Bok:978 1 4615 0057 5

Computational
Techniques for the

Summation of Series
Computational
Techniques for the
Summation of Series
Anthony Sofa
School of Computer Science and Mathematics
Victoria University
Melboume, Australia
Springer-Science+Business Media, LLC
Library of Congress Cataloging-in-Publication Data

Sofo, Anthony.Computational techniques for the summation of series/Anthony Sofo.
p. cm.
Includes bibliographical references and index.
ISBN 978-1-4613-4904-4
ISBN 978-1-4615-0057-5 (eBook)
DOI 10.1007/978-1-4615-0057-5
1. Series. 2. Summability theory. 3. Functional equations. 4. Functions of complex
variables. I. Title.
QA295.S64122003
515' .243-dc22
2003054692
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.
For their time and understanding

I dedicate this text to my supportive wife Lucia
and our wonderful children
Fiona, Jessica, Clara and Matthew.
Contents
Preface
Acknowledgments
xi
xv
1. SOME METHODS FOR CLOSED FORM REPRESENTATION
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
Search Sum and Some Relations

Binomial Summation
Riordan
Method of Jonassen and Knuth
Method of Gessel
Method of Rousseau
Hypergeometrie Form
Snake Oil Method
Some Relations
Method of Sister Celine
Method of Creative Telescoping
WZ Pairs Method
2. NON-HYPERGEOMETRIC SUMMATION
31
Introduction
31
Method
32
Brmann's Theorem and AppIication
36
Differentiation and Integration
39
VB
viii
COMPUTATIONAL TECHNIQUES FOR THE SUMMATION OF SERIES
Forcing Terms
40
Multiple Dclays, Mixed and Neutral Equations
42
Bruwier Series
43
Teletraffic Example
44
Neutron Behaviour Example
46
10
A Renewal Example
48
11
Ruin Problems in Compound Poisson Processes
50
12
A Grazing System
50
13
Zeros of the Transcendental Equation
51
14
Numerical Examples
53
15
Euler's Work
53
16
Jensen's Work
55
17
Ramanujan's Question
57
18
Cohen's Modification and Extension
57
19
Conolly's Problem
60
3. BRMANN'S THEOREM
63
Introduction
63
Brmann's Theorem and Proof

Applying Brmann's Theorem
2.1
2.2
The Remainder
67
68
Convergence Region
3.1
Extension of the Series
69
70
63
4. BINOMIAL TYPE SUMS
73
Introduction
73
Problem Statement
73
A Recurrence Relation
74
Relations Between G k (m) and Fk+ 1 (m)
81
5. GENERALIZATION OF THE EULER SUM

2
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
The k-Dominant Zeros Case

3.1
The k-System
Examples
3.2
Extension
3.3
6. HYPERGEOMETRIC SUMMATION: FIBONACCI AND RELATED

SERIES
Introduction
The Difference-Delay System
2
3
4
5
The Infinite Sum

The Lagrange Form
Central Binomial Coefficients
Related Results
Fibonacci, Related Polynomials and Products
Functional Fom1s
5.1
6
7
7. SUMS AND PRODUCTS OF BINOMIAL TYPE

1
Introduction
2
3
4
5
Technique
Multiple Zeros
More Sums
Other Forcing Terms
8. SUMS OF BINOMIAL VARIATION

1
Introduction
2
One Dominant Zero

2.1
Recurrences
2.2
Proof of Conjecture
2.3
Hypergeometric Functions
2.4
Forcing Terms
2.5
Products of Central Binomial Coefficients
3
4
Multiple Dominant Zeros

3.1
The k Theorem
Zeros
Numerical Results and Special Cases
4.1
4.2
The Hypergeometric Connection
Non-zero Forcing Terms
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
About the Author
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
SOME METHODS FOR CLOSED FORM

REPRESENTATION
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
Identities play an important role in mathematics and have been a source of

inspiration and sweat for many mathematicians over a long period of time.
Jacques Bernoulli (1654-1705), a contemporary of Newton (1642-1722), and
Leibniz (1646-1716) discovered the sum of several infinite series in closed form,
but did not succeed in finding, in closed form, the sum ofthe reciprocals ofthe
squares
"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
was to find an exact value. Euler succeeded, eventually in writing

11'2
(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 ,
Koecher [75] gave
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' ) ,
Ramanujan, see Berggren [8], evaluated
~ (4j)!(1103 + 26390j)
9801 ~
(j!)4396 4j
,
..!. _ 2y'2
11' -
J=O
Bailey [4] and coworkers wrote, (originally given by Plouffe [86])
=L
00
11'
J=o
1 [4
16j 8j
+1 -
8j
2+ 4 - 8) 1+ 5 - 8) 1]
+6 '
Amdeberhan and Zeilberger [2] published
( ) _ ~ (-1)j(j!)10(205)2 + 250)
(3 - L.64((2j + 1)!)5
)=0
+ 77)
'
Some Methodsfor Closed Form Representatioll
and Clausen [30] gave the result

2a, 2b, a
+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
Residue theory and contour integration can be gainfully employed to express

certain sums in closed form. From
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)
The residue evaluation ofthe integral

_1
2?Ti
?T
dz
sin?Tz (z2 + 1)
leads to the alternating sign identity
(-lt
?T
L n + 1= 2' Cinh?T - 1)
00
1/=1
and also we may obtain

~
1
?T-3
~ sinh 2 n;f = ~.
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
Use of Integral Equations
Cerone [23] considers an integral equation of the form
B (t) = <p ~~~ t)
B (t - u) <p (u) du
(1.4)
where B (t) is a single sex deterministic model representing births at time t,

<p (t) is a net matemity function which is of compact support and l (x) is the
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 0

.
!-lj
where
00
!-lj
= - [d:(P)]
P
P=Pi
= !e-UPi<p (u) du.

o
(1.6)
Same Methads Jar Clased Form Representatiol1
By allowing cp (t) to be exponentially constrained Cerone shows that
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
satisfies the recurrence relation
( _ 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
n=2,3,4, ... , (1.8)
J
00
uncp(U)du < 00
Mn =
are the n th moments of cp (t) . Now, in particular if cp (x) = ed (x - b) with e, b

constants and d (x) is the Dirac delta function, then cI> (p) = ee- bp and Mn =
eb n . The roots of the characteristic equation cI> (p) = 1 are given explicitly as
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].
By considering a partial fraction decomposition, such as
(1.9)
COMPUTATIONAL TECHNIQUES FOR THE SUMMATION OF SERlES
we may obtain, by the use of(1.2) and (1.9), other identities ofthe form
Ln
00
4 _
a4
= 4a4 [2 -
1fa (cot 1fa + coth 1fa)].
(1.10)
n=l
Taking the limit as a --; 0, in (1.10), confirms the result
Ln-
00
n=l
f c~~~:2 ) d:
~.
90
Secondly, (1.2) mayaiso be integrated with respect to the parameter a. From

(1.2)
[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)
where the constant of integration in (1.12) is identically zero. The identity

(1.12) is also obtained by Wheelon [109] using a different technique. Notice
that the left hand side of(1.12) may be rewritten such that
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
the m j's are the zeros 0 f the
(m) = cosmj coshmj
+1
(1.13)
A Taylor series expansion of (1.13) is
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
g' (z) z-adz

()' a > 1
9 z
(1.14)
Some Methodsfor Closed Form Representation
and ehoosing a = 4 and a = 8 in (1.14) we reeover the two results of Lord

Rayleigh. From residue ealculations and (1.14) we mayaiso give, for example
00
'\'
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
teehnique for summing series is that whieh is deseribed by Wheelon and is

worthy of a mention here, sinee we ean generalize some of his results and also
make a eonneetion with the polygamma functions, 7jJ (:r).
1.4
Wheelon 's Results
Wheelon 's method is based on the parametrie representation of the general

term of aseries, so as to produee either the geometrie or exponential series
inside one or more integral signs. The fundamental operation is contained in the
summation ofboth sides of a Laplaee transform pair with respeet to a transform
variable which is interpreted as the dummy index of summation. This operation
exhibits the desired sum as an integral of the geometrie or exponential series
eaeh of whieh may be summed in closed form. Consider the Laplaee transfonn
of a function .f (x) ,
00
F (p) = j e-xPf (x) dx

o
and if we identify the transform variable p with a dummy index of summation
n, we ean write
00
LF(n)= !L(e-xtf(x)d:r.
o
( 1.15)
As an illustration choosing f (x) = x in (1.15), leads to Euler's result (1.1).

An obvious extension is that (1.15) may be generalized to
00
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 [adl - U) + a2 (U - V)dw+ a3 (V - W) ... + ajWF,
= (j-l)!f du / dv ...
Jo
(j - 1) times
and using the relation

00
~
Aj
/ sj-1e->'sds
(j-l)!
allows a generalization of Wheelon's result as
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
lli 2~a a 3~a

a ,
a '
From(1.16) we can see that
. /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
Also, for a and j integers ~ 1 we have, because of symmetry and known

properties of the hypergeometrie function
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
and we have the identity, from Gauss's

2Fl [
j 1+11 11]
F [ 1,1,2,3, ... ,j
j 2,3,4, ... ,j+l
2Fl
summation
~ 1; j ~ 2.
1]
For a = 2, and from(1.l6)
5(2,j)
.1
() -1)!
[2j-2In2+~( ___ 1r(j-2)2j-2-r(2r-1)l

6
~., 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'
Other specific values of (1.16) may be obtained as follows
5 (6,12)
15 (~2!) [61440 In 2 + 10935 In 3 + 122571')3 - 61251]

1
[11112511]
- , 7F.6 13'6'3'2'3'6'
7 15 8 17 3 1
12.
6' 3' 6' 3' 6'
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
and again from (1.17) we have the identity
210
- 1263F2 [ 2.31'r'
d
3 ' 3
1
1]
= 3455 - 56071')3 - 700 In 3.
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
L (n + 2k) (n + 4k) =2k Lr +2k

00
Sk=
n=O
r=O
from which, we obtain the very slow converging series
Sk = 2k {'I/J (4k) - 'I/J (2k)}.

If
00
= ~ n 2 (4n2 + 1)
we may use partial fraction decomposition with polygamma functions, so that
(~) + 'I/J' (1) + 'I/J' (~)
8'I/J (1) - 8'I/J
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
Some Methods Jor Closed Form Representation
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
is a Bessel function of the first kind, of order 0:, let
The zeros, jk = jk (0:), k = 1, 2, 3, ... ofthe Besselfunction ofthefirst

kind of order 0:, are known to be real for 0: > -1 and are ordered such that
j-k = -jk andO < [11[ < [12[ < ....
The o:-Bernoulli polynomials, (or generalised Bernoulli polynomials)
Bn,ex (x) are defined by the generating function

e (x-.!)z
2
"\"'
go '2
k=O
00
(iz) = ~
B n,a ()
x Zn
I
n.
,[z[ < 2[Jd
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
Other remarkable identities involving the general sums
L Tr
00
for rEN / {O}
k=1 Jk
and the o:-Bernoulli polynomials are given in [51].
12
1.5
Binomial sums and hypergeometric funetions are intrinsieally related. It

is of fundamental importanee that binomial sums ean be generally written as
a terminating hypergeometrie series, see Roy [91]. The book, A = B, by
Petkovsek, Wilf and Zeilberger [85] expertly expounds the theory ofhypergeometrie closed form representation ofbinomial sums. The following is therefore
abrief deseription of the hypergeometrie funetion and some of it's prominent
properties. The books of Bailey [6], Slater [93] and Gaspar and Rahman [52]
eover all of the material presented here.
Ifthe ratio oftwo eonseeutive terms Tk+dTk , in aseries, is a eonstant, then
we have a geometrie series. A hypergeometrie series arises when the ratio is a
rational funetion of a positive integer k,
(1.l8)
where aI, ... , ap ; b1, ... , bq and z are eomplex and To

funetion is defined as
(ah
(a)o=l
= a (a + 1) . (a + k -
1)
= r~!~)
1. Poehhammer's
(1.19)
and henee a hypergeometrie series may be written as
pFq [ al,a2,,a p
bl, h ... , bq
Jz]
k==O
(a1)k(ap)k zk .
(b1h (bqh k!
( 1.20)
The hypergeometric series (1.20) is symmetrie both in its upper parameters

al, ... , a p and its lower parameters bl, ... , bq In general it is required that
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
Hypergeometrie functions play an important role in many fields of pure and

applied mathematics as weIl as in science. The excellent survey paper of Andrews [3] puts basic hypergeometrie functions in an applicable setting. More
reeently hypergeometric funetions led to the solution of the long standing
13
Some Methods Jor Closed Form Representation
problem of the Bieberbach conjecture by deBranges [40]; which shows that

if
f (z)
00
+L
n=2
anz n is a normalized univalent analytic function in the
unit disc, then for each n 2': 2 one has

hypergeometrie series are
oFo [
1 Fo [
z]
lanl ::;
=
n. Some elementary cases of
eZ , and
- a 1 z] = (1 - z t
Bessel functions may be expressed in the form
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 )
is called k- balanced if z = 1 and k + ao + al + ... + a q = b1 + b2 + ... + bq ;

or just balanced if k = 1; well-poised if 1 + ao = aj + b1 = ... = a q + bq ,
and very well-poised if it is well-poised and al = 1 +
There are a number
of ca ses where (1.21) with argument z = 1 can be evaluated in closed fonn
as a quotient of products of Gamma functions. Five of these cases are:
T'
1. the Gauss summation formula

2. Kummer summation formula
3. the balanced Pfaff-Saalschtz summation formula
4. the well-poised Dixon summation formula, and
5. the 2-balanced and very well-poised Dougall summation formula.
The Gauss summation formula is a limit of the Pfaff-Saalschtz summation
formula, Kummer's formula is a limit of Dixon's formula and mayaiso be
obtained from Dougall's formula. The Pfaff-Saalschtz summation formula
can be explicitly written as
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
A Tree Search Sum and Some Relations

Binomial Summation
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.
Riordan analyses (1.23) and (1.24) byrecurrences. Writing9n = 2-n (
( 1.24)
2~L
).
then
2n+1 (
hn+l=L
k=O
2n: 1)
(-l)k gk =O.
Now 9n = (2 - ~) 9n-1 and also

hn
Hence
n (fn
+ fn-r) = hn- 1 + fn-1 = nfn-l + (n nfn = (n - 1) fn-2,
fo = 1,.fI
1) fn-2, and therefore

( 1.25)
= O.
From (1.25), we have hn+1 = 0 and

f2n
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
Riordan expands on these ideas and obtains the additional identities
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
Method of Jonassen and Knuth
Jonassen and Knuth [71] consider (1.22) and by algebraic manipulations

obtain the recurrence (1.25) as folIows. From (1.22)
hence
n (In
+ In-I) = Ln
k=O
k+1
(-1)
2"
k (
2k )
k'
(1.28)
Replacing n with n - 1 in (1.28) we get
(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 ).
Let [x n) I (x) denote the coefficient of x n in I (x) ,hence
[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
when I (x) is analytic, then
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
for areal and b integer. The ratio of consecutive terms is

b
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
= 1, and hence from (1.32)

in (a, b) =
bFb-1 [
1
bb 1 ,
:~:' :~:, ... , t1- n
, b-l' b-l' ... , b-l
/ (b
a~~b-ll,
(1.33)
moreover, for the relatively simple case of b = 1
In(a,l) =
IFO[
-_n la] =(I-a)n.
Now, we concentrate on the case of b = 2 ; from (1.33)
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
+ 2) in+2 + (2n + 3) (2a -
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
hence from (1.38) and (1.37)

(1.39)
Similarly by Pfaff's reftection law
2FI
al, a~3- a2
1t] = 2
01
2FI
a~:2 1- 1]
we have from (1.3 7)

2F I [
-~n~-;:
t]
= T n 2 F I [ -n
~~;;1.
I -1.
]
Using the classical Gauss formula
F [ a 1 , a2
2 I
a3
1]
= r(a3)r(a3 -al
- (2)
r(a3- a 2)r(a3- a I)
we obtain from (1.39)

(1.40)
such that when n is odd in = 0 and when TL is even
hn
= ( 4n ) ~ ((2n)!)2 4n (2n)! = T 2n ( 2n ).
2n
4n
2nn!
(4n)!
20
Also, from (1.35) for a = 1/2 we have that (n + 2) In+2 - (n + 1) In = 0

which is identical to (1.25) and hence the Reed Dawson identity follows. For
a = 1/4, from (1.34)
and from (1.35) (n
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=
also from (1.31) hn (1/2,2) = In (1/4, 2). For b = 3, a recurrence relation,

using the Zb algorithm in Mathematica, In (a, 3) = In, of(1.31) is
2 (n
10 =
+ 3) (2n + 5) 111+3 + (n 2 (27a - 12) + n (135a - 56)

+168a - 66) 111+2 + 2 (n + 2) (3n (2 - 9a) + 11 - 54a) 111+1
+ (27a - 4) (n + 1) (n + 2) 111 = 0,
1,
!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)
and in hypergeometric notation
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~]
which has a recurrence relation

2 (2n
+ 1) 911+1 + (n + 1) (a -
4) 9n
+2=
0,90 = 1.
21
Same Methadsfar Clased Form Representatian
In the specific case of a = 4, we obtain the identity (1.27), evaluated by Riordan,

and it may be easily verified, utilizing the procedure described by Petkovsek et
al. [85], by the rational certificate function
k (1 - 2k)
= (n + 1 _ k) (2n -
R (n, k)
1)
For b = 3, a recurrence relation of (1.42), using the Zb algorithm in Mathematica, is

3 (3n
+ 4) (3n + 5) gn+2 -
{
go =
2 (n + 2)(n (27 - 2a) + 27 -. 3a) gn+1

27)(n + 1) (n + 2) gn - 6 = 0,
+ (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
and define F(x,y)
= L
2: )
(1.43)
in (y)x n . Now replace for in (y) and interchange
n~O
the order of summation, such that
1~ xL ( 2: ) C~ x)
k~O
Utilizing the identity
F (.7:, y) =
k~O (
2: )
zk
(1 - ;1;)
Vi -
Vl~4Z it follows from(1.44) that

1
= --;:;:=:::::;=::;::===;:=====;::::::
V(1 -
x) (1 - x (1 + 4y))
Ify= -1/2,F(x,-1/2) = V1~x2 andtheReedDawsonidentityfollows. If

y = -1/4,
F (x, -1/4)
L ( 2m
)
m
m~O
(~)2m
2
22
and hence,
We can generalize (1.43) a little by considering
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)
If e = -1/2 and b = 2, we get F (z, -1/2,2)

relationship
( -V
2 ) (_4)k
(2;)
k
+ bn (18 -
9b + 2b 2 )
= (1- z2r1/2 and from the
= (_l)k ( k -k1/ 2 ) (_4)k
/ )
k k-l
~r k+l 2 =~rr(2 1)
y7r r (k + 1)
k! j=o J +
the Reed Dawson identity follows. If b = 1,
F{z,-1/2,1) =
1
vr=z
= L (
z k~O
1
+ 6) .
2k
k )
(Z)k
4 '
23

which corresponds to the Vandennonde identity
2.8
Some Relations
The related surn

(1.47)
for p and q integers is an interesting one, and is briefly considered here. For
q = 1 and p = 1, (1.47) is identical to (l.31) for a = 1 and b = 1. From (1.47)
we have
S ( )=
n p,q
P p-l
[-qn, -qn, -qn, ... , -qn

1,1,1, ... ,1
(_l)P+l]
'
(1.48)
[-2n,-2n,-2n, ... ,-2n 1(_1)P+l].
( 1.49)
and some special cases, from (l.48), are
Sn (1, q) = IFo [ -qn
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
Sn (2, q) = 2Fl [ -qn 1,-qn
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 )'
[-2n + 1, ... , -2n + 1 (_l)P+l]

1, ... ,1
F:
P p-l
~ ~ (-1)' (
r~ e: r
2;
+2 ~ ( -1)' (
(-1)"
=0
'
~~
and utilizing (1.50), gives the new result

n
-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+
1, n171 \ -n, -n /1] .
Strehl offers six different proofs of(1.51) based on:

Bailey's bilinear generating function for the Jacobi polynomials in the special case when the Jacobi polynomials reduce to Legendre polynomials,
A combinatorial approach to the Bailey identity,
Legendre inverse pairs,
the Pfaff-Saalschtz identity,
Zeilberger's algorithm, and
known recurrences for the Franel and Apery numbers.
25
From (l.51), after various manipulations Strehl obtains
(1.52)
Given that 2'\1,2 = -3

from (l.52)
t (~)2
(_1)k
1] '
[ -n -n I 2 + ,\ +:\
2 Fl
(l.53)
J5 are the zeros ofthe quadratic,\2 + 3'\ + 1, then
t (~) ('\1;
k=O
+ 1)k"t,\{'2
1,2
k=O
(~)2
j=O
(1.54)
Identifying (1.53) with (1.50) for q = 1 we mayaiso give the identity
t; (n)
k
n
. k
(-1)
2 Fl
-n, -n I] = B(~,~)
+
1
-1
2n
and from (1.54) we can write, the new result
Sn (2, 1)
= {;
(n) (1 +
k
1)k f;k '\{,2

.' (k)2
j
'\1,2
where a second order recurrence of (1.54) is
+ 2) Sn+2 (2, 1) + 4 (n + 1) Sn (2, 1) = 0,

with So (2, 1) = 1 and SI (2, 1) = 0; for n odd Sn (2, 1) = 0, hence
(n
(n
+ 1) S2n+2 (2, 1) + 2 (2n + 1) S2n (2, 1) =
and by iteration
S,
2n
2.9
Let
(2 1) = (_2)n
,
rr
n -l
j=O
2j + 1,
'+ 1
J
Method of Sister Celine

n
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
Utilizing the computer package added to "Mathematica", we can generate the

recurrence
2a (n - 1)2 F (n - 2, k - 2) - (n - 1)(an - 2n - a) F (n - 2, k - 1)
- n (n - 1) F (n - 2, k) - a (n - 1) (2n - 1) F (n - 1, k - 2)
- (2n - 1) (n - an - a) F (n - 1, k - 1)
+ n (2n -
1) F (n - 1, k)
- an (n - 1) F (n, k - 1) - n 2 F (n, k)
= O.
Setting a = 0, = 1 and summing over k, we obtain a recursion equation for

fn, namely
nfn = (n -1)fn-2 ,fo = 1,JI = 0
and the Reed Dawson identity folIows.
2.10
Method of Creative Telescoping
The method of creative telescoping is described in the book of Petkovsek,

Wilf and Zeilberger [85]. It utilizes the Zb algorithm in "Mathematica" so that
the input
Zb[Binomial[n, k] ( -
~)
Binomial[2k, k], k, n, 2]
responds with a recurrence relation
(1
+ n) Sum[n]- (n + 2) Sum[n + 2]
= 0
and with initial conditions leads to the Reed Dawson identity.
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)
Calling up the WZ package in "Mathematica" we obtain the certificate function
k2
R (71 k) = ...,--------,--.,-----,
(2n-k+1)(k-2-2n)
(1.60)
Now, we define
G (71 k) = R (71 k) F (71 k) =

- (_~)k (2k)!n!2
.,
,
k!(k-1)2(2n-k+2)!
,
(1.61)
such that F (71 + 1, k) - F (71, k) = G (71, k + 1) - G (71, k) is true. Sum that

equation over all integers k, such that the right hand side telescopes to zero and
therefore
(1.62)
L F (n+1,k)= LF(n,k).
The two discrete functions F (71, k) and G (71, k) are termed the WZ pairs.
From (1.62) and with initial conditions we obtain the Reed Dawson identity.
Petkovsek et al. [85] claim that the WZ pairs method provides extra information
because of the existence of a dual WZ pair. To obtain the dual WZ pair make
(
l)"n+bk
the substitution (an + bk + c)! by (-a~-bk-c-l)! for a + b # 0 in (1.59) and

(1.61) to obtain Fand G. Next change the variables (71, k) by
F* (71, k) = G (-k - 1, -71); G* (71, k) = F (-k, -71 - 1),

(this transformation maps WZ pairs to WZ pairs), such that we obtain
F* (71 k) = (_1)n+1 2n (71 - 1)!n!2 (2k - 1 - n)!

,
4k +1 (2n - 1)!k!2
and
*
(-1t+ 1 2n+1n!3 (2k - 2 - n)!
G (n, k) =
4k (2n + I)! (k _ 1)!2
As previously, we obtain f~
= 2.:.:
term in (1.63) we shall define
( 1.63)
(1.64)
F* (71, k) and because ofthe (2k - 1 - 71)
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
F* (n + 1, k) - F* (n, k) = G* (n, k + 1) - G* (n, k);
(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
1* (2 + 2m) - 1* (2m) + F* (2m, m + 1)

= -G* (2m + 1, m + 2) -
G* (2m, m + 2) ,
and from (1.63) and (1.64) substituting for F* and G* we obtain
f * (2
) = f* (2)
m
(3m + 2) (2m + I)! (2mW

m!(4m+3)!(m+1)! .
(1.67)
Iterating the recurrence (1.67) we have
1* (2 + 2
)=
1* (2) + ~
~
)=1
(3j + 2) (2j + 1)! (2j)!2

j!(4j+3)!(j+1)!
(1.68)
and from (1.63) and (1.65) we have
1* (2) = _~ ~
3~
k~2
(2k - 3)!
4k k!2 .
(1.69)
We can put (1.69) in "Mathematica, Algebra, SymbolieSum" and obtain
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
4m (2m - I)! (2m)!2

(4m - I)!
~
~
k=m+l
(2k - 1 - 2m)!
4 k + 1 k!2
-1 V2 _ ~
- n 2 3
_ ~ (3j +2) (2j +

~
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
Some ofthese results have been published in [94].
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
l' (t) + 0:1' (t -
a)
+ f (t) + "( f
(t - a) + J f (t
+ a)
= 0
is termed a first order linear delay, or retarded, differential-difference equation

for 0: = 0, J = 0 and a > O. For 0: = 0, J = 0 and a < 0 it is termed an
advanced equation. In the case J = 0 and a > 0 it is referred to as a neutral
equation and when 0: = 0, = and a > 0, an equation of mixed type. A
great deal of the studies for the stability of differential-difference equations
necessitate an investigation of its associated characteristic function. Some of
the early work in this area has been carried out by Pontryagin [88], Wright
[114] and more recently by Cooke and van den Driessche [37] and Hao and
Brauer [61]. In this chapter we will show that, by using Laplace trans form
techniques together with a reliance on asymptotics, series representations for
the solution of differential-difference equations may be expressed in closed
form. The series, in its region of convergence, it is conjectured, applies for all
values ofthe delay parameter without necessarily relying on its association with
the differential-difference equation. Unlike some of the se ries that are listed
as high precision fraud by Borwein and Borwein [15] the series in this chapter
will be shown to be exact by the use of Brmann's Theorem. The analysis
also relies on the exact location of the zeros of the associated transcendental
characteristic function. The technique developed in this chapter is then applied
to particular examples that arise in teletraftic problems, neutron behaviour,
renewal problems, ruin problems and to a model of a grazing system. We also
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
Consider the first order linear homogeneous differential-difference equation

with real parameters a, band c and real variable t:
!,(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
The inverse Laplace trans form of (2.2) is

,E-l
(F (p))
= 1 (t)
="00 (-1)71 c71 e
-b(t-an)
(t
n!
L..t
n=O
- an
)71
H(t - an)
(2.3)
where the Heaviside unit function
H(x) = {
I,
for x 2: 0
0,
for x
<
o.
The solution to (2.1), by Laplace transform theory may be written as
'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
which suggests the solution of f (t) may be written in the form
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
After some experimentation it is conjectured from (2.5) that:

(2.6)
Vt E ~ in the region where the series on the left of(2.6) converges. Brmann's
Theorem will be used, a little later, to prove the identity (2.6). By the use of
the ratio test it can be shown that the series on the left of (2.6) converges in the
region
lacel+abl < 1.
(2.7)
In a similar fashion, the Laplace transform from (2.2) may be expressed as
F(p) _ ~
-
(1 +
b+ce-ap)-l _ ~~
.
-~~
n=O 1'=0
(n)
r
(_l)nbn-TcTe-arp
n+l'
and the inverse Laplace transform may be written as
As previous it is conjectured that
whenever the double series converges.

LEMMA 2.1
(2.7).
The poles ofthe expression (2.2) are all simplefor the inequality
34
Proo! Assume on the contrary that there is a repeated root of
P + b + ce-ap
=0
(2.9)
then by differentiation it is required that 1 - ace-ap = 0, in which case P =

In (ac) ja. Substituting in (2.9) results in In (ac) + ab + 1 = and therefore
acel+ab = 1 which violates the inequality (2.7). Hence all the zeros in (2.9)
are simple.
Now, the residue Qo of the dominant simple zero Po = ~ is
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=
we obtain, from (2.10)
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
Proof 1. and 2. can be shown to be solutions of (2.1) by substitution and

Statement 3 ofLemma 2.2 requires that we show
n
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)
Retuming briefly to (2.10), put b + c
t = -aT, so that
3!
+ ...
= 0, which implies that ~ = 0, also let
36
The inner sum, which we shall generalise in Chapter 4, is (-1

we obtain the common series
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.
Brmann's Theorem and Application
THEOREM
2.1 Let cjJ be a simple function in a domain D, zero at a point of
D, and let
e (z) =
/f f
z-
cjJ (z)
,e () = cjJ' ()"
(z) is analytic in D then Vz E D
f(z)=f()+
Ln {cjJ(z)Y dtd r-T!
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.
The v integral is taken along a contour r in D from to z, and the tintegral

along a c10sed contour C in D encircling r once positively.
We shall prove Brmann's Theorem in Chapter 3. However next, we shall
apply Brmann's Theorem to equation (2.10).
The characteristic function (2.4) may be shown to have a simple dominant
zero at p = 0 for b + c = 0 and 1 + ab > O. Thus from (2.6)
1
1 +ab'
Let t
= -aT,
ab
(2.13)
= - p, and hence from above

(2.14)
Identity (2.14) is now shown to be true by applying Brmann's Theorem. Let
eXZ
z
f (z) = - , e(z) = -(-) = eZ , cjJ (z) = ze-z,j ()=o = 1,
1-z
cjJ z
37
and we will show, in Chapter 3, that R n +1 -70 as n -7
00.
From f (t) = f:tt,
!'(t)=ext(~+
1 2) =e xt (~(.1:+1+j)tj),
1- t
(1 - t)
~
)=0
and so
f' (t){ B(t) r
= et(r+x) \[I (t), where \[I (t) =
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]
= dT"-l [et(T+X) \[I (t)]

dt r -
xr+ (7" + xr-
= et(r+x)
[(7" +
1 (
7" ; 1 )
\[1(0)
(t)
~1)
\[I'
(t)
+ (7" + :r) r3(7"-I)

2
\[I 11
(t) + ... + (7" + x) 1
X (
~ =~
2 (
7"
\[IU-2)
(t) + (7" + x)o (
~ =~
\[I(J-l)
(t)] .
Hence
Br()
((7"+x
+ (7" + xr- 2
7"
r-
(7";1 )(X+l)
~1)
+(7"+x)l
(x
+ 2) +
(7" + xr- 3 ( 7" ; 1 ) (x + 3) + ...
(~=~) (x+7"-I)(7"-2)!
+ (
If we now put y = x
~ =~
) (x + 7") (7" - I)!) .
+ r we obtain
B r (0) = yT-l (y - 7" + 1) + yr-2 (r - 1) (y - 7" + 2)
+ y1"<l (7" - 1)(7" - 2) (y - 7" + 3) + ...

+ (7" - l)!y(y - 1) + (7" - l)!y
38
= yr _
(r _1)yr-l
+ (r
_1)yr-l - (r -1) (r _ 2)yr-2
+ (r - 1) (r - 2) yr-2 - ... - (r - 1)!y + (r - 1)!y
= 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
plex variable p = x + iy, then (e 2(1-x) (x 2 +y2))2 < 1. The region is of

convergence looks like a guitar plectrum.
On the boundary p = 1, from (2.14), the series
Le
00
n=O
-(T+n) (
n.
Consider the divergent series
00
n=l
*,
may be shown to diverge.
then by the limit comparison test
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
From (2.10) and (2.2) it can be seen that
(_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.
Differentiation and Integration

Rewriting (2.10) we have that
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)
(_1)n cr!e- b (x-an) (x - ant
00
n.
n=O
and from (2.15), is equal to
.
exE
= (- c)J 1 + ab + a(
(2.17)
but from the characteristic function (2.4), since ~ is a zero
-c = (b + 0 ea~
and therefore (2.17) is equal to
(b + 0) e'c~
1 + ab + a~
(2.18)
40
which is equivalent to the right hand side of (2.16) after renaming x as t.

We mayaiso integrate (2.15) j times such that
L
n=O
00
For j
(-1)n cn eabn(t-ant+ j _ /
/
(n+j)!
........
~
et(b+~)
1+ab+a~dt.
j-times
= 1 we have, from (2.19)
where K is a constant ofintegration. Now putting t = x - a, n*

renaming the counter n *, we have
~
n=l
(2.19)
= n + 1 and
(_1)n cne -b(x-an) (x - an)n

n!
= -c [
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)n cne -b(t-an) (t _ an)n+l

(n+ I)!
= _1_
b+~
et{
_
1 + ab+a~
e-b(t+a)-a~].
(2.21)
If b + c = 0 and 1 + ab =f; 0 then ~ = 0, in which case from (2.21)
"00 bne-b(t-an) (t _ an)n+l _ _1 [ _ 1_ _ e-b(t+ a )]

L
(n + I)!
- b 1 + ab
.
n=O
5.
(2.22)
Forcing Terms
The result (2.19) may be arrived at by considering a difference-delay equation

with a forcing term. Let
f'(t)+bf(t)-bf(t-a)=
tm-1e- bt
r(rn) ,f(O)=O,
(2.23)
41
for m a positive integer. Following the procedure of the previous sections, we

have
1
F(p) = (p+b)m(p+b-be- ap )
(2.24)
where F (p) has a simple dominant pole at ~ = 0, and a pole of order m at

= -b. From these considerations we arrive at the result
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
Res(p = 0) = b"I(~+ab) and along BD and EF,
BD
=0=
J.
EF
(2.28)
42
Along D E and FG we have
J= J=
DE
limit
c---+O,R---+oo
FG
From (2.27), (2.28) and (2.29)

1
2rri
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)
and if, as previous, our conjecture is to follow then
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.
Multiple Delays, Mixed and Neutral Equations

Consider an equation with two delays
f'(t)+bf(t-a)-bf(t-)=O, f(O)=l,a,>O.
(2.32)
Taking the Laplace transform of (2.32) we obtain

1
(p) = p + be-op _ be-p
which has a simple dominant pole at ~ =

such that
F
(p) =
f: (~~; t
n=O
o. We may write F (p) in series form
(-lt (
) e-p(on+r-or)
r=O
and so using the techniques of the previous sections we have
1
- 1 - b (a - )" (2.33)
43
If (}: = 0 and = a (2.32) reduces to (2.1) and (2.33) is equivalent to (2.10). If

= -(}: (2.32) becomes a mixed equation, and (2.33) reduces to the identity
For the homogeneous neutral equation (forcing terms mayaiso be added).
f' (t)
+ bf' (t
- a)
+ cf (t -
a) = 0, f (0) = 1
we obtain
F( ) _
1
p - p + pbe- ap
+ ce- ap '
and from the methods of the previous sections

(2.34)
where
is the dominant zero of the characteristic function
9 (p) = p + (c + bp) e -ap.

Using the transformation (2.11), (2.34) reduces to the identity
(c + bO et (
c-a~(c-b~)'
and for the degenerate case of a
~~ ( 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)
where ~ is the dominant real root of ~ - vew~

of convergence of the series.
8.
= -v, a = -wand
= 0 and Ivwel < 1, is the region
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,
one negative root, plus p = 0 for 0< a < 1,

a double(repeated) root at p
= 0 for a =
one positive root, plus p = 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
which on putting t = -aT, we may write
for a
>
fr a
<1
where ~ is the positive dominant root of ~ -- 1 + e-a~ = O. Erlang considered

only the case of 0 < a < 1. In the case when a = 1 there is a double pole
which results in, from a previous statement in Section 3
lim (J (t) - 2t)
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
I: (_1)net~~(t_n)n. The W.M.C. problems group

0:::; n:::; t
[112] and Holzsager [66] both solve this problem, and in particular Holzsager
considers f (t) , Vt > o. Now, f (t) satisfies the differential-difference equation
f' (t) = f (t) - f (t - 1), t 2': 1 and using the theory of linear functional
differential equations, Holzsager shows the result (2.36). Holzsager's work
relates only to the asympttic of the finite sum whereas in this chapter it is
shown that the infinite sum is equal to the asymptotic expression for all t. We
mayaiso prove (2.36) in the following way.
positive integer
THEOREM
f (t)
2.2
(n-l e-(k-n) (k _ n)k
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
= 1 + L (ezt + L (ezr L (-~;

00
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
n = 1, 2, 3, ... and conslder the gene rat-
F(z)=l+LanZrl=l+LLe
n=1
-2n
71=0
)k
zn,letr=n-k,
, rename the counter

00
1)
= """
(ze1-Zr
~
n=O
r',
= 1- zel'
Z
46
We therefore have a pole at z = 1, hence F (z) is analytic on C : Izl S 1.

Other poles of F (z) are outside the unit circle. A Laurent expansion of F (z )
about z = 1,after putting t = 1 - z is
F (t)
1
1 - (1 - t) et
t~(l-~t+ ... )
and at t = 0, (z = 1) there is a pole of order 2. The principal part

G(z)
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'
is analytic for Izi S 1, and so converges for z

proof is complete.
= 1. Thus (2.37) follows and the

0
An elementary approach leading to (2.37) is suggested by Haigh [57]. Using

(probability) renewal theory arguments Haigh demonstrates, that, given that X
is a random positive variable, then E (X) =
00
L: Prob. (X ::: n) and if N (t) is
n=l
the number of random numbers we need to sum until we exceed some target t,
then E (N (t)) =
(2.37).
9.
[tl et-k(k t)k
L:
k=O
k! -
Cox [38] and Feller [47] then show the result
Neutron Behaviour Example
In the slowing down ofneutrons Teichmann [104] introduces Laplace transform techniques to analyze the renewal equation. This example involves the
47
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!
{(t - nuo)n + n (t - nuo)n-l}

(1-O't- 1
(l-O')t1
x e-nuo/(l-n) H (t -
nuo)
where t is lethargy and H (t - nuo) is the Heaviside function. From (2.38),

it may be shown that F (p) has a simple dominant pole at p = 0 and for
1 - 0: + 0: In 0' =1= 0 its residue contribution is A = 1 + l~a In o'. Using the
techniques developed in Section 2 it will now be shown that
~ (-lt
71=0
n!
{(t - nuo)n + n (t - nuo)n-l}

1-
1 - 0'
0:
e-nuo/(l-n)
(1 - 0') e-nt/(l-a)
(2.39)
From (2.10), for b = 0 and c = 1 we have
h(y) = ~ (-lt(y-ant
~
n!
n=O
e YT /
- - , where Tl
1 + OT]
+ e- ar/ = o.
(2.40)
Rewriting the left hand side of (2.39) gives
L
00
71=0
(_l)n {te-uO/(l-n) _ nuoe-uo/(J-o:)}n

n!
1 - 0:
1 - 0'
_ ~ (-lt {te-UO/(l-a) _ (n
~ n!
1- 0'
n=O
+ 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
=- thena = Be- B '

l-a
48
so that from (2.40)

'lte- B
e 1-<>
h(t) = - - .
1 +a'T}
From'T}+e-11Be-B = Oputa'T} = -EthenE = ae E andhenceEe- E

which is satisfied by the relationship E = aB. From (2.41)
e-at/(l-a)
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.
Identities (2.39) and (2.41) hold in the region of convergence
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
In detennining the availability of a renewed component Pages and Gondran

[83] consider the case of a constant failure rate. Given that A (t) is the availability of a Markovian component, >. is the constant failure rate, and 9 (t) is a
density function, then the integro-differential equation satisfied by A (t) is
J
t
~
A (t) = -.>.A (t) + (1 dt
A o) 9 (t)
+ .>.
9 (u) A (t - u) du, A (0)
= A o.
Taking the Laplace transform results in
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
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
in its region of convergence laAe1+aA I < 1 and Vt E R The value of the

availability limit sum is independent of the initial value Ao and the closed form
solution is independent of the value of t. It mayaiso be seen that
An
A"
L re-Art-an)
(t - an)n = L re-A(t-a(n+l)) (t - a (n + l))n
00
n=O
00
n.
n=O
n.
1
1 + aA
by putting t - a = Tin the second sumo Utilizing (2.10) and putting t

results in
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.
Ruin Problems in Compound Poisson Processes
The integro-differential equation

R' (t)
=~
{R(t) - / R(t - x) dF (x) }
(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))
Given that F is a distribution concentrated at a point a, F* is its Laplace

transform, J-L is the expectation of Fand R (0) = 1 - kJ-L, where k = ~ results
in
1 - kJ-L
R* ( ) _
(2.44)
p - p - k + ke- ap
Comparing (2.44) with (2.2), b = k, C
R (t)
= (1 - k/-L)
= -k results in
kt
:L -=r-ek(t-an)
(t - an)n H (t - an) .
n.
00
n=O
The characteristic equation p - k + ke- ap

p = 0, for 1 - ak =f:. 0 and therefore
~
~
n=O
(_k)n ek(t-an) (t _
n!
= 0 has a simple dominant zero at
ant = _1_, Vt E ~
1- ak
and in the region of convergence lake l - ak < 1.

From (2.44) a double pole occurs at p = 0 when 1 - ak = 0, therefore
(R()
11m
t
t-+oo
12.
kJ-L)) _
+ 2t(1a
2(1- kJ-L)
3
.
A Grazing System
Woodward and Wake [113] consider the differential-delay model
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
eonstants representing growing eonditions. From the work of Seetion 2, the

following results are inferred
(2.46)
where
is the dominant zero of
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
(1"1 - gt e-h -g)(t-r)
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.
Zeros of the Transcendental Equation
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,
in which ease a Re (~) = - Re (in (b + 0) + lnc. If a > 0 :::}as I~I -) 00

then Re (~) -) -00 and if a < 0 :::}as I~I -) 00 then Re (0 -) 00. Sinee (2.4)
is an analytie funetion of ~, there are therefore only a finite number of zeros
to the right of any line Re (0 = , for a > O. Similarly, if a < 0, there are a
finite number of zeros to the left of any line Re (0 = ,. We mayaiso note that
52
as a - t 0+, Re (~) - t -00 unless I~ + bl = 1, and as a - t 0-, Re (~) - t 00

unless I~ + bl = 1. A proof of the following theorem may be seen in Bellman
and Cooke [7].
THEOREM 2.3 A necessary and sulficient condition Jor the characteristic
Junction (2.4) to have zeros with negative real part is:
1. ab> 1,
2. -ab< ac < Jr12 + (ab)2 , where TJ is the zero oJ TJ

o < TJ < 7r or TJ = ~ if ab = O.
LEMMA
+ abtan TJ
= 0;
2.3 The characteristicJunclion (2.4) has al most two real zeros.
Prooj From (2.4), let z
so
= ap,
er
= ab, = ac,
G (z) = z
Putting Y (z)
> 0,
9 (ap)
= G (z) and
+ er + e- z .
= (z+!~)eZ = -1 then at the turning point z* = -
(2.49)
(1
+ er), we
Iel+ol < 1 then ifY (z*) < -1 and there
have Y (z*) = )+o, hence since,

are at most two real zeros of (2.49).
LEMMA 2.4 The transcendentalJunction (2.49) has afinite number oJcomp/ex

zeros with positive real part.
Prooj Let z
= x + iy, then from (2.49)

X + er + e- x eosy = 0
{
(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
(dZ) = (x+er)(x+er+l)+y2 >0.

d
(x + er + 1) 2 + y2
Suppose a pure imaginary zero exists, then z = iy and a manipulation of (2.50)

gives
53
increasing from 0: to
o < l (0:) < 2 (0:) < ... ,
For
so that lim
k->oo
00
there exists an increasing sequence
fh (0:) = 00 with sin vi 2 - 0: 2 > 0.
Here E (k (0:) , k+ 1 (0:)) and equation (2.49) has precisely k complex

zeros with positive real part. Also, whenever = k (0:) there exists a pair of
complex conjugate imaginary zeros iYk such that
(4k
7r
+ 1) "2 < Yk < (2k + 1) 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!
Asanexamplefor(a,b,c) = (8, -1,6),

2, from (2.52), the sum to six significant figures is
where~isthedominantzeroof(2.51).
~
= .997954 and for T =
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)
and given that for constants a and we may put
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
Euler makes several observations about (2.53) for particular cases.

1. Take the constant and the factor n to the left hand side of (2.53), then
investigate the limit as n - ? 0, such that
lnx
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 )'
where B (x, y) is the classical Beta function and from arearrangement of

(2.54)
v= ---:--a-
(2.56)
2. For = 0, from(2.53) and utilizing In (1 - av) = -a lnx, Euler obtains
~ (av)j
In (1 - av) = - L..J -.,- .
. 1
J=
J.
3. For = a, Euler writes

n
1
00
r+l
x" v )'
--=v+L..J(
r=l
II
r-l
r+ 1 . j=o (n+a(r+1))
(2.57)
55
and again taking the limit as n
-+
0, Euler obtains, from (2.57)
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)
and xC> = y, and from (2.58)
L (u(r+1))"
( + 1),. ,where uy = In y.
00
r=[)
l'
(2.59)
Multiply both sides of (2.59) by u, differentiate with respect to u, multiply

by u, and then again replacing uy = In y, Euler obtains
lny = ~ (ur)"
1-lny
L
r! '
"=1
(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
From this work, Euler states his theorem:
x8
= ~ (n + st
I-Inx
L
n!
n=O
(Inx)n
x
'
and (2.61) is identical to our equation (2.13) upon putting sa

x = e- ab .
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 -
dz n- 1 [1>' (z) (f(z)tt=a
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)
1 - wf' (z) where z = a
+ 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
= 1>' we obtain Theorem 2.4.
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)
where u = ze- z ; moreover ifwe put z = p and a = r in (2.63) we obtain

(2.14). Now, Jensen puts a = a and a = b in (2.62) and multiplies the two
series together to obtain
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.
a (a + IIr- 1b (b ~!(n - 11) t- v - 1uj ) .
J=O
v=o
Equating the coefficients of u j in (2.64), Jensen obtains
(a
+ b) (a + b + nt- 1 =
tu (~ )
a (a
+ IIr- 1 b (b + (n -
11) t- v - 1
57
and similarly, from (2.63)
which is one form of the celebrated generalization of the Binomial theorem

given by Abel in 1839. Putting x = b + n, a = 0: and = - in (2.65) we
obtain
(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
In the collected papers of Ramanujan, edited by Hardy, Seshu Aiyar and

Wilson [62] on page 332, Ramanujan states in question 738, if
lr-
1 e-x(r+l)
00 xl' (r +
0
=I 1 for x >
(2.67)
1 and also find
+ E)- <p (1)) .

E
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.
Cohen's Modification and Extension

Putting a = -1 and = -a in (2.62) and (2.63) Cohen [31], in his notation,
extends the results of the two sums
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
where w = -ze az . Cohen considers
(Dx) {(I - xat} =
t,
(_l)k (
) (1 + ak) x ak and
~)
(DX t {(l-x at }=t(-l)k(

k=O
(l+aktx ak
where D = d~' From
J
1
x a - 1 (Dxt {(I - xat}dx
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
and applying the transformation

00
[ni 8]
00
00
L L f(n,k) = LLf(n+ sk,k)

n=O k=O
n=Ok=O
he obtains the new result
= e- z
Zn { (1 - a)n }
n=O
( + l)n
= e- z IFI
~ 1 z (1 - a)] ,
which upon putting a = 1 we obtain the result (2.68) and letting ~ 00

we obtain the identity (2.69). Cohen goes on, in this article and in [32], to
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'
+ ak) xo:+ ak hence,
(~) (O'+akr- m
k=O
and similarly
We can now write, from above
ff
m=On=O
Xn~y~
t
m.n.
(-ll (
k=O
X
t,
) (0' + ak)n-rn
(-I)' ( ; ) ( +bp)m-n. (2.70)
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)
wh ich after some manipulation gives
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)
Also, from the left hand side of (2.70) we obtain
t, (~~!~nyn
~)
(0' + ak)n
+ %;1 (-bO'x)m
(2.73)
60
and applying the transformation (2.71) to (2.73), it reduces to
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
Brian Conolly [35] proposes, "for >. E [0,1] and m ~ 0, let
Sm (>') =
I: (>.n)~-m
e->.n.
n.
(2.75)
n2:1
Show that
So (>')
S3 (>')
>./ (1 - A), SI (A)
= 1,
S2 (A) = 1/ A - 1/2, and
1/>.2 - 3/ (4)') + 1/6".
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
(>.nt e->.n = _>._

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
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
which is identical to (2.76). From (2.78), for m 2: 1

Sm (,\)
f
n=l
An~m~
(-Ir (
n.
1'=0
n ) (n - rt- m +
r
in the second term, the inner sum
~(-lr( ~
)(n-rt -
=Oforn:;,m+l
and therefore from (2.80)

Sm (,\) =
A':~rn ~ (-Ir ( ~ )
n=l
(71, - rt- m , for m 2: 1.
(2.81)
1'=0
The convergence region of (2.75) for m 2: 1 is as previous however, byan

application ofthe limit comparison test and applying Stirling's approximation
for n! ,(2.75) converges on the boundary ,\ = 1, and'\ = -.278464 ....
Putting m = 1,2 and 3 into (2.81) we obtain Conolly's results. Obviously
other Sm (,\) results are available from (2.81), three other such values are;
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!
An alternative procedure, via a recurrence relation, for detennining closed form

representations of (2.75) is the following. Differentiate (2.75) with respect 1.,
algebraically manipulate the tenns, and finally we obtain the recurrence relation
d
dA (Sm (A))
1-A
+ ):Sm (A) =
~Sm-l (1.).
(2.82)
We can obtain an integrating factor of (2.82) such that
ddA (AmSm (A))
Am Sm (I.)
= Am
C~2
A) Sm-l (A) and
Am- 2 (1 - A) Sm-dA) dA,
(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))
(1_\)2 we get the additional result
S_ (A)
1- I.
= ~S-l (A),
00
= '"'
L
n=l
(A )n+l
n
e-.\n
n!
>..2
(1 _ >..)3'
(2.84)
which we shall generalize further in Chapter 5. I consider this procedure less

transparent than the first procedure leading to (2.81). The sum (2.13) will be
generalized further in Chapter 5.
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
Bnnann's Theorem [22] essentially allows, under various conditions, for

the expansion of one function in tenns of positive powers of another function. A
sketch proof ofBnnann's Theorem is given in Whittaker and Watson [110] and
at a first look it appears that the proof contains a line which seems misprinted.
The third last line on page 129 of [110] is
. . jt (~) d~
z
. .
a
JI
z
= _1
27ri.
a
f' (t) rp' (0 dt.dE,
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.
Brmann's Theorem and Proof
The following treatment is based on that given by Whittaker and Watson,

however there are some deficiencies and obscurities which we shall attempt to
63
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) is analytic in D then 't/ z E D
(z) =
(a)
+L
n
{cjJ (z)Y dr - 1
rl
dt r - 1
[1' (t){B (t)YL=a + R n + b
(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)
The ~ integral is taken along a contour f in D from a to z, and the tintegral

along a closed contour C in D encircling f once positively.
The following properties of simple functions are required for the proof of
Brrnann's Theorem and may be found in the book by Titchmarsh [106]:
PROPERTY 1 A function cjJ (z) of a complex variable z is called simple in a
domain D ifit is analytic in D and takes no value twice in D.
PROPERTY 2
If cjJ (z)
is simple in a domain D then 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
Brmann 's Theorem
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'
(t) <p' (t) dt.

<p(t)-<p(O
(3.3)
(ii) . Each~ of D hasaneighbourhoodN (~) inD. Now, byTaylor's Theorem,

for t E N (0 we have
<p (t) = <p (~)
+ (t - 0 <p' (0 + (t - 0 2 <pli (~) + (t - 0 3 <p1/I (~) + .. ,

2!
3!
and
<p' (t)
= <p' (0 + (t -
0 <pli (~)
+ (t ~!02 <p1/I (0 + (t
~!~)3 <p"" (0 + ....
Since <p (z) is simple in D and hence in the subdomain N (~),
<p (t) - <p (0
i= 0 Vt E N (0 - {O
so that
<p' (t) - <P' (0

<p(t) -<p(0
<pli (0 + ~<p1II (~)

<P' (0 + (t~!) <pli (0 .
(3.4)
Now <p' (0 i= 0 in D by property 2. Hence, by Knopp [74] page 180, the

quotient in (3.4) of power series is expressible as apower series for It - ~I
sufficiently smalI, so that it is an analytic function of t in some N of ~. Further
the quotient on the left hand side of (3.4) is a quotient of a function of t analytic
in D, with denominator non-zero in D - {O since <p is simple in D. So the
quotient is analytic in D - {O, as weil as analytic in a neighbourhood of ~.
Hence (3.4) is analytic in the whole of D.
(iii). It therefore follows that
<P' (t) - <P' (~) l' (t)
<p (t) - <p (~)
is analytic in D, being a product offunctions analytic in D. So by Cauchy's

Theorem
<p' (t) - <p' (0 f' (t) dt = 0
J.
. <p(t)-<p(O
c
'
and this with (3.3) , gives for each ~ on r
l' (0 =
_1
21ri
Jf'
(t) <P' (() dt

<p (t) - <p (0
(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)
by summing the geometrie series on the right;
For eaeh z in D we have, using (3.5)
f (z) - f (a) =
J!' (0 d~ = _1 Jd~J q;' (t) - q;' (0 dt

21ri
q; (t) - q; (~)
r
=
where
(3.6)
~ Jq;' (~) d~J f' (t) ~ (q;(~))r dt + R

c q;(t) ~ q;(t)
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)
is defined and analytic throughout D being a quotient of analytic funetions

with non-zero denominator since q; (z) - q; (a) =/::. 0 for z in D - {a}, and
being continuous at a. Taking the summation in (3.6) outside the integration,
the typical tenn is
_1 J d~ J f' (i) q; (~r q;' (~) dt

21ri
q; (tr+ 1
r
c
=
~
J f' (t) dt J
21rZ
q;(tf+l
= _1_
21ri
q; (e;r q;' (e;) de;
f' (t) () (tr+! dt q; (Zf+l - q; (af+l

(t-af+l
r+1
67
Brmann 's Theorem
=
=
[I' (t) {O(t)Y+l] 4> r(zr+

r! dtr
+1
4> (zr+ ~ [I' (t) {O (t)}r+l]
(r + I)! dtr
~~
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
Applying Brmann's Theorem
From Chapter 2, we have evaluated
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
3.1 Thejunction4>(z) = ze-zissimpleinthediscD
= {z: Izi <
~}
Proof Assume on thc contrary that 4> (z) is not simple in D, so that therc are
unequal z and z' such that
for
Izl <
~ and
Iz'l < ~ . Now

z
--, =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
Using the fact that
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
~ Iz - z'l < 1 so that we may write

1
-1'1<
z -
1-
21z - z ' I
I
and we want
1z '1 > 1 - 2'1 Iz - z'I > 1 - 2'1 > 2'1

This inequality contradicts the assumption Iz' I < 1, hence 4> (z) is both simple
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
To obtain an infinite series for f (zo) from Brmann's Theorem we need

to choose rand C so that R n + l - 0 as n - 00. Choose a fixed Zo such
that Izol :S Cl. Then r can be the line from 0 to Zo; and C can be the circle
with radius C2 > Cl centred at the origin, described once positively. From the
remainder R n + b defined in (3.2), consider, since 4> (z) = ze- z ,
4> (0 1 = I~et-~I <

14>(t)
t
where, real (t -
0 :S It -
~I
:S Itl + I~I
l:Qlereal(t-~)
Itl
< C2 + Cl = C3,
< Cl
1 4>(~)1
4>(t) - C2
eC3
and for appropriate positive constants Cl and C2,
I~ I < 1. Also
69
Brmann 's Theorem
so that
Ir/J (t) - cfJ (01
> IcfJ (t)1 - IcfJ (~)I = Ir/J (t)1
(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
[r/J(0]!1 f'(t)cfJ'(~) dtl < 27l'(C2) (Cl ec3

cfJ (t)
cfJ (t) - cfJ (0
C2
)!1 M
m
which is independent of ~. So that
IRn+11 S ~ (Cl) 27l' (C2)

27l'
0 as n
---4
(Cl eC3 )n M
nl
--> 00
since both M and m are independent of n.

In Chapter 2 we apply (3.1) and arrive at (3.7). If we put x
z =1= 0 we obtain
_1_ = ~ (nz et - nz
1- z
6
n!
= -tjz,
tt
n=O
and for the trivial case x = 0, hence t = 0 we have
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
IT+1+X (T+1+X)r (ze- I

T
+1
Ize-zllim
T--+OO
1)
(1+-+
T
Z)
+ :1:
X
r+x
1)
1+-T
+X
-x
for
70
So the series (3.7) converges Vz such that
Ize1-zl < 1
(3.8)
= x + iy , in Chapter 2 we stated that

the series converges in the region ((x 2 + y2) e2(1-x) 2 < 1 and so the curve
1
ze 1- z = 1, ((x 2 + y2) e2(1-x) 2 = 1 will separate the regions of convergel1ce
and diverges outside this region. If z
and divergence. Similarly in polar coordinates, z = re iO , the series converges

inside a certain oval whose polar equation is rel-rcosO = 1.
3.1
Extension of tbe Series
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.
From my knowledge of analysis apparantly this corollary has no counterpart

for real functions. To apply it to the series (3.7), let D be the inside ofthe oval
curve, and let the two sides of (3.7) be the two functions. They are equal in
Iz I :S Cl, a set of points S which has 0 as a limit point; and 0 is in D. So if both
sides of (3.7) are analytic in D they must be equal throughout D by the above
corollary. The left hand side of (3.7) is analytic in D, because it is analytic
everywhere except at z = 1 which is not a point of D. So now we l1eed to show
that the right hand side of(3.7) is also analytic in D. Consider
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.
By this lemma we have that F (w) is an analytic function of w in Iwl < e- 1 .

Now an analytic function of an analytic function of z is an analytic function of
z, ifthe ranges are correctly matched. In our case ze- z is an analytic function
of z in the whole z-plane, and its values w = ze- z satisfy Iwl < e- 1 if z lies in
Brmann 's Theorem
71
the oval region D. Thus F (ze- Z ) is an analytic function of z in D. Hence (3.7)

holds for all z in the disc D. It may be difficult to detennine whether (3.7) holds
for all z values on the curve. In Chapter 2 we demonstrated that (3.7) is divergent
at the point z = 1. At the other intersection 'W = -e- 1, z rv -.2784 ... , the
series (3.9) has real tenns altemating in sign which decrease in modulus by
the ratio test and te nd to zero as n ---t 00 and so the series is convergent at
z rv -.2784 ....
Chapter 4
BINOMIAL TYPE SUMS
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
Pm (11) = (- 1 ) , ' " ( __ 1

n+m. L....".
r=O
()
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
1jJ (t) = F (t)
1jJ (t - x) rjJ (x) dx
73
(4.2)
74
oeeur in a wide variety of applieations. If F (t) = 8 (t) , where 8 (t) is the

Dirae delta function, and we take the Laplaee transform of (4.2) then
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
moment of c/J (t) is given by
Mn=lim [(_l)n ddn {cf!

p--->O
pn
(pn] =
n
a +l .
n +1
(4.5)
Substituting (4.4) into (4.3) results in
1
\lI(p) - - - p -1 + e- ap
(4.6)
and an expansion of (4.6) gives
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
and Pm (n) is as given in (4.1).
3.
A Recurrence Relation
LEMMA
4.1 : A recurrence relationfor m (a) is
(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
Binomial Type Sums
Proof From (4.3) and (4.4) we may write
m(a)=~~~ [~!~: {l-~(P)}]'
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),
,
_.
[~
m.m (a) -~~~ dpm
. [dm
{
=hm
p ...... O
=J~
 (p)
1 + 1 _ (p)
dpTll
}]
 (p) }]
1 - (p)
t; ;
[ rn (
) drn-k
dk {
1
}]
dpm-k { (p)} dpk 1 _ q, (p)
. (4.10)
Hence, using (4.5) and (4.9), we may write (4.10) as
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
(1 - a) m (a) = A~O (-1)
m-k
am~k+l
(rn-k+l)!k (a), m = 1,2,3 ..
(4.12)
with o (a) = I~a' a -:f. 1,
hence the lemma is proved.

From (4.8), rn (a) may be expanded in a Maclaurin series
(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'
and this term is non-zero only for
7'
d --{1-a}~~)(O),
q=O,1,2 ....
q T
da
= q and 7' = q -
'
1, so that
76

q\m
0
I
2
3
4
5
6
7
8
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
The beta coefficients of (4.16).
Further from the right hand side of (4.12)
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)
These coefficients are demonstrated in Table 4.1.

Some observations that may be made from (4.16) and Table 4.1 are
~O) (0)
= 1,
~) (0)
~q) (0)
= r~) (0)
= q!,
~) (0) = 0 for m 2' 1,
and ~m+l) (0) = (_1)m.
Further, ~) (0) = 0 for 0< q :::; m, and so the leading power of(4.13) is
am +l .
77
Binomial Type Sums
4.1 The finite sum, Pm (n), (4.1) is a polynomial in n
THEOREM
0/ degree
m.
Proof From (4.8)

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
1 + (m+1)a 2 + -'--_--'--_-C..-.,..--'(m+l)ma 3 (a+2)

1-a 2a(1-a)2
12a 2 (I-a)3
= __
(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 ,
where the function X (a) is a polynomial in a, to be determined from the

particular m-l (a) , that is
( -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
t, (n~m )an{~ (m; 1) (-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!
I[6a k+l + 32a k+ + 8a k+ ak+4] + ... + (m + 1~!aX (a)}

2
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
Binomial Type Sums
Thus m (a) may be expressed in the form

m(a) =
~ (n~m
) {a n+2m p 1 (771)
+ a n+2m - 1P 2 (771) + .,. + a n+m+1 Pm (771) }

where the Pj (771), j = 1, 2, 3, ... ,771, are functions dependent on the fixed parameter 771 on1y. The summation indices are now adjusted to obtain coefficients
of common powers of a in the following manner,
+ ... + ~ (
n~:; 1 )
a n+m+1Pm_dm)
+~
(n~m )
a n +m + 1 Pm (m),
and so,
m (a)
(~L ) a 2m Pdm) + f (n~!~l

+ [(
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
Grouping of terms gives
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 - 1G2 (m) +

... + am+2Gm_l (m) + am+1Gm (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
and equating the powers of am +j , where j
= m + 1, m + 2, m + 3, ...
Since
n+k ) = (n+k)(n+k-l) ... (n+k-m+l),

n-m+k
m!
k=O,1,2, ... ,(m-l)
Binomial Type Sums
81
is a polynomial in n of degree m and the Fk+l (m) functions depend on the

fixed parameter m, then the right hand side of (4) is a polynomial in n of degree
m. Hence the theorem is proved.
0
4.
Relations Between G k (m) and F k +1 (m)
From (4.18) and (4.19) it can be seen that, on equating coefficients of a m +j

where j = 1,2, ... , m gives,
Gm (m)
= ( ~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
The functions Fj (m), j

from
1,2,3, ... , m in (4) can be recursively obtained

MF=G
(4.22)
where M is an (m x m) upper triangular matrix, Fand Gare (m xl) column

vectors. Similarly, from (4.20)
~m+k) (0)
Gm-k+l(m)= (m+k)!' k=,1,2,3, ... ,m,
82
putting q = m
+k
G2m- q+ I (m) =
~) (0)
I ' q=m+1,m+2, ... ,2m,
q.
and for the counter j = 2m - q + 1

~m-j+I)
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)!
This matrix setup therefore allows a recursive evaluation of the functions

Fj (m) ,j = 1,2,3, ... , m, in terms ofthe coefficients ~) (0) in the Maclaurin
series (4.13). In particular F I (m) takes the form
(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
Binomial Type Sums

m
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)
+ 1) (9n 4 + 54n 3 + 51n 2 -
58n
+ 16)
24
n (13571 7
9!
X 27 X
+ 1) (15n{; + 16571 5 + 46571 4

3
91n 2
9!
+ 126071(; + 3150715 + 840n1 3
_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!
The polynomials of(4.21).
sum (4.1) with an application of a problem using the idea of a multinomial

distribution. Moreover, the author has generalized (4.1): namely, given that
( -1 )n+m
Vrn(n,x) = (n+m)!
~(-1)
1"
(n)
(x+r)
n+m
(4.24)
for x areal number, then Vm (n, x) can be expressed as a polynomial in x and

n of degree m for both x and n.
THEOREM
andn.
4.2 Vm (n, .T) is a polynomial in x and n of degree mfor both x
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)
The result (4.27) can be integrated m times with respect to x to evaluate

Vm (n,x) as defined in (4.24). For example, integrating (4.27) and using the
initial condition VI (n, 0) = PI (n) = - n /2, from Table 4.2 for m = 1, results
in
Using this procedure we see that (4.24) is a polynomial in x and n of degree m

for both x and n. Alternatively, the recurrence relation
J
x
Vm (n, x) =
Pm
(n) -
Vm - I (n, t) dt, Vo (n, x) = 1,
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
Binomial Type Sums
v,,, (n, x)
1
(3n 2
+ 2x)
2
+ n (1 + 12x) + 12x 2 )
4!
(n 3
15n 4
+ n 2 (1 + 6x) + 11 (12x 2 + 2x) + 8x 3 )

(2 x 4!)
+ 30n3 (1 + 4x) + 5n 2 (1 + 24x (3x + 1))

8 x 6!
+
Table 4.3.
(n
The polynomials of (4.24).
2n (60.T 2 (4x
+ 1) 8 x 6!
1) + 240x 4
Chapter 5
GENERALIZATION OF THE EULER SUM
Abstract
1.
In this chapter an investigation of a generalization ofthe identity (2.13) ofChapter

2 is undertaken. The investigation will also make use ofthe finite binomial type
sums obtained in Chapter 4. A connection with renewal processes will be made.
It will be proved that generated infinite sums may be represented in c10sed form
that depend on k-dominant zeros of an associated transcendental characteristic
function.
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
(-ct e-b(t-an) (t - ant+ R- 1

(n+R-l)!
H(t-an)
(5.5)
89
Generalization ofthe Eu/er Sum
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)
which suggests the solution of f (t) may be written in the form
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)
From (5.5) and (5.8)

t _
[~l ( n + R -
f()-L
n=O
1 ) (_c)n e-b(t-an) (t - an)n+R-1

(n+R-1)!
(5.9)
90
where [x J represents the integer part of x and the residue contribution QR,k (po) ,
is given by
k=0,1,2, ... ,(R-l), (5.10)

since (5.3) has a pole of order R at the distinct dominant zero, P = Po for
1- ac =1= O. From now on we may take, without any loss of generality, b+ c = 0
and 1 + ab =1= o. These conditions simply allow the distinct dominant zero, Po,
of the characteristic function (5.6), with restrietion (5.7), to occur at Po = 0,
and therefore from (5.6) and (5.10) respectively
9 (p) = P + b - be- ap
and
k!QR,dO)
THEOREM
~]}'!I, [::' ( C~)
f) 1; ~
k
(5.11)
0, 1,2, ... , (R - 1). (5.12)
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
SR (b,a, t) = L:Tn (b,R,a, t)
(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
L: QR,k (0) (R _ k _ 1)!
(5.15)
k=O
The series (5.14) is known as an Abel type series, because of the
(t - an)n+R-l tenn and, the convergence region (5.7) may be obtained by

applying the ratio test to the term Tn (b, R, a, t) in (5.13). A proof ofthe main
Theorem 5.1 will follow shortly. Firstly we shall develop two useful recurrence
relations forthe evaluation ofthe terms QR,k (0) in (5.12) andan identity forthe
QR,k (0) tenns. Secondly, using the terms QR,k (0) we shall give some closed
fonn representations of the infinite sum (5.15). Thirdly, a recurrence relation
for the series (5.14) will be developed, and finally an induction argument on
the integer R will be applied to prove the main Theorem 5.1.
Generalization ofthe Euler Sum
2.2
91
Q R,k (0) Recurrences and Closed Forms
LEMMA
5.1 A recurrence relationfor the evaluation ofthe terms QR,k+l (0)
in (5.12) is
with
Proo! From (5.12)
Qn,o (0) =Um
p->o
[(~(
)) R]
9 p
(1
+1ab) R'
Also from (5.12)
(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
where 9 (p) is defined in (5.11) and its first derivative
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
Hence from (5.18)
(k + l)!QR,k+l (0)
. [d k
R ~~ dpk
{(
(5.19)
P ) R+l B (p) }]
9 (p)
[t, (~){ c~)rlr")
~ R~
BI")
~)]
by the Leibniz rule of differentiation, where

B(/L)
Now since
(p)
d/L
dp/L
= - B (p) .
[ dJ1.
]
( -1)/L ba/L+ 2
dp/L B (p) =
(f-t + 2)
~~
and substituting in (5.19) we find that
(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)
Proof Consider the rectangular wave

cp (x)
= -bH (a -
x)
= b( -1 + H (x -
which has a Laplace transform of<p (p) =

<p (p) can be expressed as
<pR
(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 .
Generalization ofthe Euler Sum
93
The convolution constants, cf!- in (5.20) can be evaluated recursively as follows
c~ = ', =
R=l
(.11)1;
Cf!- = L r-jCf- 1 ; R = 2,3,4, ...

j=O
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
5.3 A recurrence relationjor the evaluation ofthe terms Qn,k (0) in
(5.12) is:
QR dO)
=
x
1
(l-(-ab)R)
[1: (
R )
J
j=O
~ ( -- a)'-. ( -- ab VCI_.Qj,._ (0) -
with initial va lues
(-ab)R QR,dO)]
cR = 1 and Qo,o (0) = l.
Prao! From (5.12)
k!Qll,dO)
=~ .:;, [C_~ (p))"] ; k= 0.1.2. -- -, (R =lim
!:-.
[(1 + 1 -If> If>(p)(p) ) R]
dpk
=J~
f; dpkd R). (p)
p->O
[(
If>J
1]
(1 - If> (p}
1)
(5.21)
94

k
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 .
a'b' (3R' - IOR' +OR' + 2R)
_a 3b3 (40R 3 - 40R 2 - 16R)

+a 2 b2 (140R 2 + 36R - 8) - ab (128R + 64)
+ 16
Some values ofthe recurrence (5.12).
=;~ ~
R)
(k) d
d
j
~ r dpk-r [J (p)] dpr
k r
-
1
}
(1 - (p))j .
Utilizing Lemma 5.2, for the (k - r) th moment ofj (p) implies that
R
k!QR,k (0) =
~( ~) ~(
R
QR,k (0) =
; ) (_a)k-r (-ab)j (k - r)!CLrr!Qj,r (0)
~( ~) ~( ~
) (_a)k-r (-ab)j CLrQj,r (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.
The following lemma on a functional relationship OfQR,k (0) will be useful

in the proof of the main Theorem 5.1.
95
Generalization ofthe Eu/er Surn

LEMMA
R (1
5.4
d
+ ab) QR+I,k+1 (0) + ab db QR,k (0) = (R -
(k
+ 1)) QR,k+1 (0),
k = 0, 1,2,3, ... , (R - 1).
(5.22)
Proof From (5.12)
: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)
R (Qn+l,k (0) - Qn,k (0.
Now, the Q (0) terms may be associated by constants

QR+l,k+1 (0)
+ Cl (R -
(k
Cl,
(5.23)
C2, and C3 such that
+ 1)) QR,k+l (0)

+ C2QR,k (0) + C3QR+l,k (0) = 0; (5.24)
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'
Hence from (5.24)
QR+l,k+1 (0) -
(R-(k+1))
a
R (1 + ab) QR,k+l (0) - 1 + ab QR,k (0)
96

R
The c\osed form of (5.15).
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 - lOab + lOa 2b2) t

4(1+ab)8
+ (1+ab)3'
ba 4 (1- 8ab + 6a 2b2)

6(1+ab)7
5ba 3 (2 -7ab)t 2
(1+ab)7
ba 5 (3 - 66ab + 199a 2b2 - 72a 3b3)
3(1+ab)9
Some closed form expressions of (5.15).
(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.
For the specialized case of R = 2, b = -1 and a = A from (5.14) and

(5.15) we obtain the result, (2.84) in Chapter 2, that Conolly missed. In the
next section we give a proof of the main Theorem 5.1.
2.3
Lemma and Proof of Theorem 5.1
The following lemma will be useful for the proof of Theorem 5.1.
LEMMA
5.5 A recurrence relationfor the infinite series (5.14) is

d
R(l +ab)SR+l +ab dbSR - tSR = O.
(5.26)
97
Prao! From (5.13) and (5.14)

(5.27)
Also, from (5.l3) and (5.14)
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]
k) QR,k (0) - ab db QR,k-l (0)
tR - k
tR
x (R-k)! + (l+ab)R!QR,O(O). (5.30)

From Lemma 5.4, after adjusting the counter k
R (1
+ ab) QR+l,k (0) =
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
+ ab) R!QR,o (0) + L
k=l
tR
R! Ql,O (0) QR,O (0)
t R- k
(R _ k)!QR+l,k (0)
(5.32)
L (R _ k)! QR+l,k (0)

k=l
R
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)
which completes the proof of Theorem 5.1.
It is now worthwhile to briefly indicate a functional relationship for the

infinite sum (5.14). From the left hand side of (5.15), let
= aT, p = R -
1 and 'Y
= abeab then
99
Generalization oi the Eu/er Sum
(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.
CASE 5.1 Let
in the system (5.1), where m is a positive integer andfollowing the procedure

of Section 2, we obtain
L
DO
n=O
+R
_ 1 ) bne-b(t-an) (t - ant+ m+R - I
(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)!
'
In identity (5.1) we have that

V
v!Pm,v(-b) = lim [dd > {(p+b)m F(P)}]; v=0,1,2,3, ... ,(m-l),

p->-b
pi
1;
p,!QR,/l (0) =lim [dd ll {pR F (p)}

p-+O
pP
J
I'l = 0,1,2,3, ... , (R - 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
and Jor R = 2 and m = 3
In the degenerate case,for a = O,from (5.1) we obtain the impressive identity

00
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
R-I (-I)Jl (m)
+~
"" 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)!
in the system (5.1), where m is a positive integer andJollowing the procedure

oJ Section 2, we obtain
-b
L()
00
n=O
n (
n+ Rn - 1) L (n) (-Ir(n+m+R-l)!
(t - arr+ +
m
r=O
R- 1
m+R-I tm+R-k-1QR,k (0)

""
~
(m + R - k - I)! . (5.34)
k=O
(5.35)
= 0, 1,2, ... ,m + R -
1; m
1; R

where
101
1
F (p) = - - - - - - =
pm (p + b - be-ap)R
which has a pole of order m + R at the singularity p

by (5.11). Let t = -ax andfrom (5.34) we obtain
00
' " (-ab)n

~
(+
n
n=O
R
n
= 0, and 9 (p) is defined
-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
and hencefrom (5.2)
f(-abr'(n+~-l
)Vm+R-l(n,x)
n=O
m+R-l (_x)m+R-k-l QR,k (0)

"'"
~
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
1,)') = 1. In the special case, that

lim V'V:R-l(t
region labl < 1, since n.~oo
m+R--l n,x
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
From (5. 34),for the degenerate case a
~(_bt(n+R-1)
~
n
n=O
(5.37)
= 0, we have the identity
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'
However Jolley [70), entry 40 on page 8,
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 (1 + 6x)(n + 2x) + 8x3 )
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
wh ich mayaiso be checked on, say, 'Mathematica '.
2.5
Renewal Processes
In the theory of renewal processes, let M (t) = E (N (t)) , be the expected

number of renewals in the time interval [0, tJ such that
M (t) =
fl
n=1
J(n)
(x) dx
and the Laplace transform is
M(p) =
!(l!) ].
(5.38)
p 1 - J (p)
Also, the expected instantaneous renewal rate m (t) =

1ft M (t) exists such that
in (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
Generalization ofthe Euler Surn
M (t) , for a type 1 counter as
J
t
M (t)
1 - e- bt
+-
M (t - x) be-b(x-a) H (x - a) dx.
(5.40)
From (5.38) or (5.40) we may obtain
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)
and from (5.39) we have

in
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
Tbe k-Dominant Zeros Case

Tbe k-System
Consider, for a weIl behaved function f (t), the forced dynamical system
with constant real coefficients band c, real delay parameter a, and all initial
104
conditions at rest,
cf
(t - a)
r~o (
; )
+ r~o (
; )
bk - r f(r) (t) = w (t); t > a
<
bk - r f(r) (t) = w (t) ;
(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)
Expanding (5.43), inverting and considering the residue, we get

00
L
n=O
(_c)n e-b(t-an) (t _ an)nkH-l

k-l
~
(nk+k-1)!
= LQ(~v)e I/t
(5.44)
v=O
where
Q(~v)=lim [(p-~v)F(p)],
P-->~1/
v=0,1,2,3, ... ,k-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
which is convergent /or all values 0/ k, b, a and t in the region

(5.49)
Generalization
0/ the Euler Sum
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
IA (w)1 > IB (w)
- 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
5.7 The characteristic function
()] (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.
Proof Now, Al (w) =
1lJ
BI
has one dominant zero lying in the region
(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) =
bnke-b(t-an) (t _ ant k+k- 1

(nk + k _ 1)!
k-l
(5.52)
e~vt
~ (b+~v)k-l(k+ab+a~v)
The characteristic function (5.46) may be expressed as the product of factors

such that,
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) =
lim [(p - O:j) Fj (p)]

p.....OI.j
!im [p_-_O:_j] = ___

k_ _
qj (p)
k + ab + aO:j
p..... OI.j
and using this result, we have from (2.10) that
~ (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
Rescaling the infinite sum, by putting n

as n) gives the result,
)=0
= (n * + 1) k, (and then renaming n *
~ bnke-b(t-an~~~ ~ ~)~n -
~
n=-l
Tr
n=O
kbk
(t -
at k+ k
~ keO: jt
= ~ k+ab+ao:
j=o)
Generalization
orthe Eu/er Sum
Now letting y
=t -
107
a and, from
qj
(a j
) ,
using the fact that
eaa
= (b:a
j )
then
00
n=O
bnke-b(y-an) (y _ an)nk+k
(nk + k)!
Differentiating (5.54) with respect to y, which is permissible within the radius

of convergence (5.49), gives after some algebraic manipulation
00
n=O
bnke-b(y-an) (y _ ant k+k - l
(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
and ~v satisfies (~v

00
+ b)3 -
b3e- aev
= 0,11 = 0, 1,2. Hence we have
b3ne-b(t-an) (t _ an )3n+2
(3n+2)!
where
Xl =(x+b), X2=Xr- y2, x3=2YXI,X4=3+aXI and X5=ay.

For (a, b, t) = (.8,1,2) then
(~o, 6,
(1)
= (0, -1.2193 + 1.3668i, -1.2193 - 1.3668i)
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
0/ the Eu/er Sum
109
By the methods of the previous Section 3.1 we finally obtain

00
+R _ 1 )
n=O
b1lk e- b(t-an) (t _ antk+Rk+m-l
(nk
e-bten-r-I Pm,r
(-b)
+ Rk + m -
+ ~~:::.; e-("ttR-1L-1Q R,IL (~v)
(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
+ b)m F (P))]; r = 0, 1,2, ... , (m - 1),
~v)R F (P))];
f.L = 0,1,2, ... , (R - 1)
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
+ ~ (b + ~v)2 (2 (b + ~v) + ab2e-a~,,)2

2
[
x t - b + ~v
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
Some of these results have been published in [98] and [100).
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
In this chapter we consider a difference-delay system and by the use of Z

transfonn theory generate an infinite sum which we shall represent in closed
form by the use of residue theory. We shall compare our work to that of
lengen and lay the foundation so that our results may be further generalized in
the ensuing chapters. We shall investigate some central binomial coefficient
identities, and develop Fibonacci related polynomials, products and functional
forms.
2.
Tbe Difference-Delay System

Consider the related Fibonacci difference-delay system
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
fnz-n for those values
of z for which the infinite series converges. Taking the Z transfonn of (6.l)
111
112
and using the initial condition yields, upon rearrangement
F(z)=
z - b - cz- a
za+l
za+l - bz a - c'
(6.2)
=-~--
and expanding as aseries we have

00
F(z) =
t. (
I-ar
~ (: ~ b)Hr'
(6.3)
The inverse Z transform of(6.3) is
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)
9 (z) = za+l - bza - c
(6.7)
the characteristic function

has (a + 1) distinct zeros ~j, j = 0,1,2,3, ... , a. All the singularities in (6.2)
are therefore simple poles such that the residue, Resj ofthe poles in (6.2) may
be evaluated as follows
Resj
= lim [(z _ ~j)

z->~j
za]
za+l - bza - c
(a
~.J
+ 1) ~j -
(6.8)
ab
and hence from (6.4), (6.5) and (6.8) we have
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)
Hypergeometrie Summation: Fibonacci and Related Series
113
A Tauberian theorem [7] suggests that for n large, from (6.9)

[n/(a+l)]
' " (n
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
6.1 : For all values of a, b, c, and n
~
~
( n
~ ar
r
r=O
(~)r b(n-ar)
-:----'<-~~_+1_
(a
(6.10)
+ 1) ~o ~ ab
in the region of convergence (6.6).

ProofofTheorem 6.1. Using (6.9) and (6.10) we may write
(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
Without loss of generality let c = ab, a E ~, in (6.7) such that

~a+l
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+r(a+ l)-k Cak - a(a+l)(+r)

~O
(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
which is identical to the right hand side of (6.l0), upon replacing -a = n.

The convergence region (6.6) is obtained by applying the ratio test to the term
in the sum (6.10). The theorem is proved. Some more details may be seen in
the publication [101).
0
4.
rp
The Lagrange Form

Jensen's work [69] is related to our sum (6.10). Jensen substitutes
= (z + I t and f = (z + l) into Lagrange's formula, Theorem 2.4, and
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
lengen also substitutes cjJ = (z
(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
+~~ -.j) )} ,(6.14)

.7
equatingthe coefficients ofyj in(6.13) and (6.14) and substitutingm

lensen obtains the striking result
= 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.
Central Binomial Coefficients
For the case of a = 1, (6.7) produces the dominant zero ~o = b+~

such that for n = -a E ~,c = -x and b = 1 identity (6.10) may be written as
I(a,x) =
f( 2r+a-l )x
r=O
2Fl [ ~,:tl
and in particular I (1, x) = Jl
viI - 4x
(1+JI=4X)~~.~6)
2
4X]
:'4x occurs in the book of Wilf [111],
I (a, 1/8) = 2(3a-2)/2 (1 + vI2) l-a
117
v'2
is a very slow converging series. Replacing x by -x in

and f (1,1/8) =
(6.16) and then adding we obtain
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') .
Therefore, in general, we may obtain cIosed form expressions for binomial

sums of the type
for constants al,a2 and 0:. It also follows from (6.16) that for
have the identity
22 F1[
or
~,:tl 4X] = (l+Jf=4x) 2Fl[

1
0:
and E 3{ we
a;:,:]+
4X]
t, (
2r + ; - 1 ) x r
=(1+v'1-4X)~(
2r+0:: -l
)x
r.
(6.18)
For 0: = 2 and = 0 adding alternative terms in (6.18), we have , after using

(6.17) and some minor manipulation
From (6.16) with 0: = 1 and collecting coefficients of x we obtain the comrnon

result
(~) = (r~ ~ ) 2 2r and this result will be generalized in the next chapter.
118
The identity (6.16) may be differentiated with respect to x to obtain more

identities. If we let the operator p = x d~ f (a, x) we obtain from (6.16)
t,r (2r+;
-1
)x' ~ xc:~ra {a+ ~}6.19)

(1
0+31 4x] .
+ a) x 2F l
0+2
[ -2-' -2-
a+1
Operating on (6.19) by p again we have that
and in particular for a
00
= 1 and x = 1/8 ,L: r 2 (~) 8- r =

r=O
2~' Thus we
may, in general obtain identities of sums of central binomial coefficients;
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
+ l)j is Pochhammer's symbol.

er =
For j = 1, the constant
(2r)_1
r r+1
is known as Catalan numbers. In particular
93 (x)
= 12~x3 {1- 10x + 30x 2 -
93 (-x) =
12~x3 {-1 -
(1 - 4x)5/2} ,
lOx + 30x 2 + (1
+ 4x)5/2} ,
119
+ 93 ( - x) produces the identity
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 -
= 2 then from (6.16) we have the
0:
1 41;]
(1 _ 4x)5/2} .
1 - 4x + VI
4x
(6.20)
Multiplying both sides of (6.20) by x and then integrating, we have
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
= 1/4 we have the slow
= 2.
r=O
An alternative procedure for integrating (6.16) for 0: = 1, is the following. Take

the first term of (6.16) to the right hand side and integrate j-times such that
hj (l,x) =
~
00
27') (1')
1 JX ... lXI (1
vr=-:u - 1) Tdt
xr
l'
r-l
= xj-l
t=O
1=0
j - times
~
and consideration needs to be given to the improper integral. In particular
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
and adding produces the identity
~ ( ~~ ) (2r) (2r :7) (2r + 2)

=
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
Now, ifin (6.16) we put x
= x 2 and for (}: = 1 integrate, we have
00 (
f (1 x2) = '"'
1,
1=0
2r
2r+1 = areSln
. 2x
_x__
2r + 1
2'
and integrating onee more
(1 x 2)
=~ (
~
2,
2r )
x 2r +2
(2r + 1)(2r + 2)
In particular h (1, ~) = ~ - 1, h (1, ~)

these two cases produces the identity
?;
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 (
).
Similar identities may be obtained for other values of a. If for example a

b = 1, c = -2/27 and n = -(}: E ~ then, from (6.10)
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
form. Consider the funetion
f (x)
121
= j~~:~, then a Taylor series expansion
of f (x) about the origin allows us to write
f (x) =
aresinx
Jf=X2
(2x)2r-l
r=l r (
2; )
(6.21)
and
and
upon using the relation aresin z = -i In
(iZ + ~) . Multiplying hoth
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
Differentiating (6.22) with respeet to x, and simplifying we obtain

(6.23)
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
novsky and Chudnovsky [29], as an example they give r~
(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 -
from which upon manipulation we obtain
((2)
3 '
(6.24)
123
Integrating (6.24) three more times gives us
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
for j = 1, 2, 3, . .. and eonstants
- \0:, so that in general we may obtain identities
I:
1'=1
+ 4) arcsinx
Oe
and . Comtet [34] obtains
= 1771"4 =
r 4 (2;)
3240
5 F4 [
2'
131,1,1,1
2,2,2,2
I~]
00
however there appears to be no closed form expression of r~l
rm
r)
for m
> 4.
We ean obtain identities of this form, so that from (6.25) with x = 1/2, we
have
Other identities involving ( functions and multiple ( funetions are given by

Borwein [14]. In the next section we shall investigate the finite sum (6.9).
We shall , for a = 1, give a trigonometrie representation of (6.9), recover and
extend some results given by Binz [10], and highlight a number of interesting
applications ofthe Fibonaeei sequenee.
6.
Fibonacci, Related Polynomials and Products

Consider (6.9) for a = 1, such that from (6.1)
in+1 - bin -
Cfrt-l
= 0, io = 1
(6.26)
and
(6.27)
124

n
In
b
b2
b3
b4
b5
b6
b7
b9
+ 3b2 c + c2
+ 4b 3 c + 3bc2
+ 5b4 c + 6b2 c2 + c3
+ 6b5 c + lOb3 c2 + 4bc3
bl l
11
b" 2
+ lOb 2 c3 + c4
+ 8b7 C + 21b5 c2 + 20b3 c3 + 5bc4
biO + 9b Bc + 28b6 c2
10
Table 6.1.
+ 2bc
bB + 7b6 c + 15b4 c2
12
+c
+ 35b4 c3 + 15b2 c4 + c5
+ lOb9 c + 36b7 c2 + 56bS c3 + 35b3 c4 + 6bcS
+ llblOc + 45bBc2 + 84b6 c3 + 70b4 c4 + 21b2 c5 + c6
Polynomials ofthe finite sum (6.27).
Some polynomials of in are given in Table 6.1

The reeurrenee (6.26), for some parameter values band c may be identified
as shown in Table 6.2.
Horadam [67] and [68] has recently written alueid and pertinent examination
of Jaeobsthal representation numbers and polynomials, and funetional forms.
Hendel and Cooke [63] suggest that seeond order reeursions may be represented
by finite produets involving trigonometrie funetions. From (6.27) we may thus
write
125

b
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
Special reeurrenees and solutions.
and so after multiplying corresponding factors
bn - 2[n/2] J1 {b 2 + 4ccos (n~ 1) } .

[n/2]
In
We can see that for b = 0 and n even
In = (4ct
rr
J=1
and since from Table 6.1 we note that
rr
j=1
cos
(2r~~ 1)
In = (''' we may dcduce
(2n7rj+ 1) = 2-".
The characteristic function (6.7), for a
2~O,1
cos 2
= 1, has two zeros
= b Vb 2 + 4c
and from (6.27)
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-
~t ~ 4-" ~ (-1)' ( 2: ) (1 + x)"/'.
Subtracting (6.30) from (6.29) we have
(6.30)
127
and to identify (6.31) with (6.28) let 2f.1 = n
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.
From the left and right hand sides of (6.33) let n =
?;
[(2m+l)j2]
(2m
+r 1 - r )
Expanding the right hand side and collecting powers of x we have
[(2m+l)/2]
{; (2m +r1- r )
=
(~r
T(,,"+l)
~,"
{t, (;7:1
2 ) (
~)}
and equating coefficients of x we have the novel identity
(~;: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
The ratio of (6.34) and (6.32), after putting 2m
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
= y'f+X, which solves Krafft and Schaefer's problem after
replacing 1 + x with x*.

It is of some passing interest to note Wilf [111] derived (6.27) for c = -b.
The derivation, by Wilf, counts the number of words of n letters over an alphabet
of b letters that do not contain the substring of a word of a letters. Graham et
al. [55] discuss the continuant polynomial, K n (Xl, x2, ... , Xn), defined by
In particular, a Morse code sequence of length n that has k dashes, has n - 2k

dots and n - k symbols altogether. These dots and dashes can be arranged in
( n-k)
k
ways; therefore if we replace each dot by z and each dash by 1 we
get
t(
K n (z, z, ... , z) =
~r
) zn-2r
r=O
which is the Fibonacci sequence. There is another intriguing connection of

the Fibonacci sequence with the Lambert series. The series L (x) =
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
= =
f ~ = J5 {L (3 -vs) _L(7 -3J5)}.

j=l12j
Another lovely result conceming the summation of recipricals is given by

Melham [81].
Consider the sequence (6.26)
in+l - bin - in-l
= 0,
io
= 0, h = 1

and
gn+1 - bg n - gn-1
= 0,
go
= 2,
129
= b.
gl
The Binet forms of 1n and gn are

171
where
2a
o:n -
= 0:. and
= b + Vb 2 + 4,
9n = a
= b-
+ ,
Vb 2 + 4
and a = -1 for a > 1 and -1 < < O.

For k and m odd positive integers, the following sums are valid
and
l
:-n=l gkngk(n+m)
00
1
(a - ) gkm
[2L ( 2k ) - 4L ( 8k ) -
_1
1
.!km n=l akngkn
1'
where L (x) is the Lambert Series.

In the next section we will prove a number offunctional forms of (6.9).
7.
Functional Forms
The following lemmas are functional forms of (6.27).
LEMMA 6.1 Let n be defined by (6.27),
1-1 = 0 and 1-n = (_I)n cl-nj~t_2,

then
l: (_c)k 1n-2k = O.
Tl
(6.35)
k=1
Proof From the left hand side of(6.35)

n
2:: (_c)k 1n-2k = -C1n-2 + c21n_4 - c 1n_6

3
k==l
+ ... + ( -c )n-l 1-n+2 + ()n

-c 1-71
130
and by the definition
= -Cfn-2 + c2fn-4
+ ... + c3fn-6
- c3 fn-6
- c2 fn-4
+ Cfn-2
= 0
and the proof is complete.

LEMMA
6.2
bm fn =
f (~ )
J
j=O
(-c)j fn+m-2j, for m
Proof We shall employ an induction argument. For m

m = 1 we obtain (6.26). Consider
bm+1 In
(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
and from (6.26) substitute for bfn
= 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,
( :
collecting coefficients of ( - c)j gives us

fn+m+l
f (m:j=O
1 ) (-c)j fn+m+1-2j
= m+l
~
1=0
and the lemma is proved.
+ (_c)m+l
m;
1 )
fn-m-l
(-c)j fn+m+1-2j
131

LEMMA 6.3
= '""
L.-t (-c) k fu+v-2k
fufv
k=O
Proof From the Jeft hand side

(bfu-l + Cfu-2) fv
u-l
b (_c)k fu-l+v-2k
k=O
fufv
2.:=
u-2
+ c 2.:= (_c)k fll-2+v-2k

k=O
1l-1
= 2.:= (-cl (fu+v-2k
1l-2
- Cfu-2+v-2k)
k=O
k=O
u-l
+ c 2.:= (_c)k fu-2+v-2k
1l-1
L (_c)k fu+v-2k - CL (_c)k fll-2+v-2k

k=O
k=O
u-2
+c (-cl fu-2+v-2k,
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
2.:= (_c)k fu+v-2k + (-ct f -u+v = 2.:= (_c)k fu+v-2k

o

Two special cases are
= 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
2.:= fnfm-n = (rn + 1) fm + 2.:=
Proof From Lemma 6.3 put u

m
= 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 +
~ (-c)' Im-2k - k~l (-cl' fm-2k }
now apply Lemma 6.1, such that

m
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
and reapplying Lemma 6.1

m
[m/2]
n=O
j=I
L inim-n = (m + 1) im + L
(-c)j (m + 1 - 2j)jm-2j

LEMMA
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
also from Lemma 6.3 put u = m - 1, v = n - m - 1 hence

m-I
cim-dn-m-I
(_c)k in-2-2k
k=O
m-I
(_c)k+1 in-2-2k'
(6.38)
k=O
Adding (6.37) and (6.38), in = imin-m + cim-dn-m-I hence the lemma is

proved.
0
LEMMA
6.6
bm
= m-I
L(
j=O
m -:- 1
(-c)j im-2j'
133
Proof From Lemma 6.2 put m = m - 1 and n = 1, hence

rn-I
h -_
m-l(
L
m - 1
j=O
.and since
LEMMA
6.7
) (-c)
frn-2)
= b, the lemma is proved.

/ n2
/2
= f n-rn f n+rn + ()n+I-rn
-c
rn-I
Proof From Lemma 6.3 put u
=n =
fn-rnfn+rn
m, v
= n + m giving
n-rn
L (_c)k hn-2k.
k=O
Put n - k = n * and rename n *, giving for 0 ::; m ::; n

n
fn-mJn+rn =
(_c)n-k hkl
k=m
and specifically for m = 0,

n
f~ =
L (-ct- k hk.
k=O
Subtracting the last two sums produces
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
COMPUTATIONAL TECHNIQUES FOR THE SUMMATION OF SER/ES
LEMMA
6.8
Proof From(6.27) we have
and using Lemma 6.2

Mq
~ ~ ( p ~ r ) c"
{r, ( j
P
2r ) (-c); !p+q-2r-2; }. (640)
Now put r + j = a(constant) and equate coefficients of !p+q-2a in (6.40) and

the expression of Lemma 6.6, let p n, and put n 2m, rename the counter
and we have the result (6.39). A WZ certificate function, R (n, k), of(6.39) is
R (n, k) = k (2k - 3 - 3n) (2n - k + 1),

(n+l)(k-n-l)2
which proves that (6.39) is an identity.
Other functional identities of the Fibonacci sequence are given by Graham

et al. [55].
Chapter 7
SUMS AND PRODUCTS OF BINOMIAL TYPE
Abstract
1.
An arbitrary order forced difference-delay system is considered from which finite

binomial sums are generated. Z transform theory is then utilized to represent
the finite binomial type sums in c10sed form, moreover Zeilberger's creative
telescoping algorithm, Petkovsek's algorithm 'Hyper' and Wilf and Zeilberger's
WZ pairs method is used to certify particular instances of the identities.
Introduction
In this chapter we generalize the system of Chapter 6 and generate finite

binomial sums. We utilize Z transform theory to represent the finite binomial
sums in closed form and we can also employ Zeilberger's creative telescoping
algorithm, Petkovsek's algorithm 'Hyper' and Wilf and Zeilberger's WZ pairs
method to certify particular instances. Firstly we consider a homogeneous convoluted Fibonacci sequence and develop the general finite sum and its closed
form representation. By considering multiple zeros of an associated characteristic function we develop new identities and certify some of them by the WZ
pairs method. Secondly we generalize OUf results by considering forcing terms
of binomial type. These results and more have been detailed by Sofo [95] and
[96].
135
136
2.
Technique
Consider what we shall describe as a generalized, or convoluted, Fibonacci

sequence in, that satisfies
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
Z [in-kUn-k] = z-k F (z),
where Un -
is the discrete step function, we obtain
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
In series form, (7.3) may be expressed as

00
F(Z)=~
+r _
r
zaR+l
.
(za+l - bz a _ c)R
1) (z_b)R+r
zl-ar
er
(7.3)
(7.4)
and we may obtain the inverse Z transform of(7.4) such that
in (a, b, c, R)
[n+l-R]
= in
I:
r=O
(R +; - 1 )
Rn+-/~ 1 )
Gf bn-ar-R+l
(7.5)
Sums and Products of Binomial Type
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
- 1 in (7.10) and renaming n* as n, we have
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
,
then (7.3) has poles of order 2R. In
~(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
If R = 1, then (7.13) reduces to a result given on page 124 of Wilf's book

[111], namely
t, enr~
r )
(~1 = T2n (2n + 1) =
Q Cn~~ 1)'
sin'
(7.14)
where the trigonometrie produet is evaluated from the relation in Chapter 6.

Hence (7.13) is a generalization of(7.14). Utilizing Zeilberger's creative telescoping algorithm, described in [85] and available on 'Mathematica', we obtain
from the left hand side of(7.13) a recurrence in (R) that satisfies
4 (n + 1) (2n + 1) in+l (R) - (n
+ R) (2n + 2R + 1) in (R) = O.
(7.15)
Sums and Products ofBinomial Type
139
Iterating (7.15), we have that
rr
f (R) = 2-2n n - 1 (R + j) (2R + 1 + 2j)
(7.16)
(1+j)(1+2j)
j==O
.n
so that from (7.13) and (7.16) we obtain
Further results may be obtained as folIows. Differentiate (7.11), for R = 1, and

its trigonometrie representation with respeet to c, then substitute c = - (b/2)2
and simplify such that
(7.18)
-fI
sin 2
j==l
(~)
2n + 1
cot 2
k=l
(~)
.
2n + 1
From 'Mathematica', a recurrence relation for f~ (1) in (7.18) is
4n (2n -- 1) f~+l (1) - (n
+ 1) (2n + 3) f~ (1)
= O.
(7.19)
(_I)r = _2-2n rr (1 ~ j).(3 + 2j)
(7.20)
Iterating (7.19) and using (7.18) we have
~r
r=l
2n - r )
r
n -1
. ) (2) - 1)
J=l
and eomparing (7.18) and (7.20), we have
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 )
+(
To further illustrate the technique, from (7.13) and (7.15) with R
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
and using result (7.13) we have that
r 2n (2n + 1)2 +
ii:
sin 2
)=1
(2n1l'~ 1) t
(2n1l'~ 1)'
cot 2
k=1
(7.23)
From (7.22) and (7.23) the identity
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
is obtained and rewriting we have, using (7.21), that
(2
n
rr
1)2 = 10 n - 1 (2 + j) (5 + 2j) _
j=1 (1 + j) (2j + 1)
From (7.21) and (7.22)
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)
Sums and Products 0/ Binomial Type

(n
141
+ 1) (2n + 1) (2n + 3)
~--~--~~--~-10
and from (7.18)
~r
(~1
( 2n r- r )
rr
n -1
(2 + j) (5 + 2j)
j=1(1+j)(2j+1)
= _ 2- 2n n
(~n2 -
1) .
Simi1arly, we can show that

(7.25)
2- 2n n (8n 4 - 20n 3 - lOn 2 + 5n
_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
The 1eft hand side of (7.25) satisfies the recurrence

4n (2n - 1) (2n 2 - 5n -
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
+ 112n4 + 140n3 105
+ 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
+ 1) (2j + 3) (4j4 j (2j - 1) (4j4 n (16n 6
+ 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
can be expressed as a polynomial in n of degree 2m for m integer. By the WZ

package on 'Mathematica' the identity (7.25) may be verified by the certificate
ftmction
V(n,r) = 2(r -1) (r -1- 2n)

2 2
3 2
( 12n4 (r - 1) - 8n (r - r + 3) + n (4r - 15r - 13) )
2
+ 2rn (6r - 7) + 6r - 5r + 1
x r(2r-1-2n)(r-1-n)(n+1)(2n+3)(2n 2 -n-5) .
Similarly for the identity (7.13), for particular values of R, and by the use of
the WZ package we may obtain a rational certificate function, V (n, r, R) that
certifies the identity, in particular
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:) =,~~
and therefore (7.3) and (7.9) may be modified
z2R+l
((z-~)2(z+i))
[::, {
(z - ~t F;Z)}].
/-t
VIPR (
R'
= 0, 1,2, ... , 2R -
1,
-D ='~~l [::. { (z + ~r F;Z) }] ,

v
= 0, 1, 2, ... , R -
143
hence from (7. 11 )
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)
= 1 and R = 2, we have respectively from (7.26) that

In (1)
t, (
3n; 2r ) ( ; :
3-(3n+2) {2 3n + 1 (9n
(7.27)
+ 4) + (-lt}
and
(7.28)
3-(3n+2) { 23n +2 ( 3n: 1 )
:+
(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
+ 52) j~l+l (1)
- 8 (3n
+ 7) In (1)
= 0,
23
10 (1) = 1, h (1) = 27
and
+ 5) (3n + 4)2 In+2 (2) - 27(189n 3 + 1440n 2 + 3399n

+2348)ln+l (2) - 8 (3n + 7) (3n + 8) (3n + 10) In (2) = 0,
10 (2) = 1, h (2) = 127
729 (3n
144
5.
Other Forcing Terms
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
n /-L ) bR+/1 '

m-l-
(~)
lim [.!!!:...
2 = z---~
dzlJ, { (z _zRb)R }]
and
v!Pm,v (b) =lim [dd

z---b
In the case that R
t.(
ZV
zR 2R }].
(z _
~)
= 1, m = 1 and 2 respectively we obtain
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
For constants D:j and positive integer m we have that
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
SUMS OF BINOMIAL VARIATION
Abstract
1.
In this chapter the results of Chapters 6 and 7 will be generalized. By residue

theory and induction the author proves that infinite generated sums may be represented in closed form which depend on k dominant zeros of an associated
polynomial characteristic function. A connection between the infinite series and
generalized hypergeometrie functions will also be demonstrated.
Introduction
In this chapter we shall generalize the identities of Chapters 6 and 7. We

shall consider the infinite form of the sum (7.5) and develop a procedure for
representing it in closed form. We will prove that the closed form representation
will depend on a dominant zero of an associated characteristic function. We will
also demonstrate a connection of the infinite binomial sums with generalized
hypergeometrie functions and some of its identities. We can then develop these
ideas further and generate infinite binomial type sums which we may represent
in closed form depending on k dominant zeros of an associated characteristic
function. Particular cases of our identities may be certified by the WZ pairs
method of Wilf and Zeilberger. We shall illustrate our theoretical results with
some numerical examples. Some ofthese results have been published in [97].
2.
One Dominant Zero
Ifweconsiderthe system (7.5) ofChapter 7, with c

algebra more manageable, we obtain
F(z)
with
(z - b - bz- a)
9 (z)
= za+l
zaIHl
---;0::
(g (z))R
- bz a - b.
147
= btomake the following

(8.1)
(8.2)
148
Now 9 (z) has a + 1 distinct zeros ~j, j
= 0, 1,2, ... , a, for
(a + 1t+ 1
(abt
< 1,
(8.3)
and from residue consideration

[n+l-RJ
~
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
~ 1~ (R, n, a, b) ~ ~ Qn," ((0) (
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)
F (z) is given by (8.l) and
Tr(R,n,a,b)
= (r+~-l) (r:~a~l
)bn-ar-R+l.
(8.7)
We mayaiso note, from (8.4) and (8.5), that

[ n+l-RJ
a+l
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).
Sums ofBinomial Variation
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 -
Proof Putting J-l =

from the definition,
where
ck
<"0
1, with QR,O
a_ 1
R+l,ll- k
(C)
(8.8)
<,,0 ,
(~o) = ((a+l~~o-ab) R .
into (8.6) we obtain the expression for QR,O (~o). Also
h (z) = (a (z - ~o)
+ z) 9 (z) za+l
z (z - ~o) g' (z)
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
h(z) =~(-1)j(~0-b)(j~1)(z-~oF-2 (a+ j -1)

(z - ~of ~
~6
a- 1
= B (z).
Wenow have
150
where B(k) (z) = ~B (z), and
lim [dd: B (z )]
z
z-->o
= lim
z-->o
[f:
.
J=2
(-l)j ({o - b) (j -l)k (z - {0)j-2-k ( a + j - 1 )]

(:J
a- 1
~o
= (-l)k({o-b)(k+l)!
{~+2
(a+k-l)
a- 1
and (j - 1)k is Pochhammer's symbol. Hence we can write
(J-L
+ l)!QR,J.I+l ((0)
= R
~( ~ )
(J-L - k)!QR+l,J.I-k ((0)
(-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
where T r (R, n, a, b) is given by (8.7). A recurrence relation
(a
+ 1) bdb SR -
abRSR+l - (n
Proof The proof is as folIows.
+1-
R) SR
0/ (8.9) is
= O.
(8.10)
Sums 0/ Binomial Variation
151
J1
J1!QR,/, (~o)
AR
~Ii
aR (A - b) ~R-J _ aR (a + 1) (~o - b) ~R-J

2AR+J
0
2AR+J
0
R~R-2
~2A~+2 [A 2 (3aR - a -
8) - 2bA (3aR
a8~~i:33 [A 3 {a 2R (R -
4R)
c10sed form (8.5)
~~+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)
The Q values ofthe discrete case.
4)
+bA 2 {-a 2R(3R+ 1) + 2a(8R+ 3) -12}

+b 2A {a 2 (3R + 2) (R + 1) - 8a (R + I)}
_a 2b3 (R + 1) (R + 2) )
rahte 8.1.
+ 2a (1 -
1)
+a -
70a
+ 15ab2)
+ 12) + b2 Aa (70a
Closed form ofthe discrete case.
- 40) - 30a 2b3 )]
152
and
t, ( ~
SR+l
R ) (
~~~ )
(a+1)~
bR
bn - ar - R
~rTr(R,n,a,b)+
(n+1-R)
bR
SR
From the left hand side of(8.10)
(a+ l)b
[-~ t,rT.(R,n,a,b) + (n + ~ - R) sn]-
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) ~oQR,JL (~o)
+ (J-L
- 1) AQR,JL-l (~o)
+ (A + ab) ~o d~o QR,JL-l (~o) (8.11)

where QR,JL (~o) is defined by (8.6).
Proof From (8.6)

(8.12)
where 9 (z) is defined by (8.2) and ~o is the dominant zero of 9 (z) . From (8.12)
we can differentiate with respect to b such that
153
Sums of Binomial Variation
where A = (a
obtain
+ 1) ~O - ab. Simplifying (8.13) by adjusting the third term, we
-R
/Ll db QR,JL (~o) = -b-/L1QR,JL (~o)
+ b/L1QR+l,JL (~o)
R~o. [dJL {(Z - ~o)R-l za(R+l)

- - 11m bA z-->eo dzJL
gR+l (z)
x
Let h (z) = ~Og (z) z-a - A (z dominant zero ~o, gives us
~o) 2
h (z) = (z -
(~Og(Z)Z-a-A(Z-~o))}].
a Taylor series expansion about the
~o),
f(
-1)j ( a + ~
- 1 ) (b -
j=2
(8.14)
~o) ~~ - j (z - ~o)j - 2
and substituting into (8.14) we have
d
/Ll db QR,JL (~o)
-R
= -b-/L1QR,JL (~o) + b/L1QR+l,JL (~o)

R~o.
- - hm
bA z-->eo
[d JL {(Z - ~O)IHI za(R+l) }]

B
dzJL
gR+l (z)
J
(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.
Expanding (8.15) by the Liebniz differentiation rule
-R
/L! db QR,JL (~o) = -b-/L!QR,JL (~o) +
~~o
t (~ )
(/L - k)!QR+l,JL-k
k=O
and after evaluating !im
z-->Eu
(~o) }~~o BY)
(8.17)
BY) from (8.16) and substituting into (8.17) we obtain
b d
Rdb QR,JL (~o) = QR+l,JL (~o) _ (b -
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 -
1) Qn,J.!+1 (~o) (8.19)
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
which upon rearrangement gives
d
db Qn,J.! (~o)
~o
= bA (A + ab)
{abARQn+I,J.!+l (~o) - /-LAQn,J.! (~o)

- ab~o (R - /-L - 1) Qn,J.!+l (~o)}.
Now, since
d~o Qn,J.! (~o)
(8.20)
Ab d
db Qn,J.! (~o)
= ~5
substituting into (8.20) and rearranging we obtain
abRAQn+l,J.!+l (~o) = ab (R - /-L - 1) ~oQn,J.!+l (~o) + /-LAQn,J.! (~o)

d
+ (A + ab) ~o d~o Qn,J.! (~o)

which is the required relation (8.11) after replacing /-L with /-L - 1.
Now we give an induction argument for the right hand side of (8.5).
Sn+l = a:R {(a + 1) b:b Sn - (n

d
also, db8n = ~ dl,o Sn and
+1-
R) Sn} ,
(8.21)
Sums
0/ Binomial Variation
155
Substituting into the right hand side of (8.21), we have
R+l
= _1_ { (a + 1) b~o2 R-l

'" (
71
abR
bA
~
R - 1/L=O
t-n-R+l+/1
P,
<"0
(d~odQR,/L (~o) + (71 + 1 +~o p, - R) QR,/L (~o) )
~ (n + 1 - R) ; , (
_
1
- abRA
R-l (
n
R - 1 - p,
~_ ~ ) ~~-R+l+"QR," (~o) }
R_
)
n-R+1+/L
~o
+ p,AQR,/L (~o) + ab (71 + 1 + P,
d
(A + ab) ~o d~o QR,/L
R) QR,/L
(~o)
(~o) }
~ ab~A { ; , ab (R ~ I') ~o ( R ~ I' ) ~~-R+"QR," (~o)
+~(
~
/L=O
n
) C-R+l+/L
R-I-p,
0
x ((A + ab) ~o d~o QR,/L

In the second surn renarne p,*
(~o) + p,AQR,/L (~O)) } .
= p, + 1 (and let p,* = p,), so that we may write
~ ) ~ ~g-RQR,o (~o)
1
R-l
+ abRA { ~ ab (R - p,)
71
R _ p,
n-R+l+p
~o
QR,/L
(~o)
156
= (
~ ) ~3-RQ1,O (~o) QR,O (~o)

+ ab~A ~ (
~ ~ ) E~-R+"{ abeR - ~)EOQR," (Eo)
+ (A + ab) ~o d~o QR,/1-1 (~o) + (J-L -
1) AQR,/1-1
(~O)}
= ( ~ ) ~3-RQR+l,O (~o)
+ ab~ ~ (
~ ~ ) E~-R+"{ab(R - ~)EoQR," (Eo)
+ (A + ab) ~o d~o QR,/1-1 (~o) + (J-L -
1) AQR,/1-1
(~o) } .
Using the relationship (8.11) of Lemma 8.3 reduces the previous line to
R
~ ) ~3-RQR+l,O (~o) + ~ ( R ~ J-L ) ~~-R+/1QR+l,/1 (~o)

R
~ ( R ~ J-L ) ~~-R+/1QR+1'/1 (~o)

o
which completes the proof of the conjecture.

~o
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
R = 1, 2 and 3, n = explicitly that
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
Let T1' (R, n, a, b) be defined by (8.7) and

To (R, n, a, b) = ( R r~ 1 ) bn +1 The ratio of consecutive tenns
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-n-l R-n R-n+l
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) ]
,
~ U~~) k (lJTI) k (1~~) k'" (R+~~~-l) k
= 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
where (x)m is Pochhammer's symbol and
(a, b)
=-
(a
+ 1t+ 1
(ab t
Some particular results of (8.22) are worthy of a mention, since it may be

shown that (8.22) reduces to known hypergeometrie functions. For a = 1 and
0: = -n E ~\J- then
00
= To
1+
L (;!~~
k 2
ft (R +
0: - 1 +
j=O k-1
'
TI
k=l
j)]
(8.23)
(o:+j)
j=o
the Gauss hypergeometric series (8.23) mayaiso be written as
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
T 02n - 1r (0:) {-8 (1, b)}(1-n)/2
X{I - 8(1, b)} -R/2 pA=~ {{1 - 8(1, b)} -1/2} ,
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-
since the parameters in the hypergeometrie function

2R + 1 2R + 3
3
-----+-=1
4
4
2
'
then from Kummer's identity we have
Here the dominant zero ~o

are
= 2 (1 + J2) and some values of the infinite sum
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
=V: i" ) e{{-R+l/2
T01T2(I-2R)/2
and some values of the infinite sum are

R=
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
As in Chapter 7, if in the dynamical system we consider forcing terms of the

type
Wn
= (
n
) bn +1- R - m
m+R -1
'
then the result is the identity
~
~
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
v!Pm,v (b) =lim
z-+b
[d/1
F (z)
-d { (z - ~o) R - }] ,
z/1
Z
[~v {(Z _ b)m F (Z)}]

dz
bn - m +1+v
Sums
0/Binomia/ Variation
161
and
F (z) =
where
~o
zaIHI
(z - b)m (za+l - bz a - b)
is the dominant zero of (8.2). If R = 2 and m = 1, then
~(1';1) (:~~)
= b2a
(b:r
b2-n~+n (n ((a + 1) ~o - ab) + a (a + 1) (~o - b)),

(~o
for the degenerate case of a

identity
= 0, we obtain the interesting Binomial convolution
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
Products of Central Binomial Coefficients
In Chapter 6 we obtained identities of central binomial coefficients, we can

carry out a similar but brief examination here for the identity (8.5). From
Chapter 6, (6.10), and (8.5) putting R = 2, a = 1, n = -a, b = 1 and
c = -x then
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) =
~ (r~l ) (2;:~) (-xr
(1+4lli)
=
l-a
1 + 4x
(8.25)
{a - 1+
Jf+.4X}
and adding (8.24) and (8.25) we have

2 ~ ( 2r + 1 ) ( 4r + a ) x2r
~
r=O
= a 4F3
=
+
(8.26)
2r+ 1
2r
ili'
a+l a+2 a+3 a+4

[ -4-' 1 4 9:'
-42' 2' 2
(1+vf4Xr-
1 - 4x
(1+4lli)
l-a
1+4x
{a
{a
-1
I16x 2]
1
+ VI -
4x
-1
J1+4x
}.
From (8.24)
and collecting coefficients of x r we have the result
which we shall generalize shortly. Integrating (8.26) we have
~(
~
r=O
2r + 1 ) ( 4r + a ) x 2r + 1
2r
2r + 1 2r + 1
=
ax 4F3
a+l a+2 a+3 a+4

[ -4-' 3 4 a ' atl' -4-
2' 2'
-2-
I 16x2]
163
dividing by X and integrating, and performing this operation again we have
~
00
0:)
2r + 1 ) ( 4r +
2r
2r+l
x 2r + 1
(2r+1)3
=O:X6Fs
1 1 a+l a+2 a+3 a+4

2'2'242'2\~'atl'-4-
2'2'2'2'
I16x 2] .
Differentiating (8.26) will produce other identities, as well as putting a lgher

value of R. Retuming briefty to the relation (8.27) we will demonstrate that this
is a special case of a more general relation. From (8.24) with general Rand
= 1 we have
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
where the constants
AR,j
r - J
.
J ) 22 (r-j) A . (8.30)
R,)
are the coefficients of x in the expansion of
some values of AR,j are given in the Table 8.3.

From (8.30) for R = 1, we obtain the common result
for R = 2 we obtain the result (8.27) and for 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
The constant lambda of (8.18).
~5 )
~: 55 )
= ( r +r9 / 2 ) 22r
+20( r+7/2 )22r-2+30( r+5/2

) 22r r-2
r-l
4.
165
3.
Multiple Dominant Zeros
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)
and upon inversion

[ n-k+l
In
(~. (k~;:~l
(8.33)
Also from (8.32) we may write
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)
has exactly (a + 1) k distinct zeros ej,u for v = 0,1,2, ... , k - 1 and

v = 0, 1,2, ... ,a. From (8.32), then
k-l
in =
L L Q (ej,v) e1u where
j=O 1.1=0
(8.36)
166
.
hm
z-+<Ej,v
[ (z -
F(Z)]
~j,v)-Z
~j,v
(8.37)
k (~j,v - b)k-l ((a + 1) ~j,v - ab)

and from (8.33), (8.36) and (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
wh ich is convergent for all values of k, n, a and b in the region of convergence

(8.35), then
=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)
ProofofTheorem 8.1. The series (8.38) with (8.41) gives us

[ n-k+l]
(a+l)k
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
j=O k (~j,O - b) - ((a
+ 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
For each ofthe j factors in (8.42) we may write
za+l
F j (z) = -za-+-l---b-z-a.---b-e-27r- i-j I"-k
for j
a
(8.43)
= 0, 1,2, ... , k -1. The characteristicfunctionqj (z) in (8.42) has exactly
+ 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
ak (1' + 1) ) b71 - ak (r'+I)

k(7'+l)
k-l
0:71+1
-- L
j,O
j=O k ((a + 1)
O:j,O -
ab)
- bn
(8.45)
168
Now make the substitution n - ak = m in (8.45) giving upon simplification

00
'"
k)
m - a r
kr
+k
b-akr
b-mam+ak+l
k-l
= '"
f:o k (( a + 1) aj,O - ab)

J,O
bak
(8.46)
Newton's forward difference formula ofa function h (Xj) = hj at Xj is defined

as
f).kh j = f).k-1hj+l - f).k-1h j , k = 1,2,3, ...
and taking the first difference of (8.46) with respect to m results in, from the
left hand side
( m
r=O
akr ) b-akr _ ~ ( m - akr ) b-akr

kr+k
~
kr+k
+1-
r=O
~
(
~
r=O
m - akr ) b- akr . (8.47)

kr+k-1
Similarly from the right hand side of (8.46)
(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
and combining with (8.47) gives, after simplification
~
~
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
will now be considered. Let b be areal constant, a and k E N and z is a

complex variable.
THEOREM 8.2 (i). The function (8.2) has al least one and at most two real
zeros and a dominant zero, the one with the greatest modulus, ~o such that
~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
8.3 The characteristicfunction (8.42)

za+l _ bz ll
has a zeros on the region C : /z/

restriction (8.3).
be21rij/k = qj (z)
:s I1l~11 for each j
= 0,1,2, ... , k - 1 with
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
By Rouche's Theorem, see Tagaki [103], it is required that
Iqj (z) - A (z)1 :::;
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 :::;
8.4 The characteristic function gk (z) has ak zeros in the region
Iactll with restrietion (8.3).
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
Ia~ll In the region C,
have modulus bigger than
JB (z)J = I (za+l - bz a - be 21l'i j /k)kl
S{C~lr+b('~J+br
=
(a ~ 1) (2:: /) + 1
bk
(8.50)
Now
J9k (z) - B (z)J =
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
za (z - b). Furthermore, let
I(G (z))A~ - bk - (G (z) - Cj)kl
?; ( ~ )
k
= _bk -
S bk
(-lt cjG k- r (z)
~ ( ; ) b' [ C~ J C~\ +b)
~bk [1+ ~ (;) {C~J
c::n rl
(8.51 )
where M =
(a~l) a (2::/) >
0. By Rouche's Theorem, it is required that
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
which is satisfied by virtue of restriction (8.3). Therefore the characteristic

function (8.34) has ak zeros in the region C : JzJ ::;
modulus bigger than
Ia~ll The theorem is proved.
Ia~ll
and k zeros with
172
Consider as an example, a = 3, b = 10 and k = 6 such that restriction (8.3)

is satisfied and C : Izl ::; 7.5. The zeros of qj (z), are listed below, showing
that one dominant zero appears from each of the qj (z) , for j = 0, 1, 2, 3, 4 and
5.
qo (z)
ql (z)
q2 (z)
q3 (z)
q4 (z)
q5 (z)
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)
(a , - b)', tior J'
-ao (a, -b)
for j
< k2
=~
-aj_~ (a, -b); for j > ~
where j = 0, 1,2, ... ,k - 1.
4.1
Numerical ResuIts and Special Cases
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
The degenerate case, a
173
= 0, of(8.41) yields the result
[m-kk+lj
\;'
~
m
kr+ k-l
(8.52)
Le
mk - 1
k2
cosm
1fij (m+2)/k
)=0
7'
.)
and for k = 4, we have
["'-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
Tbe Hypergeometrie Connection
Consider the term Tr (k, n, a, b) of(8.39) with

'Tl
.1 0
(k , TL, a, b) -_ ( k TL_ 1
bn - k + 1 '
the ratio of consecutive terms

(a+l)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)
is a rational function in rand therefore the series S (k,

be expressed as a generalized hypergeometric function
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
Asimpleexampleshowsthat, from (8.40) and(8.41) fork 2, a 1, b

and n = -(} E ~, we have two distinct dominant zeros of(8.34),
~o,o
1
2x
= x-I
(1 + vI + 4x) , 6,0 = 2x (1 + VI - 4x)
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.
Non-zero Forcing Terms
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
in the region of convergence (8.35) where
v!Pm,v (b) =lim
z-+b
/1!QR,11 (~j,o)
[~v {(Z _ b)m F (z)}]
.
=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 ) .
For R = 1, k = 1, n = -0: E ~ and m = 2, from (8,53) we have
=
where
~o,o
(0: - a) ba - b2a
b2+o:e-o:
(~o,o
0,0
- b) ((a + 1) ~o,o - ab)
is the dominant zero which satisfies (8.34), and specifically for
(a,o:) = (1,0:) , 2~o,0 = b + Jb 2 + 4b, so that
Many other identities of this form may be attained by various manipulations,

one such result is
~(
~
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)
+ aly(n-l) (x + e) + ... + an-lY' (x + (n -
1) e)
+ any (x + ne) = O.
Cornptes Rendus du Congres National des Sciences, Bruxelles, pp.91-97, 1930.

[21] Bruwier, L. Sur une equation aux derivees et aux d(fferenees melees. Mathesis, Vol. 47,
pp.96-105, 1933.
[22] Brmann, B. Memoires de I'Institut, Vol.II, p.13, 1799. See also Dixon, L. ProC. London
Math. SOC. VoI. 34, pp.151-153, 1902.
[23] Cerone, P. On series involving zeros of transeendental funetions arising from Vo/terra
integral equations. Le Maternatiche, in print, 2002.
[24] Cerone, P., and Sofo, A. Summing Series Arisingfrom Integro-Differential-DifJerenee
Equations, J. Austral. Math. Soc. (Series B), Vol. 41, pp. 473-486, 2000.
[25] Cerone, P., Sofo, A. and Watson, D. Stirling related polynomials. Australian Mathernatical Society Gazette, VoI. 24, pp.66-73, 1997.
[26] Chaudhry, M. L., Harris C.M. and Marchal, W. G. Robustness ofrootfinding in single
server queueing models. ORSA Journal of cornputing, Vol. 2(3), pp.273-286, 1990.
[27] Chaudhry, M. L. On Computations ofthe Mean and Varianceofthe NumberofRenewa/s:
a Unified Approach. Journal ofthe Operational Research Society, Vol.46, pp.1352-1364,
1995.
[28] Chu Wenchang. Binomial Convolutions and Hypergeometric identWes. Rendiconti DeI
Circolo Matematico Di Palermo. Serie 11, TomoXLIIl, pp. 333-360, 1994.
[29] Chudnovsky, D. v., and Chudnovsky, G. V. Classifieation ofhypergeometrie identities
for 'Ir and other logarithms of algebraie numbers. Proc. Nat!. Acad. Sei., USA, Vol. 95,
pp. 2744-2749, 1998.
[30] Clausen, T. Ueber die Flle, wenn die Reihe von der Form ... ein Quadrat von der Form ...
hat. Journal Reine Angew, Math. 3, pp.89-91, 1828.
179
REFERENCES
[31] Cohen, M. E. On expansion problems: New classes o![ormulasfor the c1assical[unctions.

S.I.A.M. Journal ofMath. Anal., Vol. 7, pp.702-712, 1976.
[32] Cohen, M. E. Some c1asses q[generatingJunctionsfor the Laguerre and Hermite polynomials. Mathematics ofComputations, Vol. 31, pp.511-518, 1977.
[33] Cohen, M. E., and Sun, S. A note on the Jensen-Gould convolutions. Canadian Mathematical Bulletin, Vol. 23, No. 3, pp. 359-363, 1980.
[34] Comtet, L. Advanced Combinatorics. Dreidei Publishing Co. Dordrecht, p. 89, 1974.
[35] Conolly, B. Proposed problem 10638. American Mathematical Monthly, Vol. lOS, p.69,
J998.
[36] Cooke, K. L., and Grossman, Z. Discrete delay. distributed delay and stabili~v switches.
Journal ofMathematical Analysis and Applications, Vol. 86, pp.592-627, J982.
[37] Cooke, K. L., and van den Driessche, P. On zeros o[ some transcendental equations.
Funkcialaj Ekvacioj, Vol. 29, pp.77-90, 1986.
[38] Cox, D. R. Renewal Theory. Chapman and Hall Publishers, London, 1967.
[39] deBruijn, N. G. Asymptotic Methods in Analysis. North Holland, Amsterdam, 1958.
[40] deBranges, L. A proof o[the Bieberbach conjecture. Acta Math. Vol. 154, pp.137-152,
1985.
[41] Egorychev, G. P. Integral Representation and the Computation q[Combinatorial Sums.
Translations of Mathematical Monographs, Volume 59, American Mathematical Society, Rhode Island, 1984.
[42] Ekhad, S. B., and Majewicz, 1. E., A Shorl WZ-style Proof ojAbel:~ Identity, The Foala
Festschrift. Electron. J. Combin. Vol. 3, 110, 2. Research paper 16. 1996.
[43] Erlang, A. K. Telefon-Ventetider Et Slykke Sandsynlighedsregning. Mathematik
Tidsskrift B, Vol. 31, pp.25, 1920.
[44] Euler, L. De serie Lamberlina plurimisque eills insignibus proprietatibus. Acta Acad.
Sci. Petropolitanae, pp.29-51, 1779; 11(1783).
[45] Euler, L. Opera Omnia. B. G. Teubner, Leipzig, Serie I, Vol. 6, pp.350-369, 1921.
[46] Euler, L. Opera Omnia. Ser.J, Vol. 2, p.459. Speciman de usu observationum in Mathesi
pura. B. G. Teubner, Leipzig, 1913,
[47] Feiler, W. An Introduction
Sons, New York, 1971.
10
Probability Theor)' and its Applications. Jolm Wiley and
[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
[50] Frappier, C. A Generalisation of the Summation Formula

Soc., Vol. 59, pp. 315-322,1999.
~f Plana.
BuB Austral. Math.
[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,
series and products. Academic
[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
[88] Pontryagin, L. S. On the zeros 0/ some elementary transcendental./imctions. American

Mathematical Soc. Trans. Series2, VoI. 1, pp.95-110, 1955.
[89] Pyke, R. and Weinstock, R. Special Infinite Series. American Mathematical Monthly,
VoI. 67, p.704, 1960.
[90] Rayleigh, L. Theory ~f sound. Dover, New York, 1945.
[91] Roy, R. Binomial identities and hypergeometric series. American Math. Monthly, VoI.
94, pp.36-46, 1987.
[92] Riordan, 1. Combinatorial iden/ities. lohn Wiley and sons Inc., New York, 1968.
[93] Slater, L.1. Generalised hypergeometric/unctions. Cambridge University Press, London,
1966.
[94] Sofo, A. A Closed Form Representation 0/ Binomial Sums and Series. Le Matematiche.
VoI. LlV, pp. 175-186, 1999.
[95] Sofo, A. A Convoluted Fibonacci Sequence Part I. Austral. Math. Gaz. Vol. 26, pp.
66-72, 2000.
[96] Sofo, A. A Convoluted Fibonacci Sequence Part 11. Austral. Math. Gaz. VoI. 26, pp.
107-114,2000.
[97] Sofo, A. An Extension ~fJensen 's Form. Bull. Austral. Math. Soc. Vol. 63, pp. 279-298,
2001.
[98] Sofo, A. Summation o.fSeries via Lap/ace Transforms. Revista Matematica Cornplutense,
in print, 2002.
[99] Sofo, A., and Cerone, P. Generalisa/ion 0/ a Waiting-Time Relation. lournal of Mathernatical Analysis and Applications, Vol. 214, pp. 191-206, 1997.
[100] Sofo, A., and Cerone, P. Generalisa/ion
Vol. 58, pp. 359-371,1998.
0/ Euler S Identi/y. Bull. Austral. Math. Soc.
[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.
NucJear Science and Engineering, VoI. 7,
[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.
About the Author
De. Anthony Sofo is an Associate Professor in Mathematical Sciences at

Victoria Univeesity, Melbourne, Australia.
He is interested in function theoretic methods and in mathematical inequalities.
He serves as an editor in the electronic Journal of Inequalities in Pure and
Applied Mathematics http : / / j i pam. vu. edu . au.
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
Cassini 's idcntity, 133

hypergeometrie funetion, 159
Abel,57, 115
Abel type series, 39, 90
analytie eontinuation, 70
Bnnann's Theorem, x, 31, 33, 36, 63, 97
Applieation, 36, 67, 68
proof,64-67
properties, 64
remainder, 68, 69
Bemoulli, I
Q-Bernoulli numbers, 11
generalised polynomials, 11
Bessel funetions, 7, 11, 13
Beta funetion, 24, 54. 126, 158
binomial, 161
eoeffieients, 14, 111, 116, 118, 161
eonvolution, 116
expansions, 126
summation, 14
sums, 12, 117, 147
sums and produets, 135
type series, 14
type sums, 73
variation, 147
binomial eonvolution, 116
Binomial theorem, 57
Bromwieh eontour, 32, 89
Bruwier Series, 43
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
Catalan numbers, 118, 119

Cauehy,57
integral, 63, 64, 67
produet, 11
187
188
funetional fonns, 99, 111, 124, 129

Gauss, 8, 13, 19
c1assical fonnula, 19
hypergeometrie series, 158
summation fonnula, 13
Gegenbauer, 13
geometrie series, 7, 12, 66, 68
Gessel, 14, 17
grazing system, x, 31, 50
Heaviside, 32, 47, 49, 74, 89
homogeneous,32,43,135
hypergeometric
closed fonn representation, 12
conneetion, 14, 173
fonn, 18
Gauss hypergeometrie series, 158
series, 12, 13
summation, 11 1
sums, 14
hypergeometrie function, 122
hypergeometrie function, 3, 8,12,26,147,157,
158
induetion, 2, 90, 97, 130, 147, 152, 154
integral equations, 4
inverse Laplaee, 32, 33, 88
Jaeobi, 13, 56
polynomials, 24
Jacobsthal, 124
Jensen, 32, 53, 55, 56,114-116
Jonassen and Knuth, 14, 16
Lagrange
fonnula, 114, 115
inversion, 55
theorem, 63
Lagrange-Bnnann, 63
Lambert series, 54, 128, 129
Laplaee transfonn, 4,7,31,32,42,74,88,102
Laurent expansion, 46
Legendre
funetion,7
inverse pairs, 24
polynomial, 13
Leibniz, I
rule of differentiation, 92
lethargy, 47
Maclaurin series, 75, 76, 82
Mathematica, 18,20,21,26--28,142
Mean Time To Repair, 48
moments, 5, 92
multiple zeros, 135, 138, 142
neutron behaviour, x, 31
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

Bok:978 1 4615 0057 5

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Bok:978 1 4615 0057 5

Загружено:

Авторское право:

Доступные форматы

Computational

Techniques for the

Springer-Science+Business Media, LLC

Library of Congress Cataloging-in-Publication Data

For their time and understanding

1. SOME METHODS FOR CLOSED FORM REPRESENTATION

Search Sum and Some Relations

Brmann's Theorem and AppIication

Differentiation and Integration

COMPUTATIONAL TECHNIQUES FOR THE SUMMATION OF SERIES

Multiple Dclays, Mixed and Neutral Equations

Neutron Behaviour Example

Ruin Problems in Compound Poisson Processes

Zeros of the Transcendental Equation

Cohen's Modification and Extension

Brmann's Theorem and Proof

4. BINOMIAL TYPE SUMS

Relations Between G k (m) and Fk+ 1 (m)

5. GENERALIZATION OF THE EULER SUM

The k-Dominant Zeros Case

6. HYPERGEOMETRIC SUMMATION: FIBONACCI AND RELATED

The Infinite Sum

7. SUMS AND PRODUCTS OF BINOMIAL TYPE

8. SUMS OF BINOMIAL VARIATION

One Dominant Zero

Multiple Dominant Zeros

Non-zero Forcing Terms

COMPUTATIONAL TECHNIQUES FOR THE SUMMATION OF SERIES

About the Author

COMPUTATIONAL TECHNIQUES FOR THE SUMMATION OF SERIES

SOME METHODS FOR CLOSED FORM

Identities play an important role in mathematics and have been a source of

COMPUTATIONAL TECHNIQUES FOR THE SUMMATION OF SERIES

was to find an exact value. Euler succeeded, eventually in writing

Koecher [75] gave

Ramanujan, see Berggren [8], evaluated

Bailey [4] and coworkers wrote, (originally given by Plouffe [86])

Amdeberhan and Zeilberger [2] published

Some Methodsfor Closed Form Representatioll

and Clausen [30] gave the result

Residue theory and contour integration can be gainfully employed to express

The residue evaluation ofthe integral

leads to the alternating sign identity

and also we may obtain

COMPUTATIONAL TECHNIQUES FOR THE SUMMATION OF SERIES

Use of Integral Equations

Cerone [23] considers an integral equation of the form

B (t) = <p ~~~ t)

where B (t) is a single sex deterministic model representing births at time t,

L V (Pj, x) ePjt , t > 0

= !e-UPi<p (u) du.

Same Methads Jar Clased Form Representatiol1

By allowing cp (t) to be exponentially constrained Cerone shows that

satisfies the recurrence relation

n=2,3,4, ... , (1.8)

are the n th moments of cp (t) . Now, in particular if cp (x) = ed (x - b) with e, b

By considering a partial fraction decomposition, such as

COMPUTATIONAL TECHNIQUES FOR THE SUMMATION OF SERlES

1fa (cot 1fa + coth 1fa)].

Taking the limit as a --; 0, in (1.10), confirms the result

Secondly, (1.2) mayaiso be integrated with respect to the parameter a. From

where the constant of integration in (1.12) is identically zero. The identity

the m j's are the zeros 0 f the