Functions of a Complex Variable

In this chapter, we will study functions of a complex variables. The most interesting func-
tions on the complex plane are those that are holomorphic. The most important holomorphic
functions are the exponential and the logarithm function. Holomorphicity is the complex
analogue of differentiability, but for various reasons, which we shall study in great detail, a
function being holomorphic is far more restrictive than the function being differentiable.
The focal point of this chapter will be power series and functions that have local power series
expansions, i.e., analytic functions. The reason for this is two-fold:

(i). The most important functions that arise in practice are analytic.
(ii). All holomorphic functions are analytic!

The second point above will be one of the central theorems we will prove in this course.
We make some notation remarks. We will be considering functions f : U → C, where U ⊂ C
is an open set. We will usually write f = u + iv, where u := Ref and v := Imf are the real and
imaginary parts of f. Note that f is continuous iff u and v are continuous.
We begin with the definition and basic properties of holomorphic functions. We then present
the theory of power series and study various properties of analytic functions.

1 Holomorphic functions
1.1 Definitions and examples
Definition 1. We say that f is C-differentiable at the point a ∈ U if
f(a + h) − f(a)
lim → 0.
h→a h
In the case, we denote the limiting value by f 0 (a) and call it the derivative at a.
If f is C-differentiable at each point of U, then we say that f is holomorphic on U. The set
of holomorphic mappings on U is denoted H(U).
A function that is holomorphic on the whole of C is called an entire function.

Remarks.

a) If f is differentiable at a then f is continuous at a.

b) If f and g are differentiable at a then so are f ± g, cf(c ∈ C) and fg and the usual rules
of differentiation apply. The proofs of these facts follows mutatis mutandis from the
proofs of the real case.
c) If f is differentiable at a and f(a) 6= 0, then 1f is differentiable at a and its derivative at
−1
a is f(a) 2.

1
 1 Holomorphic functions

d) We can rewrite the existence of the derivative a as follows: We say that if f is differentiable
at a if there exists α ∈ C such that

f(a + h) = f(a) + αh + o(h). (1.1)

Note that the above says that f can be approximated by an affine linear map near a with
the error term being o(h).
e) If we consider f as map from U ⊂ R2 to R2 , then note that f being R-differentiable at a
is equivalent to the existence of an R-linear map df(a) such that

f(a + h) = f(a) + df(a)h + o(h), h ∈ R2 .

Combining this with 1.1 , we see that f is C-differentiable iff f is R-differentiable and the
derivative map df(a) is C-linear.
f) The previous remark shows that the chain rule is valid for a composition of C-differentiable
functions. Please write down a precise statement and give a three line proof.
Examples.
(i) The function f(z) = zn is entire and its derivative at the point a ∈ C is given by nan−1 .
The proof of this follows mutatis mutandis from the proof of the real case.
(ii) Combining the above with the preceding remarks, we see that any polynomial P(z) is
entire.
(iii) Consider the function f(z) = z. In real coordinates this the map (x, y) → (x, −y) which
is linear and hence this map is C∞ -smooth. However, for any a ∈ C, note that
f(a + h) − f(h) h
= .
h h
If h → 0 along the real axis, then the above limit → 1. On the other hand, if h → 0 along
the imaginary axis then the above limit → −1 showing that f is not C-differentiable at
any point a ∈ C.
(iv) Consider the function f(z) = |z|2 = z · z. In real coordinates, we can consider at as the
function (x, y) → x2 + y2 . This is polynomial function in x, y and hence C∞ -smooth.
Again computing as before, we see that the difference quotient if of the form
h
a+a + h.
h
As h → 0, the above difference quotient has a limit iff a = 0. Hence |z|2 is C-differentiable
only at the point 0.
The last two examples show that f being C-differentiable at a point a is a severe restriction.
The limit of the difference quotient must exist and agree on all possible approaches to the point
a. We could approach a along any line, or even complicated curves like spirals and the limiting
value must exist and agree on any such approach. This is the crucial difference between real
differentiability of a function defined on an interval and C-differentiability. The difference
between f being differentiable as a function on R2 and as a C-differentiable function is that
the map df(a) is now C-linear which means that it is conformal. This will force holomorphic
functions to behave more nicely that C∞ functions.

2
 1 Holomorphic functions

1.2 The Cauchy–Riemann equations

We would like some necessary and sufficient conditions for f to be C-differentiable at a. The
following is the famous Cauchy–Riemann equations. Let U, f, a be as before and write z as
x + iy, x, y ∈ R and f = u + iv where u and v are real-valued. Assume that a, the partial
derivatives ux , uy , vx , vy exist. We define
 
∂f 1 ∂f ∂f
(a) := −i
∂z 2 ∂x ∂y
 
∂f 1 ∂f ∂f
(a) := +i .
∂z 2 ∂x ∂y
Theorem 2. Assume that f is R-differentiable. Then the following conditions are all equiv-
alent:

(i) f is C-differentiable at a.
(ii) ux = vy and uy = −vx .
∂f ∂f
(iii) ∂y (a) = i ∂x (a).
∂f
(iv) ∂z (a) = 0.

Proof. The equivalence of (ii), (iii) and (iv) are obvious. Let us see how the map dfa acts on
the two elements 1, i ∈ C. Now 1, i are nothing but the vectors (1, 0) and (0, 1), respectively,
∂f ∂f
in R2 . This dfa (1) = ∂x (a) and dfa (i) ∂y (a). Now, dfa is C-linear iff dfa (i) = idfa (1) and
this proves the equivalence.

Remarks.
a) Any of the equivalent equations (ii) − (iv) is called the Cauchy–Riemann equations.
Note that a function being C-differentiable is equivalent to f being both R-differentiable
and satisfying the Cauchy–Riemann equation. Just one of the latter conditions p being
satisfied does not guarantee C-differentiability. For instance, the function f = |xy|
satisfies the Cauchy–Riemann equations at 0 but is not C-differentiable at 0.
b) Note that the df(a) written as a matrix (i.e., the Jacobian matrix) is
 
ux uy
.
−vx vy
The fact that this matrix is C-linear forces ux = vy and uy = −vx which is another way of
seeing that the Cauchy–Riemann equations are necessarily satisfied by a C-differentiable
map.
c) Yet an another way of seeing that the Cauchy–Riemann equations are satisfied by a
C-differentiable function is directly from definition. Let f be C-differentiable at a, then
f(a + h) − f(a)
lim ,
h→a h
can be evaluated by taking h to 0 along the real axis, i.e., the above limit is nothing but
∂f
∂x . Alternatively, we may take h → 0 along the imaginary axis, in which case the above
∂f
limit is −i ∂y which is nothing but another form of Cauchy–Riemann equations. This
also shows that f 0 (a) = ux (a) + vx (a)

3
 1 Holomorphic functions

d) If f is C-differentiable at a, then det(Jf (a)) = u2x (a) + vx (a)2 = |f 0 (a)|2 .

The above remarks motivates the question: If f satisfies the Cauchy–Riemann equations at
a then under what additional hypotheses on f does it follow that f is C-differentiable at a?
The literature on this topic is quite vast and we will be content by stating the following famous
theorem.
∂f
Theorem 3 (Looman–Menchoff). Suppose the partial derivatives of f exist on U and ∂z ≡ 0.
Then f ∈ H(U).
We end this section with the following simple consequence of the Cauchy–Riemann equation.
Proposition 4. Let f be a real-valued function and suppose that f is C-differentiable at a.
Then df(a) = 0. Consequently any real-valued holomorphic function on a domain U is
constant. If f, g ∈ H(U) that have the same real or imaginary parts, then f − g ≡ c where
c ∈ C is some constant.
Proof. If f = u + iv, v ≡ 0 then the Cacuhy–Riemann equations show that ux (a) = vy (a) = 0
and consequently f 0 (a) = ∂x
∂f
= ux (a) + ivx (a) = 0. The other parts are easy.

1.3 Holomorphic functions and conformality

The fact that the derivative of a holomorphic mappings is C-linear suggest that holomorphic
mappings should be conformal in some appropriate sense. We make this precise in this section.
First, we consider any arbitrary R-differentiable mapping f : U → Rn on an open set U ⊂ Rn .
Let I be a closed interval that has 0 as an interior point and let γ : I → U be a differentiable
curve. Then f ◦ γ is also a differentiable curve and by the chain rule

d X ∂f
n
|t=0 (f ◦ γ)(t) (γ(t)) · γi0 (0) = df(a)(γ 0 (0)).
dt ∂xi
i=1

This shows that the tangent of the curve γ at the point a := γ(a) is transformed to the tangent
vector of the curve f ◦ γ at f(a) by the map df(a).
If γ1 and γ2 are two differentiable curves that pass through the point a, we define the
(oriented) angle between γ1 and γ2 at a to be the oriented angles between the tangent vector
of γ1 at a and the tangent vector of γ2 at a.
Definition 5. We say that the map f is conformal at a if df(a) is invertible and a conformal
linear map.
Remarks.
(i) Note that his means that if γ1 and γ2 are regular curves that intersect at a then the
angles between γ1 and γ2 at a is same as the angles between f ◦ γ1 and f ◦ γ2 at f(a).
(ii) From the result5s of the previous chapter, f is conformal at a iff f is C-differentiable at
a and f 0 (a) 6= 0. In this event, if γ is a curve passing through a such that γ(0) = a
and γ 0 (0) 6= 0 then then tangent vector to f ◦ γ at f(a) is give by f 0 (a) · γ 0 (0). As
multiplication by a complex number is just a dilation followed by a rotation, it is clear
why C-differentiable mappings are conformal.

4
 2 Power series

2 Power series
So far all the examples of holomorphic functions we have given were all polynomial functions.
In this section, we develop the machinery of power series which will allow us to give more
examples of holomorphic functions. In particular, we will be able to extend the definitions of
the exponential and trigonometric functions to C. We begin with a few remarks on polynomial
and rational functions.

2.1 Absolutely convergent series

I will assume that you are familiar with the basic properties of a series of complex numbers. We
briefly recall some of the deeper results on absolutely convergent series of complex numbers.
For more detailed treatments please look at “Baby Rudin”.
P
Definition 6. We say that a series of complex numbers n cn is absolutely convergent or
P P
summable with sum s if n cn → s and n |cn | converges.
Proposition 7. Let (zn )n∈N be a sequence of complex numbers.
P∞ P∞
a) The series n=0 |zn | converges iff n=0P zn converges with sum S and for any
∞
bijective mapping σ : N → N, the series n=0 zσ(n) converges also with sum S.
In other words, a series of complex numbers is absolutely convergent iff the all
rearrangements of the series converge to the same sum.
P
b) Suppose Z+ = i∈I Ei , where I is a countable indexing set. If the series ∞ n=0 |zn |
F
P
converges then each of the series l∈Ei zl converges with sum Si and furthermore
P P
the series i∈I Si converges with sum S, where the series ∞ n=0 zn converges with
sum S. In other words, any absolutely convergent series might be summed in arbi-
trary blocks and all such sums agree. This property is called associativity.
P 0 that of the series
P
Proof. Let sn denote the n-th partial sum of the series zn and sn zσ(n) .
Pm
Let ε > 0. First choose N0 so large so that n=N0 |zn | < ε, m > N0 and then choose N1 > N0
so that σ({1, 2 . . . , N1 }) ⊃ {1, 2, . . . , N0 }. Then it is clear that |sn − sn 0 | < ε proving that the
P P
series zσ(n) converges to the same sum as zn .
The converse of this theorem is the famous theorem of Riemann on rearrangements of series.
We will sketch a prove of the result for a series of real numbers. Suppose that {zn } is condition-
zn +|zn | zn −|zn | P + P +
ally convergent. First define z+ n := 2 and z−n := 2 . The series zn and zn are
nothing but the sums of the positive and negative terms, respectively. Note that both these
series diverge (why?). Now, let a > 0. We will prove that we can find a rearrangement of {zn },
P +
say {yn } that converges to a. As zn diverges, the partial sums will eventually exceed a.
The first few terms of our rearranged series will be taken the initial terms of {z+ n } so that the
partial sums just exceed a. Now, we add the initial terms from {zn } so that the sum of the
−

rearranged series is now just < a. We then add terms from {z+ n } again so that the sum just
exceeds a again. And now terms for {z− }. . . . We repeat this process. We can do this because
P + P − n
both zn and zn diverge. It is easy to see that the rearranged series does in fact converge
to a (this uses the fact that zn → 0). All the gory details of this argument is present in “Baby
Rudin”.
The second part is left as an exercise.

5
 2 Power series

P
Example 8. We will now consider a double series n,m zn,m . So we have a infinite matrix of
complex numbers which we might sum in any way we please: row-wise, column-wise, along
diagonals, etc. There is no guarantee that all these sums agree. For instance take zn+1,n = 1
and zn,n+1 = −1 and all other entries 0. Summing by rows we get −1 and summing by
P
columns we get 1. But if we could show that in some irder m,n |zm,n | is finite then by the
previous result we can sum up the series in any way we please.

Example 9. For q ∈ C, |q| < 1, we have

∞
X ∞
X
mqm qn
= .
1 − qm (1 − qn )2
m=1 n=1

To show this, we first expand

∞
X ∞
X
qn nm qn
= q and = mqnm .
1 − qm (1 − qm )2
m=1 m=1

We will now show that (mqmn )n,m>1 is summable in C. Then the result follows from asso-
ciativity. To do this, we take absolute values to pass to a series in R+ where we can sum up
in any way we choose:

X∞ X∞ ∞
X
!
|q|m
m|q| mn
= < ∞.
(1 − |q|n )
n=1 m=1 n=1

This finishes the proof.

P P
Proposition 10. Let n an and n bn be summable with sums A and B. Then the series
P Pk
ck , where ck = n=0 an bn−k , is summable with sum AB.
P
Proof. The series ck is summable as
X X X X
|ck | 6 |an ||bm | = ( |an |)( |bn |) < ∞.
k m,n m n
P
This means that the series n,m an bm is summable. Summing first in m and then in n we
P
get AB. Grouping the terms an,m in blocks where n + m = k, we get ck .

2.2 Sequences and series of functions

Throughout this section, X will be a set and {fn }n∈N is a sequence of complex valued functions
on E.

Definition 11. We say that fn converges uniformly to f : X → C if for each ε > 0, we can find
N ∈ N such that
|fn − f| < ε,
whenever n > N.
P
We say that the series n fn converges uniformly if the sequence of partial sums converge
uniformly.

6
 2 Power series

Remarks.

(i) We have a Cauchy criterion for uniform convergence: for each ε > 0, we must have:

|fn (x) − fm (x)| < ε,

for m, n sufficiently large, independent of the point x ∈ X.

(ii) If X were a metric space then the limit of any uniformly convergent sequence (or series)
of functions on X is automatically continuous.
P
Proposition 12 (Weirstrass M-test). If |fn (x)| 6 Mn ∀x ∈ X and Mn < ∞ then the series
P
fn is uniformly convergent.
Proof. This follows in a straightforward manner from Cauchy’s criterion. For ε > 0 and
m > n, we have
Xm X
n ∞
X
fk (p) 6 |fk (p)| 6 Mk ε,
n m k=n
for n sufficiently large.

The following identity is due to Abel and will be needed in proving Abel’s convergence
theorem for power series.
Lemma 13. Let (an ), (bn ) be two sequences of complex numbers and write An = a+· · ·+an .
Then
Xk Xn
ak bk = An bn+1 − (Ak (bk+1 − bk ), n ∈ N.
n=1 k=1

Proof. Setting A0 = 0, we see that

X
n X
n
ak b k = (Ak − Ak−1 )bk
k=1 k=1
Xn X
n−1 X
n
= Ak bk − Ak bk+1 = Ak (bk − bk+1 ) + An bn+1 .
k=1 k=1 k=1

2.3 Power series

Definition 14. Let {an }n ∈ N be a sequence of complex numbers. Then the power series with
coefficients {cn } centred at a is the series of functions
∞
X
cn (z − a)n .
n=0

The domain of convergence of the power series is the set

X
E := {z ∈ C : cn (z − a)n is convergent}.

We are interested in the set E and the properties of the limit function f : E → C. For instance:

7
 2 Power series

1. Is the convergence of the power series uniform in E?

2. Is E open? Closed? Connected?

We use the notion of upper limit of a sequence of numbers.

Definition 15. Let (xn ) be a sequence of real numbers then the set of subsequential limits E
is the set of all points x in R (extended real numbers) such that some subsequence of (xn )
converges to x. We define the upper limit of (xn ) to be

lim sup xn := sup E.

n

If ρ = lim supn xn is finite then it is the unique number that satisfies the following two
properties:

1. ρ ∈ E.
2. If x > ρ, only finitely many elements of the sequence (xn ) are > x.
P
Theorem 16. Let ∞ n=0 cn (z − a) be a power series. Let R := ρ , where ρ = lim supn |cn |
n 1 1/n

(here R := 0 if ρ = ∞ and R := ∞ if ρ = 0). Then the series converges uniformly on

compact subsets of the open unit disk D(a, R) and diverges on C \ D(a, R). Furthermore,
for each z ∈ D(a, R) the convergence is absolute. Thus D(a, R) ⊂ E ⊂ D(a, R).

Proof. If |z − a| > R, then |z − a|−1 < ρ and hence there are infinitely many terms |cn |1/n that
are greater than |z−a|−1 , i.e., infinitely many terms of the form |cn (z−a)n | > 1. Therefore, the
sequence (cn (z−a)n ) cannot converge to 0 showing that z 6∈ E. On the other hand if r < R and
1/n
|z − a| 6 r then |cn (z − a)n | 6 |cn |rn =: Mn and Mn = |cn |1/n r. As lim supn |cn |1/n = 1/R,
this means that given ε > 0, for large n, |cn |1/n < R 1
+ ε and thus
r
|cn |1/n r < + rε < 1,
R
P
for ε small. Thus by the root test, Mn converges and Weirstrass’s M-test delivers the
proof.

The number R above is called the radius of convergence of the power series and the ex-
pression ρ = lim supn |cn |1/n is called Hadamard’s formula.
P
Example 17. Consider the geometric series n zn . Then each cn = 1 and hence R = 1 and
the domain of convergence is the unit disk D. The convergence is not uniform on D. For
1 P
otherwise, the limit 1−z would have to be bounded. Also note that the series n z2m has
coefficients cn = 1 if n is even and 0 otherwise, has he same radius of convergence. It is easy to
P P
see that if the series n cn zn has radius of convergence R then the series n cn wkn obtained
by substituting z = wk has radius of convergence R1/k .

Also note that

X
R = sup{r > 0 : cn (z − a)n converges for z ∈ D(a, r)}
n
X
= sup{r > 0 : |cn |rn < ∞}
n

8
 2 Power series

P P
Given two series n cn (z − a)n and n dn (z − a)n whose radii of convergence are R1 and R2 ,
respectively, we can form both the sum and Cauchy product of the two series. From the above
relations it is clear that the radius of convergence of the sum is > min(R1 , R2 ). As the Cauchy
product of two summable series is also summable, it is clear that the radius of convergence of
the Cauchy product also has radius of convergence > min(R1 , R2 ).
P zn
Example 18. The series n>1 n 2 has radius of convergence 1 and converges uniformly on D by
zn
Weirstrass M-test as | n 2 6 1
n2
.
In general, the domain of convergence will contain some points on the circle C(a, R).
P n
Example 19. The series n>1 zn has radius of convergence 1. If z = 1, the series divergence.
On the other hand, when z = −1, it converges. We will say more about this series after study
more convergence tests.

2.4 Analytic functions

P P
Given a power series cn (z−a)n , we can formally write down its derivative ncn (z−a)n−1 .
P
Proposition 20. Let a ∈ C. We define the power series S(z) = cn (z − a)n and the series
P
S 0 (z) = ncn (z − a)n−1 . Then the radius of convergence of S and S 0 are equal.
Proof. Let R be the radius of convergence of S(z) and R 0 be the radius of convergence of S 0 (z).
Let z ∈ C be such that |z − a| < R 0 . Then by the root test

1 > lim sup |ncn (z − a)n−1 |1/n > lim sup |cn (z − a)n |1/n ,

as n|z − a|n−1 > |z − a|n for n > |z − a|. From the root test, it follows that S(z) also converges.
On the other hand, suppose that S(z) converges absolutely. Choose z0 such that |z − a| <
|z0 − a| < R. By the root test, it follows that |cn (z − a)n |1/n < 1 for all sufficiently large n.
Then:

1 = lim sup |n|1/n

|z − a|
> lim sup |n|1/n
|z0 − a|
1/n
|z − a| n−1
= lim sup n
|z0 − a|
1/n
|z − a| n−1
> n cn (z − a)n−1
|z0 − a|
= lim sup |ncn (z − a)n−1 |1/n .

Hence, R 0 > R and hence R = R 0 .

P
Theorem 21. If the power series ncn (z − a)n−1 has radius of convergence R > 0, then
P
the limit function f(z) = cn (z − a)n is holomorphic in the disk D(a, R) and f 0 (z) =
P
ncn (z − a)n−1 .

9
 2 Power series

Proof. Fix z ∈ D(a, R) and let ε > 0. We have to show that

f(w) − f(z) X
− ncn (z − a)n−1 < ε,
w−z n

for w suitably close to z. We write f(z) as SN (z) + RN (z) where SN is the n-th partial sum
of the power series of f and RN is the tail and break above the above formula as the sum of
three terms I + II + III:
∞
SN (w) − SN (z) X X
N
!
RN (w) − RN (z)
− ncn (z − a)n − ncn (z − a)n + .
w−z w−z
n=0 N

The last term is

X
cn ((w − a)n−1 + w − a)n−2 (z − a) + · · · + (w − a)(z − a)n−2 + (z − a)n−1 ),
n>N

where we just expanded out ((w − a) − (z − a))n−1 . Suppose |z − a| < ρ < R and |w − a| < ρ,
P
then |III| 6 n>N n|cn |ρn−1 . Since ρ < R, this is the tail of a a convergent series, so we
may choose N so large so that |III| < ε (here it is irrelevant whether z is close to w or not).
Similarly, we can assume that N is so large that |II| < ε as well as II is also the tail of a
convergent series. Finally, it clear that for w sufficiently close to z, |I| < ε. This because SN
P
is a polynomial whose derivative is ∞ n
n=0 ncn (z − a) . Hence, we are done.

Applying the above result iteratively, we see that f is infinitely-differentiable and the n-th
derivative f is given by a power series expansion that converges in the disk D(a, R).
We might, of course, differentiate a power series term-by-term to obtain its derivative.
Repeating this process n-times, the first term of the new series is n!cn . Substituting z = a
n (a)
in this new series gives fn (a). We see that cn = f n! . This shows that the if f has a power
series expansion centred at a point a with radius of convergence R > 0, then the coefficients
of this expansion are uniquely determined.
Now, suppose that f is a holomorphic function on D(a, R) and suppose f 0 has a power series
expansion
∞
X
f 0 (z) = dn (z − a)n ,
n=0

that converges in D(a, R) then f has the power series expansion

∞
X dn
f(z) = f(a) + (z − a)n+1 .
n+1
n=0

Indeed, the RHS above is a convergent power series on D(a, R) and, calling the holomorphic
function it defines g, we see that f 0 ≡ g 0 and f(a) = g(a). Hence, the holomorphic function
f − g has a derivative that vanishes on D(a, R) and hence f ≡ g.

Definition 22. Let f : U → C be a function where U ⊂ C. We say that f is complex-analytic if

for each a ∈ U, we can find a power series centred at a with disk of convergence D(a, R) ⊂ U
whose values agree with those of f, i.e., f can be locally expressed as a power series.

10
 2 Power series

The size of the disk of convergence will, in general, vary with the point z. However, we
will later prove that it is at least d(z, ∂U). In other words, the power series expansion of an
analytic function at a point z defines the values of the function on the largest disk centred at
that point wholly contained in U.
P
If n cn (z − a)n is a power series convergent in D(a, R), it is not at all obvious that it
defines an analytic function in D(a, R). This is the content of the following
P
Theorem 23. Let f(z) = cn (z − a)n be a convergent power series in D(a, R), R > 0 and
P
let b ∈ D(a, R). Then the series ∞ n=0
f(n) (b) n
n! (z − b) has radius of convergence at least
R − |a − b| and is the power series expansion of f centred at b.

Proof. As the derivatives of a power series can be obtained by differentiating term-by-term,

we see that
X∞
n (n + m)!
f (b) = cn+m (b − a)m ,
m!
m=0

and hence
∞
X (n + m)!
|f (n)
(b)| 6 |cn+m ||b − a|m .
m!
m=0

Let r be such that |b − a| 6 r < R and take a point z ∈ D(b, R − r). We will show that the
series
∞
X f(n) (b)
(z − b)n
n!
n=0

is absolutely convergent. We compute

∞
X ∞
X
|f(n) (b)| |f(n) (b)|
|z − b|n 6 (r − |b − a|)n
n! n!
n=0 n=0
X∞
(n + m)!
6 |cn+m ||b − a|m (r − |b − a|)n
m!n!
n,m=0
X∞ X ∞
X
(n + m)!
= |ck | |b − a| (r − |b − a|) =
m n
|ck |rk < ∞.
m!n!
k=0 m+n=k k=0

P
This shows that the series ∞ n=0
f(n) (b)
n! (z − b) has radius of convergence at least R − |b − a|.
n

The sum of the series can be computed by summing up arbitrarily in blocks and a computation
similar to the one above shows that for a point z ∈ D(b, R − |b − a|), we have

X ∞
X
n f(n) (b)
f(z) = cn (z − a) = (z − b)n ,
n!
n=0

proving the result.

11
 3 Examples of Analytic functions

3 Examples of Analytic functions

3.1 Polynomials and rational functions
The simplest class of functions on C are the polynomials:

P(z) = a0 + a1 z + · · · + an zn , aj ∈ C and an 6= 0.

The number n is called the degree of P and denoted def(P). We will use the notation K[x] to
denote the ring of all polynomials in the variable x over the field K. Obviously, a polynomial
P : C → C is a continuous function. We may write P(z) = P1 (x, y) + iP2 (x, y), where P1 , P2 ∈
R[x, y] are real-valued polynomials in the variables x and y. We can also substitute z = x + iy
in the expression of P to obtain a polynomial P(x, y) ∈ C[x, y]. An interesting question: given
a polynomial P(x, y) = P1 (x, y) + iP2 (x, y), P1 , P2 ∈ R[x, y], under what conditions does it
follow that P ∈ C[z]? A detailed answer is given in the textbook. We will be content to give
a simple answer. Set x = z+z z−z
2 and y = 2i and substitute in the expression for P and check
that after simplification, there are no terms involving z.
A similar question is the following: if P ∈ C[z, z], when does it follow that P ∈ R[x, y]? Well,
writing X
P(z, z) = ak,l zk zl ,
k+l6n

we see that if P is to take only real values, it follows that P(z) = P(z) and this can happen iff
ak,l = al,k .
A polynomial is obviously a complex-analytic function on the whole of C. In fact, the
expression of the polynomial is nothing but the power series expansion around 0. This power
series has only finitely many terms. An easy way to find the power series expansion about
some other point is using Taylor’s formula. Alternatively, we could write z as (z − a) + a and
substitute in the expression for P(z) and expand out using the binomial theorem.
A rational function is any function of the type
P(z)
R(z) = .
Q(z)
Unless otherwise specified, we will always assume that any common factors have been cancelled
out. Set Z(Q) := {z ∈ C : Q(z) = 0}. On U := C \ Z(Q), R is a well-defined smooth function.
We will now show that these function are in fact complex-analytic on U. The points of the set
Z(Q) are called the poles of R. We will now facts about polynomials:

 The fundamental theorem of algebra. Any non-constant polynomial Q ∈ C[z] can be fac-
torized to the form c(z − a1 )n1 . . . (z − ak )nk where c ∈ C and ai ∈ C are the roots of
Q.
 The division algorithm.If P(z), Q(z) ∈ C[z] such that Q(z) 6= 0, them we can find q(z), r(z) ∈
C[z] such that
Q(z) = q(z)P(z) + r(z), deg(r(z)) < deg(Q(z)).

The above two results imply that R(z) is nothing but a finite sum of polynomials and simple
fractions of the type α · (z − b)−k where each b is root of Q and consequently a pole of R. It

12
 3 Examples of Analytic functions

suffices to prove that each such (z − b)−k has a power series expansion around a point a ∈ U.
Set r = |a − b| > 0. We have
1 1 1 X (z − a)n
= = z−a
=− .
z−b (z − a) − (b − a) b−a − 1 (b − a)
(b − a)n+1
1
The power series expansion of (z−a)k
is can be obtained by differentiating the above series
term-by-term k-times.

3.2 The exponential function

Let
∞
X zn
ez := ,
n!
n=0
where the radius of convergence of the series is ∞. To see thus note that for each z ∈ C, we
have:
|z|n+1 /(n + 1)! |z|
= → 0,
|z| /n!
n n
P
and we conclude by ratio test that the series n |z|n /n! is convergent for each z ∈ C.
Furthermore,

X zn X wm
ez · ew = ·
n
n! m m!
X 1 X k! n m X (z + w)k
= z w = = ez+w .
k! n!m! k!
k n+m=k k

Hence, ez = ex · ei y, when z = x + iy. Now,

X in yn
eiy = .
n
n!
The even terms give the real part of eiy and the odd terms give the imaginary part. Hence,
∞
X y2k
Re(eiy ) = (−1)k = cos y
(2k)!
k=0
X∞
y2k+1
Im(eiy ) = (−1)k = sin y
(2k + 1)!
k=0

This proves that eiy = cos y + i sin y and our definition of ez is in agreement with the earlier
definition of eit . This also motivates the following definitions of the trigonometric functions
eiz e−iz eiz − e−iz
cos z := , sin z :=
2 2
ez + e−z ez − e−z
cosh z = , sinh z = .
2 2
The following properties of the exponential are easy to see:

13
 4 Linear fractional transformations

 ez 6= 0 ∀z ∈ C.
 e−1/z = 1/ez .
 The exponential function is 2πi-periodic. In particular, the equation ez = 1 has solution
set 2πiZ.
 If ez = ew then z − w ∈ 2πiZ.
 |ez | = eRez , Imz ∈ arg ez .

The above properties of the exponential easily imply that the exponential function is a
bijective map from the strip B = {z : −π < Imz 6 π} to C \ {0} and that it is injective on all
horizontal strips of width less than 2π.

3.3 Logarithms
Definition 24. For 0 6= z ∈ C, we define log z to be any complex number w such that ew = z.

Just as in the case of arg z, this is “multi-valued” function and as before we are interested
in continuous branches. Now the exponential function restricted to R is a bijective mapping
onto R+ and hence has an inverse, the natural logarithm, Log : R+ → R. If ew = z, then
eRew = |z| and eImw ∈ arg z. Thus Rew = Log |z| and Imw could be any of the possible values
of arg z. This shows that any two possible values of log z differ by an integer multiple of 2π.

Definition 25. We define the principal branch of the logarithm by

Log z = Log |z| + i Arg z.

Thus, Log : C \ {0} → C. Moreover, the function z 7→ Log |z| is a continuous function and
thus the discontinuity of Log arises from the discontinuities of Arg. It should be clear that we
can define the notion of the continuous branch of the logarithm exactly as we did before for
the argument function. It is obvious that a domain U has a continuous branch of the logarithm
iff it has a continuous branch of the argument function.

3.4 Complex powers

We have already seen how to define powers zq , where q is a rational number. Now, we
generalize the same definition to arbitrary w. To do this, observe that if u = zw and w is an
integer then log zw satisfies the property that ew log z = (elog z )w = zw + 2πZi.

Definition 26. Let z, w ∈ C, z 6= 0, we define zw to be the set of complex numbers {ew log z where
log z is any branch of the logarithm of z.

4 Linear fractional transformations

4.1 The Riemann Sphere
As in the case of R, it is often convenient to add a point ∞ to the complex plane. We will
perform a standard construction from topology called the Alexandroff one-point compactifica-
tion. Essentially, we add a new point ∞ to C and call this space the extended complex plane

14
 4 Linear fractional transformations

or the Riemann sphere denoted C. b We equip C b with a topology that makes it a compact
topological space and such that C ⊂ C
b is a subspace.

Definition 27. The topology of C∗ := C ∪ {∞} is given by the set B that consists of any basis of
C and the sets of the form
{z : |z| > r} ∪ {∞}.

It is clear that C∗ contains C as a subspace. The compactness of C∗ follows easily from the
fact that any neighbourhood of ∞ must necessarily contain an open set of the form {z : |z| > r},
the complement of which is certainly compact.
We will now give a concrete realization of the space C∗ . Let S2 := {(x, y, z) ∈ R3 : x2 + y2 +
z2 = 1} be the unit sphere in R3 and let N := (0, 0, 1) be the north pole. We consider C as
the equator plane of S2 . In particular, the boundary of the unit D ⊂ C is the equator of S2 .
Given a point P = (X, Y, Z) ∈ S2 , we let z = x + iy ∼ (x, y, 0) be the point in C that lies on
the straight line that contains P and N. This mapping π : P 7→ z is called the stereographic
projection. To find formula, we first observe that the parametrically the line joining N to P
has the equation N + t(P − N), t ∈ R. Thus the image of P under the stereographic projection
must satisfy
(x, y, 0) = (0, 0, 1) + t((X, Y, Z − 1)),
1 X
which means that 1 + t(Z − 1) = 0 or t = 1−Z . Substituting for t, we get x = 1−Z and
Y
y = 1−Z .
On the other hand, starting with a point z = x + iy ∼ (x, y, 0) ∈ C, we can solve for X, Y, Z
in terms of x and y. We have X2 + Y 2 + Z2 = 1 and multiplying this equation by t2 and
substituting tX = x and tY = y and tZ = t − 1, we see that x2 + y2 + t2 − 2t + 1 = t2 which
shows = |z| 2+1 . Thus,
2

2x
X=
1 + |z|2
2y
Y=
1 + |z|2
|z|2 − 1
Z= 2 .
|z| + 1

The above computations proves that the stereographic projection is an injective and surjective
mapping from S2 \ N → C. Moreover, S2 is a metric space where the metric is induced from
R3 . Both the stereographic projection and its inverse are continuous and hence C has been
embedded into S2 . The north pole is to be thought of as the point at ∞ and we can now
identify C∗ with S2 .
The following properties of stereographic projection are geometrically obvious:

1. It maps the southern hemisphere to the unit disk with the south pole getting mapped
to 0.
2. The northern hemisphere minus the north pole gets mapped to the complement of the
unit disk.
3. The equator gets mapped to the unit circle.

15
 4 Linear fractional transformations

We also have the following obvious geometric fact. Our proof is analytic. For a proof using
classical geometry, see Bill Casselman’s article.

Theorem 28. Under stereographic projection, circles on S2 corresponds to circles or straight

lines in the complex plane.

Proof. The locus of points that satisfy a quadratic equation of the form

x2 + y2 + ax + by + c = 0

is either a circle, a point or the empty set. To see this, complete the squares,

(x + a/2)2 + (y + b/2)2 = a2 /4 + b2 /4 − c,

and note the three possibilities correspond respectively to a2 /4 + b2 /4 − c being strictly

positive, zero or strictly negative.
Any circle on S2 is nothing but the intersection of a plane AX + BY + C = D with S2 . Then
the sterographic projection of this circle consists of the complex numbers z = x + iy that
satisfy
2x 2y |z|2 − 1
A 2 +B 2 +C 2 = D.
|z| + 1 |z| + 1 |z| + 1
Rewriting,
(C − D)(x2 + y2 ) + 2Ax + 2By − (C + D) = 0
If C = D, the locus is a straight line. If C 6= D, then dividing by (C−D), we obtain a quadratic
and the locus must be a circle as it cannot be empty.
Conversely, starting with a circle on the plane

x2 + y2 + A 0 x + B 0 y + D 0 = 0,

we set 2A = A 0 , 2B = B 0 , C − D = 1, −(C + D) = D 0 , and the corresponding set on the sphere

is the intersection of the sphere with the plane AX + BY + CZ = D which cannot be empty
and hence must be a circle. Similarly, a straight line has the equation

A 0x + B 0y = D 0

which determines a circle on the sphere given by the intersection of the sphere with the plane
AX + BY + CZ = D, 2A = A 0 , 2B = B 0 , C = D = D 0 /2 and this plane meets the sphere in a
circle that passes through the north pole.

We will consider straight lines as circles through the point infinity. We end this section with
some remarks on other ways we could carry out stereographic projections. Call a sphere S in
R3 admissible if its north pole lies in the upper half-space H = {x3 > 0}, and, for such spheres,
denote by PS the stereographic projection from the north pole N0 of S, which identifies C∗
with S.

16
 4 Linear fractional transformations

4.2 Linear fractional transformations

The fractional linear transformations are the rational mappings of the form
az + b
T z := , ad − bc 6= 0.
cz + d
Note that if ad − bc = 0, then
adz + bd b
= .
d(cz + d) d
Thus, the condition ad − bc 6= 0 ensures that that T is a non-constant mapping.
We can consider T as a mapping from C∗ to itself by setting T (−d/c) = ∞ and T (∞) = a/c.
Thus T is a surjective mapping of C∗ . Furthermore, an easy computation shows that T is
invertible with inverse
dw − b
.
−cw + a
It is also easy to see that the composition of two linear fractional transformations is also a
linear fractional transformation. Thus the collection of all such transformations is group called
the linear group. Consider the following special linear fractional transformations

 Translations: z 7→ z + c;
 Inversion: z 7→ 1/z;
 Homothety: z → 7 αz.

It is possible to write every linear fractional transformations as a composition of the above

three simple mappings:
az + b bc − ad a
= 2 +
cz + d c (z + d/c) c
Let S and S 0 be admissible spheres, and also a rigid motion T of R3 such that S 0 := T S is also
admissible, i.e., T N0 ∈ H. Consider the composition TS 0 ,S = PS 0 ◦ T ◦ PS−1 which maps C∗ to
itself. In fact, every linear fractional transformation can be obtained this way.

Theorem 29. A complex mapping is a linear fractional transformation if and only if it can
be obtained by stereographic projection of the complex plane onto an admissible sphere,
followed by a rigid motion of the sphere which maps it to another admissible sphere,
followed by stereographic projection back to the plane.

Proof. We write the linear fractional transformation F as

ρeiθ
F(z) = + β.
z+α
If S is any admissible sphere, then translating the sphere by α gives us another admissible
sphere S 0 and it is obvious that TS 0 ,S is nothing but translation by α.
To obtain rotation, dilations and inversion, we choose S to be the unit sphere. Rotations
corresponds to the same rotations of R3 along the x3 -axis. To obtain a dilation by ρ, we move
the sphere upwards by distance ρ − 1. And to obtain inversion, we rotate the sphere around
the real-axis by an angle π. This proves the claim.

17