Antimirov A.A., Kolyshkin R. - Vaillancourt Complex Variables (1998) EXCELENTE
Antimirov A.A., Kolyshkin R. - Vaillancourt Complex Variables (1998) EXCELENTE
Antimirov A.A., Kolyshkin R. - Vaillancourt Complex Variables (1998) EXCELENTE
Preface
xi
Chapter 1. Functions of a Complex Variable 1
1.1. Complex numbers
Exercises for Section 1.1
1.2. Continuity in the complex plane
Exercises for Section 1.2
1.3. Functions of a complex variable
Exercises for Section 1.3
1.4. Analytic functions
Exercises for Section 1.4
1.5. Elementary analytic functions
Exercises for Section 1.5
vii
viii PREFACE
The authors express their thanks to Dr. Thierry Giordano, who has
used the manuscript as lecture notes since 1995. He has made many valu-
able suggestions for improving the first part of the book.
Mr. André Montpetit of the Centre de recherches mathématiques of the
Université de Montréal has been generous in offering invaluable assistance
for the composition of the text in AMS − LaT eX.
This book benefitted from the supports of the Natural Sciences and
Engineering Council of Canada, the University of Ottawa, Riga Technical
University, and the Centre de recherches mathématiques of the Université
de Montréal.
The authors express their warmest thanks to the dynamic and collab-
orative editorial and production team of Academic Press Inc.
M. Ya. Antimirov
A. A. Kolyshkin
Rémi Vaillancourt
Riga, Ottawa, 24 November 1997
CHAPTER 1
For clarity, the expressions z-plane and w-plane will be used to mean
z ∈ C and w ∈ C, respectively, when referring to different copies of C.
Two complex numbers, z1 = (x1 , y1 ) and z2 = (x2 , y2 ), are equal,
written z1 = z2 , if and only if their real and imaginary parts are equal; that
is, if and only if x1 = x2 and y1 = y2 .
z = z1 + z2 = (x1 + x2 , y1 + y2 ).
z1 + z2 = z2 + z1 ,
z1 + (z2 + z3 ) = (z1 + z2 ) + z3 ,
follow from Definition 1.1.2. The complex number zero, 0 = (0, 0), such
that z + 0 = z for all z ∈ C, is introduced in the same way as the real
number 0 in the set of real numbers.
z = z1 z2 = (x1 x2 − y1 y2 , x1 y2 + x2 y1 ).
1
2 1. FUNCTIONS OF A COMPLEX VARIABLE
A
y z = x + iy
r
θ = Arg z
0 x x
−→
Figure 1.1. The vector OA = (x, y) identified with the
complex number z = x + iy.
−→
The length r of the vector OA is called the modulus of the complex number
z and is denoted by |z|,
−→ p
|z| = |OA| = r = x2 + y 2 ≥ 0. (1.1.6)
The angle arg z is usually taken in one of the half-open intervals,
(2k − 1)π < arg z ≤ (2k + 1)π, k = 0, ±1, ±2, . . . , (1.1.7)
or
2kπ ≤ arg z < 2(k + 1)π, k = 0, ±1, ±2, . . . . (1.1.8)
The principal value of the argument of z is defined to be the angle Arg z
such that
y
tan(Arg z) = , −π < Arg z ≤ π, (1.1.9)
x
by taking k = 0 in (1.1.7), or
y
tan(Arg z) = , 0 ≤ Arg z < 2π, (1.1.10)
x
by taking k = 0 in (1.1.8).
In this book, the choice of (1.1.9) or (1.1.10) will be dictated by each
problem in hand and should be clear from the context. Generally, (1.1.9)
is used in Chapters 1, 3, 4 and 5, and (1.1.10) is used in Chapters 2, 6, 7
and 8. Most computers use the the principal value given by (1.1.9).
With the choice (1.1.9), there are three cases to be considered for Arg z:
y
(a) If x > 0 (see Fig 1.2), Arg z = Arctan .
x
y
(b) If x < 0 and y > 0 (see Fig 1.3), Arg z = Arctan + π.
x
y
(c) If x < 0 and y < 0 (see Fig 1.4), Arg z = Arctan − π.
x
1.1. COMPLEX NUMBERS 5
y y
y x
z = x + iy 0 Arg z = Arctan _x
y
Arg z = Arctan _x z = x + iy
0 x
y
z = x + iy
Arg z = Arctan _xy + π
x
0 Arctan _xy
Arctan _xy
0 x
Arg z = Arctan _xy – π
z = x + iy
y
A
z1 + z 2
A1
z1
z2 – z 1 A2
z2
0 x
B
z2 – z 1
−→
Figure 1.5. Geometric representation of the sum, OA,
−→ −→
and difference, OB = A1 A2 , of two complex numbers.
−→
OA joins the beginning and the end of the polygonal line OA1 A2 · · · An . It
follows from Fig 1.5 and formulae (1.1.12) and (1.1.13) that
−→ −→ −→ −→ −→ −→ −→
OA ≤ OA1 + OA2 , OA ≤ OA1 + OA2 + · · · + OAn ,
that is, we have the triangle inequality,
|z1 + z2 | ≤ |z1 | + |z2 |, (1.1.14)
and its generalization to n numbers,
|z1 + z2 + · · · + zn | ≤ |z1 | + |z2 | + · · · + |zn |.
These inequalities can be written in the short form
n n
X X
z
k ≤ |zk | . (1.1.15)
k=1 k=1
Equality in (1.1.14) and (1.1.15) holds only if all the complex numbers
zk lie on the same straight line in the complex plane.
Inequality (1.1.15) is basic for estimating the moduli of sums of complex
numbers and integrals of functions of a complex variable.
On the other hand, since z2 − z1 = (x2 − x1 ) + i(y2 − y1 ), then the
−→
vector OB = (x2 − x1 , y2 − y1 ) corresponds to the complex number z2 − z1 .
In this case,
−→ −→ −→ −→
OB = A1 A2 = OA2 − OA1 , (1.1.16)
−→
that is, the vector OB corresponds to the difference of the given complex
−→ −→
numbers and is represented by a difference of the vectors OA2 and OA1 .
It follows from Fig 1.5 and formula (1.1.16) that
−→ p
|z2 − z1 | = |A1 A2 | = (x2 − x1 )2 + (y2 − y1 )2 , (1.1.17)
that is, the modulus of the difference, z2 − z1 of two complex numbers is
equal to the distance between the points z1 and z2 in the complex plane.
Since the distance in C and R2 is given by the same formula, it will be seen
in the next subsections that the definition of a neighborhood of a point, the
set of interior or exterior points of a disk in C, etc., will be the same as in
R2 . Hence C and R2 have the same notions of continuity and limit, that is,
the same topology.
For example, if z0 = x0 + iy0 = constant and ρ = constant > 0, then
the formula
|z − z0 | = ρ (1.1.18)
represents the geometric locus of all the points z which are at distance ρ
from the point z0 . Thus (1.1.18) is the equation of a circle centered at z0
and of radius ρ (see Fig 1.6). If z = x + iy and z0 = x0 + iy0 , it follows
8 1. FUNCTIONS OF A COMPLEX VARIABLE
y
|z – z 0|> ρ
ρ
z0
|z – z 0|< ρ
0 x
that is, the modulus of the product of two complex numbers is equal to the
product of their moduli, while the argument of the product is equal to the
sum of their arguments. It can easily be proved by mathematical induction
that relations similar to (1.1.22) and (1.1.23) hold for any finite number of
complex numbers:
X n X
n
z1 z2 · · · zn = r1 r2 · · · rn cos θk + i sin θk ; (1.1.24)
k=1 k=1
thus
n
X
|z1 z2 · · · zn | = |z1 | |z2 | · · · |zn |, arg (z1 z2 · · · zn ) = arg zk . (1.1.25)
k=1
Similarly, if z2 6= 0,
z1 r1 cos θ1 + i sin θ1
=
z2 r2 cos θ2 + i sin θ2
r1 (cos θ1 + i sin θ1 )(cos θ2 − i sin θ2 )
= .
r2 cos2 θ2 + sin2 θ2
Multiplying the numerator out and applying trigonometric identities for
difference of angles, we have
z1 r1
= [cos (θ1 − θ2 ) + i sin (θ1 − θ2 )] . (1.1.26)
z2 r2
It follows from (1.1.26) that
z1 |z1 | z1
= = arg z1 − arg z2 ,
z2 |z2 | , arg
z2
(1.1.27)
that is, the modulus of the ratio of two complex numbers is equal to the
ratio of their moduli, and the argument of the ratio is equal to the difference
of their arguments.
Letting z1 = z2 = · · · = zn = z = r(cos θ + i sin θ) in (1.1.24), we obtain
z n = rn (cos nθ + i sin nθ); (1.1.28)
thus
|z n | = |z|n , arg z n = n arg z. (1.1.29)
1.1.5. Exponential form of complex numbers. We introduce at
this point a third form of complex numbers, called the exponential form,
even though the exponential function for a complex variable will be defined
later in Subsection 1.5.1.
Thus, to avoid breaking the logical order of presentation, we introduce
Euler’s formula,
eiθ = cos θ + i sin θ, (1.1.30)
10 1. FUNCTIONS OF A COMPLEX VARIABLE
and postpone its derivation, as (1.5.9), until Subsection 1.5.1. We shall also
assume the law of exponents (1.5.11) in the form
eiθ1 eiθ2 = ei(θ1 +θ2 ) ,
which will be proved later.
Substituting (1.1.30) in (1.1.20), we have the following definition.
Definition 1.1.8. The exponential form of the complex number
z = r(cos θ + i sin θ)
is
z = r eiθ or z = |z| ei arg z . (1.1.31)
Relations (1.1.22)–(1.1.29) can be easily obtained by means of (1.1.31).
For example, if z1 = |z1 | ei arg z1 and z2 = |z2 | ei arg z2 , then
z1 z2 = |z1 | |z2 | ei(arg z1 +arg z2 ) . (1.1.32)
Formulae (1.1.22) and (1.1.23) follow from (1.1.32).
1.1.6. Powers and roots of complex numbers.
Definition 1.1.9. Given n ∈ N, the complex number w = z 1/n is
called an nth root of the complex number z if wn = z.
We have the following theorem.
Theorem 1.1.1. A nonzero complex number z = r(cos θ + i sin θ) has
exactly n distinct nth roots given by the formula
Arg z + 2kπ Arg z + 2kπ
z 1/n = |z|1/n cos + i sin ,
n n
k = 0, 1, . . . , n − 1. (1.1.33)
Proof. Given z = r(cos θ+i sin θ) 6= 0, we determine the real numbers
ρ ≥ 0 and ϕ such that
w = ρ(cos ϕ + i sin ϕ) = z 1/n . (1.1.34)
n
It follows from the relation w = z and (1.1.28) that
ρn (cos nϕ + i sin nϕ) = r(cos θ + i sin θ). (1.1.35)
n
Thus ρ = r and
ρ = r1/n , (1.1.36)
1/n
where it is understood that the positive real value of r taken. Moreover,
cos nϕ = cos θ =⇒ nϕ = θ + 2kπ, k = 0, 1, . . . .
Thus,
θ + 2kπ
ϕ = arg w = , k = 0, 1, . . . , n − 1, (1.1.37)
n
EXERCISES FOR SECTION 1.1 11
y ––
1+i√3
–2 0 x
––
1–i√3
3−i
19. .
4 + 2i
√ √ !3
2−i 2
20. .
2
2 3−i
21. + .
1 − i 1 + 2i
3
1−i
22. .
1+i
23. Show that arg(z̄) = − arg(z), where z 6= 0.
24. Find the values of z for which Arg(z̄) = − Arg(z).
Find the complex numbers which are complex conjugates of
25. their own squares.
26. their own cubes.
Prove the following identities.
27. |z̄| = |z|.
28. z1 + z2 = z̄1 + z̄2 .
29. z1 z2 = z̄1 z̄2 .
z1 z̄1
30. = .
z2 z̄2
31. When do three points, z1 , z2 and z3 , lie on a straight line?
32. Let σ be the line segment joining the points z1 and z2 . Find the point
z which divides σ in the ratio λ1 :λ2 .
33. Show that |z1 − z2 |2 = |z1 |2 + |z2 |2 − 2<(z1 z̄2 ).
34. Prove the parallelogram law: |z1 − z2 |2 + |z1 + z2 |2 = 2(|z1 |2 + |z2 |2 ).
35. Let z1 , z2 and z3 be consecutive vertices of a parallelogram. Find the
fourth vertex z4 (opposite to z2 ).
36. Find the point in the complex plane which is symmetric to x + iy with
respect to the line y = x.
√ √
37. By which angle should the vector 3 2 + i2 2 be rotated in order to
obtain the vector −5 + i?
38. Prove the Cauchy–Schwarz inequality
p p
|z1 w1 + z2 w2 | ≤ |z1 |2 + |z2 |2 |w1 |2 + |w2 |2 ,
and generalize it to n terms, that is, |z1 w1 + z2 w2 + · · · + zn wn |.
14 1. FUNCTIONS OF A COMPLEX VARIABLE
√
3
57. 27i.
s √
4 − 3+i
58. .
1−i
s√ √
6 2−i 2
59. √ .
1+i 3
60. Prove that if z1 + z2 + z3 = 0 and |z1 |=|z2 |=|z3 |=1, then the points z1 ,
z2 , z3 are the vertices of an equilateral triangle inscribed in the unit circle
|z| = 1.
61. Let a be any nth root of unity other than 1, where n > 1. Prove that
n
1 + 2a + 3a2 + · · · + nan−1 = .
a−1
(Hint. Multiply by 1 − a.)
62. Prove that the sum of all distinct nth roots of unity is zero, and interpret
this fact geometrically.
S1 z1 z2 S2
A
0 x
y y
Ω Ω
0 x 0 x
(a) (b)
y
0 x
(c)
Note 1.2.1. The inequality |zn − a| < ε means that, for n > Nε , all
the terms of the sequence are located in the open disk Daε of center a and
radius ε.
The limit of a sequence of complex numbers is equivalent to the limit
of two sequences of real numbers as proved in the following theorem.
Theorem 1.2.1. Let {zn } be a sequence of complex numbers. A nec-
essary and sufficient condition for the existence of a limit
a + ib = lim zn , (1.2.7)
n→∞
where zn = an + ibn , is the existence of the limits
a = lim an , b = lim bn . (1.2.8)
n→∞ n→∞
1.2. CONTINUITY IN THE COMPLEX PLANE 19
y y
Ω
r2
r1 Ω
z0
π/4
0 x 0 x
(a) (b)
y y
Ω
Ω
π/3
π/6
2i
0 x 0 2 3 x
(c) (d)
3i
Ω
i
0 x
(e)
Proof. Necessity. Suppose that the limit in (1.2.7) exists, that is,
It follows from the previous theorem that the study of the properties
of sequences, {zn }, of complex numbers can be reduced to study of the
properties of pairs of sequences, {an } and {bn }, of real numbers.
N ε
~
z
y
S
z=0 x
z R
z-plane
points N (0, 0, 2), z̃ and z(x, y, 0), lie on the straight line
x1 − 0 x2 − 0 x3 − 2
= = . (1.2.14)
x−0 y−0 0−2
Expressing x2 and x3 in terms of x1 from the equation of the line and
substituting these values into the equation of the sphere, we obtain
4x
x1 = 2 .
x + y2 + 4
Similarly,
4y 2(x2 + y 2 )
x2 = and x3 = .
x2
+ y2 + 4 x2 + y 2 + 4
Since z = x + iy, we have
2(z + z̄) 2(z − z̄) 2|z|2
x1 = , x2 = , x3 = . (1.2.15)
|z|2 + 4 i(|z|2 + 4) |z|2 + 4
It follows from these formulae that to each (finite) point z = x + iy ∈ C
there corresponds a unique point z̃(x1 , x2 , x3 ) on the Riemann sphere.
Conversely, from equation (1.2.14) of the line, we have
2x1 2x2
x= , y= . (1.2.16)
2 − x3 2 − x3
Hence, to each point z̃(x1 , x2 , x3 ) on the Riemann sphere there corresponds
a unique point z = x + iy ∈ C (except for the north pole, N ).
Therefore, there is a one-to-one correspondence between the points of
the complex z-plane and the points of the sphere. The only point of the
sphere to which there does not correspond any point in the finite part of the
complex z-plane is the north pole. If we let the point z = ∞ correspond
to N , then the exterior of a disk of radius R in C corresponds to an ε-
neighborhood of N where ε decreases as R increases.
22 1. FUNCTIONS OF A COMPLEX VARIABLE
y v
u(x,y) = 0
v(x,y) = 0 u=0
0 x 0 v=0 u
(see Fig 1.12). The function (1.3.1) maps every point z of its domain of
definition in the complex z-plane to some point w of the complex w-plane,
that is, if z0 = x0 + iy0 then
w0 = f (z0 ) = u(x0 , y0 ) + iv(x0 , y0 ).
1.3.2. Limit and continuity of a function of a complex variable.
Firstly, we define the limit of a function, f (z), by means of sequences of
values of f .
Definition 1.3.2. A number w0 is called the limit of a function of a
complex variable, w = f (z), as z → z0 , if for each sequence {zn } converging
to z0 as n → ∞, the corresponding sequence, {f (zn )}, converges to w0 as
n → ∞.
Secondly, we define the limit of a function using the Cauchy “ε-δ”
terminology.
Definition 1.3.3. A number w0 is called the limit of a function w =
f (z) as z → z0 if, for every ε > 0, there exists δz0 ,ε > 0 such that, for all z
satisfying the inequality
|z − z0 | < δz0 ,ε , (1.3.2)
f (z) satisfies the inequality
|f (z) − w0 | < ε. (1.3.3)
In this case, we write
w0 = lim f (z). (1.3.4)
z→z0
It can easily be shown that the previous two definitions are equivalent.
Geometrically, inequality (1.3.2) represents the interior of the disk Dzδ0
in the z-plane while inequality (1.3.3) represents the interior of the disk
ε
Dw 0
in the w-plane (see Fig 1.13). Hence, limz→z0 f (z) = w0 if, for every
ε > 0 there exists δ = δz0 ,ε > 0 such that for all z ∈ Dzδ0 , w = f (z) ∈ Dw
ε
0
.
26 1. FUNCTIONS OF A COMPLEX VARIABLE
y v
ε
δ Dwε
Dzδ w0
0
0
z0
0 x 0 u
Note 1.3.1. It follows from Definition 1.3.3 that the limit of f (z) at z =
z0 does not depend upon the direction of the ray along which z approaches
z0 . If z approaches z0 along any ray, then as soon as it gets into the disk
Dzδ0 , the corresponding values of w gets into the disk Dw ε
0
. This fact will
often be used in this book.
The following theorem relates the convergence of a function of z ∈ C
to the convergence of two functions of (x, y) ∈ R2 .
Theorem 1.3.1. The limit of a complex function f (z) = u(x, y) +
iv(x, y) exists as z → z0 = x0 + iy0 and is equal to
w0 = u0 + iv0 = lim f (z), (1.3.5)
z→z0
if and only if the limits of its real and imaginary parts exist and are equal
to
u0 = lim u(x, y), v0 = lim v(x, y). (1.3.6)
(x,y)→(x0 ,y0 ) (x,y)→(x0 ,y0 )
Proof. (1) Suppose that the limit in (1.3.5) exists, that is, inequality
(1.3.3) is satisfied for all z satisfying (1.3.2). We rewrite (1.3.2) and (1.3.3)
in the form
p
(x − x0 )2 + (y − y0 )2 < δ (1.3.7)
and
p
(u − u0 )2 + (v − v0 )2 < ε, (1.3.8)
respectively. It follows from (1.3.8) that
|u − u0 | < ε, |v − v0 | < ε, (1.3.9)
for all (x, y) satisfying (1.3.7). But (1.3.9) implies the existence of limits
(1.3.6).
EXERCISES FOR SECTION 1.3 27
(2) Suppose that the limits in (1.3.6) exist, that is, for every ε > 0 there
δ
exists δ > 0 such that for all (x, y) ∈ D(x 0 ,y0 )
the following inequalities are
fulfilled:
ε ε
|u − u0 | < √ , |v − v0 | < √ . (1.3.10)
2 2
Inequality (1.3.8) follows from (1.3.10) for all (x, y) ∈ Dzδ0 . Hence the limit
in (1.3.5) exists.
Definition 1.3.4. A function w = f (z) is said to be continuous at the
point z0 = x0 + iy0 if
f (z0 ) = lim f (z). (1.3.11)
z→z0
1−z
8. f (z) = .
1+z
9. f (z) = z̄ − iz 2 .
10. Let z = x + iy. Express the right-hand side of f (z) = x2 − y 2 − 2y +
i(2x − 2y) in terms of z and simplify.
Find the following limits.
iz 3 − 8
11. lim .
z→2i z − 2i
z 2 + 4z − 21
12. lim .
z→3 z−3
13. lim |z|.
z→2−3i
z 2 + 3z + 2
14. lim .
z→∞ 4z 2 + 2z − 1
Find the following limits, if they exist.
|z|
15. lim .
z→0 z
|z|2
16. lim .
z→0 z
z
17. lim .
z→0 |z|
z − <z
18. lim .
z→0 =z
19. Consider the rational function
am z m + · · · + a1 z + a0
f (z) = , am 6= 0, bn 6= 0,
bn z n + · · · + b1 z + b0
and discuss the possible values of lim f (z).
z→∞
24. Let (
1+z 2
z−i , if z 6= i,
f (z) =
4i, if z = i.
(a) Prove that lim f (z) exists and determine its value.
z→i
y
z = z 0 + i ∆y
z0 z = z 0 + ∆x
0 x
Proof. It follows from the existence of limit (1.4.1) that this limit
does not depend on the direction of the ray along which ∆z → 0. We now
show that the Cauchy–Riemann equations hold.
Firstly, let ∆z = ∆x in (1.4.1), that is z → z0 along a ray parallel to
the x-axis (see Fig 1.14). Thus
1
f 0 (z0 ) = lim [u(x0 + ∆x, y0 ) + iv(x0 + ∆x, y0 )
∆x→0 ∆x
− u(x0 , y0 ) − iv(x0 , y0 )]
1 (1.4.4)
= lim [u(x0 + ∆x, y0 ) − u(x0 , y0 )]
∆x→0 ∆x
1
+ i lim [v(x0 + ∆x, y0 ) − v(x0 , y0 )].
∆x→0 ∆x
Since, by assumption, f 0 (z0 ) exists and is finite, the three limits in (1.4.4)
exist. This implies that ux and vx also exist at the point M0 = (x0 , y0 ) so
that (1.4.4) can be written in the form
0 ∂u ∂v
f (z0 ) = +i . (1.4.5)
∂x M0 ∂x M0
We see, as for (1.4.4), that the three limits in (1.4.6) exist, so that uy and
vy exist at the point M0 . Hence (1.4.6) has the form
0 ∂v ∂u
f (z0 ) = −i . (1.4.7)
∂y M0 ∂y M0
Finally, since the left-hand sides of (1.4.5) and (1.4.7) are equal, then
their right-hand sides also are equal. Equating the real and imaginary parts
of the right-hand sides of (1.4.5) and (1.4.7) we obtain the Cauchy–Riemann
equations (1.4.3).
We prove the following converse to Theorem 1.4.1.
Theorem 1.4.2. If the functions of two variables, u(x, y) and v(x, y),
are differentiable at the point (x0 , y0 ) and their partial derivatives are con-
tinuous and satisfy the Cauchy–Riemann equations (1.4.3), then the func-
tion
f (z) = u(x, y) + iv(x, y)
is differentiable at the point z0 = x0 + iy0 .
Proof. Since u and v have continuous first-order partial derivatives,
as shown in advanced calculus, we can write
∂u ∂u
u(x + h, y + k) − u(x, y) = h+ k + ε1 ,
∂x ∂y
∂v ∂v
v(x + h, y + k) − v(x, y) = h+ k + ε2 ,
∂x ∂y
where the remainders ε1 and ε2 tend to zero more rapidly than h + ik, that
is,
ε1 /(h + ik) → 0, ε2 /(h + ik) → 0, as h + ik → 0.
With the notation f (z) = u(x, y) + iv(x, y), by the Cauchy–Riemann equa-
tions, we obtain
∂u ∂v
f (z + h + ik) − f (z) = +i (h + ik) + ε1 + iε2 ,
∂x ∂x
1.4. ANALYTIC FUNCTIONS 33
and hence,
f (z + h + ik) − f (z) ∂u ∂v
lim = +i .
h+ik→0 h + ik ∂x ∂x
It then follows that f (z) is analytic.
Because of Theorems 1.4.1 and 1.4.2, the Cauchy–Riemann equations
(1.4.2) are also known as conditions of analyticity of a function.
Using the Cauchy–Riemann equations, one can express the derivative
of an analytic function in the following equivalent forms:
∂u ∂v ∂v ∂u ∂u ∂u ∂v ∂v
f 0 (z) = +i = −i = −i = +i . (1.4.8)
∂x ∂x ∂y ∂y ∂x ∂y ∂y ∂x
1.4.3. Basic properties of analytic functions. Using the expres-
sion (1.4.1) for the derivative, f 0 , of f , one can transfer some properties of
differentiable functions to analytic functions. Let D be a domain. Then we
have the following properties of analytic functions.
(1) If f (z) is analytic in D, then it is continuous in D since it follows
from (1.4.1) that
∆f (z)z = f 0 (z0 )∆z + α∆z,
0
where α → 0 as ∆z → 0; thus, ∆f (z)z0 → 0 as ∆z → 0.
(2) If f1 (z) and f2 (z) are analytic in D, then their sum, difference,
product and quotient, f1 ± f2 , f1 f2 and f1 /f2 (if f2 6= 0), are analytic in
D. Moreover,
(f1 ± f2 )0 = f10 ± f20 , (f1 f2 )0 = f10 f2 + f1 f20 ,
and 0
f1 f 0 f2 − f1 f20
= 1 , provided f2 6= 0. (1.4.9)
f2 f22
(3) Let w = f (z) be analytic in D and f 0 (z) 6= 0 in D. If ζ = g(w)
is defined and analytic on the range, G = {w = f (z); z ∈ D}, of f in
the w-plane, then the composite function g[f (z)] is analytic in D, in the
z-plane, and ζ 0 (z) is expressed by the chain rule,
dζ dζ dw
= . (1.4.10)
dz dw dz
(4) If w = f (z) is analytic in D and f 0 (z) 6= 0 at the point z0 and
hence, by continuity, in some neighborhood U of z0 , then an inverse function
z = g(w) is defined in a neighborhood of the point w0 = f (z0 ) of the range
of f over U . Moreover g is an analytic function of the complex variable w
and
1
g 0 (w0 ) = 0 . (1.4.11)
f (z0 )
34 1. FUNCTIONS OF A COMPLEX VARIABLE
(5) If the real part, u(x, y), of an analytic function, f (z), is given in a
simply connected domain D of the (x, y)-plane, then the imaginary part,
v(x, y), of f (z) is determined by the Cauchy–Riemann equations (1.4.3) to
within an arbitrary constant. In fact, we have
Z M
∂v ∂v
v(x, y) = dx + dy + C
M0 ∂x ∂y
Z M (1.4.12)
∂u ∂u
= − dx + dy + C,
M0 ∂y ∂x
where the points M0 = (x0 , y0 ) and M = (x, y) can be joined by any curve
in D. It is more convenient to join M0 and M by a polygonal line whose
segments are parallel to the x- and y-axes.
An analytic function, f , can be conveniently expressed by means of its
real part, u(x, y),
z + z̄0 z − z̄0
f (z) = 2u , − f (z0 ), (1.4.13)
2 2i
or its imaginary part, v(x, y),
z + z̄0 z − z̄0
f (z) = 2iv , + f (z0 ), (1.4.14)
2 2i
where the bar indicates complex conjugation.
(6) If
f (z) = u(x, y) + iv(x, y)
is analytic in a domain D, then the family of curves
u(x, y) = c, v(x, y) = d,
are orthogonal. In fact, by the Cauchy–Riemann equations (1.4.2), we have
∂u ∂u ∂v ∂v
∇u · ∇v = , · ,
∂x ∂y ∂x ∂y
∂u ∂v ∂u ∂v
= +
∂x ∂x ∂y ∂y
∂u ∂u ∂u ∂u
=− +
∂x ∂y ∂y ∂x
= 0,
that is, the vectors ∇u and ∇v are orthogonal. But, since these vectors are
orthogonal to the families of curves u(x, y) = c and v(x, y) = d, respectively,
these families also are orthogonal.
1.4. ANALYTIC FUNCTIONS 35
= y.
To obtain (1.5.7) we have used the approximation arctan x ∼ x if x ∼ 0.
Substituting (1.5.6) and (1.5.7) into (1.5.3) we finally obtain
z n
lim 1 + = ex (cos y + i sin y).
n→∞ n
Thus we have the following definition.
Definition 1.5.1. The exponential function, ez , is defined by the ex-
pression
ez = ex+iy = ex (cos y + i sin y). (1.5.8)
The previously used Euler’s formula,
eiy = cos y + i sin y, y ∈ R, (1.5.9)
is derived by setting x = 0 in (1.5.8).
We show that ez possesses the following four properties:
(a) For real z = x, definition (1.5.8) coincides with the usual definition
of ex .
(b) The function ez is everywhere analytic in the z-plane.
(c) The usual formula of differentiation is still valid:
d z
e = ez . (1.5.10)
dz
(d) The law of exponents holds:
ez1 ez2 = ez1 +z2 . (1.5.11)
40 1. FUNCTIONS OF A COMPLEX VARIABLE
Properties (a)–(d) are valid for both real and complex arguments of ez .
But for a complex argument, the function ez has pure imaginary period 2πi
since, by Euler’s formula (1.5.9), we have
1.5.2. Logarithm of z.
Definition 1.5.2. Given a nonzero complex number z, a complex num-
ber w such that ew = z is called a logarithm of z, written
w = log z. (1.5.13)
Suppose that w = u + iv and let z = ew = eu+iv . Then
eu = |z| =⇒ u = ln |z|,
where ln |z| is the natural logarithm, to the base e, of a real number, and
v = arg z = Arg z + 2kπ.
Thus the expression
√
Solution. We transform 3 + i to the exponential form:
√ √ 1 √ π
3 + i = 2, tan Arg 3+i = √ , Arg 3+i = .
3 6
we have
eix + e−ix eix − e−ix
cos x = , sin x = . (1.5.18)
2 2i
Now we use the analytic continuation of the right-hand sides of (1.5.18) from
the real axis to the complex plane to define cos z and sin z as functions of
the complex variable z,
which coincide with the functions cos x and sin x for real z = x. It can
be shown that this continuation is unique. We leave as an exercise for the
reader to show from (1.5.19) that cos z and sin z
(a) are analytic everywhere,
(b) satisfy the usual rules of differentiation
d d
cos z = − sin z, sin z = cos z,
dz dz
etc.
1.5. ELEMENTARY ANALYTIC FUNCTIONS 43
i
3. ee .
4. e1+(πi/4) /eπi/3 .
5. Prove that there cannot be any finite values of z such that ez = 0.
6. Show that |1 + ez | ≤ 1 + ex .
7. Prove that the function
( 4
e−1/z , if z 6= 0,
f (z) =
0, if z = 0,
satisfies the Cauchy–Riemann equations at every point of the plane without
being analytic in the whole plane.
8. Describe the limiting behavior of ez as z → ∞ along the ray arg z = α.
If −π < Arg z ≤ π, evaluate the following expressions.
9. Log(3i).
10. Log(−2i).
11. log(1 + i).
12. log(z 5 ), where z = 3 eπi/6 .
Find all the roots of the following equations.
13. ez = −4.
14. ez = 2i.
√
15. eiz = − 3 + i.
16. log z = (π/2)i.
Derive formulae for the real and imaginary parts of the following functions
and check that they satisfy the Cauchy–Riemann equations.
17. sin z.
18. cos z.
19. cosh z.
20. sinh z.
Prove the following identities.
21. cos 2z = cos2 z − sin2 z.
1
22. 1 + tan2 z = .
cos2 z
23. sin(z1 + z2 ) = sin z1 cos z2 + cos z1 sin z2 .
24. cos(z1 − z2 ) = cos z1 cos z2 + sin z1 sin z2 .
EXERCISES FOR SECTION 1.5 47
25. Show that neither sin z̄ nor cos z̄ is an analytic function of z anywhere.
26. Prove that | sin z| ≥ | sin x| and | cos z| ≥ | cos x|.
Where are the following functions analytic?
ez
27. 2 .
z (z + 1)
1
28. cos .
z
ez − 1
29. z .
e +1
ez
30. .
cos z
Find all possible solutions of the following equations.
31. cos z = i.
32. sin z = 32.
33. cosh z = 1/4.
34. sinh z = 2i.
35. Find all the zeros of the functions cosh z and sinh z.
36. Is sin |z|2 anywhere differentiable? Is it anywhere analytic?
37. Prove that all the roots of the equations sin z = a and cos z = a are
real, if −1 ≤ a ≤ 1.
38. Prove that if z ∈ C and | sin z| ≤ 1, then z ∈ R.
39. Find the principal value of ii , (1−i)2i . (Hint. By definition, z a = ea log z ,
if z ∈ C and a ∈ C.)
40. Find the real and imaginary parts of z z where z = x + iy.
If −π < Arg z ≤ π, represent the following functions in the form z = x + iy.
41. Arcsin i.
42. Arccos πi.
If −π < Arg z ≤ π, find all the values of
43. arcsin 2.
44. arccos 100.
CHAPTER 2
γ2 γ1
y v Γ1
w
z=z0+∆z
∆z ∆w
~ Γ2 ~
θ1 Θ1
z0 w0
θ2 Θ2
θ1 Θ1
0 x 0 u
θ2 − θ 1 Θ2 − Θ1
49
50 2. ELEMENTARY CONFORMAL MAPPINGS
∆w
where we have used the exponential form of the complex number ∆z . Thus
|∆w|
f 0 (z0 ) = lim exp i lim [arg ∆w − arg ∆z] , (2.1.2)
∆z→0 |∆z| ∆z→0
where θ1 is the angle between the tangent to γ1 at the point z0 and the
x-axis.
Similarly, as ∆z → 0, ∆w → 0 while w ∈ Γ1 , so that the direction of
the vector ∆w tends to the direction of the tangent to Γ1 , that is,
lim arg ∆w = Θ1 , (2.1.5)
∆w→0
where we have omitted the symbol |z0 on the right-hand side. We suppose
that |f 0 (z0 )| = k > 0. We know that a function f , which is continuous at a
point z0 , is equal to its limit at z0 plus a function g which goes to zero as
z → z0 . Then taking (2.1.9) into account, we get
|∆w|
= k + g(z) → k, as ∆z → 0. (2.1.10)
|∆z|
Thus, to within higher order infinitesimal terms with respect to |∆z|, we
have
|∆w|z0 = k|∆z|z0 , k = |f 0 (z0 )| = constant > 0. (2.1.11)
Therefore, the length of each sufficiently small vector originating from the
point z0 is dilated by the factor k = |f 0 (z0 )| under the mapping by an
analytic function w = f (z). This property is known as the property of
constant dilation.
52 2. ELEMENTARY CONFORMAL MAPPINGS
It follows from (2.1.11) that any circle with sufficiently small radius δ
centered at z0 is mapped into a circle of radius kδ centered at w0 ; each
sufficiently small triangle with a vertex at z0 is mapped into a similar
curvilinear triangle with a vertex at w0 with similarity coefficient k.
Note 2.1.1. Besides the condition f 0 (z0 ) 6= 0, in order to satisfy the
angle-preserving property and the property of constant dilation, one has to
require that the function f (z) should be univalent, that is, injective.
A function is univalent or injective if different points of the z-plane are
mapped into different points of the w-plane, that is, for every pair of points
z1 , z2 in a domain D, we have the implication
z1 6= z2 =⇒ f (z1 ) 6= f (z2 ).
The concept of univalent function and the determination of domains of uni-
valence will be illustrated later by means of examples of concrete mappings.
Definition 2.1.1. A mapping of a neighborhood of a point z0 onto a
neighborhood of a point w0 that satisfies the angle-preserving property and
the constant dilation property is called a conformal mapping.
The previous arguments lead to the following necessary and sufficient
conditions for a function f (z) to produce a conformal mapping of a domain
D:
(a) univalence condition,
(b) analyticity of f ,
(c) for all z ∈ D, f 0 (z) 6= 0.
It can be shown that the univalence of f in D implies that f 0 (z) 6= 0
everywhere in D, so that condition (c) can be omitted. The converse, in
general, is not true, that is, it does not follow that f is univalent in D if
f 0 (z) 6= 0 in D. For example, the mapping w = z 4 is not univalent on the
half-annulus,
1 < |z| < 2, 0 < Arg z < π,
because the annulus is mapped onto the domain
1 < |w| < 16, 0 < arg w < 4π,
that is, on two copies of the annulus 1 < |w| < 16, 0 < arg w < 2π, but
w0 = 4w3 6= 0 in the annulus.
y v
~
D D
w0 Γ
γ
0 x 0 u
y v w2
z2
w3
z3 z1 w1
G
D γ Γ
0 x 0 u
y v w2
z2 G
w1
z3 z1 w3
D γ Γ
0 x 0 u
y v
D2
z2 w2 Γ
z1 G2 w1
γ
G1
D1
0 x 0 u
circle). Let z1 and z2 be two points of the z-plane that are symmetric with
respect to γ (symmetry with respect to an arc of a circle will be defined
in Subsection 2.3.2). Then z1 and z2 are mapped into points w1 and w2
which are symmetric with respect to Γ (see Fig 2.5); any two sets D1 and
D2 that are symmetric with respect to γ are mapped onto sets G1 and G2
that are symmetric with respect to Γ.
5. y = 1/x2 , 1 ≤ x ≤ 4.
2
6. y = 4 − x , −1 ≤ x ≤ 0.
Find the angle through which a curve drawn from the point z0 is rotated
under the mapping w = f (z), and find the corresponding scale factor of the
transformation.
7. z0 = −1, w = z2.
8. z0 = 1 + 2i, w = z 2 + 2z.
9. z0 = −1 + i, w = 1/z.
10. z0 = −i, w = z3.
Determine all the points in the z-plane for which the following mappings
are not conformal.
11. sin z.
12. sinh z.
13. z 2 + 4z + 3.
14. z 3 + 3z 2 − 9z.
z2 + 4 3
15. , z 6= − .
2z + 3 2
2
2z + 6
16. , z 6= 2.
z−2
z 2 +2z
17. e .
4
18. ez −32z .
19. Consider the two curves
γ1 : z = t + i, 0 ≤ t ≤ 1; γ2 : z = τ + iτ, 0 ≤ τ ≤ 1.
(a) Find the point of intersection, P , of the curves and the angle, α,
between the curves at P .
(b) Find the image of each curve under the mapping w = z 2 and
determine the angle between the images of the curves in the w-
plane. Is this angle equal to α? Explain your answer.
20. Consider the two curves
γ1 : z = t, 0 ≤ t ≤ 1; γ2 : z = τ + iτ, 0 ≤ τ ≤ 1.
(a) Find the point of intersection, P , of the two curves and the angle,
α, between the curves at P .
(b) Find the image of each curve under the mapping w = z̄ and deter-
mine the angle between the images of the curves in the w-plane.
Is this angle equal to α? Explain your answer.
2.3. LINEAR MAPPING AND INVERSION 57
y
z+b
b
0 x
y y
w = az aρ
w0 = az 0
ρ
z
z0
0 x 0 x
(a) (b)
y w = e i Arg a z
Arg a z
Arg z
0 x
y v 1
C
4 0 u
~
2 C
A B
–2 ~
x B
0 1 3 5
Example 2.3.1. Find the function that maps the triangle ABC in the
eC
z-plane into a similar triangle OB e in the w-plane (see Fig 2.9), if
and
e = (0, −2), C
O = (0, 0), B e = (1, −1).
Solution. Since the triangles ABC and OB eCe are similar, then the
mapping is given by a linear function. We perform the mapping in three
stages:
(a) a parallel translation by the vector −(1 + 2i), so that the vertex
A is mapped into the origin of the w1 -plane (see Fig 2.10(a)),
(b) a rotation through the angle −π/2 (see the w2 -plane in Fig 2.10(b)),
(c) a contraction with coefficient 1/2 (see the w-plane in Fig 2.10(c)).
Hence these three steps can be described as follows:
(1) w1 = z − (1 + 2i),
(2) w2 = e−iπ/2 w1 = −i[z − (1 + 2i)],
(3) w = w2 /2 = −i[z − (1 + 2i)]/2.
v1 v2 v
u2
0
C1 C2 0 u
~
C
~
B2 B
0 B1 u1
B
A
0
R
1 z _
w1= 1/ z
0 x
– = 1/z
w=w1
−→
and if the point A approaches the center of the circle (|OA| → 0), then the
point B moves away to infinity. This means that the points 0 and ∞ are
symmetric with respect to the circle.
We prove that the inversion w = 1/z is the successive application of
two reflections:
(1) A reflection of the point z with respect to the circle |z| = 1 (see
Fig 2.12),
1
w1 = . (2.3.9)
z̄
Indeed, if z = |z| exp (i Arg z), then
1 1 i Arg z
z̄ = |z| exp (−i Arg z), = e ,
z̄ |z|
that is,
1 1
Arg = Arg z, |z| = 1;
z̄ |z̄|
thus, the points z and 1/z̄ are symmetric with respect to the circle |z| = 1.
(2) A reflection of the point 1/z̄ with respect to the x-axis:
1 1
w= = . (2.3.10)
z̄ z
Here, we have used the relation
z1 z̄1
= ,
z2 z̄2
whose proof is left to the reader.
The inversion (2.3.10) maps the interior of the upper half-disk,
|z| < 1, =z > 0,
62 2. ELEMENTARY CONFORMAL MAPPINGS
Find the image of the following regions D under the given mapping
w = f (z).
11. D = {z; <z ≥ 0}, w = iz.
12. D = {(x, y) ∈ R2 ; −∞ < x < +∞, 0 < y < 1}, w = iz + 2.
2
13. D = {(x, y) ∈ R ; 0 < x < 1, −∞ < y < +∞}, w = (1 + i)z.
14. D = {z; =z ≥ 0}, w = (1 − i)z + 1 + i.
2
15. D = {(x, y) ∈ R ; 1 ≤ x ≤ 2, 0 ≤ y ≤ 1}, w = 3z + 1.
2 2 2
16. D = {(x, y) ∈ R ; x + y = 1}, w = 2z + i.
1−i
17. D = {z; |z| < 3, 0 ≤ Arg z ≤ π/4}, w = √ z.
2
18. D = {z; |z| < 1, <z > 0}, w = 2iz + i.
Find a linear transformation w = az + b which maps the strip contained
between the given straight lines, L1 and L2 , onto the strip 0 < <w < 1
with the given normalization.
19. L1 : x = 1, L2 : x = 2, w(1) = 0.
20. L1 : y = x, L2 : y = x − 2, w(0) = 0.
Find the images of the following curves under the inversion w = 1/z.
21. |z + 1| = 1.
22. |z − 1| = 2.
23. x2 + y 2 = 4x.
24. x2 + y 2 = 6y.
25. y = x + 2.
26. y = 3x.
Find the images of the following regions under the inversion w = 1/z.
27. D = {(x, y) ∈ R2 ; 0 < x < 2, −∞ < y < +∞}.
28. D = {(x, y) ∈ R2 ; −∞ < x < +∞, 0 < y < 1}.
29. D = {(x, y) ∈ R2 ; x > 0, y < 1}.
30. D = {(x, y) ∈ R2 ; x > 1, y < 0}.
64 2. ELEMENTARY CONFORMAL MAPPINGS
(a) w1 = cz + d,
1
(b) w2 = ,
w1
a ad
(c) w = + b − w2 .
c c
Hence if a property is satisfied by a linear function or the inversion
w = 1/z, then it will also be true for the function (2.4.1).
The following theorem holds.
Theorem 2.4.1. Every circle is mapped into a circle by a linear frac-
tional transformation, provided a line is considered as a circle with radius
R = ∞.
Proof. We prove the theorem only for the inversion mapping
1
w= , (2.4.4)
z
since obviously any linear mapping preserves circles (see Section 2.3 and
Fig 2.7(b)).
Let z0 = x0 + iy0 . The Cartesian equation of the circle |z − z0 | = R in
the z-plane is
A(x2 + y 2 ) + Bx + Cy + D = 0, (2.4.5)
where A, B, C and D are real constants chosen such that (2.4.5) is the
equation of the circle (x − x0 )2 + (y − y0 )2 = R2 in R2 .
Let w = u + iv and z = x + iy. Then the equation w = 1/z can be
written in the form
1
x + iy = .
u + iv
Separating the real and imaginary parts, we obtain
u v
x= 2 , y=− 2 . (2.4.6)
u + v2 u + v2
Substituting (2.4.6) into (2.4.5), we get
1 u v
A 2 2
+B 2 2
−C 2 + D = 0,
u +v u +v u + v2
or
D(u2 + v 2 ) + Bu − Cv + A = 0. (2.4.7)
This equation is the equation of a circle in the w-plane.
There are four cases:
(a) If A 6= 0, D 6= 0, circles are mapped into circles.
(b) If A 6= 0, D = 0, circles are mapped into straight lines.
(c) If A = 0, D 6= 0, straight lines are mapped into circles.
(d) If A = 0, D = 0, straight lines are mapped into straight lines.
66 2. ELEMENTARY CONFORMAL MAPPINGS
y v
1/2 1
~ ~ u
1+i 0 B A
C
1– i ~
D C
2
A B
0 1 2 x
y v
1
0 1 x 0 u
e lies outside
Since the point w = 3 lies outside the circle |w| ≤ 1, then D
this circle.
Note that by the method used in the previous solution one can find
the image, expressed in Cartesian coordinates, of any line or circle under a
linear fractional transformation.
Example 2.4.3. Find the image of the circle |z − 1| = 1 under the
mapping (2.4.8).
Solution. Since (2.4.8) sends z = 0 to w = 1 and z = 2 to w = ∞,
we know by Theorem 2.4.1 that the circle goes into a straight line. To find
the direction of this line, we verify that the point z = 1 + i on the circle
goes to the point
2+1+i 3+i 1+i
w= = = 1 + 2i.
2−1−i 1−i 1+i
Hence the image of the circle is the vertical straight line <w = 1.
We obtain the same result by applying the method of the previous
example. It follows from (2.4.9) that the image of the circle is the curve
w−1
2
w + 1 − 1 = 1. (2.4.12)
so that, finally,
u = 1. (2.4.15)
Hence the image of the circle |z − 1| = 1 is the straight line <w = 1.
Images of straight lines and circles by elementary linear fractional trans-
formations (2.4.1) can be found, for instance, in [15], pp. 345–352, [33] and
[44], pp. 132–133. We present these formulae for reference purposes.
(a) A straight line
<(λz) = α (2.4.16)
that does not pass through the point z = −d/c (that is, if <(λd/c) 6= −α)
is mapped into the circle |w − w0 | = R, where
2aαc̄ + ad¯λ̄ + bλc̄ a
,
w0 = R = − w0 . (2.4.17)
2α|c|2 + 2< cd¯λ̄ c
In order to use formula (2.4.17) one has to determine the parameters λ and
α by means of the Cartesian equation of a line,
Ax + By + C = 0. (2.4.18)
Letting λ = λ1 + iλ2 and z = x + iy in (2.4.16), we have
<[(λ1 + iλ2 )(x + iy)] = α,
and simplifying the last relation, we get
λ1 x − λ2 y = α. (2.4.19)
It follows from (2.4.18) and (2.4.19) that
λ1 = A, λ2 = −B, α = −C,
that is,
λ = A − Bi, α = −C.
(b) The straight line
λd
<(λz) = −<
c
passing through the point z = −d/c is mapped into the straight line
ad − bc ad − bc ā
< λw̄ = < λ . (2.4.20)
c2 c2 c̄
(c) The circle |z − z0 | = r that does not pass through the point z =
−d/c, for r 6= |z0 + d/c|, is mapped into the circle |w − w0 | = R, where
¯ − c̄r2
(az0 + b)(c̄z̄0 + d) r|ad − bc|
w0 = , R = . (2.4.21)
|cz0 + d| − |c|2 r2
2 |cz0 + d|2 − |c|2 r2
70 2. ELEMENTARY CONFORMAL MAPPINGS
y v
2i
E 2
C
D 1+i
A B G
–2 0 1 2 x
0 u
Solution. First of all, it is necessary that the point, B = (2, 0), com-
mon to both circles be mapped to the point w = ∞, for, it is only in this
case that the images of both circles will be straight lines. Thus the general
form of the mapping is
az + b
w1 = ,
z−2
where a and b are arbitrary constants. For example, we can choose a = 1
and b = 0 so that
z
w1 = . (2.4.23)
z−2
As it is already clear that both circles will map into straight lines (moreover,
the images of the circles will be parallel lines because they intersect at the
point w = ∞) then it is sufficient to find the images of two points on each
circle.
We choose the two points, A = (−2, 0) and E = (0, 2), on the large
circle. Their images by transformation (2.4.23) are
A = (−2, 0) 7→ A1 = (1/2, 0), E = (0, 2) 7→ E1 = (1/2, −1/2).
EXERCISES FOR SECTION 2.4 71
v1 v
O1 A1 1/2
u1 2
0
E1 1/2 – i/2
0 u
– i C1
(a) (b)
Similarly, the images of the two points, O = (0, 0) and C = (1, 1), on the
small circle are
O = (0, 0) 7→ O1 = (0, 0), C = (1, 1) 7→ C1 = (0, −1).
Since z = −1 7→ w1 = 1/3, the region D is mapped in the w1 -plane onto
the strip that is bounded by the straight lines passing through the points
A1 , E1 and O1 , C1 (see Fig 2.16(a)).
In order to map the strip in Fig 2.16(a) onto the strip in Fig 2.16(b), it
is sufficient to perform a rotation through an angle π/2 (that is, to multiply
w1 by exp (iπ/2) = i) and a similarity with dilation factor 4:
z
w = 4eiπ/2 w1 = 4i . (2.4.24)
z−2
Note 2.4.1. The mapping (2.4.24) is not unique; one can add an ar-
bitrary real constant c to the right-hand side of (2.4.24). In this case the
strip will be shifted parallel to itself in the w-plane along the vector (c, 0).
z
5. D = {z; 1 < |z| < 2}, w= .
z+1
z
6. D = {z; 0 < =z < 1}, w= .
z−i
Find the linear fractional transformation which transforms the points z1 ,
z2 , z3 into the points w1 , w2 , w3 , respectively.
7. z1 = 1, z2 = 0, z3 = i, w1 = −1, w2 = ∞, w3 = 1.
8. z1 = i, z2 = 1 − i, z3 = 1, w1 = 0, w2 = −1, w3 = ∞.
9. z1 = 1 + i, z2 = 1 − i, z3 = −1, w1 = 0, w2 = 1, w3 = i.
10. z1 = −i, z2 = i, z3 = 0, w1 = 1, w2 = i, w3 = 1 − i.
A point z0 is called a fixed point of the transformation w = f (z) if
f (z0 ) = z0 . Find the fixed points of the following transformations.
z
11. w = .
z+2
z−i
12. w = .
z+i
13. w = 1/z.
2z + 1
14. w = .
z−2
15. w = az + b, a 6= 0.
az + b
16. w = , ad − bc 6= 0.
cz + d
Find the linear fractional transformation which maps the region D of the
z-plane onto the region G of the w-plane with the given normalization.
17. D = {z; =z ≥ 0}, G = {w; |w| ≤ 1},
with w(0) = 1, w(1) = i, w(−1) = −i.
18. D = {z; |z| ≤ 1}, G = {w; |w − 1| ≤ 1},
with w(1) = 0, w(i) = 2, w(−i) = 1 + i.
Find a linear fractional transformation which maps the region D of the
z-plane onto the region G of the w-plane.
19. D = {z; |z| < 2}, G = {w; =w > 0}.
20. D = {z; |z − 1| < 1}, G = {w; <w > 0}.
21. D = {z; |z + 1| < 2}, G = {w; <w < 0}.
22. D = {z; |z − i| < 1}, G = {w; =w > 0}.
23. Find the general form of a linear fractional transformation which maps
the upper half-plane onto itself.
24. Find the general form of a linear fractional transformation which maps
the upper half-plane onto the lower half-plane.
2.5. SYMMETRY AND LINEAR FRACTIONAL TRANSFORMATIONS 73
y v
z0
|x – z0 |
x 0 x 0 1 u
|x – –z0 |
–z
0
y v
z0
|1 – z0 |
~
A A
0 1 x 0 1 u
–z
0 |1 – –z0 |
Figure 2.18. Mapping of the unit disk onto the unit disk,
such that z0 maps to 0.
2.5.2. Mapping of the unit disk onto the unit disk. We want to
map the disk |z| ≤ 1 onto the disk |w| ≤ 1 so that a given point, z0 , of the
first disk is mapped to the center of the second disk (see Fig 2.18).
We use the fact that z0 and 1/z̄0 are symmetric with respect to the
circle |z| = 1 (see Definition 2.3.1). As z0 is mapped to w = 0, then 1/z̄0
has to map to w = ∞, which is symmetric to w = 0 with respect to any
circle centered at w = 0. Hence the desired mapping has the form
z − z0 z − z0
w=k 1 = k̃ 1 − z z̄ , k̃ = −z̄0 k, (2.5.4)
z − z̄0 0
16
P2 P1 9
–24 0 3 8 24 x
y v
/n
w = zn
π
z=
rg
π/n
A
11. Map the eccentric ring bounded by the circles |z| = 4, |z + 1| = 1 onto
the ring 1 < |w| < R. Find R.
12. Map the eccentric ring bounded by the circles |z| = 1, |z − 1| = 5/2
onto the ring 1 < |w| < R. Find R.
y v
/n
2π
w = zn
z=
Arg w = 0
rg
Arg w = 2π u
A
2π / n 0
0 Arg z = 0 x
to be understood as
eα log z ,
where α is any real number, maps the wedge 0 ≤ arg z ≤ π/α onto the
upper half-plane.
In general, the wedge 0 ≤ Arg z ≤ θ0 (if nθ0 < 2π) is mapped into the
wedge 0 ≤ Arg w ≤ nθ0 < 2π.
The wedge 0 ≤ θ ≤ Arg z ≤ 2π/n is mapped onto the whole w-plane
less the real positive axis. The rays θ = 0 and θ = 2π/n are mapped onto
this axis. To have a univalent mapping, it is necessary to cut the plane
along, say, the real axis <w = 0 (that is, to consider this line as a double
line) and assume that the ray θ = 0 is mapped onto the upper part of the
cut and the ray θ = 2π/n is mapped onto the lower part of the cut (see
Fig 2.21).
Hence the largest angle, in absolute value, measured from the positive
x-axis and such that w1 6= w2 if z1 6= z2 is the angle 0 ≤ Arg z ≤ 2π/n.
This angle defines the region of univalence of the function w = z n .
We can ask the following question “What is the image of other parts
of the z-plane?” Let us consider, for example, the wedge 2π/n ≤ arg z ≤
4π/n. As the ray θ = 2π/n is mapped onto the ray arg w = 2π and the ray
θ = 4π/n is mapped onto the ray arg w = 4π, the wedge 2π/n ≤ arg z ≤
4π/n is also mapped onto the whole complex w-plane with a cut, but, in
this case, the ray θ = 2π/n is mapped onto the upper part of the cut while
the ray θ = 4π/n is mapped onto the lower part of the cut. In general,
for the function w = z n , the whole z-plane is divided into n regions of
univalence of the form
2kπ 2(k + 1)π
≤ arg z ≤ , k = 0, 1, 2, . . . , n − 1.
n n
2.6. MAPPING BY z n AND w = z 1/n 79
y v
2π
––
3 I
2π
––
3
0 x 0 u
II
2π III
––
3
We see that each of these regions is mapped univalently onto the whole
plane, except for one cut. Such regions have a special name (see [2], pp. 98–
99).
Definition 2.6.1. A region which is mapped univalently onto the
whole plane, except for one or more cuts, by a function f (z) is called a
fundamental region of f .
2.6.2. The nth root of z. Since the power function z = wn has n
fundamental regions in the w-plane, its inverse w = z 1/n has n branches,
each of which is a function. More precisely, each branch of z 1/n ,
y C v
~ B Γ1
C ~ Γ0
z0 w0 2π
–– w0
Arg z 0 3
0 x 2π 0 u
–– 2π
~ 3 ––
Γ 3
~
~
w0 Γ2
This means that traversing the closed contour C e three times corresponds
to traversing the closed contour Γe only once in the w-plane. Similarly,
e n times for the function w = z 1/n corresponds to
traversing the contour C
e
traversing the contour Γ once.
2.6.3. Algebraic branch points.
Definition 2.6.2. A point z0 is called a branch point of the function
w = f (z) if the argument of w changes as z goes around z0 along any
sufficiently small closed contour.
Definition 2.6.3. If the increment of the argument of a function f (z)
is equal to zero when a branch point is encircled n times, n < ∞, then the
branch point is called an algebraic branch point of order n.
We see that the point z = 0 is an algebraic branch point of order n for
each branch of w = z 1/n . So is the point z = ∞. Indeed, by the inversion
1/n
z = 1/z1, z 1/n = 1/z1 . Since z1 = 0 is a branch point, then z = ∞ is also
a branch point. If we join the points z = 0 and z = ∞ by a cut, we obtain
a region of univalence for each branch of w = z 1/n . The cut can be any
arc from 0 to ∞ (see Fig 2.24). As soon as the cut is fixed, each branch of
w = z 1/n is uniquely determined.
Consider, for example, the branch w0 of w = z 1/3 ,
w0 (z) = |z|1/3 ei(Arg z/3) ,
and let z = −i. If we cut the complex z-plane along the negative real axis,
as shown in Fig 2.25(a), then −i = e−πi/2 (because −π < Arg z ≤ π) and
w0 (−i) = e−πi/6 .
If we cut the complex z-plane along the positive real axis, as shown in
Fig 2.25(b), then
0 ≤ Arg z < 2π, −i = e3πi/2 , w0 (−i) = eπi/2 = i 6= e−πi/6 .
82 2. ELEMENTARY CONFORMAL MAPPINGS
0 x
y y
0 x 0 x
(a) (b)
the third sheet to the first sheet the branch w2 continuously passes to the
branch w0 . The total angle along the closed contour on this surface is equal
to 6π.
A variant of the Riemann surface for w = z 1/3 without any abstract glue
along the cut between the third and the first sheets is shown in Fig 2.27. We
take θ = arg z and r = |z| as the horizontal and vertical axes, respectively.
If we glue such a plane along the lines θ = 0 and θ = 6π, we obtain a
cylindrical surface which represents a nonabstract variant of the Riemann
surface.
1 2
3 3 3
x 2 2
0 3 1
1 2 1
r = |z|
0 2π 4π 6π θ = arg z
y v
~ ~
B A
i π/12
B e 0 1 u
π
–
A 6
0 x
Solution. First, we map the wedge 0 ≤ Arg z ≤ π/6 onto the upper
half-plane (see Fig 2.29):
w(1) = z 6 .
In this case the point z1 = eπi/12 is mapped to the point
6
(1)
w1 = eπi/12 = eπi/2 = i,
(1)
and the point z2 = 0 is mapped to the point w2 = 0.
Then one has to map the upper half-plane =w(1) ≥ 0 onto the unit disk
(1)
|w| ≤ 1 so that the point w1 = i is mapped to the point w = 0 and the
(1)
point w2 = 0 is mapped to the point w = 1. Letting z0 = i in (2.5.3), we
obtain
w(1) − i z6 − i
w = eiα (1) = eiα 6 . (2.6.9)
w +i z +i
v1
0 u1
y v
z2
α
z1
0 x 0 u
Figure 2.30. The initial and final regions under the map-
ping in Example 2.6.3.
v1 v
2
α
θ0 α
0 u1 0 u2
(a) (b)
y v
1
2+2i
i C1
C
D
A B A1
0 2 4 x 01 1 u1
(a) (b)
y v
π/4
i
0 x 0 u
and
C = (2, 2) 7→ C1 = (0, 1).
Therefore the edges of the right angle pass through the points O1 = (0, 0),
A1 = (1, 0) and B1 = (0, 1), that is, the image of D under transformation
(2.6.14) is the first quadrant (see Fig 2.32(b)).
In order to map this quadrant onto the upper half-plane it is sufficient
to square (2.6.14):
2
2 z
w = w1 = . (2.6.15)
4−z
Example 2.6.5. Map the z-plane, with a cut from the point z = i to
the point z = ∞ making an angle π/4 with the positive x-axis, onto the
upper half-plane (see Fig 2.33).
Example 2.6.6. Map the upper half-plane, =z ≥ 0, with a cut from the
point z = i to the point z = 0, onto the upper half-plane, =w ≥ 0, without
a cut (see Fig 2.35).
Solution. If we let
w1 = z 2 ,
then the boundary of the region D in Fig 2.35 is mapped into a cut going
from the point w1 = −1 to the point w1 = 0 in the positive direction of the
88 2. ELEMENTARY CONFORMAL MAPPINGS
v1 v2
0 u2
0 u1
y v
0 x –1 0 1 u
–1 u1 0 1 u2
0
(a) (b)
u1 -axis (see Fig 2.36(a)). Indeed, the point z = i is mapped to the point
w1 = i2 = −1.
The negative real axis, −∞ < x < 0, is mapped into the lower part of a
cut going from w1 = 0 to w1 = ∞, and the positive real axis, 0 < x < +∞,
is mapped into the upper part of the cut from w1 = 0 to w1 = +∞.
EXERCISES FOR SECTION 2.6 89
Hence the cut from i to 0 in Fig 2.35 is mapped into the part of a cut
in Fig 2.36(a) that is located from w1 = −1 to w1 = 0. The remaining
mappings are elementary:
w2 = w1 + 1,
√ p
w = w1 + 1 = z 2 + 1.
y v
y2 = 3π
––
4
y1 = π
– y2
4 y1
0 x1 x2 x 0 u
y v
π
α
α
0 x 0 u
y v
2π
Arg w = 0
0 Arg w = 2π u
0 x
y v
w = Log z π/2
α
α
0 u
0 x
For example, the strip 2π ≤ =z < 4π is mapped onto the whole w-plane
with a cut along the positive real axis (compare with Fig 2.39), so that the
side y = 2π is mapped onto the upper part of the cut and the side y = 4π
is mapped onto the lower part of the cut in the w-plane.
2.7.2. The logarithm of z. We recall that the logarithm of z is given
by the formula
w = log z = Log z + 2kπi
(2.7.8)
= ln |z| + i Arg z + 2kπi.
Each branch of (2.7.8), for instance, the principal value w = Log z, is the
inverse of the exponential function z = ew , that is, w = Log z maps the
wedge 0 ≤ Arg z ≤ α < 2π onto the horizontal strip 0 ≤ =w ≤ α (see
Fig 2.40).
In particular, the wedge 0 ≤ Arg z ≤ π (that is, the upper half-plane
=z ≥ 0) is mapped onto the strip 0 ≤ =w ≤ π, and the whole complex
z-plane with a cut along the positive real axis is mapped onto the strip
0 ≤ =w < 2π (see Fig 2.41).
y v ~
~
4π w0
w = Log z
C ~
w
z0 Arg z = 0 2π 0
0 Arg z = 2π x
0 w0 u
Solution. To send the two tangent circles into parallel straight lines,
it suffices to map the point z = 2 to the point w1 = ∞ by the linear
fractional transformation
z+2
w1 = . (2.7.9)
z−2
Thus, the region D is mapped onto the strip, S, whose position is deter-
mined by the images of three boundary points, that is,
A = (−2, 0) 7→ A1 = (0, 0), B = (0, 2) 7→ B1 = (0, −1),
and
C = (4, 0) 7→ C1 = (3, 0).
94 2. ELEMENTARY CONFORMAL MAPPINGS
y v
2i B
A C
–2 0 1 3 4 x
D
0 u
v1 v2
L1 L2 π
π
– S
2
0 A1 C1 S'
3 u1
–1 B1
0 u2
(a) (b)
Hence the strip S is bounded by the two parallel lines L1 , passing through
the points A1 and B1 , and L2 , passing through the point C1 (see Fig 2.43(a)).
Finally, considering the orientation of the boundary of D and of its image,
we see that D is mapped inside the strip S of width 3 bounded by the lines
L1 and L2 .
Next, the linear transformation
π πz+2
w2 = eπi/2 w1 = i (2.7.10)
3 3 z−2
maps the vertical strip S of Fig 2.43(a) onto the vertical strip S 0 defined
by the inequations
0 ≤ =w2 ≤ π
(see Fig 2.43(b)). Finally, using the mapping properties of the exponen-
tial function, we map the strip S 0 onto the upper half-plane =w ≥ 0 (see
Fig 2.42),
π z+2
w = exp i . (2.7.11)
3 z−2
2.7. EXPONENTIAL AND LOGARITHMIC MAPPINGS 95
y v
0 1/2 1 x u
0
v1 v2
2π
π –2πh 0 u2
–e
–2πh 0 u1
(a) (b)
Example 2.7.2. Map the strip 0 ≤ <z ≤ 1 with a cut joining the points
z1 = 1/2 + ih and z2 = 1/2 + i∞, onto the upper half-plane =w ≥ 0 (see
Fig 2.44).
Solution. First, we map the strip shown in Fig 2.44 onto the strip
0 ≤ =w1 ≤ 2π by the linear transformation
Now, we determine the new position of the cut. The initial point of the cut,
z1 = 1/2 + ih, is mapped to the point w1 = 2πi(1/2 + ih) = −2πh + πi, so
that the cut joins the points −∞ + πi and −2πh + πi in the w1 -plane (see
Fig 2.45(a)).
We map the region shown in Fig 2.45(a) onto the upper half-plane
=w2 ≥ 0 with a cut along the positive real axis by
The initial point, w1 = −2πh + πi, of the cut in Fig 2.45(a) is mapped to
the point
w2 = e−2πh+πi = −e−2πh ,
and the point w1 = −∞ + πi is mapped to the point w2 = e−∞+πi = 0.
Hence, the semifinite cut in Fig 2.45(a) is mapped to the finite cut joining
the points −e−2πh and 0 in Fig 2.45(b). The boundaries =w1 = 0 and
=w1 = 2π of the region shown in Fig 2.45(a) are mapped to the upper and
lower parts of the cut joining the points w2 = 0 and w2 = ∞, respectively,
shown in Fig 2.45(b). The mapping of the region shown in Fig 2.45(b) onto
the upper half-plane is elementary:
w3 = w2 + e−2πh = e2πiz + e−2πh .
Hence p
√
w= w3 = e2πiz + e−2πh . (2.7.14)
Find the images of the following regions under the mapping w = Log z.
15. D = {z; 0 < Arg z < π/2}.
18. D = {z; 2 < |z| < 4, with the cut along the segment [2, 4]}.
Find the images of the following regions under the given mapping. (Hint:
Consider each problem as a composite mapping.)
19. D = {z; 1 < |z| < 2, 0 < Arg z < π/2}, w = Log z + 2 + i.
Letting z = reiθ and separating the real and imaginary parts of (2.8.1),
we have
a iθ 1 −iθ
w= re + e
2 r
a 1 1
= r+ cos θ + i r − sin θ
2 r r
=: u + iv.
Thus
a 1 a 1
u= r+ cos θ, v= r− sin θ. (2.8.2)
2 r 2 r
Let us find the image of the circle |z| = R. Letting r = R in (2.8.2),
we get
a 1 a 1
u= R+ cos θ, v= R− sin θ, (2.8.3)
2 R 2 R
and eliminating θ we obtain
u2 v2
a2
+
1 2 a2
1 2
= 1, (2.8.4)
4 R+ R 4 R− R
that is, the circle |z| = R is mapped onto the ellipse with semi-axes
a 1 a 1
ã = R+ , b̃ = R − . (2.8.5)
2 R 2 R
The coordinates of the foci of the ellipse are
q
c = ± ã2 − b̃2 = ±a, (2.8.6)
that is, the ellipses (2.8.4) are confocal with foci at the points ±a.
We consider the two cases: R > 1 and R < 1.
(a) The case R > 1. In this case, the points z = ±1 are located
inside the disk |z| ≤ R and therefore the mapping is conformal in the
region |z| ≥ R.
Let us find the image of the region |z| ≥ R if R > 1 (see Fig 2.46). We
first find the images of the points A, B and C on the boundary of the disk
by using formulae (2.8.3):
iπ a 1
A=R e 7→ A1 = − R+ ,0 ,
2 R
a 1
B=R eiπ/2 7→ B1 = 0, R− , (2.8.7)
2 R
a 1
C =Rei0 7→ C1 = R+ ,0 .
2 R
2.8. MAPPING BY JOUKOWSKY’S FUNCTION 99
y v
B B1
R A1 C1
A C
–1 0 1 x –a 0 a u
y v
B
A C A1 B1 C1
–1 0 1 x –a 0 a u
Since R − (1/R) > 0 if R > 1, then the point B1 is located in the upper
part of the ellipse. Therefore, going once along the circle is the same as
going once along the ellipse, as shown in Fig 2.46. Hence the exterior of
the disk is mapped onto the exterior of the ellipse.
If R → 1, it follows from (2.8.5) that
ã → a, b̃ → 0, (2.8.8)
so that the ellipse degenerates into a cut joining the foci w = −a and
w = a (see Fig 2.47). Hence the region |z| ≥ 1 is a fundamental region of
Joukowsky’s function.
(b) The case R < 1. In this case, the points z = ±1 are located
outside the disk |z| ≤ R, and therefore the mapping is conformal in the
region |z| ≤ R.
To find the image of the region |z| ≤ R < 1 (see Fig 2.48) we use the
images of the points A, B and C (formulae (2.8.7)). The difference with
100 2. ELEMENTARY CONFORMAL MAPPINGS
y v
B
R A1
A C C1
–1 0 1 x –a 0 a u
B1
the case (a) is that R − (1/R) < 0 if R < 1, and therefore the point B1
is located in the lower part of the ellipse, that is, the directions along the
circle and the ellipse in Fig 2.48 are opposite to each other. Hence the
interior of the disk |z| ≤ R is mapped onto the exterior of the ellipse, where
the lower half-disk is mapped onto the upper part of the half-plane outside
the ellipse, but the upper half-disk is mapped onto the lower part of the
half-plane outside the ellipse (see Fig 2.48).
If R → 1, the ellipse, as in the case (a), degenerates into a cut joining
the points −a and a (see Fig 2.47), but the upper semicircle ABC, in this
case, is mapped onto the lower part of the cut while the lower semicircle is
mapped onto the upper part of the cut (see Fig 2.49).
2.8.2. Examples of Joukowsky’s mapping. Joukowsky’s mapping
will be illustrated by means of examples.
Example 2.8.1. Map the open disk |z| < 1 with two cuts along the
segments [1/2, 1] and [−1, −1/2] of the real axis as shown in Fig 2.50, onto
the upper half-plane =w > 0.
2.8. MAPPING BY JOUKOWSKY’S FUNCTION 101
y v
B
1
A C
–1 _ _1 0 _1 1 x
2 2 0
–3 _ _1 _1 3 u
3 3
Figure 2.50. The initial and final regions under the map-
ping in Example 2.8.1.
v1 v2
–1 1
0 –5
u1
– –5 1– u2
4 4 0 9
9
x
(a) (b)
v2 v
B C
–1 1 1
0 u2 –1 0 1 u
– –4 –4
5 5
A D
(a) (b)
v2 v
i 1
–a a u2 0 u
–i 0
(a) (b)
Finally, we use the fact that the function inverse to Joukowsky’s func-
tion (2.8.9),
q
w = w2 + w22 − 1
s 2 (2.8.12)
2 1 16 1 1
= z+ + × z+ − 1,
5 z 25 4 z
maps the w2 -plane with a cut from w2 = −1 to w2 = 1 onto the region
|w| ≤ 1 (see Fig 2.49, where the roles of the z- and the w-planes have to
be interchanged). We find the images of the different parts of the cut in
Fig 2.52(a):
4 3 4 3
ww =−4/5 = − ± i, ww =4/5 = ± i.
2 5 5 2 5 5
Hence the left cut in Fig 2.50 is mapped onto the arc AB in Fig 2.52(b),
where A = (−4/5, −3/5) and B = (−4/5, 3/5). The right cut in Fig 2.50
is mapped onto the arc CD, where C = (4/5, 3/5) and D = (4/5, −3/5).
The upper and lower semicircles in Fig 2.50 are mapped onto the arcs AD
and BC, respectively. The desired mapping is given by (2.8.12).
Example 2.8.3. Map the exterior of the cross shown in Fig 2.53 onto
the exterior of the unit disk.
y v1
i
0
0 x –––– –––– u1
a – √a 2 +1 –1 1 √a 2+1
(a) (b)
√ √
the points − a2 + 1 and a2 + 1, respectively (see Fig 2.54). In fact, if
z = −a = eiπ a, then
z 2 + 1 = a2 e2πi + e2πi
= e2πi (a2 + 1)
and
p p
z 2 + 1 z=−a = eπi a2 + 1
p
= − a2 + 1,
√ √
where we have taken the branch of a2 + 1 for which 1 = 1.
By the symmetry principle, the lower half-plane, =z ≤ 0, with a cut
from the point z = 0 to the point z = −i is mapped onto the region
=w1 ≤ 0 by the function (2.8.13). Hence the function (2.8.13) maps the
cross in Fig 2.53(a) onto the cut shown in Fig 2.55(a).
v1 v2
–––– 0 –––– u1 u2
– √a 2 +1 √a 2+1 –1 0 1
(a) (b)
η1 arg w1 = 0
arg w1 = 2π 2
1 arg w1= 4π arg w1= 2π
1 2 2
0 ξ1
2
arg w1 = 4π 1 1
arg w1= 2π arg w1= 0
arg w1 = 2π
y v
A C
E' B' D'
0 B π x C' –1 0 1 A' u
D E
where 0 < arg ζ1 < 2π. The lower half-plane =z < 0 with the cut from
z = 0 to z = −i is mapped on the second sheet, where 2π < arg ζ1 < 4π.
The second mapping, ζ2 = ζ1 + 1, shifts both sheets of the Riemann
surface shown in Fig 2.56 to the√right by 1.
The third mapping, ζ3 = ζ2 , sends the first and second sheets of
the Riemann surface of Fig 2.55(a) onto the upper and lower half-planes,
respectively, so that the regions in Fig 2.56 are mapped onto the whole
complex plane in Fig 2.55(a) with a cut along the real axis from −∞ to
+∞. In particular, the exterior of the circle √
in Fig 2.53(b)
√ is mapped onto
the region in Fig 2.55(a) with a cut from − a2 + 1 to a2 + 1 along the
real axis.
D C
γ (α)
γ (β)
0 x
y C
γ
γ (α) γ '(t)
γ (β)
α t β 0 x
Remark 3.1.1. A path C can have multiple points, that is, γ(t1 ) =
γ(t2 ) for some t1 6= t2 (see Fig 3.2). Every point of a path can be a multiple
point, as will be illustrated in the following example.
Example 3.1.1. Consider the path C given by the continuous mapping
γ : [0, 1] → C defined by
t 7→ e2πiνt , ν ∈ R, ν 6= 0.
One sees that γ(I) is a subset of the unit circle. However, if ν = n is a
positive integer, the unit circle is traversed n times.
Example 3.1.2. Consider the path given by the mapping γ : [0, 2] → C
defined by
(
c(1 − t) + dt, 0 ≤ t ≤ 1,
γ(t) =
d(2 − t) − c(1 − t), 1 ≤ t ≤ 2.
One sees that the path C is the segment with endpoints c and d, traversed
from c to d and from d to c. One also sees that t = 1 is a point of discon-
tinuity of γ 0 .
It is important to distinguish between a path C given by γ and the
corresponding curve γ(I) which is the pointset covered by C. In fact, a
path is a parametrized curve and the parametrization is as important as
the curve itself.
We have the following terminology.
Definition 3.1.2.
(a) A path C given by γ is contained in an open set D if γ(I) ⊂ D.
(b) If I = [α, β], then γ(α) and γ(β) are the initial and terminal points
of C.
(c) C is a closed path if γ(α) = γ(β).
3.2. COMPLEX LINE INTEGRALS 111
y
C
γ (α)
1
γ1 γ (β)
γ2 1
0 ϕ x
α β ξ η
It is seen that the value of the integral depends upon the number of times
the path traverses the unit circle.
Two paths, C1 and C2 given by γ1 : [α, β] → C and γ2 : [c, d] → C,
respectively, are said to be equivalent if there exists a strictly increasing
bijection ϕ : [α, β] → [e e (see Fig 3.3) which is continuous and piecewise
α, β]
continuously differentiable together with its inverse ϕ−1 , such that
γ1 (t) = γ2 ◦ ϕ(t) =: γ2 ϕ(t) , t ∈ [α, β].
The following useful invariance theorem holds for integrals along equiv-
alent paths.
Theorem 3.2.1. Let C1 and C2 be two equivalent paths. Then
Z Z
f (z) dz = f (z) dz. (3.2.2)
C1 C2
Proof. Letting ϕ denote the bijection between the two paths and
applying the definitions, we have
Z Z β
f (z) dz = f γ1 (t) γ10 (t) dt
C1 α
3.2. COMPLEX LINE INTEGRALS 113
Z β
= f γ2 ◦ ϕ(t) γ20 ϕ(t) ϕ0 (t) dt
α
Z d
= f γ2 (u) du (u = ϕ(t))
c
Z
= f (z) dz.
C2
(c) Linearity.
Z Z Z
[af (z) + bg(z)] dz = a f (z) dz + b g(z) dz. (3.2.5)
C C C
114 3. COMPLEX INTEGRATION AND CAUCHY’S THEOREM
y
C2
C1 C
0 x
Z β
= < e−iϕ0 w(t) dt
α
Zβ −iϕ
≤ e 0 w(t) dt
α
Zβ
= |w(t)| dt.
α
Remark 3.2.1. Other types of complex line integrals have been defined.
In the language of differential geometry, one can define integrals of 0- and
1-forms along a curve C on a two-dimensional manifold.
Let C be a smooth curve given by the equation
z = γ(t), α ≤ t ≤ β.
If f (z) is a 0-form, that is, a smooth function defined on C, the line
integral of f along C is
Z Z Z β
f ds = f (z) |dz| = f (z(t))|γ 0 (t)| dt. (3.2.9)
C C α
Example 3.2.2. For each of the following curves C with initial and
terminal points (0, 0) and (1, 1), respectively, as shown in Fig 3.5, compute
the line integral
Z Z
I1 = z̄ dz = (x − iy)(dx + idy)
ZC C
Z (3.2.15)
= (x dx + y dy) + i (−y dx + x dy),
C C
where
(a) C is a segment of the straight line y = x,
3.2. COMPLEX LINE INTEGRALS 117
y
1 B
y=x
(a)
(b) (c)
y = x2
A
0 1 x
As can be seen from this example, the value of the integral depends on
the path joining the points (0,0) and (1,1).
118 3. COMPLEX INTEGRATION AND CAUCHY’S THEOREM
Example 3.2.3. For the three paths, (a), (b) and (c), joining the points
(0, 0) and (1, 1), given in the previous example and shown in Fig 3.5, com-
pute the integral
Z Z
I2 = 2
z dz = (x2 − y 2 + 2ixy)(dx + idy)
C C
Z Z
= [(x − y 2 ) dx − 2xy dy] + i [2xy dx + (x2 − y 2 ) dy]. (3.2.16)
2
C C
y ζk z k+1
ηk z n–1
zk ∆zk
z2 z n= b
z1 C
z 0= a
0 ξk x
where ∆k = zk+1 − zk .
Definition 3.2.2. Given a rectifiable curve C in C and a continuous
function f (z) defined on C, if the integral sum (3.2.17) converges to a finite
limit as max |∆zk | → 0 independently of any particular subdivision of C
and of the choice of the points ζk , then this limit is called the complex line
120 3. COMPLEX INTEGRATION AND CAUCHY’S THEOREM
z2 − 1
11. f (z) = , C : z = 1 + (1 + i)t, 0 ≤ t ≤ 1.
z2
12. f (z) = z 2 , C : z = eit , 0 ≤ t ≤ π.
13. f (z) = |z|4 , C : |z| = 4, 0 ≤ arg z ≤ 2π.
2
14. f (z) = z <z, C : z = 1 + (2 + i)t, 0 ≤ t ≤ 1.
2 it
15. f (z) = (=z) , C: z=e , −π/2 ≤ t ≤ π/2.
it
16. f (z) = z̄, C : z = 2e , 0 ≤ t ≤ π.
17. f (z) = Arg z, C : |z| = R, 0 ≤ arg z ≤ π.
18. f (z) = z̄|z|, C : |z − i| = 1, taken counterclockwise.
Log2 z
19. f (z) = , C : the line segment joining the point z = 1
z
to the point z = 2 + i.
20. f (z) = z cos z, C : the arc z = it with 0 ≤ t ≤ π.
Use the M L-inequality to obtain an upper bound for the following integrals,
where M is an upper bound for the modulus of the integrand and L is the
length of the curve of integration.
Z
1
21. dz, where C : |z − 1 + i| = 2.
C z − 1+i
Z
22. [(2 + i)z 2 + 3iz] dz, where C : |z| = 1.
C
Z
1
23. 2 (z 2 + 4)
dz, where C : |z| = 2, 0 ≤ Arg z ≤ π/4.
C z
Z z
e −1
24. dz, where C : |z| = 1, 0 ≤ Arg z ≤ π.
C z
Z
25. Find an upper bound for the integral u(z)/z 2 dz, where CR is the
CR
circle |z| = R,Z and u(z) is a continuous function which is bounded for all
z. Find lim u(z)/z 2 dz.
R→∞ C
R Z
1
26. Find an upper bound for the integral 2
Log z dz, where CR is the
CR Zz
1
semicircle |z| = R, 0 ≤ Arg z ≤ π. Find lim Log z dz.
R→∞ C z 2
R
y y
R C+ R
C–
0 x 0 x
(a) (b)
and
I ZZ
∂u ∂v
v dx + u dy = − dx dy
C R ∂x ∂y
ZZ (3.3.6)
∂v ∂v
= − dx dy = 0.
R ∂y ∂y
It then follows that I
f (z) dz = 0.
C
Note 3.3.1. If a line integral is equal to zero along every closed path
lying in a simply connected domain D, then the value of the integral does
not depend on the path joining any two points in D and lying entirely in
D (see [32], p. 510).
Since f (ζ) is continuous at the point z, then for every ε > 0 there exists
δ > 0 such that, if |4z| < δ, then
max |f (ζ) − f (z)| < ε,
ζ∈[z,z+4z]
and hence
F (z + 4z) − F (z)
− f (z) < ε.
4z
126 3. COMPLEX INTEGRATION AND CAUCHY’S THEOREM
y
C
A B Cδ
A' B' 0 1 x
We note that the integral is not equal to zero since the path |z| = 1
is not closed (see Fig 3.8). In order to close the path, one has to integrate
along: (a) the upper cut from A to B, (b) the circle Cδ , of small radius δ,
taken in the clockwise direction, and (c) the lower cuts from B 0 to A0 :
Z Z Z Z
√ √
z dz = + + z dz.
ABCδ B 0 A0 AB Cδ B 0 A0
y
γn
γ2 D
Cn C2
C
C1
γ1 x
0
The two integrals along the arcs γk and −γk add up to zero since γk is
traversed twice, but in opposite directions. Therefore from (3.3.16) we
obtain I n I
X
f (z) dz + f (z) dz = 0, (3.3.17)
C k=1 Ck
where C and all the Ck are traversed either in the positive or in the negative
direction. More specifically, formula (3.3.17) can be written in the form
I Xn I
f (z) dz = f (z) dz. (3.3.18)
C+ k=1 Ck+
Example 3.3.5. Show that Cauchy’s Theorem holds for the function
f (z) = 1/z and the closed paths |z| = 2 and |z| = 1, that is, prove that
I I
dz dz
= = 2πi. (3.3.20)
|z|=2 z |z|=1 z
y
D
z0 C
Cρ
0 x
y
D
z0
C1
C2
0 x
z0 C
2
0 2 x
Since the integral on the left-hand side of (3.4.5) does not depend on ρ,
then, in the limit as ρ → 0, the formula
I
f (z)
dz = 2πif (z0 )
C z − z0
follows from (3.4.5).
z0 C
1
0 1 x
f (z) = (z − 1)7/3 ,
the point z = 1 is a branch point and single analytic branches of f (z) exist
in each domain with a cut joining the points z = 1 and z = ∞.
where z0 , z ∈ D and the integral (3.4.17) is computed along any path lying
entirely in D, is analytic in D and F 0 (z) = f (z). Then by Theorem 3.4.3,
F 00 (z) = f 0 (z) in D. Thus, f (z) is analytic in D.
138 3. COMPLEX INTEGRATION AND CAUCHY’S THEOREM
we have
I
1 f z0 + R eiθ
f (z0 ) = R eiθ i dθ
2πi C R eiθ
Z 2π
1
= f z0 + R eiθ dθ
2π 0
I
1
= f z0 + R eiθ dl,
2πR C
which is (3.4.20).
Remark 3.4.1. Instead of formula (3.4.21), some authors (see, for ex-
ample, [42], p. 68, formula (6)) use the form
Z 2π
1
u(z0 ) = u z0 + R eiθ dθ.
2π 0
This form of the mean-value theorem for harmonic functions may be mis-
leading since u(x, y) is a function of the real variables x and y.
140 3. COMPLEX INTEGRATION AND CAUCHY’S THEOREM
z1 = 0, z2 = 2, z3 = 2 + 2i, z4 = 2i.
I 2
z cos z
3. dz, where C is the circle |z − 1| = 3/2.
C z2 − 4
I
ez (z + 4)
4. 2
dz, where C is the circle |z − 2| = 2.
C z +9
Z
dz
Evaluate the integral 2+4
along the following circles taken counter-
C z
clockwise.
5. |z − 4| = 1.
6. |z − 1| = 3/2.
7. |z + 2| = 1.
Z
z sin z
Evaluate the integral dz along the following circles taken coun-
C z3 + 8
terclockwise.
8. |z| = 1.
9. |z + 2| = 1.
10. |z − 1 − 2i| = 2.
142 3. COMPLEX INTEGRATION AND CAUCHY’S THEOREM
12. Let f (z) and g(z) be analytic in a simply connected domain D. Prove
that Z β β Z β
f (z)g(z) dz = [f (z)h(z)] − f 0 (z)h(z) dz,
α α α
where h(z) is an indefinite integral of g(z) in D and the path of integration
lies in D.
13. Use formula (3.4.8) with the circle C = {z; |z − z0 | = r} taken in the
positive direction to establish Cauchy’s estimate:
n!
|f (n) (z0 )| ≤ max |f (z)|, n = 0, 1, 2, . . . , (3.4.24)
rn |z−z0 |=r
of degree n > 0 has at least one zero. Use Liouville’s Theorem to prove the
fundamental theorem of algebra.
(Hint: Consider the function 1/p(z).)
21. Let the function f (z) be analytic in a domain D containing the closed
disk |z| ≤ r. If |f (z)| is constant on |z| = r and f (z) 6= 0 for |z| < r, show
that f (z) is constant.
22. If f (z) is analytic for |z| < 1 and |f (z)| ≤ 1/(1 − |z|), find the best
estimate of |f (n) (0)| that Cauchy’s estimate (3.4.24) will yield.
23. Show that the successive derivatives of an analytic function at a point
can never satisfy the inequality |f (n) (z)| > n!nn . Formulate a sharper
theorem of the same kind.
24. Prove that there is no function analytic in |z| ≤ 1 such that
1 1 19
|f (z)| ≤ 1 on |z| = 1, f = 0, f − = .
2 2 20
25. The function of a complex variable defined by f (z) = cos z is analytic
everywhere and satisfies the inequality | cos x| ≤ 1 for all real x. However,
it is not a constant. Is there a contradiction with Liouville’s Theorem?
144 3. COMPLEX INTEGRATION AND CAUCHY’S THEOREM
Among these four paths, there is one, called S (1) , such that
I I
f (z) dz ≥ f (z) dz , j = 1, 2, 3, 4.
(1)
S Sj
y C
A B
0 x
Let L(S) and D(T ) denote the length of S and the diameter of T , respec-
tively. Then, we have
1 1
L(Sj ) = L(S), D(Tj ) = D(T ).
2 2
Finally, by (3.5.1), we have
I I
f (z) dz ≤ 4 f (z) dz .
S S (1)
Now performing the same process on S (1) , we obtain a triangle S (2) with the
analogous properties. By induction, we get a sequence {S (n) } of closed tri-
angular paths and closed sets {T (n)}, each consisting of the region enclosed
by S (n) and its boundary. Thus we have
T (1) ⊃ T (2) ⊃ . . . , (3.5.2)
I I
f (z) dz ≤ 4 f (z) dz , (3.5.3)
S (n) S (n+1)
1
(n+1)
L S = L S (n) , (3.5.4)
2
1
(n+1)
D T = D T (n) . (3.5.5)
2
These relations imply:
I I
f (z) dz ≤ 4n
(n) f (z) dz , (3.5.6)
S S
1 n
L S (n) = L(S), (3.5.7)
2
1 n
D T (n) = D(T ). (3.5.8)
2
Since the T (n) are closed, then their intersection is non empty and consists
of a single point z0 ,
\∞
z0 = T (n) .
n=1
Let ε > 0. Since f (z) has a derivative at z0 , we can find a δ > 0 such
that Dzδ0 ⊂ G and
f (z) − f (z0 )
− f 0
(z
0 < ε,
)
z − z0
whenever 0 < |z − z0 | < δ, that is,
|f (z) − f (z0 ) − f 0 (z0 )(z − z0 )| < ε|z − z0 |, (3.5.9)
whenever 0 < |z − z0 | < δ.
146 3. COMPLEX INTEGRATION AND CAUCHY’S THEOREM
147
148 4. TAYLOR AND LAURENT SERIES
exist and are finite. This justifies the notation Sx = <S and Sy = =S
implicitely used in (4.1.4).
Thus, the convergence of a series of complex numbers can be reduced to
the convergence of two series of real numbers. Therefore, we shall use known
convergence tests for series of positive numbers, such as the comparison test,
the ratio test and the root test (see, for example, [50], pp. 20–23).
Theorem 4.1.1 (necessary condition for convergence).
P Let
{an } be a sequence of positive numbers. If the series ∞
n=1 an converges,
then lim an = 0.
n→∞
Pn
Proof. If the limit S = lim Sn of the partial sums, Sn = k=1 ak ,
n→∞
exists, then S = lim Sn−1 . Hence
n→∞
lim an = lim (Sn − Sn−1 ) = S − S = 0.
n→∞ n→∞
Theorem 4.1.2 (comparison test). Let {an } and {bn } be two sequences
of positive numbers, such that an < bn for all n ∈ N , and consider the two
series
X∞ X∞
an , bn . (4.1.6)
n=1 n=1
If the second series converges, so does the first. If the first series diverges,
so does the second.
P∞
Theorem 4.1.3 (ratio test). Let n=1 an be a series of positive num-
bers and suppose the limit L,
an+1
L = lim ,
n→∞ an
is finite. Then:
(a) If L < 1, the series converges.
(b) If L > 1, the series diverges.
(c) If L = 1, the question of convergence is open (the series may either
diverge or converge).
P
Theorem 4.1.4 (root test). Let ∞ n=1 an be a series of positive num-
bers and suppose the limit L,
L = lim a1/n
n ,
n→∞
is finite. Then:
(a) If L < 1, the series converges.
(b) If L > 1, the series diverges.
(c) If L = 1, the question of convergence is open (the series may either
diverge or converge).
4.1. INFINITE SERIES 149
Note 4.1.1. The ratio and root tests can be formulated in a more
general form by using the notion of limit superior:
an+1
lim sup , lim sup a1/n
n ,
an
which is the largest point of accumulation in case more than one such points
1/n
exist. This formulation is useful when the sequences an+1 /an and/or an
have no limit.
Example 4.1.1. Show that the series
X∞
1 1
an , where an = n [1 + (−1)n ] + n [1 − (−1)n ],
n=1
2 3
converges.
Solution. We have
2 2
a2n = , a2n+1 = 2n+1 .
22n 3
Therefore, the limit of an+1 /an as n → ∞ does not exist since the two
subsequences
a2n+2 a2n+2 a2n+1 a2n+3 a2n+3 a2n+2
= , =
a2n a2n+1 a2n a2n+1 a2n+2 a2n+1
have different limits:
a2(n+1) 2 × 2−2(n+1) 1
lim = lim = ,
n→∞ a2n n→∞ 2 × 2−2n 4
a2(n+1)+1 2 × 3−(2n+3) 1
lim = lim −(2n+1)
= .
n→∞ a2n+1 n→∞ 2×3 9
In this case,
an+1 1
lim sup = < 1,
n→∞ a n 4
P∞
so that the series n=1 an converges.
Note 4.1.2. One can ask the following question: Can a series converge
according to the ratio test but diverge according to the root test? The
answer is in the negative by the following theorem (see [29], Vol. 1, p. 437).
Theorem 4.1.5. Consider a sequence {an } of positive numbers. If the
limit
an+1
L = lim
n→∞ an
exists and is finite, then the limit
M = lim a1/n
n
n→∞
exists and is finite, and M = L.
150 4. TAYLOR AND LAURENT SERIES
P∞
Definition 4.1.3. PIf the series of absolute terms n=1 |zn | is conver-
∞
gent, then the series n=1 zn is said to be absolutely convergent. On the
other hand, a convergent series which is not absolutely convergent is said
to be conditionally convergent.
In the particular case of a sequence of real numbers we have the fol-
lowing theorem (see [29], Vol. 1, p. 418).
P∞
Theorem 4.1.6. If the series n=1 an of real numbers is absolutely
convergent, then it is convergent.
We use this theorem to prove the next one for sequences of complex
numbers.
P
Theorem 4.1.7. If the series ∞ n=1 zn of complex numbers is absolutely
convergent, then it is convergent.
Proof. We suppose that the series of absolute terms
∞
X ∞ p
X
S= |zn | = x2n + yn2
n=1 n=1
is convergent. Hence, it follows from the inequalities
p p
|xn | ≤ x2n + yn2 , |yn | ≤ x2n + yn2
and the comparison test for series of positive numbers (Theorem 4.1.2) that
the series
X∞ X∞
|xn |, |yn |
n=1 n=1
converge. Therefore, by Theorem 4.1.6, the two series
∞
X X∞
Sx = xn , Sy = yn
n=1 n=1
also converge, that is, the series
X∞ X∞
zn = (xn + iyn ) = Sx + iSy
n=1 n=1
is convergent.
For instance, the series
X∞
5n + 7i
S=
n=1
n3
is absolutely convergent since
X∞ ∞ ∞
5n + 7i X 5 X 7
|S| ≤ ≤ + < ∞.
n3 n 2 n3
n=1 n=1 n=1
4.1. INFINITE SERIES 151
diverges.
As in the real case, the converse of Theorem 4.1.7 is not true in general.
For instance, the series
X∞
2 + 3i
(−1)n
n=1
n
is convergent (its real and imaginary parts are conditionally convergent
series), but the series
X∞ ∞
|2 + 3i| √ X 1
= 13
n=1
n n=1
n
is divergent.
y y
D D
C d
D'
z0 D' C
d
0 x 0 x
(a) (b)
since the functions wn (z) are analytic in D. Since the conditions of Morera’s
Theorem 3.4.4 are satisfied, S(z) is analytic in D.
(b) Since z0 is located inside C and C is a closed set, then
min |z − z0 | = d > 0. (4.1.11)
z∈C
property. Indeed, if Sn denotes the nth partial sum of the uniformly con-
vergent series (4.1.10), then for all ε > 0 there exists N = Nε such that for
all n > Nε the inequality
|S(z) − Sn (z)| < εdk+1 (4.1.13)
holds. Then by (4.1.11) and (4.1.13), for all n > Nε we have
S(z) Sn (z) 1
(z − z0 )k+1 − (z − z0 )k+1 = |z − z0 |k+1 |S(z) − Sn (z)|
1
< k+1 εdk+1 = ε.
d
This inequality implies that series (4.1.12) is uniformly convergent on C and
in some neighborhood of C and, therefore, it can be integrated termwise
along C:
I X∞ I
k! S(z) k! wn (z)
k+1
dz = dz. (4.1.14)
2πi C (z − z0 ) n=1
2πi C (z − z0 )k+1
Since S(z) and wn (z) are analytic in D0 , and C lies inside D0 , then by
(3.4.8) for the kth derivative of an analytic function, (4.1.14) gives
∞
X
S (k) (z0 ) = wn(k) (z0 ).
n=1
0
Since z0 is an arbitrary point in D , the second statement of the theorem
is proved.
(c) Let C be an arbitrary closed path lying entirely in a simply connected
subregion of D and D0 an arbitrary simply connected closed subregion
surrounded by C at distance at least d > 0 (see Fig 4.1(b)),
d = min0 |z − ζ|. (4.1.15)
z∈D
ζ∈C
is divergent.
At first glance, this fact seems to contradict Weierstrass’ Theorem,
since the function sin nz is differentiable an arbitrary number of times in
the whole complex plane and the series
X∞
sin nz
(4.1.21)
n=1
n3
should be convergent for some values of z. We show that no such contra-
diction exists. The reason is that, if z = x + iy and y 6= 0, then
q
| sin n(x + iy)| = sin2 nx cosh2 ny + cos2 nx sinh2 ny
> | sinh2 ny| → ∞, as n → ∞,
for any y 6= 0, no matter how small. Therefore the series (4.1.21) is di-
vergent in the whole complex plane except on the real axis z = x. Since
the line z = x is not a domain, Weierstrass’ Theorem on the possibility of
differentiating the series (4.1.21) termwise cannot be applied.
In the case of series of functions of a real variable, termwise differenti-
ation is more restrictive, as stated in the following theorem.
Theorem 4.1.11. The series
X∞
un (x), a ≤ x ≤ b, (4.1.22)
n=1
converges pointwise in the half-plane 0 < <z < ∞, but not uniformly.
Solution. For any fixed z = x + iy, with x > 0,
X −(n+1)z
∞ e
|S(z) − Sn (z)| = e−kz = →0
1−e
−z
k=n+1
implies
1
(N + 1)x > − log ε = log ,
ε
that is,
1 1
N +1> log .
x ε
Hence, in the last inequality, N depends on x since N + 1 → ∞ as x → 0+.
Therefore the series (4.1.23) is not uniformly convergent in the right-hand
half-plane 0 < <z < +∞, but it can be differentiated termwise any number
of times there.
X∞
sin(in3 )
15. .
n=1
n3 + 1
X∞
sin(i/n)
16. .
n=2
n Log n
160 4. TAYLOR AND LAURENT SERIES
P∞ P∞
Let A = n=1 zn and B = n=1 ζn be two convergent series. Show that
the following relations hold.
X∞
17. A + B = (zn + ζn ).
n=1
∞
X
18. cA − dB = (czn − dζn ), where c and d are constants.
n=1
Suppose that A = lim zn .
n→∞
19. Show that |A| = lim |zn |.
n→∞
X∞
sin nz
23. . (Hint: Use Corollary 4.2.3.)
n=1
n
∞
X zn
24. .
n=1
1 + z 2n
∞
X 4n
25. .
n=1
1 + zn
X∞
n(n + 2)
27. .
n=1
(z + 1)n
∞
X π
28. sin zn.
n=1
n
X∞
sin(n|z|)
29. .
n=1
n2
4.2. INTEGER POWER SERIES 161
is called an (integer) power series. The complex numbers an are called the
coefficients of the series.
As in the real case, the following theorem plays an important role in
the investigation of the convergence of complex power series.
Theorem 4.2.1 (Abel’s Theorem). (1) If the power series (4.2.1) is
convergent at the point z1 , then it is absolutely convergent in any open disk
|z| < |z1 |. Moreover, in each closed disk |z| ≤ q|z1 |, 0 < q < 1, the series
converges uniformly and absolutely.
(2) If the series (4.2.1) is divergent at a point z2 , then it is divergent
in the region |z| > |z2 |.
Proof. Part (1). Since the series (4.2.1) is convergent at the point z1 ,
it follows from the necessary condition of convergence (see Theorem 4.1.1)
that
lim an z1n = 0. (4.2.2)
n→∞
Therefore
lim |an z1n | = 0. (4.2.3)
n→∞
It follows from (4.2.3) that there exists a positive number, M , such that
for all n = 1, 2, . . . , the inequality
|an z1n | < M (4.2.4)
|z |
is satisfied. Let z be an arbitrary interior point of the disk D0 1 , that is,
|z| ≤ q|z1 | for 0 < q < 1. Then, using (4.2.4), we have
n n
z z
|an z n | = |an | |z1 |n < M = M q n .
z1 z1
P∞ n
Therefore, P∞ the series n=1 an z is clearly majorized by the convergent
n
series
P∞ M n=1 q . Hence, by Weierstrass’ Theorem 4.1.8, the former series,
n
n=1 a n z , is absolutely and uniformly convergent in the disk |z| < q|z1 |.
Since the number q can be as close to 1 as we please, we can conclude that
series (4.2.1) is convergent for all z in the disk |z| < |z1 |.
162 4. TAYLOR AND LAURENT SERIES
y
z2
z1 R
0 x
the region |z| > R is called the radius of convergence of the power series
(4.2.1).
The ratio or root tests can be used to determine the radius of conver-
gence, that is Theorems 4.1.3 or 4.1.4, as in the real case.
Corollary 4.2.2. The radius of convergence, R, of the power series
(4.2.1) is given by the following limits, if they exist, or the limits superior:
an an
R = lim , R = lim sup (4.2.6)
n→∞ an+1 n→∞ an+1
or
1 1
R = lim , R = lim sup . (4.2.7)
n→∞ |an |1/n n→∞ |an |
1/n
Proof. We derive only the first formula in (4.2.6), for which we assume
that the limit exists. Consider the series of absolute values of the terms in
(4.2.1) for a fixed value of z:
∞
X
|an z n |. (4.2.8)
n=0
Since (4.2.8) is a series with positive numbers, then, for every fixed z, we can
use the ratio test (Theorem 4.1.3) to investigate the region of convergence
of the series.
We assume that the limit
an+1 z n+1 an+1
L = lim
= |z| lim (4.2.9)
n→∞ an z n n→∞ an
exists. In order to have convergence of (4.2.8) it suffices to satisfy the
inequality L < 1,
an+1 an
|z| lim
< 1, that is, |z| < lim . (4.2.10)
n→∞ an n→∞ an+1
We now prove that series (4.2.1) is divergent in the region |z| > R. Indeed,
since the inequality |z| > R corresponds to the inequality L > 1, it follows
from (4.2.10) that there exists a number N such that, for all n > N , the
following inequality is satisfied:
an+1 z n+1
n+1
| > |an z n |.
an z n > 1 that is, |an+1 z
164 4. TAYLOR AND LAURENT SERIES
Definition 4.2.3. The disk |z| < R, where R is the radius of conver-
gence, is called the disk of convergence of the series (4.2.1).
The series (4.2.1) can either converge or diverge at points on the bound-
ary of the disk.
For example, the series
X∞
zn
n=1
n
has radius of convergence R = 1 as can be seen from (4.2.6). On the circle
|z| = 1 where z = eiθ , the series is divergent only at the point z = 1, that
is, if θ = 0 or θ = 2π, but at all other points, z = cos θ + i sin θ, of the
circle, the series
X∞ X∞ X∞
cos nθ + i sin nθ cos nθ sin nθ
= +i (4.2.11)
n=1
n n=1
n n=1
n
is conditionally convergent.
Tests sharper than the ratio or the root tests can be used to prove the
last statement. Such tests, like the Dirichlet–Abel Test (see [29], Vol. 1,
p. 429), can be used to determine the conditional convergence of alternating
series, that is, series with terms which change signs.
Theorem 4.2.2 (Dirichlet–Abel Test). Let {an } and {bn } be two se-
quences of complex numbers such that
(a) lim an = 0,
n→∞
X∞
(b) |an+1 − an | converges, and
n=0 P
(c) the partial sums of the series bn are bounded, that is,
|b1 + b2 + · · · + bn | = |Sn | ≤ M, n = 1, 2, . . . .
Then the series
∞
X
an b n (4.2.12)
n=1
converges.
4.2. INTEGER POWER SERIES 165
P∞
Proof. With bn = (−1)n+1 the partial sums of n=1 bn are always
either 1 or 0.
is conditionally convergent.
Solution. Set an = 1/n and bn = cos nθ. Clearly, an ≥ an+1 → 0 as
n → ∞. If 0 < θ0 ≤ θ ≤ 2π − θ0 , the partial sum
n
X sin([n + 1]θ/2) − sin(θ/2)
Sn = cos kθ = (4.2.14)
2 sin(θ/2)
k=1
is bounded by
2
|Sn | < ,
2| sin(θ0 /2)|
for each n > 0. Thus, the conditions of Corollary 4.2.3 are satisfied and the
given series (4.2.13) is convergent.
is conditionally convergent.
Example 4.2.2. Show that the two series
X∞ X∞
| cos nθ| | sin nθ|
,
n=1
n n=1
n
are divergent.
Solution. If the first series would be convergent then, by the obvious
inequality
| cos nθ| ≥ cos2 nθ,
the series
X∞ X∞ X∞ ∞
cos2 nθ 1 + cos 2nθ 1 X cos 2nθ
2 = = + (4.2.15)
n=1
n n=1
n n=1
n n=1 n
EXERCISES FOR SECTION 4.2 167
would also be convergent. But this is false, since the last series converges
for θ 6= kπ but the second-last series diverges. Therefore, the two series
X∞ X∞
cos2 (nθ) | cos(nθ)|
,
n=1
n n=1
n
diverge. Thus the series
X∞
cos(nθ)
, θ 6= kπ,
n=1
n
is only conditionally convergent. The divergence of the second series follows
in the same way.
Corollary 4.2.5. The sum, S(z), of the power series (4.2.1) is ana-
lytic inside every disk |z| ≤ R1 < R that lies entirely in the disk of conver-
gence |z| < R.
Proof. Since the terms an z n of the power series are analytic in the
whole complex plane and the series (4.2.1) is uniformly convergent in the
region |z| ≤ R1 , then, by the first part of Weierstrass’ Theorem 4.1.10, S(z)
is analytic if |z| ≤ R1 < R.
Corollary 4.2.6. Power series can be differentiated and integrated
any number of times inside their disk of convergence. Moreover, the radius
of convergence of the differentiated (or integrated) series is equal to the
radius of convergence of the original series.
Proof. This fact is a consequence of the second part of Weierstrass’
Theorem 4.1.10.
∞
X
5. sin(in) (z + 2)n .
n=1
X∞
n4
6. (z − i)n .
n=1
n!
∞
X π n
7. 1+ (z + i)n .
n=1
n
∞
X
8. nn z n .
n=1
P∞
Suppose that the radii of convergence of the power series n=1 an z n and
P ∞ n
n=1 bn z are equal to R1 and R2 , respectively, where 0 < R1 < ∞ and
0 < R2 < ∞. Estimate the radius of convergence, R, of each of the following
power series.
X∞
9. (an + bn )z n .
n=1
∞
X
10. (an − bn )z n .
n=1
∞
X
11. an b n z n .
n=1
X∞
an n an bn
12. z , bn 6= 0, lim and lim exist.
b
n=1 n
n→∞ an+1 n→∞ bn+1
∞
X
13. nan z n .
n=1
∞
X
14. n k an z n , k ∈ N.
n=1
X∞
an n
15. z .
n=1
n
X∞
an n
16. z , k ∈ N.
n=1
nk
17. Does there exist a power series in powers of z that converges at z = 3+4i
and diverges at z = −3 + 3i? Explain.
Find the sum of the following series.
4.3. TAYLOR SERIES 169
∞
X
18. nz n−1 .
n=1
X∞
19. n2 z n .
n=1
X∞
zn
20. .
n=1
n+1
∞
X zn
21. .
n=2
n(n − 1)
y
D
R
CR
z z0
0 x
Therefore, one can expand the right-hand side of (4.3.4) in a power series
in (z − z0 )/(ζ − z0 ):
∞
X
1 (z − z0 )n
= . (4.3.5)
ζ − z n=0 (ζ − z0 )n+1
We prove that series (4.3.5) is uniformly convergent with respect to ζ and
z for all ζ ∈ CR and all z strictly inside the disk DzR0 . Indeed, since z is an
interior point, there exists ρ > 0 such that |z − z0 | < ρ < R and
z − z0 ρ
ζ − z0 < R < 1.
f (z) − f (z0 )
= a1 + a2 (z − z0 ) + · · · + an (z − z0 )n−1 + . . . . (4.3.11)
z − z0
Formula (4.3.11) shows that the limit
f (z) − f (z0 )
lim = a1 = f 0 (z0 )
z→z0 z − z0
exists and is finite, that is, f (z) is differentiable at z = z0 .
However, there are cases where a Taylor series expansion (4.3.10) can
be found only by using integrals.
Example 4.3.3. Find the Taylor series expansion of f (z) = e1/z with
center at z0 = 2.
Solution. Since z = 0 is the only singular point of f (z), then the
radius of convergence is equal to 2. Using (4.3.2) we obtain
I
f (n) (2) 1 e1/ζ
= dζ, (4.3.21)
n! 2πi CR (ζ − 2)n+1
EXERCISES FOR SECTION 4.3 175
z
7. , z0 = 0.
z2 + 4
z+2
8. , z0 = 0.
(z − 1)2
9. z 4 + 2z 3 − z + 1, z0 = 2.
z
10. , z0 = 2i.
(z + i)(z + 3i)
z−2
11. , z0 = −1.
(z + 3)(z − 1)
z2
12. , z0 = 3.
(z − 1)2 (z + 2)
Using Taylor series expansion for elementary functions given in Section 4.3,
solve the following problems.
13. Prove that (sin z)0 = cos z.
14. Prove that (cosh z)0 = sinh z.
z3 z5
15. Show that tan z = z + +2 + . . ., |z| < π/2.
3 15
z2 z4
16. Show that sec z = 1 + +5 + . . ., |z| < π/2.
2 24
EXERCISES FOR SECTION 4.3 177
Find the Taylor series of the given functions about the given point, z0 , and
determine their radii of convergence.
17. cos(3z − 2), z0 = 1.
18. Log(3 + z), z0 = 0.
2
19. ez +2z
, z0 = −1.
20. sin(2z − 5), z0 = −2.
Find the first three nonzero terms of the Taylor series about the given point,
z0 , and determine the radius of convergence of the series.
cos2 z
21. , z0 = 0.
1 + z2
z
22. z , z0 = 0.
e −1
23. Log(1 + cos z), z0 = 0.
1
24. , z0 = 0.
1 + cos z
25. e1/z , z0 = 1.
z
26. sin , z0 = 0.
1+z
The series
∞
X (−1)m x2m+n
Jn (x) = (4.3.26)
m=0
22m+n m!(n+ m)!
defines the Bessel function of the first kind of order n for n ∈ N. Using
(4.3.26) and properties of power series, derive the following relations.
27. [xn Jn (x)]0 = xn Jn−1 (x).
28. [x−n Jn (x)]0 = −x−n Jn+1 (x).
Using Exercises 27 and 28, show that
2n
29. Jn−1 (x) + Jn+1 (x) = Jn (x).
x
30. Jn−1 (x) − Jn+1 (x) = 2Jn0 (x).
Use Exercises 27–30 to evaluate the integrals
Z
31. J3 (x) dx.
Z
32. x3 J0 (x) dx.
178 4. TAYLOR AND LAURENT SERIES
y
D
R2
CR
R1 2
z0 CR
1
0 x
C
where
I
1 f (ζ)
cn = dζ, n = 0, ±1, ±2, . . . , (4.4.3)
2πi C (ζ − z0 )n+1
called the Laurent series of f (z) in the annulus (4.4.1).
Proof. Take a circle Cρ of radius ρ centered at z and lying entirely in-
side the annulus for ρ sufficiently small (see Fig 4.5). By Cauchy’s Theorem
4.4. LAURENT SERIES 179
y
R2
z Cρ
CR
R1 2
z0 CR
1
0 x
The integrands in (4.4.6) and (4.4.9) are the same. We show that the
paths of integration, CR1 and CR2 , in these integrals can be replaced by an
arbitrary closed path C lying entirely inside the annulus (4.4.1). Indeed,
the function f (ζ)(ζ − z0 )−n−1 is analytic in the region between CR2 and C
since its only singular point, ζ = z0 , lies outside this domain. Therefore,
by Cauchy’s Theorem for multiply connected domains,
I I
f (ζ) f (ζ)
n+1
dζ = n+1
dζ. (4.4.10)
C (ζ − z0 ) CR2 (ζ − z0 )
4.4. LAURENT SERIES 181
Substituting (4.4.10) and (4.4.11) into (4.4.6) and (4.4.9), respectively, and
substituting the results into (4.4.5), we obtain
−1
X ∞
X
f (z) = cn (z − z0 )n + cn (z − z0 )n , (4.4.12)
n=−∞ n=0
It follows from the derivation of (4.4.3) and (4.4.14) that the series
(4.4.14) is absolutely and uniformly convergent in every closed annulus
lying entirely inside the annulus (4.4.1).
Note 4.4.1. If f (z) is analytic not only in the annulus (4.4.1) but in
the whole disk D : |z − z0 | < R2 , then f (z)/(z − z0 )n+1 is analytic in D for
n = −1, −2, −3, . . . Therefore, by Cauchy’s Theorem for simply connected
domains,
I
1 f (ζ)
cn = dζ = 0, for n = −1, −2, −3, . . . .
2πi C (ζ − z0 )n+1
In this case, the Laurent series (4.4.14) reduces to the Taylor series
∞
X f (n) (z0 )
f (z) = cn (z − z0 )n , cn = .
n=0
n!
are called the principal and regular parts, respectively, of Laurent series
(4.4.14).
182 4. TAYLOR AND LAURENT SERIES
∞
X
1 n+1
= , |z| > 1. (4.4.18)
(1 − z)2 n=0 z n+2
Similarly, one can get Laurent series expansions for log (1 − 1/z), e1/z ,
sin 1/z, cos 1/z by replacing z with 1/z in (4.3.15)–(4.3.18).
To find a Laurent series for a proper rational function Pn (z)/Qm (z),
where n < m, it suffices to expand it into a sum of partial fractions and use
the Taylor series (4.4.15), (4.4.17) or the Laurent series (4.4.16), (4.4.18) or
consequences of these formulae which can be found by differentiation the
necessary number of times.
Example 4.4.1. Find the Laurent series of the function
1
f (z) =
(z − 2)(z − 3)
(a) in the annulus 2 < |z| < 3,
(b) in the region 3 < |z| < ∞.
Solution. First of all let us convince ourselves that it is possible to
find the desired expansions. The only singular points of f (z) are z = 2 and
z = 3. These points do not lie inside the annulus 2 < |z| < 3 or in the
region |z| > 3. Therefore, in both cases (a) and (b), it is possible to find
Laurent series.
4.4. LAURENT SERIES 183
A similar approach can be used for the fraction 1/(z − 2). If this fraction
is expanded in a Taylor series by means of (4.4.15), then we obtain a series
which is convergent in the disk |z| < 2, but we need a series which is
convergent in the annulus 2 < |z| < 3. Therefore we cannot use a Taylor
series in this case. A Laurent series expansion by means of (4.4.16) gives a
series which is convergent in the region |z| > 2 and, therefore, in the region
2 < |z| < 3. This is what we need. Hence
∞
X
1 1 1 2n
= = , |z| > 2.
z−2 z 1 − (2/z) n=0 z n+1
where Z π
1
Jn (x) = cos (x sin θ − nθ) dθ. (4.4.21)
2π −π
184 4. TAYLOR AND LAURENT SERIES
The function Jn (x) is called the Bessel function of the first kind of
order n, and (4.4.21) is one of its integral representations. A power series
representation of Jn (x) is given in (4.3.26).
Solution. Since z = 0 is the only singular point of the function on the
left-hand side of (4.4.20), then its Laurent series expansion in powers of z
is possible in every annulus 0 < ρ < |z| < R. To determine the coefficients
cn , we use formula (4.4.3) and take the unit circle |z| = 1 as the path of
integration. Then, z = eiθ , dz = i eiθ dθ, and
I
1 e[z−(1/z)]x/2
cn = dz
2πi |z|=1 z n+1
Z π
1 iθ −iθ
= e−i(n+1)θ ex(e −e )/2 eiθ dθ
2π −π
Z π
1
= ei(x sin θ−nθ) dθ
2π −π
Z π Z π
1 i
= cos (x sin θ − nθ) dθ + sin (x sin θ − nθ) dθ.
2π −π 2π −π
Since the last integral is equal to zero, then
Z π
1
cn = cos (x sin θ − nθ) dθ = Jn (x).
2π −π
Example 4.4.3. Prove that
∞
X
euz+(v/z) = cn z n , (4.4.22)
n=−∞
where
Z π
1
cn = e(u+v) cos θ cos [(u − v) sin θ − nθ] dθ. (4.4.23)
2π −π
Z π
1
= e(u+v) cos θ cos [(u − v) sin θ − nθ] dθ.
2π −π
cos2 z
2. .
z
ez − 1 − z
3. .
z2
1 + z 2 /2 − cos z
4. .
z4
5. z 4 sin(1/z).
6. z e1/z .
7. ez+1/z .
8. cos(1/z) cos z.
Expand each of the following functions in a Laurent series about z0 .
z+2
9. , z0 = 3.
(z − 3)3
z−1
10. , z0 = −i.
(z + i)2
11. (z + 2) sin[1/(z − i)], z0 = i.
12. (z − 1)e1/(z−2) , z0 = 2.
cos z
13. , z0 = −4.
z+4
ez
14. , z0 = −1.
(z + 1)3
Expand each function in convergent Laurent series in the given domains.
1
15. , (a) |z| < 1, (b) 1 < |z| < 2, (c) 2 < |z| < ∞.
(z + 1)(z − 2)
1
16. , (a) |z| < 2, (b) 2 < |z| < 3, (c) 3 < |z| < ∞.
(z − 3)(z + 2)
2z + 1
17. 2 , (a) |z| < 1, (b) 1 < |z| < 3, (c) 3 < |z| < ∞.
z + 4z + 3
3z − 5
18. 2 , (a) |z| < 1, (b) 1 < |z| < 6, (c) 6 < |z| < ∞.
z + 5z − 6
1
19. , 3 < |z − 1| < 4.
(z + 2)(z + 3)
z
20. 2 , 3 < |z + 2| < 6.
z + 7z − 8
1
21. , 0 < |z + i| < 2.
(z 2 + 1)2
EXERCISES FOR SECTION 4.4 187
1
22. , 0 < |z − 3| < 3.
(z 2 − 9)2
CHAPTER 5
Note that the series (5.1.1) and (5.1.3) have the same disk of convergence.
It follows that if z0 is a zero of order m of f (z), then f (z) can be
represented in the form (5.1.2) where ϕ(z) is analytic at z = z0 , and ϕ(z0 ) 6=
0 and ϕ(z0 ) 6= ∞.
The converse statement is also true. If the function f (z) is analytic at
z = z0 and is represented in the form (5.1.2), where ϕ(z0 ) 6= 0, ϕ(z0 ) 6= ∞
and ϕ(z) is analytic, then z0 is a zero of order m of f (z).
Example 5.1.1. Determine the order of the zero of
f (z) = (z − 5)100 ez . (5.1.4)
5
Solution. Comparing (5.1.2) and (5.1.4) and noting that e 6= 0 and
e5 6= ∞, we immediately see that z = 5 is a zero of order 100 of f (z) since
f (100) (5) 6= 0. Moreover, it is the only zero of f (z) since ez 6= 0 for all
z ∈ C.
Example 5.1.2. Determine the order of the zero z = 0 of
sin10 z
f (z) = . (5.1.5)
z5
Solution. To define the function ϕ(z) we write
sin10 z
f (z) = z 5 =: z 5 ϕ(z),
z 10
where limz→0 ϕ(z) = 1 6= 0. Therefore z = 0 is zero of order 5 of f (z).
5.1.2. Isolated singularities. A point a ∈ C is called a singular
point of f (z) if f is not defined at a.
For examples, the points z = 1, z = 0 and z = ∞ are singular points
of z/(z − 1)2 , (sin z)/z and z + 1, respectively.
By Liouville’s Theorem 3.4.5, the only analytic functions which do not
have any singular point in the extended complex plane are the constant
functions, f (z) = constant.
Definition 5.1.2. A singular point z0 of f (z) is called an isolated sin-
gular point if there exists δ > 0 such that f (z) is analytic in the punctured
disk 0 < |z − z0 | < δ.
Not every singular point, z0 , is an isolated singular point, as can be
seen from the following example taken from [50], p. 98, par. 5.501.
Example 5.1.3. Show that the function
∞
X
n n
f (z) = z 2 + z 4 + z 8 + · · · + z 2 + · · · = z2 (5.1.6)
n=1
has infinitely many singular points on the unit circle |z| = 1.
5.1. SINGULAR POINTS OF ANALYTIC FUNCTIONS 191
or, equivalently,
1
f (z) = , ψ(z0 ) 6= 0, ψ(z0 ) 6= ∞.
ψ(z)(z − z0 )m
Thus, by (5.1.12), z0 is a pole of order m of f (z).
that is, given any positive integer N there exists c−n 6= 0 for infinitely many
n ≥ N , then the point z0 is called an essential singularity of f (z).
The behavior of an analytic function in a neighborhood of an essential
singularity is described by the following theorem, which goes by the name
Casorati–Weierstrass, Sokhotski, or simply Weierstrass’ Theorem.
Theorem 5.1.4 (Weierstrass’ Theorem). Let z0 be an isolated essential
singularity of a function f (z) which is analytic in a punctured disk 0 < |z −
z0 | < δ. Given any ε > 0 and w ∈ C, then, in any punctured neighborhood
of z0 , there exists at least one point z such that |f (z) − w| < ε.
Proof. Assume to the contrary, that is, given a complex number w
and ε > 0, there exists δ > 0 such that for all z such that 0 < |z − z0 | < δ
the inequality
|f (z) − w| > ε (5.1.19)
is satisfied. Then, consider the auxiliary function
1
ψ(z) = . (5.1.20)
f (z) − w
By (5.1.19), ψ(z) is analytic and bounded in a punctured η-neighborhood
of z0 , that is (see Theorem 5.1.2), z0 is a removable singularity of ψ(z).
This means that, in an η-neighborhood of z0 , ψ(z) can be written in the
form
ψ(z) = (z − z0 )m κ(z), κ(z0 ) 6= 0,
where κ(z) is analytic in this neighborhood. Then it follows from (5.1.20)
that, in a punctured η-neighborhood of z0 , f (z) has the form
1
f (z) = + w, (5.1.21)
κ(z)(z − z0 )m
where 1/κ(z) is analytic in 0 < |z − z0 | < η and 1/κ(z0 ) 6= 0. But this
means that z0 is either a pole of order m of f (z) (if m > 0) or a point of
analyticity of f (z) (if m = 0). In both cases we have a contradiction with
the assumption of the theorem. This contradiction proves the theorem.
5.1. SINGULAR POINTS OF ANALYTIC FUNCTIONS 197
In the above Example 5.1.7 the function e1/z takes any complex value
w except w = 0 in any punctured δ-neighborhood of z = 0. The number
w = 0 is called an exceptional value for the function e1/z .
Example 5.1.8. The point z = ∞ is an essential singularity of the
function f (z) = cos z and for any complex w the equation cos z = w has
infinitely many solutions,
1 p
zk = Log w + w2 − 1 + 2kπ, k = 0, ±1, ±2, . . . ,
i
in any δ-neighborhood of the point z = ∞, that is, in the region δ < |z| < ∞.
Hence the function cos z does not have any exceptional values.
Note 5.1.2. It follows from Theorems 5.1.1–5.1.4 that, besides the
characterization of the isolated singularities of an analytic function pre-
sented above, there exists another equivalent characterization (see [42]),
namely, the point z0 is said to be
(a) a removable singularity if f (z) has a finite limit as z → z0 ,
(b) a pole of order m if f (z) → ∞ as z → z0 , and
(c) an essential singularity if f (z) has no finite or infinite limit as
z → z0 .
5.1.6. Behavior of an analytic function near z = ∞. We now
consider the behavior of an analytic function in a neighborhood of the
point z = ∞.
Definition 5.1.6. The point z = ∞ is said to be an isolated singular
point of an analytic function f (z) if there exists R > 0 such that there are
no singular points in the region R < |z| < ∞.
For example, z = ∞ is not an isolated singular point for the function
f (z) = 1/ sin z since the singular points, zk = kπ, k = 0, ±1, . . . , of this
function tend to ∞ as k → ±∞.
If z = ∞ is an isolated singular point of f (z) in the region R < |z| < ∞
then it can be expanded in a Laurent series
∞
X
f (z) = cn z n , (5.1.23)
n=−∞
αξ α(α − 1)ξ 2
(1 + ξ)α = 1 + + + ...
1! 2!
α(α − 1) · · · (α − n + 1) n
+ ξ + ..., |ξ| < 1, (5.1.27)
n!
we have
2 −1/2 1 2 − 21 − 23
(1 + ζ ) = 1− ζ + (ζ 2 )2
2 2!
− 21 − 23 · · · − 21 − n + 1 2 n
+ ··· + (ζ ) + . . .
n!
1 3
= 1 − ζ2 + 2 ζ4 + · · · (5.1.28)
2 2 2!
1
+ (−1)n 1 · 3 · 5 · · · (2n − 1)ζ 2n + . . .
n!2n
X∞
(2n − 1)!! 2n
= (−1)n ζ , |ζ| < 1,
n=0
2n n!
Therefore
cRn c
|ck | ≤ = k−n . (5.1.32)
Rk R
If k > n, it follows from (5.1.32) that ck = 0 since R can be taken as
large as we please and the coefficients ck are independent of R. Hence
cn+1 = cn+2 = · · · = 0 in (5.1.30), that is, f (z) is a polynomial of degree
not exceeding n.
The right-hand side of (5.1.36) is the partial fraction expansion of the given
rational function; the term
m
X
a+ c(∞)
n z
n
n=0
X µm
s X (m)
c−n
f (z)(z − zk )µk = (z − zk )µk
m=1 n=1
(z − zm )n
m6=k
(k) (k) (k)
+ c−µk + c−(µk −1) (z − zk ) + c−(µk −2) (z − zk )2 + . . .
(k)
+ c−1 (z − zk )µk −1 . (5.1.40)
(k)
To determine the coefficients c−(µk −p) , p = 0, 1, . . . , µk − 1, we differentiate
(5.1.40) p times with respect to z and take the limit as z → zk . Thus,
(k) 1
c−(µk −p) = lim [f (z)(z − zk )µk ](p) , p = 0, 1, . . . , µk − 1. (5.1.41)
p! z→zk
Setting µk − p = m in (5.1.41), we obtain
5.1. SINGULAR POINTS OF ANALYTIC FUNCTIONS 205
(k) 1
c−m = lim [f (z)(z − zk )µk ](µk −m) ,
(µk − m)! z→zk
m = 1, 2, . . . , µk . (5.1.42)
Formula (5.1.42) gives the partial fraction coefficient of the term with
denominator (z − zk )m , where µk is the multiplicity of the root zk ; in fact,
it gives the coefficients of all the terms in the partial fraction expansion
(5.1.38).
Note 5.1.3. Using formula (5.2.12) of the next section to compute the
residue at the pole of order µk − m, we can rewrite (5.1.42) in the form
(k) f (z)(z − zk )µk f (z)
c−(m+1) = Res = Res ; (5.1.43)
z=zk (z − zk )µk −m z=zk (z − zk )−m
1
21. cos .
z−i
22. tan z.
1
23. + cot2 z.
z
1 1
24. + sin .
(z − 1)10 z−1
25. Let the functions f (z) and g(z) be analytic in a domain D except
at the point z0 . Suppose that z0 is a pole of order n and m of f (z) and
g(z), respectively. Discuss the possible types of singularity of the function
f (z) + g(z) at z0 .
26. Let z0 be a singular point of f (z) and let g(z) be analytic at z0 . Find
the type of singularity of f (z)g(z) if
(a) z0 is a removable singularity.
(b) z0 is a pole of order n.
(c) z0 is an essential singularity.
27. Show that the function
X∞ n
z2
n=1
n2
is continuous in and on the unit circle, but every point of the circle is a
singularity.
where I
1 f (ζ)
cn = dζ, (5.2.2)
2πi C (ζ − z0 )n+1
and C is any closed path which contains the only singular point z0 inside
and is taken in the positive direction. In particular,
I
1
c−1 = f (ζ) dζ. (5.2.3)
2πi C
Definition 5.2.1. The coefficient c−1 of a Laurent series in a neigh-
borhood of an isolated singular point z0 is called the residue of the analytic
function f (z) at z0 and is denoted by Resz=z0 f (z).
208 5. SINGULAR POINTS AND THE RESIDUE THEOREM
By (5.2.2), we have
I
1
Res f (z) = c−1 = f (ζ) dζ. (5.2.4)
z=z0 2πi C
If z0 is a removable singularity of f (z), then the Laurent series (5.2.1) does
not contain negative powers of z − z0 and therefore substituting (5.2.1) into
(5.2.4) gives c−1 = 0.
Therefore, the residue of f (z), in general, is not equal to zero if z0 is a
pole or an essential singularity. However, the residue can be equal to zero
if the coefficient c−1 of the Laurent series is zero. For example, if z = 0
is a pole or an essential singularity and f (z) is an even function (that is,
f (−z) = f (z)), then its Laurent series contains only even (positive and
negative) powers of z and
c−1 = Res f (z) = 0.
z=0
It is essential for the sequel to compute the coefficient c−1 not by using
(5.2.4) but by either differentiating f (z) at z0 or computing c−1 by means
of some special techniques for obtaining the Laurent series expansion of
f (z).
5.2.1. Computing residues. Let z0 be a pole of order m of f (z).
Thus the Laurent series of f (z) in a neighborhood of z0 has the form
c−m c−m+1
f (z) = m
+ + ···
(z − z0 ) (z − z0 )−m+1
X ∞
c−1
+ + cn (z − z0 )n , (5.2.5)
z − z0 n=0
where c−m 6= 0. To determine c−1 , we proceed as follows:
(1) Multiply both sides of (5.2.5) by (z − z0 )m :
(z − z0 )m f (z) = c−m + c−m+1 (z − z0 ) + · · ·
∞
X
+ c−1 (z − z0 )m−1 + cn (z − z0 )n+m . (5.2.6)
n=0
(2) Differentiate (5.2.6) m − 1 times:
Thus,
c−1 = Res ez+1/z = I1 (2).
z=0
Many problems can be found, for example, in [31], pp. 79–80, num-
bers 314–336.
y
+
C∞
D
D
R
0 x
D'
+
Figure 5.1. Positive direction of curve C∞ bounding the
0
infinite domain D ⊂ D.
where C is any closed path lying in D and taken in the positive direction
with respect to the bounded region it encloses.
Note 5.2.2. If we change the direction of C in (5.2.21) we obtain the
+
path C∞ which is traversed in the positive direction with respect to the
infinite domain D0 it encloses (see Fig 5.1).
In that case, (5.2.21) becomes
I
1
Res f (z) = f (z) dz. (5.2.22)
z=∞ 2πi +
C∞
214 5. SINGULAR POINTS AND THE RESIDUE THEOREM
+
Formulae (5.2.22) and (5.2.4) are identical since the paths C∞ in (5.2.22)
and C in (5.2.4) are both taken in the positive direction, and z = ∞ and
+
z = z0 are the only singular points of f (z) inside C∞ and C, respectively.
The coefficient ck of the Laurent series (5.2.17) can be computed by
the simple formula
I
1 f (z)
ck = dz, k = 0, ±1, ±2, . . . , (5.2.23)
2πi C z k+1
where the closed path C lies entirely in the domain of analyticity, D : R <
|z| < ∞, of f (z) and is taken in the positive direction with respect to the
bounded region it encloses.
If the point z = ∞ is a zero of order k of f (z) then the Laurent series
(5.2.17) has the form
c−k
f (z) = k + c−(k−1) z −k+1 + . . . , c−k 6= 0. (5.2.24)
z
It follows from (5.2.24) that
1
f (z) = O , as k → ∞.
zk
If k = 1, then
Res f (z) = −c−1 ,
z=∞
and if k ≥ 2, then
Res f (z) = 0.
z=∞
For example, in the case of a rational function of the form
an z n + an−1 z n−1 + · · · + a0
f (z) = ,
bm z m + bm−1 z m−1 + · · · + b0
we have
an 1
f (z) ≈ , as z → ∞.
bm z m−n
Thus,
−an /bm , if m = n + 1,
Res f (z) =
z=∞ 0, if m > n + 1.
Note that, for the function f (z) = 1/z, we have
1
Res = −1 6= 0,
z=∞ z
despite the fact that z = ∞ is a point of analyticity of 1/z. Hence we have
proved the following important theorem.
5.2. THE RESIDUE THEOREM 215
y
D
zN γ z2
N
γ2
C
z1 γ
1
0 x
Using the last Theorem 5.2.2 and Definition 5.2.3 of the residue at
z = ∞, one can prove the following theorem, which is useful for evaluating
integrals.
Theorem 5.2.3. If an analytic function f (z) has a finite number of
singularities zk , k = 1, 2, . . . , N , in the complex plane, then the sum of all
the residues of f (z), including the residue at z = ∞, is equal to zero:
N
X
Res f (z) + Res f (z) = 0. (5.2.28)
z=zk z=∞
k=1
Solution. The singular points of the integrand in the disk |z| < 2 are
z = 0 and z = 1. At z = 1, the numerator is not zero and the denominator
has a zero of order 1. Hence z = 1 is a pole of order 1 of the integrand. At
z = 0, the numerator and denominator are equal to zero. However,
z2
1 − cos z = 1 − 1 − + . . . = O(z 2 ), as z → 0;
2!
5.2. THE RESIDUE THEOREM 217
thus
1 − cos z
lim = 0.
z→0 z(z − 1)
Hence z = 0 is a removable singularity and
1 − cos z
Res = 0.
z=0 z(z − 1)
Solution. The zeros of the denominator in the disk |z| < 4 are z = 0
and z = ±π. The point z = 0 is a removable singularity since z = 0 is a
zero of order 2 of the numerator and a zero of order 1 of the denominator.
Thus
z2
lim = 0.
z→0 sin z
The points z = ±π are poles of order 1 of the integrand since sin (±π) = 0,
but (sin z)0 z=±π 6= 0. Therefore, by (5.2.11) and (5.2.25),
2
z
I2 = 2πi Res + Res
z=π z=−π sin z
" #
2
z z 2
= 2πi +
cos z z=π cos z z=−π
π2
= 4πi = −4π 3 i.
cos π
Example 5.2.8. Evaluate the following integral counterclockwise:
I 2
e1/z
I3 = dz.
|z|=2 1 − z
Solution. There are two singular points in the region |z| < 2, namely,
a pole, z = 1, of order 1 and an essential singularity, z = 0, since the Laurent
2
series of e1/z contains infinitely many negative powers of z. Therefore
" #
e1/z2
I3 = 2πi Res + Res . (5.2.32)
z=1 z=0 1−z
218 5. SINGULAR POINTS AND THE RESIDUE THEOREM
Solution. The integrand has five singular points in the disk |z| < 5,
namely, the two and three zeros of the first and second factors, respectively,
in the denominator. Therefore it is more convenient to evaluate the integral
by means of Theorem 5.2.3,
z 13
I5 = −2πi Res .
z=∞ (3z 2 + 2)4 (z 3 + 3)2
~
y zk ~z
2
γk z2 Ak
γ2
zN ~z
D k
z1 γ1 C
γl Bk αk
~z ~z βk
l 1 γk
0 x
e + γ1 + · · · + γl .
Figure 5.3. The path C
Note 5.2.4. Formula (5.2.35) is true also in the case the points z̃k
are poles of any odd order (z̃k and the principal part of the Laurent series
EXERCISES FOR SECTION 5.2 221
z
2. .
(z + 1)(z − i)2
z 2 + 4z + 1
3. .
z 2 (z + 1)
z3 + 1
4. .
z(z − 1)2 (z + i)3
ez
5. .
(z − 1)(z + 3i)2
cos z
6. .
(z − 1)2 (z + 4)
Find the residue of the given functions at every finite singular point.
1
7. z .
e −1
sin z
8. .
z(z − 1)2
1 − cos z
9. .
z 2 sin z
10. z 3 e1/z .
1
11. z 2 sin .
z−1
z
12. cos .
z+2
sin z 1
13. + 3 + e1/z .
z z
1 − cos z 1 1
14. + 25 + sin .
z2 z z
15. ez/(z−1) .
1
16. cos cos z.
z
ϕ(z)
17. Let f (z) = , ϕ(z0 ) 6= 0, ψ(z0 ) = 0, ψ 0 (z0 ) 6= 0. Suppose that
[ψ(z)]2
ϕ(z) and ψ(z) are analytic at z = z0 . Find the type of singularity of f (z)
at z = z0 and Res f (z).
z=z0
The main idea in evaluating definite integrals over the real x-axis by
means of Cauchy’s Theorem and the theory of residues, in the simplest
cases, is as follows. Instead of evaluating the integral of a function f (x)
of the real variable x from −∞ to +∞, one considers the integral of f (z)
of the complex variable z along a closed path, C, consisting of a segment,
[−R, R], of the real axis and a semicircle, CR : |z| = R, 0 ≤ arg z ≤ π, in the
upper half-plane. The residue theorem is applied to f (z) over the region
bounded by C and the limit is taken as R → ∞. If |f (z)| = O 1/|z|2 , the
integral along CR tends to zero as R → ∞. Thus,
Z ∞ X
f (x) dx = 2πi Res f (z).
−∞ =z>0
and branch cuts are taken along the positive real semi-axis.
CR
–R 0 R x
where =zk > 0 since Qm (z) has no real zeros. We show that
Z
lim f (z) dz = 0. (6.1.5)
R→∞ CR
−(m−n−1)
z + a1 z −(m−n) + · · · + an z −(m−1)
=
1 + b1 z −1 + · · · + bm z −m
z=R eiθ
→ 0, as R → ∞,
because all the powers of z in the numerator and the denominator of the last
fraction are negative. Hence, as R → ∞, we get from (6.1.4) the formula
Z ∞ X
Pn (x) Pn (z)
dx = 2πi Res , (6.1.7)
−∞ Qm (x) k
z=zk Qm (z)
12
= πi lim
z→i (z + i)5
12 3π
= πi 5
= .
(2i) 8
CR
γ γ γ γ
1 2 k l
A B
–R α1 α2 0 αk αl R x
where =zk > 0 and the first integral on the left-hand side of (6.1.8) is
evaluated along the straight line subsegments from −R to R, omitting the
semicircles shown in Fig 6.2. Since m ≥ n + 2, then, as before, the integral
along the semicircle CR approaches zero as R → ∞. Moreover, since the
points αk (k = 1, . . . , l) are poles of order 1, then a Laurent series expansion
in a neighborhood of the point z = αk has the form
X ∞
Pn (z) c−1
= + cµ (z − αk )µ . (6.1.9)
Qm (z) z − αk µ=0
Hence
Z Z ∞
X
Pn (z) c−1 µ
dz = + cµ (z − αk ) dz. (6.1.10)
γk Qm (z) γk z − αk µ=0
Since z ∈ γk , then z −αk = δ eiθ and dz = δi eiθ dθ, where θ varies clockwise
from π to 0. Hence from (6.1.10) we obtain
Z Z 0 ∞ Z 0
Pn (z) δi eiθ dθ X µ
dz = c−1 iθ
+ c µ δ eiθ δi eiθ dθ
γk Qm (z) π δe µ=0 π
→ −c−1 πi + 0, as δ → 0,
where
Pn (z)
c−1 = Res . (6.1.11)
z=αk Qm (z)
Therefore, the left-hand side of (6.1.8) has a finite limit as δ → 0 and
R → ∞:
Z ∞ X
Pn (x) Pn (z)
p. v. dx = 2πi Res
−∞ Qm (x) z=zk Qm (z)
k
6.1. RATIONAL FUNCTIONS OVER (−∞, +∞) 231
l
X
Pn (z)
+ πi Res , (6.1.12)
z=αk Qm (z)
k=1
where =zk > 0. At the same time we have proved the convergence of
integral (6.1.12) in the sense of the principal value.
Note 6.1.2. Formula (6.1.12) is valid also in the case where all the
points αk are real poles of odd order 2s + 1 for some s = 1, 2, 3, . . ., and
the coefficients, c−2l , of even order in the Laurent series expansion are all
equal to zero.
and we have
Z Z 0
dz δ eiθ
= i dθ
γk (z − αk )2s+1 π (δ e )
iθ 2s+1
Z 0
= δ −2s i e−2siθ dθ
π
(
−πi, if s = 0,
=
0, if s = 1, 2, 3, . . . .
0 1
x
Example 6.1.6. Find the values of the real parameters a and c, with
c > a4 > 0, for which the Cauchy principal value
Z ∞
dx
I = p. v. 3 (x2 + 2a2 x + c)
−∞ (x − a)
is finite and evaluate I.
Solution. We expand the integrand in partial fractions,
1
f (x) =
(x − a)3 (x2 + 2a2 x + c)
(6.1.14)
A B C Dx + E
= 3
+ 2
+ + 2 .
(x − a) (x − a) x − a x + 2a2 x + c
To have a finite principal value it is necessary that B = 0; thus the integrand
is of the form (6.1.13). To use formula (5.1.42) to compute B, we multiply
both sides of (6.1.14) by (x − a)3 , differentiate the resulting equation once
with respect to x and consider the limit as x → a:
0
1
B = lim [f (x)(x − a)3 ]0 = lim
x→a x→a x2 + 2a2 x + c
2x + 2a2 2a + 2a2
= − lim 2 2 2
=− 2
x→a (x + 2a x + c) (a + 2a3 + c)2
= 0,
if a = −1. Hence I is finite if a = −1 and c > 1.
The singular points of the integrand are
√
z1 = −1 (pole of order 3), z2,3 = −1 ± i c − 1 (poles of order 1).
Using formula (6.1.12) we have
Z ∞
dx
I = p. v. 3 (x2 + 2x + c)
−∞ (x + 1)
6.2. RATIONAL FUNCTIONS TIMES SINE OR COSINE 233
1
= 2πi Res√ +πi Res 3 (z 2 + 2z + c)
z=−1+i c−1 z=−1 (z + 1)
00
1 πi 1
= 2πi + lim
(z2 + 1)3 2(z2 + 1) 2 z→−1 z 2 + 2z + c
0
πi πi 2z + 2
= − lim
(c − 1)2 2 z→−1 (z 2 + 2z + c)2
πi (z 2 + 2z + c)2 − 2(z 2 + 2z + c)(2z + 2)(z + 1)
= 2
− πi lim
(c − 1) z→−1 (z 2 + 2z + c)4
πi πi
= 2
− = 0.
(c − 1) (c − 1)2
The following lemma, due to Camille Jordan, will be used in the sequel.
Lemma 6.2.1 (Jordan’s Lemma). If a function f (z) is continuous on
a sequence of circular arcs
CRn : |z| = Rn , =z ≥ −a,
where Rn → ∞, a is fixed and
Mn = max |f (z)| → 0, as Rn → ∞, (6.2.2)
z∈CRn
Proof. Suppose that a > 0. Then, on the arc AB, −αn ≤ arg z < 0,
where αn > 0 (see Fig 6.4). Clearly, αn = arcsin (a/Rn ) → 0 as Rn → ∞.
Moreover, arcsin (a/Rn ) ≈ a/Rn for Rn large, so that αn Rn ≈ a = constant
as Rn → ∞. Since, on arc AB, −αn ≤ θ ≤ 0, then
− sin αn ≤ sin θ ≤ 0, that is, 0 ≤ − sin θ ≤ sin αn ,
and
iλz iλR (cos θ+i sin θ)
e = e n = e−λRn sin θ
≤ eλRn sin αn
≈ eλa = constant
234 6. ELEMENTARY DEFINITE INTEGRALS
y
E
Rn
R2
C 0 B
R1 − αn x
D –a A
1
y=sin x
y=2x/π
0 π/2 x
Z ∞
Pn (x)
p. v. sin λx dx
−∞ Q m (x)
X l
X
Pn (z) iλz Pn (z) iλz
= = 2πi Res e + πi Res e , (6.2.13)
z=zk Qm (z) z=αk Qm (z)
k k=1
Z ∞
Pn (x)
p. v. cos λx dx
−∞ Qm (x)
X l
X
Pn (z) iλz Pn (z) iλz
= < 2πi Res e + πi Res e . (6.2.14)
z=zk Qm (z) z=αk Qm (z)
k k=1
One can get the correct answer if the point z = 0 in Fig 6.1 is surrounded
by a semicircle Cδ = {z = δ eiθ } of radius δ < R. Then, in addition to the
integral on CR which is already evaluated, one has to evaluate the integrals
Z −δ Z Z R Z −δ Z R Z
dz dx dz
+ + = + +
−R Cδ δ z −R δ x Cδ z
Z 0
−δ R δi eiθ
= log |x|−R + log |x|δ + iθ
dθ
π δe
Z 0
= log 1 + i dθ = −πi.
π
Thus, we get I3 = πi − πi = 0, which is the correct answer.
6.3. RATIONAL FUNCTIONS TIMES EXPONENTIAL FUNCTIONS 239
X
Pn (ez ) az
= 2πi Res e . (6.3.4)
z=zk Qm (ez )
k
We evaluate each of these four integrals along the corresponding side of the
rectangle as R → ∞.
On side I, z = x, and therefore
Z Z R Z ∞
Pn (ex ) ax Pn (ex ) ax
= x
e dx → x
e dx, as R → ∞.
I −R Qm (e ) −∞ Qm (e )
and
a(R+iy) (α+iβ)(R+iy)
e = e = eαR−βy . (6.3.6)
Because α < m − n, by (6.3.5) and (6.3.6), we have the estimate
Z Z 2π
R+iy
Pn e a(R+iy)
≤ e |i| dy
Qm (eR+iy )
II 0
Z 2π
−(m−n−α)R
≈e e−βy dy → 0, as R → ∞.
0
2π III
IV II
–R 0 I R x
where
Qm (ex ) 6= 0, −k < <a < m − n, 0 < =zk < 2π.
Note 6.3.1. Formula (6.3.7) can be obtained by using the closed path
consisting of the segment [−R, R] of the real axis and the semicircle CR of
radius R shown in Fig 6.1. In fact, if zk is a singular point of the function
eaz Pn (ez ) /Qm (ez ) in the strip 0 < =z < 2π, then the points zk + 2pπi,
for p = 0, 1, . . . , are also singular points of this function.
To establish the statement of this note, for simplicity, we assume that all
the singular points, zk , of the integrand of (6.3.1) in the strip 0 < =z < 2π
are simple poles and we let Ck,p denote the circle of radius p centered at
zk + 2pπi. In this case, the integral (6.3.1) is equal to the series
X ∞ X
Pn (ez ) az
S = 2πi Res e
z=zk +2pπi Qm (ez )
p=0 k
X ∞ X I
1 Pn (eζ ) aζ
= 2πi e dζ
p=0
2πi Ck,p Qm (eζ )
k
242 6. ELEMENTARY DEFINITE INTEGRALS
iθ
∞ XZ
X Pn ezk +p e
2π
iθ
= ea(zk +2pπi+p e ) ip dθ
Q e zk +p eiθ
p=0 k 0 m
X Z 2π Pn ezk +p e
iθ
X∞
iθ
= e2paπi zk +p eiθ
ea(zk +p e ) ip dθ
p=0 k 0 Q m e
X I
2πi Pn (eζ ) aζ
= e dζ
1 − e2πai Ck,0 Qm (e )
ζ
k
2πi X Pn (ez ) az
= Res e , if =a > 0, (6.3.8)
1 − e2πai z=zk Qm (ez )
k
which is, in fact, formula (6.3.7).
If =a = 0, then the right-hand side of (6.3.8) can be obtained by
assuming that
X∞ X∞
1
e2pπai = lim e2p(α+εi)πi = .
p=0
ε→0
p=0
1 − e2παi
Finally, if =α < 0, to determine (6.3.7) one has to close the segment [−R, R]
of the real axis by a semicircle, CR , in the lower half-plane and take into
account the fact that the points zk −2pπi, for p = 0, 1, . . . , are simple poles.
6.3.2. The case of real poles. We assume that the integrand in
(6.3.1) has real simple poles at α1 , α2 , . . . , αl .
Since the function
Pn (ex )
f (x) =
Qm (ex )
is periodic of period 2πi, then the denominator vanishes at the points
α1 , α2 , . . . , αl on side I and also at the points α1 +2πi, α2 +2πi, . . . , αl +2πi
on side III of the path, C, shown in Fig 6.7(a). We bypass these points
along semicircles, γ1 , γ2 , . . . , γl , and e
γ1 , γ
e2 , . . . , e
γl , of radius δ on sides I and
III, respectively.
By the residue theorem, the value of the integral along C is
Z Z Z Z
Pn (ez ) az
+ + + e dz
I II III IV Qm (ez )
X
Pn (ez ) az
= 2πi Res e , (6.3.9)
z=zk Qm (ez )
k
where 0 < =zk < 2π.
As in Subsection 6.3.1, the integrals along sides II and IV approach
zero as R → ∞ if −k < <a < m − n, provided (6.3.2) holds.
On side I we have
6.3. RATIONAL FUNCTIONS TIMES EXPONENTIAL FUNCTIONS 243
y y
α1+2πi 2π III
2π
~
A
~
B
~ ~
γ
γ k
k
IV II II
γ γ
k k
A B
–R α1 α2 0 αk αl R x 0 αk
I
(a) (b)
Z Z R
Pn (ez ) az Pn (ex ) ax
e dz = e dx
I Qm (ez ) −R Qm (ex )
l Z
X Pn (ez ) az
+ e dz, (6.3.10)
γk Qm (ez )
k=1
where the first integral on the right-hand side is evaluated along the straight
line segments shown in Fig 6.7(a). On the arc γk , enlarged in Fig 6.7(b),
we have
z −αk = δ eiθ , θ|A = arg (z − αk )|A = π, θ|B = arg (z − αk )|B = 0,
because the arc γk is taken clockwise from A to B.
Consider the Laurent series expansion of the integrand in a neighbor-
hood of the simple pole z = αk ,
X∞
Pn (ez ) az c−1
e = + cm (z − αk )m .
Qm (ez ) z − αk m=0
Then
Z Z X∞ Z
Pn (ez ) az c−1
e dz = dz + cm (z − αk )m dz
γk Qm (ez ) γk z − αk m=0 γ k
Z 0 iθ X∞ Z 0
δ e i dθ m
= c−1 iθ
+ c m δeiθ δ eiθ i dθ
π δe m=0 π
→ −c−1 πi + 0, as δ → 0,
where
Pn (ez ) az
e
c−1 = Res .
z=αk Qm (ez )
Z Z ∞
Pn (ez ) az Pn (ex ) ax
e dz = e dx
I Qm (ez ) −∞ Qm (ex )
l
X
Pn (ez ) az
− πi Res e . (6.3.11)
z=αk Qm (ez )
k=1
The first term on the right-hand side is evaluated along the part of the
segment [−R, R], excluding the arcs γ
ek .
On the arcs e
γk ,
z − (αk + 2πi) = δ eiθ .
Thus
θ|Be = arg (z − αk − 2πi)|Be = 0, θ|Ae = arg (z − αk − 2πi)|Ae = −π,
because the arc γ e to A.
ek is taken clockwise from B e
Consider the Laurent series expansion of the integrand in a neighbor-
hood of the point αk + 2πi:
X ∞
Pn (ez ) az c̃−1
z
e = + c̃m (z − αk − 2πi)m .
Qm (e ) z − αk − 2πi
k=0
Then
Z Z −π X∞ Z 0
Pn (ez ) az δ eiθ m
e dz = c̃−1 i dθ + c̃m δ eiθ δ eiθ i dθ
γk
e Qm (ez ) 0 δ eiθ m=0 −π
→ −πic̃−1 , as δ → 0,
where
Pn (ez ) az 2πai Pn (ez ) az
c̃−1 = Res e = e Res e .
z=αk +2πi Qm (ez ) z=αk Qm (ez )
π
= .
sin πα
Example 6.3.2. Evaluate the integral
Z ∞
eax
I5 = x 2x
dx, 0 < <a < 2. (6.3.16)
−∞ 1 + e + e
Solution. The integral I5 is convergent for 0 < <a < 2 because the
integrand f (x) is like e−(2−a)x → 0 as x → ∞, since <(2 − a) > 0, and like
eax → 0 as x → −∞, since <a > 0. The poles of the function
1
f (z) =
1 + ez + e2z
are the zeros of the equation
e2z + ez + 1 = 0,
that is, (
√
z 1 3 eiθ1 ,
e =− ±i = (6.3.17)
2 2 eiθ2 ,
where
√ 2π √ 4π
tan θ1 = − 3 =⇒ θ1 = , tan θ2 = 3 =⇒ θ2 = .
3 3
Thus, there are two simple poles,
z1 = 2πi/3 and z2 = 4πi/3,
in the strip 0 < =z < 2π. Therefore, using (6.3.7), we have
2πi eaz
I5 = Res + Res
1 − e2πai z=2πi/3 z=4πi/3 1 + ez + e2z
2πi e2πai/3 e4πai/3
= +
1 − e2πai 2e4πi/3 + e2πi/3 2e2πi/3 + e4πi/3
π e−πai/3 eπai/3
=− √ √ + √ √ ,
sin πa −1 − i 3 − 1/2 + i 3/2 −1 + i 3 − 1/2 − i 3/2
where the last term is obtained by multiplying the numerator and denom-
inator of the previous term by e−πai .
Finally, noting that the second fraction inside the square brackets is
the complex conjugate of the first one, we obtain
π e−πai/3
I5 = 2< √
sin πa 3/2 + i 3/2
4π h πa πa √ i
= < cos − i sin 3−i 3
12 sin πah 3 3
π πa √ πa i
= 3 cos − 3 sin
3 sin πa 3 3
6.3. RATIONAL FUNCTIONS TIMES EXPONENTIAL FUNCTIONS 247
"√ #
2π 3 πa 1 πa
= √ cos − sin
3 sin πa 2 3 2 3
2π h π πa π πa i
= √ sin cos − cos sin
3 sin πa 3 3 3 3
2π π(1 − a)
= √ sin .
3 sin πa 3
0 ≤ =z < 2π, the simple poles of the integrand are z0 = 0 and z1 = πi.
Moreover, z0 is a real zero. Therefore, by formula (6.3.14), we have
az az
2πi e 1 + e2πai e
I= Res + πi Res
1 − e2πai z=πi e2z − 1 1 − e2πai z=0 e2z − 1
2πi eπai 1 + e2πai
= 2πai 2πi
+ πi
1−e 2e 2 (1 − e2πai )
πi e−πai + eπai
= − πai −πai
+ πi
e −e 2 (e−πai − eπai )
π cos πa
=− −π
2 sin πa 2 sin πa
π 2 cos2 (πa/2)
=−
2 2 sin(πa/2) cos(πa/2)
π πa
= − cot .
2 2
where z α−1 = e(α−1) log z , is single-valued and analytic in the region bounded
by C, except at the poles. By the residue theorem,
I X
Pn (z) α−1 Pn (z) α−1
z dz = 2πi Res z ,
C Qm (z) z=zk Qm (z)
k
that is,
Z Z Z Z
Pn (z) α−1
+ + + z dz
AB CR eA
B e Cδ Qm (z)
X
Pn (z) α−1
= 2πi Res z . (6.4.4)
z=zk Qm (z)
k
On the segment AB, z = x, and we have
Z Z R Z ∞
Pn (x) α−1 Pn (x) α−1
= x dx → x dx (6.4.5)
AB δ Q m (x) 0 Q m (x)
as R → ∞, δ → 0.
On the circle CR ,
z = Reiθ , 0 ≤ θ ≤ 2π,
and we have
Z Z 2π
Pn R eiθ α−1
= R eiθ R eiθ i dθ → 0 (6.4.6)
CR 0 Qm (R eiθ )
as R → ∞, because, by (6.4.2), m − n − α > 0.
On the segment Be A,e
α−1
z = x e2πi , z α−1 = x e2πi = e2παi xα−1 ,
and we have
Z Z δ Z 0
Pn (x) α−1 Pn (x) α−1
= e2παi x dx → e2παi x dx, (6.4.7)
eA
B e R Qm (x) ∞ Qm (x)
y
CR
Cδ A B
~ ~ x
0 A B
as δ → 0 and R → ∞.
On the circle Cδ , we have z = δ eiθ ; thus
Z Z 0
Pn δ eiθ α−1 iθ
= iθ )
δ eiθ δ e i dθ → 0 (6.4.8)
Cδ 2π Q m (δ e
as δ → 0 because Pn (0) 6= 0, Qm (0) 6= 0 and α > 0.
Hence, considering the limit as δ → 0 and R → ∞ in (6.4.4) and using
(6.4.5)–(6.4.8), we obtain
Z X
∞ Pn (x) α−1 Pn (z) α−1
1 − e2παi x dx = 2πi Res z , (6.4.9)
0 Qm (x) z=zk Qm (z)
k
where the residues are evaluated at all the poles zk in the complex plane
(we recall that the cut contains no singular points since Qm (x) 6= 0 there).
It follows from (6.4.9) that
Z ∞ X
Pn (x) α−1 2πi Pn (z) α−1
x dx = Res z , (6.4.10)
0 Qm (x) 1 − e2παi z=zk Qm (z)
k
provided
Qm (0)Pn (0) 6= 0, 0 < α < m − n, Qm (x) 6= 0 for x > 0.
Example 6.4.1. Evaluate the integral
Z ∞ α−1
x
I7 = dx, 0 < α < 1. (6.4.11)
0 x+1
Solution. The conditions Pn (x) = 1 and Qm (x) = x+ 1 are such that
(6.4.10) is true, and hence the value of I7 is found by evaluating the residue
of z α−1 /(z + 1) at the only pole z = −1 = eπi (we take 0 ≤ arg z ≤ 2π).
Thus
α−1
2πi z
I7 = Res
1 − e2παi z=−1 z + 1
α−1
2πi eπi eπαi
= 2παi
= 2πi 2παi
1−e 1 e −1
1 1
= 2πi παi = 2πi
e − e−παi 2i sin πα
π
= .
sin πα
6.4.2. The case of real poles. We suppose that the l strictly positive
real numbers, α1 < α2 < · · · < αl , are simple zeros of Qm (x).
We replace the path shown in Fig 6.8 by a closed path where the singu-
lar points, α1 , α2 , . . . , αl , are bypassed along the semicircles, γ1 , γ2 , . . . , γl ,
6.4. RATIONAL FUNCTIONS TIMES A POWER OF x 251
y y
γk
CR
γ1 γl Ak Bk
Cδ A B αk
~ x ~ x
0 ~ ~ ~ B ~ ak ~
A γ γl Ak Bk
1
~
γ
k
(a) (b)
of radius δ on the upper part of the cut and similarly along the semicircles,
e1 , e
γ γ2 , . . . , e
γl , of radius δ on the lower part of the cut (see Fig 6.9(a)).
By the residue theorem,
Z Z Z Z
Pn (z) α−1
+ + + z dz
AB CR e
BA e Cδ Q m (z)
X
Pn (z) α−1
= 2πi Res z . (6.4.12)
z=zk Qm (z)
k
The first integral on the right-hand side of (6.4.13) is evaluated along the
straight line segments on [δ, R], excluding the arcs γk . On the arc γk taken
clockwise, we have z − αk = δ eiθ where
θ|Ak = arg (z − αk )|Ak = π, θ|Bk = arg (z − αk )|Bk = 0.
Expanding the integrand in a Laurent series in a neighborhood of the simple
pole z = αk , we have
X ∞
Pn (z) α−1 c−1
z = + cµ (z − αk )µ .
Qm (z) z − αk µ=0
252 6. ELEMENTARY DEFINITE INTEGRALS
Thus,
Z Z ∞
X Z
Pn (z) α−1 c−1
z dz = dz + cµ (z − αk )µ dz
γk Qm (z) γk z − αk µ=0 γ k
Z 0 iθ X∞ Z 0
δe µ
= c−1 iθ
i dθ + cµ δ eiθ δ eiθ i dθ
π δ e µ=0 π
→ −c−1 πi + 0, as δ → 0,
where
Pn (z) α−1
c−1 = Res z .
z=αk Qm (z)
It follows from (6.4.13) that
Z Z ∞ l
X
Pn (x) α−1 Pn (z) α−1
= x dx − πi Res z , (6.4.14)
AB 0 Qm (x) z=αk Qm (z)
k=1
as δ → 0 and R → ∞.
eA
Similarly, on B e we have
Z Z δ l Z
X
Pn (x) α−1 Pn (z) α−1
= x e2πi dx + z dz. (6.4.15)
eA
B e R Qm (x) γek Qm (z)
k=1
The first integral on the right-hand side of (6.4.15) is evaluated along the
straight line segments excluding the curves γ ek . On the arc γ
ek taken clock-
wise, with center ãk = αk e2πi , we have z − ãk = δ eiθ where
θ|Bek = arg (z − ãk )|Bek = 2π, θ|Aek = arg (z − ãk )|Aek = π.
Expanding the integrand in a Laurent series in a neighborhood of the simple
pole z = ãk , we have
X ∞
Pn (z) α−1 c−1
z = + cµ (z − ãk )µ .
Qm (z) z − ãk µ=0
Thus,
Z Z X∞ Z
Pn (z) α−1 c−1
z dz = dz + cµ (z − ãk )µ dz
γ
ek Qm (z) ek z − ãk
γ µ=0 γek
Z π iθ X∞ Z π
δ e i dθ µ
= c−1 iθ
+ cµ δ eiθ δ eiθ i dθ
2π δ e µ=0 2π
→ −c−1 πi + 0, as δ → 0,
6.4. RATIONAL FUNCTIONS TIMES A POWER OF x 253
as R → ∞ and δ → 0.
Hence, considering the limit in (6.4.12) as R → ∞ and δ → 0, and
using (6.4.13)–(6.4.17), we obtain
Z ∞
2παi
Pn (x) α−1
1−e p. v. x dx
0 Q m (x)
l
X Pn (z) α−1
− πi 1 + e2παi Res z
z=αk Qm (z)
k=1
X
Pn (z) α−1
= 2πi Res z . (6.4.18)
z=zk Qm (z)
k
We thus obtain, from (6.4.18), the following formula for the evaluation of
the integral:
Z ∞ X
Pn (x) α−1 2πi Pn (z) α−1
p. v. x dx = Res z
0 Qm (x) 1 − e2παi z=zk Qm (z)
k
l
1 + e2παi X Pn (z) α−1
+ πi Res z , (6.4.19)
1 − e2παi z=αk Qm (z)
k=1
provided
zk 6∈ (0, +∞), αk > 0, 0 < α < m − n.
Note 6.4.1. The authors have not seen formula (6.4.19) in the litera-
ture for the evaluation of the previous integral.
In Problem 28.21 of [21] one finds the formula
254 6. ELEMENTARY DEFINITE INTEGRALS
Z ∞
π X
p. v. R(x) xa−1 dx = Res R(z)(−z)a−1
−∞ sin πa z=zk
k
l
X
+ π cot πa Res [R(z) z a−1 ], (6.4.20)
z=ak
k=1
It can easily be checked that the conditions are such that formula (6.4.19)
is true, and hence, by evaluating the residues of the function z iα /(z 2 − 1)
at z = 1 and z = −1, we have
" iα #
2πi eπi 1 + e2πi(iα+1) 1
I9 = = + πi
1 − e2πi(iα+1) 2(−1) 1 − e2πi(iα+1) 2
πie−πα 1 1 + e−2πα
== − + πi
1 − e−2πα 2 1 − e−2πα
−πα −2πα
−2e +1+e
=π
2 (1 − e−2πα )
2
(1 − e−πα )
=π
2 (1 − e−πα ) (1 + e−πα )
π 1 − e−πα π eπα/2 − e−πα/2
= =
2 1 + e−πα 2 eπα/2 + e−πα/2
π πα
= tanh .
2 2
Exercises for Chapter 6
Evaluate the following integrals.
Z ∞
x2
1. 4
dx.
−∞ x + 1
Z ∞
x2
2. dx.
0 x + x2 + 1
4
Z ∞
1 − cos x
3. dx.
0 x2
Z ∞
sin2 x
4. dx.
0 x2
Z ∞
x2
5. 2 2 2
dx.
−∞ (x + 1) (x + 2x + 2)
Verify the following formulae.
Z ∞
cos ax π
6. 2 +1
dx = e−a , a > 0.
0 x 2
Z ∞
dx π
7. 2 + 1)2
= .
0 (x 4
Z ∞
x − sin x π
8. 2
dx = .
−∞ x 2
256 6. ELEMENTARY DEFINITE INTEGRALS
Z ∞
cos ax π(a + 1)e−a
9. dx = , a > 0.
0 (x2 + 1)2 4
Z ∞
xα−1 π
10. dx = , 0 < α < 2.
0 1+x 2 2 sin πα
2
Z ∞
dx π
11. = 3, a > 0.
0 (x2 + a2 )2 4a
Z ∞
eax π
12. dx = , 0 < a < 1.
−∞ 1 + ex sin aπ
Z ∞
xα−1 π
13. dx = πα
, n = 1, 2, 3, . . . , 0 < α < 2n.
0 1 + x2n 2n sin 2n
CHAPTER 7
where the rational function Pn (x)/Qm (x) is not even and m ≥ n+ 2. These
integrals can be evaluated by taking the limit in (6.4.10) or (6.4.18) as
α → 1. But this procedure leads to an indefinite form 0/0 and, in general,
the limit cannot be easily found. On the other hand, these integrals can be
evaluated directly by the theory of residues.
we have
Z Z δ
Pn (x)
= (ln x + 2πi) dx
eA
B e R Qm (x)
Z 0
(7.1.5)
Pn (x)
→ (ln x + 2πi) dx,
∞ Qm (x)
as R → ∞ and δ → 0. Hence, by taking the limit in (7.1.3) as R → ∞ and
δ → 0, and using (7.1.4) and (7.1.5), we obtain
Z ∞ Z ∞
Pn (x) Pn (x)
ln x dx − (ln x + 2πi) dx
0 Q m (x) 0 Q m (x)
X
Pn (z)
= 2πi Res Log z .
z=zk Qm (z)
k
provided
bx bx
m≥n+2 and Qm 6= 0 for > 0.
x+1 x+1
Formula (7.1.7) can be considered as an indefinite integral of the function
Pn (x)/Qm (x). Therefore indefinite integrals of rational functions can be
evaluated by the theory of residues by means of (7.1.9).
7.1. RATIONAL FUNCTIONS OVER (0, +∞) 259
7.1.2. The case of positive real poles. Suppose that the real zeros
of Qm are positive and simple and are ordered as follows:
0 < α1 < α2 < · · · < αl .
We consider the path shown in Fig 6.9 and the auxiliary function (7.1.2).
The only difference between the present case and the one considered in the
previous subsection is in the evaluation of the integrals along the semicircles
γk and γek attached to the upper and lower parts, AB and B e A,
e respectively,
of the cut [0, +∞]. Hence, the sums
X l Z Xl Z
Pn (z) Pn (z)
Log z dz, Log z dz, (7.1.13)
γk Q m (z) γk
e Q m (z)
k=1 k=1
e A,
are to be added to the integrals (7.1.4) and (7.1.5) along AB and B e
respectively. The limit of the integral along γk , as δ → 0, is
Z
Pn (z)
−c−1 πi = Log z dz
γk Qm (z)
where
Pn (z)
c−1 = Res Log z . (7.1.14)
z=αk Qm (z)
ek , as δ → 0, is
The limit of the integral along γ
Z
Pn (z)
−c−1 πi = Log z dz,
γk
e Q m (z)
The formula for the evaluation of integral (7.1.1) over the positive real axis
follows from this last relation,
Z ∞ X
Pn (x) Pn (z)
p. v. dx = − Res Log z
0 Qm (x) z=zk Qm (z)
k
Xl Xl
Pn (z) Pn (z)
− Res Log z − πi Res , (7.1.16)
z=αk Qm (z) z=αk Qm (z)
k=1 k=1
where
αk > 0, =zk 6= 0 if <zk > 0, m ≥ n + 2, 0 ≤ arg z < 2π.
To the authors’ knowledge, formula (7.1.16) is not found explicitly in the
literature; however it can be obtained by the recurrence relation (7.2.2)
given in Problems 29.03 and 29.05 of [21] and derived in Subsection 7.2.1.
If all αk = 0, then (7.1.16) reduces to (7.1.6), which is given in [21],
Problem 29.01.
Example 7.1.2. Evaluate the integral
Z ∞
dx
I = p. v. .
0 (x − 1)(x2 + 1)
Solution. Using formula (7.1.16), we have
Log z 1
I = − Res + Res + Res − πi Res
z=i z=−i z=1 (z − 1)(z 2 + 1) z=1 (z − 1)(z 2 + 1)
Log i Log(−i) 1
=− + − πi 2
2i(i − 1) (−i − 1)(−2i) 1 +1
" #
iπ/2 i3π/2
Log e Log e πi
= + −
2i + 2 −2i + 2 2
262 7. INTERMEDIATE DEFINITE INTEGRALS
1 1−i π 1 + i 3π πi
= i + i −
2 2 2 2 2 2
π
=− .
4
Note 7.1.2. Since 0 ≤ Arg z < 2π in (7.1.16), then in the previous
example, Arg(−i) = 3π/2 and not −π/2. Therefore
Log i Log(−i) Log i
+ 6= 2< ,
2i(i − 1) (−i − 1)(−2i) 2i(i − 1)
although, at first glance, the second term on the left-hand side appears to
be the complex conjugate of the first one.
X
Pn (z)
= 2πi Res (Log z)p . (7.2.6)
z=zk Qm (z)
k
(a) The case p = 2. In this case we obtain from (7.2.2) and (7.2.3) that
X
0 1 0 Pn (z) 2
C2 2πiI0 + C2 (2πi) I1 = − Res (Log z) , (7.2.9)
z=zk Qm (z)
k
that is,
264 7. INTERMEDIATE DEFINITE INTEGRALS
Z ∞ Z ∞
Pn (x) Pn (x)
2πi dx + 2 ln x dx
0 Qm (x) 0 Qm (x)
X
Pn (z) 2
=− Res (Log z) . (7.2.10)
z=zk Qm (z)
k
Equating the real parts on the left- and right-hand sides in (7.2.10), we
obtain the following formula for I1 :
Z ∞
Pn (x)
I1 = ln x dx
0 Q m (x)
X (7.2.11)
1 Pn (z) 2
=− < Res (Log z) ,
2 z=zk Qm (z)
k
If we equate the imaginary parts on the left- and right-hand sides of (7.2.10),
we obtain another formula for I0 (compare with (7.1.6)):
Z ∞
Pn (x)
I0 = dx
0 Q m (x)
X (7.2.13)
1 Pn (z)
=− = Res (Log z)2 .
2π z=zk Qm (z)
k
that is,
X
2 Pn (z) 3
−4π I0 + 6πiI1 + 3I2 = − Res (Log z) , (7.2.14)
z=zk Qm (z)
k
where I0 , I1 and I2 are real integrals. Equating the real parts on the left-
and right-hand sides of (7.2.14) and using (7.1.6) we obtain a simple formula
for I2 :
Z ∞
Pn (x)
I2 = (ln x)2 dx
0 Qm (x)
X (7.2.15)
1 Pn (z) 3 2
=− < Res ((Log z) + 4π Log z) ,
3 z=zk Qm (z)
k
7.2. FORMS CONTAINING (ln x)p IN THE NUMERATOR 265
and
s Z
X Pn (z)
(Log z)p dz (7.2.17)
γk
e Qm (z)
k=1
along the semicircles γk and e γk on the upper and lower parts, AB and
Be A,
e respectively, of the cut, are added to the integral along BA in formula
(7.2.7) and along BeA e in formula (7.2.8).
As in Subsection 7.1.2, the limit of the integral along γk , as δ → 0, is
Z
Pn (z)
(Log z)p dz = −c−1 πi,
γk Q m (z)
where
Pn (z) p
c−1 = Res (Log z) . (7.2.18)
z=αk Qm (z)
The limit of the integral along γ ek , as δ → 0, is
Z
Pn (z)
(Log z)p dz = −c−1 πi,
γek Q m (z)
where
Pn (z)
c−1 = Res (Log z)p
z=αk e2πi Qm (z)
(and letting z = ζ e2πi )
Pn (ζ) h i
2πi p
(7.2.19)
= Res Log ζ e
ζ=αk Qm (ζ)
Pn (z) p
= Res (Log z + 2πi) .
z=αk Qm (z)
266 7. INTERMEDIATE DEFINITE INTEGRALS
e A.
is added to the integral (7.2.8) along B e Thus, as R → ∞ and δ → 0, we
obtain from formula (7.2.6) that
p−1
X X
Pn (z)
Cpr (2πi)p−r−1 Ir = − Res (Log z)p
r=0
z=zk Qm (z)
k
s
1X Pn (z) p p
− Res (Log z) + (Log z + 2πi) , (7.2.20)
2 z=αk Qm (z)
k=1
These integrals are computed by separating the real and imaginary parts of
specially chosen analytic functions. Integrals Il , A and B are computed by
this method in [21], Subsection 29.12, Examples 1, 3, 4 (under the assump-
tion that Qm (x) 6= 0 for x > 0). Integral Ill , to the authors’ knowledge, is
absent from the literature.
7.3. FORMS CONTAINING ln g(x) OR arctan g(x) 269
where Log(z − a) is the principal value of log(z − a) with branch cut along
the half-line [a, +∞), and m ≥ n + 2.
Proof. The function Log(z − a) is analytic in the upper half-plane if
we make a cut along the positive real axis joining the branch points z = a
and z = ∞ and assume that Arg(z − a) = 0 on the upper part of the
cut, Arg(z − a) = 2π on the lower part of the cut and Arg(z − a) = π if
z = x is any point on the real axis such that x < a. Consider a closed path
consisting of the interval [−R, R] (R > |a|) of the x-axis, a semicircle γa of
radius δ around the branch point z = a and a semicircle CR of radius R
(see Fig 7.1). The function Pn (z) Log(z − a)/Qm (z) is analytic inside the
path; therefore, by the residue theorem 5.2.2 we have
Z a−δ Z Z R Z
Pn (z)
+ + + Log(z − a) dz
−R γa a+δ CR Qm (z)
X
Pn (z)
= 2πi Res Log(z − a) . (7.3.2)
z=zk Qm (z)
k
CR
γ
a
–R a 0 R x
Thus,
Z a−δ Z a
Pn (x) Pn (x)
[ln |x − a| + iπ] dx → [ln |x − a| + iπ] dx
−R Qm (x) −∞ Qm (x)
as R → ∞ and δ → 0.
Since z − a = |x − a| on the interval a + δ ≤ x ≤ R, we have the limit
Z R Z ∞
Pn (x) Pn (x)
ln |x − a| dx → ln |x − a| dx
a+δ Qm (x) a Qm (x)
as R → ∞ and δ → 0. Hence, from (7.3.2) we have the formula
Z ∞ Z a
Pn (x) Pn (x)
ln |x − a| dx + iπ dx
−∞ Q m (x) −∞ Q m (x)
X
Pn (z)
= 2πi Res Log(z − a) , (7.3.3)
z=zk Qm (z)
k
X
Pn (z)
= < 2πi Res Log(z − a) Log(z − b) . (7.3.10)
z=zk Qm (z)
k
Substituting the value of the integral (7.3.4) into (7.3.10) and using the
relation =[if (z)] = <f (z), we obtain (7.3.7).
7.3.3. Integrals A, B, C, D. We consider integrals of the form
Z ∞ Z ∞
Pn (x) 2 2 Pn (x)
A= ln |x − a | dx, B= ln |x2 + a2 | dx,
−∞ Q m (x) −∞ Q m (x)
Z ∞ Z ∞
Pn (x) a Pn (x)
C= Arctan dx, D= Arctan x dx.
−∞ Q m (x) x −∞ Q m (x)
Since the evaluation of these integrals make use of the principal values of a
few functions, for simplicity we shall assume that a > 0.
To evaluate integral A it is sufficient to replace a by −a in (7.3.1) and
add the resulting formula to (7.3.1). As a result, we obtain the formula
Z ∞
Pn (x)
ln |x2 − a2 | dx
−∞ Q m (x)
X
Pn (z) 2 2
= < 2πi Res Log(z − a ) , (7.3.11)
z=zk Qm (z)
k
where
(
Arctan(a/x), x ≥ 0,
Arg(x + ai) =
Arctan(a/x) + π, x < 0.
Using (7.3.13) and letting R → ∞, we obtain from (7.3.12) that
Z ∞ p Z ∞
Pn (x) 2 2
Pn (x)
ln x + a dx + i Arg(x + ai) dx
−∞ Q m (x) −∞ Q m (x)
X
Pn (z)
= 2πi Res Log(z + ai) . (7.3.14)
z=zk Qm (z)
k
Equating the real parts in (7.3.14), we obtain the following formula for
evaluating integral B:
Z ∞
Pn (x)
ln(x2 + a2 ) dx
−∞ Qm (x)
X
Pn (z)
= < 4πi Res Log(z + ai) , (7.3.15)
z=zk Qm (z)
k
Using (7.3.13) and (7.3.4) and assuming that the upper limit a in (7.3.4) is
equal to zero, we can rewrite the last formula in the form
Z ∞ X
Pn (x) a Pn (z)
Arctan dx = 2π= i Res Log(z + ai)
−∞ Qm (x) x
k
z=zk Qm (z)
X
Pn (z)
− 2π= i Res Log z , (7.3.16)
z=zk Qm (z)
k
X s
1X
Res + Res
z=zk 2 j=1 z=aj
k
We cut the complex plane along the positive real axis and use the closed
path C shown in Fig 6.8 for the function
Pn (z) 1
f (z) = .
Qm (z) Log z − πi
By the residue theorem 5.2.2,
I X
Pn (z) 1
f (z) dz = 2πi Res
C z=zk Qm (z) Log z − πi
k
Pn (z) 1
+ 2πi Res ,
z=eiπ Qm (z) Log z − πi
that is,
Z Z Z Z
Pn (z) 1
+ + + dz
AB CR eA
B e Cδ Qm (z) Log z − πi
X
Pn (z) 1 Pn (−1) iπ
= 2πi Res + 2πi e . (7.4.1)
z=zk Qm (z) Log z − πi Qm (−1)
k
It can easily be shown that the integrals along the circles CR and Cδ ap-
proach zero as R → ∞ and δ → 0. On the segment AB, z = x and we
have
Z Z R Z ∞
Pn (x) 1 Pn (x) 1
= dx → dx
AB δ Qm (x) ln x − πi 0 Qm (x) ln x − πi
as R → ∞ and δ → 0. On the segment B e A,
e z = x e2πi and we have
Z Z δ Z ∞
Pn (x) 1 Pn (x dx
= dx → −
R Q m (x) ln x + 2πi − πi 0 Q m (x) ln x + πi
as R → ∞ and δ → 0. Therefore, it follows from (7.4.1) that
Z ∞
Pn (x) 1 1
− dx
0 Qm (x) ln x − πi ln x + πi
X
Pn (z) 1 Pn (−1)
= 2πi Res − 2πi
z=zk Qm (z) Log z − πi Qm (−1)
k
Proof. We use the closed rectangular path C shown in Fig 6.6, and
consider the auxiliary function
s
Pn (ez ) X 1
F (z) = z
. (7.5.2)
Qm (e ) z + (2k − 1)πi
k=−s
In the rectangle of height 2π shown in Fig 6.6, the function F (z) has poles
at the zeros, zk , of Qm (ez ) and at the singular point z = πi corresponding
to k = 0 in (7.5.2). The other singular points of (7.5.2), namely,
zk = −(2k − 1)πi, k = −s, −s + 1, . . . , s, k 6= 0,
lie outside the rectangle. Therefore by the residue theorem 5.2.2 we have
I s
Pn (ez ) X 1
z
dz
C Qm (e ) k=−s z + (2k − 1)πi
s
Pn eπi X Pn (ez ) X 1
= 2πi + 2πi Res ,
Qm (eπi ) z=zk Qm (ez ) z + (2k − 1)πi
k k=−s
It can easily be shown that the integrals along sides II and IV tend to zero
as R → ∞. On side I, z = x and we have (see (7.5.2))
Z Z R Z ∞
= F (x) dx → F (x) dx.
I −R −∞
x
Note
X 7.5.1. If Qm (e ) has simple zeros at the points x = a1 , a2 , . . . , ap ,
then Res in (7.5.1) has to be replaced by
z=zk
k
X p
1X
Res + Res .
z=zk 2 s=1 z=as
k
278 7. INTERMEDIATE DEFINITE INTEGRALS
Note 7.5.2. If s = 0 and 0 < =zk < 2π, then formula (7.5.1) becomes
Z
∞
Pn (ex ) dx Pn (−1) X Pn (ez ) 1
x 2 2
= + Res . (7.5.5)
−∞ Qm (e ) x + π Qm (−1) z=zk Qm (ez ) z − πi
k
Therefore we have to use (7.5.5) for the cases z1 = iξ1 and z2 = iξ2 . Since
Pn (−1)/Qm (−1) = 0 in the present example, we obtain
" 2 #0
ez − 1 ez − 1 ez (ez + 1) − ez (ez − 1)
=2 z
z
e +1 e +1 (ez + 1)2
ez (ez − 1)
=4 3 .
(ez + 1)
Thus we have
2b2
J = Res + Res
z=iξ1 z=iξ2 [b2 + [(ez − 1)/(ez + 1)]2 ] (z − πi)
3 3
2b2 e2iθ + 1 2b2 e−2iθ + 1
= +
4(2iθ − πi) e2iθ (e2iθ − 1) 4i(π − 2θ) e−2iθ (e−2iθ − 1)
" 3 #
b2 e2iθ + 1
=
i(2θ − π) e2iθ (e2iθ − 1)
3
b2 eiθ + e−iθ
=
i(2θ − π) eiθ − e−iθ
4b2 cos2 θ
= (since θ = Arctan b)
2i2 (θ − π/2) tan θ
2b2 1
=
π/2 − Arctan b tan θ(1 + tan2 θ)
√
2 a
= √
a Arctan a 1 + 1/a
√
2 a
= √ .
(1 + a) Arctan a
y
β
____
__ III
2 √α
IV II
–R 0 I R x
Proof. We have
Z ∞
2
P =< e−αx −iβx dx
−∞
Z ∞
2 2
= < e−β /(4α) e−α[x+βi/(2α)] dx
−∞
√ √ (7.6.3)
letting α [x + βi/(2α)] = t, dx = 1/ α dt
Z +∞+βi/(2√α )
−β 2 /(4α) 1 −t2
=< e √ e dt .
α −∞+βi/(2√α )
√
To complete the proof of (7.6.2) one has to show that βi/(2 α ) can be
discarded in (7.6.3); then, using (7.6.1), we obtain (7.6.2).
Let us consider a closed rectangular path √ in the complex plane with
base [−R, R] on the x-axis and height β/(2 α ) (see Fig 7.2).
Since the function exp −z 2 has no singular points inside the rectangle,
then by the residue theorem 5.2.2 we have
I Z Z Z Z
2 2
e−z dz = + + + e−z dz = 0. (7.6.4)
C I II III IV
Note 7.6.1. Equation (7.6.5) implies that the horizontal line of inte-
gration
β β
−∞ + i √ , +∞ + i √
2 α 2 α
can be translated parallel to the real axis. Such an operation is a particular
case of deformation of the path of integration. Another deformation will
be seen in the next section.
CR
θ0
0 A x
y
1
y=cos 2x
y=1– 4x/π
π/4
0 x
Z ∞
2
e−r cos 2θ0
cos(r2 sin 2θ0 ) − i sin(r2 sin 2θ0 )
0
√
π
× cos θ0 + i sin θ0 dr = . (7.7.7)
2
Equating the real and imaginary parts in (7.7.7) we obtain
Z ∞
2
e−r cos 2θ0 cos(r2 sin 2θ0 ) cos θ0
0
√
2
π
+ sin(r sin 2θ0 ) sin θ0 dr = (7.7.8)
2
and
Z ∞
2
e−r cos 2θ0
cos(r2 sin 2θ0 ) sin θ0
0
− sin(r2 sin 2θ0 ) cos θ0 dr = 0, (7.7.9)
which is a system of two linear equations in the unknown integrals
Z ∞
2
J1 (θ0 ) = e−r cos 2θ0 cos(r2 sin 2θ0 ) dr, (7.7.10)
Z0 ∞
2
J2 (θ0 ) = e−r cos 2θ0 sin(r2 sin 2θ0 ) dr. (7.7.11)
0
Written more concisely this system becomes
√
π
cos θ0 J1 (θ0 ) + sin θ0 J2 (θ0 ) = , (7.7.12)
2
sin θ0 J1 (θ0 ) − cos θ0 J2 (θ0 ) = 0. (7.7.13)
It follows from (7.7.9) and (7.7.13) that
√ √
π π π
J1 (θ0 ) = cos θ0 , J2 (θ0 ) = sin θ0 , 0 ≤ θ0 ≤ . (7.7.14)
2 2 4
Finally, letting θ0 = π/4 in (7.7.10), (7.7.11) and (7.7.14), we have the
formulae
Z ∞ r Z ∞ r
2 1 π 2 1 π
cos r dr = , sin r dr = ,
0 2 2 0 2 2
which coincide with formulae (7.7.1).
Z ∞
x+4
2. dx.
0 x4 + x2 + 1
Z ∞
x+1
3. dx.
0 x4 + 1
Z ∞
1
4. dx.
0 x2 + 2x + 2
Z ∞
ln x
5. dx, a > 0, b > 0.
0 (x2 + a2 )(1 + b2 x2 )
Z ∞
(1 − x2 ) ln x
6. dx.
0 (1 + x2 )2
Z ∞
x2 ln x
7. dx, ab > 0.
0 (a2 + b2 x2 )(1 + x2 )
Z ∞
(ln x)2
8. dx.
0 (x − 1)(x + a)
Z ∞
(ln x)2
9. dx.
0 x2 + x + 1
Z ∞
(1 + x2 ) (ln x)2
10. dx.
0 1 + x4
Z ∞
1
11. dx.
0 (x2
+ a2 )[(ln x)2
+ π2 ]
Prove the following formulae.
Z ∞
(ln x)3
12. dx = 0.
0 x2 + 1
Z ∞
ln x π
13. 2 + 1)2
dx = − .
0 (x 4
Z ∞
ln(x2 + 1)
14. dx = π ln 2.
0 x2 + 1
Evaluate the following integrals.
Z ∞
1
15. p. v. dx.
0 (x − 2)(x2 + 4)
Z ∞
1
16. p. v. 2 + 2x + 2)
dx.
0 (x − 4)(x
Z ∞
ln |x − 2|
17. 2 2
dx.
−∞ (x + 4)(x + 9)
286 7. INTERMEDIATE DEFINITE INTEGRALS
Z ∞
ln |x − 1|
18. dx.
−∞ x2 + 1
Z∞
x ln |x − 1| ln |x − 5|
19. dx.
−∞ (x2 + 1)(x2 + 4)
Z∞
ln |x − 2| ln |x − 4|
20. dx.
−∞ x2 + 2x + 10
Z ∞
x Arctan x
21. dx.
−∞ (x2 + 4x + 20)(x2 + 1)
Z∞
Arctan x
22. dx.
−∞ x2 + 3x + 8.5
CHAPTER 8
respectively:
• Z and Ze denote the sets of complex zeros of Qm (z) and Qm (−z) in the
upper half-plane, respectively:
Be1 = {a ∈ R, ∀ ãk ∈ A,
e cos aãk 6= 0}.
The general idea for the evaluation of these integrals is clear: one sums
residues at the zeros of Qm (x) and sin ax or cos ax. However, general
formulae to evaluate these integrals seem to be missing in the literature.
When |b| ≤ |a|, some particular cases can be found (see [23], Sections
3.743–3.749, with references to older handbooks). But in such examples
(see, for example, [18], pp. 81–82, formulae 30–32, and p. 23, formulae 36–
37), it is impossible to take the inverse Fourier sine and cosine transforms
because, in these transforms, the parameter y (here denoted b) in Iss , Isc , Ics
and Icc varies over the interval [0, +∞).
When |b| > |a|, even particular cases of the last four integrals (8.1.2),
(8.1.3) seem to be absent from the literature.
The main idea of this section is that, although the number of singular
points in these integrals is equal to infinity, these integrals can be expressed
by means of a finite number of terms, namely, by the sum of the residues
at the zeros of Qm (x). For the first two integrals (8.1.1), the sum of the
residues at the zeros of sin ax and cos ax, respectively, is equal to zero,
and the same holds in the case of the last four integrals (8.1.2), (8.1.3) if
|b| < |a|. Moreover, if |b| > |a|, the corresponding series for the last four
integrals can be expressed by a finite sum of residues at the zeros of Qm (x).
It is found that the last four integrals are equal to the sum of some function
of a and b and a 2a-periodic function of b.
8.1. RATIONAL FUNCTIONS TIMES TRIGONOMETRIC FUNCTIONS 289
We first consider the case where Pn (x)/Qm (x) is odd and Qm has no
real zeros.
Formula 8.1.1. If Qm has no real zeros and Pn (x)/Qm (x) is odd, then
Z ∞ X
Pn (x) dx Pn (z) 1
p. v. = 2πi Res ,
−∞ Qm (x) sin ax zk ∈Z Qm (z) sin az
k
m ≥ n + 1. (8.1.5)
Proof. Let
Pn (z) 1
f (z) = (8.1.6)
Qm (z) sin az
be a function of the complex variable z and let C be a closed path that
consists of parts of the segment [−Rk , Rk ] of the real axis, shown in Fig 8.1,
with |a|Rk = (2k+1)π/2, k = 0, 1, . . . , where the zeros of sin az, that is, the
points axl = lπ, l = 0, ±1, ±2, . . . , ±k, are bypassed along the semicircles
γl of radius δ in the upper half-plane, and the semicircle CRk of radius Rk .
By the residue theorem we have
Z Z Rk k Z
!
X Pn (z) 1
+ + dz
C Rk −Rk Qm (z) sin az
l=−k γl
X
Pn (z) 1
= 2πi Res , (8.1.7)
z=zk Qm (z) sin az
k
where zk are the zeros of Qm (z) that lie inside C and the integral from −Rk
to Rk is evaluated along line segments of the x-axis excluding the arcs γl .
It is shown in Lemma 8.1.1 that the integral along the arc CRk ap-
proaches zero as Rk → ∞. Since xl = lπ/a is a simple pole of f (z), then,
290 8. ADVANCED DEFINITE INTEGRALS
CR
k
γ
l
– Rk –δ δ lπ
––– Rk x
a
Proof. Since sin z = sin(x + iy) = sin x cosh y + i cos x sinh y, one has
q
| sin az| = sinh2 ay + sin2 ax. (8.1.11)
8.1. RATIONAL FUNCTIONS TIMES TRIGONOMETRIC FUNCTIONS 291
Then the moduli of the integrals with respect to the first and third intervals
are smaller than 2Cθ0 ; therefore, with θ0 sufficiently small, one can satisfy
the inequality
1 1
2Cθ0 ≤ ε if θ0 ≤ ε. (8.1.14)
3 6C
Since, by the nonheuristic part of the heuristic argument, the integral over
the interval θ0 ≤ θ ≤ π − θ0 approaches zero as Rk → ∞, there exists a
constant K such that for all k ≥ K the following inequality is satisfied:
Z π−θ0
1 ε
C dθ < . (8.1.15)
θ0 | sin az|z∈CRk 3
Therefore, for all ε > 0, there exists a constant K such that, for all k ≥ K,
Z π Z
1 Pn (z) dz
C dθ < ε or < ε. (8.1.16)
0 | sin az|z∈CRk CR Qm (z) sin az
k
The last two inequalities imply that the limit on the left-hand side in
(8.1.10) exists and is equal to zero.
292 8. ADVANCED DEFINITE INTEGRALS
provided Pn (x)/Qm (x) is even and ai /aj is not equal to a rational number
(in other words, the zeros of sin ai x and sin aj x do not coincide if i 6= j).
Integrals of the form (8.1.18), for the case l > 1, seem to be absent from
handbooks, even in the form of examples. An instance of such a formula is
the integral
Z ∞
Pn (x) x3 dx
p. v. , m ≥ n + 5, (8.1.19)
−∞ Qm (x) sin a1 x sin a2 x sin a3 x
where Pn (x)/Qm (x) is even. In this case, the finite sums, of the form
(8.1.9), of the residues at the zeros of sin a1 x, sin a2 x and sin a3 x are equal
to zero since the functions
x3 Pn (x) x3 Pn (x) x3 Pn (x)
, ,
Qm (x) sin a2 x sin a3 x Qm (x) sin a1 x sin a3 x Qm (x) sin a1 x sin a2 x
are odd.
Example 8.1.2. Derive the formula
Z ∞
1 x2 dx
I2 = p. v. 2 2 2 2
−∞ (x + α )(x + β ) sin ax sin bx
π α β
= − ,
−α2 + β 2 sinh aα sinh bα sinh aβ sinh bβ
where a, b, α, β > 0, and a/b 6∈ Q.
8.1. RATIONAL FUNCTIONS TIMES TRIGONOMETRIC FUNCTIONS 293
respectively, if
X X
p q
X l
p ≥ l, m ≥ n + 1, bk +
ck < ak ,
k=1 k=1 k=1
Pn (x)/Qm (x) is even for p − l even, and Pn (x)/Qm (x) is odd for p − l odd.
Second, we consider the case where the function Pn (x)/Qm (x) is neither
even nor odd and Qm has no real zeros.
Formula 8.1.3. If Qm has no real zeros, Pn (x)/Qm (x) is neither even
nor odd and m ≥ n + 1, then
Z ∞
Pn (x)
p. v. dx
−∞ Q m (x) sin ax
X
Pn (z) Pn (−z)
= πi Res − Res , (8.1.20)
zk ∈Z Qm (z) sin az e Qm (−z) sin az
z̃k ∈Z
k
e
where zk ∈ Z and z̃k ∈ Z.
Proof. If we represent f (x) = Pn (x)/Qm (x) as the sum of an odd
and an even function,
1 1
f (x) = [f (x) − f (−x)] + [f (x) + f (−x)]
2 2
(8.1.21)
Pen (x) Pbn (x)
=: + ,
e m (x) Q
Q b m (x)
294 8. ADVANCED DEFINITE INTEGRALS
then
Pen (x) Pbn (x)
and
e m (x) sin ax
Q b m (x) sin ax
Q
are even and odd, respectively. Therefore
Z ∞
Pbn (x)
p. v. dx = 0.
−∞ Qb m (x) sin ax
Thirdly, we show that formulae (8.1.5) and (8.1.20) are still valid if Qm
has real zeros.
8.1. RATIONAL FUNCTIONS TIMES TRIGONOMETRIC FUNCTIONS 295
then Qm (−ak ) 6= 0 since Qm (z) is neither odd nor even. Thus, if ak are
the simple real zeros of Qm (z), then the even function
ψ(z) = Qm (z)Qm (−z)
also has only simple real zeros and ψ(0) = Q2m (0) 6= 0 because sin(aak ) 6= 0
and sin(aãk ) 6= 0, where ãk are the simple real zeros of Qm (−z). If we let
the zeros of ψ(z) be â−p , â−p+1 , . . . , â−1 , â1 , â2 , . . . , âp , where â−r = −ar
for r = 1, 2, . . . , p, and consider the odd function
φ(z) = Pn (z)Qm (−z) − Pn (−z)Qm (z),
then from (8.1.23) we have
X−1 p
X
φ(z)
B= + Res
z=âk ψ(z) sin az
k=−p k=1
X
−1 p
X
φ(âk )
= +
ψ 0 (âk ) sin aâk
k=−p k=1
= 0,
because ψ (z) is odd, and hence φ(z)/[ψ 0 (z) sin az] is odd.
0
Suppose now that condition (8.1.26) does not hold. For instance, let
a1 = −a2 , hence |a1 | = |a2 |, but for the remaining values of i and j (8.1.26)
holds. Then
Qm (z) = z 2 − a21 ψm−2 (z),
where (8.1.26) holds for the polynomial ψm−2 (z). Then the even function
2
Qm (z)Qm (−z) = z 2 − a21 ψm−2 (z)ψm−2 (−z)
has a pair of double zeros at z = a1 and z = −a1 . However, in this case,
the function φ(z) contains the factor z 2 − a21 and the function f (z) is of the
form
Pn (z)ψm−2 (−z) − Pn (−z)ψm−2 (z)
f (z) = ,
(z 2 − a21 ) ψm−2 (z)ψm−2 (−z)
that is, f (z) is an odd function with only simple real poles. Therefore the
equality B = 0 is still valid.
Corollary 8.1.1. Formulae (8.1.5) and (8.1.20) still hold if Qm has
real zeros ak ∈ A for k = 1, 2, . . . , l, and a ∈ B0 .
Proof. The corollary follows from Lemma 8.1.2.
Note 8.1.2. Let f (z) denote any of the three functions
Pn (z) Pn (z) sin bz Pn (z) cos bz
, , , (8.1.27)
Qm (z) cos az Qm (z) sin az Qm (z) cos az
8.1. RATIONAL FUNCTIONS TIMES TRIGONOMETRIC FUNCTIONS 297
because sin 0 = 0.
We first consider the case where Pn (x)/Qm (x) is even and Qm has no
real zeros.
where x2k+1 = (2k + 1)π/(2a) are the zeros of cos ax. Since m ≥ n + 2,
the series (8.1.31) is absolutely convergent. Moreover, since Pn (x)/Qm (x)
is an even function of x, we can use the notation
1 Pn (x2k+1 )
= F x22k+1 . (8.1.32)
a Qm (x2k+1 )
Inserting (8.1.32) into (8.1.31), we have
∞
X
S= (−1)k+1 F x22k+1
k=−∞
−1
X ∞
X
= (−1)k+1 F x22k+1 + (−1)k+1 F x22k+1 . (8.1.33)
k=−∞ k=0
Now putting k = −l−1 in the first term on the right-hand side and changing
the summation from 0 to ∞, as k changes from −∞ to −1, we have
X∞ X ∞
S= (−1)−l F x22(−l−1)+1 + (−1)k+1 F x22k+1 = 0,
l=0 k=0
since 2
(−2l − 1)π
x22(−l−1)+1
= x2−2l−1
= = x22l+1
2a
and (−1)−l l
= (−1) . This implies (8.1.30).
Example 8.1.4. From (8.1.30) we have the formula
Z ∞
1 dx 1 1
I4 = p. v. 2 2
= 2πi Res
−∞ x + β cos bx z=βi z 2 + β 2 cos bz
1 1
= 2πi
2βi cos bβi
π
= .
β cosh bβ
This is a particular case of formula 3.743(4) in [23], p. 416, with a = 0 in
the integrand
cos (ax) 1
.
cos (bx) x2 + β 2
8.1. RATIONAL FUNCTIONS TIMES TRIGONOMETRIC FUNCTIONS 299
Using a technique similar to the one used for the derivation of formula
(8.1.18), we can easily generalize formula (8.1.30) to the following formula.
Formula 8.1.5. If Qm has no real zeros, Pn (x)/Qm (x) is even and
m ≥ n + 2, then
Z ∞
Pn (x) 1
p. v. Ql dx
−∞ Q m (x) k=1 cos ak x
" #
X Pn (z) 1
= 2πi Res Ql , (8.1.34)
zk ∈Z Qm (z)
k k=1 cos ak z
respectively, if
X X
p q
X l
m ≥ n + 1, bk +
ck < ak ,
k=1 k=1 k=1
Pn (x)/Qm (x) is even for p even, and Pn (x)/Qm (x) is odd for p odd.
A similar formula holds for the integral
Z ∞
Pn (x) xl
p. v. Ql Qp dx, (8.1.35)
−∞ Qm (x) k=1 sin ak x r=1 cos br x
if Pn (x)/Qm (x) is even, ai /aj and bi /bj are not equal to rational numbers
for i 6= j and m ≥ n + l + 2.
Note 8.1.4. In formula (8.1.35), instead of Pn (x)xl one may have
q
Y s
Y
Pn (x) sin ck x cos dk x,
k=1 k=1
if
q s l p
X X X X
q ≥ l, m ≥ n + 1, ck +
dk < ak + bk ,
k=1 k=1 k=1 k=1
Pn (x)/Qm (x) is even for l + q even, and Pn (x)/Qm (x) is odd for l + q odd.
Second, we consider the case where the function Pn (x)/Qm (x) is neither
even nor odd and Qm has no real zeros.
300 8. ADVANCED DEFINITE INTEGRALS
e
where zk ∈ Z and z̃k ∈ Z.
Proof. If Pn (x)/Qm (x) is neither even nor odd, m ≥ n + 1 and Qm
has no real zeros, then the value of the integral (8.1.29) is obtained by
representing Pn (x)/Qm (x) as the sum of an even and an odd functions.
Since the function
1 Pn (x) Pn (−x)
−
cos ax Qm (x) Qm (−x)
is odd, then its integral from −∞ to +∞ is equal to zero, so that the value
of (8.1.29) is given by (8.1.36).
Example 8.1.5. Obtain the following formula (cf. (8.1.22)):
Z ∞
dx cos a cosh a
I5 = p. v. 2 + 2x + 2) cos ax
= 2 .
−∞ (x sinh a + cos2 a
Solution. By (8.1.36) we have
1 1
I5 = πi Res + Res
z=−1+i (z 2 + 2z + 2) cos az z=1+i (z 2 − 2z + 2) cos az
1 1
= πi +
2(z + 1) cos az|z=−1+i 2(z − 1) cos az|z=1+i
π π
= +
2 cos [a(−1 + i)] 2 cos [a(1 + i)]
π
= 2<
2 cos [a(1 + i)]
1
=<
cos a cosh a − i sin a sinh a
cos a cosh a + i sin a sinh a
=<
cos2 a cosh2 a + sin2 a sinh2 a
cos a cosh a
= .
sinh2 a + cos2 a
Thirdly, we show that formulae (8.1.30) and (8.1.36) are still valid if
Qm has real zeros.
8.1. RATIONAL FUNCTIONS TIMES TRIGONOMETRIC FUNCTIONS 301
We begin with the first integral, Iss . There are two cases to be consid-
ered:
(1) |b| ≤ |a|,
(2) |b| > |a|.
In the first case, |b| ≤ |a|, the following condition is satisfied on the
arc CRk as Rk → ∞ (see Fig 8.1 and formula (8.1.11)):
Pn (z) sin bz
lim = 0. (8.1.40)
Rk →∞ Qm (z) sin az z∈CRk
Therefore the derivation procedure that has led to formulae (8.1.5) and
(8.1.20) is valid, but for the case which has led to formula (8.1.5), the
function Pn (x)/Qm (x) must be even for the first integral in (8.1.39) and
odd for the second one. Hence, we have the following pair of formulae.
Formula 8.1.7. If |b| ≤ |a|, Pn (x)/Qm (x) is even and m ≥ n+ 2, then
Z ∞ X
Pn (x) sin bx Pn (z) sin bz
p. v. dx = 2πi Res , (8.1.41)
−∞ Qm (x) sin ax zk ∈Z Qm (z) sin az
k
where zk ∈ Z, ak ∈ A for k = 1, 2, . . . , l, and a ∈ B0 .
Similarly, if |b| < |a|, Pn (x)/Qm (x) is odd and m ≥ n + 1, then
Z ∞ X
Pn (x) cos bx Pn (z) cos bz
p. v. dx = 2πi Res . (8.1.42)
−∞ Qm (x) sin ax k
zk ∈Z Qm (z) sin az
In formulae (8.1.41) and (8.1.42), the sum of the residues at the points
ak ∈ A is equal to zero by Lemma 8.1.2.
If the function Pn (x)/Qm (x) is neither even nor odd, the term
Pn (z) sin bz
Res
zk ∈Z Qm (z) sin az
Therefore the derivation procedure which has led to formulae (8.1.5) and
(8.1.20) can be used here. However, in this case, the symmetry breaks down
because of the factor eibz and because the series of residues at the zeros,
zn = nπ/a, n = 0, ±1, ±2, . . . , of sin az is not zero for the first integral
in (8.1.39). Hence, the following term is added to the right-hand side of
(8.1.41):
" ∞
#
X Pn (kπ/a) eibkπ/a
S1 := = πi
Qm (kπ/a) a(−1)k
k=−∞
∞ (8.1.45)
π X k Pn (kπ/a) bkπ
= (−1) cos
a Qm (kπ/a) a
k=−∞
l
X
Pn (z) eibz
+ πi Res + S1 , (8.1.46)
ak ∈A Qm (z) sin az
k=1
− π ≤ x ≤ π. (8.1.48)
We shall use the values of the series (8.1.48) outside the interval [−π, π].
Since each term of the series (8.1.47) is a 2π-periodic function because
cos k(x + 2π) = cos kx, then its sum S(x), which is equal to the right-hand
side of (8.1.47) in the interval −π ≤ x ≤ π, must be 2π-periodic, that is,
X∞ √
(−1)k cos kx π cosh (x − 2pπ) α 1
= √ √ − , (8.1.49)
k2 + α 2 α sinh π α 2α
k=1
with
−π ≤ x − 2pπ ≤ π, p = 0, ±1, ±2, . . . .
Example 8.1.6. Derive Formula 3.743(1) in [23], p. 416:
Z ∞
sin bx dx π sinh (bβ)
I6 = p. v. 2 2
= ,
−∞ sin ax x + β β sinh (aβ)
0 < b ≤ a, <β > 0. (8.1.50)
Solution. The formula follows from formula (8.1.41) since the inte-
grand satisfies the condition of validity of this formula. Thus,
sin bz 1
I6 = 2πi Res
z=βi sin az z 2 + β 2
sin (bβi) 1
= 2πi
sin (aβi) 2βi
π sinh (bβ)
= , 0 < b ≤ a, <β > 0.
β sinh (aβ)
Formula (8.1.41) cannot be used for the case 0 < a < b, but, for all
a > 0 and b > 0, one can use formula (8.1.46). Thus
ibz
e 1
I6 = = 2πi Res + S1
z=βi sin az z 2 + β 2
e−bβ 1
= = 2πi + S1 (8.1.51)
i sinh aβ 2βi
π e−bβ
=− + S1 ,
β sinh aβ
where
∞
π X (−1)k bkπ
S1 = cos
a kπ 2 a
k=−∞ a + β2
∞
a X (−1)k bkπ
= 2 cos
π aβ a
k=−∞ k 2 +
π
8.1. RATIONAL FUNCTIONS TIMES TRIGONOMETRIC FUNCTIONS 305
h √ i
πa α1 + √π
cosh bπ α/a
√ − α1 , −π ≤ bπ/a ≤ π,
α sinh π α
= h √ √ i
a 1 + √π
cosh (bπ α/a−2π α )
√ − 1
, π ≤ bπ/a ≤ 3π
π α α sinh π α α
√
by (8.1.47) and (8.1.49) with p = 1 and α = aβ/π. Thus
π cosh bβ , −a ≤ b ≤ a,
β sinh aβ
S1 = (8.1.52)
π cosh (b−2a)β , a ≤ b ≤ 3a.
β sinh aβ
so that
∞
π X Pn (kπ/a) bkπ
S1 = (−1)k cos
a Qm (kπ/a) a
k=−∞
(
X Pn (z) sin bz X Pn (z) eibz
= 2πi Res − = 2πi Res
zk ∈Z Qm (z) sin az zk ∈Z Qm (z) sin az
k k
X l )
Pn (z) eibz
+ πi Res ,
ak ∈A Qm (z) sin az
k=1
(8.1.55)
with
bπ
−π ≤ ≤ π, that is, −a ≤ b ≤ a, m ≥ n + 2.
a
The series on the left-hand side of (8.1.55) does not change if we replace
bπ/a with bπ/a − 2pπ for p = 0, ±1, . . . , that is, S1 does not change under
306 8. ADVANCED DEFINITE INTEGRALS
X
Pn (z) sin (b − 2pa)z
S1 = 2πi Res
zk ∈Z Qm (z) sin az
k
( )
X Pn (z) ei(b−2pa)z
−= 2πi Res +πi Res ,
zk ∈Z ak ∈A Qm (z) sin az
k
− a ≤ b − 2pa ≤ a. (8.1.56)
Hence, (8.1.56) gives the formula for the sum, S1 , of the series.
Substituting the value for S1 from (8.1.56) into (8.1.46), we obtain a
formula to evaluate the first integral in (8.1.39), which is valid for any
relation between a and b (b > 0).
where zk ∈ Z, ak ∈ A and a ∈ B0 .
with (2p − 1)a ≤ b ≤ (2p + 1)a. It follows from (8.1.58) that, if p = 0, then
π sinh bβ
I6 = , 0 ≤ b ≤ a,
β sinh aβ
8.1. RATIONAL FUNCTIONS TIMES TRIGONOMETRIC FUNCTIONS 307
which coincides with the value (8.1.50) found before. It follows from (8.1.58),
in the case p = 1, that
Formula 8.1.10. If (2p − 1)a < b < (2p + 1)a for p = 0, ±1, ±2, . . . ,
Pn (x)/Qm (x) is odd and m ≥ n + 1, then
Z ∞ X
Pn (x) cos bx Pn (z) cos (b − 2pa)z
p. v. dx = 2πi Res
−∞ Qm (x) sin ax zk ∈Z Qm (z) sin az
k
(
X
+< 2πi Res +πi Res
zk ∈Z ak ∈A
k
)
Pn (z) 1 ibz
i(b−2pa)z
e −e , (8.1.59)
Qm (z) sin az
∞
π X k+1 Pn (kπ/a) bkπ
S2 = (−1) sin
a Qm (kπ/a) a
k=−∞
X
Pn (z) cos (b − 2pa)z
= 2πi Res (8.1.60)
zk ∈Z Qm (z) sin az
k
( )
X
Pn (z) ei(b−2pa)z
−< 2πi Res +πi Res ,
zk ∈Z ak ∈A Qm (z) sin az
k
Example 8.1.7. If (2p − 1)a < b < (2p + 1)a, derive the following
formula:
Z ∞
x cos bx
I7 = p. v. 2 + β 2 sin ax
dx
−∞ x
z cos (b − 2pa)z
= 2πi Res 2
z=βi z + β 2 sin az
(8.1.61)
z 1 ibz i(b−2pa)z
+ < 2πi Res 2 e −e
z=βi z + β 2 sin az
cosh (b − 2pa)β 1 h i
=π +π e−bβ − e−(b−2pa)β .
sinh aβ sinh aβ
Solution. The formula follows from (8.1.59). In fact, if p = 0, we
have
cosh bβ
I7 = π , −a < b < a,
sinh aβ
which coincides with formula 3.743(3) in [23], p. 416. If p = 1,
π
I7 = sinh (b − 2a)β + e−bβ , a < b < 3a.
sinh aβ
where, in the first integral, the function Pn (x)/Qm (x) is even and m ≥ n+2,
and, in the second integral, this function is odd and m ≥ n + 1. These
integrals are evaluated as the integrals Iss and Isc (8.1.39) in the previous
Subsection 8.1.4.
If |b| < |a|, we have the following formulae for Icc and Ics .
Formula 8.1.12. If |b| ≤ |a|, Pn (x)/Qm (x) is even and m ≥ n + 2,
then
Z ∞ X
Pn (x) cos bx Pn (z) cos bz
p. v. dx = 2πi Res , (8.1.64)
−∞ Qm (x) cos ax zk ∈Z Qm (z) cos az
k
Z ∞ X
Pn (x) cos bx Pn (z) cos (b − 2pa)z
p. v. dx = (−1)p 2πi Res
−∞ Qm (x) cos ax zk ∈Z Qm (z) cos az
k
(
X
+< 2πi Res +πi Res
zk ∈Z ak ∈A
k
)
Pn (z) 1 ibz
e − (−1)p ei(b−2pa)z , (8.1.66)
Qm (z) cos az
Z ∞ X
Pn (x) sin bx Pn (z) sin (b − 2pa)z
p. v. dx = (−1)p 2πi Res
−∞ Qm (x) cos ax zk ∈Z Qm (z) cos az
k
(
X
+= 2πi Res +πi Res
zk ∈Z ak ∈A
k
)
Pn (z) 1 ibz
p i(b−2pa)z
e − (−1) e . (8.1.67)
Qm (z) cos az
While deriving (8.1.66) one finds the sum of the following series:
∞
π X Pn (2k + 1)π/(2a) b(2k + 1)π
S3 = (−1)k sin
a Qm (2k + 1)π/(2a) 2a
k=−∞
X
Pn (z) cos (b − 2pa)z
= (−1)p 2πi Res − (−1)p (8.1.68)
zk ∈Z Qm (z) cos az
k
( )
X Pn (z) ei(b−2pa)z
×< 2πi Res +πi Res ,
zk ∈Z ak ∈A Qm (z) cos az
k
Similarly, while deriving (8.1.67) one finds the sum of the following
series:
∞
π X k+1 Pn (2k + 1)π/(2a) b(2k + 1)π
S4 = (−1) cos
a Qm (2k + 1)π/(2a) 2a
k=−∞
X
Pn (z) sin (b − 2pa)z
= (−1)p 2πi Res − (−1)p (8.1.69)
zk ∈Z Qm (z) cos az
k
( )
X
Pn (z) ei(b−2pa)z
×= 2πi Res +πi Res ,
zk ∈Z ak ∈A Qm (z) cos az
k
e−aβ π e−aβ 2π
= = 2πi = = 2aβ .
2 cos(aβi) cosh aβ e +1
Example 8.1.11. Compute the Fourier cosine transform of
x
tan ax,
x2 + β 2
that is,
Z ∞
x sin ax
I11 (y) = p. v. 2 + β 2 cos ax
cos xy dx, y > 0.
0 x
Solution. We have
Z ∞ Z ∞
1 x sin (y + a)x 1 x sin (y − a)x
I11 (y) = p. v. 2 + β 2 ) cos ax
dx − p. v. 2 + β 2 ) cos ax
dx
4 −∞ (x 4 −∞ (x
1
=: (A1 − A2 ),
4
(8.1.71)
which defines A1 and A2 . To evaluate A1 we use formula (8.1.67) with
b = y + a:
βi sin[(y + a − 2pa)βi]
A1 = (−1)p 2πi
2βi cos(aβi)
h i
βi 1 i(y+a)βi p i(y+a−2pa)βi
+ = 2πi e − (−1) e ,
2βi cos(aβi)
where 2(p − 1)a < y < 2pa for p = 1, 2, 3, . . ., or
π n
A1 = (−1)p+1 sinh[(y + a − 2pa)β]
cosh aβ
o
+ e−(y+a)β + (−1)p+1 e−(y+a−2pa)β . (8.1.72)
To evaluate A2 it suffices to replace y + a by y − a (or y by y − 2a) and p
by p − 1 in (8.1.72):
π n
A2 = (−1)p sinh[(y + a − 2pa)β]
cosh aβ
o
+ e−(y−a)β + (−1)p e−(y+a−2pa)β , (8.1.73)
where 2(p − 1)a < y < 2pa for p = 1, 2, 3, . . .. Combining (8.1.71), (8.1.72)
and (8.1.73), we have
π n
I11 (y) = 2(−1)p+1 cosh[(y + a − 2pa)β]
4 cosh aβ
o
+ e−(y+a)β − e−(y−a)β , (8.1.74)
8.1. RATIONAL FUNCTIONS TIMES TRIGONOMETRIC FUNCTIONS 313
π cosh βy
= 2aβ , 0 < y < 2a.
e +1
(8.1.75)
This result is the same as in [18], p. 23, Sect. 1.6, formula (34), provided
we add the restrictive inequality 0 < y < 2a to this formula.
It is easy to verify that the inverse Fourier cosine transform of I11 (y)
of (8.1.74) is
Z
2 ∞ x
I11 (y) cos xy dy = 2 tan ax,
π 0 x + β2
as it should be.
Example 8.1.12. Compute the Fourier cosine transform of
x
cot ax,
x2 + β 2
that is,
Z ∞
x cos ax
I12 (y) = p. v. cos xy dx, y > 0.
0 x2 + β 2 sin ax
Solution. We have
Z ∞ Z ∞
1 x cos (y + a)x 1 x cos (y − a)x
I12 (y) = p. v. 2 + β 2 ) sin ax
dx + p. v. 2 + β 2 ) sin ax
dx
4 −∞ (x 4 −∞ (x
1
=: (B1 − B2 ),
4
(8.1.76)
which defines B1 and B2 . To evaluate B1 we use formula (8.1.59) with
b = y + a:
h i
cos[(b − 2pa)βi] 1 ibβi i(b−2pa)βi
B1 = πi + < πi e −e
sin(aβi) sin(aβi)
π n o
= cosh[(b − 2pa)β] + e−bβ − e−(b−2pa)β ,
sinh aβ
where 2(p − 1)a < y < 2pa for p = 1, 2, 3, . . ., or
π n o
B1 = sinh[(y + a − 2pa)β] + e−(y+a)β . (8.1.77)
sinh aβ
314 8. ADVANCED DEFINITE INTEGRALS
where 0 < a < π/2, Pn (x) and Qm (x) are real polynomials of the real
variable x, of degrees n and m, respectively, m ≥ n, Pn (x)/Qm (x) is an
even function of x and Qm (x) 6= 0 for real x.
If Pn (x)/Qm (x) is neither even nor odd, it suffices to consider its even
part since the integral of the odd part is equal to zero.
If 0 < a < π/2, the function x2 − 2ax sin x + a2 has no real zeros since
|(x2 + a2 )/(2ax)| ≥ 1, but if a = π/2, its only real zero is x = π/2.
8.2.1. The particular case Pn (x)/Qm (x) ≡ 1. We first consider the
simple case Pn (x)/Qm (x) ≡ 1 and obtain the following formula.
Formula 8.2.1. Derive the formula
Z ∞
dx π 1 + sin a π
I11 = 2 2
= , 0<a< . (8.2.2)
−∞ x − 2ax sin x + a a cos a 2
This integral was computed for the first time in the monograph [47],
p. 181, formula (2.4.41), in a roundabout way by determining the minimal
eigenvalue of a boundary-value problem in two ways.
Proof. Considering the transformation
1 1
= 2
x2 − 2ax sin x + a2 a cos2 x + (x − a sin x)2
1 1 1
= + ,
2a cos x a cos x − i(x − a sin x) a cos x + i(x − a sin x)
we have
Z ∞ Z ∞
1 dx 1 dx
I11 = p. v. ix − ix) cos x
+ p. v. −ix + ix) cos x
2a −∞ (a e 2a −∞ (a e
(and putting x = −t in the second integral)
Z ∞
1 dx
= p. v. ix − ix) cos x
.
a −∞ (a e
(8.2.3)
Set
1
f (z) = , z ∈ C. (8.2.4)
(a eiz
− iz) cos z
Let Rs = (s + 1)π for s ∈ N and consider the “rectangular” closed path,
C, consisting of those segments of the real interval [−Rs , Rs ] where the
points ak = (2k + 1)π/2 for k = 0, ±1, ±2, . . . , ±s, are bypassed along
the semicircles γk of radii δ in the upper half-plane, and the sides of the
rectangle As Bs Cs Ds , with vertices
As = (Rs , 0), Bs = (Rs , Rs ), Cs = (−Rs , Rs ), Ds = (−Rs , 0)
shown in Fig 8.2. By the residue theorem we have
316 8. ADVANCED DEFINITE INTEGRALS
y
(–Rs , Rs) (Rs , Rs)
Cs Bs
γ
k
Ds As
(–R s ,0) a –s a –k 0 ak a s (Rs ,0) x
Z Z Rs s Z
X
1
+ + dz
ηs −Rs γk (a eiz − iz) cos z
k=−s
X 1
= 2πi Res , (8.2.5)
p
z=zp (a eiz − iz) cos z
ηs = As Bs ∪ Bs Cs ∪ Cs Ds , (8.2.6)
zp are the zeros of the function ϕ(z) = aeiz − iz inside C, and the integral
from −Rs to Rs is evaluated along the line segments of the x-axis excluding
the arcs γk .
It will be shown in Lemmas 8.2.1 and 8.2.2 that the integral along ηs
in (8.2.5) approaches zero as Rs = (s + 1)π → ∞ and that there are no
zeros of ϕ(z) = a eiz − iz in the region =z ≥ 0. Consequently, the sum on
the right-hand side of (8.2.5) is equal to zero.
Since ak is a simple pole of f (z), using a Laurent series in a neighbor-
hood of ak we obtain
Z
lim f (z) dz = −πi Res f (z) (8.2.7)
δ→0 γk z=ak
obtain
Z Rs
1 dx
I11 = s→∞
lim ix − ix) cos x
δ→0
a
−Rs (a e
X∞
πi 1
= Res + Res
a z=ak z=−ak (a eiz − iz) cos z
k=0
∞
πi X (−1)k+1 (−1)k+1
= −
a a eiak − iak a e−iak + iak
k=0
since e±iak = ±i sin ak = ±i(−1)k (8.2.8)
∞
πX 1 1
= (−1)k+1 +
a a(−1)k − ak a(−1)k − ak
k=0
X∞ ∞
2π (−1)k 2π X (−1)k ak + a
= =
a ak − a(−1)k a a2k − a2
k=0 k=0
π 1
= + tan a ,
a cos a
where ak = (2k + 1)π/2. In order to derive (8.2.8) we have used Formulae
1.421(1) and 1.422(1) in [23], p. 36, namely,
∞
πx 4x X 1
tan = (8.2.9)
2 π (2k − 1)2 − x2
k=1
and
∞
1 4 X (−1)k+1 (2k − 1)
= . (8.2.10)
cos(πx/2) π (2k − 1)2 − x2
k=1
This completes the proof of (8.2.2).
Lemma 8.2.1. The following integral along the path ηs given by (8.2.6)
approaches zero as s → ∞:
Z
dz π
lim = 0, 0<a< . (8.2.11)
s→∞ η (a eiz − iz) cos z 2
s
Next, we obtain an upper bound for the integral along ηs on the right-hand
side of (8.2.12). We have
q
| cos z|z∈ηs = sinh2 y + cos2 x , (8.2.13)
(x,y)∈ηs
and
|a eiz − iz|z∈ηs ≥ |iz| − |a| |eiz | z∈ηs
π π 1
≥ Rs − e−y ≥ (s + 1)π − = s+ π.
2 2 2
Consequently,
1 1
|f (z)|z∈η ≤ .
s (s + 1/2)π | cos z|
On the segment As Bs ,
thus we have
Z Z (s+1)π
|dz| dy
= p
As Bs | cos z| 0 sinh2 y + 1
Z (s+1)π
dy
=
0 cosh y (8.2.15)
(s+1)π
= 2 arctan (ey )
0
π π π
→2 − = , as s → ∞.
2 4 2
hence, we have
Z Z −(s+1)π
|dz| dx
= q
Bs Cs | cos z| (s+1)π 2
sinh (s + 1)π + cos2 x
(and putting x = (s + 1)πt)
Z −1 (8.2.16)
dt
= (s + 1)π q
1 2
sinh (s + 1)π + cos2 (s + 1)πt
→ 0, as s → ∞.
Lemma 8.2.2. If 0 < a < π/2 and z = x + iy, then the function
ϕ(z) = a eiz − iz = y + a e−y cos x + i a e−y sin x − x
has no zeros in the upper half-plane =z ≥ 0.
Proof. The equation ϕ(z) = 0 is equivalent to the following system
of equations:
y + a e−y cos x = 0, (8.2.17)
−y
ae sin x − x = 0. (8.2.18)
Let y ≥ 0. Then e−y ≤ 1 and it follows from (8.2.17) that cos x ≤ 0 so that
π 3π
+ 2kπ < x < + 2kπ, k = 0, ±1, . . . .
2 2
From (8.2.18) we have
|x| = a e−y | sin x|. (8.2.19)
If y ≥ 0, equation (8.2.19) does not have a solution because on the left-hand
side |x| satisfies the inequality |x| ≥ π/2, while on the right-hand side we
have a e−y | sin x| < π/2 since 0 < a < π/2, e−y ≤ 1, and | sin x| ≤ 1.
8.2.2. The case for general Pn (x)/Qm (x). We now turn to the gen-
eral case of integral Iqp . If m = n then
To evaluate S6 in closed form, we use the formulae (see [21], pp. 296–297)
∞
X X
f (n) = −π Res [f (z) cot πz], (8.2.22)
z=ζk
n=−∞ k
and
∞
X X
n f (z)
(−1) f (n) = −π Res , (8.2.23)
z=ζk sin πz
n=−∞ k
where ẑk are the zeros of Qm (z̃)(z̃ 2 − a2 ) and z̃ = (2z + 1)π/2. It follows
from (8.2.24) that
π2 X Pn (z̃) a cos πz + z̃
S6 = − Res . (8.2.25)
a z=ẑk Qm (z̃) (z̃ 2 − a2 ) sin πz
k
8.2. FORMS CONTAINING (x2 − 2ax sin x + a2 )−1 321
Let us compute separately the sum of the residues in (8.2.25) at the points
z̃ = ±a, that is, at the points ẑ1 = a/π − 1/2 and ẑ2 = −(a/π + 1/2):
Pn (z̃) a cos πz + z̃
Res + Res
z=ẑ1 z=ẑ2 Qm (z̃) (z̃ 2 − a2 ) sin πz
Pn (a) a cos πẑ1 + a Pn (−a) a cos πẑ2 − a
= +
Qm (a) 2aπ sin πẑ1 Qm (−a) (−2aπ) sin πẑ2
(since Pn (x)/Qm (x) is even)
Pn (a) cos(a − π/2) + 1 cos(a + π/2) − 1
= +
Qm (a) 2π sin(a − π/2) 2π sin(a + π/2)
Pn (a) 1 + sin a
=− .
Qm (a) π cos a
Then (8.2.25) can be written as
π2 X Pn (z̃) a cos πz + z̃
S6 = − Res
a z=z̃k Qm (z̃) (z̃ 2 − a2 ) sin πz
k
π Pn (a) 1 + sin a
+ , (8.2.26)
a Qm (a) cos a
where z̃k are all the zeros of Qm (z̃) and z̃ = (2z + 1)π/2. Upon substitution
of (8.2.26) into (8.2.20) we obtain the formula for the evaluation of the
integral Iqp :
Z ∞
p Pn (x) dx
Iq = 2 − 2ax sin x + a2
−∞ Q m (x) x
2πi X Pn (z)
= Res
a z=zk Qm (z) (a eiz − iz) cos z
k
(8.2.27)
π2 X Pn (z̃) a cos πz + z̃
− Res
a z=z̃k Qm (z̃) (z̃ 2 − a2 ) sin πz
k
π Pn (a) 1 + sin a
+ ,
a Qm (a) cos a
where zk are the zeros of Qm (z) in the upper half-plane and z̃k are all the
zeros of Qm (z̃) with z̃ = (2z + 1)π/2.
Note 8.2.1. Although formula (8.2.27) is derived under the condition
that m ≥ n + 2, in fact it is still valid if Pn (x)/Qm (x) ≡ 1. In this case ẑk
are the zeros of z̃ 2 − a2 , that is, ẑ1 = a/π − 1/2, ẑ2 = −(a/π + 1/2) and we
obtain Z ∞
dx π 1 + sin a
Iqp = 2 2
= ,
−∞ x − 2ax sin x + a a cos a
which is formula (8.2.2).
322 8. ADVANCED DEFINITE INTEGRALS
where, in general, ϕ(z) is the entire function ϕ(z) = h sin az + z cos az. We
also consider similar integrals, Iϕc and Iϕs , where dx is replaced by cos bx dx
and sin bx dx, respectively.
8.3. FORMS CONTAINING (h sin ax + x cos ax)−1 323
The entire functions ϕ(z) considered in this section generally come from
the solution of Sturm–Liouville differential equations:
d du
k(x) − q(x)u = −λ2 ρ(x)u, (8.3.2)
dx dx
over the interval 0 < x < L or 0 < x < ∞, with appropriate boundary
conditions, where k(x) > 0, ρ(x) > 0, q(x) ≥ 0 are given functions, which
are continuous on the closed interval [0, L], and k 0 (x) is continuous on the
open interval (0, L). If the interval is semi-infinite, 0 ≤ x < ∞, then |u| is
bounded at infinity.
For given boundary conditions, the eigenvalues of (8.3.2) on a finite
interval are the roots of some equation
ϕ(λ) = 0. (8.3.3)
It is known [7] that, under these conditions, equation (8.3.3) has only simple
real roots. In this case, the entire function, ϕ(z), in the integral (8.3.1) has
only real zeros.
On the infinite interval 0 ≤ x < ∞, one considers equation (8.3.2) with
boundary conditions of the first, second or third kind at x = 0 and the
condition |u(x)| < M , M = constant, as x → ∞. Depending upon the
behavior of the functions k(x) and ρ(x) in (8.3.2) as x → ∞, this problem
may have a discrete or a continuous spectrum (see [5]).
If (8.3.2) has only one regular singular point in the finite complex plane,
that is, the functions k(z), q(z) and ρ(z) are analytic, k 0 (z)/k(z) has one
simple pole, and q(z)/k(z) and ρ(z)/k(z) have only one pole whose order is
not higher than two, then the eigenfunctions are (apart from a factor z α )
entire functions with simple real zeros.
The function ϕ(z) was taken to be sin az and cos az in Section 8.1
and a eiz − iz (without zeros in the upper half-plane) in Section 8.2. In
the present section, ϕ(z) will be taken to be h sin az + z cos az, with h >
0 and a > 0. In Section 8.4, ϕ(z) will be taken to be Jp (az)/z p and
Jp+ν (az)/z p−l Jl+ν (bz), where Jp (z) is the Bessel function of the first kind
of order p for p = 0, 1, . . ..
In the following two subsections we consider integrals of the form
Z ∞
Pn (x) dx
Iϕ = p. v. ,
−∞ Q m (x) ϕ(x)
Z ∞
c Pn (x) cos bx
Iϕ = p. v. dx, (8.3.4)
−∞ Q m (x) ϕ(x)
Z ∞
Pn (x) sin bx
Iϕs = p. v. dx,
−∞ Q m (x) ϕ(x)
324 8. ADVANCED DEFINITE INTEGRALS
where ϕ(x) = h sin ax + x cos ax, h > 0, a > 0, Pn (x) and Qm (x) are
polynomials of degrees n and m, respectively, and m ≥ n + 1. Using the
methods of Section 8.1, we express these integrals as finite sums of residues
at the zeros of the polynomial Qm (x).
8.3.1. The integral Iϕ . Consider the integral
Z ∞
Pn (x) dx
Iϕ = p. v. , (8.3.5)
−∞ Qm (x) h sin ax + x cos ax
where the degrees of Pn and Qm satisfy m ≥ n + 1 and h > 0.
The transcendental equation
h sin aλ + λ cos aλ = 0 (8.3.6)
arises in the determination of the eigenvalues, λn , of the following Sturm–
Liouville boundary value problem:
d2 u
+ λ2 u = 0, 0 < x < a, (8.3.7)
dx2
du
u|x=0 = 0, + hu = 0. (8.3.8)
dx x=a
Since (8.3.7) is a particular case of (8.3.2), the entire function
ϕ(z) = h sin az + z cos az (8.3.9)
has only simple real zeros.
First, we consider the case where Qm (x) 6= 0 for real x and Pn (x)/Qm (x)
is an odd function so that the integrand in (8.3.5) is even. If Pn (x)/Qm (x)
is neither even nor odd, then it can be represented as the sum of an even
and an odd function. Moreover, the integral of the quotient of an even
function with the odd function ϕ(x) is zero.
Set
Pn (z) 1
f (z) = , z ∈ C. (8.3.10)
Qm (z) h sin az + z cos az
Recalling that h > 0, we may rewrite (8.3.6) for real x in the form
h tan ax + x = 0 (8.3.11)
and consider the real roots, νk , of this last equation. It is easily seen by
examining the graphs of y = tan ax and y = −x/h that νk satisfy the
inequality
kπ (k + 1)π
< νk < , k = 0, 1, 2, . . . ,
a a
and that −νk also is a root.
Let C be a closed path consisting of the segment [−Rk , Rk ] of the real
axis, where aRk = kπ for k = 1, 2, 3, . . . and the zeros, νl , l = 0, ±1, . . . , ±k,
8.3. FORMS CONTAINING (h sin ax + x cos ax)−1 325
y
(–Rk , Rk ) (Rk , Rk )
Ck Bk
γ
l
Dk Ak
(–R k ,0) ν–k ν–l 0 νl νk (Rk ,0) x
Z Z Z
C |dz|
|Iηk | ≤ + + . (8.3.14)
Rk Ak Bk Bk Ck Ck Dk | cos az| |h tan az + z|
On the segment Ak Bk , z = Rk + iy and |dz| = dy; thus we have
q
1
| cos az| z∈A B = sinh2 ay + 1 = cosh ay > eay ,
k k 2
and
|h tan az + z|z∈A = |h tan(kπ + iay) + Rk + iy|
k Bk
Xl
Pn (νk ) 1 (8.3.18)
= πi
Qm (νk ) (h sin az + z cos az)0 z=νk
k=−l
Pn (0) 1
= πi = 0,
Qm (0) ha + 1
is zero since
Pn (−νk ) Pn (νk )
=− ,
Qm (−νk ) Qm (νk )
the derivative of the odd function h sin az + z cos az is even, and Pn (0) = 0.
(For a similar computation, see formula (8.1.9) in Subsection 8.1.2.) Hence
Sl does not contribute to the right-hand side of (8.3.12).
Therefore, taking the limit in (8.3.12) as Rk → ∞, δ → 0, we obtain
Z ∞
Pn (x) dx
p. v.
−∞ Qm (x) h sin ax + x cos ax
X
Pn (z) 1
= 2πi Res , (8.3.19)
z=zk Qm (z) h sin az + z cos az
k
π
= .
h sinh a + cosh a
Second, we consider the case where Qm (x) has the simple real zeros
x = ak for k = 1, 2, . . . , l, and ak 6= νk . By bypassing the zeros ak on the
segment [−Rk , Rk ] along semicircles, δk , in the upper half-plane, we find
that the term
X l
Pn (z) 1
A = πi Res (8.3.20)
z=ak Qm (z) h sin az + z cos az
k=1
It follows from Lemma 8.3.1 that if the real zeros, x = ak , of Qm (x) are
simple and ak 6= νk , then the integral (8.3.5) can be evaluated by (8.3.19).
8.3.2. The integrals Iϕc and Iϕs . We consider integrals of the form
Z ∞
Pn (x) cos bx
Iϕc = p. v. dx,
−∞ Qm (x) h sin ax + x cos ax
Z ∞ (8.3.21)
Pn (x) sin bx
Iϕs = p. v. dx,
−∞ Qm (x) h sin ax + x cos ax
Z ∞
Pn (x) cos bx
p. v. dx
−∞ Qm (x) h sin ax + x cos ax
X
Pn (z) cos bz
= 2πi Res (8.3.22)
z=zk Qm (z) h sin az + z cos az
k
and
Z ∞
Pn (x) sin bx
p. v. dx
−∞ Qm (x) h sin ax + x cos ax
X
Pn (z) sin bz
= 2πi Res . (8.3.23)
z=zk Qm (z) h sin az + z cos az
k
The sum of residues at the real zeros, ak , is equal to zero by virtue of
Lemma 8.3.1.
One can, however, compute the integrals (8.3.21) also in the case where
|b| ≤ |a| (as in Subsections 8.1.2–8.1.4) by using Jordan’s Lemma and letting
cos bx = < eibx and sin bx = = eibx . In this case, the symmetry is lost and
the answer is given by the sum of two terms: a finite sum of residues at
the zeros of the polynomial Qm (z) and an infinite series (that is, the sum
of the residues at the zeros, νk , of the function ϕ(x) = h sin ax + x cos ax).
Equating these answers to the right-hand sides in (8.3.22) and (8.3.23), one
can get equations for the determination of the sum of these series in the
following form (for a similar detailed computation, see Subsection 8.1.2):
X
Pn (z) cos bz
<D = < 2πi Res (8.3.24)
z=zk Qm (z) ϕ(z)
k
and ( #
X
Pn (z) sin bz
=D = = 2πi Res , (8.3.25)
z=zk Qm (z) ϕ(z)
k
where
X ∞
X
Pn (z) eibz Pn (νk ) eibνk
D = 2πi Res + 2πi ,
z=zk Qm (z) ϕ(z) Qm (νk ) ϕ0 (νk )
k k=1
and
∞
X Pn (νk ) cos νk
S8 =
Qm (νk ) ϕ0 (νk )
k=1
( )
X Pn (z) sin bz − eibz
== i Res , (8.3.27)
z=zk Qm (z) ϕ(z)
k
(−1)k z 2k+p
∞
X
Jp (z) = , |z| < ∞. (8.4.4)
k!(k + p)! 2
k=0
Z Rk k Z
X Z
Pn (z) zp
+ + dz
−Rk γl ηk Qm (z) Jp (az)
l=−k
X
Pn (z) zp
= 2πi Res , (8.4.14)
z=zk Qm (z) Jp (az)
k
is added to the right-hand side of (8.4.14). Here Jp0 (s) denotes the derivative
of Jp (s) with respect to s. Using formula (8.4.4) we obtain
X −1
(−1)k (2k + p) z 2k+p−1
∞
αp 1
Ak = 0 k = αpk
Jp (aαk ) 2 k!(k + p)! 2 z=aαk
k=0
X ∞ k 2k+p−1 2k−1 −1
1 (−1) (2k + p)(a/2) αk
= ,
2 k!(k + p)!
k=0
that is, Ak is an odd function of αk .
Since, by assumption, Pn (αk )/Qm (αk ) is even, then the function under
the summation sign in (8.4.18) is an odd function of αk , the term with k = 0
being absent. Hence, Sl = 0. Therefore, considering the limit in (8.4.14)
as Rk → ∞, δ → 0, we obtain a formula for evaluating the integral (8.4.5)
in the form
Z ∞ X
Pn (x) xp Pn (z) zp
p. v. dx = 2πi Res , (8.4.19)
−∞ Qm (x) Jp (ax) k
z=zk Qm (z) Jp (az)
where Qm (x) 6= 0 for real x, zk are the zeros of Qm (z) in the upper half-
plane, m ≥ n + p + 1, and the function Pn (x)/Qm (x) is even. Formula
(8.4.19) has not been found in the literature even in the form of examples.
Example 8.4.1. Compute the integral
Z ∞
1 x
I = p. v. 2 + 1 J (ax)
dx.
−∞ x 1
π
= ,
I1 (a)
where (see formula (8.4.4))
∞
X 1 a 2k+1
I1 (a) =
k!(k + 1)! 2
k=0
is the modified Bessel function of the first kind of order 1.
Secondly, we consider the case Qm (x) = 0 for real x at the points
x = ak , k = 1, . . . , l, where all the zeros ak are simple and ak 6= αk . By
bypassing the zeros ak on the segment [−Rk , Rk ] along semicircles δk in the
upper half-plane, we find that the term
Xl
Pn (z) zp
A = πi Res (8.4.20)
z=ak Qm (z) Jp (az)
k=1
is even. Although each of the functions Jl+ν (bz) and Jp+ν (az) has a branch
point at z = 0 (and z = ∞) for non-integer ν, their ratio, as one can see
from (8.4.22), has no branch point and is a meromorphic function.
In contrast with Subsections 8.1.2–8.1.4, we restrict ourselves to the
case |b| ≤ |a|. Let zk be the zeros of Qm (z) in the upper half-plane, and let
ak be the simple real zeros of Qm (z) such that ak 6= αk , where αk 6= 0 are
real zeros of Jp (az) (the function Jp (az) does not have other zeros). In this
case a computation similar to the one in the previous subsection leads to a
formula expressing the integral (8.4.21) through a finite sum of the residues
at the zeros of Qm (z):
Z ∞
Pn (x) xp−l Jl+ν (bx)
p. v. dx
−∞ Qm (x) Jp+ν (ax)
X
Pn (z) z p−l Jl+ν (bz)
= 2πi Res . (8.4.23)
z=zk Qm (z) Jp+ν (az)
k
By Lemma 8.4.1, the sum of the residues at the simple zeros ak is equal to
zero.
336 8. ADVANCED DEFINITE INTEGRALS
respectively.
Proof. The proof is in two parts.
(1) Let z̃k be a zero of multiplicity nk of f (z). Then f (z) can be
represented in the form
f (z) = (z − z̃k )nk f1 (z), (9.1.3)
where f1 (z) is analytic at z̃k , and f1 (z) 6= 0, f1 (z) 6= ∞ in some neighbor-
hood of z̃k . Taking the logarithm of f (z),
log f (z) = nk log(z − z̃k ) + log f1 (z), (9.1.4)
and differentiating the result, we obtain
f 0 (z) nk f 0 (z)
ϕ(z) := = + 1 . (9.1.5)
f (z) z − z̃k f1 (z)
337
338 9. FURTHER APPLICATIONS OF THE THEORY OF RESIDUES
Since f1 (z̃k ) 6= 0 and f1 (z̃k ) 6= ∞, then f10 (z)/f1 (z) can be expanded in a
Taylor’s series about z̃k . Thus the function f10 (z)/f1 (z) is the regular part
of the Laurent series for ϕ(z) at z̃k and nk /(z − z̃k ) is its principal part. It
is seen that z̃k is a simple pole of ϕ(z) and
Res ϕ(z) = nk ,
z=z̃k
X X
= nk − pk = Zf − Pf .
k k
y
z0
D
0 x
(a)
v γ v
A
w0 w0
0 u
γ
0 u
(b) (c)
Case 1. The function w = f (z) maps the closed path C, in the z-plane
as shown in Fig 9.1(a), into the closed path γ, in the w-plane, not enclosing
the point w = 0, as shown in Fig 9.1(b). Suppose that the point z0 ∈ C
is mapped to the point w0 ∈ γ. As z0 traverses C once in the positive
direction, w0 traverses γ an integer number of times in the positive or
negative direction. However, the number
arg f (z)|z=z0 = arg w0
does not change as z0 goes once or several times along C. In fact, arg w0
increases (up to the point A) then decreases and when w0 returns to its
initial position, arg w0 returns to its initial value. In this case (9.2.2) gives
1
Zf − Pf = VarC arg f (z) = 0. (9.2.4)
2π
9.2. THE ARGUMENT PRINCIPLE 341
Case 2. The function f (z) maps the closed path C into the closed path
γ which encloses the point w = 0, as shown in Fig 9.1(c). In this case,
VarC arg f (z) increases by 2π every time w0 traverses γ in the positive
direction and decreases by 2π every time w0 traverses γ in the negative
direction. Hence, we have
VarC arg f (z) = 2πM, (9.2.5)
where the number M = M+ − M− is as defined in (9.2.1). It follows from
(9.2.2) and (9.2.5) that
Zf − Pf = M, (9.2.6)
that is, the difference between the number of zeros and the number of poles
of f (z), counting orders, in a simply connected domain D bounded by the
path C (on which f (z) has no zeros nor poles) is equal to the number of
times γ is traversed as C is traversed once in the positive direction.
If the function w = f (z) has no poles in D, then Pf = 0 and formula
(9.2.6) reduces to
Zf = M. (9.2.7)
Definition 9.2.2. The index of a point z0 with respect to a closed
curve C in the z-plane is the integer defined by the equation
I
1 dz
n(C, z0 ) = . (9.2.8)
2πi C z − z0
The index is also called the winding number of C with respect to z0 .
One can see that our definition of M is n(γ, 0) in the w-plane, where
γ is the image of C under the mapping w = f (z).
Note 9.2.1. The fact that the path γ is traversed more than once in
the w-plane as C is traversed once in the z-plane means that the w-plane is
considered as a Riemann surface with a corresponding cut. The first time
around γ is made on the first sheet of this surface; the second time around
γ is made on the second sheet, and so on.
Example 9.2.1. Find the number of zeros of w = z 2 − 0.5 in the disk
D : |z| ≤ 1 bounded by the path C : |z| = 1.
Solution. Since the equation of C is z = eiθ , then the image of C is
γ : w = u + iv = e2iθ − 0.5,
that is,
u = cos(2θ) − 0.5, v = sin 2θ, 0 ≤ 2θ ≤ 2π. (9.2.9)
342 9. FURTHER APPLICATIONS OF THE THEORY OF RESIDUES
To show that the term on the right-hand side of (9.3.3) is equal to zero, we
consider the function
g(z)
w(z) = 1 + . (9.3.4)
f (z)
Since, by assumption, |f (z)| > |g(z)| for z ∈ C, it follows from (9.3.4) that
g(z)
|w(z) − 1| = ≤ ρ0 < 1, z ∈ C. (9.3.5)
f (z)
Inequality (9.3.5) implies that w(z) maps the path C onto the path γ, which
lies entirely inside the disk |w − 1| ≤ ρ0 < 1 (see Fig 9.2).
Therefore, γ does not enclose the point w = 0. Thus (see Fig 9.1(b))
g(z)
VarC arg w = VarC arg 1 + = 0,
f (z)
and formula (9.3.3) becomes
ZF = Zf .
Example 9.3.1. Find the number of zeros of the polynomial
F (z) = z 10 − 7z 6 − 2z + 1
inside the unit disk D : |z| ≤ 1.
Solution. Let F (z) = f (z) + g(z), where
f (z) = −7z 6 + 1 and g(z) = z 10 − 2z.
Then, for every z on the unit circle C : |z| = 1
|f (z)| = | − 7z 6 + 1| ≥ | − 7z 6 | − 1 = 7 − 1 = 6,
and
|g(z)| = |z 10 − 2z| ≤ |z 10 | + |2z| = 1 + 2 = 3.
Hence
|f (z)| > |g(z)| > 0, z ∈ C.
v C
ρ
0
γ
0 1 u
The fundamental theorem of algebra (see Exercise 20, Section 3.4) fol-
lows simply from Rouché’s Theorem, as shown in the following example.
Example 9.3.2. Use Rouché’s Theorem to prove that a polynomial of
degree n,
p(z) = z n + a1 z n−1 + a2 z n−2 + · · · + an ,
has exactly n zeros.
Solution. Let
f (z) = z n , g(z) = a1 z n−1 + a2 z n−2 + · · · + an ,
and consider the path CR : |z| = R.
We have
|f (z)| = Rn , z ∈ CR ,
and
|g(z)| = |a1 z n−1 + a2 z n2 + · · · + an |
≤ |a1 |Rn−1 + |a2 |Rn−2 + · · · + |an |
=: g̃(R).
Since
Rn
lim= +∞,
R→∞ g̃(R)
7. Prove the following form of Rouché’s Theorem: Suppose f (z) and g(z)
are meromorphic in a neighborhood of the closed disk |z − a| ≤ R with no
zeros or poles on the circle C : |z − a| = R. If Zf , Zg (Pf , Pg ) are the
number of zeros (poles) of f (z) and g(z), respectively, inside C, counting
orders, and if
|f (z) + g(z)| < |f (z)| + |g(z)|
on C, then
Zf − Pf = Zg − Pg .
f (z) f (z)
(Hint: The inequality
+ 1 < + 1 does not hold on C if f (z)
g(z) g(z) g(z)
f (z)
is real. Then define a branch of log as a well-defined primitive for
g(z)
[f (z)/g(z)]0
.)
f (z)/g(z)
√ We note that the quotient in (c) is equal to 2π for a circle |z| = R, and
2/2 for a square centered at the origin.
9.4.2. Partial fraction expansion theorem. We prove the follow-
ing theorem, which is a particular case of a theorem proved in [41], p. 219.
Theorem 9.4.1 (partial fraction expansion). Suppose that a
meromorphic function f (z) satisfies the condition
I
f (ζ)
lim dζ = 0 (9.4.6)
n→∞ C ζ − z
n
where z is an arbitrary but fixed point lying inside the closed path Cn and
distinct from any poles zk of f (ζ). The integrand in (9.4.7) has simple poles
at ζ = z and at ζ = zk inside the region Gn bounded by Cn . Therefore, by
the residue theorem we have
I X
1 f (ζ) f (ζ)
dζ = f (z) + Res . (9.4.8)
2πi Cn ζ − z ζ=zk ζ − z
zk ∈Gn
However,
f (ζ) f (ζ)
Res = lim (ζ − zk )
ζ=zk ζ − z ζ→zk ζ −z
1
= lim [(ζ − zk )f (ζ)]
(zk − z) ζ→zk
1
= Res f (ζ)
zk − z ζ=zk
1
= Res f (z).
zk − z z=zk
Then (9.4.8) can be written in the form
I X
1 f (ζ) 1
dζ = f (z) + Res f (z). (9.4.9)
2πi Cn ζ − z zk − z z=zk
zk ∈Gn
That is, one first computes the terms related to the poles inside C1 ; then
one adds to the partial sum the terms related to the poles lying between
C1 and C2 , and so on.
9.4. SIMPLE-POLE EXPANSION OF MEROMORPHIC FUNCTIONS 349
Im ζ
Sn γn Rn
Cn
–γ n 0 γn Re ζ
Pn –γ n Qn
Solution. To use (9.4.5) one has to show that (9.4.10) satisfies (9.4.6).
For Cn we take the square Pn Qn Rn Sn with vertical sides through the points
±γn = ±(2n + 1)π/2 (see Fig 9.3). Letting ζ = ξ + iη, we have
I
dζ
|In | : =
Cn (ζ − z) sin ζ
I
|dζ|
≤ (9.4.11)
Cn |ζ − z| | sin ζ|
I
1 |dζ|
≤ ,
dn Cn | sin ζ|
where
q
dn = min |ζ − z|, | sin ζ| = sinh2 η + sin2 ξ , (9.4.12)
Cn
X∞
1 1 1
= Res
sin z n=−∞ z − nπ z=nπ sin z
∞
X (−1)n
=
n=−∞
z − nπ
X∞ ∞
X
(−1)n (−1)n
= + (9.4.16)
n=0
z − nπ n=1 z + nπ
∞
1 X n 1 1
= + (−1) +
z n=1 z − nπ z + nπ
X∞
1 (−1)n
= + 2z .
z n=1
z − n2 π 2
2
Z 1
t − z/γn cosh γn sinh γn
= 2i dt
−1 (t − z/γn ) + 1 cosh2 γn − cos2 γn t
2
Z 1
1 cos γn t sin γn t
+ 2i dt. (9.4.19)
−1 (t − z/γn )2 + 1 cosh2 γn − cos2 γn t
Since γn = (2n + 1)π/2, then for all t ∈ [−1, 1] we have
cosh γn sinh γn cos γn t sin γn t
lim = 1, lim = 0.
n→∞ cosh2 γn − cos2 γn t n→∞ cosh2 γn − cos2 γn t
Since the integrand in (9.4.19) is continuous on the interval −1 ≤ t ≤ 1 for
1 ≤ n < ∞, then the limit and the integration can be interchanged and we
have
Z Z Z 1
cot ζ t
lim + dζ = 2i 2
dt
n→∞ Pn Qn Rn Sn ζ −z −1 t + 1 (9.4.20)
= 0.
Similarly,
Z Z
cot ζ
+ dζ
Qn Rn Sn Pn ζ −z
Z γn
cot (γn + iη) cot (γn − iη)
=i − dη
−γn γn + iη − z γn − iη + z
Z γn n
1
=i 2 2 2
γn cot (γn + iη) − cot (γn − iη)
−γn γn + η + 2izη − z
o
+ (z − iη) cot (γn + iη) + cot (γn − iη) dη
Z γn
1 cosh η sinh η
= 2γn 2 2 2 2 dη
−γn γn + η + 2izη − z cosh η − cos γn
2
Z γn
z − iη cos γn sin γn
+2 2 2 2 2 dη
−γn γn + η + 2izη − z cosh η − cos γn
2
→ 0, (9.4.21)
as n → ∞. The second last integral is zero since the integrand is odd and
the limits of integration are symmetric. Thus (9.4.20) and (9.4.21) imply
(9.4.6). Hence, (9.4.17) can be expanded in partial fractions by means of
(9.4.5) as follows:
∞
X 1
cot z = Res cot z
n=−∞
z − nπ z=nπ
N
X 1
= lim
N →∞ z − nπ
n=−N
N N
1 X 1 X 1
= lim + +
N →∞ z n=1 z − nπ n=1 z + nπ
N
X
1 2z
= + lim
z N →∞ n=1 z 2 − n2 π 2
X∞
1 1
= + 2z 2 − n2 π 2
. (9.4.22)
z n=1
z
We note that the series (9.4.22) and (9.4.16) are absolutely and uni-
formly convergent in any disk |z| ≤ R with deleted rings |z − nπ| ≤ δ
(n = 1, 2, 3, . . .) for arbitrary large R, since the series in (9.4.5) can be
majorized by the convergent series of positive terms
X∞
1
2 2
,
n=1
|n π − R|
because
1 1 1
≤ 2 2 ≤
.
|n2 π 2 − z| n π − |z| 2 2
|n π − R|
The proof of (9.4.4) for the function f (z) = tan z can be done similarly;
one has to take squares, Pn Qn Rn Sn , with vertical sides through the points
±nπ (n = 1, 2, 3, . . .).
Remark 9.4.1. We note that a more general formula than (9.4.5) is
given in Problem 27.02 on p. 262 in [21], namely,
" n
#
X
lim f (z) − g(z; zk ) = 0, (9.4.23)
n→∞
k=1
then
1
g(z; zk ) = Res f (z)
z − zk z=zk
and formula (9.4.23) is transformed into (9.4.5). However, the derivation
of (9.4.23) in [21] is done under more stringent conditions than for (9.4.4),
namely,
I
|dz|
lim |f (z)| = 0. (9.4.24)
n→∞ C
n
|z| +1
It can be shown that 1/ sin z satisfies (9.4.24), but tan z and cot z do not
satisfy this condition. Therefore, in [42], p. 268, one proves the bound-
edness of | cot z| on the paths Cn (that is, condition (9.4.2)), and then
f (z) = cot z − 1/z is expanded in partial fractions by means of (9.4.2).
b 1 , b2 , . . . , bn , . . . ,
exists, is finite and is not equal to zero. If (9.5.3) has no nonzero finite
limit, then the infinite product (9.5.2) is said to be divergent and has no
numerical value.
9.5. INFINITE PRODUCT EXPANSION OF ENTIRE FUNCTIONS 355
where
log(1 + ak ) = Log(1 + ak ) + 2mπi (9.5.16)
and
Log(1 + ak ) = ln |1 + ak | + i Arg(1 + ak ),
− π < Arg(1 + ak ) ≤ π. (9.5.17)
For finite n, one can take any branch of log(1 + ak ) in (9.5.15), that is, any
fixed value of m in (9.5.16), since e2mπi = 1. For instance, taking m = 1,
we have
log(1 + ak ) = ln |1 + ak | + i Arg(1 + ak ) + 2πi. (9.5.18)
However, it should be taken into account that the necessary condition
(9.5.14) for convergence must be satisfied. Therefore
lim Arg(1 + ak ) = Arg 1
k→∞
= 0.
If the series
X∞ ∞
X
Log(1 + ak ) = ln |1 + ak | + i Arg(1 + ak ) (9.5.19)
k=1 k=1
is convergent, then the series, with the kth term given by (9.5.18), is diver-
gent, since
2π + 2π + 2π + · · · + 2π + . . . → ∞, as n → ∞.
Hence, in this case, one should take
n
Y X
n
(1 + ak ) = exp Log(1 + ak ) (9.5.20)
k=1 k=1
358 9. FURTHER APPLICATIONS OF THE THEORY OF RESIDUES
that is,
∞
X
log Q = lim log(1 + ak ).
n→∞
k=1
Hence, the convergence of the series (9.5.20) is necessary and sufficient for
the convergence of the infinite product (9.5.13). Since ak → 0 as k → ∞,
then
log(1 + ak ) ∼ ak , as k → ∞.
P∞ P∞
Thus, both series k=1 log(1 + ak ) and k=1 ak either diverge or converge.
Therefore, we have proved the following theorem.
Theorem 9.5.3. If all ak > 0, then a necessary and sufficient condition
for theP
convergence of the infinite product (9.5.13) is the convergence of the
∞
series k=1 ak .
9.5. INFINITE PRODUCT EXPANSION OF ENTIRE FUNCTIONS 359
where the integral approaches zero uniformly in any disk |z| ≤ R not con-
taining the disks |z − zk | ≤ δ. Supppose, moreover, that f (0) 6= 0. Then
f (z) has the infinite product representation
∞
Y nk
z
f (z) = f (0) 1− , (9.5.24)
zk
k=1
which is uniformly convergent in any bounded region of the complex plane.
Proof. It follows from Theorem 9.1.1 that the logarithmic derivative
F (z) = f 0 (z)/f (z) has simple poles at the zeros zk of the entire function
f (z) and does not have any other poles. Since the order of the zero zk is
nk , then, by (9.1.2),
Res F (z) = nk . (9.5.25)
z=zk
Substituting (9.5.25) into (9.4.5), we obtain
∞
X nk d
F (z) = = log f (z), (9.5.26)
z − zk dz
k=1
and integrating F (z) along any arbitrary path joining the the origin to any
point z and not passing through any zeros of f (z), we obtain
∞
X z=z
log f (z) − log f (0) = nk log(z − zk )
z=0
k=1
∞ (9.5.27)
X z
= nk log 1 − .
zk
k=1
Then (9.5.24) follows by taking the exponential of (9.5.27).
Note 9.5.2. Formula (9.5.24) has not been found in the literature,
where one uses (9.4.2) instead of (9.4.5) for expanding F (z) into partial
fractions in the form
∞
Y
zf 0 (0)/f (0) z
f (z) = f (0) e 1− ez/zk . (9.5.28)
zk
k=1
In this expression, each factor
z
1− ez/zk
zk
is repeated nk times, where nk is the order of the zero zk .
Example 9.5.3. Expand in an infinite product the function
(
(sin z)/z, z 6= 0,
f (z) = (9.5.29)
1, z = 0.
9.5. INFINITE PRODUCT EXPANSION OF ENTIRE FUNCTIONS 361
m
= m cot z − .
z
Since this function is similar to the one in the previous example, condition
(9.5.23) is satisfied and one can use (9.5.24). Thus, we have
m ∞
sin z Y z m
= 1−
z kπ
k=−∞
k6=0
n
z m Y z m
Y n
= lim 1− 1+
n→∞ kπ kπ
k=1 k=1
Y∞ m
z2
= 1− 2 2 ,
k π
k=1
that is,
m ∞
Y m
sin z z2
= 1− 2 2 . (9.5.32)
z k π
k=1
This formula can be found by raising (9.5.30) to the power m since it can
easily be proved that
"∞ #m Y ∞ m
Y z2 z2
1− 2 2 = 1− 2 2 .
k π k π
k=1 k=1
n
Y m Y
n m
z z
= lim 1− 1+
n→∞ (2k − 1)π/2 (2k − 1)π/2
k=1 k=1
Y∞ m
4z 2
= 1− .
(2k − 1)2 π 2
k=1
Thus,
∞
Y m
m 4z 2
cos z = 1− . (9.5.34)
(2k − 1)2 π 2
k=1
Q∞
8. Prove that the infinite product k=1 1 + ak is absolutely convergent if
Q
and only if the infinite product ∞k=1 1 + |ak | is convergent.
Q∞ 2
9. Prove that sinh z = k=1 1 + k2zπ2 .
CHAPTER 10
where f (z) = Pn (z)/Qm (z), Pn (z) and Qm (z) are polynomials of degrees
n and m, respectively, m ≥ n + 2 (also m ≥ n + 1 for the series S3 and S4 ).
We also consider the summation of series of the form
∞
X ∞
X
S7 = f (γk ), S8 = f (γk ) eiγk a , (10.1.2)
k=−∞ k=−∞
where γk are the real roots of some transcendental equation (for instance,
the zeros of an entire function). Finally we consider the summation of S7
where γk are the complex roots of a transcendental equation (for instance,
the roots of the equation sinh z ± z = 0).
Series of the form (10.1.1) are considered in the literature (see, for
example, [21], pp. 241–247, [44], pp. 188–191, [51]), but a systematic study
of the series (10.1.2) seems not to have been done. There are two examples
in [21] for the case where γk are the roots of the equation tan x = x.
The summation of the series (10.1.1) and (10.1.2) is done by means of one
common method based on the following theorem.
then
X X
Pn (z) F 0 (z) Pn (z) F 0 (z)
Res =− Res , (10.1.4)
z=γk Qm (z) F (z) z=zk Qm (z) F (z)
k k
where zk are the zeros of the polynomial Qm (z) and zk 6= γl for all k and l.
Proof. By the residue theorem,
I
Pn (z) F 0 (z)
dz
Ck Qm (z) F (z)
X X
Pn (z) F 0 (z)
= 2πi Res + Res , (10.1.5)
z=zk z=γk Qm (z) F (z)
k k
where γk are the poles of F 0 (z)/F (z) and zk are the zeros of Qm (z) inside
the path Ck . Considering the limit of (10.1.5) as k → ∞ and using (10.1.3),
we obtain (10.1.4).
By choosing F (z) properly, one can find formulae for evaluating the
sums S1 to S6 .
10.2. Summation of S1
We obtain a formula for the summation of series of the form
X∞
Pn (k)
S1 = , m ≥ n + 2, (10.2.1)
Qm (k)
k=−∞
y
Ek Dk
Ck
2k + 1
_ ______ 0 2k + 1 x
______
2 2
Ak Bk
10.3. Summation of S2
To evaluate the sum
∞
X Pn (k)
S2 = (−1)k , m ≥ n + 2, (10.3.1)
Qm (k)
k=−∞
Let us use the same system of paths shown in Fig 10.1. To use Theo-
rem 10.1.1 one has to prove that
I
Pn (z) dz
lim = 0. (10.3.3)
k→∞ C Qm (z) sin πz
k
so that the singular points of f (z), z = ±ai, are simple poles. Using (10.3.5)
we obtain
X∞
(−1)k 1 π 1 1
=− 2 − Res + Res
k 2 + a2 2a 2 z=ai z=−ai z 2 + a2 sin πz
k=1
1 π 1 1
=− 2 − +
2a 2 2ai sin πai 2ai sin πai
1 π 1
=− 2 + .
2a 2 a sinh πa
Therefore
∞
X (−1)k 1 π
=− 2 + . (10.3.7)
k 2 + a2 2a 2a sinh πa
k=1
∞
X (−1)k
. (10.3.8)
(k 2 + a2 )2
k=1
∞
X
(−1)k d 1 π
= − √ √
(k 2 + a2 )2 dα 2α 2 α sinh (π α )
k=1 (10.3.9)
1 1 π π 2 cosh πa
= − 3+ 2 + .
2a a 2a sinh πa 2a sinh2 πa
∞
X (−1)k
. (10.3.10)
k4
k=1
10.3. SUMMATION OF S2 371
To derive (10.3.11) one uses the identity 2 sinh2 πa = cosh 2πa − 1 and
Maclaurin’s series for the functions cosh y and sinh y.
P∞ n 3
In [21], in Problem
P∞ 30.03(8) on p. 297, instead of k=1 (−1) /n
n 4
one should read n=1 (−1) /n since theP given answer (apart
P∞ fromn the3
sign) coincides with (10.3.11). The series ∞ 3
n=1 1/n and n=1 (−1) /n
cannot be evaluated in closed form by means of (10.3.5). However, these
can be evaluated by means of a partial fraction expansion of the logarithmic
derivative of the gamma function (see the hint for Problem 30.10 on p. 299
in [21]).
(10.3.15)
d4 1
= 4
dz 1 − z3!2 + z5!4 − . . . z=0
(we have kept only the terms up to z 4 since higher powers of z, after
differentiation, will disappear as z → 0). Expanding the function on the
right-hand side of (10.3.15) in a Maclaurin series (in even powers of z since
the function is even), we have
1
f (z) = z2 z4
1− 6 + 120 − ... (10.3.16)
= 1 + a2 z 2 + a4 z 4 + . . . ,
where
1 00 1
a2 = f (0), a4 = f (4) (0).
2! 4!
Our aim is to compute a4 . It follows from (10.3.16) that
z2 z4
1 = 1 + a2 z 2 + a4 z 4 + . . . 1 − + − ... . (10.3.17)
6 120
Equating the coefficients of z 2 and z 4 in (10.3.17), we obtain
1 1
0 = a2 − , that is, a2 = ,
6 6
a2 1 7
0 = a4 − + , that is, a4 = .
6 120 360
10.3. SUMMATION OF S2 373
Thus,
7 · 4! d4 z1
f (4) (0) = = 4 . (10.3.18)
360 dz1 sin z1
z=0
Substituting (10.3.18) into (10.3.14) we obtain, as in the previous example,
that
X∞
(−1)k 7π 4
= − .
k4 720
k=1
we have
X∞
(−1)k 1 z (6)
2 = −π lim
k6 6! z→0 sin πz
k=1
π6 z (6)
=− lim
6! z→0
" sin z !# (10.3.21)
π 6
d6
1
=−
6! dz 6 1 − z3!2 + z5!4 − z6
7! + ...
z=0
π6
= − f (6) (0).
6!
To derive (10.3.21) one uses the substitution πz = z1 and replaces z1 by z
again. Thus we have
1
f (z) = z2 z4 z6
1− 3! + 5! − 7! + ...
= 1 + a2 z + a4 z + a6 z 6 + . . . .
2 4
Hence
2 4 6
z2 z4 z6
1 = 1 + a2 z + a4 z + a6 z + . . . 1 − + − + ... .
3! 5! 7!
Equating the coefficients of z 2 , z 4 and z 6 , we have the system
1
0 = a2 −
3!
374 10. SERIES SUMMATION BY RESIDUES
a2 1
0 = a4 − +
3! 5!
a4 a2 1
0 = a6 − + − ,
3! 5! 7!
whose solution is
1 7 31 f (6) (0)
a2 = , a4 = , a6 = = .
6 360 15120 6!
Substituting f (6) (0) into (10.3.21) we obtain
∞
X (−1)k 31π 6
=− ,
k6 30240
k=1
and
∞
(
X Pn (k) ikb 0 ≤ b ≤ 2π, if m ≥ n + 2
S4 = e , (10.4.2)
Qm (k) 0 < b < 2π, if m = n + 1,
k=1
The modulus of the integral on the left-hand side of (10.4.4) has the
bound
I
Pn (z) eiaz
|Jk | := dz
Ck Qm (z) sin πz
I
Pn (z) |eiaz |
≤ |dz|
Ck Qm (z) | sin πz|
I (10.4.5)
1 e−ay
≤ |A| p
|dz|
Ck |z| | sin πz|
Z
1 e−ay
= |A| p
|dz| ,
Ak Bk ∪Bk Dk ∪Dk Ek ∪Ek Ak |z| | sin πz|
and
10.4. SUMMATION OF S3 AND S4 377
∞
X X
Pn (k) Pn (z) eiaz
(−1)k cos ak = −π< Res , (10.4.9)
Qm (k) z=zk Qm (z) sin πz
k=−∞ k
and
∞
X X
Pn (k) Pn (z) ei(b−π)z
cos bk = −π< Res , (10.4.11)
Qm (k) z=zk Qm (z) sin πz
k=−∞ k
and
X ∞
Pn (0) Pn (k)
+ (−1)k cos ak
2Qm (0) Qm (k)
k=1
π X Pn (z) eiaz
=− < Res , (10.4.13)
2 z=zk Qm (z) sin πz
k
and
378 10. SERIES SUMMATION BY RESIDUES
∞
X Pn (k)
Pn (0)
+ cos bk
2Qm (0) Qm (k)
k=1
π X Pn (z) ei(b−π)z
=− < Res , (10.4.15)
2 z=zk Qm (z) sin πz
k
where 0 ≤ b ≤ 2π if m ≥ n + 2 and 0 < b < 2π if m = n + 1, and
Pn (k)/Qm (k) is odd in (10.4.14) and even in (10.4.15).
Example 10.4.1. Evaluate the series
X∞
k sin kx
, α > 0. (10.4.16)
k 2 + α2
k=1
Solution. Since
k/(k 2 + α2 ) = O(1/k),
that is, m = n+1, we can use formula (10.4.14) on the interval 0 < x < 2π:
X∞
k sin kx π z ei(x−π)z
= − = Res + Res
k 2 + α2 2 z=iα z=−iα z 2 + α2 sin πz
k=1
i(x−π)iα
π 1 e 1 e−i(x−π)iα
=− = −
2 2 sin πiα 2 sin πiα (10.4.17)
h i
π 1 1 (π−x)α
=− = e − e−(π−x)α
4 sinh πα i
π sinh α(π − x)
= , 0 < x < 2π.
2 sinh πα
This result coincides with Formula 1.445(1) in [23], p. 40.
Example 10.4.2. Evaluate the series
X∞
cos kx
, α > 0. (10.4.18)
k 2 + α2
k=1
Solution. Since
1/(k 2 + α2 ) = O(1/k 2 ),
that is, m = n + 2, we can use formula (10.4.15) on the interval 0 ≤ x ≤ 2π:
X∞
cos kx 1 π 1 ei(x−π)z
= − − < Res + Res
k 2 + α2 2α2 2 z=iα z=−iα z 2 + α2 sin πz
k=1
i(x−π)iα
1 π e e−i(x−π)iα
=− 2 − < +
2α 2 2iα sin πiα 2iα sin πiα
1 π h i
=− 2 + eα(π−x) + e−α(π−x)
2α 4α sinh πα
10.4. SUMMATION OF S3 AND S4 379
π cosh α(π − x) 1
= − 2, 0 ≤ x ≤ 2π.
2α sinh πα 2α
This result coincides with Formula 5.4.5(1) in [38], p. 730. Note that for
Formula 1.445(2) in [23], p. 40, the open interval 0 < x < 2π can be
replaced by the closed interval 0 ≤ x ≤ 2π.
and
∞
X Pn (2k + 1)
(−1)k cos (2k + 1)a
Qm (2k + 1)
k=−∞
10.4. SUMMATION OF S3 AND S4 381
( )
π X Pn (z) eiaz
= < Res , (10.4.24)
2 z=zk Qm (z) cos πz/2
k
and
∞
X Pn (2k + 1)
(−1)k cos (2k + 1)a
Qm (2k + 1)
k=0
( )
π X Pn (z) eiaz
= < Res , (10.4.30)
4 z=zk Qm (z) cos πz/2
k
and
X∞
Pn (2k + 1)
cos (2k + 1)b
Qm (2k + 1)
k=0
( )
π X Pn (z) ei(b−π/2)z
= < i Res , (10.4.32)
4 z=zk Qm (z) cos πz/2
k
and
∞
X Pn (k)
S6 = (−1)k , m ≥ n + 1, (10.5.2)
Qm (k)
k=1
where Pn (x)/Qm (x) is neither even nor odd.
In [1] Sect. 6.8, p. 264, rational series are summed by means of polygamma
functions which are defined as follows. The logarithmic derivative of the
gamma function
d[log Γ(z)] Γ0 (z)
ψ(z) = =
dz Γ(z)
is called the ψ or digamma function. The nth derivatives of the digamma
functions for n = 0, 1, 2, . . . are called polygamma functions. The expansion
of the digamma function in partial fractions is given in [1], p. 259, formula
(6.3.16)):
∞
X z
ψ(z + 1) = −γ +
n=1
n(n + z)
∞ (10.5.3)
X 1 1
= −γ + − ,
n=1
n n+z
for z 6= −1, −2, −3, . . ., where
γ = −ψ(1) = 0.577215665 . . . (10.5.4)
is Euler’s constant. Differentiating (10.5.3) we have
∞
X ∞
X
0 1 00 1
ψ (z + 1) = , ψ (z + 1) = −2 , (10.5.5)
n=1
(n + z)2 n=1
(n + z)3
and so on. In [1], values of the polygamma functions ψ (n) (z) for real z and
n = 0, 1, 2, 3 are listed in Tables 6.1 and 6.2, pp. 267–271, and values of
the digamma function ψ(z) for complex values of z are listed in Table 6.8,
pp. 288–293. We shall restrict ourselves to an example, similar to the one
given in [1], p. 264.
Suppose one has to evaluate the series
∞
X ∞
X A(n)
u(n) = , (10.5.6)
n=1
B (n)B2 (n)
n=1 1
384 10. SERIES SUMMATION BY RESIDUES
where
B1 (n) = (n + α1 )(n + α2 ) · · · (n + αm ),
(10.5.7)
B2 (n) = (n + β1 )2 (n + β2 )2 · · · (n + βr )2 ,
and A(n) is a polynomial in n whose degree does not exceed m + 2r − 2 and
the constants αi and βi are distinct. Expanding u(n) in partial fractions,
we obtain
Xm Xr Xr
ak b1k b2k
u(n) = + + , (10.5.8)
n + αk n + βk (n + βk )2
k=1 k=1 k=1
where
m
X r
X
ak + b1k = 0, (10.5.9)
k=1 k=1
since the sum of residues of the analytic function u(z) in C is equal to zero.
Substituting (10.5.8) into (10.5.6), we have
X∞ Xm X∞
1 1 1
u(n) = ak − +
n=1 n=1
n + αk n n
k=1
Xr X∞
1 1 1
+ b1k − + (10.5.10)
n=1
n + βk n n
k=1
r
X ∞
X 1
+ b2k ,
n=1
(n + βk )2
k=1
It can be shown that the series (10.5.13) is uniformly convergent for all
x > 0 (see, for example, [32], p. 819). Hence we have
∞
X ∞
X
(−1)k+1 (−1)k
β 0 (x) = , β 00 (x) = 2 . (10.5.14)
(x + k)2 (x + k)3
k=0 k=0
Note that by means of (10.5.12) and the tables for ψ(x) in [1], one can
evaluate β(x), β 0 (x) and β 00 (x).
As an example, we evaluate the following series in closed form:
∞
X ∞
X
n A(n)
(−1) u(n) = (−1)n , (10.5.15)
n=0 n=0
B1 (n)B2 (n)
where A(n), B1 (n) and B2 (n) are the same as in (10.5.7), but here the
degree of the polynomial A(n) is at most m + 2r − 1. The expansion of the
rational function u(n) in partial fractions is similar to (10.5.8):
∞
X m
X X∞ r
X X∞
(−1)n (−1)n
(−1)n u(n) = ak + b1k
n=0 n=0
n + αk n=0
n + βk
k=1 k=1
r ∞ (10.5.16)
X X (−1)n
+ b2k .
n=0
(n + βk )2
k=1
∞
X m
X r
X r
X
(−1)n u(n) = β(αk ) + b1k β(βk ) + b2k β 0 (βk ). (10.5.17)
n=0 k=1 k=1 k=1
and
∞
X
S8 = f (γk ) eiγk a , (10.6.2)
k=−∞
where f (z) = Pn (z)/Qm (z), Pn (z) and Qm (z) are polynomials of degrees n
and m, respectively, m ≥ n + 2, and γk are the zeros of an entire function.
The formulae of this section have not been found in the literature. We shall
consider two cases.
Case 1. We consider the case where γk are the roots of the equation
tan x = −Cx, C = constant, C ≥ −1. (10.6.3)
This last equation is a particular case of the equation
λ2 − h1 h2
cot λl = , h1 ≥ 0, h2 ≥ 0. (10.6.4)
λ(h1 + h2 )
Equation (10.6.4) is of the form
h2 λ
cot λl = − , that is, tan λl = − ,
λ h2
if h1 → ∞ and coincides with (10.6.3) if C = 1/h2 and h1 = 0. Equation
(10.6.4) has only real roots λ = λn , since these roots are the eigenval-
ues of the following self-adjoint Sturm–Liouville problem (see [14], p. 256,
Problem 112):
X 00 (x) + λX = 0, 0 < x < l, (10.6.5)
0 0
X (0) − h1 X(0) = 0, X (l) + h2 X(l) = 0. (10.6.6)
In a note to the table of the first seven roots of (10.6.3) in [14], p. 684, it is
stated that “all the roots of (10.6.3) are real if C ≥ −1.” Negative values
of C occur in Sturm–Liouville problems for spheres.
To sum the series
X∞
S7 = f (γk ),
k=−∞
Hence, substituting (10.6.8) into (10.1.4) and taking the fact into account
that z = 0 is also a pole of F 0 (z)/F (z), S7 is evaluated in closed form,
∞
X Pn (γk )
Qm (γk )
k=−∞
!
X Pn (z) cos z + C(cos z − z sin z)
= − Res − Res , (10.6.10)
z=0 z=zk Qm (z) sin z + Cz cos z
k
z3
1 z2 z − 3! + ...
= − lim
z→0z 2 + a2 z − z3!3 + · · · − z 1 − z2!2 + . . .
1 sin ai 1 sin ai
− −
2 sin ai − ai cos ai 2 sin ai − ai cos ai
sinh a 3
= − 2.
a cosh a − sinh a a
This answer coincides with
P∞ [21], Problem 30.09(2),
P∞ if, changing the lower
limit, in [21] one takes k=−∞ instead of k=1 .
where tan γk = γk , γk 6= 0.
Solution. One can find the sum of the above series by letting a → 0
in the formula of Example 10.6.1. However, the computation of the limit
is not simple. Therefore we use formula (10.6.10) directly with C = −1,
∞
X 1 1 z sin z
= − Res 2
γk2 z=0 z sin z − z cos z
k=−∞
sin z
= − Res
z(sin z − z cos z)
z=0
00
1 z 2 sin z
= − lim
2! z→0 sin z − z cos z
3 5
00
1 z 2 z − z3! + z5! − . . .
=−
2 z − z3!3 + z5!5 − · · · − z 1 − z2!2 + z4!4 − . . .
z=0
" #00
3 1 − z3! + z5! − . . .
2 4
=− 2
2 1 − z10 + . . .
z=0
3 d2 f (z)
=− ,
2 dz 2 z=0
(10.6.13)
where
z2 z4
1− + − ...
f (z) = 3!
z2
5!
= 1 + a2 z 2 + . . . , (10.6.14)
1− 10 + ...
10.6. SERIES INVOLVING REAL ZEROS OF ENTIRE FUNCTIONS 389
Since tan z is an odd function, one sees that γ−k = −γk < 0 is a root of
tan z = z if γk is a root. Hence (10.6.17) reduces to
∞
X 1 1
2 = , γk > 0. (10.6.18)
γk 10
k=1
and
γ1 ≈ 4.49 ≈ 3π/2, γ2 ≈ 7.73 ≈ 5π/2, . . . , γk ≈ (2k + 1)π/2, . . . .
y
y=x
y = tan x
y = tan x
0 π 3π/2
γ1 x
Hence,
X∞ X∞
1 1
≈
γk2 (2k + 1)2 π 2 /4
k=1 k=1
"∞ #
4 X 1
= 2 −1 .
π (2k + 1)2
k=0
where tan γk = γk , γk 6= 0.
Solution. Using formula (10.6.10) with C = −1, we have
∞
X 1 1 z sin z
= − Res 4
γk4 z=0 z sin z − z cos z
k=−∞
(4)
1 z 2 sin z
= − lim
4! z→0 sin z − z cos z
10.6. SERIES INVOLVING REAL ZEROS OF ENTIRE FUNCTIONS 391
(4)
z3 z5
1 z2 z − 3! + 5! − ...
= − lim z3 z5 2 z4
4! z→0 z − − · · · − z 1 − z2! +
3! + 5! 4! − ...
2 4
(4)
1 z 3 1 − z3! + z5! − . . .
= − lim z3 z5 z7
4! z→0 3 − 30 + 840 − . . .
" 2
#(4)
z4
1 1 − z6 + 120 − ...
= − lim 1 z2 z4
4! z→0 3 − 30 + 840 − ...
1 d f (z)
4
=− , (10.6.20)
4! dz 4 z=0
where
z2 z4
1− + − ...
f (z) = 1
6
z2
120
z4
= 3 + a2 z 2 + a4 z 4 + . . . (10.6.21)
3 − 30 + 840 − ...
and
1 00 1 (4)
a2 =
f (0), a4 = f (0). (10.6.22)
2! 4!
It follows from (10.6.21) that
z2 z4
1− + − ...
6 120
1 z2 z4
= 3 + a2 z 2 + a4 z 4 + . . . − + − . . . . (10.6.23)
3 30 840
In (10.6.23), equating the coefficients of z 2 and z 4 , respectively, for a2 and
a4 we have
1 1 a2 1
− =− + , that is, a2 = − ,
6 10 3 5 (10.6.24)
1 3 a2 a4 1
= − + , that is, a4 = − .
120 840 30 3 175
Substituting (10.6.22) and (10.6.24) into (10.6.20) we obtain
∞
X 1 1
4 = ,
γk 175
k=−∞
that is,
∞
X 1 1
= = 0.002857 . . . . (10.6.25)
γk4 350
k=1
392 10. SERIES SUMMATION BY RESIDUES
This result can also be easily obtained by means of formula (10.6.10). The
difference between (10.6.26) and (10.6.25) is about 15%, and it reduces
almost to zero if one takes the sum of the first four terms of the following
series (see, for example, Table 5 on p. 757 in [14]):
∞
X 4
X
1 1
4 ≈
γk γk4
k=1 k=1
1 1 1 1
= + + +
(4.4943)4 (7.7253)4 (10.9410)4 (14.0662)4
= 0.0028327 . . . .
This result differs from the exact value 0.002857 only by 1%. This difference
can be further decreased if we add the sum of the terms of the asymptotics
for the nth term, starting with n = 5:
X∞ ∞
1 16 X 1
= 0.002827 +
γk4 π4 (2k + 1)4
k=1 k=5
16 π 4 1 1 1 1
= 0.002827 + 4 −1− 4 − 4 − 4 − 4
π 96 3 5 7 9
= 0.0028539.
To evaluate the series
∞
X Pn (γk ) iaγk
S8 = e , (10.6.27)
Qm (γk )
k=−∞
and
∞
X X
Pn (γk ) Pn (z) eiaz F 0 (z)
sin aγk = −= Res , (10.6.31)
Qm (γk ) z=zk Qm (z) F (z)
k=−∞ k
for |z| < ∞, that is, Jν0 (z)/Jν (z) is a ratio of two entire functions.
394 10. SERIES SUMMATION BY RESIDUES
where J2 (βk ) = 0.
Formula (10.6.33) occurs in hydrodynamical problems (see [46], p. 245,
Formula 6.52).
Solution. By formula (10.6.32), we have
∞
X
1 1 J20 (z)
= − Res 2
β2 z=0 z J2 (z)
k=−∞ k
X ∞ 2k+1
X∞
2k+2 00
1 k (k + 1)(z/2) k (z/2)
= − lim z (−1) (−1)
2! z→0 k!(k + 2)! k!(k + 2)!
k=0 k=0
" # 00
4 z 2
1 1 − 3! 2 + ...
= − lim
2! z→0 1 − 1 z 2 + . . .
2 3! 2
1
= − f 00 (0),
2
(10.6.34)
where
4
z 2
1− 3! 2 + ...
f (z) =
1 1 z 2 (10.6.35)
2 − 3! 2 + ...
2
= 2 + a2 z + . . . ,
and a2 = f 00 (0)/2. It follows from (10.6.35) that
4 z 2 1 1 z 2
1− + ... = − + . . . 2 + a2 z 2 + . . . . (10.6.36)
3! 2 2 3! 2
Equating the coefficients of z 2 in (10.6.36) we obtain
4 1 a2 2 1 1
− · = − · , that is, a2 = − .
6 4 2 3! 4 6
Then from (10.6.34) we obtain
∞
X ∞
X
1 1 1 1
2 = , or 2 = .
βk 6 βk 12
k=−∞ k=1
where J0 (γk ) = 0.
where
z2
1− 8 + ...
f (z) = z2
1− 4 + ... (10.6.40)
2
= 1 + a2 z + . . . ,
and a2 = f 00 (0)/2. It then follows from (10.6.40) that
z2 z2
1− + . . . = 1 + a2 z 2 + . . . 1 − + ... .
8 4
Equating the coefficients of z 2 , we obtain
1 1 1 1
− = a2 − , that is, a2 = , f 00 (0) = .
8 4 8 4
Thus,
X∞ X∞
1 1 1 1
= and = ,
γk4 16 γk4 32
k=−∞ k=1
where J0 (γk ) = 0.
where f (z) = Pn (z)/Qm (z), Pn (z) and Qm (z) are polynomials of degrees
n and m, respectively, m ≥ n + 2, and γk are the complex roots of the
equations
sinh z + z = 0, (10.7.2)
sinh z − z = 0. (10.7.3)
The roots of (10.7.2) and (10.7.3) appear in the solution of the biharmonic
equation
∆∆u = 0, (10.7.4)
where
∂2 ∂2
∆= 2
+ 2 (10.7.5)
∂x ∂y
is the Laplacian, for the case of an infinite strip in elasticity problems
(see, for example, [43], pp. 330–333, and [48], pp. 26–37). Complex roots
of z tan z = c occur in dielectric spectroscopy (see [22]). A quasi-global
selective method of solution of elementary transcendental equations based
on the iteration theory of Fatou and Julia can be found in [16] and [39]
and will be covered in Chapter 11.
10.7. SERIES INVOLVING COMPLEX ZEROS OF ENTIRE FUNCTIONS 397
00
z2
1 z 2 + 2 + . . .
= lim z 3
2 z→0 2z + 6 + . . .
2
!00
1 2 + z2 + . . .
= lim
2 z→0 2 + z62 + . . .
" #0
4
1 3z + . . .
= lim
2 z→0 2 + z2 + . . .2
6
1 4 · 22 1
= · = .
2 3 · 24 6
If γk = ak + ibk is a root of the equation sinh z + z = 0, then −ak ± ibk and
ak − ibk are also roots of the same equation. Hence,
X∞
1 1 1 1 1
2
+ 2
+ 2
+ 2
= ,
(ak + ibk ) (ak − ibk ) (−ak − ibk ) (−ak + ibk ) 6
k=−∞
that is,
" ∞
#
X 1 1
4< = .
(ak + ibk )2 6
k=−∞
Finally,
∞
X a2k − b2k 1
= ,
(a2k + b2k )2 24
k=−∞
where γk = ak + ibk , ak > 0, bk > 0, and sinh γk + γk = 0.
∞
X sin kx
5. , 0 ≤ x ≤ 2π.
1
k3
∞
X cos kx
6. , 0 ≤ x ≤ 2π.
1
k4
∞
X 1
7. 6, where tan γk = γk , γk 6= 0.
γ
−∞ k
∞
X 1
8. , where J1 (αk ) = 0, k 6= 0.
α2
−∞ k
CHAPTER 11
11.1. Introduction
In interactive or automatic scientific computation, one looks for adapted
methods to solve specific problems that occur in the applications. In this
chapter, which follows [16], [39] and the references therein, we present a
combination of global and local iterative methods to find selective roots of
elementary transcendental equations,
F (z, c) − z = 0, c ∈ C.
Such equations occur in two-point boundary value problems, which could
be called complex Sturm–Liouville eigenvalue problems, after separation
of variables in initial-boundary value problems in physics and engineering.
Examples of such equations are found in dielectric spectroscopy, scattering
problems for metallic grooves, and orbit determination.
It will be shown that the iteration functions in question,
zn+1 = F (zn , c),
have very few attractive fixed points, z = F (z, c), and very few critical
values. Geometric considerations will identify bounded regions of the plane
which contain the attractive fixed points and the critical values of F . More-
over, the attractive fixed points of all but a few branches of the inverse,
F −1 , of F have relatively large basins of attraction. An application of the
Fatou–Julia iteration theory for entire and meromorphic functions will en-
sure convergence to the specified roots, while attempting to avoid attractive
cycles. In the presence of slow convergence near multiple zeros, Steffensen’s
procedure or an interpolation scheme will accelerate convergence. In cases
where the specified roots are known to lie in convex regions, good starting
values can be supplied for an efficient use of a fast convergent local method,
such as Newton’s method.
In Section 11.2, transcendental equations are derived from the bound-
ary conditions of some Sturm–Liouville problems in the complex plane.
Section 11.3 presents basic concepts of the Fatou–Julia theory which will
401
402 11. NUMERICAL SOLUTIONS OF TRANSCENDENTAL EQUATIONS
1 + z tan z = 0. (11.4.6)
0.0 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 3-2 3 3 3 3 3 3 3 3 3 3 3 3 3
0.2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3
0.4 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2-3 3 3 3 3 3 3 3 3 3 3 3 3 3 3
0.6 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2-3 0 3 3 3 3 3 3 3 3 3 3 3 326
0.8 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2-2 3 3 3 3 3 3 3 3 3 3 3 3 3 3
1.0 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 224 3 3 3 3 3 3 3 3 3 4 3 32432
1.2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3 3 3 3 3-3 3 32424
1.4 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2-22424 3 3 3 3 3 3 326 2 2242424
1.6 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2242424 02222 422222428 3-3322424
1.8 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 232242425222222 5222211-3 330 3-324
2.0 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 6-2-4-5222222222222 2-4 326 43224
2.2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 3-3-1 0222222222222-3 4-4 2 324-2
2.4 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2-2 3-6-2-322222222222224 3 3 4 4 3 3
2.6 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 3 32326 126 5322222222226 2 423 3 3 3-3
2.8 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 224 2 3 326-422-32222-422 333 42626-3-3 4
3.0 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 22222-324 022222222-3-322222222262638-1 3 0
3.2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 22222 6222222222222222222222222222232-424 3
3.4 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 22222222222222222222222222222222222222222 0-22892
3.6 2 2 2 2 2 2 2 2 2 2 2 2 222 322222222 2222222222222222222222222222222222230 322
3.8 2 2 2 2 2 2 2 2 2 2 2 22222 3 3222222 22222222222222222222222222222222222222222
4.0 2 2 2 2 2 2 2 2 2 22222222222 3222222222222222222222222222222222222222222222222
4.2 2 2 2 2 2 22222 322222222 22222222222 1 422222222222222222222222222222222222222
4.4 3222222222222 3 222222222222222222222222222222222222222222222222222222222222222
4.6 3222222222222222222222222222222222222222222222222222222222222222222222222222222
4.8 3 322222222222222222222222222222222222222-1222222222222222222222222222222222222
5.0 3 3 422222222222222222222222222222222222222222222222222222222222222222222222222
5.2 3 3 322222222222222222222222222222222222222222222222222222222222222222222222222
5.4 3 3 3222222222222222222222222222222222222 2222222222222222222222222222222222222
5.6 3 3 32222 4 5222222222222222222222222222222222222222222222222222222222222222222
5.8 3 3 3 3 4 422222222222222222222222222222222222222222222222222222222222222222222
6.0 3 3 3 3222222222222222222222222222222222222222222222222222222222222222222222222
6.2 3 3 3 3 3 2222222222222222222222222222222 0 02222222222222222222222222222222222
6.4 3 3 3 3 32222222222222222222222222222222222222222222222222222222222222222222222
6.6 3 3 3 3 3 3 4222222222222222222222222222222222222222222222222222222222222222222
6.8 3 3 3 3 3 3 3 3 222222222222222222222222222222222222222222222222222222222222222
7.0 3 3 3 3 3 3 322222222222222222222222222 722222222222222222222222222222222222222
7.2 3 3 3 3 3 322222222222222222222222222 6 622222222222222222222222222222222222222
7.4 3 3 3 3 3 3222222222222222222222222 5221122222222222222222222222222222222222222
7.6 3 3 3 3 3 3 32222222222 5 4 4 422 5 5 42222222222222222222222222222222222222222
7.8 3 3 3 3 3 3222222222222 4 4 4 4 422 6222222222222222222222222222222222222222222
8.0 3 3 3 3 3 3222222 5 4 4 4 4 4 4 42222 02222222222222222222222222222222222222222
8.2 3 3 3 3 3 3222222 4 4 4 4 4 4 4 42222-42222222222222222222222222222222222222222
8.4 3 3 3 3 3 32222 4 4 4 4 4 4 422222222222222222222222222222222222222222222222222
8.6 3 3 3 3 3 3 4 4 4 4 4 4 4222222222222222222222222222222222222222222222222222222
8.8 3 3 3 3 3 322222222 5222222 522222222222222222222222222222222222222222222222222
9.0 3 3 3 3 3 322222222222222222222222222222222222222222222222222222222222222222222
9.2 3 3 3 3 3 3 3222222222222222222222222222222222222222222222222222222222222222222
9.4 3 3 3 3 3 3 3 22222222222222222**2222222222222222222222222222222222222222222222
9.6 3 3 3 3 3 3 3 3 322222222222222-42222222222222222222222222222222222222222222222
9.8 3 3 3 3 3 3 3 32222222222222222222222222222222222222222222222222222222222222222
10.0 3 3 3 3 3 3 3 3 322222222222222222222222222222222222222222222222222222222222222
10.2 3 3 3 3 3 3 3 3 32222 3 2222222222222222222222222222222222222222222222222222222
10.4 3 3 3 3 3 3 3 3 3 3222222222222222222222222222222222222222222222222222222222222
10.6 3 3 3 3 3 3 3 3 3 322 422222222222222222222222222222222222222222222222222222222
10.8 3 3 3 3 3 3 3 3 3 3222222222222222222222222222222222222222222222222222222222222
11.0 3 3 3 3 3 3 3 3 3 3222222222222222222222222222222222222222222222222222222222222
Im z
z
–f 0 f Re z
f<k
Im z
z
–f 0 f Re z
f=k
Im z
z
f>k –f 0 f Re z
are as listed in Table 2. Moreover the only finite critical points of c cot z
and c tan z are the transcendental critical points ±ic, and the only critical
points of (1/c) cos z and (1/c) sin z are the algebraic critical points kπ and
(2k + 1)π/2, respectively, with algebraic critical values ±1/c.
The level curves, | sin z| = constant and | cos z| = constant, bounding
A are shown in Fig 11.4.
With these considerations we have the following theorem.
Theorem 11.5.1. For any given value c 6= 0 the iteration function
zn+1 = T (zn , c) has two attractive fixed points, and these are in B, if B ⊂ A,
and only if B ∩ A = 6 ∅, where A is the region of attractivity of T and
B = T (A). The two fixed points, if any, are in the forward orbits of the two
transcendental critical points of T for cot and tan, and of the two algebraic
critical values of T for cos and sin, respectively.
The proof of the first part follows from the fact that the mapping
T : A → A ∩ B is contracting. The proof of the last part follows from
Theorem 11.3.1 and Remark 11.3.1.
11.5. ALMOST GLOBAL ITERATION FUNCTIONS 413
Im z
2.4
1.6 2.2
1.4 2.0
1.8
1.2 1.6
1.0 1.4 1.0
0.8 1.2 0.8
0.6
0.6
0.4
0.4
0.2
0.2 Re z
0
−π/2 0 π/2 π
sin z cos z
The equations for the double roots of the equations z = c cot z and
z = tan z are
z + sin z cos z = 0 and z − sin z cos z = 0, (11.5.6)
respectively. Table 3 lists the first seven double roots of these equations
and the values of c for which double roots exist.
Similarly, the equations for the double roots of the equations cz = cos z
and cz = sin z are
z tan s = −1 and z cot z = 1, (11.5.7)
respectively. Tables 4 and 5 lists the first seven double roots of these equa-
tions and the values of c for which double roots exist.
It is to be noted that there are no roots of multiplicity higher than 2,
except for sin z = cz, c = 1 where the origin is a root of multiplicity 3; but
in this case only two roots bifurcate as c moves away from the value 1 since
the third one remains at the origin.
By drawing the images of the real c-axis and branch cuts joining the
branch points cj and c̄j through infinity on the Riemann c-sphere for each
of the four multiple-valued mappings
c → {z : z tan z = c}, c → {z : z cot z = c}, (11.5.8)
and
c → {z : cos z = cz}, c → {z : sin z = cz}, (11.5.9)
one obtains rough graphs of the images of the four quadrants, I, II, III
and IV , of the c-plane into the four regions Ik , IIk , IIIk and IVk as
shown in Figs. 11.5 to 11.8, respectively. For each k = 1, 2, . . ., the union
Ik ∪IIk ∪IIIk ∪IVk forms a fundamental region of the functions c = z tan z,
c = z cot z, c = (1/z) cos z, and c = (1/z) sin z. In Figs. 11.5 and 11.6 one
obtains the corresponding regions in the second, third and fourth quadrants
of the z-plane by reflection through the origin and reflection in the real axis
since z tan z and z cot z are even and real functions, that is,
−z tan (−z) = z tan z
and x tan x is real for real x, and similarly for the second function; this fact
will also be used later in Table 8.
Now with
p
t = (ic − z)(ic + z) and t = cz ± (cz)2 − 1, (11.5.10)
the branches T −1 (t, c) = T −1 (z, c) of the inverses of T (z, c) = c cot z and
T (z, c) = c tan z are
1 1 1
log t + kπ and log − , (11.5.11)
2i 2i t
11.5. ALMOST GLOBAL ITERATION FUNCTIONS 415
will converge to s(k) for any z0 ∈ Sl \ A, provided s(k) is the unique root in
Sl . This follows from the fact that T −1 is contracting in Sl \ A.
The uniqueness can been seen from Figs. 11.5 to 11.8, except in the
following cases. In Fig 11.5, when c ∈ II, the roots s(1) and s(k) , k > 1,
could possibly lie in the same vertical strip of width 2π, and similarly for
s(1) and s(k) , k > 1, in Fig 11.6, when c ∈ IV .
11.5. ALMOST GLOBAL ITERATION FUNCTIONS 417
Im z
II1
2 p
p 4
p 3
2
1
I1 II 2 I2 II 3 I3 II 4 I4
0 π/2 π 3π/2 2π 5π/2 3π Re z
Im z
IV1
2
q q
q 3 4
1 2
Figure 11.6. The figure shows the regions IIIk and IVk ,
k ≥ 1, which are images of the lower half of the c-plane
into the first quadrant of the z-plane by the multiple-
valued mapping c → {z; z cot z = c}. The points q1 =
0, q2 , q3 , . . . , are the double roots of the equation z cot z =
c. The region IV1 is unbounded.
A similar situation occurs for s(1) in Fig 11.7; but in this case, this
difficulty will be resolved by means of Newton’s iteration in the next section.
Any double root of the given equations lies on the boundary of the
region A, and the oval O is tangent to that boundary at that point. At
418 11. NUMERICAL SOLUTIONS OF TRANSCENDENTAL EQUATIONS
Im z
ξ1
III1 IV1 Re z
II1 I1 ξ2 ξ3 ξ4 ξ5
Im z
Re z
0 η2 η3 η4
Im z
ξ1
π Re z
π/2 ξ2 ξ3 2π
b = Im c
II Γ I
1/2
–1/e 0 G e
–1/π – 1/2π a = Re c
A
III –1/2 B
IV
most one indifferent fixed point. These points lie in the Julia set of Fc for
the corresponding c.
11.5.3. The exponential equation. We turn now to the exponential
equation
1
z = ez =: E(z, c). (11.5.18)
c
Here the region of attractivity A is the unit disk in the z-plane for c lying
outside the region G ∪ Γ shown in the right-hand part of Fig 11.10, where
the boundary curve Γ is defined by the relation
1 ξ
Γ = c; c = e , |ξ| = 1 .
ξ
One sees that the oval O has reduced in this case to the unit circle.
Since E(z, c) has only one transcendental critical point, z = 0, and no
algebraic critical points, the iteration
zn+1 = E(zn , c)
can have at most one attractive fixed point. Hence for any c 6∈ G ∪ Γ, the
iteration started at z0 = 0, will converge to the first root, s(1) , of (11.5.18)
inside the unit disk.
The only double root of (11.5.18) is z = 1, and this occurs only when
c = e. By considering the images of the real and imaginary axes, a = <c and
b = =c, respectively, under the multiple-valued mapping c → {z; ez = cz},
one obtains Fig 11.11.
The inverse E −1 of E is given by the logarithm
E −1 (z, c, k) = ln z + ln c + 2kπi, k ∈ Z, (11.5.19)
422 11. NUMERICAL SOLUTIONS OF TRANSCENDENTAL EQUATIONS
where the branch cut of ln z and ln c are taken appropriately along the
negative or positive real axes in the z-plane and c-plane, respectively. For
c in a given quadrant, one can choose a value of k such that the values of
(11.5.19) will lie in a horizontal strip of width 2π which covers only one
image of the given quadrant, except for the first images I1 , II1 , III1 and
IV1 .
In case of the first images, for example, for c ∈ III in the small region
A ⊂ G, E −1 (z, c, k) has two attractive fixed points, s(1) ∈ III1 ∩ A0 and
s(2) ∈ III2 . In such case, to have convergence to s(1) one needs a good
starting value, say, z0 ∈ A0 . The same holds for c ∈ B ⊂ G and for c ∈ II
and c ∈ I inside regions which are symmetric to A and B with respect to
the axis <c.
Otherwise, for c ∈ G \ A ⊂ III, (11.5.19) will converge to s(1) for
z0 ∈ III1 \ A0 , as is illustrated in Fig 11.11. The same holds for c ∈ G and
in the other three quadrants.
y = Im z
2π
II 2 IV 3
a=0
I2 III 3
π
b=0
IV1
a=0
B'
III1 A' a=e b=0
x = Re z
II 1 1
a=0
I1
–π b=0
IV2 II 3
a=0
III2 I3
–2π
method may fail to improve the estimate, may converge to another root or
may be divergent.
Here, we explain the instability as the iterates get closer to a double
root. Let us suppose that z̃ is an indifferent fixed point of Fc̃ (z), that is,
Fc̃ (z̃) = z̃ and |Fc̃0 (z̃)| = 1,
implying that
cos z̃ = c̃z̃ and sin z̃ = −c̃ eiϕ . (11.7.2)
For any angle ϕ ∈ [−π, π), z̃ is on one of the ovals of Fig 11.9. Notice that
z̃ tends to the double root ξi in the oval as ϕ → 0.
Let c ≈ c̃. We want to find a root z of Ec (z) = 0 and write
e = c − c̃, u = z − z̃.
424 11. NUMERICAL SOLUTIONS OF TRANSCENDENTAL EQUATIONS
Thus,
Ec (z) = cos z − cz
(11.7.3)
= cos(z̃ + u) − (c̃ + e)(z̃ + u).
Solving for e,
c̃ (eiϕ − 1)u − 12 c̃z̃u2 + O(|u|3 )
e= ,
z̃ + u
and replacing 1/(z̃ + u) by its Taylor expansion in powers of u ,
1 1 u u2
= − 2 + 3 + O(|u|3 ),
z̃ + u z̃ z̃ z̃
we obtain
c̃ (eiϕ − 1)u 1 (eiϕ − 1)
e= − c̃ + u2 + O(|u|3 ). (11.7.4)
z̃ 2 z̃ 2
11.7. INTERPOLATION NEAR A DOUBLE ROOT 425
As z̃ approaches the double root, the first term tends to zero, and for z̃ = ξi
and c̃ = ci we have e ≈ −c̃u2 /2, that is,
c − ci ≈ −c̃(z − ξi )2 /2.
This means that in a neighborhood of a double root, very small differences
in c result in large differences in z, causing instability.
If greater precision than the one obtained by iterating Fc or Gc and
improved by Steffensen’s formula is desired, we can use interpolation for
solving (11.7.1) near a double root.
The double root ξi and its corresponding ci are known. Suppose we
have c ≈ ci and want to find the root z ∗ of cos z = cz. We start with a
very good estimate, z0 , of z ∗ , and choose four points around z0 , namely,
z1 = z0 + δ, z2 = z0 − δ, z3 = z0 + iδ, z4 = z0 − iδ,
for a small value of δ. We use (11) to compute c1 , . . . , c4 and construct an
interpolating polynomial (with complex coefficients) that verifies
P (ci ) = zi , i = 0, 1, 2, 3, 4, (11.7.5)
and interpolate for c,
z̃0 = P (c).
We take z̃0 as the new z0 and repeat the procedure until | cos z̃0 − cz̃0 | is
sufficiently small or two consecutive values of z̃0 are close enough.
It is crucial to start the process with a very good estimate for z ∗ . To
compute the initial value z0 , we consider
Ec (z) = cos z − cz
(11.7.6)
= cos(ξi + u) − (ci + e)(ξi + u).
Replacing cos(ξi + u) by its third-order Taylor expansion,
cos ξi 2 sin ξi 3
cos(ξi + u) = cos ξi − (sin ξi )u − u + u + O(|u|4 ),
2 6
and recalling that
cos ξi = ci ξi and − sin ξi = ci ,
we get
Ec (z) = Ec (ξi + u)
ci ci (11.7.7)
= − u3 − ξi u2 − eξi − eu + O(|u|4 ),
6 2
and write
ci 3 ci
Qc (u) = − u − ξi u2 − eu − eξi . (11.7.8)
6 2
∗
To find an estimate of z we calculate the three roots u1 , u2 and u3 of the
polynomial Qc (u). Suppose that the two roots that give the two smallest
values for |Ec (ξi + uj )| are u1 and u2 . Then we have the two starting values
426 11. NUMERICAL SOLUTIONS OF TRANSCENDENTAL EQUATIONS
has only one imaginary root iy(, d) for each given real values of and d,
and no multiple roots. This root is an attractive fixed point of
iyn − id
iyn+1 = arcsin ,
that is,
y−d
y = arcsinh
s
2
y − d y − d
= log + + 1.
31. 1.
33. 0.
35. 1.
37. The relative positions of the images of −z and z̄ on the Riemann sphere
with respect to the image of z are the points lying in the same plane parallel
to the z-plane which are diametrically opposite to z and symmetric to z
with respect to the real axis, respectively.
39. Let zn = xn + iyn and α = c + id. Then
lim xn = c and lim yn = d.
n→∞ n→∞
Therefore,
lim = α =⇒ lim |zn | = |α|.
n→∞ n→∞
However, if lim |zn | = |α|, then it is possible, for example, that lim = −c
n→∞ n→∞
and lim = −d. In this case
n→∞
23. Using the Cauchy–Riemann equations, we can write f 0 (z) in the form
f 0 (z) = ux − iuy = vy + ivx
Since f 0 (z) ≡ 0 in D then
ux ≡ 0, uy ≡ 0.
Therefore, u ≡ C1 . Similarly, v ≡ C2 . Hence, f (z) = u+iv = C1 +iC2 = C.
√
41. −i ln 2 − 1 + 2πk, k = 0, ±1, . . . .
√
43. π/2 + 2πk − i ln 2 + 3 , k = 0, ±1, . . . .
(u − 1/4)2 + v 2 = 1/16
Then
u = (a1 + ia2 )x + b1 + ib2 .
Hence u = a1 x+b1 and 0 = a2 x+b2 . Since a2 x+b2 = 0, then w = a1 x+b1 ,
where a1 and b1 are real.
436 ANSWERS TO ODD-NUMBERED EXERCISES
1. <w > 0.
3. Put z = x + iy. Then
x2 + y 2 − 1 2y
w = u + iv = −i .
(x − 1)2 + y 2 (x − 1)2 + y 2
Show that the semi-infinite ray x = 0, −∞ < y < 0 is mapped onto the
semicircle u2 + v 2 = 1, v > 0 in the w-plane and the semi-infinite ray
y = 0, 0 < x < +∞ is mapped onto the two rays: −∞ < u < −1, v = 0
and 1 < u < +∞, v = 0. The image of D is the domain
e = {(u, v) ∈ R2 ; v > 0, u2 + v 2 > 1}.
D
5. (u − 4/3)2 + v 2 = 4/9.
7. Let w = (az + b)/(cz + d). Since z2 = 0 is mapped into w2 = ∞, then
w(0) = b/d = ∞ if d = 0, b 6= 0. Let a/c = α, b/c = β. Then we have a
system α + β = −1, α − iβ = 1, whose solution is α = −i, β = −1 + i.
Hence, w = (i − 1)/z − i.
9. w = [(6 + 7i)z + 1 − 13i]/(17z − 3 + 5i).
11. z1 = 0, z2 = −1.
13. z1,2 = ±1.
15. z = b/(a − 1) if a 6= 1; if a = 1 then any z ∈ C is a fixed point if and
only if b = 0. If a = 1, b 6= 0 there are no fixed points.
17. w = (−z + i)/(z + i).
19. Suppose
z = 2 7→ w = ∞, z = 2i 7→ w = 0, z = −2 7→ w = 1.
z = 0 7→ w = ∞, z = −3 7→ w = 0, z = −1 − 2i 7→ w = i.
11. w = −e−2z .
√
13. w = e3iz .
e = {w; −∞ < <w < +∞, 0 < =w < π/2}.
15. D
e = {w; 0 < <w < ln 2, 0 < =w < π}.
17. D
e = {w; 2 < <w < 2 + ln 2, 1 < =w < π/2 + 1}.
19. D
e = {w; −∞ < <w < +∞, −π/2 < =w < 0}.
21. D
23. w = Log z.
25. w = −(i/π) Log eiπ/4 z + i .
1. The whole w-plane with a cut joining the points −1 and 1 along the real
u-axis.
3. The upper half-plane =w > 0 with the cut [1, +∞) ∪ (−∞, −1] along
the extended real axis. This is easily visualized on the Riemann sphere.
5. The domain between the segment [−1/2(R + 1/R), 1/2(R + 1/R)] of the
real u-axis and the lower part of the ellipse
4u2 4v 2
2
+ = 1.
(R + 1/R) (R − 1/R)2
u2 v2
+ = 1.
cosh π/2 sinh2 π/2
2
13. w = cos z.
15. w = − cos [−iπ(z − 2)].
ANSWERS TO ODD EXERCISES FOR SECTION 3.4 439
Hence the function ef (z) is uniformly bounded in the whole complex plane,
and by Liouville’s Theorem 3.4.5, it is constant in C. Therefore f (z) is
constant in C.
17. Applying Cauchy’s estimate for n = 2, 3, . . . at every point z ∈ C, we
get
|f (n) (z)| < M/rn−1 , for all r > 0 and n = 2, 3, . . . .
00
Integrating the equation f (z) = 0, we obtain that f (z) is a polynomial of
degree at most 1. Assuming that there exist positive constants M and R
such that |f (z)| ≤ M |z|n if |z| ≥ R, one can show that, in this case, f (z)
is a polynomial of degree at most m.
19. u(x, y) = <f (z) = ex cos y. If there is a maximum or a minimum at a
point (x, y) inside R, then we have
ux = ex cos y = 0 and uy = −ex sin y = 0.
Since the system of equations ux = 0, uy = 0 has no real solutions, u(x, y)
has no maxima nor minima inside R. Finally,
umax = e at x = 1, y = 0, and umin = −e at x = 1, y = π.
21. Consider the function g(z) = 1/f (z). Since f (z) 6= 0 for |z| < r, it
follows from the maximum modulus principle that the maximum of |g(z)|,
that is, the minimum of |f (z)|, is assumed on |z| = r. On the other hand,
the maximum of |f (z)| is assumed on |z| = r. Since |f (z)| = constant
on |z| = r, the maximum and the minimum of |f (z)| coincide. Therefore
ANSWERS TO ODD EXERCISES FOR SECTION 4.1 441
f (z) = constant.
23. It follows from Cauchy’s estimate that
|f (n) (z)| ≤ n!M/rn , where M = max |f (ζ)|.
|ζ−z|=r
Suppose |f (n) (z)| > n!nn . Then we should have n!nn < n!M/rn , that is,
nn < M r−n . The last inequality cannot be satisfied for all n since nn grows
faster than r−n for any fixed 0 < r < 1.
25. There is no a contradiction with Liouville’s Theorem since | cos z| is
not bounded in C. Therefore Liouville’s Theorem cannot be applied.
∞
2 1 X (z + 1)2n
19. ez +2z
= , R = ∞.
e n=0 n!
cos2 z 2 7z 4
21. = 1 − 2z + − . . . , R = 1.
1 + z2 3
z2 z4
23. Log(1 + cos z) = Log 2 − − − . . . , R = π.
4 96
3e(z − 1)2
25. e1/z = e − e(z − 1) + + . . . , R = 1.
2
31. −2J2 (x) − J0 (x) + C.
1 1 n 1 1 1
+ + + ... z + ···+ 1 + + + ... + ...
n! n!(n + 1)! 2! 2!3! z
X∞
= In (2)z n ,
n=−∞
where
∞
X (x/2)2k+n
In (x) =
k!(k + n)!
k=0
15. We have
∞ ∞
1X 1 X zn
(a) (−1)n+1 z n − ;
3 n=0 3 n=0 2n+1
∞ ∞
1 X (−1)n+1 1 X zn
(b) − ;
3 n=0 z n+1 3 n=0 2n+1
∞ ∞
1 X (−1)n+1 1 X 2n
(c) + .
3 n=0 z n+1 3 n=0 z n+1
17. We have
∞ ∞
1X 5 X (−1)n z n
(a) − (−1)n z n + ;
2 n=0 2 n=0 3n+1
∞ ∞
1 X (−1)n 5 X (−1)n z n
(b) − + ;
2 n=0 z n+1 2 n=0 3n+1
∞ ∞
1 X (−1)n 5 X (−1)n 3n
(c) − + .
2 n=0 z n+1 2 n=0 z n+1
ANSWERS TO ODD EXERCISES FOR SECTION 5.2 445
X∞ X∞
(−1)n 3n (−1)n (z − 1)n
19. n+1
− .
n=0
(z − 1) n=0
4n+1
1 i 3 i(z + i) 5(z + i)2
21. − + + − − + ....
4(z + i)2 4(z + i) 16 8 64
1
7. Res = 1, k = 0, ±1, ±2, . . . .
z=2kπi ez
−1
9. The residues are
1 − cos z 1
Res = ,
z=0 z 2 sin z 2
(
1 − cos z 0, n = 2k,
Res = k = ±1, ±2, . . . .
z=nπ z 2 sin z −2/(n2 π 2 ), n = 2k − 1,
11. We have
2 1 2 1
z sin = [(z − 1) + 2z − 1] sin
z−1 z−1
2 1
= [(z − 1) + 2(z − 1) + 1] sin
z−1
2 1 1 1
= [(z − 1) + 2(z − 1) + 1] − + + ... .
z − 1 3!(z − 1)3 5!(z − 1)5
2 1 1 5
Hence, Res z sin =1− = .
z=1 z−1 3! 6
sin z 1
13. Res + 3 + e1/z = 1.
z=0 z z
z/(z−1) 1+1/(z−1) 1 1
15. e =e =e 1+ + + ... .
1!(z + 1) 2!(z + 1)2
Hence, Res ez/(z−1) = e.
z=1
Hence
∞
X ∞
1 (−1)n z 2n X 1
e1/z = .
z 2 + 4 n=0 22n+2 n=0 n!z n
Multiplying the series and collecting the terms containing 1/z, we obtain
∞
e1/z 1 X (−1)n (1/2)2n+1 1 1
Res 2 = = sin .
z=0 z + 4 2 n=0 (2n + 1)! 2 2
The sum of the residues at the three singular points, z = 2i, z = −2i and
z = 0, is equal to 0.
29. 0.
31. 0.
33. 0.
448 ANSWERS TO ODD-NUMBERED EXERCISES
13. −π/4.
15. −π/16.
π π
17. ln 8 − ln 13.
20 30
π2 π 1 π2 π
19. − ln 26 + Arctan ln 2 + ln 29 − Arctan(2) ln 29
8 6 5 6 6
π 2
− Arctan ln 5.
6 5
19π 19π 21π 2
21. − ln 2 + ln 29 + Arctan .
377 754 754 5
π cos 2a cosh 2a
5. .
2(sinh2 2a + cos2 2a)
π sinh bα sinh bβ
7. − .
(β 2 − α2 ) α sinh aα β sinh aβ
−aα
π e e−aβ
9. − .
(β 2 − α2 ) cosh aα cosh aβ
π 1 1
11. − .
3 b sinh a + cosh a b sinh 2a + 2 cosh 2a
πβ
13. , where I2 (z) is the modified Bessel function of the first kind of
I2 (aβ)
order 2.
451
452 BIBLIOGRAPHY
[43] I. N. Sneddon, The use of integral transforms, McGraw-Hill, New York, 1972.
[44] M. R. Spiegel, Theory and problems of complex variables, Schaum’s Outline Se-
ries, Schaum Publishing Co., New York, 1964.
[45] A. Sveshnikov and A. Tikhonov, The theory of functions of a complex variable,
Mir, Moscow, 1978. (Russian).
[46] S. M. Targ, Basic problems in the theory of laminar flows, GTTI, Moscow–
Leningrad, 1951 (Russian).
[47] E. F. Tsarkov, Random perturbations of functional differential equations, Zinātne,
Riga, 1989 (Russian).
[48] Ya. S. Uflyand, Integral transforms for elasticity problems, Soviet Academy of
Science Publ. House, Moscow–Leningrad, 1963 (Russian).
[49] L. A. Vainstein, Theory of diffraction and factorization method, Sovetskoye Ra-
dio, Moscow, 1966 (Russian).
[50] E. T. Whittaker and G. N. Watson, A course of modern analysis, 4th ed., Uni-
versity Press, Cambridge, 1963.
[51] A. D. Wunsch, Complex variables with applications, Addison-Wesley, Reading,
MA, 1994.
Index
455
456 INDEX
value of a function, 25
variation of arg f (z), 351