Академический Документы
Профессиональный Документы
Культура Документы
Computational
Techniques for the
Summation of Series
Anthony Sofa
School of Computer Science and Mathematics
Victoria University
Melboume, Australia
ISBN 978-1-4613-4904-4
2003 Springer Science+Business Media New York
Originally published by Kluwer Academic/Plenum Publishers, New York in 2003
Softcover reprint of the hardcover 1st edition 2003
10
A c.I.P. record for this book is available from the Library of Congress
All rights reserved
No part of this book may be reproduced, stored in a retrieval system, or transmitted in any form
or by any means, electronic, mechanical, photocopying, microfilming, recording, or otherwise,
without written permission from the Publisher, with the exception of any material supplied
specifically for the purpose of being entered and executed on a computer system, for exclusive
use by the purchaser of the work.
Contents
Preface
Acknowledgments
xi
xv
Some Methods
Introduction
1.1
Contour Integration
1.2
Use of Integral Equations
1.3
Wheelon 's ResuIts
1.4
1.5
Hypergeometrie Functions
3
4
7
12
A Tree
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9
2.10
2.11
14
14
15
16
17
17
18
21
23
25
26
26
2. NON-HYPERGEOMETRIC SUMMATION
31
Introduction
31
Method
32
36
39
VB
viii
Forcing Terms
40
42
Bruwier Series
43
Teletraffic Example
44
46
10
A Renewal Example
48
11
50
12
A Grazing System
50
13
51
14
Numerical Examples
53
15
Euler's Work
53
16
Jensen's Work
55
17
Ramanujan's Question
57
18
57
19
Conolly's Problem
60
3. BRMANN'S THEOREM
63
Introduction
63
67
68
Convergence Region
3.1
Extension of the Series
69
70
63
73
Introduction
73
Problem Statement
73
A Recurrence Relation
74
81
87
Introduction
87
I-Dominant Zero
2.1
The System
2.2
QR,k (0) Recurrences and Closed Forms
Lemma and Proof of Theorem 5.1
2.3
Extension of Results
2.4
87
87
91
96
99
Contents
IX
2.5
3
Renewal Processes
3
4
5
5.1
6
7
Introduction
2
3
4
5
Technique
Multiple Zeros
More Sums
Other Forcing Terms
3
4
102
103
103
107
108
111
111
111
113
114
116
120
123
129
135
135
136
138
142
144
147
147
147
149
152
157
160
161
165
166
169
172
173
174
References
177
185
Index
187
Preface
Over the last twenty years or so, work on the c10sed form representation
of sums and series has prospered and flourished. Recently, two great books
dealing with the summation of series have appeared on the market. The book
"A = B" by Petkovsek, Wilf and Zeilberger (1996) expounds the theory of
hypergeometric summation and has given a great impetus to research in this
area. The more recent book, "Hypergeometric summation: an algorithmic
approach to summation and special function identities" by Koepf (1998) gives
up-to-date algorithmic techniques for summation and examples are worked out
using Maple programs. Another large group working in the area of c10sed
form representation for sums and series is the Centre for Experimental and
Constructive Mathematics (CECM) in Canada, led by the brilliant Professor
J. Borwein. Many good papers may be viewed on the CECM website at:
http://www.cecm.sfu.ca/.
The main aim of this book is to present a unified treatment of summation
of sums and series using function theoretic methods. We develop a tecimique,
based on residue theory, that is useful for the summation of series of both
non-hypergeometric and hypergeometric type.
This book is intended to complement the books of Koepf and Petkovsek,
Wilf and Zeilberger, it gives an extra comprehensive perspective on the many
methods and procedures that are available for the summation of series. To
the author's knowledge, no book of this type exists which attempts to give a
link, by developing a comprehensive method, between non-hypergeometric and
hypergeometric summation.
The book has intentionally not been written as an algorithmic approach to
summation, no doubt this will be done by other authors. In particular the book
develops computational techniques for the summation of series.
To put the book into context, Chapter I is an introductory one in which
some methods for closed form summation are given. Methods dealing with
residue theory are discussed and various results are extended. The second part
XI
XII
ofChapter 1 investigates a partieular tree search sum and in the process develops
some new identities.
Chapter 2 investigates non-hypergeometrie summation. We introduce our
function theoretic methods by considering a first order differential-difference
equation. By the use ofLaplace transform theory, an Abel type series is generated and summed in closed form. The series, it will be shown, arises in a number
of different areas including teletraffic problems, neutron behaviour, renewal processes, risk theory, grazing systems and demographie problems. Related works
to this area of study will be considered, including Euler's and Jensen's investigation, Ramanujan's question, Cohen's modification and Conolly's problem.
In Chapter 3, we give a detailed proofofBrmann's Theorem and apply it
to the Abel type series generated in Chapter 2.
In Chapter 4, we consider Binomial type sums with some parameters and
prove that they may be represented in polynomial form. The binomial type
sums will be applied in the generalisation of the results obtained in Chapter 2.
Moreover, the binomial type sums have a connection with Stirling numbers of
the second kind and may be applied to problems in multinomial distributions.
The results of Chapter 2 are generalised in Chapter 5. The generalised
series, it will be proved, may be expressed in closed form which depend on
the dominant zero of an associated transcendental function. A connection with
renewal processes will also be made. We then prove that more general Abel
type series may be expressed in closed form which depend on a multiple number
of dominant zeros of an associated transcendental function.
In Chapter 6, a first order difference-delay system is considered and by the
use of Z -transform theory generate an infinite series which it will be proved,
by the use of residue theory, may be represented in closed for. The work of
Jensen will be considered and some work on central binomial coefficients will
be undertaken. A development of Fibonacci and related polynomials will also
be given.
In Chapter 7, we consider arbitrary order forced difference-delay systems
from which finite binomial type sums are generated. By considering multiple
zeros of an associated polynomial characteristic function, many binomial type
sums are represented in closed form.
In Chapter 8, we extend the results of Chapter 6. In particular, binomial
type series with free parameters are expressed in closed form that depend on
the dominant zero of an associated polynomial characteristic function. A connection between the binomial sums and generalised hypergeometrie functions
is made and particular cases, including Kummer's identity, are extracted. We
further generate binomial type sums and prove that they may be represented
in closed form in terms of k-dominant zeros of an associated characteristic
function. Some numerical results and special cases are discussed.
PREFACE
X111
The present book is intended for use in the fields of applied mathematics,
analysis, non-hypergeometric and hypergeometric summation, summation of
series and automated techniques.
Melbourne
March 2003.
A.
SOFO
Acknowledgments
I would like to thank Associate Professor Pietro Cerone for the many hours
of constructive discourse on this subject.
My thanks also go to Professor George Anastassiou, from the University of
Memphis, USA, and to my colleagues from the Research Group in Mathematical
Inequalities and Applications at Victoria University, Professor S.S. Dragomir
and Dr. 1. Roumeliotis.
xv
Chapter 1
Abstract
1.
1.1
This ehapter eonsists of two seetions. The first seetion is abrief overview of
some methods, basieally ones dealing with residue theory, whieh are useful for
the summation of series and their representation in closed form. Some results
given by various authors are generalized and extended.
In the seeond seetion a partieular tree search sum, with some variations is
eonsidered. A number of techniques are utilized, reeurrences and automated
proeedures, that are useful in determining its closed form representation. Some
related results, whieh are new, are also presented.
So me Methods
Introduction
"Ifsomebody should succeed", wrote Bemoulli, "in finding what till now withstood our efforts and communicate it to us, we shall be obliged to hirn". The
problem came to the attention of Euler (1707-1783). He found various expressions for the desired sum, definite integrals and other representations, none of
which satisfied hirn. He used the integral representation to compute the sum,
S, numerically to sevel1 places, yet this is only an approximate value, his goal
(1.1)
8=6'
Euler [46], moreover wrote "There are many properties ofnumbers with wh ich
we are weil acquainted, but which we are not yet able to prove; only observations
have led us to their knowledge. Hence we see that in the theory of numbers,
which is still ve,y impelject, we can place our highest hopes in observations;
they will lead us continually to new properties which we shall endeavour to
prove afterwards. The kind of knowledge which is supported by observations
and is not yet proved must be carefully distinguished from truth; it is gained by
induction as we usually say. Yet we have seen cases in wh ich mere induction led
10 error. Therefore, we shall take great care not to accept as true such properties
ofnumbers which we have discovered by observation and wh ich are supported
by induction alone. Indeed, we shall use such a discovery as an opportunity
to investigate more exactly the properties discovered and to prove or disprove
them; in both cases we may learn something useful". One may imagine the
excitement and sense of achievement when Pythagoras (c.580 B.C.-c.500 B.C.)
first wrote that for a right angle triangle
a 2 = b2 + c2 ,
L (-l)J L -1 - 2 L (-1)1
00
( 5) _ -5
- 2 j=I
P(
2' )
j-I
k=I
00
k2
j=I
j5
2' ) ,
~ (4j)!(1103 + 26390j)
9801 ~
(j!)4396 4j
,
..!. _ 2y'2
11' -
J=O
=L
00
11'
J=o
1 [4
16j 8j
+1 -
8j
2+ 4 - 8) 1+ 5 - 8) 1]
+6 '
( ) _ ~ (-1)j(j!)10(205)2 + 250)
(3 - L.64((2j + 1)!)5
)=0
+ 77)
'
+b
+ b + 1/2, 2a + 2b
where pFq is the hypergeometric function. Some ofthe main techniques, dealil1g
mai111y with residue theory, in the il1vestigatiol1 of the representatiol1 of series
in closed form are now discussed.
1.2
Contour Integration
L
00
(-1) n f (n)
=-
n=-oo
L Re s (?T
j
CSC
?T Z f (z ))
where Resj are the residues at the poles of f (z), we may obtain some classical
results, namely
00
1
1
(1.2)
2
2 = - 2 [?Ta coth ?Ta - 1]
n +a
2a
n=1
and
~ (2 1 2)2
~
n=l
+a
[?T
2
2]
- coth?Ta + (?Tcosech?Ta) - 2' .
= -4a12
(1.3)
2?Ti
?T
dz
sin?Tz (z2 + 1)
(-lt
?T
L n + 1= 2' Cinh?T - 1)
00
1/=1
Flajolet and Salvy [48] apply contour integral methods to obtain some Euler
sums, in particular they recover the alternating term identity, without the use of
residue theory
For another function of the form f (z) = 7rCO;~ 'Ir Z they recover some results,
one of which was originally given by Ramanujan
coth mr = 19n 7
n=l
n7
56700 .
The strength ofthe Flajolet Salvy paper is that it expouses a method and shows
many connections of identities with the logarithmic derivative of the Gamma
function 'ljJ (z) , the zeta function, harmonie numbers and double infinite sums.
The book by Egorychev [41] also contains investigations on the problem of
finding integral representations for, and computing finite and infinite sums asrising in various practical fields. The results (1.2) and (1.3) are also obtained
and extended by Cerone [23] using different methods. The method is described
as folIows.
1.3
B (t - u) <p (u) du
(1.4)
survivor function which gives the prob ability of surviving to age x of a newbom.
The Inverse Laplace transform of (1.4) is
,,),+ioo
B x (t)=_l
2ni
ePtV(p,x)dp
1 - <P (p)
(1.5)
")'-lOO
where
ePX
00
= v (p, x)
l(x)
l(x)
and <P (p) is the Laplace transform of <p (x) . Assuming that <P (p)
simple roots, Pj, which are the only poles in (1.5) then
V (p, x)
J e-PUdu
B x (t) =
_;::,-X- ; - - - : - _
1 has
!-lj
where
00
!-lj
= - [d:(P)]
P
P=Pi
(1.6)
So
= 2:::: ~ = cp (0+) ,
2
. J-lj
SI
'"
= ~ PjJ-lj =
.1
Mo + 1
2 (Mo - 1)
(1.7)
and, in general
Sn=2::::-1j
p'] J-lj
( _ M )S
IOn
n-l (
= '"
~
-1
k=2
S
M
n-k k
(n _ k)!
)n+k
(-lt M n -
+ (n - 1)I(
)'
. 1 - Mo
where
J
00
uncp(U)du < 00
Mn =
Pj =
Ine - 27rij .
b
,] =0,1,2, ...
and, from (1.6) J-lj = b. Using (1.7) with a = l~\~ gives, after some simplification, the result (1.2). Other identities, similar to (1.2), may be evaluated
from (1.8) for n=2,3,4, . .. , or indeed by differentiating (1.2) with respect to the
parameter a. The following two points are certainly worthy of mention. Firstly,
replacing a with ia in (1.2) gives the result
2:::: 2
n=1 n 00
= - 2 [1 2a
7racot7ra].
(1.9)
we may obtain, by the use of(1.2) and (1.9), other identities ofthe form
Ln
00
4 _
a4
= 4a4 [2 -
(1.10)
n=l
Ln-
00
n=l
f c~~~:2 ) d:
~.
90
[ln {Sina1fa1f}] .
(1.11)
Integrating both sides of (1.11) with respect to the parameter a and interchanging
sum and integral results in
fin ( 1+
~~) = In (Si~1fa1f) ,
(1.12)
11 (1 + na22)
00
n=l
= sinh a1f.
a1f
The summation of zeros of other transcendental functions have also been considered by several other authors. For instance, Lord Rayleigh [90], obtained
00
,
!--, m4
)=1
1
12
00
" m j-8
andL..
frequency function
J=l
33
35(12)2
Where
+1
(1.13)
g{m)
=2-
m4
(3 +
2m 8
7! -
16m 12
3.11! +
16m 16
l5!" -
16 2m 20
5.19! +
163 m 24
24!
and since 9 (m) is an even function in m then -mj and imj are also zeros of
(1.13). Ifwewrite
S ( ) - ~ -a - _1
a - L mj - 2 .
j=l
1fZ
(1.14)
'\'
i...J
)=1
-12
mj
2641
34650(12)2
00
an
d '\'
i...J
J=l
-16
mj
12343
2002000(12)2'
An h
ot er operatlOna
1.4
LF(n)= !L(e-xtf(x)d:r.
o
( 1.15)
L-F_(_n)-;-k =
1
jxk-le-axL(e-Xr f(x)d;z:.
(n+a)
(k-l)!
o
The integral representation of (1.15) may be so chosen to allow for denominators with rational and irrational aigebraie functions and linear faetors, and the
numerator may be so chosen to allow for algebraie, exponential, trigonometrie, inverse trigonometrie, logarithmie, Bessel and Legendre functions. The
convolution theorem may be beneficially exploited, so that we may write, for
.>2
J_
1
ala2 a 3.. a j
u
= (j-l)!f du / dv ...
Jo
(j - 1) times
~
Aj
/ sj-1e->'sds
(j-l)!
L:00
8(a,j)
1
/
(j-l)!
j- - -
n=O
TI
(an
+ k)
x=O
(1 - x)j-l
-'------'-dx (1.16)
l-x a
k=l
=
for a E
and j
F [
J-:-; j+ 1 j
.
= 2, 3, 4, ....
1,.!,~,~, ... ,i
a '
. /1 (1 1__x)j-l
_,
xa dx -
,[
J+I FJ
(1.l7)
a fu
a , ... ,
1,~,~,~, ... , ~
Ha l l i ~
a '
x=O
a '
a , ...
ili
a
For specific values of a and j various listings of (1.16) occur in the works of
Jolley [70], Hansen [60] and Gradshteyn and Ryzhik [53]. We mayaiso obtain
some other interesting cases as folIows. From (1.16)
8(1 ')
,J = (j-l)(j-l)! = j! H
j 1+11 11]
F [ 1,1,2,3, ... ,j
j 2,3,4, ... ,j+l
2Fl
summation
~ 1; j ~ 2.
1]
5(2,j)
.1
() -1)!
~., J+1
. .F
J
[ 1,
J.
and hence
3 F2
1,1,1
$,~
r=l
~,21, ~, ...
,j
~
2'
I]
1 =
j2 j
, ... ,
In2+
1] ,
j-2
~(-1)
r (
j - 2 ) 1 - 2- r l
T
T'
5 (6,12)
Non integer values of a mayaiso be considered and hence (1.16) may be related to the polygamma functions. The following two examples are given;
5(~,8) = ~~gi!~ andfrom(1.17)wehave
9
p, [ 1,2,4,6,8,10,12,14.,16/ 1] = 67864
8
3,5,7,9,11,13,15.17
45045'
also
210
- 1263F2 [ 2.31'r'
d
3 ' 3
1
1]
10
Numerical estimates ofthe integral (1.16) may be determined for those values
of a and j which do not permit an analytical solution of the integral;
S ( 1 9)
r-..J
. ,.
00001315 = ~
9!
Po [ 1, 10,20,30,40,50,60,70,80,90 1 1]
11,21,31,41,51,61,71,81,91
.
10 9
Using this procedure Wheelon also sums the double infinite series
00
00
~ ~
~f:r (4n -
1)2m+l
7f
1
= - - -ln 2 and
8
'
(
L L nm(n()n+m
-1
= ~ In 2 +m)2
6
00
00
m=l n=l
~)
4
both of which agree with the results obtained by Bromwich [19]. A similar
summation procedure, to that given by Wheelon, has been developed by MacFarlane [79], which depends upon the properties ofthe Fourier-Mellin transformation. From Wheelon's work, we may now see a connection with (1.16) and
the polygamma functions, 'I/J (x). From (1.16), let j = 2 and a =
,k E N
in which case we may write, by partial fraction decomposition
4k2
2k-l
Sk=
n=O
r=O
= ~ n 2 (4n2 + 1)
2+((2) -7fcoth"i'
7f
A great deal of exciting work has also recently been carried out by Borwein
and his coworkers [13] on symbolically discovered identities with special and
other functions. Flajolet and Salvy [48], by the use of residue theory also obtain
identities involving special functions. Other transform techniques also provide
a rich source of possibilities for investigating sums which may be represented
11
in closed fonn; Z transfonn techniques are widely used and a general method
may be seen in the books of Jury [72] and Vieh [107].
Frappier, (see for example [49] and [50]), has also embarked on some exciting
work in the summation of series. Most recently, Frappier [51] has investigated
the representation of generalised Bernoulli polynomials and series in closed
form.
The following results are given in [51].
DEFINITION 1.1 Given that
"\"'
go '2
k=O
00
(iz) = ~
B n,a ()
x Zn
I
n.
The Bn,o (x) are rational functions of 0: and for:z: = 0, Bn,o (0) are the
o:-Bernoulli numbers. Also Bn,O) (x) = B n (.7:) are the classical Bernoulli
polynomials and B n, ( _ ~) (x) = E n (x) are the classical Euler polynomials.
Using the Cauchy product and the integral representation ofthe Bessel function
of the first kind, the following identity was proved in [51].
For 0: a complex number, and not a negative integer
L Tr
00
k=1 Jk
12
1.5
Hypergeometrie Functions
(ah
(a)o=l
= a (a + 1) . (a + k -
1)
= r~!~)
1. Poehhammer's
(1.19)
pFq [ al,a2,,a p
bl, h ... , bq
Jz]
k==O
(a1)k(ap)k zk .
(b1h (bqh k!
( 1.20)
bl, ... , bq rt 0, -1, -2, ... , sinee otherwise the denominators in the series will
eventually beeome zero. If for some j, aj = -n then all terms with k > n
will vanish, so that the series will terminate. In the non-terminating ease, the
ratio test yields the radius of eonvergenee, whieh is infinite for p < q + 1, 1
for p = q + 1 and 0 for p > q + 1. Moreover, if p = q + 1 then there will be
absolute eonvergenee for Izl = 1 if
13
f (z)
00
+L
n=2
oFo [
1 Fo [
z]
lanl ::;
=
eZ , and
- a 1 z] = (1 - z t
G0
Fj [ a
~ 1 1- :2]
(a + 1) J er (z)
and the 2Ft series is the classical Gauss series with Gegenbauer, Cebysev,
Legendre and Jacobi polynomials as terminating cases. It is weil known in
the theory of hypergeometrie functions that the confluent 1Ft function can be
obtained from the Gaussian 2Fl function by a limit process called confluence.
The hypergeometrie function
q+l
al, ... ,a q I ]
F.q [ ao,
b1, b2, .. , bq z
(1.21 )
T'
14
in particular if c
= a + b + 1 we have
Hypergeometrie sums are often met in the form of combinatorial sums with
binomial coefficients. Evidently, one hypergeometric sum may have many
representations as a sum with binomial coefficients. Saalschtz's summation,
for example, may be written as
(a+k-l)(c-a-b+n-k-l)
c-a-b-l
k
t;
(C+k-l)
c-b
n
The excellent book by Koepf[76] lists many more examples and implementation
of Binomial type series.
The book by Egorychev [41] also lists a great number ofBinomial type series.
In the next section, rather than detail the theory and practice of summation
ofbinomial series in closed form, we will consider a particular sum, with some
variations, and investigate its solution through various procedures, including
the automated approaches described by Petkovsek, Wilf and Zeilberger [85].
2.
2.1
The sum, with some variations and relations, which we shall explore in detail,
arises in the work of Jonassen and Knuth [71] in an algorithm known as tree
search and insertion. In particular the sum is
(1.22)
We shall explore (1.22) and survey severaI methods of finding a closed form
solution. We shall compare the analytical techniques of Riordan, Jonassen
and Knuth, Gessel, Rousseau, the hypergeometrie connection, the generating
functionology method of Wilf and the automated approaches of Sister Celine,
Zeilberger and the WZ pairs method.
15
2.2
Riordan
Under the heading ofInverse Relations, Riordan [92] considers the identities
(1.23)
_ 2n+1 ( 2n +
hn+l k
k=O
1) (--1)
"2
k (
2k ) _
k
- O.
( 1.24)
2~L
).
then
2n+1 (
hn+l=L
k=O
2n: 1)
(-l)k gk =O.
Hence
n (fn
fo = 1,.fI
= O.
2n-12n-32n-5"'~=(_1)n(
2n
(2n)!
22nn!2
2n - 2 2n - 4
n-l (
= II
J=O
1)
1 - 2n - 2j
-1/2)
n
(1.26)
16
L
k=O
2n )
k+1
2n-l (
(-1)
2"
2k )
k
=
k (
22n+1
(2n)
n
and
t,e::l1)(~lr en ~ 2~~1 en
Riordan attributes the identities (1.23) and (1.24) to Reed Dawson. Another
interesting identity related to (1.23) and which may be evaluated by inverse pair
relations is
(1.27)
2.3
hence
n (In
+ In-I) = Ln
k=O
k+1
(-1)
2"
k (
2k )
k'
(1.28)
(n -
1)
(/n-I
+ In-2l ~
ta (
k:
1) (~1)' ( 2: ) -
ln-I. (1.29)
Subtracting (1.29) from (1.28) we obtain the recurrence relation (1.25) and
hence identity (1.26) folIows.
17
2.4
Method of Gessel
Ihis method is given on page 3 of the Greene and Knuth [56] book and
is described as folIows. Replace k with n - k, that is change the order of
summation, in (1.22) such that
fn
(n)
L
k
k=O
n
(_1)n-k (
2
2nn -- 2kk ).
[x n] (1 - 2xt
[yn-k] (1
=( ~)
+ y)2n-2k = ( 2~ ~k
(_2)k
) (_2)k
= [yn) yk (1 + y)2n-2k
and therefore
But since
[x k] f (x) 9 (y)k =
f (g (y))
k=O
In =
and the solution follows
In
={
2-n (
0,
2.5
n )
n/2
'
for n even
( 1.30)
for n odd.
Method of Rousseau
Ihis method is also described in the book of Greene and Knuth [56] and
essentially it identifies the coefficient in a polynomial expansion. From
[:cO]
(X + ~1) 2k = (2k)
k
'
18
t. (~ )(~1 r D
(1 -
(x ~ ~)2) n
(X +
2k
and
hence
2.6
Hypergeometrie Form
Here we consider a slightly more general version of (1.22) in terms of hypergeometric notation. Let
in (a,b) = t
k=O
(~
) (_a)k (b:)
= tTk
(1.31)
k=O
Tk+1
ab (k - n)
Tk = (b - 1) b-1 (k +
Yl (k+;t)
j=l
TI
j=2
To
--b-'-) ,
1) 2 -b--1-(
(1.32)
+ b-1
bFb-1 [
1
bb 1 ,
/ (b
a~~b-ll,
(1.33)
In(a,l) =
IFO[
in (a,2) =
2 Fl [
(1.34)
!,;n /4a]
and a recurrence relation for (1.34) obtained from the Zb algorithm in Mathematica, is
(n
10 (a, 2) = 1, Ida, 2) = 1 -
1) in+l
2a.
+ (n + 1) (1 -
4a) in
= 0,
(1.35)
19
We can see from (1.35) that for two special cases of a = 1/2 and a = 1/4 the
recurrence relation (1.35) becomes manageable. From (1.34) let a = 1/2 such
that
( 1.36)
and replacing k with n - k we have
1 )=
in (2,2
T02 F I
[ -n~~n-n 11]2'
(_l)n ( 2n )
TO=:2
n'
( 1.37)
There is an identity, due to Gauss, see Graham, Knuth and Patashnik [55], which
states
2FI
al, a~3- a2
1t] = 2
01
2FI
a~:2 1- 1]
-~n~-;:
t]
= T n 2 F I [ -n
~~;;1.
I -1.
]
F [ a 1 , a2
2 I
a3
1]
= r(a3)r(a3 -al
- (2)
r(a3- a 2)r(a3- a I)
hn
= ( 4n ) ~ ((2n)!)2 4n (2n)! = T 2n ( 2n ).
2n
4n
2nn!
(4n)!
20
fn
+ 2) In+2 -
~ (2n
1)
rr
= 2-n n - 1 (2 j +
. 0
J. + 1
+ 3) In+1
= 0, hence
r (~ + n)
= T 2n ( 2n )
, ;;;;:
n'
n.y7r
J=
2 (n
10 =
!I =
1 - 3a,
12 =
1 - 3a + 15a2
(1.41 )
The recurrence (1.41) does not lend itselfto easy closed form evaluations for
any special values of a. Retuming, briefty, to the identity (1.27), we begin with
the general form
(1.42)
9n (a,
b)- F1
- b b-l
For b = 1, 9n (a, 1)
[1,t=i'~=~'
... 'b~1,-n
b-1 b-2 b-3
1
b ' b ' b , ... , b
= In (a, 1) = (1 9n(a,2)
la(b-1)b-1]
----'----:-bb-'---
a)2 . For b = 2,
2F1 [1'i n
I~]
+ 1) 911+1 + (n + 1) (a -
4) 9n
+2=
0,90 = 1.
21
k (1 - 2k)
= (n + 1 _ k) (2n -
R (n, k)
1)
+ 4) (3n + 5) gn+2 -
{
go =
+ (4a -
1, gl = 1 - ~a
and again it does not lend itselfto easy closed form evaluations for any special
values of a.
2.7
Snake on Method
This method is described on page 126 of the book by Wilf [111]. Let
in (y) =
t (~ )
yk (
k=O
= L
2: )
(1.43)
n~O
1~ xL ( 2: ) C~ x)
k~O
F (.7:, y) =
k~O (
2: )
zk
(1 - ;1;)
Vi -
= --;:;:=:::::;=::;::===;:=====;::::::
V(1 -
x) (1 - x (1 + 4y))
F (x, -1/4)
L ( 2m
)
m
m~O
(~)2m
2
22
and hence,
in (b, c) =
t (~ )
bk
(1.45)
%)
k=O
and define
(1.46)
n~O
Putting (1.45) into (1.46) and interchanging the order of summation, we have
(l+z(b-l))C
(1 - z)c+l
For einteger (1.45) will always have a closed form solution, For example, with
e = 3, we have
t, (~ )
bk
~ (b3n 3 + 3b2n 2 (3 -
b)
( -V
2 ) (_4)k
(2;)
k
+ bn (18 -
9b + 2b 2 )
/ )
k k-l
~r k+l 2 =~rr(2 1)
y7r r (k + 1)
k! j=o J +
F{z,-1/2,1) =
1
vr=z
= L (
z k~O
1
+ 6) .
2k
k )
(Z)k
4 '
23
2.8
Some Relations
S ( )=
n p,q
P p-l
(_l)P+l]
'
(1.48)
( 1.49)
1 1]
= { 0 if qn E Z+
1 if qn
=0
and
S (2)
n P,
P p-l
1,1,1, ... ,1
It is known that
and therefore
Sn (3, 2) = ( 3n
. ) Sn (2,2) ;
n
however for P 2 4, deBruijn [39] showed that (1.49) cannot be expressed as a
ratio of products of factorials, and Graham et al. [54] also showed this by an
application ofthe multidimensional saddle point method. We can deduce, from
(1.48) the identity
I]
- 1 =
qn
2 +1
B (2+ qn l-qn)
2
'
(1.50)
24
where B (x, y) is the Beta function. From (1.4 7) and (1.48) we mayaiso deduce
that
8 2n+1 (p, 1)
~ ~' (-1)'
=
8 2n (p, 1)
n : 1 )'
F:
P p-l
~ ~ (-1)' (
r~ e: r
2;
+2 ~ ( -1)' (
(-1)"
=0
'
~~
-1
~ ()
r (
2n )
22n n!f
..;rr + (-1 r
(~)
( 2n )
The sum (1.47) may, for specific cases of p and q, be written as a recurrence
relation. Another related sum is given by Strehl [102], whom in an infonnative
paper shows that, for all natural numbers n
4 F3 [
n+
25
(1.52)
t (~)2
(_1)k
1] '
[ -n -n I 2 + ,\ +:\
2 Fl
(l.53)
t (~) ('\1;
k=O
+ 1)k"t,\{'2
1,2
k=O
(~)2
j=O
(1.54)
t; (n)
k
n
. k
(-1)
2 Fl
-n, -n I] = B(~,~)
+
1
-1
2n
Sn (2, 1)
= {;
(n) (1 +
k
'\1,2
(n
and by iteration
S,
2n
2.9
Let
(2 1) = (_2)n
,
rr
n -l
j=O
2j + 1,
'+ 1
J
In = LF(n,k)
k=O
(1.55)
26
where
F (n, k)
=(
~ ) (~1)k (
2; ).
(1.56)
Since the ratio oftwo subsequent terms of(1.22) is a rational function in both n
and k then (1.56) is a proper hypergeometric function. Following Sister Celine
[85] we require non-trivial solutions of the recurrence
2
L L ai,j (n) F (n - j, k - i) = O.
(1.57)
i==O j==O
+ n (2n -
1) F (n - 1, k)
- an (n - 1) F (n, k - 1) - n 2 F (n, k)
= O.
2.10
~)
Binomial[2k, k], k, n, 2]
(1
+ n) Sum[n]- (n + 2) Sum[n + 2]
= 0
2.11
WZ Pairs Method
This method certifies a given identity as weIl as having some spin-offs. Given
the identity (1.22) we may write
n
'L F (n,k)=l
k2:0
(1.58)
27
where
F(n,k)=
~n ) (_~)k (
(2: )
2: ) 4 (_~)k (2k)!n!24n
n _
(271 - k)!k!3
(1.59)
k2
R (71 k) = ...,--------,--.,-----,
(2n-k+1)(k-2-2n)
(1.60)
Now, we define
(1.61)
l)"n+bk
*
(-1t+ 1 2n+1n!3 (2k - 2 - n)!
G (n, k) =
4k (2n + I)! (k _ 1)!2
As previously, we obtain f~
= 2.:.:
( 1.63)
(1.64)
k2:0
f~=
k2:[nt
F*(n,k),
1
(1.65)
28
where [xl represents the integer part of x. Now, we need to sum over k, the
recurrence
(1.66)
since the right hand side of (1.66) does not disappear, we sum for k 2' 1 + [~],
this however gives us an extra term, and distinguishing for n odd and n even,
we obtain
F* (n + 2, k) - F* (n, k)
= G* (n + 1, k + 1) - G* (n + 1, k) + G* (n, k + 1) - G* (n, k).
For n even, let n = 2m, and surnming for k
2' 2 + m, we obtain
G* (2m, m + 2) ,
f * (2
) = f* (2)
m
(1.67)
1* (2 + 2
)=
1* (2) + ~
~
)=1
(1.68)
1* (2) = _~ ~
3~
k~2
(2k - 3)!
4k k!2 .
(1.69)
1* (2) =
} - In V2.
(1. 70)
(We mayaiso obtain (1.70) by starting with identity 2.5.16 in the book by Wilf
[111]). Now from (1.70), (1.68), and (1.65) we obtain
~
~
k=m+l
(2k - 1 - 2m)!
4 k + 1 k!2
-1 V2 _ ~
- n 2 3
J=1
1)! (2j)!2
(1.71)
j!(4j+3)!(j+1)! .
29
From (1.67) and (1.70) we also obtain 1* (0) = -ln y'2 and from (1. 71) putting
k* = k - m and renaming k* we have the new result
00
k=l
(2k - I)!
22k (m + k)!2
(4m - I)!
{ln4 _ 4 r~ (3j + 2) (2j + I)! (2 jW }
(2m - I)! (2m)!2
L
j! (4j + 3)! (j + I)!
.
)=0
Chapter 2
NON-HYPERGEOMETRIC SUMMATION
1.
Introduction
Differential-difference equations occur in a wide variety of applications including: ship stabilization and automatie steering [82], the theory of electrical
networks containing lossless transmission lines [17], the theory of biological
systems [16], and in the study of distribution of primes [108]. The equation
a)
+ f (t) + "( f
(t - a) + J f (t
+ a)
= 0
31
32
investigate, briefly, equations with forcing terms, and equations with multiple
delays, mixed and neutral equations. The fundamental series obtained in this
chapter has also been investigated, using different methods than the author, by
Euler, Jensen and Ramanujan. We shall describe their techniques and give in
detail, a description of Cohen's modification and extension, and a solution to
ConOllY'S problem.
2.
Method
!,(t)+bl(t)+cl(t-a)=O,
o ~ t < a.
1'(t)+bl(t)=O, /(0)=1,
(2.1)
Taking the Laplace transform of (2.1) and using the initial condition, results in
(F (p))
= 1 (t)
-b(t-an)
(t
n!
L..t
n=O
- an
)71
H(t - an)
(2.3)
H(x) = {
I,
for x 2: 0
0,
for x
<
o.
'1'+ioo
1 (t)
~
21TZ
ept F (p) dp
for an appropriate choice of'Y such that all the zeros ofthe characteristic function
9 (p) = p + b + ce -ap
(2.4)
are contained to the left of the line in the Bromwich contour. Now, using the
Residue Theorem
1 (t)
L residues of (e
Pt
F (p) )
33
Non-Hypergeometric Summation
where the sum is over all the characteristic zeros Pr of 9 (p) and Qr is the
residue of F (p) at P = Pr. The poles of the expression (2.2) depend on the
zeros of the characteristic function (2.4), namely, the roots of 9 (p) = O. The
dominant zero Po of 9 (p) has the greatest real part and therefore asymptotically
f (t) rv QoePot , and so from (2.3),
f (t)
(t - an t
L (-1 )"cne-b(t-an)
,
n.
00
H(t - an)
rv
Qoe Pot .
(2.5)
n=O
lacel+abl < 1.
(2.7)
F(p) _ ~
-
(1 +
b+ce-ap)-l _ ~~
.
-~~
n=O 1'=0
(n)
r
(_l)nbn-TcTe-arp
n+l'
(2.7).
The poles ofthe expression (2.2) are all simplefor the inequality
34
P + b + ce-ap
=0
(2.9)
1
h
-a~ _
~' w ere ~ + b + ce
- 0,
1 + a b + a"
and so the expressions (2.6) and (2.8) become
~
(-1 )ncne-b(t-an) (t - an )n
n!
n=O
e~t
1 + ab + a~
(2.10)
whenever the single and double series converge in a mutual region. Using the
transformation
oon
0000
LL!(n,r) = LL!(n+r,r)
n=r=O
(2.11)
n=Or=
LEMMA 2.2
1. The single and double sum in (2.10) are solutions to (2.1) in their region 01
convergence.
2. The closedlorm expression in (2./0) is a solution to (2.1)lor t
a.
3. The single and double sums in (2.10) are equal in their mutual region
convergence, which is no larger than that given by (2.7).
01
35
Non-Hypergeometric Summation
00
LL
(n)
n=01'=O
(_l)nl/!~~"c1'
(t _
n!
art
00
(_l)"cne-b(t-an)
(t -
n!
n=O
ant
'
so that expanding the left hand side and summing each colurnn from the left
hand side results in
C2 (t-2a)2[1
O!
c3 (t - 3a) 3 [ 1
3!
O!
2!
b(t-2a)1
1!
b (t - 3a) 1
1!
b2 (t-2a)2
2!
b2 (t - 3a) 2
2!
b3 (t-2a)3
3!
b3 (t - 3a) 3
3!
+ ...
+ ...
+ ...
c (t - a) e-b(t-a)
I!
c2 (t - 2a)2 e- b(I-2a)
2!
c3 (t - 3a)3 e- b(t-3a)
t = -aT, so that
3!
+ ...
36
t , and hence
1
I>-abt=-b'
n=O
1+a
00
(2.12)
Brmann's Theorem [110] will now be used to prove the explicit form ofrelationship (2.6).
0
3.
THEOREM
D, and let
e (z) =
/f f
z-
cjJ (z)
,e () = cjJ' ()"
f(z)=f()+
r=l
where
Rn+l = 27ri
Jr dv
[!'(t){B(t)rJt=+ Rn+1
[</J(v)]n f'(t)</J'(V)
c cjJ(t) cjJ(t) _ cjJ(v)dt.
1
1 +ab'
Let t
= -aT,
ab
(2.13)
eXZ
z
f (z) = - , e(z) = -(-) = eZ , cjJ (z) = ze-z,j ()=o = 1,
1-z
cjJ z
37
Non-Hypergeometric Summation
00.
!'(t)=ext(~+
1 2) =e xt (~(.1:+1+j)tj),
1- t
(1 - t)
~
)=0
and so
00
I: (x + 1 + j) t j . The
j=O
coefficients in this expression are the same as those in a Taylor series expansion
\[I(j) (0) = (:1: + 1 +.j) j!. Now let
Br(t)
dr - 1
dtr- 1 [!'(t) {B(t)Y]
= et(r+x)
[(7" +
1 (
7" ; 1 )
\[1(0)
(t)
~1)
\[I'
(t)
\[I 11
X (
~ =~
2 (
7"
\[IU-2)
~ =~
\[I(J-l)
(t)] .
Hence
Br()
((7"+x
+ (7" + xr- 2
7"
r-
(7";1 )(X+l)
~1)
+(7"+x)l
(x
+ 2) +
(~=~) (x+7"-I)(7"-2)!
+ (
If we now put y = x
~ =~
+ r we obtain
38
= yr _
(r _1)yr-l
+ (r
= yr = (x + rf.
Hence it follows that
~
1- z
= 1 + ~ {ze-Zr (x
~
r!
+ rf .
r=l
A modification of this sum also appears as a problem in the work of Polya and
Szeg [87].
By the ratio test the infinite sum (2.14) converges in the region
< 1,(or label+abl < 1 for (2.13, and so considering p as a corn-
Ipe1-pl
Le
00
n=O
-(T+n) (
n.
00
n=l
*,
on utilizing Stirling's approximation n! '" (;) n v27fn as n ........ 00. The divergence of the above series can also be ascertained from the closed form
representation ofthe right hand side in (2.14).
The characteristic function (2.4) may be shown to have a dominant double
zero at p = 0 for b + c = 0 and 1 + ab > O. From the general theory of linear
functional differential equations [58] it follows that there exists constants a and
such that
lim (f (t) - at) =
t-+oo
From residue theory, the constants a and can be shown to be ~ and ~ respectively, in which case
. (f (t) - -2t)
2
= -3'
a
hm
t-+oo
39
Non-Hypergeometric Summation
(_l)nbn-TcTt n
n!
This result can be ascertained directly from the differential-difference equation
(2.1).
The sum (2.13) is known as an Abel type series. Similar Abel type sums
mayaiso be expressed in closed form by utlising the W Z certification method,
as highlighted at the end of chapter one. In particular, the interested reader may
refer to the recent works ofMajewicz [80] and Ekhad and Majewicz [42].
4.
1 + ab
Differentiating both sides of (2.15) with respect to
+ a(
(2.15)
t, j times we have
(b + 0] etE
1 + ab + a~
(2.16)
On the left hand side (2.16) put n - j = n * and rename n * = n, also put
t + aj = x, then the left hand side becomes
(-c)
00
n.
n=O
.
exE
= (- c)J 1 + ab + a(
(2.17)
-c = (b + 0 ea~
and therefore (2.17) is equal to
(b + 0) e'c~
1 + ab + a~
(2.18)
40
L
n=O
00
For j
(-1)n cn eabn(t-ant+ j _ /
/
(n+j)!
........
~
et(b+~)
1+ab+a~dt.
j-times
n=l
(2.19)
= n + 1 and
e~(x-a)
+ Ke-b(x-a)
(b +~) (1 + ab + aO
1
'
to evaluate the constant K, adjust the counter on the left hand side and use the
result (2.10); on the right substitute (2.18) for - c, such that K = _ e-~~;O
Thus from (2.20)
00
n=O
= _1_
b+~
et{
_
1 + ab+a~
e-b(t+a)-a~].
(2.21)
5.
(2.22)
Forcing Terms
f'(t)+bf(t)-bf(t-a)=
tm-1e- bt
r(rn) ,f(O)=O,
(2.23)
Non-Hypergeometric Summation
41
1
F(p) = (p+b)m(p+b-be- ap )
(2.24)
00
bne-b(t-an) (t _ an)n+m
(n+m)!
_
- bm (1
m-l
"P
+ ab) + ~
m,v ( -
v=O
b)
tm-v-1 -bt
(m -
lJ -
I)!
(2.25)
where
IJ!Pmv(-b)=lim
,
p---.-b
{~[
b 1
]},1J=0,1,2, ... ,m-1.
dpv p + - be-ap
For m = 1, (2.25) gives the result (2.22), and for m = 3 we have the result
Now let us consider the case where m may be a rational number. As an example
ifm = a/ then (2.24) has a simple dominant pole at E. = 0 and a branch point
at p = -b. Also, from (2.24), f (t) = ,[-1 (F (p)) where
and the contour C in (2.26) is the usual part circle with radius R, small radius, c,
at the branch point p = - band naming, anticlockwise, the points AB D E FG A.
Now
f J+ J+ J+ J+ J J
+
AB
BD
DE
EF
FG
= 27riRes (p = 0),
(2.27)
GA
BD
=0=
J.
EF
(2.28)
42
J= J=
DE
limit
c---+O,R---+oo
FG
J
R-b
1
1
= bo/ (1 + ab) - :;
AB
e-(x+b)tdx
x o / (x + bea(x+b)) .
(2.29)
00
x=o
e-(x+b)tdx
xo/ (x + bea(x+b)) ,
(2.30)
f:
n=O
bne-b(t-an)
(t - ant+~
r (n + ~ + 1)
1
1
= bo / (1 + ab) - :;
J
00
x=o
e-(x+b)tdx
x o / (x
+ bea(x+b)) ? (2.31 )
however, since the integral in (2.30) is improper and divergent then (2.31) is
not an identity. A similar improper divergent integral (2.30) may be obtained
for any real number m.
6.
f'(t)+bf(t-a)-bf(t-)=O, f(O)=l,a,>O.
(2.32)
(p) =
f: (~~; t
n=O
(-lt (
) e-p(on+r-or)
r=O
1
- 1 - b (a - )" (2.33)
43
Non-Hypergeometric Summation
f' (t)
+ bf' (t
- a)
+ cf (t -
a) = 0, f (0) = 1
we obtain
F( ) _
1
p - p + pbe- ap
+ ce- ap '
(c + bO et (
c-a~(c-b~)'
~~ ( n )
L. L.
n=Or=O
= 0 we have
(ct)"
b
(_b)n = e- ct /(1+b).
r!
1+b
A number of examples will next be investigated in which the methods of the previous sections are applicable and in which the identity (2.13) and its variations
can be extracted.
Some ofthese results have been published in [24].
7.
Bruwier Series
Bellman and Cooke [7] refer to
44
as the Bruwier series, see [20] and [21], which is a solution to the advanced
equation
(2.35)
!,(t)-vj(t+w)=O, j(O)=1.
Comparing (2.35) with (2.1) it can be seen that b = 0,
from the series at (2.13)
8.
= -v, a = -wand
Teletraffic Example
Erlang [43], see also Brockmeyer and Halstrom [18], considers the delay in
answering of telephone calls. The problem is to determine the function j (t) ,
representing the probability ofthe waiting time not exceeding time t. Hence for
an M / M /1 regimen, Erlang shows
J
00
j(t)=
j(t+y-a)e-Ydy.
y=o
The probability that, at the moment a call arrives, the time having elapsed
since the preceding call confined between y and dy, is e-Ydy. The probability
that the waiting time of the preceding call has been less than t + Y - a is
j (t + Y - a), where a is the connection time of a call. Differentiating the
integral equation with respect to t and partially integrating the result gives the
differential-difference equation (2.1) with b = -1 and C = 1, from which (2.10)
folIows. It may be shown that the characteristic function (2.9), with b = -1
and c = 1, has the following real root distribution:
One root at p
= 0 for a ~ 0,
= 0 for a =
1 and,
> l.
The following results apply for all real values of t, in the region of convergence lae1-al < 1.
for a
>1
for a
<1
45
Non-Hypergeometric Summation
which on putting t = -aT, we may write
for a
>
fr a
<1
t->oo
= ~.
3
(2.36)
This fact has also been noted, in a different context, by Feller [47]. Bloom
[12] proposes the problem of evaluating lim (J (t) - 2t) given that, for t a
t-->oo
THEOREM
f (t)
2.2
J~~
Proof Let an =
ing function
I:n
k!
k=O
e-(k-n)(k_n(
k!
k=O
00
00
'
00
00
n=1
(2.37)
.
-(k-n)
(k
k! -n
e-(k-n) (k _ n)k
k!
n=1 k=1
00
00
n=1
r=O
k=l
00
= """
(ez)n + """
(ezt (e- Zrl ~
~
n=O
)k
zn
n=lk=O
=1+ L(ez)n+ L L
00
F(z)=l+LanZrl=l+LLe
n=1
-2n
71=0
)k
zn,letr=n-k,
1)
= """
(ze1-Zr
~
n=O
r',
= 1- zel'
Z
46
F (t)
1
1 - (1 - t) et
t~(l-~t+ ... )
2
(1-z)2
4
3(1- z)
Hence
1+~anZn_; 2n+
00
(F-G)(z)
=
13 + ""
W
00
n=l
00
2) zn
2)
a - 2n - - zn
n
3'
00
n=l
the number of random numbers we need to sum until we exceed some target t,
then E (N (t)) =
(2.37).
9.
L:
k=O
k! -
In the slowing down ofneutrons Teichmann [104] introduces Laplace transform techniques to analyze the renewal equation. This example involves the
47
Non-Hypergeometric Summation
Placzek function
1 - D-(Hp)uo
F( ) '
p - (1 + p) (1 - 0') - 1 + e - (Hp )uo
(2.38)
before inversion, where 0' is a constant depending on the mass of the moderating
nuclei and Uo = - In 0' is the maximum lethargy change in a single collision.
Keane [73] obtains
t = ~ (-lt ent/(l-a)
j()
f;:o
(l-O')n!
(l-O')t1
x e-nuo/(l-n) H (t -
nuo)
~ (-lt
71=0
n!
1 - 0'
0:
e-nuo/(l-n)
(1 - 0') e-nt/(l-a)
(2.39)
h(y) = ~ (-lt(y-ant
~
n!
n=O
e YT /
- - , where Tl
1 + OT]
+ e- ar/ = o.
(2.40)
L
00
71=0
+ 1) uoe-uo/(l-a)}n e-~
1= h (t)
0'
~
- h (t - uo) e- 1 - a
(2.41)
Now it is required to show that (2.41) is identically equal to the right hand side
of (2.39). Let
y=
te-uo/(l-a)
1-0'
uoe-uo/(l-a)
1-a
andB
Uo
48
e 1-<>
h(t) = - - .
1 +a'T}
e-a(t-uo)/(l-a) e-uo/(l-a)
1 + 1 ~a In a
1 + 1~a In a
(1 - a) e-at/(l-a)
1+
a
1-a
na
= Be- B,
' as reqmred.
l-l"-~"o I < 1.
I~e
1-a
From (2.38) a double pole occurs at p = 0 when 1 - a
(I (t) _ 1-a
2at ) =
lim
t--+oo
10.
2a (2a
+ a In a = 0, therefore
+ 1).
3(1-a)
A Renewal Example
J
t
~
A (t) = -.>.A (t) + (1 dt
A o) 9 (t)
+ .>.
= A o.
1: (A (t)) = A o + (1 - A o) 9 (p) .
p +.>. -.>.g (p)
Considering the case of constant repair time, that is Mean Time To Repair,
M.T.T.R., is a, then 9 (t) = t5 (t - a) , where t5 (t) is the Impulse function,
resulting in
1: (A(t))
AG + (1 - AG) e- ap
p +.>.
00
= """
(2.42)
- .>.e- ap
.>.n
~ (p+ .>.t+1
{A e- apn + (1 - A)e-ap(n+l)}
0
49
Non-Hypergeometric Summation
and by inversion
An
n! [Aoe-A(t-an) (t - an)n H (t - an)
00
A (t) = L
n=O
+ (1 -
A o) e-A(t-a(n+l)) (t - a (n
+ 1)t H (t -
a (n
+ 1))]
where H (x) is the Heaviside function. From (2.42) the residue at the dominant
zero p = 0, of the characteristic equation p + A - Ae -ap = for a > and
1 + aA :/= 0, is l;aA' hence, by utilizing the results of the previous sections,
the result becomes
+ (1 -
A o) e- A(t-a(n+1)) (t - a (n
+ 1) )n]
= _l_
I +aA
n=O
00
n.
n=O
n.
1
1 + aA
00
(-1 t (aAe aA
n=O
n.
= -ar
r (r+n) n
-aAT
n ( l)r (
= _e_
_ =e-aATLL---00
1 + aA
n=Or=O
n!
n
r
(aAt(r+rt
whenever the double sum converges. From (2.42), a double pole occurs at
= when 1 + aA = 0, resulting in
(A()
I1m
t
t---"oo
+ -2t)
a
2 (3A o - 2) .
3
50
11.
=~
(2.43)
is derived by Tijms [105] and FeIler [47] and has applications in collective risk
theory, storage problems and scheduling of patients. Here, 0: is the Poisson
parameter and Cl a positive rate. Taking the Laplace transform of (2.43), it
follows that
*
R (0)
1
R (p) = .E (R (t)) = 1 _ ~ (1 _ F* (p))
R (t)
= (1 - k/-L)
= -k results in
kt
:L -=r-ek(t-an)
(t - an)n H (t - an) .
n.
00
n=O
~
~
n=O
(_k)n ek(t-an) (t _
n!
ant = _1_, Vt E ~
1- ak
(R()
11m
t
t-+oo
12.
kJ-L)) _
+ 2t(1a
2(1- kJ-L)
3
.
A Grazing System
w' (t)
+ (rl
- g) w (t)
+ ge-rT1w (t -
T) = 0
(2.45)
describing a linear continuous grazing system. w' (t) represents the change of
pasture mass over time t, rl is a constant grazing pressure, 9 and T are positive
51
Non-Hypergeometric Summation
h (p)
= p - 9 + Tl + ge-r(p+rJ)
(2.47)
and in fact for t > 7, the right hand side of (2.46) is a solution of (2.45).
Woodward and Wake deseribe a neutral stability eondition for
(2.48)
and applying (2.48) to (2.47) gives us the dominant zero,
(2.46) reduees to
n=O
n!
(t _ 7t =
= 0, in which case
1
1+7(1"1-g)'
representing a eonstant steady state solution for 1"1 > g. The eharaeteristie
funetion (2.47) has a double pole for p = -7"1 and g7 = 1, in which ease, from
residue theory
lim
t->oo
(w (t) - -2t) = 2.
7
Other examples oeeur in stoehastic processes, see Hall [59] and in the demographie problem of a counter model with fixed dead time, see Biswas [11].
13.
Equation (2.4) is the transcendental funetion assoeiated with the differentialdifferenee equation (2.1). The zeros of tbis funetion are weil doeumented and
sinee many research papers bave been interested in the stability of the solution
oftbe differential-differenee equation, eonditions are given for the existence of
eomplex eonjugate roots with negative real part. Firstly, t; satisfies (2.4) if and
only ifln (~ + b) = In ( -ce-a~) ; ~ eomplex, henee
a~
+ In (~ + b) = (2n + 1) 'Tri + In c,
52
1. ab> 1,
+ abtan TJ
= 0;
so
= ap,
er
= ab, = ac,
G (z) = z
Putting Y (z)
> 0,
9 (ap)
= G (z) and
+ er + e- z .
(2.49)
(1
+ er), we
Prooj Let z
(2.50)
y - e-xsiny = 0
The zeros ofG (z) depend continuously on , and for > 0 all zeros will be in
the half plane Re (z) ::; . IfG (z) = G' (z) = 0 there will be a double zero at
z + 1 + er = 0 and therefore zeros cannot bifurcate or merge, as varies, in the
half plane x > -1. Utilizing similar arguments to that of Cooke and Grossman
[36] it can be seen that if z = z () is an isolated simple zero with Re (z) 2: 0,
then it moves to the right ofthe half plane for increasing , since
dz = _ dG / d =
z + er
and
d
dG / dz
(z + 1 + er)
Re
(x + er + 1) 2 + y2
53
Non-Hypergeometric Summation
increasing from 0: to
o < l (0:) < 2 (0:) < ... ,
For
so that lim
k->oo
00
(4k
7r
k = 0, 1,2,3, ....
It appears, from (2.50), that a zero must remain in the region where sin Y >
and cos Y < 0 and in the specific case where 0: = then = Yk = (4k + 1) ~;
k = 0, 1,2,3, . .. .
0
14.
Numerical Examples
The zeros of the characteristic function (2.4) can be located using Mathematica. Let p = x + 'iy then
Re(x,y) =O=x+b+ce-axcosay
Im(x,y) = 0
=y-
(2.51 )
ce-axsinay
and iffor any x, y is a solution then so is -y. Therefore non-real zeros occur
in complex conjugate pairs. Putting t = -aT in (2.10), we may write
1 + ab +!l~
e-abT~ (~~t ~ (~
(2.52)
) Gf
(T+r)1!
where~isthedominantzeroof(2.51).
~
1.050472.
In the next section we look at the related works of Euler, Jensen and Ramanujan. We shall describe their techniques and give in detail, a description of
Cohen's modification and extension, and a solution to Conolly's problem.
15.
Euler's Work
Eu1er's work is related to our equation (2.10); his work was published in
Latin in 1779 and we give an English translation of the main points that are
pertinent to our work. Euler [44], see also [45], considers aseries given by
54
Lambert and investigates several of its notable properties. Euler rewrites the
Lambert series in the form
r+l
00
S=l+nv+nL(v
r
r=l
r-l
),II(n+(j+1)a+(r- j ))
+ 1 . j=o
(2.53)
xa
x = (a - ) vx a +
(2.54)
then S = x n .
In terms ofthe Pochhammer symbol, as defined by (1.19) and which was not
available to Euler, we may rewrite (2.53) as
vr+1
S=1+nv+n~(r+1)!(t-t-)
00
r(n+a+r)
a-
r
r+l
00
r-l
V+L(v 1),II((j+1)a+(r- j ))
r= I
+ ..J=0
v r + 1 (a -
00
v+
(a+ar)
a-
r (a+r)
a-
r+l
00
= v+ L
v
(a-r
_ (r + I)!
r-l
(2.55)
~ (r + 1) rB ( ~~; , r )'
v= ---:--a-
(2.56)
~ (av)j
In (1 - av) = - L..J -.,- .
. 1
J=
J.
r=l
II
r-l
r+ 1 . j=o (n+a(r+1))
(2.57)
55
Non-Hypergeometric Summation
-+
1 _ ~ v (O'v (1' + 1)
nx - L
(1'+1.
)'
,.=0
moreover, substituting In x
xC>
Now substitute av =
y =
'U
= vxC> we obtain
=~
;':0
(Ct/I (1' + 1W
(1' + I)!
(2.58)
L (u(r+1))"
( + 1),. ,where uy = In y.
00
r=[)
l'
(2.59)
(2.60)
in its region of convergence lul < e- l , Euler gives some numerical values
of(2.60) and in particular for Iny = 1/2, from (2.60) he obtains
u r + 1 (1' + Ir
1 1
where u = -e- 2 .
L
r!
2
1 = '"
00
,'=0
x8
= ~ (n + st
I-Inx
L
n!
n=O
(Inx)n
x
'
16.
(2.61)
= -t
and
Jensen's Work
Jensen's [69] work is related to our equation (2.10); his work was published
in French in 1902 and we give an English translation of the main points that are
pertinent to our work. Jensen, who appears to have been unaware of Euler's
work, obtains the equivalent ofidentity (2.13) by an application ofthe Lagrange
inversion formula. Lagrange's inversion formula, circa 1770, arose as a tool in
the solution of implicit equations or the reversion of series.
56
THEOREM
2.4 :
1> (z) =
1> (a) +
L:,
00
n=l
dn -
where z = a + w f (z) .
An alternative form ofTheorem 2.4, given by Jensen, who says he leamed it
from Hermite, circa 1881, but apparently known to Jacobi, at least as early as
1826 is;
THEOREM
2.5 :
n
_oowndn
-;J dzn [g (z) (f(z)) lz=a
g(z)
+ wf
n=O
(z) .
If we take the partial derivative with respect to w in Theorem 2.5 and put
Jensen puts 1> (z) = eCiz' f (z) = ez , a = 0,
1>( a) = and obtains, from Theorems 2.4 and 2.5,
9
CiZ _ ~ a (a + j)j-l uj d
e - ~
.,
an
j=o
J.
~ = ~ (a+j)ju j
1- z
j!
J=O
(2.62)
(2.63)
e(a+b)z =
(a + b) (a
(2.64)
J.
j=O
= ~ a (a
+ j)j-l u j
j=o
+ ~, + j)j-l uj
.,
J.
f (f (~ )
~ b (b + j)j-l u j
.,
j=o
J.
J=O
v=o
(a
+ b) (a + b + nt- 1 =
tu (~ )
a (a
+ IIr- 1 b (b + (n -
11) t- v - 1
Non-Hypergeometric Summation
57
(x
~( ~
+ o:t =
0:
(a - vt- 1 (x
+ vt-
(2.66)
and in a note dedicated to the memory of Abel, Lie noted that (2.66) is in fact
a special case of a formula noted by Cauchy:
(x
17.
+ 0: + n)" -
(x
(
+ o:t = n-l
~ ~
0: (0:
+n -
v)n-v-l (x
+ vt .
Ramanujan's Question
lr-
1 e-x(r+l)
00 xl' (r +
<p (:1:) = ~ - - ' - - - ' - - - -
r!
1'=0
show that <p (x) = 1 for 0 < x < 1 and that <p (x)
lim (<P (1
0-->0
=I 1 for x >
(2.67)
1 and also find
We can see that (2.67) is identical to (2.62) when 0: = 1 and = 1. B.C. Berndt
[9] gives an excellent short historical account of (2.62) and (2.63).
18.
I
L: -wn-(1' -+n!-ant- - - = e- z and
oo
(2.68)
n=O
~ w n (1
+ an)n
n!
n=O
1 + az
(2.69)
58
t,
(_l)k (
) (1 + ak) x ak and
~)
(l+aktx ak
J
1
x=O
Cohen obtains
which corrects a minor mi sprint error in his equation (1.9). Cohen then considers
~zn~(_l)k( n) (l+akt
k
k+
n=O k=O
L..J n! L..J
~zn{(l-at}
( + 1)
n=O
n
L..J
[ni 8]
00
00
= e- z
Zn { (1 - a)n }
n=O
( + l)n
= e- z IFI
~ 1 z (1 - a)] ,
Non-Hypergeometric Summation
59
obtain other results, especially related to Laguerre, Herrnite and other special
functions of mathematical physics. It appears that other results mayaiso be
obtained by gainfully employing the ideas ofCohen. The following is one such
result. Let
(xD) = xD {xO: (1 - xar} =
(_l)k (
k=O
(xDr-m{xO:(1-xar}],c=l = t(-l)k
) (0'
(~) (O'+akr- m
k=O
and similarly
ff
m=On=O
Xn~y~
t
m.n.
(-ll (
k=O
X
t,
) (0' + ak)n-rn
On the right hand of (2.70) we can use the transformation suggested by Cohen
[32], namely
CXJ
l: l:
f(n,k) =
n=Ok=O
oom
l: l:
m=O 1'=0
f (m, p) =
00
CXJ
l: l:
k=On=O
f(n+k,k)
0000
l: l:
1'=0 m=O
(2.71 )
f (n
+ p, p)
00 2:00 (_l)k+ p yk x p
2:
k=Op=O
k!p!
(0'
+ ak) k-p ey(,,+a.k)+x(J!..I!..e.)
___
+bp
,,+ak
+ bp
.
(2.72)
t, (~~!~nyn
~)
(0' + ak)n
+ %;1 (-bO'x)m
(2.73)
60
1 _ abxy
(2.74)
From (2.72) and (2.74) we obtain the result, whenever the double series converges,
I: I:oo
OO
x (a_+_ak)
(_I)k+ p yk p
k=Op=O
kIpf
+ bp
k-p e y(<>ak)+x(
+be)
be
aak
1-~
- (l+i)(I+:)'
19.
Conolly's Problem
Sm (>') =
I: (>.n)~-m
e->.n.
n.
(2.75)
n2:1
Show that
So (>')
S3 (>')
= 1,
The infinite sum (2.75), for m = 0, is a very specialized case ofthe generalized sum (2.13). For t = 0, b = -1, and a = >. we have, from (2.13)
So (>') =
~
~
n=1
(2.76)
with convergence region IAe l ->'1 < 1, which is different than that given by
Conolly and indeed, we have previously shown that (2.76) diverges on the
boundary >. = 1, and converges on the boundary >. = -.278464 ....
From (2.75), writing the exponential in series form, we have
Sm(>')
= LL
00
00
n=1 r=O
(-Ir (>.nt+ r- m
11
n.r.
(2.77)
Expanding the double infinite sum in (2.77) term by term, and then summing
diagonally from the top left hand corner, thus collecting coefficients of An - m ,
61
Non-Hypergeometric Summation
we have
(2.78)
For the special case ofm
= 0, from (2.78)
(2.79)
and since it is weil known that the inner sum is equal to n!, we have from (2.79)
00
,\
So (A) = .w
"" An = 1-'\
n=l
f
n=l
An~m~
(-Ir (
n.
1'=0
n ) (n - rt- m +
r
~(-lr( ~
)(n-rt -
=Oforn:;,m+l
A':~rn ~ (-Ir ( ~ )
n=l
(2.81)
1'=0
S4 (A) =
S5 (,\) =
1
,\4 -
1
,\3 -
7
8,\2
11
+ 36'\
- 4"
15
85
25
1
16,\3 + 216,\2 - 288'\ + 5!
62
and
1
31
S6 (I.) = 1.5 - 32,V
575
415
+ 12961.3 - 3456A2
137
1
+ 7200A - 6!
d
dA (Sm (A))
1-A
+ ):Sm (A) =
~Sm-l (1.).
(2.82)
Am Sm (I.)
= Am
C~2
(2.83)
given that for A = 0 the constant of integration is zero and for m = 0, we have
the value So (A) = >"1 (1 - >"). Putting m = 1,2 and 3 into (2.83) we obtain
Conolly's results. For m = 0, and using (2.82) we have
dA (So (1.))
and since d~ (So (A))
S_ (A)
1- I.
= ~S-l (A),
00
= '"'
L
n=l
(A )n+l
n
e-.\n
n!
>..2
(1 _ >..)3'
(2.84)
Chapter 3
BRMANN'S THEOREM
Abstract
1.
A detailed proof of Brmann's Theorem is given in this chapter. For the principle
series (2.14), obtained in Chapter 2, it will be shown that the remainder term of
the series vanishes for large n, and discuss in detail its radius of convergence.
Introduction
. . jt (~) d~
z
. .
a
JI
z
= _1
27ri.
a
rp (t) -- rp (~)
and it may seem that rp' (0 in the numerator ought to be cjJ' (t) arising out of
applying Cauchy's integral to the f' (~) function in the left hand side. This point
will be cIarified in the author's proof. Whittaker and Watson, list Bnnann's
Theorem and Lagrange's Theorem as two separate results, however according
to Henrici [64], except for matters of notation they are identical. A formal
proof of the Lagrange-Bnnann Theorem, that is different than the author's, is
given by Henrici [65] and apart of his proof is based on elementary notions
in the theory of matrices. We also show that the remainder tenn of Bnnann 's
Theorem when applied to the sum (2.14), obtained in Chapter 2, goes to zero.
2.
64
clarify. In our proof we shall suppose that cjJ (a) = 0, which involves no loss of
generality, but does shorten the forrnulae. If cjJ (a) =F 0 we simply replace cjJ (z)
by cjJ (z) - cjJ (a) throughout the theorem as set out below.
THEOREM 3.1 Let cjJ (z) be a simple function in a domain D, zero at a point
a of D, and let
z-a
1
B(z) = cjJ(z) ,B(a) = cjJ'(a)'
If f
(z) =
(a)
+L
n
{cjJ (z)Y dr - 1
rl
dt r - 1
(3.1)
r=l
1
where Rn+! = 27ri
r r [cjJ(o]n
f'(t)cjJ'(O
cjJ (t)
cjJ (t) _ cjJ (~) dt.
Jr d~ Jc
(3.2)
If cjJ (z)
# 0 in D.
PROPERTY 3 The inverse function cjJ -1 (w) exists and is simple in D w , where
D w is the map of D in the w-plane by w = cjJ (z).
PROPERTY 4 A domain D means an open connected set ofpoints in the plane;
that is , every point of D has a neighbourhood, ~, in D and every two points of
D can be joined by a continuous curve in D.
Proof ofBrmann 's Theorem. The proof will be given in five parts.
(i). Let Cw be the map of C by w = cjJ (z), and D w the map of D .
Since cjJ is simple, Cw is a closed contour and D w is a domain containing
it. Also since C encircles every ~ off, Cw encircles cjJ (~) for all such~. Now
9 (w) = f' (cjJ-l (w)) isanalyticin D w , since cjJ-l is analytic in D w by property
3, and l' is analytic in D the map of D w by z = cjJ-l (w) . So Cauchy's integral
gives, for every ~ of r
g(cjJ(~
= _1
27ri
j9(S)dS
c",
s-cjJ(~)
= _1 j9(cjJ(t)cjJl(t)dt
27ri
cjJ(t)-cjJ(~)
65
upon changing the variable by s = <p (t), so that t goes along C. But 9 (<p (t)) =
f' (<p -1 (<p (t))) = f' (t) and similarly for 9 (<p (~)) ; so that, for every ~ on r
l' (0 =
_1
21ri
J f'
(3.3)
3!
and
<p' (t)
= <p' (0 + (t -
0 <pli (~)
+ (t ~!02 <p1/I (0 + (t
i= 0 Vt E N (0 - {O
so that
(3.4)
J.
. <p(t)-<p(O
c
'
l' (0 =
_1
21ri
Jf'
(3.5)
66
(iv). Since ais on rand not on C, q; (t) =/::. q; (a) for t on C, since q; is simple.
But q; (a) = 0; so q; (t) =/::. 0 for ton C. For sueh t
1
[(q;(o)n]
1 n-l
q; (t) - q; (~) 1 - q; (t)
= q; (t)?;
(q;(~))r
q; (t)
f (z) - f (a) =
r
=
where
(3.6)
21rZ r
1 J'
q; (~) d~
R = 21ri
f'(t)q;(~t
q; (tr [q; (t) _ q; (e;)] dt
r
c
and this agrees with the expression for R n +1 stated in the theorem at (3.2).
(v) . The function
() (z)
z-a
q; (z)
z-a
_ q; (a) , () (a)
= q; (z)
= q;' (a)
~
J f' (t) dt J
21rZ
q;(tf+l
= _1_
21ri
67
=
=
~~
t=a
t=a
by Cauchy's integral for the r th derivative. This gives the expression (3.1)
stated in the theorem by means of (3.6), after replacing r - 1 instead of rand
so completing the proof.
0
2.1
2: (ze-Zr (x+r,
)"
eXZ _
-1+
1- z
00
(3.7)
7'.
1'=1
and the hypothesis ofBrmann's Theorem requires that 4> be simple in a domain
D containing the origin, and that f be analytic in D. Now
f (z)
e=
1- z' 4>(z)
o(a) =
1
4>' (a)
= ze- z ,
O(z)
= 4>(z) = e
and
= 1 for a = O.
For f (z) to be analytic in D the disc Izl < 1 will be adequate. The definition
of a simple function is given by property 1, and the following lemma proves
that 4> is simple.
LEMMA
~}
Proof Assume on thc contrary that 4> (z) is not simple in D, so that therc are
for
Izl <
~ and
--, =e
z
z _ z'
z-z'
1
--:; =
<.
2:
00
- - ,- = eZ - z - 1 =
1'=1
2:
00
r=1
z - z' )
I
r.
1'-1
(z -
z')
r.
slllce z - z
l'
'
=I 0,
68
and so
-I'I :S L
Z
r=2
00
~ :S
r-l
r=2
L
00
= 1+
r.
r=l
00
Iz - z' I
00
Iz - z' I
r=2
r- l
r.
2r~1 we have
since the geometrie series is convergent, and by the ratio test we require
-1'1<
z -
1-
21z - z ' I
I
and we want
and analytic in D. All conditions ofBfmann's Theorem are now met. Now
we need to show that the remainder goes to zero for large n.
0
2.2
The Remainder
0 :S It -
~I
:S Itl + I~I
l:Qlereal(t-~)
Itl
< C2 + Cl = C3,
< Cl
1 4>(~)1
4>(t) - C2
eC3
I~ I < 1. Also
69
so that
Ir/J (t) - cfJ (01
(1 -
IcfJ (~)I)
IcfJ (t) 1
(1 - C2 = m > 0, say.
ThefunctionscfJ (z) = ze- z andf (z) = rzz are analytic in D = {z : Izl < H
> C2 e -
Cl e C3 )
C2
and SO are cfJ' (z) and l' (z) . Consequently cfJ' (z) is continuous on the compact
set r, and so bounded on r. Similarly l' (t) is continuous on the compact set
C, and so bounded on C. So there is an M independent of ~ and t such that
1cfJ' (~) l' (t) I S M for ~ on rand t on C. The inner integral in (3.2) has modulus
r
IJe
)!1 M
m
---4
(Cl eC3 )n M
nl
--> 00
= -tjz,
tt
n=O
1
1- z
~ (nze-zt
n!
n=O
1+z
3.
+ z2 + z3 + z4 + z5 + z6 + . .. .
Convergence Region
For convergence of (3.7) we apply the ratio test, such that
lim
r--+oo
+1
Ize-zllim
T--+OO
1)
(1+-+
T
Z)
+ :1:
X
r+x
1)
1+-T
+X
-x
for
70
Ize1-zl < 1
(3.8)
3.1
The series (3.7) cannot hold outside the region described by (3.8). To investigate if the series holds everywhere inside this region, which is where both
sides of (3.7) have meaning, we need corollary (ii), the principle of analytic
continuation, as given on page 89 ofTitchmarsh [106].
COROLLARY 3.1 lftwo fimctions are ana/ytic in a domain D, and are equal at
the points 0/a set S which has a limit point in D, then they are equal throughout
D.
F(w)
= ~ (n+xtwn
W
n=O
n!
(3.9)
where w = ze- z , and F (w) is therefore the sum of apower series which, by
the use ofthe ratio test, converges in the disc Iwl < e- 1 . The following lemma
is required and may be found on page 66 ofTitchmarsh [106].
LEMMA 3.2 Apowerseries represents an analytic/unction inside its circle 0/
convergence.
71
Chapter 4
Abstract
1.
A procedure which will allow a specific finite Binomial type sum to be expressed
in c10sed polynomial form will be developed in this chapter. The Binomial type
sum will be useful in the next chapter, where the results of Chapter 2 will be
generalized.
Introduction
In the next chapter we shall extend, in various directions the results ofChapter
2. In the course of these investigations we shall come across the finite sum
(
)n+m
()
n
r
T''' + m .
(4.1)
Therefore the major aim ofthis chapter is to develop a procedure that will allow
us to express (4.1), and its generalisation, in closed polynomial form. Initially
an infinite double sum will be obtained by the consideration of a Volterra integral
equation and the double sum then expanded to obtain coefficients in terms of a
recurrence relation, which upon further expansion will lead to sums of the form
(4.1). We shall then prove that the finite sum (4.1) can be written as a polynomial
in 11 of degree m, and a procedure for the evaluation ofthe polynomial will be
given.
2.
Problem Statement
Volterra integral equations of the form
J
t
73
(4.2)
74
1
\lI(p) = 1- cf! (p)'
(4.3)
where \lI (p) and cf! (p) are the Laplaee transforms of't/J (t) and c/J (t) respeetively.
Now, eonsider the rectangular wave c/J (t) = H (a - t) = 1- H (t - a) , where
H (t) is the Heaviside funetion, and taking the Laplaee transform of c/J (t) we
have
cf!(p) = 1- e- ap
(4.4)
p
and the
n th
(pn] =
n
a +l .
n +1
(4.5)
1
\lI(p) - - - p -1 + e- ap
(4.6)
It may be notieed from (4.7) that \lI (p) ean be written in the form
L m (a) pm
00
\lI (p) =
m=O
where
L an+mPm (n)
00
m (a) =
(4.8)
n=O
3.
A Recurrence Relation
LEMMA
(1 - a) m (a)
with o (a)
m-I
= L
k=O
(_l)m-k (~~~;)!da), m
I~a' a # 1
= 1,2,3, ...
75
m=1,2,3,...
(4.9)
and itis easy to see from (4.9)and (4.4) thato(a) = l-~(O) = l~a' a-:f.l.
From (4.9),
,
_.
[~
. [dm
{
=hm
p ...... O
=J~
<I> (p)
1 + 1 _ <I> (p)
dpTll
}]
<I> (p) }]
1 - <I> (p)
t; ;
[ rn (
) drn-k
dk {
1
}]
dpm-k {<I> (p)} dpk 1 _ q, (p)
. (4.10)
m!m (a) =
Lrn (-1)
k=O
k ( m ) arn - k+ 1 k!
k
k
da)
m- + 1
(4.11 )
and from (4.11) the m (a) are given by the recurrence relation
rn-I
m-k
am~k+l
(4.12)
(4.13)
and the coefficients ~i) (0) can be calculated from the recurrence relation in
(4.12) as folIows. From the left hand side of(4.12)
q
dq
-{(1-a)rn(a)}='"
da q
L
r=O
7'
7'
d --{1-a}~~)(O),
q=O,1,2 ....
q T
da
= q and 7' = q -
'
1, so that
76
Table 4.1.
0
I
I
2!
3!
4!
5!
6!
7!
8!
I
0
0
-I
-6
-36
-240
-1800
-15120
-141120
2
0
0
0
I
14
150
1560
16800
191520
3
0
0
0
0
-I
-30
-540
-8400
-126000
4
0
0
0
0
0
I
62
1806
40824
5
0
0
0
0
0
0
-I
-126
-4914
6
0
0
0
0
0
0
0
I
254
7
0
0
0
0
0
0
0
0
-I
t;
dq [m-I
am-k+1
da q
(_l)m-k (m _ k + l)!k (a)
= ~ (_l)m-k
6
k=O
m-I
q-m+k-1
q ()
+t;(-l)rn-k?;
) (q-m+k-I) (a)
k
m-k+l-q+r
(m~k+1_q+r)!kr)(a),
(4.15)
and setting a = 0 in (4.14) and (4.15) gives, after equating the right hand sides
and rearranging, the recurrence relation for the coefficients of the Maclaurin
series (4.13),
~) (0) = ~ (_l)m-k (
q_ m
~k_ 1 )
kq-m+k-I) (0),
q
= 0, 1,2,....
(4.16)
~O) (0)
= 1,
~) (0)
~q) (0)
= r~) (0)
= q!,
Further, ~) (0) = 0 for 0< q :::; m, and so the leading power of(4.13) is
am +l .
77
THEOREM
0/ degree
m.
m (a)
=L
an+m~:!+m) (0)
(n
n=l
+ m)!
00
)n+m
+ ""
-a
"" (-1)"
6
(n+m)! 6
n=m+l
7"=0
()
n
r
rn+m. (4.17)
From (4.12)
(_I)m (1 - a) m (a)
am+1o (a)
(m
aml (a)
+ 1)!
m!
+ ... +
a m - 12 (a)
a3m_2 (a)
3!
(m - 1)!
a 2 m_l
2!
am- 2 3 (a)
- ---,------'--'(m - 2)!
(a)
,m = 1,2,3, ....
(-1)m(m+l)!(1-a) ()
m+l
m a
a
= __
(m
(m
+ 1) m (m -
1) a 4 (1
24a 3 (1 - a) 4
+ 2a)
+ 1) m (m - 1)( m - 2) a 5 (6 + 32a + 8a 2 - a3 )
3!5!a 4 (1
.~.
a)""
(m + 1)!a m X (a)
+ ... + 2am- 1 (1 - a) m ,
( -1) m (m + I)! (1 - a) m+ 1 ()
am +1
m-l
= ( 1 - a)
(m+l)a
2 (1 - a)2-m
(m+l)ma(a+2)
+ -'----'------;,--~
12 (1 - a)3-m
(m + 1) m (m - 1) a (1 + 2a)
+-'--------~-~-~
24(1 - a)4-m
78
(m 1) m (m - 1) (m - 2) a (6 32a 8a 2 - a3 )
+~--~----------~~--------~
3!5! (1 - a)5-m
+ ... +
(m + l)!a m X (a)
2
andso
(-l)m(m+l)! ()
a m +1
m a
_m; 1E (m ; 2)
(~2+ 1) (m; 3) Hl' [2a'+1 + ak+2]
+ ~~
1) y: (m; 4)
+ +
+
~
5)
~
(_a)k+l
k=O
+ m
+ m (m
(_l)k [ak+l
(m -
2a k 2]
k=O
+
X
m (m
1) (m - l)(m - 2)
3!5!
3 _
(m + 1)!m (a)
=
k=O
(_I)k
(m k
t, (n~
m )
a+ + {a
n m
m- l
[-1 + 1_ :21)
m;
(m+l)m(m-1)(m-2)
3!5!
+ a m-2[(m
+
+ ...
(m
(m+1)(m-2)
- 1 - ~--":""':""-----'2
(m+l)m(m-l)]
12
+ . ..
+ ... + a
o}
[... J
79
~ (n~m
) {a n+2m p 1 (771)
+ ... + ~ (
n~:; 1 )
a n+m+1Pm_dm)
+~
(n~m )
a n +m + 1 Pm (m),
and so,
m (a)
771
71 )
a2m
+ ( ~l,
+~
( n-m+2
n +2
~
a 2m - l ] P2 (771)
) a n +m + 1 F (771)
2
n=m
+ [(
2;L~ 22
) a 2m
~)
+ ... + (
+~
( n +n 771
~
-1
1)
n=m
+ [(
)an+m+1Pdm)
n=m
2: ~ 11 ) a2m
am+2] P m - 1 (771)
n+m+ 1F
+ ... + (
%;l (
+ ...
()
m-I 771
~t
) a m +1] Pm
~ 771 )
(771)
an +m + 1 Pm (771).
(4.18)
80
m (a)
~ an +m +1 [ (
L.t
n=m
+ ... + (
n + 1 ) F (m)
n-m+l
1
+(
+(
n : : ; 1 ) Fm-dm)
n + 2 ) F (m)
n-m+2
2
n: m ) Fm (m)]
+ a2m Gl (m)
andso
m (a)
=
n=m+l
+ ... +
an+m [( n: m ) Fdm)
+(
(n::;2 ) Fm - 1 (m)+
n:!
~1 )
(n:~~l
F2 (m)
) Fm (m)]
+ I: an+mGm_n+l (m)
m
(4.19)
n=l
where the functions G j (m) ) like Fj (m) are dependent only on the fixed
parameter m. From the right hand side of(4.17) and (4.19) it may be seen that
= m + 1, m + 2, m + 3, ...
Since
81
4.
= ( ~l
Fm (m)
Gm- l (m) = ( m)
0
Fm- l (m) + (m+1)
1
Fm (m)
0
F (m) + (m+l)
1
F3 (m) + ...
( m)
G 2 (m) =
+(
2m - 3 )
m - 3
Fm-l (m)
( 2m - 2 )
m _ 2
Fm (m)
and therefore
Gk(m)=L...- (m+
j j ) Fj +k(m),k=1,2,3, ... ,m.
)=0
(4.22)
82
putting q = m
+k
G2m- q+ I (m) =
~) (0)
I ' q=m+1,m+2, ... ,2m,
q.
Gj (m) = (2m _ j
(0)
+ 1)!'
J = m, m - 1, ... ,2,1,
where the ~) (0) values can be obtained from (4.16). Therefore (4.22), may
be written as MF = B, B is a (m x 1) vector and in explicit form,
C;)
(mi I )
em-I)
m-I
(7;)
m- 2)
m-2
~m)(O)
FI (m)
F2
(2m)!
~m-l)(O)
(m)
(2m-I)!
=
(mi I )
(r;; )
(m;-2)
(mi I )
(r;; )
(m)
~:,,+3)(O)
F m - I (m)
~'+2)(O)
F m -2
Fm(m)
(m+3)!
(m+2)!
~'+l)(O)
(m+l)!
(m
+ l)IF
. 1 (m ) --
1+m + 1 (m + 1) m
2- 12
+0
+ (m+1)m(m-1)(m-2)
6!
+ ...
(4.23)
and for a particular value of m, that same number of terms are used on the right
hand side of(4.23). Some values of F I (m) are
Fd1)
= -2'
Fd2)
1
12' Pt (3)
= 0,
1
Fd4) = - 6!' Pt (5) = 0, ....
Various closed form polynomial representations of(4.1) are given in Table 4.2.
Cerone, Sofo and Watson [25], have shown a connection of the finite sum
(4.1) with Stirling numbers of the second kind and an association of the finite
83
Pm (n)
--
n (3n + 1)
4!
_n 2 (n + 1)
2 X 4!
11
n (15n 3
_n 2 (n
n (63n 5
_n 2 (n
Table 4.2.
42n
+ 16)
58n
+ 16)
24
n (13571 7
9!
X 27 X
91n 2
9!
_71 2 (71
2)
6!
+ 315n 4 + 315n:\ 23
2)
6!
+ 1) (371 2 + 771 16
+ 30n 2 + 5n -
X 28 X
234571 3
1O!
17n:J
+ 540n 2 64871 2
404n - 144)
+ 548n
- 144)
10!
Vrn(n,x) = (n+m)!
~(-1)
1"
(n)
(x+r)
n+m
(4.24)
andn.
84
Proo! The following result is needed and is quoted by FeIler [47] on page 65.
i = 0, 1,2, ... ,n - 1
i = n.
(4.25)
From (4.24),
Vo (n, x)
(4.26)
Changing the order of summation on the right hand side of (4.26), gives upon
using the above result (4.25)
Vo(n,x) =
(-Ir t(-lr
n.
r=O
(n) (x+r)n = 1.
r
(4.27)
J
x
Vm (n, x) =
Pm
(n) -
may be used to evaluate Vm (n,x). The Table 4.3lists some ofthe Vm (n,x)
in closed form.
In the next chapter we shall extend the results of Chapter 2, and also utilize
the polynomials obtained in this chapter.
85
v,,, (n, x)
1
(3n 2
+ 2x)
2
+ n (1 + 12x) + 12x 2 )
4!
(n 3
15n 4
+
Table 4.3.
(n
2n (60.T 2 (4x
+ 1) 8 x 6!
1) + 240x 4
Chapter 5
Abstract
1.
Introduction
We shall consider a forced differential-difference equation, of arbitrary order, and by the procedure developed in Chapter 2, generate sums which, by
the use of residue theory, may be represented in closed. As in Chapter 2 the
generalized identity, as it will be shown, depends on a dominant zero of an
associated transcendental characteristic function. We shall also develop recurrence relations for use in determining specific closed form express ions of the
infinite sumo We shall employ an induction argument to prove the closed form
representation of the infinite sum and then give a functional relations hip of the
sumo An extension to our main results will be indicated and in the process
utilize the finite binomial type sums obtained in Chapter 4; a connection with
renewal processes can also be made.
2.
2.1
I-Dominant Zero
The System
Consider, for a weIl behaved function f (t), the forced dynamical system
with constant real coefficients band c, real delay parameter a, and aB initial
87
88
conditions at rest,
t (R~
n==O
)CR-nf: (n)bn-rf(r)(t-(R-n)a)=w(t);
r
r==O
for t > Ra
for 0< t S Ra.
(5.1 )
In the system (5.1) w (t) is a forcing term, t a real variable, and R is a positive
integer, being the order of the differential-delay equation. Taking the Laplace
transform of (5.1) and utilizing the property
we obtain
From (5.2)
F(p) =
(p
W(p)
+ b + ce-ap)R
(5.3)
where F (p )and W (p) are the Laplace transforms of f (t) and w (t) respectively. Equation (5.3) may be expanded in series so that
F(p) =
W(p)
(p + b)R
(1 + c;~;;)
=~(n+R-1)W(p)(-ce-aPt
;:0
(p + b)n+R
(5.4)
To bring out the essential features of our results we may choose the forcing
term w (t) = 8 (t), the Dirac delta function, such that W (p) = 1. Substituting
for W (p) into (5.4) and taking the inverse Laplace transform, we have
f(t)=~( n+~-l)
x
(5.5)
89
where, H (x) is the unit Heaviside step function. The inverse of(5.3), a solution
ofthe system (5.1) by Laplace transform theory mayaiso be written as
,+ioo
=~
2m,
f(t)
ePtF(p)dp,
for an appropriate choice of'Y such that all the zeros of the characteristic function
91(p)=p+b+ce- ap
(5.6)
are contained to the left ofthe line in the Bromwich contour, and F (p) is defined
by (5.3). Now by the Residue Theorem
f(t)
= Lresiduesof
(ePtF(p)
where the sum is over all the characteristic zeros Pr of 91 (p) and Qr is the
contribution of the residues in F (p) at p = Pr. The zeros of the characteristic
function (5.6) with restriction
(5.7)
are all distinct. The poles of the expression (5.3) depend on the zeros of the
characteristic ftmction (5.6), namely, the zeros of 91 (p) . The dominant distinct
root Po, of 91 (po) = 0 is defined as one with the greatest real part and therefore
we have that asymptotically
R-1
f (t) '"
k=O
tR-k-1ePot
QR,k (po) (R _ k - 1)!'
(5.8)
[~l ( n + R -
f()-L
n=O
(5.9)
90
where [x J represents the integer part of x and the residue contribution QR,k (po) ,
is given by
and
k!QR,dO)
THEOREM
f) 1; ~
k
(5.11)
5.1 Let
_ ( n + R - 1 ) bne-b(t-an) (t - ant+ R- 1
Tn(b,R,a,t)n
(n+R-l)!
and
(5.13)
00
(5.14)
n=O
which is convergent for all values ofb, R, a and t in the region of convergence
(5.7). Then by the suggestive behaviour of (5.9)
R-l
SR (b, a, t) =
t R- k- 1
(5.15)
k=O
2.2
91
LEMMA
in (5.12) is
with
p->o
[(~(
)) R]
9 p
(1
+1ab) R'
(k
+ l)!QR,k+l (0)
Cfp) n] ; ~
)~ [,~ {;
R lim
p-,O
(5.17)
0, 1,2, ... , (R - 1)
[~ {pR-l (g (p) -
pg' (P))}]
(g (p) )R+l
'
dpk
g' (p)
= 1 + abe- ap .
Letting h (p) = 9 (p) - pg' (p) we find that h (0) = 0 and h' (0) = 0 and
therefore expanding h (p) as a Taylor series about p = 0 we may write, from
(5.17)
(k
where
+ l)!Qn,k+l (0) =
dk {( p )R+lh(p)}]
~~ [ dpk
9 (p)
7
(5.18)
92
(k + l)!QR,k+l (0)
. [d k
R ~~ dpk
{(
(5.19)
P ) R+l B (p) }]
9 (p)
~ R~
BI")
~)]
Now since
(p)
d/L
dp/L
= - B (p) .
[ dJ1.
]
( -1)/L ba/L+ 2
dp/L B (p) =
(f-t + 2)
~~
(k
+ 1) QR,k+dO)
= R
Lk
(-1)/Lba/L+ 2 (f-t + 1)
(+ 2)!
f-t
/L=O
QR+l,k-/L (0)
which completes the proof of Lemma 5.1. The following lemma regarding
moments of the generator function cp (x) will be proved and required in the
evaluation of another comprehensive recurrence relation for the contribution
QR,k (0) to the residues.
0
5.2 ThenthmomentoftheRthconvolutionofcp(x)
is (-ab)R (-ar n!C!;.
LEMMA
= -bH(a-x)
= -bH (a -
x)
= b( -1 + H (x -
(p) =
bR
(-1
bR
bR (-a
-1
b( -He-np)
p
a,
th
The R
convolution of
+pe-ap)R
+ 2:
00
(-~r)
(5.20)
r)R
,.
r=O
f ( -ap)~,)
+
r=O r
1.
R -- "
1 2 3 ...
R = (-ab)R
f cf!-
r=O
(-apf .
93
c~ = ', =
R=l
(.11)1;
and they are polynomials in R of degree r; in fact they are related to Stirling
polynomials ofthe second kind so that cf? = (-Ir Pr (R) where Pr (R) are
the polynomials fully described in Chapter 4. The n th moment of the R th
convolution can be obtained by differentiating (5.20) n times with respect to p,
so that
and therefore
d71
lim [If>R(p)]
p->o dp71
= (-ab)R(-atn!C;;,
hence the proof of the lemma is complete. This lemma may now be used to determine arecurrence for Qn,k (0) wh ich, it is argued to be more computationally
0
efficient than directly using (5.12).
LEMMA
(5.12) is:
QR dO)
=
x
1
(l-(-ab)R)
[1: (
R )
J
j=O
(-ab)R QR,dO)]
k!Qll,dO)
!:-.
[(1 + 1 -If> If>(p)(p) ) R]
dpk
=J~
p->O
[(
If>J
1]
(1 - If> (p}
1)
(5.21)
94
k!QR,k (0)
1
(1
2 (1
Table 5.1.
+ ab)R+l
{Rb
2}
Rba 3
R+2 {ab(3R - 1) - 4}.
12{1+ab)
+ ab)R
8 (1
Rba 5
240 (1 + ab)RH
"{
Rba R+'
96 (I + ab) "
Rb~ RH { 2 - 4abR + a 22
b R (R - 1) } .
+ ab)
a3b3~15R3_30R2+5R+2)
}
-a 2b (120R 2 + 40R - 16) + ab (200R + 56) - 48 .
+ 16
=;~ ~
R)
(k) d
d
j
~ r dpk-r [<I>J (p)] dpr
k r
-
1
}
(1 - <I> (p))j .
Utilizing Lemma 5.2, for the (k - r) th moment of<I>j (p) implies that
R
k!QR,k (0) =
~( ~) ~(
R
QR,k (0) =
~( ~) ~( ~
using the fact that Cf} = 1 and taking the term at j = R, r = k to the left hand
side we obtain the recurrence (5.21). Now using recurrence (5.16), or (5.21),
we can list some values of Q R,k (0) as given in Table 5.1.
95
R (1
5.4
d
(k
(5.22)
:i
I
db {k.QR,k
(O)} -:i
- db {.
;~ [~(~)R]}
dpk 9 (p)
.
Interchanging the order of differentiation in the second term, and after some
simplification, we obtain
:i
I
_.
[~{ _RpR~g
(p)}]
db {k.QR,k (On -;~ dpk
gR+I (p)
and since
d ()
9 (p) - P
db g P =
b
we have
d
R. [d k {pR
pR+I}]
k! db QR,d O) = -t;;~ dpk gR (p) - gR+I (p)
,
and using (5.12)
d
b db Qn,k (0)
+ Cl (R -
(k
Cl,
(5.23)
the (R - (k + 1)) factor in Q R,k+ 1 (0) is required since it does not contribute
for R = k + 1. From Table 5.1 we choose three k values and substitute the
respective Q (0) values in (5.24), then solving for Cl, C2, and C3 we obtain
Cl
= - R (1 + ab) , C2 = - 1 + ab'
and
C3
= 1 + ab'
QR+l,k+1 (0) -
(R-(k+1))
a
R (1 + ab) QR,k+l (0) - 1 + ab QR,k (0)
96
1
1 + ab'
ba 2
(1+ab)2
t2
2 (1
ba 2t 2
t3
6(1+ab)4
t4
24 (1
Table 5.2.
+ ab)3
+ (1+ab)5 -
+ ab)5
5ba 4
3ba 2t
2 (1 + ab)4
ba 3 (1 - 2ab)
2(1+ab)5 .
ba 3 (4 - Uab) t
6(1+ab)6
5ba 2t 3
12 (1 + ab)6
+ (1+ab)3'
5ba 3 (2 -7ab)t 2
(1+ab)7
ba 5 (3 - 66ab + 199a 2b2 - 72a 3b3)
3(1+ab)9
(5.25)
Substituting (5.23) in (5.25), and after some minor manipulation, we obtain the
result (5.22), and the proof of Lemma 5.4 is complete. Using the QR,k (0) in
Table 5.1, some closed formrepresentation ofthe infinite series (5.15) are given
in Table 5.2.
2.3
The following lemma will be useful for the proof of Theorem 5.1.
LEMMA
(5.26)
97
d
db SR
1 + ab
= - b - L nTn
00
(5.28)
tSR
n=O
Now multiplying (5.27) and (5.28) by R (1 + ab) and ab respectively and substituting into the left hand side of (5.26), gives
+ ab) L nTn
00
(1 +ab)tSR - a (1
n=O
+ a (1
+ ab)
L nT
00
n -
(1
+ ab) tSR =
n=O
which is identical to the right hand side of (5.26) and the proof is complete.
Pra%/ Theorem 5.1. The proof of Theorem 5.1 will involve an induction
argument on the parameter R. Lemma 5.5 proves the left hand side of(5.15).
For the basis, R = 1, a proof of (5.15) has been given in Chapter 2. For R = 2,
a proof of (5.15), by Brmann's Theorem has been given by Sofo and Cerone
[99]. The induction argument for the right hand side of(5.15) will involve the
recurrence relation (5.26). From (5.26)
SR+l
R (1
~ ab)
(5.29)
[tSR - ab :b SR]
[R-l
1
= R(1 +ab) t
t;
tR - k - 1
(R - k _1)!QR,d O)
R-l
L (R t_ k _ I)! dbd QR,k (0)1
R- k- 1
- ab
k=O
98
= R (1
+ ab)
[Rt R
R! QR,O (0)
t;
R (R - k) t R- k
(R _ k)! QR,k (0)
R
tR- k d
- ab [ ; (R _ k)! db QR,k-dO)
where the counter in the third term has been adjusted. Now collecting tenns in
(5.29) we have that
SR+l = R (1
+ ab) [R; [ (R -
d]
tR - k
tR
R (1
d
(R - k) QR,k (0) - ab db QR,k-l (0)
(5.31)
so that by substituting (5.31) into the square bracket of(5.30) we have that
tR
(1
k=l
tR
t R- k
(R _ k)!QR+l,k (0)
(5.32)
t R- k
where Ql,O (0) is identified in (5.16) or Table 5.1. By the convolution nature
ofthe Q (0) tenns we may write equation (5.32) as
tR
R!QR+l,O (0)
+L
k=l
t R- k
(R _ k)!QR+l,k (0)
= aT, p = R -
1 and 'Y
= abeab then
99
(jp(7)
(-1T (7 + nt+ P
I
,p=0,1,2,3 ...
n.
(7)
+ "Wp (7 + 1) = T(}p-I
DO
n=O
and
(jp
(7).
(5.33)
Pyke and Weinstock [89] gave a functional relationship of (5.33) for the case
of R = 1 only. Sofo and Cerone [99] have given a proof of the functional fonn
(5.33) for the general case of integer R.
2.4
Extension of Results
The dynamical system (5.1) may take other functional values of the forcing
tenns w (t), other than J (t), such that consequent results of (5.15) may be
extended. We shall consider two other cases.
L
DO
n=O
+R
(n+m+R-l)!
R-I
;;:0
= '"
(0)
tR-/l-IQ
R,/l
(R-I'l-l)!
m-I
tm-v-I
+ '"
Pmv(-b).
~(m-v-l)!
'
1;
and
F (p) =
For R
DO
'"
~
n=O
1
(p
+ b)m (p + b -
be-ap)R
= 1 and m = 2 we have
bne-b(t-an) (t _ ant+ 2
(n
+ 2)!
e- bt [
= - -beab
1 + abe ab ]
be ab
+ --,,----b2 (1 + ab)
100
n=O
n + R _ 1 ) blle-bttn+m+R-I
n
(n+m+R-l)!
(R) tm-v-l
"" R v v
~
b + v! (m - v - I)!
(_I)R
m-l
f::o
+~
"" bm+Jl t-t! (RJl-
tR-Jl-1
t-t - I)!'
where (x)p is known as Pochhammer's symbol. The identities (5.1) and (5.15)
may be differentiated and integrated with respect to t to produce more identWes.
CASE
5.2 Let
w(t)
tm -
(m -I)!
-b
L()
00
n=O
n (
n+ Rn - 1) L (n) (-Ir(n+m+R-l)!
(t - arr+ +
m
r=O
R- 1
(5.35)
= 0, 1,2, ... ,m + R -
1; m
1; R
101
1
F (p) = - - - - - - =
pm (p + b - be-ap)R
(+
n
n=O
R
n
-1) (_lt+m +R - 1
(n+m+R-1)!
t, (~ )
(_lf(x+r)n+m+R-l
m+R-l (_x)m+R-k-l QR,k (0)
2::
ak(m+R-k-l)!
k=O
where the inner sum on the lefl hand side is the general polynomial investigated
in Chapter 4, namely
(_lt+ m + R - 1
Vm+R-dn,x) = (n+m+R-1)! ~
n )
(-1) (x+r)
n+m+R-l
f(-abr'(n+~-l
)Vm+R-l(n,x)
n=O
k=O
ak(m+R-k-I)!'
(5.36)
By an application of the ratio test the infinite sum in (5.36) converges in the
ofx = 0 (a # 0), Vm+R-l (n, 0) = Pm+R-l (n), where Pm+R-l (n) are the
polynomials in Chapter 4, described by (4.1), and therejorefi'om (5.36)
2::
00
n=O
(_
l)n( n+R-l)p
() __ QR,m+R-l(O)
m+R-l n
n
a m +R - 1 .
aJ
~(_bt(n+R-1)
~
n
n=O
(5.37)
tn
t(-l
(n+m+R-1)!
r=O
m+R-l
2::
k=()
r (n)
r
CkQR,k (0)
(m+R-k-1)!'
102
since on the lejt, the inner sum is unity for n = and zero otherwise and on the
right, using (5.35) QR,k (0) = 1for k = 0, and zero otherwise. Some examples
are now illustrated. Putting a = -ab in (5.36), we havefor the case of R = 1
andm = 1 that
2x
a
L an (n + 2x) = -+
1- a
(1 - a)
00
2'
n=O
00
and when
= 0, n~o na n =
f:
(1_0'0')2'
o(;::o)J,
na n =
which is obviously incorrect. Jolley gives
n=O
no other entries of this form, apart from Chapter 2, (2.12). We have given a
general method for closed form representations. For R = 2 and m = 2 we
have that
gives the listing
L an (n + 1) (n
00
n=O
8x 3
= (1 -
24ax 2
2+ (1 a)
3+
a)
+ 4) 4a (a 2 + 4a + 1)
4 +
"
(1 - a)
(1 - at
4ax (5a
2.5
Renewal Processes
fl
n=1
J(n)
(x) dx
M(p) =
!(l!) ].
(5.38)
p 1 - J (p)
! (p)
1-
J (p)
1ft M
(t) , whenever
(5.39)
M (p) , in (p) and f (p) are, respectively, the Laplace transforms of M (t) , m (t)
and J (t) . Feller [47] obtains the average number of registrations of an event
103
J
t
M (t)
1 - e- bt
+-
M (t - x) be-b(x-a) H (x - a) dx.
(5.40)
1
M (p) = p (p + b _ be-ap)
and upon inversion the result is identical to that given by (5.34) with R = 1
and m = 1. Chaudhry [27] considers various other forms ofthe function J (p)
from which expected instantaneous renewal rates are evaluated. For the shifted
exponential function
f (t) =
(1- e-b(t-a
J)
H(t - a)
p) = p2
b -ap
e
+ bp _ be-ap
where
(5.41)
From the work ofthe previous section we may write
m(t)
rv
00
n ( n ) (_1)n+"bn+1(t_a(T+1))n+r+l
~~
LL T
(n+T+1)!
n=Or=O
where ~j, j = 0,1 are the two distinct dominant zeros of the characteristic
function 9 (p) in (5.41). In the next part we shall prove closed form representations of infinite sums which depend on k dominant zeros of an associated
characteristic function.
3.
3.1
with constant real coefficients band c, real delay parameter a, and all initial
104
conditions at rest,
cf
(t - a)
r~o (
; )
+ r~o (
; )
<
(5.42)
t ::; a.
In the system (5.42) w (t) is a forcing term, t a real variable, and k is a positive
integer. If we let w (t) = <5 (t) , for illustration, and use the methods of the
previous section we obtain
F(p)=
(p + b) + ce-ap
(5.43)
L
n=O
(5.44)
v=O
where
Q(~v)=lim [(p-~v)F(p)],
P-->~1/
(5.45)
and ~v are defined as the k dominant distinct zeros ofthe characteristic function
9 (p) = (p + b)k + ce-aP. To simplify the algebra let us take c + bk = 0, which
allows one dominant zero of the characteristic fWlction
(5.46)
with k + ab > 0, to occur at the origin. The condition k + ab #- 0 will ensure
the distinct nature ofthe zeros of(5.46). From these considerations and (5.44)
we have the following theorem.
THEOREM
5.2 Let
T n (k, b, a, t) =
bnke-b(t-an) (t - antk+k-l
(nk
+k _
I)!
and
(5.47)
00
S(k,b,a,t) = LTn(k,b,a,t)
(5.48)
n=O
Generalization
105
Then
k-l
S(k,b,a,t) = LQ(~II)e~vt,
(5.50)
11=0
where Q (~II) is d~fined in (5.45) and ~II are the k dominant distinct zeros of the
characteristic fimction (5.46).
The following two lemmas, regarding the loeation of dominant zeros, will
be useful in the proof of Theorem 5.2.
LEMMA 5.6 The characteristic function (5.46) has k simple dominant zeros
lying in the region r : Ipl < k~ab; a, b > 0 and k is a positive integer.
Proof We have previously defined a dominant zero as the one with the greatest
real part. It is known, see [7], that (5.46) has an infinite number of zeros lying
in the left (or right) half plane. Using the same method as deseribed in Chapter
2, Lemma 2.3, it ean be shown that (5.46) has at most three (and at least one)
real zeros with restrietion (5.49) one ofwhieh is at the origin, ~o = 0. Applying
Rouehe's Theorem it is required to show that
- A (w)1
forw =p+b,
A
(w)
= wk , B
(w)
= wk _
bkeab-aw
in the region r' : Iwl < k+;ab. Now A (w) has k zeros lying in the region r'
and since jwkj > j_bkpab-awj implies that (I.: + 2ab)h' > (ab)k; then B (w)
has k dominant zeros lying in the region r' and henee (5.46) has k dominant
zeros lying in r.
0
LEMMA
()] (p)
=p+b -
be(21rij-ap)/k
(5.51)
for j = 0, 1, 2, 3, ... , k - 1 has one dominant zerofor each j lying in the region
as defined in Lemma 5.6.
1lJ
BI
(w)
=W
r' and
be(21rij+ab-aw)/k.
Therefore Iwl > l_be(21ri j +ab-aw)/kl implies that (k + 2ab) > ab, henee
BI (w) has one zero lying inside the region
and it follows that for j
0,1,2,3, ... , k - 1, (5.51) has one dominant zero lying in the region r.
0
r'
106
Proof ofTheorem 5.2. Firstly, we evaluate Q (~v) from (5.45) and from (5.50)
we may write
00
S (k, b, a, t) =
k-l
(5.52)
e~vt
~ (b+~v)k-l(k+ab+a~v)
gk (p)
= (p + b)k -
bke- ap
k-l
= II (p + b -
be(21ri j -ap)/k) =
j=O
k-l
II qj (p).
j=O
Lemmas 5.6 and 5.7 show that the dominant zeros, O'j, of qj (O:j) for each
j = 0, 1,2,3, ... , k - 1 are the same as the k dominant zeros of gk (p) . Using
(5.45), the contribution n (O:j) to each ofthe factors qj (O:j) is
n (O:j) =
p..... OI.j
~ (be27rij/kt e-b(t- T ) (t
~
n!
Tr
keOl. jt
(5.53)
k+ab+aO:j
n=O
for each k = 1,2,3, ... and j = 0, 1,2,3, ... , k - 1. Note that the sum (5.53)
may in fact be a complex number. The summation of all the k dominant zeros,
for each ofthe factors qj (O:j) implies from (5.53) that
k-l
k OI.t
""'
e J
~
k+ab+ao:'
j=O
)
I:I:
(be21rij/kt e-b~t,-T)
j=On=O
=
L b e-b(t-ann. (t _ k L. e27rijn/k.
00
!!!!)n k-l
)=0
~ bnke-b(t-an~~~ ~ ~)~n -
~
n=-l
Tr
n=O
kbk
(t -
at k+ k
~ keO: jt
= ~ k+ab+ao:
j=o)
Generalization
Now letting y
=t -
107
a and, from
qj
(a j
) ,
eaa
= (b:a
j )
then
00
n=O
bnke-b(y-an) (y _ an)nk+k
(nk + k)!
n=O
(nk
+k -
1)!
ay
e
k-l J
j=O (b+oj)
(k+ab+aaj)
k-l
'~
"
(5.55)
Renaming y as t shows that (5.55) is the same as (5.52) since by Lemmas 5.6
and 5.7, aj = ~v for j = 0,1,2,3, ... , k - 1; lJ = 0,1,2,3, ... , k - 1, and
therefore Theorem 5.2 is proved. Some examples are now given to illustrate
the above theorem.
0
3.2
Examples
(i) . For k even there are 2 real dominant distinct zeros and (k - 2) complex
conjugate zeros of the characteristic function (5.46) that need to be considered
for determining the right hand side of (5.52). Consider, in particular, the case
k = 2, then
For (a, b, t) = (.1,2,2) then (~o, 6) = (0, -4.5053) and the sum takes the
value, to four significant digits, .2272.
(ii) . For k odd there are 3 real dominant distinct zeros and (k - 1) complex
conjugate zeros ofthe characteristic function (5.46) that need to be considered
for determining the right hand side of (5.52). Consider, in particular, k = 3, in
this case there will be one real zero ~o and two complex conjugate zeros
= (x
+ iy),
~I = (x - iy)
108
+ b)3 -
b3e- aev
b3ne-b(t-an) (t _ an )3n+2
(3n+2)!
where
(~o, 6,
(1)
and the sum takes the value, to four significant digits, .2769. Again the previous
results (5.52) may be extended in various directions, we briefly mention one
extension.
3.3
Extension
Consider, for a weil behaved function j (t), the forced dynamical system
with constant real coefficients band c, real delay parameter a, and all initial
conditions at rest,
t(
j=O
E(
r=O
~.) bk(R-j)
J
Rk )
b Rk - r j(r)
r=O
(jk )
r
(t) = w (t);
blk-rj(r)
(t - (R - j)a) = w(t);
t > Ra
0< t
Ra.
(5.56)
In the system (5.56) w (t) is a forcing term, t a real variable, and Rand kare
m - 1 fi
. . .mtegers. Let w (t ) = e-btt
posItive
(rn-I)!'
or m = 1, 2 ,3, ... and tak'mg t he
Laplace transform of (5.56) we have
Generalization
109
+R _ 1 )
n=O
(nk
e-bten-r-I Pm,r
(-b)
+ Rk + m -
(m - r - I)!
- ~
r=O
I)!
(5.58)
(R - f.L - I)!
~ ~
v=O IL=O
where
lim [ddr ((p
r!Pm,r (-b)
p-->-b
pr
lL ((p plL
lim [dd
P-->(v
~v)R F (P))];
F (p) is defined by (5.57) and ~/J' V = 0,1,2,3, ... , k - 1 are the k dominant
zeros ofthe characteristic function (5.46). For (R, k, m) = (2,2,3) we have,
from (5.58)
b2ne-b(t-an) (t _ an)2n+6
00
L(n+1)
(2n+6)!
n=O
1
[b2eabt (t + 4a) + 4 (1 + a 2b2e ab )]
2 (b 2 eab )3
e("t
2-a2b2e-a~,/]
2 (b + ~I/) + ab 2 e- a(v
'
where~v are the two dominant zeros of(~v + b)2 _b 2 e- a (v = O. The degenerate
case of a = 0, implies that the transcendental function (5.46) reduces to a
polynomialin p of degree k. Specifically for (a, R, k, m) = (0,2,2,3) we have
the identity
(bt)2n+G
1
L (n + 1) (2n
+ 6)! = 2 [(bt)2 + 4 + btsinh (bt) - 4cosh (bt)] .
00
n=O
Chapter 6
HYPERGEOMETRIC SUMMATION:
FIBONACCI AND RELATED SERIES
Abstract
1.
A first order difference-delay system is considered and by the use of Z transform theory generate an infinite sum which by the use of residue theory may be
represented in closed form. Related works to this area of study are considered
and some central binomial coefficient identities are also given. A development
of Fibonacci and related polynomials is undertaken together with products and
functional forms.
Introduction
2.
fn+l - bfn
=cfn-a
fn+l - bfn - 0,
= 0,
n 2:: a
n <a
(6.1)
with initial condition fo = 1, band c are real constants and a and n are positive
integers including zero. The Z transfonn of a sequence {fn} is a function F (z)
of complex variable z defined by F (z)
00
= Z [fn] = I:
n=O
of z for which the infinite series converges. Taking the Z transfonn of (6.l)
111
112
F(z)=
z - b - cz- a
za+l
za+l - bz a - c'
(6.2)
=-~--
F(z) =
t. (
I-ar
~ (: ~ b)Hr'
(6.3)
in =
[n/(a+l)]
~ ar
G)
b(n-ar)u (n - ar)
(6.4)
(n ~ ar ) Gf b(n-ar)
where U (n - ar) is the discrete step function and [x] represents the integer
part of x. The inverse Z transform of (6.3) mayaiso be expressed as
in
2~i /
zn-l
F(z) dz =
znResj
J=O
(F ;z))
(6.5)
where C is a smooth Jordan curve enclosing the singularities of (6.2) and the
integral is traversed once in an anticlockwise direction around C. It mayaiso be
shown that in (6.5) there is no contribution from the integral around the contour.
For the restriction
c(a + 1t+1
<1
b(abt
(6.6)
(6.7)
za]
za+l - bza - c
(a
~.J
+ 1) ~j -
(6.8)
ab
in =
L (n -r ar ) (:)r
b(n-ar) = L
r=O
b
j=O (a + 1) ~j
[n/(a+l)]
cn+l
<"j
ab
(6.9)
113
r=O
~ ar
r
(~)r b(n-ar)
n+l
rv
..,...--~-"-o_
(a
+ 1) ~o ~
ab
where ~o is the dominant zero of (6.7), defined as the one with the greatest
modulus.
3.
ThelnfinHeSum
THEOREM
~
~
( n
~ ar
r
r=O
(~)r b(n-ar)
-:----'<-~~_+1_
(a
(6.10)
+ 1) ~o ~ ab
(a
+ 1) ~o
ab
and so
00
'"'
~
~ (n
+ ar)
( )
r=[(n+l)/a]
a
cn+1
(~)r b(n-ar) = ~ '"'
-:-----,'"Je-.'_ _
b
J~l (a + 1) ~j ~ ab
~
Thus
'00
"
~
r=[(n+l)/a]
(~lr+l
aT+'r~l~n
r
a
(:)rb(n-ar)=",
b
~ (a
J=1
b=_o_
a + ~'
cn+1
'"j
+ 1) ~
~ ab
114
also, let n = - a and substitute in the left hand side of (6.10) such that
t, (
~ t,
-a (~+ r) )
arb-a(+r)
(-1)' ( a + ar/ r -
~ ( -1 r
00
xa
( a + arr+ r -
(6.11)
1) (0;+;8 f+'J
a'
t;
a(+r)
1 )
(a; ar )
Now, expand (6.11) term by term and sum the convergent double sum diagonally
from the top right hand corner, thus gathering inverse powers of ~o, such that
~ca(+r)~(_lr-k(
~ 0
r=O
~
k=O
a(+r-k) )
a ( + r - k) - k
X (
a (
+r -
k)
+r -
r- k
k-
1)
and after some lengthy but straightforward algebra we may simplify the double
sumto
4.
rp
115
obtains
j ) yj
f_n-.
(n+
j=on+J
.7
= (z+l)n
where y = z (z + 1) - . Similarly, from the Lagrange alternate formula, Theorem 2.5, lensen obtains
~(n+j)yj=
~
j
(z+lt+ 1 =
(z+lt
l-(-l)z
l-(-Z)'
J-O
(z
(6.12)
z+l
+ 1t+ m
and
f = (z + 1) , and obtains
+ l)m (z + l)n
1 - (z~l)
(6.13)
(n+m+ j )yj
j=O
J
fyj
j=O
{t ~ (
n +.j ) ( m
J
+ J
j=O
= m- k,
lengen notes that (6.15) is analogous to Abel's identity, see Chapter 2, and is
comprised in the more general formula, duc to M.I.G. Hagen,
a (p + q - nd) + bnq (
(p + q) (p - nd) q
]J
+q
a + bj
( q + jd ) ( p - j )
-.i=o(q+jd)(p-jd)
j
n-j
lengen also used (6.15) and the method of recurrences to prove the novel identity
116
which Cohen and Sun [33] gave a more general form by the use of general series transformations. Chu Wenchang [28] obtained even more general
forms of binornial convolution identities by the use of power series expansions and convolutions. We can now show that our surn (6.10) is the same as
Jensen's identity (6.12) and that (6.10) holds for all real parameter values. Let
c = ab, a = -, ab = y and z = a~g, frorn(6.7)wehavethat~o = b (z + 1)
and y = z/ (z + l), and substituting in (6.10) we obtain Jensen's identity
(6.12). Ifwe let = 0, (6.12) reduces to the binornial theorem. To show that
(6.12) holds for all real parameter values, we first write
~
00
+ r
r
zr
(z + I)r+n+l
1 - ( - 1) z
and putting the denominator in the left hand side in series form we can write
expanding the double series and sumrning diagonally from top left hand corner
we obtain powers of z, namely
and since the inner sum is equivalent to ( - 1) r, the parameters n and are
arbitrary and therefore belong to the set of real numbers. In the next section we
shall investigate rnany interesting cases of the identity (6.10) and an associated
related result. We shall investigate the case of a = 1.
5.
I(a,x) =
f( 2r+a-l )x
r=O
2Fl [ ~,:tl
viI - 4x
(1+JI=4X)~~.~6)
2
4X]
117
v'2
g (O:,X )
(
=~
~
4r +2r 1) x
0: -
2r
'r=O
so that
g (0: 1/8) =
2(20'-1)/2 ( ( 1
J372r
J3
-0'
+ (1 + v'2)
J2
1-0') .
for constants al,a2 and 0:. It also follows from (6.16) that for
have the identity
22 F1[
or
0:
and E 3{ we
a;:,:]+
4X]
t, (
2r + ; - 1 ) x r
=(1+v'1-4X)~(
2r+0:: -l
)x
r.
(6.18)
(~) = (r~ ~ ) 2 2r and this result will be generalized in the next chapter.
118
t,r (2r+;
-1
0+31 4x] .
+ a) x 2F l
0+2
[ -2-' -2-
a+1
00
2~' Thus we
t,
rk
2r +;
1 ) xr
for k integer. Similarly, (6.16) mayaiso be integrated to obtain more identities. Integrating the right hand side of (6.16) will necessitate branch cuts and
singularities dependent on the value of a. So that putting a = 1 will simplify
the integral. From (6.16)
9j(x)
=~
r-O
(2;) :+:). = ]
(rx
t=O
...
1~,
t=o
j - times
"'-v--'
where (r
(2r)_1
r r+1
93 (x)
93 (-x) =
12~x3 {-1 -
(1 - 4x)5/2} ,
lOx + 30x 2 + (1
+ 4x)5/2} ,
119
so that 93 (x)
~ ( ;~ )
+ 1) (2::: 2) (21' + 3)
(21'
= _1_ {(I
240x 3
Other values of
identity
0:
~(21'+l
~
r=O
l'
may be utilized; if
)xr=
[1,~
2F l
+ 4x)5/2 _ 20x -
0:
1 41;]
(1 _ 4x)5/2} .
1 - 4x + VI
4x
(6.20)
L:
oo
1) - + = -4x1{I - 2x - ~4x}
MCr = e
r12 = 2:tiCr
x
21' +
xr
l'
r=O
l'
vi -
r: l )
where the coefficient
modified Catalan numbers, and in particular for
converging series
( 21' + 1 ) 4- r
~
l'
1'+2
may be thought of as
= 2.
r=O
~
00
27') (1')
1 JX ... lXI (1
vr=-:u - 1) Tdt
xr
l'
r-l
= xj-l
t=O
1=0
j - times
~
1{I- - - + -
h3 (1 x) = "
x2
24
x
2
3x 2
4
+ (lOx -
+ x 2 In 2
1) VI
41;
24
-x 21 n ( 1+ ~)}
1- x
'
1{I + - + -
h3 (1 -x) = ,
x2
24
3x 2
+ x 2 In24
(lOx
+ 1) vT+4X _ x 2 1n (1 + VI + 4X)}
24
120
1 {1
2x2 12
+ 2 +x
21 2
n
+ ~4v'I+4X -
- (10x
3x 2
(lOx - 1) vT=4X
24
x 2 1n ((1
+ V1 +4x)
(1
+ V1- 4X))}
i, t /16X2] .
x 2 43
F [ 1, 1,
4
2, Q2' 3
00 (
f (1 x2) = '"'
1,
1=0
2r
2r+1 = areSln
. 2x
_x__
2r + 1
2'
(1 x 2)
=~ (
~
2,
2r )
x 2r +2
(2r + 1)(2r + 2)
?;
00
= V1 -
4x 2 + 2xarcsin2x - 1.
= 1 - J2 - In ( J2 -
1) and adding
4r )
2- 4r
7r
1
1
- - - - - - l n h-l
2r (4r+1)(4r+2)-4 J2 2 (
).
5.1
3'
0+1 0+2
-3-' -3o 0+1
2'-2-
= 2,
/~] .
Related Results
The previous results that involve the aresin x function suggests we may
00
design a related sum of the form r~l (~!) which may be expressed in closed
f (x)
121
f (x) =
aresinx
Jf=X2
(2x)2r-l
r=l r (
2; )
(6.21)
and
and
sides of(6.21) by 2x and differentiating any number oftimes will produee other
identities. Differentiating (6.21) onee, and simplifying we have
f 1 ()
X
2::
00
(2x )2r
r=l (
:r
{
ares in x }
= - - x+---===
2;) 1-
x2
Jf=X2'
(6.22)
so that,
7r
= -2 + 1'
and
122
partieularly,
and
i)
f ( =~
2 2
~
(-Irr
= ~ (~ln (v'5- 1
) -3).
( 2;)
25 v'5
2
More general identities ofthe form (6.22) and (6.23) have been given by Chud00
(5;) = ~ + 1. In
their reeent paper, Chudnovsky and Chudnovsky [29] obtain a master theorem
from whieh they ean, amongst other manipulations, derive explieit expressions
of eontiguous generalized hypergeometrie funetions. The identity (6.21) may
also be integrated repeatedly. Integrating (6.21) onee we have
gl
(x) =
L
00
r=l
r2
(2 )2r
(x 2r ) = 2 (arcsinx)2
r
and
gdl) =
gl (i)
gl
1T 2
"2 =
3( (2),
= 2(ln ( J2 _ 1)) 2 ,
( ~)
2
_ 1T 2 _
- 18 -
((2)
3 '
(6.24)
123
g4 (x) =
(2x)2r+3
I:
00
~~.
r=l r 2 (
) 4 (2r
+ 3) (21' + 2) (2r + 1)
x (2x 2 + 3) (arcsinx)2
3
. 2~ (11x 2
+
9
and g4 (1) = 52~2
22r
00
1'=1 r 2 (
85x 3
8x
27 - 9
(6.25)
0e7!"2
+ ,
2; ) (2r + 1) j
I:
1'=1
+ 4) arcsinx
Oe
= 1771"4 =
r 4 (2;)
3240
5 F4 [
2'
131,1,1,1
2,2,2,2
I~]
00
rm
r)
for m
> 4.
We ean obtain identities of this form, so that from (6.25) with x = 1/2, we
have
6.
in+1 - bin -
Cfrt-l
= 0, io = 1
(6.26)
and
(6.27)
124
In
b
b2
b3
b4
b5
b6
b7
b9
+ 3b2 c + c2
+ 4b 3 c + 3bc2
+ 5b4 c + 6b2 c2 + c3
bl l
11
b" 2
+ lOb 2 c3 + c4
biO + 9b Bc + 28b6 c2
10
Table 6.1.
+ 2bc
bB + 7b6 c + 15b4 c2
12
+c
+ 35b4 c3 + 15b2 c4 + c5
125
name
gen. fune.
zeros
Fibonaeci
z2 - z - 1
lv'5
-2-
laeobsthal Z2 - Z - 2
2, -1
Pell
Z2 - 2z - 1
-2
Fennat
Z2 - 3z + 2
2, 1
-I
Cebysev
2z + 1
1, 1
Table 6.2.
Z2 -
solution of (6.26)
7s {(I+2v'5f+l -
e-
2v'5f+l}
~(2"-(-1)")
J2
~ { (1 + J2)"+l - (1 - J2(+l }
2n~ J -
n+l
In
In = (4ct
rr
J=1
rr
j=1
cos
(2r~~ 1)
(2n7rj+ 1) = 2-".
2~O,1
cos 2
= b Vb 2 + 4c
b - Vb22
+ 4C) n+1}
(6.28)
126
and putting b = (x - 1)2 and c = bx, we obtain the result obtained by Binz
[10]; namely
[n/2]
( n
~r
) x T (x - 1)2n-2T
= x 2 ~ 1 {(x (x -
1)t+ 1 - (1 - xt+ 1 }
We can integrate, for 0 ::; x ::; 1, the result given by Binz, therefore producing
the new identity
where B (x, y) is the Beta function and r (x) is the classical Gamma function.
Differentiating (6.28) with respect to c and substituting we obtain, in an easier
manner, yet another result quoted by Binz [10]
We may derive (6.28) from a slightly different viewpoint and also in the process
give a solution to a problem posed by Krafft and Schaefer [78]. Consider the
two binomial expansions
c-
(6.30)
127
f"2 1 l
T
~ (~: 11
) (1
+ 1 so that
+ xF
(l+rf+l - e-~f+l
JI+X
where fv l is the least integer not smaller than v. From (6.28) let b
c = x/4, hence
(l+~f+l
(6.32)
- e-rf+1
vI +x
f"2 1 l
2- n
L.
;.:
j=O
11 ) (1
1 and
(6.33)
+ x)j .
2m + 1 so that
(~r = T(2m+l)?; ( ~~:: ) (1 + xr.
?;
[(2m+l)j2]
(2m
+r 1 - r )
[(2m+l)/2]
{; (2m +r1- r )
=
(~r
T(,,"+l)
~,"
{t, (;7:1
2 ) (
~)}
(~;:12) (~)
J=r
2 (m-r)+1
(2m+rl-r).
Now, to solve the problem ofKrafft and Schaefer, add (6.29) and (6.30) so timt
fnt1l
T
j=O
(n;l) (l+x)j
(1 + ~)71+1 + (1- ~) n+l
J
(6.34)
128
am
f ( ;~ ) (1 +
m-I
( 2m)
j~O 2j + 1 (1 + x)
and !im am
m->oo
)=0
x)j
= n + 1 is
=Yl+X
j
{I
+ (I-v1+X)2m
}
l+v'l+x
(I-v1+X)
m
l+v'l+x
2
( n-k)
k
get
t(
K n (z, z, ... , z) =
~r
) zn-2r
r=O
00
L:
r=l
l~:T'
lxi< 1, was presented by Lambert circa 1771, and has been studied extensively.
A closed form representation of the Lambert series may be useful, because of
its possible importance in prime number theory. For the Fibonacci sequence
(6.26) with b c 1 it may be shown, see Knopp [74], that
= =
= 0,
io
= 0, h = 1
gn+1 - bg n - gn-1
= 0,
go
= 2,
129
= b.
gl
where
2a
o:n -
= 0:. and
= b + Vb 2 + 4,
9n = a
= b-
+ ,
Vb 2 + 4
and
l
:-n=l gkngk(n+m)
00
1
(a - ) gkm
[2L ( 2k ) - 4L ( 8k ) -
_1
1
.!km n=l akngkn
1'
7.
Functional Forms
The following lemmas are functional forms of (6.27).
l: (_c)k 1n-2k = O.
Tl
(6.35)
k=1
k==l
130
= -Cfn-2 + c2fn-4
+ ... + c3fn-6
- c3 fn-6
- c2 fn-4
+ Cfn-2
= 0
6.2
bm fn =
f (~ )
J
j=O
(t, (7)
bfn+m - bc (
~ O.
= 0,
(6.36)
fn
fn, for
(-c); fn+m-2 j )
7)
+b (_c)m-l ( m
fn+m-2
~1)
+ bc2 (
fn-m+2
) fn+m-4
+ ...
+ b (_c)m fn-m
= fn+m+l
- Cfn+m-l -
+ c2 (
+ ... + (_c)m-l
{fn+m-l - Cfn+m-3}
)
{fn+m-3 - Cfn+m-5}
( m
~1)
+ (_c)m
{fn-m+3 - cfn-m+d
) {fn-m+l - cfn-m-d,
( :
f (m:j=O
1 ) (-c)j fn+m+1-2j
= m+l
~
1=0
+ (_c)m+l
m;
1 )
fn-m-l
(-c)j fn+m+1-2j
131
= '""
L.-t (-c) k fu+v-2k
fufv
k=O
fufv
2.:=
u-2
1l-1
1l-2
- Cfu-2+v-2k)
k=O
k=O
u-l
1l-1
2.:=
k=O
all of the second and third terms are annihilated except for the k = u - 1 tenn,
in which case
u-l
k=O
k=O
= u,
f~
2. for v = u
+ 1,
1. for v
LEMMA
=L
k=O
(_c)'L-k
f1lfu+l =
u
I:
k=O
hk and
(-c)
u-k
f1+2k
6.4
[m/2]
n=O
j=1
= rn m
(-c)j (rn
n, v
(m-n
.~ fnfm-n =.~ {;
+1 -
= n, hence
(_c)k fm-2k
2j) fm-2j.
132
m{m
m (_c)k im-2k }
t; (_c)k im-2k - k~I
t.
{Im +
Linim-n
n=O
(m + 1) im - L
L
(_c)k im-2k
n=Ok=n+I
m
(m + 1) im - L j (-c)j im-2j
j=I
[m/2]
n=O
j=I
L inim-n = (m + 1) im + L
(-c)j (m + 1 - 2j)jm-2j
6.5
= Imin-rn + C/rn-lln-m-l
From Lemma 6.3 put u = m, v = n - m hence
In
Proof
=L
imin-m
(_c)k in-2k
(6.37)
k=O
cim-dn-m-I
(_c)k in-2-2k
k=O
m-I
(_c)k+1 in-2-2k'
(6.38)
k=O
6.6
bm
= m-I
L(
j=O
m -:- 1
(-c)j im-2j'
133
h -_
m-l(
L
m - 1
j=O
.and since
LEMMA
6.7
) (-c)
frn-2)
/2
= f n-rn f n+rn + ()n+I-rn
-c
rn-I
=n =
fn-rnfn+rn
m, v
= n + m giving
n-rn
L (_c)k hn-2k.
k=O
fn-mJn+rn =
(_c)n-k hkl
k=m
f~ =
L (-ct- k hk.
k=O
f~ - fn-mJn+m =
k=O
k=m
L (-cr- k hk - L (-ct- k hk
rn-I
(_c)"-k hk
k=O
(_cr+ 1- m
m-I
(_c)m-l-k hk;
k=O
identifying the last sum as f~-l we have the result and hence the proof ofthe
lemma. For m = 1, and c = 1, we have Cassini's identity, namely
f~ = fn-dn+l
+ (-Ir
o
134
LEMMA
6.8
~ ~ ( p ~ r ) c"
{r, ( j
P
Chapter 7
SUMS AND PRODUCTS OF BINOMIAL TYPE
Abstract
1.
Introduction
136
2.
Technique
t(
J=o
. ) (_c)R-j
t ( ~ ) (_b)R-r in+r
t (r
j ) (_b)j-r in+r-(R-j)a =
r=O
r=O
W n;
~aR
n< aR
W n;
(7.1)
with a and R integer, band c real and W n is a discrete forcing term. A method of
analyzing the solution of system (7.1) is by the use of Z transform techniques.
Let W n = 0, i R-l = 1 and all other initial conditions of the system (7.1)
be zero. If we now take the Z transform of (7.1), utilize the two Z transform
properties
and
where Un -
R
.
R-'
R ( j ) (z - b)J (-cz- a) J } = Z.
F(z) { ~
(7.2)
From (7.2)
F (z) =
z
(z - b - cra)R
F(Z)=~
+r _
r
zaR+l
.
(za+l - bz a _ c)R
1) (z_b)R+r
zl-ar
er
(7.3)
(7.4)
in (a, b, c, R)
[n+l-R]
= in
I:
r=O
(R +; - 1 )
Rn+-/~ 1 )
Gf bn-ar-R+l
(7.5)
137
where [xl represents the integer part of x. The inverse Z transfonn of(7.3) may
also be expressed as
where C is a smooth Jordan curve enclosing the singularities of(7.3) and Resj
is the residue ofthe poles of(7.3). The residue, Resj, of(7.6) depend on the
zeros ofthe characteristic function in (7.3), namely
g(z) = za+l - bz a - c.
(7.7)
Now, g (z) has a + 1 distinct zeros ~j, j = 0, 1,2,3, ... ,a, for
c =I- _aa (_b_)a+l
a+l
therefore the singularities in (7.3) are all poles of order R. We may now write
(7.6) as
(7.8)
where
p,lQR
,11
(~.)
J
.
= hm
z---(.j
[d-dzl1l1 { (z -
~.) RF(Z)}]
J
(7.9)
for each j = 0,1,2,3, ... , a, and F (z) is givcn by (7.3). Combining the
expressions in (7.5) and (7.8) we have that
and putting n = n* (a
an alternate fonn
+ 1) + R
138
The case of distinct zeros has been examined in Chapter 6, hence we shall
briefty investigate the case of multiple zeros. In doing so we shall recover a
result given by Wilf [111], and describe a generalization of this result. The
wi pairs method ofWilf and Zeilberger will be employed to certify particular
instances of identities that we shall generate.
3.
Multiple Zeros
r+
When the characteristic function (7.7) has double (repeated) zeros, which
will be the case for c = _aa (a!l
this case we may write from (7.11)
1
,
~(R+r-l) (n(a+l)+R-l-ar) (
R+r-l
-aa
(a+lt+ 1
)r
ReSj (F;Z))
=b- n(a+l)tz n
(7.12)
)=0
where the Resj must take into account the repeated zeros of (7.7). For a = 1,
c = - (b/2)2 and, from (7.3),
zR+l
F(z)---~
- (z - b/2)2R
which has poles of order 2R at z = b/2. Utilizing (7.8), (7.9) and (7.12) we
have
2n
L ( : ) ( ~~ ~ ~ ~ ~ ) .
/1=0
t, enr~
r )
Q Cn~~ 1)'
sin'
(7.14)
+ R) (2n + 2R + 1) in (R) = O.
(7.15)
139
rr
(7.16)
(1+j)(1+2j)
j==O
.n
-fI
sin 2
j==l
(~)
2n + 1
cot 2
k=l
(~)
.
2n + 1
+ 1) (2n + 3) f~ (1)
= O.
(7.19)
(7.20)
~r
r=l
2n - r )
r
n -1
. ) (2) - 1)
J=l
rr
2-2n n -1 (1 + j) (3 + 2j)
.
j(2j-l)
J=l
rr
j=l
1)
+
7rj
sin 2 (2n
7rk
tcot 2 (2n
k=l
1)'
(7.21)
140
r) (~
2n; 1 )
+(
t, (r;
1 ) ( 2n:: 1-
= 2- 2n { ( 2n: 1 )
= 2-2n
fi
j=O
+2(
= 2 we obtain
(7.22)
2n (1 )}
(2 + j) (5 + 2j)
(1 + j)(2j + 1)'
Writing
t, ( ~
r
(~1 ) r
1 ) ( 2n r+: 1- r )
~
( 2n r- r )
=f::o(2n+l-r)
(-1)
4
r 2n (2n + 1)2 +
ii:
sin 2
)=1
(2n1l'~ 1) t
(2n1l'~ 1)'
cot 2
k=1
(7.23)
2n
fi
j=O
+ j) (5 + 2j)
(1 + j)(2j + 1)
(2
ii:
sin2
(~)
2n + 1
j=1
cot 2
k=1
(~)
2n + 1
2n (2n
+ 1)2
(2
n
rr
1)2 = 10 n - 1 (2 + j) (5 + 2j) _
j=1 (1 + j) (2j + 1)
n (4n 2
3
1) =
rr
n -1
j=1
(1
rr
n -1
(1
j=1
+ j) (3 + 2j).
j (2j - 1)
+ j)(3 + 2j)
. (2 . - 1)
J J
(7.24)
141
+ 1) (2n + 1) (2n + 3)
~--~--~~--~-10
~r
(~1
( 2n r- r )
rr
n -1
(2 + j) (5 + 2j)
j=1(1+j)(2j+1)
= _ 2- 2n n
(~n2 -
1) .
_2-2n
+ 2)
15
rr
n-1 ( . + 1) (2 . + 3) (2 '2 - . - 5)
J
J
J
J
.
.
j (2j - 1) (2]2 - 5j - 2)
)=1
2) In+1
+ (n
and hence
n (8n 4 - 20n 3 - lOn 2 + 5n + 2)
15
+ 1) (2n + 3) (2n 2 -
-11
(j
j=l
n -
5) In
= 0
+ 1)(2j + 3) (2j2 -
j - 5)
j(2j-1)(2]2-5j-2)
Similarly
=
=
2- 2n n (16n 6
rr
_2- 2n n -1 (j
112n 5
+ 1) (2j + 3) (4j4
)=1
21n 2 - 28n -
-12j3 - 31j2
2)
+ 18j + 35)
j(2j-1)(4j4_28]3+29]2+28j+2)
'
and
rr
n -1
)=1
(j
+ 18j + 35)
28]3 + 29]2 + 28j + 2)
112n 5 + 112n 4 + 140n 3 - 21n 2 12j3 - 31P
28n --
2)
=~----------------------------~
105
142
In general
2~n ~ rm (
2n r- r )
(~1
2r(2n+1-r)(4r-5-6n)
d
(n,r, 1) = (2n+3)(2n+1-2r)(n+1-r) an
)
V n, r, 4 =
4.
2r(2n+4-r)(4nr+lOr-6n2-23n-14)
(n + 4) (n + 9) ( 2n + 1 - 2r ) (n + 1 - r ) '
More Sums
Since (7.7) has at most three real zeros we may obtain further results as
follows. Consider multiple zeros of (7.7) for a = 2 and c = -4 (b/3)3 such
that 9 (z)
such that
(z - ~)2 (z
+ ~)
F(z)=
MIQ2R.,
c:) =,~~
z2R+l
((z-~)2(z+i))
[::, {
(z - ~t F;Z)}].
/-t
VIPR (
R'
= 0, 1,2, ... , 2R -
1,
= 0, 1, 2, ... , R -
143
L..J
( R
+r
- 1 ) ( 3n
1'=0
-L L
_
2R-l
]=0 /1=0
+L
a
Q2R,/1
+R
- 1 - 2r ) (_4)1' b3n
R +r - 1
27
(2b)(3n+R-1) (2b)3n+/1
3
2R - 1 - P
3
L PR,v (_~)
3
R-l
j=O v=O
For R
(_~)3n+/1
+R _ 1 )
R - 1 - 11
3n
. (7.26)
t, (
3n; 2r ) ( ; :
3-(3n+2) {2 3n + 1 (9n
(7.27)
+ 4) + (-lt}
and
(7.28)
:+
(3n
+ 1)
3n
3F2 [
1)
3;;4 +
+2
+ 23n+3 (3n + 1) +
1~C~::~3~~;:n
1 1]
(-~t
(3n
+ 131) }
From 'Hyper', in 'Mathematica' a recurrence relation for (7.27) and (7.28) is,
respectively
729 (3n
+ 4) In+2 (1)
- 27 (21n
- 8 (3n
+ 7) In (1)
= 0,
23
10 (1) = 1, h (1) = 27
and
144
5.
We can now consider the system (7.1) with non zero forcing terms. Consider
a forcing term of the form, (other forms mayaIso be taken).
Wn
= (
n
) bn +1- R - m
m+R-l
with all initial conditions zero and m a positive integer, again the results of the
previous section are applicable. For the purpose of demonstration let a = 1
and c = - (b/2)2 so that from (7.12)
~ ( R +r -
~
r=O
1) (
1- (-1)
2n + R + m r )
R+m+r-l
_ -2n 2R-l
- b
Q2R,/1
/1=0
(~)
2
v=o
/-L.'Q 2R,1J,
n
) (~) 2n-R+m+/1
2R - 1 - /-L
2
+~
~ (b)
~ m,v
where
(~)
lim [.!!!:...
2 = z---~
dzlJ, { (z _zRb)R }]
and
t.(
ZV
zR 2R }].
(z _
~)
2n:;2- r )
(~lr
4(2n-l)+T 2n (2n+5)
rr
n-l
j=l
rr
n-l
2j + 1 + 7.2- 2n
7 + 2j.
2j - 1
j=l 5 + 2j
145
n-l
= (_1)m2- 2n (2n + 2m + 1) + 4 L
D:jn j
j=O
and for m = 0 reduces to identity (7.14); moreover a recurrence for the left
hand side is
4 (2n
+ 2m + 1) In+l
+ 2m + 3) In
4m (6n + 2m + 5)
- (2n
(2n
+ m + 2)
( 2n
+ m + 2)
m
f, = 1
,0
Many ofthe Binomial coefficient identities ofthis chapter and some in Chapter 1 may be proved by using the Paule and Schorn [84] algorithms in Mathematica or by using the Maple software developed by Koepf [77].
In the next chapter we shall develop new identities for the infinite representation ofthe sum (7.5).
Chapter 8
Abstract
1.
Introduction
2.
F(z)
with
(z - b - bz- a)
9 (z)
= za+l
zaIHl
---;0::
(g (z))R
- bz a - b.
147
(8.2)
148
(a + 1t+ 1
(abt
< 1,
(8.3)
~
r=O
(r
+R -
1) (
n - ar ) bn-ar-R+l
r+R-l
~ ~ Q R,/1 (Co)
(
<"J
~ ~
j=O /1=0
n
)
R - 1 - f.L
i"~-R+1+/1.
<"J
(8.4)
Ifwe define the dominant zero ~o of 9 (z) in (8.2) as the one with the greatest
modulus, we conjecture (and shortly prove) that
R_
~ _ I' ) (~-R+1+"
(8.5)
where
J.L!QR,/1 (~o)
.
=z-+~o
hm
(z)}] ,
[d/1
F -d { (z - ~o) R z/1
Z
J.L
= 0,1,2, ... ,R -
I,
(8.6)
Tr(R,n,a,b)
= (r+~-l) (r:~a~l
)bn-ar-R+l.
(8.7)
Tr (R, n,a, b)
r=O
00
r=[n+2-RJ
a+l
Tr (R, n, a, b)
= L QR,/1 (~o)
R-l
R _ ~ _ f.L
~~-R+1+/1
/1=0
and therefore
The conjecture (8.5) will be proved later, firstly we give a recurrence for the
evaluation of (8.6) and a recurrence for (8.5).
2.1
149
Recurrences
A recurrence relation for the evaluation of QR,1l (~o) in (8.6) is now given.
LEMMA
(J-l
8.1
+ 1) QR,Il+1 (~o)
=R(~0-b)~(-1)k(k+1) (a+k+1)Q.
c2
<"0
J-l
k=O
= 0, 1,2, ... ,R -
where
ck
<"0
1, with QR,O
a_ 1
R+l,ll- k
(C)
(8.8)
<,,0 ,
(~o) = ((a+l~~o-ab) R .
h (z) = (a (z - ~o)
+ z) 9 (z) za+l
and 9' (z) = za-l ((a + 1) z - ab). It can be seen that h (~o) =
and
h' (~o) = 0, hence expanding h (z) in a Taylor series about z = ~o we find
that
Wenow have
150
lim [dd: B (z )]
z
z-->o
= lim
z-->o
[f:
.
J=2
= (-l)k({o-b)(k+l)!
{~+2
(a+k-l)
a- 1
(J-L
+ l)!QR,J.I+l ((0)
= R
~( ~ )
(-1)k({0-b)(k+1)! (a+k-1)
(~+2
a- 1
and upon simplification we obtain (8.8) hence, the lemma 8.1 is proved.
We can now list some values of QR,I' ((0) in Table 8.1, where for ease,
= (a + 1) (0 -
ab.
Using the values OfQR,J.I ({o) in Table 8.1, some closed form express ions of
(8.5) are listed in Table 8.2.
The following lemma gives a recurrence relation for the left hand side of the
identity (8.5).
LEMMA
8.2 Let
I:T (R,n,a,b)
00
SR =
(8.9)
r=O
(a
+ 1) bdb SR -
abRSR+l - (n
+1-
R) SR
0/ (8.9) is
= O.
(8.10)
151
J1
J1!QR,/, (~o)
AR
~Ii
R~R-2
~2A~+2 [A 2 (3aR - a -
8) - 2bA (3aR
a8~~i:33 [A 3 {a 2R (R -
4R)
~~+1 [.~]
~~+1 [
-;r2 n+
~~+J
A3
[(
n)
2
~~+J
A4
3na(A-b)
2A
+ 12}
a(a+1)(~o-b)]
A
a(2(a-l)A 2 -bA(5a-2)+3ab 2 )]
2A2
n)
2a (A - b) ( n)
3 +
A
2
na ((l1a - 8) A 2 - 2bA (13a - 4)
+
6A2
+ 12~:J (A 3 (12a 2 - 30a + 12)
[(
_bA 2 (52a 2
rahte 8.2.
+ 3 (R + 1) ab 2)
4)
rahte 8.1.
+ 2a (1 -
1)
+a -
70a
+ 15ab2)
+ 12) + b2 Aa (70a
152
and
t, ( ~
SR+l
R ) (
~~~ )
(a+1)~
bR
bn - ar - R
~rTr(R,n,a,b)+
(n+1-R)
bR
SR
(a+ l)b
abR [ -
(a+1)~
(n+1-R)
bR ~ rTr (R, n, a, b) +
bR
SR
-(n+1-R)SR=0,
which is the right hand side of (8.10) and the proof is complete.
2.2
Proof of Conjecture
The proof ofthe conjecture (8.5) will involve an induction argument on the
parameter R. For the basis, R = 1, (8.5) was proved in Chapter 6.
Firstly the following recurrance relation for QR,JL (~o) is given
LEMMA
8.3
abRAQR+l,JL
(~o)
= ab (R -
+ (J-L
- 1) AQR,JL-l (~o)
153
where A = (a
obtain
-R
+ b/L1QR+l,JL (~o)
~o) 2
h (z) = (z -
(~Og(Z)Z-a-A(Z-~o))}].
~o),
f(
-1)j ( a + ~
- 1 ) (b -
j=2
(8.14)
~o) ~~ - j (z - ~o)j - 2
d
/Ll db QR,JL (~o)
-R
- - hm
bA z-->eo
(8.15)
where
BJ.-
h(z)
(z -
~o)
(8.16)
= I>-l)j ( a+ j -1 ) (b j=2
J
~o)~tj (z - ~O)j-2.
-R
~~o
t (~ )
(/L - k)!QR+l,JL-k
k=O
z-->Eu
(8.17)
b d
t;rdQR+l,JL (~o)
~o) ~ (_l)k
QR,JL (~o)
( a + k + 1 ) t:-kQ
A L k+ 2
k=O
<"0
R+l,JL-k
(t: )
<,,0 .
(8.18)
154
Now (8.18) and (8.8) suggest that the Q (~o) tenns may be related by an expression of the fonn
Qn+l,J.!+1 (~o) =
Cl
d
db Qn,J.! (~o)
+ C2/-LQn,J.! (~o)
+ C3 (R - /-L -
for /-L = 0,1,2, ... , R - 1. The constants Cl, C2, and C3 can be evaluated by
forming three simultaneous equations and using the Q (~o) values, given in
Table 8.1, we may write (8.19) as
d
db Qn,J.! (~o)
~o
= bA (A + ab)
Now, since
(8.20)
Ab d
db Qn,J.! (~o)
= ~5
+1-
R) Sn} ,
(8.21)
Sums
0/ Binomial Variation
155
R+l
t-n-R+l+/1
P,
<"0
~ (n + 1 - R) ; , (
_
1
- abRA
R-l (
n
R - 1 - p,
~_ ~ ) ~~-R+l+"QR," (~o) }
R_
)
n-R+1+/L
~o
d
(A + ab) ~o d~o QR,/L
R) QR,/L
(~o)
(~o) }
+~(
~
/L=O
n
) C-R+l+/L
R-I-p,
0
~ ) ~ ~g-RQR,o (~o)
1
R-l
+ abRA { ~ ab (R - p,)
71
R _ p,
n-R+l+p
~o
QR,/L
(~o)
156
= (
1) AQR,/1-1
(~O)}
= ( ~ ) ~3-RQR+l,O (~o)
+ ab~ ~ (
1) AQR,/1-1
(~o) } .
Using the relationship (8.11) of Lemma 8.3 reduces the previous line to
R
to
The degenerate case, for a = 0, of the identity (8.5) can be noted. Firstly,
= 2b and J-L!Q R,/1 (2b) = 1 for J-L = 0 and zero otherwise, hence (8.5) reduces
nI=R (
n+;
R )
= 2n +1- R ,
r=O
which is not Gosper summable, as defined by Petkovsek et al. [85], however by
the WZ pairs method a rational certificate function is V (n, r) = 2(r-n~2+R).
It is ofpassing interest only, to note that the dominant zero ~o, may be set to
unity, in which case the closed forms in Table 8.2 of the identity (8.5) would
be simplified. Ifwe put, from (8.2), 9u (z) = za+l - bz a - 1 + b, then for
1 < b < (1 + ~), 9u (z) has a unit distinct dominant zero ~o. From (8.5) for
157
0:
E ~
~ ( -1 r ( (a + 1) rr +
~(-lr (r+1)
L..J
1'=0
r'
and putting Au
1 )
0: -
((a+l)1'+O:)
+1
l'
(~a~
1'=0
l'
+2
r
) ( (a + 1) l'
l'
+ 0: + 1
+2
bQ +2
!~/
ba+l
{
+ 1)( 1 -
a (a
Al;
~ (-Ir
+ 1, we have
(~)r
= lP+ 1
L..J
a (1 - b)
Au
Cl'
b) }
'
(~)1'
ba + 1
a(a+l)
3aa(a+l)(1-b)
}
-2- 2A u
a(2(a-l)Ar,-(5a-2)bAu+3ab2)
2Ab
"fi3
u
and in general all other R value identities can be obtained from (8.5).
2.3
Hypergeometrie Functions
T1'+l (R, n, a, b) =
T1'(R,n,a,b)
fr (1' + j+~~~-l
)=0
R.
(1'+l)rt(r+~)s
b)
a,
J=O
is a rational function in l' and therefore the series SR (n, a, b) of (8.9) may be
expressed as a generalized hypergeometric function
at
R+a-n-l
1 '-a:t:l, ... ,
a+l
710 a +1F.a [ -a:t:l'n -n
2-n
a-l-n
-> --a, -a-"'" - , - , -
IS (a b) ]
,
= To L..J
k=O
(_)1) (l-n)
a kak
(2-n)
a
'"
(a-l-n)
kak
sk(a,b)
I
k.
(8.22)
158
(a, b)
=-
(a
+ 1t+ 1
(ab t
00
= To
1+
L (;!~~
k 2
ft (R +
0: - 1 +
j=O k-1
'
TI
k=l
j)]
(8.23)
(o:+j)
j=o
J
1
To
B (R+n n-R)
2
'
(1 - t)(n-R-2)/2 t(R+n-2)/2 (1 -
(1 b) t)(1-R-n)/2 dt
t=o
which is valid for 18 (1, b) I < 1 and B (x, y) is the Beta function. The difference
in the two top terms of the hypergeometrie function (8.23) is one-half, hence
there exists a quadratic transformation, see [1], connected with the Legendre
function, Pt:. Using the identity on page 562 of Abramowitz and Stegun [1] we
may write (8.23) as
where 8 (1, b) E (-00,0) and r (x) is the Gamma function. We may write the
identity (8.5) as
~(
r=O
+r -
1 ) (
=L
R-1
QR,JJ
-0: - r
R+r-1
(~o)
JJ=O
) b- n - r - R +1
--t_
=TOr(0:){~}(1-n)/2{
21-n
fL
~3-R+l+JJ
~}-R/2
1+ b
1-n{{ 1 + ~}-1/2}
b
PR- 1
159
for b > 4, 2~o = b+ Vb 2 + 4b and QR,/1 (~o) is defined by (8.6). Other specific
cases of (8.23) are as follows.
(i). For b = 4, S (1,4) = -1, a = 3/2, and from page 557 of Abramowitz
and Stegun [1]
2R+1 2R+31
4
"2
= T o2-(2R+l)/4r
2R-1
(~) Vii {
-4-
R=l
R=2
R=3
- (2 + v'2)
2 (4 3) {I + v'2)1/2
3{2+v'2)
2 (45) v'2 (1 + v'2)1/2
8 (1 + v'2) 1/2
R=4
-5
4 7v'2 (1
+ v'2) 1/2
160
(ii) . For b = 4, S (1,4) = -1, a = 1/2, and from page 557 of Abramowitz
and Stegun [1] we have
~
r=O
( R + r - 1 ) ( -1/2 - r ) 41/ 2 r
R+r-l
= To 2Ft
=;
1- 1]
2R4-1 2R4+1
QR,p (eo) ( R
r- R
T01T2(I-2R)/2
R=2
R=3
R=4
(1 + v'2) 1/2 (1 + v'2) 1/2 (v'2 - 2) 3v'2 (1 + v'2)1/2 (v'2 - 2) 5v'2 (1 + v'2// 2 .
4
43
2 (45)
47
2.4
Forcing Terms
= (
n
) bn +1- R - m
m+R -1
'
~
~
r=O
( r
+R- 1 ) (
r
n - ar
r+R+m-l
~ QR,/1 (~o) (
R_
) bn-ar-R-m+l
~ _ JL ) ~~-R+l+/1
/1=0
+ m-I
" P (b)
~ m,v
v=o
n
m-l- v
where
JL!QR,/1 (~o)
.
=z-+~o
hm
z-+b
[d/1
F (z)
-d { (z - ~o) R - }] ,
z/1
Z
bn - m +1+v
Sums
0/Binomia/ Variation
161
and
F (z) =
where
~o
zaIHI
(z - b)m (za+l - bz a - b)
~(1';1) (:~~)
= b2a
(b:r
n+l-m-R ( 1'+R-l ) (
l'
~l(_I)R(
6
)
n
r+R+m-l
R+V-l) (
1/
v=o
+
- b) (( a + 1) ~o - ab)
~
(-1)" (m+/-l-l
6
/-l
/),=0
2F]
[ R, R
~: ~n -
n
)
m-l-v
) (
R-l-/-l
) 2n-R+1+/1
11_ 1]
- (n+l)B(R+m,n+2-R-m)"
which for specific values of m and R may be certified by the WZ pairs method
of Wilf and Zeilberger.
2.5
f (a, J;) =
~(
(1+lfFr=
= a
~:: )
r; 1 ) (
1-4x
2H [
{a-l+
at ~ Qt
l
(8.24)
xT
4x] ,
Jr=4X}
162
f(a,-x) =
(1+4lli)
=
l-a
1 + 4x
(8.25)
{a - 1+
Jf+.4X}
= a 4F3
=
+
(8.26)
2r+ 1
2r
ili'
(1+vf4Xr-
1 - 4x
(1+4lli)
l-a
1+4x
{a
{a
-1
I16x 2]
1
+ VI -
4x
-1
J1+4x
}.
From (8.24)
~(
~
r=O
2r + 1 ) ( 4r + a ) x 2r + 1
2r
2r + 1 2r + 1
=
ax 4F3
2' 2'
-2-
I 16x2]
163
~
00
0:)
2r + 1 ) ( 4r +
2r
2r+l
x 2r + 1
(2r+1)3
=O:X6Fs
2'2'2'2'
I16x 2] .
0:
~(r+R-1)
~
r
r=O
(2r+R-l
r+R-l
)xr=
l j4X]
[q,Rt
1
(8.28)
[R21]
AR ,J'x j
'\'
1....
j=o
= ----::--.......,.,(1- 4x)R-l/2
and
~(r+R-3/2)(4Xr=
1
.
(1 _ 4x)R-l/2
(8.29)
From (8.28) and (8.29) adjust and then equate the coefficients of x j gives
( r
+R r
1 ) ( 2r + R - 1 )
r+R-l
[R 21]
" (r + R - 3(2 ~
j=O
AR,j
r - J
.
J ) 22 (r-j) A . (8.30)
R,)
= 6,
164
Table 8.3.
( r
R\j
12
20
30
30
90
20
42
210
140
56
420
560
70
10
72
756
1680
630
II
90
1260
4200
3150
252
12
110
1980
9240
11550
2772
13
132
2970
18480
34650
16632
924
14
156
4290
34320
90090
72072
12012
15
182
6006
60060
210210
252252
84084
3432
16
210
8190
100100
450450
756756
420420
51480
~5 )
~: 55 )
= ( r +r9 / 2 ) 22r
4.
165
3.
In this section we shall generalize the identity (8.5). We shall prove that a
generated infinite series may be represented in closed form that depend on k
dominant zeros of an associated polynomial characteristic function. Consider
the delay system
t (~ )
J
j=O
(_b)k- j in+j
(_b)k- j in+j =
n 2 ak
(8.31 )
t (~ )
j=O
+ Cin-ak = Wn;
TL< ak
W n;
with c + bk = and all initial conditions at rest, except for ik-l = 1. As in the
first section, we set W n = and take the Z transform of(8.31) such that
z
F (z) = -(z---b)-;-k---(-bz---a--"")k
(8.32)
In
(~. (k~;:~l
(8.33)
Cis a smooth Jordan curve enclosing the singularities of (8.32) and Resj,v is
the residue at the poles. The characteristic function
(8.34)
with restriction
(a + l)a+l
(abt
}k
<1
(8.35)
in =
j=O 1.1=0
(8.36)
166
.
hm
z-+<Ej,v
[ (z -
F(Z)]
~j,v)-Z
~j,v
(8.37)
[!.!.=lli]
(a+l)k
"
r=O
n - akr
kr+k-1
bn-akr-k+l
k-l a
-L
L
k-l
j=O v=o k (~j,v - b)
((a + 1) ~j,v ~n+l
J,V
ab)
(8.38)
Ifwe let ~j,o,j = 0, 1,2, ... , k - 1 be the k dominant zeros of(8.34) then we
have the following theorem.
3.1
The k Theorem
THEOREM
8.1 Let
J: (k
r
, n,
a,
b) = (
n - akr ) bn-akr-k+l
kr + k - 1
and
(8.39)
00
S(k,n,a,b)
= LTr(k,n,a,b)
(8.40)
r=O
=L
k-l
S(k,n,a,b)
~n+l
k 1 j,O
.
j=ok(~j,o-b) - ((a+1)~j,0-ab)
(8.41)
r=O
L
00
Tr(k,n,a,b)+
T
= [!!.Etll]
(a+1)k
Tr (k, n,a,b)
k-l
_"
~n+l
k 1
j,O
+ 1) ~j,O -
ab)
167
and hence
k-l
00
~n+l
L
Tr(k,n,a,b)=-LL
k-l},11
r=[n+a.k+l]
j=O 11=1 k (~j,1I - b)
((a + 1) ~j,1I - ab)
(a.+1)k
The characteristic function (8.34) may be expressed as the product of factors
such that
k-l
k-l
gk (z) =
(za+l - bz a - be27rijlk) =
qj (z).
(8.42)
j=O
j=O
rr
rr
za+l
F j (z) = -za-+-l---b-z-a.---b-e-27r- i-j I"-k
for j
a
(8.43)
+ 1 distinct zeros for each j, of which 0:j ,0 shall indicate the dominant zero,
the one with the largest modulus, which may be complex. All the singularities
in (8.43) are simple and therefore for each j, F j (z) has simple poles. Utilizing
the result (6.10), from (8.43) and evaluating its residue, we may write
r=O
e27rijrlk ( n - ar ) bn - ar =
l'
0:j,t 1
(a + 1) O:j,O - ab
(8.44)
for each j = 0,1,2, ... , k - 1, and note that (8.44) may in fact be a complex
number. Summing (8.44) over j gives
k-l 00
L L e27rijr,/k
j=O r=O
n - ar
k-l
bn - ar = L
(171+1
J,O
(a
+
1)
j = o }0:' 0
l'
ab
Rescaling the left hand side by l' = (1'* + 1) k, ( and then replacing 1'* by
results in , after changing the order of summation
l' )
00
k-l
(
)
k-l
71+1
"" "" e 27rij (r+l)
n - ak (1' + 1) b71 - ak (r+l) = ""
O:j,O
r~lf::o
k(r+1)
~(a+1)O:j,0-ab'
~( n~
r=O
0:71+1
-- L
j,O
j=O k ((a + 1)
O:j,O -
ab)
- bn
(8.45)
168
'"
k)
m - a r
kr
+k
b-akr
b-mam+ak+l
k-l
= '"
bak
(8.46)
( m
r=O
+1-
r=O
~
(
~
r=O
(8.48)
where Al< = (a + 1) aj,O - ab. From the characteristic function qj (z) in (8.42)
aj,o (aj,o - b) - be21rijJk = 0 we may write (8.48) as
I:
j=O
b-m+k-1a}:o+1
kAu (aj,o - b)k-l
~
~
r=O
m - akr ) bm-akr-k+l
kr+k-1
k-l
I: k (aj,o - b) -
am+1
k 1 j,O
j=O
((a
+ 1) aj,O -
ab)
(8.49)
Since the dominant zeros aj,o,j = 0,1,2, ... , k - 1 of qj (z) are the same as
the dominant zeros ~j,O of 9k (z) in (8.34) then upon renaming m as n in (8.49)
Theorem 8.1 is proved since (8.49) and (8.41) are identical. Note that putting
k = 1 in (8.41) yields the result (8.5) for R = 1.
0
4.
169
Zeros
Some properties ofthe zeros ofthe characteristic functions (8.2);
g(z) = za+l - bz a - b
and (8.34);
gk (z)
= (za (z -
b))k - bk
~o > bfor b > 0 and /~o / > Ia~1 Ifor b < 0 and the restriction (8.3),' namely
I < 1.
I(a+l)a+1
(aW
(ii) Thefunction (8.34) has at least one and at mostfour real zeros.
Proof (i). The characteristic function (8.2) with restriction (8.3) has a + 1
distinct zeros, for the derivative of 9 (z) cannot vanish coincidentally with
9 (z) . The fact that a related function to (8.2) has distinct zeros appears to have
been reported first by Bailey [5). Let G (z) = za (z - b), hence G (z) = b, and
the tuming point ofG (z), away from the origin occurs at z = a~l' The graphs
of G (z), for b > 0 and b < 0 indicate therefore that (8.2) has at least one and
at most two real zeros. In both cases of b > 0 and b < 0 it will be shown in the
next theorem that the dominant zero, ~o, the one with the greatest modulus, of
(8.2) is always real, such that (0 > b for b > 0 and /(0 / > Ia~l I for b < 0 and
the restrietion (8.3) with areal.
(ii). In a similar fashion it may be seen that 9k (z) has at least one and at
most four real zeros. Let Gk (z) = (za (z - b))k , hence Gk (z) = bk and the
tuming point of G k (z) away from the origin occurs at z = a~1 . The graph of
G k (z) for b > 0 ( the case of b < 0 follows in a similar fashion ), indicates that
gk (z) has at least one and at most four real zeros.
0
THEOREM
be21rij/k = qj (z)
The study of the zeros of (8.42), (8.34) and (8.2) is important in the area
of queuing theory, and severeal papers have been devoted to this study, see far
170
example, Chaudhry, Harris and Marchal [26] and Zhao [115]. Their studies
have concentrated, amongst other things, on robustness of methods for locating
zeros inside a unit circle. The location of dominant zeros of (8.2), (8.34) and
(8.42) is of prime importance.
Proof LetA(z) = -bza.ThenA (z) hasazerosintheregionCand IA (z)1 :::;
( ab )a . Now
b a+l
IA (z)1
hence
and
which is satisfied since (8.3) applies, therefore the theorem is proved. Also from
(8.42), letting j = 0 gives the characteristic function (8.2). Theorem 8.2 now
follows since at least one zero of (8.2) must be real, it is evident that ~o > b for
b > 0 and I~o I > a~ll for b < O. Note that the restriction (8.3) is imperitive
for Theorem 8.3 to apply. If for example a = 1, b = 1/2 and k = 1 which
indicates that (8.3) is not satisfied, then qo (z) = z2 - z/2 - 1/2 gives the two
zeros as z = {-1/2, 1}, neither ofwhich are in the region C : Izl :::;
0
i.
THEOREM
C :
Izl :::;
Proof Let B (z) = (qj (z))k = (za+1 - bz a - be27rij/k)k . Utilizing Theorem 8.3, B (z) has therefore ak zeros in the region C and therefore k zeros
Sums
0/Binomial Variation
171
S{C~lr+b('~J+br
=
(a ~ 1) (2:: /) + 1
bk
(8.50)
Now
I(G (z))k - bk -
= 0,1,2, ... , k
= be 2 1l'ij / k, such that
for every j
Cj
(za+l - bz a - be 21l'ij/k
- 1, and G (z)
fl
?; ( ~ )
k
= _bk -
S bk
c::n rl
(8.51 )
where M =
J9k (z) - B (z)J ::; JB (z)J and upon using (8.50) and (8.51) we have that
bk [1 + (1 + M)k - M k ] ::; bk
1::;
[(1 + M)k]
ab ) (2a + 1 ) }
{( a+1
a+1
a
Ia~ll
172
10.0100
1O.0051+0.0086i
9.9951+0.0087i
9.9900
9.995J-0.0087i
10.0051-0.0086i
0.4798-0.8944i
0.7697-0.6786i
0.9669-0.3668i
-0.5136+0.8375i
0.9669+0.3668i
0.7697+0.6786i
-0.9696
-0.9157-0.3231i
-0.7589-0.6127i
1.0372
-0.7589+0.6 J27i
-0.9157+0.323Ji
0.4798+O.8944i
0.1412+0.9933i
-0.2031 +0.9708i
-0.5136-0.8375i
-0.203 J-0.9708i
0.1412-0.9933i
The dominant zeros of qj (z) are listed in the first column and all have modulus
bigger than 7.5. These dominant zeros are exactly the same k dominant zeros
of (8.34). It appears that the zeros, aj (a, b) of function (8.42) can be related
for b > 0 and b < O. It may be shown that the following relationships hold.
(i). For all values of k and a even, aj (a, b) = -aj (a, -b),
(ii). For k odd and a odd, aj (a, b) f:. -aj (a, -b), and
(iii). For k even and a odd,
n,
-u.j+~
aj (a, b)
for j
< k2
=~
4.1
In the following numerical results, the dominant zeros ~j,O are evaluated from
gk (z) in (8.34). It mayaiso be noted that for k ~ 3 the dominant zeros occur
in complex conjugate pairs. The numerical results are given to four significant
digits.
k
10
-10
10
identity (8.41)
~j,O
= 9.8979
6,0 = 10.0981
~o,o
~o,o
= -8.8730
= -10.5329+.7826i
6,0 = -1O.5329-.7826i
~1,0
~o,o
10.0981
6,0
9.9511+.0883i
=
~2,0 = 9.9511-.0883i
299.9899
-30.0005
29.9998.
Sums
0/ Binomial Variation
173
[m-kk+lj
\;'
~
m
kr+ k-l
(8.52)
Le
mk - 1
k2
cosm
1fij (m+2)/k
)=0
7'
.)
["'-3]
~
~
r=O
( 4rm
+ 3 ) = 4~ (2
m _
2~+1 sm
. m7r)
4 .
Using the WZ pairs method of Wilf and Zeilberger a rational function proof
certificate, Vk (m, r) , for k = 1 and 2 of (8.52) is respectively
VI (m,r)
4.2
= 2 (r
r
- 1-
) and V2 (m,r) =
rn
(r - 1) (2r - 1)
(
)"
m 2r - 2 - rn
(k , TL, a, b) -_ ( k TL_ 1
bn - k + 1 '
(k,TL,a,b)
Tr (k, TL, a, b)
TI
.
)=0
ak-I
Tr+l
(r+l)
(r + j+k-n-l)
(a+l)k
k-2
jQo (r+~)jQo(r+2k-j-l)
TL,
a, b) of (8.40) may
k-n-l
k-n
k-n-l+(a+l)k-I
(a+l)k' (a+l)k''
(a+l)k
n I-n
ak-l-n 2k-l 2k-2
2k-I-(k-2)
- ak' -ak' ... ' ---ar-, -k-' --,;-, ... ,
k
where
s (a, b, k) =
(a
+ It+ 1 )
(ab)a
s(a,b,k)
S(a,b,k)]
174
~o,o
1
2x
= x-I
and therefore
00
""""
r=O
(4 ++
r
2r
(}
1
x2r = (} P.
4 3
0+1 -4-'
0:+2 0:+3
0:+4 / 16x 2]
-4-'
4 ' -4~ 0:+ 1 Q
2' 2 ' 2
e-o:
j,O
- j=O 2 (x~j,O
+ 1) (2x~j,0 -
1)'
Specifically with (} = 1,
f (~~! ~ )
r=O
x 2r
= 4 F3
~l2,1,
i, 1~~
2
/16X 2]
vI - 4x + VI + 4x - 2
+ 4x - (1 + 4x)) (VI - 4x - (1 - 4x) )
(vI
=~~==~------~~====~------~
1(1
VI -
= 4x
1)
VI + 4x
4x -
= ~
( + 1)
4r
2r
2r
which confirms the result obtained in Chapter 6. Again the identities may be
differentiated and integrated to produce more results.
5.
If we consider the system (8.31) with all initial conditions at rest and with a
forcing term of the form
Wn
=(
~1)
bn - m +1
for m integer and follow the procedure of the previous seetion we obtain the
identity
r=O
( r
+R r
1 ) (
n - akr
) bn-akr-kR-m+l
kr + Rk + m - 1
rn-I
= '~
" Pm,1I (b)
11=0
f;
R-Ik-I
+~
QR,I-'
n v
m -1-
(~j,o)
bn - m +1+11
R - 1 - I-L
n-R+l+1-' (8.53)
~j,O
175
z-+b
/1!QR,11 (~j,o)
.
=z-+~j,O
lIm
and
F(z) =
dz
[d-d
11
zl1
(z - ~j,o) R -(z)
- }]
Z
zakR+l
R'
(z - b) m ( (za (z _ b)) k - bk ) .
=
where
~o,o
(0: - a) ba - b2a
b2+o:e-o:
(~o,o
0,0
~(
~
r=O
+1
r
) ( 4r
2r
+4
+2
) x-
2r = :. ((_X )~ _(_X
)~).
+
2
X -
References
[I] Abramowitz, M., and Stegun, I. Handbook o{ Mathematicalfunctions. Dover Publications Inc. New York. 1970.
[2] Amdeberhan, T., and Zeilberger, D. Hypergeometric series acceleration via the WZ
method. Electronic Journal ofCombinatorics, Vol. 3,1996.
[3] Andrews, G. E. Applications ofbasic hypergeometricfunctions. S.I.A.M. Review, Vol.
16, pp.441-484, 1974.
[4] Bailey, D. H., Borwein, 1. M., Borwein, P. B., and Plouffe, S. The questfor Pi. The
Mathematicallntelligencer, Vol. 19, pp.50-56, 1997.
[5] Bailey, N. T. J. On Queueingprocesses with bulk service. Journal ofthe Royal Statistical
Society, 816, 1954, pp.80-87.
[6] Bailey, W. N. Generalised hypergeometrie series. Stechert-Hafner Publishing Co., New
York.1964.
[7] Bellman, R., and Cooke, K. L. Differential Di(ference Equations. Academic Press, New
York,1963.
[8] Berggren, L., Borwein, J. M., and Borwein, P. B. Pi: A source book. Springer-Verlag.
Berlin, 1997.
[9] Berndt, B. C. Ramanujan 's notebooks, Part 1. Springer-Verlag, New York, 1985.
[10] Binz, J. C. Closedform o{ two sums. Problem 1498 and solution by J. Rosenberg.
Mathematics Magazine, Vol. 70, 1997, pp.201-214.
[11] Biswas, S. Stochastic Processes in Demography and Applications. John Wiley and Sons,
New York, 1988.
[12] Bloom, D. M. Advanced problem 6652. American Mathematical Monthly, Vol. 98,
pp.272-273, 1991.
[13] Borwein, J. Three adventures: Symbolically discovered identitiesfor Zeta(4N+3) and
Iike matters. 9th International conference on: Formal Power Series and Algebraic Combinatorics. Vienna, July, 1997.
177
178
[14] Borwein, J. M., and Girgensohn, R. Evaluation 0/ TripIe Euler Sums. Sirnon Fraser
University. Departrnent ofMathernatics and Statistics, Technical Report, 1996.
[15] Borwein, J. M., and Borwein, P. B. Strange series and high precision/raud. Arnerican
Mathernatical Monthly, Vol. 99, pp.622-640, 1992.
[16] Brauer, F., and Zhien, M. Stability 0/ stage-struetured population models. Journal of
Mathernatical Analysis and Applications, Vol. 126, pp.301-315, 1987.
[17] Brayton, R. K. Bifureation 0/ periodie solutions in a non linear difJerenee-difJerential
equation o/neutral type. Quarterly of Applied Mathernatics, Vol. 24, pp.215-224, 1966.
[18] Brockrneyer, E., and Halstrorn. H. L. The life and works
1948.
0/ A. K.
Erlang. Copenhagen,
[19] Brornwich, T. J. I'a. Theory o/infinite series. Macrnillan and Cornpany Ltd. London,
1947.
[20] Bruwier, L. Sur /'equation /onetionelle
y(n)
(x)
1) e)
+ any (x + ne) = O.
179
REFERENCES
10
[48] Flajolet, P., and Salvy, B. Euler sums and contour integral representations. Experimental
Mathematics, Vol. 7, pp,15-35, 1998.
[49] Frappier, C. Representation Formulas Jor Entire Functions o[ Exponential Type and
Generalised Bernoulli Polynomials. 1. Austral. Math. Soc. (Series A), Vol. 64, pp. 307316,1998.
180
~f Plana.
[51] Frappier, C. Generalised Bernoulli Polynomials and Series. BuH. Austral. Math. Soc.,
Vol. 61, pp. 298-304, 2000.
[52] Gaspar, G., and Rahman, M. Basic hypergeometric series. Cambridge University Press,
Cambridge,1990.
[53] Gradshteyn, 1. S., and Ryzhik, I. M. Tabte
Press, New York, 1994.
~r integrals,
[54] Graham, R. L., Grotschel, M., and Lovasz, L. Handbook ~fcombinatorics, Vol. ll, North
Holland, Elsevier, New York, 1995.
[55] Graham, R. L., Knuth, D. E., and Patashnik, O. Concrete mathematics. Addison Wesley
Publishing Co. Massachusetts, 2nd. Ed., 1994.
[56] Greene, D. H., and Knuth, D. E. Mathematicsfor the analysis ofalgorithms. Birkhuser,
Boston, 3rd. Ed. 1990.
[57] Haigh, J. Not only e, but e t . The Mathematical Gazette, Vol. 78, pp.54-56, 1994.
[58] Haie, 1. K., and Ycrduyn Lunel, S. M.lntroduction to Functional Differential Equations.
Springer-Verlag, New York, 1993.
[59] Hall, P. Introduction to the Theory ofCoverage Processes. John Wiley and Sons, New
York,1988.
[60] Hansen, E. R. A table ~rseries and integrals. Prentice HaBlnc. Englewoodcliffs, N. J.,
1975.
[61] Hao, D. Y., and Brauer, F. Analysis of a characteristic equation. Journal of Integral
Equations and Applications, Vol. 3, pp.239-253, 1991.
[62] Hardy, G. H., Seshu Aiyar, P. V., and Wilson, B. M. (Editors). Collected Papers
Srinivasa Ramanujan. Chelsea Publishing Company, New York, reprinted 1962.
~r
[63] Hendel, R. J., and Cooke, C. K. Recursive properties oftrigonometric products. Applications ofFibonacci Numbers, Editors G. E. Bergum, A. N. Philippou and A. F. Horadam.
Kluwer Academic Publishers, Netherlands, Vol. 6, 1996, pp. 201- 214.
[64] Henrici, P. Applied and Computational Complex Analysis Volume 1. John Wiley and
Sons, New York. 1974.
[65] Henrici, P. An Algebraic Proof of the Lagrange-Brmann Formula. Journal of Mathematical Analysis and Applications, Vol. 8, pp.218-224, 1964.
[66] Holzager, R. Solution to Asymptotic Linearity. American Mathematical Monthly, Vol.
99, pp.878-880, 1992.
[67] Horadam, A. F. lacobsthal Representation numbers. The Fibonacci Quarterly, Vol. 34,
pp.40-53,1996.
REFERENCES
181
[68] Horadam, A. F. lacobsthal Polynomials. The Fibonacci Quarterly, Vol. 35, No. 2, pp. 137149,1997.
[69] Jensen, 1. L. W. V. Sur une identite d'Abel et sur d'autres formules analogues. Acta
Mathematica, Vol. 26, pp.307-313. 1902.
[70] Jolley, L. B. W. Summation ofseries. Dover Publications, New York, 1961.
[71] Jonassen, A. T., and Knuth, D. E. A trivial algorithm whose analysis isn 't. Journal of
computer and system sciences, Vo!. 16, pp.301-322, 1978.
[72] Jury, E. l. Theory and application of the Z trans(orm method. lohn Wiley and sons !ne.
New York, 1964.
[73] Keane, A. Slowing down from an energy distributed neutron soune. Nuclear Science
and Engineering, VoI. 10. pp.1l7-1l9, 1961.
[74] Knopp, K. TheOlY and Applicatiolls of Infinite Series. Blackie and Son, London, 2 nd
Edition, 1963.
[75] Koecher, M. Klassische elementaire analysis. Birkhuser, Boston, 1987.
[76] Koepf, W. Hypergeometric Summation. An Algorithmic Approach to Summation and
Special Function Identities. Vieweg Publications, Germany, 1998.
[77] Koepf, W. Power Series in Computer Algebra. 1. Symbolic Computation, VoI. 13. pp.
581-603,1992.
[78] Kram, 0., and Schaefer, M. American Mathematical Monthly, Vol. 104, p.871, Prob.
No.10625, 1997.
[79] Macfarlane, G. G. Summation ofinfinite series. Phi\. Mag., VoI. 40, pp.188, 1949.
[80] Majewicz, J. E. WZ-style Certification and Sister Celine 's Technique o(Abel-type Sums.
J. Difference Equations Appl., Vol. 2. pp. 55-65. 1996.
[81] Melham, R.S. A Generalization of a result of Andl'/?-leannin concerning summation ol
reciprocals. Portugaliae Mathematica, Vol. 57. pp. 45-47, 2000.
[82] Minorsky, N. Nonlinear Oscillations. D. Van Nostrand Company, Inc., Princeton, 1962.
[83] Pages, A., and Gondran, M. System Reliability Evaluation and Prediction in Engineering, North Oxford Academic, 1986.
[84] Paule, P., and Schom, M. A Mathematica Version ojZeilberger's Aigorithmfor Proving
Binomialldentities. J. Symbolic Computation, Vol. 20. pp. 673-698,1995.
[85] Petkovsek, M., Wilf, H. S., and Zeilberger, D. A=B. A. K. Peters, Wellesley, Massachusetts, 1996.
[86] Plouffe, S. On the Rapid Computation ol Various Polylogarithmic Constants. Math.
Comp., VoI. 66(1), pp.903-913. 1997.
[87] Polya, G. and Szeg, G. Alifgaben und Lehrsatze alls der Analysis. Springer-Verlag,
Berlin, Dover, New York, 1945.
182
[101] Sofo, A., and Cerone, P. On a Fibonacci Rela/ed Series. The Fibonacci Quarterly. VoI.
36, pp. 211-215, 1998.
[102] Strehl, V. Binomial iden/ilies- Combina/orial and algori/hmic aspec/s. Discrete mathematics, VoI. 136, pp.309-346, 1994.
[I03] Tagaki, H. Queueing Analysis, A Foundation ofPerformance Evaluation, Vol.l, Vacation
and Priority System, Part I. North Holland, Amsterdam, 1991.
[104] Teichmann, T. Slowing down
pp.292-294, 1960.
0/ neutrons.
[105] Tijrns, H. C. Slochaslic Modelling and Analysis: A Compulaliona/ Approach. lohn Wiley
and Sons, New York, 1986.
[106] Titchmarsh, E. C. The /heory o//une/ions. Oxford University Press, Second Edition,
1939.
REFERENCES
183
[107] Vieh, R. Z transform theory and applications. D. Reidel Publishing Co., Boston, 1987.
[108] Wheeler, F. S. Two differenlial-difference equalions arising in number Iheory. Trans. of
the Ameriean Math. Soe., Vol. 318, pp. 491-523, 1990.
[109] Wheelon, A. D. On Ihe summalion ofinftnite series in closedform. Journal of Applied
Physics, Vol. 25, pp.lt3-lt8, 1954.
[110] Whittaker, E. T., and Watson, G. N. A Course o[Modern Analysis. Cambridge University
Press, Cambridge, Fourth Edition, reprinted 1978.
[111] Wilf, H. S. Generating[unclionology. Academic Press Inc., New York, 1994.
[112] W.M.C. problem group. Solution to Asymplolic Linearily. American Mathematical
Monthly, Vol. 99, pp.877-878, 1992.
[I 13] Woodward, S. 1. R., and Wake, G. C.A Differential-Delay Model ofPaslure Accumulation
and Loss in Conlrolled Grazing Systems. Mathematical Biosciences, Vol. 121, pp.37-60,
1994.
[114] Wright, E. M. Slability crileria and Ihe real rools of a lranscendenlal equation. Journal
Soc. Indust. Appl. Math., Vol. 9, pp.136-148, 1961.
[115] Zhao, Y. Analysis o[lhe Gl x /M/c model. Questa, VoI. 15, pp.347-364, 1994.
185
Index
Ztransfonn,x, 111,112,135--137,165
Cebysev, 125
Cebysev, 13
theorem, 65
central binomial eoeffieients, 111, 116, 118, 161
charaeteristie funetion, 31-33, 36, 38, 43, 89,
104, 105, 107, 112, 125, 137, 147,
165,169,170
combinatorial sums, 14
complex variable, 38, 64,111,169
complex zeros, 52, 53
Conolly's problem, x, 32, 53, 60
contour integration, 3
convolution, 7, 92, 93, 98, 116, 161
creative teleseoping, 26, 135, 138
dclays,42
mixed,42
multiple, 32, 42
neutral,42
dilTcrcnee-delay system, x, 111, 135
dilTerential-difference equation, 31, 32, 39, 44,
45,48,51,87,88
Dirae delta funetion, 5, 74, 88
diseretc step funetion, 112, 136
divergent, 42, 71
divergent series, 38
dominant simple zero, 34
dynamical system, 87, 99, 103, 108, 160
Euler, 1,53
classical polynomials, 11
generalization of the Euler Sum, 87
resul!, 7
sums, 3
Fcrmat, 124
Fibonaeei, x, 111, 124
products, 123
sequenec, 123, 128, 134-136
finite binomial, x, 73, 87, 135
foreing tenns, 40, 43, 88, 99,104,108, \36, 144,
160,174
187
188
Newton, I
forward difference fonnula, 168
numerical estimatcs, 10
Polya and Szcg, 38
Pell, 124
Pfaff's rellection law, 19
Pfaff-Saalschtz idcntity, 24
Pfaff-Saalschtz summation, 13
Placzck function, 47
Pochhammer, 12,54, 100, 118, 150, 158
Poisson process, 50
poles, 3, 33, 46, 89,137,138,165
simple, 112, 167
polygamma functions, 7, 9, 10
quadratic transformation, 158
Ramanujan, 2, 57
rational certificate function, 21,142,156
Rayleigh, Lord, 6, 7
recurrancc relation, 152
ReedDawson, 16,20-22,26,27
renewal, 31, 46, 48, 103
renewal proeess, x, 87, 102
Residue
Theorem, 32, 89
thcory,3
residue, 3, 7, 33, 47, 49,89,90, 104, 112, 137,
148,165
Riordan, 14-16,21
Rouche 's Theorem, 105, 170, 171
Rousseau, 14, 17
Ruin problem, 31, 50
simple function, 36, 64, 67
Sister Celine, 14,25,26
Snake Oil Method, 21
steady statc, 51
Stirling approximation, 38, 61
Stirling numbers, x, 82
Taylorseries,6,37,91,I21,I49,I53
teletraffic, x, 31
transcendental equation, 51
transcendental function, 6, 31, 51, 52, 109
tree scareh, I, 14
triangular matrix, 81
Vandermonde idcntity, 23
Volterra integral, 73
wavc
rectangular, 74, 92
Wilf, 12, 14,21,26,28,116,128,135,138,147,
161,173
INDEX
WZ pairs, 14,26,27, 135, 138, 147, 156, 161,
173
Zb algorithm, 18, 20, 21, 26
Zeilbcrger,2, 12, 14,26, 135, 138, 147, 161, 173
189
algorithm, 24
zeros, 6, 25, 32-34, 51-53, 89, 103-107, 109,
112, 124, 137, 138, 142, 148, 165169,172
complcx, 52, 53