) = 4a4b- 10a2b + a4 + 8a2b . "Some Sums Containing the Greatest Integer Function.b. Summary. Heaslet. where (a. 2. 42 (1974). (b7h2b3.10) gives (4. (hjh2.2k + 1).4). 4. Hans Rademacher and Emil Grosswald/'Dedekind Sums.c. we restate the main results proved above.1). 1939. pp.SOME SUMS CONTAINING THE GREATEST INTEGER FUNCTION 84 FEB.hk(h.k)= ((hlh2. J. V. New York.2ab-a2 .u{j:h2. . McGraw-Hill.14) is rather complicated.] . L Carlitz. The writer has proved [1] that the sunn a-1 (4. 1972.s=1 satisfies (4. for any n.1)(2h.l) (k .-. 41-52." Abbandlungen aus dem Mathematischen Seminar der Universitat Hamburg. "Inversions and Generalized Dedekind Sums. • r.k)= I REFERENCES 1. 3.. (5.12) 2 £ [ h L l l ^ R ^ n=o m1. in (5.1) (2hk -h-k+1) k-1 12h2 £ k— 1 ^ [ — . It will now be clear how to evaluate k—1 (4. Uspensky and M. 1977 Comparison with (4.l)(k- 4k(k-l) "£ h-1 3 [|£] 2 s=1 2 = (h . A.lk(k- V2J^h1 V2^2hih2 + | k(k - - tk(k-1)(2k-1). Elementary Number Theory. Vol.s.h3.13) S(b.3) it is assumed that ft.2) = (h.hnfk)= j. h-1 r2 [ ^ ] + 12k2 J2 s2 [ ^ ] > r=1 (5. s=1 .k) = \k(k- V(2k .5).W= 7. t=0 In (5.. 20 (1975).2).hn.

Our contribution consists of making available a description and proof of the method which is self-contained. At the present time. so is never allowed a "pass. a calculation which resembles the very aspect of the present method which we might hope to eliminate. In the case of binary representations. 5. In Bouton's Nim it is possible to determine correct play solely on the basis of the cardinalities of the constituent piles by way of their binary representations. In his turn. Generating formulas have been found for these pairs. As in Bouton's Nim. in fact. Worth Carolina 27607 Our aim in what follows is to show how Fibonacci representations play a role in determining winning moves for Wythoffs Nim [ 1 . but for the convenience of the reader. the method we are about to describe is an explicit consequence of the material in [4] and [5]. a succession of divisions provides the needed representations. Raleigh. On each turn a player may move as in Bouton's Nim. a process which. there is a residual disparity.. As has been stated." All counters picked up in one turn must come from a single pile. we briefly recount the rules of play.5). provided he picks up an equal number of counters from the two piles. but they involve computations with irrational quantities. only two piles are involved.2). As has been known since Wythoffs original paper [ 1 ] . An analogous approach to playing Bouton's Nim would require the inductive generation of safe configurations up to an appropriately advanced stage. The particulars of these two games can be found in the preceding references. this method places the two games on an equal footing with respect to the computation of play. Ostensibly. The existence and/or determination of an analogous algorithm for the determination of Fibonacci representations is therefore a related question of interest in what follows. related to the Fibonacci representations. This will remove the necessity for a good deal of background which might otherwise exclude this information from many to whom this same information will be comprehensible in its present form. the traditional play of Wythoffs Nim requires having or computing the table of "safe" pairs to an appropriately advanced position. 2] very analogous to the role of binary representations in Bouton's Nim [ 3 ] . Recent researches into Fibonacci representations [4. the winner is the player who makes the last move. QP. but has the added option of picking up from both piles. and on the same level as Bouton's treatment [1]. In Wythoffs Nim. a player must pick up at least one counter. An immediate and explicit by-product of these investigations is a method of computing moves for Wythoffs Nim using Fibonacci representations. the strategy for Wythoffs Nim consists of always leaving the opponent one of a sequence of pairs (1. . eliminating the necessity of computing tables of powers of 2. 7 and more] have coincidentally turned up the safe pairs for Wythoffs Nim as being quite fundamental to the analysis of the Fibonacci number system. The player who makes the last move (picking up the last counter) is declared the winner. ( 4 . 7 ) . although altogether well-defined and straightforward. The number of counters picked up is constrained above by the size of the selected pile. In contrast. which are defined inductively. but is otherwise an open choice on each move. (3. In Bouton's Nim (usually referred to simply as nim) two players alternate picking up from a given collection of piles of counters (such as stones or coins). 6. although the selection of pile can be changed from turn to turn. Thus a certain inequity is seen to exist with regard to the computation of play for the two games. no other information is necessary.WYTHOFF'SNIM AND FIBONACCI REPRESENTATIONS ROBERT SiLBER Worth Carolina State University. the determination of Fibonacci representations requires a listing or calculation of Fibonacci numbers to an appropriately advanced stage. is quite complex in that case.

a < b. Repeat for the residue n . F3 = 2 and so forth.0) a safe pair. A Fibonacci representation consists of a finite sequence of zeroes and ones. then m < n iff the left-most position in which the canonical Fibonacci representations of m and n differ contains a one in the representation of n and a zero in that of m. driving the residue to zero.d) by a legal move. provided he continues to play correctly. Theorem 2. Obviously every positive integer is either an A-number or a B-number. Notice that if (a. Theorem 1. the right shift of the canonical representation of a yields the second canonical representation of b — a. the left shift of that representation is the representation obtained by adjoining a zero at the right end. Thus 100000 denotes F6 =8. Lemma 1. We now prove some lemmas which will considerably expedite the proofs of the theorems. The second canonical form turns out to exist and to be unique for each positive integer and is the representation resulting from the following algorithm: given n. A Fibonacci representation is said to be second canonical if it satisfies condition (i) of the canonical representation and in addition (ii)' the right-most one of the representation is in an odd-numbered position. (12. then there exists a safe pair (a.0). read positionally from right to left. Let n be a positive integer which is represented canonically (not necessarily second canonically) in Fibonacci.F^ and continue to repeat.1010100). The canonical and second canonical Fibonacci representations are lexicographic. The right shiftIs the representation obtained by deleting the digit at the right end.1 and then add 1 "in Fibonacci. A one in \\\eitfl position denotes the presence of the Ith Fibonacci number F. If the reader is willing to accept the assertions we have made concerning the canonical and second canonical Fibonacci representations. F2 = 1. Thus.b) by a legal move is unsafe. since in canonical Fibonaccithisis written (101010. We shall prove the following two theorems. it is clear that the winning strategy for Wythoffs Nim consists of always leaving one's opponent a safe pair. If m and n are positive integers.b) is a safe pair other than (0. If (c. We begin with some definitions. The number represented is determined by summing the Fibonacci numbers whose presence is indicated by a one. the F.d) is an unsafe pair." meaning that any pair of adjacent ones is rounded up to a single one in the position immediately to the left of the pair. The same is also true of the second canonical Fibonacci representations. . then every pair (c. where we adopt the convention Fl = 1. We introduce one last bit of terminology for convenience. The canonical form turns out to exist and to be unique for each positive integer and is the representation resulting from the following algorithm: given n. If (a. for example. otherwise we call n a B-number. The pair (a.b) is any safe pair different from (0. Let a and b be positive integers.86 WYTHOFF'S NIM AND FIBONACCI REPRESENT ATI ONS [FEB. not used correspond to zeroes in the canonical representation. but never both. determine the largest F^ < n and place a 1 in position k.b) derivable from (c.) Of course. What this amounts to is this: just as in the case of conventional bases. which means the following. Clearly uniqueness is not generally assured. one can determine the larger of two numbers "at sight" by comparing their representations.0). A pair which is not safe is unsafe. We shall call n an A-number if the right-most one in the representation of n occurs in an even-numbered position. we can produce an otherwise complete account of the winning strategy for Wythoffs Nim. Given any Fibonacci representation. determine the canonical representation of n .b) will be called a safe pair if the following two conditions are satisfied: (i) the smaller numbers is an /l-number and (ii) the canonical Fibonacci representation of b is equal to that of a with a zero adjoined at the right end. Since (0. 100000 is the canonical representation and 10101 is the second canonical representation.0) is a safe pair. In the case of 8. 20) is a safe pair. (Be sure to denote the number 1 by F2 instead of Fu if it is needed.b) is a safe pair. We begin by reviewing some basic facts about the Fibonacci number system. For convenience we also agree to consider (0. 11000 denotes f 5 + F4 = 8 and 10101 denotes F5 + F3 + Fx= 8. b must be a ^-number.-. A Fibonacci representation is said to be canonical if it satisfies two conditions: (i) the representation contains no adjacent ones and (ii) Fx is not present (although F2 may be) meaning that there is not a one in the first position. Thus the first player who can establish a safe pair is the winner.d) which is derived from (a. If (a.

If the smaller number of the pair is an /l-number. By Lemma 1. Therefore. then an =ATand bn = a~n = n~. Each positive integer belongs to exactly one safe pair. If a ? b we assume a < b. reduce b to a~. If we put together the proof of Theorem 2 and the statement of Lemma 2. Since no Anumber can be a ^-number. It follows that there can be no more than one safe pair (an. Proof of Theorem 1. 2. Represents and b canonically. Corollary. bnl By the corollary. Let/7? = b .denote the left shift. Thus we need only show that the right shift of a represents b . By the definition of a safe pair. If (a. Let (a. there is exactly one safe pair (an. Now let <. If m < n then am < an and bm < bn . 0„ Given a pair (a.an. since by definition. so m < n and (a. Proof.*a) = (an.b) cannot be safe because of Lemma 2. reduce/? to ~a. represents and b canonically in Fibonacci.1977] WYTHOFF'S NIM AND FIBONACCI REPRESENTATIONS 87 Proof. it is clear that b = ~b+T .a <a~. Proof. if a and a are represented canonically. Clearly the right shift of any /l-number is in second canonical form. reduce the larger to that quantity represented by the right shift of the canonical Fibonacci representation of the smaller. Fibonacci representations are lexicographic. By the rules of the game. Lemma 2. (am. If both an and bn are reduced to obtain a pair (a. If a is an /l-number and b < a~.an = n. given any positive integer nf set an = ATand bn =/7~with n in second canonical form to obtain a safe pair (an. am <an = a. and since b is a ^-number in canonical form. An equal reduction in b necessarily produces bm.b) and (a'. the resultant pairstill contains bn so cannot be safe by the preceding corollary.a = b''. proceed to 1. bn) with bn .a. If a = b. both ?and a' are the second canonical representation of /7.0). the pair can be reduced to the safe pair (0. If neither 0. determine the second canonical Fibonacci representation of the positive difference of the members of the pair. Each one in the representation of b is equivalent to a pair of ones in the two immediately adjacent positions on its right.b). 3. so that (a.a = n for a given n. 1. Lemma 2 has the following corollary. But the latter is unique and thus the canonical representations of the /l-numberss and a'are identical.a. which is the desired conclusion. Suppose that (a. then b .a'' = n. with an and bn in canonical form. From the preceding remark. and if in addition the larger number of the pair exceeds that quantity represented by the left shift of the canonical representation of the smaller. Proof of Theorem 2. reduce it to the latter quantity. a legal move must either reduce an or bn alone or reduce both an and bn by the same amount. concede (if you think your opponent knows what he is doing). (This is simply a way of saying that Fn = Fn--j + Fn-2 if /7 >3. b is the canonical representation of a. If n is rep- resented in second canonical form.) Suppose we denote the right shift operation by ->. bm) is the unique safe pair with bm~ 3m= m. If a is an /4-number and b > a~. bn) be a safe pair. Let (an.a = bn . likewise \ox bn.a. bn) with b . A left shift on this representation will produce an /l-number and another . Otherwise.a > 0 and n = a~. so that a can be reduced to am.b) is a safe pair. Corollary. we arrive at the following specific algorithm for playing Wythoff s Nim. 1 nor 2 holds. Clearly the left shift of any second canonical representation is the canonical representation of an /l-number. then b .b') be safe pairs with b .an = n. bn) such that bn .b) is an unsafe pair.b). the position of each one is 3 or greater. If the smaller number of the pair is a ^-number. If an alone is reduced. If a is a ^-number. For each positive integer n. this corollary yields another. so this equation is the same as b = a +a.

335-345. pp. 10. 4. 11." The Fibonacci Quarterly. Suppose a=9. pp. 11. A. Borosh. pp. Suppose a = 10. A Game with a Complete Mathematical Theory.b = 20. 1977 left shift will produce a ^-number. "Characterization of the Set of Values f(n)= [na]. which is the key to actual computation of play. 175f. a = 100100 and/? = 1000100. 10. This gives the canonical representation 101010 for as and 1010100 for/? 8 . Vol. We are indebted to Mr. A. " A Modification of the Game of Nim. References [11. 1963. The result is the safe pair (6. 380-384. Vol. 1 8 1 190. b= 15. In canonical form. 15.." 7. 4. 15). p. 1976). pp. 337-386. 1973. pp.S. R. J. n = 1." The Fibonacci Quarterly. 35-39. We therefore reduce /? to ^obtaining the safe pair (9. but more complicated than the second canonical approach herein. 12. pp. II.10). "Some Arithmetic Functions Related to Fibonacci Numbers. L. "Fibonacci Nim. Fraenkel. 13. I. pp. pp. Combinatorial Theory. Domoryad. 14. Hoggatt. Scoville and T. Martin Gardner who furnished the additional reference [10] upon reading a preprint of this manuscript. C. Wythoff. pp. 10. ******* . Vol. 527-530. b = 32. A. 1-28. L Bouton. L. Vaughn. Vol. Thus a is a ^-number (case 1) so we reduce/? t o ? = 10010. a = 10001000 and b = 10101000. "A Generalization of Wythoff's Game. Scoville and V. 1973. Mathematical Games and Pastimes. Jr. Yaglom and Yaglom. Vol. 1967. 1901-1902. 2. Vol. A. 2. FINAL NOTE. a/." NieuwArchiefvoorWiskunde. "Lucas Representations.b) by the reduction of a and b by an equal amount. A Series. No." The Fibonacci Quarterly. pp. " A Generalization of Wythoff's Game. 1972. 1. Connell. 5. R. Vol. In this connection reference [13] is usable.88 WYTHOFF'S NIM AND FIBONACCI REPRESENTATIONS FEB. Y. G. E. L Carlitz. The canonical representation of 7 is 10100 so the second canonical representation of 8 is 10101.. 7. 29-42. Vol. Fraenkel and I. We conclude by illustrating each of these cases. Levitt and M. Bulletin.a = 8. W. 20). " A Fibonacci Property of Wythoff Pairs. 1964. Challenging Mathematical Problems with Elementary Solutions. San Francisco. September. 7. 3. 62-65. 13] are cited in [10]. Vol. 8. 1972. the two of which constitute a safe pair obtainable from (a. However. 9. 2nd Series. Here a is an /^-number with t= 1000100. S.. N. R. Vol. Here a is an /1-number with a~= 100010000 > /? (case 3). Silber.-•••/' Discrete Mathematics. Then a= 100010 and b= 1010100. Then in canonical form. et. 10." Canadian Math. 9-13. 199-202. 1972." Ann a/sofMathematics. Carlitz. The connection with Fibonacci representations we have discussed is given (in very definite form) in [12] and is generalized in [10]." The Fibonacci Quarterly. REFERENCES 1. No. J." The Fibonacci Quarterly. Holden-Day. Suppose a = 24. 12.2. "Addendum to the Paper 'Fibonacci Representations'.." The Fibonacci Quarterly. "Fibonacci Representations. 4 (Nov. which is obtained by reducing both 24 and 32 by 12. Macmillan Co. L. Carlitz. M. in neither case is the connection with second canonical representations discussed. a/.. Carlitz. Whinihan. Hence we compute b . 1972. Shimshoni. 6. Vol. 2. "Nim. et. 1907. which is less than b by inspection (case 2). P. yielding the safe pair (12. 1959 pp.

Lamar University. Whitney. by arranging D in increasing order. WHITNEY Lock Haven State College.ADVANCED PROBLEMS AND SOLUTIONS Edited by RAYMOND E. Define the binary dual. b. North Carolina. Mathematics Department. Lock Haven. b < a + b. Lewisburg. Proposers should submit solutions or other information that will assist the editor. SOLUTIONS UNITY WITH FIBONACCI H-247 Proposed by G.in the expansion of (x + x + l)n. Form a sequence. c such that a < b +c. Wulczyn. Lock Haven. NOTE: The elements of D result from interchanging t and x in a binary number. Lock Haven. „a. Pennsylvania. Duke University. Sum the series s. a/ e {0. D. respectively..h.) Verify that an =bn-i + bn for each n = 1. Fr. there exist an infinite number of positive nonsquare integers. •••. defined by . Durham.1}. Pennsylvania. is the coef- H-270 Proposed by L.b^c (b+c- x y z a)!(c + a. B-271 Proposed by R.e L. Whitney. Show that for each Fibonacci number. Pennsylvania 17745. Beaumontf Texas The sequences {dn\ n=l ^nd [bn]n=g [n/3] . Let D denote the complement of D*.2. Find *With respect to the set of positive integers. Pennsylvania 17745 Send all communications concerning Advanced Problems and Solutions to Raymond E. (More precisely. H-269 Proposed by George Berzsenyi.a. n > 0 >. such that 89 . as follows: D =<t\t = \ \ (a/+ 21). Carlitz. This department especially welcomes problems believed to be new or extending old results. Lock Haven State College.c)! ' where the summation is over all non-negative a. c < a +b. D. Bucknell University.b)! (a + b . Lock Haven State College. To facilitate their consideration. ["fl [n/2] k=0 k=0 (n > 1) k=0 (n > 0) are obtained as diagonal sums from Pascal's triangle and from a similar triangular array of numbers formed by the coefficients of powers of A. [Sn}n=i. solutions should be submitted on separate signed sheets within two months after publication of the problems. [ " ficient o f x k in (x2 + x+ 1)n. a formula for5^ .

then there exists an integer A/such that V$-Yn-lYn+1 = N(-1)n +yn-2< an d if Vo a n d / / are inte- . Hence there are at least two infinite chains of Fibonacci numbers which are congruent to 1 modulo F'r thusmakingZTin (Dan integer. = 1. By use of the above identity for Fibonacci numbers and the well known relation Yn = Fn-1Y0+ FnYU we first establish that N = Y0+Y0Y1-Y2 • Indeed. •••} satisfies the equation yn =yn-i gers. 1. F24n+22> F24n+14. D in (1) is a non-square integer. F„ 89. or 6. Solution by the Proposer. = More generally. FJO 55.2Fn)YoYl = (y20+y0yi-y*)(-Dn • . St Lawrence University. Beaumont.(Fn-2Y0+ + = (F _1 . F24n+13.. •••. Prove this statement and show that N cannot be of the form 4k + 2. Example: Fr = F4 = 3 Congruence Table Modulo 9 112 88 7 3 5 8 4 3 7 1 8 0 6 4 1 5 6 2 8 1 0 Fr+S can be chosen as any one of the Fibonacci numbers F 24n + 1. Parker. THE VERY EXISTENCE H-248 Proposed by F. a congruence table modulo F2 of the Fibonacci numbers is not only periodic but each period starts 1. Yn'Yn-lYn+l = <Fn-lYO+ 2 FnY 1>2 . F24n+2. F22 = 17711 F23 = 28657. D. F 24n + 18. y-j. 4. Also. 1) = It is assumed as well known that every positive integer is a divisor of an infinite number of Fibonacci numbers. F13= 233.Fn„2Fn)Yo 2 (F - = -l-ir1Y20-(-1)nY21-l-ir1y0y1 Fn-1Y 1>(Fnyo Fn-jFn + jly'j+fFJj. and show that 4N terminates in 0. F 24n+23- = 7 = 336 = 880 = 6032 = 741895 = 34853280 = 91247072 Also solved by P.Since the difference of two positive squares is never one. Texas. F18 = 2584. if a sequence [yQ.90 ADVANCED PROBLEMS AND SOLUTIONS F2r+s-F2rD [FEB. D D D D D D D = = = 8. Solution by G. New York. F 24n+6r F 24n+1Qf F6 F 24n+11. F2+s- (1) 1 = (Fr+S+ 1)(Fr+s- F2D. A well known identity for the Fibonacci numbers is rf-Fn-lFn+l = ~(-1)n and a less well known identity for the Lucas numbers is L2n-Ln^Ln+1 5(-1)n. Bruckman. Lamar University. + Fn+lY 1> - Fn. Berzsenyi.

Parker. J.e.1977] ADVANCED PROBLEMS AND SOLUTIONS 91 It is easy to see that/I/ is odd unless both y^and j/7 are even. University of Illinois at Chicago. Thus the coefficients an satisfy a second-order difference equation which is both linear and homogeneous. A. Howell. Lawrence University. in which case it is a multiple of 4. n -z—J — £ • — . -f- 1 n X Xj-X2 Thus.+ anxn +-) we say that ag= 1. Chicago. the last digit of 4N is 0. we present both solutions. New York.: If .+ bkxk 1 + ox+fix2 Since two quite different solutions were offered by the Proposer and by P. where x-j and^2 are solutions to the equation / +ax+fix2 = 0. Shannon. J.d0 + d1x + .. U. then an takes the easier form of an = (1 + 3n)xr). 4N is not of the following forms and 4(5t + 2) = 20t + 8 = 10(2t) + 8 4(5t + 3) = 20t + 12 = 10(2t +D + 2. J.+ bkxk . we can evaluate the constants cj and C2 to get _ 2x2+x1 2X1+X2 n g X X2-X-.C. there exists no integer t such that y20 + Yoyi-Y21 vl + VoVi-V2 or = 5t + 2 = 5t + 3. Biggs. Since ag= 1 and 5/ = . B rue km an.4 or 6. Since we require the values for an for which 1 = (1 + ax + $x2)(ao + a1x + a2X2 + . I vie. Find an explicit formula for the coefficients of the Maclaurin series for b0 + b1x + . D. 1 or 4 (mod 5).. P.k). Furio Alberti. Peck. and an + aan-i+Pan-2 = 0. Thus it cannot be of the form 4k + 2. The following little-known determinant theorem was brought to my attention by Dr. C. and the Proposer.+ dn xn + . Bruckman. and fix2 I + ax + d n = u J2 1 n a n-i°i> i=0 where r= min (n. The general solution is = dp C -iX 1 1 CpX n . a-j = . Therefore.I.a . B. i. We first get a Maclaurin series for the reciprocal of 1 + ax + fix2. Also solved by A.a . Solution by P. Canton. Illinois. By a case-by-case examination of the congruences of yg and y 7 (mod 5) one can also establish that yo+VoYl-y2i = 0. St. Consequently. . Bruckman. FOLK-LACJRIN H-249 Proposed by F. Solution by the Proposer.C. we have b0 + b1x + . If the roots xj and X2 are equal.

Then A(n)(a2n+2~ 1) + B(n)(a2n+1 -a) = C(n)(a-$)an. Then = . This gives a^ = 0. Put k AoM = S a xJ j > ak ^ 0. (*) Let r be a fixed positive integer and a > 1 a fixed real number. then Cn = A0 B0 0 (-1)" A1 B1 B0 A2 B2 B1 An-l Bn-1 B n-2 An Bn B n-1 Bn Bi Rn+1 O O O (n+1)X(n+1) In our particular problem. = 0.= ArM A0(n) Proof.2. also. In the given equation A(n)Fn+1+B(n)Fn = C(n). . Assume that A0(n)am+A1(n)a(r~1)n + . We may assume that AQ(n) 4 0. n = 0.-).2. where the Aj(n) are polynomials in n. Solution by the Proposer. B-j = a. 1. 0= %(l-y/5). (-1)n bo 1 0 bi a 1 b2 $ a b3 0 $ 0 0 0 0 V 1 0 0 bn 0 0 1 a (n+1)X(n+1) GROWTH RATE H-250 Proposed by L Carlitz. Show that if AM = Fn+1+B(n)Fn = C(n) (n = 0. B(n). CL= W+yj5). where the Fn are the Fibonacci numbers and A(n). An = bn.(S)C(n)an . with bn = 0 if n > k. Bn~ 0\\ n > 3. Durham.92 ADVANCED PROBLEMS AND SOLUTIONS FEB. 1. thus proving the theorem. We shall prove the following more general result. ••• . BQ= 1. C(n) are polynomials. then = B(n) = CM AM = 0.(a. -). Therefore.+ Ar(n) = 0 (n = 0. 1. Duke University. 1977 B0 + B^x + B2x2 + and BQ ^ 0.(A(n) + aBM) = 0.] + aB(n))a2n . put Fn = ^ i p . Theorem. B2 = $. so that (a2AM Hence by the theorem [Continued on page 96. North Carolina. Divide both sides of (*) by n am and let n -+ °°.

. 2.S.1 = 0. New Mexico 87108. 709 Solano Dr. Also a and/? designate the roots (1 +sfs)/2 and (1 . FQ= 0. Find a closed form for n J2 an-k(ak+k) k=0 93 . •••. a2. — be the sequence 1. Albuquerque. LetP-j. Albuquerque. F7 = 1 and Ln+2 = Ln+1 + Ln.a is an integer for/7 = 0.. Establish a closed form for n £ F2kTn-k k=1 +Tn+1. Find the ratio of lengths QfQ2/PiP2B-349 Proposed by Richard M. Hoggatt. University of New Mexico.b b. Albuquerque. 1.be the intersection of segments/V7P/+3 and Pj+2Pj+4 (subscripts taken modulo 5). •••.c c. P5 be the vertices of a regular pentagon and let Q. unless otherwise specified. Hillman. respectively. Dover. Let an be as in B-349. Jr. Hoggatt. 2.E. Give a recursion formula for the an and express the generating function E n=0 anxn as a quotient of polynomials. B-348 Proposed by Sidney Kravitz..x . i. Solutions should be received within four months of the publication date. P. DEFINITIONS The Fibonacci numbers Fn and the Lucas numbers Ln satisfy Fn+2 = Fn+f + Fn. B-347 Proposed by Verner E. Li= 1. where Tk is the triangular number (k+22) = (k + 2)(k+1)/2. P.ELEMENTARY PROBLEMS AND SOLUTIONS Edited By A. California. LQ = 2. University of New Mexico. San Jose State University. San Jose. Albuquerque. Mew Mexico 87131 Send all communications regarding Elementary Problems and Solutions to Professor A. HILLMAN University of New Mexico. Each solution or problem should be on a separate sheet (or sheets). Show that a.. New Jersey. let 3.x2 .e. New Mexico. Let 30. Grass/. 1.\f5)/2. New Mexico. Let a. 3. PROBLEMS PROPOSED IN THIS ISSUE B-346 Proposed by Verner E. of* 2 -x -1 = 0.7 be the greatest integer in 7 + (n/2).. b. San Jose State University. 2. San Jose. California.. B-350 Proposed by Richard M. Grassl. and c be the roots of x3 . •••. a-j. 3. Jr. Preference will be given to those typed with double spacing in the format used below.

(-1)r(a2n tfn+2' + t. Texas. B-351 Proposed by George Berzsenyi. Hunter. A. Fun with Figures. J. The identity is a restatement of l-jg of Hoggatt's Fibonacci and Lucas Numbers with (k. This has two possible solutions: S = 7. VARIATIONS ON AN OLD THEME B-323 Proposed by J. Ralph Garfield. Freitag. SOLUTIONS FRONTPAGE ALPHAMETIC B 322 Proposed by Sidney Kravitz. Solve the following alphametic in which no 6 appears: A R K I N A L D E R S A L L E A L L A D I (All the names are taken from the front cover of the April.b) = ar~br a-b 2n+r a a-b -b2n+r FrF2n +r Also solved by Richard Blaze/. Dover. If S = 7. Lamar University. Carl F. Then S + 3 = L + 10. Brady. A. B. L = zero. Milsom. 1975 Fibonacci Quarterly. and the Proposer. Peck. so R = 8. Lamar University. Milsom. Frank Higgins. C. Beaumont Texas. John W. and K = 2.94 ELEMENTARY PROBLEMS AND SOLUTIONS [FEB.2"*2' _ (. Herta T. C. F2n+r . Canada. Beaumont. John W. Sahib Singh. Peck.»rb2n _ {_ [a2n+2r + b2n+2r-(ab)rb2n + b2n - 2(ab)n] J^nj (a-bj 2 1 - (ab)ra2n] 2 (a . Wray G. A = 1. Trigg. Graham Lord. San Diego. in the case (a) in which n is even and the case (b) in which n is odd. It may be proven directly by using the Binet-formulas: = i _ _ [a2n+2r (a-b)2 _L_ 2n+2r + b - 2(ab) n+r . Toronto. the subsequent values follow immediately. A. New Jersey. Moore.n+r). Gregory Wulczyn. D = 9. Prove that F4 = 3 is the only Fibonacci number that is a prime congruent to 3 modulo 4. Also solved by Richard Blaze/. California. . B. namely: N = 4. Thus the reconstructed addition is 18254 + 10938 + 71003 = 100195. whereupon R + 2 = 1'0. Bob Prielipp.1 = 5. L = 2. Ontario.n) replaced by (n. E = 3.) Solution by Charles W. Shallit. L = zero and S = 9. and the Proposer. H.(-1)rF2n = FrF2n+r Prove that • Solution by George Berzsenyi.

There does not exist a single-valued function G which satisfies the equation since if x = a. -2. Jr. For n = 1 and/7 =2 we obtain the equations 2 .Lr=0. J. Ralph Garfield. Universite Laval.3ands = 2. Oshkosh. Universite Laval. B.>j5)/2r Prove that there does not exist an even single-valued function G such that x + G(x2) = G(ax) + G(bx) on -a < x < a .1)(Lr. Roanoke. Frank Higgins. Thusr= 0. in fact. Let a = (1 + \Js)/2 and b = (1 .odd. reduced to (Lr. We now show that each pair is.4) = 0for/. J. since L2r = Lr . (Note that G need not be even. Moore. F3n+2 = F6F3n„3 n . IMPOSSIBLE FUNCTIONAL EQUATION B-325 Proposed by Verner E. ON THE SUM OF DIVISORS B-326 Based on the Solution to B-303 by David Zeitlin. Solution by Graham Lord. Quebec. San Jose. Illinois. Frank Higgins. 1. Solution by Bob Prielipp. Carl F. Bob Prielipp.2)(Lr3) = 0 for r even and to (Lr. Using the Binet form we have .Lr\r\ the second equation we have L2r = 10 . In B-260 it was shown that o(mn) > o(m) + o(n) for m > 7 and n > 1. one finds that a = G(ab) and for x = b that/? = G(ab). FINISHING TOUCHES ON A LUCAS IDENTITY B-327 Proposed by George Berzsenyi. Canada. Also solved by George Berzsenyi. o(m) + o(n) > 2^/o(m)o(n). Solution by Frank Higgins. California. Frank Higgins. C. Brady. -1. Beaumont. are the only possible pairs of solutions. Prove that o(mn) > 2^fo(m)o(nT for m > 1 and n > 1. Shallit and Sahib Singh. Replacing s by 2 . and the Proposer. By the arithmetic mean-geometric mean inequality. Minneapolis. Solution by Graham Lord. Texas. Wisconsin. l/l/ray G. Quebec. Graham Lord. Freitag.2. Canada.Lr = s and 2 . San Jose State University. let o(n) be the sum of the positive integral divisors of n. Hoggatt. Virginia. The desired result follows immediately. respectively. for all positive integers n.2(~1)r. Sahib Singh. F 3n+2 = knFn~i (mod 5). Lamar University.(-V Fn->i + F5F3n„4 = F6-Fn-V[5F2n_1 + 3(-L)n~1J + 5F3n-4 (mod 5). the two results together violate the single-valuedness. 1.2Lr + L2r = 3s . Determine a constant k such that. respectively. Also solved by Herta T. a solution for all positive integers n.WLr + 3Lf which. Gregory Wulczyn.) Also solved by George Berzsenyi. and the Proposer. For positive integers n. Freitag.. Minnesota. Find all integral values of r and s for which the equality •£ e. A.)(-ihn = s"Ln i=0 holds for all positive integers n.1977] ELEMENTARY PROBLEMS AND SOLUTIONS 95 FIBONACCI CONGRUENCE B-324 Proposed by Herta T. The University of Wisconsin. Peck. Naperville. Shallit.

2 .] ADVANCED PROBLEMS AND SOLUTIONS a2A(n) + aB(n) = 0 (a-p)C(n) = 0 A(n) + aB(n) = 0 It follows at once that AM = B(n) = CM = 0 (n > 0). Ralph Garfield. . Bruckman. Fmi tag. Also solved by P. 1. Tracy and P. 1977 sn(an+Pn) = snLn = £ (?) (-l)jLrj = £ (?) M J V / ' /=0 /=0 + H {ni)(-Vi<&r)i = (1-ar)" + (1-$r)n from which it is readily verified that r . and the Proposer.1 . . are solutions. 1.96 ELEMENTARY PROBLEMS AND SOLUTIONS FEB. ******* . respectively. [Continued from page 92. 2.0. Also solved by Herta T. 3 and 5 = 0. It is evident That a similar result holds for the Lucas numbers and similar sequences of numbers.

Swamy L. Jr. P. Glabe *H. P. A. M. F. Carlitz G. DeSales McNabb L. H. Thoro H. Umansky *L. Cocuzza W. Giuli G. N. Ledbetter George Ledin. California UNIVERSITY OF SANTA CLARA Santa Clara. Ossiander Fanciulli Pietro M. Cox D. T. Alexanderson *J. A. Jr. Rothwell H. L. S. J. E. Horadam Virginia Kelemen R. *A. Diem N. Jr. C. Paul F. R. Serkland A. Taylor *D. L. Yang Charles Ziegenfus ACADEMIC OR INSTITUTIONAL MEMBERS DUKE UNIVERSITY Durham. E. E. G. Burton L. North Carolina ST.SUSTAINING IYI£!VIB£R§ *H. California WASHINGTON STATE UNIVERSITY Pullman. Rondeau F. Styles M. M. L. P. Leonard. Kimberling *Kenneth Kloss J. R. Alder G. Winans E. W. K. Cosick. T. Beverage *Marjorie Bicknell-Johnson C. Arkin LeorvB&nkoff Murray Berg Gerald Bergum David G. E. Washington THE BAKER STORE EQUIPMENT COMPANY THE CALIFORNIA MATHEMATICS COUNCIL . J. A. Harris Frank Higgins A. Robbins M. *Sister M. Mamuscia E. Manning * James Maxwell R. Singmaster C. H. Frankel C. California SAN JOSE STATE UNIVERSITY San Jose. Fielder ^Charter Members Harvey Fox E. T. Kelisky C. Westbury Raymond Whitney Paul Willis C. E. Hoggatt. Y. Walton J. C. DeLeon J. D. Daykin M. Walker Marcellus Waddill J. H. *C. J. D. Alfred Brousseau John L. S. O'Donnell F. Ercolano D. N. Jr. Gould Nicholas Grant William Greig David Harahus V. Jr. C. R. R. T. A. G. Meissner W. McConnell. P. L. Shumaker D. Hillman *Verner E. Byrd C. Gardner R. Lahr J. Jr. MARY'S COLLEGE St. E. Draim J. C. Brown. Desmond Harvey Diehl M. Farmer D. F. L. Seielstad C. Risueno T. Bridger *Bro. Shannon J. Chakerian P. Long D. B. Mary's College. L. California SACRAMENTO STATE COLLEGE Sacramento.

electronically sealed over rigid board equipped with a clear label holder extending 2 3/4' high from the bottom of the backbone. round cornered." The name.The binder is made of heavy weight virgin vinyl. FIBONACCI QUARTERLY. fitted with a IV2" multiple mechanism and 4 heavy wires. The price per binder is $3. Mathematics Department. There is a small pocket on the spine for holding a tab giving year and volume. This binder is described by the company producing it as follows: ". Calif. The color of the binder is dark green. 95053. is printed in gold on the front of the binder and the spine. Santa Clara.BINDERS NOW AVAILABLE The Fibonacci Association is making available a binder which can be used to take care of one volume of the publication at a time. University of Santa Clara.. These latter will be supplied with each order if the volume or volumes to be bound are indicated. .50 which includes postage (ranging from 50^ to 80^ for one binder). All orders should be sent to: Professor Leonard Klosinski.. The tabs will be sent with the receipt or invoice.