Chapter 4

Elements of Complex Analysis

4.1 T h e A l g e b r a of C o m p l e x N u m b e r s

4.1.1 Basic Definitions

Complex numbers occur naturally in several areas of physics and engineer-
ing, for example, in the study of Fourier series and transforms and related
applications to signal processing and wave propagation. They also appear
in the mathematical models of quantum mechanics. Still, the original moti-
vation to introduce complex numbers was the study of roots of polynomials.
A polynomial p = p(x) in the v a r i a b l e x is an expression

anxn + an-ixn~x H aix + a0.

The numbers a,j are called the c o e f f i c i e n t s of the polynomial; if an ^ 0,

then n is called t h e degree of the polynomial. A polynomial equation
is p(x) = 0, and a root of p is, by definition, a solution of this equation.
So far, our underlying assumption was that the reader has some ba-
sic familiarity with the construction of the real numbers. At this point,
though, we will go back to the foundations of the theory of real numbers.
In the late 1800's, German mathematicians G E O R G FERDINAND LUDWIG
P H I L I P P C A N T O R (1845-1918) and JULIUS WILHELM RICHARD DEDEKIND
(1831-1916) put the construction of the real number system on a precise
mathematical foundation by combining, in a rather sophisticated way, set
theory and analysis. A modern approach that we will outline next, is more
algebraic and leads to the complex numbers in a natural way. The remark

God made the integers and all the rest is the work of man,
attributed to the German mathematician LEOPOLD KRONECKER (1823-
1891), suggests the set N = {1,2,3,...} of positive integers as the starting

179
180 Algebra of Complex Numbers

point of the construction.

The set N is naturally equipped with two binary operations, addition
(a,b) —
I > a + b, and multiplication (a,b) >—> ab. These operations are
associative:

(a + b) + c = a + (b + c), (ab)c = a(bc),

and distributive:

a(b + c) = ab + ac.

To solve the linear equation x + a = b for every a, b G N the set of

positive integers is extended to the set Z = {0, ± 1 , ± 2 , . . . } of all integers
by introducing the special number 0 so that a + 0 = a for all a and adjoining
to every non-zero a G N the additive inverse —a so that a + (—a) = 0.
To solve the linear equation ax = b for every a, b G Z, the set of integers
is extended to the set Q of rational numbers, that is, expressions of the
form p/q = pq_1, where p, q G Z and q ^ 0. For every non-zero p/q G Q,
the element q/p satisfies {p/q)(q/p) — 1 and is called t h e m u l t i p l i c a t i v e
i n v e r s e of p/q.
To solve the quadratic equation x2 — 2 = 0 and other general polynomial
equations with coefficients in Q, the set of rational numbers is extended
by creating and including the algebraic irrational numbers, that is, the
numbers such as \/2 or v^4, that can be roots of polynomial equations with
coefficients in Q. The result is the algebraic closure of Q.
It turns out there are other irrational real numbers that are not alge-
braic, that is, are not roots of any polynomial with rational coefficients.
These irrational numbers are called t r a n s c e n d e n t a l . To put it differently,
the ordered set of algebraic numbers contains many holes, and these holes
are filled with transcendental numbers. It was only in 1844 that the French
mathematician J O S E P H LIOUVILLE (1809-1882) showed the existence of
such numbers by explicitly constructing a few. Later, it was proved that
the two familiar numbers, n and e, are also transcendental. Together, the
algebraic and transcendental numbers make up the r e a l numbers. The
rigorous definition of a real number is necessary to put calculus on a sound
basis.
Cantor defined a real number as the limit of a sequence of rational num-
bers. The operations of addition, subtraction, multiplication, and division
are introduced on the set of real numbers in an obvious way as the limits
of the corresponding sequences; the non-trivial part is the proof that the
 Basic Definitions 181

definitions do not depend on the approximating sequences. The set of real

numbers is denoted by R. Geometrically, we represent E as points on a
one-dimensional continuum, known as the real line. The construction of
Cantor ensures that the set of real numbers, also known as the real line,
does not contain any holes.
Finally, to solve equation x2 + 1 = 0 and other similar polynomial
equations with real coefficients but without real solutions, the set of real
numbers is extended by adjoining to M an element i (also denoted by j
in some engineering books) such that i2 = - 1 . This element is called
the imaginary u n i t . The resulting extension is denoted by C and is the
collection of the expressions x + iy, where x, y € M.. The elements of C
are called complex numbers. As far as solving polynomial equations, we
are all set now: the Fundamental Theorem of Algebra states that every
polynomial with coefficients in C has at least one root in C; in other words,
the set C is algebraically closed.
If z = x + iy G C, and i , i / £ l , then x, denoted by 3?z, is called t h e
r e a l p a r t of z. Also, y = 9 z is called t h e imaginary p a r t of z, and
~z = a — hi, the complex conjugate of z. By definition,

(a + bi) + (c + di) = (a + c) + (b + d)i, (a + bi)(c+di) = (ac — bd) + (ad + bc)i,

where the multiplication rule follows naturally from the distributive law
and the equality i2 = —1.
EXERCISE 4.1.l. c (a) Verify that "Siz = (z + z)/2, Sz = (z - z)/(2i).
(b) Verify that the complex conjugate of the sum, difference, product, and
ratio of two complex numbers is equal to the sum, difference, product, and
ratio, respectively, of the corresponding complex conjugates: for example,
z\Z2 = z\Z2- (c) Verify that if P = P(z) is a polynomial with real coeffi-
cients, then P(z) = P(z), and therefore the non-real roots of this polynomial
come in complex conjugate pairs, (d) Conclude that a polynomial of odd
degree and with real coefficients has at least one real root.
Complex numbers appear naturally in computations involving square
roots of negative numbers. Some records indicate that such computa-
tions can be traced to the Greek mathematician and inventor HERON OF
ALEXANDRIA (c.10 - c.70 AD). In 1545, the Italian mathematician G E R O -
LAMO CARDANO (1501-1576) published the general solutions to the cubic
(degree three) and quartic (degree four) equations. The publication boosted
the interest in the complex numbers, because the corresponding formulas
required manipulations with square roots of negative numbers, even when
182 Algebra of Complex Numbers

the final result was a real number; for one example of this kind, see Exer-
cise 4.1.5 on page 185 below. It still took some time to get used to the new
concept, and to develop the corresponding theory. In fact, the term "imag-
inary" in connection with the complex numbers was introduced around
1630 by Descartes, who intended the term to be derogatory. In 1777, Euler
suggested the symbol i for y/— 1, and the complex numbers started to get
the respect they deserve. The foundations of modern complex analysis were
laid during the first half of the 19th century. By the end of the 19th cen-
tury, it was already impossible to imagine mathematics without complex
numbers. Part of the reason could be that, as the French mathematician
JACQUES SALOMON HADAMARD (1865-1963) put it, the shortest path be-
tween two truths in the real domain passes through the complex domain.
In the following sections, we will see plenty of examples illustrating this
statement.
Since solving polynomial equations was the main motivation for the
introduction of complex numbers, let us say a bit more about these equa-
tions. Given a polynomial, the objective is to find an algebraic formula
for the roots, that is, an expression involving a finite number of additions,
subtractions, multiplications, divisions, and root extractions, performed on
the coefficients of the polynomial. The formula x = (—b± \/b2 — 4ac)/(2a)
for the roots of the quadratic equation ax2 + bx + c = 0 was apparently
known to ancient Babylonians some 4000 years ago. The formulas of Car-
dano for equations of degree three and four are much more complicated but
still algebraic. Ever since the discovery of those formulas, various mathe-
maticians tried to extend the results to equations of degree five or higher,
until, around 1820, the Norwegian mathematician NIELS HENRIK ABEL
(1802-1829) proved the non-existence of such algebraic representations for
the solutions of a general fifth-degree equation. In 1829, the French math-
ematician EVARISTE GALOIS (1811-1832) resolved the issue completely by
proving the non-existence of an algebraic formula for the solution of a gen-
eral polynomial equation of degree five or higher, and also describing all
the equations for which such a formula does exist. Note that both Abel
and Galois were under 20 years of age when they made their discoveries.
The solutions of a polynomial equation can exist even without an alge-
braic formula to compute them; the fundamental theorem of algebra ensures
that a polynomial of degree n and with coefficients in C has exactly n roots
in C, and there are many ways to represent the roots using infinitely many
operations of addition, subtraction, multiplication, and division, performed
on the coefficients. Such representations lead to various numerical methods
 The Complex Plane 183

of solving the polynomial equations, but these topics fall outside the scope
of our discussions. For more on the history of complex numbers, see the
book An Imaginary Tale: The Story of \/—I by P. J. Nahin, 1998.

4.1.2 The Complex Plane

The field C can be represented geometrically as a set of points in the plane,
with the real part along the horizontal axis, and the imaginary part along
the vertical axis. This representation identifies a complex number z = x+iy
with either the point (x,y) or the vector r — xi + yj in the Euclidean
space R 2 ; see Figure 4.1.1. The upper (lower) half-plane contains complex
numbers with positive (negative) imaginary part. Similarly, the r i g h t or
l e f t half-plane refers to complex numbers with positive or negative real
parts, respectively.

z — x + iy xi + yj

Fig. 4.1.1 Complex plane and K 2

In polar coordinates,

z = r(cos# + ism0), (4.1.1)

where r = y/x2 + y2 = \z\ is the modulus or a b s o l u t e value of z, and

9 = arg(z) is the argument of z. Similar to the polar angle at the origin,
the argument is not defined for z = 0.
Note that if 9 is the argument of z, so is 9 + 2nk for every integer k.
Accordingly, the p r i n c i p a l value of t h e argument Arg(z) is defined as
the value of argz in the interval (—n, w].
EXERCISE 4.1.2? Verify the following formula for the principal value of the
argument:
1
tan (y/x), x>0,y^0;
1
Arg(z) = - 7T + ta,n~ (y/x), x<0,y>0; (4.1.2)
1
—TT + t&n~ (y/x) x < 0, y < 0.
184 Algebra of Complex Numbers

EXERCISE 4 . 1 . 3 . C
Using the suitable trigonometric identities, verify that if
z\ = n (cos #1 + i sin 6\) and z2 = r2(cos 82 4- i sin 62), then

z\Z2 = rir2(cos(0j + 02) + ism(9i + 02)),

^ = rM cos(0! - B2) + i sin(0i - 92)) • (4 L3)
'
z2 r2\ >

Some sources call the left picture in Figure 4.1.1 the Argand diagram,
in honor of the Swiss-born non-professional mathematician J E A N - R O B E R T
ARGAND (1768-1822), who published the idea of the geometric interpre-
tation of complex numbers in 1806, while managing a bookstore in Paris.
Mathematical formulas are neither copyrightable nor patentable, and the
names of those formulas are often assigned in an unpredictable way. Ar-
gand's diagram is one such example: in 1685, when complex number were
much less popular, a similar idea appeared in a book by the English math-
ematician JOHN WALLIS (1616-1703).
The following result is known as E u l e r ' s formula:

cos6 + isin0 = ei9. (4.1.4)

In this case, the name is true to the fact: the result was first published
by L. Euler in 1748. At this point, we will take (4.1.4) for granted and
only mention that (4.1.4) is consistent with (4.1.3); later on, we will prove
(4.1.4) using power series. A particular case of (4.1.4), em + 1 = 0, collects
the five most important numbers in mathematics, 0,1, e, n, i, in one simple
equality.
An immediate consequence of (4.1.4) is that multiplication of a complex
number z by el$ is equivalent to a rotation of z in the complex plane by 0
radians counterclockwise. Note also that e%e = el(s+2n). As a result, a time-
varying periodic quantity A is conveniently represented as A(t) = AoeluJt,
where AQ is the amplitude, and u is the angular frequency (so that 27r/w
is the period). Below, we will use such representations in the analysis of
electrical circuits and planar electromagnetic waves.
EXERCISE 4.1.4? Let z\ = x\ +iyi, 22 = ^2 +iy2 be two complex numbers,
and ri = xii + yij, r2 = X2 i + 2/2 J, the corresponding vectors, (a) Verify
that Z1Z2 = (7*1 • rg) + i((ri x r 2 ) • k), where, as usual, k = i x j . (b)
Express the angle between the vectors r\ and r2 in terms of Arg(z\) and
Arg[z,2)- Hint: draw a picture; the angle is always between 0 and n.
One application of (4.1.4) is computing r o o t s of complex numbers.
 The Complex Plane 185

Writing

z = |z|e(ArSW+2-fc)\

we find

z lM = | z |l/m c (ArgW/m+2 W fc/m)i ) fc = 0 , . . . , m - l . (4.1.5)

T h u s , the m - t h roots of a complex number have exactly m distinct values.

On the complex plane, these values are at t h e vertices of a regular m-gon.
F O R E X A M P L E , taking z = 1 and m = 3 so t h a t Arg(z) = 0, we find t h e
three values of \/l: 1, ( - 1 + iy/3)/2, ( - 1 - iy/Z)/2.

E X E R C I S E 4.1.5. (a)c Find all the values of \/&i and \/64i and draw them
in the complex plane. (b)A Find all the values of y 1 + iy/3 + y 1 — iy/i.
Hint: one of them is \/6-' simply square the expression.

Another application of (4.1.4) is deriving certain trigonometric identi-

ties. By taking the n - t h power on b o t h sides, we get de M o i v r e ' s f o r m u l a :

(cos# + ism8)n = cosn8 + isinn8; (4.1.6)

of course, t h e French m a t h e m a t i c i a n A B R A H A M D E M O I V R E (1667-1754),

who published (4.1.6) in 1722, did n o t use (4.1.4) in his derivations.
By equating the real and imaginary p a r t s of (4.1.6), we get the expres-
sions for cos nQ and sin nO in terms of the products of sin 6 and cos 6. F O R
E X A M P L E , n = 2 yields the familiar results: cos 28 = cos28 — sin 2 8, sin 28 =
2 sin 6 cos 8.

E X E R C I S E 4.1.6. B Find similar expressions for cos3$ and sin 3^.

Yet another application of (4.1.4) is illustrated by the following exercise.

E X E R C I S E 4 . 1 . 7 . c Evaluate the indefinite integral f e ~ 3 x cos 2x dx without

integration by parts. Hint: note that the integrand is di(e<-~3+2^x). Integrate
the complex exponential as if it were real, and them compute the real part of the
result.

We conclude this section with a few definitions related t o sets in t h e

complex plane; these definitions a r e identical t o those on page 121. As
with ordinary points in the plane, \z\ — z<z\ is t h e distance between z\ a n d
Z2, and the set {z : \z—ZQ\ < r} is an open d i s k with center a t z0 and radius
r > 0; similarly, {z :\z — ZQ\ < r} is a c l o s e d d i s k . A n e i g h b o r h o o d of a
point z is a n open disk centered a t z. A point in a set is called i n t e r i o r if
186 Algebra of Complex Numbers

there exists a neighborhood of the point that lies entirely in the set. F O R
EXAMPLE, the center of the disk (open or closed) is an interior point of the
disk.
A point P is called a boundary point of a set if every neighborhood of
the point contains at least one point that is not in the set, and at least one
point that belongs to the set and is different from P. FOR EXAMPLE, the
boundary points of the set {z : [^r — 2r0| < r} are exactly the points of the
circle {z :\z — ZQ\ = r}.
A point P is called an i s o l a t e d point of a set if there exists a neighbor-
hood of P in which P is the only point belonging to the set. FOR EXAMPLE,
the set {z : z = a + ib} in the complex plane, where a, b are real integers,
consists entirely of isolated points.
A set is called

• Bounded, if it lies entirely inside an open disk of sufficiently large radius.

• Closed, if it contains all its boundary points.
• Connected, if every two points in the set can be connected with a
continuous curve lying completely in the set.
• Open, if every point belongs to the set together with some neighborhood.
In other words, all points of an open set are interior points.
• Domain, if it is open and connected.
• Simply connected, if it has no holes. More precisely, consider a simple,
closed, continuous curve that lies entirely in the set (see page 25); such
a curve encloses a domain (recall that we take for granted the Jordan
curve theorem, see page 123). The set is simply connected if this domain
lies entirely in the set.

The c l o s u r e of a s e t is the set together with all its boundary points.

The complement of a set are all the points that are not in the set.
FOR EXAMPLE, the closure of an open disk {z : \z\ < 1} is the closed disk
{z : \z\ < 1}, and the complement of that open disk is the set {z : \z\ > 1}.
Every disk, open or closed, is both connected and simply connected, while
the set {z : 0 < \z\ < 1} is open, connected, but not simply connected
because the point z = 0 is missing from the set.
EXERCISE 4.1.8. c Give an example of a set in C that is not bounded, is
neither open nor closed, is not connected, but is simply connected.
 AC Circuits 187

4.1.3 Applications to Analysis of AC Circuits

In this section, we continue to use i as the notation for the imaginary unit.
The basic components of an alternating current (AC) circuit are the resistor
R, the capacitor C, and the inductor L; see Figure 4.1.2. The main facts
about electric circuits are summarized in Section 8.4 in Appendix. If A is
a quantity, such as current or voltage, changing periodically in time with
period T = l/v, then we represent this quantity as A(t) = Aoe%^tJr^a\
where 4>o is the initial phase, and u> is the underlying angular frequency,
related to the usual frequency v by w = 2nv (a household outlet in the US
has v = 60 Hz or 60 cycles per second; in Europe, the standard is / = 50
Hz). Since taking the real part of A(t) brings us back to physical reality and
can be done at any moment, we will work only with complex currents and
voltages.
Denote by Iy and Vy the current through and the voltage across the
element Y, respectively, with Y being a resistor R, a capacitor C, or an
inductor L. We assume that all the elements are linear, so that
• by Ohm's Law, see page 173, IR — VR/R;
• by the definition of the capacitance, C = qc/Vc, where qc{t) =
JQ Ic(s)ds is the charge; thus, Ic = CdVc/dt;
• by Faraday's Law, see page 165, VL = Ldli,{t)/dt.
EXERCISE 4.1.9. c Taking I(t) - I0ei"t+i4'0, verify that (a) The current
through the resistor is in phase with the voltage; (b) The current though
the capacitor is ahead of, or leads, the voltage by the phase n/2; (c) The
current through the inductor is behind, or lags, the voltage by the phase
7T/2. Hint: i = e i 7 r / 2 .

Consider the series circuit on the left-hand side of Figure 4.1.2, with
E(t) = Eoei<-UJt+'t'0K All the elements of the circuit have the same cur-
rent I(t) passing through them; we take I(t) = Joe""'*. Then we have
VR(£) = I{t)R (the voltage across R is in phase with the current),
Vc(t) = (l/wC)I(t)e~in/2 (the voltage across C is behind the current by
n 2
7r/2), and Vi(t) = Lu>I(t)e / (the voltage across L is ahead of the cur-
rent by 7r/2). The vector diagram corresponds to time t = 0; for t > 0
the diagram rotates counterclockwise with angular speed u> = 2-KV; lin-
ear frequency v = 60 Hz, corresponds to 60 full turns per second. Since
188 Algebra of Complex Numbers

"VL
Ic
E = VR + Vc + VL
IR E
•+J •— •+T *—
VR I
I = IR + IC + IL
Vc
II

Series Parallel

Fig. 4.1.2 RCL Circuits

VR + VC + VL = E, we have E(t) = (R+l/(iu}C) + iuL)I(t), and therefore

E0
h -., tan^>o = ^ - . (4.1.7)
^/R2 + ^L2__i_7y

EXERCISE 4.1.10. B (a) Verify (4-1.7). Hint: Write (R+l/(iwC) + iu>L) in

the complex exponential form; keep in mind that \/i = —i. (b) Verify that, for
every input E(t), the current I(t) in the series circuit satisfies

LI"(t) + RI'{t) + I(t)/C = E'(t). (4.1.8)

Substitute E(t) = E0ei{"t+'t'o) and I(t) = I0eiuJt to recover (4.1.7). Hint: if

q is the charge, then E = EL + ER + EC = Lq" + Rq' + q/C; I = q'.
For fixed Eo, the largest value of IQ = EQ/R is achieved at the reso-
nance frequency U>Q — \/\/LC; the corresponding phase shift <po at this res-
onance frequency is zero. According to (4.1.8), a series circuit is a damped
harmonic o s c i l l a t o r , with damping proportional to R. The "ideal" series
circuit with R = 0 is a pure harmonic oscillator and has infinite current at
 AC Circuits 189

the resonance frequency wo = l/VLC.

Now let us consider the parallel RCL circuit on the right-hand side of
Figure 4.1.2. This time, the voltage E(t) is the same across all compo-
nents; we take E(t) = EoeluJt. For the currents, we have / R ( £ ) = E(t)/R
(in phase with E), Ic{t) = E(t)uiCein^2 (ahead of E by TT/2), IL(t) =
(E(t)/(wL))e~™/2 (behind E by 7r/2). The vector diagram corresponds to
t = 0; for t > 0, the diagram rotates counterclockwise with angular speed
w. The total current is I = IR + Ic + IL = (1/-R + iwC + l/{iuL))E{t).
Taking J(t) = / 0 e i ( a " + * o ) , we conclude that

1 ( „ 1 N2
R2 + \wC- — ) E0, tan^o = fl(wC - l/(wL)). (4.1.9)

EXERCISE 4.1.11. 5 (a,) Ven/i/ (4.1.9). (b) Verify that, for every input
voltage E{t), the current I(t) in the parallel circuit satisfies

CE"(t) + E'(t)/R + E(t)/L = /'(*). (4.1.10)

Substitute E(t) = £ 0 e i M ) and I{t) = I^^+M to recover (4.1.9).

Unlike the series circuit, the resonance frequency a>o = l/y/LC now cor-
responds to the smallest absolute value IQ = EQ/R of the current for given
EQ\ the corresponding phase shift <j>o at the resonance frequency is zero. Ac-
cording to (4.1.10), a parallel circuit is a damped harmonic o s c i l l a t o r
with damping proportional to 1/R. The "ideal" parallel circuit with R = oo
is a pure harmonic oscillator and has zero total current at the resonance
frequency OJQ = 1/y/LC.
The above analysis also demonstrates that the effective resistance,
called reactance, of the capacitor and the inductor is, respectively,
Xc = l/(iuiC) and XL = iu>L. The usual laws for series or par-
allel connection apply: the total effective resistance Z, known as the
complex impedance, satisfies Z = R + Xc + X^ in the series circuit and
1/Z = (1/R) + (l/Xc) + (l/XL) in the parallel circuit.
EXERCISE 4.1.12.3 Sketch the graph of \Z\ as a function of w for
(i) the series circuit; (ii) the parallel circuit.
190 Functions of a Complex Variable

4.2 Functions of a C o m p l e x Variable

4.2.1 Continuity and Differentiability

The study of functions of a complex variable is called complex a n a l y s i s .
As in the usual calculus, we define a complex function, that is, function
/ = f(z) of a complex variable z as a rule that assigns to every complex
number z at most one complex number f(z). This definition is especially
important to keep in mind: rules that assign several values to the same com-
plex number z also appear in complex analysis and are called m u l t i - v a l u e d
functions.
A polynomial function p — p(x) = X3fc=o akxk °f a r e a l variable x
easily extends to the complex plane by replacing a; with z and allowing the
coefficients afc to be complex numbers. Similarly, a r a t i o n a l function,
that is, a ratio of two polynomial functions, extends to the complex plane
except at the points where the denominator vanishes. For other functions of
the real variable, the replacement of x with z is not as straightforward, and
a special theory of the functions of a complex variable must be developed.
Definition 4.1 A function / = f(z) is continuous at ZQ if / is defined
in some neighborhood of ZQ and

lim f(z0 + z) = f(z0).

\z\—>0

A function / = f(z) is d i f f e r e n t i a b l e at ZQ if / is defined in some neigh-

borhood of ZQ and there exists a complex number, denoted by f'(zo), so
that

lim /(«> + * ) - / ( « > ) = f{zo).

EXERCISE 4 . 2 . 1 . C Verify that a function differentiable at ZQ is continuous

at ZQ.

EXERCISE 4.2.2? Which of the following functions are differentiable at zero:

f(z) = z\f{z) = Z(z)J(z) = \z\J(z) = \z\2?
EXERCISE 4.2.3. c Verify that if f(z) = zk for a positive integer k, then
f'{z) = kzk-1.
Definition 4.2 A function is called a n a l y t i c a t a p o i n t if it is differ-
entiable in some neighborhood of the point. A function is called a n a l y t i c
i n a domain if it is analytic at every point of the domain. In general, we
 Cauchy-Riemann Equations 191

say that a function is a n a l y t i c if it is analytic somewhere. An e n t i r e

function is a function that is analytic everywhere in the complex plane.
Remark 4.1 Sometimes, the word holomorphic is used instead of ana-
lytic.

EXERCISE 4.2.4? (a) Convince yourself that a function is analytic at a point

if and only if the function is analytic in some neighborhood of the point.
Hint: a neighborhood of a point is an open set; see page 186. (b) Verify that
a polynomial is an entire function, (c) Verify that a rational function
f(z) = P(z)/Q(z), where P,Q are polynomials, is analytic everywhere ex-
cept at the roots ofQ(z).

4.2.2 Cauchy-Riemann Equations

Let / = f(z), z = x + iy, be a function of a complex variable. Being a
complex number itself, f(z) can be written as

f(z)=u(x,y) + iv(x,y) (4.2.1)

for some real functions u, v of two real variables x, y. F O R EXAMPLE, if

f(z) = z2, then/(,z) = (x+iy)2 = (x2~y2) + 2ixy, so that u(x,y) = x2-y2
and v(x, y) = 2xy.
EXERCISE 4.2.5.
C
Find the functions u,v if f(z) = z3.
The objective of this section is to investigate the connection between
differentiability of the complex function / and differentiability of the func-
tions u, v. The motivation for this investigation comes from the following
exercise, showing that differentiability of a function of a complex variable
requires more than mere differentiability of the real and imaginary parts.
EXERCISE 4.2.6. C (a) Show that the function f(z) = u(x,y) + iv(x,y) is
continuous at the point ZQ = XQ + iyo, in the sense of Definition 4-1, if and
only if both functions u,v are continuous at (xo,yo), as real functions of
two variables, (b) Show that if the function f = f(z) is differentiable, then
the corresponding functions u, v are also differentiable, as functions of x
and y. (c) Show that the function f{z) — x is not differentiable anywhere,
in the sense of Definition 4-1-
To understand what is going on, let us assume that / = f(z) is dif-
ferentiable at ZQ SO that, according to Definition 4.1, we have f(zo +
z) - /(*>) = */'(*>) + e(z), where \e(z)\/\z\ - • 0, \z\ - • 0. Writing
192 Functions of a Complex Variable

e(z) = E\(x,y) + ie2(x,y), f'(zo) = A + iB, r = \z\, we find that, for

all x, y sufficiently close to zero,
(u(x0 + x,y0 + y) -u(x0,yo)) + i(v(x0 + x,y0 + y) -v(x0,y0))
= {x + iy)(A + IB) + ei(x,y) + ie2(x,y),
where \ek(x,y)/r\ —> 0, r —* 0, for k = 1,2.
EXERCISE 4.2.7. c By comparing the real and imaginary parts in the
last equality, convince yourself that the functions u, v are difjerentiahle at
{xo,yo) and the following equalities hold:

A = —{x0,y0) = —(x0,y0), B = - — (x0,y0) = ~(x0,y0).

We now state the main result of this section.

Theorem 4.2.1 A function f = f(z) is difjerentiahle at the point ZQ =
^o + iya if and only if the functions u = 3?/, v = S / are difjerentiahle at
the point (xo,yo) and the equalities

dx dy' dy dx
hold at the point (xo,yo).
Equalities (4.2.2) are known as the Cauchy-Riemann equations. Riemann
introduce many key concepts of complex analysis in 1851 in his Ph.D. dis-
sertation; his advisor was Gauss. And, as we mentioned earlier in connec-
tion with the Cauchy-Bunyakovky-Schwartz inequality, one should not be
surprised to see the name of Cauchy attached to an important result.
EXERCISE 4.2.8. (a)B Prove the above theorem. Hint: You already have
the proof in one direction, and you reverse the arguments to get the other direc-
tion. (b)c Verify that the derivative f'(z) can be written in one of the four
equivalent ways:
,. . _ du .dv _ dv .du du .du dv .dv
dx dx dy dy dx dy dy dx'
(c)B Verify that if the function f is analytic in the domain G, then each
of the following implies that f is constant in G: (i) f'(z) = 0 in G; (ii)
Either 3?/ or 9 / is constant in G; (Hi) \f(z)\ is constant in G.
We will see later that if a function / = f(z) is differentiable (analytic)
in a domain, then it is infinitely differentiable there. Then the functions u
 Cauchy-Riemann Equations 193

and v have continuous partial derivatives of every order. Recall that the
alternative notation for the partial derivative du/dx is ux, and when the
second-order partial derivatives are continuous, we have uxy = uyx. Then
from (4.2.2) we find uxx = vyx and
Similarly, vxx + vyy = 0, that is, both u and v are harmonic functions.
Two harmonic functions that satisfy (4.2.2) are called conjugate. We
conclude that a function f = f(z) is analytic in a domain G if and only if
the real and imaginary parts of f are conjugate harmonic functions. This
remarkable connection between analytic and harmonic functions leads to
numerous applications of complex analysis in problems such as the study
of two-dimensional electrostatic fields and two-dimensional flows of heat
and fluids. We discuss mathematical foundations of these applications in
Problem 5.3 on page 433. The lack of a three-dimensional analog of complex
numbers is one of the reasons why the corresponding problems in three
dimensions are much more difficult.
If u and v are conjugate harmonic functions, then, because of (4.2.2),
one of the functions uniquely determines the other up to an additive con-
stant. Similarly, either the real part u or the imaginary part v specify
the corresponding analytic function / = u + iv uniquely up to an addi-
tive constant. F O R EXAMPLE, if u(x,y) = xy, then, by the first equal-
ity in (4.2.2), ux = y = vy, so that v(x,y) = y2/2 + h(x) and hence
vx = h'(x). By the second equality in (4.2.2), vx = —uy = —x, that is,
h'{x) = —x and v(x, y) = (y2 — x2)/2 + c, where c is a real number. Note
that vxx + vyy = —1 + 1 = 0, as it should; this is a useful computation to
ensure that the computations leading to the formula for v are correct. With
z = x + iy, the corresponding function f(z) is recovered from the equality

f(z) = u((z + z)/2, (z - z)/2i) + iv({z + 2)/2, (z - z)/2i),

which in this case results in f(z) = —iz2/2 + ic, where c is a real number.
This answer is easy to check: f(z) = —iz2/2 + ic = xy + i(y2 — x2)/2 =
u(x,y) + iv(x,y).
EXERCISE 4.2.9F Find all functions f = f(z) that are differentiable for all
z and have 9 / ( z ) = x2 — xy — y2. Check that your answer is correct.
If we write z = r(cos6 + isin#) in polar coordinates, then f(z) =
u(r,9) +iv(r,6).
EXERCISE 4.2.10. B (a) Verify that in polar coordinates equations (4-2.2)
194 Functions of a Complex Variable

imply

ur = r~1ve, vr = -r~1ue, (4.2.3)

and conversely, (4-2.3) imply (4-2.2). (b) Verify that each of the following
functions satisfies (4-2.3): (i) f(z) = raelaS, a a real number; (ii) f{z) =
lnr + i6
The usual rules of differentiation (for the sum, difference, product, ratio,
and composition (chain rule) of two functions, and for the inverse of a
function) hold for the functions of complex variable just as for the functions
of real variable. If the function / = f(x) is differentiable, the chances
are good that the corresponding function / = f(z) is analytic, and to
compute the derivative of f(z), you treat z the same way as you would
treat x. The functions f{z) that contain z, $tz, $Sz, arg(z), and \z\ require
special attention because they do not have clear analogs in the real domain.
Analyticity of such functions must be studied using the Cauchy-Rieman
equations. F O R EXAMPLE, the function f(z) = z is not analytic anywhere,
because for this function ux = 1, vy = — 1, and so ux ^ vy.

EXERCISE 4.2.11. c Check whether the following functions f are analytic.

If the function is analytic, find the derivative, (a) f(z) = (z)2, (b) f(z) =
3z/3fcz, (c) f{z) = 8fcz3 - iSz3, (d) f(z) = z2/(l - z2).

4.2.3 The Integral Theorem and Formula of Cauchy

In this section, we study integrals of analytic functions and establish two
results from which many of the properties of the analytic functions follow.
We start with integration in the complex plain. Consider a curve C in
R 2 denned by the vector-valued function r(t) = x(t)i + y(t)j, a < t <
b. Equivalently, we can define this curve using a complex-valued function
z = z(t), t £ [a, b], by setting z(t) ~ x(t) + iy(t). As usual, we write
z(t) = x(t) + iy(t), provided the derivatives of x and y exist.
Assume that the curve C is piece-wise smooth, that is, consists of finitely
many smooth pieces; see page 28 for details. Let / = f(z) be a function,
continuous in some domain containing the curve C. We define the integral
Jc f(z)dz of / along C by

/ f(z)dz = f f(z(t))z(t)dt. (4.2.4)

JC Ja
In what follows, we consider only piece-wise smooth curves. The line inte-
 The Theorem and Formula of Cauchy 195

gral over a closed curve C (that is, a curve for which r(a) = r(b)) is often
denoted by §c . The letter z is not the only possible notation for the variable
of integration; in particular, we will often use the Greek letter ( when z is
being used for other purposes.
EXERCISE 4 . 2 . 1 2 . C (a) Writing f(z) = u(x,y) + iv(x,y), verify that

/ f(z)dz = / (udx - vdy) + i / (vdx + udy). (4.2.5)

In particular, by (3.1.17) on page 131, change of orientation of the curve

C reverses the sign of fc f(z)dz. (b) Verify that the length of the curve can
be written as fc \dz\.
EXERCISE 4.2.13? Let f(z) = (z — z0)n, where ZQ is a fixed complex number
and n, an integer: n = 0, ± 1 , ± 2 , . . . . Let C be a circle with radius p,
center at z§, and orientation counterclockwise. Verify that §c f(z)dz = 2-ni
if n = — 1 and fc f(z)dz = 0 otherwise. Hint: z(t) = zo + peu, 0 < t < 2TT.
In other words, you show in Exercise 4.2.13 that

U h
-dz ={—' (4.2.6)
Jfz-z0\=p (z ~ zo) n
[0, n = 0 , - 1 , ±2, ± 3 , . . . .
Recall that, for a real-valued continuous function h = h(x), we have
| fa h(x)dx\ < (b — a) maxx£[a]f,] |/i(:r)|, and the easiest way to prove this
inequality is to apply the triangle inequality to the approximation of the
integral according to the rectangular rule and then pass to the limit. We
will now derive a similar inequality for the complex integral (4.2.4).
By the left-point rectangular rule, we have
b "-1

/ f(z(t))z{t)dt= lim y]/(z(tfc))i(*fc)(tfc+i-tfc), (4.2.7)

Ja fc=0

where a = to < h < ... < tn — b. Note that |i(£fc)| = ll^(*fc)|| where
r = r(t) is the vector function that defines the curve, and we saw on page
29 that
n-l -fe

lim y)||r(tfc)||(tfc+i-tfe)= / ||r(t)||dt = L c (a,6),

maxltfc+i-tfcHOj^ Ja

the length of the curve C. We then apply the triangle inequality on the
right-hand side of (4.2.7) and denote by max z e c \f{z)\ the maximal value
196 Functions of a Complex Variable

of |/(.z)| on the curve C. The result is the following inequality for the
integral (4.2.4):

L f(z)dz < Lc(a,b) max\f(z)\. (4.2.8)

Note that max z e c \f(z)\ = maxa<t<b \f(z(t))\; since the functions / = f(z)
and z = z(t) are both continuous and the interval [a, b] is closed and
bounded, a theorem from the one-variable calculus ensures that the maxi-
mal value indeed exists. If you remember that the curve C is only piece-wise
smooth, and insist on complete rigor, apply the above argument to each
smooth piece of the curve separately and then add the results.
We will now look more closely at the line integrals along closed curves.
Recall (page 25) that a curve C, defined by a vector-valued function r =
r(t), a < t < b, is called simple closed if the equality r(ti) = rfa) holds for
h = a, i 2 = b and for no other £1,^2 € [a, b\. By default, the orientation of
such a curve is counterclockwise: as you walk along the curve, the domain
enclosed by the curve stays on your left.

EXERCISE 4.2.14? Let f be a function, analytic in a domain G, and letC be

a simple, closed, piece-wise smooth curve in G so that the domain enclosed
by C lies entirely in G. Assuming that the derivative / ' of f is continuous
in G, show that §c f(z)dz = 0. Hint: use Green's formula to evaluate the line
integrals in (4-2.5), then use the Cauchy-Riemann equations (4-2.2).
As with line integrals of real-valued vector functions, we say that the
function / = f(z) has the path independence property in a domain G
of the complex plane if / is continuous in G and §c f(z)dz = 0 for every
simple closed curve C in G (recall that we always assume that C is piece-wise
smooth).
EXERCISE 4.2.15. c Show that if the function f has the path independence
property in G and C\, C2 are two curves in G with a common starting point
and with a common ending point, then L f(z)dz — Jc f(z)dz. Hint: make
a closed curve by combining C\ and C2.
EXERCISE 4 . 2 . 1 6 . C
Verify that the function f(z) = 1 has the path indepen-
dence property in every domain, and therefore Jc,z z ,dz = Z2 — z\, where
C(.z\iz2) is a curve that starts at z\ and ends at Z2- Hint: use (4-2.5) and
the result about path independence from vector analysis.
We will show next that a continuous function with the path indepen-
 The Theorem and Formula of Cauchy 197

dence property in a simply connected domain has an anti- derivative there.

Theorem 4.2.2 Assume that a continuous function f = f(z) has the

path independence property in a domain G, and assume that the domain G
is simply connected (see page 186). Then there exists an analytic function
F = F(z) such that F'(z) = f(z) in G.

Proof. Define the function F(z) = Jc,z z* /(()d£, where C(zo, z) is a curve
in G, starting at a fixed point ZQ £ G and ending at z £ G. By Exercise
4.2.15 this function is well defined, because the integral depends only on
the points ZQ,Z and not on the particular curve. Define the number A =
A(z, Az) by

A_Fi,+££-m_m (4.2.9)
We need to show that limA2-+o A = 0 for every z £ G, which is equivalent
to proving that F'(z) = f(z). The result will also imply that F is analytic
inG.
Let C = C(z, z + Az) be any curve that starts at z and ends at z + Az.
Using the result of Exercise 4.2.16, we have f(z) = f(z)(l/Az) fcd£ =
(1/Az) fc f(z)dC, because f(z) is constant on C. Therefore,

z
^ JC(z,z+Az)

and, by (4.2.8), \A\ < (Lc/\Az\)max(£c |/(C) - f(z)\- Since we are free to
choose the curve C(z, z + Az), let it be the line segment from the point z
to the point z + Az. Then Lc = \Az\. Since / is continuous at z, for every
e > 0, there exists a 6 > 0 so that \Az\ < 6 implies |/(C) — f(z)\ < e for all
C, £ C. As a result, for those Az, \A\ < e, which completes the proof. •

We will now state the first main result of this section, the Integral Theo-
rem of Cauchy, which says that the functions having the path independence
property in a simply connected domain are exactly the analytic functions.
Theorem 4.2.3 (The Integral Theorem of Cauchy) A continuous
function f has the path independence property in a simply connected domain
G if and only if f is analytic in G.
In Exercise 4.2.14, you already proved this theorem in one direction
(analyticity implies path independence) under an additional assumption
that / ' is continuous; this is exactly what Cauchy did in 1825. This is not
198 Functions of a Complex Variable

the best proof, because the definition of an analytic function requires only
the existence of the derivative.
In 1900, the French mathematician EDOUARD GOURSAT (1858-1936)
produced a proof that does not rely on the continuity of / ' . He published
the result in 1900 in the very first issue of the Transactions of the American
Mathematical Society (Volume 1, No. 1, pages 14-16; the paper is in
French, though). Goursat's proof is too technical to discuss here and is
more suitable for a graduate-level course in complex analysis.
The converse statement (path independence implies analyticity) was
proved by the Italian mathematician GIACINTO MORERA (1856-1909). We
almost have the proof of this result too: with Theorem 4.2.2 at our disposal,
all we need is the infinite differentiability of analytic functions, which we
will establish later in this section.

Remark 4.2 Assume that the function f is continuous in the closure

of G (that is, f is defined on the boundary of G and, for every ZQ on the
boundary of G, f{z) —> f(zo) as z —> ZQ, Z 6 G.) If f is analytic in G
and G is simply connected and bounded with a piece-wise smooth boundary
Co, then §c f{z)dz = 0. Indeed, continuity of f implies that, for every
e > 0, there exists a simple, closed, piece-wise smooth path Ce lying inside
of G such that | §c f(z)dz — §c f{z)dz\ < s. By the Integral Theorem of
Cauchy, §c f{z)dz — 0.

The Integral Theorem of Cauchy says nothing about domains that are
not simply connected; Exercise 4.2.13, in which G is the disk with the center
removed, shows that anything can happen in those domains. If the holes
in G are sufficiently nice, then we can be more specific.

Theorem 4.2.4 Assume that the boundary of G consists ofn+1 closed,

simple, piece-wise smooth curves CQ, ... ,Cn so that C\,... ,Cn do not have
points in common and are all inside the domain enclosed by CQ. Assume
that f is a function analytic in G and is continuous in the closure of G.
Then

I f(z)dz + J2(£ f(z)dz = 0, (4.2.10)

J Co k=1 JCk

where the curve CQ is oriented counterclockwise, and all other curves, clock-
wise, so that, if you walk in the direction of the orientation, the domain G
is always on the left.
 The Theorem and Formula of Cauchy 199

Proof. Remember that the change of the orientation reverses the sign of
the line integral; see Exercise 4.2.12, page 195. Connect the curves Cfc,
k = 1 , . . . ,n to Co with smooth curves (for example, line segments), and
apply the Integral Theorem of Cauchy to the resulting simply connected
domain (draw a picture). Then note that the integrals over the connecting
curves vanish (you get two for each, with opposite signs due to opposite
orientations), and you are left with (4.2.10). •

EXERCISE 4.2.17? Use (4.2.10) to show that if f is analytic in G andCi,C2

are two simple closed curves in G with the same orientation (both clockwise
or both counterclockwise) so that Ci is in the domain enclosed by C\, then
§c f(z)dz = §c f(z)dz. Note that G does not have to be simply connected.
Hint: consider the domain G\ that lies in between C\ and C2.
We now use the "analyticity implies path independence" part of the In-
tegral Theorem of Cauchy to establish the I n t e g r a l Formula of Cauchy.

Theorem 4.2.5 (Integral Formula of Cauchy) Let f = f(z) be an

analytic function in a simply connected domain G, and C, a simple closed
curve in G, oriented counterclockwise. Then the equality

4 211
i^-hfM* <' »
holds for every point z inside the domain bounded by C.

Proof. Step 1. Fix the curve C and the point z. Since / is analytic in
G, the function /(£) = /(C)/(C — z) is analytic in the domain G with the
point z removed. Let Cp be a circle with center at z and radius p small
enough so that Cp lies completely inside the domain bounded by C. We
orient the circle counterclockwise and use the result of Exercise 4.2.17 to
conclude that §c f{QdC, = §Cp f(()d(.
Step 2. By (4.2.6), f{z) = (1/2TTJ) JC (f(z)/{C, - z))d( (keep in mind
that the variable of integration is (, whereas z is fixed). Combining this
with the result of Step 1, we find:

^iP^dC- m = -L / /«>-/(*>«. (4.2.12)

S M ; V ;
2-Ki Jc C - z 2ni JCp C-z

Step 3. Note that the left-hand side of (4.2.12) does not depend on p.
200 Functions of a Complex Variable

To complete the proof of (4.2.11), it is therefore enough to show that

lira d> /(C) - /(*)-dC = 0,

P-O 2m JCo <-z
and the argument is very similar to the proof of Theorem 4.2.2; see the
discussion following equation (4.2.9). Indeed, the continuity of / implies
that, for every e > 0, we can find p > 0 so that \f(z) — /(C)| < £ as long as
|C ~ z\ < P- Using inequality (4.2.8) and keeping in mind that LQP = 2-Kp
and |C — z\ = p when £ G Cp, we find:

/(C) ~ f(z) ,r
2m Jc — 7 "C < e, (4.2.13)

which completes the proof. •

Similar to Remark 4.2, if the domain G is bounded and has a piece-wise

smooth boundary CQ, and the function / is analytic in G and is continuous
in the closure of G, then, for every z £ G, we have

' ( "-55jL^* (4 2 14)

'-
EXERCISE 4.2.18. (a)B Fill in the details leading to inequality (4.2.13).
c
(b) What is the value of the integral on the right-hand side of (4-2.11) if
the point z does not belong to the closure of the domain enclosed by C.
Hint: 0 by the Integral Theorem of Cauchy.
EXERCISE 4.2.19. A (a) Use the Integral Formula of Cauchy to prove
the mean-value p r o p e r t y of the analytic functions: if f is analytic in
a domain G (not necessarily simply connected), ZQ is a point in G, and
{z : \z — zo\ < p} is a closed disk lying completely inside G, then

1 /"27r
/(Z ) =
° 2W /(*0+Pe")d*- ( 4 - 2 - 15 )

Hint: Use (4-2.14) in the closed disk; ( = z0 + pezt. (b) In (4-2.15, can you
replace the average over the circle with the average over the disk? Hint: yes.
An almost direct consequence of representation (4.2.11) is that an an-
alytic function is differentiable infinitely many times. More precisely, if
/ = fiz) i s analytic in a domain G (not necessarily simply connected), the
 The Theorem and Formula of Cauchy 201

n - t h order derivative / ( " ) of / exists at every point in G a n d

as long as the closed disk {z : \z - (\ < p] lies entirely in G; recall t h a t

0! = 1 and, for a positive integer k, k\ = 1 • 2 • . . . • k. Using the result of
Exercise 4.2.17, we can replace the circle of radius p in (4.2.16) with any
simple, closed, piece-wise smooth curve, as long as the domain bounded by
t h a t curve contains the point z and lies entirely in G.
An informal way to derive (4.2.16) from (4.2.11) is to differentiate
(4.2.11) n times, bring the derivative inside the integral, and observe t h a t
dn(Q - z)~l/dzn = n!(C - z)~n~l. A rigorous argument could go by induc-
tion, with b o t h the basis a n d induction step computations similar t o the
proof of (4.2.11). This proof is rather lengthy and does not introduce any-
thing new to our discussion; we leave the details to the interested reader.
Later, we will discuss an alternative rigorous derivation of (4.2.16) using
power series; see Exercise 4.3.8(c) on page 211 below.
T h e consequences of (4.2.16) are far-reaching indeed, as demonstrated
by the following results. T h e best p a r t is, you can now easily prove all of
t h e m yourself.

E X E R C I S E 4.2.20. (a)B Complete the proof of the Cauchy Integral Theo-

rem by showing that a continuous function that has the path-independence
property in a simply connected domain is analytic there. Hint: by Theorem
4-2.2, there exists an analytic function F so that F'(z) = f(z) for all z in G. By
(4-2.16), all derivatives of F are continuous in G, and you have f'{z) = F"(z).
(bf Use (4.2.16) to prove C a u c h y ' s I n e q u a l i t y : 7 / 1 / ( 0 1 < M when
\Q — z\ = p, then

\fin\z)\ < ^ 1 . (4.2.17)

pn
Hint: use (4-2.8). Also note that since the function f is continuous, such a
number M always exists. (c)B Now use (4-2.17) to prove L i o u v i l l e ' s
Theorem: A bounded entire function is constant. (That is, if a function is
analytic everywhere in the complex plain and is bounded, then the function
must be constant.) Hint: Taking n = 1 and p arbitrarily large, you conclude
that f'(z) = 0 for all z. Then recall part (c) of Exercise 4-2.8- (d)B Finally,
use Liouville's Theorem to prove the F u n d a m e n t a l Theorem of A l g e b r a :
every polynomial of degree n > 1 with complex coefficients has at least one
202 Functions of a Complex Variable

root. Hint: if the polynomial P = P(z) has no roots, then l/P(z) is a bounded
entire function, hence constant — a contradiction.
The Liouville theorem is due to the same Joseph Liouville who discov-
ered transcendental numbers; he has one more famous theorem, related
to Hamiltonian mechanics. The first proof of the fundamental theorem of
algebra appeared in 1799 in Gauss's Doctoral dissertation; as with other
important theorems, there had been numerous unsuccessful attempts at the
proof prior to that.
EXERCISE 4.2.21."4 Let us say that a function f = f{z) is analytic at the
point z = oo if and only if h(z) — f(l/z) is analytic at z = 0. Prove that
if an entire function is analytic at z = oo, then the function is everywhere
constant.

4.2.4 Conformal Mappings

No discussion of analytic functions is complete without mentioning confor-
mal mappings.
As the name suggests, a conformal mapping is a mapping that preserves
(local) form, or, more precisely, angles. We will see that an analytic function
with non-zero derivative defines a conformal mapping. Before giving the
precise definitions, let us recall how to compute the angle between two
curves in M2.
As a set of points, a curve in M2 is defined in one of the two ways: (a) by
a vector-valued function r(t) = x(t) i + y(t) j ; (b) as a level set of a function
F = F(x,y), that is, a set {(x,y) : F(x,y) = const.}. If r\ = ri(t) and
**2 = T"2(«) define two smooth curves and ri(£ 0 ) = rz(so), then the angle 9
between the curves at the point of intersection is defined by cos 8 =\u\ (to) •
U2{sa)\: it is either the angle between the unit tangent vectors or n minus
that angle, whichever is smaller. If the differentiable functions F = F(x, y)
and G = G(x,y) define two curves so that F(x0,yo) — G(xo,yo) and the
point (xo,j/o) is n ° t critical for F and G, then the angle 0 between the
curves at the point of intersection satisfies

a \VF{xo,yo)-VG{x0,yo)\
\\VF(x0,y0)\\\\VG(x0,yo)W

EXERCISE 4.2.22? Show that conjugate harmonic functions have orthogonal

level sets. In other words, let u, v be two differentiable functions satisfying
the Cauchy-Riemann equations (4-2.2) oft page 192, dndCiif CVf two curves
 Conformal Mappings 203

corresponding to some level sets of u and v. Show that, at every point of

intersection, the angle between Cu and Cv is TT/2. Hint: use (4.2.2) on page
192.

In the complex plane, the vector representation of a smooth curve is

equivalent to denning a continuously differentiable complex-valued function
z = z(t) of a real variable t, so that z(t) ^ 0; the tangent vector to the
curve is represented by the function z(t). Using the properties of complex
numbers, we conclude that if z\ — z\(€) and z2 = z2(s) represent two
smooth curves and zi(to) = z2(s0), then, according to Exercise 4.1.4 on
page 184, the angle 8 between the curves is

6 = min (|Arg(ii(t 0 )) - Arg(i 2 (s 0 ))|,7r - |Arg(ii(t 0 )) - Arg(i 2 (s 0 ))|).

(4.2.18)
We will now see that an analytic function / — f(z) with non-zero deriva-
tive does not change the angle between two curves. Indeed, let z\ — z\(t)
and z2 = Z2(s) represent two smooth curves. The function / maps these
curves to wi(t) = f{z\(t)) and w2(s) = f(z2(s)). By the chain rule,
Mt) = f'(zi(t))zi(t), w2(s) = f'(z2(s))z2(s). If z* = Zl(t0) = z2(s0)
is the point of intersection of the original curves, then f(z*) is the point of
intersection of the images of these curves under / . By Exercise 4.1.4, the
argument of the product of two non-zero complex numbers is equal to the
sum of the arguments, so that, under the mapping / , all smooth curves that
pass through the point z* are mapped onto curves w(t) = f(z(t)), which
are turned by the same angle Arg(/'(z*)). Then relation (4.2.18) implies
that the angle between the original curves at the point z* is the same as the
angle between the images at the point f(z*). Note that the above calcula-
tions do not go through if f'(z*) — 0, because in that case the argument of
f'(z*) is not defined.

EXERCISE 4.2.23? The derivative of the analytic function f(z) = z2 is equal

to zero when z = 0. By considering two lines passing through the origin,
show that this function doubles angles between curves at z = 0. Hint: a line
through the origin is defined by Arg(z) = const, and Arg(z2) = 2Arg(z).
Recall that, to every mapping from R 2 to E 2 , we associate the Jacobian,
the function describing how the areas change locally at every point (see page
148 for a three-dimensional version). If written f{z) = u(x,y) + iv(x,y),
every function of a complex variable defines a mapping from E 2 to E 2 by
sending a point (x,y) to the point (u(x,y),v(x,y)).
204 Functions of a Complex Variable

EXERCISE 4.2.24. B Using the Cauchy-Riemann equations (4-2.2), verify

that the Jacobian of the mapping defined by an analytic function f = f(z)
is equal to \f'(z)\2.
Beside the conservation of angles, another important property of the
mapping defined by an analytic function with non-zero derivative is the
uniform s c a l i n g at every point, that is, the linear dimensions near every
point z are changed by the same factor | / ' ( z ) | near w = f{z) in all direc-
tions. This property is the consequence of the definition of the derivative:
if wo = f(zo), u> = f(z), Az = z~ ZQ, AW = w — wo, then, for \Az\ close to
zero, we have (Aw/Az) « f'(zo) or \Aw\ « |/'(^o)| |Az|, and the direction
from ZQ to z or from WQ to w does not matter.
By definition, a mapping is called conformal at a point if it preserves
angles and has uniform scaling at that point. We just saw that an analytic
function defines a conformal mapping at all points where the derivative of
the function is non-zero.
EXERCISE 4.2.25. B Let f = f(z) be an analytic function in a domain G
and f'{z) ^ 0 in G. (a) Denoting by G the image of G under f, verify that
the area of G is Jf \f'(z)\2dA. (b) Denoting by C the image under f of a
G
piece-wise smooth curve C in G, show that the length of C is J \f'(z)\ \dz\.
c
Figuring out how a particular function / transforms a certain domain or
a curve is usually a matter of straightforward computations. In doing these
calculations, one should keep in mind that, while a conformal mapping pre-
serves the local geometry, the global geometry can change quite dramatically.
F O R EXAMPLE, let us see what the mapping f(z) = 1/z, which is confor-
mal everywhere except z = 0, does to the family of circles \z — ic\2 = c 2 ,
where c > 0 is a real number. It is convenient to consider two different
complex planes: where the function / is defined, and where the function /
takes its values. Since z = x + iy denotes the generic complex number in
the complex plane where / is denned, it is convenient to introduce a differ-
ent letter, w, to denote the generic complex number in the complex plane
where / takes its values. The equation of the circle is x2 + y2 — 2cy = 0.
The relation between z and w is w = 1/z = (x — iy)/(x2 + y2). When z
is on the circle, we have x2 + y2 = 2cy, and then w = x/(2cy) — i/(2c).
In other words, if a point z is on the circle \z — ic\2 = c 2 , then the point
w = 1/z satisfies Sw = —l/(2c); you should convince yourself that, as we
take different points z on the circle, we can get all possible values of the
 Conformal Mappings 205

real part of w. Since the collection of all w with the same imaginary part
is a line parallel to the real axis, we conclude that the function f(z) = \jz
maps a circle \z — ic\2 — c2 to the line {w : Q(w) = - l / ( 2 c ) } .
EXERCISE 4.2.26F (a) Verify that the family of circles \z - c\2 = c2, c> 0,
c ^ 0 is orthogonal to the family of circles \z — ic\2 = c 2 , c £ WL, c > 0
(draw a picture), (b) Verify that the function f(z) = \/z maps a circle
\z — c\2 = c2 to the line line 3?w = l/(2c). Again, draw a picture and
convince yourself that all the right angles stayed right, (c) What happens if
we allow c < 0 ?

For more examples of conformal mappings, see Problem 5.4, page 434.
The following theorem is one of the main tools in the application of
complex analysis to the study of Laplace's equation in two dimensions.

Theorem 4.2.6 Let G and G be two domains in M2 and let f(z) —

u(x, y) + iv(x, y) be a conformal mapping of G onto G. IfU = U(£, n) is a
harmonic function in G, then the function U(x,y) = U(u(x,y), v(x, y)) is
a harmonic function in G.

The domain G is often either the upper half-plane or the unit disk with
center at the origin. Note the direction of the mapping: for example, to
find a harmonic function in a domain G given a harmonic function in the
unit disk, we need a conformal mapping of the domain G onto the unit disk.
EXERCISE 4.2.27? (a) Prove Theorem 4-2.6 in two ways: (i) by interpreting
U as the real part of analytic function F{f{z)), where F is an analytic
function with real part U; (b) by showing, with the help of the Cauchy-
Riemann equations, that

uxx + um = (UK + £/„„) (u2 + u2y).

(b) Suppose you can solve the Dirichlet problem V2U = 0, U\dD = 9, for
every continuous function g when D is the unit disk. Let G be a bounded
domain with a smooth boundary dG, and f : G —» D, a conformal mapping
of G onto D. How to solve the Dirichlet problem V2V = 0, V\QG = h, for
a given continuous function h ?
206 Power Series

4.3 Power Series and Analytic Functions

4.3.1 Series of Complex Numbers

A series of complex numbers ck, k > 0 is the infinite sum Co + c\ + c-i +... =
^ fc>0 Cfc. The sum ]Cfc=icfc' n ^ 1, is called the n-th partial sum of
the series. The series is called convergent if the sequence of its partial
sums converges, that is, if there exists a complex number C such that
linin^oo \C — X^fe=oCfcl = 0- The series is called a b s o l u t e l y convergent
if the series ^fc>olcfcl converges. A series that converges but does not
converge absolutely is said to converge c o n d i t i o n a l l y . Note that the
sequence Yjk=i \ck\: n > 1, of the partial sums of the series ^2k>0\ck\
is non-decreasing and therefore converges if and only if Y^k=i lcfcl — C
for some number C independent of n. As a result, we often indicate the
convergence of Ylk>o \c^\ bY writing £ f e > 0 lcfcl < °°.
EXERCISE 4 . 3 . 1 . C
Show that (a) the condition limn^oo \cn\ —> 0 is neces-
sary but not sufficient for convergence of the series ^Zk>0 ck; (b) absolute
convergence implies convergence, but not conversely.
Recall that, for a sequence of real numbers an, n>0, the upper l i m i t
is denned by

l i m s u p a n = lim supofc,
n
n -*°°k>n

where sup means the least upper bound. Since the sequence An =
sup fc>n ak, n > 0, is non-increasing, the upper limit either exists or is equal
to +oo. For a convergent sequence, the upper limit is equal to the limit of
the sequence. Similarly, the lower l i m i t liminf„a„ = linin^oo inffc>nafc,
where inf is the greatest lower bound, is a limit of a non-decreasing se-
quence; this limit is either —oo or a finite number, and, for a convergent
sequence, is equal to the limit of the sequence.

EXERCISE 4.3.2. Verify that, for every sequence {an, n > 0} of real
numbers, (a) the sequence {An, n > 0}, defined by An = sup fc>n afc is non-
increasing, that is, An+i < An for all n > 0; (b) the sequence {Bn, n > 0},
defined by Bn = inf k>n ^fc is non-decreasing, that is, Bn+\ > Bn for all
n > 0. Hint. Use the following argument: if you remove a number from a finite
collection, then the largest of the remaining numbers will be as large as or smaller
than the largest number in the original collection; the smallest number will be as
small as or larger.
 Series of Complex Numbers 207

The following result is known as the r a t i o t e s t for convergence.

Theorem 4.3.1 Define

„ ,. I c n+l| T v • c lc«+ll
K = hm sup ^-j—j-, L — hm inf ——p.
n
n jCn| |Cn|
Cfc
Then the series J2k>o

• Converges absolutely if K < 1;

• Diverges if L > 1;
• Can either converge or diverge, if K > 1 or L < 1.

Proof. If K < 1, then, by the definition of limsup, there exists a q G (K, 1)

and a positive integer N so that | c n + 1 | < q\cn\ for all n > AT. Then
|c;v+fc| < qk\cN\, k>l, and
AT-l JV-l , ,

q
fc>o fe=o fc>i fc=o

If L > 1, then, by the definition of liminf, there exists a q € (1,L) and

a positive integer N so that |c n +i| > q\cn\ for all n > N. Therefore,
limn^oo \cn\ ^ 0, and the series diverges.
The following exercise completes the proof of the theorem. •

EXERCISE 4.3.3. B Construct three sequences J2k>ock > 3 — 1>2,3, so

t/iai 2Zfc>o cfe converges absolutely, X]fc>o cfc converges conditionally, and
Sfc>o cfc diverges, while the ratio test in all three cases gives K = 2 and
L = 0.

The following result is known as the root t e s t .

Theorem 4.3.2 Define

M = limsup | c n | 1 / n .
n

Then the series J2k>0 ck

• Converges absolutely if M < 1;

• Diverges if M > 1;
• Can either converge or diverge, if M = 1.
208 Power Series

Proof. If M < 1, then, by the definition of limsup, there exists a g e (M, 1)

and a positive integer N so that Icnl1/™ < q, that is, \cn\ < qn for all n> N.
Then
N-l N-l N
cf
E M ^ E M + E <z = E M + rr^ < °°- V
fc>0 fc=0 fe>JV fc=0
If M > 1, then, by the definition of limsup, there exists a q £ (1,K) so
that \cn\ > qn for infinitely many n. Then limn_>oo \cn\ ^ 0 and the series
diverges.
The following exercise completes the proof of the theorem. •

B
EXERCISE 4.3.4. Construct three sequences Ylk>ock > 3 ~ 1)2,3, so
that X3fc>o cfc converges absolutely, Ylk>o ck converges conditionally, and
/Cfc>o cfc diverges, while the root test in all three cases gives M = 1.

4.3.2 Convergence of Power Series

In this section, we will see that analytic functions are exactly the functions
that can be represented by convergent power series. We start with the
general properties of power series.
A power s e r i e s around (or at) point ZQ £ C is a series

z0)k = a0 + ai(z - zQ) + a2{z - z0)2 H

E
^ak(z- , (4.3.1)
fc>0
where ak,k > 0, are complex numbers. The following result about the
convergence of power series is usually attributed to Cauchy and Hadamard.
Theorem 4.3.3 For the power series (4-3.1), define the number R by

R = \ „. , (4.3.2)
hm sup J a „ I 1 /"' >
with the convention 1/ + oo = 0 and 1/0 = +oo. If R = 0, then (4-3.1)
converges only for z = ZQ. If R = +oo, then (4-3.1) converges for all
z e C . If 0 < R < +oo, then (4-3.1) converges absolutely for \z — zo\ < R
and diverges for \z — ZQ\ > R.
An alternative representation for R is

JR = limsupr^-r (4.3.3)
n |a«+i|
 Convergence 209

Proof. Define z = z — ZQ. For z ^ 0, apply the root test to the resulting
series (4.3.1):

K = limsup \anzn\^n = \z\ limsup | a „ | 1 / n = \z\/R.

n n
If R = 0, then K — +oo for all z ^ 0, which means that the series diverges
for z 7^ 0. If R = +oo, then K = 0 for all 5, which means that the series
converges for all z. If 0 < R < +oo, then K < 1 for |z| < R and if > 1 for
\z\ > i?, so that the result again follows from the root test.
Representation (4.3.3) follows in the same way from the ratio test. •

EXERCISE 4.3.5. C Verify representation (4.3.3).

Definition 4.3 The number R introduced in Theorem 4.3.3 is called
the r a d i u s of convergence of the power series (4.3.1), and the set {z :
\z — zo| < R} is called the d i s k of convergence.

EXERCISE 4.3.6. c (a) Verify that the power series X)fc>o a*zk an
d
k
^ f e > 0 ka\iz have the same radius of convergence, (b) Give an example
of a power series that converges at one point on the boundary of the disk of
convergence, and diverges at another point. Explain why the convergence
in this example is necessarily conditional (in other words, explain why the
absolute convergence at one point on the boundary implies absolute conver-
gence at all points of the boundary).
In practice, it is more convenient to compute the radius of convergence
of a given series by directly applying the ratio test or the root test, rather
than by formulas (4.3.2) or (4.3.3). F O R EXAMPLE, to find the radius of
convergence of the power series
(-l)"z3"+1n!
(4.3.4)
n>l
(2n) n

f-ll"z 3 n + 1 n'
L anc
we write c„ = •*—fan)" *

where we used lim n _ >00 (l + n -1 )™ = e. The ratio test guarantees conver-

gence if |z| 3 /(2e) < 1 or \z\ < \/2e. Therefore, the radius of convergence is
\/2e. Direct application of either (4.3.2) or (4.3.3) is difficult because many
of the coefficients an in the series (4.3.4) are equal to zero.
210 Power Series

EXERCISE 4.3.7. c Find the radius of convergence of the power series

7.2n+1(dnV
E ( 5 + ( _ l n «
fc>l

Let us now state and prove the main result of this section.

Theorem 4.3.4 A function f = f{z) is analytic at the point ZQ if and

only if there exists a power series Ylk>oak(z ~ z°)k vy^1 some radius of
convergence R > 0 so that f(z) = ^2k>0 ak{z — zo)k for all \z — zo\ < R.

Proof. Step 1. Let us show that an analytic function can be written as a

convergent power series. Recall that, by definition, / = / ( z ) is analytic at
ZQ if and only if / has a derivative in some neighborhood of ZQ. Therefore,
there exists an R > 0 so that / is analytic in the open disk GR = {z :
\z — zo\ < R}. Take z G GR and find a number p so that \z — zo| < P < R-
Then the circle Cp with center at ZQ and radius p encloses z and lies entirely
in GR (draw a picture). By the Cauchy Integral Formula,

Next, we note that \z — ZQ\ < |£ — ZQ\ = p for C, € Cp and use the formula
for the geometric series to write

(z - Z0)

We now plug the result into (4.3.5) and find

z0)k dC (4.3.7)

Finally, let us assume for the moment that we can integrate term-by-term
in (4.3.7), that is, first do the integration and then, summation; this is
certainly true for sums with a finite number of terms, but must be justified
for (4.3.7), and we will do the justification later. Then the term-by-term
integration results in the equality

/(z) = J> f e (z - z0)fe, (4.3.8)

fc>0
 Convergence 211

where

By (4.3.6), the radius of convergence of the power series in (4.3.8) is at least

p. This shows that an analytic function can be represented by a convergent
power series.
Step 2. Let us show that a convergent power series defines a continuous
function. Define g(z) = J2k>oak(z ~ zo)k, \z — ZQ\ < R. We will show
that the function g is continuous for \z — ZQ\ < R, that is, for every z\ in
the disk of convergence and for every e > 0, there exits a S > 0 so that
\g(zi) — g(z)\ < E as long as \z\ — z\ < S and \z — ZQ\ < R; the interested
reader can then use similar arguments to prove that g is analytic. To prove
the continuity of g, fix a z\ with \z\ — ZQ\ < R and an e > 0. Next, let
us define r = (R+\z\ — zo\)/2 and find TV so that J2k>N+i \ak\rk < e / 4 -
Such an N exists because r < R and the power series converges absolutely
inside the disk of convergence. Now consider gN(z) = J2k=i akZk- Being a
polynomial, gN(z) is a continuous function at z\, and therefore there exists
a <5i > 0 so that \gw(z\) — g(z)\ < e/2 as long as \z — z\\ < 6\. Finally,
define S as the minimum of Si and (R — \z\ — ZQ\)/A. By construction, if
l^i — z\ < S, then \ZQ — z\ < r (draw a picture). Now let us look at the
value of \g{zi) — g(z)\ for \z — z\\ < 6. By the triangle inequality,

\g(zi)-g(z)\<\gN(z1)-gN(z)\+ ^ \ak\ \zi-zQ\k+ ]T \ak\ \z-zQ\k;

k>N+l k>N+l
(4.3.10)
by the choice of N and J, |gjv(zi) — 9JV(Z)| < e/2, and, by the choice of r,

X) \ak\\zi-z0\k+ ]T \ak\\z-zQ\k <2 J2 \ak\rk<e/2.

k>N+l k>N+l k>N+l

As a result, \g(zi) — g(z)\ < e, which proves the continuity of g.

The following exercise completes the proof of the theorem. •

EXERCISE 4.3.8/ 1 (a) Use the same arguments as in Step 2 above to show
that g(z) = J2k>o ° fe ( z ~ z°)k l s differentiable for \z — z 0 | < R and g'{z) =
Sfc>o kak{z—zo)k~l • Thus, g is indeed an analytic function for \z—ZQ\ < R.
Hints: (i) you can differentiate the polynomials; (ii) by Exercise 4-3.6, the power
series ^2k>0 fcafcZ*-1 has the radius of convergence equal to R. (b) Use the same
arguments to justify the switch of summation and integration in (4-3.7).
Hints: (i) you can do this switching when the number of terms is finite, (ii)
212 Power Series

The value of \z — zo|/|z — C| is constant for all £ on Cp and is less than one,
which allows you to make the sum ^2k>N arbitrarily small, (c) Show that the
n-th derivative g^ of g(z) = 2fc>o a fc( z — z°)k ea;*s*s inside the disk of
convergence and satisfies g^(zo) = n)an. Then use (4-3.8) and (4-3.9) to
give the rigorous proof of (4-2.16) on page 201. Hint: part (a) of this exercise
shows that you can differentiate the power series term-by-term as many times as
you want.
Corollary 4.1 Uniqueness of power s e r i e s . (a) Assume that the
a z z k k
two power series Y^T=o k( ~ o) , Sfclo bk(z — zo) converge in the some
neighborhood of ZQ and Y?kLoak(z ~ zo)k = J2T=o^k(z — zo)k for a^ z * n
that neighborhood. Then a^ = b^ for all k > 0.
(b) If f is analytic at ZQ, then

f(z) = f(zo) + f ; ^ - ^ - (z - z0)k (4.3.11)

fc=i

for all z in some neighborhood of ZQ. Therefore, if a function has a power

series representation at a point ZQ, this power series is necessarily (4-3.11).
EXERCISE 4.3.9; A Prove both parts of Corollary 4-1-
As in the ordinary calculus, the series in (4.3.11) is called the Taylor
s e r i e s for / at ZQ\ when z§ = 0, we also call it the Maclaurin s e r i e s .
It appears, though, that the original idea to represent real functions by a
power series belongs to neither Taylor nor Maclaurin and can be traced back
to Newton. The English mathematician BROOK TAYLOR (1685-1731) and
the Scottish mathematician COLIN MACLAURIN (1698-1746) were the first
to make this idea clear enough and spread it around, and thus got their
names attached to this power series representation, even in the complex
domain.
The methods for finding the power series expansion of a particular com-
plex function are the same as for the real functions. With these methods,
you never compute the derivatives of the function. One of the key relations
is the sum of the geometric series:

r ^ = E 2 *- w <:L- (4-3-12)
fc>o
F O R EXAMPLE, let us find the Taylor series for f(z) = 1/z 2 at point z 0 = 1-
We have f(z) = -g'(z), where g{z) = 1/z. Now, 1/z - 1/(1 + (z -
1)) = Efe>o( _ 1 ) f c ( z ~ 1)fc> w h e r e w e u s e d (4-3.12) with -(z - 1) instead
 Exponential Function 213

of z. Differentiating term-by-term the power series for g gives f{z) =

l k+lk z 1 k 1 1 fe fc fc w h e r e i n
Ek>i(- ) ( ~ ) ~ = E f c >o(- ) ( + 1)(* - !) . the last
equality we changed the summation index to start from zero. The series
converges for \z — 1| < 1.
EXERCISE 4.3.10. c Find the power series expansions of the following
functions at the given points, and find the radius of convergence: (a)
f(z) — z + z + l at ZQ = i. Hint: put z = i+(z — i) and simplify. Alternatively,
put w = z — i so that z = w + i, and find the expansion of the resulting function in
powers of w; then replace w with (z — i). (b) f(z) = l/(z + 1z + 2) at ZQ = 0
Hint: use partial fractions: z2 + 2z + 2 = (z + l) 2 + 1 = (z + 1 — i)(z + 1 + i),
f(z) = A/(z + 1 - i) + B/(z + l + i).
Given an analytic function, you usually do not need the explicit form of
its Taylor series to find the r a d i u s of convergence of the series. Indeed,
by Theorem 4.3.4, the radius must be the distance from ZQ to the closest
point at which / is not analytic (think about it...). F O R EXAMPLE, the
function f(z) = l / ( z 2 + 2z + 2) is not analytic only at points z = l±i, and
both points are y/2 away from the origin. As a result, the Maclaurin series
for / has the radius of convergence equal to \pl.
EXERCISE 4.3.11. c Consider the function f(z) = ( . 2w2_5s at the point
ZQ = 1 + 2i. (a) Without computing the series expansion, find the radius of
convergence of the Taylor series of f at ZQ. (b) Find the series expansion
and verify that it has the same radius of convergence as you computed in
part (a).

4.3.3 The Exponential Function

For z e C, we define

ez = Zw^ = E[-Tdrw>C0SZ
V
=
^(kr-
V
(4 3 13)
--
fc>0 fc>0 ' fe>0 '

EXERCISE 4.3.12. (a)c Verify that all three series in (4-3.13) converge
for every z € C. (b)B Verify the two main properties of the exponential
function:

ezi+z* =e'ie*2t (e«y==e*)

214 Power Series

E u l e r ' s formula
eie = cos6 + ism6, (4.3.14)
and the relations
eiz — e~*z eiz 4- e~iz
sinz= — , cosz= . (4.3.15)
2i 2

We use the Euler formula to evaluate the complex exponentials: it fol-

lows from (4.3.14) that, for z = x + iy,
ez = ex(cosy + isiny), \ez\ — ex. (4.3.16)
Relation (4.3.16) shows that, as a function of complex variable, ez can take
all complex values except zero and is a periodic function with period 2m.
Similarly, (4.3.15) implies that sinz and cosz can take all complex values.
In particular, the familiar inequalities ex > 0, |sinx| < 1, |cosa;| < 1 that
are true for real x, no longer hold in the complex domain.
EXERCISE 4 . 3 . 1 3 . B (a) Verify that the mapping defined by the exponential
function f(z) = ez is conformal everywhere in the complex plane. (b)Find
the image of the set {z : Uz > 0, 0 < Sz < 7r} under this mapping.
Two other related functions are the hyperbolic s i n e sinh and
hyperbolic cosine cosh:
ez — e~z ez -\- e~z
sinhz = , coshz = . (4.3.17)

EXERCISE 4 . 3 . 1 4 . B (a) Verify that

z2k+l z2k
sinh z — > 7— -77, cosh z = > -^rm •

(b) Verify that

sin(a; + iy) = sin x cosh y + i cos x sinh y, sinh(,z) = —i sin(iz),
cos(a: + iy) = cos a; coshy — isinx sinh y, cosh(z) = cos(iz).

The n a t u r a l logarithm In z of z ^= 0 is defined as the number whose

exponential is equal to z. If a = In z, then, because of (4.3.14), a+2ni is also
a natural logarithm of z. In other words, In z has infinitely many values,
and, since z — \z\elaTe^z\ we have lnz = ln|z| +iaxg(z). The p r i n c i p a l
value of the natural logarithm is defined by Ln z = In \z\ + i Arg(z), which
 Laurent Series 215

is specified by the condition Lnl = 0. More generally, by fixing the value

of ln z at one point ZQ ^ 0, we define a branch of the natural logarithm.
For example, condition l n l = 2m results in a different branch / = f(z) of
ln z, such that f(i) = (5ir/2)i. The values of different branches at the same
point vary by an integer multiple of 27ri.
EXERCISE 4.3.15. C (a) Use the Cauchy-Riemann equations in polar coordi-
nates to verify that the derivative of Ln z (and, in fact, of every branch of
the natural logarithm) is 1/z. (b) Verify that —Ln(l — z) = Y^k>\ zk/k- (If
everything else fails, just integrate term-by-term the expansion of 1/(1 — z).)
Using the exponential and the natural logarithm, we define the complex
power of a complex number:

zw = ewlnz. (4.3.18)

While the natural logarithm Inz has infinitely many different values, the
power can have one, finitely many, or infinitely many values, depending on
w. Quite surprisingly, a complex power of a complex number can be real:
jt = ei(ni/2+2*ki) = e-^/2-27Tfc) k = Q )± 1 )± 2 ) ... f

which was first noticed by Euler in 1746. By selecting a branch of the nat-
ural logarithm, we select the corresponding branch of the complex power.
For example il — e - 7 r / 2 corresponds to the principal value of the natural
logarithm. What is the value of il if we take ln 1 = —2-KI>.
EXERCISE 4.3.16. (a)c Verify that if m is an integer, and w = 1/m,
then the above definition of zw is consistent with (4-1.5), page 185. (b)B
For what complex numbers w does the expression zw have finitely many
different values? Hint: recall that the exponential function is periodic with period
2m. (c)A Assume that in (4-3.18) we take the principal value of the natural
logarithm. Verify that, for every complex number w, the corresponding
function f(z) = zw is analytic at every point z / 0 , and f'(z) = wzw~1 •
Convince yourself that this is true for every branch of the complex power.

4.4 Singularities of Complex Functions

4.4.1 Laurent Series

We know that a function analytic at a point ZQ can be written as a Taylor
series that converges to the values of the function in some neighborhood
216 Singularities of Complex Functions

of ZQ. The proof of this series representation essentially relies on the fact
that a neighborhood of a point is a simply connected set. It turns out
that a somewhat similar expansion exists in domains that are not simply
connected. This expansion is known as the Laurent series, the study of
which is the main goal of this section. The French mathematician PIERRE
ALPHONSE LAURENT (1813-1854) published the result in 1843.
Before we state the result, let us make a simple yet important observa-
tion that will be an essential part of many computations to follow.
Recall that the geometric series formula
.. oo

fc=0

is true for \z\ < 1. On the other hand, writing

1 1
1-z *(l-i)
and replacing z with 1/z in (4.4.1), we can write

1- Z~ £" Zk+l '

fc=0

which is now true for \z\ > 1. Therefore, we have two representations for
1/(1 — z), one, for small \z\, and the other, for large.
We will now state and prove the main result of this section. Recall that
the Taylor series is written in an open disk {z : \z — ZQ\ < R}, which is the
basic example of a simply connected domain; we allow R = oo to include
the whole plane. Similarly, the basic domain that is not simply connected
is an annulus, a set of the type {z : R\ < \z — ZQ\ < R}, where ZQ is a fixed
complex number and, for consistency, we allow i?i = 0 and/or R = oo; the
Latin words anulus and anus mean "a ring". The Laurent series expansion
is written in an annulus.
T h e o r e m 4.4.1 Laurent s e r i e s expansion. If a function f = f (z) is
analytic in the annulus G = {z : R\ < \z — ZQ\ < R}, then, for all z £ G,
oo

f(z)= J2 ^(z-z0)k
k= — oo

7 r^ + 7 r + c0 + ci(z - z0) + c2(z - zQ)2 + ...,

(Z - Z0y (2 - Zo)
(4.4.2)
 Laurent Series 217

where

Ck = : (
ti f '» '-^fc+i d C, fc = 0 , ± l , ± 2 , . . . , (4.4.3)
27riJCo(C-z0)fe+1'
and Cp is a circle with center at ZQ and radius p so that Ri < p < \z — zo\;
the circle is oriented counterclockwise. Representation (4-4-%) *s unique:
if f(z) = EfeL-oo ck(z ~ zo)k = Efcl-oo ak(z ~ zo)k in G, then ak = ck for
all k.
Proof. We present the main steps of the proof. The details are in the
exercise below.
Step 1. Fix the point z. Let Cr be a circle with center ZQ and radius
r so that p < r < R and \z — ZQ\ < r (draw a picture). Combining the
proof of Theorem 4.2.4, page 198, with the Integral Formula of Cauchy, we
conclude that

with both Cr and Cp oriented counterclockwise.

Step 2. We already know that

h i ^-C - J>(. - »)>, * - JL £ j^JC (4.4.5)

see the proof of Theorem 4.3.4, page 210.
Step 3. By Exercise 4.2.17, page 199, we conclude that
/(C) ,r i / /(C) ,, r
Tcr (C - zo)k+1
2™ Jr * 2ni JCp (C - zo) f c + 1
Step 4- Let us show that

h I {&« - t ( 7 ^ - — Si £'<«« - *>"«•

fc=i
(4.4.6)
Once again, in (4.4.6) we have positive integer k.
To establish (4.4.6), we write
1 1
C- z z-zQ-{Q-z0) (2 _ ZQ) ^ _ c ^

_y (C-*o)fc _f>(C-*o) fc - 1
218 Singularities of Complex Functions

Then

_m = f /(OK-*)*-1 (447 )

and it remains to integrate this equality.

Step 5. We combine the results of the above steps to get both (4.4.2)
and (4.4.3). Note that any simple, closed, piece-wise smooth curve can be
used instead of Cp in (4.4.3), as long as the curve is completely inside G,
the closed disk {z : \z — ZQ\ < R\\ is inside the domain enclosed by the
curve, and the point z is outside that domain.
Step 6. To prove the uniqueness, we multiply the equality
c z z k
'E'kL-oc i*( - o) = T,'kL-ooak(z-zo)k by (z-zo)-™-1 for some integer
m and integrate both sums term-by-term over the circle Cp. By (4.2.6), page
195, all integrals become zero except for those corresponding to k = m, and
we get cm = am. •

EXERCISE 4.4.1. A (a) Verify (4-4-4)• Hint: connect the circles Cr and Cp
•with a line segment that does not pass through the point z and write the Integral
Theorem of Cauchy in the resulting simply connected domain. Then simplify the
result, keeping in mind that you integrate along the line segment twice, but in
opposite directions, and that the orientation of Cp is clockwise. (b) Justify
the term-by-term integration in (4-4-V- Hint: note that, for £ e Cp, we have
|(C — zo)/(z — zo)\ — p/\z — zo\ < 1. Then use the same argument as in the
proof of Theorem 4-3-4, Page 210. (c) Fill in the details in the proof of the
uniqueness of the expansion. Hint: once again, the key step is justifying the
term-by-term integration, and once again, you use the same arguments as in the
previous similar cases.

Note that if, in the above theorem, the function is analytic for all z
satisfying \z — ZQ\ = Ri, then we can decrease R\. Similarly, we can increase
R if / is analytic for all z satisfying \z — ZQ\ = R (make sure you understand
this). In other words, with no loss of generality, we will always assume
that the annulus Ri < \z — zo\ < R is maximal, that is, each of the sets
|z — 2o| = Ri and \z — z§\ = R (assuming R < oo) contains at least one
point where the function / is not analytic. There are several types of such
points.

Definition 4.4 A point ZQ € C is called an i s o l a t e d s i n g u l a r point

or an i s o l a t e d s i n g u l a r i t y of a function / if the function / is not ana-
 Laurent Series 219

lytic at ZQ and there exists a S > 0 so that the function / is analytic in the
region {z : 0 < \z — ZQ\ < 6}.
An isolated singular point ZQ is called

• removable, if the function can be defined at zo so that the result is an

analytic function at ZQ.
• a pole of order k, if k is the smallest value of the positive integer
power n with the property that function (z — zo)nf(z) has a removable
singularity at ZQ.
• an e s s e n t i a l s i n g u l a r i t y , if it is neither a removable singularity nor
a pole.

A pole of order 1 is called simple.

Without going into the details, let us mention that the point z = 0 is not an
isolated singularity of f(z) = z 1//2 in the sense of the above definition, but
rather a branching p o i n t of order two. The reason is that the square root
yfz has two different values in every neighborhood of z = 0. As a result,
in any neighborhood of zero, there is no unique number assigned to z1^2
and f(z) = z1/2 is not a function in the sense of our definition. Similarly,
z = 0 is a branching point of order three for the function f{z) = z - 1 / 3
(the fact that / is unbounded near z = 0 is not as important as the three
different values of yfz in every neighborhood of z = 0), and z — 0 is a
branching point of infinite order for the function f(z) = \nz. The study
of branching points and the related topics (multi-valued analytic functions,
Riemann surfaces, etc.) is beyond the scope of our discussions.

Note that the closed disk {z : \z — ZQ\ < Ri} can contain several
points where / is not analytic, and ZQ is not necessarily one of them.
In the special case of the Laurent series with R\ — 0, ZQ is an isolated
singular point of the function / , with no other singular points in the do-
main {z : 0 < \z — ZQ\ < R} for some R > 0. The corresponding Lau-
rent series is called the expansion of / a t (or around) t h e i s o l a t e d
s i n g u l a r p o i n t ZQ. This expansion has two distinct parts: the r e g u l a r
p a r t , consisting of the terms with non-negative k, J2k>ock(z ~~ zo)k, and
the p r i n c i p a l p a r t , consisting of the terms with the negative values of k,
2fc<o Ck(z ~ zo)k- A s the names suggest, the regular part is a function that
is analytic at ZQ (this follows from Theorem 4.3.4), and the principal part
determines the type of the singularity at ZQ (this follows from the exercise
below).
220 Singularities of Complex Functions

EXERCISE 4 . 4 . 2 . C (a) Let ZQ be an isolated singular point of the function

f = f(z), and consider the corresponding Laurent series Y^'kL-oo ck(z—zo)k
converging for 0 < \z — zo\ < R. Verify that ZQ is
(i) a removable singularity if and only if Ck = 0 for allk < 0 Hint: this pretty
much follows from Theorem 4-3.4- Equivalently, ZQ is a removable singularity
if and only if there exists a 6 > 0 so that the function f(z) is analytic and
bounded for 0 < \z — ZQ\ < S.
(ii) a pole of order N if and only C-N ^ 0 and Ck = 0 for all k < —(N + 1)
Hint: for the proof in one direction, multiply the Laurent series by (z — zo)N; for
the proof in another direction, multiply f by (z — zo)N and use Theorem 4-3-4-
(Hi) an essential singularity if and only if Ck ^ 0 for infinitely many k < 0.
Hint: by definition, the essential singularity is the only remaining option.
(b) Verify that the point ZQ = 0 is
(i) a removable singularity for the functions f(z) = sin z/z,
f(z) = (ez — 1) 2 /(1 — cosz) and f(z) — (zcosz — sin z)/(z sin z);
(ii) a second-order pole for the function f(z) = (1 — cosz)/(ez - l ) 4 ;
(Hi) an essential singularity for the function f(z) — e 1//z ;
(iv) not an isolated singularity for the function f(z) — l / s i n ( l / z ) .
As with the Taylor series, we usually do not use the formula (4.4.3) to
find the coefficients of the Laurent series, and use other methods instead.
F O R EXAMPLE, consider the function

««) — ( I ^ I ) -
This function is analytic everywhere except at the point ZQ = 1. Let us find
the Laurent series for / at ZQ. We have z/(z — 1) = (z - 1 + l ) / ( z — 1) =
1 + (z — 1 ) _ 1 and so

f(z) = sin(l + (z-l)-1) = s i n l cos((2 - 1) _ 1 ) + sin((z - l ) - 1 ) cosl,

where we use the formula for the sine of the sum. Then we use the standard
Taylor expansions for the sine and cosine to conclude that

f{Z) = Sln
' go <* - W * ) ' + C°S %tz~ D^(2 f c + 1)!;
one could write this as a single series, but it will not add anything essential
to the final answer. We therefore conclude that ZQ — 1 is an essential
singularity of / .
 Laurent Series 221

EXERCISE 4.4.3. c Find the Laurent series for the function

(2z + 5\
f(z) = cos
JK
' \z + 2 J
around the point ZQ = —2. What is the type of the singularity of f at ZQ?
When writing the Laurent series expansion, one should pay attention to
the domain in which this expansion should hold. Recall that the expansion
is written in an annulus {z : Ri < \z — ZQ\ < R}, where ZQ is a fixed complex
number, and none of the singular points of / should be inside this annulus.
The expansion in such an annulus has the form Y^kL-oo ck(z ~ zo)k- The
point ZQ does not have to be a singular point of f, but at least one singular
point of f must be in the set {z : \z — ZQ\ = Ri}, and, unless R is infinite, at
least one singular point of f must be in the set {z : \z — ZQ\ = R} (otherwise,
we are able to expand the annulus). The Laurent series that converges in
the disk {z : \z — ZQ\ < R} is the same as the Taylor series at ZQ. As a result,
the same function can have different expansions in different domains, even
when the point ZQ is the same: recall that (1 - z)~l = Ylk>o zk ^ or \z\ < 1
and (1 - z)-1 = E f c ^ o * - * - 1 for \z\ > 1.
As A DIFFERENT EXAMPLE, consider the function

The Laurent series of this function at ZQ = 0 is f(z) = (2z)~1 +

Ylk>o 2~k~2zk (check it), and ZQ is a simple pole (a pole of order one).
The expansion is true for 0 < \z\ < 2. Similarly, the Laurent series in the
domain {z : 2 < \z — 2|} is computed as follows: 1/z = (z — 2 + 2)""1 =
(z - 2 ) - x ( l + 2/(z - 2))- 1 , and so

!{Z) = _
2(1^2) +^0(z- 2)fc+! = W=Y) +
g (z - 2)*+i • (4A9)

EXERCISE 4.4.4. C Find the Laurent series of the function f from (4-4-8) in
the domain {z : 0 < \z — 2| < 2}.
It follows from (4.4.8) that ZQ = 2 is a pole of order one; this is also
what you should conclude from the previous exercise. On the other hand,
the expansion in (4.4.9) contains infinitely many negative powers of (z — 2).
Should we conclude from (4.4.9) that ZQ = 2 is an essential singularity of / ,
and how do we reconcile these seemingly contradictory conclusions? After
looking more closely at (4.4.8), we realize that (4.4.8) is not a Laurent series
222 Singularities of Complex Functions

of / at zo = 2; the Laurent series at the point ZQ = 2 must converge when

0 < \z — 21 < R for some R, and that is the series you compute in Exercise
4.4.4.
Finally, let us emphasize once again that a Laurent series does not have
to be around a singular point. We go back to the equality (1 — z ) _ 1 =
Sfc>o • 2 ~ 1_fc , which is true for \z\ > 1. Even though we have infinitely many
negative powers of z, zero is not an essential singularity of f(z) = 1/(1 — z),
because, for f(z) — 1/(1 — z), the point ZQ = 0 is not a singularity at all!
EXERCISE 4.4.5. G Find the expansion of the function from (4-4-8) in the
domain {z : 1 < \z + 1| < 3}. Hint: use geometric series; your answer should
be Xlfcl-oo ak(z + 1) for suitable numbers ak-
EXERCISE 4.4.6^ We say that the point z = oo is an isolated singular point
of the function f = f(z) if and only if the point z = 0 is an isolated singular
point of the function h(z) = f(l/z). Show that z = oo is (a) a removable
singularity of a rational function P(z)/Q(z) if the degree of P is less than
or equal to the degree of Q; (b) A pole of order n for a polynomial of degree
n; (c) an essential singularity for f(z) = sin 2.

4.4.2 Residue Integration

Let ZQ be an isolated singular point of the function / = f(z) and let
oo

f{z)= Y, ck(z-z0)k (4.4.10)

fc= —oo

be the corresponding Laurent series expansion of / at ZQ. The coefficient

c_i in this expansion is called the r e s i d u e of the function / = f(z) at the
point zo and is denoted by Kesf(z). As we saw in the previous section,
Z = ZQ

there could be several different expansions of / in powers of (z — ZQ); the

expansion we use in (4.4.10) is around the point z§ and must converge when
0 < \z — ZQ\ < R for some R.
The Latin word residuus means "left behind," and we will see next that
c_i is the only coefficient in the expansion (4.4.10) that contributes to the
integral of / along a closed curve around ZQ.
Recall that

I dz |2*i, n=l,
Jc(z0) (* - zo)n [0, n = 0,-1,±2,±3,...,
 Residue Integration 223

where C(zo) is a simple, closed, piece-wise smooth curve enclosing the point
zo and oriented counterclockwise; see Exercises 4.2.13 and 4.2.17. Term-
by-term integration of (4.4.10) over the curve C(ZQ) then results in

<p f(z)dz = 2iric-1=2iri Ties f(z). (4.4.12)

z
JC{z0) =z°

Note also that we get (4.4.12) after setting k = — 1 in formula (4.4.3) for
the coefficients Cfc (remember that, for our integration purposes, the curve
C(ZQ) can be replaced with a circle centered at ZQ). Consequently, c_i is the
only coefficient in the Laurent series expansion contributing to the integral
over a closed curve around the singular point.
The more general result is as follows.
Theorem 4.4.2 Let G be a simply connected domain. Assume that the
function f = f(z) is analytic at all points in G except finitely many points
z\,..., zn. If a simple, closed, piece-wise smooth curve C in G encloses the
points z\,..., zn and is oriented counterclockwise, then

I f(z)dz = 2m V Res f(z). (4.4.13)

Jc fiz=Zk
EXERCISE 4.4.7. (a)B Prove the above theorem. Hint: surround each zk
with a small circle that stays inside the domain bounded by C, then apply Theorem
4-2-4, page 198, in the (not simply connected) domain bounded by C and the n
circles around z\,..., z„. (b)c Explain why both the Cauchy Integral Formula
(4-2.11), page 199, and formula (4-2.16), page 201, are particular cases of
(4-4-13). Hint: for (4-2.16), write the Taylor expansion and integrate term-by-
term.
The idea of the residue integration is to find the residues of a function
without computing any integrals, and then use the above theorem to evalu-
ate the integrals of the function over different closed curves. Let us mention
that if the curve does not enclose any singular points of the function, then
the integral along the curve is zero by the Integral Theorem of Cauchy.
If the curve passes through a singular point, then, in general, the integral
along such a curve is not defined (remember that, to define the integral, we
require the function to be continuous at all points on the curve).
COMPUTING THE RESIDUES. The residue is a certain coefficient in the
Laurent series expansion at an isolated singular point, and there are three
types of isolated singular points: removable singularity, pole, and essential
224 Singularities of Complex Functions

singularity. Computation of the residue at a removable singularity or a pole

does not require the explicit knowledge of the Laurent series expansion.
If ZQ is a removable singularity, then, according to Exercise 4.4.2, c_i =
Oand Res/(z) = 0.
2=20

If ZQ is a simple pole of / , then

Res/(z) = lim ((* - z0)f(z)), (4.4.14)

because the Laurent series around a simple pole is

oo
J2cn(z-z0)k.
C-l
f(z) = ^^
Z — Zn
+
U
k=0

An immediate consequence of (4.4.14) is the following result.

EXERCISE 4.4.8F Assume that f(z) = h(z)/g(z), where h,g are analytic at
ZQ, h(zo) ^ 0, g(z0) = 0, g'(zo) / 0. Show that

Res/(z) = ^ . (4-4.15)

Hint: use (4-4-14) and note that g'(zo) = limz_»z0(<?(z) — g(zo))/(z — zo) =
\imz^zo g(z)/(z - z0).
Usually, formula (4.4.15) is easier to use than (4.4.14). F O R EXAMPLE,
if f(z) = (z + 5)/(z3-z), then

2Se-s1^=^iL=6/2-3-
If z0 is a pole of order N > 1, then the function (z — z0)N f(z) has a
removable singularity at ZQ and

(z - z0)Nf(z) = J2 ck(z - z0)k+N = C-N + C-N+1(z - zo)

k=-N
+ ... + c_ 2 (z - z0)N-2 + d(z - zo)"-1 + -..

We differentiate this equality N - 1 times with respect to z, so that all the

terms on the right that are before c_i(z — ZQ)N~X disappear, while the term
c-i(z — ZQ)N~X becomes (N — l)!c_i. We then set z = ZQ so that all the
 Residue Integration 225

terms after (TV — l)!c_i disappear as well. As a result,

1 / dN~l \

In particular, if f(z) = g(z)(z — ZQ)~N, where the function g is analytic at

ZQ and g(zo) ^ 0, then

£s/^=(]^i)!^'1)^)- (4 417)
-
FOR EXAMPLE, if / ( z ) = (z6 + 3z 4 + 2z - \)/{z - l ) 3 , then TV = 3 and

Res/(z) = - ( 6 • 5z 4 + 3 • 4 • 3z 2 )| 2 = 1 = 33.
z=i 2

If z — ZQ is an essential singularity, then the computation of the residue

usually requires the computation of the corresponding terms in the Laurent
series. F O R EXAMPLE, if f(z) = 2 2 sin(l/,z), then f(z) = z2(l/z-l/(6z3) +
l/(120z 5 ) - . . . ) and so Res/(z) = - 1 / 6 .
z=0
C
EXERCISE 4.4.9. Taking C = {z : \z - 5i| = 5}, verify that

£ -dz = 87T2.
ic e z + 1
Hint: how many poles are enclosed by C?
One of the most elegant applications of residue integration is evaluation
of real integrals. In what follows, we describe several classes of real integrals
that can be evaluated using residues.
The easiest class is integrals involving rational expressions (sums, dif-
ferences, products, and ratios) of simp and cos<p, integrated from 0 to 2n:

j H(cos ip, sin (p) dip.

Jo
Such integrals are immediately reduced to complex integrals over the unit
circle. Indeed, writing z = el,p, we have z going around the unit circle
counterclockwise; dz = iet<pdip, so that dip = dz/(iz); also, by the Euler
formula, cos<p = (z + z~1)/2, simp = (z — z~1)/(2i). As a result,

z2 + 1 z2 - 1 \ dz
/ H(cos ip, simp) dip = / HI
Jo J\z\=i \ 2z 2iz I iz
226 Singularities of Complex Functions

assuming that the integral on the left is well-defined. Then integration

is reduced to computing the residues inside the unit disk for the modified
integrand; the computations can be rather brutal, even for seemingly simple
functions H.
F O R EXAMPLE, let us evaluate the integral

-I
* 2 + COSI/5 ,
•;—7T~.—dip.
4 + 3sinv? ^

Step 1. We write z = eitfi, cosip = (z2 + l)/(2z), simp = (z2 - l)/(2iz),

dip = dz/(iz), and so

z2 + 4z + 1
Ay
J\z\ = l z(3z2 + 8iz - 3)

Step 2. The function/(z) = ( z 2 + 4 z + l ) / ( z ( 3 z 2 + 8 i z - 3 ) ) we are integrating

has three simple poles: ZQ = 0, zi, and Z2, where zi,2 are the solutions of
3z 2 + 8iz - 3 = 0. By the quadratic formula, zi, 2 = (-4i ± V - 1 6 + 9)/3 =
i(—4± V7)/3; only z\ = i(—4 + v / 7)/3 has \zi\ < 1 and therefore lies inside
the unit circle. As a result,

I = 2ni Res/(z)+Res/(z)
1
z=0 z=zi

Step 3. We compute the residue at zero by formula (4.4.14):

1
Res/(z) - ( z / ( z ) ) | z = 0
z=zo O

We compute the residue at Z\ by formula (4.4.15). It is more convenient to

compute the derivative of the denominator by the product rule:
(z(3z 2 + 8iz - 3))' = (3z 2 + 8iz - 3 ) + z(6z + 8i), because the first term on
the right vanishes at z\. As a result,

z 2 + 4zi + 1 - 4 + V7 .
Res KHz) — -—; —, where z\ = 1.
*=zi ' 2zi(3zi+4i)' 3
 Residue Integration 227

We simplify the expression as follows:

,2 , , , ~(16-8>/7 + 7 ) + i l 2 ( - 4 + > / 7 ) + 9
z1 + A7
4zi + 1 =
_ - 1 4 + 8^7 (-A + V7)
—
— ~ r~ *& „ j

-4+v/7\ -14 + 8 ^
2z1(3zi+4i) = -2^7

As a result,

Res/(z) = - - 2 t - = .
z=zi 3 ^/7
5£ep 4- We now get the final answer:

47T
7= 2 ^ 1 - -+ - - ^
77

EXERCISE 4.4.10. C ' W%/IOM£ using a computer, verify that

/•27I-
1 + 2 sin co , 2-K
aw = — .
0 5 + 4 cos (p 3

/f Cfl^vX.

\ ,» —•• 1 x

-R 0 R -R ~P O R
(a) (b)
Fig. 4.4.1 Computing Real Integrals

Next, we consider integrals of the type

P(x)
L -00 Q(x)
dx, (4.4.18)

where P, Q are polynomials with real coefficients and no common roots, the
polynomial Q has no real roots, and the degree of Q is at least two units
higher than the degree of P (to ensure the convergence of the integral). In
particular, the degree of Q must be even; think about it.
228 Singularities of Complex Functions

Such integrals are evaluated by integrating the function P(z)/Q(z) over

the curve CR on Figure 4.4.1(a), and then passing to the limit R —> oo. The
integral over CR is determined by the residues of P(z)/Q(z) at the roots of
Q that have positive imaginary part, and the difference of degrees of P and
Q ensures that the integral over the semi-circle tends to zero as R increases.
If R is sufficiently large, then the curve CR encloses all the roots of Q with
positive real part.
Compared to the integration of the trigonometric expressions, we have
more complicated theoretical considerations, such as passing to the limit as
R —> oo, but often much easier computations of the residues.
F O R EXAMPLE, consider the integral

/= r ±
J_00(2 + 2x + x*)Z-
Step 1. We consider the complex integral
IR =
"CR(2
dz
+ 2Z + Z2 2\)2 '
Jc
where the curve CR is from Figure 4.4.1(a), and R is sufficiently large.
Step 2. The function f(z) = (2 + 2z + z2)~2 we are integrating has
two second-order poles zit2 at the roots of the polynomial z2 + 2z + 2 =
(z + l ) 2 + 1, and these roots are z\ = — 1 + i, z2 = —\ — i. The root in the
upper half-plane is z\ = — 1 + i, and z\ is also inside the domain enclosed
by CR if R > y/2. As a result,

IR = 2ni Res f(z), z\ — — 1 + i.

Step 3. We notice that f(z) = (z - z{)~2{z - z<i)~2 and then find the
residue at z\ by formula (4.4.17) with N = 2 :
d(z - z2)-2
Res f(z) = - 2 ( z i - z2)~3 = -2(2i)~A = -t/4;
dz
recall that i _ 1 = —i. Accordingly, IR = n/2 for all R > y/2.
Step 4- We have IR = lR,r + lR,c, where IR^ is the integral over the real
axis from —R to R, and lRth is the integral over the circular arc. Then / =
limfi-,00 lRtT. As for i# iC , we note that \z\ = Ron the arc, and, for R > 10,
we have \z2 + 2z + 2| > \z\2 - 2\z\ -2 = R2-2R-2> R{R - 4) > R2/2.
Since the length of the arc is nR, inequality (4.2.8) on page 196 implies
ATTR
\IRA < - ^ 4 - -» 0, R - oo.
 Residue Integration 229

As a result,

lim IR =
R—+00

EXERCISE 4A.11F (a) Without using a computer, verify that

dx wy/3
I _00 (4 + 2x + x2)3 72 '

(b) Convince yourself that a computation of (4-4-18) using a mirror image

of the curve CR in the lower half-plane leads to the same final answer. Hint:
the reason is that the roots of Q come in complex conjugate pairs.
The same method applies to the computation of integrals

f°° P(x) f°° P(x)

/ cos(a:r) dx and / sinfaa;) , dx, (4.4.19)
7-oo Q{x) 7-oo Q(x)
where a is a real number, P, Q are polynomials without common roots, the
degree of Q is at least two units bigger than the degree of P, and Q has no
real roots. Assuming that a > 0, you evaluate the integral

W)dx (4A20)
OO
Aax
/

using the steps described above, and then take the real or imaginary part
of the result. The integral in (4.4.20) can exist even when the degree of Q
is only one unit bigger than the degree of P , see Problem 5.8 on page 436.
EXERCISE 4.4.12. c (a) Evaluation of the integrals in (4-4-19) will not go
through if, instead of eiaz you try to work directly with cos az or sin ax.
What goes wrong? Hint: look carefully at cosaz and sin az when \z\ = R. (b)
How should you change the integral in (4-4-20) if you want to evaluate
(4-4-19) with a > 0 by integrating over a semi-circle with Sz < 0? Hint:
you should go with e~iax. (c) Without using a computer, verify that

f
J —t
1+x
COS X
2
IT
dx = —.

Then ask your computer algebra system to evaluate this integral.

Some real integrals are evaluated using residues and integration over the
230 Singularities of Complex Functions

curve CR^ in Figure 4.4.1(b). FOR EXAMPLE, let us evaluate the integral

cos (3a;) — cos a;

dx. (4.4.21)
Jo
c
Step 0. EXERCISE 4.4.13. Verify that the integral converges. Hint:
f?° dx/x2 for every 5 > 0, and, for x near 0, cos(ax) = 1 — a x /2 + ....
Step 1. By the Integral Theorem of Cauchy, we have

J Z
CR,P

Step 2. The integrals over the parts of CR,P on the real axis result in

f—
—pp e pSix cix
3 i z _ eix rpR -Zix „ix
dx
/ P
x
*
rR / „3ix i „-3tx pix I p-ix\
= / f^—£r J J <fa - 2f, P -> o, it -> oo.
Step 3. The integral over the semi-circle of radius R tends to zero as
R —> oo. All you need to notice is that, for z — R(cos8 + ism8), we have
\piaz\ __ p — aRs'm& < i

Step 4- The integral over the semi-circle of radius p tends to — 2n as

p —> 0. Indeed, the Taylor expansion shows that, near z = 0, we have
z~2{e3lz —elz) = 2iz~l +g(z), where g(z) is analytic at z = 0. The integral
of g will then tend to zero, and the integral of 2 i z - 1 , once you take into
account only half of the circle, the clock-wise orientation and the factors of
i, produces 27r.
Step 5. By combining the results of the above steps, we conclude that

f
Jo
cos(3a;) — cos x ,
T ax = —n.

EXERCISE 4.4.14. (a)A Provide the details in the steps 3-5 above. (b)G Is
it possible to evaluate (4-4-&1) by integrating (cos(3;z) — cosz)/,? 2 over the
curve in Figure 4-4-l(a)?
With all these new techniques of integration, we should never forget the
basic rules related to symmetry. First and foremost, the integral of an odd
function (that is, a function / = f(x) satisfying f(x) = —f(—x)) over a
 Power Series and ODEs 231

symmetric interval is zero. F O R EXAMPLE,

./_«, 1 + 2x6 + 5x8

even though some computer algebra systems do not recognize this. Second,
if the original integral is not over the region we want, but otherwise is of the
suitable type, we can try and use symmetry to extend the integration to
the large interval. One basic example is /0°° F(x)dx = (1/2) J^° F(x)dx,
if the function F is even, that is, F(x) = F(—x). Still, as (4.4.21) shows,
sometimes it is better to keep the original interval. More subtle symmetry
can happen with trigonometric functions. F O R EXAMPLE, verify that
F* 1 — cos <p 1 r2v 1 — cos <p •K
V
J0 5 + 3cos<^ ~ 2 JQ 5 + 3cosy> ^ ~~ 3 '
There exist many other classes of real integrals that are evaluated using
residues, but most of the examples include roots and natural logarithms.
Evaluation of such integrals relies on the theory of multi-valued analytic
functions and is discussed in a special Complex Analysis course.

4.4.3 Power Series and Ordinary Differential Equations

Many problems in mathematics, physics, engineering, and other sciences
are reduced to a linear second-order ordinary differential equation (ODE)

A(x)y"(x) + B(x)y'(x) + C(x)y(x) = 0. (4.4.22)

Even though the solution of such equations is usually not expressed in terms
of elementary functions, a rather comprehensive theory exists describing
various properties of the solution. This theory is based on power series and
was developed in the second half of the 19th century by the German math-
ematicians LAZARUS FUCHS (1833-1902) and G E O R G FROBENIUS (1849-
1917). In what follows, we will outlines the main ideas of the theory; we
will need the results later in the study of partial differential equations.
We assume that the functions A, B, C can be extended to the complex
plane, and, instead of (4.4.22), we consider the equation in the complex
domain

w"{z) + p(z)w'(z) + q(z)w(z) = 0 (4.4.23)

for the unknown function w of the complex variable z, where p(z) =

B{z)/A(z), q(z) = C(z)/A(z).
232 Singularities of Complex Functions

To proceed, we need the following two equalities:

oo oo
£ ak(z - z0)k = Y, °n+N(z - z0)n+N, N>1, (4.4.24)
fc=JV n=0

and
oo \ / oo \ oo / fc \

( ] T ak(z - z0)k I ] T h{z - z0)k I = S 1 S a™6*-™ I (z - z

o)h-
k=0 J \k=0 J k=0 \m=0 )
(4.4.25)
EXERCISE 4.4.15? (a) Verify (4.4.24). Hint: set n = k-N, so that k = n+N.
(b) Verify (4-4-25). Hint: to get the idea, look at the first few terms by writing
(a0 + ai(z — zo) + a2(z - z$) + .. .)(6o + b\(z - z0) + bi(z — z%) + ...) and then
multiplying through.
If the functions p, q are analytic in a neighborhood of a point ZQ, then
the point z$ is called r e g u l a r for equation (4.4.23), and it is natural to
expect that all solutions of (4.4.23) are also analytic functions in some
neighborhood of ZQ. Indeed, writing
00 00 00
P(z) = ^2pk(z- z0)k, q(z) = ^qk(z-z0)k, w(z) = ^wk{z - z0)k
fc=0 k=0 k=0

and substituting into (4.4.23), we find:

w"(z) + p(z)w'(z) + q(z)w(z)

00 / 00 \ / 00 \

= J2 Kk - l)wk(z - z0)k-2 + £>*(* ~ z^ E kw z

^ ~ 2°)fc_1
fc=2 \fc=0 / \fc=l /

+ £>(*-*>)* $ > f c ( z - z 0 ) f c ) =^((k + l)(k + 2)wk+2

U=0 / \k=0 J k=0 V
fc+1 k \
+ ^ mwmpk+i-m + ^ wmqk-m J (z - z0)k = 0,
m=l m=0 /
(4.4.26)

where the last equality follows from (4.4.24) and (4.4.25).

EXERCISE 4.4.16.B Verify (4.4.26).
By Corollary 4.1 on page 212, we conclude that the coefficients wk are
 Power Series and ODEs 233

computed recursively by

Wk+2 = mw w
~(k + l)(k + 2) ( ^ mPk+i-m + Yl ™qk-m J (4.4.27)

for k = 0,1, 2, — The values of Wfc, fc > 2, are uniquely determined by the
given initial conditions wo = W(ZQ) and w\ = W'(ZQ). These computations
suggest that the following statement is true.

T h e o r e m 4.4.3 Assume that the functions p = p{z) and q — q(z) are

analytic in some domain G of the complex plain, and ZQ € G. Then, given
u>o and w\, equation (4-4-23) with the initial conditions W(ZQ) = wo and
w'(zo) = w\ has a unique solution w = w(z), and the function w = w(z) is
analytic in the domain G.

The proof of this theorem is beyond the scope of our discussion; for
proofs and extensions of this and many other results in this section, an
interested reader can consult, for example, Chapter 4 of the book Theory
of Ordinary Differential Equations by E. A. Coddington and N. Levinson,
1955.
EXERCISE 4.4.IIP Write the expressions for W2,wz, and w± without using
the Y^j sign.
Next, we will study the solutions of equation (4.4.23) near a s i n g u l a r
point, that is, a point where at least one of the functions p, q is not analytic.
As the original equation (4.4.22) suggests, singular points often correspond
to zeroes of the function A = A(z). A singular point ZQ is called a r e g u l a r
s i n g u l a r p o i n t of equation (4.4.23) if the functions (z — zo)p(z) and (z —
z 2
o) q(z) are both analytic at ZQ. In other words, the point ZQ is a regular
singular point of (4.4.23) if and only if p(z) — B(z)/(z — ZQ) and q(z) =
C(z)/(z — ZQ)2 for some functions B,C that are analytic at ZQ.
Accordingly, we now consider the equation

(z - z0)2w"(z) + (z- z0)B(z)w'(z) + C(z)w(z) = 0, (4.4.28)

and assume that the point ZQ is a regular singular point of this equation.
Being a linear second-order equation, (4.4.28) has the general solution

w(z) = AiWi{z) + A2W2(z), (4.4.29)

where Ai,A2 are arbitrary complex numbers, and W\, W2 are two linearly
independent solutions of (4.4.28); see Exercise 8.2.1, page 455. In what
234 Singularities of Complex Functions

follows, we will compute W\ and W2 using power series. While in the

regular case both W\ and W2 are analytic at ZQ, for equation (4.4.28) we
usually have at most one of the functions W\, W2 analytic at the point ZQ.
We start by looking for a solution of (4.4.28) in the form
00

w(z) = J2wk(z - z0)k+,i, (4.4.30)

fc=0

where fi and u>k, k > 0, are unknown complex numbers and WQ ^ 0. In

other words, we choose (J, so that the function (z — zo)~^"w(z) is analytic
and non-vanishing at z = ZQ.
Similar to (4.4.26), we write B(z) = £ £ L 0 M - 2 - *o)k, C(z) =
X^fcLocfc(z — zo)h, a n d substitute into (4.4.28) to conclude that
00

(z - zoY Y, ((k + n)(k + fi- l)iufe

fc=
k ° (4-4.31)
k
+ ^ ((m + /x)6fc_m + ck-m)wmj (z - z0) = 0.
m=0

EXERCISE 4.4.18.c Verify (44.31).

Similar to (4.4.27), we get a recursive system to find Wk'.
k
{k + n){k + fx- l)wk +^2{(m + n)bk-m + ck-m)wm = 0, (4.4.32)
m=0

k = 0,1,2,.... For k = 0, (4.4.32) yields (/i(/x - 1) + fib0 + c0)w0 = 0, or,

since we assumed that wo ^ 0,

fj,2 + (b0 - l)/x + co = 0. (4.4.33)

Equation (4.4.33) is called the i n d i c i a l equation of the differential equa-

tion (4.4.28). We will see that the indicial equation indicates the general
solution of (4.4.28) by providing the roots (i. Equation (4.4.33) has two so-
lutions m,H2, a n d there are two main possibilities to consider: (a) /J,\ — ^2
is not an integer; (b) \x\ — /i2 is an integer; this includes the possible double
root Hi = /i2 = (1 — bo)/2. The reason for this distinction is that, for k > 1,
equation (4.4.32) is ((/x + k)2 + (b0 - l)(fx + k) + c0)uik = Fk(w0,. •• ,Wk-\)
for some function Fk- As a result, if (fi + k)2 + (bo — l)(n + k) + Co j= 0
for every k > 1, then, starting with WQ = 1 and two different values of /J,,
we can get two different sets of the coefficients Wk, and the two linearly
 Power Series and ODEs 235

independent solutions W i , ^ , both in the form (4.4.30). If we have two

identical values of fi or if (fi + N)2 + (b0 — 1)(M + N) + c 0 = 0 for some
integer N > 1, then only one of the functions Wi,W^ will be of the form
(4.4.30), and extra effort is necessary to find the other function.
IF THE ROOTS MliM2 OF (4.4.33) DO NOT DIFFER BY AN INTEGER,
then the two linearly independent solutions W\,W2 of (4.4.28) are
oo

Wn(z) = (z-zor«Y, '«(z-Zo^>

Wk n=1
>2, (4.4.34)
fc=0

where Wfc,„, k > 1, n = 1,2, are determined recursively from (4.4.32) with
wo,n = 1. Note that this is the case when &o>co a r e rea -l a n d MiiA^ are
complex so that Mi = M2-
I F THE ROOTS MI>M2 OF (4.4.33) SATISFY MI - /z2 = N > 0, where AT
is an integer, then one solution of (4.4.28) is
00
Wi(z) = (z - z 0 ) Ml 5 > f c ( z - z 0 ) fc , (4.4.35)
fc=0

where Mi is the larger root of the indicial equation (4.4.33) and Wk, k > 1,
are determined from (4.4.32) with M = Mi and u>o = 1. To find W2, we use
L i o u v i l l e ' s formula:

Wx(z) W2(z)
exp| - / -W-dz), (4.4.36)
W{(z) W^z) V Jc(zi,zz) -2 - ZQ J

where the left-hand side is a two-by-two determinant, z\ is a fixed point

in the neighborhood of ZQ, and and C(zi,z) is a piece-wise smooth path
from z\ to z so that ZQ is not on C{z\,z). For the proof of this formula,
see an ODE textbook, such as Theory of Ordinary Differential Equations
by E. A. Coddington, and N. Levinson, 1955. By assumption, the function
B(z)/(z — ZQ) is analytic away from z§, and so the value of the integral
does not depend on the particular path. After some computations, (4.4.36)
yields

W2(z) = W1(z) JJ H(z)dz,H(z)

H(z)dz,H(z) =
= ww^-
~-)exvl-
)exp\- jJ -j^-dz
C(z2,z)
Z2,Z) \ C(zi,z)
C{zltz] /
(4.4.37)
236 Singularities of Complex Functions

EXERCISE 4.4.19/ 4 Verify (44.37). Hint: the determinant in (4-4-36) can be

written as the square ofWi(z) times the derivative of W2(z) /W\{z); C(z2,z) is a
path similar to C{z\,z), starting at zi-
By assumption, B{z)/(z - z0) = b0/(z - z0) + b\ + b2(z - z0) + •.., so
that

/ — — dz = B0 ln(z - ZQ) + ip(z),

J Z- ZQ
C(zuz)
where In z is some branch of the natural logarithm, and the function <p is
analytic at ZQ. Recalling that W\ (z) = {Z — ZQ)'11 (1+U>IZ + . ..), we conclude
that
H(z) = (z - zo)~^2,il+bo^ip(z), where the function ip is analytic at z0 and
ip(z0) ^ 0. By assumption, \i\ — fi2 = N, while equation (4.4.33) implies
Ml + A*2 = 1 - bo. Therefore, 2/ix + b0 = 1 + N. Writing
oo

i>(z) = Y,Mz-zo)k (4.4.38)

fc=0

and integrating (4.4.37), we conclude that

W2(z) = (z- z0)^h(z) + VJV WI(Z) ln(z - z0), (4.4.39)

where the function h is analytic at z0, and ipN is the coefficient of zN in

the expansion (4.4.38). Note that

• ipo ^ 0, while, for N > 0, ipN = 0 is a possibility.

• If N > 0, then h(z0) =ipo^0.
In particular, if the indicial equation (4-4-33) has a double root, then one
of the solutions of (4-4-28) always has a logarithmic term.
EXERCISE 4.4.20. (a)A Verify (44.39).
Hint: H(z) = X ^ o ^(z — zo) ~~ ~ , and ipN is the coefficient of {z — zo)~x.
(b)c Consider B e s s e l ' s d i f f e r e n t i a l equation

z2w"{z) + zw'(z) + (z2 - q)w(z) = 0, (4.4.40)

where q is complex number; the equation is named after the German as-
tronomer and mathematician FRIEDRICH WILHELM BESSEL (1784-1846).
Verify that this equation has a solution that does not have a singularity
at ZQ = 0 if and only if q = N2 for some non-negative integer N, and,
if it exists, this non-singular solution is unique up to a constant multiple.
 Power Series and ODEs 237

Hint: the inidicial equation is p.2 — q = 0; if fi = s + it for some real s, t, then

2M = zszu = zseitlnz = z3(cos(t\nz) + isin(£ln.z)), see page 215. Then use
(44.34), (4-4.35), and (44.39).
EXERCISE 4 . 4 . 2 1 . C Let w = w(z) be a solution of (44.23). Show that the
function <p(z) = w(l/z) is a solution of

<p"(z) + (2Z-1 - z-2p(l/z))<f'(z) + z-4q{l/z)<p(z) = 0. (4.4.41)

By definition, the point ZQ = 00 is a r e g u l a r s i n g u l a r p o i n t of equa-

tion (4.4.23) if and only if the point ZQ = 0 is a regular singular point of
equation (4.4.41).

EXERCISE 4.4.22.B Verify that equation

z(l - z)w"(z) + [c- (a + b+l)z]w'(z) - abw(z) = 0, a,b,c€C, (4.4.42)

has exactly three regular singular points at ZQ = 0,1,00.

Equation (4.4.42) is called the hypergeometric d i f f e r e n t i a l
equation, and has a special significance: for many second-order linear
ordinary differential equations with at most three regular singular points,
including, if necessary, ZQ = 00, and no other singular points, the solution
is expressed in terms of the solutions of (4.4.42).
We now list some other particular equations of the type (4.4.23) or
(4.4.28); all of them arise in many mathematical, physical, and engineer-
ing problems, and we will encounter some of the equations later in our
discussion of partial differential equations:

(1 - z2)w"(z) + (fi + uz) w'(z) + X w{z) = 0, (4.4.43)

2
z w"(z) + vzw'(z) + \w(z) = 0, (4.4.44)
w"(z) + vzw'(z) + X w{z) = 0, (4.4.45)
2
w"{z) - {z + X)w{z) = 0, (4.4.46)
z w"(z) + {v- z)w'{z) + X w(z), (4.4.47)

w"(z) + vzw{z) = Q, (4.4.48)

w"(z) + w'(z) + (z + X)w{z) = 0, (4.4.49)
2n
w"(z) + (X- z )w(z) = 0, (4.4.50)
w"(z) + {v cos(z) + X)w(z) = 0, (4.4.51)
238 Singularities of Complex Functions

(z2 - a2)(z2 - b2)w"(z) + z(2z2 -b2- c2)w'(z)

- ({n(n + \)z2 - (b2 + c2)p))w{z) = 0, (4.4.52)
w"(z) + w'(z) + (nz-1 + vz-2)w{z) = 0, (4.4.53)

\sur(a;z) cos2 (ax) )

*""(*) ~ (-7JT-, + X, , + A) W(Z) = 0. (4.4.55)

\sinh (az) cosh (az) J
In these equations, n is a non-negative integer, a,b,c,p are real numbers,
fi, v, A are complex numbers. For each equation, the reader is encouraged to
do the following: (a) find all singular points (the point ZQ = oo must always
be investigated), (b) find the first few terms in the expansion of the two
linearly independent solutions, around ZQ = 0 and around all other finite
regular singular points, if any.
For certain values of the parameters, the above equations can have espe-
cially interesting properties. FOR EXAMPLE, consider (4.4.47) with positive
integer v = m + 1 and non-negative integer A = n; m, n = 0,1, 2, Writ-
ing w(z) = X^fcLo wkzk a n d substituting into the equation, we find that

wk+i = {(k - n)/(k + l)(fc + m + l))wk

(verify this!). Then wn+i — 0 and therefore u>k = 0 for all k > n. In
other words, a solution of (4.4.47) is a POLYNOMIAL. This remarkable fact
certainly deserves special recognition, and (4.4.47) with v = m + 1 and
A = n is called L a g u e r r e ' s d i f f e r e n t i a l equation, after the French
mathematician EDMOND LAGUERRE (1834-1894); the polynomial solution
of (4.4.47) with v — 1, A = n, and WQ — 1 is called Laguerre' s polynomial
of degree n and denoted by Ln(z).
EXERCISE 4.4.23. c Verify that if Ln(z) satisfies zw" + (1 - z)w' + nw = 0,
then the k-th derivative L„ (z) of Ln(z), k < n, satisfies
zw" + {k + 1 - z)w' + (n - k)w = 0.
EXERCISE 4.4.24. (a)c Find the Laguerre polynomials Ln for n =
B
0,1,2,3,4. (b) The Laguerre differential equation has another solution
with a singularity at ZQ = 0. Find the type of this singularity for different
m and n. (c)A Verify that equation (4-4-43) has a polynomial solution for
the following values of /x, v, and X:
(i) n = 0, v = —1, A = n2: Chebyshev's d i f f e r e n t i a l equation of the
first kind, after the Russian mathematician PAFNUTI L'VOVICH CHEBY-
 Power Series and ODEs 239

SHEV (1821-1894), see also Problem 6.3 on page 439;

(ii) /! = 0, v = —2, A = n(n+ 1 ) : L e g e n d r e ' s d i f f e r e n t i a l equation,
after the French mathematician ADRIEN MARIE LEGENDRE (1752-1833);
(Hi) /i = 0, v = - 3 , A = n ( n + 2 ) : C h e b y s h e v ' s d i f f e r e n t i a l e q u a t i o n
of the second kind;
(iv) /j, = b — a, v = —(a + 6 + 2), A = n(n + a + b + 1 ) : J a c o b i ' s
d i f f e r e n t i a l e q u a t i o n , after C. G. J. Jacobi. Note that this includes
the previous three equations as particular cases.
(d)A Verify that equation (4-4-45) with u — — 2 and A = 2n, known
as H e r m i t e ' s d i f f e r e n t i a l e q u a t i o n , after the French mathematician
C H A R L E S H E R M I T E (1822-1901), has a polynomial solution.

Some equations have several names. For example, (4.4.47) with

general complex v a n d A is known as t h e c o n f l u e n t h y p e r g e o m e t r i c
d i f f e r e n t i a l e q u a t i o n , and, with real v a n d real negative A, as Kummer' s
d i f f e r e n t i a l e q u a t i o n , after the German mathematician E R N S T E D -
UARD K U M M E R (1810-1893). B y t h e general terminology, an XYZ func-
tion or an XYZ polynomial is a certain solution of t h e XYZ equation. T h e
book Orthogonal Polynomials by G. Szego, re-published by t h e AMS in
2003, a n d t h e Handbook of Differential Equations by D. Zwillinger, 1997,
provide more information on t h e subject. Keep in mind that the notations
and terminology can vary from source to source.

To conclude this section, we note t h a t t h e power series m e t h o d can work

for linear equations of any order. FOR EXAMPLE, consider t h e equation
w'"{z) = zw(z), with initial conditions w(0) = w'(0) = 0, w"(0) = 12. T h e
reader is encouraged t o verify t h a t w(z) = 6 z 2 + z6/20 + z 1 0 / 1 4 4 0 0 +
How will t h e answer change if w'(l) = 5 rather t h a n w'(0) = 0?
For certain NONLINEAR EQUATIONS, t h e power series m e t h o d can still
be used, because t h e product of two power series is again a power series. Of
course, there is no longer any hope of getting nice recursive relations of t h e
t y p e (4.4.27). T h e following analog of Theorem 4.4.3 holds; see Theorem
6 in Section 11, C h a p t e r 3, of t h e book Ordinary Differential Equations by
G. Birkhoff a n d G.-C. Rota, 1969.

T h e o r e m 4 . 4 . 4 If F — F(w, z) is a function of two complex variables,

analytic at (u>o, ZQ), then there exists a neighborhood of the point ZQ in which
the solution w = w(z) of the initial value problem w'(z) — F(w(z),z),
W(ZQ) — ZQ, exists, is unique, and is an analytic function of z.

Even if t h e function F is analytic everywhere, the solution of w'(z) —

240 Singularities of Complex Functions

F(w(z), z) may fail to be analytic everywhere. F O R EXAMPLE, the solution

of the equation w'{z) = w2, w(0) = 1, is w(z) = 1/(1 — z). The reader is
encouraged to derive this representation of w using power series.