DELYRA - Solutions For Complex Calculus Mathematical Methods For Physics and Engineering

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 323

Mathematical Methods

for
Physics and Engineering
Volume 1S

Solutions for
Complex Calculus
Solutions for

Complex Calculus

Jorge L. deLyra
Copyright c 2018–2019 Jorge L. deLyra

Feedback to the author via


delyra@latt.if.usp.br

Translated from the Portuguese


by
M. A. S. deLyra

Translation revised by the author

ISBN: 978-1793016805

Version 1.1 (15Feb19)

QFTL
JLDL
Dedicated to Cida,
for all her support,
all her patience and
her constant company
through all these years.
Preface

This book contains detailed solutions of all the problems proposed at the
end of the chapters of the corresponding volume of the series of textbooks
“Mathematical Methods for Physics and Engineering”. The solution were
elaborated by the author himself, they are detailed and commented, and
many times constitute extensions or variations of the material covered in the
text. The set of problems proposed in the text should be considered as an
integral part of the text, since the best use of the text presupposes that the
reader does work at the problems. In this volume the problems are presented
and solved in the order in which they appear in the corresponding volume of
the text, and are organized sequentially in chapters, according to the chapters
of the text. The original propositions of the problems are repeated in this
volume, thus making it reasonably (but not completely) self-contained, and
the problems are numbered exactly as in the text.
The problems are of many different types, with varying levels of diffi-
culty and complexity. They include many things, from simple examples and
calculations used as reinforcement exercises, going through problems that
complete and extend the material presented in the text, and proceeding as
far as to include theorem proofs not given in the text. The most difficult prob-
lems, some of which extend significantly the content of the text, as marked
as “challenge problems”. In some cases the results of the problems are used
in subsequent parts of the text, and such problems are marked as “reference
problems”. In this way, this companion volume to the corresponding volume
of the original text can serve not only as a support asset to the teachers that
use it in their courses, but also as extended material for students who are
interested in extending and deepening their knowledge of the subject.
This is Volume 1S, which is dedicated to the complex calculus. It includes
117 problems with complete solutions, organized according to the 16 chapters
of the corresponding volume of the text [1].

vii
viii

Acknowledgements
The author would like to thank the contributions of the several teaching
assistants that acted in the courses on Mathematical Physics under his su-
pervision, as well as to the many students that took the courses, throughout
the years of development of this text, and that in a way played the role of
guinea pigs during that development. Without such able and willing test sub-
jects, it would not have been possible to develop the text to the considerable
extent which was achieved.
Many students and teaching assistants contributed by finding errors in
the text and in the equations, both in the text itself and in the solutions to
the problems, but the main contribution of the students is always to read
and use the text with a critical mind-set, as well as to ask intelligent and
challenging questions, of which there were many. This certainly contributed
to add both quality and scope to the text.
Contents

1 Number, the Language of Science 1

2 The Simplicity of Complex Numbers 15

3 Elementary Functions, but Not Quite 35

4 Even Less Elementary Functions 47

5 Geometrical Aspects of the Functions 63

6 Border Effects in Capacitors 75

7 Complex Calculus I: Differentiation 95

8 Complex Calculus II: Integration 109

9 Complex Derivatives and Integrals 137

10 Complex Inequalities and Series 151

11 Series, Limits and Convergence 169

12 Representation of Functions by Series 189

13 Convergence Criteria and Proofs 213

14 Laurent Series and Residues 223

15 Calculation of Integrals by Residues 231

16 Residues on Riemann Surfaces 281

ix
x
Solutions 1

Number, the Language of


Science

We present here complete and commented solutions to all problems proposed


in Chapter 1 of the text. For reference, the propositions of the problems are
repeated here. The problems are discussed in the order in which they were
proposed within the problem set of that chapter.

Problem 1. Consider the set I of the integers. Show that it does not
satisfy some of the 11 properties of a field of numbers, among those that do
not relate only to the operation of addition.

Complete Solution:

Since I is contained in R (I ⊂ R), both share the same arithmetic rules,


and R satisfies all the 11 properties, it follows that I will satisfy all those
involving only the integers. Thus we see that only the questions related to
the closure need to be analyzed in detail. Since the sum and the product of
integers produce other integers, the two operations are closed in I. It remains
to be seen whether the identity elements and the inverse elements are in I.
Since 0 and 1 are integers, the identity elements are in I. Since I includes the
negative integers, the inverse of the sum presents no problems. Therefore,
the only property that fails is the existence of the inverse for the product.

Problem 2. Consider the set Q of the rational numbers. Show that it


satisfies all the 11 properties of a field of numbers.

Complete Solution:

1
2 SOLUTIONS 1

Since Q is contained in R (Q ⊂ R), both share the same arithmetic rules,


and R satisfies all the 11 properties, it follows that Q will satisfy all those
that involve only the rational numbers. Thus we see that only the questions
related to the closure need to be analyzed in detail. Since I ⊂ Q and we have
already seen in a previous problem that for I the only property that does not
hold is the existence of the inverse of the product, it is enough to look at this
property. Since by definition Q is the set of numbers of the form p/q with p
and q integers (p ∈ I and q ∈ I), and with q 6= 0, we see that it was built in
such a way that any non-zero element has a multiplicative inverse. If we have
the element p/q with q 6= 0 by definition, and p 6= 0 so that zero is excluded,
then the inverse element is given by q/p with p 6= 0, which by definition is in
Q. Thus we see that Q, just as R does, satisfies all the 11 properties.

Problem 3. Show that Q is a dense subset of R, considering the integers,


the arithmetic averages of pairs of these numbers, the arithmetic averages of
pairs of the resulting numbers, and so on ad infinitum.

Complete Solution:

In order to show that Q is dense in R, we will show that there is no real


interval (x−δ/2, x+δ/2) around a real number x, of non-zero length δ, within
which there is no rational number. Let us do this by reductio ad absurdum,
imagining for a moment that there is such an interval, with δ > 0. If it exists,
then it certainly does not contain any integer numbers, since the integers are
rational.
Let us consider the integer q0− which is immediately to the left of the
interval, and the integer q0+ which is immediately to the right of the interval.
These two numbers form another interval which contains the interval  (x −
δ/2, x + δ/2). We have, of course, for the size of the interval q0− , q0+ , that
q0+ − q0− = 1. Let us now consider the number q = q0+ + q0− /2, which is
rational, since q0+ and q0− are rational and the  rationals form a field. This
− +
number is the midpoint of the interval q0 , q0 , and it is therefore contained
therein. This number cannot fall within the interval (x−δ/2, x+δ/2), because
this would violate the hypothesis on this interval. Therefore it must fall to
the left of the interval, between q0− and x− δ/2, or to the right of the interval,
between x + δ/2 and q0+ . One and only one of these two cases can be realized.
In the first case, let us define the new points q1+ = q0+ and q1− = q, so
that q1+ is to the right on the interval (x − δ/2, x + δ/2), while q1− is to the
left, and so these two new points form a new interval q1− , q1+ containing
the interval (x − δ/2, x + δ/2). Similarly, in the second case let us define the
NUMBER, THE LANGUAGE OF SCIENCE 3

new points q1+ = q and q1− = q0− , so that q1+ is to the right of the interval
(x − δ/2, x + δ/2), while q1− is to the left, and so these two new points form
an interval q1− , q1+ containing the interval (x − δ/2, x + δ/2). Whatever the
case may be, these two points form a new interval q1− , q1+ that contains
(x − δ/2, x + δ/2) and whose length is 1/2, that is, half the length of the
previous interval.
If we repeat this procedure indefinitely, we generate a sequence of intervals
(qn , qn+ ), for n ∈ {0, 1, 2, 3, . . . , ∞}, each nested into the previous one, whose

lengths are given by 1/2n , and all of which contain the interval (x − δ/2, x +
δ/2), which has a length δ > 0. However, in the limit n → ∞ we have
that 1/2n → 0, so that there certainly is some value of n for which the
interval (x − δ/2, x + δ/2) does not fit within the interval (qn− , qn+ ). This is
therefore absurd, because by construction all the intervals in this sequence
contain (x − δ/2, x + δ/2). It follows that there can be no real interval
(x − δ/2, x + δ/2), with δ > 0, containing no rational number. We also see
that, given a real number x and a distance ǫ > 0, there is always a rational
number at a distance less than ǫ = δ/2 from x.

Problem 4. Consider a regular polygon with N sides, in which the distance


from the center to a vertex is 1.

(a) Calculate the perimeter of the polygon as a function of N .

(b) Show that, in the limit N → ∞, the perimeter approaches the value
2π.

Complete Solution:

Consider the unit circle with an inscribed regular polygon, as illustrated in


the diagram of Figure 1.1. The complete angle 2π is divided into N equal
parts, so that we have θ = 2π/N .

(a) The perimeter of P the polygon is given by P = N ℓ, where we have,


by elementary trigonometry,
 
ℓ θ
= sin ⇒
2 2
π
P = 2N sin .
N
4 SOLUTIONS 1

θ = 2π
N 1

θ/2

ℓ/2

Figure 1.1: The unit circle with the inscribed regular N -polygon.

(b) In the limit N → ∞ the argument of the sine approaches zero. Using
the fact that

sin(α)
lim = 1,
α→0 α

we have that
N π
lim sin = 1 ⇒
N →∞ π N
P
lim = 1.
N →∞ 2π

We have therefore that, in the N → ∞ limit, P = 2π.

Problem 5. Consider the complex numbers defined as ordered pairs of real


numbers, and the definitions of the operations of addition and multiplication
on them. We call this structure C.

(a) Show that (0, 0) is the identity element of the addition and that (1, 0)
is the identity element of the multiplication.
(b) Verify that the complex numbers and the two operations satisfy all the
11 properties of a field of numbers.
(c) Verify that C reduces to the field R of the real numbers for the subset
of the complex numbers for which the second element of the ordered
pair is zero.
NUMBER, THE LANGUAGE OF SCIENCE 5

(d) Consider the complex number (0, 1), which is a very special case. Show
that (0, 1)2 = (−1, 0).

Complete Solution:

Since each complex number z is an ordered pair of real numbers (x, y), all
proofs are made by reduction to the corresponding properties of real numbers.
It is thus necessary to use the definitions of the two operations for complex
numbers. If we have that z1 = (x1 , y1 ) and z2 = (x2 , y2 ) then the sum is
given by z1 + z2 = (x1 + x2 , y1 + y2 ), and the product is given by z1 z2 =
(x1 x2 − y1 y2 , x1 y2 + x2 y1 ).

(a) In the case of the sum, making x1 = 0 and y1 = 0, we have that


z1 + z2 = (0 + x2 , 0 + y2 ), and therefore, using the properties of real
numbers, we conclude that z1 + z2 = (x2 , y2 ) = z2 . It follows that
(0, 0) is the identity element of the sum. In the case of the product,
making x1 = 1 and y1 = 0, we have that z1 z2 = (x2 − 0 y2 , y2 + 0 x2 ),
and therefore, using the properties of real numbers, we conclude that
z1 z2 = (x2 , y2 ) = z2 . It follows that (1, 0) is the identity element of the
product.
(b) Let us consider each of the properties in turn.

Operation of Addition:
Closure: by construction, the sum of two complex numbers is a
complex number, that is, an ordered pair of real numbers.
Existence of the identity element: as shown above, there is
a complex number (0, 0) which is the identity element of the
sum.
Existence of the inverse: given an arbitrary complex number
(x, y), we can construct the complex number (−x, −y), since
the real numbers are a field. Using the definition of the sum,
we have that (x, y) + (−x, −y) = (0, 0), thus showing that
(−x, −y) is the additive inverse of (x, y).
Commutativity: it suffices to reduce the problem to the use of
the commutativity of the sum for the real numbers,

z1 + z2 = (x1 + x2 , y1 + y2 )
= (x2 + x1 , y2 + y1 )
= z2 + z1 .
6 SOLUTIONS 1

Associativity: it suffices to reduce the problem to the use of the


associativity of the sum for the real numbers,

z1 + (z2 + z3 ) = z1 + (x2 + x3 , y2 + y3 )
= (x1 + [x2 + x3 ], y1 + [y2 + y3 ])
= ([x1 + x2 ] + x3 , [y1 + y2 ] + y3 )
= (x1 + x2 , y1 + y2 ) + z3
= (z1 + z2 ) + z3 .

Operation of Multiplication:
Closure: by construction, the product of two complex numbers
is a complex number, that is, an ordered pair of real numbers.
Existence of the identity element: as shown above, there is
a complex number (1, 0) which is the identity element of the
product.
Existence of the inverse: given the non-zero complex number
(x, y), that is, one for which we do not have that x = y = 0,
we can construct the complex number
 
x −y
, .
x2 + y 2 x2 + y 2
Using then the definition of the product, we have that
 
x −y
(x, y) ,
x2 + y 2 x2 + y 2
 
x (−y) −y x
= x 2 −y 2 ,x +y 2
x + y2 x + y 2 x2 + y 2 x + y2
 2
x + y 2 −xy + yx

= ,
x2 + y 2 x2 + y 2
= (1, 0),

thus showing that the number we have constructed is the mul-


tiplicative inverse of (x, y).
Commutativity: it suffices to reduce the problem to the use the
commutativity of the product for the real numbers,

z1 z2 = (x1 x2 − y1 y2 , x1 y2 + x2 y1 )
= (x2 x1 − y2 y1 , x2 y1 + x1 y2 )
= z2 z1 .
NUMBER, THE LANGUAGE OF SCIENCE 7

Associativity: it suffices to reduce the problem to the use of the


associativity of the product for the real numbers,

z1 (z2 z3 ) = z1 (x2 x3 − y2 y3 , x2 y3 + x3 y2 )
= (x1 [x2 x3 − y2 y3 ] − y1 [x2 y3 + x3 y2 ]
, x1 [x2 y3 + x3 y2 ] + y1 [x2 x3 − y2 y3 ])
= (x1 [x2 x3 ] − x1 [y2 y3 ] − y1 [x2 y3 ] − y1 [x3 y2 ]
, x1 [x2 y3 ] + x1 [x3 y2 ] + y1 [x2 x3 ] − y1 [y2 y3 ])
= (x1 [x2 x3 ] − x1 [y2 y3 ] − y1 [x2 y3 ] − y1 [y2 x3 ]
, x1 [x2 y3 ] + x1 [y2 x3 ] + y1 [x2 x3 ] − y1 [y2 y3 ])
= ([x1 x2 ]x3 − [x1 y2 ]y3 − [y1 x2 ]y3 − [y1 y2 ]x3
, [x1 x2 ]y3 + [x1 y2 ]x3 + [y1 x2 ]x3 − [y1 y2 ]y3 )
= ([x1 x2 ]x3 − [y1 y2 ]x3 − [x1 y2 ]y3 − [y1 x2 ]y3
, [x1 x2 ]y3 − [y1 y2 ]y3 + [x1 y2 ]x3 + [y1 x2 ]x3 )
= ([x1 x2 − y1 y2 ]x3 − [x1 y2 + y1 x2 ]y3
, [x1 x2 − y1 y2 ]y3 + [x1 y2 + y1 x2 ]x3 )
= (x1 x2 − y1 y2 , x1 y2 + x2 y1 )z3
= (z1 z2 )z3 .

Distributivity of the product with respect to the sum:


It suffices to reduce the problem to the use of the distributivity
for the real numbers,

z1 (z2 + z3 ) = z1 (x2 + x3 , y2 + y3 )
= (x1 [x2 + x3 ] − y1 [y2 + y3 ]
, x1 [y2 + y3 ] + y1 [x2 + x3 ])
= (x1 x2 + x1 x3 − y1 y2 − y1 y3
, x1 y 2 + x1 y 3 + y 1 x2 + y 1 x3 )
= ([x1 x2 − y1 y2 ] + [x1 x3 − y1 y3 ]
, [x1 y2 + y1 x2 ] + [x1 y3 + y1 x3 ])
= ([x1 x2 − y1 y2 ], [x1 y2 + y1 x2 ]) +
+([x1 x3 − y1 y3 ], [x1 y3 + y1 x3 ])
= z1 z2 + z1 z3 .

(c) Since we have that z = (x, y), it is clear that the subset z = (x, 0) is
identical to the set of the real numbers. It only remains to show that the
8 SOLUTIONS 1

operations between these numbers are reduced to the usual operations


with real numbers. If we have z1 = (x1 , y1 ) and z2 = (x2 , y2 ), making
y1 = 0 and y2 = 0 we have, starting from the definition of the sum,

z1 + z2 = (x1 + x2 , 0 + 0),

which is a real number given by x = x1 + x2 , which is therefore the


usual real operation of addition. In the case of the product we have

z1 z2 = (x1 x2 − 0 × 0, x1 × 0 + x2 × 0),

which is a real number given by x = x1 x2 , which is therefore, once


again, the usual real operation of multiplication.

(d) Directly using the definition of the product, for ı = (0, 1) multiplied by
itself, we have, making x = 0 and y = 1,

(0, 1)(0, 1) = (0 − 1, 0 + 0),

which is the real number (−1, 0). It follows therefore that we do have
that ı2 = −1.

Problem 6. Consider the complex field C. Find a representation of this


field by 2 × 2 real matrices, through the steps below.

(a) Find a 2 × 2 real matrix whose square is the negative of the identity
2 × 2 real matrix.

(b) Identifying the complex number (1, 0) with the 2 × 2 identity matrix
and the complex number (0, 1) with the matrix found above, write an
arbitrary complex number z = (x, y) in terms of these matrices.

(c) Write the arbitrary complex number z = (x, y) as a real 2 × 2 matrix,


and verify explicitly that this new representation preserves the form of
the arithmetic operations on complex numbers.

Complete Solution:

As a vector space, the complex numbers can be described in terms of a basis


of two elements, the real unit (1, 0) and the imaginary unit (0, 1). In terms of
NUMBER, THE LANGUAGE OF SCIENCE 9

real 2 × 2 matrices, it is reasonable to expect that the real unit is represented


by the identity matrix. It remains then for us to find a matrix representation
for the imaginary unit.

(a) One must keep in mind that the solution to this problem may very well
not be unique. The condition that a generic 2 × 2 matrix has as its
square the negative of the identity matrix is
 2  
a b −1 0
= ⇒
c d 0 −1
    
a b a b −1 0
= ⇒
c d c d 0 −1
a2 + bc b(a + d)
   
−1 0
= ⇒
c(a + d) d2 + bc 0 −1
a2 + bc = −1,
d2 + bc = −1,
b(a + d) = 0,
c(a + d) = 0.

Considering the four equations which determine the coefficients, from


the first two it can be concluded that a2 and d2 must have the same
value. Moreover, since these quantities are positive, and the right-
hand sides of these equations are negative, it is necessary that neither
b nor c be zero, otherwise it would not be possible to satisfy these two
equations. If we now consider the last two equations, which should
result in zero, since neither b nor c are zero, it is necessary that a + d
be zero, that is, d = −a. With this, only an equation remains to be
satisfied,

a2 + bc = −1.

Since we still have three unknowns, there are many ways to satisfy
this equation so that we can simply choose one, which is enough for
our purposes. In the interest of simplicity, and because of the fact
that the positive quantity a2 cannot contribute to generate a negative
number on the right-hand side, let us pick a = 0, which also implies
that d = 0. With this, we see that our matrix will have zero diagonal,
in a complementary way with respect to the identity matrix. Since the
10 SOLUTIONS 1

equation bc = −1 now remains, we have that c = −1/b. Again in the


interest of simplicity, we will choose c = 1 and b = −1, in such a way
that our matrix has for all its elements either zero or numbers with
absolute value 1, just as the identity matrix. The resulting matrix is
then
 
0 −1
,
1 0

which we will use to represent the imaginary unit. It is easy to check


directly that the square of this matrix is in fact the negative of the
identity matrix.

(b) Let us now consider an arbitrary complex number z = (x, y). In our
usual algebraic notation it can be written as z = x + ıy. Making
the identification of the real unit with the identity matrix, and of the
imaginary unit with the matrix determined above, we can write that
   
1 0 0 −1
z=x +y ,
0 1 1 0

which is a real 2 × 2 matrix representation of an arbitrary complex


number.

(c) Manipulating the expression above with the usual matrix rules, we can
write that
 
x −y
z= .
y x

Still using the usual matrix rules, we can now show that this struc-
ture gives us the correct way to add two complex numbers z1 and z2 ,
resulting in a complex number z = (x, y),
   
x1 −y1 x2 −y2
z1 + z2 = +
y1 x1 y2 x2
 
(x1 + x2 ) −(y1 + y2 )
=
(y1 + y2 ) (x1 + x2 )
 
x −y
= ⇒
y x
x = x1 + x2 ,
y = y1 + y2 .
NUMBER, THE LANGUAGE OF SCIENCE 11

The same can be verified in the case of the multiplication of z1 and z2 ,


resulting in a complex number z = (x, y),
  
x1 −y1 x2 −y2
z1 z2 =
y1 x1 y2 x2
 
(x1 x2 − y1 y2 ) −(x1 y2 + x2 y1 )
=
(x1 y2 + x2 y1 ) (x1 x2 − y1 y2 )
 
x −y
= ⇒
y x
x = x1 x2 − y 1 y 2 ,
y = x1 y 2 + x2 y 1 .

As we see, this new structure preserves the arithmetic operations on


complex numbers in their original form.

Problem 7. (Challenge Problem) Show that the field of rational num-


bers Q is countable, building explicitly a one-to-one relation between Q and
N.

Complete Solution:

Let us start by treating separately the number 0, which becomes the first in
the sequence. Also, we will limit ourselves to just enumerating the positive
rational numbers, which is equivalent to doing the same for all the negative
rational numbers. That is enough, because once this is done it is easy to
enumerate all the positive and negative rational numbers, by simply merging
(interspersing) the two sequences. We must now decide what to do with the
numbers of the form p/q, with p ∈ I, q ∈ I, p 6= 0 and q 6= 0. We will
organize these numbers using an infinite table as illustrated in Table 1.1,
with the values of p listed along the horizontal, those of q listed along the
vertical, and the numbers p/q placed in the corresponding table elements.
Let us now consider going through this table along successive diagonals,
each containing a finite number of elements, so that the following sequence
is defined, starting with the number zero that we separated from the rest,
           
1 1 2 1 2 3
0, , , , , , ,
1 2 1 3 2 1
       
1 2 3 4
, , , ,... .
4 3 2 1
12 SOLUTIONS 1

p = 1 p = 2 p = 3 p = 4 p = 5 p = 6 ···

q=1 1/1 2/1 3/1 4/1 5/1 6/1 · · ·

q=2 1/2 2/2 3/2 4/2 5/2 6/2 · · ·

q=3 1/3 2/3 3/3 4/3 5/3 6/3 · · ·

q=4 1/4 2/4 3/4 4/4 5/4 6/4 · · ·

q=5 1/5 2/5 3/5 4/5 5/5 6/5 · · ·

q=6 1/6 2/6 3/6 4/6 5/6 6/6 · · ·


.. .. .. .. .. .. .. . .
. . . . . . . .

Table 1.1: A table of the values of p and q and the corresponding ratios.

With this we are effectively enumerating, that is, constructing a sequence


indexed by the natural numbers, all the table elements, which includes all
possible positive rational numbers. Of course, there is some repetition in the
sequence, because when we simplify fractions there are several that repre-
sent the same number. For example, both the second element of the above
sequence, (1/1), and the sixth element, (2/2), represent the number 1. How-
ever, it is easy to overcome this problem by simply adopting in the construc-
tion algorithm of the sequence, the procedure of skipping a number whenever
it is equal to some previous number already included in the sequence. By
doing this we obtain the sequence
           
1 1 2 1 3 1
0, , , , , , ,
1 2 1 3 1 4
         
2 3 4 1 5
, , , , ,... .
3 2 1 5 1

Including now the negative numbers, we have the following sequence


       
1 1 1 1
0, 1, −1, ,− , 2, −2, ,− ,
2 2 3 3
       
1 1 2 2
3, −3, ,− , ,− ,
4 4 3 3
NUMBER, THE LANGUAGE OF SCIENCE 13
       
3 3 1 1
,− , 4, −4, ,− ,... ,
2 2 5 5

that is sufficient to display a bijection between Q and N, thus showing that


the two sets have the same number of elements.

Historical note: this construction is due to the great German mathemati-


cian Georg Cantor.

Problem 8. (Challenge Problem) Show that the field of real numbers


R is not countable. In order to do this, by reductio ad absurdum, show that,
if we assume that a sequence enumerating the real numbers is presented, then
it is always possible to construct a real number that is none of the elements
in that sequence.
Hint: use the decimal representation of the real numbers.

Complete Solution:

We will limit ourselves to showing that it is not possible to enumerate the real
numbers contained within the interval [0, 1), that is, greater than or equal
to zero and strictly smaller than 1. This is enough because if one cannot
enumerate these real numbers, then of course one cannot enumerate all the
real numbers. Let us do this by reductio ad absurdum, showing that, given
any enumeration that is presented, we can construct a real number that is
not included in that enumeration.
Any real number in [0, 1) can be represented in decimal notation as

0.d1 d2 d3 . . . ,

where each di is a digit, that is, an integer between 0 and 9. Of course, a


change in any single digit of the number is enough to change the number. Let
us assume for a moment that a sequence of these real numbers that represents
an enumeration of them has been presented to us, that is, a sequence that
contains all possible real numbers within [0, 1), as shown in the table below,

0.d1,1 d1,2 d1,3 d1,4 d1,5 d1,6 . . .


0.d2,1 d2,2 d2,3 d2,4 d2,5 d2,6 . . .
0.d3,1 d3,2 d3,3 d3,4 d3,5 d3,6 . . .
0.d4,1 d4,2 d4,3 d4,4 d4,5 d4,6 . . .
0.d5,1 d5,2 d5,3 d5,4 d5,5 d5,6 . . .
14 SOLUTIONS 1

0.d6,1 d6,2 d6,3 d6,4 d6,5 d6,6 . . .


... .

In order to construct a real number contained in [0, 1), described in the form
0.d1 d2 d3 . . ., which is none of the elements listed in this infinite sequence, it
suffices to proceed as follows: choose for d1 any digit that is not d1,1 , for
d2 any digit that is not d2,2 , for d3 any digit that is not d3,3 , and so on ad
infinitum. The result is a real number contained in [0, 1) that is none of the
numbers presented in the sequence. It follows that the sequence presented
cannot be a complete list of all real numbers contained in [0, 1), because there
is at least one of these numbers that is not included in the sequence. In fact,
there are many, because in each of the choices that we made we could use
any of 9 among the 10 digits available.

Historical note: this construction is due to the great German mathemati-


cian Georg Cantor.
Solutions 2

The Simplicity of Complex


Numbers

We present here complete and commented solutions to all problems proposed


in Chapter 2 of the text. For reference, the propositions of the problems are
repeated here. The problems are discussed in the order in which they were
proposed within the problem set of that chapter.

Problem 1. Starting from the Euler formula, prove the DeMoivre theorem.

Complete Solution:

The DeMoivre theorem states that for any real θ,

[cos(θ) + ı sin(θ)]n = cos(nθ) + ı sin(nθ).


We will prove this result by using the Euler formula, which was proved in
the text, and which states that for any real θ,

eıθ = cos(θ) + ı sin(θ).


It suffices to write the left-hand side of the DeMoivre equation in terms of
the exponential with imaginary argument,
 n
[cos(θ) + ı sin(θ)]n = eıθ
= eı(nθ) ,
where we used the properties of the exponential. Using again the Euler for-
mula, this time for the case of the exponential with the imaginary argument
ı(nθ), we have

15
16 SOLUTIONS 2

[cos(θ) + ı sin(θ)]n = cos(nθ) + ı sin(nθ),

which establishes the DeMoivre theorem.

Problem 2. Show that the constant function w(z) = 1 is analytic, that


is, that it satisfies the Cauchy-Riemann conditions. Starting from this fact,
prove by finite induction that the function w(z) = z n , with n a positive
integer, is analytic on the entire complex plane.

Complete Solution:

For the function w(z) = 1 we have that u(x, y) = 1 and v(x, y) = 0. It follows
therefore that all partial derivatives of u(x, y) and v(x, y) are zero, so that
∂u ∂v
= 0 = ,
∂x ∂y
∂u ∂v
= 0 = − ,
∂y ∂x
that is, the Cauchy-Riemann conditions are satisfied for any values of x and
of y, and thus the function is analytic in the entire complex plane. Note that
we have the same results for the case of the function w(z) = 0 as well. We will
now extend this result to all the powers of z, by finite induction. In order to
do this, we assume the result for the case n − 1, that is, for wn−1 (z) = z n−1 ,

wn−1 (z) = un−1 (x, y) + ıvn−1 (x, y),


∂un−1 ∂vn−1
= ,
∂x ∂y
∂un−1 ∂vn−1
= − .
∂y ∂x
Let us now consider the next case, wn (z) = z n = zwn−1 (z), for which we
therefore have

wn (z) = un (x, y) + ıvn (x, y)


= (x + ıy)[un−1 (x, y) + ıvn−1 (x, y)]
= [xun−1 (x, y) − yvn−1 (x, y)] +
+ı[xvn−1 (x, y) + yun−1 (x, y)] ⇒
un (x, y) = xun−1 (x, y) − yvn−1 (x, y),
vn (x, y) = xvn−1 (x, y) + yun−1 (x, y).
THE SIMPLICITY OF COMPLEX NUMBERS 17

Calculating the partial derivatives involved in the first condition, for the case
n, we have
∂un ∂un−1 ∂vn−1
= un−1 + x −y ,
∂x ∂x ∂x
∂vn ∂vn−1 ∂un−1
= x + un−1 + y .
∂y ∂y ∂y
Using now the Cauchy-Riemann relations for the case n − 1 in the equation
above, we get
∂un ∂un−1 ∂vn−1
= un−1 + x −y ,
∂x ∂x ∂x
∂vn ∂un−1 ∂vn−1
= un−1 + x −y ⇒
∂y ∂x ∂x
∂un ∂vn
= ,
∂x ∂y
thereby establishing the first condition for the case n. Doing the same for
the case of the second condition we have
∂un ∂un−1 ∂vn−1
= x − vn−1 − y ,
∂y ∂y ∂y
∂vn ∂vn−1 ∂un−1
= vn−1 + x +y .
∂x ∂x ∂x
Using now the Cauchy-Riemann relations for the case n − 1 in the equation
above, we get
∂un ∂un−1 ∂vn−1
= x − vn−1 − y ,
∂y ∂y ∂y
∂vn ∂un−1 ∂vn−1
= −x + vn−1 + y ⇒
∂x ∂y ∂y
∂un ∂vn
= − ,
∂y ∂x
thus establishing the second condition for the case n. As we have shown that
the case n − 1 implies the case n, and that the conditions hold for the case
n = 0, it follows that the functions wn (z) = z n are analytic throughout the
complex plane, for every positive value of n.
Note that this proof could be greatly simplified through the use of the
theorem, that we have seen previously, that tells us that the product of two
analytic functions is also an analytic function within the common domain of
18 SOLUTIONS 2

analyticity of the two functions. Once we have established that the function
w1 (z) = z is analytic throughout the complex plane, it follows that w2 (z) =
w1 (z)w1 (z) is also analytic throughout the complex plane. Iterating this
argument several times, we have that if w1 (z) = z and wn−1 (z) = z n−1
are analytic throughout the complex plane, then it follows that wn (z) =
w1 (z)wn−1 (z) is also analytic throughout the complex plane. This again
extends our result by finite induction, for any positive integer value of n.

Problem 3. Complete the proof given in the text that the function w(z) =
1/z is analytic. From this fact prove, by finite induction, that the function
w(z) = 1/z n , with n a strictly positive integer, is analytic on the whole
complex plane except at the origin z = 0.

Complete Solution:

As noted in the text, we have for the complex function w(z) = 1/z

w(z) = u(x, y) + ıv(x, y),


x
u(x, y) = ,
x + y2
2
y
v(x, y) = − 2 .
x + y2

The partial derivatives involved in the first Cauchy-Riemann condition are

∂u 1 2x2
= −
∂x x2 + y 2 (x2 + y 2 )2
−x2 + y 2
= ,
(x2 + y 2 )2
∂v −1 2y 2
= +
∂y x2 + y 2 (x2 + y 2 )2
−x2 + y 2
= ,
(x2 + y 2 )2

from which follows the first relation,

∂u ∂v
= .
∂x ∂y

The partial derivatives involved in the second Cauchy-Riemann condition are


THE SIMPLICITY OF COMPLEX NUMBERS 19

∂u 2xy
= − ,
∂y (x2 + y 2 )2
∂v 2xy
= ,
∂x (x + y 2 )2
2

from which follows the second relation,

∂u ∂v
=− .
∂y ∂x

Thus we see that the conditions are satisfied in the whole complex plane
except for the point z = 0, at which none of the partial derivatives is well
defined. It follows that the function w(z) = 1/z is analytic in the whole
complex plane except at the point z = 0. We will now extend this result
to all the negative powers of z, by finite induction. For this, we assume the
result for the case n − 1, that is, for wn−1 (z) = 1/z n−1 ,

wn−1 (z) = un−1 (x, y) + ıvn−1 (x, y),


∂un−1 ∂vn−1
= ,
∂x ∂y
∂un−1 ∂vn−1
= − .
∂y ∂x

Let us now consider the next case, wn (z) = 1/z n = wn−1 (z)/z, for which we
have therefore

wn (z) = un (x, y) + ıvn (x, y)


x − ıy
= [un−1 (x, y) + ıvn−1 (x, y)]
x2 + y 2
xun−1 (x, y) + yvn−1 (x, y)
= +
x2 + y 2
xvn−1 (x, y) − yun−1 (x, y)
+ı ⇒
x2 + y 2
xun−1 (x, y) + yvn−1 (x, y)
un (x, y) = ,
x2 + y 2
xvn−1 (x, y) − yun−1 (x, y)
vn (x, y) = .
x2 + y 2

Calculating the partial derivatives involved in the first condition, for the case
n, we have
20 SOLUTIONS 2

∂un −2x(xun−1 + yvn−1 ) un−1


= 2 2 2
+ 2 +
∂x (x + y ) x + y2
x ∂un−1 y ∂vn−1
+ 2 2
+ 2 2
x + y ∂x x + y ∂x
2 2
−2x un−1 − 2xyvn−1 + (x + y )un−1 2
= +
(x2 + y 2 )2
x ∂un−1 y ∂vn−1
+ 2 2
+ 2 2
x + y ∂x x + y ∂x
2 2
(−x + y )un−1 − 2xyvn−1
= +
(x2 + y 2 )2
x ∂un−1 y ∂vn−1
+ 2 2
+ 2 2
,
x + y ∂x x + y ∂x
∂vn −2y(xvn−1 − yun−1 ) un−1
= 2 2 2
− 2 +
∂y (x + y ) x + y2
x ∂vn−1 y ∂un−1
+ 2 − 2
x + y 2 ∂y x + y 2 ∂y
−2xyvn−1 + 2y 2 un−1 − (x2 + y 2 )un−1
= +
(x2 + y 2 )2
x ∂vn−1 y ∂un−1
+ 2 2
− 2 2
x + y ∂y x + y ∂y
2 2
(−x + y )un−1 − 2xyvn−1
= +
(x2 + y 2 )2
x ∂vn−1 y ∂un−1
+ 2 2
− 2 2
.
x + y ∂y x + y ∂y

Using now the Cauchy-Riemann relations for the case n − 1 in the equation
above, we get

∂un (−x2 + y 2 )un−1 − 2xyvn−1


= +
∂x (x2 + y 2 )2
x ∂un−1 y ∂vn−1
+ 2 2
+ 2 2
,
x + y ∂x x + y ∂x
∂vn (−x2 + y 2 )un−1 − 2xyvn−1
= +
∂y (x2 + y 2 )2
x ∂un−1 y ∂vn−1
+ 2 + 2 ⇒
x + y 2 ∂x x + y 2 ∂x
∂un ∂vn
= ,
∂x ∂y
THE SIMPLICITY OF COMPLEX NUMBERS 21

thereby establishing the first condition for the case n. Doing the same for
the case of the second condition we have
∂un −2y(xun−1 + yvn−1 ) vn−1
= 2 2 2
+ 2 +
∂y (x + y ) x + y2
x ∂un−1 y ∂vn−1
+ 2 2
+ 2 2
x + y ∂y x + y ∂y
2 2
−2xyun−1 − 2y vn−1 + (x + y )vn−1 2
= +
(x2 + y 2 )2
x ∂un−1 y ∂vn−1
+ 2 2
+ 2 2
x + y ∂y x + y ∂y
2 2
(x − y )vn−1 − 2xyun−1
= +
(x2 + y 2 )2
x ∂un−1 y ∂vn−1
+ 2 2
+ 2 2
,
x + y ∂y x + y ∂y
∂vn −2x(xvn−1 − yun−1 ) vn−1
= 2 2 2
+ 2 +
∂x (x + y ) x + y2
x ∂vn−1 y ∂un−1
+ 2 2
− 2 2
x + y ∂x x + y ∂x
2 2
−2x vn−1 + 2xyun−1 + (x + y )vn−1 2
= +
(x2 + y 2 )2
x ∂vn−1 y ∂un−1
+ 2 2
− 2 2
x + y ∂x x + y ∂x
2 2
(−x + y )vn−1 + 2xyun−1
= +
(x2 + y 2 )2
x ∂vn−1 y ∂un−1
+ 2 2
− 2 2
.
x + y ∂x x + y ∂x
Using now the Cauchy-Riemann relation for the case n − 1 in the equation
above, we get

∂un (x2 − y 2 )vn−1 − 2xyun−1


= +
∂y (x2 + y 2 )2
x ∂un−1 y ∂vn−1
+ 2 2
+ 2 2
,
x + y ∂y x + y ∂y
∂vn (x2 − y 2 )vn−1 − 2xyun−1
= − +
∂x (x2 + y 2 )2
x ∂un−1 y ∂vn−1
− 2 2
− 2 2

x + y ∂y x + y ∂y
22 SOLUTIONS 2

∂un ∂vn
= − ,
∂y ∂x

thus establishing the second condition for the case n. As we have shown that
the case n − 1 implies the case n, and that the conditions hold for the case
n = 1, it follows that the complex functions wn (z) = 1/z n are analytic in
the whole complex plane except at the origin z = 0, for all strictly positive
values of the integer n.
Note that this proof could be greatly simplified through the use of the
theorem, which we have seen previously, that tells us that the product of two
analytic functions is also an analytic function within the common domain of
analyticity of the two functions. Once we have established that the function
w1 (z) = 1/z is analytic in the whole complex plane except for the origin, it
follows that w2 (z) = w1 (z)w1 (z) is also analytic in the whole complex plane
except at the origin. Iterating this argument several times, we have that
if w1 (z) = 1/z and wn−1 (z) = 1/z n−1 are analytic in the whole complex
plane except at the origin, then it follows that wn (z) = w1 (z)wn−1 (z) is also
analytic in the whole complex plane except at the origin. This again extends
our result by finite induction, for any strictly positive value of the integer n.

Problem 4. Consider the movement of a solid disk which rotates around


its axis with a constant angular speed ω. Consider a point of this disk at a
distance ρ from the center, whose initial position forms an angle θ0 with the
x axis of a system of Cartesian coordinates (x, y) on the plane of the disk.

(a) Given that (x, y) are the coordinates of a point of the disk in this
coordinate system, represent the position of the point as a complex
number z = (x, y), giving z as a function of the time t and of the initial
position θ0 of the point.

(b) Give the complex velocity ż of the point of the disk as a function of the
time t and of the initial position θ0 of the point. Also, write ż in terms
of z.

(c) What is the effect on the two-dimensional vector z = (x, y) of the


operation of multiplying it by ı = (0, 1)?

(d) Give the complex acceleration z̈ of the point of the disk as a function
of the time t and of the initial position θ0 of the point. Also, write z̈ in
terms of z.
THE SIMPLICITY OF COMPLEX NUMBERS 23

ρ
θ
0 x

Figure 2.1: The disk rotating at a constant angular velocity ω.

(e) Rewrite the results for z, ż and z̈ using the exponential form of the
complex numbers.

Complete Solution:

Consider the disk, its axis, a point on the disk and the coordinate system as
illustrated in Figure 2.1.

(a) The position vector of the point on the disk is given by ~r = (x, y). In
terms of ρ and θ, where ρ is a constant and θ(t) = θ0 + ωt, we have

x(t) = ρ cos(θ0 + ωt),


y(t) = ρ sin(θ0 + ωt).

We can represent the position by means of the complex number

z(t) = ρ [cos(θ0 + ωt) + ı sin(θ0 + ωt)] .

(b) Having the representation of the position in terms of z(t), it suffices to


differentiate with respect to the time in order to obtain the correspond-
ing representation of the velocity ż(t),

z(t) = ρ [cos(θ0 + ωt) + ı sin(θ0 + ωt)] ⇒


ż(t) = ρω [− sin(θ0 + ωt) + ı cos(θ0 + ωt)] .
24 SOLUTIONS 2

Note that we can write, by putting a factor of ı in evidence, that

ż(t) = ıρω [cos(θ0 + ωt) + ı sin(θ0 + ωt)]


= ıρωz(t).

(c) From the physics of the problem we know that the velocity is perpen-
dicular to the position vector. This means that ız is perpendicular to z,
therefore the multiplication by ı rotates the vector by an angle of π/2
in the positive direction, that is, counterclockwise. The orthogonality
can be verified directly by making the scalar product (which is not the
same as the complex product) of the vector ız and the vector z.

(d) Having the representation of the velocity in terms of ż(t), it suffices to


differentiate with respect to the time in order to obtain the correspond-
ing representation of the acceleration z̈(t),

ż(t) = ρω [− sin(θ0 + ωt) + ı cos(θ0 + ωt)] ⇒


z̈(t) = −ρω 2 [cos(θ0 + ωt) + ı sin(θ0 + ωt)] .

Note that we can immediately write that

z̈(t) = −ρω 2 z(t),

which shows that the acceleration is in the direction opposite to the


position vector. It is a centripetal acceleration.

(e) Using the polar representation of the complex numbers and the Euler
formula, we can write immediately that

z(t) = ρ eı(θ0 +ωt) ,


ż(t) = ıρω eı(θ0 +ωt) ,
z̈(t) = −ρω 2 eı(θ0 +ωt) .

As one can see, in this way the derivatives and corresponding physical
consequences are all quite immediately visible.
THE SIMPLICITY OF COMPLEX NUMBERS 25

Problem 5. Consider the complex number z = exp(ıθ), where θ is a real


number.
(a) Determine the real and imaginary parts of z and of its complex conju-
gate z ∗ .

(b) Calculate the absolute value of z, that is, |z| = z ∗ z.
(c) Determine the complex number z 2 , the square of z, using the exponen-
tial form, and write its real and imaginary parts.
(d) Calculate z 2 starting from z, without using the exponential form, that
is, first writing z in terms of its real and imaginary parts, and then
taking the square.
(e) Comparing the results obtained in the two previous items, derive the
trigonometric identities for the sine and the cosine of the double arc.

Complete Solution:

Consider the complex number z = exp(ıθ), which can be written in terms of


its real and imaginary parts as z = u(θ) + ıv(θ).
(a) Using the Euler formula, we find that the real and imaginary parts of
z are

u(θ) = cos(θ),
v(θ) = sin(θ).

Thus we see that the real part u′ (θ) and the imaginary part v ′ (θ) of z ∗
are

u′ (θ) = cos(θ),
v ′ (θ) = − sin(θ).

(b) We can calculate the absolute value of z in two ways,


q
|z| = cos2 (θ) + sin2 (θ)
= 1,

|z| = eıθ e−ıθ

= eıθ−ıθ

= e0
= 1.
26 SOLUTIONS 2

(c) Calculating the square of z, using the exponential form, we have


 2
z2 = eıθ
= eı2θ
= cos(2θ) + ı sin(2θ).

(d) Calculating the square of z, using the form u + ıv, we have

z 2 = [cos(θ) + ı sin(θ)]2
= cos2 (θ) − sin2 (θ) + ı [2 cos(θ) sin(θ)] .
 

(e) Comparing the results of the previous two items, we get the following
trigonometric identities,

cos(2θ) = cos2 (θ) − sin2 (θ),


sin(2θ) = 2 sin(θ) cos(θ).

Problem 6. Show that the sum-function w(z) = f (z)+g(z), of two analytic


functions f (z) and g(z), is also analytic in the common domain of analyticity
of f (z) and g(z).

Complete Solution:

Consider the sum-function w(z) = f (z) + g(z), where f (z) = uf (x, y) +


ıvf (x, y) and g(z) = ug (x, y) + ıvg (x, y). Within the common domain of
analyticity of f (z) and g(z), we have that all partial derivatives exist, and
that they satisfy
∂uf ∂vf
= ,
∂x ∂y
∂uf ∂vf
= − ,
∂y ∂x
∂ug ∂vg
= ,
∂x ∂y
∂ug ∂vg
= − .
∂y ∂x
We therefore have for the sum-function w(z) = u(x, y) + ıv(x, y), where by
definition of the sum of complex numbers u(x, y) = uf (x, y) + ug (x, y) and
v(x, y) = vf (x, y) + vg (x, y), within this domain, that
THE SIMPLICITY OF COMPLEX NUMBERS 27

∂u ∂uf ∂ug
= +
∂x ∂x ∂x
∂vf ∂vg
= +
∂y ∂y
∂v
= ,
∂y

which shows that the first Cauchy-Riemann condition is satisfied for w(z).
Similarly, for the other condition,
∂u ∂uf ∂ug
= +
∂y ∂y ∂y
∂vf ∂vg
= − −
∂x ∂x
∂v
= − .
∂x
It follows therefore that w(z) is analytic in the common domain of analyticity
of f (z) and g(z).

Problem 7. Show that the product-function w(z) = f (z)g(z), of two an-


alytic functions f (z) and g(z), is also analytic in the common domain of
analyticity of f (z) and g(z).

Complete Solution:

Consider the product-function w(z) = f (z)g(z), where f (z) = uf (x, y) +


ıvf (x, y) and g(z) = ug (x, y) + ıvg (x, y). Within the common domain of
analyticity of f (z) and g(z), we have that all partial derivatives exist, and
that they satisfy
∂uf ∂vf
= ,
∂x ∂y
∂uf ∂vf
= − ,
∂y ∂x
∂ug ∂vg
= ,
∂x ∂y
∂ug ∂vg
= − .
∂y ∂x

We therefore have for product-function w(z) = u(x, y) + ıv(x, y), where by


definition of the product of complex numbers
28 SOLUTIONS 2

u(x, y) = uf (x, y)ug (x, y) − vf (x, y)vg (x, y),


v(x, y) = uf (x, y)vg (x, y) + vf (x, y)ug (x, y),

within this domain, that


∂u ∂ug ∂uf ∂vg ∂vf
= uf + ug − vf − vg
∂x ∂x ∂x ∂x ∂x
∂vg ∂vf ∂ug ∂uf
= uf + ug + vf + vg
∂y ∂y ∂y ∂y
   
∂vg ∂uf ∂ug ∂vf
= uf + vg + vf + ug
∂y ∂y ∂y ∂y
∂ ∂
= (uf vg ) + (vf ug )
∂y ∂y

= (uf vg + vf ug )
∂y
∂v
= ,
∂y
which shows that the first Cauchy-Riemann condition is satisfied for w(z).
Similarly, for the other condition,
∂u ∂ug ∂uf ∂vg ∂vf
= uf + ug − vf − vg
∂y ∂y ∂y ∂y ∂y
∂vg ∂vf ∂ug ∂uf
= −uf − ug − vf − vg
 ∂x ∂x  ∂x  ∂x 
∂vg ∂uf ∂ug ∂vf
= − uf + vg − vf + ug
∂x ∂x ∂x ∂x
∂ ∂
= − (uf vg ) − (vf ug )
∂x ∂x

= − (uf vg + vf ug )
∂x
∂v
= − .
∂x
It follows therefore that w(z) is analytic in the common domain of analyticity
of f (z) and g(z).

Problem 8. Show that the composite function w(z) = f [g(z)] obtained


by the composition of two analytic functions f (z) and g(z) is also analytic,
within the domain that is given by the intersection of the image of g(z) with
the analyticity domain of f (z).
THE SIMPLICITY OF COMPLEX NUMBERS 29

Hint: consider the composition of two complex functions involving three


complex planes, so that one of the functions maps the first plane onto the
second and the other function maps the second plane onto the third.

Complete Solution:

Consider the composite function w(z) = f [g(z)], where g(z) = ug (x, y) +


ıvg (x, y). In the analyticity domain of g(z) we have that

∂ug ∂vg
= ,
∂x ∂y
∂ug ∂vg
= − .
∂y ∂x

Similarly, in the domain of analyticity of f (z) we have that f (z) = uf (x, y) +


ıvf (x, y) and that

∂uf ∂vf
= ,
∂x ∂y
∂uf ∂vf
= − .
∂y ∂x

Let us now consider the portion of the image of g(z) that falls within the
domain of analyticity of f (z), that is, the intersection of the image of g(z)
with the analyticity domain of f (z). Within this region we have that f [g(z)]
has real and imaginary parts given by

f (z) = uf (ug , vg ) + ıvf (ug , vg ).

Since f (g) is analytic with respect to the complex variable g, which has real
and imaginary parts given by ug and vg , we have that

∂uf ∂vf
= ,
∂ug ∂vg
∂uf ∂vf
= − .
∂vg ∂ug

We therefore have for w(z) = u(x, y) + ıv(x, y), where by the definition of the
composition of two functions u(x, y) = uf (ug , vg ) and v(x, y) = vf (ug , vg ),
within this region, by the chain rule, that
30 SOLUTIONS 2

∂u ∂uf ∂ug ∂uf ∂vg


= +
∂x ∂ug ∂x ∂vg ∂x
∂vf ∂vg ∂vf ∂ug
= + (−1) (−1)
∂vg ∂y ∂ug ∂y
∂vf ∂ug ∂vf ∂vg
= +
∂ug ∂y ∂vg ∂y
∂v
= ,
∂y

where in this last passage we use the chain rule in the reversed direction.
This shows that the first Cauchy-Riemann condition is satisfied for w(z).
Similarly, for the other condition,

∂u ∂uf ∂ug ∂uf ∂vg


= +
∂y ∂ug ∂y ∂vg ∂y
∂vf ∂vg ∂vf ∂ug
= − −
∂vg ∂x ∂ug ∂x
 
∂vf ∂ug ∂vf ∂vg
= − +
∂ug ∂x ∂vg ∂x
∂v
= − ,
∂x
where once again we use the chain rule in the reversed direction in the last
passage. It follows therefore that w(z) is analytic in the intersection of the
image of g(z) with the analyticity domain of f (z).

Problem 9. Starting from the fact that, for any complex numbers z1 , z2
and z3 , it is true that

z3 = z1 + z2 ⇒
|z3 | ≤ |z1 | + |z2 |,

show that if we have a sum of n complex numbers zi ,


n
X
zs = zi
i=1
= z1 + z2 + z3 + . . . + zn−1 + zn ,

it follows that
THE SIMPLICITY OF COMPLEX NUMBERS 31

n
X
|zs | ≤ |zi |.
i=1

Complete Solution:

Consider the basic triangle inequality |zc | ≤ |za | + |zb |, valid for any complex
numbers za , zb and zc , and the sum
n
X
zs = zi .
i=1

We may write this sum as


n
X
zs = z1 + zi ,
i=2

and using the basic inequality we have


n
X
|zs | ≤ |z1 | + zi .
i=2

We may now write for the remaining sum


n
X n
X
zi = z2 + zi ,
i=2 i=3

from which it follows that


n
X n
X
zi ≤ |z2 | + zi ,
i=2 i=3

and therefore that


n
X
|zs | ≤ |z1 | + |z2 | + zi .
i=3

Iterating this procedure for each possible value of i in the sum, we end up
getting as the final result the general triangle inequality
n
X
|zs | ≤ |zi |.
i=1

Note that this remains true even if the sum extends indefinitely to infinity,
that is, for n → ∞, provided that the infinite sums involved converge.
32 SOLUTIONS 2

Problem 10. (Challenge Problem) Consider the complex number given


by z = exp(ı2π/N ), where N is a strictly positive integer.

(a) Determine the real and imaginary parts of the powers z k of z, for k ∈
{1, . . . , N }, and write z k in terms of their real and imaginary parts,
leaving the result in terms of k and N .

(b) Determine the sum of all the N numbers z k , that is, calculate

N
X
S= zk .
k=1

Hint: draw vectors z k in the complex plane, for a definite and not too
large value of N , for example, for N = 6, N = 7 or N = 8.

Complete Solution:

Consider the complex number z = exp(ı2π/N ), where N > 0.

(a) The N powers z k , for k ∈ {1, . . . , N }, are given by z k = exp(ı2πk/N ).


Using the Euler formula, we see that their real and imaginary parts are
given by
h i
ℜ z k = cos(2πk/N ),
h i
ℑ z k = sin(2πk/N ).

(b) Let us now calculate the sum of all these N vectors in the complex
plane,

N
X
S= zk .
k=1

Since z is a unit vector in the complex plane, which makes an angle


θ = 2π/N with the real axis, each z k is also a unit vector, but pointing
in a different direction. Since for k = N we have that z N = exp(ı2π) =
1, and each multiplication by z corresponds to a rotation by the angle
θ, we see that these N vectors point in N directions equally spaced by
THE SIMPLICITY OF COMPLEX NUMBERS 33

y
2 1

3 6
x

4 5

Figure 2.2: The unit circle with N equally spaced unit vectors, in the case
N = 6.

the angle θ. We can therefore draw all N vectors along the unit circle,
as exemplified in the diagram of Figure 2.2.
This figure suggests that the sum is zero by symmetry. In the cases
where N is even, like the one shown in the figure, that is immediate,
since the vectors cancel off in pairs. In the cases where N is odd, the
result is still suggested by the figure, but it is not so immediate. We can
see that in fact the sum is zero for any N , using the following reduction
ad absurdum argument. Imagine for a moment that the sum is not zero.
It follows that it is a complex number and is therefore represented by
some non-zero vector in the complex plane, such as the lower vector
shown in dashed line in Figure 2.2. If we now rotate all the N vectors
by the angle θ, in the positive direction, we note that the figure formed
by them remains invariant, and therefore has the same sum as before.
However, by this same rotation the sum-vector is rotated by the angle θ,
and does not remain invariant, which is absurd. It follows that the sum
must be a vector which is invariant under a rotation through an angle
θ 6= 0. But the only vector that has this property is the zero vector,
and hence we conclude that S = 0. We can express these same ideas
arithmetically. Since the rotation by θ corresponds to a multiplication
by z, we have that the process described can be represented as

N
X
S = zk ⇒
k=1
34 SOLUTIONS 2

N
X
zS = z k+1 ,
k=1

where z 6= 0 and z 6= 1. Now, it is easy to see that, in the lower sum,


z N +1 is equal to z, because

z N +1 = eı2π(N +1)/N
= eı2π eı2π/N
= eı2π/N
= z,

so that we have for the sum after the rotation by θ implemented by the
multiplication by z,

N
X N
X −1
z k+1 = z k+1 + z N +1
k=1 k=1
N

X
= z+ zk
k ′ =2
N
X
= zk .
k=1

This shows that the sum is invariant. It follows from the preceding
equations that zS = S, where z 6= 1. The only way to satisfy this
equation is to have S = 0, which thus establishes the required result.
There are other ways to prove this relation for all values of N , that will
be seen later on, in this series of books.
Solutions 3

Elementary Functions, but


Not Quite

We present here complete and commented solutions to all problems proposed


in Chapter 3 of the text. For reference, the propositions of the problems are
repeated here. The problems are discussed in the order in which they were
proposed within the problem set of that chapter.

Problem 1. By means of a transformation of variables in the complex


plane, from the Cartesian coordinates (x, y) to the polar coordinates (ρ, θ),
write the two Cauchy-Riemann conditions in terms of the variables ρ and θ,
that is, in terms of derivatives with respect to these variables, where

x = ρ cos(θ),
y = ρ sin(θ).

Complete Solution:

Recalling that we have the functions u(x, y) and v(x, y), and that by means
of the transformation of variables

x(ρ, θ) = ρ cos(θ),
y(ρ, θ) = ρ sin(θ),

we also have the functions u(ρ, θ) and v(ρ, θ), we calculate the partial deriva-
tives of u and v with respect to ρ and θ, using the chain rule. The first
derivative that appears in the first Cauchy-Riemann condition is

35
36 SOLUTIONS 3

∂u ∂u ∂x ∂u ∂y
= +
∂ρ ∂x ∂ρ ∂y ∂ρ
∂u ∂u
= cos(θ) + sin(θ).
∂x ∂y

The other derivative that appears in this first condition is

∂v ∂v ∂x ∂v ∂y
= +
∂θ ∂x ∂θ ∂y ∂θ
∂v ∂v
= − ρ sin(θ) + ρ cos(θ).
∂x ∂y

Using now the Cauchy-Riemann conditions, in their Cartesian form, in order


to write the derivatives of v in terms of derivatives of u, we can write this as

∂v ∂u ∂u
= ρ sin(θ) + ρ cos(θ)
∂θ ∂y ∂x
 
∂u ∂u
= ρ cos(θ) + sin(θ) .
∂x ∂y

Comparing this with the previous equation we have that

∂u 1 ∂v
= ,
∂ρ ρ ∂θ

which is the first Cauchy-Riemann condition in polar coordinates. Examining


now the other relevant derivatives, we have

∂v ∂v ∂x ∂v ∂y
= +
∂ρ ∂x ∂ρ ∂y ∂ρ
∂v ∂v
= cos(θ) + sin(θ),
∂x ∂y

and
∂u ∂u ∂x ∂u ∂y
= +
∂θ ∂x ∂θ ∂y ∂θ
∂u ∂u
= − ρ sin(θ) + ρ cos(θ).
∂x ∂y

Using now the Cauchy-Riemann conditions, in their Cartesian form, in order


to write the derivatives of u in terms of derivatives of v, we can write this as
ELEMENTARY FUNCTIONS, BUT NOT QUITE 37

∂u ∂v ∂v
= − ρ sin(θ) − ρ cos(θ)
∂θ ∂y ∂x
 
∂v ∂v
= −ρ cos(θ) + sin(θ) .
∂x ∂y
Comparing this with the previous equation we have that
1 ∂u ∂v
=− ,
ρ ∂θ ∂ρ
which is the second Cauchy-Riemann condition in polar coordinates.

Problem 2. Show that the complex function w(z) = z is analytic, that
is, that it satisfies the Cauchy-Riemann conditions throughout the complex
plane, except for the origin z = 0.

Complete Solution:

Writing z in polar coordinates, z = ρ exp(ıθ), we have for the function w(z) =



z
√ ıθ/2
w(z) = ρe
    
√ θ θ
= ρ cos + ı sin .
2 2
We can now verify the analyticity using the polar version of the Cauchy-
Riemann conditions. For the first condition we have
 
∂u 1 θ
= √ cos ,
∂ρ 2 ρ 2
 
1 ∂v 1 θ
= √ cos ,
ρ ∂θ 2 ρ 2
so that the first condition is satisfied throughout the whole complex plane ex-
cept for the point ρ = 0, which corresponds to z = 0, at which the derivatives
do not exist. For the second condition we have
 
1 ∂u 1 θ
= − √ sin ,
ρ ∂θ 2 ρ 2
 
∂v 1 θ
= √ sin ,
∂ρ 2 ρ 2
so that the second condition is also satisfied, once again throughout the whole
complex plane except for the point z = 0.
38 SOLUTIONS 3

Problem 3. Given the complex number z = x + ıy, consider the complex


function w(z) = u(x, y) + ıv(x, y) in each case below. In each case, determine
the real part u(x, y) and the imaginary part v(x, y) of w(z), and show that
the function w(z) satisfies the two Cauchy-Riemann conditions.

(a) w(z) = exp(z).

(b) w(z) = cos(z).

(c) w(z) = sin(z).

(d) w(z) = cosh(z).

(e) w(z) = sinh(z).

Complete Solution:

In each case, we simply write explicitly the real and imaginary parts of
the function w(z) = u(x, y) + ıv(x, y), and calculate the appropriate par-
tial derivatives of u(x, y) and v(x, y).

(a) w(z) = exp(z):

w(z) = ex [cos(y) + ı sin(y)]


= ex cos(y) + ı ex sin(y) ⇒

∂u ∂v
= ex cos(y) =,
∂x ∂y
∂u ∂v
= − ex sin(y) = − .
∂y ∂x

(b) w(z) = cos(z):

eız + e−ız
w(z) =
2
e−y eıx + ey e−ıx
=
2
e−y [cos(x) + ı sin(x)] + ey [cos(x) − ı sin(x)]
=
2
= cos(x) cosh(y) − ı sin(x) sinh(y) ⇒
ELEMENTARY FUNCTIONS, BUT NOT QUITE 39

∂u ∂v
= − sin(x) cosh(y) = ,
∂x ∂y
∂u ∂v
= cos(x) sinh(y) = − .
∂y ∂x

(c) w(z) = sin(z):

eız − e−ız
w(z) =

e−y eıx − ey e−ıx
=

e−y [cos(x) + ı sin(x)] − ey [cos(x) − ı sin(x)]
=

= sin(x) cosh(y) + ı cos(x) sinh(y) ⇒

∂u ∂v
= cos(x) cosh(y) = ,
∂x ∂y
∂u ∂v
= sin(x) sinh(y) = − .
∂y ∂x

(d) w(z) = cosh(z):

ez + ez
w(z) =
2
e eıy + e−x e−ıy
x
=
2
e [cos(y) + ı sin(y)] + e−x [cos(y) − ı sin(y)]
x
=
2
= cosh(x) cos(y) + ı sinh(x) sin(y) ⇒

∂u ∂v
= sinh(x) cos(y) = ,
∂x ∂y
∂u ∂v
= − cosh(x) sin(y) = − .
∂y ∂x

(e) w(z) = sinh(z):

ez − ez
w(z) =
2
40 SOLUTIONS 3

ex eıy − e−x e−ıy


=
2
ex [cos(y) + ı sin(y)] − e−x [cos(y) − ı sin(y)]
=
2
= sinh(x) cos(y) + ı cosh(x) sin(y) ⇒

∂u ∂v
= cosh(x) cos(y) = ,
∂x ∂y
∂u ∂v
= − sinh(x) sin(y) = − .
∂y ∂x

Problem 4. Show that the complex function of the variable z = x + ıy


given by w(z) = x2 + ıy 2 is not analytic. Show the same for the complex
function w(z) = z ∗ z.

Complete Solution:

In the case of the function w(z) = x2 + ıy 2 we have that u = x2 and v = y 2 ,


so that we have the partial derivatives

∂u
= 2x,
∂x
∂v
= 2y,
∂y
∂u
= 0,
∂y
∂v
= 0.
∂x
Thus we see that the first Cauchy-Riemann condition is not satisfied, al-
though the second is. In the case of the function w(z) = z ∗ z = x2 + y 2 we
have that u = x2 + y 2 and v = 0, so that we have the partial derivatives

∂u
= 2x,
∂x
∂v
= 0,
∂y
∂u
= 2y,
∂y
∂v
= 0.
∂x
ELEMENTARY FUNCTIONS, BUT NOT QUITE 41

Thus we see that in this case neither of the Cauchy-Riemann conditions is


satisfied. In fact, we can see that in both cases there is a single isolated point
where the two conditions are satisfied, the point z = 0. However, this is not
enough for us to say that the functions are analytic at that point, because the
analyticity condition at a given point is, in fact, that the Cauchy-Riemann
conditions be satisfied at least in an infinitesimal neighborhood of that point,
and not just at the point.

Problem 5. In each case below, determine the domain of analyticity of


the function, that is, determine the points of the complex plane where the
function is not analytic.

(a) w(z) = tan(z).

(b) w(z) = cot(z).

(c) w(z) = sec(z).

(d) w(z) = csc(z).

Complete Solution:

In each case, we must determine the location of all possible zeros in denom-
inator, as well as other sources of non-analyticity, if any.

(a) w(z) = tan(z): since we have that

sin(z)
w(z) = ,
cos(z)

where both the sin(z) and the cos(z) are analytic functions throughout
the complex plane, we only need to find the set of all the zeros of the
function cos(z). This function can be written as

cos(z) = cos(x) cosh(y) − ı sin(x) sinh(y),

where both the real part and the imaginary part should be zero. In the
real part we have that cosh(y) never vanishes, so that it is necessary
that cos(x) vanish. However, at the points where cos(x) is zero, the
function sin(x) appearing in the imaginary part is not zero, hence it is
42 SOLUTIONS 3

necessary that sinh(y) vanish in the imaginary part. The only point at
which sinh(y) vanishes is y = 0, so that the zeros of the function cos(z)
are all on the real axis, being characterized by cos(x) = 0 and y = 0. In
other words, the only zeros of the complex function cos(z) are the real
zeros of the real function cos(x). It follows that the complex function
w(z) = tan(z) is analytic in the entire complex plane except for the set
of points given by y = 0 and

π
x= + nπ,
2

where n is an arbitrary positive or negative integer.

(b) w(z) = cot(z): since we have that

cos(z)
w(z) = ,
sin(z)

where both the sin(z) as the cos(z) are analytic functions throughout
the complex plane, we only need to find the set of all the zeros of the
function sin(z). This function can be written as

sin(z) = sin(x) cosh(y) + ı cos(x) sinh(y),

where both the real part and the imaginary part should be zero. An
analysis similar to the one that was developed in the previous item
implies that, in this case, we should have sin(x) = 0 and y = 0. In
other words, the only zeros of the complex function sin(z) are the real
zeros of the real function sin(x). It follows that the complex function
w(z) = cot(z) is analytic in the entire complex plane except for the set
of points given by y = 0 and x = nπ, where n is an arbitrary positive
or negative integer.

(c) w(z) = sec(z): since we have that

1
w(z) = ,
cos(z)

where cos(z) is an analytic function in the whole complex plane, we


only need to find the set of all the zeros of the function cos(z). The
analysis of the problem is thus identical to that of the first item, so that
ELEMENTARY FUNCTIONS, BUT NOT QUITE 43

the complex function w(z) = sec(z) is analytic in the entire complex


plane except for the set of points given by y = 0 and

π
x= + nπ,
2

where n is an arbitrary positive or negative integer.

(d) w(z) = csc(z): since we have that

1
w(z) = ,
sin(z)

where sin(z) is an analytic function in the whole complex plane, we only


need to find the set of all the zeros of the function sin(z). The analysis
of the problem is thus identical to that of the second item, so that the
complex function w(z) = csc(z) is analytic in the entire complex plane
except for the set of points given by y = 0 and x = nπ, where n is an
arbitrary positive or negative integer.

Problem 6. (Challenge Problem) Determine the branch points, the


branch cuts and the Riemann leaves of the functions that follow. Cut and
paste the leaves to produce Riemann surfaces on which the functions are
completely well defined and analytic, except for the singularities at the branch
points.

(a) w(z) = z 2 − 1.

(b) w(z) = 1/ z 2 − 1.

Hint: consider the behavior of the functions along circles around the points
where the square root vanishes.

Complete Solution:

In both cases to be examined, we can write that


p p
z 2 − 1 = (z − 1)(z + 1),

so that it is clear that the two special points to consider during the analysis
are z = 1 and z = −1. We can see this because if we have, for example,
z ≈ 1, it follows that z + 1 ≈ 2 is essentially a constant, and therefore we
44 SOLUTIONS 3

θ′ θ θ′
0 x
−1 1

Figure 3.1: The singular points and the three test circles.

effectively have a square root of the variable z ′ = z − 1. In order to make


the analysis, we will go around these two points along the circles illustrated
in the diagram of Figure 3.1.
Adopting the polar representation for z, that is z = ρ exp(ıθ), we will
first write the quantity z 2 − 1 in polar form, z 2 − 1 = λ exp(ıα). We thus
have

z 2 − 1 = λ eıα
= ρ2 e2ıθ − 1
= ρ2 cos(2θ) − 1 + ı ρ2 sin(2θ) .
   

We can now calculate λ,


2  2
λ2 =
 2
ρ cos(2θ) − 1 + ρ2 sin(2θ)
= ρ4 + 1 − 2ρ2 cos(2θ) ⇒
p
λ = ρ4 + 1 − 2ρ2 cos(2θ).

Similarly, we can now calculate exp(ıα),

ρ2 cos(2θ) − 1 ρ2 sin(2θ)
eıα = p +ıp .
ρ4 + 1 − 2ρ2 cos(2θ) ρ4 + 1 − 2ρ2 cos(2θ)
ELEMENTARY FUNCTIONS, BUT NOT QUITE 45

(a) In this case we have 2
√ w(z) = z − 1, which is expressed in terms of λ
and α as w(z) = λ exp(ıα/2). In order to go around the big circle
in the complex plane shown in the diagram of Figure 3.1, so as to
verify whether or not there is a change of leaf when we do this, we will
calculate exp(ıα) in the limit ρ → ∞,

"
ρ2 cos(2θ) − 1
lim eıα = lim p +
ρ→∞ ρ→∞ ρ4 + 1 − 2ρ2 cos(2θ)
#
ρ2 sin(2θ)
+ıp
ρ4 + 1 − 2ρ2 cos(2θ)
"
cos(2θ) − 1/ρ2
= lim p +
ρ→∞ 1 + 1/ρ4 − 2 cos(2θ)/ρ2
#
sin(2θ)
+ıp
1 + 1/ρ4 − 2 cos(2θ)/ρ2
 
cos(2θ) sin(2θ)
= √ +ı √
1 1
2ıθ
= e .

Similarly, we have in this limit λ = ρ2 . Thus we see that in this limit


we simply have that w(z) = ρ exp(ıθ), that is, the function approaches
the identity. Thus we see that there is no change of leaf when we vary θ
from 0 to 2π with ρ → ∞, so that there is no branch cut that extends to
infinity. On the other hand, if we approach the point z = 1, p we can write
the function in terms of the variable z ′ =√ z − 1 as w(z) = z ′ (z ′ + 2).
In the limit in which z ′ → 0 it approaches 2z ′ , which can be written in
terms of a polar coordinate system (ρ′ , θ ′ ) centered at the
√point z = 1, as
illustrated in the diagram of Figure 3.1, since w(z) ≈ 2ρ′ exp(ıθ ′ /2).
Thus we see that, if we follow an infinitesimal circle around z = 1,
with ρ′ constant and θ ′ going from 0 to 2π, there is indeed a change of
leaf. It follows that there is a branch cut connected to the point z = 1,
which, however, does not extend to infinity. A similar analysis around
the point z = 1 leads to the same conclusions for that point. It follows
that there is a single branch cut of finite extent, which connects the
two branch points z = 1 and z = −1. Any simple curve that connects
the two points can be chosen as the branch cut, but the simplest choice
is the real segment [−1, 1].
46 SOLUTIONS 3

The Riemann surface consists therefore of two copies of the complex


plane, except for the branch points z = 1 and z = −1, and with a cut
from one of these points to the other. Each side of this cut in one of
the two planes is continuously connected to the other side of the cut in
the other plane, so that by crossing the branch cut one changes from
one plane to the other.

w(z) = 1/ z 2 − 1, which is expressed in terms of
(b) In this case we have p
λ and α as w(z) = 1/λ exp(−ıα/2), which already shows that the
analysis is not very different from that of the previous case. Further-
more, we can reduce this problem explicitly to the previous problem,
because we have that
1
w(z) = √
2
z −1

z2 − 1
=
z2 − 1

z2 − 1
= ,
(z − 1)(z + 1)
where the function in denominator is analytic and single-valued, so
that the structure of branch cuts and points is determined only by the
numerator. There are, of course, additional singularities at z = 1 and
z = −1, due to the zeros in denominator, but this does not change the
structure of the leaves or of the Riemann surface. It follows that in this
case we also have that z = 1 and z = −1 are two branch points, with
a branch cut which connects the two points.
The Riemann surface is the same as the one in the previous item, so
that it consists of two copies of the complex plane, except for the branch
points z = 1 and z = −1, and with a cut from one of these points to
the other. Each side of this cut in one of the two planes is continuously
connected to the other side of the cut in the other plane, so that by
crossing the branch cut one changes from one plane to the other.
Note that it is not really necessary to take the limits ρ → ∞ and z → 1 or
z → −1 in order to do this analysis. The limits simplify the analysis, and
are sufficient due to the topological nature of the problem, with continuous
simple curves that connect points or that extend to infinity, but they are not
really essential. For the analysis it suffices to consider closed curves that go
independently around each one of the two branch points, and that go around
the two points at once.
Solutions 4

Even Less Elementary


Functions

We present here complete and commented solutions to all problems proposed


in Chapter 4 of the text. For reference, the propositions of the problems are
repeated here. The problems are discussed in the order in which they were
proposed within the problem set of that chapter.

Problem 1. Calculate, giving their real and imaginary parts, the numbers
that follow.

(a) z = 1ı .

(b) z = eı .

(c) z = sin(ı).

(d) z = cos(ı).

Complete Solution:

In each case, the idea is to reduce the given expressions to forms that are
already known, using the elementary functions that were discussed in the
text, as well as their properties.

(a) z = 1ı : taking logarithms we have that ln(z) = ı ln(1), where ln(1) =


0 + 2nπı for some integer n, where n = 0 corresponds to the usual leaf
of the logarithm function. It follows that we have ln(z) = 2nπ and

47
48 SOLUTIONS 4

therefore that z = exp(2nπ), where n is an arbitrary integer, and the


particular value n = 0 corresponds to the particular value z = 1 for
this imaginary power.

(b) z = eı : using the Euler formula we have that z = cos(1) + ı sin(1),


where cos(1) and sin(1) are well-defined real values of these functions.

(c) z = sin(ı): by the algorithmic definition of the function sin(z), we have


that

eıı − e−ıı
z =

e−1 − e1
=
2ı 
ı 1
= e− ,
2 e

which is purely imaginary and is written in terms of the well-defined


real number e. Note that this result can also be written as ı sinh(1).

(d) z = cos(ı): by the algorithmic definition of the function cos(z), we have


that

eıı + e−ıı
z =
2
e−1 + e1
=
2 
1 1
= e+ ,
2 e

which is real and is written in terms of the well-defined real number e.


Note that this result can also be written as cosh(1).

Problem 2. Consider the complex function w(z) = z 1/n = n
z where n > 1
is an integer.

(a) Show that w(z) is a multivalued function, with n different values for
each z.

(b) Show that these n values are evenly distributed along a circle in the
complex plane.
EVEN LESS ELEMENTARY FUNCTIONS 49

(c) Determine the radius of this circle and the angles corresponding to each
one of the n possible values.
(d) Build a Riemann surface with n leaves to represent the domain of the
function. Determine the singular point.
(e) Show that the function is analytic throughout this Riemann surface.

Complete Solution:

In order to define the complex function w(z) = z 1/n , it is necessary to use


the logarithm function, in which we must explicitly consider all the leaves of
the Riemann surface. Thus, with z = ρ exp(ıθ) where θ ∈ [−π, π], we have
that
1
w(z) = e n ln(z) , where
ln(z) = ln(ρ) + ıθ + ık2π,

where k is an arbitrary integer, and ln(ρ) is the usual real logarithm function
of the positive real quantity ρ.

(a) Writing w(z) explicitly, we have


1
w(z) = e n [ln(ρ)+ıθ+ık2π]
e n ln(ρ) e(ı n ) e(ı n 2π)
1 θ k
=
= ρ n e(ı n ) e(ı n 2π) .
1 θ k

Since θ is between −π and π, and n > 1, θ/n never makes a complete


turn around the origin, so that both the factor involving ρ and the first
imaginary exponential are single-valued complex functions. On the
other hand, the second exponential attributes to the function various
values, as k varies along the integers. However, since both n and k
are integers, a periodicity is established in this function, because its
value for k = 0 coincides with its value for k = n, since in this case
the argument of the exponential is 2πı. In fact, all values of k given by
k = mn, where m is an arbitrary integer, result in the same value for
the function, including the case of negative multiples of n. Thus, there
are only n different values of the function, which can be represented,
for example, by k ∈ {0, . . . , n − 1}. Thus we see that the function
w(z) = z 1/n is a function of n values, which assigns n different complex
values to each value of z.
50 SOLUTIONS 4

(b) As we already have the n values of the function explicitly written in


polar form,

w(z) = ρ n e[ı( n + n 2π)] , for k ∈ {0, . . . , n − 1},


1 θ k

we verify immediately that the radius ρ1/n does not depend on k, so


that it is the same for all the n values of k, which are therefore located
on the circle with this radius. Furthermore, each one of the values has
an angle equal to the previous one increased by the constant quantity
2π/n, so that the n values are equally spaced along the circle.

(c) As we have already seen in previous items, the radius is ρ1/n and the
n angles are given by

θ 2π
+ k , for k ∈ {0, . . . , n − 1}.
n n

(d) The singular point is the point z = 0, where the angles that differentiate
the Riemann leaves from one another are undefined. Therefore, in order
to build the Riemann surface, it is necessary to remove this point from
the domain, and consider as leaves n superposed copies of the complex
plane, each one with the origin removed. We assign to each of these n
leaves a value of k in {0, . . . , n − 1}. Each of these leaves must be cut
from the origin to infinity, along an arbitrary simple curve, the same
for all leaves, for example the negative real semi-axis. Then, one side
of each of these cuts is glued to the opposite side of the cut on the next
leaf, thus forming a kind of spiraling surface with n turns. Finally, the
remaining sides of the first and last leaves are glued together to form
the Riemann surface that represents the domain of the function, and
over which the image of the function can be represented completely and
unambiguously.

(e) Again, we start from the explicit polar form of the function,

w(z) = ρ n e[ı( n + n 2π)]


1 θ k

    
1 θ k θ k
= ρ n cos + 2π + ı sin + 2π
n n n n
   
1 θ k 1 θ k
= ρ cos
n + 2π + ıρ sin
n + 2π ,
n n n n
EVEN LESS ELEMENTARY FUNCTIONS 51

and we use the polar version of the Cauchy-Riemann conditions in order


to check the analyticity. We have for the relevant expressions involving
the partial derivatives,
 
∂u 1 1 −1 θ k
= ρ n cos + 2π ,
∂ρ n n n
 
1 ∂v 1
−1 1 θ k
= ρn cos + 2π ,
ρ ∂θ n n n
 
1 ∂u 1
−1 (−1) θ k
= ρ n sin + 2π ,
ρ ∂θ n n n
 
∂v (−1) 1 −1 θ k
− = ρ n sin + 2π ,
∂ρ n n n

from which we find that the two Cauchy-Riemann conditions are satis-
fied for all values of ρ and θ except for ρ = 0, because in this case the
derivatives do not exist, since the exponent (1/n) − 1 is negative, given
that n > 1. Furthermore, we see that the two conditions are satisfied
for any value of k, from which it follows that the function is analytic
over the entire Riemann surface.

Problem 3. For each one of the inverse functions listed below, prove that
they can be written in terms of the logarithm as shown in each case.
 √ 
(a) sin−1 (z) = −ı ln ız ± 1 − z 2 .
 √ 
(b) cos−1 (z) = −ı ln z ± z 2 − 1 .
 
(c) tan−1 (z) = ı2 ln ıı +
−z
z .

 √ 
(d) sinh−1 (z) = ln z ± z 2 + 1 .
 √ 
(e) cosh−1 (z) = ln z ± z 2 − 1 .
 
(f) tanh−1 (z) = 1
2 ln 1 +z .
1−z

Complete Solution:
52 SOLUTIONS 4

In each case, we write z as the corresponding direct function of w, and isolate


the exponential of w in terms of z, which always reduces to finding the roots
of a quadratic equation. We use the Baskara formula for this, since it only
depends on the operations of the field and on the square root function, both of
these being concepts that are transparently generalized from the real context
to the complex context.

(a) w(z) = sin−1 (z): we have that z = sin(w), and therefore

eıw − e−ıw
z = ⇒

−ıw
e −e
ıw
− 2ız = 0 ⇒
( eıw )2 − 2ız ( eıw ) − 1 = 0

p
( e ) = ız ± 1 − z 2 ⇒
ıw
 p 
w(z) = −ı ln ız ± 1 − z 2 .

(b) w(z) = cos−1 (z): we have that z = cos(w), and therefore

eıw + e−ıw
z = ⇒
2
−ıw
e +e
ıw
− 2z = 0 ⇒
( eıw )2 − 2z ( eıw ) + 1 = 0 ⇒
p
(e ) = z ±
ıw
z2 − 1 ⇒
 p 
w(z) = −ı ln z ± z 2 − 1 .

(c) w(z) = tan−1 (z): we have that z = tan(w), and therefore

1 eıw − e−ıw
z = ⇒
ı eıw + e−ıw
eıw − e−ıw − ız eıw − ız e−ıw = 0 ⇒
(1 − ız) ( eıw )2 − (1 + ız) = 0 ⇒
r
1 + ız
(e ) =
ıw

1 − ız
 
ı 1 + ız
w(z) = − ln
2 1 − ız
 
ı ı+z
= ln .
2 ı−z
EVEN LESS ELEMENTARY FUNCTIONS 53

(d) w(z) = sinh−1 (z): we have that z = sinh(w), and therefore

ew − e−w
z = ⇒
2
ew − e−w − 2z = 0 ⇒
( ew )2 − 2z ( ew ) − 1 = 0 ⇒
p
( ew ) = z ± z 2 + 1 ⇒
 p 
w(z) = ln z ± z 2 + 1 .

(e) w(z) = cosh−1 (z): we have that z = cosh(w), and therefore

ew + e−w
z = ⇒
2
ew + e−w − 2z = 0 ⇒
( ew )2 − 2z ( ew ) + 1 = 0 ⇒
p
( ew ) = z ± z 2 − 1 ⇒
 p 
w(z) = ln z ± z 2 − 1 .

(f) w(z) = tanh−1 (z): we have that z = tanh(w), and therefore

ew − e−w
z = ⇒
ew + e−w
ew − e−w − z ew − z e−w = 0 ⇒
(1 − z) ( ew )2 − (1 + z) = 0 ⇒
r
w 1+z
(e ) = ⇒
1−z
 
1 1+z
w(z) = ln .
2 1−z

Each of these functions has multiple values, which can be determined by the
analysis of the infinitely many Riemann leaves of the logarithm function, as
well as of the two Riemann leaves of the square root function, which are
explicitly represented in these formulas, where appropriate, by the symbol
±.
54 SOLUTIONS 4

Problem 4. Starting from the definition of the function Γ(x) in terms of


a parametric integral, show that it can be written in each one of the three
forms below.
Z ∞
(a) Γ(x) = dt e−t e(x−1) ln(t) .
0

dt −t x ln(t)
Z
(b) Γ(x) = e e .
0 t
Z t=∞
(c) Γ(x) = d[ln(t)] e−t ex ln(t) .
t=0

Complete Solution:

We start from the original parametric integral that defines Γ(x),


Z ∞
Γ(x) = dt e−t tx−1 .
0

(a) First, we recall the definition of a power with an arbitrary real exponent,
in terms of the logarithm function. Since the function in which we are
interested is real, we take the n = 0 leaf of the logarithm. We have
therefore

tx−1 = Ze(x−1) ln(t) ⇒



Γ(x) = dt e−t e(x−1) ln(t) .
0

(b) Separating the argument of the second exponential in two parts in the
result of the previous item, we have that

e(x−1) ln(t) =e(−1) ln(t) ex ln(t)


1 x ln(t)
= e ⇒
Zt ∞
dt −t x ln(t)
Γ(x) = e e .
0 t

(c) Considering now that

d ln(t) 1
= ,
dt t
EVEN LESS ELEMENTARY FUNCTIONS 55

we have that dt/t = d[ln(t)], so that we obtain


Z t=∞
Γ(x) = d[ln(t)] e−t ex ln(t) ,
t=0

where the integration interval is still written in terms of t. Note that


by setting ξ = ln(t), that is, t = exp(ξ), this can also be written as
Z ∞
Γ(x) = dξ e− exp(ξ) exξ .
−∞

Problem 5. Calculate explicitly the following real values of the function


Γ(x).

(a) Γ(1).

(b) Γ(2).

(c) Γ(1/2).

Answer: π.

Complete Solution:

This involves the explicit calculation, in some particular cases, of the para-
metric integral that defines Γ(x),
Z ∞
Γ(x) = dt e−t tx−1 .
0

(a) For x = 1 we have


Z ∞
Γ(1) = dt e−t t(1−1)
Z0 ∞
= dt e−t
0
∞
= − e−t
0
= −(0 − 1)
= 1,

that is, Γ(1) = 1.


56 SOLUTIONS 4

(b) For x = 2 we have


Z ∞
Γ(2) = dt e−t t(2−1)
Z0 ∞
= dt t e−t .
0

In this case we integrate by parts,


Z ∞
Γ(2) = dt t e−t
0
∞ Z ∞
−t
= −t e + dt e−t
0 0
∞
= −(0 − 0) − e−t
0
= −(0 − 1)
= 1,

that is, Γ(2) = 1.

(c) For x = 1/2 we have


Z ∞
Γ(1/2) = dt e−t t1/2−1
Z0 ∞
= dt e−t t−1/2
Z0 ∞
dt
= √ e−t .
0 t

Making the transformation of variables t = ξ 2 , that is, ξ = t, with
dt = 2ξ dξ, we have


dt
Z
Γ(1/2) = √ e−t
t
Z0 ∞
2ξ dξ −ξ 2
= e
0 ξ
Z ∞
2
= 2 dξ e−ξ .
0
EVEN LESS ELEMENTARY FUNCTIONS 57

Since the integrand is an even function, we can write the integral in the
form
Z ∞
2
Γ(1/2) = 2 dξ e−ξ
0
Z ∞
2
= dξ e−ξ
−∞

= π,

where we used the result for the Gaussian integral over the whole real
line, which is obtained in another problem of this chapter, in the case

α = 1. In short, we have that Γ(1/2) = π.

Problem 6. Show that Γ(z + 1) = zΓ(z) for any complex z except for the
origin z = 0 and the negative integers.

Complete Solution:

We start from the original parametric integral that defines Γ(x), generalized
to a complex argument z, which we can do without problems since x or z are
only fixed parameters for this integration, which remains a real integration
on t,
Z ∞
Γ(z) = dt e−t tz−1 .
0

The only concern we must have is that the integral exist, and as we have
seen in the text it exists so long as ℜ(z) > 0. We simply write Γ(z + 1), and
integrate by parts, to obtain
Z ∞
Γ(z + 1) = dt e−t tz
0
∞ Z ∞
−t z
= −e t + dt e−t ztz−1
0 0
∞ Z ∞
= − e−t ez ln(t) +z dt e−t tz−1
0 0
∞
= − e−t e(x+ıy) ln(t) + zΓ(z),
0

where we wrote z as x + ıy in the exponent of the second exponential in


the integrated term. Joining the two exponentials in the integrated term, we
have for that term
58 SOLUTIONS 4

− e−t e(x+ıy) ln(t) = − e−t+x ln(t)+ıy ln(t)


= − e−t+x ln(t) eıy ln(t) .

In the limit t → ∞ the second exponential, which has as real and imaginary
parts real periodic functions, simply oscillates, remaining with absolute value
limited by 1. In this same limit the argument of the first exponential goes
to −∞, because the ln(t) goes to ∞ much more slowly than −t goes to −∞.
Thus we see that this first exponential goes to zero, as well as the entire
integrated term, when t → ∞. On the other hand, in the limit t → 0 we have
that ln(t) goes to −∞, so that the second exponential still remains limited
and oscillates. In this other limit the argument of the first exponential goes
to −∞ since x > 0, in which case this exponential also vanishes. Thus we see
that the integrated term vanishes in both limits, so long as ℜ(z) > 0. Going
back to the previous equation, we therefore have that

Γ(z + 1) = zΓ(z),

for ℜ(z) > 0. Using the procedure described in the text, we can now use this
property to extend the analytic definition of this function to successive strips
of the complex plane with x < 0. Of course, in this way the property remains
valid throughout the region to which it is possible to extend the function.
For example, for x > −1 we can write that

Γ(z + 1)
Γ(z) = ,
z

since the right-hand side of this equation is well defined for x > −1. Of
course, due to the zero in denominator, this definition cannot be used for the
case z = 0, at which point the function remains undefined, having there an
isolated singularity. Continuing with the process for x > −2 we can write
that

Γ(z + 2)
Γ(z) = ,
z(z + 1)

which remains undefined at z = 0 and at z = −1. Continuing this process


indefinitely, we can extend the function to the whole complex plane except
for the zero and the negative integers, which is also the domain where the
property Γ(z + 1) = zΓ(z) holds.
EVEN LESS ELEMENTARY FUNCTIONS 59

Problem 7. Consider the function


(t − t0 )2
 
f (t) = exp − ,
2τ 2
where t0 is a real constant and τ 6= 0 is a strictly positive real constant, and
consider also the definition of the average value of another function g(t) in
the statistical distribution defined by f (t), which is given by
Z ∞
g(t)f (t) dt
−∞
hgi = Z ∞ .
f (t) dt
−∞

(a) Calculate the average value of t, that is, hti.

Answer: t0 .

(b) Calculate the dispersion of t, that is the quantity


p
σt = h(t − hti)2 i.

Answer: τ .

Complete Solution:

These are two examples of calculation of average values via


Z ∞
2 2
dt g(t) e−(t−t0 ) /(2τ )
hgi = −∞ Z ∞ .
2 2
dt e−(t−t0 ) /(2τ )
−∞

By means of a simple transformation of variables, we can see that the integral


in the denominator is an example of Gaussian integral, whose value is given
by the result that is obtained in another problem of this chapter. Making
ξ = t − t0 we have that dt = dξ, and therefore that
Z ∞ Z ∞
2 2 2 2
dt e−(t−t0 ) /(2τ ) = dξ e−ξ /(2τ )
−∞ −∞
p
= (2τ 2 )π

= 2π τ.
60 SOLUTIONS 4

Here we used the result for the Gaussian integral that is obtained in another
problem of this chapter, with α = 2τ 2 . Let us give √ the name N to this
normalization denominator, that is, we have that N = 2π τ .

(a) We must calculate the following integral,



1
Z
2 /(2τ 2 )
hti = dt t e−(t−t0 ) .
N −∞

Making the transformation of variables ξ = t − t0 , t = ξ + t0 , dt = dξ,


we have

1 ∞
Z
2 2
hti = dξ (ξ + t0 ) e−ξ /(2τ )
N −∞
1 ∞ t0 ∞
Z Z
2 2 2 2
= dξ ξ e−ξ /(2τ ) + dξ e−ξ /(2τ ) .
N −∞ N −∞

The first of these two integrals is zero because it is the integral of an


odd function on a symmetric domain. The second integral is again
simply N , so that we have for the average value

hti = t0 .

(b) We must calculate the following integral,

σt2 = (t − t0 )2
1 ∞
Z
2 2
= dt (t − t0 )2 e−(t−t0 ) /(2τ )
N −∞
1 ∞
Z
2
= dt (t − t0 )2 e−β(t−t0 ) ,
N −∞

where 2
√ we defined the parameter β = 1/(2τ ), that is, we have that τ =
1/ 2β. We can now write this integral as a derivative with respect to β
of another integral, which ends up being once again the normalization
integral of value N ,
Z ∞
1 ∂ 2
σt2 = − dt e−β(t−t0 )
N ∂β −∞
1 ∂N
= − .
N ∂β
EVEN LESS ELEMENTARY FUNCTIONS 61

Writing N = 2π τ in terms of β and taking the derivative, we have

1 ∂ √ 
σt2 = − 2π τ
N ∂β
√ 
1 ∂ π
= − √
N ∂β β
√ 1 1
= π √ .
2N β 3
p
Substituting again the value of N = π/β in terms of β we have
r
√ β 1 1
σt2 = π √
π 2 β3
1 1
= √
2 β2
1
=

= τ 2.

It follows therefore that we have for the dispersion around t0

σt = τ.

Problem 8. (Challenge Problem) Show that, for any real, strictly pos-
itive number α,
Z ∞  2
t √
exp − dt = απ.
−∞ α

Hint: calculate the square of the integral, and transform to a polar coordi-
nate system on the plane.

Complete Solution:

We will calculate the following integral, in which the integration variable can
have any name, so that we chose the name x for it,
Z ∞
2
I= dx e−x /α ,
−∞
62 SOLUTIONS 4

where α > 0. Following the hint given, we will calculate I 2 , because it is


easier to do this than to calculate I directly. Since we will have the product
of two integrals, we need to use two different integration variables, to avoid
confusion, so we chose x and y,
Z ∞ Z ∞
2 2
I2 = dx e−x /α dy e−y /α
Z−∞
∞ Z ∞
−∞
2 2
= dx dy e−(x +y )/α .
−∞ −∞

We now make a transformation in the integration variables, interpreting the


pair (x, y) as Cartesian coordinates in an infinite plane, and changing to the
corresponding polar coordinates r and θ, where

x = r cos(θ),
y = r sin(θ).

The transformation of the integration element is dx dy = r dr dθ, and we also


have that x2 + y 2 = r 2 , so that we can write the integral on the plane R2 in
the form
Z
2 2 2
I = dx dy e−(x +y )/α
2
ZR
2
= r dr dθ e−r /α
R2
Z 2π Z ∞
2
= dθ dr r e−r /α .
0 0

The integral over θ is now immediate, and the integral over r can be done by
reversing the chain rule, due to the additional factor of r which appeared in
the integrand. Thus we have
Z 2π Z ∞
2 2
I = dθ dr r e−r /α
0 0
(−α) −r2 /α ∞

= 2π e
2 0
= −πα(0 − 1)
= απ.

We have therefore that I 2 = απ, and hence it follows that I = απ.
Solutions 5

Geometrical Aspects of the


Functions

We present here complete and commented solutions to all problems proposed


in Chapter 5 of the text. For reference, the propositions of the problems are
repeated here. The problems are discussed in the order in which they were
proposed within the problem set of that chapter.

Problem 1. Let us call stationary those points where a function f (x, y)


of two variables (x, y) has its two first partial derivatives with respect to
these variables equal to zero. Assuming that at least one of the two second
partial derivatives is non-zero, a stationary point of this type can be a point
of local maximum of the function, a point of local minimum of the function,
an inflection point or a saddle point. At a point of local minimum the two
second partial derivatives are strictly positive, at a point of local maximum
the two second partial derivatives are strictly negative, at an inflection point
one of the second derivatives is zero while the other is strictly positive or
strictly negative, and at a saddle point one of the second derivatives is strictly
positive while the other is strictly negative. Prove that the functions u(x, y)
and v(x, y) which constitute an analytic function cannot have any points of
local minimum, local maximum, or inflection, but only saddle points.

Complete Solution:

As noted in the text, the functions u(x, y) and v(x, y) which are the real and
imaginary parts of an analytic function are always harmonic functions, that
is, continuous and differentiable functions that satisfy the Laplace equation
in two dimensions,

63
64 SOLUTIONS 5

∂2 ∂2
f (x, y) + f (x, y) = 0,
∂x2 ∂y 2
at all points where the function is analytic, where we are using f as a generic
name for either u or v. It follows that there can be no point where the two
second partial derivatives are strictly positive, in which case the sum would be
strictly positive, and this equation could not be satisfied, which eliminates
the possibility of points of local minimum. The same can be said for the
case in which the two second partial derivatives are strictly negative, which
eliminates the possibility of points of local maximum. Furthermore, if one of
the two second partial derivatives is zero it follows that the other must also
be zero, which eliminates the possibility of any inflection points as described
in the statement of the problem. Therefore, the only possibilities are that
both second derivatives be zero, or that one be strictly positive while the
other be strictly negative. As we assumed by definition that at least one of
the two partial second derivatives is non-zero, the only remaining possibility
is that there are saddle points.

Problem 2. Consider an arbitrary analytic function w(z) = u(x, y) +


ıv(x, y) of the complex variable z = x + ıy.
(a) Write the gradient vectors in the (x, y) plane of the two harmonic func-
tions u(x, y) and v(x, y).
(b) Show that these two gradient vectors are orthogonal to each other, at
any point of the (x, y) plane where w(z) is analytic.
(c) Show that these two gradient vectors have the same absolute value at
every point of the (x, y) plane where w(z) is analytic.

Complete Solution:

We have to deal here with some basic concepts of the vector calculus in two
dimensions.
(a) We have for the gradient vectors,
 
~ ∂u ∂u
∇u = , ,
∂x ∂y
 
~ ∂v ∂v
∇v = , .
∂x ∂y
GEOMETRICAL ASPECTS OF THE FUNCTIONS 65

Since the function is analytic, we can use the Cauchy-Riemann con-


ditions in order to write this only in terms of partial derivatives of
u(x, y),

 
~ ∂u ∂u
∇u = , ,
∂x ∂y
 
~ ∂u ∂u
∇v = − , .
∂y ∂x

(b) Calculating the dot product of the two gradient vectors, using the latter
form for them, we have

   
~ · ∇v
~ ∂u ∂u ∂u ∂u
∇u = , · − ,
∂x ∂y ∂y ∂x
∂u ∂u ∂u ∂u
= − +
∂x ∂y ∂y ∂x
= 0.

It follows therefore that the two gradient vectors are orthogonal to one
another at all points where the Cauchy-Riemann conditions hold, that
is, at all points where the function w(z) is analytic.

(c) Calculating the absolute value of the gradient vector in each case, using
the latter form for them, we have

s 2
∂u 2
 
~ | = ∂u
|∇u +
∂x ∂y
s 2  2
~ | = ∂u ∂u
|∇v − +
∂y ∂x
s 
∂u 2 ∂u 2
 
= + .
∂x ∂y

It follows therefore that the two gradients have the same absolute value,
at all points where the Cauchy-Riemann conditions hold, that is, at all
points where the function w(z) is analytic.
66 SOLUTIONS 5

Problem 3. Consider the analytic function w(z) = 1/z as a mapping from


the complex plane z = (x, y) onto the complex plane w = (u, v).

(a) Show that the unit circle is mapped onto itself, but not as the identity
map.

(b) Find the fixed points of the mapping, that is, points that are mapped
to themselves.

Answer: ±1.

(c) Show that the interior of the unit circle is mapped onto its exterior,
and vice-versa.

Complete Solution:

Writing the function w(z) = 1/z in the polar representation, where z =


ρ exp(ıθ), we have

w(z) = u + ıv
1 −ıθ
= e .
ρ

(a) Making ρ = 1 we also have that 1/ρ = 1, so that we are on the unit
circle in both the (x, y) plane and the (u, v) plane, which makes it clear
that the unit circle of (x, y) is mapped on the unit circle of (u, v). In
detail, we have the mapping

z = cos(θ) + ı sin(θ) −→
w = cos(θ) − ı sin(θ),

which shows that there is a reversal of the orientation of the unit circle,
since θ is replaced by −θ by the mapping, so that the mapping is not
the point-to-point identity on the circle. Nevertheless, the unit circle
of (x, y) is, in its entirety, mapped onto the unit circle of (u, v).

(b) The fixed points of the mapping are those at which w = z, that is, u = x
and v = y. Of course this is only possible if ρ = 1, because otherwise
the distance from the point to the origin in the (x, y) plane would not
be preserved when one maps the point onto the corresponding point in
the (u, v) plane. In other words, the fixed points are necessarily on the
GEOMETRICAL ASPECTS OF THE FUNCTIONS 67

unit circle. The condition u = x is an identity, but the condition v = y


implies that

sin(θ) = − sin(θ),

which in turn implies that sin(θ) = 0, and therefore that θ = 0 or


θ = π. It follows that the two fixed points of the mapping are z = 1
and z = −1.

(c) This is immediate because, if z is within the unit circle, then ρ < 1,
which implies that 1/ρ > 1, so that w is outside the unit circle. In
other words, the interior of the unit circle of (x, y) is mapped onto the
exterior of the unit circle of (u, v). The reverse situation is also true
because, if z is outside the unit circle, then ρ > 1, which implies that
1/ρ < 1, so that w is within the unit circle. In other words, the exterior
of the unit circle of (x, y) is mapped onto the interior of the unit circle
of (u, v).

Problem 4. Consider the complex function

1
w(z) = z + ,
z
where z = x + ıy and w = u + ıv, seen as a transformation that maps the
complex (x, y) plane onto the complex (u, v) plane.

(a) Show that this transformation maps the unit semicircle with positive
imaginary part of the (x, y) plane onto the real segment (−2, 2) of the
(u, v) plane.

(b) Show that this transformation maps the radii of the unit disk that fall
strictly within this upper unit semicircle onto curves starting in the
real segment (−2, 2), extending to infinity on the (u, v) half-plane with
negative imaginary part.

(c) Show that the curves mentioned in the previous item intersect the seg-
ment (−2, 2) perpendicularly, on the (u, v) plane.

(d) Show that the interior of the semicircle is mapped onto the whole lower
half-plane in the (u, v) plane.
68 SOLUTIONS 5

Complete Solution:

Writing explicitly u and v in terms of polar variables ρ and θ on the (x, y)


plane we have
1
w(z) = z +
z
1
= ρ eıθ + e−ıθ
ρ
   
1 1
= ρ+ cos(θ) + ı ρ − sin(θ) ⇒
ρ ρ
 
1
u(ρ, θ) = ρ+ cos(θ),
ρ
 
1
v(ρ, θ) = ρ− sin(θ).
ρ
(a) The upper unit semicircle is described by ρ = 1 and θ ∈ [0, π], so that
we have that w is real, given by w = 2 cos(θ). In other words, we have
v = 0, so that we are on the real axis of (u, v), and u ∈ [−2, 2], because
the cosine varies from −1 to 1. It follows that the semicircle is mapped
onto this segment.
(b) Let us start with some particular cases for the radius. In the case θ = 0
and ρ ∈ [0, 1], which is one of the two limiting cases of those described
in the statement of the problem, we have that the cosine is 1 and that
the sine is zero, so that w is real,
 
1
w = ρ+ .
ρ

We see that v = 0, so that we are on the real axis of (u, v), and that
for ρ → 1 we have that u → 2, while for ρ → 0 we have that u → ∞, so
that this radius is mapped onto the real semi-axis [2, ∞) of the (u, v)
plane. On the other hand, in the particular case of the vertical radius
given by θ = π/2 and ρ ∈ [0, 1] we have that the cosine vanishes and
that the sine is 1, so that w is purely imaginary,
 
1
w = ρ− ı.
ρ

We see that u = 0, so that we are on the imaginary axis of (u, v), and
that for ρ → 1 we have that v → 0, while for ρ → 0 we have that
GEOMETRICAL ASPECTS OF THE FUNCTIONS 69

v → −∞, so that the radius is mapped onto the negative imaginary


semi-axis of the (u, v) plane. Since the transformation from (x, y) to
(u, v) is clearly symmetric by reflection about the axis y, whereby the
imaginary part of w does not change, but the real part changes sign,
it suffices to examine in detail the interval [0, π/2] of values of θ. The
cases θ = 0 and θ = π/2 have been already analyzed, corresponding
respectively to the part of the positive real semi-axis above u = 2 and to
the negative imaginary semi-axis of the (u, v) plane. For values within
the interval we have cos(θ) > 0 and sin(θ) > 0, so that
 
1
u(ρ, θ) = ρ+ cos(θ),
ρ
 
1
v(ρ, θ) = ρ− sin(θ),
ρ

where u is always positive for any value of ρ ∈ [0, 1], while v is always
negative. As we have seen, when ρ → 1 we have that v vanishes and
u is on the real interval [−2, 2] of the (u, v) plane, showing that all the
curves in the image begin at this real segment. On the other hand,
when ρ → 0 we have that u → ∞ while v → −∞, so that the curves
extend to infinity in the lower right quadrant of the (x, y) plane. The
slope of these curves, which are described by the parameter ρ with θ
constant, is given by

∂v ∂v ∂ρ
=
∂u ∂ρ ∂u
(1 + 1/ρ2 ) sin(θ)
=
(1 − 1/ρ2 ) cos(θ)
(ρ2 + 1) sin(θ)
= .
(ρ2 − 1) cos(θ)

Taking the limit ρ → 0, corresponding to the asymptotic limit for the


curves in (u, v), we have that the slope has the limit

∂v sin(θ)
lim = −
ρ→0 ∂u cos(θ)
= − tan(θ).

The slopes of the curves are therefore all negative, and interpolate
continuously between the positive real semi-axis for θ = 0 and the
70 SOLUTIONS 5

negative imaginary semi-axis for θ = π/2. By symmetry, a similar


situation happens for negative x and u, on the other halves of the
respective planes.

(c) As we saw in the previous item, the slope of the curves in (u, v) is given
by

∂v (ρ2 + 1) sin(θ)
= 2 ,
∂u (ρ − 1) cos(θ)

where both the cosine and the sine are non-zero for θ ∈ (0, π), excluding
θ = 0 and θ = π, provided that we also have that θ 6= π/2. Taking
the limit ρ → 1, which corresponds to the real segment (−2, 2) on the
(u, v) plane, we have that

∂v
lim = ∓∞,
ρ→1 ∂u

where the sign depends on whether we are within the interval θ ∈


(0, π/2) (negative), or within the interval θ ∈ (π/2, π) (positive). This
shows that the curves tend to be vertical as we approach the horizontal
segment (−2, 2) on the (u, v) plane, and are therefore perpendicular
to it. This result can be extended to the case θ = π/2, because in
this case the slope is ∓∞, with a sign that depends on how we take
the limit to the point π/2, whatever the value of ρ, and therefore we
have the negative imaginary semi-axis, which is vertical and therefore
perpendicular to the segment. Note that we can take the two limits,
ρ → 1 and θ → π/2, in any order, always obtaining the same result.
However, the result cannot be extended to the cases θ = 0 and θ = π,
because in these cases the results depend on the order in which we
take the two limits. If we take the limit ρ → 1 first, the result for the
slope is ∓∞, indicating the perpendicularity, but if we take the limit
θ → 0 first, the result is zero, indicating a curve that tends to become
horizontal and therefore parallel to the segment.

(d) As we have seen in the previous items, we can build an infinite family of
curves that covers completely and continuously the domain, namely the
set of all the radii of the upper unit semicircle, so that the corresponding
family of curves on the image covers in a complete and continuous
way the entire lower half-plane, stretching to infinity in all directions
and continuously interpolating between the real positive semi-axis and
GEOMETRICAL ASPECTS OF THE FUNCTIONS 71

the real negative semi-axis. It follows that the interior of the upper
semicircle of the (x, y) plane is mapped to the entire lower half-plane
of the (u, v) plane.

Problem 5. Consider the two oriented curves C1 and C2 in the complex


(x, y) plane, which intersect at a certain point, and the transformation defined
by the analytic function w(z), which maps these two curves onto two other
curves C1′ and C2′ in the complex (u, v) plane, as discussed in the text. Assume
that the gradients of u(x, y) and v(x, y) are not zero at the intersection point.
Consider two infinitesimal variations of z at the intersection point in the (x, y)
plane, denoted in vector language by dz ~ 1 and dz
~ 2 , each one tangent to and
pointing in the positive direction of the corresponding curve.

(a) Show that the sine of the angle θ between the two curves in the (x, y)
plane is given by

dx1 dy2 − dx2 dy1


sin(θ) = ,
~ 1 dz
dz ~2

where dz~ 1 = dx1 + ıdy1 and dz


~ 2 = dx2 + ıdy2 . Show in the same way
that the sine of the angle θ between the corresponding curves C1′ and

C2′ in the complex (u, v) plane, at the intersection point, is given by

 du1 dv2 − du2 dv1


sin θ ′ = ,
~ 1 dw
dw ~ 2

~ 1 = du1 + ıdv1 and dw


where dw ~ 2 = du2 + ıdv2 .

Hint: consider using vector (cross) products.

(b) Prove, using the analyticity properties of the transformation function


w(z), that sin(θ ′ ) = sin(θ). Together with the result that was shown
in the text, cos(θ ′ ) = cos(θ), this suffices to ensure that θ ′ = θ, thus
completing the proof started in the text.

Complete Solution:
72 SOLUTIONS 5

(a) Considering the two infinitesimal displacements dz ~ 1 = dx1 + ıdy1 and


~ 2 = dx2 +ıdy2 on the (x, y) plane, we can build a vector perpendicular
dz
to the plane by means of the cross or vector product dz ~ 1 × dz
~ 2 . On
the one hand, the component of this product in the positive direction
along the normal to the plane can be written as dx1 dy2 − dx2 dy1 . On
the other hand, the absolute value of the product, which is equal to the
absolute value of this component, can be written as dz ~ 1 dz~ 2 sin(θ).
It follows that we can write for the sin(θ), including the sign associated
with the orientation of the normal to the plane,

dx1 dy2 − dx2 dy1


sin(θ) = .
~ 1 dz
dz ~2

Since the two planes have the same structure, it is clear that this same
argument, applied to the corresponding displacements dw ~ 1 = du1 +ıdv1
and dw~ 2 = du2 + ıdv2 on the (u, v) plane, implies the corresponding
relation for θ ′ ,

 du1 dv2 − du2 dv1


sin θ ′ = .
~ 1 dw
dw ~ 2

(b) Let us now calculate sin(θ ′ ) using the formula derived in the previous
item. Using the expression of the differentials of u(x, y) and v(x, y) in
terms of the variations of x and y, we can write dw ~ 1,2 as functions of
the corresponding (dx, dy)1,2 , as was done in the text. Omitting, for
the moment, the indices that identify the curves, we have

~ = (du, dv)
dw
 
∂u ∂u ∂v ∂v
= dx + dy, dx + dy .
∂x ∂y ∂x ∂y

Using the Cauchy-Riemann conditions we can write this solely in terms


of the partial derivatives of u,
 
~ = ∂u ∂u ∂u ∂u
dw dx + dy, − dx + dy .
∂x ∂y ∂y ∂x

As shown in the text, we have for the absolute values in the denominator
of our formulas, for each one of the two curves,
GEOMETRICAL ASPECTS OF THE FUNCTIONS 73

~ = |∇u
dw ~ .
~ | dz

We can also calculate the product in the numerator involving the vari-
~ 1 and dw
ations dw ~ 2 . Of course, this time we need to keep the indices
of the curves explicitly at every step, and we have

du1 dv2 − du2 dv1


  
∂u ∂u ∂u ∂u
= dx1 + dy1 − dx2 + dy2 +
∂x ∂y ∂y ∂x
  
∂u ∂u ∂u ∂u
− dx2 + dy2 − dx1 + dy1
∂x ∂y ∂y ∂x
∂u ∂u ∂u ∂u
= − dx1 dx2 + dx1 dy2 +
∂x ∂y ∂x ∂x
∂u ∂u ∂u ∂u
− dx2 dy1 + dy1 dy2 +
∂y ∂y ∂y ∂x
∂u ∂u ∂u ∂u
+ dx1 dx2 − dx2 dy1 +
∂x ∂y ∂x ∂x
∂u ∂u ∂u ∂u
+ dx1 dy2 − dy1 dy2 .
∂y ∂y ∂y ∂x

Just as it was the case in the text, we find that due to the Cauchy-
Riemann conditions several of the terms that appear cancel each other
off. This time only the mixed products are left, so that we have

∂u 2
 
du1 dv2 − du2 dv1 = (dx1 dy2 − dx2 dy1 ) +
∂x
 2
∂u
+ (dx1 dy2 − dx2 dy1 )
∂y
2
~ | (dx1 dy2 − dx2 dy1 ) .
= |∇u

We can now assemble our formula for sin(θ ′ ),

du1 dv2 − du2 dv1


sin θ ′

=
~ 1 dw
dw ~ 2
2
~ | (dx1 dy2 − dx2 dy1 )
|∇u
=
~ |2 dz
|∇u ~ 1 dz
~2
74 SOLUTIONS 5

(dx1 dy2 − dx2 dy1 )


=
~ 1 dz
dz ~2
= sin(θ),

where we are assuming that |∇u~ | 6= 0. We have therefore that sin(θ ′ ) =


sin(θ), and as was shown in the text, we also have that cos(θ ′ ) = cos(θ).
This is enough to ensure that θ ′ = θ, thus completing the demonstration
that the transformations generated by analytic functions preserve the
angles between curves, and are therefore conformal transformations, at
~ | 6= 0, and therefore |∇v
points where |∇u ~ | 6= 0.
Solutions 6

Border Effects in Capacitors

We present here complete and commented solutions to all problems proposed


in Chapter 6 of the text. For reference, the propositions of the problems are
repeated here. The problems are discussed in the order in which they were
proposed within the problem set of that chapter.

Problem 1. Consider the complex variables z = x + ıy and w = u + ıv,


and the analytic function

w(z) = z + ez ,

interpreted as a mapping or transformation from the complex plane z = (x, y)


onto the complex plane w = (u, v).

(a) Show that this transformation maps the straight line y = 0 of the (x, y)
plane onto the straight line v = 0 of the (u, v) plane.

(b) Show that this transformation maps the two semi-axes of the (x, y)
plane given by y = ±π with x ≥ 0 onto the two semi-axes of the (u, v)
plane given by v = ±π with u ≤ −1, respectively, and that it maps
the two semi-axes of the (x, y) plane given by y = ±π with x ≤ 0 onto
these same two semi-axes of the (u, v) plane.

(c) Show that this transformation maps the strip given by the interval
−π ≤ y ≤ π of the (x, y) plane onto the entire (u, v) plane. In order to
do this consider the mappings of families of curves.
~ | and/or |∇v
(d) Calculating |∇u ~ |, show that this transformation is analyti-
cally invertible within the strip of the (x, y) plane between y = −π and

75
76 SOLUTIONS 6

y = π, including the boundary of this region, except for the two points
(0, π) and (0, −π) located on this boundary.

Complete Solution:

Writing the analytic function w(z) = z + exp(z) explicitly in terms of real


functions, with z = x + ıy, we have

w(z) = u(x, y) + ıv(x, y)


= x + ıy + ex [cos(y) + ı sin(y)] ⇒
x
u(x, y) = x + e cos(y),
v(x, y) = y + ex sin(y).

(a) Making y = 0, we immediately have that v = 0, while for u we have


that

u(x, 0) = x + ex .

We have therefore the line v = 0 in the image plane, or a part of it. It


is easy to see that if we make x → ∞, we will have u → ∞, because
both the term x and the exponential exp(x) diverge to ∞. On the other
hand, if we make x → −∞, the exponential vanishes quickly, and we
have that the limit is determined by the term x, that is, we have that
u → −∞. We see in this way that we travel along the whole straight
line v = 0 in the image plane, that is, that the line y = 0 of the (x, y)
plane is mapped on the line v = 0 of the (u, v) plane.

(b) Making y = ±π, we have that sin(y) = 0 and that cos(y) = −1, so that
we have for u and v,

u(x, ±π) = x − ex ,
v(x, ±π) = ±π.

With this we can already see that we are on the straight lines v = ±π
in the image plane. We now need to determine the maximum value of
u(x), so that we compute its derivative and equate it to zero,
∂u
= 1 − ex = 0 ⇒
∂x
x = 0.
BORDER EFFECTS IN CAPACITORS 77

Calculating also the second derivative, and applying at the point x = 0,


we have
∂2u
= − ex ⇒
∂x2
∂2u
(0) = −1 < 0.
∂x2
Thus we see that the point of local maximum of the function is at x = 0,
where the function has the value u = −1, and where the second deriva-
tive is negative, characterizing in this way a point of local maximum.
It follows that the image can be represented by v = ±π and u ≤ −1,
which are the two semi-axes that constitute the image of the straight
lines y = ±π.

(c) One way to do this is to choose a family of curves in the domain, and
show that the corresponding family of curves on the image plane covers
the entire complex plane. A family of curves that we can consider in
the domain is the set of straight lines with constant y, and x going from
−∞ to ∞. This set covers the strip which constitutes the domain, when
y ranges from −π to π, so that the value of y in this interval serves as
a continuously varying parameter that describes the set of curves. The
specific cases y = 0 and y = ±π have been studied in previous items.
Moreover, if we examine the equations of the transformation from (x, y)
to (u, v), we find that there is a reflection symmetry in y, because the
exchange of the sign of y implies the exchange of the sign of v, without
changing u. What this means is that the straight lines with y > 0
are mapped onto curves in the upper half-plane of the (u, v) plane, and
those with y < 0 are mapped onto curves in the lower half-plane. Thus,
it suffices to analyze in detail the case y > 0. Therefore, let us begin
by examining here one more particular case, the case y = π/2. Making
y = π/2, we have that sin(y) = 1 and that cos(y) = 0, so that we have
for u and v,

u(x, π/2) = x,
π
v(x, π/2) = + ex .
2

In the limit in which x → −∞, we have that v quickly approaches


π/2, while u always equals x. We see therefore that in this case the
mapping tends rapidly to the identity, so that in this limit the strip is
78 SOLUTIONS 6

mapped onto itself. On the other hand, in the limit in which x → ∞,


we still have that u = x, but now v begins to grow exponentially with
x. The curve in the (u, v) plane therefore has an exponential shape,
and extends to infinity. Its slope is given by

∂v ∂v
=
∂u ∂x
= ex ,

which tends to ∞ when x → ∞. It follows that in this limit the angle


between the tangent to the curve and the u axis is π/2. In all other
cases in which 0 < y < π/2 the corresponding curve will also extend
to infinity, but will asymptotically make angles between 0 and π/2
with the axis u, with positive tangent. If we have values of y in this
interval, both cos(y) and sin(y) will be positive and non-zero, and we
will therefore have

u(x, y) = x + ex cos(y),
v(x, y) = y + ex sin(y),

where both u and v tend to ∞ when x → ∞, so that again these curves


extend to infinity in the (u, v) plane. The slope of the curves is given
by

∂v ∂v ∂x
=
∂u ∂x ∂u
ex sin(y)
= .
1 + ex cos(y)

In the limit x → ∞ the slope tends to

∂v ex sin(y)
lim = lim
x→∞ ∂u x→∞ 1 + ex cos(y)
sin(y)
=
cos(y)
= tan(y),

which for the interval of values of y in consideration here is positive,


indicating an angle between 0 and π/2 between the tangent to the curve
BORDER EFFECTS IN CAPACITORS 79

and the u axis. In the complementary case, if we have values of y in


the interval (π/2, π), we still have sin(y) strictly positive, but in this
case cos(y) will be negative and not zero, and we will therefore have

u(x, y) = x + ex cos(y),
v(x, y) = y + ex sin(y),

where v still tends to ∞, but u tends to −∞ when x → ∞, so that


again these curves extend to infinity in the (u, v) plane, but now with
negative slope, contained in the second quadrant. The slope of the
curves is given by the same formula as before, and in the limit x → ∞
the slope tends once again to tan(y), that for the interval of values of y
in consideration here is this time negative, indicating an angle between
π/2 and π between the tangent to the curve and the u axis. Thus we see
that the part with negative x of the strip in the (x, y) plane is mapped
essentially onto the same strip in the (u, v) plane, while the part with
x and y both positive in the strip in the (x, y) plane is mapped on the
part of the upper half-plane of the (u, v) plane which is outside of the
strip. The curves continuously change between the semi-axis defined
by v = 0 and u > 0 and the semi-axis defined by v = π and u < 0. A
mapping symmetric to this one exists in the lower half-plane. Thus we
see, by continuity, that the whole (u, v) plane is mapped from the strip
−π ≤ y ≤ π in the (x, y) plane.
(d) Calculating the gradient of u(x, y), as well as its absolute value, we
have

~
∇u(x, y) = [1 + ex cos(y)] x̂ − [ ex sin(y)] ŷ ⇒
2
~
∇u(x, y) = [1 + ex cos(y)]2 + [ ex sin(y)]2
= 1 + e2x + 2 ex cos(y).

It is not necessary to independently calculate the gradient of v(x, y),


since the two gradients have the same absolute value. The question now
is to determine at what points this expression vanishes. The easiest way
to do this is to check in what conditions the gradient is zero, because
its modulus is zero if and only if it is zero itself. In this case we have
the two conditions

1 + ex cos(y) = 0,
ex sin(y) = 0.
80 SOLUTIONS 6

From the second condition we can conclude that sin(y) must vanish,
because the exponential of a real number is always strictly positive, so
that we have that y = kπ, where k is some integer. The values falling
within the interval of interest, y ∈ [−π, π], are y = 0 and y = ±π. In
the first case, the first condition above is reduced to 1 + exp(x) = 0,
which has no solution for x, because the exponential of a real number
is always positive. There remains therefore the alternative y = ±π,
for which the first condition reduces to 1 − exp(x) = 0, which is only
satisfied for x = 0. It follows that we have only two points belonging to
the strip where the two gradients are zero, at which the transformation
fails to have an analytic inverse, and therefore is not conformal: the
points (0, π) and (0, −π).

Problem 2. Show that the integral


Z −1
du
,
−u0 1 − ex(u)
where

u(x) = x − ex ,

is finite for any finite value of the constant u0 in the interval [1, ∞).

Complete Solution:

Since we have that u(x) = x − exp(x), it follows that we also have that

du = dx − ex dx
= (1 − ex ) dx.

In addition to this, the value u = −1 corresponds to x = 0, and denoting by


x0 the finite value of x corresponding to the finite value u0 = x0 − exp(x0 ),
we have that
Z −1 Z 0
du 1 − ex
= dx
−u0 1 − e
x(u) 1 − ex
Z−x 0
x0
= dx
0
= x0 .

It follows that the integral is finite for any finite value of x0 , which is the
case for any value of u0 in the given interval.
BORDER EFFECTS IN CAPACITORS 81

Problem 3. Consider the complex function w(z) = z 2 where z = x + ıy


and w = u + ıv.

(a) Find the equipotential curves of u and v, that is, those in which these
quantities are constant.
~ u = −∇u
(b) Calculate the electric-field vectors E ~ and E
~ v = −∇v.
~

(c) Show that E~u · E~ v = 0 at all points, where the dot represents the
dot-product of vectors.

(d) Show that E ~u = E ~ v at all points, where the absolute values shown
are vector magnitudes in the usual sense, that is,
q
~u =
E ~u · E
E ~u.

~v .
and similarly for E

(e) Show that the equipotential curves of u and v are the integral curves
~ v and E
(field lines) of E ~ u respectively.

Complete Solution:

Since we have that w(z) = z 2 , with z = x + ıy and w = u + ıv, it follows that

w(z) = (x2 − y 2 ) + ı2xy ⇒


2 2
u(x, y) = x − y ,
v(x, y) = 2xy.

(a) The curves u = u0 are given by p x2 − y 2 = u0 . If the constant u0 is


positive, then we have that x = ± u0 + y 2 , which is a pair of hyperbo-
las asymptotic to the straight lines x = ±y, located in the half-planes
x > 0 and
√ x < 0. If the constant u0 is negative, then we have that
y = ± −u0 + x2 , which is a pair of hyperbolas asymptotic to the
straight lines x = ±y, located in the half-planes y > 0 and y < 0. The
curves v = v0 are given by y = v0 /(2x), for any sign of v0 . In the case in
which v0 > 0 this defines a pair of hyperbolas asymptotic to the coor-
dinate axes, located in the quadrants (x > 0, y > 0) and (x < 0, y < 0).
In the case in which v0 < 0 this defines a pair of hyperbolas asymptotic
to the coordinate axes, located in the quadrants (x > 0, y < 0) and
(x < 0, y > 0).
82 SOLUTIONS 6

(b) Calculating the gradients we have

~ u = −∇u(x,
E ~ y)
= −(2x, −2y)
= −2(x, −y),
~ ~
Ev = −∇v(x, y)
= −(2y, 2x)
= −2(y, x).

(c) Calculating directly we have

~u · E
E ~ v = 4 (x, −y) · (y, x)
= 4 (xy − yx)
= 0.

(d) Calculating directly we have

~u
p
E = 2 x2 + (−y)2 ,
~v
p
E = 2 y 2 + x2 ⇒
~u
E = E~v
p
= 2 x2 + y 2 .

(e) First of all we must determine the integral curves of the fields, or “lines
~ u , for example, these curves are given by rela-
of force”. In the case of E
tions such as y = f (x) or x = f (y). More generally, we can parametrize
the curves through an arc length parameter along them, so that we have
for their description the pair of functions [x(t), y(t)]. The tangent to
the curve is given by a derivative with respect to t, which is therefore
proportional to the field E ~ u . In short, the integral curves of E~ u are
determined by

~ u = −2(x, −y)
E
 
dx dy
= A , ,
dt dt

where A is a real number, which can depend on the position along


the curve, that is, on t. Equating the components of the vectors, we
therefore have the pair of differential equations
BORDER EFFECTS IN CAPACITORS 83

dx
−2x = A ,
dt
dy
2y = A .
dt

Dividing the second equation by the first we eliminate all references to


t, thus obtaining the differential equation

y dy
− = ⇒
x dx
dy dx
= − ⇒
y x
ln(y) = B − ln(x) ⇒
C
y = ,
x

where we have the integration constant C = exp(B). Thus we see that


~ u are in fact associated with the equipotentials
these integral curves of E
of v(x, y), it being enough to identify C as v0 /2, in order to complete
the association. Repeating the argument in the case E ~ v we have

~ v = −2(y, x)
E
 
dx dy
= A , .
dt dt

Equating the components of the vectors, we have therefore the pair of


differential equations

dx
−2y = A ,
dt
dy
−2x = A .
dt

Dividing the second equation by the first we obtain the differential


equation

x dy
= ⇒
y dx
x dx = y dy ⇒
x2 = B + y 2 ⇒
2 2
x −y = B,
84 SOLUTIONS 6

where we have the integration constant B. Thus we see that these


integral curves of E~ v are in fact associated with the equipotentials of
u(x, y), it being enough to identify C as u0 , in order to complete the
association.

Problem 4. Consider the complex function w(z) = z where z = x + ıy
and w = u + ıv.
(a) Write the real part u(x, y) using polar coordinates and show that it is
the potential of the electrostatic problem of a semi-infinite grounded
metal plate whose cross section takes up the negative real semi-axis.
Note: the conducting plate being grounded means that it is at zero
electrical potential.
(b) For angles θ 6= ±π determine the behavior of the potential for ρ → ∞.
~ = −∇u.
(c) Calculate the electric-field vector E ~ Write the result in terms
of the variables ρ and θ.

Answer:

~ = − cos(θ/2)
E √ x̂ −
sin(θ/2)
√ ŷ.
2 ρ 2 ρ

~ for ρ constant and θ → ±π.


(d) Calculate the limits of E
(e) After calculating these limits, determine the limits of the results ob-
tained when we make ρ → ∞.
(f) Calculate the surface charge density σ on the plate, using the fact that
σ
En = ,
2ε0
where En is the component of the electric field that is normal to the
plate, pointing away from it. Determine at which point σ(ρ) has a
singular behavior.
(g) Since we are, in fact, looking at a two-dimensional section of a three-
dimensional problem, consider that the negative real semi-axis is a slice
of unit width of the infinite half-plane. Calculate the total electric
charge within this slice between the origin ρ = 0 and a particular value
ρ0 of ρ.

Answer: −2ε0 ρ0 .
BORDER EFFECTS IN CAPACITORS 85

Complete Solution:

Using the polar representation z = ρ exp(ıθ), we have for the function w(z) =

ρ exp(ıθ/2), where we will use for θ the interval [−π, π].

(a) We have for the real part of w = u + ıv,


u(ρ, θ) = ρ cos(θ/2),

which of course is a solution of the Laplace equation. The condition


u(ρ, θ) = 0 is satisfied in two cases, ρ = 0, which is the origin of the
complex plane, and cos(θ/2) = 0, that within the chosen interval of
values of θ implies that θ = ±π. It follows that this function is zero
on the negative real semi-axis, including the origin. It can therefore
be interpreted as the potential of a grounded electrically conducting
semi-infinite plate, whose section is placed on this semi-axis.

(b) If we have θ 6= ±π, then the potential u(ρ, θ) is not zero, so that we
have


lim u(ρ, θ) = lim ρ cos(θ/2)
ρ→∞ ρ→∞
= ∞,

that is, the potential goes to infinity as a square root of the distance
from the origin. A potential that tends to infinity at infinity, instead
of tending to zero, is characteristic of distributions of sources (electric
charges) which extend to infinity, in this case along the half-plane where
the plate is.

(c) We can calculate E ~ = −∇u~ in two ways, using polar coordinates or


Cartesian coordinates for the ∇ ~ operator. Let us start making the
calculation in polar coordinates, which is simpler but less familiar. In
~
this case we have for the operator ∇,

~ ∂ 1 ∂
∇ = ρ̂ + θ̂ ,
∂ρ ρ ∂θ

where ρ̂ and θ̂ are the two versors of the coordinate system, so that we
~ with extreme simplicity,
obtain for E,
86 SOLUTIONS 6

~ = −∇u
E ~
∂u(ρ, θ) 1 ∂u(ρ, θ)
= −ρ̂ − θ̂
∂ρ ρ ∂θ
cos(θ/2) sin(θ/2)
= − √ ρ̂ + √ θ̂.
2 ρ 2 ρ

We can also do the calculation in the Cartesian system. For this, we


must write u in terms of x and y, which may be done by means of the
use of the formula of the cosine of the half-arc,


u(x, y) = ρ cos(θ/2)
r
√ 1 + cos(θ)
= ρ
r r 2
ρ ρ+x
=
2 ρ
1 √
= √ ρ + x,
2
p
where ρ = x2 + y 2 . We can now easily calculate the two partial
derivatives with respect to x and to y, recalling that

∂ρ x
= ,
∂x ρ
∂ρ y
= ,
∂y ρ

so that we have

~ = −∇u
E ~
∂u(x, y) ∂u(x, y)
= −x̂ − ŷ
∂x ∂y
 
1 1 x 1 1 y
= −x̂ √ √ 1+ − ŷ √ √
2 2 ρ + x ρ 2 2 ρ + x ρ
1 1 1 1
= −x̂ √ p [1 + cos(θ)] − ŷ √ √ sin(θ)
8ρ 1 + cos(θ) 8ρ 1 + cos θ
1 p 1 sin(θ)
= −x̂ √ 1 + cos(θ) − ŷ √ √ .
8ρ 8ρ 1 + cos θ
BORDER EFFECTS IN CAPACITORS 87

We can write this result once again in terms of θ/2, using the trigono-
metric formulas of the double arc,

cos(θ) = cos2 (θ/2) − sin2 (θ/2),


sin(θ) = 2 sin(θ/2) cos(θ/2),

thus obtaining

~ = − cos(θ/2)
E √ x̂ −
sin(θ/2)
√ ŷ.
2 ρ 2 ρ

This is curiously similar to the previous result, written in the basis


(ρ̂, θ̂) instead of (x̂, ŷ). In fact, it is so similar that, since these are two
very different bases, one may have the impression that one of the two
results must be wrong. However, it can be shown that these two bases,
in fact, satisfy the curious identity

x̂ cos(θ/2) + ŷ sin(θ/2) = ρ̂ cos(θ/2) − θ̂ sin(θ/2),

which is somewhat unexpected, but true. Check it out!

(d) For θ → ±π, we have that θ/2 → ±π/2, and therefore that cos(θ/2) →
0 and sin(θ/2) → ±1. Using each of the two ways in which we wrote
~ we therefore have
E,

~ cos(θ/2) sin(θ/2)
E = −
√ ρ̂ + √ θ̂
2 ρ 2 ρ
1
→ ± √ θ̂,
2 ρ
~ = − cos(θ/2)
E √ x̂ −
sin(θ/2)
√ ŷ
2 ρ 2 ρ
1
→ ∓ √ ŷ.
2 ρ

The sign difference is due to the fact that, for θ = ±π, we have that
~ is normal to the plate, and that in
θ̂ = −ŷ. Thus we see that the field E
either one of the two sides it points towards the plate, which indicates
the existence of negative electrical charges on the plate.
88 SOLUTIONS 6


(e) Since the absolute value of the field is proportional to 1/ ρ, it follows
that in the limit ρ → ∞ along the plate the field goes to zero. Thus, it
is zero if we distance ourselves infinitely from the edge, despite being
singular on the edge itself, ρ = 0.

(f) The field component normal to the plate is given by the component θ̂
or ŷ on the plate, with the appropriate sign, which is determined by
the external normal to the material of the plate, which in this case is
negative,

−1
En = √ .
2 ρ

Since we have that σ = 2ε0 En , it follows that

−ε0
σ(ρ) = √ .
ρ

Just like the field, this surface charge density is singular on the edge
of the plate at ρ = 0, where it diverges to infinity, and goes to zero as
ρ → ∞.

(g) On a unit-width strip, the total charge is given by the integral of σ(ρ)
on ρ, from the edge ρ = 0 to some value ρ0 of ρ,

Z ρ0
Q = dρ σ(ρ)
0
Z ρ0
1
= −ε0 dρ √
0 ρ
 ρ0

= −ε0 2 ρ
√ 0
= −2ε0 ρ0 .

As we already expected, this total charge is negative. Thus we see that,


although the charge density is singular on the edge, the total charge on
a finite area that includes the edge is finite.
BORDER EFFECTS IN CAPACITORS 89

Problem 5. (Challenge Problem) Consider a capacitor formed by two


semi-infinite plates. Each plate has an edge with the form of an infinite
straight line and the two plates are placed forming a wedge with an angle
θ0 ≪ π/4, with the two edges placed together but electrically isolated from
one another. Determine the capacitance per unit area for a unit-width slice
of the plates located at distances between r1 > 0 and r2 > r1 from the
edges. Do this by means of a conformal transformation in the complex plane,
that maps an infinite capacitor on this wedge capacitor, except for the point
corresponding to the edge of the plates.
Answer:
ε0 ln(r2 ) − ln(r1 )
.
θ0 r2 − r1

The solution to this problem was found by Prof. Henrique Fleming. It will
be addressed again later in this series of books, through the use of other
techniques.

Complete Solution:

First, let us describe the conformal map that solves the problem. Let us
consider the complex planes z = x + ıy and w = u + ıv, and the mapping
w(z) = exp(αz/2), where α is a strictly positive real constant. It follows that
we have for u(x, y) and v(x, y),

u(x, y) = eαx/2 cos(αy/2),


v(x, y) = eαx/2 sin(αy/2).

We will determine the images in the collection of straight lines x ∈ (−∞, ∞),
y = y0 of the (x, y) plane. If we describe the (u, v) plane in terms of polar
variables,
p
ρ = u2 + v 2
= eαx/2 ,
α
θ = y,
2
we verify that these lines have as images the semi-axes that form angles
θ = αy0 /2 with the axis x, on which x → −∞ corresponds for ρ → 0 and
x → ∞ corresponds to ρ → ∞, that is, ρ ∈ (0, ∞), and the semi-axes
extend from the origin to infinity, but do not include the origin. Thus we
90 SOLUTIONS 6

see that this transformation takes a capacitor of horizontal parallel plates in


the (x, y) plane onto a capacitor with angled plates in the (u, v) plane, which
do not touch at the origin due to the fact that it is outside the domain of
the mapping. Of course we will have a singular behavior at the origin of the
(u, v) plane. We can therefore start from a pair of plates located at y = ±1
with opposite electrical potentials,

x ∈ (−∞, ∞),
y = ±1,
V0
φ(x, 1) = ,
2
V0
φ(x, −1) = − ,
2
which represents an infinite parallel-plate capacitor, in which there is a poten-
tial difference V0 . In terms of dimensional variables X = xd/2 and Y = yd/2,
such that Y = d/2 corresponds to y = 1, and where d is the distance between
the capacitor plates, we can write a complex potential Φ(X, Y ) on the (x, y)
plane,

V0
Φ(X, Y ) = z
2
V0
= (x + ıy)
2  
V0 2X 2Y
= +ı
2 d d
V0
= (X + ıY ).
d

Since the function w(z) = z is analytic, both the real part and the imaginary
part of this function are solutions of the Laplace equation in the (x, y) plane.
The imaginary part of Φ(X, Y ) is the electrical potential inside the capacitor
in the (x, y) plane,

V0 Y
φ(X, Y ) = ,
d

which also satisfies the boundary conditions, because Φ(X, d/2) = V0 /2 and
Φ(X, −d/2) = −V0 /2. It follows that the corresponding complex potential
in the (u, v) plane is given by Φ′ (w) = Φ[z(w)], where z(w) is the inverse
mapping of the mapping w(z), that is, z(w) = 2 ln(w)/α, so that we have
BORDER EFFECTS IN CAPACITORS 91

V0 2 ln(w)
Φ′ (w) =
2 α
ln(ρ) + ıθ
= V0 .
α
This is an analytic function in the (u, v) plane, so that its real and imaginary
parts are solutions of the Laplace equation in that plane. We can now see
that the imaginary part of this expression represents the potential within the
capacitor in the (u, v) plane, because we have that

θ
φ′ (ρ, θ) = V0 ,
α
so that for θ = α/2 we have the constant potential V0 /2, and for θ = −α/2
we have the constant potential −V0 /2. Thus, the potential difference in the
image capacitor is also V0 , and the angle between the plates is α, so that we
have α = θ0 , and therefore we have
θ
φ′ (ρ, θ) = V0 .
θ0
Note that the potential does not depend on ρ at all. The next step is to
calculate the electric field inside the capacitor. For this, it is easier to use
the expression of the gradient in polar coordinates. In this case we have for
~
the operator ∇,

~ = ρ̂ ∂ + 1 θ̂ ∂ ,

∂ρ ρ ∂θ
~ = −∇φ
so that we to obtain E ~ ′ , very simply,

~ = ρ̂ 0 − 1 θ̂ V0 1 .
E
ρ θ0
V0
= − θ̂.
θ0 ρ
Thus we see that the electric field has no radial component, but only the
angular component, in the direction of −θ̂. Furthermore, this component
does not depend on θ, but only on ρ. It follows that it is very simple to
see that the component of the electric field in the direction of the external
normal at the top plate is given by
V0
En = ,
θ0 ρ
92 SOLUTIONS 6

so that the surface charge density is given by

σ(ρ) = ε0 En
ε0 V0
= .
θ0 ρ

We must now calculate the total amount of charge contained in the top plate
for values of ρ between r1 and r2 . Note that the hypothesis that θ0 be small
is only necessary for us to actually cut out this part from the full capacitor,
and still be able to neglect the edge effects. We therefore have for the charge,
considering a unit-width slice of the capacitor in the direction perpendicular
to the complex plane,
Z r2
Q = dρ σ(ρ)
Zr1r2
ε0 V0
= dρ
r1 θ0 ρ
Z r2
ε0 V0 1
= dρ
θ 0 r1 ρ
 r2
ε0 V0
= ln(ρ)
θ0 r
 1
ε0 V0 r2
= ln .
θ0 r1

Since the capacitance is given by C = Q/V0 and the capacitor area is pro-
portional to r2 − r1 , we have for capacitance C per unit area

ε0 ln(r2 /r1 )
C =
θ0 r2 − r1
ε0 ln(r2 ) − ln(r1 )
= .
θ0 r2 − r1

Note that the result diverges if we make r1 → 0, because in this case we


approach the singularity at the common edge of the two plates. We can
recover the result for a parallel plate capacitor, in the approximation in which
the edge effects are neglected, if we make both r1 and r2 very large, while
we make θ0 ever smaller, in such a way that the distance between the plates,
which is given approximately by the arc d = r1 θ0 , remains constant. Making
r2 = r1 + ∆r, for a constant ∆r, we have in the limit in which r1 → ∞
BORDER EFFECTS IN CAPACITORS 93

ε0 ln(1 + ∆r/r1 )
C =
θ0 ∆r
ε0 ∆r/r1

θ0 ∆r
ε0
= ,
r1 θ0
where we see appearing in the denominator the distance d between the plates.
Since this is the capacitance per unit area, for a capacitor with total area A
we therefore have
ε0 A
C= ,
d
which is the usual result for this type of approximation.
94 SOLUTIONS 6
Solutions 7

Complex Calculus I:
Differentiation

We present here complete and commented solutions to all problems proposed


in Chapter 7 of the text. For reference, the propositions of the problems are
repeated here. The problems are discussed in the order in which they were
proposed within the problem set of that chapter.

Problem 1. Prove that the Leibniz formula for the derivative of a product
applies to the product of two analytic functions. In other words, given two
analytic functions w1 (z) and w2 (z), each one of which has, therefore, a well-
defined complex derivative, show that the derivative of the product-function
w(z) = w1 (z)w2 (z) is given by
dw dw1 dw2
(z) = (z) w2 (z) + w1 (z) (z).
dz dz dz

Complete Solution:

The simplest way to prove the validity of the Leibniz rule for complex deriva-
tives is to start from the definition of the derivative of the product-function
w(z) = w1 (z)w2 (z),
dw(z) w(z + δz) − w(z)
= lim
dz δz→0 δz
w1 (z + δz)w2 (z + δz) − w1 (z)w2 (z)
= lim
δz→0
 δz
w1 (z + δz)w2 (z + δz) − w1 (z + δz)w2 (z)
= lim +
δz→0 δz

95
96 SOLUTIONS 7

w1 (z + δz)w2 (z) − w1 (z)w2 (z)
+ ,
δz

where we added and subtracted the term w1 (z + δz)w2 (z) in the numerator,
in order to isolate the variations of w1 (z) and w2 (z). We therefore have,
making this separation,

dw(z) w1 (z + δz)w2 (z + δz) − w1 (z + δz)w2 (z)


= lim +
dz δz→0 δz
w1 (z + δz)w2 (z) − w1 (z)w2 (z)
+ lim

δz→0 δz 
w2 (z + δz) − w2 (z)
= lim w1 (z + δz) +
δz→0 δz
 
w1 (z + δz) − w1 (z)
+ lim w2 (z)
δz→0 δz
 
w2 (z + δz) − w2 (z)
= w1 (z) lim +
δz→0 δz
 
w1 (z + δz) − w1 (z)
+ lim w2 (z)
δz→0 δz
dw2 (z) dw1 (z)
= w1 (z) + w2 (z).
dz dz
It is also possible to write the complex derivative of w = u + ıv in terms of
the partial derivatives of u and v, for example with respect to x, to do the
same with w1 and w2 , and to manipulate these expressions in order to reduce
the problem to the use of the Leibniz rule for these real functions, for which
we already know that it is valid.

Problem 2. Starting from the complex derivative

d −1 1
z = − 2,
dz z
prove by finite induction the more general formula, for a positive integer n,

d −n n
z = − n+1 .
dz z

Complete Solution:

Let us use the fact, assumed known, that


COMPLEX CALCULUS I: DIFFERENTIATION 97

dz −1
= −z −2 .
dz
In order to show the general result by induction, we assume its validity in
the case n − 1, that is, we assume that

dz −(n−1)
= −(n − 1)z −n .
dz
We now consider the derivative of z −n , write this function as the product of
z −1 with z −(n−1) , and use the Leibniz rule for this product,

d z −1 z −(n−1)
 
dz −n
=
dz dz
dz −1 dz −(n−1)
= z −(n−1) + z −1 .
dz dz
Using the results that have been assumed, we have

dz −n
= −z −2 z −(n−1) − z −1 (n − 1)z −n
dz
= −z −(n+1) − (n − 1)z −(n+1)
= −nz −(n+1) .

which proves the case n, and therefore the general case, that is, we do have,
in fact, that

dz −n
= −nz −(n+1) ,
dz
for all n > 0.

Problem 3. Starting from the known real derivatives, derive the formulas
for the derivatives of the analytic functions that follow.

(a) w(z) = cosh(z).

(b) w(z) = sinh(z).

(c) w(z) = cos(z).

(d) w(z) = sin(z).


98 SOLUTIONS 7

Complete Solution:

In each case, we write the function in terms of known real functions, and
choose an appropriate direction for the variation dz in the complex plane, in
order to simplify the calculation, since the derivative is independent of the
direction of variation.
(a) w(z) = cosh(z): we make dz = dx, and get

ez + e−z
cosh(z) =
2
ex eıy + e−x e−ıy
= ⇒
2
d cosh(z) ex eıy − e−x e−ıy
=
dz 2
ez − e−z
=
2
= sinh(z).

(b) w(z) = sinh(z): we make dz = dx, and get

ez − e−z
sinh(z) =
2
ex eıy − e−x e−ıy
= ⇒
2
d sinh(z) ex eıy + e−x e−ıy
=
dz 2
z
e +e −z
=
2
= cosh(z).

(c) w(z) = cos(z): we make dz = ıdy, and get

eız + e−ız
cos(z) =
2
e e + e−ıx ey
ıx −y
= ⇒
2
d cos(z) − eıx e−y + e−ıx ey
= ı−1
dz 2
eız − e−ız
= −

= − sin(z).
COMPLEX CALCULUS I: DIFFERENTIATION 99

(d) w(z) = sin(z): we make dz = ıdy, and get

eız − e−ız
sin(z) =

eıx e−y − e−ıx ey
= ⇒

d sin(z) − eıx e−y − e−ıx ey
= ı−1
dz 2ı
eız + e−ız
=
2
= cos(z).

Problem 4. Derive the second-order complex differential equations and the


auxiliary conditions that are satisfied by the following analytic functions.

(a) w(z) = cosh(z).

(b) w(z) = sinh(z).

(c) w(z) = cos(z).

(d) w(z) = sin(z).

Complete Solution:

In each case, we start from the known derivatives of each function, and build a
second-order differential equation satisfied by it. Since each complex function
is reduced to the corresponding real function on the real axis, it suffices to
look for the auxiliary conditions on the real axis.

(a) w(z) = cosh(z): the first two derivatives are given by

d cosh(z)
= sinh(z),
dz
d sinh(z)
= cosh(z) ⇒
dz
d2 cosh(z)
= cosh(z).
dz 2

We have therefore the differential equation and the auxiliary conditions


100 SOLUTIONS 7

d2 f (z)
− f (z) = 0,
dz 2
f (0) = 1,
df (z)
(0) = 0.
dz

(b) w(z) = sinh(z): the first two derivatives are given by

d sinh(z)
= cosh(z),
dz
d cosh(z)
= sinh(z) ⇒
dz
d2 sinh(z)
= sinh(z).
dz 2

We have therefore the differential equation and the auxiliary conditions

d2 f (z)
− f (z) = 0,
dz 2
f (0) = 0,
df (z)
(0) = 1.
dz

(c) w(z) = cos(z): the first two derivatives are given by

d cos(z)
= − sin(z),
dz
d sin(z)
= cos(z) ⇒
dz
d2 cos(z)
= − cos(z).
dz 2

We have therefore the differential equation and the auxiliary conditions

d2 f (z)
+ f (z) = 0,
dz 2
f (0) = 1,
df (z)
(0) = 0.
dz
COMPLEX CALCULUS I: DIFFERENTIATION 101

(d) w(z) = sin(z): the first two derivatives are given by

d sin(z)
= cos(z),
dz
d cos(z)
= − sin(z) ⇒
dz
2
d sin(z)
= − sin(z).
dz 2

We have therefore the differential equation and the auxiliary conditions

d2 f (z)
+ f (z) = 0,
dz 2
f (0) = 0,
df (z)
(0) = 1.
dz

Problem 5. Show that the chain rule applies to the composition of analytic
functions. In other words, show that, if f (z) and g(z) are two analytic
functions, and w(z) = f (g(z)) is the composite function, then
 
dw(z) df (z) dg(z)
= (g) .
dz dz dz

Complete Solution:

In this case the simplest thing to do is to work out the direct proof, similar
to what is done in the real case. We have that w(z) = f [g(z)], so that by
definition the derivative is written as
dw(z) w(z + δz) − w(z)
= lim
dz δz→0 δz
f [g(z + δz)] − f [g(z)]
= lim .
δz→0 δz
While g(z + δz) is not equal to g(z), that is, so long as g(z) is not constant in
a neighborhood of the point z, we can multiply and divide by the difference
of these two values, thereby obtaining
 
dw(z) f [g(z + δz)] − f [g(z)] g(z + δz) − g(z)
= lim .
dz δz→0 g(z + δz) − g(z) δz
102 SOLUTIONS 7

Note that the case in which g(z) is constant within the neighborhood must
be examined separately, but it is trivial, because in this case g(z) = g0 and
thus f [g(z)] = f (g0 ), which is also constant. It follows that the derivative
of w(z) with respect to z is zero at this point, as is the derivative of g(z)
with respect to z, so that the chain rule is satisfied, whatever the value of
the derivative of f (g) with respect to g, as long as all involved functions are
differentiable, which they always are, since they are analytic at the point z.
Since all the derivatives exist, we can separate the two ratios in the above
formula, into two different limits,
  
dw(z) f [g(z + δz)] − f [g(z)] g(z + δz) − g(z)
= lim lim
dz δz→0 g(z + δz) − g(z) δz→0 δz
df (z) dg(z)
= (g) ,
dz dz
so that the chain rule is proved.

Problem 6. Consider the following analytic function as a conformal trans-


formation, that is, a transformation that preserves the angles between ori-
ented curves that intersect each other, mapping the (x, y) plane onto the
(u, v) plane,

w(z) = u(x, y) + ıv(x, y)


1
= z+ .
z

~ u = −∇u
(a) Calculate the electric field vectors associated with u and v, E ~
~ ~
and Ev = −∇v.

(b) Show that E ~ u = 0 at the points (−1, 0) and (1, 0) of the complex
~ v = 0 at these points.
plane (x, y), and that therefore we also have E

(c) Show that a radius inside the semicircle of the (x, y) plane with θ very
small and ρ ∈ [ρ0 , 1], where ρ0 > 0 is a small number, is mapped on
a curve which is very close to the positive real semi-axis of the (u, v)
plane, a curve which is, however, still perpendicular to the real segment
(−2, 2) of the (u, v) plane, crossing this segment at a point near (2, 0).

(d) Show that in the θ → 0 limit the curve tends to the positive real semi-
axis [2, ∞), which is obviously not perpendicular to the segment (−2, 2)
of the (u, v) plane.
COMPLEX CALCULUS I: DIFFERENTIATION 103

(e) Show that at the points (−1, 0) and (1, 0), the complex derivative of
the function w(z) vanishes, that is, that

dw
(−1, 0) = 0,
dz
dw
(1, 0) = 0.
dz

Just as an informative note, let us recall that at the points where we have
~ u = 0 the proof that the conformal transformation preserves angles does
E
not apply, so that the preservation of angles does not hold for curves that
cross at these points. Since the derivative of the function is zero at these
points, it is not analytically invertible at them, which means that the inverse
function of w(z) has singularities at these points. It can be said that these
points are singular points of the conformal transformation, in a sense that is
not the same as the notion of singularity of the complex function, which is
analytic at these points. It is the inverse of this analytic function that has
singularities at these points.

Complete Solution:

Writing explicitly u(x, y) and v(x, y) we have


1
w(z) = z +
z
x − ıy
= x + ıy + 2 ⇒
x + y2
 
1
u(x, y) = x 1 + 2 ,
ρ
 
1
v(x, y) = y 1 − 2 ,
ρ
where, in order to simplify the differentiations, we wrote the expression in
terms of ρ2 = x2 + y 2 .

(a) In order to obtain the fields, in Cartesian coordinates, we calculate the


relevant partial derivatives, of which two are naturally dependent on
the other two, due to the Cauchy-Riemann conditions, and we thus
obtain

2x2
 
∂u 1
= 1+ 2 − 4
∂x ρ ρ
104 SOLUTIONS 7

ρ4 + ρ2 − 2x2
=
ρ4
ρ4 − x 2 + y 2
= ,
ρ4
∂u 2xy
= − 4 ,
∂y ρ
∂v 2yx
= ,
∂x ρ4
2y 2
 
∂v 1
= 1− 2 + 4
∂y ρ ρ
4 2
ρ − ρ + 2y 2
=
ρ4
ρ4 − x 2 + y 2
= ⇒
ρ4
~u = ρ4 − (x2 − y 2 ) 2xy
E 4
x̂ − 4 ŷ,
ρ ρ
~v = 2xy ρ − (x − y 2 )
4 2
E x̂ + ŷ.
ρ4 ρ4

(b) For the field vectors to be zero, all their components must be zero.
From two of the components it follows that we must have xy = 0,
which means that x = 0, y = 0 or both. From the other two compo-
nents we have that ρ4 = x2 − y 2 . Thus we see that we cannot have
x = 0, because this would reduce this equation to y 4 = −y 2 , which is
impossible because y is real. Thus, it is necessary that y = 0, and it
follows therefore that x4 = x2 , from which it follows that x2 = 1 and
therefore that x = ±1. We have therefore that the two gradient vectors
are zero at only two points, (1, 0) and (−1, 0).

(c) Writing u(x, y) and v(x, y) completely in terms of the polar coordinates
ρ and θ, we have
 
1
u(ρ, θ) = ρ+ cos(θ),
ρ
 
1
v(ρ, θ) = ρ− sin(θ).
ρ

For θ ≪ 1 and ρ ∈ [ρ0 , 1] with ρ small and positive, we have that


COMPLEX CALCULUS I: DIFFERENTIATION 105
 
1
u(ρ, θ) = ρ+ (1 − θ 2 /2), where
ρ
(ρ + 1/ρ) ∈ [2, U1 ],
 
1
v(ρ, θ) = ρ− θ, where
ρ
(ρ − 1/ρ) ∈ [−U2 , 0],

where for ρ0 near zero the values of U1 and U2 are positive and large,
but finite. Thus we see that in the θ → 0 limit the coordinate u of
the image in the (u, v) plane tends to (ρ + 1/ρ), which is within the
interval [2, U1 ], while the corresponding coordinate v tends to zero.
Therefore, the curve on the image tends to approach the interval [2, U1 ]
on the positive real semi-axis of the (u, v) plane. On the other hand,
we can determine the slope of the tangent to the curve on the image,
calculating the variations of u and v, when we vary ρ with a fixed value
of θ,
 
1
du = 1 − 2 cos(θ)dρ,
ρ
 
1
dv = 1 + 2 sin(θ)dρ ⇒
ρ
du 2
ρ − 1 cos(θ)
= .
dv ρ2 + 1 sin(θ)

For θ small but non-zero, the second ratio is a large but finite number,
so that in the ρ → 1 limit, in which we approach the point defined by
u = 2 cos(θ) and v = 0, we have for this ratio of variations

du
= 0,
dv

which means that we approach a point of the segment [−2, 2] such that
u tends to remain constant along the curve while v varies, that is, the
curve is orthogonal to the segment. For θ ≪ 1 the contact point is
given by (2 − θ 2 , 0), that is, it is a point within the segment and very
close to the point (2, 0).

(d) As shown in the previous item, in the θ → 0 limit the coordinate u of


the image in the (u, v) plane tends to (ρ + 1/ρ), which is in the interval
106 SOLUTIONS 7

[2, U1 ], while the corresponding coordinate v tends to zero. If we make


ρ0 → 0 after that, we have that

U1 = ρ0 + 1/ρ0
→ ∞,

so that in the θ → 0 limit the image is the real semi-axis [2, ∞).
(e) We can simply calculate the derivative in the usual way,
dw(z) 1
= 1 − 2,
dz z
so that the derivative is zero only for z 2 = 1, that is, for z = 1 and z =
−1, which correspond to the points (1, 0) and (−1, 0) on the complex
(u, v) plane.

Problem 7. Starting from the known real derivatives, derive the formula
for the complex derivative of the analytic function w(z) = ln(z). Show ex-
plicitly that this derivative gives the same result on all the infinitely many
leaves of the Riemann surface of the logarithm function, thus mapping them
all onto a function that has a Riemann surface with a single leaf, which is
simply the complex plane except for one point.

Complete Solution:

The definition of the logarithmic function in terms of known real functions,


including the discrimination of the leaves of the Riemann surface, by means
of an integer n, in terms of the polar representation z = ρ exp(ıθ), is given
by
w(z) = ln(z)
= ln(ρ) + ıθ + n(ı2π).
We can choose as the direction for the variation of z either the radial direction,
using dz = (dρ) exp(ıθ), or the angular direction, using dz = ıρ exp(ıθ)(dθ).
In the first case we have for the derivative
dw(z) d ln(ρ)
= e−ıθ
dz dρ
e−ıθ
=
ρ
1
= .
z
COMPLEX CALCULUS I: DIFFERENTIATION 107

Since this does not depend on n, it follows that the result is the same for all
the leaves of the Riemann surface. Taking the derivative in the other way we
have
dw(z) e−ıθ d(ıθ)
=
dz ıρ dθ
e−ıθ
=
ρ
1
= .
z
As expected, the result is the same, and again the dependence on n vanishes.
108 SOLUTIONS 7
Solutions 8

Complex Calculus II:


Integration

We present here complete and commented solutions to all problems proposed


in Chapter 8 of the text. For reference, the propositions of the problems are
repeated here. The problems are discussed in the order in which they were
proposed within the problem set of that chapter.

Problem 1. For each function w(z) below, make an analysis of the possible
singularities in order to determine whether or not one can use the Cauchy-
Goursat theorem in order to determine that, on the unit circle C,
I
w(z) dz = 0.
C

z2
(a) w(z) = .
z−3

(b) w(z) = z ez .

1
(c) w(z) = .
z 2 + 2z + 2

(d) w(z) = sech(z).

(e) w(z) = tan(z).

(f) w(z) = ln(z + 2).

109
110 SOLUTIONS 8

Complete Solution:

In each case, we list the singularities and determine the applicability or nor
of the Cauchy-Goursat theorem for the integral over the unit circle.

(a) Function:

z2
.
z−3

The function has only one singularity, a simple pole at z = 3. This


point is outside the unit circle, therefore the theorem applies and the
integral is zero.

(b) Function:

z ez .

The function has no singularities, therefore the theorem applies and


the integral is zero.

(c) Function:

1
.
z 2 + 2z + 2

We have for the roots of the polynomial in the denominator, by the


Baskara formula,

−2 ± 4−8
z± =
2
−2 ± 2ı
=
2
= −1 ± ı.

The numbers −1 ± ı have absolute value ρ = 2, and hence are outside
the unit circle, so that the theorem applies and the integral is zero.

(d) Function:

sech(z).
COMPLEX CALCULUS II: INTEGRATION 111

In this case we need to recall that


1
sech(z) = ,
cosh(z)
cosh(z) = cosh(x) cos(y) + ı sinh(x) sin(y).

Now we must determine where cosh(z) can be zero. For this, both the
real part and the imaginary part should vanish. Since cosh(x) is never
zero, from the real part we conclude that we must have cos(y) = 0.
Due to this, we know that sin(y) 6= 0, so that, from the imaginary part,
we should have sinh(x) = 0, which implies that x = 0. Thus, all zeros
are on the imaginary axis. Their positions along the axis are given
by cos(y) = 0, which implies that the zeros nearest to the origin are
y = π/2 and y = −π/2. Since π/2 > 1, these zeros are outside the unit
circle, so that the theorem applies and the integral is zero.

(e) Function:

tan(z).

In this case we need to recall that

sin(z)
tan(z) = ,
cos(z)
cos(z) = cos(x) cosh(y) − ı sin(x) sinh(y).

By means of an analysis similar to that used in the previous item,


we conclude that the zeros are only those of the function cos(x), with
y = 0. The two zeros closest to the origin are therefore (π/2, 0) and
(−π/2, 0). Again, since π/2 > 1, these zeros are outside the unit circle,
so that the theorem applies and the integral is zero.

(f) Function:

ln(z + 2).

The function has only one singularity, a branch point at z = −2, which
is outside the unit circle. It follows that the theorem applies and the
integral is zero. Note that the integral can be on any one of the infinite
Riemann leaves of the logarithm, so long as it is entirely on a single
leaf.
112 SOLUTIONS 8

Problem 2. Consider an arbitrary closed simple curve C on the (x, y)


plane, with the condition that it do not pass through the point z = 0. Show
that
dz
I
2
= 0,
C z
in the following cases.
(a) If the interior of the curve does not contain the origin z = 0.
(b) If the interior of the curve contains the origin z = 0.
Note: a closed simple curve is a closed curve that does not intersect itself.

Complete Solution:

We must calculate on simple closed curves the integral


dz
I
2
,
C z
whose integrand has only one singularity at z = 0. Of course, the curve
cannot pass exactly over the point z = 0, because in this case the integral
would not be well defined.
(a) If the curve does not contain the point z = 0, then the integrand
is analytic on the curve and within it, so that the Cauchy-Goursat
theorem applies and the integral is zero.
(b) If the curve contains the point z = 0, then we can use the Cauchy-
Goursat theorem in order to deform the curve until it coincides with
the unit circle, without this changing the value of the integral. Then
it suffices to calculate the value on the unit circle, using the polar
representation z = ρ exp(ıθ),
Z 2π
dz ρı eıθ
I
2
= dθ
C z 0 ρ2 e2ıθ
Z 2π
ı
= dθ e−ıθ
ρ 0
ı −ıθ 2π

= ıe
ρ 0
−1 −ı2π
− e0

= e
ρ
= 0,
COMPLEX CALCULUS II: INTEGRATION 113

due to the periodicity of the exponential with imaginary argument. It


follows therefore that the integral is zero in this case as well.

Problem 3. Consider an arbitrary closed simple curve C on the (x, y)


plane, with the condition that it do not pass through the point z = 0. Show
that
dz
I
3
= 0,
C z

in the following cases.

(a) If the interior of the curve does not contain the origin z = 0.

(b) If the interior of the curve contains the origin z = 0.

Note: a closed simple curve is a closed curve that does not intersect itself.

Complete Solution:

We must calculate on simple closed curves the integral


dz
I
3
,
C z

whose integrand has only one singularity at z = 0. Of course, the curve


cannot pass exactly over the point z = 0, because in this case the integral
would not be well defined.

(a) If the curve does not contain the point z = 0, then the integrand
is analytic on the curve and within it, so that the Cauchy-Goursat
theorem applies and the integral is zero.

(b) If the curve contains the point z = 0, then we can use the Cauchy-
Goursat theorem in order to deform the curve until it coincides with
the unit circle, without this changing the value of the integral. Then
it suffices to calculate the value on the unit circle, using the polar
representation z = ρ exp(ıθ),


dz ρı eıθ
I Z
= dθ
C z3 0 ρ3 e3ıθ
Z 2π
ı
= dθ e−2ıθ
ρ2 0
114 SOLUTIONS 8

ı ı −2ıθ 2π

= e
ρ2 2 0
−1 −ı4π
− e0

= 2
e

= 0,

due to the periodicity of the exponential with imaginary argument. It


follows therefore that the integral is zero in this case as well.

Problem 4. Consider an arbitrary closed simple curve C on the (x, y)


plane, with the condition that it do not pass through the point z = 1. Show
that
dz
I
2
= 0,
C (z − 1)

in the following cases.

(a) If the interior of the curve does not contain the point z = 1.

(b) If the interior of the curve contains the point z = 1.

Note: a closed simple curve is a closed curve that does not intersect itself.

Complete Solution:

We must calculate on simple closed curves the integral

dz
I
.
C (z − 1)2

If we make the transformation of variables z ′ = z − 1, then the condition on


the curve, which is whether or not it contains the point z = 1, becomes the
corresponding condition of whether or not it contains the point z ′ = 0, and
the integral is simplified to

dz ′
I
′2
,
C z

whose integrand has only one singularity at z ′ = 0. Of course, the curve


cannot pass exactly over the point z ′ = 0, because in this case the integral
would not be well defined.
COMPLEX CALCULUS II: INTEGRATION 115

(a) If the curve does not contain the point z ′ = 0, then the integrand
is analytic on the curve and within it, so that the Cauchy-Goursat
theorem applies and the integral is zero.

(b) If the curve contains the point z ′ = 0, then we can use the Cauchy-
Goursat theorem in order to deform the curve until it coincides with
the unit circle, without this changing the value of the integral. Then
it suffices to calculate the value on the unit circle, using the polar
representation z ′ = ρ exp(ıθ),


dz ′ ρı eıθ
I Z
= dθ
C z ′2 0 ρ2 e2ıθ
Z 2π
ı
= dθ e−ıθ
ρ 0
 2π
ı −ıθ
= ıe
ρ 0
−1 −ı2π
− e0

= e
ρ
= 0,

due to the periodicity of the exponential with imaginary argument. It


follows therefore that the integral is zero in this case as well.


Problem 5. Consider the analytic function w(z) = z.

(a) Calculate the integral


I
w(z) dz,
C

on the unit circle C.

(b) Calculate the integral of this function over a curve that goes twice
around the unit circle.

(c) Examine the situation for the integrals over other circular contours
centered at the origin, in order to determine whether it is possible to
say something about the values of these integrals based on the results
obtained here.
116 SOLUTIONS 8

Hint: use polar coordinates.

Complete Solution:

Using the polar representation z = ρ exp(ıθ), we have for the function



w(z) = z
√ ıθ/2
= ρe .

(a) Let us calculate on the unit circle ρ = 1, the integral,




I Z
w(z) dz = dθ ıρ eıθ ρ eıθ/2
C 0
Z 2π
= ı dθ e3ıθ/2
0
−2ı 3ıθ/2 2π

= ı e
3 0
2 3ıπ 0

= e −e
3
4
= − .
3
As one can see, the result is not zero. Note that when we go around the
circle we change from one leaf of the Riemann surface of the function

z to the other, so that this integral is not actually on a closed contour
within the domain of the function.
(b) Going around the unit circle twice, we have


I Z
w(z) dz = dθ ıρ eıθ ρ eıθ/2
C 0
Z 4π
= ı dθ e3ıθ/2
0
−2ı 3ıθ/2 4π

= ı e
3 0
2 6ıπ 0

= e −e
3
= 0.

We see that this time the result is zero. Note that when we go twice
around the circle we change twice the leaf of the Riemann surface of the
COMPLEX CALCULUS II: INTEGRATION 117


function z, thus returning to the original leaf, so that this time the
integral actually is over a closed contour, with regard to the domain of
the function.

(c) The change to circles with other radii simply includes a factor of ρ3
multiplying the integral, which does not introduce any divergence, and
in fact makes the integral vanish for ρ → 0. On the other hand, if we
make n turns on the unit circle, we obtain

n2π

I Z
w(z) dz = dθ ıρ eıθ ρ eıθ/2
C 0
Z n2π
= ı dθ e3ıθ/2
0
−2ı 3ıθ/2 n2π

= ı e
3 0
2 3nıπ 0

= e −e
3
2
= [(−1)n − 1] .
3

If n is even the result is zero, and if n is odd the result is −4/3. There-
fore, the result is zero every time we return to the same point on the
Riemann surface, exchanging leaves an even number of times, and thus
closing the contour.

Problem 6. Consider the analytic function w(z) = 1/ z.

(a) Calculate the integral


I
w(z) dz,
C

on the unit circle C.

(b) Calculate the integral of this function over a curve that goes twice
around the unit circle.

(c) Examine the situation for the integrals over other circular contours
centered at the origin, in order to determine whether it is possible to
say something about the values of these integrals based on the results
obtained here.
118 SOLUTIONS 8

Hint: use polar coordinates.

Complete Solution:

Using the polar representation z = ρ exp(ıθ), we have for the function


1
w(z) = √
z
1 −ıθ/2
= √ e .
ρ
(a) Let us calculate on the unit circle ρ = 1, the integral,

1
I Z
w(z) dz = dθ ıρ eıθ √ e−ıθ/2
C 0 ρ
Z 2π
= ı dθ eıθ/2
0
 2π
= ı(−2ı) eıθ/2

 0
= 2 eıπ − e0
= −4.

As one can see, the result is not zero. Note that when we go around the
circle we change from one leaf of the Riemann surface of the function

1/ z to the other, so that this integral is not actually on a closed
contour with regard to the domain of the function.
(b) Going twice around the unit circle, we have

1
I Z
w(z) dz = dθ ıρ eıθ √ e−ıθ/2
C 0 ρ
Z 4π
= ı dθ eıθ/2
0
 4π
= ı(−2ı) eıθ/2

0
= 2 e2ıπ − e0
= 0.

We see that this time the result is zero. Note that when we go twice
around the circle we change twice the leaf of the Riemann surface of the
COMPLEX CALCULUS II: INTEGRATION 119


function 1/ z, thus returning to the original leaf, so that this time the
integral actually is over a closed contour, with regard to the domain of
the function.

(c) The change to circles with other radii simply includes a factor of ρ
multiplying the integral, which does not introduce any divergence, and
in fact the integral vanishes for ρ → 0. On the other hand, if we make
n turns on the unit circle, we obtain

n2π
1
I Z
w(z) dz = dθ ıρ eıθ √ e−ıθ/2
C 0 ρ
Z n2π
= ı dθ eıθ/2
0
 n2π
= ı(−2ı) eıθ/2
nıπ 0
0
= 2 e −e
= 2 [(−1)n − 1] .

If n is even the result is zero, and if n is odd the result is −4. So the
result is zero every time we return to the same point on the Riemann
surface, exchanging leaves an even number of times, and thus closing
the contour.

Problem 7. Consider the analytic function w(z) = ln(z).

(a) Calculate the integrals


I
w(z) dz,
C

on the unit circle C.

(b) Calculate the integral of this function over a curve that goes twice
around the unit circle.

(c) Examine the situation for the integrals over other circular contours
centered at the origin, in order to determine whether it is possible to
say something about the values of these integrals based on the results
obtained here.
120 SOLUTIONS 8

Hint: use polar coordinates.

Complete Solution:

Using the polar representation z = ρ exp(ıθ), we have for the function

w(z) = ln(z)
= ln(ρ) + ıθ,

where we indicate that we are initiating the integral on the central leaf of the
Riemann surface, with n = 0.

(a) Let us calculate on the unit circle ρ = 1, the integral,

I Z 2π
w(z) dz = dθ ıρ eıθ [ln(ρ) + ıθ]
C 0
Z 2π
2
= ı dθ θ eıθ .
0

The integral can be made by parts, integrating the exponential and


differentiating the factor θ. Doing this we obtain

I  2π Z 2π
w(z) dz = −(−ı)θ e ıθ
+ dθ (−ı) eıθ
C 0 0
 2π Z 2π
= ıθ eıθ −ı dθ eıθ
0 0
 2π
= ı2π e ı2π
− ı(−ı) e ıθ
0
= ı2π − eı2π + e0
= ı2π.

As one can see, the result is not zero. Note that when we go around the
circle we changed from a leaf of the Riemann surface of the function
ln(z) to the next, so that this integral is not actually on a closed contour
with regard to the domain of the function.

(b) Going twice around the unit circle, we have


COMPLEX CALCULUS II: INTEGRATION 121
I Z 4π
w(z) dz = dθ ıρ eıθ [ln(ρ) + ıθ]
C 0
Z 4π
2
= ı dθ θ eıθ .
0

Integrating by parts we have


I  4π Z 4π
w(z) dz = −(−ı)θ e ıθ
+ dθ (−ı) eıθ
C 0 0
 4π Z 4π
= ıθ eıθ −ı dθ eıθ
0 0
 4π
= ı4π e ı4π
− ı(−ı) e ıθ
0
= ı4π − eı4π + e0
= ı4π.

Again the result is not zero. Note that when we go twice around the
circle we change twice from one leaf of the Riemann surface of the
function ln(z) to the next, and therefore the contour again is not closed
with regard to the domain of the function.

(c) The modification to circles with other radii adds to the previous inte-
gral, for the case of the integration on the circle with a single turn, the
integral of the real part of the function,
Z 2π Z 2π
dθ ıρ e ln(ρ) = ıρ ln(ρ)
ıθ
dθ eıθ
0 0
 2π
= ıρ ln(ρ)(−ı) e ıθ
0
0

= ρ ln(ρ) e − e
ı2π

= 0.

Thus we see that the real part of the original integral never contributes
to the final result. Clearly, this also applies to integrals that make any
number of turns around the origin. Note that the factor in front of the
integral is not divergent, since the limit of ρ ln(ρ) when ρ → 0 is in
fact zero. On the other hand, if we make n turns on the unit circle, we
obtain
122 SOLUTIONS 8
I Z n2π
w(z) dz = dθ ıρ eıθ [ln(ρ) + ıθ]
C 0
Z n2π
2
= ı dθ θ eıθ .
0

Integrating by parts we have

I  n2π Z n2π
w(z) dz = −(−ı)θ e ıθ
+ dθ (−ı) eıθ
C 0 0
 n2π Z n2π
= ıθ eıθ −ı dθ eıθ
0 0
 n2π
= ın2π e ın2π
− ı(−ı) e ıθ
0
0
= ın2π − e ın2π
+e
= ın2π.

We see that the integral increases indefinitely to ı∞ if we make more


and more turns around the origin in the positive direction. If we make
an infinite number of turns in the negative direction, then the integral
goes to −ı∞. This time there is no way to return to the same leaf of
the Riemann surface, except if the total number of turns is zero, that
is, if n = 0.

Problem 8. Consider the proof of Green’s theorem in two dimensions that


was presented in the text. Consider therefore an infinitesimal rectangle of
dimensions dx and dy, a vector field (u, v) with Cartesian components u(x, y)
and v(x, y), the line integral of (u dx + v dy) over the perimeter of the rect-
angle, and the surface integral
 
∂v ∂u
− dx dy
∂x ∂y

over the area of the rectangle. In the case of the line integral, consider that
each value of u and v is associated with a link, like those which form the
perimeter of the rectangle, and represent these values as functions of the
midpoint of each link. For example, in the first part of the integral we have
u(x + dx/2, y), and so on.
COMPLEX CALCULUS II: INTEGRATION 123

(x, y + dy) (x + dx, y + dy)

dy

(x, y) (x + dx, y)
dx

0 x

Figure 8.1: A rectangular plaquette with sites and links.

(a) Show the central fact of the theorem, that is, that in this infinitesimal
rectangle, with the perimeter positively oriented,

(4)  
X ∂v ∂u
(u dx + v dy) = − dx dy.
∂x ∂y
links

(b) Determine to which point one should associate each partial derivative
in the most natural and symmetric way possible, and show that the
expression
 
∂v ∂u

∂x ∂y

is naturally associated with the center of the rectangle, which is also


called, in some circumstances, a plaquette.

Complete Solution:

We are to calculate the line and surface integration elements, involving the
vector w
~ = (u, x) and the position ~z = (x, y), on the rectangle shown in
Figure 8.1, made with four links of a Cartesian grid, or lattice.

(a) In order to build the line integral element on the perimeter of the rect-
angle, we associate the vector components to the links, whose positions
124 SOLUTIONS 8

are represented by the midpoints of the links, as shown with crosses in


Figure 8.1, so that we have, using the positive direction of the perime-
ter,

(4) (4)
X X
w
~ · d~z = (u dx + v dy)
links links
= u (x + dx/2, y) dx + v (x + dx, y + dy/2) dy +
−u (x + dx/2, y + dy) dx − v (x, y + dy/2) dy
= dy [v (x + dx, y + dy/2) − v (x, y + dy/2)] +
−dx [u (x + dx/2, y + dy) − u (x + dx/2, y)]
v (x + dx, y + dy/2) − v (x, y + dy/2)
= dy dx +
dx
u (x + dx/2, y + dy) − u (x + dx/2, y)
−dx dy .
dy

We now recognize in these two ratios, in the continuum limit, in which


dx and dy both go to zero, the crossed partial derivatives of u and
v, that we also associate with the midpoint of each link on which a
derivative is taken,

v (x + dx, y + dy/2) − v (x, y + dy/2)


lim
dx→0 dx
∂v
= (x + dx/2, y + dy/2) ,
∂x
u (x + dx/2, y + dy) − u (x + dx/2, y)
lim
dy→0 dy
∂u
= (x + dx/2, y + dy/2) .
∂y

As one can see, these partial derivatives end up associated to the center
of the plaquette. We therefore have that our integration line element
can be written in terms of a surface integration element, which estab-
lishes the central fact of the theorem,

(4)
X
(u dx + v dy)
links
v (x + dx, y + dy/2) − v (x, y + dy/2)
= dy dx +
dx
COMPLEX CALCULUS II: INTEGRATION 125

u (x + dx/2, y + dy) − u (x + dx/2, y)


−dx dy
dy
 
∂v ∂u
= dx dy (x + dx/2, y + dy/2) − (x + dx/2, y + dy/2) .
∂y ∂y

(b) As shown above, if we proceed as symmetrically as possible, the partial


derivatives are both associated to the point (x + dx/2, y + dy/2), which
is the center of the rectangle, or plaquette, as shown in the diagram of
Figure 8.1. It follows that, if each vector component associated with
a link is positioned at the midpoint of that link, and each derivative
associated with a link is positioned at the midpoint of that link, then
the surface integral element is associated to the central point of the
plaquette,

(4)  
X ∂v ∂u
(u dx + v dy) = dx dy − (x + dx/2, y + dy/2) .
∂y ∂y
links

Problem 9. Consider the integral of an analytic function given by


Z B
I= z n dz,
A

for n a non-negative integer, between the points A = (0, 0) and B = (1, 1).

(a) Calculate the integral along the path formed by a straight segment
which links the point A directly to the point B.

(b) Calculate the integral along the path formed by a straight segment from
A to (1, 0) and another straight segment from (1, 0) to B.

(c) Calculate the integral along the path formed by a straight segment from
A to (0, 1) and another straight segment from (0, 1) to B.

Complete Solution:

In each case, the integrals in each segment can be parametrized in terms of


either x or of y, where z = x + ıy.

(a) On the path formed by a straight segment connecting the points A and
B we have that x = y and therefore that z = x + ıx and dz = (1 + ı)dx,
so that the integral I is given by
126 SOLUTIONS 8
Z 1
I = (1 + ı)dx [(i + ı)x]n
0
Z 1
n+1
= (1 + ı) dx xn
0
n+1  1
n+1 x
= (1 + ı)
n+1 0
(1 + ı)n+1
= .
n+1

(b) On the path formed by a straight segment from A to (1, 0) and another
straight segment from (1, 0) to B, we have that on the first segment
z = x and dz = dx, and that on the second segment z = 1 + ıy and
dz = ıdy, so that the integral I is given by

Z 1 Z 1
I = dx xn + ıdy (1 + ıy)n
0 0
xn+1 1 (1 + ıy)n+1 1
 
= +
n+1 0 n+1 0
1 (1 + ı)n+1 1
= + −
n+1 n+1 n+1
(1 + ı)n+1
= .
n+1

(c) On the path formed by a straight segment from A to (0, 1) and another
straight segment from (0, 1) to B, we have that on the first segment
z = ıy and dz = ıdy, and that on the second segment z = x + ı and
dz = dx, so that the integral I is given by

Z 1 Z 1
n
I = ıdy (ıy) + dx (x + ı)n
0 0
(ıy)n+1 1 (x + ı)n+1 1
 
= +
n+1 0 n+1 0
ın+1 (1 + ı)n+1 ın+1
= + −
n+1 n+1 n+1
(1 + ı)n+1
= .
n+1
COMPLEX CALCULUS II: INTEGRATION 127

Problem 10. Consider the integral of a complex function, which is not


analytic, given by
Z B
I= z ∗ dz,
A

between the points A = (−1, 0) and B = (1, 0).

(a) Calculate the integral along the path formed by a straight segment
which links the point A directly to the point B.

(b) Calculate the integral along the arc of the unit circle on the upper
half-plane connecting the points A and B.

(c) Calculate the integral along the arc of the unit circle on the lower half-
plane connecting the points A and B.

Complete Solution:

In each case, the integrals in each segment or arc can be parametrized in terms
of either x or of θ, where we have that either z = x + ıy or z = ρ exp(ıθ).

(a) On the path formed by a straight segment connecting the points A and
B we have that z = x and therefore that dz = dx and that z ∗ = x, so
that the integral I is given by
Z 1
I = dx x
−1
x2 1

=
2 −1
= 0.

(b) On the path formed by the arc of the unit circle that connects the
points A and B on the upper half-plane we have that z = exp(ıθ) and
therefore that dz = ı exp(ıθ)dθ and that z ∗ = exp(−ıθ), so that the
integral I is given by
Z 0
I = ı eıθ dθ e−ıθ
π
Z 0
= ı dθ
π
128 SOLUTIONS 8

0
= ıθ
π
= −ıπ.

(c) On the path formed by the arc of the unit circle that connects the
points A and B on the lower half-plane we have that z = exp(ıθ) and
therefore that dz = ı exp(ıθ)dθ and that z ∗ = exp(−ıθ), so that the
integral I is given by
Z 0
I = ı eıθ dθ e−ıθ
−π
Z 0
= ı dθ
−π
0
= ıθ
−π
= ıπ.

Problem 11. Consider the unit-side square and its diagonal going from
the point (0, 0) to the point (1, 1) in a Cartesian coordinate system (x, y).
Consider approximating this diagonal by a “staircase” that consists of hori-
zontal and vertical segments, all the steps being equal, going from the point
(0, 0) to the point (1, 1). Consider the limit in which the size of the steps
goes to zero.

(a) What is the limit of the distance between any point on the “staircase”
and the diagonal of the square, when the size of the steps goes to
zero? The “staircase” can be considered as a good representation of
the diagonal in this limit, in terms of the points of the plane that make
up each object?

(b) What is the limit for the total length of the “staircase”, when the size
of the steps goes to zero? The “staircase” can be considered as a good
representation of the diagonal in this limit, if we now think in terms of
the measure associated with each object?

Complete Solution:

If we have N steps, we will have 2N segments of the same length, half of


them horizontal and the other half vertical. Each segment will have an end
COMPLEX CALCULUS II: INTEGRATION 129

y
1

0 1/N 1 x

Figure 8.2: The “staircase” as an approximation of the diagonal of the square.

located on the diagonal, and the other end away from it. One can draw the
steps either above or beneath the diagonal, without this changing anything
in the answers of the problem. The length of each segment is 1/N , as can be
easily seen by drawing the elements involved, as shown in Figure 8.2.

(a) For each segment, the point farthest from the diagonal is the end that
is away from it. This distance is one half of the diagonal
√ of the small
square defined by the steps, and it is therefore 1/( 2N ). Since this
distance goes to zero as N → ∞ and all other points are at distances
shorter than this from the diagonal, it follows that the limit of the
distance between an arbitrary point on the “staircase” and the diagonal
of the square, when the size of the steps goes to zero, is zero.
It follows therefore that all the points of the plane constituting the
“staircase” converges to the set of points of the plane forming the di-
agonal of the square, and the two sets become identical in the limit. In
this geometrical sense, in which we did not concern ourselves with mea-
surements, we can therefore say that the “staircase” can be considered
as a good representation of the diagonal in the N → ∞ limit.

(b) Since each segment has length 1/N and there are 2N of them, the total
length of all the chained segments is 2, which is equal to the sum of two
sides of the square and that does not depend on N . It follows that the
limit of this length when N → ∞ is equal to 2, and therefore it is not
equal
√ to the length of the diagonal of the square, which has the value
2.
130 SOLUTIONS 8

The value 2 is the linear measure associated with the √ set of points in
the plane constituting the “staircase”, while the value 2 is the linear
measure associated with the set of points of the plane that constitutes
the diagonal of the square. Although the two sets of points become
identical in the limit, the two measures do not coincide, so that in terms
of the measure associated with each object the “staircase” cannot be
considered a good representation of the diagonal in the N → ∞ limit.
Extending the concepts that are at stake here to the case of curves in
the plane, what we learn from this problem is that, in order for us to be
able to properly represent the measure (that is, the length) of a simple
closed curve in the plane, we must use a polygon whose vertices are all
on the curve. Only in this way the total length of the polygon will have
as its limit the length of the curve, in the limit where the size of the
segments of the polygon goes to zero.

Problem 12. (Challenge Problem) Consider a closed simple curve C in


the plane, the interior of which is the surface S. For simplicity, assume that
the curve is differentiable. A certain maximum size is given, represented by
the strictly positive real number ǫ, which can be chosen as small as needed.

(a) Construct a set of rectangles and triangles, all with vertical and hor-
izontal dimensions smaller than ǫ, which is almost entirely contained
within the interior of the curve, in the sense that its boundary is a
polygon with all its vertices located on the curve.

(b) Given a continuous and differentiable vector field w


~ = (u, v), construct
a discrete representation of the surface integral
Z
~ ×w
∇ ~ da,
S

using this set of rectangles and triangles. Show that in the limit ǫ → 0
this discrete representation approaches the integral.

(c) Given the same vector field w,


~ construct a discrete representation of
the line integral
I
w ~
~ · dℓ,
C
COMPLEX CALCULUS II: INTEGRATION 131

Figure 8.3: A regular discretization of the surface S.

using the polygon which is the external boundary of the set of rectangles
and triangles. Show that in the limit ǫ → 0 this discrete representation
approaches the integral.

(d) Use these constructions to elaborate a more complete proof of Green’s


theorem, that is, in our vector language, of the result
I Z
w ~ =
~ · dℓ ~ ×w
∇ ~ da.
C S

Consider in detail the cancellations of integration elements, in order to


transform the surface integral into a line integral.

Complete Solution:

(a) Cover the curve C and its interior S with a regular lattice formed
of squares with sides smaller than ǫ, as illustrated in the diagram of
Figure 8.3.
We will now elaborate an algorithm for the construction of a lattice,
that is, of a set of adjacent triangles and rectangles, which is contained
within the surface S and whose boundary is a polygon with all of its
vertices on the curve C. For each square, there are three disjoint pos-
sibilities: either it is completely outside the curve, or it is completely
inside the curve, or it is cut by the curve. If the curve passes only
132 SOLUTIONS 8

111
000
000
111 111
000
0001
1110
0
1 11
00
0011
1100
00
11
000
111 0001
1110 0011
1100
000
111 0001
1110 0011
1100
000
111 0001
1110 0011
1100
000
111 000
111
0001
1110
0
1 0011
1100
0011
1100

Figure 8.4: Characterization of the possibilities in the discretization of the


curve C.

through a single point of the square, we consider that it is completely


inside or completely outside the curve, as appropriate. We do the same
if the curve touches the square only at two vertices of one side of the
square. In all other cases, since the curve is differentiable, we can
choose ǫ sufficiently small, so that the curve always cuts the square
exactly twice, which can occur only in one of the three distinct forms
illustrated in the diagram of Figure 8.4.
If a square is completely inside of the curve, we included it in our set
of rectangles and triangles. If it is completely outside, we exclude it.
If it is cut twice by the curve, then we interconnect the two points of
intersection using a straight segment. This straight segment divides the
square into two parts, one part being essentially all outside the curve
and the other essentially all inside it, except for a small region contained
between the segment and the curve. We can choose the region to keep
in our set as follows: if one of the two regions is entirely outside of the
curve, then we exclude it and keep the other region; if one of the two
regions is entirely within the curve, then we keep it and exclude the
other region.
Our set now consists of squares and those regions that were retained,
some of which already have a triangular shape. Those which are already
triangles are included in our set without further modifications. The
others can be divided by means of horizontal or vertical segments, in
two or three disjoint parts, which consist of a triangle and one or two
rectangles. Having made this division, we include all the other elements
in our set, which now consists of squares (that is, a particular case of
rectangle), rectangles and triangles. By construction, all these elements
have dimensions smaller than ǫ, and its boundary is a polygon with all
vertices on the curve, as requested.
COMPLEX CALCULUS II: INTEGRATION 133

(b) Given a continuous and differentiable vector field w~ = (u, v), we can
build on each square or rectangle of our set the quantity
 
δv(x, y) δu(x, y)
δIS = δa − ,
δx δy

where δa = δxδy is the area element associated with the rectangle. On


the triangles is not necessary to calculate in detail the contributions to
the surface integral because, as noted in the text, the total area of these
triangles tends to zero in the integration limit, because it fits within a
strip with infinitesimal width along the perimeter of S, whose area goes
therefore to zero. Since the surface integral is two-dimensional, the total
contribution of this sub-domain of measure zero is zero. Adding the
contributions of all rectangles of our set we therefore have a Riemann
partition of the integral over the area S,
 
X δv(x, y) δu(x, y)
IS = δa − .
δx δy
latt

In the limit ǫ → 0 the dimensions δx and δy of all rectangles go to


zero, and the union of all of them covers completely the domain of
the integral, while the ratios of variations in the above formula tend
to the corresponding partial derivatives, since u(x, y) and v(x, y) are
differentiable, so that we have that
 
∂v(x, y) ∂u(x, y)
Z
lim IS = da −
ǫ→0 ∂x ∂y
ZS
= ~ ×w
∇ ~ da.
S

(c) Given the same vector field w,~ we can build on the boundary of each
square or rectangle of our set the quantity

δIC = u(x, y) δx + v(x + δx, y) δy − u(x, y + δy) δx − v(x, y) δy,

which is a sum over all four sides of the rectangle where δx and δy are
the length elements of the sides of the rectangle. As for the triangles,
each one has a horizontal side, a vertical side and a sloping side. By
construction, all the horizontal or vertical sides are strictly within the
134 SOLUTIONS 8

(x, y) (x, y) (x, y) (x, y)

Figure 8.5: Classification of the triangles on the boundary of S.

surface S, while all the sloping sides are part of the boundary of our
lattice. The triangles may be oriented in one of four ways that are
shown in the diagram of Figure 8.5.
Using as an example a triangle with the first orientation, we can write
on it the quantity

δIC = u(x, y) δx − v(x, y) δy + w ~


~ · δℓ,

~ is the displacement associated with the sloping side, with the


where δℓ
proper positive orientation for the triangle. A quantity of this type
can be written for every triangle, varying only the application point of
u(x, y) and v(x, y) and the orientation of each segment. For example,
for the second type of orientation we have

δIC = u(x, y) δx + v(x + δx, y) δy + w ~


~ · δℓ,

and so on. In all cases, the sloping sides, which are associated with the
terms w ~ are part of the outer boundary of the lattice, while the
~ · δℓ,
other two sides are internal. It turns out that, as noted in the text, all
internal sides are traversed twice, once in each direction, so that the
two identical contribution to the line integral have opposite signs, and
therefore all cancel off. There remains therefore the outer boundary
of the lattice, which is a polygon consisting of the sloping sides of the
triangles and occasionally some side of a square, which happens to be
located on a vertical or horizontal part of the curve. It follows that we
have a discrete representation of the line integral, which is given by

~
X
IC = w
~ · δℓ,
poly

which is a sum over a polygon with all vertices on the curve. In the
limit ǫ → 0 the dimensions of all the squares and triangles go to zero,
COMPLEX CALCULUS II: INTEGRATION 135

and therefore the sides of this polygon go to zero, so that the polygon
approaches the curve with the correct measure, and hence we have that

~
X
lim IC = lim w
~ · δℓ
ǫ→0 ǫ→0
poly
I
= w ~
~ · dℓ.
C

(d) We started by showing that, in the limit ǫ → 0, the quantities δIC and
δIS on one of the rectangles of our lattice are equal, just as was done
in the text. We started by writing δIC in the following way,

δIC = [v(x + δx, y) − v(x, y)] δy − [u(x, y + δy) − u(x, y)] δx.

Since u(x, y) and v(x, y) are differentiable, these variations can be writ-
ten in the limit ǫ → 0, in terms of the appropriate partial derivatives,

δv(x, y) δu(x, y)
δIC = δxδy − δyδx
δx δy
 
δv(x, y) δu(x, y)
= − δa,
δx δy

where δa = δxδy. This establishes that δIC = δIS in each of the


rectangles of the lattice. If we add these quantities over all rectangles,
we will have a corresponding equality for the two sums.
X X
δIC = δIS .
rect rect

It only remains for us to consider the triangles of our lattice. Since


from the point of view of the surface integral they are a zero-measure
subset, we can add the contributions of the triangles to the sum on the
right-hand side, without changing its value in the limit. On the other
hand, if these same contributions are added to the sum on the left-hand
side, then it has as its limit the line integral. It follows that we can
write that the equality holds in the limit for the sums over the entire
lattice
X X
δIC = δIS ,
latt latt
136 SOLUTIONS 8

and therefore, taking the integration limit, we have Green’s theorem,


that is, in our vector language,
I Z
w ~ =
~ · dℓ ~ ×w
∇ ~ da.
C S
Solutions 9

Complex Derivatives and


Integrals

We present here complete and commented solutions to all problems proposed


in Chapter 9 of the text. For reference, the propositions of the problems are
repeated here. The problems are discussed in the order in which they were
proposed within the problem set of that chapter.

Problem 1. Calculate the integral

zn
I
dz,
C (z − z0 )2

for a non-negative integer n, on an arbitrary closed curve C that makes a


single turn around the point z0 . Determine for which combinations of values
of n and z0 the result is zero.

Complete Solution:

Let us recall that one of the Cauchy integral formulas, that for the first
derivative, tells us that

1 f (z)
I
f ′ (z0 ) = dz ⇒
2πı C (z − z0 )2
f (z)
I
2
dz = 2πıf ′ (z0 ),
C (z − z0 )

so long as the curve C contain z0 , so that we can use this relation in order
to calculate the integral

137
138 SOLUTIONS 9

zn
I
dz,
C (z − z0 )2

by simply making f (z) = z n . We have therefore

zn
I
(n−1)
2
dz = 2πınz0 .
C (z − z0 )

This vanishes if n = 0, for any value of z0 , or if z0 = 0 with n > 1. In the


special case in which we have z0 = 0 with n = 1, the value is 2πı, because in
this case we have the integral around a simple pole,

1
I
dz = 2πı.
C z

Problem 2. Calculate the integral



z
I
n
dz,
C (z − z0 )

for a non-negative integer n, on an arbitrary closed curve C that makes a


single turn around the point z0 and does not contain the origin. Calculate
the z0 → 0 limit of the result obtained. How can we interpret the result in
this limit?
Hint: use the Cauchy integral formulas.

Complete Solution:

Let us recall the Cauchy integral formulas, which give the derivatives of an
analytic function in terms of integrals,

n! f (z)
I
n′
f (z0 ) = dz ⇒
2πı C (z − z0 )n+1
f (z) 2πı n′
I
n+1
dz = f (z0 ),
C (z − z0 ) n!

so long as the curve C contain z0 , where n ≥ 0. It follows that we can use


this relation in order to calculate the integral

z
I
n
dz,
C (z − z0)
COMPLEX DERIVATIVES AND INTEGRALS 139

where n ≥ 0 and the contour does not contain the origin, which is the branch

point of the square root, by simply adopting f (z) = z. Note that for n = 0
the integrand is analytic in C, so that in this case the integral is zero on the
contour described, by the Cauchy-Goursat theorem. Hence, we can limit the
values of n in the following discussion to n ≥ 1. We have therefore
√ √
z 2πı dn−1 z
I
n
dz = (z0 ).
C (z − z0 ) (n − 1)! dz
In the particular case n = 1 we have the function itself instead of a derivative,
√ √
and hence the result is simply 2πı z0 . Of course, since the function z has
a Riemann surface with two leaves, and since the contour is all on the same
leaf due to the fact that the branch point is outside the contour, we have
in fact two possible answers with opposite signs, one on each one of the two
leaves. The same is true for the other values of n. What remains to be done
here is to calculate all the derivatives and try to write a general form for
them. Making a table of the first few cases we have
√ 0′ √
z = z,
√ 1′ 1 1
z = √ ,
2 z
√ 2′ 1 (−1) 1
z = √ 3,
2 2 z
√ 3′ 1 (−1) (−3) 1
z = √ 5,
2 2 2 z
√ 4′ 1 (−1) (−3) (−5) 1
z = √ 7,
2 2 2 2 z
√ 5′ 1 (−1) (−3) (−5) (−7) 1
z = √ 9,
2 2 2 2 2 z
√ 6′ 1 (−1) (−3) (−5) (−7) (−9) 1
z = √ 11 ,
2 2 2 2 2 2 z
... = ... .
Collecting the factors of (−1) in the numerator and of 2 in the denominator,
separating a square root and a factor of (−1), and noting that there appear
in the numerator the double factorials of odd numbers, we can write this
in the following way, which displays clearly the regularity of almost all the
factors in these results,
√ 0′ √ (−1)0 (−1) 1
z = − z ,
20 z0
140 SOLUTIONS 9

√ 1′ √ (−1)1 1
z = − z ,
21 z 1
√ 2′ √ (−1)2 1!! 1
z = − z ,
22 z2
√ 3′ √ (−1)3 3!! 1
z = − z ,
23 z3
√ 4′ √ (−1)4 5!! 1
z = − z ,
24 z4
√ 5′ √ (−1)5 7!! 1
z = − z ,
25 z5
√ 6′ √ (−1)6 9!! 1
z = − z ,
26 z6
... = ... .
In order to facilitate the manipulation of the double factorials, we add ad-
ditional factors so that they appear in all the cases, starting with 1!! = 1 in
the first line, thus getting
√ 0′ √ (−1)0 1!! 0
z = − z 0 z ,
2 (1)(−1)
√ 1′ √ (−1)1 3!! −1
z = − z 1 z ,
2 (3)(1)
√ 2′ √ (−1)2 5!! −2
z = − z 2 z ,
2 (5)(3)
√ 3′ √ (−1)3 7!! −3
z = − z 3 z ,
2 (7)(5)
√ 4′ √ (−1)4 9!! −4
z = − z 4 z ,
2 (9)(7)
√ 5′ √ (−1)5 11!! −5
z = − z 5 z ,
2 (11)(9)
√ 6′ √ (−1)6 13!! −6
z = − z 6 z ,
2 (13)(11)
... = ... .
We now see that, in the case of the derivative of order k, the double factorial
which appears is (2k + 1)!!, and that the two additional factors in the denom-
inator are (2k + 1) and (2k − 1), which are never zero and whose product is
2

(2k) − 1 . Therefore, we can already write the general case, for k ≥ 0,
√ k′ √ (−1)k (2k + 1)!! −k
z = − z k z
2 [(2k)2 − 1]
COMPLEX DERIVATIVES AND INTEGRALS 141

√ (−1)(k+1) (2k + 1)!! 1


= z .
2k [(2k)2 − 1] zk

The square root that appears multiplying these results clearly shows that they
all have a Riemann surface with two leaves. We can write the result solely in
terms of factorials instead of double factorials, multiplying and dividing by
(2k)!!, in order to obtain

√ k′ √ (−1)(k+1) (2k + 1)!!(2k)!! 1


z = z
2k (2k)!! [(2k)2 − 1] zk
√ (−1)(k+1) (2k + 1)! 1
= z (2k) .
2 k! [(2k)2 − 1] z k

Applying this at z0 , for k = n − 1, with n ≥ 1, we have



z 2πı √ (−1)n (2n − 1)! 1
I
n
dz = z 0 (2n−2) 2 (n−1)
C (z − z0 ) (n − 1)! 2 (n − 1)! [(2n − 2) − 1] z
0
√ πı(−1)n n2 (2n − 1)! 1
= z0 (2n−3)
2 (n!)2 (2n − 1)(2n − 3) z (n−1)
0
√ πı(−1)n n(2n)! 1
= z0 (2n−2) 2 (n−1)
.
2 (n!) (2n − 1)(2n − 3) z
0

Note that, since the factors in denominator never vanish, this result is cer-
tainly well defined for n ≥ 1. If we make z0 → 0, the contour C, which must
contain z0 but cannot contain the origin, must pass through an increasingly
narrow region between 0 and z0 . This means that in this limit the contour is
forced to approach indefinitely the two singularities, the pole at z0 and the
branch point at 0. As a consequence, it is reasonable to expect divergences
in these results, in this kind of limit. However, the case n = 1 is different
from the others, because in this case we have that

z √
I
lim dz = lim 2πı z0
z0 →0 C z − z0 z0 →0
= 0,

which happens due to the fact that, although the square root has a point
of singularity at the origin, it is not a pole but a branch point at which the
function itself exists and is zero, although it is not analytic. For all other
values of n the derivatives of the square root are involved, which are in fact
divergent at the origin, so that in this case we have
142 SOLUTIONS 9

z
I
lim dz
z0 →0 C (z − z0 )n
√ πı(−1)n n(2n)! 1
= lim z0 (2n−2) 2 (n−1)
z0 →0 2 (n!) (2n − 1)(2n − 3) z
0
πn(2n)! 1
= ı(−1)n (2n−2) 2
lim (n−1/2) ,
2 (n!) (2n − 1)(2n − 3) z0 →0 z
0

where for n > 1 the numerical fraction displayed is a positive real number.
Writing z0 in polar form, in terms of ρ0 and θ0 , we have that

z
I
lim dz
z0 →0 C (z − z0 )n

πn(2n)! e−ı(n−1/2)θ
= ı(−1)n lim
2(2n−2) (n!)2 (2n − 1)(2n − 3) z0 →0 ρ(n−1/2)
0
n −ı(n−1/2)θ πn(2n)! 1
= ı(−1) e (2n−2) 2
lim (n−1/2) ,
2 (n!) (2n − 1)(2n − 3) 0 ρ
ρ →0
0

where the remaining limit goes to real positive infinity, and we are assuming,
for simplicity, that the limit is taken in the complex plane with θ constant,
that is, along straight lines through the origin. The result is therefore infinite
in magnitude, with a certain phase, which depends on the angle θ,

z
I
lim dz = ı(−1)n eı(−n+1/2)θ ∞
z0 →0 C (z − z0 )n

= eıπ/2 e−ıπn eı(−n+1/2)θ ∞


= eı(−n+1/2)(π+θ) ∞,

where we represented by ∞ the product of the divergent limit and the positive
real quantity multiplying it. Note that the limit depends on θ, and if it
changes during the limit the limit can, for example, spiral out to infinity,
instead of going along straight lines.

Problem 3. Calculate the integral

sin(z)
I
2
dz,
C (z − z0 )

on an arbitrary closed curve C that makes a single turn around the point z0 .
Determine for which values of z0 the result is zero.
COMPLEX DERIVATIVES AND INTEGRALS 143

Hint: use the Cauchy integral formulas.

Complete Solution:

Let us recall that one of the Cauchy integral formulas, that for the first
derivative, tells us that

1 f (z)
I

f (z0 ) = dz ⇒
2πı C (z − z0 )2
f (z)
I
2
dz = 2πıf ′ (z0 ),
C (z − z0 )

so long as the curve C contain z0 , so that we can use this relation in order
to calculate the integral

sin(z)
I
dz,
C (z − z0 )2

by simply making f (z) = sin(z). We have therefore

sin(z)
I
dz = 2πı cos(z0 ).
C (z − z0 )2

This vanishes where cos(z0 ) vanishes, and as we have seen before, the only
zeros of cos(z0 ) are the real zeros that we already know quite well. Thus, the
result of the integral is zero for
π
z0 = + nπ,
2
where n is an arbitrary integer.

Problem 4. Calculate the integral

sin(z)
I
3
dz,
C (z − z0 )

on an arbitrary closed curve C that makes a single turn around the point z0 .
Determine for which values for z0 the result is zero.
Answer: −ıπ sin(z0 ).
Hint: use the Cauchy integral formulas.

Complete Solution:
144 SOLUTIONS 9

Let us recall that one of the Cauchy integral formulas, that for the second
derivative, tells us that

2 f (z)
I
′′
f (z0 ) = dz ⇒
2πı C (z − z0 )3
f (z)
I
3
dz = πıf ′′ (z0 ),
C (z − z0 )

so long as the curve C contain z0 , so that we can use this relation in order
to calculate the integral

sin(z)
I
dz,
C (z − z0 )3

by simply making f (z) = sin(z). We have therefore

sin(z)
I
3
dz = −πı sin(z0 ).
C (z − z0 )

This vanishes where sin(z0 ) vanishes, and as we have seen before, the only
zeros of sin(z0 ) are the real zeros that we already know quite well. Thus, the
result of the integral is zero for

z0 = nπ,

where n is an arbitrary integer.

Problem 5. Prove by finite induction the Newton binomial formula,


n
n
X n!
(a + b) = ak bn−k ,
k!(n − k)!
k=0

where a and b are any complex numbers. Make sure that the proof holds for
complex numbers.

Complete Solution:

We start by showing that the Newton binomial formula is valid for the initial
case n = 1, that is trivial. Putting n = 1 in
n
X n!
(a + b)n = ak bn−k
k!(n − k)!
k=0
COMPLEX DERIVATIVES AND INTEGRALS 145

we obtain
1
1
X 1!
(a + b) = ak b1−k ⇒
k!(1 − k)!
k=0
1 1!
a+b = a0 b1−0 + a1 b1−1
0!(1 − 0)! 1!(1 − 1)!
= a0 b1 + a1 b0
= b + a,

which is nothing more than a simple identity that is clearly true for both real
numbers and complex numbers. Let us now assume the result for the case
n − 1,
n−1
n−1
X (n − 1)!
(a + b) = ak bn−1−k ,
k!(n − 1 − k)!
k=0

and show that this implies the case n. In order to do this, we multiply both
sides of this formula by (a + b), thus obtaining

(a + b)n
n−1
X (n − 1)!
= (a + b) ak bn−1−k
k!(n − 1 − k)!
k=0
n−1 n−1
X (n − 1)! X (n − 1)!
= ak+1 bn−1−k + ak bn−k .
k!(n − 1 − k)! k!(n − 1 − k)!
k=0 k=0

In order to show that the result holds for complex numbers, it suffices to note
that everything that is being done here consists of arithmetic operations of
the field. In the second sum the factor ak bn−k is already in the form we
want. In the first sum we make the transformation of variables k′ = k + 1,
k = k′ − 1, so that this factor takes the correct form, thus obtaining
n n−1
n
X (n − 1)! k ′ n−k ′
X (n − 1)!
(a + b) = ′ ′
a b + ak bn−k ,
(k − 1)! (n − k )! k!(n − 1 − k)!
k ′ =1 k=0

where we can now change the name k′ back to k. The last term of the first
sum and the first term of the second sum must be considered separately, but
for all the other terms the two sums can now be united in a single sum, thus
resulting in
146 SOLUTIONS 9

(a + b)n
(n − 1)! (n − 1)! 0 n
= an bn−n + a b +
(n − 1)!(n − n)! 0!(n − 1)!
n−1  
X 1 1
+ (n − 1)! + ak bn−k
(k − 1)!(n − k)! k!(n − 1 − k)!
k=1
n−1  
n 0 0 n
X k n−k
= a b +a b + (n − 1)! + ak bn−k
k!(n − k)! k!(n − k)!
k=1
n−1
X (n − 1)!
= a0 bn + (k + n − k) ak bn−k + an b0 .
k!(n − k)!
k=1

As one can see, we get a factor of n that completes the factorial of n in the
numerator. It is easy to verify that the first and last terms can be included
in the sum as the terms k = 0 and k = n respectively, so that we in fact have
the result
n
X n!
(a + b)n = ak bn−k .
k!(n − k)!
k=0

Since all this derivation did not use anything other than the operations and
properties of the fields, and since they all hold for complex numbers as well
as for real numbers, it follows that this result holds for all complex numbers
a and b, and for any natural number n.

Problem 6. Prove by finite induction the Cauchy integral formula for the
n-th derivative of an analytic function,
n! f (z)
I
n′
f (z0 ) = dz,
2πı C (z − z0 )n+1
that is, assume the formula for the case n − 1 and show that this implies that
it holds for the case n.
Hints: do not separate the expression (z − z0 ) during the calculations; write
δz = z1 − z0 , recalling that δz will be eventually taken to zero; and write the
difference z − z1 as z − z1 = (z − z0 ) − δz.

Complete Solution:

We will prove the general Cauchy formula for the n-th derivative of an ana-
lytic function f (z), by means of finite induction. This means that we have
the result for the case n = 0, which is the Cauchy formula
COMPLEX DERIVATIVES AND INTEGRALS 147

1 f (z)
I
f (z0 ) = dz,
2πı C z − z0
and that we will assume the validity of the case n − 1,
(n − 1)! f (z)
I
(n−1)′
f (z0 ) = n
dz,
2πı C (z − z0 )

from which we will prove the case n. In order to do this, we use this formula
to calculate f (n−1)′ (z) at two points, z0 and z1 , and set up the ratio which,
in the limit (z1 − z0 ) → 0, is the definition of n-th derivative f n′ (z),

δf (n−1)′ (z) f (n−1)′ (z1 ) − f (n−1)′ (z0 )


=
δz z1 − z0
I  
(n − 1)! 1 1
= − f (z) dz
2πı(z1 − z0 ) C (z − z1 )n (z − z0 )n
(n − 1)! (z − z0 )n − (z − z1 )n
I
= f (z) dz.
2πı(z1 − z0 ) C (z − z1 )n (z − z0 )n

We will now write everything in terms of δz = (z1 − z0 ), using the fact that
we can write that (z − z1 ) = (z − z0 ) − δz,

δf (n−1)′ (z)
δz
(n − 1)! (z − z0 )n − [(z − z0 ) − δz]n
I
= f (z) dz
2πıδz C (z − z0 )n [(z − z0 ) − δz]n
(n − 1)!
= ×
2πıδz
I (z − z )n − Pn n! k k n−k
0 k=0 k!(n−k)! (−1) (δz) (z − z0 )
× f (z) dz,
C (z − z0 )n [(z − z0 ) − δz]n
where we used in the numerator the Newton binomial formula, applied to
[(z − z0 ) − δz]n ,
n
n
X n!
[(z − z0 ) − δz] = (−1)k (δz)k (z − z0 )n−k .
k!(n − k)!
k=0

We now observe that, in the numerator of the resulting formula, the first
term cancels with the term k = 0 of the sum, so that only the terms k ≥ 1
of this sum remain, and we therefore have
148 SOLUTIONS 9

δf (n−1)′ (z)
δz
I − Pn n! k k n−k
(n − 1)! k=1 k!(n−k)! (−1) (δz) (z − z0 )
= f (z) dz
2πıδz C (z − z0 )n [(z − z0 ) − δz]n
I Pn n! k+1 (δz)k−1 (z − z )n−k
(n − 1)! k=1 k!(n−k)! (−1) 0
= n n
f (z) dz,
2πı C (z − z0 ) [(z − z0 ) − δz]
where we canceled a factor of δz. Let us now observe that, in the denomi-
nator, the δz → 0 limit produces a finite and non-zero result, whereas in the
numerator the same limit produces zero for all terms of the sum except for
the term k = 1, so that we can write that

f n′ (z0 )
δf (n−1)′ (z)
= lim
δz→0 δz
I Pn n! k+1 (δz)k−1 (z − z )n−k
(n − 1)! k=1 k!(n−k)! (−1) 0
= lim n n
f (z) dz
2πı δz→0 C (z − z0 ) [(z − z0 ) − δz]
Pn n! k+1 (δz)k−1 (z − z )n−k
(n − 1)! k=1 k!(n−k)! (−1)
I
0
= lim n n
f (z) dz
2πı C δz→0 (z − z0 ) [(z − z0 ) − δz]
n! 2 n−1
(n − 1)! 1!(n−1)! (−1) (z − z0 )
I
= lim n n
f (z) dz
2πı C δz→0 (z − z0 ) [(z − z0 ) − δz]
n(n − 1)! 1
I
= lim n
f (z) dz
2πı C δz→0 (z − z0 )[(z − z0 ) − δz]
n! f (z)
I
= dz,
2πı C (z − z0 )n+1
where we can pass the limit into the integral because the integration contour
C is fixed, it does not change at all when we take the δz → 0 limit, and thus
we prove that
n! f (z)
I
f n′ (z0 ) = dz,
2πı C (z − z0 )n+1
as was our intention. Since we already know that the case n = 0 holds, and
since we have just shown that the case n − 1 implies the case n, it follows by
finite induction that the result holds for all values of n.

Problem 7. Consider an analytic function f (z). Suppose that we differen-


tiate it a certain number n of times, and then integrate it the same number
COMPLEX DERIVATIVES AND INTEGRALS 149

of times in the sense of indefinite integration or primitivization, thus obtain-


ing an analytic function g(z). Write the most general possible form of the
difference g(z) − f (z).

Complete Solution:

Let us start by examining the problem in the case n = 1. The analytic


function g(z) which is the primitive of the derivative of the analytic function
f (z) may differ from f (z) by a complex constant C0 , which is the integration
constant, so that in this case we can have, in the most general case possible,

g(z) − f (z) = C0 .

If we have n = 2 and we take two derivatives followed by two integrations, the


two first derivatives, g ′ (z) and f ′ (z) also differ in this way, say by a constant
C1 , g ′ (z) − f ′ (z) = C1 , but when we make the second integration on g′ (z),
the term C1 is integrated to C1 z, and a new integration constant appears,
which we may call C0 . Thus, in this case the most general possible difference
is a first degree polynomial, that is,

g(z) − f (z) = C0 + C1 z.

Further, examining the case n = 3, in this case the three integrations produce,
on the way back, in the first step a constant C2 , in the second step the
combination C1 + C2 z, and in the third step the combination C0 + C1 z +
C2 z 2 /2, where we now have three arbitrary integration constants. It is clear
now that the generalization to the case n produces a polynomial of order
n − 1, that is,
C2 2 C3 3 C4 3 Cn−1 n−1
g(z) − f (z) = C0 + C1 z + z + z + z + ... + z .
2 3! 4! (n − 1)!

It is interesting to note that the set of functions which appear naturally in


this process is the set of function z n /n! that we got here. However, since all
integration constants are arbitrary, we can write, in a completely equivalent
way, that the most general possible difference is given by
n−1
X
g(z) − f (z) = Ak z k ,
k=0

for n arbitrary complex constants Ak , k ∈ {0, . . . , n − 1}.


150 SOLUTIONS 9
Solutions 10

Complex Inequalities and


Series

We present here complete and commented solutions to all problems proposed


in Chapter 10 of the text. For reference, the propositions of the problems are
repeated here. The problems are discussed in the order in which they were
proposed within the problem set of that chapter.

Problem 1. Starting from the discrete triangle inequality for an arbitrary


integer N ,

N
X N
X
zn ≤ |zn |,
n=1 n=1

show that the corresponding inequality for complex integrals over finite in-
tegration contours C holds,
Z Z
f (z) dz ≤ |f (z)| |dz| ,
C C

so long as the two integrals involved exist.


Hint: use the Riemann definition of the integrals.

Complete Solution:

According to the definition of a complex Riemann integral, we divide the


curve C in N segments of equal length |δz|, and then we write that

151
152 SOLUTIONS 10

Z N
X
f (z) dz = lim f (zn ) δzn ,
C δz → 0
n=1
N →∞

where each zn is within the corresponding segment described by δzn . Using


the triangle inequalities for these finite sums we have now
N
X N
X
f (zn ) δzn ≤ |f (zn ) δzn |
n=1 n=1
XN
= |f (zn )| |δzn |.
n=1

Since this holds for any value of N , it also holds in the limit in which N goes
to infinity, so long as the sums converge, so that we can write that

Z N
X
f (z) dz = lim f (zn ) δzn
C δz → 0
n=1
N →∞
N
X
= lim f (zn ) δzn
δz → 0
n=1
N →∞
N
X
≤ lim |f (zn )| |δzn |
δz → 0
n=1
N →∞
Z
= |f (z)| |dz| ,
C

where we use once again, in this last passage, the definition of the Riemann
integral. So long as the integrals exist, that is, so long as the indicated limits
exist, we therefore obtain the result
Z Z
f (z) dz ≤ |f (z)| |dz| .
C C

Problem 2. Starting from the Cauchy integral formulas that give the
derivatives of an analytic function f (z) in terms of contour integrals,
n! f (z)
I
n′
f (z0 ) = dz,
2πı C (z − z0 )n+1
COMPLEX INEQUALITIES AND SERIES 153

where the integration contour C contains z0 , show that the inequalities

n!
f n′ (z0 ) ≤ |f |
r0n C

hold, where the indicated average is taken on a circular contour C of radius


r0 centered at z0 .

Complete Solution:

Starting from the Cauchy integral formulas that give the derivatives of an
analytic function f (z) in terms of contour integrals,

n! f (z)
I
n′
f (z0 ) = dz,
2πı C (z − z0 )n+1

and taking C as a circle of radius r0 centered at z0 , on which we have that


|z − z0 | = r0 and dz = r0 ı exp(ıθ) dθ, we can write that

n! f (z)
I
n′
f (z0 ) = dz
2πı C (z − z0 )n+1
n! f (z)
I
≤ dz
2π C (z − z0 )n+1
n! |f (z)|
I
= |dz|
2π C |z − z0 |n+1
n! 2π |f (z)|
Z
= r0 dθ
2π 0 r0n+1
 Z 2π 
n! 1
= |f (z)| dθ
r0n 2π 0
n!
= |f | ,
r0n C

that is, we conclude that in fact the inequalities

n!
f n′ (z0 ) ≤ |f | ,
r0n C

hold, where the indicated average is taken on the circular contour C of radius
r0 centered at z0 .
154 SOLUTIONS 10

Problem 3. Consider the Taylor series of the functions exp(x), sin(x) and
cos(x), for real x, around x = 0.

(a) Write each one of the series as an infinite sum of a general term.

(b) Write the series of exp(ıx), with x real, separating its real and imagi-
nary parts.

(c) Identify the real and imaginary parts of the series, and write the re-
sulting relation between the functions involved.

Answer: exp(ıx) = cos(x) + ı sin(x).

Complete Solution:

Taking the appropriate derivatives and applying at the point x = 0, it is not


difficult to verify that the series are as follows,

1 2 1 1 1
exp(x) = 1 + x + x + x3 + x4 + x5 + . . . ,
2 3! 4! 5!
1 3 1 5 1 7 1 9
sin(x) = x − x + x − x + x + ... ,
3! 5! 7! 9!
1 2 1 4 1 6 1 8
cos(x) = 1 − x + x − x + x + ... .
2 4! 6! 8!

(a) Systematizing the above sequences of coefficient, we can write the series
in explicit form,


X 1 n
exp(x) = x ,
n!
n=0

X (−1)k
sin(x) = x2k+1 ,
(2k + 1)!
k=0

X (−1)k
cos(x) = x2k .
(2k)!
k=0

(b) We have for the series of exp(ıx), separating terms with even n from
those with odd n,
COMPLEX INEQUALITIES AND SERIES 155


X 1 n n
exp(ıx) = ı x
n!
n=0
∞ ∞
X 1 2k 2k X 1
= ı x + ı2k+1 x2k+1
(2k)! (2k + 1)!
k=0 k=0
∞ ∞
X 1 k X 1 k
= ı2 x2k + ı ı2 x2k+1
(2k)! (2k + 1)!
k=0 k=0
∞ ∞
X (−1)k X (−1)k
= x2k + ı x2k+1 ,
(2k)! (2k + 1)!
k=0 k=0

where we used the fact that ı2 = −1, and the sums are now explicitly
real.

(c) As discussed above, we have that

∞ ∞
X (−1)k 2k
X (−1)k
exp(ıx) = x +ı x2k+1 ,
(2k)! (2k + 1)!
k=0 k=0

where the real and imaginary parts are written explicitly. We now
identify the first series as the Taylor series of cos(x), and the second as
the Taylor series of sin(x). Since the series are faithful representations
of the respective functions, we conclude that holds between them the
identity

exp(ıx) = cos(x) + ı sin(x),

which is the Euler formula.

Problem 4. Consider the function f (z) = cos(z).

(a) Expand the function around the point z = π/2.

(b) Identify the coefficients of the terms of the resulting series and write
the trigonometric identity that follows from them.

Answer: cos(z) = sin(π/2 − z).


156 SOLUTIONS 10

Complete Solution:

We will expand in a power series around z0 = π/2 the function

w(z) = cos(z).

(a) Calculating the first few derivatives we have

w0′ (z) = cos(z),


w1′ (z) = − sin(z),
w2′ (z) = − cos(z),
w3′ (z) = sin(z),
w4′ (z) = cos(z),
w5′ (z) = − sin(z),
... ... .

Systematizing the answers separately for the cases of even n and odd
n, we can write that

w(2k)′ (z) = (−1)k cos(z),


w(2k+1)′ (z) = (−1)k+1 sin(z),

for k ∈ {0, 1, 2, 3, . . . , ∞}. Applying this at z0 = π/2 we have

w(2k)′ (π/2) = (−1)k cos(π/2)


= 0,
(2k+1)′
w (π/2) = (−1)k+1 sin(π/2)
= (−1)k+1 .

We have therefore for the Taylor series of the function around π/2,


X (−1)k+1
w(z) = (z − π/2)2k+1 ⇒
(2k + 1)!
k=0

X (−1)k
cos(z) = − (z − π/2)2k+1 .
(2k + 1)!
k=0
COMPLEX INEQUALITIES AND SERIES 157

(b) If we compare the result above with the series of sin(z) around z = 0,


X (−1)k
sin(z) = z 2k+1 ,
(2k + 1)!
k=0

we verify that the coefficients and the powers involved are the same,
so that it follows that cos(z) = − sin(z − π/2) or, to put it in a more
familiar form for the case in which we have a real z, cos(z) = sin(π/2 −
z).

Problem 5. Consider the function f (z) = sinh(z).

(a) Expand the function around the point z = ıπ.

(b) Identify the coefficients of the terms of the resulting series and write
the identity between functions that follows from them.

Answer: sinh(z) = sinh(ıπ − z).

Complete Solution:

We will expand in a power series around z0 = ıπ the function

w(z) = sinh(z).

First of all, let us recall that the hyperbolic functions are given by the series

X 1
cosh(z) = z 2k ,
(2k)!
k=0

X 1
sinh(z) = z 2k+1 ,
(2k + 1)!
k=0

and that for purely imaginary arguments it follows therefore that



X 1 2k 2k
cosh(ıy) = ı y
(2k)!
k=0

X (−1)k 2k
= y
(2k)!
k=0
= cos(y),
158 SOLUTIONS 10


X 1
sinh(ıy) = ı2k+1 y 2k+1
(2k + 1)!
k=0

X (−1)k
= ı y 2k+1
(2k + 1)!
k=0
= ı sin(y).
(a) Calculating the first few derivatives we have

w0′ (z) = sinh(z),


w1′ (z) = cosh(z),
w2′ (z) = sinh(z),
w3′ (z) = cosh(z),
w4′ (z) = sinh(z),
w5′ (z) = cosh(z),
... ... .

Systematizing the answers separately for the cases of even n and odd
n, we can write that

w(2k)′ (z) = sinh(z),


w(2k+1)′ (z) = cosh(z),

for k ∈ {0, 1, 2, 3, . . . , ∞}. Applying this at z0 = ıπ we have

w(2k)′ (π/2) = sinh(ıπ)


= ı sin(π)
= 0,
(2k+1)′
w (π/2) = cosh(ıπ)
= cos(π)
= −1.

We have therefore for the Taylor series of the function around ıπ,

X (−1)
w(z) = (z − ıπ)2k+1 ⇒
(2k + 1)!
k=0

X 1
sinh(z) = − (z − ıπ)2k+1 .
(2k + 1)!
k=0
COMPLEX INEQUALITIES AND SERIES 159

(b) If we compare the result above with the series of sinh(z) around z = 0,
which was given before, we verify that the coefficients and the powers
involved are the same, so that it follows that sinh(z) = − sinh(z − ıπ),
that is, we have that sinh(z) = sinh(ıπ − z).

Problem 6. (Challenge Problem) For each of the following functions,


use its Taylor series around z = 0 in order to explicitly write the real and
imaginary parts of the function in explicit form, in terms of other known
functions.

(a) sin(z).

Answer: sin(x) cosh(y) + ı cos(x) sinh(y).


(b) cos(z).

Answer: cos(x) cosh(y) − ı sin(x) sinh(y).


(c) sinh(z).

Answer: sinh(x) cos(y) + ı cosh(x) sin(y).


(d) cosh(z).

Answer: cosh(x) cos(y) + ı sinh(x) sin(y).

Hints: it will be necessary to use the Newton binomial formula, and to figure
out how to make changes of summation variables in infinite double sums with
two indices.

Complete Solution:

Let us begin by recalling that the functions listed are given by the series

X (−1)k
sin(z) = z 2k+1 ,
(2k + 1)!
k=0

X (−1)k 2k
cos(z) = z ,
(2k)!
k=0

X 1
sinh(z) = z 2k+1 ,
(2k + 1)!
k=0

X 1
cosh(z) = z 2k .
(2k)!
k=0
160 SOLUTIONS 10

Moreover, in all cases it will be necessary to use the Newton binomial formula
for (x + ıy)n , separately for the cases of even n and odd n,

n
X n!
(x + ıy)n = xn−l (ıy)l
l!(n − l)!
l=0
n
X ıl n!
= xn−l y l ⇒
l!(n − l)!
l=0
2k
X ıl (2k)!
(x + ıy)2k = x2k−l y l ,
l!(2k − l)!
l=0
2k+1
X ıl (2k + 1)!
(x + ıy)2k+1 = x2k+1−l y l ,
l!(2k + 1 − l)!
l=0

where we wrote separately the cases n = 2k and n = 2k + 1. In each one of


these cases, it is necessary to separate the sum over l in two, one for even l
and another for odd l, in order to allow us to separate the real and imaginary
parts. Let us do first l = 2j and then l = 2j + 1. Since l ranges from 0 to n,
that is, from 0 to 2k or from 0 to 2k + 1, the terms with even l range from
0 to 2k, regardless of what the parity of n may be. Thus we see that in this
case we have l = 2j with j going from 0 to k. However, the terms with odd l
go from 1 to 2k + 1 if n is odd, but only from 1 to 2k − 1 if n its even. Thus,
if either l or n are odd, we have l = 2j + 1 with j going from 0 to k, but if
l is odd and n even, we have l = 2j + 1 with j going from 0 to k − 1. This
means, in particular, that in the latter case the value k = 0 is not included.
We therefore have, for even n,

2k
2k
X ıl (2k)!
(x + ıy) = x2k−l y l
l!(2k − l)!
l=0
k
X ı2j (2k)!
= x2k−2j y 2j +
(2j)!(2k − 2j)!
j=0
k−1
X ı2j+1 (2k)!
+ x2k−2j−1 y 2j+1
(2j + 1)!(2k − 2j − 1)!
j=0
k
X (−1)j (2k)!
= x2k−2j y 2j +
(2j)!(2k − 2j)!
j=0
COMPLEX INEQUALITIES AND SERIES 161

k−1
X (−1)j (2k)!
+ı x2k−2j−1 y 2j+1 ,
(2j + 1)!(2k − 2j − 1)!
j=0

with real and imaginary parts properly separated, and in which the second
term exists only for k ≥ 1. For odd n we have
2k+1
X ıl (2k + 1)!
(x + ıy)2k+1 = x2k+1−l y l
l!(2k + 1 − l)!
l=0
k
X ı2j (2k + 1)!
= x2k+1−2j y 2j +
(2j)!(2k + 1 − 2j)!
j=0
k
X ı2j+1 (2k + 1)!
+ x2k+1−2j−1 y 2j+1
(2j + 1)!(2k + 1 − 2j − 1)!
j=0
k
X (−1)j (2k + 1)!
= x2k+1−2j y 2j +
(2j)!(2k + 1 − 2j)!
j=0
k
X (−1)j (2k + 1)!
+ı x2k−2j y 2j+1 ,
(2j + 1)!(2k − 2j)!
j=0

where the real and imaginary parts are properly separated. We can write the
final form of these two expressions in terms of the difference (k − j),
k
2k
X (−1)j (2k)!
(x + ıy) = x2(k−j) y 2j +
(2j)![2(k − j)]!
j=0
k−1
X (−1)j (2k)!
+ı x2(k−j)−1 y 2j+1 ,
(2j + 1)![2(k − j) − 1]!
j=0
k
X (−1)j (2k + 1)!
(x + ıy)2k+1 = x2(k−j)+1 y 2j +
(2j)![2(k − j) + 1]!
j=0
k
X (−1)j (2k + 1)!
+ı x2(k−j) y 2j+1 .
(2j + 1)![2(k − j)]!
j=0

(a) sin(z): in this case we have that


X (−1)k
sin(z) = z 2k+1 ,
(2k + 1)!
k=0
162 SOLUTIONS 10

and hence we have only odd powers of z, and therefore we use the
version of the binomial formula odd n, thus obtaining
∞ k
X (−1)k X (−1)j (2k + 1)!
sin(z) = x2(k−j)+1 y 2j +
(2k + 1)! (2j)![2(k − j) + 1]!
k=0 j=0
∞ k
X (−1)k X (−1)j (2k + 1)!
+ı x2(k−j) y 2j+1
(2k + 1)! (2j + 1)![2(k − j)]!
k=0 j=0
∞ X
k
X (−1)k+j
= x2(k−j)+1 y 2j +
(2j)![2(k − j) + 1]!
k=0 j=0
∞ X
k
X (−1)k+j
+ı x2(k−j) y 2j+1 .
(2j + 1)![2(k − j)]!
k=0 j=0

Let us now consider how to make a change of variables in these sums,


in order to switch to using the variable l = k − j, that is, to eliminate
k by means of k = l + j, since now k appears only in this particular
combination. The problem is to determine the set of the pairs (j, l)
which runs through exactly the same set of possibilities as (j, k), with
the limits specified for j and k. In order to do this, let us observe the
diagram in Figure 10.1, where axes are placed with the values of k, j
and l, and where the points that correspond to each term of the double
series are marked.
The description of this set of points in the (j, k) coordinate system
corresponds to going through all the points along vertical segments,
ranging from j = 0 to the diagonal line which represents the equation
j = k. In this case we can have any value of k ∈ [0, ∞), and values of
j ∈ [0, k]. Moreover, the same set of points can be traversed along the
horizontal lines, each interpreted as a copy of the l axis, as shown in
the figure, where we see that l = k − j. In this case, j can take any
value in [0, ∞), and for each value of j we have that l ∈ [0, ∞), so that
in the (j, l) coordinate system neither coordinate is limited. It follows
that we can write
∞ X

X (−1)l+2j
sin(z) = x2l+1 y 2j +
(2j)!(2l + 1)!
l=0 j=0
∞ X ∞
X (−1)l+2j
+ı x2l y 2j+1
(2j + 1)!(2l)!
l=0 j=0
COMPLEX INEQUALITIES AND SERIES 163

4
0 1 2 3 4 5 l
3

0 1 2 3 4 5 6 7 k

Figure 10.1: Diagram depicting the change of summation variables in the


double series, in the case of the sine.


" # ∞ 
X (−1) l X 1
= x2l+1  y 2j  +
(2l + 1)! (2j)!
l=0 j=0
"∞ # ∞ 
X (−1)l X 1
+ı x2l  y 2j+1  ,
(2l)! (2j + 1)!
l=0 j=0

where we see that, because of the change of variables, it was possible


to separate each double series into the product of two simple series.
We can now identify each of the four series as the Taylor series of
known functions. Those in which the terms exchange sign represent
trigonometric functions, and the others represent hyperbolic functions,
so that we have

sin(z) = sin(x) cosh(y) + ı cos(x) sinh(y),

just as was previously derived by other means, using the Euler formula
164 SOLUTIONS 10

instead of the series representation.

(b) cos(z): in this case we have that


X (−1)k
cos(z) = z 2k ,
(2k)!
k=0

and hence we have only even powers of z, and therefore we use the
version of the binomial formula for even n, noting that the second term
of this formula exists only for k ≥ 1, and thus obtaining

∞ k
X (−1)k X (−1)j (2k)!
cos(z) = x2(k−j) y 2j +
(2k)! (2j)![2(k − j)]!
k=0 j=0
∞ k−1
X (−1)k X (−1)j (2k)!
+ı x2(k−j)−1 y 2j+1
(2k)! (2j + 1)![2(k − j) − 1]!
k=1 j=0
∞ X
k
X (−1)k+j
= x2(k−j) y 2j +
(2j)![2(k − j)]!
k=0 j=0
∞ X
k−1
X (−1)k+j
+ı x2(k−j)−1 y 2j+1 .
(2j + 1)![2(k − j) − 1]!
k=1 j=0

In the first term the same transformation of variables that we used


before can be used. However, in the second term we have that j ranges
from 0 to k − 1, instead of k, and that k ≥ 1. Thus, in this term we
make the change of variables l = k − j − 1, that is, k = l + j + 1, which
is illustrated by the diagram of Figure 10.2.
Since this diagram is similar to the previous one, and considering the
new variation intervals which are shown, it is not difficult to see that
in this case once more the (j, l) coordinate system variables start at 0
and are not limited. It follows that we can write

∞ X

X (−1)l+2j 2l 2j
cos(z) = x y +
(2j)!(2l)!
l=0 j=0
∞ X ∞
X (−1)l+2j+1
+ı x2l+1 y 2j+1
(2j + 1)!(2l + 1)!
l=0 j=0
COMPLEX INEQUALITIES AND SERIES 165

4
0 1 2 3 4 5 l
3

0 1 2 3 4 5 6 7 8 k

Figure 10.2: Diagram depicting the change of summation variables in the


double series, in the case of the cosine.

" ∞
# ∞ 
X (−1)l 1
X
= x2l  y 2j  +
(2l)! (2j)!
l=0 j=0
"∞ # ∞ 
X (−1)l X 1
−ı x2l+1  y 2j+1  ,
(2l + 1)! (2j + 1)!
l=0 j=0

where we see that, because of the change of variables, it was possible


to separate each double series into the product of two simple series.
We can now identify each of the four series as the Taylor series of
known functions. Those in which the terms exchange sign represent
trigonometric functions, and the others represent hyperbolic functions,
so that we have

cos(z) = cos(x) cosh(y) − ı sin(x) sinh(y),

just as was previously derived by other means, using the Euler formula
166 SOLUTIONS 10

instead of the series representation.


(c) sinh(z): in this case we have that

X 1
sinh(z) = z 2k+1 ,
(2k + 1)!
k=0

and hence we have only odd powers of z, and therefore we use the
version of the binomial formula for odd n, thus obtaining
∞ k
X 1 X (−1)j (2k + 1)!
sinh(z) = x2(k−j)+1 y 2j +
(2k + 1)! (2j)![2(k − j) + 1]!
k=0 j=0
∞ k
X 1 X (−1)j (2k + 1)!
+ı x2(k−j) y 2j+1
(2k + 1)! (2j + 1)![2(k − j)]!
k=0 j=0
∞ X
k
X (−1)j
= x2(k−j)+1 y 2j +
(2j)![2(k − j) + 1]!
k=0 j=0
∞ X
k
X (−1)j
+ı x2(k−j) y 2j+1 .
(2j + 1)![2(k − j)]!
k=0 j=0

In this case the change of variables in each sum is exactly the same
as that used in the case of the function sin(z), and therefore it follows
that we can write
∞ X

X (−1)j
sinh(z) = x2l+1 y 2j +
(2j)!(2l + 1)!
l=0 j=0
∞ X ∞
X (−1)j
+ı x2l y 2j+1
(2j + 1)!(2l)!
l=0 j=0
"∞ # ∞ 
X 1 X (−1)j
= x2l+1  y 2j  +
(2l + 1)! (2j)!
l=0 j=0
"∞ # ∞ 
X 1 X (−1)j
+ı x2l  y 2j+1  ,
(2l)! (2j + 1)!
l=0 j=0

where we see that, because of the change of variables, it was possible


to separate each double series into the product of two simple series.
COMPLEX INEQUALITIES AND SERIES 167

We can now identify each of the four series as the Taylor series of
known functions. Those in which the terms exchange sign represent
trigonometric functions, and the others represent hyperbolic functions,
so that we have

sinh(z) = sinh(x) cos(y) + ı cosh(x) sin(y),

just as was previously derived by other means, using the Euler formula
instead of the series representation.

(d) cosh(z): in this case we have that



X 1
cosh(z) = z 2k ,
(2k)!
k=0

and hence we have only even powers of z, and therefore we use the
version of the binomial formula for even n, noting that the second term
of this formula exists only for k ≥ 1, and thus getting

∞ k
X 1 X (−1)j (2k)!
cosh(z) = x2(k−j) y 2j +
(2k)! (2j)![2(k − j)]!
k=0 j=0
∞ k−1
X 1 X (−1)j (2k)!
+ı x2(k−j)−1 y 2j+1
(2k)! (2j + 1)![2(k − j) − 1]!
k=1 j=0
k
∞ X
X (−1)j
= x2(k−j) y 2j +
(2j)![2(k − j)]!
k=0 j=0
∞ X
k−1
X (−1)j
+ı x2(k−j)−1 y 2j+1 .
(2j + 1)![2(k − j) − 1]!
k=1 j=0

In this case the change of variables in each sum is exactly the same
as that used in the case of the function cos(z), and therefore it follows
that we can write
∞ X

X (−1)j
cosh(z) = x2l y 2j +
(2j)!(2l)!
l=0 j=0
∞ X ∞
X (−1)j
+ı x2l+1 y 2j+1
(2j + 1)!(2l + 1)!
l=0 j=0
168 SOLUTIONS 10


" # ∞ 
X 1 X j
(−1) 2j 
= x2l  y +
(2l)! (2j)!
l=0 j=0
"∞ # ∞ 
X 1 X (−1)j
+ı x2l+1  y 2j+1  ,
(2l + 1)! (2j + 1)!
l=0 j=0

where we see that, because of the change of variables, it was possible


to separate each double series into the product of two simple series.
We can now identify each of the four series as the Taylor series of
known functions. Those in which the terms exchange sign represent
trigonometric functions, and the others represent hyperbolic functions,
so that we have

cosh(z) = cosh(x) cos(y) + ı sinh(x) sin(y),

just as was previously derived by other means, using the Euler formula
instead of the series representation.
Solutions 11

Series, Limits and


Convergence

We present here complete and commented solutions to all problems proposed


in Chapter 11 of the text. For reference, the propositions of the problems are
repeated here. The problems are discussed in the order in which they were
proposed within the problem set of that chapter.

Problem 1. Consider an arbitrary series whose terms are all positive real
numbers, such as

X
S∞ = |ak |.
k=0

Show that, if this series is limited from above, that is, if there is a positive
real number SM such that S∞ ≤ SM , then the series necessarily converges
to some positive real number less than or equal to SM .
Hint: the statement S∞ ≤ SM actually means that all the partial sums SN
of this series are limited in this way, for all N .

Complete Solution:

If we consider the partial sums of the series S∞ ,


N
X
SN = |ak |,
k=0

then the condition of the existence of an upper limit SM is that SN ≤ SM


for all N . Thus, if the upper limit exists, then the entire sequence of partial

169
170 SOLUTIONS 11

sums is contained within the closed interval [0, SM ], that is, 0 ≤ SN ≤ SM


for all N . We should now consider the special cases in which the limit of the
sequence is 0 (which only happens if all terms are zero) or SM . In any of
these two cases, the limit exists and hence the convergence of the sequence
is established. In any other case, both the complete sequence and its limit,
if it exists, must be contained in the open interval (0, SM ).
Now, since the series is a sum of positive numbers, when we add more
numbers to the sum it can never decrease, so that the sequence of real num-
bers SN is an increasing sequence, and therefore ordered. Thus, if this se-
quence goes above the point SM /2 at some stage, then it can no longer be
below this point later on. If the limit of the sequence is the point SM /2 then
the convergence of the sequence is established. In any other case, if it goes
above the point SM /2 then there is a certain value N1 of N such that for
N above this value all the rest of the sequence is contained in the interval
(SM /2, SM ).
We therefore have two remaining possibilities, one and only one of which
must be true: either the sequence never goes above the point SM /2, in which
case it is all contained in the interval (0, SM /2), or it goes above the point
SM /2, in which case the part of the sequence above N1 is all contained in
the interval (SM /2, SM ). In any of these two cases, the part of the sequence
for N > N1 is always contained in an open interval of length SM /2. This
implies, of course, that if there is a limit S∞ then it is also contained in this
interval, and that the distance from each point of the sequence to this limit
is smaller than SM /2, that is, we have the condition that
SM
N > N1 =⇒ |S∞ − SN | < .
2
If we repeat this whole argument, starting from this point, we conclude that
either the sequence converges to one end of the intervals involved, or there is
a value N2 of N such that
SM
N > N2 =⇒ |S∞ − SN | < .
22
Iterating the argument k times, we conclude that either the sequence con-
verges to one end of the intervals involved, or there is a value Nk of N such
that
SM
N > Nk =⇒ |S∞ − SN | < k .
2
Making the value of k large enough, the quantity SM /2k can be made as
small as one likes because the limit of 1/2k when k → ∞ is zero. We can
SERIES, LIMITS AND CONVERGENCE 171

now establish the criterion of the convergence of the series. Given a strictly
positive real number ǫ, we determine a value of k(ǫ) such that
SM
< ǫ,
2k(ǫ)
which is always possible to do. Now taking the value Nk corresponding to
this k(ǫ), we have that
SM
N > Nk(ǫ) =⇒ |S∞ − SN | < < ǫ,
2k(ǫ)
which establishes the convergence of the series. Actually, the easiest way
to establish the convergence in this series is to use the Cauchy convergence
criterion for the sequence SN , because in this way there is no need to explicitly
mention the limit S∞ of the series. In a sense, this is what we are doing here
indirectly. What the Cauchy criterion says is that, if the sequences of the
remaining elements become all infinitely close to each other, as one progresses
along the sequence, then this is sufficient to ensure that the series converges.
In a way, this is what we are doing here, when we show that the infinite final
part of the sequence can be placed within a sequence of nested intervals whose
lengths go to zero in the N → ∞ limit. The existence of the limit S∞ is a
consequence of the fact that the completeness of the real line is equivalent
to the statement that this infinite sequence of nested intervals, known as
Bolzano nested intervals, must all contain a single real number, which is the
limit.

Problem 2. Show that the real numerical series of positive terms



X 1
S∞ =
(2k + 1)(2k + 2)
k=0

converges to a finite value, limiting it from above by an asymptotic integral


of the real function 1/k 2 , such an integral being finite. Recall that we can
ignore some initial terms of the sum without affecting the convergence of the
series, if this is necessary for the argument.

Complete Solution:

Consider the infinite sum given by



X 1
S∞ = .
(2k + 1)(2k + 2)
k=0
172 SOLUTIONS 11

S
1
k2

0 1 2 3 4 5 k

Figure 11.1: Maximizing the series by an asymptotic integral.

Separating the first two terms we can write this as



1 1 X 1
S∞ = + + .
2 12 (2k + 1)(2k + 2)
k=2

Replacing the factors (2k + 1) and (2k + 2) by 2k, we can limit the sum from
above, obtaining the relation

1 1 X 1
S∞ < + +
2 12 (2k)2
k=2

7 1 X 1
= + .
12 4 k2
k=2

Let us now compare the remaining sum with the definite integral of the real
function 1/k 2 between k = 1 and k → ∞, as illustrated by the diagram in
Figure 11.1.
Considering the height under the curve at the point k = 2, which is equal
to the first term of the sum, and a unit-width rectangle with that height
between the points k = 1 and k = 2, we see that the area of this rectangle
is equal to the first term of the sum, and that it is contained entirely below
the graph of the function between these two points, a region whose area is
given by the definite integral of the function between the two points. In other
words, we have that
 Z 2
1 1
< dk.
k2 2 1 k 2
SERIES, LIMITS AND CONVERGENCE 173

Repeating this construction between the points k = 2 and k = 3, we obtain


in the same way
 Z 3
1 1
2
< 2
dk.
k 3 2 k

Now adding these two inequalities, we obtain the inequality


3 Z 3
X 1 1
< dk,
k2 1 k 2
k=2

and iterating indefinitely the procedure for increasing values of k, we have in


the asymptotic limit that
∞ Z ∞
X 1 1
2
< dk.
k 1 k2
k=2

Since the integral can be calculated and is finite, we obtain the inequality
∞ ∞
X 1 −1
<
k2 k 1
k=2
= 1.

It follows therefore that we have for our previous inequality



7 1X 1
S∞ < +
12 4 k2
k=2
7 1
< +
12 4
5
= .
6
We have therefore the upper bound S∞ < 5/6, and since S∞ is a sum of
positive terms, it follows that it must converge to some value between 0 and
5/6.

Problem 3. Show that the real numerical series of positive terms



X 1
S∞ =
n=0
n + 1
174 SOLUTIONS 11

S
1
n

0 1 2 3 4 5 n

Figure 11.2: Minimizing the series by an asymptotic integral.

diverges to positive infinity, limiting it from below by as asymptotic integral


of the real function 1/n, such an integral being infinite. Recall that we can
ignore some initial terms of the sum without affecting the convergence of the
series, if this is necessary for the argument.

Complete Solution:

Consider the infinite sum given by



X 1
S∞ = .
n+1
n=0

Through a simple change of variables in the index of the sum, we can also
write this as

X 1
S∞ = .
n
n=1

Let us now compare this sum with the definite integral of the real function
1/n between n = 1 and n → ∞, as illustrated by the diagram in Figure 11.2.
Considering the height under the curve at the point n = 1, which is equal
to the first term of the sum, and a unit-width rectangle with that height
between the points n = 1 and n = 2, we see that the are of the rectangle is
equal to the first term of the sum, and that it entirely contains the region
below the graph of the function between these two points, a region whose
area is given by the definite integral of function between the two points. In
other words, we have that
SERIES, LIMITS AND CONVERGENCE 175

2 
1 1
Z
dn < .
1 n n 1

Repeating this construction between the points n = 2 and n = 3, we obtain


in the same way
Z 3 
1 1
dn < .
2 n n 2

Now adding these two inequalities, we obtain the inequality

3 2
1 1
Z X
dn < ,
1 n n=1
n

and iterating indefinitely the procedure for increasing values of n, we have


for any maximum value N of n, which can be as large as we want,

N N
1 1
Z X
dn < .
1 n n=1
n

Since the integral can be calculated, we obtain the inequality


N N
X 1
ln(n) < ⇒
1 n
n=1
ln(N ) < SN .

Since the asymptotic N → ∞ limit of the left-hand side of this last inequality
grows without limit, the right-hand side necessarily does the same. It follows
therefore that the sum S∞ diverges to positive infinity.

Problem 4. Consider the function f (z) = 1/(1 + z).

(a) Expand the function f (z) in a power series around z = 0.

(b) Identify the convergence disk of the power series.

(c) Change to the variable w = 1 + z and write a power series for the
function g(w) = 1/w.

(d) Identify the convergence disk in terms of w.


176 SOLUTIONS 11

(e) Show that this power series is a Taylor series of g(w) and identify the
point z0 around which it is developed.

Complete Solution:

We have the analytic function


1
f (z) = .
1+z
(a) Making a table of the first few derivatives we have

1
f 0′ (z) = ,
(1 + z)
−1
f 1′ (z) = ,
(1 + z)2
2
f 2′ (z) = ,
(1 + z)3
−3!
f 3′ (z) = ,
(1 + z)4
4!
f 4′ (z) = ,
(1 + z)5
−5!
f 5′ (z) = ,
(1 + z)6
... ... .

We can write the all derivatives in a single explicit form as

(−1)n n!
f n′ (z) = ,
(1 + z)n+1

which applied at z0 = 0 gives us

f n′ (0) = (−1)n n!,

so that we have for the Taylor series of the function


X
f (z) = (−1)n z n .
n=0
SERIES, LIMITS AND CONVERGENCE 177

(b) Since the function has a single point of singularity, which is at z = −1,
the convergence disk is centered at z0 = 0, with unit radius. Note that
this series converges within this disk, although its coefficients do not
go to zero at all for large values of n.

(c) Making a change of variables w = 1 + z we have for the function

f (z) = g(w)
1
= ,
w

and, by simple composition of functions, we get the power series


X
g(w) = (−1)n (w − 1)n .
n=0

(d) Since the series for g(w) is in fact the same series we saw before for
f (z), it converges at the same points where that series did converge,
that is, within a disk of unit radius. In terms of w the disk is centered
at w0 = 1.

(e) Making a table of the first few derivatives of g(w) we have

1
g 0′ (w) = ,
w
−1
g 1′ (w) = ,
w2
2
g 2′ (w) = ,
w3
−3!
g 3′ (w) = ,
w4
4!
g 4′ (w) = ,
w5
−5!
g 5′ (w) = ,
w6
... ... .

We can write all the derivatives in a single explicit form as

(−1)n n!
gn′ (w) = .
wn+1
178 SOLUTIONS 11

Since the series is written in terms of the powers of (w − 1), it is clear


that the reference point is w0 = 1, where the center of the convergence
disk is indeed located, as we have seen. Applying the above derivatives
at w0 = 1 we have

gn′ (1) = (−1)n n!,

so that we have for the Taylor series of the function g(w)


X
g(w) = (−1)n (w − 1)n ,
n=0

which is in fact the same series that we obtained by simple composition


of functions.


Problem 5. Consider the function w(z) = 2/ 4 − z, where z = x + ıy is
a complex number, while x and y are real numbers.

(a) Write the Taylor expansion of w(z) around z = 0, that is, the Maclaurin
expansion of the function. Write explicitly the general term of the
series.

(b) Determine and describe the interval of convergence of the series on the
real x axis, that is, for y = 0.

(c) Use this √


series in order to define a method for calculating the irrational
number 2 via a limit involving only rational numbers. Show that each
term of the series is a rational number.

(d) Using this series,


√ identify an infinite sequence of rational numbers that
converges to 2.

Complete Solution:

We will expand in a power series around z0 = 0 the function

2
w(z) = √ .
4−z
SERIES, LIMITS AND CONVERGENCE 179

(a) Calculating the first few derivatives we have

2
w0′ (z) = √ ,
4−z
1 2
w1′ (z) = √ ,
2 4 − z3
13 2
w2′ (z) = √ ,
2 2 4 − z5
135 2
w3′ (z) = √ ,
2 2 2 4 − z7
1357 2
w4′ (z) = √ ,
2 2 2 2 4 − z9
13579 2
w5′ (z) = √ ,
2 2 2 2 2 4 − z 11
... ... .

Systematizing the answers, we can write that

(2n − 1)!! 2
wn′ (z) = n √ 2n+1
2 4−z
(2n + 1)!! 2
= √ .
2 (2n + 1) 4 − z 2n+1
n

Applying this at z√0 = 0, and choosing the leaf of the Riemann surface
corresponding to 4 = +2 as the location of the point z0 , we have

(2n + 1)!! 2
wn′ (0) = √
2n (2n + 1) 42n+1
(2n + 1)!! 2
=
2n (2n + 1) 22n+1
(2n + 1)!!
=
23n (2n + 1)
(2n + 1)!!(2n)!!
=
23n (2n)!!(2n + 1)
(2n + 1)!
= 4n
2 n!(2n + 1)
(2n)!
= .
24n n!
180 SOLUTIONS 11

We have therefore for the Taylor series of the function around zero,


X (2n + 1)!!
w(z) = zn
n=0
23n n!(2n+ 1)

X (2n)!
= zn.
24n (n!)2
n=0

(b) Since the function has a single point of singularity located at z = 4,


the convergence disk is a disk of radius 4 centered at the origin. The
intersection of this disk with the real axis is the open interval (−4, 4),
that is, in the case y = 0 the series converges for −4 < x < 4.

(c) If we calculate the function at the point z = 2, which is well within the
real interval of convergence of the series, we get

2
w(2) = √
2

= 2,

so that the calculation


√ of the series for z = 2 gives us an algorithmic
way to calculate 2,


X (2n)! n
w(2) = 2 ⇒
n=0
2 (n!)2
4n


√ X (2n)!
2 = .
2 (n!)2
3n
n=0

Since the factorial (2n)! is an integer, the numerator of each term of


the series is an integer. Since both 23n and (n!)2 are also integers, the
denominator is also an integer, and it follows that each term of the series
is a rational number. Since the sum of rational numbers is rational,
given that the rationals form a field, each partial sum of the series is
a rational number. The sequence of partial sums forms, therefore, a
sequence
√ of rational numbers that converges to the irrational number
2.
SERIES, LIMITS AND CONVERGENCE 181


Problem 6. Consider the function w(z) = 3/ 9 − z, where z = x + ıy is
a complex number, while x and y are real numbers.

(a) Write the Taylor expansion of w(z) around z = 0, that is, the Maclaurin
expansion of the function. Write explicitly the general term of the
series.

(b) Determine and describe the interval of convergence of the series on the
real x axis, that is, for y = 0.

(c) Use this √


series in order to define a method for calculating the irrational
number 3 via a limit involving only rational numbers. Show that each
term of the series is a rational number.

(d) Using this series,


√ identify an infinite sequence of rational numbers that
converges to 3.

Complete Solution:

We will expand in a power series around z0 = 0 the function


3
w(z) = √ .
9−z

(a) Calculating the first few derivatives we have

3
w0′ (z) = √ ,
9−z
1 3
w1′ (z) = √ ,
2 9 − z3
13 3
w2′ (z) = √ ,
2 2 9 − z5
135 3
w3′ (z) = √ ,
2 2 2 9 − z7
1357 3
w4′ (z) = √ ,
2 2 2 2 9 − z9
13579 3
w5′ (z) = √ ,
2 2 2 2 2 9 − z 11
... ... .

Systematizing the answers, we can write that


182 SOLUTIONS 11

(2n − 1)!! 3
wn′ (z) = n √ 2n+1
2 9−z
(2n + 1)!! 3
= √ .
2 (2n + 1) 9 − z 2n+1
n

Applying this at z√0 = 0, and choosing the leaf of the Riemann surface
corresponding to 9 = +3 as the location of the point z0 , we have

(2n + 1)!! 3
wn′ (0) = n √ 2n+1
2 (2n + 1) 9
(2n + 1)!! 3
= n 2n+1
2 (2n + 1) 3
(2n + 1)!!
=
2n 32n (2n + 1)
(2n + 1)!!(2n)!!
=
2 32n (2n)!!(2n + 1)
n

(2n + 1)!
=
22n 32n n!(2n + 1)
(2n)!
= .
62n n!
We have therefore for the Taylor series of the function around zero,

X (2n + 1)!!
w(z) = zn
n=0
2n 32n n!(2n
+ 1)

X (2n)! n
= z .
62n (n!)2
n=0

(b) Since the function has a single point of singularity located at z = 9,


the convergence disk is a disk of radius 9 centered at the origin. The
intersection of this disk with the real axis is the open interval (−9, 9),
that is, in the case y = 0 the series converges for −9 < x < 9.
(c) If we calculate the function at the point z = 6, which is well within the
real interval of convergence of the series, we get

3
w(6) = √
3

= 3,
SERIES, LIMITS AND CONVERGENCE 183

so that the calculation


√ of the series for z = 6 gives us an algorithmic
way to calculate 3,


X (2n)! n
w(6) = 6 ⇒
6 (n!)2
2n
n=0

√ X (2n)!
3 = .
n=0
6n (n!)2

Since the factorial (2n)! is an integer, the numerator of each term of


the series is an integer. Since both 6n and (n!)2 are also integers, the
denominator is also an integer, and it follows that each term of the series
is a rational number. Since the sum of rational numbers is rational,
given that the rationals form a field, each partial sum of the series is
a rational number. The sequence of partial sums forms, therefore, a
sequence
√ of rational numbers that converges to the irrational number
3.

Problem 7. Consider the function w(z) = 3/ 4 − z, where z = x + ıy is
a complex number, while x and y are real numbers.

(a) Write the Taylor expansion of w(z) around z = 0, that is, the Maclaurin
expansion of the function. Write explicitly the general term of the
series.

(b) Determine and describe the interval of convergence of the series on the
real x axis, that is, for y = 0.

(c) Use this √


series in order to define a method for calculating the irrational
number 3 via a limit involving only rational numbers. Show that each
term of the series is a rational number.

(d) Using this series,


√ identify an infinite sequence of rational numbers that
converges to 3.

Complete Solution:

We will expand in a power series around z0 = 0 the function


3
w(z) = √ .
4−z
184 SOLUTIONS 11

(a) Calculating the first few derivatives we have

3
w0′ (z) = √ ,
4−z
1 3
w1′ (z) = √ ,
2 4 − z3
13 3
w2′ (z) = √ ,
2 2 4 − z5
135 3
w3′ (z) = √ ,
2 2 2 4 − z7
1357 3
w4′ (z) = √ ,
2 2 2 2 4 − z9
13579 3
w5′ (z) = √ ,
2 2 2 2 2 4 − z 11
... ... .

Systematizing the answers, we can write that

(2n − 1)!! 3
wn′ (z) = n √ 2n+1
2 4−z
(2n + 1)!! 3
= √ .
2 (2n + 1) 4 − z 2n+1
n

Applying this at z√0 = 0, and choosing the leaf of the Riemann surface
corresponding to 4 = +2 as the location of the point z0 , we have

(2n + 1)!! 3
wn′ (0) = √
2n (2n + 1) 42n+1
(2n + 1)!! 3
=
2n (2n + 1) 22n+1
3 (2n + 1)!!
=
2 23n (2n + 1)
3 (2n + 1)!!(2n)!!
=
2 23n (2n)!!(2n + 1)
3 (2n + 1)!
=
2 24n n!(2n + 1)
3 (2n)!
= .
2 24n n!
SERIES, LIMITS AND CONVERGENCE 185

We have therefore for the Taylor series of the function around zero,


3 X (2n + 1)!!
w(z) = zn
2 n=0 23n n!(2n + 1)

3 X (2n)! n
= z .
2 24n (n!)2
n=0

(b) Since the function has a single point of singularity located at z = 4,


the convergence disk is a disk of radius 4 centered at the origin. The
intersection of this disk with the real axis is the open interval (−4, 4),
that is, in the case y = 0 the series converges for −4 < x < 4.

(c) If we calculate the function at the point z = 1, which is well within the
real interval of convergence of the series, we get

3
w(1) = √
3

= 3,

so that the calculation


√ of the series for z = 1 gives us an algorithmic
way to calculate 3,


3 X (2n)! n
w(1) = 1 ⇒
2 n=0 24n (n!)2

√ 3 X (2n)!
3 = .
2 24n (n!)2
n=0

Since the factorial (2n)! is an integer, the numerator of each term of


the series is an integer. Since both 24n and (n!)2 are also integers, the
denominator is also an integer, and it follows that each term of the series
is a rational number. Since the sum of rational numbers is rational,
given that the rationals form a field, each partial sum of the series is
a rational number. The sequence of partial sums forms, therefore, a
sequence
√ of rational numbers that converges to the irrational number
3.
186 SOLUTIONS 11

Problem 8. Consider the function w(z) = 5/ 22 + z, where z = x + ıy is
a complex number, while x and y are real numbers.

(a) Calculate the Taylor expansion w(z) around z = 0, that is, the Maclau-
rin expansion of the function. Write explicitly the general term of the
series.

(b) Determine and describe the interval of convergence of the series on the
real x axis, that is, for y = 0.
(c) Use this √
series in order to define a method for calculating the irrational
number 5 via a limit involving only rational numbers. Show that each
term of the series is a rational number.

(d) Using this series,


√ identify an infinite sequence of rational numbers that
converges to 5.

Complete Solution:

We will expand in a power series around z0 = 0 the function


5
w(z) = √
22 + z
5
= √ .
4+z
(a) Calculating the first few derivatives we have

5
w0′ (z) = √ ,
4+z
(−1) 5
w1′ (z) = √ 3,
2 4+z
(−1) (−3) 5
w2′ (z) = √ 5,
2 2 4+z
(−1) (−3) (−5) 5
w3′ (z) = √ 7,
2 2 2 4+z
(−1) (−3) (−5) (−7) 5
w4′ (z) = √ 9,
2 2 2 2 4+z
(−1) (−3) (−5) (−7) (−9) 5
w5′ (z) = √ 11 ,
2 2 2 2 2 4+z
... ... .
SERIES, LIMITS AND CONVERGENCE 187

Systematizing the answers, we can write that

(−1)n (2n − 1)!! 5


wn′ (z) = n √ 2n+1
2 4+z
(−1)n (2n + 1)!! 5
= √ 2n+1 .
2n (2n + 1) 4+z

Applying this at z√0 = 0, and choosing the leaf of the Riemann surface
corresponding to 4 = +2 as the location of the point z0 , we have

(−1)n (2n + 1)!! 5


wn′ (0) = n √ 2n+1
2 (2n + 1) 4
(−1)n (2n + 1)!! 5
=
2n (2n + 1) 22n+1
5 (−1)n (2n + 1)!!
=
2 23n (2n + 1)
5 (−1)n (2n + 1)!!(2n)!!
=
2 23n (2n)!!(2n + 1)
5 (−1)n (2n + 1)!
=
2 24n n!(2n + 1)
5 (−1)n (2n)!
= .
2 24n n!

We have therefore for the Taylor series of the function around zero,


5 X (−1)n (2n + 1)!! n
w(z) = z
2 n=0 23n n!(2n + 1)

5 X (−1)n (2n)! n
= z .
2 24n (n!)2
n=0

(b) Since the function has a single point of singularity located at z = −4,
the convergence disk is a disk of radius 4 centered at the origin. The
intersection of this disk with the real axis is the open interval (−4, 4),
that is, in the case y = 0 the series converges for −4 < x < 4.

(c) If we calculate the function at the point z = 1, which is well within the
real interval of convergence of the series, we get
188 SOLUTIONS 11

5
w(1) = √
5

= 5,

so that the calculation


√ of the series for z = 1 gives us an algorithmic
way to calculate 5,


5 X (−1)n (2n)! n
w(1) = 1 ⇒
2 24n (n!)2
n=0

√ 5 X (−1)n (2n)!
5 = .
2 n=0
24n (n!)2

Since (−1)n and the factorial (2n)! are integers, the numerator of each
term of the series is an integer. Since both 24n and (n!)2 are also
integers, the denominator is also an integer, and it follows that each
term of the series is a rational number. Since the sum of rational
numbers is rational, given that the rationals form a field, each partial
sum of the series is a rational number. The sequence of partial sums
forms, therefore, a√sequence of rational numbers that converges to the
irrational number 5.
Solutions 12

Representation of Functions
by Series

We present here complete and commented solutions to all problems proposed


in Chapter 12 of the text. For reference, the propositions of the problems are
repeated here. The problems are discussed in the order in which they were
proposed within the problem set of that chapter.

Problem 1. Consider the Taylor series of the function w(z) = exp(z)


around z0 = 0.

(a) Differentiate the series term by term and show that it remains invariant.

(b) Integrate the series term by term and show that it remains invariant
except for a complex constant.

Complete Solution:

The Taylor series of w(z) = exp(z) around z0 = 0 is



X 1 n
w(z) = z .
n!
n=0

(a) Differentiating the series term-by-term we have


dw(z) X 1 dz n
=
dz n=0
n! dz

189
190 SOLUTIONS 12


X n n−1
= z
n!
n=1

X 1
= z n−1
(n − 1)!
n=1

X 1 m
= z ,
m=0
m!

where we made the change of variables m = n − 1 in the sum. Thus we


see that we get the same sum with which we began, that is, the series
remains invariant.

(b) Integrating the series indefinitely term-by-term we have

z ∞ z
1
Z Z
′ ′
X n
dz ′ z ′

dz w z =
n!
n=0

X 1
= C0 + z n+1
(n + 1)n!
n=0

X 1
= C0 + z n+1
n=0
(n + 1)!

X 1 m
= C0 + z ,
m!
m=1

where we made the change of variables m = n + 1 in the sum. Thus


we see that, except for the term m = 0, which instead of 1 is now an
arbitrary constant, we get the same sum with which we began, that is,
the series remains invariant.

Problem 2. Consider the Taylor series of the function w(z) = sin(z)


around z0 = 0.

(a) Differentiate the series term by term and show that the resulting series
is that of the function cos(z).

(b) Integrate the series term by term and show that, except for a complex
constant, the resulting series is that of the function − cos(z).
REPRESENTATION OF FUNCTIONS BY SERIES 191

Complete Solution:

The Taylor series of w1 (z) = sin(z) and w0 (z) = cos(z) around z0 = 0 are,
respectively,

X (−1)k
w1 (z) = z 2k+1
(2k + 1)!
k=0

X (−1)k
w0 (z) = z 2k .
(2k)!
k=0

(a) Differentiating the series of w1 (z) term-by-term we have


dw1 (z) X (−1)k dz 2k+1
=
dz (2k + 1)! dz
k=0

X (−1)k (2k + 1) 2k
= z
(2k + 1)!
k=0

X (−1)k
= z 2k ,
(2k)!
k=0

which is directly the series of w0 (z), without any need to change vari-
ables in the sum.

(b) Integrating the series of w1 (z) indefinitely term-by-term we have

z ∞ z
(−1)k
Z Z
X 2k+1
dz ′ w1 z ′ dz ′ z ′

=
(2k + 1)!
k=0

X (−1)k
= C0 + z 2k+2
(2k + 2)(2k + 1)!
k=0

X (−1)k
= C0 + z 2k+2
(2k + 2)!
k=0

X (−1)j−1
= C0 + z 2j
(2j)!
j=1

X (−1)j
= (C0 + 1) − z 2j ,
(2j)!
j=0
192 SOLUTIONS 12

where we made the transformation of variables j = k + 1 in the sum.


Thus we see that, except for an arbitrary constant, we obtain the series
of w0 (z).

Problem 3. Consider the function f (z) = 1/(1 − z).

(a) Write the Maclaurin series of f (z) as the sum of a general term.

(b) Determine the convergence disk of this power series.

(c) Differentiate the power series in order to obtain a representation of the


function 1/(1 − z)2 .

(d) Differentiate the power series once again in order to obtain a represen-
tation of the function 1/(1 − z)3 .

Complete Solution:

Let us consider the function f (z) = 1/(1 − z).

(a) Making a table of the first few derivatives we have

1
f 0′ (z) = ,
(1 − z)
1
f 1′ (z) = ,
(1 − z)2
2
f 2′ (z) = ,
(1 − z)3
3!
f 3′ (z) = ,
(1 − z)4
4!
f 4′ (z) = ,
(1 − z)5
5!
f 5′ (z) = ,
(1 − z)6
... ... .

Systematizing the results, we have

n!
f n′ (z) = ,
(1 − z)n+1
REPRESENTATION OF FUNCTIONS BY SERIES 193

and applying this at z = 0 we have

f n′ (0) = n!,

so that the Maclaurin series of the function is



X
f (z) = zn.
n=0

(b) Since the function has a single point of singularity at z = 1, the con-
vergence disk is centered at zero and has radius 1.
(c) Differentiating the series term-by-term we obtain

df (z) 1
=
dz (1 − z)2

X dz n
=
dz
n=0
X∞
= nz n−1
n=1
X∞
= (m + 1)z m ,
m=0

where we made the transformation of variables m = n − 1. Note that


this series is also convergent strictly within the convergence disk deter-
mined before, although in this case the coefficients of the powers do, in
fact, increase with m.
(d) Differentiating the series term-by-term once again we obtain

d2 f (z) 2
=
dz 2 (1 − z)3

X dz n
= (n + 1)
n=0
dz

X
= (n + 1)nz n−1
n=1
X∞
= (m + 2)(m + 1)z m ,
m=0
194 SOLUTIONS 12

where we again made the transformation of variables m = n − 1. It


follows that we have
1 d2 f (z) 1
=
2 dz 2 (1 − z)3

X (n + 2)(n + 1)
= zn.
2
n=0

Note that this series is also convergent strictly within the convergence
disk determined before, although in this case the coefficients of the
powers grow even faster with m.

Problem 4. Consider the function f (z) = 1/(1 + z).


(a) Write the Maclaurin series of f (z) as the sum of a general term.
(b) Determine the convergence disk of this power series.
(c) Integrate the power series from 0 to z in order to obtain a representation
of the function ln(1 + z).
(d) Compare the result with the Taylor series of ln(1 + z) around the point
z = 0.

Complete Solution:

Let us consider the function f (z) = 1/(1 + z).


(a) Making a table of the first few derivatives we have
1
f 0′ (z) = ,
(1 + z)
−1
f 1′ (z) = ,
(1 + z)2
2
f 2′ (z) = ,
(1 + z)3
−3!
f 3′ (z) = ,
(1 + z)4
4!
f 4′ (z) = ,
(1 + z)5
−5!
f 5′ (z) = ,
(1 + z)6
... ... .
REPRESENTATION OF FUNCTIONS BY SERIES 195

Systematizing the results, we have

(−1)n n!
f n′ (z) = ,
(1 + z)n+1

and applying this at z = 0 we have

f n′ (0) = (−1)n n!,

so that the Maclaurin series of the function is


X
f (z) = (−1)n z n .
n=0

(b) Since the function has a single point of singularity at z = −1, the
convergence disk is centered at zero and has radius 1.

(c) Integrating the series term-by-term from 0 to z we obtain

Z z
dz ′ f z ′

= ln(1 + z)
0
∞ Z z
X
n
n
= (−1) dz ′ z ′
n=0 0
∞ z
X 1 n
n+1
= (−1) z′
n=0
n+1 0

X (−1)n
= z n+1
n+1
n=0

X (−1)m−1 m
= z ,
m=1
m

where we made the transformation of variables m = n + 1.

(d) In order to assemble the Taylor series of g = ln(1 + z) around z0 = 0,


we make a table of the first few derivatives,
196 SOLUTIONS 12

g0′ (z) = ln(1 + z),


1
g1′ (z) = ,
(1 + z)
−1
g2′ (z) = ,
(1 + z)2
2
g3′ (z) = ,
(1 + z)3
−3!
g4′ (z) = ,
(1 + z)4
4!
g5′ (z) = ,
(1 + z)5
−5!
g6′ (z) = ,
(1 + z)6
... ... .

Systematizing the results, we have g0′ (z) = ln(1 + z) and, for n > 0,

(−1)n−1 (n − 1)!
gn′ (z) = ,
(1 + z)n

so that, applying the results at z = 0, we have g 0′ (0) = 0 and, for


n > 0,

gn′ (0) = (−1)n−1 (n − 1)!,

so that the Maclaurin series of the function is



X (−1)n−1
g(z) = zn,
n
n=1

which is, in fact, exactly the same series that we obtained before.

Problem 5. Consider the following differential equation for a real function


f (x), with k 6= 0,

∂2
f (x) − k2 f (x) = 0.
∂x2
Assume that f (x) can be faithfully represented by a power series,
REPRESENTATION OF FUNCTIONS BY SERIES 197


X
f (x) = an xn ,
n=0

substitute this power series in the equation, differentiating it term by term,


and manipulate the summation indices in order to write the equation as a
certain power series equaled to zero. Determine which should be the relations
between the coefficients an of the series above so that it is indeed a solution of
the equation. Pay particular attention to the first two coefficients. Determine
in this way two different functions that are solutions of the equation and
identify these functions.

Complete Solution:

We will solve, using the representation of functions by power series, the equa-
tion
∂2
f (x) − k2 f (x) = 0,
∂x2
where k 6= 0. This technique is known as the Frobenius method, and it aims
at finding one or more solutions of the equation that have the property of
being regular at the origin, that is, they do not have singularities at this
point. Let us assume for f (x) the form of a power series around x = 0,

X
f (x) = an xn ,
n=0

let us assume that this series is convergent in an open neighborhood around


x = 0, and therefore uniformly convergent, so that it can be differentiated
term-by-term. Substituting the series in the equation we have
∞ ∞
∂2 ∂2 X X
f (x) − k2 f (x) = n
an x − k 2
an x n
∂x2 ∂x2 n=0 n=0

X ∞
X
n−2 2
= n(n − 1)an x −k an x n
n=2 n=0
= 0.

In order to be able to compare the powers in the two terms, and write the
resulting equation as a single power series, let us make a transformation in
198 SOLUTIONS 12

the summation variable in the first term, using m = n − 2, that is, n = m + 2,


in order to obtain
∞ ∞
∂2 X X
f (x) − k2 f (x) = (m + 2)(m + 1)am+2 xm − k2 an x n
∂x2
m=0 n=0

X
(n + 2)(n + 1)an+2 − k2 an xn
 
=
n=0
= 0.

In the resulting equation we have a powers series equaled to zero. Since the
series of the identically zero function is the only power series in which all
the coefficients are zero, and since there is a bijection between the functions
and the power series, the series above must be the one associated with the
identically zero function, that is, all its coefficients must be zero. Another
way of saying the same thing is to note that the set of powers is a complete
basis for the space of all functions f (x) which are regular at the origin, so
that the only way in which a linear combination of elements of this basis can
be zero is that all its coefficients be zero. One way or another, it follows that

(n + 2)(n + 1)an+2 − k2 an = 0 ⇒
2
(n + 2)(n + 1)an+2 = k an ⇒
k2
an+2 = an ,
(n + 2)(n + 1)

for all values of n. What this establishes is what is called a recurrence


relation for the coefficients an of the series. Note, however, that the first two
coefficients, a0 and a1 , never appear on the left-hand side of this equation,
and therefore they are not determined by it. It follows therefore that we have
two constants that remain indeterminate, that is, they are arbitrary, as is to
be expected for the general solution of an ordinary differential equation of
the second order. Writing explicitly the first few equations in this recurrence
relation we have

k2
a2 = a0 ,
(2)(1)
k2
a3 = a1 ,
(3)(2)
k2
a4 = a2 ,
(4)(3)
REPRESENTATION OF FUNCTIONS BY SERIES 199

k2
a5 = a3 ,
(5)(4)
k2
a6 = a4 ,
(6)(5)
k2
a7 = a5 ,
(7)(6)
... = ... .

Note that each coefficients is related with the one two steps behind, such
that a0 determines all the coefficients with even n, and a1 all the coefficients
with odd n. We therefore have two separate sequences of equations, each
associated with one of the two arbitrary constants. We can solve these step-
two recursion relations, thus finding all coefficients in terms of a0 and a1 ,
by iterating the relations, that is, by substituting each one in the later ones
until all coefficients have been written in terms of the first. Doing this for
the even case as an example, we have
k2
a2 = a0 ,
(2)(1)
k2
a4 = a2 ,
(4)(3)
k2
a6 = a4 ,
(6)(5)
k2
a8 = a6 ,
(8)(7)
... = ... ⇒
k2
a2 = a0 ,
(2)(1)
k4
a4 = a0 ,
(4)(3)(2)(1)
k4
a6 = a2 ,
(6)(5)(4)(3)
k4
a8 = a4 ,
(8)(7)(6)(5)
... = ... ⇒
k2
a2 = a0 ,
(2)(1)
k4
a4 = a0 ,
(4)(3)(2)(1)
200 SOLUTIONS 12

k6
a6 = a0 ,
(6)(5)(4)(3)(2)(1)
k6
a8 = a2 ,
(8)(7)(6)(5)(4)(3)
... = ... ⇒
k2
a2 = a0 ,
(2)(1)
k4
a4 = a0 ,
(4)(3)(2)(1)
k6
a6 = a0 ,
(6)(5)(4)(3)(2)(1)
k8
a8 = a0 ,
(8)(7)(6)(5)(4)(3)(2)(1)
... = ... .

Therefore, it is easy to induce the general solution, which is, for n = 2j,
k2j
a2j = a0 ,
(2j)!
thereby determining all coefficients with even n in terms of a0 . In an entirely
analogous way, we have for the odd case
k2j
a2j+1 = a1 .
(2j + 1)!
It follows that we obtained in this way two independent solutions which can
now be written explicitly in terms of power series. In order to generate one
of the two solutions, we choose a1 = 0 and keep a0 arbitrary, which makes
all the coefficients with odd n equal to zero, thereby generating a series with
only even powers, and thus an even solution. In order to generate another
solution we choose a0 = 0 and keep a1 arbitrarily, thereby generating both
an odd series and an odd solution,

X k2j
f1 (x) = a0 x2j ,
(2j)!
j=0

X k2j
f2 (x) = a1 x2j+1 ⇒
(2j + 1)!
j=0

X 1
f1 (x) = a0 (kx)2j ,
(2j)!
j=0
REPRESENTATION OF FUNCTIONS BY SERIES 201


a1 X 1
f2 (x) = (kx)2j+1 ,
k (2j + 1)!
j=0

where we can divide by k so long as this constant is not zero, as is explicited


in the proposition of the problem. We now recognize the two resulting power
series as the Taylor series of the functions cosh(kx) and sinh(kx), respectively.
It follows that the general solution of the equation is given by

f (x) = α cosh(kx) + β sinh(kx),

where α and β are two arbitrary real constants. Note that, although this
problem was formulated in the real context, since we are dealing with power
series, all that we did here holds in exactly the same way in the complex
context.

Problem 6. (Challenge Problem) Consider the question of the calcu-


lation of the sucessive derivatives of the product of two analytic functions,
f1 (z) and f2 (z), that is, derivatives of f (z) = f1 (z)f2 (z), where f1 (z) and
f2 (z) are known only in terms of their expressions in series around a reference
point z0 .

(a) Using the Leibniz rule, write the first, second and third derivatives of
f (z) in terms of the derivatives of f1 (z) and f2 (z).

(b) From this kind of experimentation, induce a general formula for the
n-th derivative of f (z), in terms of a sum involving the appropriate
combinatorial factors.

(c) Prove by finite induction the formula that was induced, that is, as-
sume that the formula is true for the case n − 1 and show that, as a
consequence of this, the case n also holds.

(d) Consider now the product f (z) of the two functions, represented by the
product of the Taylor series of f1 (z) and of f2 (z). Collect the terms
with a definite power n of z − z0 and write a general formula for the
coefficient of the power n.

(e) Use the results to prove that the Taylor series of the function f (z) that
is the product of f1 (z) and f2 (z) is obtained as the product of the two
series.
202 SOLUTIONS 12

Complete Solution:

Let us consider that f (z) = f1 (z)f2 (z), where each of these functions may be
represented by a Taylor series around a certain point z0 ,

X f n′ (z0 )
f (z) = (z − z0 )n ,
n=0
n!

X f1n′ (z0 )
f1 (z) = (z − z0 )n ,
n!
n=0

X f2n′ (z0 )
f2 (z) = (z − z0 )n .
n=0
n!

(a) Let us calculate the first few derivatives of the product f (z), using the
Leibniz rule repeatedly. By doing this and using primes to indicate the
derivatives with respect to z, we obtain

f 0′ (z) = f10′ (z)f20′ (z),


f 1′ (z) = f11′ (z)f20′ (z) + f10′ (z)f21′ (z),
f 2′ (z) = f12′ (z)f20′ (z) + 2f11′ (z)f21′ (z) + f10′ (z)f22′ (z),
f 3′ (z) = f13′ (z)f20′ (z) + 2f12′ (z)f21′ (z) + 2f11′ (z)f22′ (z) + f10′ (z)f23′ (z),
f 4′ (z) = f14′ (z)f20′ (z) + 3f13′ (z)f21′ (z) + 6f12′ (z)f22′ (z) +
+3f11′ (z)f23′ (z) + f10′ (z)f24′ (z),
... ... .

As one can see, in general the result is a sum of bilinear terms in


derivatives of f1 (z) and f2 (z), and the orders of the derivatives have a
structure similar to that of the powers in the Newton binomial formula.
The coefficients are also those that appear in the binomial expansion.

(b) Assuming that the structure is in fact the same as that of the formula
of the binomial expansion, we can induce that the general formula is

n
n′
X n! (n−i)′
f (z) = f (z)f2i′ (z).
i!(n − i)! 1
i=0

(c) The formula proposed above reduces, in the case n = 0, simply to


f (z) = f1 (z)f2 (z), and in the case n = 1 to
REPRESENTATION OF FUNCTIONS BY SERIES 203

f ′ (z) = f1′ (z)f2 (z) + f1 (z)f2′ (z),

which we already know to be true. Let us assume now that this formula
applies to the case n − 1,

n−1
X (n − 1)! (n−1−i)′
f (n−1)′ (z) = f (z)f2i′ (z),
i!(n − 1 − i)! 1
i=0

and simply take an additional derivative on each side, using for this the
Leibniz rule once again. Doing this we have

n−1
X (n − 1)! h (n−i)′ (n−1−i)′ (i+1)′
i
f n′ (z) = f1 (z)f2i′ (z) + f1 (z)f2 (z)
i!(n − 1 − i)!
i=0
n−1
X (n − 1)! (n−i)′
= f1 (z)f2i′ (z) +
i!(n − 1 − i)!
i=0
n−1
X (n − 1)! (n−1−i)′ (i+1)′
+ f (z)f2 (z)
i!(n − 1 − i)! 1
i=0
n−1
X (n − 1)! (n−i)′
= f1 (z)f2i′ (z) +
i!(n − 1 − i)!
i=0
n
(n − 1)! (n−j)′
(z)f2j′ (z)
X
+ f
(j − 1)!(n − j)! 1
j=1
n−1
X (n − 1)! (n−i)′
= f1n′ (z)f20′ (z) + f1 (z)f2i′ (z) +
i!(n − 1 − i)!
i=1
n−1
X (n − 1)! (n−i)′
+ f (z)f2i′ (z) + f10′ (z)f2n′ (z)
(i − 1)!(n − i)! 1
i=1
= f1n′ (z)f20′ (z) + f10′ (z)f2n′ (z) +
n−1
X  (n − 1)! 
(n − 1)! (n−i)′
+ + f (z)f2i′ (z)
i!(n − i − 1)! (i − 1)!(n − i)! 1
i=1
= f1 (z)f20′ (z)
n′
+ f10′ (z)f2n′ (z) +
n−1
X 
(n − 1)!(n − i) (n − 1)!i (n−i)′
+ + f (z)f2i′ (z)
i!(n − i)! i!(n − i)! 1
i=1
204 SOLUTIONS 12

= f1n′ (z)f20′ (z) + f10′ (z)f2n′ (z) +


n−1
X (n − 1)!(n − i + i) (n−i)′
+ f1 (z)f2i′ (z)
i!(n − i)!
i=1
n−1
X n! (n−i)′
= f1n′ (z)f20′ (z) + f (z)f2i′ (z) + f10′ (z)f2n′ (z)
i!(n − i)! 1
i=1
n
X n! (n−i)′
= f (z)f2i′ (z),
i!(n − i)! 1
i=0

where we made the transformation of variables j = i + 1, i = j − 1 in


one of the sums, subsequently returning the variable name to i. Thus
we see that the case n − 1 does in fact imply the case n, and therefore
that it is true in general, by finite induction, that
n
X n! (n−i)′
f n′ (z) = f (z)f2i′ (z).
i!(n − i)! 1
i=0

(d) Representing each of the two factor-functions of the product by its


series, we have for the product-function

f (z) = f1 (z)f2 (z)


∞ ∞
X f1n1 ′ (z0 ) X f2n2 ′ (z0 )
= (z − z0 )n1 (z − z0 )n2
n =0
n 1 ! n =0
n 2 !
1 2
∞ X

X f1n1 ′ (z0 )f2n2 ′ (z0 )
= (z − z0 )n1 +n2 .
n1 !n2 !
n1 =0 n2 =0

It is now necessary to collect all the terms in which the total power
n1 + n2 is a constant, n1 + n2 = n, where n is a new variable, which
clearly varies from 0 to ∞, being equal to 0 only if n1 = 0 = n2 . Given
a certain constant value for n, the smallest possible value for n1 is 0, in
which case n2 = n, and the highest possible value for n1 is n, in which
case n2 = 0. Thus, given a constant value for n, there are n + 1 possible
pairs of values for n1 and n2 . We can index these values by the variable
n2 , which we now call i, and which varies from 0 to n. In this case, we
have for n1 the corresponding values n − i. Thus, the sums can now
be described by the pair of variables i and n, instead of n1 and n2 . We
therefore have, changing n2 to i, n1 to n − i and n1 + n2 to n,
REPRESENTATION OF FUNCTIONS BY SERIES 205

∞ X
n (n−i)′
X f (z0 )f i′ (z0 )
f (z) = 1 2
(z − z0 )n .
(n − i)!i!
n=0 i=0

(e) We can write the result of the previous item as follows,


" n #
X 1 n
X n! (n−i)′ i′
f (z) = (z − z0 ) f (z0 )f2 (z0 ) ,
n! i!(n − i)! 1
n=0 i=0

where we multiplied and divided by n! and moved out of the sum on


i some quantities that do not depend on i. We recognize now in the
second sum the n-th derivative of the product-function f (z), as it was
previously derived, applied at the point z0 . We therefore have the
Taylor series of f (z),


X f (n)′ (z0 )
f (z) = (z − z0 )n ,
n!
n=0

which proves that the product of the two Taylor series, that of f1 (z)
and that of f2 (z), in fact produces the Taylor series of f (z), which has
its coefficients related to those of the two factor-series by the Leibniz
rule.

Problem 7. (Challenge Problem) Assuming that f2 (z) does not have


any zeros in the common region of analyticity of f1 (z) and f2 (z), initially
show that if f (z) is the ratio of the two analytic functions f1 (z) and f2 (z),

f1 (z)
f (z) = ,
f2 (z)

then the Taylor series S(z) of f (z) is such that S2 (z)S(z) = S1 (z), where
S1 (z) is the Taylor series of f1 (z) and S2 (z) is the Taylor series of f2 (z), all
with respect to the same reference point z0 . Prove this fact without using
the explicit expression of the coefficients of S(z) in terms of the coefficients
S1 (z) and S2 (z).
Hint: use the result for the product of two analytic functions, which was
proved before.
206 SOLUTIONS 12

Then consider the question of the calculation of the derivatives of f (z) where
f1 (z) and f2 (z) are known only in terms of their expressions in series. The
idea is to try to repeat the inductive-deductive scheme that was used in the
corresponding deduction for the product of two analytic functions, and to use
the result to explicitly show how the Taylor series of f (z) is in fact obtained
from the Taylor series of f1 (z) and f2 (z).
Hint: consider simplifying the problem by first trying to solve the case where
f1 (z) = 1, that is, first try to solve the problem of finding the Taylor series
of the multiplicative inverse of a function,

1
f (z) = ,
f2 (z)

which naturally means that

f (z)f2 (z) = 1.

After that one may consider combining this solution with that of the problem
of the product, which was solved before.
Warning: this problem has a very difficult combinatorial part and, for the
time being, a truly complete solution is not available.

Complete Solution:

(A bit incomplete) Let us now consider the problem of the series in the
case of the division, where we have the series of f1 (z) and f2 (z), and we want
to determine the ratio of the two series, assuming that f2 (z) has no zeros
within the region of interest,

f1 (z)
f (z) = .
f2 (z)

One objective of this problem is to show that the series which results from
the division of the series of f1 (z) and f2 (z) is in fact the Taylor series of
the ratio-function f (z). This is a fact that we can prove based on what we
already know about the case of the multiplication. Let S(z) be the series of
f (z), S1 (z) the series of f1 (z) and S2 (z) the series of f2 (z). Note that, by
the definition of the division, determining the ratio-series

S1 (z)
S(z) =
S2 (z)
REPRESENTATION OF FUNCTIONS BY SERIES 207

means determining, given the series S1 (z) and S2 (z), a series S(z) such that

S(z)S2 (z) = S1 (z),

where the multiplication of the series should be done by collecting powers, as


we did in the case of the multiplication. From the case of the multiplication
we already know that the fact that f (z)f2 (z) = f1 (z) is equivalent to the
fact that S(z)S2 (z) = S1 (z). Hence, we can elaborate a proof by reductio ad
absurdum. If there are functions f (z), f1 (z) and f2 (z), each with its Taylor
series, S(z), S1 (z) and S2 (z), such that

S1 (z)
S(z) 6= ,
S2 (z)

it follows, from the definition of the division, that S(z) does not satisfy the
corresponding product, namely that

S(z)S2 (z) 6= S1 (z).

However, for the corresponding functions it is certainly true that

f1 (z)
f (z) = ⇒
f2 (z)
f (z)f2 (z) = f1 (z),

which, as we know from the case of the multiplication, implies that

S(z)S2 (z) = S1 (z),

contradicting the previous conclusion. It follows that there can be no func-


tions f (z), f1 (z) and f2 (z), each with its Taylor series, S(z), S1 (z) and S2 (z),
such that
f1 (z)
f (z) = ,
f2 (z)

while
S1 (z)
S(z) 6= .
S2 (z)

It follows therefore, by reductio ad absurdum, that it is always true that


208 SOLUTIONS 12

f1 (z)
f (z) = ⇒
f2 (z)
S1 (z)
S(z) = ,
S2 (z)
that is, that the power series obtained by the ratio of two Taylor series is
always the Taylor series of the ratio-function.
The other objective of the problem is the determination of the coefficients
of the ratio-series S(z) in terms of the coefficients of the series S1 (z) and
S2 (z). This is a considerably more complex combinatorial problem than that
of the case of the multiplication. Given the above result, we can determine
these coefficients in two ways: on the one hand, we can simply take directly
multiple derivatives of the function
f1 (z)
f (z) = ,
f2 (z)
subsequently applying the results at z0 in order to obtain the coefficients
of S(z); on the other hand, we can start from S1 (z) and S2 (z), and solve
the problem of dividing one series by the other, thereby determining the
coefficients of S(z). In what follows we will explore this second possibility,
for two reasons: first, because it is the less immediate and familiar alternative;
second, because there is the possibility that it is in fact the simplest one. If
we consider the function h(z) which is the multiplicative inverse of f2 (z), we
can reduce this problem to an earlier problem, that of the multiplication of
two series, since
1
h(z) = ⇒
f2 (z)
f (z) = f1 (z)h(z).
Since this inversion is a particular case of the division, we already know
that the series obtained by the inversion of the series of f2 (z) is in fact the
Taylor series of h(z). In addition, we already know the formula that gives
us the coefficients of the product-series. Let us therefore focus our efforts on
explicitly getting the coefficients of the series of h(z), that is, on the inversion
problem. If we have the series

X hn′ (z0 )
h(z) = (z − z0 )n ,
n!
n=0

X f2n′ (z0 )
f2 (z) = (z − z0 )n ,
n=0
n!
REPRESENTATION OF FUNCTIONS BY SERIES 209

the first of which is our unknown, while the second is known, we can write
the expression h(z)f2 (z) = 1 as follows, using the series to represent the
functions,

1 = h(z)f2 (z)
∞ ∞
X hn0 ′ (z0 ) n0
X f2n2′ (z0 )
= (z − z0 ) (z − z0 )n2
n0 ! n2 !
n0 =0 n2 =0
∞ X ∞
X hn0 ′ (z0 )f2n2 ′ (z0 )
= (z − z0 )n0 +n2 .
n0 !n2 !
n0 =0 n2 =0

Just as we did before in the case of the product of two series, we can now
collect terms with constant n0 + n2 . Thus, the sums can now be described
by the pair of variables i = n2 and n = n0 + n2 , instead of n0 and n2 . We
therefore have, changing n2 to i, n0 to n − i and n0 + n2 to n,
∞ X
n
X h(n−i)′ (z0 )f i′ (z0 )
1 = 2
(z − z0 )n
(n − i)!i!
n=0 i=0

" n #
X 1 X n!
= (z − z0 )n h(n−i)′ (z0 )f2i′ (z0 ) .
n=0
n! i!(n − i)!
i=0

Comparing powers of (z −z0 ) on both sides of this equation, we conclude that


for n = 0 this implies simply that 1 = h(z0 )f2 (z0 ), that is, h(z0 ) = 1/f2 (z0 ).
For all other cases we have that the coefficients of the series on the right-hand
side of the expression above must vanish, that is, we have that
n
X n!
h(n−i)′ (z0 )f2i′ (z0 ) = 0,
i!(n − i)!
i=0

for n ≥ 1. This is an infinite set of equations, determining the derivatives of


h(z) at the point z0 in terms of the derivatives of f2 (z) at the same point.
However, note that the unknowns, that is, the various derivatives of h(z), of
orders 0 to n, applied at the point z0 , are mixed in these equations. In order
to solve these equations for h(n−i)′ (z0 ), it is necessary to proceed iteratively
using, to start the process, the already known case n = 0. It is because of this
fact that this problem is considerably more complex than the corresponding
problem in the case of the multiplication. Writing the first few cases we have

h1′ (z0 )f20′ (z0 ) + h0′ (z0 )f21′ (z0 ) = 0,


h2′ (z0 )f20′ (z0 ) + 2h1′ (z0 )f21′ (z0 ) + h0′ (z0 )f22′ (z0 ) = 0,
210 SOLUTIONS 12

h3′ (z0 )f20′ (z0 ) + 3h2′ (z0 )f21′ (z0 ) + 3h1′ (z0 )f22′ (z0 )+
+h0′ (z0 )f23′ (z0 ) = 0,
h4′ (z0 )f20′ (z0 ) + 4h3′ (z0 )f21′ (z0 ) + 6h2′ (z0 )f22′ (z0 )+
+4h1′ (z0 )f23′ (z0 ) + h0′ (z0 )f24′ (z0 ) = 0,
... ... ,

where from now on, in order to simplify the expressions, we will cease to
indicate explicitly that all functions and derivatives are applied at z0 , making
this fact implicit. Isolating the derivative of h(z) of highest order in each case,
and including again the case n = 0 we have

1
h0′ = ,
f2
1
h1′ = h0′ f21′ ,
 

f2
1
h2′ = 2h1′ f21′ + h0′ f22′ ,
 

f2
1
h3′ = 3h2′ f21′ + 3h1′ f22′ + h0′ f23′ ,
 

f2
1
h4′ = 4h3′ f21′ + 6h2′ f22′ + 4h1′ f23′ + h0′ f24′ ,
 

f2
... ... .

Substituting now the known value h0′ everywhere, we have

1
h0′ = ,
f2
1
h1′ = − 2 f21′ ,
f2
 
1 1 2′
h2′ = − 2h1′ f21′ + f2 ,
f2 f2
 
1 1 3′
h3′ = − 2′ 1′
3h f2 + 3h f2 +1′ 2′
f ,
f2 f2 2
 
1 1 4′
h4′ = − 4h3′ f21′ + 6h2′ f22′ + 4h1′ f23′ + f2 ,
f2 f2
... ... ,

where we now have the value of h1′ determined. Substituting this value
everywhere we have
REPRESENTATION OF FUNCTIONS BY SERIES 211

1
h0′ = ,
f2
1
h1′ = − 2 f21′ ,
f2
1  1′ 2 1
h2′ = 2 3 f2 − 2 f22′ ,
f2 f2
 
1 1 1′ 2′ 1 3′
h3′ = − 2′ 1′
3h f2 − 3 2 f2 f2 + f ,
f2 f2 f2 2
 
1 1 1′ 3′ 1 4′
h4′ = − 3′ 1′ 2′ 2′
4h f2 + 6h f2 − 4 2 f2 f2 + f ,
f2 f2 f2 2
... ... ,

where we now have the value of h2′ determined. Substituting this value
everywhere we have
1
h0′ = ,
f2
1
h1′ = − 2 f21′ ,
f2
1  2 1
h2′ = 2 3 f21′ − 2 f22′ ,
f2 f2
1  3 1 1
h3′ = −6 4 f21′ + 6 3 f22′ f21′ − 2 f23′ ,
f2 f2 f2

1 1 2 1  2
h4′ = 4h3′ f21′ + 12 3 f21′ f22′ − 6 2 f22′ +
 

f2 f2 f2

1 1′ 3′ 1 4′
− 4 2 f2 f2 + f ,
f2 f2 2
... ... ,

where we now have the value of h3′ determined. Substituting this value
everywhere we have
1
h0′ = ,
f2
1
h1′ = − 2 f21′ ,
f2
1  1′ 2 1
h2′ = 2 3 f2 − 2 f22′ ,
f2 f2
1 3 1 1
h3′ = −6 4 f21′ + 6 3 f22′ f21′ − 2 f23′ ,
 
f2 f2 f2
212 SOLUTIONS 12

1  1′ 4 1  1′ 2 2′ 1
h4′ = 24 5 f 2 − 36 4 f2 f2 + 8 3 f21′ f23′ +
f2 f2 f2
1  2′ 2 1
+6 3 f2 − 2 f24′ ,
f2 f2
... ... ,

where we now have the value of h4′ determined. The problem that remains
open is to systematize and generalize this type of procedure to the general
case.
Solutions 13

Convergence Criteria and


Proofs

We present here complete and commented solutions to all problems proposed


in Chapter 13 of the text. For reference, the propositions of the problems are
repeated here. The problems are discussed in the order in which they were
proposed within the problem set of that chapter.

Problem 1. Use the ratio-limit test in order to show that the Maclaurin
series of each of the following functions is convergent, and determine in each
case the radius of the convergence disk.

(a) w(z) = cos(z).

(b) w(z) = sin(z).

(c) w(z) = cosh(z).

(d) w(z) = sinh(z).

(e) w(z) = ln(1 + z).

Complete Solution:

(a) The Maclaurin series of w(z) = cos(z) is given by


X (−1)k
cos(z) = z 2k .
(2k)!
k=0

213
214 SOLUTIONS 13

Taking the ratio of the absolute values of two successive terms we have

|tk+1 | (2k)!|z|2k+2
=
|tk | (2k + 2)!|z|2k
|z|2
= .
(2k + 2)(2k + 1)

Given an arbitrary finite z, this ratio goes to zero in the k → ∞ limit,


and is therefore smaller than 1. It follows that the series converges on
the whole complex plane, and that the radius of the convergence disk
centered at z0 = 0 is infinite.

(b) The Maclaurin series of w(z) = sin(z) is given by



X (−1)k
sin(z) = z 2k+1 .
(2k + 1)!
k=0

Taking the ratio of the absolute values of two successive terms we have

|tk+1 | (2k + 1)!|z|2k+3


=
|tk | (2k + 3)!|z|2k+1
|z|2
= .
(2k + 3)(2k + 2)

Given an arbitrary finite z, this ratio goes to zero in the k → ∞ limit,


and is therefore smaller than 1. It follows that the series converges on
the whole complex plane, and that the radius of the convergence disk
centered at z0 = 0 is infinite.

(c) The Maclaurin series of w(z) = cosh(z) is given by



X 1
cosh(z) = z 2k .
(2k)!
k=0

Taking the ratio of the absolute values of two successive terms we have

|tk+1 | (2k)!|z|2k+2
=
|tk | (2k + 2)!|z|2k
|z|2
= .
(2k + 2)(2k + 1)
CONVERGENCE CRITERIA AND PROOFS 215

Given an arbitrary finite z, this ratio goes to zero in the k → ∞ limit,


and is therefore smaller than 1. It follows that the series converges on
the whole complex plane, and that the radius of the convergence disk
centered at z0 = 0 is infinite.

(d) The Maclaurin series of w(z) = sinh(z) is given by


X 1
sin(z) = z 2k+1 .
(2k + 1)!
k=0

Taking the ratio of the absolute values of two successive terms we have

|tk+1 | (2k + 1)!|z|2k+3


=
|tk | (2k + 3)!|z|2k+1
|z|2
= .
(2k + 3)(2k + 2)

Given an arbitrary finite z, this ratio goes to zero in the k → ∞ limit,


and is therefore smaller than 1. It follows that the series converges on
the whole complex plane, and that the radius of the convergence disk
centered at z0 = 0 is infinite.

(e) The Maclaurin series of w(z) = ln(1 + z) is given by


X (−1)n+1
ln(1 + z) = zn .
n
n=1

Taking the ratio of the absolute values of two successive terms we have

|tk+1 | n|z|n+1
=
|tk | (n + 1)|z|n
n
= |z|.
n+1

The ratio involving n goes to 1 from below in the n → ∞ limit. It fol-


lows that the series converges provided that |z| < 1. We have therefore
that the radius of the convergence disk centered at z0 = 0 is equal to
1, which is consistent with the fact that the function has a singularity
at z = −1.
216 SOLUTIONS 13

Problem 2. Consider the series of positive real numbers



X
S∞ = |an |,
n=0

where an are complex numbers.

(a) Write in symbolic mathematical language, involving a positive real


number ǫ, the convergence condition of the series.

(b) Write in symbolic mathematical language the condition that the limit
of |an | be zero when n goes to infinity.

(c) Show that if the series converges, then we do have this value for the
limit, that is,

lim |an | = 0.
n→∞

Hint: one can try using the method of reductio ad absurdum, but this
time a constructive proof is simpler.

Complete Solution:

(a) If we consider the partial sums of the series,

N
X
SN = |an |,
n=0

then we have that the series is given by the limit

S∞ = lim SN .
N →∞

The convergence condition is therefore the expression of the existence


of this limit, that is, it is the condition that, given any real number
ǫ1 > 0, there is a value N (ǫ1 ) of N such that

N > N (ǫ1 ) =⇒ |S∞ − SN | < ǫ1 .


CONVERGENCE CRITERIA AND PROOFS 217

(b) The condition that |an | should go to zero for large values of n can be
expressed as the limit

lim |an | = 0.
n→∞

The expression of the existence of this limit is the condition that, given
any real number ǫ2 > 0, there is a value n(ǫ2 ) of n such that

n > n(ǫ2 ) =⇒ |an | < ǫ2 .

(c) If we consider that the difference of two successive partial sums of the
series, SN +1 −SN , is exactly the term |aN +1 | of the series, and therefore
is positive, we can write that

|aN +1 | = |SN +1 − SN |
= |(SN +1 − S∞ ) − (SN − S∞ )|
≤ |SN +1 − S∞ | + |SN − S∞ |,

where we used the triangle inequality. Let us now assume that the
series is convergent and show that this implies that the N → ∞ limit
of |aN | is zero. Thus, we are given a strictly positive real number ǫ2 ,
and we need to show that there is an N (ǫ2 ) that satisfies the condition
related to this limit. Let us start by choosing for ǫ1 the value ǫ2 /2.
Since the series converges, for any value of ǫ1 , including this one, there
is an N (ǫ1 ) such that N > N (ǫ1 ) implies that

|SN − S∞ | < ǫ1 .

Since this holds for any N > N (ǫ1 ), it also applies to the value N + 1,
so that we have

|SN − S∞ | < ǫ1 ,
|SN +1 − S∞ | < ǫ1 ,

and adding these two inequalities we therefore obtain the inequality

|SN +1 − S∞ | + |SN − S∞ | < 2ǫ1 .


218 SOLUTIONS 13

As we saw earlier, the left-hand side of this inequality is greater than


or equal to |aN +1 |, while 2ǫ1 = ǫ2 , so that we can write

|aN +1 | ≤ |SN +1 − S∞ | + |SN − S∞ | < ǫ2 .

This is true for any value of N larger than the value N (ǫ1 ) which we
know to exist due to the convergence of the series. Since N > N (ǫ1 )
implies that N + 1 > N (ǫ1 ) + 1, we have that the above inequality is
valid for any value of N that is greater than N (ǫ2 ) = N (ǫ1 ) + 1. In
summary, we have shown that, given a strictly positive real number ǫ2 ,
there is a value of N (ǫ2 ), the value N (ǫ1 ) + 1, such that

N > N (ǫ2 ) =⇒ |aN | < ǫ2 ,

as we wanted to prove. Note that this proof works equally well for the
series without the absolute values, that is, for the series given by the
partial sums

N
X

SN = an .
n=0

This is true because in this case we have that SN ′ ′


+1 − SN = aN +1 ,
and if we take absolute values on both sides we therefore have that

|SN ′
+1 − SN | = |aN +1 |, which is exactly the relation we need to use for
the proof. The rest of the proof proceeds without any modification.

Problem 3. Consider the series of positive real numbers



X
S∞ = |an |,
n=0

where an are complex numbers.

(a) Write in symbolic mathematical language, involving a positive real


number ǫ, the convergence condition of the series.

(b) Show that if the series converges, then there is a real number A such
that |an | < A, for all n.

Hint: use the method of reductio ad absurdum.


CONVERGENCE CRITERIA AND PROOFS 219

Complete Solution:

(a) If we consider the partial sums of the series,

N
X
SN = |an |,
n=0

then we have that the series is given by the limit

S∞ = lim SN .
N →∞

The convergence condition is therefore the expression of the existence


of this limit, that is, it is the condition that, given any real number
ǫ > 0, there is a value N (ǫ) of N such that

N > N (ǫ) =⇒ |S∞ − SN | < ǫ.

(b) Assuming that the series converges, let us consider the negation of the
thesis, that is, let us assume that there is no real number A such that
|an | < A, for all n. It follows that, given an arbitrary strictly positive
real number ǫ, there is at least one term N (ǫ) of the series such that

aN (ǫ) ≥ ǫ,

because otherwise there would be a real number A, the number A = ǫ,


such that |an | < A, for all n. Thus, given the number 1, there is a term
N (1) such that

aN (1) ≥ 1,

given the number 2, there is a term N (2) such that

aN (2) ≥ 2,

and so on for all strictly positive integers n. Thus, there is an infinite


subset of terms of the series that has the property that
220 SOLUTIONS 13

aN (n) ≥ n,

for n ∈ {1, 2, 3, . . .}, and adding the two sides of this inequality for n
going from 1 to N , we have that

N
X N
X
aN (n) ≥ n.
n=1 n=1

The sum on the right-hand side of this inequality is the sum of an


arithmetic progression which has a well-known value,

N
X N (N + 1)
aN (n) ≥ .
n=1
2

Since the limit of this expression for N → ∞ is infinite, we see that


there is a subset of terms of the series whose sum diverges to infinity.
Since all terms of the series are positive, this sum of part of the terms is
certainly less than or equal to the sum of the whole series, and it follows
therefore that the complete series diverges to infinity, which contradicts
the hypothesis. Thus, it cannot be true that there is no real number
A such that |an | < A, for all n, and it thus follows that there is such
a number A. To make this argument a little more precise, given the
partial sum

N
X
aN (n) ,
n=1

which has a finite number of terms, we can consider the maximum


among the integers N (n), for n = 1, . . . , N , which we will denote as
NM . Consider the partial sum SNM of the series. Since it is a sum of
terms in ascending order, it follows that it contains all the terms of the
sum above. Since all terms are positive, it is therefore greater than or
equal to the sum above, that is, we can write that

N
X N (N + 1)
S NM ≥ aN (n) ≥ .
n=1
2
CONVERGENCE CRITERIA AND PROOFS 221

Thus, given an arbitrary integer N , one can always find a partial sum
SNM of the series, where NM is some function of N , such that

N (N + 1)
S NM ≥ .
2

That is, the partial sums of the series grow without limit and there-
fore the series does not converge. Note that because all terms of the
series are positive, it follows that NM increases with N . That is, as
we increase N , the corresponding partial sums SNM are progressively
further along into the sequence of partial sums.

Problem 4. Consider the Maclaurin series of the complex function


p
w(z) = z + ǫ2 ,

where ǫ is a strictly positive real number.

(a) Write the series as an infinite sum of a general term, involving factorials,
double factorials and powers.

(b) Use the ratio-limit test in order to establish the convergence of the
series, and calculate the radius of the convergence disk.

(c) Consider the limit in which ǫ → 0. What is the convergence domain of


the series in this case?

Complete Solution:


(a) Taking the first few successive derivatives of w(z) = z + ǫ2 and ap-
plying at z0 = 0, we obtain

w0′ (0) = +ǫ
11
w1′ (0) = +
2 ǫ
2′ 11 1
w (0) = −
2 2 ǫ3
113 1
w3′ (0) = +
2 2 2 ǫ5
1135 1
w4′ (0) = −
2 2 2 2 ǫ7
222 SOLUTIONS 13

11357 1
w5′ (0) = +
2 2 2 2 2 ǫ9
113579 1
w6′ (0) = −
2 2 2 2 2 2 ǫ11
... ... .

Systematizing these results we can induce the general term of the series,
and write therefore that

X (−1)k+1 (2k + 1)!! zk
w(z) = .
2k (2k + 1)(2k − 1)k! ǫ2k−1
k=0

It is not difficult to verify that this general expression reproduces all


the particular cases listed above.

(b) Calculating the ratio of the absolute values of two successive terms we
have

|tk+1 | 2k (2k + 3)!!(2k + 1)(2k − 1)k! |z|k+1 ǫ2k−1


=
|tk | 2k+1 (2k + 1)!!(2k + 3)(2k + 1)(k + 1)! |z|k ǫ2k+1
(2k + 3)!!(2k − 1)k! |z|
=
2(2k + 3)!!(k + 1)! ǫ2
(2k − 1) |z|
=
2(k + 1) ǫ2
k − 1/2 |z|
= .
k + 1 ǫ2

The k → ∞ limit of the ratio involving k is 1, this limit being reached


from below. It follows that the k → ∞ limit of the ratio of successive
terms is less than 1, so that the series is convergent, provided that
|z| < ǫ2 . Thus, the convergence disk centered at z0 = 0 has radius
equal to ǫ2 .

(c) In the limit ǫ → 0 the convergence disk tends to have zero radius and
hence reduces to a single point, the point z = 0. In this case the series
converges at a single point, its point of reference z0 = 0, despite the
fact that the function has a singularity at that point.
Solutions 14

Laurent Series and Residues

We present here complete and commented solutions to all problems proposed


in Chapter 14 of the text. For reference, the propositions of the problems are
repeated here. The problems are discussed in the order in which they were
proposed within the problem set of that chapter.

Problem 1. Consider the function f (z) = cos(z)/(z − z0 ), for an arbitrary


value of z0 .

(a) Write the Taylor series of cos(z) around z0 .

(b) Write the Laurent series of f (z) around z0 .

(c) Determine the radii of the convergence ring of this series.

(d) Determine the value of the residue of f (z) at the point z0 .

Answer: cos(z0 ).

Complete Solution:

(a) The derivatives of even order of the function cos(z) are the cos(z) itself
with alternating signs, and those of odd order are sin(z), also with
alternating signs. Therefore, we have for the Taylor series of cos(z)
around z0

X (−1)k cos(z0 )
cos(z) = (z − z0 )2k +
(2k)!
k=0

223
224 SOLUTIONS 14


X (−1)k sin(z0 )
− (z − z0 )2k+1
(2k + 1)!
k=0

X (−1)k
= cos(z0 ) (z − z0 )2k +
(2k)!
k=0

X (−1)k
− sin(z0 ) (z − z0 )2k+1 .
(2k + 1)!
k=0

Note that another way to obtain this result is to use the trigonometric
identity relative to the cosine of the sum of two angles,

cos(z) = cos(z0 + z − z0 )
= cos(z0 ) cos(z − z0 ) − sin(z0 ) sin(z − z0 ),

using then the usual Maclaurin series of cos(z) and sin(z), applied at
the point z − z0 .

(b) Using this series expansion in the formula of f (z), and separating the
first two terms of the sums we have


X (−1)k
cos(z0 )
f (z) = − sin(z0 ) + cos(z0 ) (z − z0 )2k−1 +
z − z0 (2k)!
k=1

X (−1)k
− sin(z0 ) (z − z0 )2k .
(2k + 1)!
k=1

This is the Laurent series of the function f (z) around z0 .

(c) Since cos(z) is analytic throughout the complex plane, the only singu-
larity of f (z) is the simple pole at the point z0 . It follows that the
convergence ring of the Laurent series is the whole complex plane ex-
cept for the point z0 . We have therefore that the inner radius of the
convergence ring is zero and that the outer radius is infinite.

(d) The residue of f (z) at the point z0 is the coefficient of the term of the
Laurent series that contains the power (z − z0 )−1 , and hence the value
of the residue is cos(z0 ).
LAURENT SERIES AND RESIDUES 225

Problem 2. Consider the function f (z) = ln(z)/(z − z0 ), for a z0 whose


real part is strictly positive. Consider just the n = 0 leaf of the Riemann
surface of the logarithm.

(a) Write the Taylor series of ln(z) around z0 .

(b) Write the Laurent series of f (z) around z0 .

(c) Determine the radii of the convergence ring of this series.

(d) Determine the value of the residue of f (z) at the point z0 .

Answer: ln(z0 ).

Complete Solution:

(a) Since the derivatives of the logarithm function always have the same
value on all the leaves of the Riemann surface, the leaf only needs to be
chosen in fact for the first term of the series. Since only the n = 0 leaf
should be considered, the first term is simply ln(z0 ). The derivatives
of the function ln(z) are negative powers with alternating signs and
factorials, so that we have for the Taylor series of ln(z) around z0


X (−1)k+1
ln(z) = ln(z0 ) + (z − z0 )k .
k=1
kz0k

Observe how for z0 = 1 this reduces to the usual and more familiar
series of the logarithm.

(b) Using this series expansion in the formula of f (z), and separating the
first two terms of the sum, we have


X (−1)k+1
ln(z0 ) 1
f (z) = + + (z − z0 )k−1 .
z − z0 z0 kz0k
k=2

This is the Laurent series of the function f (z) around z0 .

(c) Since the logarithm function has a single point of singularity at z = 0,


the outer radius of the convergence ring can only be extended to the
value |z0 |. On the other hand, the value of the inner radius is zero.
226 SOLUTIONS 14

Thus, the convergence ring is the disk of radius |z0 | centered at z0 ,


with the exclusion of the center, that is, of the point z0 itself. Note
that the limitation that the real part of z0 be strictly positive is not
really necessary, it just avoids the singular point z = 0 and simplifies
the analysis because it leaves this convergence ring entirely contained
in the n = 0 leaf of the Riemann surface.

(d) The residue of f (z) at the point z0 is the coefficient of the term of the
Laurent series that contains the power (z − z0 )−1 , and hence the value
of the residue is ln(z0 ).

Problem 3. Consider the function f (z) = 1/Pn (z), where Pn (z) is a poly-
nomial of degree n,

Pn (z) = a0 + a1 z + a2 z 2 + . . . + an−1 z n−1 + an z n ,

with complex coefficients a0 , . . . , an , where an 6= 0 and n > 0, and where


the coefficients are such that the polynomial has n distinct complex roots
z1 , . . . , zn .

(a) Write the first three terms of the Laurent series of f (z) around zn .

(b) Determine the radii of the convergence ring of this series.

(c) Determine the value of the residue of f (z) at the point zn .

Answer:

n−1
1 Y 1
.
an zn − zk
k=1

Complete Solution:

We are considering the function f (z) = 1/Pn (z), where Pn (z) is a polynomial
of degree n,

Pn (z) = a0 + a1 z + a2 z 2 + . . . + an−1 z n−1 + an z n ,

with complex coefficients a0 , . . . , an , where an 6= 0 and n > 0, these coeffi-


cients being such that the polynomial has n distinct complex roots z1 , . . . , zn ,
so that it can be factored as
LAURENT SERIES AND RESIDUES 227

n
Y
Pn (z) = an (z − zk ).
k=1

It follows that f (z) has exactly n singularities, which are simple poles at the
positions z1 , . . . , zn .
(a) In order to develop a representation in series of f (z) around zn , we begin
by noting that, since all the roots of Pn (z) are distinct and therefore
f (z) has n distinct simple poles, the function

w(z) = (z − zn )f (z)
z − zn
=
Pn (z)
1
=
Pn−1 (z)

is analytic at zn , and therefore has a Taylor series around that point.


We will use the fact that the polynomial Pn−1 (z) can be written as
n−1
Y
Pn−1 (z) = an (z − zk ).
k=1

Calculating the coefficients of the first three terms of the Taylor series
of w(z) around zn we have
1
w0′ (zn ) = ,
Pn−1 (zn )

−Pn−1 (zn )
w1′ (zn ) = 2 ,
Pn−1 (zn )

2[Pn−1 (zn )]2 − Pn−1 (zn )Pn−1
′′ (z )
n
w2′ (zn ) = 3 (z ) ,
Pn−1 n

so that we have for the three initial terms of the Taylor series of w(z)

1 P ′ (zn )
w(z) = − n−1
2 (z ) (z − zn ) +
Pn−1 (zn ) Pn−1 n

2[Pn−1 (zn )]2 − Pn−1 (zn )Pn−1
′′ (z )
n
+ 3 (z ) (z − zn )2 +
2Pn−1 n
+... .
228 SOLUTIONS 14

It follows that we have for the first three terms of the Laurent series of
f (z)

1 1 P ′ (zn )
f (z) = − n−1
2 (z ) +
Pn−1 (zn ) z − zn Pn−1 n

2[Pn−1 (zn )]2 − Pn−1 (zn )Pn−1
′′ (z )
n
+ 3 (z − zn ) +
2Pn−1 (zn )
+... .

(b) Since all the roots of the polynomial Pn (z) are distinct, and therefore
all the poles of f (z) are simple, there is a non-zero distance |zk − zn |
from the pole zn to each of the poles zk , with k ∈ {1, . . . , n − 1}. We
denote the least of these n − 1 distances by |zm − zn |, where zm is the
position of the corresponding pole. We can extend the convergence
disk of the Taylor series of w(z) around zn only until we reach this
singularity, which is the one nearest to zn . Thus, the convergence disk
of this series has radius |zm − zn |. It follows that the convergence ring
of the Laurent series of f (z) has outer radius given by |zm − zn | and
zero inner radius, consisting of the disk of radius |zm − zn | centered at
zn , with the exclusion of the center, that is, of the point zn itself.

(c) The residue of f (z) at the point zn is the coefficient of the term of the
Laurent series that contains the power (z − zn )−1 , and hence the value
of the residue is given by

n−1
1 1 Y 1
= .
Pn−1 (zn ) an zn − zk
k=1

Consider the function f (z) = sin z 2 /z 2 , for z 6= 0.



Problem 4.

(a) Write the Taylor series of sin z 2 around z = 0.




(b) Write a series based on the series obtained above, which represents the
function f (z).

(c) Determine the convergence ring or disk of this series, noting that the
series of sin(z) converges on the entire complex plane.

(d) Calculate the limit of f (z) when z → 0.


LAURENT SERIES AND RESIDUES 229

(e) How should we define the value of f (z) at z = 0, in such a way that it
becomes analytic at that point?

Complete Solution:

Let us recall that the Taylor series of sin(z) around z = 0 is



X (−1)k
sin(z) = z 2k+1 .
(2k + 1)!
k=0

(a) Simply composing the series of sin(z) with the function z 2 we have the
Taylor series


 X (−1)k
sin z 2 = z 4k+2 .
(2k + 1)!
k=0

(b) Considering now the function f (z) = sin z 2 /z 2 , for z 6= 0, we can




write for it the series



X (−1)k
f (z) = z 4k .
(2k + 1)!
k=0

(c) Note that this series has no negative powers of z, and therefore is a
Taylor series and not a Laurent series. Thus, the series automatically
extends the domain of analyticity of the function, relative to the initial
definition, which cannot be used at z = 0. Since the series of sin(z)
is convergent on the whole complex plane, it follows that this series is
also convergent on the entire complex plane.

(d) If we calculate the limit z → 0 using the series to represent the function,
it immediately follows that only the first term is non-zero, and therefore
that we have

X (−1)k
lim f (z) = lim z 4k
z→0 z→0 (2k + 1)!
k=0

X (−1)k
= lim z 4k
(2k + 1)! z→0
k=0
= 1.
230 SOLUTIONS 14

(e) If we define the function at z = 0 by the continuity criterion, that is,


f (0) = 1, which is the value returned by the Taylor series at that point,
it follows that the function is represented by the series throughout the
complex plane, and since a convergent power series converges to an
analytic function, it follows that the resulting function is analytic on
the whole complex plane.
Solutions 15

Calculation of Integrals by
Residues

We present here complete and commented solutions to all problems proposed


in Chapter 15 of the text. For reference, the propositions of the problems are
repeated here. The problems are discussed in the order in which they were
proposed within the problem set of that chapter.

Problem 1. Calculate by residues the following integral,


Z ∞
1
2 + 2x + 2
dx.
−∞ x

Consider the following steps.

(a) Factor completely the polynomial in the denominator.

(b) Determine how to close the contour and what are the relevant singu-
larities.

(c) Show that the integral over the additional part of the contour, used to
close it, vanishes.

(d) Calculate the relevant residues and use the residue theorem in order to
find the value of the integral.

Answer: π.

Complete Solution:

231
232 SOLUTIONS 15

We must calculate the asymptotic integral


Z ∞
1
I= 2
dx.
−∞ x + 2x + 2

(a) We use the Baskara formula in order to find the roots of the polynomial,

−2 ± 4 − 8
x± =
2√
−2 ± ı 4
=
2
= −1 ± ı,

so that using a complex variable z in place of x, the polynomial can be


factored as

z 2 + 2z + 2 = [z − (−1 + ı)][z − (−1 − ı)],

and therefore the integral can be written as


Z ∞
1
I= dz,
−∞ [z − (−1 + ı)][z − (−1 − ı)]

which still extends over the real axis in the complex z plane.
(b) We can close the contour using an arc that goes to infinity, both in the
upper half-plane and in the lower half-plane. If we use the arc in the
upper half-plane, the relevant singularity is z = −1 + ı. If we use the
arc in the lower half-plane, the relevant singularity is z = −1 − ı. In
what follows we will use the first alternative.
(c) The additional integral IR is defined on a semi-circle C of radius R,
with θ going from 0 to π, in the limit in which we make R → ∞. Using
z = R exp(ıθ) and thus dz = ız dθ on the arc, we have for this integral

1
Z
IR = dz 2
z + 2z + 2
ZCπ
1
= dθ ız 2
z + 2z + 2
Z0 π
1
= dθ ıR eıθ 2 ı2θ
0 R e + 2R eıθ + 2
Z π
ı 1
= dθ eıθ ı2θ .
R 0 e + 2 eıθ /R + 2/R2
CALCULATION OF INTEGRALS BY RESIDUES 233

We see that in the R → ∞ limit the integrand is simplified, so that we


have

ı π 1
Z
lim IR = lim dθ eıθ ı2θ
R→∞ R→∞ R 0 e + 2 eıθ /R + 2/R2
ı π 1
Z
= lim dθ eıθ ı2θ
R→∞ R 0 e
ı π 1
Z
= lim dθ ıθ .
R→∞ R 0 e

Since the exponential with imaginary argument has unit absolute value,
the remaining angular integral is the integral of a limited function over
a finite domain, and therefore is finite and independent of R. It follows
that the limit vanishes due to the factor of R in the denominator, and
therefore we have

lim IR = 0.
R→∞

Another way to see this, which is often useful, is to calculate the abso-
lute value of the integral in this limit,
Z π
1 1
lim |IR | = lim dθ eıθ ı2θ
R→∞ R→∞ R e + 2 e /R + 2/R2
ıθ
Z 0π
1 1
≤ lim dθ eıθ
R→∞ R 0 | eı2θ + 2 eıθ /R + 2/R2 |
1 π 1
Z
= lim dθ ı2θ
R→∞ R 0 |e |
1 π
Z
= lim dθ
R→∞ R 0
π
= lim
R→∞ R
= 0.

Since the absolute value of IR is positive and is bounded from above by


zero in the limit, it follows that the limit of the absolute value is zero.
If the absolute value tends to zero, it is necessary that the number also
tend to zero, so that once again we end up with

lim IR = 0.
R→∞
234 SOLUTIONS 15

(d) The contour is being traversed in the positive direction, and the residue
ξ⊕ of the relevant pole can be calculated by the limit

z − (−1 + ı)
ξ⊕ = lim
[z − (−1 + ı)][z − (−1 − ı)]
z→(−1+ı)
1
= lim
z→(−1+ı) z − (−1 − ı)
1
=
−1 + ı + 1 + ı
1
= .

It follows that we have for the integral

I = 2πıξ⊕
= π.

Problem 2. Consider the following integral, to be calculated by residues,


Z ∞
x2
dx.
0 x4 + 5x2 + 4
Consider the following steps.
(a) Discover how to extend the integral to the interval (−∞, ∞).
(b) Factor completely the polynomial in the denominator.
(c) Determine how to close the contour in the complex plane and what are
the relevant singularities.
(d) Show that the integral over the additional part of the contour, used to
close it, vanishes.
(e) Calculate the relevant residues and use the residue theorem in order to
find the value of the integral.

Answer: π/6.

Complete Solution:

We must calculate the asymptotic integral


Z ∞
x2
I= dx.
0 x4 + 5x2 + 4
CALCULATION OF INTEGRALS BY RESIDUES 235

(a) Since both the numerator and the denominator are even functions of
x, the integrand is even, and therefore if we extend the integral to the
whole real axis, then we will have exactly doubled the result. Therefore,
we can write for the integral


1 x2
Z
I= dx.
2 −∞ x4 + 5x2 + 4

(b) We use the Baskara formula in order to find the roots of the quadratic
polynomial in x2 ,


−5 ± 25 − 16
x2± =
√2
−5 ± 9
=
2
−5 ± 3
= .
2

As one can see, the two results are negative, because we have the pos-
sibilities −8/2 = −4 and −2/2 = −1. It follows that the roots for x are
all imaginary, x = ±2ı and x = ±ı, so that using a complex variable z
in place of x, the polynomial can be factored as

z 4 + 5z 2 + 4 = (z − 2ı)(z + 2ı)(z − ı)(z + ı),

so that the integral can be written as


1 z2
Z
I= dz,
2 −∞ (z − 2ı)(z + 2ı)(z − ı)(z + ı)

which still extends over the real axis in the complex z plane.

(c) We can close the contour using an arc that goes to infinity, both in the
upper half-plane and in the lower half-plane. If we use the arc in the
upper half-plane, the relevant singularities are z = ı and z = 2ı. If we
use the arc in the lower half-plane, the relevant singularities are z = −ı
and z = −2ı. In what follows we will use the first alternative.
236 SOLUTIONS 15

(d) The additional integral IR is defined on a semi-circle of radius R, with


θ going from 0 to π, in the limit in which we make R → ∞. Using
z = R exp(ıθ) and thus dz = ız dθ on the arc, we have for this integral

π
z2
Z
IR = dz
0 z 4 + 5z + 4
Z π
z2
= dθ ız 4
0 z + 5z + 4
Z π
R2 eı2θ
= dθ ıR eıθ 4 ı4θ
0 R e + 5R2 eı2θ + 4
Z π
ı eı3θ
= dθ ı4θ .
R 0 e + 5 eı2θ /R2 + 4/R4

We see that in the limit R → ∞ the integrand is simplified, so that


calculating the absolute value of the integral in this limit, we have

π
1 eı3θ
Z
lim |IR | = lim dθ .
R→∞ R→∞ R 0 eı4θ + 5 eı2θ /R2 + 4/R4
π eı3θ
1
Z
≤ lim dθ ı4θ
R→∞ R 0 | e + 5 eıθ /R2 + 4/R4 |
Z π
1 1
= lim dθ ı4θ
R→∞ R 0 |e |
1 π
Z
= lim dθ
R→∞ R 0
π
= lim
R→∞ R
= 0.

Since the absolute value of IR is positive and is bounded from above by


zero in the limit, it follows that the limit of the absolute value is zero.
If the absolute value tends to zero, it is necessary that the number also
tend to zero, so that we have

lim IR = 0.
R→∞

(e) The contour is being traversed in the positive direction, and the residues
ξ1 and ξ2 of the relevant poles of the integrand can be calculated by
means of the limits
CALCULATION OF INTEGRALS BY RESIDUES 237

z 2 (z − ı)
ξ1 = lim
z→ı (z − 2ı)(z + 2ı)(z − ı)(z + ı)

z2
= lim
z→ı (z − 2ı)(z + 2ı)(z + ı)
−1
=
(−ı)(3ı)(2ı)
ı
= ,
6
z 2 (z − 2ı)
ξ2 = lim
z→2ı (z − 2ı)(z + 2ı)(z − ı)(z + ı)

z2
= lim
z→2ı (z + 2ı)(z − ı)(z + ı)
−4
=
(4ı)(ı)(3ı)
−ı
= .
3

It follows that we have for the integral

1
I = (2πı)(ξ1 + ξ2 )
2 
ı ı
= πı −
6 3 
1 2
= −π −
6 6
π
= .
6

Problem 3. Consider the following integral, to be calculated by residues,


Z ∞
x2
4 2
dx.
−∞ x + 3x + 2

Consider the following steps.

(a) Factor completely the polynomial in the denominator.

(b) Determine how to close the contour in the complex plane and what are
the relevant singularities.

(c) Show that the integral over the additional part of the contour, used to
close it, vanishes.
238 SOLUTIONS 15

(d) Calculate the relevant residues and use the residue theorem in order to
find the value of the integral.
√ 
Answer: π 2 − 1 .

Complete Solution:

We must calculate the asymptotic integral


Z ∞
x2
I= 4 2
dx.
−∞ x + 3x + 2

(a) We use the Baskara formula in order to find the roots of the quadratic
polynomial in x2 ,

−3 ± 9−8
x2± =
2

−3 ± 1
=
2
−3 ± 1
= .
2

As one can see, the two results are negative, because we have the pos-
sibilities −4/2 = −2 √and −2/2 = −1. It follows that the roots for x are
all imaginary, x = ± 2ı and x = ±ı, so that using a complex variable
z in place of x, the polynomial can be factored as
 √  √ 
z 4 + 3z 2 + 2 = z − 2ı z + 2ı (z − ı)(z + ı),

so that the integral can be written as



z2
Z
I= √  √  dz,
−∞ z − 2ı z + 2ı (z − ı)(z + ı)

which still extends over the real axis in the complex z plane.

(b) We can close the contour using an arc that goes to infinity, both in
the upper half-plane and in the lower half-plane. If we use the arc√ in
the upper half-plane, the relevant singularities are z = ı and z = 2ı.
If we use the arc in
√ the lower half-plane, the relevant singularities are
z = −ı and z = − 2ı. In what follows we will use the first alternative.
CALCULATION OF INTEGRALS BY RESIDUES 239

(c) The additional integral IR is defined on a semicircle of radius R, with


θ going from 0 to π, in the limit in which we make R → ∞. Using
z = R exp(ıθ) and thus dz = ız dθ on the arc, we have for this integral

π
z2
Z
IR = dz
0 z 4 + 3z + 2
Z π
z2
= dθ ız 4
0 z + 3z + 2
Z π
R2 eı2θ
= dθ ıR eıθ 4 ı4θ
0 R e + 3R2 eı2θ + 2
Z π
ı eı3θ
= dθ ı4θ .
R 0 e + 3 eı2θ /R2 + 2/R4

We see that in the limit R → ∞ the integrand is simplified, so that


calculating the absolute value of the integral in this limit, we have

π
1 eı3θ
Z
lim |IR | = lim dθ .
R→∞ R→∞ R 0 eı4θ + 3 eı2θ /R2 + 2/R4
π eı3θ
1
Z
≤ lim dθ ı4θ
R→∞ R 0 | e + 3 eıθ /R2 + 2/R4 |
Z π
1 1
= lim dθ ı4θ
R→∞ R 0 |e |
1 π
Z
= lim dθ
R→∞ R 0
π
= lim
R→∞ R
= 0.

Since the absolute value of IR is positive and is bounded from above by


zero in the limit, it follows that the limit of the absolute value is zero.
If the absolute value tends to zero, it is necessary that the number also
tend to zero, so that we have

lim IR = 0.
R→∞

(d) The contour is being traversed in the positive direction, and the residues
ξ1 and ξ2 of the relevant poles of the integrand can be calculated by
means of the limits
240 SOLUTIONS 15

z 2 (z − ı)
ξ1 = lim √  √ 
z→ı z − 2ı z + 2ı (z − ı)(z + ı)

z2
= lim √  √ 
z→ı z − 2ı z + 2ı (z + ı)

−1
= √  √ 
1 − 2 ı 1 + 2 ı(2ı)
−ı
=  √ 2
12 − 2 2
ı
=
(2 − 1)2
ı
= ,
2 √ 
z 2 z − 2ı
ξ2 = lim
√ √  √ 
z→ 2ı z − 2ı z + 2ı (z − ı)(z + ı)

z2
= lim
√ √ 
z→ 2ı z + 2ı (z − ı)(z + ı)
−2
= √  √  √ 
2 2 ı 2−1 ı 2+1 ı
−ı
= √ √ 2 
2 2 − 12

−ı 2
=
2(2 − 1)

−ı 2
= .
2

It follows that we have for the integral

I = (2πı)(ξ1 + ξ2 )
√ !
ı 2ı
= 2πı −
2 2
 √ 
= −π 1 − 2
√ 
= π 2−1 .

Problem 4. Calculate by residues the following real asymptotic integral,


CALCULATION OF INTEGRALS BY RESIDUES 241


sin(x)
Z
dx.
−∞ x2 + 4x + 5

Consider the following steps.

(a) Write sin(x) in terms of the complex exponentials exp(±ıx). Note that
we will have two integrals to calculate.

(b) Factor completely the polynomial in the denominator.

(c) Determine how to close the contour in each case, and what are the
relevant singularities.

(d) Show that the integral over the additional parts of the contours, used
to close them, vanish.

(e) Calculate the relevant residues and use the residue theorem in order to
find the value of the original integral.

Answer: −π sin(2)/e.

Complete Solution:

We must calculate the asymptotic integral


Z ∞
sin(x)
I= 2
dx.
−∞ x + 4x + 5

(a) Writing the integral in terms of a complex variable z, and decomposing


the sine into complex exponentials, we have


sin(z)
Z
I = dz
−∞ z2
+ 4z + 5
Z ∞
1 eız − e−ız
= 2
dz
−∞ 2ı z + 4z + 5
1 ∞ eız 1 ∞ e−ız
Z Z
= dz − dz,
2ı −∞ z 2 + 4z + 5 2ı −∞ z 2 + 4z + 5

where the integrals still extend over the real axis in the complex z plane.
242 SOLUTIONS 15

(b) We use the Baskara formula in order to find the roots of the polynomial,

−4 ± 16 − 20
x± =
√2
−4 ± ı 4
=
2
= −2 ± ı,

so that, using the complex variable z in place of x, the polynomial can


be factored as

z 2 + 4z + 5 = [z − (−2 + ı)][z − (−2 − ı)],

and therefore the integral can be written as

1 1
I = I⊕ − I⊖
2ı Z 2ı
1 ∞ eız
= dz +
2ı −∞ [z − (−2 + ı)][z − (−2 − ı)]
1 ∞ e−ız
Z
− dz,
2ı −∞ [z − (−2 + ı)][z − (−2 − ı)]

where the two integrals still run over the real axis in the complex z
plane.

(c) Each of the two integrals has to be closed in a different way. Since we
have that z = x+ıy, it follows that for the first integral the exponential
appearing in the numerator is

eız = eı(x+ıy)
= eıx e−y ,

wherein the first exponential, with an imaginary argument, is a limited


function, and the second goes to zero if y → ∞, but diverges if y → −∞.
Thus we see that this first integral must be closed by an arc in the upper
half-plane, where y > 0, and that in this case the relevant singularity
is at z = −2 + ı. For the other integral the situation is reversed due
to the inversion of the sign in the argument of the exponential, and
therefore it will have to be closed by the lower half-plane, and the
relevant singularity is at z = −2 − ı.
CALCULATION OF INTEGRALS BY RESIDUES 243

(d) Let us start with the first of the two integrals. The additional integral
I⊕R is defined on a semicircle of radius R, with θ going from 0 to π,
in the limit in which we make R → ∞. Using z = R exp(ıθ) and thus
dz = ız dθ on the arc, we have for this integral
π
eız
Z
I⊕R = dz
z 2 + 4z + 5
Z0 π
eız
= dθ ız 2
0 z + 4z + 5
Z π
eıR[cos(θ)+ı sin(θ)]
= dθ ıR eıθ 2 ı2θ
0 R e + 4R eıθ + 5
Z π
ı eıR cos(θ) e−R sin(θ)
= dθ eıθ ı2θ .
R 0 e + 4 eıθ /R + 5/R2

We see that in the limit R → ∞ the integrand is simplified, so that we


have, calculating the absolute value of the integral in this limit,

π
1 eıR cos(θ) e−R sin(θ)
Z
lim |I⊕R | = lim dθ eıθ
R→∞ R→∞ R 0 eı2θ + 4 eıθ /R + 5/R2
1 π eıR cos(θ) e−R sin(θ)
Z
≤ lim dθ eıθ
R→∞ R 0 | eı2θ + 4 eıθ /R + 5/R2 |
1 π e−R sin(θ)
Z
= lim dθ
R→∞ R 0 | eı2θ |
Z π
1
= lim dθ e−R sin(θ)
R→∞ R 0
1 π
Z
≤ lim dθ
R→∞ R 0
π
= lim
R→∞ R
= 0.

In these passages we replaced sin(θ) by its minimum value within the


integration interval, [0, π], in order to maximize the real exponential.
Note that in this interval the sine is always positive, and that its min-
imum value is zero. Since the absolute value of I⊕R is positive and
is bounded from above by zero in the limit, it follows that the limit
of the absolute value is zero. If the absolute value tends to zero, it is
necessary that the number also tend to zero, so that we have
244 SOLUTIONS 15

lim I⊕R = 0.
R→∞

For the second integral everything takes place analogously but with θ
ranging from π to 2π, so that we have


e−ız
Z
I⊖R = dz
π z 2 + 4z + 5
Z 2π
e−ız
= dθ ız 2
π z + 4z + 5
Z 2π
e−ıR[cos(θ)+ı sin(θ)]
= dθ ıR eıθ 2 ı2θ
π R e + 4R eıθ + 5
Z 2π
ı e−ıR cos(θ) eR sin(θ)
= dθ eıθ ı2θ .
R π e + 4 eıθ /R + 5/R2

Considering now the R → ∞ limit of the absolute value of the integral,


1 e−ıR cos(θ) eR sin(θ)
Z
lim |I⊖R | = lim dθ eıθ
R→∞ R→∞ R π eı2θ + 4 eıθ /R + 5/R2
1 2π e−ıR cos(θ) eR sin(θ)
Z
≤ lim dθ eıθ
R→∞ R π | eı2θ + 4 eıθ /R + 5/R2 |
1 2π eR sin(θ)
Z
= lim dθ
R→∞ R π | eı2θ |
Z 2π
1
= lim dθ eR sin(θ)
R→∞ R π

1 2π
Z
≤ lim dθ
R→∞ R π
π
= lim
R→∞ R
= 0.

In these passages we replaced sin(θ) by its maximum value within the


integration interval, [π, 2π], in order to maximize the real exponential.
Note that in this interval the sine is always negative, and its maximum
value is zero. Since the absolute value of I⊖R is positive and is bounded
from above by zero in the limit, it follows that the limit of the absolute
CALCULATION OF INTEGRALS BY RESIDUES 245

value is zero. If the absolute value tends to zero, it is necessary that


the number also tend to zero, so that we have

lim I⊖R = 0.
R→∞

(e) The contour of the integral I⊕ is being traversed in the positive di-
rection, and the residue ξ⊕ of the relevant pole in this case can be
calculated by the limit

[z − (−2 + ı)] eız


ξ⊕ = lim
z→(−2+ı) [z − (−2 + ı)][z − (−2 − ı)]
eız
= lim
z→(−2+ı) z − (−2 − ı)

eı(−2+ı)
=
−2 + ı + 2 + ı
e−2ı−1
= .

But the contour of the integral I⊖ is being traversed in the negative


direction, and the residue ξ⊖ of the relevant pole in this case can be
calculated by the limit

[z − (−2 − ı)] e−ız


ξ⊖ = lim
z→(−2−ı) [z − (−2 + ı)][z − (−2 − ı)]
e−ız
= lim
z→(−2−ı) z − (−2 + ı)

e−ı(−2−ı)
=
−2 − ı + 2 − ı
− e2ı−1
= .

It follows that we have for the original integral

1 1
I = I⊕ − I⊖
2ı 2ı
1 1
= 2πıξ⊕ − (−2)πıξ⊖
2ı 2ı
e−2ı−1 − e2ı−1
= π +π
2ı 2ı
246 SOLUTIONS 15

π e2ı − e−2ı
= −
e 2ı
π
= − sin(2).
e
Problem 5. Calculate by residues the following integral, where a > 0,
Z ∞
cos(ax)
dx.
0 x2 + 1
Consider the following steps.

(a) Discover how to extend the integral to the interval (−∞, ∞).
(b) Consider another integral similar to this extended integral, but with
cos(ax) replaced by sin(ax). What is its value?
(c) Combine the two integrals in order to write a third one, whose calcu-
lation is simpler.
(d) Determine how to close the contour and what are the relevant singu-
larities. Consider the relevance of the sign of a.
(e) Show that the integral over the additional part of the contour, used to
close it, vanishes.
(f) Calculate the relevant residues and use the residue theorem in order to
find the value of the integral.

Answer: π exp(−|a|)/2.

Complete Solution:

We must calculate the asymptotic integral


Z ∞
cos(ax)
I= dx.
0 x2 + 1
(a) Since both the numerator and the denominator are even functions of
x, the integrand is even, and therefore if we extend the integral to the
whole real axis, then we will have exactly twice the values as the result.
Thus we can write for the integral

1 cos(ax)
Z
I= dx.
2 −∞ x2 + 1
CALCULATION OF INTEGRALS BY RESIDUES 247

(b) Replacing the cosine by a sine in the integral above,


1 sin(ax)
Z
I′ = dx,
2 −∞ x2 + 1

we have the integral of an odd function over a symmetric domain, and


the result is therefore zero, that is, I ′ = 0.

(c) Since we have I ′ = 0, we can write that I = I +ıI ′ , and we can therefore
write for I

1 ∞ cos(ax) 1 ∞ ı sin(ax)
Z Z
I = dx + dx
2 −∞ x2 + 1 2 −∞ x2 + 1
1 ∞ cos(ax) + ı sin(ax)
Z
= dx
2 −∞ x2 + 1
1 ∞ eıax
Z
= dx.
2 −∞ x2 + 1

Writing the integral in terms of a complex variable z we have

1 ∞ eıaz
Z
I = dz
2 −∞ z 2 + 1
1 ∞ eıaz
Z
= dz,
2 −∞ (z − ı)(z + ı)

where the integral still extends over the real axis in the complex z
plane, and where we factored the polynomial in the denominator, using
its roots z = ±ı.

(d) The integral has to be closed from above or from below, depending on
the sign of a. Since we have z = x + ıy, it follows that the exponential
appearing in the numerator is

eıaz = eı(ax+ıay)
= eıax e−ay ,

where the first exponential, with imaginary argument, is a limited func-


tion when y → ±∞. However, the behavior of the second exponential
depends on the sign of a. On the one hand, when a > 0 it goes to zero
if y → ∞, but diverges if y → −∞. On the other hand, if a < 0 the
248 SOLUTIONS 15

situation is reversed. Thus, we see that for a > 0 the integral must be
closed by an arc in the upper half-plane, where y > 0, and in this case
the relevant singularity is at z = ı. As for the case a < 0, it will have to
be closed by an arc in the lower half-plane and the relevant singularity
is at z = −ı. Note that for a = 0 one can close the contour in any of
these two ways, and therefore it is not necessary to consider this case
separately. It can be included in either of the two other cases, so that
we can consider as our two alternatives a ≥ 0 and a ≤ 0.
(e) Let us start with the first of the two integrals, that is, the case a ≥ 0.
The additional integral I⊕R is defined on a semicircle of radius R, with
θ going from 0 to π, in the limit in which we make R → ∞. Using
z = R exp(ıθ) and thus dz = ız dθ on the arc, we have for this integral
Z π
eıaz
I⊕R = dz 2
z +1
Z0 π
eıaz
= dθ ız 2
0 z +1
Z π
eıaR[cos(θ)+ı sin(θ)]
= dθ ıR eıθ
0 R2 eı2θ + 1
Z π
ı eıaR cos(θ) e−aR sin(θ)
= dθ eıθ .
R 0 eı2θ + 1/R2

We see that in the R → ∞ limit the integrand is simplified, so that we


have, calculating the absolute value of the integral in this limit,

π
1 eıaR cos(θ) e−aR sin(θ)
Z
lim |I⊕R | = lim dθ eıθ
R→∞ R→∞ R 0 eı2θ + 1/R2
1 π eıaR cos(θ) e−aR sin(θ)
Z
≤ lim dθ eıθ
R→∞ R 0 | eı2θ + 1/R2 |
1 π e−aR sin(θ)
Z
= lim dθ
R→∞ R 0 | eı2θ |
Z π
1
= lim dθ e−aR sin(θ)
R→∞ R 0
1 π
Z
≤ lim dθ
R→∞ R 0
π
= lim
R→∞ R
= 0.
CALCULATION OF INTEGRALS BY RESIDUES 249

In these passages we replaced sin(θ) by its minimum value within the


integration interval, [0, π], in order to maximize the real exponential.
Note that in this interval the sine is always positive, and that its min-
imum value is zero. Since the absolute value of I⊕R is positive and
is bounded from above by zero in the limit, it follows that the limit
of the absolute value is zero. If the absolute value tends to zero, it is
necessary that the number also tend to zero, so that we have

lim I⊕R = 0.
R→∞

For the other case, that is, for a ≤ 0, everything takes place similarly,
but with θ ranging from π to 2π, so that we have


eıaz
Z
I⊖R = dz
π z2 + 1
Z 2π
eıaz
= dθ ız 2
π z +1
Z 2π
eıaR[cos(θ)+ı sin(θ)]
= dθ ıR eıθ
π R2 eı2θ + 1
Z 2π
ı eıaR cos(θ) e−aR sin(θ)
= dθ eıθ .
R π eı2θ + 1/R2

Considering now the R → ∞ limit of the absolute value of the integral,


1 eıaR cos(θ) e−aR sin(θ)
Z
lim |I⊖R | = lim dθ eıθ
R→∞ R→∞ R π eı2θ + 1/R2
1 2π eıaR cos(θ) e−aR sin(θ)
Z
≤ lim dθ eıθ
R→∞ R π | eı2θ + 1/R2 |
1 2π e−aR sin(θ)
Z
= lim dθ
R→∞ R π | eı2θ |
Z 2π
1
= lim dθ e−aR sin(θ)
R→∞ R π

1 2π
Z
≤ lim dθ
R→∞ R π
π
= lim
R→∞ R
= 0.
250 SOLUTIONS 15

In these passages we replace sin(θ) by its maximum value within the


integration interval, [π, 2π], in order to maximize the real exponential.
Note that in this interval the sine is always negative, and that its max-
imum value is zero. Since the absolute value of I⊖R is positive and
is bounded from above by zero in the limit, it follows that the limit
of the absolute value is zero. If the absolute value tends to zero, it is
necessary that the number also tend to zero, so that we have

lim I⊖R = 0.
R→∞

(f) When a ≥ 0 the integration contour is being traversed in the positive


direction, and the residue ξ⊕ of the relevant pole in this case can be
calculated by the limit

(z − ı) eıaz
ξ⊕ = lim
z→ı (z − ı)(z + ı)
eıaz
= lim
z→ı z + ı
eıaı
=
ı+ı
e−a
= .

However, when a ≤ 0 the integration contour is being traversed in the


negative direction, and the residue ξ⊖ of the relevant pole in this case
can be calculated by the limit

(z + ı) eıaz
ξ⊖ = lim
z→−ı (z − ı)(z + ı)
eıaz
= lim
z→−ı z − ı
e−ıaı
=
−ı − ı
ea
= − .

It follows that we have for the original integral in the case a ≥ 0,


CALCULATION OF INTEGRALS BY RESIDUES 251

1
I = (2πı)ξ⊕
2 
e−a

= (πı)

e−a
= π ,
2

while in the case a ≤ 0 we have

1
I = (−2πı)ξ⊖
2  a
e
= (−πı) −

ea
= π .
2
Note that the two results in fact coincide in the case a = 0. We can
write the result for any value of a, in terms of |a|, thus obtaining

π e−|a|
I= .
2

Problem 6. Calculate by residues the following real asymptotic integral,


Z ∞
cos(x)
2
dx.
−∞ x + 4x + 5

Consider the following steps.

(a) Write cos(x) in terms of the complex exponentials exp(±ıx). Note that
we will have two integrals to calculate.

(b) Factor completely the polynomial in the denominator.

(c) Determine how to close the contour in each case, and what are the
relevant singularities.

(d) Show that the integrals over the additional parts of the contours, used
to close them, vanish.

(e) Calculate the relevant residues and use the residue theorem in order to
find the value of the original integral.

Answer: π cos(2)/e.
252 SOLUTIONS 15

Complete Solution:

We must calculate the asymptotic integral


Z ∞
cos(x)
I= 2
dx.
−∞ x + 4x + 5

(a) Writing the integral in terms of a complex variable z, and decomposing


the cosine into complex exponentials, we have

cos(z)
Z
I = dz
−∞ z2
+ 4z + 5
Z ∞
1 eız + e−ız
= 2
dz
−∞ 2 z + 4z + 5
1 ∞ eız 1 ∞ e−ız
Z Z
= dz + dz,
2 −∞ z 2 + 4z + 5 2 −∞ z 2 + 4z + 5

where the integral still extends over the real axis in the complex z plane.

(b) We use the Baskara formula in order to find the roots of the polynomial,

−4 ± 16 − 20
x± =
√2
−4 ± ı 4
=
2
= −2 ± ı,

so that, using the complex variable z in place of x, the polynomial can


be factored as

z 2 + 4z + 5 = [z − (−2 + ı)][z − (−2 − ı)],

and therefore the integral can be written as

1 1
I = I⊕ + I⊖
2Z 2
1 ∞ eız
= dz +
2 −∞ [z − (−2 + ı)][z − (−2 − ı)]
1 ∞ e−ız
Z
+ dz.
2 −∞ [z − (−2 + ı)][z − (−2 − ı)]
CALCULATION OF INTEGRALS BY RESIDUES 253

(c) Each of the two integrals has to be closed in a different way. Since we
have z = x + ıy, it follows that for the first integral the exponential
appearing in the numerator is

eız = eı(x+ıy)
= eıx e−y ,

wherein the first exponential, with an imaginary argument, is a limited


function, and the second goes to zero if y → ∞, but diverges if y → −∞.
Thus, we see that this first integral must be closed by an arc in the upper
half-plane, where y > 0, and that in this case the relevant singularity
is at z = −2 + ı. For the other integral the situation is reversed due
to the inversion of the sign in the argument of the exponential, and
therefore it will have to be closed by the lower half-plane, and the
relevant singularity is at z = −2 − ı.
(d) Let us start with the first of the two integrals. The additional integral
I⊕R is defined on a semicircle of radius R, with θ going from 0 to π,
in the limit in which we make R → ∞. Using z = R exp(ıθ) and thus
dz = ız dθ on the arc, we have for this integral
Z π
eız
I⊕R = dz 2
z + 4z + 5
Z0 π
eız
= dθ ız 2
0 z + 4z + 5
Z π
eıR[cos(θ)+ı sin(θ)]
= dθ ıR eıθ 2 ı2θ
0 R e + 4R eıθ + 5
Z π
ı eıR cos(θ) e−R sin(θ)
= dθ eıθ ı2θ .
R 0 e + 4 eıθ /R + 5/R2

We see that in the R → ∞ limit the integrand is simplified, so that we


have, calculating the absolute value of the integral in this limit,

π
1 eıR cos(θ) e−R sin(θ)
Z
lim |I⊕R | = lim dθ eıθ
R→∞ R→∞ R 0 eı2θ + 4 eıθ /R + 5/R2
1 π eıR cos(θ) e−R sin(θ)
Z
≤ lim dθ eıθ
R→∞ R 0 | eı2θ + 4 eıθ /R + 5/R2 |
π
1 e−R sin(θ)
Z
= lim dθ
R→∞ R 0 | eı2θ |
254 SOLUTIONS 15

1 π
Z
= lim dθ e−R sin(θ)
R→∞ R 0
1 π
Z
≤ lim dθ
R→∞ R 0
π
= lim
R→∞ R
= 0.

In these passages we replaced sin(θ) by its minimum value within the


integration interval, [0, π], in order to maximize the real exponential.
Note that in this interval the sine is always positive, and that its min-
imum value is zero. Since the absolute value of I⊕R is positive and
is bounded from above by zero in the limit, it follows that the limit
of the absolute value is zero. If the absolute value tends to zero, it is
necessary that the number also tend to zero, so that we have

lim I⊕R = 0.
R→∞

For the second integral everything takes place analogously but with θ
ranging from π to 2π, so that we have


e−ız
Z
I⊖R = dz
π z2
+ 4z + 5
Z 2π
e−ız
= dθ ız 2
π z + 4z + 5
Z 2π
e−ıR[cos(θ)+ı sin(θ)]
= dθ ıR eıθ 2 ı2θ
π R e + 4R eıθ + 5
Z 2π
ı e−ıR cos(θ) eR sin(θ)
= dθ eıθ ı2θ .
R π e + 4 eıθ /R + 5/R2

Considering now the R → ∞ limit of the absolute value of the integral,


1 e−ıR cos(θ) eR sin(θ)
Z
lim |I⊖R | = lim dθ eıθ
R→∞ R→∞ R π eı2θ + 4 eıθ /R + 5/R2
1 2π e−ıR cos(θ) eR sin(θ)
Z
≤ lim dθ eıθ
R→∞ R π | eı2θ + 4 eıθ /R + 5/R2 |

1 eR sin(θ)
Z
= lim dθ
R→∞ R π | eı2θ |
CALCULATION OF INTEGRALS BY RESIDUES 255

1 2π
Z
= lim dθ eR sin(θ)
R→∞ R π

1 2π
Z
≤ lim dθ
R→∞ R π
π
= lim
R→∞ R
= 0.

In these passages we replaced sin(θ) by its maximum value within the


integration interval, [π, 2π], in order to maximize the real exponential.
Note that in this interval the sine is always negative, and that its max-
imum value is zero. Since the absolute value of I⊖R is positive and
is bounded from above by zero in the limit, it follows that the limit
of the absolute value is zero. If the absolute value tends to zero, it is
necessary that the number also tend to zero, so that we have

lim I⊖R = 0.
R→∞

(e) The contour of the integral I⊕ is being traversed in the positive di-
rection, and the residue ξ⊕ of the relevant pole in this case can be
calculated by the limit

[z − (−2 + ı)] eız


ξ⊕ = lim
z→(−2+ı) [z − (−2 + ı)][z − (−2 − ı)]
eız
= lim
z→(−2+ı) z − (−2 − ı)

eı(−2+ı)
=
−2 + ı + 2 + ı
e−2ı−1
= .

However, the contour of the integral I⊖ is being traversed in the nega-


tive direction, and the residue ξ⊖ of the relevant pole in this case can
be calculated by the limit

[z − (−2 − ı)] e−ız


ξ⊖ = lim
z→(−2−ı) [z − (−2 + ı)][z − (−2 − ı)]
e−ız
= lim
z→(−2−ı) z − (−2 + ı)
256 SOLUTIONS 15

e−ı(−2−ı)
=
−2 − ı + 2 − ı
− e2ı−1
= .

It follows that we have for the original integral

1 1
I = I⊕ + I⊖
2 2
1 1
= 2πıξ⊕ + (−2)πıξ⊖
2 2
e−2ı−1 − e2ı−1
= πı − πı
2ı 2ı
π e2ı + e−2ı
=
e 2
π
= cos(2).
e

Problem 7. Consider the following real asymptotic integral, to be calcu-


lated by residues,
Z ∞
cos(x + 1)
2
dx.
−∞ x + 4x + 5

Consider the following steps.

(a) Determine and describe one or more closed contours in the complex
plane that can be used to compute this integral.

(b) Show that the integrals over the additional parts of the contours, used
to close the original contour over the real line, vanish.

(c) Calculate the residues of the relevant singularities and use the residue
theorem in order to find the value of the integral.

Answer: π cos(1)/e.

Complete Solution:

We must calculate the asymptotic integral


Z ∞
cos(x + 1)
I= 2 + 4x + 5
dx.
−∞ x
CALCULATION OF INTEGRALS BY RESIDUES 257

Writing the integral in terms of a complex variable z, and decomposing the


cosine into complex exponentials, we have
Z ∞
cos(z + 1)
I = 2 + 4z + 5
dz
−∞ z
Z ∞
1 eı(z+1) + e−ı(z+1)
= dz
−∞ 2 z 2 + 4z + 5
1 ∞ eı(z+1) 1 ∞ e−ı(z+1)
Z Z
= dz + dz,
2 −∞ z 2 + 4z + 5 2 −∞ z 2 + 4z + 5

where the integral still extends over the real axis in the complex z plane. We
now use the Baskara formula in order to find the roots of the polynomial,

−4 ± 16 − 20
x± =
2

−4 ± ı 4
=
2
= −2 ± ı,

so that, using the complex variable z in place of x, the polynomial can be


factored as

z 2 + 4z + 5 = [z − (−2 + ı)][z − (−2 − ı)],

and therefore the integral can be written as


1 1
I = I⊕ + I⊖
2 2
1 ∞ eı(z+1)
Z
= dz +
2 −∞ [z − (−2 + ı)][z − (−2 − ı)]
1 ∞ e−ı(z+1)
Z
+ dz,
2 −∞ [z − (−2 + ı)][z − (−2 − ı)]

where the two integrals still extend over the real axis in the complex z plane.

(a) Each of the two integrals has to be closed in a different way. Since we
have z = x + ıy, it follows that for the first integral, the exponential
appearing in the numerator is

eı(z+1) = eı(x+1+ıy)
= eı(x+1) e−y ,
258 SOLUTIONS 15

wherein the first exponential, with an imaginary argument, is a limited


function, and the second goes to zero if y → ∞, but diverges if y → −∞.
Thus, we see that this first integral must be closed by an arc in the upper
half-plane, where y > 0, and that in this case the relevant singularity
is at z = −2 + ı. For the other integral the situation is reversed due
to the inversion of the sign in the argument of the exponential, and
therefore it will have to be closed by the lower half-plane, and the
relevant singularity is at z = −2 − ı.
(b) Let us start with the first of the two integrals. The additional integral
I⊕R is defined on a semicircle of radius R, with θ going from 0 to π,
in the limit in which we make R → ∞. Using z = R exp(ıθ) and thus
dz = ız dθ on the arc, we have for this integral
Z π
eı(z+1)
I⊕R = dz 2
0 z + 4z + 5
Z π
eı(z+1)
= dθ ız 2
0 z + 4z + 5
Z π
eı{R[cos(θ)+ı sin(θ)]+1}
= dθ ıR eıθ 2 ı2θ
0 R e + 4R eıθ + 5
Z π
ı eı[R cos(θ)+1] e−R sin(θ)
= dθ eıθ ı2θ .
R 0 e + 4 eıθ /R + 5/R2

We see that in the R → ∞ limit the integrand is simplified, so that we


have, calculating the absolute value of the integral in this limit,
π
1 eı[R cos(θ)+1] e−R sin(θ)
Z
lim |I⊕R | = lim dθ eıθ
R→∞ R→∞ R 0 eı2θ + 4 eıθ /R + 5/R2
1 π eı[R cos(θ)+1] e−R sin(θ)
Z
≤ lim dθ eıθ
R→∞ R 0 | eı2θ + 4 eıθ /R + 5/R2 |
1 π e−R sin(θ)
Z
= lim dθ
R→∞ R 0 | eı2θ |
Z π
1
= lim dθ e−R sin(θ)
R→∞ R 0
1 π
Z
≤ lim dθ
R→∞ R 0
π
= lim
R→∞ R
= 0.
CALCULATION OF INTEGRALS BY RESIDUES 259

In these passages we replace sin(θ) by its minimum value within the


integration interval, [0, π], in order to maximize the real exponential.
Note that in this interval the sine is always positive, and that its min-
imum value is zero. Since the absolute value of I⊕R is positive and
is bounded from above by zero in the limit, it follows that the limit
of the absolute value is zero. If the absolute value tends to zero, it is
necessary that the number also tend to zero, so that we have

lim I⊕R = 0.
R→∞

For the second integral everything takes place analogously but with θ
ranging from π to 2π, so that we have


e−ı(z+1)
Z
I⊖R = dz
π z 2 + 4z + 5

e−ı(z+1)
Z
= dθ ız
π z 2 + 4z + 5

e−ı{R[cos(θ)+ı sin(θ)]+1}
Z
= dθ ıR eıθ
π R2 eı2θ + 4R eıθ + 5

e−ı[R cos(θ)+1] eR sin(θ)
Z
ı
= dθ eıθ .
R π eı2θ + 4 eıθ /R + 5/R2

Considering now the R → ∞ limit of the absolute value of the integral,


1 e−ı[R cos(θ)+1] eR sin(θ)
Z
lim |I⊖R | = lim dθ eıθ
R→∞ R→∞ R π eı2θ + 4 eıθ /R + 5/R2
1 2π e−ı[R cos(θ)+1] eR sin(θ)
Z
≤ lim dθ eıθ
R→∞ R π | eı2θ + 4 eıθ /R + 5/R2 |
1 2π eR sin(θ)
Z
= lim dθ
R→∞ R π | eı2θ |
Z 2π
1
= lim dθ eR sin(θ)
R→∞ R π

1 2π
Z
≤ lim dθ
R→∞ R π
π
= lim
R→∞ R
= 0.
260 SOLUTIONS 15

In these passages we replace sin(θ) by its maximum value within the


integration interval, [π, 2π], in order to maximize the real exponential.
Note that in this interval the sine is always negative, and its maximum
value is zero. Since the absolute value of I⊖R is positive and is bounded
from above by zero in the limit, it follows that the limit of the absolute
value is zero. If the absolute value tends to zero, it is necessary that
the number also tend to zero, so that we have

lim I⊖R = 0.
R→∞

(c) The contour of the integral I⊕ is being traversed in the positive di-
rection, and the residue ξ⊕ of the relevant pole in this case can be
calculated by the limit

[z − (−2 + ı)] eı(z+1)


ξ⊕ = lim
z→(−2+ı) [z − (−2 + ı)][z − (−2 − ı)]

eı(z+1)
= lim
z→(−2+ı) z − (−2 − ı)

eı(−1+ı)
=
−2 + ı + 2 + ı
e−ı−1
= .

However, the contour of the integral I⊖ is being traversed in the nega-


tive direction, and the residue ξ⊖ of the relevant pole in this case can
be calculated by the limit

[z − (−2 − ı)] e−ı(z+1)


ξ⊖ = lim
z→(−2−ı) [z − (−2 + ı)][z − (−2 − ı)]

e−ı(z+1)
= lim
z→(−2−ı) z − (−2 + ı)

e−ı(−1−ı)
=
−2 − ı + 2 − ı
− eı−1
= .

It follows that we have for the original integral


CALCULATION OF INTEGRALS BY RESIDUES 261

1 1
I = I⊕ + I⊖
2 2
1 1
= 2πıξ⊕ + (−2)πıξ⊖
2 2
e−ı−1 − eı−1
= πı − πı
2ı 2ı
π eı + e−ı
=
e 2
π
= cos(1).
e

Problem 8. Calculate by residues the following integral,



4
Z
dθ.
0 5 + 4 sin(θ)

Consider the following steps.

(a) Write sin(θ) in terms of the complex exponentials exp(±ıθ).

(b) Change variables in the integral, from θ to z = exp(ıθ). For this


purpose write dθ in terms of dz. Also determine what is the integration
contour in the complex z plane.

(c) Write the transformed integral as the integral of a rational function,


that is, the ratio of two polynomials on z.

(d) Factor completely the polynomial in the denominator.

(e) Determine which are the relevant singularities.

(f) Calculate the relevant residues and use the residue theorem in order to
find the value of the integral.

Answer: 8π/3.

Complete Solution:

We will calculate by residues the real integral



4
Z
I= dθ .
0 5 + 4 sin(θ)
262 SOLUTIONS 15

(a) Writing the sine in terms of complex exponentials we have

1  ıθ 
sin(θ) = e − e−ıθ
2ı  
1 1
= z− ,
2ı z

where the last form is valid provided that it be restricted to the unit
circle in complex z plane, with z = ρ exp(ıθ) and ρ = 1.
(b) Writing z in the polar representation, we have

z = ρ eıθ ⇒
dz = ız dθ,

so that we have
1
dθ = dz.
ız
This holds independently of the value of ρ but, as we saw above, only
for ρ = 1 the sine can be written simply in terms of z, so that the
integration contour in the complex z plane is the unit circle.
(c) Making in the integral the transformations indicated above, we obtain

1 1
Z
I = 4 ız dθ
0 ız 5 − 2ı(z − 1/z)
Z 2π
1 1
= 4 dz
ı 5z − 2ı(z 2 − 1)
I0
1
= 4 dz ,
C 5ız + 2z 2 − 2

where the closed contour C is the unit circle.


(d) It is now necessary to factor the polynomial in the denominator. Using
the Baskara formula it is not difficult to see that the two roots are −2ı
and −ı/2, so that we have for the integral

1
I
I = 4 dz 2
2z + 5ız − 2
IC
1
= 4 dz .
C 2(z + 2ı)(z + ı/2)
CALCULATION OF INTEGRALS BY RESIDUES 263

(e) Only the pole given by z = −ı/2 lies within the integration contour, so
that only its residue ξ will contribute to the integral.

(f) The relevant residue of the integrand can be calculated very simply
through the limit

(z + ı/2)
ξ = lim
2(z + 2ı)(z + ı/2)
z→−ı/2
1
= lim
z→−ı/2 2(z + 2ı)
1
=
−ı + 4ı
1
= .

It follows that the value of the integral is given, without any difficulty,
by

I = 4(2πı)ξ

= .
3

Problem 9. Calculate by residues the following integral,


Z π
1
2 dθ.
−π 1 + sin (θ)

Consider the following steps.

(a) Write sin2 (θ) in terms of the complex exponentials exp(±ıθ).

(b) Change variables in the integral, from θ to z = exp(ıθ). For this


purpose write dθ in terms of dz. Also determine what is the integration
contour in the complex z plane.

(c) Write the transformed integral as the integral of a rational function,


that is, the ratio of two polynomials on z.

(d) Factor completely the polynomial in the denominator.

(e) Determine which are the relevant singularities.


264 SOLUTIONS 15

(f) Calculate the relevant residues and use the residue theorem in order to
find the value of the integral.

Answer: π 2.

Complete Solution:

We will calculate by residues the real integral


Z π
1
I= dθ .
−π 1 + sin2 (θ)

(a) Writing the sine in terms of complex exponentials we have

1  ıθ 
sin(θ) = e − e−ıθ
2ı  
1 1
= z− ,
2ı z

where the last form is valid provided that it be restricted to the unit
circle in complex z plane, with z = ρ exp(ıθ) and ρ = 1. We have
therefore for the square of the sine

1 2
 
2 −1
sin (θ) = z−
4 z
 
1 2 1
= − z −2+ 2 .
4 z

(b) Writing z in the polar representation, we have

z = ρ eıθ ⇒
dz = ız dθ,

so that we have

1
dθ = dz.
ız

This holds independently of the value of ρ but, as we saw above, only


for ρ = 1 the sine can be written simply in terms of z, so that the
integration contour in the complex z plane is the unit circle.
CALCULATION OF INTEGRALS BY RESIDUES 265

(c) Making in the integral the transformations indicated above, we obtain

π
1 4
Z
I = ız dθ
−π ız 4 − (z − 2 + 1/z 2 )
2
Z π
1 z
= 4ı ız dθ 2 2
z z − 6 + 1/z 2
I−π
z
= 4ı dz 4 ,
C z − 6z 2 + 1

where the closed contour C is the unit circle.

(d) It now necessary to factor the polynomial in the denominator. Using


the Baskara formula for the quadratic polynomial on z 2 it √is not difficult
to see that the two roots are given in terms of z 2 by 3 ± 2p2. It follows

that the four roots of the polynomial are given by z = ± 3 ± 2 2, so
that we have for the integral

z
I
I = 4ı dz
z4 − 6z 2 + 1
IC
= 4ı dz ×
C
z
×  p √  p √  p √  p √  .
z− 3+2 2 z + 3+2 2 z− 3−2 2 z + 3−2 2

(e) Aspone can see, all four poles are on the real axis. If we take the square of
√ √
± 3 ±√2 2, we obtain 3 ± 2 2, and taking the square again we obtain
17 ± 12 2. It is thus clear that the cases in which the remaining sign is
positive fall outside the unit circle. Calculating the√numbers involved
in the other case in an approximate way, we have 12 2 ≈ 16.97, which
makes it clear that in this case the two roots are inside the unit circle.
We therefore conclude that only p two of the four poles contribute to the

integral, those given by z = ± 3 − 2 2.

(f) The relevant residues of the integrand can be calculated very simply
through the limit
 p √ 
z z∓ 3−2 2
ξ± = lim
√ √  2 √   p √  p √ 
z→± 3−2 2 z − 3 + 2 2 z− 3−2 2 z+ 3−2 2
266 SOLUTIONS 15

z
= lim
√ √  √   p √ 
z→± z2 − 3 + 2 2
3−2 2 z± 3−2 2
p √
± 3−2 2
=  √  √   p √ 
3−2 2 − 3+2 2 ±2 3 − 2 2
1
= √ √ 
2 3−2 2−3−2 2
−1
= √ ,
8 2
which ends up being, as we see, the same result for the two poles,

ξ⊕ = ξ⊖
1
= − √ .
8 2
It follows that the value of the integral is given, without greater diffi-
culty, by

I = 4ı(2πı)(ξ⊕ + ξ⊖ )
 
(−1)
= (−8π) 2 √
8 2

= π 2.

Problem 10. Consider the complex function w(z) = 1/z, and the unit
circle in the complex z plane.
(a) Calculate the closed-contour integral of w(z) on the complete unit cir-
cle.
(b) Calculate the open integral of w(z) on the upper semicircle of the unit
circle, that is, the integral over the circle between the points (−1, 0)
and (1, 0).
(c) Calculate the Cauchy principal value of the closed-contour integral of
w(z) on the upper semicircle of the unit circle, closed by the real line
segment between the points (1, 0) and (−1, 0).
(d) Show that the complex version of the criterion for the Cauchy principal
value is equivalent to the real version of the criterion, that is, to the
use of the symmetric limit defined in the text for the calculation of the
real integral between −1 and 1.
CALCULATION OF INTEGRALS BY RESIDUES 267

Complete Solution:

We must calculate several definite integrals of the complex function w(z) =


1/z, on curves in the complex plane.

(a) Let us calculate the integral of the function on the unit circle of the
complex z plane,

1
I
dz = 2πı,
C z

which can be obtained immediately in several different ways, for exam-


ple using the Cauchy integral formula,

1 f (z)
I
f (z0 ) = dz ,
2πı z − z0

with f (z) = 1 for all z, and z0 = 0.

(b) Let us calculate the open integral of w(z) on the upper semicircle C⊕ (1)
of the unit circle, that is, the integral over the unit circle between the
points (1, 0) and (−1, 0). Using z = ρ exp(ıθ), with ρ = 1 and θ going
from 0 to π, we have dz = ız dθ, and therefore

π
1
Z
IC⊕ (1) = dθ ız
0 z
Z π
= ı dθ
0
= ıπ.

(c) Let us calculate the Cauchy principal value of the closed-contour inte-
gral of w(z) on the contour C formed by upper semicircle of the unit
circle, closed by the real line segment between the points (−1, 0) and
(1, 0). The residue at the pole is 1, and since we are calculating the
principal value, we take half of its value,

1
I
I = dz
C z
= πı.
268 SOLUTIONS 15

(d) Let us consider the upper unit semicircle C⊕ (1) and a small circle of
radius ρ centered at the origin, formed by the upper semicircle C⊕ (ρ)
and by the lower semicircle C⊖ (ρ). By its definition in the complex
plane, the Cauchy principal value is the average of the following two
integrals,
Z −ρ Z 1
1 1 1 1
Z Z
I⊕ = dz + dx − dz + dx ,
C⊕ (1) z −1 z C⊕ (ρ) z ρ z
Z −ρ Z 1
1 1 1 1
Z Z
I⊖ = dz + dx + dz + dx ⇒
C⊕ (1) z −1 z C⊖ (ρ) z ρ z
I⊕ + I⊖
I =
2
Z −ρ Z 1
1 1 1 1
Z
= 2 dz + 2 dx + 2 dx +
2 C⊕ (1) z −1 z ρ z
!
1 1
Z Z
+ dz − dz
C⊖ (ρ) z C⊕ (ρ) z
Z −ρ Z 1
1 1 1
Z
= dz + dx + dx +
C⊕ (1) z −1 z ρ z
!
1 1 1
Z Z
− dz − dz .
2 C⊕ (ρ) z C⊖ (ρ) z

Both I and the integral over C⊕ (1) are already known. Let us calculate
the integrals over C⊕ (ρ) and C⊖ (ρ),
π
1
Z
IC⊕ (ρ) = dθ ız
0 z
Z π
= ı dθ
0
= ıπ,
Z 2π
1
IC⊖ (ρ) = dθ ız
π z
Z π
= ı dθ
0
= ıπ,

so that these two integrals cancel off in the expression for I, and we
have therefore
CALCULATION OF INTEGRALS BY RESIDUES 269

−ρ 1
1 1 1
Z Z Z
I = dz + dx + dx
C⊕ (1) z −1 z ρ z
−ρ 1
1 1
Z Z
= ıπ + dx + dx
−1 z ρ z
= ıπ,

from which it follows that

−ρ 1
1 1
Z Z
dx + dx = 0.
−1 z ρ z

The ρ → 0 limit, which defines the Cauchy principal value in the com-
plex plane, corresponds therefore to the limit

Z −ρ 1 
1 1
Z
lim dx + dx = 0,
ρ→0 −1 z ρ z

which is the corresponding definition of the Cauchy principal value in


the real case.

Problem 11. Consider the complex function w(z) = 1/z, and the real axis
in the complex z plane. Calculate the integral of the function on the real axis
according to the criterion of the Cauchy principal value, that is, according
to the definition
Z ∞
I = dz w(z)
−∞
Z R
= lim dz w(z),
R→∞ −R

with respect to the asymptotic part, and the Cauchy criterion discussed in
the text for the integration through the singularity at the origin. Do this in
the two ways listed below and interpret the meaning of the Cauchy principal
value of all the integrals that appear.

(a) Do the calculation directly by elementary means, that is, do the real
integral using the fundamental theorem of the calculus, for finite R,
and then take the limit R → ∞.
270 SOLUTIONS 15

(b) Close the contour in the complex plane by means of a circular arc, with
finite radius R, consider the integral over this arc of circle, use the
residue theorem in order to calculate the integral, and then take the
limit R → ∞.

Answer: I = 0.

Complete Solution:

(a) Let us calculate directly the integral of w(z) on the real axis. Using the
criterion of the Cauchy principal value for both the asymptotic part
and the singular region around zero, we have, with z = x + ıy and
y = 0,
Z ∞
1
I = dx
−∞ x
Z −ǫ Z R 
1 1
= lim lim dx + dx .
R→∞ ǫ→0 −R x ǫ x

Making the transformation of variables x′ = −x in the first integral we


obtain
Z ǫ Z R 
′ 1 1
I = lim lim dx ′ + dx
R→∞ ǫ→0 R x ǫ x
 Z R Z R 
1 1
= lim lim − dx′ ′ + dx
R→∞ ǫ→0 ǫ x ǫ x
Z R Z R 
1 1
= lim lim dx − dx
R→∞ ǫ→0 ǫ x ǫ x
= 0,

independently of either R or ǫ. Note that it is not even necessary to do


the integration, which in any case can easily be done,
Z R Z R 
1 1
I = lim lim dx − dx
R→∞ ǫ→0 x x
 ǫ  ǫ 
R R
= lim lim ln − ln
R→∞ ǫ→0 ǫ ǫ
= lim lim ln(1)
R→∞ ǫ→0
= 0.
CALCULATION OF INTEGRALS BY RESIDUES 271

It follows, of course, that we can simply take the two limits involved
and therefore that the answer is I = 0.

(b) Let us calculate the open integral of w(z) on the upper semicircle
C⊕ (R), of radius R, of the circle of radius R, that is, the integral
over the circle of radius R, in the positive direction, between the points
(R, 0) and (−R, 0). Using z = R exp(ıθ), with θ going from 0 to π, we
have that dz = ız dθ, and therefore that
π
1
Z
IC⊕ (R) = dθ ız
0 z
Z π
= ı dθ
0
= ıπ.

We can now use this semicircle in order to close the integral IR of w(z)
on the real axis from −R to R, and so we have a closed contour integral,
whose value is the sum of the integral IR we want to calculate with the
integral IC⊕ (R) we have just calculated, that is, IR + IC⊕ (R) = IR + ıπ.
We now use the residue theorem in order to calculate the integral of
w(z) on this closed contour. Following the criterion of the Cauchy
principal value when the integral passes over the pole at z = 0, we have
that the value of the residue is 1, and therefore that the closed-contour
integral has the value 2ıπ(1/2) = ıπ, that is, we have that IR +ıπ = ıπ,
from which it follows that IR = 0. This is independent of R, and thus
holds for all R, so that, making R → ∞, we have the answer I = 0.

Problem 12. Calculate by residues the following integral,


Z ∞
1
dx.
−∞ cosh(x/2)

Consider the following steps.

(a) Write cosh(x/2) in terms of the real exponentials exp(±x).

(b) Extend the variable x to the complex plane of z = x + ıy, and verify to
what the integral reduces over two infinite straight lines: the real axis
y = 0 and the line y = 2π.

(c) Consider how to close a contour which comprises these two straight
lines, in order to enable the calculation of the integral.
272 SOLUTIONS 15

(d) Locate the singularities of the function being integrated and determine
which ones are relevant.

(e) Show that the integrals over the two vertical segments which are in-
cluded in order to close the contour vanish when these additional seg-
ments are taken to x → ±∞.

(f) Assume that the relevant pole is of the first order and calculate the
residue by means of the limit that applies to simple poles. Use the
residue theorem in order to find the value of the integral.

Answer: 2π.

(g) (Challenge Item) Find the Laurent series of the function being inte-
grated, around the relevant pole, including a sufficient number of terms
to determine the order of the pole and the value of the residue.

Complete Solution:

We must calculate the following asymptotic integral,


Z ∞
1
I= dx.
−∞ cosh(x/2)

(a) Writing the hyperbolic cosine in terms of real exponentials we have


2
Z
I = dx
−∞ ex/2 + e−x/2

2 ex/2
Z
= dx
−∞ ex + 1


2 ex
Z
= dx.
−∞ ex + 1

(b) Extending the integral to the complex plane of z = x + ıy we have

ez/2
Z
Iz = 2 dz,
C ez + 1

on some contour C yet to be specified. On the real axis y = 0 the


integral is reduced, of course, to that from which we began,
CALCULATION OF INTEGRALS BY RESIDUES 273

I = I0

ex/2
Z
= 2 dx.
−∞ ex + 1

However, on the axis y = 2π we have for the integral,


ex/2 eıπ
Z
I2π = 2 dx
−∞ ex e2ıπ + 1
Z ∞
ex/2
= −2 dx
−∞ ex + 1
= −I0 ,

since exp(2ıπ) = 1 and exp(ıπ) = −1.

(c) If we close the contour formed by the straight lines y = 0 and y = 2π by


two vertical segments of finite length 2π, which are taken to infinity, one
in each direction, and we traverse the resulting contour in the positive
direction, the bottom line y = 0 will contribute I0 to the integral, and
the top line y = 2π will contribute −I2π = I0 , that is, the same value,
because this line is being traversed in its negative direction. If the
two finite segments did not contribute to the integral in the limit in
which they are taken to infinity, then we would be able to calculate the
original integral using this closed contour C, that is,

1 ez/2
I
I = 2 z
dz
2 C e +1
ez/2
I
= z
dz.
C e +1

(d) The singularities of the integrand, considering that the exponential


function is analytic on the whole complex plane, and that the integrand
is given by

ez/2
,
ez + 1

are given by the zeros in the denominator of this ratio. The zeros
are therefore at the points defined by exp(z) = −1, that is, at the
coordinates x and y given by
274 SOLUTIONS 15

ex eıy = ex [cos(y) + ı sin(y)]


= −1 ⇒
x
e cos(y) = −1,
ex sin(y) = 0.

From the second relation, since the real exponential never vanishes, we
conclude that it is necessary to have y = kπ, for some integer value of
k. This implies that the cosine is equal to (−1)k , and therefore, from
the first relation, that

ex (−1)k = −1.

Since the real exponential is never negative, it follows that k must be


odd, and hence it follows that x = 0, so that this equation can be
satisfied. In conclusion, the singularities are on the imaginary axis, at
the coordinates x = 0 and y = (2k + 1)π, with integer k. Of all these
infinitely many singularities only one is within the contour C described
above, the one with k = 0, located at (0, π).

(e) The additional integrals over the segments that go to infinity, located
at ±x0 , are given by


ez/2
Z
Ix 0 = ıdy
0 ez + 1
Z 2π
1
= ı dy
0 e + e−z/2
z/2
Z 2π
1
= ı x /2
dy.
0 e 0 eıy/2 + e−x0 /2 e−ıy/2

The exponentials with imaginary arguments in y are limited functions,


and when we make x0 → ±∞ the exponentials of x0 go to zero or to
infinity, depending on the signs involved. From the two terms in the
denominator, one always goes to zero, and can be disregarded when
calculating the limit. The other term always goes to infinity, and the
bounded oscillating function that multiplies it is never zero. Thus we
see that in both cases the limit of the above integral is zero,

lim Ix0 = 0.
x0 →±∞
CALCULATION OF INTEGRALS BY RESIDUES 275

(f) The residue ξ of the pole located at z = (0, π) can be calculated using
the limit

(z − ıπ) ez/2
ξ = lim .
z→ıπ ez + 1

Analyzing this limit, we see that we have a zero in the numerator and
a zero in the denominator, while the exponential in the numerator has
a finite and non-zero limit. There are several ways of calculating this
limit, including the use of the l’Hôpital rule (in fact, due to Bernoulli).
What we will do here is to calculate the limit of the inverse ratio of the
two factors that cancel each other out, using the Taylor series of the
exponential around the point ıπ. It is not difficult to verify that this
series is given by


X −1
ez = (z − ıπ)n ,
n!
n=0

where we have that



z
X 1
e +1=− (z − ıπ)n ,
n=1
n!

and therefore that



ez + 1 X 1
= − (z − ıπ)n−1
z − ıπ n!
n=1

X 1
= − (z − ıπ)n .
(n + 1)!
n=0

This makes it very easy to take the limit, because only the first term of
the remaining series is non-zero in the limit, and we have immediately
that

ez + 1
lim = −1.
z→ıπ z − ıπ

From this it also follows that


276 SOLUTIONS 15

z − ıπ 1
lim =
z→ıπ ez + 1 −1
= −1.

Finally, it follows that we have for the original limit

(z − ıπ) ez/2
ξ = lim
z→ıπ ez + 1
= −e ıπ/2

= −ı.

We can now calculate the integral,

ez/2
I
I = z
dz
C e +1
= (2ıπ)ξ
= (2ıπ)(−ı)
= 2π.

(g) In order to formalize and clarify what we just did, we calculate the
initial part of the Laurent series of the integrand with respect to the
point z0 = ıπ. In its complex form, the integrand can be written as

ez/2
f (z) =
ez + 1
1
=
ez/2 + e−z/2
1
=
2 cosh(z/2)
X∞
= an (z − ıπ)n ,
n=−∞

where we wrote the general form of the representation in terms of the


Laurent series. Recall now that, as shown above, the limit

ez/2
lim (z − ıπ) z = lim (z − ıπ)f (z)
z→ıπ e + 1 z→ıπ
CALCULATION OF INTEGRALS BY RESIDUES 277

exists, is finite and non-zero. It follows from this fact that the largest
negative power present in the Laurent series of f (z) is n = −1, because
if there were a negative power greater than this, the limit would not be
finite. In other words, the existence of this limit shows that the pole is
of the first order, that is, a simple pole. Thus, we can write for f (z)

g(z)
f (z) =
z − ıπ
X∞
= an (z − ıπ)n ⇒
n=−1

X
g(z) = an (z − ıπ)n+1
n=−1
X∞
= an−1 (z − ıπ)n
n=0
X∞
= bn (z − ıπ)n ,
n=0

where bn = an−1 are now the coefficients of the power series around
z0 = ıπ which represents g(z), that we now realize is a Taylor series,
since g(z) is an analytic function at z0 , if we define it at that point by
continuity. We have, moreover, that

g(z) = (z − ıπ)f (z)


(z − ıπ)
= ,
2 cosh(z/2)

so that we can construct the Taylor series of g(z), calculating its deriva-
tives at the point z0 , which in each case will involve the calculation of
a limit for z → z0 . However, before doing this it is interesting to deter-
mine the parity properties of this function, with respect to the variable
z ′ = z−ıπ. Let us start by determining the parity of cosh(z/2). Making
z = z ′ + ıπ we have for this function
′ ′
2 cosh(z ′ /2 + ıπ/2) = ez /2+ıπ/2 + e−z /2−ıπ/2
′ ′
= ez /2 eıπ/2 + e−z /2 e−ıπ/2
′ ′
= ı ez /2 − ı e−z /2
= 2ı sinh(z ′ /2).
278 SOLUTIONS 15

Thus we see that cosh(z/2) is an odd function with respect to the


variable z ′ , and therefore so is f (z). Since we have that g(z) = z ′ f (z),
it follows that g(z) is an even function of z ′ , so that the Taylor series
of g(z) around z0 = ıπ only has the even powers,


X
g(z) = b2k (z − ıπ)2k .
k=0

Writing the function and its series in terms of z ′ , in order to facilitate


the derivation of the coefficients, we have

z′
g z′

=
2ı sinh(z ′ /2)

X 2k
= b2k z ′ .
k=0

The first coefficient is simply b0 = a−1 , that is the residue of the pole
and that, as we have already seen, can be obtained by the limit

z′
g z′

lim = lim
z →0 ′ z →0 2ı sinh(z ′ /2)

1
= lim
z ′ →0 ı cosh(z ′ /2)
1
=
ı
= −ı,

where we used the l’Hôpital rule. Calculating the first derivative of


g(z ′ ) we have

1 z ′ cosh(z ′ /2)
g′ z ′

= −
2ı sinh(z ′ /2) 4ı sinh2 (z ′ /2)
2 sinh(z ′ /2) − z ′ cosh(z ′ /2)
= .
4ı sinh2 (z ′ /2)

If we now calculate the z ′ → 0 limit of the derivative, which is once


more indeterminate, we have
CALCULATION OF INTEGRALS BY RESIDUES 279

2 sinh(z ′ /2) − z ′ cosh(z ′ /2)


g′ z ′

lim = lim

z →0 ′
z →0 4ı sinh2 (z ′ /2)
cosh(z ′ /2) − cosh(z ′ /2) − z ′ sinh(z ′ /2)/2
= lim
z ′ →0 4ı sinh(z ′ /2) cosh(z ′ /2)
−z sinh(z ′ /2)/2

= lim
z ′ →0 4ı sinh(z ′ /2) cosh(z ′ /2)
−z ′
= lim
z ′ →0 8ı cosh(z ′ /2)
= 0,

as predicted, and where we used once again the l’Hôpital rule. Calcu-
lating the second derivative of g(z) we have

cosh(z ′ /2) − cosh(z ′ /2) − z ′ sinh(z ′ /2)/2


g′′ z ′

= +
4ı sinh2 (z ′ /2)
2 sinh(z ′ /2) − z ′ cosh(z ′ /2) cosh(z ′ /2)
−2
4ı sinh3 (z ′ /2) 2
−z ′ sinh2 (z ′ /2) 4 sinh(z ′ /2) cosh(z ′ /2) − 2z ′ cosh2 (z ′ /2)
= −
8ı sinh3 (z ′ /2) 8ı sinh3 (z ′ /2)
−z ′ sinh2 (z ′ /2) − 4 sinh(z ′ /2) cosh(z ′ /2) + 2z ′ cosh2 (z ′ /2)
= .
8ı sinh3 (z ′ /2)

If we now calculate the z ′ → 0 limit of the second derivative, which is


once more indeterminate, we have

g′′ z ′

lim

z →0
−z ′ sinh2 (z ′ /2) − 4 sinh(z ′ /2) cosh(z ′ /2) + 2z ′ cosh2 (z ′ /2)
= lim
z →0 ′
8ı sinh3 (z ′ /2)
− sinh2 (z ′ /2) −z ′ sinh(z ′ /2) cosh(z ′ /2) +
 
 −2 cosh2 (z ′ /2) −2 sinh2 (z ′ /2) + 
2 ′ ′ ′ ′
+2 cosh (z /2) +2z cosh(z /2) sinh(z /2) +
= lim
z ′ →0 12ı sinh2 (z ′ /2) cosh(z ′ /2)
− sinh(z ′ /2) − z ′ cosh(z ′ /2) − 2 sinh(z ′ /2) + 2z ′ cosh(z ′ /2)
= lim
z ′ →0 12ı sinh(z ′ /2) cosh(z ′ /2)
− cosh(z ′ /2)/2 − cosh(z ′ /2) −z ′ sinh(z ′ /2)/2 +
 

− cosh(z ′ /2) +2 cosh(z ′ /2) +z ′ sinh(z ′ /2) +


= lim 2 2
z ′ →0 6ı cosh (z ′ /2) + 6ı sinh (z ′ /2)
280 SOLUTIONS 15

− cosh(z ′ /2)/2 + z ′ sinh(z ′ /2)/2


= lim
z →0 6ı cosh2 (z ′ /2) + 6ı sinh2 (z ′ /2)

−1/2
=

ı
= ,
12

where we used the l’Hôpital rule twice. Therefore we have the first two
non-zero terms of the Taylor series of g(z) around z0 = ıπ,

ı
g(z) = −ı + (z − ıπ)2 + . . . ,
24

and it finally follows that we have the first two non-zero terms of the
Laurent series of f (z) around z0 = ıπ,

ı ı
f (z) = − + (z − ıπ) + . . . ,
z − ıπ 24

where the residue ξ = −ı is now in clear view.


Solutions 16

Residues on Riemann
Surfaces

We present here complete and commented solutions to all problems proposed


in Chapter 16 of the text. For reference, the propositions of the problems are
repeated here. The problems are discussed in the order in which they were
proposed within the problem set of that chapter.

Problem 1. Consider the first complex integral that was discussed in the
text,

z
I
I¯ = dz ,
C z2 + 1
on the closed contour C defined in the text.

(a) Show with all rigor that the part of the integral over the segment Cε of
the contour is zero in the limit ε → 0.

(b) Show with all rigor that the part of the integral over the segment CR
of the contour is zero in the limit R → ∞.

Hint: take absolute values and use the triangle inequalities.

Complete Solution:

The complex integral to be discussed is



z
I
I¯ = dz 2 .
C z +1

281
282 SOLUTIONS 16

(a) On the segment Cε of the contour we have

2π √
z
Z
I¯Cε = dz ,
0 z2 +1

where z = ε exp(ıθ) and thus dz = ız dθ. Taking absolute values and


using the triangle inequalities we have

2π √
| z|
Z
I¯Cε ≤ |dz| ,
0 |z 2 + 1|
√ √
where |dz| = εdθ and | z| = ε. For the denominator we can write,
using once again the triangle inequalities,

1 = (z 2 + 1) − z 2 ⇒
2 2
1 ≤ z +1 + z ⇒
2 2
z +1 ≥ 1−ε .

We can therefore upper-bound the integrand and write

2π √
ε
Z
I¯Cε ≤ dθ ε .
0 1 − ε2

Taking now the ε → 0 limit we obtain


ε3/2
Z 
lim I¯Cε ≤ dθ lim .
ε→0 0 ε→0 1 − ε2

Since the integral that remains is limited and the limit shown is zero,
we have that

lim I¯Cε = 0,
ε→0

and therefore that

lim I¯Cε = 0,
ε→0

which is the required result.


RESIDUES ON RIEMANN SURFACES 283

(b) On the segment CR of the contour we have

2π √
z
Z
I¯CR = dz 2 ,
0 z +1

where z = R exp(ıθ) and thus dz = ız dθ. Taking absolute values and


using the triangle inequalities we have

2π √
| z|
Z
I¯CR ≤ |dz| 2 ,
0 |z + 1|
√ √
where |dz| = Rdθ and | z| = R. For the denominator we can write,
using once again the triangle inequalities,

z 2 = (z 2 + 1) − 1 ⇒
2 2
z ≤ z +1 +1 ⇒
2 2
z +1 ≥ R − 1.

We can therefore upper-bound the integrand and write



R
Z
I¯CR ≤ dθ R .
0 R2 − 1

Taking now the R → ∞ limit we obtain


R3/2
Z 
lim I¯CR ≤ dθ lim .
R→∞ 0 R→∞ R2 − 1

Since the integral that remains is limited and the limit shown is zero,
we have that

lim I¯CR = 0,
R→∞

and therefore that

lim I¯CR = 0,
R→∞

which is the required result.


284 SOLUTIONS 16

Problem 2. Consider the second complex integral that was discussed in


the text,
ln(z)
I
I¯ = dz ,
C 1 + z + z2
on the closed contour C defined in the text.

(a) Show with all rigor that the part of the integral over the segment Cε of
the contour is zero in the limit ε → 0.

(b) Show with all rigor that the part of the integral over the segment CR
of the contour is zero in the limit R → ∞.

Hint: take absolute values and use the triangle inequalities.

Complete Solution:

The complex integral to be discussed is


ln(z)
I
I¯ = dz .
C 1 + z + z2

(a) On the segment Cε of the contour we have


ln(z)
Z
I¯Cε = dz ,
0 1 + z + z2

where z = ε exp(ıθ) and thus dz = ız dθ. Taking absolute values and


using the triangle inequalities we have


|ln(z)|
Z
I¯Cε ≤ |dz| ,
0 |1 + z + z 2 |

where |dz| = εdθ. For the logarithm we can write, using once again the
triangle inequalities,

ln(z) = ln(ε) + ıθ ⇒
|ln(z)| ≤ |ln(ε)| + θ.

For the denominator we can write, using yet again the triangle inequal-
ities,
RESIDUES ON RIEMANN SURFACES 285

1 = (1 + z + z 2 ) − z − z 2 ⇒
2 2
1 ≤ 1+z+z + |z| + z ⇒
2 2
1+z+z ≥ 1−ε−ε .

We can therefore upper-bound the integrand and write


|ln(ε)| + θ
Z
I¯Cε ≤ dθ ε
0 1 − ε − ε2
Z 2π Z 2π
ε |ln(ε)| ε
= dθ 2
+ dθ θ .
0 1−ε−ε 0 1 − ε − ε2

Taking now the ε → 0 limit we obtain


Z 2π  Z 2π 
ε |ln(ε)| ε
lim I¯Cε ≤ dθ lim + dθ θ lim .
ε→0 0 ε→0 1 − ε − ε2 0 ε→0 1 − ε − ε2

Since the integrals that remain are limited and the limits shown are
zero, we have that

lim I¯Cε = 0,
ε→0

and therefore that

lim I¯Cε = 0,
ε→0

which is the required result.

(b) On the segment CR of the contour we have


ln(z)
Z
I¯CR = dz ,
0 1 + z + z2

where z = R exp(ıθ) and thus dz = ız dθ. Taking absolute values and


using the triangle inequalities we have


|ln(z)|
Z
I¯CR ≤ |dz| ,
0 |1 + z + z 2 |
286 SOLUTIONS 16

where |dz| = Rdθ. For the logarithm we can write, using once again
the triangle inequalities,

ln(z) = ln(R) + ıθ ⇒
|ln(z)| ≤ ln(R) + θ.

For the denominator we can write, using yet again the triangle inequal-
ities,

z 2 = (1 + z + z 2 ) − z − 1 ⇒
2 2
z ≤ 1+z+z + |z| + 1 ⇒
2 2
1+z+z ≥ R − R − 1.

We can therefore upper-bound the integrand and write


ln(R) + θ
Z
I¯CR ≤ dθ R
0 R2 − R − 1
Z 2π Z 2π
R ln(R) R
= dθ 2 + dθ θ 2 .
0 R −R−1 0 R −R−1

Taking now the R → ∞ limit we obtain


Z 2π 
R ln(R)
lim I¯CR ≤ dθ
lim 2 +
R→∞ 0 R→∞ R − R − 1
Z 2π 
R
+ dθ θ lim 2 .
0 R→∞ R − R − 1

Since the integrals that remain are limited and the limits shown are
zero, we have that

lim I¯CR = 0,
R→∞

and therefore that

lim I¯CR = 0,
R→∞

which is the required result.


RESIDUES ON RIEMANN SURFACES 287

Problem 3. Consider the third complex integral that was discussed in the
text,

I
I¯ = dz ,
C z2 +1
with 0 < α < 1, on the closed contour C defined in the text.

(a) Show with all rigor that the part of the integral over the segment Cε of
the contour is zero in the limit ε → 0.

(b) Show with all rigor that the part of the integral over the segment CR
of the contour is zero in the limit R → ∞.

Hint: take absolute values and use the triangle inequalities.

Complete Solution:

The complex integral to be discussed is



I
I¯ = dz 2 ,
C z +1
where 0 < α < 1.

(a) On the segment Cε of the contour we have



Z
I¯Cε = dz ,
0 z2 + 1

where z = ε exp(ıθ) and thus dz = ız dθ. Taking absolute values and


using the triangle inequalities we have


|z α |
Z
I¯Cε ≤ |dz| ,
0 |z 2+ 1|

where |dz| = εdθ and |z α | = εα . For the denominator we can write,


using once again the triangle inequalities,

1 = (z 2 + 1) − z 2 ⇒
2 2
1 ≤ z +1 + z ⇒
2 2
z +1 ≥ 1−ε .
288 SOLUTIONS 16

We can therefore upper-bound the integrand and write


εα
Z
I¯Cε ≤ dθ ε .
0 1 − ε2

Taking now the ε → 0 limit we obtain


ε1+α
Z 
lim I¯Cε ≤ dθ lim .
ε→0 0 ε→0 1 − ε2

Since the integral that remains is limited and the limit shown is zero,
we have that

lim I¯Cε = 0,
ε→0

and therefore that

lim I¯Cε = 0,
ε→0

which is the required result.

(b) On the segment CR of the contour we have



Z
I¯CR = dz ,
0 z2 + 1

where z = R exp(ıθ) and thus dz = ız dθ. Taking absolute values and


using the triangle inequalities we have


|z α |
Z
I¯CR ≤ |dz| ,
0 |z 2 + 1|

where |dz| = Rdθ and |z α | = Rα . For the denominator we can write,


using once again the triangle inequalities,

z 2 = (z 2 + 1) − 1 ⇒
2 2
z ≤ z +1 +1 ⇒
2 2
z +1 ≥ R − 1.
RESIDUES ON RIEMANN SURFACES 289

We can therefore upper-bound the integrand and write



Z
I¯CR ≤ dθ R .
0 R2 −1

Taking now the R → ∞ limit we obtain


R1+α
Z 
lim I¯CR ≤ dθ lim .
R→∞ 0 R→∞ R2 − 1

Since the integral that remains is limited and the limit shown is zero,
given that α < 1, we have that

lim I¯CR = 0,
R→∞

and therefore that

lim I¯CR = 0,
R→∞

which is the required result.

Problem 4. Calculate by residues the real asymptotic integral given by


Z ∞
α + βx
I= dx ,
0 1 + x + x2 + x3
for arbitrary real α and β. Justify explicitly every step of the procedure.
Answer: π(α + β)/4.

Complete Solution:

We must calculate the integral


Z ∞
α + βx
I= dx .
0 1 + x + x2 + x3
The denominator can be easily factored, for example by using the formula
for the sum of a geometric progression, resulting in

1 + x + x2 + x3 = (x + 1) x2 + 1


= (x + 1)(x − ı)(x + ı).


290 SOLUTIONS 16

Since there is in the integrand no singularity of the branch-point type, we


consider the complex integral

(α + βz) ln(z)
I
I¯ = dz
1 + z + z2 + z3
IC
(α + βz) ln(z)
= dz ,
C (z + 1)(z − ı)(z + ı)

on the closed contour C described in the text. Let us now discuss the two
arcs. On the segment Cε we have z = ε exp(ıθ) and thus dz = ıε exp(ıθ)dθ,
and therefore we obtain

(α + βz) ln(z)
Z
I¯Cε = dz
0 1 + z + z2 + z3
Z 2π
≈ dz (α + βz) ln(z)
0
Z 2π  
= ı dθ ε eıθ α + βε eıθ [ln(ε) + ıθ]
0
Z 2π  Z 2π 
= ıα dθ eıθ ε ln(ε) − α dθ θ eıθ ε +
0 0
Z 2π  Z 2π 
2ıθ 2 2ıθ
+ıβ dθ e ε ln(ε) − β dθ θ e ε2 .
0 0

Taking now the ε → 0 limit we obtain

lim I¯Cε
ε→0
Z 2π  Z 2π 
= ıα dθ eıθ
lim ε ln(ε) − α dθ θ eıθ lim ε +
0 ε→0 0 ε→0
Z 2π  Z 2π 
2ıθ 2 2ıθ
+ıβ dθ e lim ε ln(ε) − β dθ θ e lim ε2 .
0 ε→0 0 ε→0

The remaining integrals are all limited, and the limits shown are zero, so that
we have

lim I¯Cε = 0.
ε→0

On the segment CR we have z = R exp(ıθ) and thus dz = ıR exp(ıθ)dθ, and


thus we obtain
RESIDUES ON RIEMANN SURFACES 291


(α + βz) ln(z)
Z
I¯CR = dz
0 1 + z + z2 + z3
Z 2π
(α + βz) ln(z)
≈ dz
0 z3
Z 2π 
ıθ α + βR e
ıθ [ln(R) + ıθ]
= ı dθ R e
0 R3 e3ıθ
Z 2π  Z 2π 
−2ıθ ln(R) −2ıθ 1
= ıα dθ e 2
−α dθ θ e 2
+
0 R 0 R
Z 2π  Z 2π 
ln(R) 1
+ıβ dθ e−ıθ −β dθ θ e−ıθ .
0 R 0 R

Taking now the R → ∞ limit we obtain

lim I¯CR
R→∞
Z 2π  Z 2π 
ln(R)
−2ıθ −2ıθ 1
= ıα dθ e lim 2
−α dθ θ e lim 2 +
0 R→∞ R 0 R→∞ R
Z 2π  Z 2π 
ln(R) 1
+ıβ dθ e−ıθ lim −β dθ θ e−ıθ lim .
0 R→∞ R 0 R→∞ R

The remaining integrals are all limited, and the limits shown are zero, so that
we have

lim I¯CR = 0.
R→∞

The discussion of the two straight segments of the contour is exactly the same
as the one that was made explicitly in the text in a case similar to this one,
from which it results that we have for the real integral


I= .
−2πı

All that remains to be done here is the calculation of I¯ by means of the


residue theorem. Since the relevant singularities are three simple poles, we
obtain for each of them, by the method of the limit,

(α + βz) ln(z)
b1⊖1 = lim (z + 1)
z→−1 (z + 1)(z − ı)(z + ı)
(α + βz) ln(z)
= lim
z→−1 (z − ı)(z + ı)
292 SOLUTIONS 16

(α − β) ln(−1)
=
(−1 − ı)(−1 + ı)
(α − β) ln[exp(ıπ)]
=
(1 + ı)(1 − ı)
ıπ(α − β)
= ,
2
(α + βz) ln(z)
b1⊕ı = lim (z − ı)
z→ı (z + 1)(z − ı)(z + ı)
(α + βz) ln(z)
= lim
z→ı (z + 1)(z + ı)
(α + ıβ) ln(ı)
=
(ı + 1)(ı + ı)
(α + ıβ) ln[exp(ıπ/2)]
=
(1 + ı)2ı
ıπ(α + ıβ)
=
4ı(1 + ı)
π(α + ıβ)(1 − ı)
=
8
π[(α + β) − ı(α − β)]
= ,
8
(α + βz) ln(z)
b1⊖ı = lim (z + ı)
z→−ı (z + 1)(z − ı)(z + ı)
(α + βz) ln(z)
= lim
z→−ı (z + 1)(z − ı)
(α − ıβ) ln(−ı)
=
(−ı + 1)(−ı − ı)
(α − ıβ) ln[exp(3ıπ/2)]
= −
(1 − ı)2ı
3ıπ(α − ıβ)
= −
4ı(1 − ı)
3π(α − ıβ)(1 + ı)
= −
8
3π[(α + β) + ı(α − β)]
= − .
8

Thus we have for the integral I¯

I¯ = 2πı(b1⊖1 + b1⊕ı + b1⊖ı )


RESIDUES ON RIEMANN SURFACES 293

ıπ(α − β)
= 2πı +
2

π[(α + β) − ı(α − β)] 3π[(α + β) + ı(α − β)]
+ −
8 8
    
1 1 3 1 3
= 2πı ıπ(α − β) − − + π(α + β) −
2 8 8 8 8
α+β
= −π 2 ı .
2
We have therefore for the original real integral


I =
−2πı
−π 2 ı α + β
=
−2πı 2
π(α + β)
= .
4
As expected, this is a real result, which is positive for positive α and β, a
fact that serves as a partial confirmation of the calculations. We can make
another partial verification of the correctness of the result since in the case
α = 1, β = 1 this integral reduces to one which was calculated in the text of
the previous chapter, and that actually has the value π/2 as is the case here.

Problem 5. Calculate by residues the real asymptotic integral given by


Z ∞

I= dx ,
0 (x + 1)2

for a real constant α in the open interval (0, 1). Justify explicitly every step
of the procedure.
Answer: πα/[sin(πα)].

Complete Solution:

We must calculate the integral




Z
I = dx
1 + 2x + x2
Z0 ∞

= dx .
0 (x + 1)2
294 SOLUTIONS 16

Since the integrand has a branch point at z = 0, we consider the complex


integral

I
I¯ = dz
1 + 2z + z 2
IC

= dz
C (z + 1)2
eα ln(z)
I
= dz ,
C (z + 1)2
on the closed contour C described in the text. Note that here we have a
double pole. Let us now discuss the two arcs. On the segment Cε we have
z = ε exp(ıθ) and thus dz = ıε exp(ıθ)dθ, and therefore we obtain
Z 2π

I¯Cε = dz
0 1 + 2z + z 2
Z 2π
≈ dz z α
0
Z 2π
= ı dθ ε eıθ εα eıαθ
0
Z 2π 
= ı dθ eı(1+α)θ
ε1+α .
0
Taking now the limit we obtain
Z 2π 
lim I¯Cε = ı dθ e
ı(1+α)θ
lim ε1+α .
ε→0 0 ε→0

The remaining integral is limited and the limit shown is zero, so that we have

lim I¯Cε = 0.
ε→0

On the segment CR we have z = R exp(ıθ) and thus dz = ıR exp(ıθ)dθ, and


therefore we obtain
Z 2π

I¯CR = dz
0 1 + 2z + z 2
Z 2π

≈ dz 2
0 z
Z 2π
Rα eıαθ
= ı dθ R eıθ 2 2ıθ
0 R e
Z 2π 
−ı(1−α)θ 1
= ı dθ e 1−α
.
0 R
RESIDUES ON RIEMANN SURFACES 295

Taking now the limit we obtain


Z 2π 
¯ −ı(1−α)θ 1
lim ICR = ı dθ e lim .
R→∞ 0 R→∞ R1−α

The remaining integral is limited and the limit shown is zero, since 0 < α < 1,
so that we have

lim I¯CR = 0.
R→∞

The discussion of the two straight segments of the contour is exactly the same
as the one that was made explicitly in the text in a case similar to this one,
from which it results that we have for the real integral

I= .
1 − e2ıπα
All that remains to be done here is the calculation of I¯ by means of the
residue theorem. Since the relevant singularity is a pole of order two, we
consider the function φ(z) given by

φ(z) = (z + 1)2
(z + 1)2
= zα.

We have therefore that the residue b1 is the Taylor coefficient of order 2−1 = 1
of this function, around the point z = −1, that is, we have that
dz α

b1 =
dz −1

Calculating the derivative we obtain



α−1
b1 = αz
−1

(α−1) ln(z)
= αe
−1
(α−1) ln(−1)
= αe
= α e(α−1) ln[exp(ıπ)]
= α e(α−1)ıπ
= α e−ıπ eıπα
= −α eıπα .
296 SOLUTIONS 16

Thus we have for the integral I¯

I¯ = −2ıπα eıπα .

It follows therefore that we have for the original real integral


I =
1 − e2ıπα
−2ıπα eıπα
=
1 − e2ıπα
2ıπα
=
e ıπα − e−ıπα
πα
= .
sin(πα)
As expected, this is a real, positive and finite result, since 0 < α < 1, which
serves as a partial confirmation of the calculations. Furthermore, in the case
α = 0 the result above exists and is equal to one, as we can see by taking
the limit α → 0. In this case the integral can be done by elementary means,
resulting in fact the value one, which once again confirms the correctness of
the result.

Problem 6. Calculate by residues the real asymptotic integral given by


Z ∞

I= dx ,
0 1 + βx + x2
for real constants 0 < α < 1 and 0 ≤ β ≤ 2. Justify explicitly every step of
the procedure. Discuss in particular the cases β = 0 and β = 2.
Answer:
sin[α(π − θ⊕ )]
I = π ,
sin(πα) sin(θ⊕ )
p
sin(θ⊕ ) = 1 − (β/2)2 ,
cos(θ⊕ ) = −β/2.

Complete Solution:

We must calculate the integral




Z
I= dx .
0 1 + βx + x2
RESIDUES ON RIEMANN SURFACES 297

Since the integrand has a branch point at z = 0, we consider the complex


integral

I
I¯ = dz
C 1 + βz + z 2
eα ln(z)
I
= dz
C 1 + βz + z 2
eα ln(z)
I
= dz ,
C (z − z⊕ )(z − z⊖ )

in the closed contour C described in the text. By the Baskara formula the
two roots of the polynomial in the denominator are given by
p
−β ± β 2 − 4
z± =
2p
−β ± ı 4 − β 2
= ⇒
2s
 2
β β
z⊕ = − + ı 1 − ,
2 2
s  2
β β
z⊖ = − − ı 1 − ,
2 2

since 0 ≤ β ≤ 2. Note that the two roots, and therefore the two singularities
of the integrand, are located on the unit circle, with the same negative real
parts and imaginary parts of opposite signs.
Let us now discuss the two circular segments of the contour. On the
segment Cε we have z = ε exp(ıθ) and thus dz = ıε exp(ıθ)dθ, and therefore
we obtain
Z 2π

I¯Cε = dz
0 1 + βz + z 2
Z 2π
≈ dz z α
0
Z 2π
= ı dθ ε eıθ εα eıαθ
0
Z 2π 
= ı dθ eı(1+α)θ
ε1+α .
0

Taking now the ε → 0 limit we obtain


298 SOLUTIONS 16

Z 2π 
lim I¯Cε = ı dθ e
ı(1+α)θ
lim ε1+α .
ε→0 0 ε→0

The remaining integral is limited and the limit shown is zero, so that we have

lim I¯Cε = 0.
ε→0

On the segment CR we have z = R exp(ıθ) and thus dz = ıR exp(ıθ)dθ, and


thus we obtain
Z 2π
¯ zα
ICR = dz
0 1 + βz + z 2
Z 2π

≈ dz 2
0 z
Z 2π
Rα eıαθ
= ı dθ R eıθ 2 2ıθ
0 R e
Z 2π 
−ı(1−α)θ 1
= ı dθ e 1−α
.
0 R

Taking now the R → ∞ limit we obtain


Z 2π 
1
lim I¯CR = ı dθ e−ı(1−α)θ lim .
R→∞ 0 R→∞ R1−α

The remaining integral is limited and the limit shown is zero, since 0 < α < 1,
so that we have

lim I¯CR = 0.
R→∞

The discussion of the two straight segments of the contour is exactly the same
as the one that was made explicitly in the text in a case similar to this one,
from which it results that we have for the real integral


I= .
1 − e2ıπα

All that remains to be done here is the calculation of I¯ by means of the


residue theorem. In the case β = 2 we have a double pole whose residue
was already calculated in the previous problem, b1 = −α exp(ıπα). In any
RESIDUES ON RIEMANN SURFACES 299

other case we have two simple poles and we can calculate the residues by the
method of the limit, thus obtaining

b1⊕ = lim (z − z⊕ )
z→z⊕ (z − z⊕ )(z − z⊖ )

= lim
z→z⊕ (z − z⊖ )
α
z⊕
= ,
z⊕ − z⊖

b1⊖ = lim (z − z⊖ )
z→z⊖ (z − z⊕ )(z − z⊖ )

= lim
z→z⊖ (z − z⊕ )
α
z⊖
= .
z⊖ − z⊕
We have therefore for the sum of the two relevant residues for the calculation
¯
of the integral I,
α − zα
z⊕ ⊖
b1⊕ + b1⊖ =
z⊕ − z⊖
eα ln(z⊕ ) − eα ln(z⊖ )
= .
z⊕ − z⊖

We now observe that we can write z⊕ as eıθ⊕ , where


s  2
β
sin(θ⊕ ) = 1− ,
2
β
cos(θ⊕ ) = − ,
2
and that for z⊖ , which is the complex conjugate of z⊕ , we can write eıθ⊖ ,
where
s  2
β
sin(θ⊖ ) = − 1 − ,
2
β
cos(θ⊖ ) = − .
2
The fact that z⊖ = z⊕ ∗ is equivalent to the relation θ = 2π − θ between
⊖ ⊕
the angles, so that we can write for the sum of the residues
300 SOLUTIONS 16

eıαθ⊕ − eıαθ⊖
b1⊕ + b1⊖ =
eıθ⊕ − eıθ⊖
eıαθ⊕ − eıα(2π−θ⊕ )
=
eıθ⊕ − eı(2π−θ⊕ )
e ⊕ − e2ıπα e−ıαθ⊕
ıαθ
=
eıθ⊕ − e−ıθ⊕
e−ıπα eıαθ⊕ − eıπα e−ıαθ⊕
= eıπα
eıθ⊕ − e−ıθ⊕
eıα(θ ⊕ −π) − e−ıα(θ⊕ −π)
= eıπα
eıθ⊕ − e−ıθ⊕
sin[α(π − θ⊕ )]
= − eıπα .
sin(θ⊕ )
Note that in the limit β → 2, which corresponds to θ⊕ → π, in which we
have

sin(θ⊕ ) ≈ π − θ⊕ ,
sin[α(π − θ⊕ )] ≈ α(π − θ⊕ ),

we obtain for this sum of residues


sin[α(π − θ⊕ )]
lim (b1⊕ + b1⊖ ) = − eıπα lim
β→2 β→2 sin(θ⊕ )
α(π − θ⊕ )
= − eıπα lim
β→2 (π − θ⊕ )
= −α eıπα
= b1 .

Thus we see that in this limit, in which the two simple poles are united at the
position z = −1, we recover the residue of the double pole b1 calculated in
the previous problem. Returning to the general case we have for the integral

sin[α(π − θ⊕ )]
I¯ = −2ıπ eıπα .
sin(θ⊕ )
It follows therefore that we have for the original real integral

I =
1 − e2ıπα
eıπα sin[α(π − θ⊕ )]
= −2ıπ
1 − e2ıπα sin(θ⊕ )
RESIDUES ON RIEMANN SURFACES 301

2ı sin[α(π − θ⊕ )]
= −π
e−ıπα
−e ıπα sin(θ⊕ )
sin[α(π − θ⊕ )]
= π .
sin(πα) sin(θ⊕ )
As expected, this is a real and positive result and, since 0 < α < 1 and
0 ≤ β ≤ 2, implying that π/2 ≤ θ⊕ ≤ π. This serves as partial confirmation
of the calculations. These inequalities also imply that the result is obviously
finite in all cases except in the case β = 2, which corresponds to θ⊕ = π.
In this particular case we consider again the limit in which β → 2 and thus
θ⊕ → π, in which we have, as we have seen before,
sin[α(π − θ⊕ )]
lim I = π lim
β→2 β→2 sin(πα) sin(θ⊕ )
α(π − θ⊕ )
= π lim
β→2 sin(πα)(π − θ⊕ )
πα
= .
sin(πα)
This is in fact the same result that was obtained in the previous problem for
this case. The case in which β = 0 and thus θ⊕ = π/2 can be calculated
directly, and we obtain in this case
sin(απ/2)
I = π
sin(πα)
sin(απ/2)
= π
2 sin(απ/2) cos(απ/2)
π
= .
2 cos(απ/2)
This is in fact the same result that was obtained in the text for this case.

Problem 7. Calculate by residues the real asymptotic integral given by


Z ∞

I= dx ,
0 1 + 2x2 + x4
for a real constant α in the open interval (0, 3). Justify explicitly every step
of the procedure.
Answer: −π(α − 1)/[4 cos(πα/2)].

Complete Solution:
302 SOLUTIONS 16

We must calculate the integral




Z
I= dx .
0 1 + 2x2 + x4

Since the integrand has a branch point at z = 0, we consider the complex


integral

I
¯
I = dz
C 1 + 2z 2 + z 4
eα ln(z)
I
= dz
C (1 + z 2 )2
eα ln(z)
I
= dz ,
C (z − ı)2 (z + ı)2

in the closed contour C described in the text, where we show the two double
poles of the integrand.
Let us now discuss the two circular segments of the contour. On the
segment Cε we have z = ε exp(ıθ) and thus dz = ıε exp(ıθ)dθ, and thus we
obtain
Z 2π
¯ zα
ICε = dz
0 1 + 2z 2 + z 4
Z 2π
≈ dz z α
0
Z 2π
= ı dθ ε eıθ εα eıαθ
0
Z 2π 
= ı dθ eı(1+α)θ
ε1+α .
0

Taking now the ε → 0 limit we obtain


Z 2π 
lim I¯Cε = ı dθ eı(1+α)θ
lim ε1+α .
ε→0 0 ε→0

The remaining integral is limited and the limit shown is zero, so that we have

lim I¯Cε = 0.
ε→0

On the segment CR we have z = R exp(ıθ) and thus dz = ıR exp(ıθ)dθ, and


thus we obtain
RESIDUES ON RIEMANN SURFACES 303



Z
I¯CR = dz
0 1 + 2z 2 + z 4
Z 2π

≈ dz 4
0 z
Z 2π
Rα eıαθ
= ı dθ R eıθ 4 4ıθ
0 R e
Z 2π 
1
= ı dθ e−ı(3−α)θ 3−α
.
0 R

Taking now the R → ∞ limit we obtain


Z 2π 
1
lim I¯CR = ı dθ e−ı(3−α)θ lim .
R→∞ 0 R→∞ R3−α

The remaining integral is limited and the limit shown is zero, since 0 < α < 3,
so that we have

lim I¯CR = 0.
R→∞

The discussion of the two straight segments of the contour is exactly the same
as the one that was made explicitly in the text in a case similar to this one,
from which it results that we have for the real integral


I= .
1 − e2ıπα
All that remains to be done here is the calculation of I¯ by means of the
residue theorem. Since the two relevant singularities are poles of order two,
we consider in the case of the pole z⊕ = ı the function φ⊕ (z) given by


φ⊕ (z) = (z − ı)2
(z − ı)2 (z + ı)2

= .
(z + ı)2

We have then that the residue b1⊕ is the Taylor coefficient of order 2 − 1 = 1
of this function, around the point z = ı, that is, we have that



d
b1⊕ = .
dz (z + ı)2 ı

Calculating the derivative we obtain


304 SOLUTIONS 16

αz α−1 2z α
 
b1⊕ = −
(z + ı)2 (z + ı)3 ı#
"
α e(α−1) ln(z) 2 eα ln(z)

= −
(z + ı)2 (z + ı)3 ı
α e(α−1) ln[exp(ıπ/2)] 2 eα ln[exp(ıπ/2)]
= −
(2ı)2 (2ı)3
α eıπ(α−1)/2 eıπα/2
= − −ı
4 4
ı
= (α − 1) eıπα/2
.
4

In the case of the pole z⊖ = −ı we consider the function φ⊖ (z) given by


φ⊖ (z) = (z + ı)2
(z − ı)2 (z + ı)2

= .
(z − ı)2

We have then that the residue b1⊖ is the Taylor coefficient of order 2 − 1 = 1
of this function, around the point z = −ı, that is, we have that



d
b1⊖ = .
dz (z − ı)2 −ı

Calculating the derivative we obtain

αz α−1 2z α
 
b1⊖ = −
(z − ı)2 (z − ı)3 −ı
" #
α e(α−1) ln(z) 2 eα ln(z)
= −
(z − ı)2 (z − ı)3 −ı

α e(α−1) ln[exp(3ıπ/2)] 2 eα ln[exp(3ıπ/2)]


= 2

(−2ı) (−2ı)3
α e3ıπ(α−1)/2 e3ıπα/2
= − +ı
4 4
ı 3ıπα/2
= − (α − 1) e .
4
We have therefore for the sum of the two residues that must be used to
calculate the integral I¯
RESIDUES ON RIEMANN SURFACES 305

ıh i
b1⊕ + b1⊖ = (α − 1) eıπα/2 − (α − 1) e3ıπα/2
4
ı(α − 1) eıπα h −ıπα/2 i
= e − eıπα/2
4
(α − 1) eıπα  πα 
= sin .
2 2
Thus we have for the integral I¯

I¯ = 2ıπ(b1⊕ + b1⊖ )
 πα 
= ıπ(α − 1) e ıπα
sin .
2
It follows therefore that we have for the original real integral

I =
1 − e2ıπα
eıπα  πα 
= ıπ(α − 1) sin
1 − e2ıπα 2
1  πα 
= ıπ(α − 1) −ıπα sin
e  − eıπα 2
πα
π(α − 1) sin 2
= −
2 sin(πα)
 πα 
π(α − 1) sin
= −  πα  2  πα 
4 sin cos
2 2
π(α − 1)
= −  πα  .
4 cos
2
As expected, this is a real result. In addition, although not obvious at first
glance, it can be verified that it is positive for 0 < α < 3. In order to do this,
we separate this interval into two parts, one in which 0 < α < 1 and another
in which 1 < α < 3. The point α = 1 needs to be considered separately, in
which case both the numerator and the denominator vanish. For 0 < α < 1
the numerator is strictly negative and argument of the cosine varies from
0 to π/2, so that the cosine is strictly positive. It follows that the result is
strictly positive in this case. For 1 < α < 3 the numerator is strictly positive,
and the argument of the cosine varies from π/2 to 3π/2, so that the cosine
is strictly negative. It follows once again that the result is strictly positive
in this case. In order to examine the case α = 1 we take the limit α → 1,
getting
306 SOLUTIONS 16
 πα 
π(α − 1) sin
lim I = − lim 2
α→1 α→1 2 sin(πα)
π(α − 1) 1
= − lim
α→1 2 (π − πα)
1
= .
2
Again, we have a positive result. It follows that we have a strictly positive
result for every value of α in (0, 3). This serves as a partial confirmation of
the result, since the real original integral is clearly positive.

Problem 8. (Challenge Problem) Calculate by residues the real asymp-


totic integral given by
Z ∞
ln(x)
I= dx 2 .
0 x +1
Justify explicitly every step of the procedure.
Answer: 0.
Hint: consider a contour contained in the upper half-plane.

Complete Solution:

We must calculate the integral


Z ∞
ln(x)
I = dx 2
x +1
Z0 ∞
ln(x)
= dx .
0 (x − ı)(x + ı)
Since the integrand has a branch point at z = 0, we consider the complex
integral
ln(z)
I
I¯ = dz 2
z +1
IC
ln(z)
= dz ,
C (z − ı)(z + ı)
in the closed contour C described in Figure 16.1, contained in the upper
half-plane. Note that it is not necessary to consider in detail the cut, which
should only be taken out of the way.
RESIDUES ON RIEMANN SURFACES 307

CR
C
ı

C⊖ Cε C⊕
ε R
x
−ı

Figure 16.1: The integration contour in the complex plane, showing also the
branch cut.

Let us now discuss the two semicircular segments. On the segment Cε we


have z = ε exp(ıθ) and thus dz = ıε exp(ıθ)dθ, and thus we obtain
Z π
ln(z)
I¯Cε = dz 2
z +1
Z0 π
≈ dz ln(z)
0
Z π
= ı dθ ε eıθ [ln(ε) + ıθ]
0
Z π  Z π 
= ı dθ e ıθ
ε ln(ε) − dθ θ eıθ
ε.
0 0

Taking now the ε → 0 limit we obtain


Z π  Z π 
lim I¯Cε = ı dθ eıθ
lim ε ln(ε) − dθ θ e ıθ
lim ε.
ε→0 0 ε→0 0 ε→0

The remaining integrals are limited, and the limits shown are zero, so that
we have

lim I¯Cε = 0.
ε→0

On the segment CR we have z = R exp(ıθ) and thus dz = ıR exp(ıθ)dθ, and


thus we obtain
308 SOLUTIONS 16

π
ln(z)
Z
I¯CR = dz
z2 + 1
Z0 π
ln(z)
≈ dz 2
0 z
π
ln(R) + ıθ
Z
= ı dθ R eıθ
R2 e2ıθ
0Z π  Z π 
−ıθ ln(R) −ıθ 1
= ı dθ e − dθ θ e .
0 R 0 R

Taking now the R → ∞ limit we obtain


Z π  Z π 
¯ −ıθ ln(R) −ıθ 1
lim ICR = ı dθ e lim − dθ θ e lim .
R→∞ 0 R→∞ R 0 R→∞ R

The remaining integrals are limited, and the limits shown are zero, so that
we have

lim I¯CR = 0.
R→∞

Let us now consider the segment C⊕ of the contour. In this case we have
z = x and dz = dx, so that we obtain immediately
R
ln(z)
Z
I¯C⊕ = dz
ε z2 + 1
Z R
ln(x)
= dx 2 .
ε x +1

In the limit in which ε → 0 and R → ∞ this reduces to our original real


integral. Let us now consider the segment C⊖ of the contour. In this case we
also have z = x and dz = dx, so that we obtain
Z −ε
¯ ln(z)
IC⊖ = dz 2
−R z +1
Z −ε
ln(x)
= dx 2 ,
−R x +1

where the signs and the order of the extremes are exchanged in relation to our
original real integral. Of course, all values of x are negative in this interval,
which would be a problem for the real version of the logarithm, but is not
a problem here because this is the complex version of this function. Let us
make the transformation of variables x → −x, to obtain
RESIDUES ON RIEMANN SURFACES 309

ε
ln(−x)
Z
I¯C⊖ = − dx
R (−x)2 + 1
R
ln[x exp(ıπ)]
Z
= dx
ε x2 + 1
Z R
ln(x) + ıπ
= dx .
ε x2 + 1
Adding all portions of the integral and taking the limit, we therefore have,
since the integrals on the semicircles vanish,
Z R Z R
¯ ln(x) ln(x) + ıπ
lim lim IC = dx 2 + dx
ε→0 R→∞ ε x + 1 ε x2 + 1
Z ∞ Z ∞
ln(x) 1
= 2 dx 2 + ıπ dx 2
0 x +1 0 x +1
π 2
= 2I + ı ,
2
where the value of the additional integral that appears here was calculated
in the previous chapter, and turns out to be π/2. We can omit the limits on
the left-hand side of the equation, with the understanding that the integral
should be calculated by residues, including only the pole located at z = ı.
We have therefore for our original real integral I,

I¯C π2
I= −ı .
2 4
All that remains to be done here is the calculation of I¯ by means of the
residue theorem. Since the only relevant singularity is a simple pole, we can
get the residue by the method of the limit,
ln(z)
b1+ = lim (z − ı)
z→ı (z − ı)(z + ı)
ln(z)
= lim
z→ı (z + ı)
ln(ı)
=

ln[exp(ıπ/2)]
=

ıπ
=

π
= .
4
310 SOLUTIONS 16

Thus we have for the integral I¯


π
I¯ = 2ıπ
4
ıπ 2
= .
2
It follows therefore that we have for the original real integral

I¯C π2
I = −ı
2 4
ıπ 2 π2
= −ı
4 4
= 0.

As expected, this is a real and finite result. There was no need for the
result to be positive, since the logarithm changes sign within the integration
interval. Here we see that the negative part of the integral in (0, 1) exactly
cancels the positive part in (1, ∞).
Bibliography

[1] J. L. deLyra, Complex Calculus, vol. 1 of Mathematical Methods for


Physics and Engineering. Amazon, first ed., 2018. ISBN-13: 978-
1793012050.

311
Index

Chapter 1, 1 Problem 5, 55
Problem 1, 1 Problem 6, 57
Problem 2, 1 Problem 7, 59
Problem 3, 2 Problem 8, 61
Problem 4, 3 Chapter 5, 63
Problem 5, 4 Problem 1, 63
Problem 6, 8 Problem 2, 64
Problem 7, 11 Problem 3, 66
Problem 8, 13 Problem 4, 67
Chapter 2, 15 Problem 5, 71
Problem 1, 15 Chapter 6, 75
Problem 2, 16 Problem 1, 75
Problem 3, 18 Problem 2, 80
Problem 4, 22 Problem 3, 81
Problem 5, 25 Problem 4, 84
Problem 6, 26 Problem 5, 89
Problem 7, 27 Chapter 7, 95
Problem 8, 28 Problem 1, 95
Problem 9, 30 Problem 2, 96
Problem 10, 32 Problem 3, 97
Chapter 3, 35 Problem 4, 99
Problem 1, 35 Problem 5, 101
Problem 2, 37 Problem 6, 102
Problem 3, 38 Problem 7, 106
Problem 4, 40 Chapter 8, 109
Problem 5, 41 Problem 1, 109
Problem 6, 43 Problem 2, 112
Chapter 4, 47 Problem 3, 113
Problem 1, 47 Problem 4, 114
Problem 2, 48 Problem 5, 115
Problem 3, 51 Problem 6, 117
Problem 4, 54 Problem 7, 119

312
Problem 8, 122 Problem 2, 216
Problem 9, 125 Problem 3, 218
Problem 10, 127 Problem 4, 221
Problem 11, 128 Chapter 14, 223
Problem 12, 130 Problem 1, 223
Chapter 9, 137 Problem 2, 225
Problem 1, 137 Problem 3, 226
Problem 2, 138 Problem 4, 228
Problem 3, 142 Chapter 15, 231
Problem 4, 143 Problem 1, 231
Problem 5, 144 Problem 2, 234
Problem 6, 146 Problem 3, 237
Problem 7, 148 Problem 4, 240
Chapter 10, 151 Problem 5, 246
Problem 1, 151 Problem 6, 251
Problem 2, 152 Problem 7, 256
Problem 3, 154 Problem 8, 261
Problem 9, 263
Problem 4, 155
Problem 10, 266
Problem 5, 157
Problem 11, 269
Problem 6, 159
Problem 12, 271
Chapter 11, 169
Chapter 16, 281
Problem 1, 169
Problem 1, 281
Problem 2, 171
Problem 2, 284
Problem 3, 173
Problem 3, 287
Problem 4, 175
Problem 4, 289
Problem 5, 178 Problem 5, 293
Problem 6, 181 Problem 6, 296
Problem 7, 183 Problem 7, 301
Problem 8, 186 Problem 8, 306
Chapter 12, 189
Problem 1, 189
Problem 2, 190
Problem 3, 192
Problem 4, 194
Problem 5, 196
Problem 6, 201
Problem 7, 205
Chapter 13, 213
Problem 1, 213

313

You might also like