Chapter 2

Analytic Functions

2.1 Functions of a Complex Variable

The concept of a complex number z was introduced in Chapter 1 in order to solve cer-
tain algebraic equations. We shall now study functions f (z) defined on these complex
variables. Our objective is to mimic the concepts, theorems, and mathematical struc-
ture of calculus; we want to differentiate and integrate f (z). The notion of a derivative
is far more subtle in the complex case because of the intrinsically two-dimensional
nature of the complex variable, and the exposition of this point will consume all of
Chapter 2. The payoff is enormous, however, and the remainder of the book will be
devoted to developing the mathematical consequences and demonstrating their appli-
cations to physical problems.
Let us begin with a careful review of the basics. Recall that a function f is a rule
that assigns to each element in a set A one and only one element in a set B. If f
assigns the value b to the element a in A, we write
b = f (a)
and call b the image of a under f . The set A is the domain of definition of f (even
if A is not a domain in the sense of Chapter 1), and the set of all images f (a) is the
range of f . We sometimes refer to f as a mapping of A into B.
Here we are concerned with complex-valued functions of a complex variable, so
that the domains of definition and the ranges are subsets of the complex numbers. If
f (z) is expressed by a formula such as
z2 − 1
f (z) = ,
z2 + 1
then, unless stated otherwise, we take the domain of f to be the set of all z for which
the formula is well defined. (Thus the domain for this f comprises all z except for
±i.)
If w denotes the value of the function f at the point z, we then write w = f (z).
Just as z decomposes into real and imaginary parts as z = x +i y, the real and imaginary

From Chapter 2 of Fundamentals of Complex Analysis with Applications to Engineering, Science, and Mathematics,
Third Edition. Edward B. Saff, Arthur David Snider. Copyright © 2003 by Pearson Education, Inc. All rights reserved.
53
 54 Analytic Functions

parts of w are each (real-valued) functions of z or, equivalently, of x and y, and so we

customarily write
w = u(x, y) + iv(x, y),

with u and v denoting the real and imaginary parts, respectively, of w. Thus a complex-
valued function of a complex variable is, in essence, a pair of real functions of two real
variables.

Example 1
Write the function w = f (z) = z 2 + 2z in the form w = u(x, y) + iv(x, y).
Solution. Setting z = x + i y we obtain

w = f (z) = (x + i y)2 + 2(x + i y) = x 2 − y 2 + i2x y + 2x + i2y.

Hence w = (x 2 − y 2 + 2x) + i(2x y + 2y) is the desired form.

Unfortunately, it is generally impossible to draw the graph of a complex function;
to display two real functions of two real variables graphically would require four di-
mensions. We can, however, visualize some of the properties of a complex function
w = f (z) by sketching its domain of definition in the z-plane and its range in the
w-plane, and depicting the relationship as in Fig. 2.1.

Figure 2.1 Representation of a complex function.

Example 2
Describe the range of the function f (z) = x 2 + 2i defined on the closed unit disk
|z| ≤ 1.
Solution. We have u(x, y) = x 2 and v(x, y) = 2. Thus as z varies over the
closed unit disk, u varies between 0 and 1, and v is constant. The range is therefore
the line segment from w = 2i to w = 1 + 2i.

54
2.1 Functions of a Complex Variable 55

Figure 2.2 Mapping of a semidisk under f (z) = z 3 .

Example 3
Describe the function f (z) = z 3 for z in the semidisk given by |z| ≤ 2, Im z ≥ 0 [see
Fig. 2.2(a)].

Solution. From Sec. 1.5 we know that the points z in the sector of the semidisk
from Arg z = 0 to Arg z = 2π/3, when cubed, cover the entire disk |w| ≤ 8. The
cubes of the remaining z-points also fall in this disk, overlapping it in the upper half-
plane, as depicted in Fig. 2.2(b).

The function f (z) = 1/z is called the inversion mapping. It is an example of a
one-to-one function because it maps distinct points to distinct points, i.e., if z 1  = z 2 ,
then f (z 1 )  = f (z 2 ).

Example 4
Show that the inversion mapping w = 1/z corresponds to a rotation of the Riemann
sphere by 180◦ about the x1 -axis (see Fig. 1.21 on page 44).
Solution. Let Z = (x1 , x2 , x3 ) denote the stereographic projection of the point
z and W = (x̂1 , x̂2 , x̂3 ) denote the projection of 1/z. We need to show that W can be
obtained by a 180◦ rotation of Z about the x1 -axis.
Referring to the formulas (1) in Sec. 1.7 we have
2 Re(z) 2 Im(z) |z|2 − 1
x1 = , x 2 = , x 3 = ;
|z|2 + 1 |z|2 + 1 |z|2 + 1
2 Re(1/z) 2 Im(1/z) |1/z|2 − 1
x̂1 = , x̂ 2 = , x̂ 3 = .
|1/z|2 + 1 |1/z|2 + 1 |1/z|2 + 1
Strictly speaking, this function is not defined at z = 0, but in light of the discussion of the point
at ∞ in Sec. 1.7, it is natural to define f (0) = ∞ and, moreover, to regard f as a function in the
extended complex plane
C = C ∪ {∞}, with f (∞) = 0.

55
 56 Analytic Functions

Since Re(1/z) = (Re z)/|z|2 and Im(1/z) = −(Im z)/|z|2 , we get after simplifi-
cation that
2 Re z −2 Im z 1 − |z|2
x̂1 = = x 1 , x̂ 2 = = −x 2 , and x̂ 3 = = −x3 .
1 + |z|2 1 + |z|2 1 + |z|2
A rotation about the x1 -axis preserves x1 while negating x2 and x3 ; so indeed W is the
stated rotation of Z .

One nice consequence of this example is the fact that an inversion mapping pre-
serves the class of circles and lines (see Prob. 17)

EXERCISES 2.1

1. Write each of the following functions in the form w = u(x, y) + iv(x, y).

(a) f (z) = 3z 2 + 5z + i + 1 (b) g(z) = 1/z

z+i 2z 2 + 3
(c) h(z) = 2 (d) q(z) =
z +1 |z − 1|
(e) F(z) = e3z (f) G(z) = e z + e−z

2. Find the domain of definition of each of the functions in Prob. 1.

3. Describe the range of each of the following functions.

(a) f (z) = z + 5 for Re z > 0

(b) g(z) = z 2 for z in the first quadrant, Re z ≥ 0, Im z ≥ 0
1
(c) h(z) = for 0 < |z| ≤ 1
z
π
(d) p(z) = −2z 3 for z in the quarter-disk |z| < 1, 0 < Arg z <
2
4. Show that the inversion mapping w = f (z) = 1/z maps

(a) the circle |z| = r onto the circle |w| = 1/r ;

(b) the ray Arg z = θ0 , −π < θ0 < π , onto the ray Arg w = −θ0 ;
(c) the circle |z − 1| = 1 onto the vertical line x = 1/2.

5. For the complex exponential function f (z) = e z defined in Sec. 1.4:

(a) Describe the domain of definition and the range.

(b) Show that f (−z) = 1/ f (z).
(c) Describe the image of the vertical line Re z = 1.
(d) Describe the image of the horizontal line Im z = π/4.
(e) Describe the image of the infinite strip 0 ≤ Im z ≤ π/4.

56
2.1 Functions of a Complex Variable 57

6. The Joukowski mapping is defined by

 
1 1
w = J (z) = z+ .
2 z
Show that

(a) J (z) = J (1/z).

(b) J maps the unit circle |z| = 1 onto the real interval [−1, 1].
(c) J maps the circle |z| = r (r > 0, r = 1) onto the ellipse

u2 v2
  2 +   2 = 1,
1
2 r + 1
r
1
2 r − 1
r

which has foci at ±1.

7. A function of the form F(z) = z + c, where c is a complex constant, generates

a translation mapping. Sketch the image of the semidisk |z| ≤ 2, Im z ≥ 0, [see
Fig. 2.2(a)] under F when (a) c = 3; (b) c = 2i; (c) c = −1 − i.
8. A function of the form G(z) = eiφ z, where φ is a real constant, generates a rotation
mapping. Sketch the image of the semidisk |z| ≤ 2, Im z ≥ 0 [see Fig. 2.2(a)]
under G when (a) φ = π/4; (b) φ = −π/4; (c) φ = 3π/4.
9. A function of the form H (z) = ρz, where ρ is a positive real constant, generates a
magnification mapping when ρ > 1 and a reduction mapping when ρ < 1. Sketch
the image of the semidisk |z| ≤ 2, Im z ≥ 0 [see Fig. 2.2(a)] under H when (a)
ρ = 3; (b) ρ = 1/2.
10. Let F(z) = z + i, G(z) = eiπ/4 z, and H (z) = z/2. Sketch the image of the
semidisk |z| ≤ 2, Im z ≥ 0 [see Fig. 2.2(a)] under each of the following composite
mappings:

(a) G(F(z)) (b) G(H (z))

(c) H (F(z)) (d) F(G(H (z)))

11. Let F(z) = z − 3, G(z) = −i z, and H (z) = 2z. Sketch the image of the circle
|z| = 1 under each of the following composite mappings:

(a) G(F(z)) (b) G(H (z))

(c) H (F(z)) (d) F(G(H (z)))

12. A function of the form f (z) = az+b, where a and b are complex constants, is called
a linear transformation. Show that every linear transformation can be expressed as
the composition of a magnification (or reduction; Prob. 9), a rotation (Prob. 8), and
a translation (Prob. 7). Deduce from this that a linear transformation maps lines to
lines and circles to circles. [HINT: Write a in polar form.]

57
 58 Analytic Functions

13. Show that the function w = z 2 maps

(a) the line x = 1,
(b) the hyperbola x y = 1, and
(c) the circle |z − 1| = 1,
into a parabola, a straight line, and the cardioid w = 2(1 + cos θ )eiθ , respectively,
in the w-plane.
14. (Rotation of the Riemann Sphere) Referring to Fig. 1.21 and Eqs. (1) in Sec. 1.7 for
stereographic projection, show each of the following:
(a) The mapping w = eiφ z corresponds to a rotation of the Riemann sphere about
the x3 -axis through an angle φ.
(b) The mapping w = −1/z corresponds to a 180◦ rotation of the Riemann sphere
about the x2 -axis (the imaginary axis).
15. Show that the mapping w = (1 + z)/(1 − z) corresponds to a 90◦ counterclockwise
rotation of the Riemann sphere about the x2 -axis.
16. Describe the mapping w = (1 − i z)/(z − i) in terms of a suitable rotation of the
Riemann sphere.
17. Use the result of Example 4 and properties of stereographic projection to show that
the inversion w = 1/z maps any circle in the z-plane to either a circle or a line in
the w-plane, and the same holds for the mapping of any line in the z-plane. (Regard
∞ as a point on every line.)

2.2 Limits and Continuity

As we observed in Chapter 1, the definition of absolute value can be used to designate
the distance between two complex numbers. Having a concept of distance, we can
proceed to introduce the notions of limit and continuity.
Informally, when we have an infinite sequence z 1 , z 2 , z 3 , . . . of complex numbers,
we say that the number z 0 is the limit of the sequence if the z n eventually (i.e., for
large enough n) stay arbitrarily close to z 0 . More precisely, we state

Definition 1. A sequence of complex numbers {z n }∞

1 is said to have the limit z 0
or to converge to z 0 , and we write

lim z n = z 0
n→∞

or, equivalently,
zn → z0 as n → ∞,
if for any ε > 0 there exists an integer N such that |z n − z 0 | < ε for all n > N .

58
2.2 Limits and Continuity 59

Figure 2.3 A convergent sequence.

Geometrically, this means that each term z n , for n > N , lies in the open disk of
radius ε about z 0 (see Fig. 2.3).

Example 1
Find the limit (if it exists) of the sequence
 n
i 2 + in
(a) z n = ; (b) z n = ; (c) z n = i n .
3 1 + 3n
Solution. We use methods familiar from elementary calculus that can be rigor-
ously justified from Definition 1.
(a) Since |(i/3)n | = 1/3n → 0, it follows (see Prob. 5) that
 n
i
lim = 0.
n→∞ 3

(b) Dividing numerator and denominator by n we get

2 + in (2/n) + i 0+i i
= → = as n → ∞.
1 + 3n (1/n) + 3 0+3 3
(c) The sequence i n consists of infinitely many repetitions of i, −1, −i, and 1. Thus
this sequence does not have a limit.

A related concept is the limit of a complex-valued function f (z). Roughly speak-

ing, we say that the number w0 is the limit of the function f (z) as z approaches z 0 , if
f (z) stays arbitrarily close to w0 whenever z is sufficiently near z 0 . In precise terms
we give

Definition 2. Let f be a function defined in some neighborhood of z 0 , with the

possible exception of the point z 0 itself. We say that the limit of f (z) as z
approaches z 0 is the number w0 and write

lim f (z) = w0
z→z 0

59
 60 Analytic Functions

or, equivalently,
f (z) → w0 as z → z0,
if for any ε > 0 there exists a positive number δ such that

| f (z) − w0 | < ε whenever 0 < |z − z 0 | < δ.

Geometrically, this says that any neighborhood of w0 contains all the values as-
sumed by f in some full neighborhood of z 0 , except possibly the value f (z 0 ); see
Fig. 2.4.

Figure 2.4 Mapping property of a function with limit w0

as z → z 0 .

Example 2
Use Definition 2 to prove that limz→i z 2 = −1.
Solution. We must show that for given ε > 0 there is a positive number δ such
that
|z 2 − (−1)| < ε whenever 0 < |z − i| < δ.
So we express |z 2 − (−1)| in terms of |z − i|:

z 2 − (−1) = z 2 + 1 = (z − i)(z + i) = (z − i)(z − i + 2i).

It follows from the properties of absolute value derived in Sec. 1.3 (in particular, the
triangle inequality) that
 
 2 
z − (−1) = |z − i||z − i + 2i| ≤ |z − i|(|z − i| + 2). (1)

Now if |z − i| < δ the right-hand member of (1) is less than δ(δ + 2); so to ensure that
it is less than ε, we choose δ to be smaller than each either of the numbers ε/3 and 1:
ε
|z − i|(|z − i| + 2) < (1 + 2) = ε.

3

60
2.2 Limits and Continuity 61

There is an obvious relation between the limit of a function and the limit of a
sequence; namely, if limz→z 0 f (z) = w0 , then for every sequence {z n }∞ 1 converging to
z 0 (z n  = z 0 ) the sequence { f (z n )}∞
1 converges to w0 . The converse of this statement
is also valid and is left as an exercise.
The condition of continuity is expressed in

Definition 3. Let f be a function defined in a neighborhood of z 0 . Then f is

continuous at z 0 if
lim f (z) = f (z 0 ).
z→z 0

In other words, for f to be continuous at z 0 , it must have a limiting value at z 0 ,

and this limiting value must be f (z 0 ).
A function f is said to be continuous on a set S if it is continuous at each point
of S.
Clearly the definitions of this section are direct analogues of concepts introduced in
elementary calculus. In fact, one can show that f (z) approaches a limit precisely when
its real and imaginary parts approach limits (see Prob. 18); similarly, the continuity of
the latter functions is equivalent to the continuity of f . Because of the analogy, many
of the familiar theorems on real sequences, limits, and continuity remain valid in the
complex case. Two such theorems are stated here.

Theorem 1. If limz→z 0 f (z) = A and limz→z 0 g(z) = B, then

(i) lim ( f (z) ± g(z)) = A ± B,

z→z 0

(ii) lim f (z)g(z) = AB,

z→z 0

f (z) A
(iii) lim = if B  = 0.
z→z 0 g(z) B

Theorem 2. If f (z) and g(z) are continuous at z 0 , then so are f (z) ± g(z) and
f (z)g(z). The quotient f (z)/g(z) is also continuous at z 0 provided g(z 0 )  = 0.

(Theorem 2 is an immediate consequence of Theorem 1.)

One can easily verify that the constant functions as well as the function f (z) =
z are continuous on the whole plane C. Thus from Theorem 2 we deduce that the
polynomial functions in z, i.e., functions of the form

a0 + a1 z + a2 z 2 + · · · + an z n ,

61
 62 Analytic Functions

where the ai are constants, are also continuous on the whole plane. Rational functions
in z, which are defined as quotients of polynomials, i.e.,
a0 + a1 z + · · · + an z n
,
b0 + b1 z + · · · + bm z m
are therefore continuous at each point where the denominator does not vanish. These
considerations provide a much simpler solution for problems such as Example 2, as
we illustrate next.

Example 3
Find the limits, as z → 2i, of the functions f 1 (z) = z 2 − 2z + 1, f 2 (z) = (z + 2i)/z,
and f 3 (z) = z 2 + 4 /z(z − 2i).
Solution. Since f 1 (z) and f 2 (z) are continuous at z = 2i, we simply evaluate
them there, i.e.,
lim f 1 (z) = f 1 (2i) = (2i)2 − 2(2i) + 1 = −3 − 4i,
z→2i
2i + 2i
lim f 2 (z) = f 2 (2i) = = 2.
z→2i 2i
The function f 3 (z) is not continuous at z = 2i because it is not defined there (the
denominator vanishes). However, for z  = 2i and z  = 0 we have
(z + 2i)(z − 2i) z + 2i
f 3 (z) = = = f 2 (z),
z(z − 2i) z
and so
lim f 3 (z) = lim f 2 (z) = 2.

z→2i z→2i

Note that in the preceding example the discontinuity of f 3 (z) at z = 2i can be

removed by suitably defining the function at this point [set f 3 (2i) = 2]. In general, if
a function can be defined or redefined at a single point z 0 so as to be continuous there,
we say that this function has a removable discontinuity at z 0 .
We shall see that limits involving infinity are very useful in describing the behavior
of certain sequences and functions. We say “z n → ∞” if, for each positive number
M (no matter how large), there is an integer N such that |z n | > M whenever n > N ;
similarly “limz→z 0 f (z) = ∞” means that for each positive number M (no matter how
large), there is a δ > 0 such that | f (z)| > M whenever 0 < |z − z 0 | < δ. Essentially
we are saying that complex numbers approach infinity when their magnitudes approach
infinity. Therefore
z iz − 2 i z 3 + 3i
lim = ∞, lim = , lim = ∞.
z→3i z 2 +9 z→∞ 4z + i 4 z→∞ z 2 + 5z

In fact, the concept of a point at infinity, as introduced in Sec. 1.7, quantifies this notion
nicely: see Probs. 23–25.

62
2.2 Limits and Continuity 63

In closing this section, we wish to emphasize an important distinction between

the concepts of limit in the (one-dimensional) real and complex cases. For the latter
situation, observe that a sequence {z n }∞
1 may approach a limit z 0 from any direction in
the plane, or even along a spiral, etc. Thus the manner in which a sequence of numbers
approaches its limit can be much more complicated in the complex case.

EXERCISES 2.2

1. Sketch the first five terms of the sequence (i/2)n , n = 1, 2, 3, . . ., and then describe
the convergence of this sequence.
2. Sketch the first five terms of the sequence (2i)n , n = 1, 2, 3, . . ., and then describe
the divergence of this sequence.
3. Using Definition 1, prove that the sequence of complex numbers z n = xn + i yn
converges to z 0 = x0 + i y0 if and only if xn converges to x0 and yn converges to y0 .
[HINT: |xn −x0 | ≤ |z n −z 0 |, |yn −y0 | ≤ |z n −z 0 |, and |z n −z 0 | ≤ |xn −x0 |+|yn −y0 |.]
4. Prove that z n → z 0 if and only if z n → z 0 as n → ∞.
5. Prove that limn→∞ z n = 0 if and only if limn→∞ |z n | = 0 using Definition 1.
6. Prove that if |z 0 | < 1, then z 0n → 0 as n → ∞. Also prove that if |z 0 | > 1, then the
sequence z 0n diverges.
7. Decide whether each of the following sequences converges, and if so, find its limit.
 
i i
(a) z n = (b) z n = i(−1) n (c) z n = Arg −1 +
n  n  n
n(2 + i) 1−i 2nπi
(d) z n = (e) z n = (f) z n = exp
n+1 4 5
8. Use Definition 2 to prove that limz→1+i (6z − 4) = 2 + 6i.
9. Use Definition 2 to prove that limz→−i 1/z = i.
10. Use Theorem 1 to prove Theorem 2.
11. Find each of the following limits.

z2 + 3
(a) lim (z − 5i)2 (b) lim
z→2+3i z→2 iz
z2 +9 z2 + i
(c) lim (d) lim 4
z→3i z − 3i z→i z − 1
(z 0 + z)2 − z 02  
(e) lim (f) lim z 2 − 1
z→0 z z→1+2i

12. Show that the function Arg z is discontinuous at each point on the nonpositive real
axis.

63
 64 Analytic Functions

13. Let f (z) be defined by

2z/(z + 1) if z = 0,
f (z) =
1 if z = 0.

At which points does f (z) have a finite limit, and at which points is it continuous?
Which of the discontinuities of f (z) are removable?
14. Prove that the function g(z) = z̄ is continuous on the whole plane.
15. Prove that if f (z) is continuous at z 0 , then so are the functions f (z), Re f (z),
Im f (z), and | f (z)|. [HINT: To show that | f (z)| is continuous at z 0 use inequality
(2) in Sec. 1.3.]
16. Let g be a function defined in a neighborhood of z 0 and let f be a function defined
in a neighborhood of the point g(z 0 ). Show that if g is continuous at z 0 and f is
continuous at g(z 0 ), then the composite function f (g(z)) is continuous at z 0 .
17. Let f (z) = x 2 /(x 2 + y 2 ) + 2i. Does f have a limit at z = 0? [HINT: Inves-
tigate { f (z n )} for sequences {z n } approaching 0 along the real and imaginary axes
separately.]
18. Let f (z) = u(x, y) + iv(x, y), z 0 = x0 + i y0 , and w0 = u 0 + iv0 . Prove that

lim f (z) = w0
z→z 0

if, and only if,

lim u(x, y) = u 0 and lim v(x, y) = v0 .

x→x0 x→x0
y→y0 y→y0

[HINT: First show that f (z) → w0 if and only if f (z) → w0 as z → z 0 , and then
use Theorem 1.]

19. Use Prob. 18 to find limz→1−i x x 2 + 3y + i x y.
20. Use Prob. 18 to prove that f (z) = e z is continuous everywhere.
21. Find each of the following limits:

(a) lim e z (b) lim e z − e−z

z→0 z→2πi  
z2 + π 2
(c) lim (z + 1)e z (d) lim exp
z→πi/2 z→−πi z + πi

22. Show that if limn→∞ f (z n ) = w0 for every sequence {z n }∞ 1 converging to z 0 (z n  =

z 0 ), then limz→z 0 f (z) = w0 . [HINT: Show that if the latter were not true, then one
could construct a sequence {z n }∞ 1 violating the hypothesis.]
23. Show that the definition of “z n → ∞” in the text is equivalent to the following: A
sequence of complex numbers {z n }∞
1 is said to have the limit ∞ if lim χ [z n , ∞] =
n→∞
0, where χ denotes the chordal distance (chi metric) (see page 49).

64
2.3 Analyticity 65

24. Show that the definition of “limz→z 0 f (z) = ∞” in the text is equivalent to: The
limit of f (z) as z → z 0 is ∞ if limz→z 0 χ [ f (z), ∞] = 0, where χ denotes the
chordal distance (as in the previous problem).
25. Find each of the following limits involving infinity.

z2 + 9 3z 2 − 2z
(a) lim (b) lim
z→2i 2z 2 + 8 z→∞ z2 − i z + 8

3z
(c) lim (d) lim (8z 3 + 5z + 2) (e) lim e z
z→5 z 2 − (5 − i)z − 5i z→∞ z→∞

2.3 Analyticity
Now that we have a secure notion of functions of a complex variable, we are ready
to turn to the main topic of this book—the theory of analytic functions. Before we
proceed with the rigorous exposition, however, it will prove useful for the reader’s
perspective if we give an informal preview of what it is we want to achieve.
So far we have viewed a complex function of a complex variable, f (z), as nothing
more than an arbitrary mapping from the x y-plane to the uv-plane. We have individual
names for the real and imaginary parts of z (x and y, respectively) and for the real
and imaginary parts of f (u and v); and any pair u(x, y) and v(x, y) of two-variable
functions gives us a complex function (u + iv) in this sense. But notice that there is
something special about the pair

u 1 (x, y) = x 2 − y 2 , v1 (x, y) = 2x y,

as opposed to (say)

u 2 (x, y) = x 2 − y 2 , v2 (x, y) = 3x y;

namely, the complex function u 1 + iv1 treats z = x + i y as a single “unit,” because

it equals x 2 − y 2 + i2x y = (x + i y)2 and thus it respects the complex structure of
z = x + i y. However (apparently, at least), the formulation of u 2 + iv2 requires us to
break apart the real and imaginary parts of z. √
In (real) calculus we don’t deal with functions that look at a number like 3 + 4 2
and square the 3 but cube the 4. The interesting calculus functions treat the number
as an indivisible module. We seek to classify the complex functions that behave this
same way with regard to their complex argument. Thus we want to admit functions
such as
z = x + iy (admissible),
z 2 = x 2 − y 2 + i2x y (admissible),
z 3 = x 3 − 3x y 2 + i(3x 2 y − y 3 ) (admissible),
y
1/z = x 2 +y
x
2 − i x 2 +y 2 (admissible),

65
 66 Analytic Functions

and their basic arithmetic combinations (sums, products, quotients, powers, and roots)
but ban such functions as
Re z = x (inadmissible),
Im z = y (inadmissible),
x 2 − y 2 + i3x y (inadmissible).
Notice that we will have to ban the conjugate function z̄, because if we admit it we
will open the gate to x [= (z + z̄)/2] and y [= (z − z̄)/2i]:
z̄ = x − i y (inadmissible).
Similarly, admitting the modulus |z| would be a mistake as well, since z̄ = |z|2 /z:
|z| (inadmissible).
One could criticize our “inadmissible” classification of u 2 +iv2 = x 2 − y 2 +i3x y,
because we have not yet proved that it cannot be written in terms of z alone. The
following computation is instructive: we set
x = (z + z̄)/2, y = (z − z̄)/2i (1)
in u 2 + iv2 and obtain, after some algebra,
(z + z̄)2 (z − z̄)2 (z + z̄) (z − z̄)
u 2 + iv2 = x 2 − y 2 + i3x y = − + i3
4 (−4) 2 2i
5 1
= z 2 − z̄ 2 .
4 4
Now we see that if we admit u 2 + iv2 , we would have to admit z̄ 2 [since it equals
5z 2 − 4(u 2 + iv2 )] and its undesirable square root z̄.

Example 1
Express the following functions in terms of z and z̄:
x − 1 − iy
f 1 (z) = , f 2 (z) = x 2 + y 2 + 3x + 1 + i3y.
(x − 1)2 + y 2
Solution. Using relations (1) we obtain
z + z̄ z − z̄
−1−i
f 1 (z) =  2 2i
2  
z + z̄ z − z̄ 2
−1 +
2 2i
z̄ − 1 1
= = ,
z z̄ − z − z̄ + 1 z−1

   
(z + z̄)2 (z − z̄)2 z + z̄ z − z̄
f 2 (z) = + +3 + 1 + i3
4 4i 2 2 2i
= z z̄ + 3z + 1.

66
2.3 Analyticity 67

Clearly we will want to accept f 1 as admissible, but the presence of z̄ in f 2 dis-

qualifies it. However, this procedure—the disqualification of functions with z̄ in their
formulas—does not lead to a workable criterion. For instance, who among us would
recognize that the function

z 2 z̄ 2 + z 2 + z̄ 2 − 2z̄z 2 − 2z̄ + 1
10z̄ + z z̄ 2 − 2z z̄ − 5z̄ 2 + z − 5

has a canceling common factor of (z̄ − 1)2 in its numerator and denominator and thus
is admissible?
The function e z is even more vexing. The definition we have adopted separates the
real and imaginary parts of z:

e z = e x (cos y + i sin y). (2)

But recall (Example 1, Sec. 1.4) that this definition was shown to be consistent with
the Taylor expansion for e x ,

z2 z3
ez = 1 + z + + + ··· . (3)
2! 3!
As the right-hand side of (3) appears to respect the complex structure of z, we suspect
e z to be admissible. This is indeed the case, but we postpone the official verification
until the next section.
Over the next four chapters we will see that the criterion we are seeking—the test
that will distinguish the admissible functions from the others—can be expressed sim-
ply in terms of differentiability. The following definition is a straightforward extension
of the definition in the real case and appears innocuous enough.

Definition 4. Let f be a complex-valued function defined in a neighborhood of

z 0 . Then the derivative of f at z 0 is given by
df f (z 0 + z) − f (z 0 )
(z 0 ) ≡ f  (z 0 ) := lim ,
dz z→0 z
provided this limit exists. (Such an f is said to be differentiable at z 0 .)

The catch here is that z is a complex number, so it can approach zero in many
different ways (from the right, from below, along a spiral, etc.); but the difference
quotient must tend to a unique limit f  (z 0 ) independent of the manner in which z →
0. Let us see why this notion disqualifies z̄.

Example 2
Show that f (z) = z is nowhere differentiable.

67
 68 Analytic Functions

Figure 2.5 Horizontal and vertical approach to zero of z.

Solution. The difference quotient for this function takes the form

f (z 0 + z) − f (z 0 ) (z 0 + z) − z 0 z
= = .
z z z
Now if z → 0 through real values, then z = x (see Fig. 2.5) and z = z, so
the difference quotient is 1. On the other hand, if z → 0 from above, then z = iy
and z = −z, so the quotient is −1. Consequently there is no way of assigning a
unique value to the derivative of z̄ at any point. Hence z̄ is not differentiable.

A similar analysis demonstrates that neither x, y, nor |z| is differentiable (see
Prob. 4).
Let us reassure ourselves that the elementary functions such as sums, products,
and quotients of powers of z are differentiable.

Example 3
Show that, for any positive integer n,
d n
z = nz n−1 . (4)
dz
Solution. Using the binomial formula (Prob. 27 in Exercises 1.1) we find
n(n−1) n−2
(z + z)n − z n nz n−1 z + z (z)2 + · · · + (z)n
= 2
.
z z
Thus
d n n(n − 1) n−2
z = lim [nz n−1 + z z + · · · + (z)n−1 ] = nz n−1 .

dz z→0 2

Notice that the proof was just the same as for the real-variable case. In fact the
validity of any of the following rules can be proven from Definition 4 by mimicking
the corresponding proof from elementary calculus.

68
2.3 Analyticity 69

Theorem 3. If f and g are differentiable at z, then

( f ± g) (z) = f  (z) ± g  (z), (5)

(c f ) (z) = c f  (z) (for any constant c), (6)

( f g) (z) = f (z)g  (z) + f  (z)g(z), (7)
 
f g(z) f  (z) − f (z)g  (z)
(z) = if g(z)  = 0. (8)
g g(z)2
If g is differentiable at z and f is differentiable at g(z), then the chain rule holds:
d
f (g(z)) = f  (g(z))g  (z). (9)
dz

The reader will also not be surprised to learn that differentiability implies continu-
ity, as in the real case (see Prob. 3).
It follows from Example 3 and rules (5) and (6) that any polynomial in z,
P(z) = an z n + an−1 z n−1 + · · · + a1 z + a0 ,
is differentiable in the whole plane and that its derivative is given by
P  (z) = nan z n−1 + (n − 1)an−1 z n−2 + · · · + a1 .
Consequently, from rule (8), any rational function of z is differentiable at every point
in its domain of definition. We see then that for purposes of differentiation, polynomial
and rational functions in z can be treated as if z were a real variable.

Example 4
Compute the derivative of
 100
z2 − 1
f (z) = .
z2 + 1
Solution. Unless z = ±i (where the denominator is zero), the usual calculus
rules apply. Thus
 99
 z2 − 1 (z 2 + 1)2z − (z 2 − 1)2z (z 2 − 1)99
f (z) = 100 2 = 400z 2 .

z +1 (z + 1)
2 2 (z + 1)101

As we demonstrate in Prob. 10, it is possible for a complex function to be differ-

entiable solely at isolated points. Of course, this also occurs in real analysis. Such
functions are treated there as exceptional cases, while the general theorems usually
apply only to functions differentiable over open intervals of the real line. By analogy,
then, we distinguish a special class of complex functions in

69
 70 Analytic Functions

Definition 5. A complex-valued function f (z) is said to be analytic on an open

set G if it has a derivative at every point of G.

We emphasize that analyticity is a property defined over open sets, while differen-
tiability could conceivably hold at one point only. Occasionally, however, we shall use
the abbreviated phrase “ f (z) is analytic at the point z 0 ” to mean that f (z) is analytic
in some neighborhood of z 0 . A point where f is not analytic but which is the limit of
points where f is analytic is known as a singular point or singularity. Thus we can
say that a rational function of z is analytic at every point for which its denominator
is nonzero, and the zeros of the denominator are singularities. If f (z) is analytic on
the whole complex plane, then it is said to be entire. For example, all polynomial
functions of z are entire.
As we shall see in the next few chapters, analyticity is the criterion that we have
been seeking, for functions to respect the complex structure of the variable z. In fact,
Sec. 5.2 will demonstrate that all analytic functions can be written in terms of z alone
(no x, y, or z̄).
When a function is given in terms of real and imaginary parts as u(x, y)+iv(x, y),
it may be very tedious to apply the definition to determine if f is analytic. In the next
section we will establish a test that is easier to use. Also, we will verify the analyticity
of e z . We will then have no further occasion for using the substitution method based
on Eqs. (1).

EXERCISES 2.3
1. Let f (z) be defined in a neighborhood of z 0 . Show that finding
lim [ f (z 0 + z) − f (z 0 )] /z
z→0

is equivalent to finding
lim [ f (z) − f (z 0 )]/(z − z 0 ).
z→z 0

2. Prove that if f (z) is differentiable at z 0 , then

f (z) = f (z 0 ) + f  (z 0 )(z − z 0 ) + λ(z)(z − z 0 ),
where λ(z) → 0 as z → z 0 .
3. Prove that if f (z) is differentiable at z 0 , then it is continuous at z 0 . [HINT: Use the
result of Prob. 2.]
4. Using Definition 4, show that each of the following functions is nowhere differen-
tiable.
Some authors use the words holomorphic or regular instead of analytic. The terminology “ana-
lytic function” was first used by Marquis de Condorcet (1743–1794.)

70
2.3 Analyticity 71

(a) Re z (b) Im z (c) |z|

5. Prove rules (5) and (7).

6. Prove that formula (4) is also valid for negative integers n.
7. Use rules (5)–(9) to find the derivatives of the following functions.
−6
(a) f (z) = 6z 3 + 8z 2 + i z + 10 (b) f (z) = z 2 − 3i
z2 − 9 (z + 2)3
(c) f (z) = 3 (d) f (z) =
i z + 2z + π z2 + i z + 1
4
4
(e) f (z) = 6i z 3 − 1 (z 2 + i z)100
8. (Geometric Interpretation of f  ) Suppose that f is analytic at z 0 and f  (z 0 ) = 0.
Show that
| f (z) − f (z 0 )|   
lim = f (z 0 )
z→z 0 |z − z 0 |
and
lim {arg [ f (z) − f (z 0 )] − arg (z − z 0 )} = arg f  (z 0 ).
z→z 0
Thus, on setting w = f (z) and w0 = f (z 0 ) we see that for z near z 0 , the mapping
f dilates distances by the factor | f  (z 0 )|:
|w − w0 | ≈ | f  (z 0 )| × |z − z 0 |.
Also, f rotates vectors emanating from z 0 by an angle of arg f  (z 0 ):
arg (w − w0 ) ≈ arg (z − z 0 ) + arg f  (z 0 ).
In other words, for z near z 0 the mapping w = f (z) behaves like the linear trans-
formation
w = f (z 0 ) + f  (z 0 ) (z − z 0 )
= c + eiφ ρ (z − z 0 ) .
(See Prob. 12, Exercises 2.1.)
9. For each of the following determine the points at which the function is not analytic.
1 i z 3 + 2z
(a) (b)
z − 2 + 3i z2 + 1
3z − 1 −2
(c) (d) z 2 2z 2 − 3z + 1
z2+z+4
10. Let f (z) = |z|2 . Use Definition 4 to show that f is differentiable at z = 0 but is not
differentiable at any other point. [HINT: Write
|z 0 + z|2 − |z 0 |2 (z 0 + z) z 0 + z − z 0 z 0
=
z z
z
= z 0 + z + z 0 .]
z

71
 72 Analytic Functions

11. Discuss the analyticity of each of the following functions.

z z 3 + 2z + i
(a) 8z̄ + i (b) (c) (d) x 2 − y 2 + 2x yi
z̄ + 2 z−5
   
x y
(e) x 2 + y 2 + y − 2 + i x (f) x + 2 + i y −
x + y2 x 2 + y2

|z| + z
(g) |z|2 + 2z (h)
2

12. Let P(z) = (z − z 1 )(z − z 2 ) · · · (z − z n ). Show by induction on n that

P  (z) 1 1 1
= + + ··· + .
P(z) z − z1 z − z2 z − zn
[NOTE: P  (z)/P(z) is called the logarithmic derivative of P(z).]
13. Let f (z) and g(z) be entire functions. Decide which of the following statements are
always true.

(a) f (z)3 is entire. (b) f (z)g(z) is entire.

(c) f (z)/g(z) is entire. (d) 5 f (z) + ig(z) is entire.
(e) f (1/z) is entire. (f) g(z 2 + 2) is entire.
(g) f (g(z)) is entire.

14. Prove L’Hôpital’s rule: If f (z) and g(z) are analytic at z 0 and f (z 0 ) = g(z 0 ) = 0,
but g  (z 0 ) = 0, then
f (z) f  (z 0 )
lim =  .
z→z 0 g(z) g (z 0 )
[HINT: Write
f (z) f (z) − f (z 0 )  g(z) − g(z 0 )
= .]
g(z) z − z0 z − z0

15. Use L’Hôpital’s rule to find limz→i (1 + z 6 )/(1 + z 10 ).

√ √
16. Let f (z) = z 3 + 1, and let z 1 = (−1 + 3i)/2, z 2 = (−1 − 3i)/2. Show that
there is no point w on the line segment from z 1 to z 2 such that

f (z 2 ) − f (z 1 ) = f  (w) (z 2 − z 1 ).

This shows that the mean-value theorem of calculus does not extend to complex
functions.
17. Let F(z) = f (z)g(z)h(z), where f , g, and h are each differentiable at z 0 . Prove
that

F  (z 0 ) = f  (z 0 )g(z 0 )h(z 0 ) + f (z 0 )g  (z 0 )h(z 0 ) + f (z 0 )g(z 0 )h  (z 0 ).

Guillaume De L’Hôpital (1661–1704) wrote the first textbook on differential calculus.

72
2.4 The Cauchy-Riemann Equations 73

2.4 The Cauchy-Riemann Equations

The property of analyticity for a function indicates some type of connection between
its real and imaginary parts. The precise expression of this kinship is easily derived,
as we shall see shortly, by letting z approach zero from the right and from above in
Definition 4. In this section we shall explore the nature of this relationship.
If the function f (z) = u(x, y) + iv(x, y) is differentiable at z 0 = x0 + i y0 , then
the limit
f (z 0 + z) − f (z 0 )
f  (z 0 ) = lim
z→0 z
can be computed by allowing z (= x +iy) to approach zero from any convenient
direction in the complex plane. If it approaches horizontally, then z = x, and we
obtain
u (x0 + x, y0 ) + iv (x0 + x, y0 ) − u (x0 , y0 ) − iv (x0 , y0 )
f  (z 0 ) = lim
x→0 x
   
u (x0 + x, y0 ) − u (x0 , y0 ) v (x0 + x, y0 ) − v (x0 , y0 )
= lim + i lim .
x→0 x x→0 x
(It may be helpful to consult Fig. 2.5 again.) Since the limits of the bracketed ex-
pressions are just the first partial derivatives of u and v with respect to x, we deduce
that
∂u ∂v
f  (z 0 ) = (x0 , y0 ) + i (x0 , y0 ) . (1)
∂x ∂x
On the other hand, if z approaches zero vertically, then z = iy and we have
   
 u (x0 , y0 + y) − u (x0 , y0 ) v (x0 , y0 + y) − v (x0 , y0 )
f (z 0 ) = lim + i lim .
y→0 iy y→0 iy
Hence
∂u ∂v
f  (z 0 ) = −i (x0 , y0 ) + (x0 , y0 ) . (2)
∂y ∂y
But the right-hand members of Eqs. (1) and (2) are equal to the same complex number
f  (z 0 ), so by equating real and imaginary parts we see that the equations
∂u ∂v ∂u ∂v
= , =− (3)
∂x ∂y ∂y ∂x
must hold at z 0 = x0 + i y0 . Equations (3) are called the Cauchy-Riemann equations.
We have thus established

Theorem 4. A necessary condition for a function f (z) = u(x, y) + iv(x, y) to

be differentiable at a point z 0 is that the Cauchy-Riemann equations hold at z 0 .
Consequently, if f is analytic in an open set G, then the Cauchy-Riemann
equations must hold at every point of G.

Augustin-Louis Cauchy (1789–1867), Bernhard Riemann (1826–1866).

73
 74 Analytic Functions

There’s an easy way to recall the Cauchy-Riemann equations. Simply remember

that the horizontal derivative must equal the vertical derivative and write
∂f ∂f
= or
∂x ∂(i y)
∂(u + iv) 1 ∂(u + iv)
= ,
∂x i ∂y
and equate the real and imaginary parts:
∂u ∂v ∂v ∂u
= , =− .
∂x ∂y ∂x ∂y
Example 1
Show that the function f (z) = (x 2 + y) + i(y 2 − x) is not analytic at any point.
Solution. Since u(x, y) = x 2 + y and v(x, y) = y 2 − x, we have
∂u ∂v
= 2x, = 2y,
∂x ∂y
∂u ∂v
= 1, = −1.
∂y ∂x
Hence the Cauchy-Riemann equations are simultaneously satisfied only on the line
x = y and therefore in no open disk. Thus by Theorem 4 the function f (z) is nowhere
analytic.

To be mathematically precise, we point out that the Cauchy-Riemann equations
alone are not sufficient to ensure differentiability; one needs the additional hypothesis
of continuity of the first partial derivatives of u and v. The complete story is given in
the following theorem.

Theorem 5. Let f (z) = u(x, y) + iv(x, y) be defined in some open set G con-
taining the point z 0 . If the first partial derivatives of u and v exist in G, are
continuous at z 0 , and satisfy the Cauchy-Riemann equations at z 0 , then f is dif-
ferentiable at z 0 .
Consequently, if the first partial derivatives are continuousand satisfy the
Cauchy-Riemann equations at all points of G, then f is analytic in G.

Proof. The difference quotient for f at z 0 can be written in the form

f (z 0 + z) − f (z 0 )
z
[u (x0 + x, y0 + y) − u (x0 , y0 )] + i [v (x0 + x, y0 + y) − v (x0 , y0 )]
= (4)
x + iy
Albeit far from obvious, it has been shown that the continuity assumption can be removed in this
part of the theorem.

74
2.4 The Cauchy-Riemann Equations 75

where z 0 = x0 + i y0 and z = x + iy. The above expressions are well defined if

|z| is so small that the closed disk with center z 0 and radius |z| lies entirely in G.
Let us rewrite the difference

u (x0 + x, y0 + y) − u (x0 , y0 )

as

[u (x0 + x, y0 + y) − u (x0 , y0 + y)] + [u (x0 , y0 + y) − u (x0 , y0 )] . (5)

Because the partial derivatives exist in G, the mean-value theorem says that there is a
number x ∗ between x0 and x0 + x such that
∂u ∗
u (x0 + x, y0 + y) − u (x0 , y0 + y) = x x , y0 + y .
∂x
Furthermore, since the partial derivatives are continuous at (x0 , y0 ), we can write
∂u ∗ ∂u
x , y0 + y = (x0 , y0 ) + ε1 ,
∂x ∂x
where the function ε1 → 0 as x ∗ → x0 and y → 0 (in particular, as z → 0). Thus
the first bracketed expression in (5) can be written as
 
∂u
u (x0 + x, y0 + y) − u (x0 , y0 + y) = x (x0 , y0 ) + ε1 .
∂x
The second bracketed expression in (5) is treated similarly, introducing the func-
tion ε2 . Then working the same strategy for the v-difference in Eq. (4), we ultimately
have
 
∂u ∂v ∂u ∂v
f (z 0 + z) − f (z 0 ) x ∂x + ε1 + i ∂x + iε3 +y ∂y + ε2 + i ∂y + iε4
= ,
z x + iy
where each partial derivative is evaluated at (x0 , y0 ) and where each εi → 0 as z →
0. Now we use the Cauchy-Riemann equations to express the difference quotient as
∂u
x ∂x + i ∂∂vx + iy ∂u
∂x + i ∂∂vx λ
+ , (6)
x + iy x + iy
where λ := x (ε1 + iε3 ) + y (ε2 + iε4 ). Since
λ x y
≤ |ε1 + iε3 | + |ε2 + iε4 |
x + iy x + iy x + iy
≤ |ε1 + iε3 | + |ε2 + iε4 |,
we see that the last term in (6) approaches zero as z → 0, and so
f (z 0 + z) − f (z 0 ) ∂u ∂v
lim = (x0 , y0 ) + i (x0 , y0 ) ;
z→0 z ∂x ∂x

75
 76 Analytic Functions

i.e., f  (z 0 ) exists.

It follows from Theorem 4 that the nowhere analytic function f (z) of Example 1
is, nonetheless, differentiable at each point on the line x = y.
As promised in Sec. 1.4, we now offer one last vindication of our definition of the
complex exponential, by demonstrating its analyticity.

Example 2
Prove that the function f (z) = e z = e x cos y + ie x sin y is entire, and find its deriva-
tive.
Solution. Since we have ∂u/∂ x = e x cos y, ∂v/∂ y = e x cos y, ∂u/∂ y =
−e x sin y, and ∂v/∂ x = e x sin y, the first partial derivatives are continuous and satisfy
the Cauchy-Riemann equations at every point in the plane. Hence f (z) is entire. From
Eq. (1) we see that
∂u ∂v
f  (z) = +i = e x cos y + ie x sin y.
∂x ∂x
Not surprisingly, f  (z) = f (z).

As a further application of these techniques, let us prove the following theorem
whose analogue in the real case is well known.

Theorem 6. If f (z) is analytic in a domain D and if f  (z) = 0 everywhere in

D, then f (z) is constant in D.

Before we proceed with the proof, we observe that the connectedness property of
the domain is essential. Indeed, if f (z) is defined by
0 if |z| < 1,
f (z) =
1 if |z| > 2,
then f is analytic and f  (z) = 0 on its domain of definition (which is not a domain!),
yet f is not constant.
Proof of Theorem 6. Since f  (z) = 0 in D, we see from Eqs. (1) and (2) that all
the first partial derivatives of u and v vanish in D; that is,
∂u ∂u ∂v ∂v
= = = = 0.
∂x ∂y ∂x ∂y
Thus, by Theorem 1 in Sec. 1.6 (see page 40), we have u = constant and v = constant
in D. Consequently, f = u + iv is also constant in D.

One easy consequence of Theorem 6 is the fact that if f and g are two functions
analytic in a domain D whose derivatives are identical in D, then f = g + constant in
D (see Prob. 7).

76
2.4 The Cauchy-Riemann Equations 77

Using Theorem 6 and the Cauchy-Riemann equations, one can further show that
an analytic function f (z) must be constant when any one of the following conditions
hold in a domain D:
Re f (z) is constant;
Im f (z) is constant;
| f (z)| is constant. (7)
The proofs are left as problems.

EXERCISES 2.4
1. Use the Cauchy-Riemann equations to show that the following functions are no-
where differentiable.

(a) w = z̄ (b) w = Re z (c) w = 2y − i x

2. Show that h(z) = x 3 + 3x y 2 − 3x + i(y 3 + 3x 2 y − 3y) is differentiable on the

coordinate axes but is nowhere analytic.
3. Use Theorem 5 to show that g(z) = 3x 2 + 2x − 3y 2 − 1 + i(6x y + 2y) is entire.
Write this function in terms of z.
4. Let
(x 4/3 y 5/3 + i x 5/3 y 4/3 )/(x 2 + y 2 ) if z = 0,
f (z) =
0 if z = 0.
Show that the Cauchy-Riemann equations hold at z = 0 but that f is not differen-
tiable at this point. [HINT: Consider the difference quotient f (z)/z for z → 0
along the real axis and along the line y = x.]
2 −y 2
5. Show that the function f (z) = e x [cos(2x y) + i sin(2x y)] is entire, and find its
derivative.
6. If u and v are expressed in terms of polar coordinates (r, θ ), show that the Cauchy-
Riemann equations can be written in the form
∂u 1 ∂v ∂v 1 ∂u
= , =− .
∂r r ∂θ ∂r r ∂θ
[HINT: Consider the difference quotient ( f (z)− f (z 0 ))/(z−z 0 ), as z → z 0 = r0 eiθ0
along the ray arg z = θ0 and along the circle |z| = r0 .]
7. Show that if two analytic functions f and g have the same derivative throughout a
domain D, then they differ only by an additive constant. [HINT: Consider f − g.]
8. Show that if f is analytic in a domain D and either Re f (z) or Im f (z) is constant
in D, then f (z) must be constant in D.
 
9. Show, by contradiction, that the function F(z) = z 2 − z  is nowhere analytic be-
cause of condition (7).

77
 78 Analytic Functions

10. Show that if f (z) is analytic and real-valued in a domain D, then f (z) is constant
in D.
11. Suppose that f (z) and f (z) are analytic in a domain D. Show that f (z) is constant
in D.
12. Show that if f is analytic in a domain D and | f (z)| is constant in D, then the
function f (z) is constant in D. [HINT: | f |2 is constant, so ∂| f |2 /∂ x = ∂| f |2 /∂ y =
0 throughout D. Using these two relations and the Cauchy-Riemann equations,
deduce that f  (z) = 0.]
13. Given that f (z) and | f (z)| are each analytic in a domain D, prove that f (z) is
constant in D.
14. Show that if the analytic function w = f (z) maps a domain D onto a portion of a
line, then f must be constant throughout D.
15. The Jacobian of a mapping

u = u(x, y), v = v(x, y)

from the x y-plane to the uv-plane is defined to be the determinant

 ∂u ∂u 
 
 
 ∂x ∂y 
 
J (x0 , y0 ) :=  ,
 ∂v ∂v 
 
 
∂x ∂y
where the partial derivatives are all evaluated at (x0 , y0 ). Show that if f = u + iv
is analytic at z 0 = x0 + i y0 , then J (x0 , y0 ) = | f  (z 0 )|2 .
16. The notion of analyticity as discussed in the preceding section requires that the
function f (x, y) = u(x, y) + iv(x, y) can be written in terms of (x + i y) alone,
without using z̄ = (x − i y). To make this concept more explicit, we introduce the
change of variables
ξ = x + iy x = (ξ + η)/2
or, equivalently,
η = x − iy y = (ξ − η)/2i
producing the function

f˜(ξ, η) := f (x(ξ, η), y(ξ, η)).

(a) Using the chain rule show formally that

   
∂ f˜ 1 ∂u ∂v i ∂v ∂u
= + + − ,
∂ξ 2 ∂x ∂y 2 ∂x ∂y
   
∂ f˜ 1 ∂u ∂v i ∂u ∂v
= − + + .
∂η 2 ∂x ∂y 2 ∂y ∂x
That is, “z-bar” is barred!

78
2.5 Harmonic Functions 79

(b) Since η is the same as z̄, the statement “ f is independent of z̄” is equivalent
to
∂ f˜ ∂ f˜
= = 0.
∂η ∂ z̄
Show that this condition is the same as the Cauchy-Riemann equations for f .

2.5 Harmonic Functions

Solutions of the two-dimensional Laplace equation

∂ 2φ ∂ 2φ
∇ 2 φ := + 2 =0 (1)
∂x 2 ∂y
are among the most important functions in mathematical physics. The electrostatic
potential solves Eq. (1) in two-dimensional free space, as does the scalar magnetostatic
potential; the corresponding field in any direction is given by the directional derivative
of φ(x, y). Two-dimensional fluid flow problems are described by such functions
under certain idealized conditions, and φ can also be interpreted as the displacement
of a membrane stretched across a loop of wire, if the loop is nearly flat. In the next
section we shall discuss equilibrium temperature distributions as models for solutions
to Eq. (1).
One of the most important applications of analytic function theory to applied math-
ematics is the abundance of solutions of Eq. (1) that it supplies. We shall adopt the
following standard terminology for these solutions.

Definition 6. A real-valued function φ(x, y) is said to be harmonic in a domain

D if all its second-order partial derivatives are continuous in D and if, at each
point of D, φ satisfies Laplace’s equation (1).

The sources of these harmonic functions are the real and imaginary parts of ana-
lytic functions, as we prove in the next theorem.

Theorem 7. If f (z) = u(x, y) + iv(x, y) is analytic in a domain D, then each

of the functions u(x, y) and v(x, y) is harmonic in D.

Proof. In a later chapter we shall show that the real and imaginary parts of any
analytic function have continuous partial derivatives of all orders. Assuming this fact,

Marquis Pierre-Simon de Laplace, 1749–1827.

79
 80 Analytic Functions

we recall from elementary calculus that under such conditions mixed partial derivatives
can be taken in any order; i.e.,
∂ ∂u ∂ ∂u
= . (2)
∂y ∂x ∂x ∂y
Using the Cauchy-Riemann equations for the first derivatives, we transform Eq. (2)
into
∂ 2v ∂ 2v
= − .
∂ y2 ∂x2
which is equivalent to Eq. (1). Thus v is harmonic in D, and a similar computation
proves that u is also.

Conversely, if we are given a function u(x, y) harmonic in, say, an open disk, then
we can find another harmonic function v(x, y) so that u + iv is an analytic function of
z in the disk. Such a function v is called a harmonic conjugate of u. The construction
of v is effected by exploiting the Cauchy-Riemann equations, as we illustrate in the
following example.

Example 1
Construct an analytic function whose real part is u(x, y) = x 3 − 3x y 2 + y.
Solution. First we verify that
∂ 2u ∂ 2u
+ = 6x − 6x = 0,
∂x2 ∂ y2
and so u is harmonic in the whole plane. Now we have to find a mate, v(x, y), for u
such that the Cauchy-Riemann equations are satisfied. Thus we must have
∂v ∂u
= = 3x 2 − 3y 2 (3)
∂y ∂x
and
∂v ∂u
=− = 6x y − 1. (4)
∂x ∂y
If we hold x constant and integrate Eq. (3) with respect to y, we get
v(x, y) = 3x 2 y − y 3 + constant,
but the “constant” could conceivably be any differentiable function of x; it need only
be independent of y. Therefore, we write
v(x, y) = 3x 2 y − y 3 + ψ(x).
We can find ψ(x) by plugging this last expression into Eq. (4);
∂v
= 6x y + ψ  (x) = 6x y − 1. (5)
∂x

80
2.5 Harmonic Functions 81

Figure 2.6 Level curves of real and imaginary parts of z 2 .

This yields ψ  (x) ≡ −1, and so ψ(x) = −x + a, where a is some (genuine) constant.
It follows that a harmonic conjugate of u(x, y) is given by

v(x, y) = 3x 2 y − y 3 − x + a,

and the analytic function

 
f (z) = x 3 − 3x y 2 + y + i 3x 2 y − y 3 − x + a ,

which we recognize as z 3 − i(z − a), solves the problem.

This procedure will always work for an u(x, y) harmonic in a disk, as is shown in
Prob. 20. Thus we can learn a great deal about analytic functions by studying harmonic
functions and vice versa.
The harmonic functions forming the real and imaginary parts of an analytic func-
tion f (z) each generate a family of curves in the x y-plane, namely, the level curves or
isotimic curves
u(x, y) = constant (6)
and
v(x, y) = constant. (7)
If u is interpreted as an electrostatic potential, then the curves (6) are the equipoten-
tials. If u is temperature, (6) describes the isotherms.
For the function f (z) = z 2 = x 2 − y 2 + i2x y, the level curves u = x 2 − y 2 =
constant are hyperbolas asymptotic to the lines y = ±x, as shown in Fig. 2.6(a). The
curves v = 2x y = constant are also hyperbolas, asymptotic to the coordinate axes; see
Fig. 2.6(b).
We caution the reader that finding a harmonic conjugate in an arbitrary domain may not always
be possible. See Prob. 21 for an example of this unfortunate circumstance when the domain is a
punctured disk.

81
 82 Analytic Functions

Figure 2.7 Level curves of Fig. 2.6 superimposed.

Figure 2.8 Level curves of real and imaginary parts of z 3 .

Figure 2.9 Level curves of real and imaginary parts of 1/z.

82
2.5 Harmonic Functions 83

Figure 2.10 Level curves of real and imaginary parts of e z .

Notice that if the two families of curves are superimposed as in Fig. 2.7 they appear
to intersect at right angles. The same effect occurs with the level curves for the analytic
functions z 3 (Fig. 2.8), 1/z (Fig. 2.9), and e z (Fig. 2.10). This is no accident; the level
curves of the real and imaginary parts of an analytic function f (z) will always intersect
at right angles—unless f  (z) = 0 at the point of intersection. This can be seen from
the Cauchy-Riemann equations as follows.
Recall that the vector with components [∂u/∂ x, ∂u/∂ y] is the gradient of u and
is normal to the level curves of u. Similarly, [∂v/∂ x, ∂v/∂ y] is normal to the level
curves of v. The scalar (dot) product of these gradient vectors is
∂u ∂v ∂u ∂v ∂v ∂v ∂v ∂v
+ = − =0
∂x ∂x ∂y ∂y ∂y ∂x ∂x ∂y
by the Cauchy-Riemann equations. Thus if these gradients are nonzero, they are per-
pendicular, and hence so are the level curves. Level curves of harmonic functions and
their harmonic conjugates intersect at right angles.
The following examples illustrate how analytic function theory can be used to
solve Laplace’s equation in regions whose boundaries are identifiable as level curves.

Example 2
Find a function φ(x, y) that is harmonic in the region of the right half-plane between
the curves x 2 − y 2 = 2 and x 2 − y 2 = 4 and takes the value 3 on the left edge and the
value 7 on the right edge (Fig. 2.11).
Solution. We recognize x 2 − y 2 as the real part of z 2 , so the boundary curves are
level curves of a known harmonic function. To meet the specified boundary conditions,
we add some flexibility by considering
   
φ(x, y) = A x 2 − y 2 + B = Re Az 2 + B , A, B real,

and adjust A and B accordingly. When x 2 − y 2 = 2, we require φ = 3;

A(2) + B = 3.
When x 2 − y 2 = 4, we want φ = 7;
A(4) + B = 7.

83
 84 Analytic Functions

Figure 2.11 Laplace’s equation for the region of Example 2.

Solving for A and B we find the solution to be

 
φ(x, y) = 2 x 2 − y 2 − 1.

This example was clearly contrived. In Chapters 3 and 7 we shall consider more
profound applications of this idea.

EXERCISES 2.5
1. Verify directly that the real and imaginary parts of the following analytic functions
satisfy Laplace’s equation.
1
(a) f (z) = z 2 + 2z + 1 (b) g(z) =
z
(c) h(z) = e z

2. Find the most general harmonic polynomial of the form ax 2 + bx y + cy 2 .

3. Verify that each given function u is harmonic (in the region where it is defined) and
then find a harmonic conjugate of u.

(a) u = y (b) u = e x sin y

(c) u = x y − x + y (d) u = sin x cosh y
2
(e) u = ln |z| for Re z > 0 (f) u = Im e z

4. Show that if v(x, y) is a harmonic conjugate of u(x, y) in a domain D, then every

harmonic conjugate of u(x, y) in D must be of the form v(x, y) + a, where a is a
real constant.
5. Show that if v is a harmonic conjugate for u, then −u is a harmonic conjugate for v.
6. Show that if v is a harmonic conjugate of u in a domain D, then uv is harmonic
in D.

84
2.5 Harmonic Functions 85

7. Find a function φ(x, y) that is harmonic in the infinite vertical strip

{z : − 1 ≤ Re z ≤ 3}

and takes the value 0 on the left edge and the value 4 on the right edge.

8. Suppose that the functions u and v are harmonic in a domain D.

(a) Is the sum u + v necessarily harmonic in D?

(b) Is the product uv necessarily harmonic in D?

(c) Is ∂u/∂ x harmonic in D? (You may use the fact—which we will prove in
Chapter 4—that harmonic functions have continuous partial derivatives of all
orders.)

9. Find a function φ(x, y) that is harmonic in the region of the first quadrant between
the curves x y = 2 and x y = 4 and takes the value 1 on the lower edge and the value
3 on the upper edge. [HINT: Begin by considering z 2 .]

10. Show that in polar coordinates (r, θ ) Laplace’s equation becomes

∂ 2φ 1 ∂φ 1 ∂ 2φ
+ + 2 = 0.
∂r 2 r ∂r r ∂θ 2

11. Let f (z) = z + 1/z. Show that the level curve Im f (z) = 0 consists of the real axis
(excluding z = 0) and the circle |z| = 1. [The level curves Im f (z) = constant can
be interpreted as streamlines for fluid flow around a cylindrical obstacle.]

12. Prove that if r and θ are polar coordinates, then the functions r n cos nθ and r n sin nθ ,
where n is an integer, are harmonic as functions of x and y. [HINT: Recall De
Moivre’s formula.]

13. Find a function harmonic inside the wedge bounded by the nonegative x-axis and
the half-line y = x (x ≥ 0) that goes to zero on these sides but is not identically
zero. [HINT: See Prob. 12.] The level curves for this function can be interpreted as
streamlines for a fluid flowing inside this wedge, under certain idealized conditions.

14. Suppose that f (z) is analytic and nonzero in a domain D. Prove that ln | f (z)| is
harmonic in D.

85
 86 Analytic Functions

15. Find a function φ(z), harmonic within the annulus (ring domain) bounded by the
concentric circles |z| = 1 and |z| = 2, such that φ = 0 on the inner circle and
φ(2eiθ ) = 5 cos 3θ on the outer circle. [HINT: Think of z n and z −n .]
16. Find a function harmonic outside the circle |z| = 3 that goes to zero on |z| = 3 but
is not identically zero. [HINT: See Prob. 14]
17. Find a function φ(x, y) harmonic in the upper half-plane Im z > 0 and continuous
on Im z ≥ 0 such that
(a) φ(x, 0) = x 2 + 5x + 1 for all x.
(b) φ(x, 0) = 2x 3 /(x 2 + 4) for all x.
[HINT: φ = Re[2z 3 /(z 2 + 4)] won’t work because 2z 3 /(z 2 + 4) is not analytic at
z = 2i in the upper half-plane. Instead, write
2x 3 x2 x2 x2
= + = 2 Re ,
x2 +4 x − 2i x + 2i x + 2i
which suggests the proper choice for φ.]
18. Show that if φ(x, y) is harmonic, then φx − iφ y is analytic. (You may assume that
φ has continuous partial derivatives of all orders.)
19. Find a function φ(z) harmonic outside the unit circle |z| = 1, satisfying
φ(eiθ ) = cos2 θ, 0 ≤ θ ≤ 2π,
such that φ(r eiθ ) approaches the constant value 1/2 along all large radii r . [HINT:
Recall that z −n is analytic outside the unit circle and goes to zero along large radii r .]
20. By tracing the steps in Example 1, show that every function u(x, y) harmonic in
a disk has a harmonic conjugate v(x, y). [HINT: The only difficulty which could
occur is in the step corresponding to Eq. (5), where in order to find ψ  (x) we must
be certain that all appearances of the variable y cancel. Show that this is guaranteed
because u is harmonic.]
21. Show that although u = ln |z| is harmonic in the complex plane except at z = 0 (i.e.,
in the domain C\{0}), u does not have a harmonic conjugate v throughout C\{0}.
In other words, show that there is no function v such that ln |z| + iv(z) is analytic in
C\{0}. [HINT: Show that if ln |z| + iv(z) is analytic in C\{0}, then v(z) = Arg z + a
except along the nonpositive real axis.]
22. Show that if φ(x, y) and ψ(x, y) are harmonic, then u and v defined by
u(x, y) = φx φ y + ψx ψ y

and
 
v(x, y) = 1
2 φx2 + ψx2 − φ y2 − ψ y2

satisfy the Cauchy-Riemann equations.

86
2.6 *Steady-State Temperature as a Harmonic Function 87

Figure 2.12 Slab of thermally conducting material.

Figure 2.13 Sinks and sources.

2.6 *Steady-State Temperature as a Harmonic

Function
It is useful to have a familiar physical model for harmonic functions as an aid in visu-
alizing and remembering their properties. The equilibrium temperatures in a slab, as
we shall see, fill this role nicely.
Figure 2.12 depicts a uniform slab of a thermally-conducting material, such as
a copper plate or a ceramic substrate for microelectronic circuitry. It has constant
thickness, so its top and bottom surfaces lie parallel to the x y-plane. We assume that
these surfaces are also insulated, and no heat flows in the vertical direction. As a result
the equilibrium temperature T is a function of x and y;
T = T (x, y).
This temperature distribution is maintained by heat sources (or sinks) and insula-
tion placed around the edges, so that the isotherms appear as illustrated in Fig. 2.13.
Now once the temperature has reached equilibrium, T (x, y) will be a harmonic
function:
∂2T ∂2T
+ = 0. (1)
∂x2 ∂ y2
The physical reason for this is as follows. Focusing attention on a small square in the
slab as depicted in Fig. 2.14, we call upon Fourier’s law of heat conduction, which

87
 88 Analytic Functions

states that the rate at which heat flows through each side of the square is proportional
to the rate of change of temperature in the direction of the flow. Thus the flow through
AB and C D is proportional to ∂ T /∂ x, and that through BC and AD is proportional
to ∂ T /∂ y. (In fact the constant of proportionality, which depends on the cross-section
area and the material, is negative, since heat flows from hot to cold!) The heat flows
are depicted as entering the square through AB and AD and exiting through BC and
C D. Therefore, the net outflow of heat is proportional to
   
∂ T  ∂ T  ∂ T  ∂ T 
− + − .
∂ x C D ∂ x  AB ∂ y  BC ∂ y  AD
For small dimensions s the difference in the first derivatives can be approximated
by the second derivative, and the net outflux is proportional to

∂2T ∂2T
s + s. (2)
∂x2 ∂ y2
At equilibrium the temperature has settled; the square has finished cooling down (or
heating up), and the net outflux will be zero. Dividing expression (2) by s, then, we
conclude that T (x, y) satisfies Eq. (1).
The fact that harmonic functions arise as temperature distributions permits us to
anticipate some of their mathematical properties. For example, look at the isothermal
curves in Fig. 2.15. They indicate a “hot spot” in the interior of the slab. This cannot
occur at equilibrium, because heat would flow away from the hot spot and it would cool
down. Of course, this pattern could be maintained by an external source underneath
the slab, but we have precluded this by assuming that such sources are located only
on the edge. We conclude that the temperature distribution can never exhibit such an
interior maximum. The rigorous formulation and generalization of this observation to
harmonic functions is identified in Chapter 4 as the maximum principle, which says
that a harmonic function cannot take its maximum in the interior of a region, except in
the trivial case when it is constant throughout.
As another example consider the following experiment. The edges of the slab in
Fig. 2.16 are maintained at fixed temperatures by external heat sources. except for a
small section along which we can control the temperature to our liking (using some
type of adjustable furnace). Then, on physical grounds we would expect to be able,
by turning up the furnace sufficiently, to raise the temperature of an arbitrary interior
point to any specified value—although, of course, we couldn’t guarantee to replicate a
whole pattern of temperatures across the slab.
Our expectation is premised on the intuitive feeling that the interior temperatures
are completely determined by the edge temperature distribution. This is actually an
instance of the boundary value property of harmonic functions, and in fact in Chapter
4 we shall study Poisson’s formulas, which express the explicit relationship between
the interior and boundary values of such functions for certain geometries.
Note also that the thermodynamic reality of a zero absolute temperature inhibits our ability to
cool interior points arbitrarily.

88
2.6 *Steady-State Temperature as a Harmonic Function 89

Figure 2.14 Heat flow.

Figure 2.15 Isotherms.

Figure 2.16 Adjustable boundary temperatures.

89
 90 Analytic Functions

EXERCISES 2.6
1. Using only your physical intuition, sketch the family of isotherms that you would
expect to see at equilibrium for slabs with edge temperatures maintained as shown
in Fig. 2.17.

Figure 2.17 Isotherm constructions (Prob. 1).

2. Sketch the isotherms for the edge-temperature distribution in Fig. 2.18. Does this
configuration violate the maximum principle?

Figure 2.18 Isotherm construction (Prob. 2).

3. Sketch the isotherms for the edge-temperature distribution in Fig. 2.19. Does this
configuration violate the maximum principle?

90
2.7 *Iterated Maps: Julia and Mandelbrot Sets 91

Figure 2.19 Isotherm construction (Prob. 3).

2.7 *Iterated Maps: Julia and Mandelbrot Sets

What happens if one enters a number into a calculator, pushes a function key like x 2 ,
and then pushes it again and again? The calculator squares the number, then squares
the result, then squares that result, and so on; one has iterated the function f (x) = x 2 .
Now this iteration process can be very interesting when it is performed with complex
numbers; the sequence of points z 0 , f (z 0 ) , f ( f (z 0 )) , f ( f ( f (z 0 ))), . . . becomes an
orbit in the complex plane.
If we iterate the complex function f (z) = z 2 , it is easy to predict many of the
orbits. When the starting point or “seed” z 0 lies within the unit circle, that is |z 0 | < 1,
the orbit stays bounded (because the squares get smaller in modulus) and converges to
z = 0. If |z 0 | > 1, the iterates get larger in modulus and the orbit is unbounded.
Example 1 describes a test for bounded orbits that is easy to apply.

Example 1
Show that if
(i) f (z) is analytic in a neighborhood of z = ζ ,
(ii) f (ζ ) = ζ , and
(iii) | f  (ζ )| < 1,
then there is a disk around ζ with the property that all orbits launched from inside the
disk remain confined to the disk and converge to ζ .
Solution. Since
 
 f (z) − f (ζ ) 
lim   = | f  (ζ )| < 1

z→ζ z−ζ
and f (ζ ) = ζ , we can pick a real number ρ lying between | f  (ζ )| and 1 such that
| f (z) − f (ζ )| ≡ | f (z) − ζ | ≤ ρ|z − ζ | (1)

91
 92 Analytic Functions

for all z in a sufficiently small disk around ζ . Such a disk meets the specifications;
indeed, if any point z 0 in this disk is the seed for an orbit z 1 = f (z 0 ), z 2 = f (z 1 ), . . .,
then by (1) we have

|z n − ζ | ≤ ρ|z n−1 − ζ | ≤ · · · ≤ ρ n |z 0 − ζ |.

Since ρ < 1, the point z n lies closer to ζ than z n−1 and, in fact, limn→∞ z n = ζ .

If f (ζ ) = ζ , then ζ is called a fixed point of the function f . A fixed point meeting

the conditions of Example 1 is called an attractor, and the set of seed points whose
orbits converge to ζ is called its basin of attraction. Thus ζ = 0 is an attractor for
f (z) = z 2 (since 0 = 02 = f (0) and | f  (0)| = 0 < 1) whose basin is the open disk
|z| < 1. Example 1 shows that every attractor has a basin containing, at least, a small
disk.
The other fixed point for f (z) = z 2 , ζ = 1, is a repellor. Its properties are
explored in Prob. 2.
For the function f (z) = z 2 , if z 0 lies on the unit circle |z 0 | = 1, so does the entire
orbit launched from z 0 . In fact if z 0 = 1 or −1 the orbit quickly settles down to the
fixed point z = 1. If z 0 = ei2π/3 (a primitive cube root of unity in the parlance of
Sec. 1.5) the orbit oscillates between two points e±i2π/3 and is called a 2-cycle, with
period 2; note that this is equivalent to saying that ei2π/3 is a fixed point of f ( f (z)).
The seed ei2π/7 gives birth to a 3-cycle, and ei2π/15 to a 4-cycle. Do you see the pattern
(cf. Prob. 6)?
It can be shown that the seed choice z 0 = eiα2π , for irrational α, generates an orbit
whose points never repeat (cf. Prob. 4) and, in fact, permeate the unit circle densely.
So the unit circle, which separates the seeds of orbits converging to zero from those of
unbounded orbits, contains a variety of orbits itself.

Definition 7. The filled Julia set for a polynomial function f (z) is defined to
be the set of points that launch bounded orbits through iteration of f ; the Julia
set is the boundary of the filled Julia set.

So the Julia set for z 2 is the unit circle, and the filled Julia set is the closed unit
disk.
The Julia set for the function f (z) = z 2 − 2 consists of the real interval [−2, 2]
(which is already “filled”). Indeed, one immediately sees that if −2 ≤ x ≤ 2, then
0 ≤ x 2 ≤ 4 and −2 ≤ x 2 − 2 ≡ f (x) ≤ 2, so orbits launched from [−2, 2] remain
bounded. The proof that other values of z are seeds of unbounded orbits will have to
wait until we have studied the Joukowski transformation (Prob. 8a, Exercises 7.7).

G. Julia investigated these sets in 1918.

92
2.7 *Iterated Maps: Julia and Mandelbrot Sets 93

From these considerations, one might conclude that the Julia sets for all functions
of the form f (z) = z 2 + c are disks, segments, or other mundane configurations.
Nothing could be further from the truth! An astonishing assortment of exotic patterns
result when we locate the seeds of bounded orbits of z 2 + c, for various complex
values of the constant c. Fig. 2.20 displays some of the dragons, rabbits, fern leaves,
and shoreline patterns that have been discovered to be Julia sets, together with the
corresponding values of c. Many of these patterns are fractals; objects that typically
have dimension neither one nor two, but some fraction in between. Some also enjoy
the property of self-similarity, in that if one zooms in to see the details of a small
subset, a replica of the original pattern reappears.
The most enjoyable way to explore Julia sets is with software. Although func-
tional iteration is trivial to code, the graphics displays are best left to experts, and the
references at the end of this chapter contain pointers to some websites and software
packages that provide this facility. We invite the reader to try to replicate the designs
depicted in Fig. 2.20, using software.
Note that some of the filled Julia sets consist of single connected components,
while the others appear to be totally disconnected. In 1982, Benoit Mandelbrot was
inspired to investigate which values of c give rise to (filled) Julia sets that are con-
nected. The answer was the astonishing Mandelbrot set, depicted in Fig. 2.21. The
point c lies in the Mandelbrot set if the filled Julia set for f (z) = z 2 + c is connected;
it lies outside if the filled Julia set is disconnected. It is amusing to use software to
see what happens to the Julia sets for a family of values of c tracing a path from the
interior to the exterior of the Mandelbrot set; the Julia sets “vaporize” as they meta-
morphose from connected to disconnected. (See references.) It should not surprise the
reader that some phase transitions in physics have been modeled using the Mandelbrot
set.
The interplay between rigorous, theoretical analyses of these sets and computer
experimentation has generated some interesting insights into the latter. For example,
we can’t trust a computer to tell us if an orbit is unbounded, because the computer can
only distinguish a finite set of numbers; eventually it will recycle (or overflow). So we
have to make a judgment as to how many iterations we must simulate before deciding
that an orbit is unbounded. Secondly, virtually every time the computer iterates the
function f (z) it commits a roundoff error, and these errors accumulate as we simulate
long orbits; so we have to carefully evaluate the credibility of our calculations. Despite
these forebodings, in 1987 Hammel, Yorke, and Grebogi proved that every computed
orbit is arbitrarily close to some true orbit!
The Julia sets (and their analogs) for more complicated functions are subjects of
continuing mathematical research. The software listed at the end of the chapter will
guide the reader to explore the beautiful convergence patterns resulting from iterating
the complex trigonometric functions (defined in Chapter 3) and others. In fact a in-
ternational exhibition of patterns generated using the Mandelbrot set, titled “Frontiers
of Chaos,” toured many museums in the late 1980s; see the reference by Peitgen and
Richter.

93
 94 Analytic Functions

Figure 2.20 Julia sets.

Figure 2.21 Mandelbrot set.

94
Summary 95

EXERCISES 2.7

1. Find the fixed points and attractors for the function f (z) = z 2 + c, in terms of the
real constant c.
2. If f (z) is entire and ζ is a fixed point of f such that | f  (ζ )| > 1, then ζ is said to
be a repellor for f . Show that there is a disk around ζ such that all orbits launched
from within the disk, other than the 1-cycle launched from ζ itself, eventually leave
the disk.
3. For each of the following functions, determine the fixed points and decide which
are attractors, repellors (see Prob. 2), or neither.

(a) f (z) = z 2 + z + 1 (b) f (z) = z 3 + z 2 + (3/4)z − 1/4 .

4. Prove that if α is a real irrational number, the seed z 0 = ei2π α generates an orbit
under f (z) = z 2 whose points never repeat.
5. For the function f (z) = 1/(z + 1), determine the fixed points and decide which are
attractors, repellors (see Prob. 2), or neither.
n −1)
6. Derive the formula z 0 = ei2π/(2 for a seed launching an n-cycle for orbits
formed by iterating f (z) = z 2 .
7. The seed point z 0 = ei2π/5 launches a 4-cycle for iterates of f (z) = z 2 , but it does
not fit into the pattern of Prob. 6. Explain.
8. Use software to generate the Julia sets in Fig. 2.20.
9. Determine the filled Julia set for the mapping f (z) = αz, where α is a complex
constant. Consider separately the cases where |α| ≤ 1 and |α| > 1.
10. In general, Newton’s method for approximating the zeros of the entire function F(z)
can be described as forming the orbits of the function
F(z)
f (z) = z − .
F  (z)

Show that the fixed points of f (z) are the same as the zeros of F(z), with the
possible exception of the points where F  (z) = 0. Then show that every zero of
F(z) (other than ones where F  (z) = 0) is an attractor for f (z).

SUMMARY
A complex-valued function f of a complex variable z = x + i y can be considered
as a pair of real functions of two real variables in accordance with f (z) = u(x, y) +
iv(x, y). The definitions of limit, continuity, and derivative for such functions are
direct analogues of the corresponding concepts introduced in calculus, but the greater

95
 96 Analytic Functions

freedom of z to vary in two dimensions lends added strength to these conditions. In

particular, the existence of a derivative, defined as the limit of

f (z + z) − f (z)
as z → 0,
z
implies a strong relationship between the functions u and v, namely, the Cauchy-
Riemann equations
∂u ∂v ∂u ∂v
= , =− .
∂x ∂y ∂y ∂x
If the function f is differentiable in an open set, it is said to be analytic. This
property can be established by showing that the first-order partial derivatives of u and
v are continuous and satisfy the Cauchy-Riemann equations on the open set. Analyt-
icity of a function f is the mathematical expression of the intuitive condition that f
respects the complex structure of z; i.e., f can be computed using x and y only in the
combination (x + i y). If f is given in terms of z alone, the basic formulas of calculus
can be used to find its derivative.
The real and imaginary parts of an analytic function are harmonic; i.e., they satisfy
Laplace’s equation
∂ 2φ ∂ 2φ
+ = 0,
∂x2 ∂ y2
and their second-order partial derivatives are continuous. Furthermore, the level curves
of the real part intersect those of the imaginary part orthogonally. Given a harmonic
function u(x, y) in a disk it is possible to construct another harmonic function v(x, y)
so that u(x, y) +iv(x, y) is analytic in that disk; such a function v is called a harmonic
conjugate of u. Harmonic functions can be physically interpreted as equilibrium tem-
perature distributions.

Suggested Reading
In addition to the references following Chapter 1, the following texts, articles, web-
sites, and software packages may be helpful for special topics:

Harmonic Functions
[1] Davis, H., and Snider, A.D. Introduction to Vector Analysis, 7th ed. Quant
Systems, Charleston, SC, 1994.

[2] Hille, E. Analytic Function Theory, Vol. II. Chelsea, New York, 1973.

[3] Snider, A.D. Partial Differential Equations: Sources and Solutions, Prentice-
Hall, Upper Saddle River, NJ, 1999.

96
Chapter 2 – Suggested Reading 97

Julia and Mandelbrot Sets

[1] Devaney, R. L. A First Course in Chaotic Dynamical Systems, Addison-Wesley
Publishing Co., Reading, MA, 1992.

[2] Hammel, S. M., Yorke, J. A., and Grebogi, C. “Do numerical orbits of chaotic
dynamical processes represent true orbits?,” J. of Complexity 3 (1987), 136–145.

[3] Peitgen, H.-O. and Richter, P. H. The Beauty of Fractals, Springer-Verlag, Ber-
lin, 1986. (Describes the exhibit “Frontiers of Chaos.”)

[4] http://math.bu.edu/DYSYS/explorer/tour4.html, a web site maintained at Boston

University by R. L. Devaney, contains much material on the Julia and Mandel-
brot sets. The “evaporating” Julia sets mentioned in the text are visible here.

[5] http://www.unca.edu/ mcmcclur/java/Julia/ allows the user to click on an arbi-

trary value for c and view the Julia set for z 2 + c.

97
98