,
WITH APPLICATIONS OF FIBONACCI TYPE
P.S.BRUCKMAN
Concord, California 94521
and
1.1 GOOD
Virginia Polytechnic Institute and State University, Blacksburg, Virginia 24082
Bromwich [1] f p. 24, attributes the formula
(1 )
_i_+
_ * ! _ + _ * ! _ + ... + J^* A
l_
1 x
1x%
1x4
1x*
1x2n
~
1x*n
to Augustus de Morgan, together with the corresponding sums of the infinite series, namely x(1  x)1 if * < 1, and
(1  x)'1 if * > 1. As far as the authors know, the following generalization has not yet appeared in print.
...
f ,„,",*"•"<"»>>*•»"<»<>,,
,
"
iS ft'rV**',"•*'/ '"' X",""
« _JL_
xV
r"" .
_
2
™N+1
,„,,,
(m=2,3,~J.
™N+1
xm
ym
To see this, note that the expression
n
n
izm
+z2m
+
...
n
n+1
+ z<rn1)™ ][1+zn,
j1
n+1
+z2m
7™
si
...adjnf/
+
=
m
is equal to the sum
(3)
(1
<7
(m~U
7™
I
„z
<
Jf
m
(1z
)(1z
}
of all those powers ofz where the powers are multiples of m" but not multiples of tnn+ . Therefore
y
_£? (1zm
n=0 (1Z<»
)(1z
On replacing z by y/x we obtain (2), and,
(m
m
,_j
^!l
L_
(\z\<1).
1Zm
)
on allowing N to tend to infinity we obtain, if \x\ ^ \y \,
V Ccv)mnfxmn(m1)vmn(m1)]
fm = 2 3 «J
= min(abs)/x,W
Z^
„
n+1
Xy
' '
'
n
n+1
n=0 (Xm ym )(Xm
ym
)
where min (abs) fx,^ signifies x or ^depending on whether \x \ <  /  o r  x  >  /  * respectively.
To obtain examples, let a and h be positive integers and let un be the denominator of the (n  Vth convergent of the
continued fraction
, , ,
(A)
]L A A
a+ a+ a+
so that
193
194
A GENERALIZATION OF A SERIES OF DE MORGAN
(5)
[0
un  p^L .
j(a2+4b)
where
r
y  a + J(a2+4b)
n . aJ(a +4b)
V
2
'
~~
2
Now putx = % m&y = Vk in (2), where k is a positive integer, and we obtain
5
(B)
f
(HI*'''''**'**.,, ,
U
~
n=u
kmnukmn+1
W*"™
g
u
kmn+*
When £ = /the formula simplifies somewhat. When a = b= 1, then un = Fn, the /?f/* Fibonacci number.
Some specM rases nf these formulae are
^
(7)
+£
F+ r
(8)
r
(9)
l
12^
1 +
r~ +=
~4
k ^ k
+
k
2
+
r
~2
+
+
^ ,....,
1
^Z^
(x =
2fy=hm=M))]
(a = b =
k*.1,m=2mHi),nN^—);
'
'8
. . .
= i
Zl^I
fe6*/,/»=3in«)f«yV^A
(where £j = 7, L2 =3, L3 =4, L4 = 7, are the Lucas numbers),
F
JL+FJL+
(10)
f± + ... = tz.s/E
L3
L9
L21
10 '
OO
(11)
OO
Y\
Lk3n
y » fAr3n
= (J57)k ^
=
^ 5  1)k
Further generalization. Formulae (2) and (4) can be further generalized to
M
1
„
K.
'
where \xn\
*0Ko
and j ^ o }
= y
Xn+lYn*nVn+1
Km_
+
f^0 <xnYn)(Xn+1Vn+l)
X
mYm
'
are any sequences of real or complex numbers, with xntynVn,
and
oo
Mil
1
'
x
V
n+lYnXnYn+1
titXnYnHXn+1Yn+l)
=
I y0/(x„  y0)
X x0/(x,y0)
if y„/x„  0
ii X„/yn * 0
'
/7=0
Although (12) is more general than (2), its proof is obvious. A special case of (13), after a change of notation, is
(14)
y
, ~
Ms(n)[xs(n+1hs(n)
 ys(n+1hs(n}]
fxs(n)_ys(n)j[xs(nH)_ys(n^1)j
__
if \x\ > \y\ and s(n) *• «> as n » <*>.
Now put A = 5, y= r\ and we obtain
d5)
Y
~!
nm
and in particular
ys(m)
xs(m) _ ys(m)
(h)s(n)us(n+ihs(n)
r^
s
Us{n)Us(n+1) Us(m)
1976]
WITH APPLICATIONS OF FIBONACCI TYPE
/Fs(0)
h~~^ra^7~MT)
n=0
For example , \U(n)=
195
Fn+1,
p°
and \is(n)=
€n
V
hi)
Fn

where
=*BL=J
e « J
1 if /? = 0 or 1 (mod 3)
Ln+1,
(18,
f
JL^.^/.
Putting s M = fa * 7^r in (16) gives
~
F(n+i)kF(n+2)k
Fj?
P u t t i n g ^ = (n +2)c and yn = 1 in (13) gives, after a change of notation,
(2o>
J2 — ( ^ . n i ^ i i — . i—
c
t^2
ft\VLmxn
=8
<n+1)t
,
c
(n ~1)((n
(n 1)t
(C > D.
C
+ 1) 1)
2 1
in (13) gives
yn=e~ +
(2D
V
=_dl
L
£<
cosh (2n+Vtcosh
t
2$lnh2{
'
Historical note. The formula
(22)
^3
2
=
2
~ 4 ~ 4 J 4 " 4.14.194 " ' " '
2
where 14 = 4  2, 194 = 14  2,  , was drawn to the attention of I. J. Good by Dr. G. L Camm in November
1947. (The sequence 4, 14, 194, — occurs also in tests for primality of the Mersenne numbers [ 4 ] , p. 235.) The
similar formula

(23)
(rVan
r^J(l_
+ _!_ + . \
where
r > I
ax = 2 m
^  +j ,1
an+1
= a£2,
30 = /,
Pn =
a1a2of1
was given in [ 3 ] ; and formula (8) in [2] and [ 6 ] . Hoggatt [5] then noticed that
T
^n
1
F k2
k2»
"
could similarly be summed. All these results follow from deMorgan's formula. S. J. Good noticed the generalization
(4) in November 1947, but at that time did not see its application to the Fibonacci and similar sequences and therefore withheld its publication. P. S. Bruckman independently, and recently, noticed the more general formula (14).
Alternate methods of proof appear in [ 7 ] .
b) the number of steps in this Harris algorithm. T. 5. N(a. 272276. 2\vn ~ 2\n (n > 0). MacMiilan. p. 434. 1974). H237. 8. L .. No. In case r = 0 . 3. vn := 2vn„1 + vn2 (n > 1). J.9. 1970). Clarendon Press. Example: 83 = 471+4.b) resp. I. D. Vol. In this note we prove that H(a. M. Knuth2 also) modified the Euclidean algorithm (= algorithm by greatest integers) for finding the gcd of two odd integersa>A > 1. E. 4. 2\r define the integersq.47) = 3. 300. 14. I. v1 := 1. 3.b) is sometimes much larger than E(a. 12. Harris1 (see D. Akkkkhk ON THE HARRIS MODIFICATION OF THE EUCLIDEAN ALGORITHM G.b) the number of steps in the algorithm by greatest resp. H. 47 = 95 + 2. N(a. divider by its highest power of 2 and obtain*? (say). Vol. No. • . nearest integers for a > b > 0. 2\ cn (n > 0). An Introduction to the Theory of Numbers. _ Cn+2 = 3cn+i . Germany V. "The Generalized Serial Test and the Binary Expansion of sjl] Corrigendum and Addition. p.2cn. obviously . London. Vol. Grover. I I'A.bk) = k.. proceed with/?. Denote by H(a. 1 Good. E(cn+1. Denote by E(a. 2. Millin. "Seminumerical Algorithms. For every integer k > 0 there exist odd integersa k >b k >Q with E(ak. C. 309. pp.bk) < 5. Jr. p. Let . 1 2 The Fibonacci Quarterly. 4 (Dec./ (k > 0) gives Theorem 1." AddisonWesley Pub. Hoggatt. 2nd Ed. Jr. E. The conditionsa = bq + r. Hardy and E.1. 6.r uniquely. G. Hannover. then_ ton+U Vn) = h *n < 3n"1. An Introduction to the Theory of Infinite Series. 2. 7. 1 (February. Vol. N.J. Bromwich. RIEGER Institut fii'rMathematik. \c\ instead oia. pp.f A. E. The Fibonacci Quarterly. Let vo '= 0." The Fibonacci Quarterly." 1 Roy. pp. Technische Universitat Hannover. since .b) < E(a. //(83. Wright. V. V. Oxford. According to Kronecker. Private communication (14 December. The Art of Computer Programming. \r\ <b. 346. 12. 102103. 1969. and Marjorte Bicknell. Good and T. 131 (1968). Hoggatt. 1931 ." The Fibonacci Quarterly.196 A GENERALIZATION OF A SERIES OF DE MORGAN WITH APPLICATIONS OF FIBONACCI TYPE OCT 1976 REFERENCES 1. 1974).. H(ak. Statist Soc. 316 [continued on page 200. Vol. c0 := I cn+i = 2cn + 5 (n > 0).b.b) always. "A Reciprocal Series of Fibonacci Numbers. 3 (Oct. "A Primer for the Fibonacci Numbers. Vol. the choice ak = ck/ bk = ck.b) and sometimes much smaller than N(a. No. Advanced problem No. 1974).] .b).cn)<5 (n > 01 r u Furthermore. stop. 1938. 9 = 19. A. In caser^O. No. Part XV: Variations on Summing a Series of Reciprocals of Fibonacci Numbers.
Chapman College.V K^n 1}kr (2n + 1)[(2nr)!] 2n+12r Proof.) [(vk+1]n+1sp2siM. s=1 The following Theorem 2 gives an explicit expression for these polynomials: Theorem 2. (1). Px (x) = x. (1) and (2) are easily shown to be identical. USN Program for Afloat College Education. so that P5 M = 25xs + 25(. 2 + 2>5 713xs + 5713(1)kx* + 13x. (x) = 5x> + 3(~ 1)kx. s=1 where s7 Pn <M = V Rs^'ryc2s'1'2f(1)kr (2s1)[(2s2r)!] r=0 Rearranging terms and changing the variable of summation by t = s. assume that (1) and (2) express identical polynomials P2s+lM for all s <n [ 2 ] . +5211x3 + 11(1)kx. it will be shown that the resulting expression for P2n+i(*) is identical to that as determined by Eq.Vkx3 + 5x. Thus. (n2:.rof summation on t and s obtain: 197 I eliminating r.. in p2n+iM = 5°x2n+iY._'. P9(x) = 5Ax9°+3253(1)kx1 = 55xll+5*>11(1)kx9 PJx) PJx) 6 l3 5 k 11 = 5 x +5 13(1) x PJx) + 3352x5 +23'52(Vkx3 + 22«5311x1 +527.(x) = 125X1 + 175(1)kxs + 70x3 + 7(1)kx.£ (/. Eq. Using the second principle of mathematical induction. Calif. (1) becomes p2n+iM = 5 V + 7 . Substituting the expression (2) into the righthand side of Eq.is) [<i)k+iin+isp2siM. The polynomials P3 (x) obtained by substituting n = 1 into Eqs. 92668 Denote the polynomials defined indirectly and recursively by my Theorem [1] by ?2n+iM P. P. Orange. BEVERAGE Civilian Instructor. n (OS P^ <(v\ . and interchanging the order . +23256x11 + 5611(1)kx9 +2325sxn + 233527(~1)kx5 + 22527x3 +3>5(Vkx . (2).POLYmmiALSP2n+lM SATISFYING P2n+l(Fk) = F(2n+l)k DAVID G.11(1)kx5 5 9 2 3 k n + 5 13x + 2 3>5 13(1) x = 5nx15 +3*51(Vkx13 +32x. Theorem [1] may be written as Theorem 1.
XT 2. = F(2nH)k l 2 " ( 2 t ^ r • Q. obtains three summations: one n+1 E tn+2r.^ ' . 5 f nkr " (2n + 1)[(2n .t . 3. { ' lSkr V (2n + 3)[(2n + 2 .5 + 1)!(n + s .1)!(s . each summation has the common r 2n+5 2r factor 5n+2~r(1)k x ~ + 1 as follows: w i t h the range of summation overlapping forr = 2ton n+1 E btn+2r. obtain p2n+lM . Eq. and thus by a well known property of polynomials. respectively. Proof. (5) = [5x2 + 2(1)k] P2n+sM P2n+3(x)  .198 ?OLWtiOM\AlSP2n+1MSM\SF\mGP2nH(Fk} (3) = 5"x2n+1 P2n+1M 5U1)knkt+n+1 .r)!] r!(2n+12r)! 2n+12r * Replacing r by r . Each of the polynomials is of degree (2m + 1)(2n + 1). the polynomials have identical values for an infinite number of arguments.1 and r . (5) above. (4) becomes (2) as desired.2)! . and noting the first term is represented with r= 0.t.£ [OCT. k= 1. 2.P2n+1(x) Proof. changing the variable of summation to r = n . namely. and since the same Fibonacci number.2 in the second and third summations. Theorem 3. E (4) 0tm^(n+t)l <i^<> Finally.=tH ' Substituting the latter for Q in (3) above and simplifying.r)!] r!(2n+32r)! 2n+52r * Rn+1~n *1*fr*D 2(2" + 3)[(2n +2( 1} ~ r!(2n+T^2F)!~ 5 r)!] „2n+32r * n E r=0 qnr. 5 n+1 4.r)! 2n+52r ( JJ + la * T^JT!(2n + 5~2r)! * r=1 . F(2m+D(2n+i)k> is obtained f o r x = F^. Theorem 4. P(2m+ D (2n+D M = P2m+1 (P2n+1M). the polynomials in x are identical [ 3 ] . and multiplying by [5x2 + 2(1)k]. where Q (2s1)[(s =E + t1)!] ^i^f= 1)!(ns+1)!(n+s)! S=t+1 Expression Q may be summed using the antidifference method as Q = + nTl 1 __ti\n (1) (n + t)! (nt)(2n)!(ntD! s^n+1 ~(~1)s(s + tW (n— t)(n . —.( iikr 1J ~ (2n + 3)[(2n +2r)!] ~ni2n^r^jr * 2n+52r n+2 + V trn+2r( i\kr 2(2n +3)[(2n + 3 . Substituting (1) into the righthand member of Eq.
1)1 + I (2n_2r < + J)(2n_2r) + 4r(2n.34m + 2r\n + 5r* .8m2 .+f(f*)f by applying the chain rule. one obtains: n+1 E ^ 5n+2r( 1}kr2n+52r 2 ( \ ) (2n +3)[(2n+2.2r)l + 52r)(2n+42r) + 2r(2n + r!(2n+52r)I 3)(2n+3r)r(r1)(2n+1)] +48n2+ 94n .1)!(2n +52r)\ ~ (r.1976] POLYNOMIALS P^i+/W SATISFYING ^ ^ / ^ .r)l] „2n+52t X (r2)!(2n+52r)! r=2 Collecting the overlapping portion of the summations in a single summation and simplifying the remaining individual terms. /WW^li. and n +2. one obtains four summations with like general terms inx with the range of summation overlapping from r= 1 to n . / ? +5? ? " ?2r)l . [5x2 + 4(~1)k]P'2n+lM (7) + 5xP'2n+iM 5(2n + 1)2P2n+lM = 0. 2(2n+3)[(2n+3 r)!] (2n+1)[(2n+2. For k odd.T2n+1 ( ^ / ) or (V JL T2n+1 ( .2)!(2n +5. 1. 52r)f Substituting this simplified result for the brace in (6) and noting the three individual terms in (6) from the summation general summation term with r= 0.D! 1! 0! (2n)f (2n +I)2 + \x2n+1~2r ' 5(1)kn(2n+1)[n!]x n! 0! 5nH(2n+1)3[(2n)!]x2n+1 __ 5(1)kn(2n + 1)2(n!)x 01 (2n + 1)! nf 1! The expression within the brace of the summation is easily shown to be zero.2r)l I (6) +Sn+2x^^ + Sn+U1)k(2n + Shc^+3 + (2n+5)(^1)knx. The expression within the brace of Eq. . Proof.I Factoring out the common factors.r)f] [ ) r!(2n + 3. where T2n+iM is the Chebyshev polynomial of the first kind [4].2rH (r. Proof.{li2+ "+*Z!.X ^ (^ J . (6) becomes (2n + 2r)![(2n+3)(2n = !??+l~)rT7 ttn* r!(2n + 5. The polynomials P211+1M satisfy P2n+1M = jL. and ^ +I W . Theorem 6. Differentiating (2) and substituting into the lefthand member of (7).1 in the summation formed from 4(1)kP'2n+i(x).2r+ 1)1 r=1 + 5n+1(2n + 1)[(2n)!]x2nH Of (2n . (6) becomes P2n+sM as expressed by (2). respectively.5^"U2n + 1M:Vkr(2nr)l r! (2n . the lefthand member of (7) becomes y L t ^ .r)l] . multiplying the binomial [5x + 4(1) ] appropriately in (7) to form two summations and changing the index of summation r to r.^ t t n ^ / A 199 n+2 E rn+2n l U nkr " (2n + 1)[(2n + 2. Eq. Theorem 5. and the remaining six individual terms are easily shown to be zero also.35r + 60] )U = r!(2n .r+1) +(2n+l2r) 5(1)kn(2n + 1)[(n + 1)!]4x + 5n+1 (2n+1)[(2n)!]x2n+1 (n .i ^ x ) according to k odd or even. [<2n+5H2n+4r)(2n+3r)l « {2" +r!(2n ?.
1).2\k. obtain s (1z2)T2n+l(z)z.9 p. rk uniquely. if ls\yk := 2*2 +rj.hs)=l Nothing is known about the average size of H(a. 2/f rk+1 (j = k1. p. The case for k even may be handled similarly. 773795.(2/\Js)z.hs := ys in case 2\s and gs :=xs+f. 4. 3. xk := 3yk+2 . 0 < rk+t < 2vk+1. David G. 1971). 781]. yk) > k.6.1) are the beginning of &. (j = k1. define the integers rk+j. 4... .f2n+l(z)+(2n+1)2T2n+lM 0 defining the required polynomials [4: 22. hs := ys+f in case 2 Jf s.S." The Fibonacci Quarterly. 5 (Dec. ifkkkkkk [Continued from page 196. National Bureau of Standards. H(gs. 15. .] K:=4k + 3. 541544. U. This proves Theorem 2. L E. No. algorithm by greatest and by nearest integers forx k .k2. D. yk. Applied Math Series 55. Nathan Jacobson. The defining equations forx k .2j(yk. John Wiley & Sons. 1960. Vol. 9.b). For an arbitrary integers > 0.2 #> IS = 2K. For every integer s > 0 there exist odd integers gs > hs > 0 with E(gs. yk) = 2. hs) > N(gs. j= 2J[r/~2J[j.2/\xk.~.then22 < yk.1. k . yk and therefore N(xk. Dept.k2. 1. Dickson. the conditions L&t k>0. 1 gives Let 2r1+r2 l* ». q (j= 1. Handbook of Mathematical Functions. . ^ 0 < 2r1 < 2K. Vol. 2. 1). Th.The defining equation for xk gives H(xk. \etgs :=xs. Van Nostrand. pp. "A Polynomial Representation of Fibonacci Numbers.. h$) >s. K Vj^rj + vjrj^ . p. New First Course in the Theory of Equations. Commerce. Vk+I?k + Vkrk+1 = 2K.200 POLYNOMIALS/^/WSATISFYING P2n+l(Fk) =P(2n+1)k OCT. 1976 Substituting into (7) and changing the variable* to z by x . Beverage. 195. 9. REFERENCES 1. Lectures in Abstract Algebra. pp. rj := 2rj+7 + r/+2 0 < 2rj+1 < rj. Then 2rk+i <rk. 2.
Once incidence subgroups are defined the result follows quite easily. where \Fn}= {1.1. of Tn is an incidence subgroup if (a) The relations defining H can be given entirely by specifying the domain for each a. = which is an incidence subgroup. That is Tn= a. Let K be any field with more than two elements and let K* denote the nonzero elements of K.c s F r. The key definition is as follows. (c) The two possibilities for a. of Tn such t h a t 5 ' = T'n is Fn+2. Our result is the following: Proposition 1.e F*. Definition.FIBONACCI NUMBERS AND UPPER TRIANGULAR GROUPS LOUIS SHAPIRO Howard University. Washington. We define Tn to be the group of all nonsingular n x n upper triangular matrices over K. 20001 In this note we call attention to the curious fact that the Fibonacci numbers arise when we look at that familiar example from group theory. A subgroup.f.j e K\. 201 an^K*.} is the sequence of Fibonacci numbers.5.8.S. By way of example we have is an incidence subgroup of Td.0..f e F Since He Tn we automatically have a.3.f e F if i < For instance n a...2) and (1_„3) entries are dependent. .b. The number of incidence sugroups.. Then it is easily shown that T'n ^(atj)\an = I a. (b) The two possibilities for each a.f = 0 whenever / >j. i (atj)\ajj= 0 if/>/. the n X n nonsingular upper triangular matrices.are an = 1 or a./ = 0 or a. H. is a subgroup but not an incidence subgroup since the (1. { ( 0 / A ]  a.h e K \0 0 1 is not a subgroup.2.f when / < / are <?. 0. We let G' denote the commutator subgroup of G.
Mathematics of Choice. <j. I. Incidence subgroups are themselves an interesting topic. The term comes from incidence algebra as used in the study of locally finite partially ordered sets./>/ G F is specified for each i. Niven./= 1 or a/. The number of normal incidence subgroups of T'n is given by the Catalan numbers.202 FIBONACCI NUMBERS AND UPPER TRIANGULAR GROUPS OCT. To complete the proof we need the standard result (for instance see Niven [1]) that the number of sequences of n plus and minus signs with no two minus signs adjacent is Fn+2. pp. 2. Alternatively both F* b G F] and H. New York."/ 3 /^... \0 al+1ii+1)' along the main diagonal is either Hu H2 or T2 then 5. This yields S'= T'n.i+t 0 1 0 1 Using block multiplication and the above computation we have / / Cl2 0 1 0 0 AlB~lAB  Cfn C2n = and such matrices will not yield all of T'n. If the number of incidence subgroups of T'n were known it might be useful in determining the number of finite TQ topologies.j = 0 for/ >j. 6 (1955). 454464. Thus if no two consecutive entries on the main diagonal are specified as 1's then S'= T'n.M. Note that the commutator U 7/ U / / \0 1l\0 11 ~ \0 l)\0 1 l \ O l ) \ 0 1 l = \ 0 1 ) Now let bin I>2n 3in\ 32n A = 1 bi.  a e F* b e f ) are nonabelian.. REFERENCES 1. 1976 We must have TnDS D T'n so that if S = j (a/j) I we then have a.* e /f* Suppose we specify 1 = a/f = aj+ij+i. 5253. Similarly \0 1J \0 1) [O 1) \0 1j [ 0 1 ) and we can generate T\ by choosing a appropriately. . "Sylow pSubgroups of the General Linear Groups Over Fields of Characteristic/?. However this is an unsolved problem for/7 larger than nine. pp. Random House. If K is finite then most normal and all characteristic subgroups of T'n are incidencesubgroups (see Weir [2]). !f every 2 x 2 block. Weir.S. 1965. The center or commutator subgroup of any incidence subgroup is itself an incidence subgroup. The following facts are known. and for each /we must specify either a.i+1 ' 1 ai. A. a//G Kfor/ Proof. A.
3.r. 1973. and so on until the last value. After/77 iterations of Formula (3). Here we use the usual convention that f nk \ = 0 when n is a positive integer and k > n. such as the binary system. [ 1 ] . then r 2.V n(n.1. Ohio 44115 Abel. We first establish the following triple summation expansion: ( £'/ r i=1 \n J r « ( j1 . aj. ar be complex numbers with n not equal to a nonnegative integer. about 150 years ago gave the first proof of the Binomial Theorem for the case of an arbitrary complex exponent. we can apply Formula (3) to the summation under c on the right side of (3). so that the right side of (3) becomes. r r n E <bl + 'l> z?) = E (*M*I> M = *»1 > M which is precisely the left side of (3). > 0. forj = 2. all hold.k $ 0. Let/7. —. ar. letZj = 0 and z = H i E a* 9=1 for/ > 2. 203 . is taken on.. then the following Multinomial Expansion holds: n\ (2) ( V * l . m > 0 and not too large. Since n .T2 EE (0'' k=1 \ \ nk a £ * + " ««/ / if the inequalities (!) all hold. we obtain 1 Received by the editors in July. If the inequalities \aj\ < \ai+B2 +  (!) + *jl\. 1. Indeed. as well as Formula (7). by (1). to be of use in the representation of integers in specialized arithmetical systems.ON THE INFINITE MULTINOMIAL EXPANSION1 DAVID LEE H1LL1KER The Cleveland State University.V(n~n1n2 {L * 1 L n^+1) "777^7^77— nlan2 a2 ji^nn^n^^n^ 1 *' ^ " " where the summation is an iterated summation taken under all n. Formula (3) is of interest in its own right This author has found it. where i first takes on the value r . From Abel's result one can deduce various versions of the Multinomial Expansion. This iterative process can be continued. In this note we shall derive one such form. Cleveland.
c 2r _ 3 2. Then. r. so that the first two inequalities in (5) are possible. . •a"^~^+a" .1. [53.. by (5). and with f(1) = I That is. the subscripts are uniquely determined: r1.1 We now extend the range of fi2i.0 . to include 0. It now follows from (6) that 2> i=1 E n(n 1)~'(nz2 Q \0 I .. Our inequalities (1) imply that the quantities ax + a2 +» + a/. this result being valid for all r > 1.I f ° r 1 < i < rn. (5) The only restriction on /w is that/w < r . for 1 <J<r.v " . and various Multinomial Expansions (Hilliker [ 3 ] . as usual. so that Formula (4) becomes n9.(""^jrH ^ • • ^ / . Volumes I and II.e ' k ^ Here the indices vary over J 2 < c. . for 0 < i < m.2. . . [6] and the present paper) will continue (Hilliker [ 7 ] . the Multinomial Theorem. < r. subject to the restrictions (1). and.1. the logarithmic function is given by l o g z = log z + i argzwith argz < n. %2r3 = 2. zn is defined to be that branch of the function f(x) = en . now follows with a change of notation. we are employing the usual convention that the empty sum is 0 and the empty product is 1. Our sequence of expository papers on the Binomial Theorem. by (5). [8]). Another version of the Multinomial Theorem is (7) t" ] = (D ii <i)' r i~1 / H E ("*)*/ E 2> k=1 nk *' £=7£ / *a * . o g z defined over the complex zplane with the nonpositive real axis included. . The Multinomial Expansion (2). ^ 2 < fi2/>/ < c2/7 . A good source for the Binomial Theorem and the Multinomial Theorem is Chrystal's/4/#0i&/*a [ 2 ] . for 1 < / < r . * »\ . [ 4 ] . Then.204 ON THE INFINITE MULTINOMIAL EXPANSION [OCT. ADDENDUM. C 2m + i~7 (4) + E E U)("C2) . valid under the conditions (1).2 (6) £3f~a /=/ e k * * . We let m = r . the summation under k reduces toasingle term k = r.+ D Jt 2 V ^/"^22^ I /72. Here. Here. for r > 2. L 7 < fi2/>2 < °° .2.
David Lee Hilliker. Chrystal. David Lee Hilliker. naturally. 1 (1974). 6. 1881.7 1 . 8. Englewood Cliffs. David Lee Hilliker. " 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. David Lee Hilliker." Second paper. Vol. New York. pp. 4 (1973). G." The Mathematics Student. I. Vols.. The proof for complex exponents also appears in Abel's Oeuvres. 1961. Vol XLII. Markushevich. and III. For more on this. may. 6 5 . 311. New Jersey. I. 8 pp. "Contributions of Cavalieri. David Lee Hilliker.1976] ON THE INFINITE MULTINOMIAL EXPANSION 20B are not 0. 219. Chelsea. Theory of Functions of a Complex Variable. . PrenticeHall." The Fibonacci Quarterly (to appear). in regard to Newton's Discovery of the Binomial Theorem. Also reprinted by Dover. Textbook of Algebra. No. 4 pp. Vol. New York." The Mathematics Student. No. I and II. 3. "On the infinite Multinomial Expansion." second part.see Markushevich [ 9 ] . "Contributions to Archimedes. "Contributions of Pascal. 4. 6 pp. "Contributions of Newton. Crelle's Journal. "On the Multinomial Theorem. and also for a development of the Binomial Theorem. 2(1974). Volumes I. 1965. be ignored. I.1' The Mathematics Student. the Maclaurin expansion (1+z>n E (l)zk k=0 for /7 and z complex and with \z\ < 1. p. Christiania. No. The Fibonacci Quarterly (to appear). This is a reprint of works of 1886 and 1889. XL. XL. A." fourth part. 2. " A Study in the History of Analysis up to the Time of Leibniz and Newton. 7." First part. David Lee Hilliker. Vol. When n is a (negative) integer these restrictions which guarantee singlevaluedness. XLI. 5. 1964. " 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. 9." third part. Niels Henrick Abel. Translated from the Russian. We shall need to assume that they are not negative real numbers. " 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. II. I. REFERENCES 1. 1826. 1 (1972). that is." The Mathematics Student. No. p.
m. zj = I was not adequate. tc\w. we now discuss the trivial solutions to C\). z0 to k). vf w. tb\v. to (1) is called a (nontrivial) primitive solution if and only if there is no t > 1 such that ta\u. . Thus for the nonzero trivial solutions with x0 = 0. 4. 1 TRIVIAL SOLUTIONS For completeness as well as for illustrating a simple case of primitive solutions. the finding of the primitive parameters which augment to primitive solutions is equivalent to the original problem. we say y0. 206 and L = [n. We have a general algorithm which sometimes augments primitive parameters to primitive solutions regardless of the choice of m. trivial. For some types all the primitive parameters augment to all the primitive solutions.THE SUM OF TWO POWERS IS A THIRD.k].e. z0. Los Angeles. and the nonzero elements are both powers with common exponent the least common multiple of their corresponding exponents (x0 corresponds to n. Thus the possible candidates for a nonzero. zQ form a primitive solution if and only if there is no integer d > 1 such that d \y0 and d \zQ where L = [m. namely. one where (**. 2. We classify the types of applications of this algorithm based on the greatest common divisor of the exponents. z0 is zero. z0.. HISTORY OF THE PROBLEM In 1964 we obtained the computer solution (1176)2 + (49)3 = (35)4 and from this one example we discovered our method of augmentation as well as a type of exponents for which we determined all primitive solutions. where n. California 90032 1. Strauss pointed out to us that this method could be applied successfully (i. we could extend this approach to a Diophantine equation with more powers on the left than two but only one power on the right. y0. yielding solutions) to another type. INTRODUCTION We seek integer solutions to the Diophantine equation xn+ym (1) = zk. and turn our attention elsewhere. c = L/k. Thus we give a new definition which reduces to the old when appropriate. B. k]. Without gain of generality (an expression of Professor DeHardt). Recently Beerenson [2] has found a similar method for finding integer solutions for this first type. G. primitive solution s are:K0 =±hz* ±1. where a = L/n. Definition: A solution u. y0 to/??. At this time we found that Basu [1] and others had found rational solutions for the first type mentioned above. m=n=2. k. SOMETIMES R. At this later time we found that Teilhet [8] in 1903 used the method of augmentation for a special case k = 3.x0. zo> = fro. where xQy0z0 = 0. y0. K8LLGR0VE California State University. the zero case. Then exactly one of x0. y0. PRIMITIVE SOLUTIONS AND THE CLASSIFICATION SCHEME The computer example indicated to us that the usual definition of primitive solution x0. uvw^O. Subsequently that year Professor E. b = L/m. m and k are positive integers. yj = (*<>. Let us call the case x = y = z = Q. n. For the type which includes the famous case n = m = k > 2.
y. v = y 0z%.0 .. Suppose the conditions hold. i ? j. e = D/m.y0.\t. zdn = ZQ171 = z®.k) = 1. If A. thenz^ is one if all a/= 1 f o r / V & and is the product of a. q Q v = yxzl.m. we can show there are no nontrivial solutions if F.. v > 0. d Let/7 be a prime such thatp \\t andp \\z 0 and qf(L Thenp \xj and/?e\y. w >k 0.G. The type (n. if nontrivial solutions exist.w is not a primitive solution.butnooneof/7.w.LT.LT. holds. = 1 or there is a solution g\ to Dg\ = . The smallest such positive solution \%<k/(D. v = y04 and w = akkakk_\ Hence (w/ak)k = akZ] a^ajZQ a^a^Zg . This is a solution to (1) but it may not be primitive whenz 0 $ 1.w form a primitive solution to (1) then there is a unique ordered pairx 0 .v. Theorem 1. uvw ? 0.ast 1. The type (n. \\(n.^ 1.k)> 2. is squarefree and (a. we have p'p9 p9\\z0 Thus Dg =i and = pq .k] = L = [n. p.v. there are an infinite number of primitive solutions.OCT. Thusf : = z 0 andx 0 =xt andj/ 0 =yx./r is relatively prime to the other two.\/. THE METHOD OF AUGMENTATION Let D = [m. 1 < / <k and ai$\.v. T\\mxQ=u/td are primitive parameters.K 0 which are primitive parameters so that u. and k = 2j. m or k is relatively prime to the others.w is an augmentation o f x 0 . where each a. y t are primitive paramu = xxzd. Theorem 2. 1976 THE SUIV! OF TWO POWERS IS A THIRD.k). pL\wk.LT. If (h. Thus pL\u". If positive integers u. This conjecture is true for/7 < 25000 [ 7 ] .m. Then u = x0zd0. where d = D/n and e = D/m. Definition: A primitive solution u. then there is an augmentation to a positive primitive solution for (1) if and only if for each /.exp D g. pL\vm. Suppose there is an /^ 1 < / < / r . as far as we know. Then v = and w s a OzO' VozO* k*' where 7T is the product of the positive k roots of a.) wherein the conjecture states that there are no nontrivial solutions. If n = 2h. \ig>k/(D. m = 2i.m. Strauss) all possible solutions can be obtained by augmentation.n] throughout this section. SOMETIMES 207 The case n = m = k > Z \$ referred to as Fermat's Last Theorem (F. Then for each prime p dividing a.. for all i. and u. then we have a proper augmentation. Definition: Positive integers x0 and y 0 are primitive parameters for (1) if and only if there is no f > 1 such that td\x0 and te\yQ .k) = 2 has not yet been completely resolved. then gD > [D. i f k. Sn all other cases. y0. y0 for (1) if and only if u=x0zd. is the only known type which sometimes yields a finite number of primitive solutions. andy 0 =v/f eters and Let t be the largest positive integer for which td\u a n d *e\v and d=D/n./ ' (mod k) and gt is the smallest such positive solution. aj)= 1 for each i <k? j <k. This is the "first type" referred to in Section 2./77. where/ = 0 if q <Q and/= 1 if Q < q. Proof Suppose we have a primitive solution u > 0. (mod k). This contradicts the condition x. fj= Dgj + i. where pq\\(w/ak).i) > 2. otherwise. and u. We conjecture the same holds for (hj) > 2 and (ij) > 2.i) = (h. to (1) is an augmentation of primitive parameters x0. and if (h. then (E. or one. u = x th . 5. such that a.j) = (ij) = 1. yQ are primitive parameters for (1) a n d x " +y™ is written as akakZ1i — a^ai.. either a.are primitive parameters. Proof. then (1) for this type of exponents can be reduced to F.k) [2. and we write bc as/? exp c.w\% an augmentation of x0. 1 <i <k. Suppose^. We complete our classification scheme by mentioning the remaining type where one of n. Ifz 0 > 1. 51]..exp f\.m.k].
a = Sd. For n = k = 2. e then p \xo.m = 2f k = 2t. m = 2i./ (mod 2t). 20. Theorem 3. or more augmentations of Pythagorean triples. SOMETIMES [OCT. Since L . and 2sg. For completeness we give a proof for a theorem in the literature [4] because it is easy and not too accessible. and if From a primitive Pythagorean triple [3.k]/D for » L/D = S. y0 are primitive parameters for n = 2.VS some integer S. by adding r + s to r—s. z0.so g. primitive parameters then there is an odd /such that a/^ 1. yo are not primitive parameters. we use the fact [3.s from r + s and rename.k) = [D.= . (s. k = 4. If n = 2h. we obtain i / f + vt=wt. then all the primitive solutions to (1) are obtained by augmentation of primitive Pythagorean triples./ (mod 2t) has no solution. and the corresponding odd indexed a. (s. Thenxt= 2rs.j) = (i. where a = L/n.. Theng. 191] that* 4 +y* = z2 has only trivial solutions. Sf n = 2.t) = 1. THE TYPE (n.d < dSd = ad. 5 forn . In the latter case subtract/. pb\yozeo and pc\w.i) = (h. and 2gj = . But if y0 +y0s is not a square for. \\n = 2. y0.<S . Proof \fx0.amam1 2 1Lemma 2.e < eS .e = h . k = 2by augmentation. andg. But k/(D.e. 2 Lemma 1.k) = 2 Here n = 2h.f 0/ y0. Theorem: lfx2t + y2t = z2 has a nontrivial solution then t is odd and ut + vt=wthasa Proof.c = L/k If pj(a. m = 2$. and there is no solution to 2sg. We do not give the proof since it repeats a third time essentially the proofs of the two lemmas. V. k = 2. = 1. then there is a prime/? such that pa\x0zd0. g. rs)= I (In this case the new and old definition of primitive are equivalent). y0 for primitive param 2 2 2t 2t1 0 V0 = a2ta2t1 "'a2a1 under the conditions of Theorem 2.m. k = 2j. 6.m . 24. andx0 and y0 are not primitive parameters. then when written in the above form a. hence r + s = u* and rs = vf.=i (mod 2t) can be solved for all needed even i.b = L/m. b = Se. \\p\aj.k = 2t. If t is even. y0.for any i.1. but either2r= wt or2s = wt.m = 2. k = 2t. If this is not a primitive solution. 190].208 THE SUM OF TWO POWERS IS A THIRD.1 1 form some odd i. 5. and there is no augmentation.j) = 1 then there are an infinite number of primitive solutions to (1) obtained by none. = 1 for all odd / and 2g. Proof. eters. m = 2s. p. p. z0 is a primitive solution to (1) with n = 2. (h. p. m = 2i. then/7 exp#/z 0 .. But (r + s. In the former case. Lebesque nontrivial solution.= . If xl + y\ is not a square.t) = 1.^. .k). y* = 2 r .s [3.similarly p \yo an&xo. k = 2j. and 7. then all primitive solutions to (1) are obtained from primitive solutions to (1) with n = 2.2s. then/7 /{z0 and/7ax0 and/?^y 0 . except as proper or improper augmentations. we can use x0. NOTE: Our method does not distinguish solutions 15. m = 4 we use a general modification of Theorems 1 and 2 using X + z a a O~xO .< k/(D.dlZQ. 190] x0. 1 </<k. one. A. then. Since and Pa\*OzO' pexpg./ (mod 2t) can be solved for all even /. then a.
pp. it can be shown there is ai? and c if and only if (h. pp. 9. Mr.. REFERENCES 1./M/. M. thus for nontrivial solutionsz = u. 737.i) = (bj) = (ij) = 1. L E. J. p. k = 5 and hence uh = ±3a3. By Legendre's result. 1903. Dickson. History of the Theory of Numbers. Then we augment by 74 and obtain the solution we found on the computer. M." Ball. If u = v. n = 2. Thus we consider the former case only. then v = 0. then z = 0. L E. ibid. k is relatively prime to the other two. 3.m =3. 766.E. has also found an algorithm for obtaining rational solutions for this remaining case. 25. L.Note for b = 4 c = 9. Mow any solution u. V.242 = 72 from the Pythagorean triple. P. hencey =2. u = ±v. m=3i. Pollack.z = t l Proof. and by Legendre's result. 8. Vol. 138139. "Cube somme de dans Carres. Dickson. k = 4 then we write z 4 x2 = y3. (1940). 1960. Selfridge and B. Moreover. then the conditions of Theorem 2 are always met and every set of primitive parameters augment. and u3 + v3 = 2z3. H.z3 +x= v3.9]. If/7 is a prime greater than 3 an6pd\\a. Axel Thue (entry 236). Vol. p. Lebesque (entry 31).1976] THE SUM OF TWO POWERS IS A THIRD.C. . Legendre (entry 184). W. from (u. k = 6] for this type. p. I I . Hardy and E. 54 (1970). ibid. 24. we obtain z\ + (u)3 = 2\/3.Y.m. w to (1) for n = 2h. Sf z = u. including some differential equations.D. Dickson. 27. Oxford. 5. Using this hard to obtain result we give a proof of a theorem in the literature [6. and j\d and [n. Calcutta Math. Soc. A..k IS RELATIVELY PRIME TO THE OTHER TWO We know how to solve only n = 2h. 10. 8.. 9. namely that x3 + y3 = 2z3 implies x = ±y. z3 + x) = 2. Teilhet. A. 2. One case can be obtained by the other by replacing x by x. w1 = ±a.3if k = 6j is a solution to the case n = 2." Notices of AMS. G.u = +1. N. and j\ h and j\c and h \3b and i\2c and h \ 1 + 3c and /1 / + 2b. and if u = v. Thus a = 2 3 . v.k) = 1 BUT NONE OF n. The only nontrivial primitive solutions t o * 2 + y3 = z6 Mex = ±3. Lind (entry 224). He has made a real gain with his general approach because it not only applies to general Diophantine equations of this type but also applies to many other problems as well. "Fermat's Theorem is True for any Exponent up to 25000. 97. m . "On the Equation xn ± yn =zm. Dickson. but x is a solution if and only if x is. 210.v) = 1. and / = 0. SOMETIMES 209 7. Basu.k]\6d b c and this is not a primitive solution. M." Llntermediaire desMathematicens. 573. Now z3 x = 2y3 x3 +x = 4v3 and or z 3 x = 4u3 and z 3 +x = 2\/3 .6 case. Wright. 1520. and.n. Jim Grant.. "On a Diophantine Equation. R. For example..L. U.m. N. L E. p.z3 +x)= 1 o r Z In the former case. Then by adding. i\2d. 6." The Mathematical Gazette. vl = 2a2. Conversely if these conditions are met then there is a solution. An Introduction to the Theory of Numbers. x = 0. s®. \/±f.A.y = 2. Fourth Edition. for example. p. then h\3d. 1952. Chelsea. 6 case as well as 8.z3 x = u3. 5.z = ±L Q. L E. Then v = u. ThueLind Theorem. THE TYPE (m. we obtain 8. 7. 24 being relatively prime are primitive parameters and 54 . when the equation is written with the special exponent term being the only term of one side of the equation. Beerensson. 4.m = 3. 7. student. 11 (1964). F. G. We assume a result of Legendre [ 5 ] . m. First we note (z3 x. p. 580. THE REMAINING CASE AND SUMMARY The remaining case when one of n. z = u orz = u. Therefore for nontrivial solutions (z2 x.
. Uk1 = /. then U0 is invertible in Rkxk. akl Let S(aj.. then the sequence is said to be of finite period. Solutionsto this problem have previously been given in special cases: Vinson [7. is generalized to sequences of arbitrary order over an arbitrary (not necessarily commutative) ring. A solution is given in the present paper for linear recurrent sequences of arbitrary order over an arbitrary commutative ring with an identity. and Wyler [10. *). Let (w) <ES(ai. ekAk~1 = (1. if (wa+v.consecutive elements of (w).associated with the list fa. —. then (w) is said to be purely periodic. wn+ki) = (w0.I f t n e r e ' s a l' s t of k consecutive elements of (w) that is equal to a preceding list of A. Thus. wa+i. 1)e/? . both the firstequality and Un = UQA11 are valid. (In case/r = 7. By finite induction on /?.THE RANK AND PERIOD OF A LINEAR RECURRENT SEQUENCE OVER A RING DONALD W. A sequence (w): wQ. Similarly. 210 . since U0 is invertible. Theorem 2] for the modular Fibonacci sequence. .ArXA: be the companion matrix of (ai. a^1 The Lucas sequence of S(a1f . 0 0 ak aki \ 0  / a7 0 . '••.. — is equal to a preceding term. then W&. Theorem 1]. . —. In particular. wa+klh with a+v> a> 0. Theorem 3] and Barner [1. Uk2=0. wu —of elements in R that satisfy the recurrence wn+k = wn+kiai + ~' + wnak for/7 > 0 is said to be a (right) linear recurrent sequence associated with the list fa/. PERIODIC LINEAR RECURRENT SEQUENCES Let R be an associative ring with an identity 1. Provo. ak) such that ug = Q. a^J be the collection of all linear recurrent sequences over R associated with (a^f —.. let Un e Rkxk be the matrix with ekAf~1+n as its ith row. a^J and let A / 0 =I / 0 0 •». (wn. O. a theorem of Wyler [10./ wa+v+k1> = (WQ. a^) is the sequence (u): u0. .fy be elements of /?. the matrix A is said to be periodic if some term of the sequence /.. Utah 84602 INTRODUCTION Two problems from the theory of linear recurrent sequences are considered in this paper. —. for secondorder sequences over a commutative ring. • •. Ward [9] for modular integral sequences of arbitrary order in case the characteristic polynomial of the recurrence has distinct roots. k. * . 1). Since the rows of U0 are of the form €k = O. In this case (w) is said to be periodic of index a and period v. akl Then for A > O. . If the index a= 0.. Let (w)^S(aj. —. —. and let a/.. thenuo = /). and the second equality holds. */. 0. " v ak). Lemma 1. A2.. Specifically. —.. 1.ROBINSON Brigham Young University. wki)An = (w0< '" .Wki)UQ Un. For A7 a nonnegative integer and e^ = (0. The first is to establish the existence of the rank of the Lucas sequence over an arbitrary ring with an identity. ». . / = 7. Theorem 4] for secondorder sequences over a commutative ring with odd prime power characteristic. A. then An = U~Q Un. 7. is periodic of period v. Proof. ekA = (O. uu . is the first such list. The second problem is to determine the period of a purely periodic Lucas sequence as a function of its rank..
Proof First. (ii) The Lucas sequence ofS(aj. Proof. a+v > a > 0 Then and (w) is periodic. . then by Lemma 2. Since (u) is periodic. Then the following statements are equivalent: (i) Every sequence of S(aj. ak) is periodic. —. Lemma 2. —. (iv) => (i). ak) be purely periodic. —. suppose BA = 0. If p = 0. That Is. Specifically. then (u) is not periodic. —. Let the Lucas sequence of Sh]. Then Ua+v ~ Ua. a^) is purely periodic. Av = UQUV= L If Icjj] = A . respectively. —. Hence. (ii) The Lucas sequence ofShi. and ak is right invertible. Let (wo. (i) => (ii)and (iii) => (iv) are trivial. the index and period of A Lemma 3.. Then Uv = UQ for v > 0. Then there exists a unique nonnegative integer p such that un = 0. That is. Proof. >4 is periodic. Indeed. . the main result given there follows immediately from Lemma 3. un+k?l = O if and only if n is a multiple of p. ak) is periodic. ". . let(u) be periodic. The following statements are equivalent: (i) Every sequence otSfaj. M/£.OCT. (w0. 1976 THERANKAND PERIOD OF A LiWEAR RECURRENT SEQUENCE OVER A RIIUG \iAa = Aaf a+v>a>Q. un+ki) = ekAnf and un+kiekA' = 0\s provided.Let(u) e $(aj.. then A is said to be periodic of index a and period v. then A is right cancellable. then (u) is periodic if and only \iup+k1 is of finite order in the unit group of R. O. every (w) e S(aj. Since Ak = Iak + Aak1 + . ••. Reference is made at this point to DeCarli [ 4 ] . Wki)Aa. Wkl)AV = (W0f . akCkj = 1. and suppose ak is not a right zero divisor in /?. Let the Lucas sequence (u) e Stai..+ Ak~1ai . It is clear that the index of (w) is at most the index of A and that the period of (w) divides the period of A Moreover. 2. If p > 0. —. (iv) a^ is not a right zero divisor in /?. then Bak = 0 and B = 0. Theorem l. (i) => (ii) is trivial. Theorem 1]. the index and period of the Lucas sequence are./j and /Wj is purely periodic. (iii) ak is right invertible in /?. (iii) => (i). is now extended. a+V ^Ac U0A"+P == U0u A° a+v > a > 0. suppose = ekAnAl = ekA'An . (ii) =• (iii).ak)bet\\e Lucas sequence.. a^) is purely periodic.. Therefore. Wk^i)Aa+v = (wo. (ii) => (iii). (iii) A is periodic. •••. THE RAWK OF THE LUCAS SEQUENCE A result of Wyler [8. a matrix characterization of the condition un = O. for secondorder recurrences over a commutative ring. un+k2 un+kiek = (O. ak) be periodic. a+v Also since ak is not a right zero divisor. Let Aa+V = Aa. a^) is periodic. 211 is the first such term. a+V A Aa+V and and = A" . —. > a > 0. then by direct calculation.
) We show that every such/7 is a multiple of p.j = e'. in the unit group of R. then UQ ek = e\.*/p+*/f and wp = (wp. —. then n s not a right zero divisor in R. if WQ = UQA" for some t <ER. Suppose^ is not a zero divisor in R. Thus. On the other hand.and [OCT. ~<. Then UQUUO = An = AxApq = AxUo1sqU0. and 5 = up+ki is of finite order in the unit group of R. Since ak is not a right zero divisor in R.0 and un+ki = t. Therefore. if sq = 1. by Lemma 3. Then vU0An = vtU0 = 0. if (u) is periodic. Indeed. O) <= Rk and e^ = (O. That is. VUQ = 0 and v = 0../ = /. suppose An = Uo1tUo. On the other hand. un+k2 . and let 5 and j3be the orders of (1)k~1 and Up+kj. Indeed.1. by definition of p. then choose p = 0.. Also since </ / 7 ^ = t = d„Sq. a^l If WQ = 0. in this case t = = 0 if and un+kj. un+k2 = 0 implies n = 0. if M/0 = = (wo.(3) \sthe order of uj££J/p. In this case.y) and [x. •»«. Indeed.1. then it is purely periodic and Av= l. Then (u) is periodic if and only if ak is of finite order in the unit group of /?. •••./ wp+ki)e'i Consequently. (Compare [3. un = 0. it follows thatX = 0 and n = pq. then by check ing the first row of this matrix. a/d and let (w) e S(aj. respectively. Q. Consequently. q Then Ds = It Since d.) 3. Since e£ = U0ep where the prime denotes transpose. Then (i) v=p$=(k. Thereforelun+k.i)APc'j = (w0. then as in the proof of Lemma 3.y] denote the greatest common divisor and least common multiple of the positive integers* and y. —. Suppose a/< is not a right zero divisor in R. then djj = 0 for / ^ j. . Second. Letp be the rank of the Lucas sequence (u)<ES(aj. . upq+k. un+k2 ~ 0 f ° r s o m e n > 0. 0 < X < p. —. If un = 0. ••. Also. (u) is periodic. where sq is not a right zero divisor. then the condition is satisfied vacuously for every positive n and p = 1. RELATIONS BETWEEN THE RANK AND PERIOD In this section R is a commutative ring with identity 1. Let the Lucas sequence (u)^S(aj. Theorem 2. Let e. £. —. Therefore. = (I. suppose vt = 0 for v e R. . —. un = 0. x then */// = f//7 = d.Hence. .iUo= UQA". the Lucas sequence (u) is not periodic. a^l Corollary 1. un+k2 n only if A = U~QXUQ\W some te/?.2 upq = 0. Define [djj] q = D = U0AxUo1. let v be the period of (u). .. (x. Wklfe'iUpk1 = ^oup+k1 • 0. Conversely. ak) be of rankp > 0. Of 1) e /?*.p] (ii) (k. . That is. A is right cancellable. let A? = UQ} SU^with s = up+k1 • Then Apq = Uo1sqU0 and = 0. then Wp = 0. the last statement of the theorem is demonstrated. n = pq + \. Consequently. —. then APq = UQUQUQ = I and. I = Apq = U01sqU0t lsq = I. if An' = U'QUUO. The existence of p is now demonstrated.212 THE RANK AND PERIOD OF A LINEAR RECURRENT SEQUENCE OVER A RING f o r / = £7. D = Id and A = U^dilo. Ape'7 = U'o'up+kjUQe'j = WQ1 up+^j^ = U"01 ekup+k. suppose un = 0. Theurem 1]. by Lemma 2.v > 0. say. Av = UQ Wo and v= pq for some q. Finally. Proof. •. the desired p exists and is unique. t e R.js = 0 for / i=j. In this case. and ^et P be the least such n. sq .$)[&. then w/p= 0. The nonnegative integer p of Theorem 1 is called the rank of the Lucas sequence associated with (aj.k./ wk. (If /r = 1.
Moreover. Theorem 2] or Herrick [5. l / if 2\p/2 .P)[8. since $/(k. suppose (u) is periodic of period v. . ]32 if 2\p and j3= 4 if 2 j p . the relation between the rank and period of the Fibonacci sequence modulo a prime may now be given. Theorem 3]. and /32 if and only if 2p. Now. ^ = UJup+kfUo and = up+k^l ((1)k1ak)p = d e t ^ = (up+k. in particular. (ii) v = 2p if 2  p . = On the other hand since A<* . 2. Clearly p/2 )upH . Let the conditions be as in Theorem 2.) Also. ./. and suppose k is a prime. let 0. then v\p$. v= 2pif 2p/2 and v = pit 2{p/2. since [2. (See Barner [1.fj/p. Therefore. (up*M)v/P.p) = p/(kM Since 8/(b.) Corollary (i) v= 3. 1. Therefore. ak is of finite order if and only if Up+k. suppose p is even and let A be the companion matrix associated w i t h ^ = 7. by Corollary 3. then /3= 4 and. (u)\$ periodic if and only if ak is of finite order. k = 2 and 5 = 2.P) = [8. then Proof. 5. Let the Fibonacci sequence reduced modulo an odd prime be of rank p and period v.f is of finite order. The second part of (i) is an extension of a result of Ward [9] for modular integral sequences. j34. (See for example Wall [8. v.J)k then pfi = (k. 1. k[8. 3.Ap/2)Ap/2)Ap/2 = M]AP/2)AP = (ad]A Since the off diagonal elements of Ap/2 are not zero and are the negatives of the off diagonal elements of adj Ap/ * then it follows that Up+i = (1)P/2. Consequently. then (k.4p. This corollary includes several facts that have been previously observed for some special sequences. 2  p / 2 . Then if u[p%pJ/p M . Thus. by Theorem 1. 2. = (up+k.p] if u^kpJ/p = 1.p]/p. If pis odd. (See Vinson [7. and let 5 and j3be the orders of (1)k~1ak and Up+ki. Theorem 3]. (iii) P\k if and only if 5  p . a2 = 1./p ^/^ /2 ( _?. « . v= pj3. — be the sequence of Fibonacci numbers reduced modulo m > 2. 2 ( p / 2 .p]/p = 1. Proof. and 5 . P+T P Consequently. In this case. Therefore. (See also Robinson [6].lup+^lf . respectively.1)k .) Corollary 4. For example.pJ • Finally.2 . the order of ((Vk1ak)p is 5 /(8.p) = [8. In particular.1976] THE RANK AND PERIOD OF A LINEAR RECURRENT SEQUENCE OVER A RING 213 Since Ft is commutative. Let R be the ring of integers modulo an odd prime.p]/p. In particular 2\p. k = 2. (iii) i/= p if 2  p . Theorem 4]. Then (i) v = 4p if 2{p.Y T. in the unit group of R Since I = Av = <AP)v/p then §\vlp. (_1}p/2Ap/2 = mAP/2)AP/2= (M. In other words.$) is the order of uj$. Then (ii)0A:5. ax = a2 = 1. (ii) v = [&. by Theorem 2(i).) Corollary (i) 5  i .p] Let the conditions be as in Theorem 2. The first part of (i) in Theorem 2 is due to Carmichael [ 2 ] .
67 (1960). "The Modular Period of a Linear Integral Recurrence. Dade. Wall. 28." Quart J. 2. pp. 1973). R. and u2 = 1 implies either u = 1 or u = . "Periodicity Over the Ring of Matrices." The Fibonacci Quarterly. 1976 A slight extension of the foregoing argument provides another proof of the main theorem of Wyler [10]. D. 7. Math. Math. 72(1965). 3745. pp. 4. Ward. Then the sequence (log Fn ) is uniformly distributed mod 1 (abbreviated u. 180183. W. 343372. Barner. mod 1. 5. In fact. R. 0." The Fibonacci Quarterly. stands for the sequence of Fibonacci numbers. J. REFERENCES 1.1 . pp. 500506.214 THE RANK AND PERIOD OF A LINEAR RECURRENT SEQUENCE OVER A RING OCT. pp. 97104. 11. 1962.d. "Divisors of Recurrent Sequences. No. W. Vol. D.] . D. 6. 8. pp. 11. 48 (1920). "On SecondOrder Recurrences.. M. Robinson." Monatsh. 1 (Feb. Vol. p." Amer. Heine Angew. 525532. 5 (Dec. Valais. 1963). . Taussky. 214/215 (1964). See [ 2 ] . Math. Ward. 3. 1. "On the Periodicity of the Terminal Digits in the Fibonacci Sequence. Monthly. DeCarli. Carmichael. Theorem 4] is valid for every purely periodic secondorder Lucas sequence over a commutative ring with 1 satisfying the following two properties: 1 + 1 is not a zero divisor. "The Relation of the Period Modulo m to the Rank of Apparition of m in the Fibonacci Sequence. J.. L. 9." unpublished manuscript dated Europe. No. pp. Nowsetf/ n = l o g F / / j t .pp. E. 3. ******* LETTER TO THE EDITOR GENERALIZED FIBONACCI NUMBERS AND UNIFORM DISTRIBUTION MOD 1 L KUBPERS Mollens. Proof. D. pp. 535538. 10. Math. mod 1). Monthly. Math. 466468. No. "Fibonacci Series Modulo m. JulyOctober. 1973). 1. 2. 5 (Dec. "Zur FibonacciFolge modulop. Theorem 1. D. Let k be an integer different from 0. (Fn) or 1. C. Vol. "On Sequences of Integers Defined by Recurrence Relations. Robinson. which tends to [Continued on page 253. van der Corput: Let (un) be a sequence of real numbers. Then un+. "The Fibonacci Matrix Modulo m. (Feb. Vinson. Switzerland In the following I want to comment on a paper by William Webb concerning the distribution of the first digits of Fibonacci numbers [1] and to give a partial answer to some questions raised by the author. 2936. and M. 69 (1965). Wyler. restriction to Fibonaccirelated sequences makes it possible to obtain a number of results. 1." J. Wyler [10. G. D..un = log F'n§ ." The Fibonacci Quarterly. K. 5.d. Herrick. 1963)."Amer. pp. In fact." The Fibonacci Quarterly. then the sequence (un) is u. 0.log F1n/k = I log F ^± . 1. If Jim^ (un+1un) exists and is irrational. Vol. No. We apply a classic result of J.
As this paper shows. 6] have determined. by varying methods. a complete solution to it does not seem to be available. seems to have received less attention. It differs from the others in that it contains the inverse powers of 4p + /. They are ± (»"' ^ / ^ s i n h r / y cothnr ^ j$ r4p~1 ' *£ »»*<»">% (1)rHr4p~3 y . Kantorovich and Krylov in [2] proposed a method of solution but the suggested method contains.SAYER University of Bristol. it is cioselv related to the Riemann Zeta function $(4p + 1). England 1.2. Once this error is removed the method can be applied to solve the problem. Initially. Sandham and Watson in [3. however. Bristol. 5. which leads to the sums of the various series. The last series of the group. while Phillips. we state and solve the Neumann problem for the rectangle and then subsequently in Section 3 make appropriate use of the solution to obtain the various results. Their inversion. an error of principle. the sums of the series. liTRODUCTIOfy In this paper we are concerned with the summation of a number of series. All results are obtained by considering the Neumann problem for the rectangle.THE SUMS OF CERTAIN SERIES CONTAINING HYPERBOLIC FUNCTIONS F.P. therefore. Further. Certain of the above series have been extensively discussed in the past Results for particular values of/7 are given by Ramanujan in [ 4 ] . 3. as we shall show. 2. sums for general p. satisfy recursive relations containing binomial coefficients. where they are not identically zero. Although this problem is of an elementary nature and is in fact discussed in both contemporary and established literature on Laplace's equation. involves the Bernoulli and the lesser known Euler numbers.inhnr h and y (2r + V4p~1 ' ' % j ^ (2r + I)4***cosh (2r + 1)± y (1)r(2r+1)4p~1 h mhVr+Vf 24pcothrcoth2nrf ~ r=1 ' where p = 1. 215 . P=1 The recursive relations are themselves of interest and can be inverted. Thus if we write (~1)p(4p)! y ^4P122p2 f  T4p1 (1)r 4 P /ginhnr r then t(V 2 p ) V / " / n=1.
b) .216 THE SUMS OF CERTAIN SERIES CONTAINING HYPERBOLIC FUNCTIONS [OCT. F(y) and G(y) are known functions and the subscripts* and y are used to denote partial differentiation.2) and Ux(0. where U and V are functionsofxandy.1) ^xx +< for 0 < x < a.3). 2abA = f 0 (fg)dx .y) . 2.2) <t>y(x. We now briefly describe the method used by Kantorovich and Krylov in [2].Vy(x.5).1) to (2. where fix). a i.0) . t>yy .b) = OforO<x<a.b) = g(x)+2Av for O < x < a Vx(0. Vx(a. given functions £& Fand £ satisfying (2. Using (2. and therefore the functions U and V may not exist.4)  & ds = 0. while the functions (/and V each satisfy (2.3) 4>x(0.y) = G(y)2Aa for 0 < y < b. while Vsatisfies (2. Uy(x.4) the existence of U requires a f (fg)dx = 0.e. The condition (2. the existence of V requires b f o (FG)dy = 0.0) = f(x). <j)y(x. Yet the difficulty is readily overcome.O Uy(x.4) we require for the existence of U and V a f 0 and  g(x) + 2Ab . It is necessary for a solution that (2. the original Neumann problem is replaced by two other Neumann problems.yJ = F(y).y) = Ux(a.4) is equivalent to a (2.y) = F(y).yJ satisfying (2. We write 0 = A(x2y2) + U+K where A is some constant to be found. <px(a.0) = f(x).. THE NEUMANN PROBLEM FOR THE RECTANGLE This problem requires the determination of a function </>(x.1) and the further conditions: Ux{0.y) = £7 for 0<y < b.Vy(x.y)= Ux(a.0) = Vy(x. g(x). By virtue of (2. f c where c is the boundary of the rectangle. it does not necessarily follow that the integrals a b f(fg)dx and 0 f(FG)dy 0 are each zero.5) b f (fg)dx+ f 0 0 (FG)dy = 0. one for U and the other for V.fix) \dx = 0. It is evident that if we can find U and V we shall fulfill the conditions imposed on 0 by (2. We put $ = U + V.3) and Vy(x.0 0<y < h (2.1)# (2.1). However. 0 Likewise. d/dn denotes differentiation with respect to the outward normal to c and s refers to arc length. (2.b) = g(x) for O < x < a (2.y) = G(y) for 0 < y < b. We choose the function U so that it satisfies (2. Thus.
Gr (r= 0.3)1 Jin . 2.2. the Fourier coefficients fr and />are all zero. where (3. t4 In particular !r(1) = h1)»l 4M so that putting n= 1 in (3.2) and (2. APPLICATION OF THE SOLUTION TO THE NEUMANN PROBLEM We puta = i& = 7rand define functions (p(x. . while gr = Gr = lr(n) (r= 1.3.2)(4n .2) leads to the result (3. over the range 0 < x < a and Fr. we deduce for them that f(x) and Fly) are both identically zero.(„} )cosh/y cos^cosh/x cos/zl the C4n (n = 1. 3. Thus. In addition glx) = 2n { fr+ix)4n'1 + fr.3) we find E t i\r+l ~~V ( \ cos" ry cos rx + cosh rx cos ry § ' "FliShnr . 1. by 2$(x.5) shows that these two expressions for A are consistent Having found A.4n) = (x + /y)4n + (x (3.3) 4>(x. 2.F)dy.1876] THE SUMS OF CERTAIN SERIES CONTAINING HYPERBOLIC FUNCTIONS f { G(y)2Aa .ix)4n~f} and G(y) = 2n \ fr+iy)4n'1 + friy)4n~1} . 1.1) iy)4n. respectively.4) :: lr{n) = ( = ^ r 2 7r4n'322n(4n)(4n 1) + % (4n . —).1) and (3. •••) are the Fourier cosine coefficients of F(y) and G(y) over 0 <y <b. In fact. where n= 1.1 ^ — Jt icos^f . Further. •••) are the Fourier cosine coefficients for f(x) and #W.2..1)(4n .F(y) } dy = 0 or 2abA = f 0 217 (G . # r fr = 0. we can now follow the procedure given in [2] to determine U and V. It is readily verified that these functions satisfy (2.4n). • •) being constants which have yet to be determined.1) . using (2.3).6) 0 = A(x* y2) +1/2fQy+1/2F0x+ V —* = —± 1 cos r^ a nrslnh ^ r a °° b i£?rcosh ^ ~F r cosh 2E fexjA + V .yAn) = < * „ + £ !. 0 Equation (2. % nrsinh G5 * b where r"r. 2.1).y.2) lr(n) = Re4^ f [fr+ix)4n'1t fr ix)4n'1]eirxdx 0 using the result a 2abA = f (fg)dx o we find that the constants vanishes and hence with the help of (2. it can be verified directly that to within an arbitrary constant <£ is given by \gr cosh 3K = fr cosh QL (b y) \ (2.6) we can write (3. Successive integration by parts of (3.y.
Putting* = y = 0in (3.8).7) / 1 \n~An 02n (3 10)  *» .11) (x+iy)4n + (xiy)4n = 2c4n + 2 V lr(n) l£0.6) and (3. JihTl^&ilim*4l = ^n(y^an(y)+yPnM+qnM x2 c V2 _ V * / r (n) 4 Ld 4n I cosh ry coss rx + cosh rx cos ry f r sinh m .11) and simplifying we obtain (4n + 1)(4n + 2) ^rsinhnrZ* 7t4pi22p2r4p2 I ' *' ' . A t cosh V c p i / * * cosh rx cos / y ] J r=1 ^ + (4n + 1)(4n+2)(4n+3)(4n+4) J xPn(y) + Qn(y) + yPn (x) + qn (x).9) fan = lr(n + i) + (~1)*1 +4)(4n + 3)(4n+2)(4n + 1) ^ a2(n + 1) .6) that (3.8) a4p.10) and lr(n) by results (3. therefore. Hence equating to zero the coefficient of x2y2 we deduce with the aid of (3. qn(x). This is an identity.sh ry cos rx ^cosJLaf^o»flJ ** rsinhnr r=l where c4n is given by (3. P =. r=1 wherepn(x).6) [ r r * ^ 10 * aj¥T\ • r r4n2 J where.218 THE SUMS OF CERTAIN SERIES CONTAINING HYPERBOLIC FUNCTIONS [ 0 CT. (3.3).0ijkhr • Thus we have (3. . We integrate Eq. [4p4n_2 ) . Repeated application of (3. Noting the result contained in (3.9) we can write this equation in the alternative form <j>(x. n. for example./.4) yields lrM .3) and (3. These integrations will introduce arbitrary functions of x and y.y.5) to 0 = c4n+4 +^ p  (** .3) twice with respect to x and twice with respect to y.. Pn(y) and Qn(y) are arbitrary functions which may depend on n.Bx*y2 +y*cj x + (4n + 1){4n+2M4n+3)(4n + 4) <xPn(y) +Qn(y) + ypn(x) + qn(x)c4n ^f\ . We have.2. it follows a4p+2(n +U s (4n+4)(4n +3)(4n+2)(4n + 1)a4p2(n) and hence from (3. a2(n) = (1)n7T4n~322n4n(4n (3. Using this last result. We now proceed to find the constants C4n occurring in (3.(irl'41 2 * ^ 6 (3.2<n) = {ir*+i*4n4p+122n2p+2(4p2)! .4n+4) = T l Ir(„ + 1) + (=1^1 ^ (n + .c4n ^f\ This reduces with the help of (3.7) and more generally (3.1.
13) can be put in the alternative form 1 (4n+2)l Multiplying both sides of this equation by x 1J (4n+2)l " 2r LJ _ T~"% 2J T '4p1 (4p)!(4n+2~4p)! and summing from n = 1 to <*> yields (4p)!(4n+24p)! ) ^ (4p)l [ ) £ i (4k + 2)! After some manipulation we obtain °° V1 ^w (3.15) leads after some simplification to T «~' ' ^ Z ('ft**' 2 D(24p2c>1 ._ Upi This is the first of our results. To do this. Using the expansion of cosechx given in [1] Eq. we observe that (3.12) we deduce In a similar manner if we p u t * = y = 7rin (3.1976] THE SUMS OF CERT AIM SERIES COWTAIMIWG HYPERBOLIC FUf\SCTIOWS 219 Thus if we write oo (3.. We now show how this recursive relation can be inverted to give F ^ / in terms of the Bernoulli numbers. 2 P=1 where 2a = 1 + i.q .2 (us) n = lz.11) and define S4p~t by oo S4p1 = P 1 1 4p (1) ~ * .D ( J ) BqB2p. P = 1.1 ( 7)P(4p. q=0 It should be noted that BQ is taken as .—r^~ cosh x .14) A = UD1 77iT ^ f4/?// 7 ..2.2 2p+2 (4p)! Y ^^ " Ap1 then n (3.cos x = 1+ ^cosechcurcosech/cur. (3.! =  T L •1)r ~ .12) T4P.  sinh m then it follows T4Pi satisfies the recursive relation /= £ (4»4.1 while the Bernoulli numbers are defined here by ? Vj'i>""'*&e —/ p=i With the help of (3.16) £ [*£2)s4pt'*>(*!>+3) .
" (3. From (3.20) (4n4p2) £ Q *P1 = (4» + U(4n+2) { 1 + (~1)nH21'2n] . This leads to ^"U + ln^h2"1] (3. = (.6) and (3. Since (see [1]! 2p xz*xhx= £ r . E n {p=1 4p 1 where as before 2a .16) in the form V S4p ~1 L* (4n 4p + 2)!(4p)! = I / —L l—\ 2 \ (4n)! (4n + 2)! \ and following a procedure similar to that for T^p/ we find J^ > x4p S4Pi * .19) n (3. &4Pi can also be expressed in terms of the Bernoulli numbers.18) gives with the aid of (3.20) we deduce ±\»»r»>~\&\t^&\\±&& n=1 or. By writing (3.coth ax coth lax ./ / * V * jf^ p=0 we have <vp+q {%) *qB2P~q «pi**f: q=0 giving V JL.„ 2 1 cosh x + cos x .13) and (3./.19) % J * <2r+1)4p1 (3.If V 4 p * 122p+4(4p)! V .2 cos ax + 2\ " * TZ^Z—TZrz * coshxcos* ) . ~ *4p WW (4p)! ) (k=0 .220 THE SUMS OF CERTAIN SERIES CONTAINING HYPERBOLIC FUNCTIONS [OCT. y = n in (3./ 2 = .• (2r+V4p1 Writing Q4p. after some manipulation.16). .18) =T2l2r+l(n) U —±.11) and subtract from twice the result the expressions obtained by putting* = y = 0 and x = y = n. S4P~i can also be expressed in terms of the Bernoulli numbers.2 cosh ax .y . This is the third of the recursive relations and may be compared directly in form with (3. n\i\ [ } r=1 coth nr 4f>1 = P4p2_4p1 Z * i TT* (Dq+ Bg%2pg jLi (2q)!(4p2q)! ' q=Q We next put x = 0.1 + i.
.tanh ^ tanh '^. are the Euler numbers and E0 is taken as unity.iy) " ] = Y* ^P^1 " sinh m /sinh ry sin rx + sinh rx sin ry] r=1 Puttings = y = n/2 in (3.1)i[(x + iy) ~ 4n 2 . '(2g)!(4p2g)l Bg B B " *"> and hence by (3.(4p+2U ^ r . Following a procedure similar to earlier ones.21) can be expressed as •y ( coth y.1)(22«.2 coth ax coth fax . = (IPfail A Q 4P1 ^.26) and hence R4p+1 . we can deduce from (3. Recalling the expansions for coth x. £ £ E M.19) (3 a . f ^ ^(2r+1>l 2T1 (24p2q _ 1)(22q _ 1} The expression in (3.25) '^^EUAK* The quantities /?4 P j can be expressed in terms of the Euler numbers (see [1]).f " .s e c M X4 * X > „ + .i .2 cosec ax cosech fax . we obtain (3.11) can be differentiated as many times as we wish with respect t o * and y at points within the rectangle.23) and defining /?4P„3 by (3.(x .2). Differentiating once with respect to x and once with respect to y gives oo (3..« ^ ^ 7 .1) 2 .24) R4p.3 = / .<? (24P2«.1976] THE SUMS OF CERTAIN SERIES COWTAIWIWG HYPERBOLIC FUNCTIONS 221 The righthand side of (3.coth l~. —.(4P .23) 4n 2 (2n)(4n . 'jref "* 2 * 2" lp+1 Since oo sec* =E ^ x2« (2q)l ' q=0 where Elf E2.s V> / 11 .25) that 1.fa~ 2)1 E S ^ "T ~0 (2r+ D4"3 cosh ^ n4p322p1 +1)f yields (3. cosech x already used and noting that in [1] for tanhx we obtain after some manipulation n .
Putting* = 7 r .28) xy = r 2ft 2^ ™(~1) ~ —  sinh ry sin rx + sinh rx sin ry I i Hence differentiating once with respect t o * and then y we have. / = 0 in (3. Using the results of Section 2.sinh m A/// If we differentiate (3. i ™+ a* x+*n fax J4$T ~ 1 2 T i + „ „ ax T ~Jm Y T~ . .28) (2p + 1) times with respect to x and (2p + 1) times with respect to y then f o r * y = it/ 2 we find ± J ^ l l = O P=1.iy)4n+2 X J % ii>"+1P y2) + Y" (1)r ^ o s h ^ cos ry .. .29) and d e f i n i n g # 4 ^ / by (3.3D p s . nn+ /'ax z r p H sinh nr "1 i+ (vnHr2n> P=1 From the recurrence relation (3..7)n<n4n22n(x2 > i d H ^ l 92n2p+2.2.I (x + iy)4n+2 2 % + (x.THE SUMS OF CERTAIN SERIES CONTAINING HYPERBOLIC FUNCTIONS 222 00 /o 97 v V (3 27) (W 2p n X^ (1rEqE2pq „4p+U4p~3 [OCT.29) vanishes when x = y = 0..28) (2p) times with respect to x and 2p times with respect to y leads to f" (2r+1)4p~1(1)r % cosb(2r+1)Z =0 P= < 7 2 < <~" We now proceed to find the sum of the last of the series referred to in the Introduction.23) yields (3.1 J. • . t smh nr *^ r=/ Likewise differentiating (3.31) we deduce \T* M 2r M*P+I P=I and hence x p .cosh ry cos rx ! **> r sinh nr r=1 \ = (. on putting* = y = n/2 ~ tr+f '"ES.30) M4p+1 = (1)p+1^4p"12'2p+2(4p)! leads to £ / * M / * coshr^ = J2 M*p+i {4n4p2) (3.n1 9 n S ^ / ^ W ^ / .. ? * ' h <*»«!»!>' p' °'l2'  2 Putting /? = / i n (3. it can be shown for n = 1. 2.AnU (4n+2\\ r o=1 P=1 ) The constant appearing in the Neumann solution is determined here to be zero by observing that each side of (3.
r*/ u * °° ( coth^2"" 4p coth2ttr ) y ^ f/A # cosh HT* / . I am told by Professor H. Edited by G.(if gg 2p2 E 2 (2mP22S~+2). Ramanujan.16). Boordhoff Ltd.17) can be deduced from (3. 1927. H. London Math. is contained. Math. ACKNOWLEDGEMENTS I am indebted to Professor L." Proc. Ltd. itkkkkkk . 2. P. 3. H. 430436. Benster). pp. Cambridge University Press. Collected Papers. 1 I'A. 3 (1928). 4. Society. pp. et al. G. 4 (1929). Watson. 297299." Journal London Math. in one of Ramanuj:. I also thank Bruce Berndt for drawing my attention to the papers by Phillips and Sandham. 6. "Note on Summation of Series. Halberstam. It also follows for/7 > 3 we can obtain a good approximation to .30). G. Netherlands (1958). R. REFERENCES 1. "Theorems Stated by Ramanujan (II). Sandham. pp. Second Edition. Kantorovich and Krylov. N. in a somewhat disguised form.v s notebooks. Phillips. 5 (1954). This result. Approximate Methods ofHigher Analysis (translated by Curtis D. 223 B B2 ° ^° Since 00 / .. obtain the required sum. IVIacMillan & Co. 114116. Amer. . pp. Bromwich. Soc. S. Gronigen.1976] THE SUMS OF CERTAIW SERIES CONTAINING HYPERBOLIC FUNCTIONS h1)S »++. Hardy.V 1 } 2 f ~i r**'sinhnr ~ %1 } &+1 \ we can. F." J. "Some Infinite Series. An Introduction to the Theory of Infinite Series.cothJJ Za 4p+1 in terms of the Bernoulli numbers. Carlitz for pointing out to me how the result (3. pp. 5.. 79. noting (3. 216225. (1926). Society. E.
If the generating function of Sx is CM. so that when/=4. we have every fourth column of the convolution array for the Catalan numbers. leading to interrelationships between the sequences Sj. — }• appears. or.j= 5. 2. 0. Notice that S_3 contains the elements of the first convolution of S2. For/ = 2. 1. then the generating function for SL3 is MC(x).i . the rows of Pascal's triangle appear on and below the main diagonal. We form a series of nxn matrices P/j by placing every /^column of the convolution triangle for the sequenceS/ on and below the main diagonal. —. taken with alternating signs. As a second example.2 . —I which contains the Catalan numbers or Sx. the seventh columns of the convolution array for£ 4 . we use convolution arrays of the sequences S/ to form a sequence of matrix inverses. in the new notation. . 1 . which are the reciprocals of the column generators for Pascal's triangle. we have every eighth column of the array for£ 4 .1.3 0 .3 for the form ®fP$j verifies that these results continue. 0.2. every third column of the array forS 0 . for Pij. These results continue for P^j in Table 1. we return to the sequences S/ arising from the first column of PJ1 as in [1]. we have every column of the convolution triangle for the sequence $~3 = 4 1. are (1 .j. 1.SEQUENCES OF MATRIX INVERSES FROM PASCAL. AND MARJORIE BICKNELL San Jose State University. If the generating function for£ 2 is Dfx).J . AND RELATED CONVOLUTION ARRAYS V. the sixth columns of the array for£ 3 . disregarding signs. Now. every column of the eonvolution array for the sequence S_3 = 4 1. E.6. HCIGGATT. and zeroes elsewhere. Inspecting Table 1. 14. these columns are also the columns of the convolution triangle for the sequenceS~f = j 1. the matrix Pii3 would contain every third column of the convolution array for the Catalan sequence Sx written in triangular form. for/ = 3. . For/ . When/25 1. PjiPoj.4 gives fourth columns for S2. for/ = 2.2. and for/= 8.1. The proofs involve generating functions for the columns of infinite matrices. Here. ••• I.5 .4 2 . 1 . . then the generating function for£„ 3 is 1/D2(x). California 95192 A sequence of sequences 5/ arising from the first column of matrix inverses of matrices containing certain columns of Pascal's triangle provided a fruitful study in [1]. which is Pascal's triangle. 224 . j = 3. we look at the form of the elements of each matrix inverse. every third column of Pascal's triangle. For/ = 1. SEQUENCES OF MATRIX INVERSES In this paper.1 . JR. we have every second column of the convolution array for S„2. We call the inverse ofPfj the matrix PjJ and record these inverses in the tables that follow.. and f o r / * 7. we have alternate columns of Pascal's triangle. . every fourth column of the array for£ 0 . First. alternating signs included. When / = 1. every third column of the array forSLj//= 4. let us analyze the results.) For/ = 3f we ha?/e every third column of the convolution triangle forSit while/. the fifth columns of the array forS 2 . notice that each array contains columns of the convolution array for its leftmost column. (In fact. following the initial term. and have diverse applications. where we do not adjust for the triangular form. of/ every/^column of the convolution array for the sequence So =  1 . CATALAN. or of S22! taken with alternating signs and with one additional term preceding the sequence.. . The column generators. San Jose. Then. to relate to the matrix/5/ from [1] which was formed by writing the// + 1)st columns of Pascal's triangle on and below the main diagonal. we have every second column of the array for S . . or alternate columns of the convolution triangle for S0.x) . oxSx. 0. disregarding the alternating signs of the array. 7. These results continue. Then.
AND RELATED CONVOLUTION ARRAYS 225 Table 1.1 7 1 1 1 2 5 1 5 20 9 1 14 75 54 13 1 42 275 273 104 17 1 .0 NonZero Elements of the Matrices PQ.1 1 Nion*Zero Elements of P~* and/*/. ! / f 1 1 0 0 0 1 2 1 0 PlJ hi 1 1 1 3 3 1 4 1 1 1 1 1 1 3 6 10 1 5 15 1 7 1 1 2 5 14 1 4 14 48 1 7 35 1 10 1 1 3 12 BB 273 1 1 1 2 5 14 1 2 5 14 1 1 1 2 3 5 9 14 28 1 1 1 2 4 5 14 14 48 1 3 9 1 4 1 1 5 20 1 7 1 1 7 35 1 10 1 1 1 5 25 130 1 9 63 700 . CATALAN.4 0 8 1 13 1 117 .SEQUENCES OF MATRIX INVERSES FROM PASCAL.and PQJ pD~1 'to P oJ hj 1 1 1 1 1 1 1 2 1 3 3 4 6 B 10 1 1 2 B 14 1 3 9 1 5 20 m 1 1 3 12 BB 1 1 4 22 140 1 4 18 88 1 7 42 1 B 30 1 9 72 200 1 4 10 1 7 1 10 1 13 1 5 1 1 1 1 1 1 1 1 1 2 3 4 S 1 3 6 10 1 4 10 5 1 1 1 1 1 1 3 6 10 1 5 15 1 7 1 1 4 10 20 1 7 28 1 10 1 1 5 15 35 1 9 45 1 13 1 1 1 1 1 1 •1 1 1 1 1 1 1 1 1 Table 1.
[OCT.SEQUENCES OF MATRIX INVERSES FROM PASCAL.j J 1 4 1 1 1 3 12 55 1 2 7 30 1 3 12 1 4 1 1 7 1 1 1 3 12 55 1 3 12 55 1 5 25 1 7 1 1 1 1 3 12 55 1 4 18 88 1 7 42 1 10 1 1 1 1 1 3 5 12 25 55 130 1 9 1 63 13 1 6 1 33 11 1 182 88 16 1 1 1 1 3 6 12 33 55 182 1 11 1 88 16 1 1 2 1 2 1 3 3 1 1 4 10 20 1 5 20 75 1 3 0 1 5 10 1 7 1 28 10 1 1 9 54 13 1 1 1 1 1 1 3 7 1 T2 42 13 1 55 245 117 19 1 1 1 1 1 1 1 3 8 12 52 15 1 55 320 150 22 1 6 4 7 1 1 22 49 13 140 357 130 19 1 1 7 5 8 1 35 68 15 1 285 606 270 22 1 1 . Table 1.2 NonZero Elements of P^j and P2j H 1 1 1 1 2 5 1 1 2 0 0 0 I 1 3 1 1 1 1 1 4 2 5 14 1 1 5 3 12 55 2.
/ > 0.and P3J p 1 sJ 227 1 3 1 3 15 1 4 1 1 5 30 1 7 1 1 7 1 49 10 1 1 1 9 72 13 1 To generalize. AND RELATED CONVOLUTION ARRAYS p 3j 1 2 3 10 3 1 4 1 1 3 0 1 1 5 5 1 7 1 1 4 6 4 1 7 1 21 10 1 1 5 15 35 1 9 1 45 13 1 1 1 1 4 2 9 22 140 52 1 1 1 3 4 22 15 140 91 1 1 1 4 4 22 22 140 140 1 1 1 4 5 22 30 140 200 I 1 1 2 6 1 5 27 11 1 14 110 77 16 1 1 1 1 4 6 22 39 140 272 1 11 1 99 16 1 1 1 1 3 1 7 12 42 13 1 55 245 117 19 1 1 1 1 4 7 22 49 140 357 1 13 1 130 19 1 /= 7 1 1 1 1 4 8 60 15 1 22 140 456 165 22 1 1 1 1 4 8 1 22 60 15 1 140 456 165 22 1 /= 1 1 1 1 9 5 1 35 81 17 285 759 221 25 1 1 1 1 1 4 9 22 72 17 1 140 570 204 25 1 /= 1 /= 2 /= 5 1 1 2 7 30 1 1 1 2 5 1 1 0 0 0 1 1 1 1 1 Fable 1.by ...3 NonZero Elements of P^. zeroes everywhere above its main diagonal.1976] CATALAN. on and below its main diagonal. P"jj contains the sequence S/// along its first column and the/ f/? columns of the convolution triangle for the sequence £///. / > 2. with. of course. taken with alternating signs. The sequences S/../. The sequences S. are all related to the sequences S. were explored in [1].
and 1. For example P'J^ contained the same elements as Pft3 except for the alternating signs./ . Then the row sums are R(x) = SjM+xS'f'M+xZsf*1 *••• = Sj(x)/[1xSlM] by summing the infinite geometric series. LetSffx) denote the generating function for the sequence Sf. using 5 x 5 matrices. so that R(x) = Sf (x) upon simplication. There are other patterns which occur for the matrices P~Jm Except for the alternating signs. •••. is G(x). [OCT. but beginning from the first convolution.2. so that the kth column o f P / ^ i s t h e (3k + 1)st column of the array. with S(O) = I It will follow that . if we form the matrix P*'*3 from the (3k + 2)nd columns of the convolution array forSff P*f*31 has the same elements as P^*3 taken with alternating signs.30.42. •••} . Form ^ j to contain every third column of the convolution array f o r S / . or SJ. then the generating function for 5 . We easily prove that Theorem. Proof. i> 2. i> 1.90 110 54 12 1 " 1 13  7  = 7 1 2 5 14 _ 42 r = 1 3 9 28 I 90 0 1 5 20 75 0 1 6 27 110 0 0 1 8 44 0 0 0 1 Q 0 0 0 11 i j 0 0 1 9 54 0 0 0 1 12 0 0 0 0 1 Notice that we can consider n x n submatrices of the infinite matrices of this paper. PROOF OF RESULTS AND FURTHER APPLICATIONS Now. 14. Furthermore. But.  I. . row sums of P. if we knowthat/4# = £ b y generating functions. if the initial one is deleted. For example. has a generating function which is the reciprocal of that for Sn./ is identical to the (i . 320. p*1 1.t h e r o w s u m s o f / ^ are j 1.143. 2. Let A be an infinite matrix such that all of its nonzero elements appear on and below its main diagonal. 7. . if the generating function for S. since for infinite matrices A. 9. by [ 1 ] . Let S. because each n x n matrix is the same as the n x n block in the upper left in the respective infinite matrix. 5. and C. j = 2i + 1. Lemma. sixth. taken with alternating signs. then it must follow that AB = C forn x n matrices^. 2. 2. . o r S ^ .. We write the Lemma. / BSiMxSpUx). and P*f3 has the same elements ^p\s. third. and the row sums of P33 are  l . a n d l e t ^ x / 7 bethenxn matrix formed from the upper left corner of A Let B and C be infinite matrices with Bnxn and Cnxn the/7 x n matrices formed from their respective upper left corners. If AB = C. andS_. where Pff3 contained the zeroeth.(x) be the generating function for the sequence S/. The successive.j f o r / = 2i + 1.228 SEQUENCES OF MATRIX INVERSES FROM PASCAL. Also. B. 1. notice that the row sums of Pij are  1. and let Sk I Sk~i mean that5^7 is the solution to Sk(x/s(x» = S(x). the sequence £_. and C. columns of the convolution triangle for Sf so that its k**1 column was the (3k)th column of the convolution array.1. s. 52.\% identical to P.2h& convolution of the sequence S/_y./ \%1/G'(x). or SI. this property still holds if we form Pfj from any set ofjth columns of the convolution triangle for S. we establish firmly the matrix inverse results of this paper. then Anxn&nXn = CnXn> Returning for a moment to Tables 1.3 P ~ 1 2 0 1 0 0 0 0 0 0 5 14 5 20 1 8 0 1 0 0 _42 75 44 11 1_ 3 0 1 0 0 0 0 0" 0 9 28 6 27 1 9 0 1 0 0 L.1 = stf so that. Similarly.j are sf. 2. B. Pjj.3.
but this continues as 1 Sjx/1/SJx)) 1 SJx)' and thus we can generally say 1 Sm(x) Notice that.x3g(x)B3(x). if S0 (x) = 1/(1 . fm+k(x).x2g(x)B2(x). AMD RELATED CONVOLUTION ARRAYS 229 <u>2 I w j 4WQ I w_ j i 6 _ Now. then Sn(x) satisfies / 1xSnn(x) S„M 1 Sm+1(x) or Sn(x)xS^7(x) 1= Now.(x) From this we can show 1 Sjx) as follows: 1+x. {S%M. •) That is.S_2(x) = l/SJxhS. Am+2k(x). ) and (Am(x).xg(x)B(x). are inverses if g(x)B(x)A(xB(x))f(xB(x)) = 1. Am+k(x).SllM.~)1 is given by (A(x)/A6(x)/A11(x)/h where SJx/1/AUx)) = 1/A(x) so we go down five sequences from S2(x): S2(x). xf(x)A(x). x2f(x)A2(x). r i 1 0 0 0 6 1 0 0 33 11 1 0 182 88 18 1 o o cr 1 1 0 3 . 1/A(x) is k steps down the descending chain of sequences from f(x). x3f(x)A3 (x).) The two infinite matrices (fm(x).6 1 [12 3311 55 182 88 o 0 0 0 0 1 0 16 U Lemma. it turns out that s >Masfcj = 1 .S*M. 1 * Sjx) SjxSjx)) * S2(x) = S2(x) is trivial. J .x).S_Jx) = 1/SJx). Let us examine the two together.Jx) = 1/S2(x) so that 1/AM = SJx).Sx(x). are matrix inverses if A(x)f(xAk(x)) = 1 or f(x/[1/Ak(x)]) = 1/A(x). fm+2k(x). ) and (g(x). This verifies that 1 1 3 12 55 o o o oi".1876] CATALAN. let us look at our general (Pascal) problem. Two infinite matrices (f(x).S0(x). (We denote each matrix by giving successive column generators.
then glx)Bm(x)Am(x(B(x))f(xBM) = 1 so it seems to naturally break into two separate parts: (1) B(x)A(x(B(x)) = 1 (2) gMf(xBM) = 1. Generally speaking. upon proper processing. let S* =  1. This matrix uses elements from the central column of Pascal's triangle and the columns parallel to it.5 1 If we now go to themth columns.(k+1)xSkk~1M = t±± gM = 14xSjx) = 4/SJx)3. 6.3 \ 0 2 \ 0 0 0 0 1 4 5 .4. originally gave us the Catalan sequence. 55.1. we use / = Sjx)xSjx) and we can write fM = Sjx). which. BM = 1/SdM. 20.(1^14x ' )/2x. 14. 1 2 6 10 20 0 1 3 10 35 0 0 0\~.J. [OCT. 28. the Catalan sequence. A(x) . g(x) = 1 2x. lead to our sequence^ = J 1. 84.J. 2x(1x) 1x B(x) —± = f B(x) = 1 x. Let us consider S* = \ 1. . AM = SJx). 220. 1. LztS*(x) be the generating function forS*.230 SEQUENCES OF MATRIX INVERSES FROM PASCAL. j 1. 12. Consider the central column of Pascal's triangle. but (2) is something new when combined with (1) since the above has to hold for ail m ^ 0 . There are several interesting applications. — } . we take fM = S*kM. 0 0 0 \ 1 0 0 = 4 1 0 / 56 5 1 / / 1 0 / 2 1 0 . 3.2. 2. gM = 1 . where we already know how to solve (1). and BM = 1/Sk(x). the diagonal of Pascal's triangle which led to the sequence S3. 273. AfxJ = Sjx)f and let Sl(x) = C(x) be the generating function for the sequence Sx =  1. Let fM he the generating function for the central column of Pascal's triangle. 15. so that B(x)A(x(B(x)) = /._1_ _ * " CM  1 CM 3 _2 Sjx) ^ I Sjx) = Now. —J.5.. for the 5 x 5 case. and has inverse whose columns have the coefficients from the generating functions. 70. Here. For example. AM = Sk(x). and take fM = 1/sJ14x.k. upon proper processing. . 3.Now f(x(B(x)) = 77^=r7l^=sr= so thatg(x)f(xB(x)) = 1 also. the diagonal of Pascal's triangle. Then. gM = 13xC(x) B(x) = 1xC(x) .   which. and take f(x) = Sjx). .
••) and (G(x). 231 The two infinite matrices m (f(x)A (x). G(x) = g(x)Bm(x). 1. ~).3.. 8. With application to the sequences S/ of this paper.1976] CATALAW. J. 5594. 16. 168). 1/Sk^(x). where Fix) = f(x)Xn(x).1 which is generated by n=0 Let Hfix) = S(x).given by Carlitz [4].yJ1 + 6x+~x~~*)/2.2 = 0. g(x) = (3 + x . A (x) =11+ x)/(1 . MO) = B(O) = 1.. 1. .3(2n . p. 4. AND RELATED CONVOLUTION ARRAYS Lemma. B(x) = [~(1+x) + s/1 + This arises from the triangular matrix (from a paper by Alladi [61) 1 1 1 1 3 1 5 1 7 1 5 1 13 7 Y u 1 V x =u+ v+y where the column generators are successively given by / 1x' x(lfx) Ux)\ ' xn(1+x)n '"' (1Xf+1' The lemmas of this section also apply to some other interesting sequences. simultaneously. The Lemma is the same as considering the two infinite matrices (F(x).11.1. .x2F(x)A2k(x). x2g(x)Bm+2k(x). we can take fix) = DofxK g(x) = 1(k+ 1)Skk~1(x). (See Riordan [2]. .3. Ak(xBk(x))f(xBk(x))Am(xBk(x))Bm(x)g(x) = 1 or [A(xBk(x))B(x)]m = 1 and f(xBk(x))g(x) = 1.obeys In + 1)hn . and Blx) A(x) = Sk(x)..903. while H ir^k)= which is the generating function for the sequence H(g(x. J are inverses \if(xBk(x))g(x) = 1 and A(xBk(x))B(x) = 1.45.x).x).4.. 126.xf(x)Am+k(x). The above lemma can also be illustrated by taking f(x) = 1/(1 .xG(x)Bk(x).xF(x)Ak(x). bn. Then H* f(x) = S(x) means that f(xS2(x)) = S(x).1)hn1 + (n .11. Suppose we take the sequence J1. 2.2)bn. There is another sequence from the same article by Carlitz. S(O) = I which is the generating function for  1. J.) and (g(xBm(x).x2G(x)B2k(x).xg(x)BmH(x). x*f(x)Am+2k(x).21. but first we note^ =  l . \.45. 818. and 6x+x2]/2x. 1.197. where S(O) = 1 and S(x) satisfies f(xS(x)) = S(x). = hix) >  l .
1658. "The Generating Functions for the Vertical Columns of (N + 1)Nomial Triangles.x)S(x) . Bruckman. . No. We also note that. which is such a prolific result that it is the content of another paper [ 3 ] . c 1+x )/2xisthe iTih. Hoggatt. 3. p. S = { 1.2 . if Pjjte the matrix formed by moving each column oiPjj up to form a rectangular array. S = { 1. REFERENCES 1. 2 2xS S2 = { 1. 14. 11. 242. = { 0. CATALAN. Carlitz. E. 516. Let f(x) generate the odd numbers.494. where S(x) generates  1. Jr. Private Communication. 9. Then the solution to f(x($(x)) = S(x). .3294.494. Vol. . where C(x)= (1 . 168. 6. PjjP*j contains Pascal's triangle written in rectangular form. 3. 14.. 14.\ . 135142. 3 S = { 1.given by Carlitz [ 4 ] . 9. we should be able to establish the recurrence. Jr. j . 4." the publication dates of references 1 and 2 were inadvertently interchanged..2xS2 (x) + (1.37. 1. "Pascal. . which has generating functions 1+x (1x)2 1 +3x + 5x2 +7xz 1+xS(x) (1xS(x))2 = " *. V. Jr.14. .1. U S ISO. No. No..1 2 .\ with all vertical sums equalling zero." unpublished Master's Thesis. Krishnaswami Alladi. 357367. and General Sequence Convolution Arrays in a Matrix. 5. 0.3294.7 4 . S(O) = 7." The Fibonacci Quarterly. and Marjorie Bicknell. No. Reference 4 should have read: Michael Rondeau. is the sequence  1. Catalan. 3 x*S \ = { 0.} xS = { 0.1 . With the methods of this section.1 5 ° = { . Paul Bruckman has proved the matrix theorems used in this section in [ 5 ] . 7. and Marjorie Bicknell. 1973). 1976).2xS2 (x) +x2S3 (x) 0 = x2S3 (x) .3. . NOTE: In the paper. V. See [7] also. . 4 (December 1975).3 . 14. "The HConvolution Transform. pp. Hoggatt.. 79. 2." The Fibonacci Quarterly.SEQUENCES OF MATRIX INVERSES FROM PASCAL.J . . E. 0.3. ••}. 13. San Jose State University. Hoggatt." The Fibonacci Quarterly. "Enumeration of TwoLine Arrays. As verification. 2 (April. 3294.52 n=*0 From this. 79. 4 (Dec.7 9 . and General Sequence Convolution Arrays in a Matrix. and Paul S. "Pascal. John Wiley and Sons.» <w . 3316.) «" \ (1+x) 2 which comes from Riordan [ 2 ] .1 . Private Communication 6. Vol. 0. 69. 69. it is easily shown that. 3294. 2 (April.4 9 4 . John Riordan. AND RELATED CONVOLUTION ARRAYS 232 OCT. pp. 0. December.1 4 . . . .494. 1968. to appear. L. • } . 14. 516. . . Vbl. 0. 79. 0. pp." The Fibonacci Quarterly. 3. •• \ . Combinatorial Analysis.sJ1 4x generator for the Catalan sequence.4 8 4 . 1973.494. 1976 We solve the quadratic 2xS2 (x+DS+1 =0 + 2 S(x) = 1 *£j1+2x+x 4x 8x SM= / + w / 4xf r + * L . Catalap.. Vol. . "Catalan and Related Sequences Arising from Inverses of Pascal's Triangle Matrices. 79. 1976). There is another application. •••} . . E. b{Xi 1 + xS(x) = S(x) . Paul Bruckman. V.
Since (1) P (z) = kzk~2 233 (zkf±) . —. Table 2 gives approximate values of these \ e e= 0. limits of quotients of terms.g.Zk~1 . Let k > 1. Theorem annuli A.1 are distinct and lie in the intersection of the two X0 < \Z\ < \ and t Xx .j < k.ON A GENERALIZATION OF THE FIBONACCI NUMBERS USEFUL IN MEMORY ALLOCATION SCHEMA. expressing the elements in terms of certain sums of binomial coefficients and sums of powers of roots. and another expression giving the numbers as sums of binomial coefficients which lie on a line of rational slope falling across Pascal's triangle. all of the zeroes of are rational.1. The generalized numbers may be expressed as the coefficients of a rational generating function where the denominator of the rational function involves the trinomial Zk Z ~ . Complete information about the sequence then follows.. None of the zeroes of p(Z) are rational. 100. Tables useful for storage design are included. A certain linear recurrence relation arises in the theory of memory allocation schema which generalizes the linear recurrence defining the Fibonacci numbers. one in terms of linear combinations of the powers of the roots of the trinomial. and limits of roots. From this fact follows two expressions for the numbers themselves.nk / fkj lim < 1< \ and = 0.1. k= 1. k > 1.2. f^ = 1. e.Z "f . = \e(k). This latter data depends upon the location of the roots of this trinomial: all complex zeroes lie in the intersection of two annuli in the complex plane. All of the k zeroes of Z . A theorem is proven showing that all the zeroes of this polynomial lie in the intersection of two annuli. we number the lemmas accordingly. The proofs of these theorems depend upon two sequences of lemmas. k > 0 HELAMAW ROLFE PRATT FERGUSON Brigham Young University.7 < \Z 1  < / + X0 .k>\. Then lim XJk) = 1.Zk~1 .1.1. Define fkff1 = fken.Zk~1 . Lemma (1) p (Z) Al.1 = 0. where Xe= \e(k) is the largest (positive) real solution of rk + (1)erk'1 . / + fk. See Table 1 and Figure 1 for explicit numbers and visulization of the following central theorems. OR ALL ABOUT THE ZEROES OF Zk . Proof.1. Prove. Theorem B. The associated linear recurrence relation involves the polynomial Zk . The former expression gives complete information on the limit of successive quotients. Utah 34602 ABSTRACT A generalization of the Fibonacci numbers arises in the theory of dynamic storage allocation schema. 20. 0 < \ thpt!' = \(k) < 2. let p(Z) = Zk . Let k > 1. those bearing more directly upon Theorem A or B.
234
ON A GENERALIZATION OF THE FIBONACCI NUMBERS
Figure 1. The Two Annuli Theorem
(The shaded region represents the region in which all of the complex zeroes of Zk 
[OCT.
Zk~1
1 must lie.)
1976]
USEFUL IU MEIVfORY ALLOCATIOW SCHEMA
Table 1
The Sequences f^/7 = h,r
WithfkJ=
nk
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
nk
0,j <k; fkrk = 1,k > 1
2
3
4
5
131072
1597
262144
2584
524288
4181
0
0
0
0
1
1
1
1
2
3
4
5
7
10
14
19
26
36
50
69
95
131
181
250
345
476
657
907
0
0
0
0
0
32768
65536
0
0
1
1
2
3
5
8
13
21
34
55
89
144
233
377
610
987
1
1048576
6765
2097152
10946
0
0
0
1
1
1
2
3
4
6
9
13
19
28
41
60
88
129
189
277
406
595
872
4194304
17711
1278
8388608
28657
1873
0
1
2
4
8
16
32
64
128
256
512
1024
2048
4096
8192
16384
16777216
46368 2745
33554432
75025 4023
67108864 121393 5896
134217728 196418 8641
1252
268435456 317811 12664
536870912 514299 18560
2385
1073741824 832040 27201
3292
1728
2
3
4
5
6
8
11
15
20
26
34
45
60
80
106
140
185
245
325
431
571
756
6
1
0
0
0
0
0
0
0
0
0
0
0
0
0
8
0
0
0
0
0
0
0
0
2
3
2
4
3
5
4 2
6
5 3
7
6 4
9
7 5
12
8 6
16
21 10 7
27 13 8
34 17 9
43 22 11
55 28 14
71 35 18
92 43 23
119 53 29
153 66 36
196 83 44
251 105 53
9
0
0
0
0
0
0
0
0
0
2
3
4
5
6
7
8
9
10
12
15
19
24
30
10
11
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
2
3
4
5
6
7
8
9
10
11
13
16
2
3
4
5
6
7
8
9
10
11
236
ON A GENERALIZATION OF THE FIBONACCI NUMBERS
Table 2
\€ = \€(k), e = 0,1 is the Largest Positive Real Root of rk + (l)erk~1
The roots are truncated to 25 decimal places; see [3].
k
~
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
100
\(k)

I
\(k)
2.0000000000000000000000000
1.618033988749894B482045868
1.4655712318767680266567312
1.3802775690976141156733016
1.3247179572447460259609088
1.2851990332453493679072604
1.255422871076846543 2050014
1.2320546314285722959319676
1.2131497230596399145540815
1.1974914335516807746915412
1.1842763223508938723515139
1.1729507500239802071448788
1.1631197906692044180088153
1.1544935507090564328867379
1.1468540421995067272864110
1.1400339374770049101652704
1.1339024903348373489121350
1.1283559396916029856471042
1.1233108062463267587889592
1.1186991080522260494554442
1.034
Figure 2. Combined graph of xk  xk~1
[OCT.
0.0000000000
~
0.6180339887498948482045868
0.7548776662466927600495088
0.8191725133961644396995711
0.8566748838545028748523248
0.8812714616335695944076491
0.8986537126286992932608757
0.9115923534820549186286736
0.9215993196339830062994303
0.9295701282320228642044130
0.9360691110777583783971914
0.9414696173216352043780467
0.9460285282856136156355381
0.9499283999636198830314051
0.9533025374016641591079826
0.9562505576379890668254960
0.9588484010075613716652026
0.9611549719964985735216646
0.9632166633389015467989664
0.9650705109167162350928078
0.9930
1 = y for k even and odd. There is a local minimum atx =
1976] USEFUL IN MEMORY ALLOCATION SCHEMA 237 we see that the roots of piJ(Z) are 0 with multiplicity k2and (k .j+i • Proof. There exist numbers. Proof. such that all of the zeroes of p(Z) Z ./+/' /= 1. Comparing radii of these circles there will be a nonvacuous intersection i f r < 1+r1~k o r i f r < X j where \ is the largest positive root of p(Z). Combined Graph ofxk + xk~11 = y for k Even and Odd.k)/k. Locating these roots gives the inequalities above and noting that \1~k = 1 + X 0 X f " * = \x .1 < 1 + X0 centered at 1./ j . From Lemma A1 we have proven that p(Z) and p (Z) are relatively prime (C is algebraically closed) which is sufficient for the roots to be distinct. Since p(Q) $ 0. We note that in addition to nonreal complex zeroes occurring in conjugate pairs. 0 < X0 < 1 < X.1\ = r1~k\ Thus any zero lies on the intersection of the two circles Z = rand \Z . The second case of \Z\=r lying inside \Z. Corollary Al.k1 with equality iff \/<j is the complex conjugate of \k.1 lie in an annulusX0 < \Z\ <X X centered atO and in an annulusXi . Since p(Z) is rnonic with integer coefficients any rational root must be agaussian integer. From the relation Zk'Uz . . the roots are distinct. There is a local maximum and minimum a t * = (1 .1 orr> X0 where X0 is the largest positive root of q(Z) = Zk + Z ~1~1. exactly two roots are real if k is even and exactly one is real if k is odd.. FigureS.1 <\Z.1\ = r1"k with fixed centers.1)= 1 it is easy to infer that Z cannot be integral. Lemma A2. Define the collection of zeroes of p(Z) to be Then [Z^] = k.2.1 bounds the radius 1~k Corollary A2. and we can order them \KA < l**. There are two cases of empty intersection: one circle lying wholly inside the other.. Then Xj(k) is real and  X ^ /  <\i(k) for 1 <j<k.rr""^ yields 0 < rk + rk~1 .e. i. k > 1.Z ~1 .l)/k\N\\\\ multiplicity 1. any complex zero Z of p(Z) has norm Z = r > 0 andp(Z) = 0 gives \Z. both rational. (!). S e t X ^ ^ ^ X / ^ ^ X y . < 2 dependent only upon k.
*• 1<J<k Proof The sum telescopes. q(7) = 7 < k . c Yl of \ i\the nth powers\ l\ be any (complex) linear combination of the zeroes ofp(Z).1~r so that kin I 1T) < Ink. This follows directly from Corollary A2 and the usual absolute value inequalities. Xt is. The purpose of this simple Lemma is to motivate the next Lemma. Let k > 3. forp^fx)=xk xk'! . §'mep(1) = 7 we need only show that p(k1/k)>Q. Then£~///: < \<k) < 1. the largest positive real zero of the sum is k . Lemma A 5. Then ><ifc<' But this gives 0 <kk1(lM1 = p(k1/k). Note that if k is even that . It is clear that k1/k 7. we have ln[kfl) >lnk1/k. Lemma A4.7. 1+ X) PjM*xk . Then./ for k an integer larger than two. But then / + . A = c < k max L erf} < A\1 .X 0 is the smallest real root oip(Z). Then Proof \<\t(k)<k1/k. Let* >2.238 ON A GENERALIZATION OF THE FIBONACCI NUMBERS [OCT. Proof For q(z) = xk+xk'11/k it is sufficient to show that q(k~ ) k1/k <k gives < 0. Lemma A6. This Lemma gives information on the rate of growth of the integers f^n. exp is order preserving so that 1l>k~1/k. Rewriting. from the proof of Lemima A2 the largest possible real root oip(Z). Proof. for <k t<J<k 11<J<k ^' I Proof. 1+ But t 7 + ^7T~T 2(k mhj<lnl< =4r+ihr+JrT+>~> 2 3 7) 2k ' 2k For k> 3 it is clear that ' 2k "" _ + " = klil 1+ _ 4r_ + _ T£T + rrr T ' " 2k ' 3k ' 4k ( 7 . Let Lemma A3.
G (1) the generating function for fk/n.k+j .. Let fk.nt" kM = n>Q is a rational function of £ In fact. AT. From the initial conditions.n = fk.1976] USEFUL IN MEMORY ALLOCATION SCHEMA k Lemma Proof. fk/J = 0.n1 + fkjik • Then (7) f nf + Y* fntn 53 Gk(t)= k<n<2k n>2k and (8) tGk(t)= £ f^it"+ J2 ^1*" k<n<2k (9) n>2k tkGk(t) = 0+Y1 n>2k f nktn. by referring to the relation .. The development concludes the proof of Theorem A and the second limit of Theorem B. Y*fk.ki (4') j~ Akij which. and letting f^X^. fkfk = 1 m hwe fk._£ . We now proceed to the rationality of the generating function. f •Hi ft 1  (2) Uk(V 1tr (3) — 1<J<k i\ k jt where the Xkj are as in Corollary A1 and (4) AkJ = BkJ4j with (5) Proof.. Lemma Bl. the two closed form expressions for its coefficients and the limit of successive ratios. j<k. with X^~* = Xkj .E *u ^ KJ ]<j<k ' From Lemma A1. lim 1/k k 239 k = I This follows from ^lim^ (lnk)/k = 0.1 yields (5).we have k .n be defined as in Theorem B. <3'» BkJ = k\kJlkj) Given equations (2) and (3). . we have 1 . For k>\.I 0<j<k (6) 'fk. viz.
Eq.240 ON A GENERALIZATION OF THE FIBONACCI NUMBERS [OCT. E KJ = I 1< <k /</<* J Since. by Lemma A1 the Xkj are all distinct we have the partial fractions decomposition stated in the Lemma.1 1ttk k ft n nXkjt) 1<J<k where A^. Clearly. £ fK„ ("'J*'*") • * 0^n<(nk)/k Proof. (3).Zk"~1 = 0. . Lemma d5) B2. ' From Eq. From the relation (6) we have the equation (10) tGkM fnitn + t*6k(t)°GkM $^ k<n<2k E f nt"' k<n<2k Isolating Gk(t) and noting that «D '*" E f n*n~ E k<n<2k W k<n<2k we have •A (12) Gk(t) = (13)  . Let k > 1. (14) n xkJ = (Dk~1 . (2) in Lemma B1 we have (16) 0*M = (u) f* Etv/***'/ (18) £i** s>0 ^— 1(t + tk) £ M r*'*» 0<m<s s> 0 0<m<s (20) n>0 0<m<(nk)/k Thus (15) follows from the definition of G/tM Note that i f £ = 7./ are the solutions of Zk .
The Numbers/^. the case of Fibonacci numbers yields the familiar (22) f 2gn 0<m<(n~2)/2 corresponding to summing the binomial coefficients lying upon lines of slope 1 through Pascal's triangle. Let k > 1. See Figure 4. Proof.10= 9 . (3). (See Lemma B2.5 = 16 hj = 8 1 2 3 1 s h/'' i^<s¥ ^^""\yy\M .E ^ (26) = E'" n>k n>0 E **. (24) Gk(t) = tk E Akrj Y. 1*q<k (25) . as Sums of Binomial Coefficients Lying Upon Lines of Slope k— 7 through Pascal's Triangle.1976] USEFUL IN MEMORY ALLOCATION SCHEMA 241 (2D 0<m<n1 corresponding to summing Pascal's triangle horizontally.S' y' / fdld = 14/ /8 I / ' 9 )/ 10 1 11 1 12 1 13 21 1 4 10 20 35 56 84 120 165 1 5 15 35 70 126 210 1 6 21 56 126 1 7 28 Figure 4. Then (23) 1<j<k where the \k. In general one sums along lines of slope k .. Ktitn ..JKI.1.••'" i yi / 15 s* l' 28 36 45 55 66 i/'' y f3..j are the zeroes of k\kij(k1) zkzk1i. If k = 2. From Eq.. I i fl. E **JKJ • 1<j'<k .) Lemma B3.
90691 0.10331 ±i0.) * 1 2 3 4 5 6 7 8 9 10 hc.52431 0. k.61803 1.0 « 1 and 1k Vt.63526 0. Clear HO.—.86603 0. In fact.1 %n+1 = 1</<k f k. E BkM:.23279 ±i0. .88127 1.19749 0. Lemma A8.61803 0.50000 ±i0.26935 ±i0.93677+ J0.38028 0.jJ= h 2. A more complete table of these roots. ~e/? where H<KnH = (l + M.21315 0.90173 ±10.73720 ±i0.93358 0.k.39863 ±10.(Each nonreal complex root has absolute value r such that 1 + r1~k > r or r < X ^ ^Xifk).67137 ±i0.41683 ±i0.21945 ±i0.79855 ±i0. Let k > 1.92957 Lemma B4. 10 (The zeroes are listed in decreasing order of modulus.42110 1. K of the Polynomial Zk .Zk~1 .242 ON A GENERALIZATION OF THE FIBONACCI NUMBERS [OCT.23205 0.10935 ±i0. Proof. . Then fkiwH f T k.86082 ± iO.78019 ±i0.9144 7 0.n k where X^max is the largest positive real root of Z .84192 0.70533 0. Table 3 Real and Complex Zeroes Rounded to Five Places.Z ~ .32472 0.91159 1. n 1<j<k k k 1 Define X ^ m a x to be the zero of Z .max Proof.66236+ i0./. X#./ with largest absolute value.00000 1.56 228 1.1.85224 ± iO.2.82964 0. Then (30) X eM = Jim M.20 to 28 significant digits is available upon request .utv 1<j<k k. But from the proof of Lemma A2.33435 1.79255 1. F\xk> 1.X^max n +" /k.n+1 = 1+Vin< VhO* 1 .25542 0.57522 0.68720 0.^ — . ^k.61578 ± iO.9 5648 0.94058 0.k 2. each sum in the quotient has one or two terms depending upon whether X^max is real or complex and in the latter case the limit need not exist.Zk~1 .37333 ±i0.n>1/(1~k). X ^ m a x is real and is equal to \k.28520 0. Then (27) Jim_ .46557 0.) Since the Lemma follows.n y* (28) . ^^k From Lemma B3.2.23216 ±i0.n (29) > 1<&k ~ A/r.81917 1.92442 0.max * A *'max ' X K .57531 0.7 for k= 1.78485 0.max 1 Letting S7+ °°.
itkkkkkk . B. Hays. REFERENCES 1. ACKNOWLEDGEMENT We would like to thank Ernest Chou. Charles Hart. In other words X1 (k) converges monotonically to 1 3S K —> oo . Proof. and other members of our graduate Numerical Analysis class for the computer programming behind the tables. [2]. to appear. We acknowledge the BYU Scientific Computing Center for the use of their 28digit IBM 7030 accumulator. r js. H. and Kenneth Rees.s> 7. T.lHrk'1 + rk2s + . (31) USEFUL IN MEMORY ALLOCATION SCHEMA 243 For k>0 XJkl >\Jk+V. D.+ = r^J rs rsk2+sk1). litr = XJk). Thus/*s>0. 1974.1976] LemmaA9. Ferguson.s we have (32) (rk+1sk+1) rs But the lefthand side is positive because it equals _ (rksk) rs (33) rk + (s. R. private communication.s = \Jk + U Thenr>1. 2. An Analysis of Dynamic Storage Allocation Schemes Based on GeneralizedFibonacci Sequences. Robinson. and r(rk_rk1_1)= sk+1 0f _sk_1 = Q Subtracting the second equation from the first and dividing through by r. W. P. Norman.
e. L2j2. then the Fibonacci numbers. 2.2) L0 = 1. Scoville have introduced and studied the arithmetic functions a and b which are defined by (1.. Hoggatt. are also given by (2.. and throughout this paper. = 3. In the present paper we show that the functions has the following property: L e t / > 0 and let/7 be an integer with n > F21.1) Fn = ^ ~ ^  (n = 0. VALUES OF THE FUNCTION aWHICH ARE CONGRUENT MODULO A FIXED FIBONACCI NUMBER We shall require_a few facts about the Fibonacci and Lucas numbers.is any real number.4) and (1.1 = 0).1). we have similar results for the Fibonacci numbers of odd index. and for the Lucas numbers. The equalities (1. WaM ^a(n . Thus we have 0 < j x \ < 1.VF2/) /'mod Fq+i) Finally. if a(n) ^a(n ..3). Carlitz. L. and kx\ =x fx] is called the fractional part of x. defined by (1. The functions a and h satisfy many relations which follow from (1.A NOTE ON SOME ARITHMETIC FUNCTIONS CONNECTED WITH THE FIBONACCI NUMBERS THERESA P. If a= 1M1 + V ^ a n d / 3 = 1M1 .1. Jr.5) b(n) = a(n)+n=a2(n)+1 (n>1) ab(n) = a(n) + b(n) = bain)+ 1 (n > 1).4) (1. that a(n+kF2j) for k = 0. [x] is defined to be the greatest integer less than or equal tox. then a(n + L2jF2j) = a(n) + L2jF2j 1. . a(n) = [on]. along with many other properties of a and b. a and /3 are the roots of the equation x2 x . F1 = 1.F2j)(mod F2J+1) then a(n) = a(n + kF2j) (mod F2j+i) for k = 0. .3) b(n) = [a2nj. e. Here. juxtaposition of functions indicates composition.2. (1. North Carolina 27706 1.. Fn = Fn^ + Fn2 (n > 2) Lt Ln = Lni + Ln2 (n > 2)..1) F0 = 1. We give conditions on n for deciding whether or not a(n)^a(n + (L2/ . and R.E. Durham. = a(n) + kF2j+i L2j2.g. and n > 0 . [4]. V. In fact.sf5) (i. INTRODUCTION AND PRELIMINARIES The Fibonacci numbers are defined as usual by (1.1. Hoggatt. Recall that if A.F2j) fmod F2J+1). and the Lucas numbers are defined by (1. where a = 1A(1 +s/5).5) are given in [1]. In addition. In [1] and [2]. Jr. E. which may be found in V.1.) 244 . VAUGHAN Duke University. we have the stronger result.
1 (v) a(n + L2jF2j) = a(n) + L2jF2j+i . L2j+1 .DF2j) = a(n) + (L2j .then (iv) Ife < / . Suppose that/? > F2j.6 Theorem.2) are given by Ln=an+Pn (2.. F2J. f 2 / > / . a(n . 1076 A WOTE 0!\l SOME ARITHMETIC FUNCTIONS CONNECTED WITH THE FIBONACCI NUMBERS 241 and the Lucas numbers defined by (1. 1. e < \$fl+1 (i) (ii) a(n+kF2j+1) for k = 0.0 2 y + j 3 4 / .2). 1. 2. it is easy to check that aFn = Fn+1$n (2. then we have a(n + L2j+iF2j+1) = a(n) + L2j+1F2j+2 (iv) Ife > (34j+2.~) and aLn = Ln+1F1(1 (2.(j > 0) and that a(n . 7 .\on\. 1. 2.1 Theorem.L2j+i. a(n + kF2j) = a(n) + kF2j+1 2/ L2j2 4/ (iii) If e > //3 f/3 .2) (n = 0. Then (i) 6 < \&fii(1 (ii) + and that p) a(n+kL2J) = a(n) + kL2J+1 for k = 0.OCT.J . Let/? be a positive integer.4) (n = 0.Fy) ? a(n) .. .L2j+i)? a(n) .L2j+2. . Write m = [an] and e= j an I.1)F2j+1 .1. Let n be a positive integer. Using (2. Suppose that/? > F2j+i and a(n . then a(n + (L2j . then a(n + F2jL2j) (v) = a(n) + F2jL2j+1 + 1 a(n + (F2J + 1)L2j) = a(n) + (F2j + 1)L2j+1 + 1. and set/?? = [on] and e.1 = a(n) + kF2j+2 4yV 2 (iii) If e < j3 " . 1.~).1)F2j+1 a(n + (L2j ~ 1)F2j) = a(n) + (L2j .1. 4/ (iii) If e < (3 .F2j+1)?a(n) Then F2j+2.3) (n = 0. Let/7 be a positive integer.F2j+j.5 Theorem. and set/?? = [an] and e= ja/? J\ Suppose that/? > L2j(j>0) a(n .L2j) ? a(n) . 2. •».2.2 . then a(n + L2i+1F2j+<[) = a(n) + L2j+iF2j+2+ (v) 1 a(n + (L2j+1 + DF2j+i) = a(n) + (L2j+1 + 1)F2j+2 +1. 2. Let/? be a positive integer. then a(n + F2jL2j) = a(n) + F2jL2j+1 (iv) If e > $4j. . Then (i) 1fl e > (ii) for k = 0..8 Theorem. + p) The main results of the present paper are given in the next four theorems.1) and (2. with m = [an] and e= \an\. Then (i) (ii) e> a(n + kL2j+1) Suppose that /? > L2l+i (j > 0) and that 1$2j(1+$2) = a(n) + kL2j+2 for k = ft f. . 1.
L2j. (b) a(kF2j+1) = kF2j+2 and a((L2j+1 + 1)F2j+1) = (L2j+1 + 1)F2j+1 + 1. where the last equality follows from (2.F2j)] = [an . we have kf3 < 1.2. Proof.*™ (iii) If e > a(n + (F2j+1 .. L2j.5 and 2. To see (ii). 2.7. then a(n + (F2j+1 .3) we have a(kF2j) = [kaF2j] = [k(F2j+1 fh].1. •»..DL2j+i) = a(n) + (F2j+1 .3) of the functions.5. . suppose that k > 0 is any integer.1 (v) a(n + F2j+ iL2j+i) = a(n) + F2j+1 L2j+2 . Now the assumption a(n . lfPd+pl+tf1*2. while L2j(52j > 1.3). PROOFS We prove in detail only Theorems 2.1.1 and a(L2jF2j) = L2JF2J+J2. Then (a) a(kF2j) = kF2j+1 .52.1 for k = I 2. L2j+1 1.DL2J+1) = a(n) + (F2j+1 .(F2j+1 . using (2.DL2j+2 2j (iv) If £ < / .. .246 A NOTE ON SOME ARITHMETIC FUNCTIONS [OCT.3)) . we have a(n) = m.2).1 = kL2j+2 a(F2j+<iL2j+i) = F2J+IL2J+2 2.F2j+i implies that e + fl and this proves part (i). 2j Then for all k= 1.9 Theorem. Then >1 (by (2. (c) (d) and for k= a(kL2j) = kL2j+1 and for k = 1. From the definition (1. we have the following theorem (which is not quite a direct corollary of the preceding results). Let/ > 0 be any integer.. The proofs of all four parts are very similar.1)L2j+2 . and a(n .2.P21)] .aF2j] = [m + e. It is then obvious how to prove Theorems 2. This proves (a). that L2JP2i = 1 + f1 > h while (L2j1)$2i $4i$2i = 1+ and since /3 < 1. It is easy to check. Proof of Theorem 2.2.F2j) ? a(n) . and in fact.F2j) = [a(n . By (1.F2j a((F2j + 1)L2j) = (F2j + 1)L2j+1 + 1 a(kL2jH) 1 for k = 1.8 are given In §3. and we prove only (a). 3. 2.tmF2j+1 2 + (e + $ hl . F2j+1 . The proofs of Theorems 2. we have j32y > (l4j\ so that (L2j1)$2j < 1.$ (1 + p) + fj+2.6 and 2.8. . It is natural to ask about the values of a(kFm) and a(kLm).
1)F2j) = a(n) + (L2j  1)F2jH.9) It is easy to compute that 0 < ej3 4 y < / .8) holds.(L2j .2) > kf1 > 0 (k = 0.5) that L2jp2j = 1 * j34y Then a(n + L2jF2j) = [m + e+ L2j(F2j+1 $2j)l = [m + L2jF2M We must show that (3.1)$2i] .3)) 2i = [m+kF2j+i + ek$ ] .j3 .2.f1 (3.1)P > 0.3) (k = 0. and this proves (3.8) <0 2j since 0 < e < 1 and (L2j .2). we have/34y < j32y foraSI/> 0. It is evident that if (3. then 0 < e(L2j1)p2j' < 1 is equivalent to 0 < e(1P2j' (3. we have L2j$2j = (a2i+$2h$2i (3. we need only show that for all k satisfying 0 < k < L2j .5) and so = 1 + $4i (L2lnf! = 1+f1 f1. to see (v). we have a(n + (L2j . then a(n + (L2j ..7) + p4J < e < 1 (since we always have 0 < e< 1).6) + p4J) < 1 orequivalently.1)F2j) = [m +e+(L2j . I . This proves (iii). if we can show (L2j1)$2j (3.1)F2j) = a(n) + (L2j . To see (iii) and (iv). Finally.1)F2jH . It is clear that if (3. It suffices to show / .1.4) < I the inequality (3. 0 < 1p2I (3. If 0 < e .3) will follow.4).1976] CONNECTED WITH THE FIBONACCI NUMBERS a(n + kF2j) = [on + k(aF2j)J = fm + e + k{F2jHP2J)] (by (2.6) (and hence (3. we have e > 1 . then we have a(n + (L2j . But since (L2IHP2J = 1$2j + &4i. 1+ efl] .(L2j . By (2.7)) does not hold. To prove (ii). (k + 1)P2J < / (3. L2j .2) orequivalently. I ». This proves (iv). we have from (3. Clearly. then we must have e(L2j1)$2i (3.1)F2j+1 ..1) < U y By (i).1)P2j < 1. L2I . Since j3 < 1.2). we have 0 < ek$21 (3.
But also. this gives e > Vi. 0 4 y < j3 4 < %.aL2j] (by (2.15) that if 0 < e < j3 4/ . 1 (Feb.5. 2 1 4 Proof of Theorem 2. 1972).13) 1)\$\2H(1 0 < e + (F2j + $2) < I We first note that. we have a(n) = m. pp. In view of (1. and this proves (iv). 1 L Carlitz.. It is clear from (3.10) OCT. No. E. Vol.2. On the other hand. Hoggatt.. REFERENCES 1. Then the assumption a(n . since o3 = .16) 0 < e + k\$\2H(1+$2) and (ii) is proved. 1976 . Boston.L2j)?a(n) . Scoville. using (i).8 all remain true if we substitute the function b for the function a wherever it appears. To see (v). and (3. P. and since/ > 0. This completes the proof of Theorem 2. As in the proof of Theorem 2. Hoggatt. Clearly (3. 2942. and R. V.7. 1969. a(n + kL2j) = [m + e+k(L2j+1 . since/3 < 0.9) follows. Jr. we have (3. 4.11) 2H e+$ (1 f))l + $2) < 0.14) 1+f Then we have = / . and T. R. This proves (v) and completes the proof of Theorem 2.P2h 1(1 + a(n .6 < /3 < .5.1. 10. we need to show that (3.17) 0 < e+F2j\$\2i~1(1 < 1 + $2) = e+(1 .4)).12) + $2) and this proves (i).5 . "Some Arithmetic Functions Connected with the Fibonacci Numbers. Scoville. Vol. "Lucas Representations.£ = <kzl .U2I+1 . if e> j3 J. we have (3. ickkkkkk . it is clear that Theorems 2. (3.7 so that P < /2and /3 < %. 1972)." to appear.j32y.. a a Then. No." The Fibonacci Quarterly.3).L2/) = [m + e. we have (since/ > 0) 0 < e+(F2j1)\$\2i1(l < \$\2hUi+f) + $2) + (i$4i)\$\2H(i+$2)= i$4i < u It follows that if 0 < k < F2j . As before. 2. To see (ii). Houghton Mifflin Co. then e+F2j\$\2>1(1+$2) = e+(1f>) > 1.11) is the same as e < \$\2H(1 (3. . By (i) we know e > 7 . L Carlitz. V. we first have. for any integer k > 0.$4j) < 1 and (iii) follows. E. then (3.L2j+i is equivalent to (3. and = [m + e. 1 (Feb. Fibonacci and Lucas Numbers.$2h 1(1 + $2))]. 10. Jr. "Fibonacci Representations.18) (F2j+1)\$\2H(1+$2) = (1~&4i) + \$\2H(1 + $2) > 1 and it follows that (v) holds.248 A NOTE ON SOME ARITHMETIC FUNCTIONS CONNECTED WITH THE FIBONACCI NUMBERS (3. Vaughan.1 . 7. 128. pp." The Fibonacci Quarterly.
2) gives no information about A(2ml By (1. k=0 J ^ k=0 v J so that (1. Let VThen £ 'W(Z)U/ t 0<2k<n (K2k<n V ' J A2(2m iD = (2m + DA.1) we have ^E^MO35S^*(2)2(*i)+3ii^(.)(*02 k=0 k=1 k=0 k=0 k=0 k=0 Supported in part by GP37924.SOME BINOMIAL SUMS LCARLITZ Dyke University. Mr*"{(. Put n+1 where it is understood that (_". Gould [ 1 ] ..)}'.£ /«'"{(.•.) '(nit) (n '° >°> Consideration of this sum was suggested by the following problem proposed by H. However (1.) (:)V. therefore (1.3) A(2m + 1) = 0. (2m + 1). W.2) AM = (1)nA(n).(. Nortli Carolina 27706 1. Since AUT..„). 24S k=0 . Durham... It is noted that this result does not hold for even n.
4) becomes (1.5) S2(n) = (iF+'Sjfn) and (1. In particular we have /.)' it is clear that (1.6) AM = 2S0(n)3{l + (1)n)s1(n).)^ f'*(i)(n:'){(ni')(*%)}*<*> k=0 so that (2.1) where SJn) = T0(n) TjnjSJn).8) SQ(2m) = (1)m (M[ ' . we have . In the next place.». p. In order to evaluate Sx (2m) we proceed as follows. (n) + 3S2 (n).4) A(n) = 2S0 (n) .6Sl (2m) j A(2m + 1) = 2S0(2m + 1). 2. 13]. However St (n) does not seem to be known. 243]) that S0 (2m + 1) = 0. while (1. _ n+1 2 k=0 n+1 k=0 n+1 k=0 so that (1. [3. Thus if we put n+1 k=0 2 k=0 n+1 S2(n)=Z(V* (»)(. n+1 wE'^UH'T)' k=0 Now T i<»> = k=0 H<i>k[nk){nl1){ki) .E^(^)2(^+. We have n+1 2 n r s + n+1 ^Er^(. p.3S.)}.250 SOME BINOMIAL SUMS [OCT.) {(^ ')(. 7 \ f A(2m) = 2S0 (2m) . It is well known (see for example [2.
1) yields (2.:?)(r)} k=0 k=0 ^ (i)k[nk){nk++\){nV)+^nt .)(j:!){(Br)*(. (2. For/7 =2/77. Therefore T7(2m) = 0and (2.1876] SOU/IE BINOMIAL SUMS 251 k=0 n+1 ? = ^"* !>"*(*.2) Tjfn) = (1}nHTi(ni. (2. indeed each sum on the right vanishes.:?)2.)(:.)(.(ri»± ^*( fl *)(**?)f(2^)(**! r/r E r/^(. £=0 For/7 = 2m + 1. (2.:.)(r) + M^ £ A:=0 J M.4) gives no information about 7"0^m * / j . however.*(.:!)} A:=0 ' x N k=0 ' (i>k{nk)[nk++2i)2• Hirz k=0 so that (2.:?)2 £=0 M/«E^(.)2M^i:^*(.4) { / + (1)"\T0(n) = 2(Vn f ) C// ( J ) ( T ) ( 2 * \) *M/»I:^(. v k=o that is. )(T)U) .){(.4) becomes .3) Sj2m) = Tf(2m)S0(2m). In the next place To*)1 (i>k[:)[nv)2= X k=0 )[::i)2 t <'?*( „:k N k=0 r//'i:M/f(j)(n.)(.)(j.
a.10) .3) by (a)jX^/j! and summing over/.252 SOME BINOMIAL SUMS [OCT.8). by (1.5). we get as.2r .2) is automatically satisfied.1). 243] that 2m U.9) Finally. it follows that E (a)i(ab)j(ach m+1)/(F+J7T 1=0 OO E r=0 J = V (ill 4* il J V <Or(a+J)r(b + ca+1)r ** r!(b+1)r(c+1)r 1=0 OO . (2. Note that we have not evaluated 5j (2m + 1). 2m (2. p.2) c+d = n + a + b+ 7. Multiplying both sides of (3. (a)2r(b + ca + 1)r . .b)j(a . ^ ' " r!(b+1)r(c+1)r * ^ j=0 r=0 CO (a + 2r)j •• _ ^ / „ r (a)2r(b+ca+ 1)r X jl l^'1' rl(b+1)r(c+Vr r=0 . 4* k \ 1\ k k+1 M f m!m!(m + D! k~u and 2m 0 V * / i)k ( 2m \ (2m+ 2 Y = (D™ 2(3m+2)! 1} ( u Z^{ { k ) [ k+1 I m!m!(m+1)!(2m + 1) ' l?i\ U /} ' k=0 x Substituting from (2.7) in (2.5) (Vk 2T0(2m) = 2 £ ( *" ) ( 2m+ 1 k ) ( 2m + k +] ) k=0 2m 0 + k=0 It is known [3.1) in the following way: H o\ 1 ' v ^ H)r(a+j)r(b + ca+1)r 2# r!(b + l)r(c + 1)r = (a . (2. 9]: V * (nk(a)k(h)k L> k!(c)k(d)k k=0 C\ i) = (ca)n(cb)n (C)n(cab)n ' where (a)k = a(a+l)(a + k. (a)0 = 1 and (3. We rewrite (3.D.6) and (2.8) S? (2m) = (1)m . A(2m) = 2(1)m I M l ^ l L .3) and (1.<ir !$£&* • Therefore by (2. For completeness we give a simple proof of (1. (ml)3 (2m + 1) This completes the evaluation of the sum A(2m). We assume Saalschutz's theorem [2.— 7 ^ — 7 r ' m!m!(m . ram .1)!(2m + 1) (2.. p. 3.7).c)j (b + l)j(c + 1)j the condition (3.9).6) and (2.6) and (2.
H. solution. Let/?. Amen Math. Let/7 and q be nonnegative integers. Monthly.2.] i '•• '•¥• as n + °°.1976] SOME BINOMIAL SUMS Mow take a = n and we get (34) K6A > V (ntj(nb)j(ncij / _ A x 2^ ji(b + 1)j(c + 1)j . p + 2q. E 2395. mod 1.w i t h Hn= qFn1+pFn2 (n>3). the sequence (un) is u. . p* and q* be nonnegative integers. [Continued from Page 214. 1935.5) yields 2m E <»m ( T ) ( 2 m / ? / C ) ( 2m]+cC (36) ) /=o /_fjm (2m)!(3m +b + c)!(2m +b+ c)l m!(m+b)(m+c)!(2m+b)!(2m+c)! f ForZ? = c = 0weget (1. Generalized Hypergeometric Functions.6). 1966.c be nonnegative integers.8). we have —j^ as n » oo. { For/7 = 2m and x = 1. Then the sequence p.] A?* + <7*. or(Hn). Bailey. q. (3.n 2p+3q.4) reduces to { bl ^ j!(b+1)j(c { + 1)j U m!(b + 1)m(c+1)m Now let b. Cambridge.  and (H%) the sequence /?*. For if \/n = log H Vn+1Vn ~* \ log n . q.1)r r!(b + Vr(c +1)r r ( 1J . 75. H1 = p. 80 (1973).L. 2. 3. . p* + 2q*. L J. Let (Hn) be the sequence p.for/? = 0. p+q. Since this limiting value is an irrational number. W. REMARK. £ = / w e get (2. p. Then (3. Slater. Generalized Hypergeometric Series. (7* [Continued on Page 276.for. p+2q. 1146. . p. 2p* + 3q*. Cambridge.& = c = / we get (2. 80 (1973). p + q.7). REFERENCES 1. '" = 1. Theorem 2. 2p+3q. q. (~n)2r(b+c+n .d. H2 = q possesses the property shown in Theorem 1.
.1) a<i+a2 + ~' + ak * fl it is well known and easy to prove that the number of such compositions is equal to the binomial coefficient /n+k. Then clearly (1. 2S4 '*' . zk = n.1) and (1.6) y" nm0 1 c(n)xn _ _ t+22tv'*L // where Supported in part by NSF grant GP3724X1. this reduces to (1.k)xn. Durham. s^) such that (1. We shall show that (1.k) denote the number of solutions of (1.1) and (1. .4) c <n. p.kt).k) denote the number of solutions of (1.RESTRICTED COMPOSITIONS L CARL8TZ Duke University.2) in nonnegative a.3) c(n.k) = 0 (k > 2n + 1). Let c(n. Also it is evident from the definition that (1. a2.1) when we require that (1. INTRODUCTION A composition of the integer n into k parts is defined [1.1) is equal to ( " .k=0 1 + JT(i)l*!£ Forz = 7. be strictly positive then of course the number of solutions of (1. North Carolina 27706 1.k) = c~(nk. 107] as the number of ordered sets of nonnegative integers (ai.1\ [ k1 ) • If we require that the a.2. : ! ) • In the present paper we consider the problem of determining the number of solutions of (1.2) in positive a/ and let c~(n.~.5) E 1 c(n.k).2) ait ai+1 (i = t.
1(0) = 1 k is less immediate.2) ca(n.1) and (1. Clearly n n € (2. M 1*' This is also proved independently.k) denote the number of solutions of (1.k) It is convenient to define the following refinements Qic(n.k) = } £ ca(n. let ca(n. a=1 .1) c(n.10) x2i~1 / .xn • 0 We show that the radius of convergence of the first is at least V%\ the radius of convergence of the second is also probably > 1/2 but this is not proved. Ia(n.with a. 1878 RESTRICTED COMPOSITIOWS c(n) = Y l c(n>k)> (1.> 0..k) = £ ca(n.k).2) that oo <bk0c. oo Fa(x.k)xnzk = 7+ £ —i .9) F(n) = J2 c(n.3) and (1.y) = £ FafcUy* . n=1 it follows from (2. > 0 when the number of parts is unrestricted.k) = Y! a(nM c(n.k) AS^D ca(n. The generating function for (1. it follows from (1.^ satisfies the recurrence (2. = a.8) c(n.OCT.k1) (k > 1).kl.2) with a.k) = 2 ^ cb(n a.k) is defined as the corresponding number when the a.k)xn.k). It is proved that oo £ c(n)xn = (1.6) that (1.k) and F(n.2) in positive integers a. h^a If we put.7) 255 c(0) = 1  k=1 Thus c(n) is the number of solutions of (1.1) and (1. GEiyERATIPJG FUiCTlOWS FOR c(n. fork> 7.V inx)^ 0 (1x2H)(1x2j) t It is of some interest to determine the radius of convergence of the series oo (LID oo J2 c(n)xtl < 0 X ! *(n. a=1 a^O The enumerantc a fr7. 2.
* . Using (2.Y) * E a=1 OO (xy.y) = £ (D'~1 1=1 JU^1x'y $kj(x. Iterating (2.5) reduces to k (2.7) $*fc/. In particular.a F Tt b(*.a = r^a1 it is clear that (2.v) = jz^r *kl(x. F o r / = 7.If = E hta OO F b=1 a b(x. (2.V ( jzLatb b=1 \ *Y so that (2.x2Y) ^ (k > 2) and generally s = £ (~VH j=1 *k(*.4) <bk(x. 9o0c.y) = J2 (xy.y) = j&j ^k^(xf1)^k^(xfxy) (k > 1).xk1.1hk = £ k= ° 1 nJT M (1)1*!iL 1x' ' Ji = 1 ~ Wb ) I .E V * . for$ = Ar .y) <k> 1).256 RESTRICTED COMPOSITIONS Fa(x. bta Then OO *k(*.1) + (1)k1<f>1(x.3) *k(x.D+J2 {»•> si *H(X. $kj(x.V E M .4) may be replaced by k (2. this gives k „j f : zk f *k(x. Since ®r(x. we get ®k(*.k. (k > V.3).6).k.k.y) JS^r1x'y $H<X'1> + <1>5®ks<x.y) = 1.k) = xa E F b<x.V [QLCT. this becomes k1 (2.5) **fr^ = ( 1)H E n where it is understood that  (2.8) $>klx.6) ^ ixJy F kl(x.D} i = k=0 * •* ' ^ i •* ' j=1 *—* 1x'  / and therefore (2.D + ®k2(x.D = Skm1 where 5 ^ / is the Kronecker delta.D j $k2(x. + E lD1^.7.xsv) (k > s).D (k > 1).
.k ca(n. .8) when y = 1. 3. Note that the LHS of (2./ x'z' JlllL 7+ M We now put 1*' PkM 1 (2.8).9) £ c(n.k=0 .»wfc« E M. Pjx) = 2x\ ' Pjx) = x4 +xs +4x6 . (2*8) can be written in the more explicit form (2.." ^f E Hence. by (2.1).13) is equal to (2..(J.(»' . X ).12) Pk(x) = Ys ("1>H [ y 1 MHX'PHM (k > V. In the next place.k)x"yazk . M where rk i = The first few values of P^ (x) are Pjx) = I Mk L ' J M/M*/ Pjx) = x.M . \>./ x'z' M 1*' This evidently reduces to (2. V . E **fcj** . (2. GENERATING FUNCTION FOR c(n) AND RELATED FOWCTIOiyS Forz= 1. by (2. oo *=' .13) £*kMfc**=* ™M L^K. (2.1976] RESTRICTED COIVfPOSITfOiS In view of (2.V V {t)J / „.5).8) reduces to .k)xnzk = ".14) £ 2 n1 a.10) •D1 k=0 x'z' Mo = / k Mk = (1xHl X 2.1) _ „  (2.11) ' 1x1 where Clearly k Mk oo P JMw The Pk(x) are polynomials inx that satisfy k (2.E ^ E tv"*!*. i+Y. .
oo (3.1) = .1) by oo (3./ in (2.8). A" thus d'(n) is the number of odd divisors of n less the number of even divisors. where m is odd. Thus we may replace (3.6) d'fjkfnj) (n > 1).a=1 n=1 we have therefore oo (3.4) z1 J^$k(x. where d(n) is the number of divisors of /?.l) = k= ° " 1JbdW 1 Since OO OO k=0 OO n.2) d'(n)=Y. M where cfO) = I It is also of some interest to take z = . s=0 j\m j\m so that (3.1) £  *k(*.258 RESTRICTED COMPOSITIONS [OCT.k.k=1 n=1 j\n Put H (3. and r>0.7) = """^ ' 1 OO 1 = L oo 1X1 1 • .(1) .3) din) = (rDd(m)._£ • h1)H ^!__ =1 We have x i 1 x j=1 ~' j.5) 1+YJ c(n)x " = " = /=/ ' ~ I*' n1 It follows that cfn) satisfies the recurrence n cfn) = £ (3. We get ] £ fVk^k(x. For/7 = 2rm.
It is also of interest to take y =1 c*{3) = I c*{4) = 0. 1+ .k).aFl = 1 + J2 c*(n)xn n1 . where n (3.fc/. . c*(6) = in (2.8) c*(n)xn 1+Y\ = ' .7 .^ 7 1x> so that (3.1) = — [ — — k=1 so that 1+ 2 (3.*!7 1 1*' * 1+2Zdo<n)x" = ^ 1 + J2d(n)x» 1 2.1976] RESTRICTED COfVIPOSITiOWS Since oo oo J2 (Vk*k{x. 1Ld(n)xn 1 This yields the recurrence n (3.. c*{2) = .^ J ? 1+Y. Yl k. ' k=1 —'— + ? E (1)'' .9) c*(n) + Y^ d(j)c*(nj) = 0 (n > 1).10) £ ^k(K1) = ' — k=0 1x' If we take)/ =z = 1 in (2. M where c*(0)= I The first few values oH*(n) are c*(1) = 1. c*(5) = 1.a=1 we get oo (3.7) c*(n) = t1)kca(n.13) we get oo J2 (Vk<S>k(x. .k)xn J2 n.1) k=0 = 1+ oo {1)kca(n.k.D = T ./ / z = / we get £ (D1 ^~ E*. F o r / .13).11) £ k=0 k (D ^k(x.
it is of some interest to derive them independently.k) IkUWc a(nfk) Whilegeneratingfunctionsfors (n.k) can be obtained from those for c(n.1) ®k(*.D .1)a+kca(n.k 4.f) + (1)k'1<!>t(x.k)x .y) = J^ Fa0c.t) ix'y (k > 1).H(x. a=0 ~ca(n.3).k) = ]£ cb(n .V ( rz~r .k).J2 ca{n.260 RESTRICTED COMPOSITIONS [OCT. Put oo oo n Fafok) .y) = 1.£ hh*.12) n=1 a.VQkt(x.k)y' . exactly as in Section 2. n=0 Then. where (4. b?a so that Ta(n.k) = xa "2l'Fb(x.k) by using (1.y) = ] T j=1 hi J^l— <f>k_j0c.11) is equal to oo 1 + £ xn £ (.a.2) S=1L *k(KY) " £ pi r *kj(*.xk~1y} J fx y Since ®i(x.k.Mb ) h^a b=0 Thus (4. Note that the LHS of (3.j ~ .k1) bfa and oo (x y> = **  H oo (xy)a a=0 T H b(*.y) .y> = J2 <xy>a = jz~ > a=0 we get (4. (3. .3) W0(x. iteration yields k 1 ~ :(x.k).k) and ca(n.2) reduces to k (4. GENERATING FUNCTION FOR c(n. ®k(x.QkiOc. Fory = /. (4.V = 8k/0 1x1 (k > 1).xy) (k > 1).4) QkOctJ+E pi It follows that J=J£ ^ 4 S 4>.k) and Fa(n. As above. where dQ(n) denotes the number of odd divisors of n.
(4.7) <j(kj) Let jk + 1/2jlj.7) reduces to _ _ Pk(V = kPktd). so that (4.k) = 0 it follows \\\%\Pk(x) begins with a term i n .' 12x 1. . moreover the coefficient of this term is 1 for k odd and 2 for /r .10) 1 K ~ .E * y k=0 \ = i. c(n. 1=1 also it is clear from the definition that (4. and so d e g f y M = M(k 1). it follows that the degree of the/ f / ? term on the right of (4. by (2. j M so that E Jk*k k=0 Thus yk = 2k~ 1. Pk(x) = F o r * = / .1976] RESTRICTED COMPOSITIONS <45> E *k<x. J (k > 2n + 1).ikk = 261 —^~J—— k=0 ^E^ y —M Now put 1x' Pk(x) /=/ zk 1*' so that _ 4 6) P <. Then we have k1 yk = E ?*/= E v 1=1 (k > 1)  ro This gives E vkxfl ? .4).9) Pk(1) = kL Also it is easy to show by induction that _ deg/JW < %j(j n Indeed.D + %(kj)(kj1 denote the coefficient otx^'*' ' in Pk(xl k 1) = M(k V. assuming that this holds f o r / < k.7) Pk(x) = 2 r ~ / j / ^ [/ ] MhPHM (k > 1).k> (4. r even and positive._ **M. Since.Mt • The Pk(x) are polynomials in x that satisfy the recurrence k (4.8) xkPk(x).
It would be interesting to know if they are positive.16) and (2.12) as GO (4.6) and (4.k=1 This is of course equivalent to (2.y) .k).12) that (4. 2 n=0 so that/?f/7. the equivalence of (4.Ykk = ^Z 1 ~ M • 1*' We may rewrite (4.5).k) = J2P("J.9) follows easily.6) that (4.11) c(n.17) *k(x.£/isthe number of partitions (in the usual sense) of/7 into parts < k. Since ~caln. k) (k > 0).14) Ca<n. It is clear from the recurrence (4.xi and 157 = j P<n. Note that it follows from (4.3) we have LliL£ . we have (4. Hence.4  (4.16) 1 c(n + k.262 RESTRICTED COMPOSITIONS [OCT. If we put Pk(x) = Ys 1(k>J.£ J=Htl FgM .2). n./tf = c(n + k.k)xn. . /=? (4.k) = ca+i(n+k.k)xnzk = ./ x'z t„ +VJ^fr )'*£.5) and (4. replacingz byxz in (4.7) that all the coefficients are integers.12) 2 *k(*.9). kfttKl) • J Returning to (4.k)xnyazk = M _ 1 + J2 E By (1.4  (4.13) 1 + J] I(n. we have ** This gives i <»H .k)xn+kzk 1+ £ 1 > / „.15) ~ x £7/j. 1 t <»H . it follows from (4.
As for FfW.2) + x(n.1) = 1 (n > 1).k) < v(n.k).2) n 0 is at least V2.3) + c(n.n2ssn J2 °(n u 1k(nv 2). nx+n2=n ^ c~(n.2) + 2 c(n u 1)c(n2.k) in terms of c(nfk) can be obtained by considering the possible location of zero elements. It follows that c(n) < 2n~1 (5.k) denote the number of solutions in nonnegative integers. Then vM = (°kZ)) . 1) (n > 2). we shall get crude upper bounds forc(n) and c~(n). Let v(n>k) denote the number of solutions in positive integers of n = ax+ a2 + ••• + ak and let v(n. 1)+ ^2 c(n u 1)c(n2.t)zk = 1+z + 1 m oo <bk(Xf1)ik 1~z V 1 It is easily verified that this is in agreement with (2. There may be a zero on the extreme left or the extreme right. ~c(n. nl+n2s=n It follows that J^f$k(x. 1) + 2 ni"f.k)("?_.1976] RESTRICTED COMPOSITIONS 203 Sn addition to (4.8) and (4.k) < v(n. Thus we get relations such as the following.k) •= 0 (k > 2n + 1).Vzk = 1+z + (1+z)*J^<l>k(xJ)zk + (1+z)>zlY. we get 2n+1 F(n) < J2 k=l so that { 2n 2n k=0 k=0 V) = £ ( "r) <£ n +kk  .2) + 2c(n.1) (n > 1). 1).3) + 2c(n.1)^ y+(1+z)'zYE*k(x. c (n.k). also there may be one or more zeros on the inside. v(n. GEWERATliG FU!\fCTIOf\IS FOR cfn) AND F(n) We may not put z = 1 in (4.4) + 2c(n.15) another relation expressing c (n. We can get around this difficulty in the following way.1)*kl *'" (1+z)* ] T <$>k(x.1) = c(n.2) = c(n.') Clearly c(n. c (n.3) = c(n. To begin with.1) = I h~(n.5) since the righthand side then becomes meaningless. since c (n. 5.®k(x.0) =J(0. so that the radius of convergence of E c(n)x (5. 1(0.5).4) = c(n.
6) 1 1ax) Y. 1958.4) o is at least 1/8. i x2ji (1x2h1Hlx2i) Forx = & we get 1/2 . J6_ . Next consider 2k y k t1\H PI I Z' =Y f 1x' J M \ Z2'~1 _ .* * ) 1Z+X2H(ZX) 2H M 2j1 2i <lx )(1x > Thus (4. New York.Vzk (5. John Riordan. Presumably these bounds are by no means best possible. REFERENCE 1. It seems likely that the radius of convergence of (5. Wiley.5) becomes Yi ®k(*. 1976 ~c(n) < 23n. 1/8 ('})('!) I'iK'i) 1/32 . *kkkkkk .4) is probably somewhat greater than 1/a.3) OCT.264 RESTRICTED COMPOSITIONS (5.V I . _32_ . .63 [>A)(>A)" Thus the radius of convergence of (5. An Introduction to Combinatorial Analysis.X 2 H Z2 ' \ <*> . 3 105 31. _ 4_ .4) is about Vz. Hence the radius of convergence of J2 c(n)xn (5.5) 0 /_ y* iz+x2H(zx) (1x2H)(1x2i) 1 It is now permissible to \etz> 1. We get oo V ) c (n)xn = (5.
SUMS OF COMBINATION PRODUCTS IVIYRONTEPPER 195 Dogwood.2 + 1.2—#1. 265 .Xr) <x 1<r r denote the sum of all products of the form xj 9X2" — mxr. Consider ( £' n /=/ \ I n \ n n n 2 E/ EE^E^E'l\M J '. A4.34 24 3»—*#»r3.3 + 1. A. and n > r > 2. We now develop a formula for A%. Illinois 60466 INTRODUCTION The combinations of the integers 1. .4.2.4 4 We will be interested in developing methods for evaluating sums of the form 1. Park Forest. 52 *i** = 12+13+ 1 4 + 23+ 2 4 + 3 4 and ^ x x i* = 12+13 + 2.4 and 1. where *1 < x2 < "' < xrt X VX2> '"*xr ^  1. 4 can be represented by the following diagram: ^2. We let £) K (xt.^12^*124 ^2.n\. For example.3 + 2 4 + 3.X2.3 + 1.24 + 1.3—^2.3. We also provide a general approach for finding >4^ when/7 > r>5.4 + 2.23. 2.3. 2.3.=1 W /•/ M Thus. We define n Anr = Ys (X X 1° 2 X } t>2 r< ' m] *7"23' Mn In this paper we develop formulas for A^ A3. 3..4 + 2.4 .
5+4. £ / / = M2 Y^ii 12 = 2.3+2. J2 U = 12+1.3 + 2.4+1. we have Theorem 1.3 + 1.3+2.4+3. i<j Now. 1)(2n + 1) Thus. with /7 = 3. We could also find /</ by using the method of undetermined coefficients.5 = 85.266 SUMS OF COMBINATION PRODUCTS Now. Say n > 2. /</ [OCT.3+1. . We begin by assuming that Y^ (*1 x r) x1<"<xr Mn is a polynomial of degree 2r in n (we assume that the coefficient of n° is zero): 2 ^ T / / = An4 +Bn3 + Cn2 +Dn.3 = 1 1 .5 + 3.4+2. M. = 12+1.3 2 j t 1 3 f 3 3 .3) = 3 ' 3 4 .4 = 35. Mn Thus. £ / / = 12+L3 + 2. Then /</ \/=/ / /=/ For example.5 + 2.2 + 1. 2(1.4+3.3 > .4 + 2.
Hence. Then a\+a +a B*(a\+a\+a%). we have Theorem 2. Then r2+r22+ . . 2(35) = /f»44 /£43 +CV+D1.C. lithe product 114 appears in the sum . 412.. 214. rn are the roots of xn = Df and n > 3.r .* r£ = 0.« A j=1 kt • A!_ E «*+ E «** E vk. Generalizing Theorem 1.4For example. Then I n y n 2 For example. equal the / root of the above equation. if the product 1 24 appears in the sum n n n EEE i=l j=1 VK k=1 the following products also appear: 142. Say a. letting a. EEX>i=1 . aj e I ai. = 0. A .D should provide the required answer.B. we have Theorem 3. Consider: (i'Yt/Y t»W£')'£££*• We consider n n n E E E ** /«/ / = . by Theorem 2.2*+C22+D>2. * = / to be a sum of products having three factors. 2C = 2 2 Say B = C = 0.. These may be considered as rearrangements of 1 24. We note that the number of permutations of three distinct objects taken three at a time is six. We now develop a formula for A^ . letting a/. H E f £. 2(2) = A2« +B.. * £ . Thus. Then. all factors equal all factors different two factors equal Now. say x3 + Bx2 + Cx + D = 0. 421 . 2(85) = AB4 +B& + C§2 +DS. a2. an I and n > 2.= i2 *E^(E'VX> Similarly. .///...1976] SUSVIS OF COMBINATION PRODUCTS 267 Thus. B. . Solving this system forA. Say ru r2. . 2(11) = A24 + BWl+ CV + D>1. 241. 2( JL+L +J_\=( 1 + l+lV f i+l+L): \ 12 13 23 / I 2 3 / \ 4 9 / Now. letting a.
34) ( Yl # ] +2^ n)( E /)  We now give an alternate derivation of the formula for £^ ijk i<j<k Mn Consider: 3(12) + 4(12 + 1 3 + 23) + 5(1 2 + 1 3 + 1 4 + 23 + 24 + 34) = (1 23) + (124 + 134 + 234) + (1 25 + 1 35 + 1 45 + 235 + 245 + 345).3.2.4+1. / 3 . Then w=2 #*= E w=2 Av+tf £ />'• Kj Mw .3+1. we have Theorem 4.( jr 6(1.4+2. n E£ £ ^ ri M k=i the following products also appear: 1*41. Say n > 3. two of which are of one kind. n1 £ (2) i<j<k Mn Thus. Thus. by Theorem 1.2.268 SUMS OF CON BIN ATI ON PRODUCTS n n [ 0 CT. Say n > 3. we conjecture E Ukand we have i<j<k Mn Theorem 5. 4M. with n= 4. This suggests that E '7*5E'>**E i<s<k Mn j<j M2 ii+ ~'+n E v • /</ Mf i<j M3 Thus. is three. We note that the number of permutations of three objects. Then Mn For example. we conjecture. n n n / J n \ k = n i3+6 k+I 3 n 2 3\ E E E 'i i=i j=i k=t ( n 5 H i<j<k Mn \ \ j i<j<k / E n32* \ /»/ / n i3n* \ = lLi3+6 i=1 i '3i r 1=1 n n2 E i=1 E 'J [ E i=i iik + 3Y^i(12+22+'''hn2}3(P+2*+ E M i<j<k \ i=1 j \ i=1 + nz) i=1 ) i=1 Mn Thus.
5n4 +n3 + 5n2 i<j<k • f Mn We can prove Theorem 5 by using Theorem 4 and the following formulas: n n n 24 2 Y^'= "2 +n' 3 i=i Y1p =n * + ^j i=i i=1 + 4 j * 1=1 J2 +2n. i=1 . ]£ ^ (3) t<J<k<Z Mn Thus. we conjecture.19761 SUIVIS OF COMBINATION PRODUCTS 200 1L1 ijk = Y\ (3i5 +5/4 . Then (n . An5. */ Xj<<xr/ 'w Thus.Si2 . Say A? > 4. Theorem J . Consider: 4(1 23) + 5(1 23 + 124 + 1 34 + 234) = (1234)+ (1235 + 1245 + 1345 +2345) This suggests that ^ ijkz =4 i<j<k<z M n J2 '*k + 5 Yl Kj<k ijk + ~ + n ^ j<j<k M3 ijk .?. . i<J<k M4 Mn1 Thus. w <*&• (yw ~ yw ~ 3w* y~ y . by Theorem 4. Then Conjecture 24 //*« . S a y n > r > 3 .2 i ) 3ns . Conjecture 2 and Theorem 1 provide a recursive method for determining A^. Then n1 E £ */.. we have Conjecture D.!"0*1*.i)i = nn'1 + £ (li'A?1. Say n > 2. we conjecture.E ^ x1<<xr w=r1 r_1 E . n1 E /</<*<£ Mn and we have n1 I] (w+1> £ ft w=3 i<j<k Mw 6 5 3 .i = 5 4+ +/1 s+ 4 3 2 E '^ = E ( f "" ' ¥ ) r ^" " f " /</<*<« i=1 Mn Comparing (2) and (3) we have Conjecture 2. /B = n +2n +n2 * * ' 1=1 C. We now develop a formula for A%.t/u2 w w=3 1./ 3 . A%. 6.
1)(n 2) [n . Theorem An3"1nn"4 n1 + .270 SUMS OF COMBINATION PRODUCTS Proof.r derivative avaluated at zero.« + (1) ' A ThUS + (~1)1 Anf1 1 n nZ 1n ^ nn~2 + (V2 A n2"1 n n ~2 . comparing the coefficients of the above two equations.1)! = (n . x3 =2 where Q V(V. + An21n2+ An1xn~1+ An2xn'2+ + Ann_1 11 + Ann = f(D 1 n 2 1 Ap" + A2 2 ~ + ••• + A^_j2 Ar\nn1 + An2nn2+~ Ann + + Ann_in1+Ann = = +Ann_1x+Ann. F.E v E ' . n E Proof. (5) (7) *i * £ W(V<Q>v=*2 Tx = (6) [1] T [1+2] I 22 = —l I r d.7 £ *> £ *•"£ *> x *2 x2=v1 xz=v2 Q :. n [OCT.(n ~ 1)] = nn~1 3 + (1) E. Then f[nrl(Q) .HOlH^fOk* . A Curiosity. . xx1 (4) f(x) . E *>£** E > E xx=1 Q xx=2 xt1 x i~3 x31 x2=1 x2~1 x2=2 xz~1 £ *i x*=3 E ** £ ** £ ^•••^w^.»_ (n f)l fnrl ' ' J where r (0) denotes the n . Thus. where the A" can be solved for by Cramer's rule.Q) = (n¥ Y* x. by Taylor's formula. Proof. Say f(x) = (x + Dfx + 2) •• (x + n) = (x + n)!/xL Then f(x) = xn+An7xn~1 is a polynomial of degree n. x21 x ^ . 7. The A" can be solved for by Cramer's rule.J Tz =[1+2 This suggests + 3] + [2(1)+ 3(1+2)]  x^f Q { 3[2(1)]\ T +[2(1)] L*\ l/J = [1 + 2 + 3] + [(21) + (31) + (32)]  [(321)]. + . Also. = <x + n)!/xl. TnmfM=xn+ LetfM = (x+ 1)(x+2)(x+n) An1l"i ' 1 A? = (n + 1)!1. (n . x .. G. Theorem 8.=v x47 Xj7 x.+ Ann_ . the theorem is proved.. 1" f(2)2n f(n)nn. Sayn>r> 7 and f(x) = (x + n)!/x!. Now f(x) f[2l <°>x2 + •••+ f@£ /l n! Thus.M £ . Let Q Q Xj~. TQ = J2 xt=1 Thus.
Hence.E *. by (8). Let's see. We leave the proof to the reader.1976] SUSVIS OF COMBIWATiOiy PRODUCTS 271 Conjecture 4. n xl1 x21 xt=3 x2=2 x3 = 1 n i1 j—1 i=3 j=2 k=1 We note that T3 = T2 3 + 3[1 + 2] .1 . Thus.3[2(1)1 = T233T2 and T4 = T3 4+4[1 + 2 + 3] . Say Q > 1. By (5) we know that Tt = .3 and F4 = 3T3 . Then.4.7 /? n i—1 n ^ =x. T3 = 2T2 . This suggests Theorem 9. .I T2=3 + 2=l T3 =6+116 = 1..1 and the theorem is proved. we have Conjecture 6. This suggests Theorem 10.1 .e.S . Then (8) TQ = (QDTQ^Q. Say k is a fixed integer greater than or equal to two and 7V./ .xE=7 *> zz^z i=2 j=1 i=2 2 = 1 J\n(n+1) f ~2\l n(n + 1)(2n + 1)\ } 6 2 \ ~ which agrees with Theorem 1. n r„ = E '"W • / / We note that from Conjecture 4. Similarly. (6). from Conjecture 5 and the above theorem. i. E *• x x =2 x2 = 7 fl ••A"i= x 2 =2 x27 E *« E ** E x* xx=n1 and x^J Xj/ x2=n2 x 3=113 x3 = 7 *n2"1 E x"? XnI^l Conjecture 5. E *. ? x. A n xx~~1 n A xt1 x21 "2 . We might hope that the TQ represent a new species of number. Say Q > 1. (7) we have T i =.1 . from (5). E *.4[2(1) + 3(1 +2)] + 4  3[2(D] \= T344T3. n E (D'AP = . Then TQ = .E *.2 E *. T^ = ./ = . Proof (induction).
and IVSARJORSE BfCKNELL San Jose State University. in [1] Good shows that d) ±i. Method I. »'•&•'•£• which suggests that 1 Fr (2) 2 4 2n 2n From [3]. PART XV VARIATIONS ON SUMMING A SERIES OF RECIPROCALS OF FIBONACCI NUMBERS V. we can prove (2) by mathematical induction. we could show that n ' nn + °° F n J. However.2m+1 L 1 + AJZ1.. Write out the first few terms of (1). If we compute the limit as/? >• <*> for (2). San Jose. . then we have the infinite sum of (1).Thus.n an '^ff' In"**?' where a = (1+V5)/2. 1 = 0.ar .^ t(fi/ajn since fa. « » . to derive the sum of a series whose terms are reciprocals of Fibonacci numbers such that the subscripts are terms of geometric progressions. im (apjar+gL..yjS )/2.. in general. we write (3) from which it follows that LmLm+i L.OO Jim ^ Ln+r/Ln n^non a „n « ar.^ „ . JR.. 1 + 1 + 3 i + 2 T = 2 T ' .i=Ji 2n n=0 a problem proposed by Millin [2]..j3 = (1 >/5)/2 are the roots ofx2 xurn L _JL= . 1 + 1.. California 95192 It is not easy.§) = y/5 andj3/a< 1.2. The limits used above can be easily derived from the wellknown r. This particular series can be summed in several different ways.«> /? Jirn^ Fn+r/Fn 272 . . Now. 1.A PRIMER FOR THE FIBONACCI NUMBERS. HOGGATT. E.+ n n 2 2 = (Vm = 1+ 2—~f 2nH 2"+1 since FmLm = F2m. 1 + 1+jJ. for (see [3]) where a= (1 + \j5)/2f which simplifies to (7 .m (av 1+JteZ.. In an entirely similar manner.
L =31 2 £=! 2n .p n+1 p even.7. L nF n — 1 F n+i 3 .o .1 _ 1 r n 2 ? 2n 2nf n+1 2 2 2nl1 _n n+1 2n where we need L nF n = F n+1 2 2 1 which follows from [ 3 ] .+ 4. [4] 2 1 Fm+p + Pmp = LmFp.V 5 ) / 2 . 8 .. + Fl p odd.= 21 21 suggests F (6) * + 1 + . we obtain 2 + 1/a2= (7 >/5)/2. 2n used by Good [ 1 ] . so that F nH 2 2~F2= F n 2 2L2n • Method III... + L.+ .4 _ 34 *L== A4 _F9_ 21 21 FB and n l.. V2 .2.. since . p = 2 . F nH 2 2»F2»2+1 F 2 2»+1 We need to establish that F n 2 2L2n*1 = F 2n+K2 which follows from (see [ 3 ] . wherem+p = 2n+1 . 50_ = 33_ 'l i = 3 15. . ieie A PRIMER FOR THE FIBONACCI gyUEViBERS 273 Method II. Returning to the first few terms of (1).+. Examining the first terms of (1) yet again.= 2+ n 2n 2 If we take the limit as n><*> of the righthand side of (4).2.42 n F pl2n 2.4a4l±&Z=j£ >°°\ I F?n 2 if indeed (7) 1+1+. m . = FpLm. n n ..OCT. m = 2 . we notice that 50 = . We can prove (4) by induction. [ 4 ] ) (5) where m + p = 2 Fm+pFm.2 . F n / <F 2^HF2»J/F2«+1 _2: F n+1 2 L _ ..p = 2. Establishing (6) by induction involves showing that r%n 4 . where the limit as n > <*> of the righthand side is 3 . p=2nI Method IV.. F6 21 21 Z Fft ' which suggests (4) 2 "~2 4. Proceeding in a similar manner. mp=1. 50 = 7 .+ 1 Thus one expects 2 1. m = 2n.1/a= ( 7 .m I AJ^}.
274 A PRIMER FOR THE FIBONACCI NUMBERS L 2"F2n+1~1 [OCT.< * > i s 5 .( a + l ) = 4 . Fm+p + Fmp = LpFm> n+7 n n \Nherem+p = 2 +1. which simplifies to (7 .a.1)] = 2 +J5 [1/(^/5a)(a)] = 2+1/a 2 . . [4] (8) P even.a 2 = 5 . We again return to the early terms of (1). ^ 4 + 1 = ~T 21 F. mp=1.p = 2.= 21 " q Fl0 FB suggests (9) 1 + 1 + .1)] .s/b )l 2.. 7 + 1' = _ 02J+. From the form of (9) and earlier experience.J0+1 = ^. ^ J L * . one expects F n L n — 1 = F n+1 2n+2 2n 2 + 2 which follows from (8).y/b a. f. L ^ p^ becoming 6 . Again looking at the early terms of (1). Method V.+ oo p^ JL. The limit as/7 » *> of the righthand side of (10) follows from Jim = n^oo Jim f^ n. where/w +p = 2n+1 +2. Method VII. Method VI. m . + _ L l 2 = 5 _ *j* 2n 2n where the limit of the righthand side a s / 7 . we inspect the early terms of (1) to observe 76 » • • which has the form of 7 F> (10) *..* .J _ = 2 T * » £ ' * * » * 1 = 2i L»+L*+L* F6 Fl6 Fu Fl6 + * Assume that (11) Since E + V?29Ji = 2* j=o L„n +Ln +L„ 2"4 2'8 2"12 F n 2 Jim *°° n ±E=L Fm = ++L4+I j5ar the limit as n > <*> of the righthand side of (11) becomes 2 + j5(a4+a8+ar12+•') = 2 + j5[1/(a +0 = 2 + ^/5or4[1/(12 2 or4)] = 2 + J5f1/(a4 + 1)(a .a again. m=2n +2 and p = 2n. p=2 . but we proceed in a different manner. m=2 + 1. One last time.R IT6" L n = ff. 50. 2 + ±1+A_ J 13 21 2 + _ 20 J+. F n+1 " 2 H which follows from [ 3 ] .2 +1 *7f The proof of (10) by induction depends upon the identity L n L n 1 = L nH 2 +i 2 ~ 2 H which follows readily from (3).
every even number greater than zero can be written as (2j+ 1)2'. Method IX. (11) can be proved by induction if the identity (12) Ln(L/n + L. and proceeded as in [ 6 ] ."* . I. Good [7] uses the identity D (xy)2 /(x^y2") = m i n y _ y .n +~°+L4+1) 4 2ni (2n4) (2n8) = L n+i + L n+1 2n l4 2n 8 ++L4 is known.2kn ~ ^2k ~ 2 to sum the numerator of (11). We could also have used /1Q\ V^ / L  2k(n+1) ~ L. . 2 2 + 1/a =2 + (3. since a" = (Ln + Fny/5)/2.1176] A PRIMER FOR THE FIBONACCI NUMBERS since a2 = a+ 1 a 2 + 1 = a+2 = L±j£ and 2 Also. £ ^ . this is . On the other hand. Method VIII. " 4 " > E 1/F oi 2 1=0 J L n = V}+'2"2 +L n 2"4 ++L2+ 1 2" but L2m + L2m2 + ~"l'L2 = ^2m+1 ~~ 1 Thus E ^'F2r H5) 1+ f~.s/5)/2. L Carlitz [8] uses n=o i=o a2 $2 hi a2 $2 i=i\.y/5)/2. Starting with the first few partial sums. — n=1__ where x = (1 + ^/5)/2 and y = (1 . (See [5]). + 2 = 5^ = s/5a. J.s/5)/2 = (7 . n1 ^ 2 .= A ?0 so that Jim A = 1+s/5/a= (7sj5)/2. Generally. Here. This is not quite complete by itself.Han £ . i^M . Method X.. Thus. so that this is oo (*vnz« / oo (2j+i)2l ]+i i=l\j=0 but clearly.=o 2 but (a/3) = 1.. . and the above becomes a = (3 + sj5)/2.
and Marjorie Bicknell. 193196. 346. The Fibonacci Quarterly. pp. Ruggles. 309. Jr. .log — — n— — ——— . HoughtonMifflin. Jr. Millin. /?*. to appear. " A Reciprocal Series of Fibonacci Numbers/' The Fibonacci Quarterly. Hoggatt. kkkkkkk [Continued from Page 253. q*. Hoggatt. G. "Fibonacci Summations. J. q. 6.276 A PRIMER FOR THE FIBONACCI NUMBERS OCT. " A Generalization of a Series of De Morgan with Applications of Fibonacci Type. 3 (Oct. Shannon's solution in the April 1976 Admwiced Problem Section solution to H237. S. 1. E. 4 (Dec. Vol. Vol. Hn and H% have the same meaning as in Theorem 2. mod 1. private communication. We have Wn+1Wn • Hn+1 . n n ^ (q+q*) + —(P+P*)Fn1 . 14. 3 (Oct. "Some Fibonacci Results Using FibonacciType Sequences. V. 1974). I. Let/7. A. J. 1969. Vol. 1974). 3." The Fibonacci Quarterly. 7580. pp. Bruckman. p. pp. 5." The Fibonacci Quarterly.2 (n > 3) and so we see that *n+lxn [Continued on Page 281. By the definitions of Hn and H„ we have Hn + H* = (q+qVFnj + (p + p*)Fn. 1976 Method XI. 8. Hn+1 = log . No. 4. L Carlitz. Vol. Proof. No. mod 1. 1. For yet another method see A.] » ((M nu +H^u* )/(H +H urn )) +u*n = log n+1 n+1 n n F . p." The Fibonacci Quarterly.. 2 (April 1963). 1976). 2. 6769. Boston.d. I. E. V.. 12. I.1 +pFn„2 ~ q(Fn/Fn^)+p q(Fn„f/Fn2) +P Fn1 Fn2 goes to 1+x/T 2 as/7 »°° Theorem 3.7 7 — "Hog 7J17 which tends to 2 l o g l ^ as/7 *°° for Mn+1 „ qFn+PFn1 qFn. 12. Then the sequence (xn) = (\Q%(Hn + H*)) is u. Ken Siler. D. 3 (Oct." The Fibonacci Quarterly. D. No. 7. 1963). Good and P. Problem H237.d. REFERENCES 1. No. Good. No. Proof. Vol.] Then the sequence (wn) = (\oqHnH*) is u. Fibonacci and Lucas Numbers. " A Reciprocal Series of Fibonacci Numbers with subscripts k2n.
The situation considered by Fibonacci involves two types of rabbit which will be denoted B and F (for baby and female. Family Tree of Fibonacci's Rabbits 277 . Binghamton. a sequence of generations of rabbits is formed as follows: Each individual of type B in the nth generation matures to become an individual of type F i n the (n + 1)st generation. Wew York 13901 In 1969. each individual of type F in the nth generation gives birth to an individual of type B in the (n + l)st generation. see Figure 1. Also. and survives to become an individual of type F in the (n + l)st generation. A family tree may be drawn which represents this process.A MODEL FOR POPULATION GROWTH DAVID A.KLARWER State University of Wew York. In this note we describe how one might treat a generalization of these problems. generation number 0 Figure 1. we state the problems of Fibonacci and Parberry and note what they have in common. respectively). First. Parberry [1] posed and solved an interesting problem in population growth analogous to the rabbit problem considered by Fibonacci. Starting with one individual of type B.
when m = 2. . • .(n) denote the number of individuals of type / in the nth generation. and put f(n) = fj(n) + . —. Let/77 and n denote natural numbers. —. n can only mature to become diatoms of type Sm+j+i.j< fjfor/7 = Of 1. m split to form two new diatoms. individuals of type i. there is an initial population containing f. where a diatom of type Sm+n+i is defined to be of type Sj. For example.278 A MODEL FOR POPULATION GROWTH [OCT. (Thus. but diatoms of typeSm+j for/ = 7.j individuals of t y p e / in the (n + 1)stgeneration (1 <i. Family Tree of Diatoms The problems of Fibonacci and Parberry have common features which are embodied in the following generalization... . Diatoms of type Sj for / = 7.+ ft(n). Let f. and let $1* ' " / Srrif Sm+1 * "V $m+n denote a classification of the diatoms. There is a finite set T  7. and each individual of type / in the nth generation gives rise to f. Changes in classification along with reproduction are assumed to take place at regular intervals which will be called generations. generation number Figure 2. n = 1. Parberry considered populations of diatoms. one of type Sj and the other of type Sj+f. Also. t \ of types of individuals. The sequences . the family tree of diatoms descending from one individual of type Sj is shown in Figure 2. onecelled algae whose reproductive capabilities can be classified according to size and maturity. = f2(0)). f.
. this implies (6) f. called the minimal polynomial Qi F. if so. FncF(F) (4) = 0 ior n = 0. This polynomial. It is well known and easy to prove that there is a polynomial. Recall the CayleyHamilton Theorem: Every square matrix M satisfies its characteristic equation. where / denotes the t x t identity matrix. 1. Hence.) satisfies this difference equation. •" . j < t.V individuals of type / in the (n .j(n) denote the (jj) (5) th ajXt" at. also. hence. Using (2). then cp(F) is the allzero matrix. —) satisfies a certain difference equation. Let cp(x) = xfand let f. this may be succinctly expressed as (2) l(n) = f(n 1)F. This may be expressed in terms of matrices as filfl2'"flt f21 f22ht ffjfn) ft(n)] = [f7(n .l ) + . the matrix involved is fO 1 ' / 7 (?!) . any linear combination of these sequences also satisfies this difference equation.1) ft(n . this polynomial may be used in place of cp(x) to obtain a lower order difference equation. I . 1. 1.1976] A MODEL FOR POPULATION GROWTH (fi(n):n = 0. what is their rate of growth.) satisfies the same difference equation given in (5). Thus. Then (4) implies fjj(n + t) aif. In particular. The matrix F may satisfy a polynomial equation with degree less than t.1)st generation gives rise to fy individuals of type k\x\ the/7 generation. if we form the polynomial cp(x) = det (xl .+ atfj(n) for A7 = 0. ••• and / < i. the sequence (f(n):n = 0. 279 1.1)] f t1 f t2 or. and so on. with an obvious notationa! convention. how should they be calculated. having minimal degree such that F satisfies the corresponding polynomial equation. 1. Each of the fj(n . There is a very simple theory which explains these things. the sequence (f(n) : n = 0. Since each of the sequences (fij(n):n = 0..F). —) also satisfies this difference equation.j(n + t1) atfjj(n) = 0  for n = 0.~) are of interest: How are they related. hence. unique up to a constant factor. n entry of F for/7 = 0.(n): n = 0. an easy induction argument gives (3) J(0)Fn. is a factor of Cf(x). f(n) = Now (3) can be used to show that each of the sequences (f.(n + t) = a1fj(n + t . 1. summing on/we have (1) fk(n) = f1(n)f1k + +ft(n)ftk. 1. Returning to Fibonacci's problem. I .
280
A IVIODEL FOR POPULATION GROWTH
The minimal polynomial of this matrix is A2  X  1: Hence, fin), the number of rabbits in then
satisfies
(7)
fin +2) f(n + 1)  fin) = 0
[OCT.
generation
for/7 = Q, 7, •••; also, we have f(0) = f(l) = 1, so this gives Fibonacci's sequence 1,1, 2, 3, 5, 8, —.
A more realistic model of a rabbit population would reflect the fecundity of the female depending on her
age. For example, type 1 matures to become type 2; type 2 has a litter of 3 type 1's and matures to become
type 3; type 3 has a litter of 4 type 1's and matures to become type 4; type 4 has a litter of 2 type 1's and
dies. The matrix involved is
0
3
4
2
(8)
1
0
0
0
0
1
0
0
0
0
1
0
and the characteristic equation is
x4 3x2 4x2
= (x+1)(x3 x2
2x2).
Suppose the initial population consists of one rabbit of type 1, then the family tree shown in Figure 3 results.
generation number
DOO© ©OO© ©©©@© C
Figure 3
The number of rabbits in the n
(9)
th ,
generation satisfies
fin + 4) = 3 fin +2)+ 4f(n + 1) + 2f(n)f
but it is easy to check that the initial conditions
HO) = f(1) = 1, f(2) = 4,
and
f(3) = 8
give rise to a sequence 1, 7, 4, 8, 18,  which satisfies the lower order difference equation
(10)
fin + 3) = fin +2) + 2f(n +1)+ 2f(n).
1976]
AfVIODEL FOR POPULATIONGROWTH
281
This relation arose because
x3 ~x2
2x2
is a factor of
x4  3x2
4x2.
Since
x3x2~2x2
has a real zero 9 between 2.2 and 2.3, it follows that
fin) > (2.2)n
for all sufficiently large n.
REFERENCE
1. Edward A. Parberry, "A Recursion Relation for Populations of Diatoms," The Fibonacci Quarterly, Vol.
7, No. 4 (Dec. 1969), pp. 449456.
[Continued from Page 276.]
which tends to
„
.log 1 ^ ^
as n » <*> and this completes the proof.
In addition we want to mention another interesting property possessed by the sequences of the previous
theorems. This property can be shown by applying a result of Vanden Eynden (see [2] p. 307): Let (Cn) be a
sequence of real numbers such that the sequence (Cn/m) is u.d. mod 1 for all integers m > 2. Then the sequence ([Cn]) of integral parts is u.d. in the ring of integers Z .
Theorem 4. The sequences
([\mFn/kll
([\o$HnH*])
and
f/log (Hn + H*n)])
are u.d. in Z .
Proof.
It is easily seen that for all nonzero integers m the expressions
i l o g F1n/k,
1 log (HnH*)
and
± log (Hn + H*n)
satisfy the condition in van der Corput's Theorem.
REFERENCES
1. William Webb, "Distribution of the First Digits of Fibonacci Numbers,*' The Fibonacci Quarterly, Vol. 13,
No. 4 (Dec. 1975), pp. 334336.
2. L. Kuipers and H. Niederreiter, "Uniform Distribution of Sequences," 1974.
ADVANCED PROBLEMS AND SOLUTIONS
Edited by
RAYMOND E. WHITNEY
Lock Haven State College, Lock Haven, Pennsylvania 17745
Send all communications concerning Advanced Problems and Solutions to Raymond E. Whitney, Mathematics
Department, Lock Haven State College, Lock Haven, Pennsylvania 17745. This department especially welcomes
problems believed to be new or extending old results. Proposers should submit solutions or other information that
will assist the editor. To facilitate their consideration, solutions should be submitted on separate signed sheets within two months after publication of the problems.
H264 Proposed by L Carlhz, Duke University, Durham, North Carolina.
Show that
mr
XT* ls+i\(m+nsi+l\
i M
/ )[
i=0
ns
ST*
" Ls
i=0
=
ns
)
I r +i \ I m +n  r  / + 1 \
[ i J [
mr
'
H265 Proposed by V. E. Hoggatt, Jr., San Jose State University, San Jose, California.
Show that F
3
k„7
= 0 (mod 3*), where k > 1.
H266 Proposed by G. Berzsenyi, Lamar University, Beaumont, Texas.
Find all identities of the form
k=0
with positive integral r, s and t
SOLUTIONS
TRIPLE PLAY
H238 Proposed by L Carlitz, Duke University, Durham, North Carolina.
Sum the series
S=
xmynzp ,
£
m,n,p=0
where the summation is restricted to m,n,p such that
m < n +p,
n < p+m,
p < m + n.
Solution by D. Russell, Digital Systems Lab, Stanford, California.
If m + n + p is even, then either (1) exactly one of m, n, or p is even, or (2) all of/??, n, and/7 are even. In either
case, m +n  p is also even. Leta= 1Mm + n  p), and similarly l e t £ = 1Mn +p  m) and c= 1Mp +m  n); because
of the restrictions, all of a, b, c are nonnegative. Then m = a + c, n = a +b, and p =b + c, and
xmynzp
= xa+cy*+t>zt>+c =
(xy)a(yz)b(xz)c.
This is a general term of the generating function
Teven
t
7
(1xy)(1yz)(1xz)
282
2 One of the four squares on the righthand side of (2) must be 1 . Consider the terms where m + n + p is odd and m + n + p > 3 (no terms exist with m +n+p = \). A. or p is odd. FERMAT' INEQUALITY H239 Proposed by D. in the restriction m <n + p. equality may not hold.7. c> d. Let the smaller prime be a2 +b2. Prove that \c \ a d\< b\ 1 100' Also.OCT. For the last part of the problem. The terms xmynzp = (xyz)xm'yn'zp' = xyzTeyen and all terms of 7"0dd a re easily seen with m + n+p odd are thus terms of the generating function T0^ to satisfy the restrictions with m + n +p odd. and/? are odd. Now rand / a r e approximately equal m&p2/px p2 ~ r2. n. Bruckman. U_ d\ < _i_ \a b\ " 100 ' follows. then it is well known that each prime is of the form 4A?7 + 1 and each has a unique expression as the sum of two integer squares.be)2 = (ac . it is obviously not the first one on the top line or the last one on the bottom line. b. c. It is well known that (1) (a2 + b2)(c2 + d2) = (ac + bd)2 + (ad . Finkel. since then one side of the relation would be even and the other would be odd. Hence. Solution by the Proposer. Here 641 = 252 + 4 2 and ab = 100. P. n. Brady.177s5 51€2 +89*. .I lhmxm yn zp satisfies the restrictions and / T J ' ^ ^ ' ^ P ' is even. 0. then m . But if m <n + p. and the larger prime be c2 +d2. a simple calculation leads to = 80831112 + 13941802 .* Hence the result. andtf as positive. Recreations in the Theory of Numbers. or (2) all of m. Takings. 175. Let m' = m . If a Fermat number 2 + 1 is a product of precisely two primes.b2r')2 > 1. New York. Klamkin. 1976 ADVAWCED PROBLEMS AMD SOLUTIONS 283 and it is easily seen that all tem$xmynzp of 7"even satisfy the restrictions and that m+n+p\s even. In either case. Lossers.310. pp. Brooklyn.7 <n +p 2. Either (1) exactly one of m. Since c = ar and d= hr'. Clearly ac > bd and thus (a2r. W. given that 2 2 ' + 1 = (274.417). let the smaller prime b e ^ = a2 + b2 and the larger prime be p2 = c2 +d2 = a2r2 +b2(r')2 .r l = ~ ' ab .I p'= p . Let c/a = r and d/b = r'.177) (67. and the Proposer. No other Fermat number can have an ab product as low as 100.700. (abr'abr)2 = P or I1 r ' . Since m + n + p is either odd or even. *Beiler. 143.721) and that 274. the sum S is given by q 7 +xyz " (1xy)(1yz)(1xz) ' Also solved by P.bd)2 + (ad + be)2. a > b.280. Shannon. n' = n . M. Then (2) (a2 +b2)(c2 +d2) = (a2r + b2rT +(abrfabr)2 = (a2rb2rr)2 + (abrJ' + abr)2 . The smallest Fermat number which is a product of exactly two primes is 2 2 5 + 1 = (641 )(6.421. express the 14digit prime as a sum of two squares.
s=0 oo oo.n.n.s=0 Then efx)e(y)efz) e(xyz) V ~ . i=0 where W / = f / . .p) S(m.qj).» f / .$* i+j=n i+k=p qrni+(ni)(pi) — i=0 (q)i(q)ni(q)Pi H so that BjxMiMil eixyzi = y L m.. It is well known and easy to show that oo E M_n_ n e(x) = (q)„ e(ax) ' n=0 where (a)n = (1 .p=0 xmynzP „ <q>m V .. REMARK. Durham.qn'1 a). . qmixm(vzr (q>m(q)i E Xmynzp m.n.? V .n. Since vm.n.(1 .mi+jk s i (q)i(q)j(q>k ' .f"n J. oo oo r=0 s=0 r=0 r=0 so that E n qr xrys (q)r(q)s = e(x)efy) efxy) ' r. (q)o = 1 Show \\\dXS(m.p=0 = V* V ^ m.284 ADVANCED PROBLEMS AND SOLUTIONS [OCT.p=0 The stated result follows at once.q)(1 .a)(1 .p) is symmetric in m. Put ff (1qnxr1 n=0 e(x) = = T jr *~i Wn n=0 . EGAD H240 Proposed by L Carlitz.p) = (q)n (q)p mi+(ni)(pi) . (a)0 = 1.n_p _ ^ V L _ (q)m(q)n(Q)P s(mnp) ' " P r.qa) . . Solution by the Proposer.k=0 m. Duke University.i=0 n (q>m e(x)efyz) e(y)e(z) e(xyz) e(yz) qjkvizk _ (q)j(q)k ' J v* .p. North Carolina* Let m\n(n. T] .n. It follows that oo r.
145. 1xe n n ^ j 2k7T K u n . .Z . mn ? 0.V= I Also solved by C Bridger. { 1} / n=0 it follows from (*) that 6d?!L — m.1976] ADVANCED PROBLEMS AMD SOLUTIONS 285 q1/2n(n1)zn (a)n J _ . n. New York. Integral Calculus. Smith. In the Pellian equationx 2 . tn even.1 n1 i n 1 X ~ 2* k=0 n1 2irk(n1). Pennsylvania. Using partial fractions.Dy2 = 1. New York. Prove that nJ _/ v 1 7 hT. Wulczyn. we have n1 1 _ 1 *  y* _ 1 k=0 Ak x  e — ' As1) [e. Lewisburg.V (1\» e(z) . Bucknell University. see Edwards. p. [m2tn + 3mt2n+4](m2tn+3m) = 4D. PELLMELL H243 Proposed by G. [(mt 2 n + 1)(mtn + 2)+2] y= tn. 1. Wulczyn. College of Insurance. Show that for each triangular number tn = ^(n+ 1) there exist an infinite number of nonsquare positive integers D such that t*+r. (2) m odd. and not an integer square since the difference of these two integer squares is never one.nj)) ** (q)s(q)ms(q)ns(q)Ps s=0 ^ HARMONIC H241 Proposed by R. letx = tmtn+i.n.P=0  ij. Solution by the Proposer. Bucknell University. Lewisburg. The lefthand side is congruent to zero modulo 4 for the conditions (1) m an even integer. tn odd. (3) m odd. P. \ ' [ xe"^' j n n1 Ls k=0 2nk(n1) n ** k=0 i xe~» 2lTk 1xe n since (n.. Solution by G. Garfield.t^D = 1. Pennsylvania. [(mt2 + 1)(mt2n + 2)2] = 4t2nD.k0 —Wlk S=0 so that min (m. ne~"* I 27Tk. Vol.g.. and the Proposer. Hence D is an integer.
B336 Proposed by Herta T.HILLIVSA1 University of New Mexico. Show that TO is an integral multiple n 7 of2 ~ . 53. Lewisburg. respectively. Let/? =4k+ 1 and let h be the largest integer with 2b + 1 <n. Show that there are infinitely many points with b o t h * and / rational on the ellipse 25x2 + 16y2 = 82. 7 = 0. Freitag.. Hillman. (For example. Roanoke. Preference will be given to those typed with double spacing in the format used below. Albuquerque. Pennsylvania. Also a and b designate the roots (1 + >/5 )/2 and (1 . Slippery Rock. Let Qo= 7 = CLi and Qn+2 = 2Qn+i + #/?• Show t h a t i Y ^ . E. B339 Proposed by Gregory Wulczyn. S. Slippery Rock State College. New Mexico. Obtain a closed form for nk ]T (Fi+kLj + i=0 FjLj+k). 709 Solano Dr. ofx 2 x Z. Roanoke. Albuquerque. 29. Lamar University. the powers of u are used as either Fibonacci or Lucas subscripts.F. / = 7. after the binomial expansion has been performed. Ff = 1 and Ln^2 = Ln+1 + Lnv LQ = 2. Are all the terms prime in the sequence 11. B338 Proposed by George Berzsenyi. P. Each solution or problem should be on a separate sheet (or sheets). New Mexico 87131 Send ail communications regarding Elementary Problems and Solutions to Professor A. New Mexico 87108. Solutions should be received within four months of the publication date.ELEMENTARY PROBLEMS AND SOLUTIONS Edited by A.334 Proposed by Phil Man a. 17. Albuquerque. Freitag.. Brady. DEFINITIONS The Fibonacci numbers Fn and the Lucas numbers Ln satisfy F n+2 = F n+1 + F n* F0 = 0. Texas. 3. Virginia.yjb )/2. Virginia. Establish the validity of E.5 for/7 > 0? B335 Proposed by Herta T. Pennsylvania. Cesko's symbolic FibonacciLucas identity (2u + 7)n = u3n. B337 Proposed by Wray G. Beaumont. PROBLEMS PROPOSED IN THIS ISSUE B. 2. when n = 2 one has both 4F2+4FX+F(} = F6 and 4L2 +4LX+ L0 = L6 .1) is a perfect square for/7 = 1.2un . ••*.) 286 . Bucknell University. Let k and n be positive integers. defined by u0 = 11 and un+<i .
Wynnewood. Polytechnic of the South Bank.B. Martin C. New York. Find some positive integers n and r such that the binomial coefficient ( n ) is divisible by n + 1. Let k be a constant and let j an i be defined by an = ani+an. Bruckman. r = 21. Harvey J. 7. 0£zero. W= 4. Quebec.OCT. Also solved by the Proposer. Hunter. 21. then / = 2. Paul S. 41. . One can show that n + 1 must have at least three prime factors. n = 83. Hindin. and the sum of the digits in the hundreds' column is<30. With ao . Ontario. R = 1. Solution by Graham Lord. Solve the doublytrue alphametic ONE ONE ONE TWO THREE EIGHT Unity is not normally considered so. California. 14. Shannon. 18. 1978 ELEMENTARY PROBLEMS AND SOLUTIONS 287 SOLUTIONS SPECIAL BINOMIAL COEFFICIENTS B310 Proposed by Daniel Finke/.0 = 2. 41. Ralph Garfield. 22. Canada. 23. Weiss. 13. But if N = 2. G. or 9. r = 15. San Diego. H. Gregory Wulczyn. T+1 = E. 37.A. but here our ONE is prime! Solution by Charles W. r = 36. A.22. since ONE is prime. 7. Universite Laval. n = 69. Then 4E + O = T+ 10k. David Zeitlin. 19. then T=%. John W. If E = 3. For/7 < 100. and the Proposer. a1 = I Find n\\mJan/Fn). Milsom. then 7"= 6. 14. England. Trigg. and G = 3. 15. A. Mike Hoffman. Toronto. Hence. and 0 = 8. £ £ 9 . London. and ONE is prime. If £ = 7. If £ = 9. Fun with Figures. whereupon N = 2 or 5. 40. n = 29. with 2r < /?. then 0 = zero. 20. 40. n = 89. r = 6. Frank Higgins.23.42. N = 5. Freitag.0 and 3/ = 1 then an = Fn + (Fn+f . Pennsylvania. 44. H = 9. 38. DOUBLYTRUE FIBONACCI ALPHABETIC B312 Proposed by J. ANONHOMOGENEOUS RECURRENCE B311 Proposed by Jeffrey Shal/it. Peck. Charles Chouteau.24. Solution by David Singmaster. Herta T. so E= 3. / = 2. Brooklyn.2 + K a0 = 0. n = 59. C. which is not acceptable. r = 12.H= 7. 39. Also solved by George Berzsenyi. Larry Zimmerman. I find the following solutions for f^/7 + 1) [ n). Consequently.1)k (use induction) and so the limit is 1 + ak.
Lm+12~ Lm (mod 10). Also solved by Frank Higgins. which implies2/?^ = ±2 (mod 12). Roanoke. LUCAS NUMBERS ENDING IN THREE B314 Proposed by Herta T. Peck. Martin C. for all natural k. If Ln is replaced by an + (3n then MM becomes \n(lax)(l 0x) which is the same as \n(1xx2). Shannon. and the Proposer. it is also easy to observe that £ ^ 3 (mod 10) iff m s ±2 (mod 12).. MM = \n(1 xx2l (1+2x)/(1xx2). California. Universite Laval. Weiss. Gregory Wulczyn. Weiss.e. Charles Chouteau. 2. University of Illinois. David Zeitlin.B. Graham Lord. Hin din. Mike Hoffman.. it follows that L k = * <mod 10 >*for a!l prime/? > 5. Hoggatt. David Zeitlin.. Jr. and the Proposer. G. Solution by Graham Lord. Also solved by Hai Vo Ba. G. Madeleine Hatzenbuehler and George Berzsenyi (jointly). Martin C. Hence. and the Proposer.m Also solved by Paul S. Quebec. pk = ± 1 (mod 6). A. Chicago Circle. 1i76 The unique reconstruction of the addition is: 857 + 857+857 + 648 + 69177 = 72396.p=±1 (mod 6). we readily find that the cycle has length 12 (i. B rue km an. Mm). RichardBlazej. Show that L u>  3 (mod 1 ° ) f o r a" primesp > 5.A. If we now write down the Lucas sequence (mod 10). John W. A. San Jose. Bruckman. Combining this with the first result above... EXPONENTIATING LUCAS INTO FIBONACCI B313 Proposed by VernerE. eM(x) s i/(i_x_X2) = +F^x Fi + FzX2 + . Herta T. Hence.thatis Ft + F2x + Fdx2 + . California. Virginia. Freitag. San Jose. /. California State University. Ralph Garfield. Shannon. Weiss. M'M = Lt + L2x + L3x2 •/••• = Integrating. Let MM = Lxx + (L2/2)x2 + (Ld/3)x3 + . Milsom. Shannon. C. .288 ELEMENTARY PROBLEMS AND SOLUTIONS OCT. Freitag. Bruckman. . Sahib Singh. HenceeM(x)\$1/(1xx2). For all primesp >b. G. Harvey J. Solution by Martin C. A. Illinois. Show that the Maclaurin series expansion fore **' is Fx + F2x + F3x2 * » . Solution by Paul S. PaulS..