You are on page 1of 7

Chapter 9: Analytic Functions and Their Taylor Expansions

The class of analytic functions can be considered as a substantual extension to the


class of polynomials. Given a polynomial function p(z) and a point z0 in the complex
plane, we have the Taylor expansion

N p(n) (z0 )
p(z) = cn (z − z0 )n , where cn = ,
n= 0 n!

which is a finite series, that is, a series with finitely many terms. If we allow such an
expansion to be an infinite series, then we can deal wih a class of functions substantially
larger than polynomials. We say that a complex function f is analytic at a point z0 if
there is a power series expansion (with a positive radius of convergence) at z0 :
∞
f (z) = cn (z − z0 )n . (9.1)
n= 0

(In the future we will see that the identity cn = f (n) (z0 )/n! still holds.) Now (9.1) means
N
f (z) = lim cn (z − z0 )n .
N→ ∞ n= 0

This tells us that analytic functions can be locally approximated by polynomials. This
inspires us to put down the following definition: a complex function f (z) is analytic at z0
if it can be uniformly approximated by polynomials over a disk, in other words, there exists
r > 0 such that, given any ε > 0, there exists a polynomial p(z) such that |f (z) − p(z)| < ε
for all z with |z − z0 | ≤ r, that is, for all z ∈ D(z0 , r).

We begin with the general study of power series of the form appearing on the right
hand sude of (9.1). Let us assume that there exists R > 0 such that the sequence {|cn |Rn }
is bounded, that is, there exists M > 0 such that |cn |Rn ≤ M for all n. Take any positive
number r strictly less than R, that is, 0 < r < R. Then, for all z ∈ D(z0 , r),

|cn (z − z0 )n | = |cn ||z − z0 |n ≤ |cn |rn = |cn |Rn (r/R)n ≤ M (r/R)n ≡ Mn

Since r < R, we have r/R < 1 and hence


∞ ∞
Mn = M (r/R)n = M (1 − (r/R))−1 < ∞
n= 0 n= 0

1
∞ n
The Weierstrass M –test tells us that the series n= 0 cn (z − z0 ) converges uniformly on
D(z0 , r). We have proved the following:

Fact. If the sequence {|cn |Rn }n≥0 is bounded, then, for each positive r < R, the
∞ n
series n= 0 cn (z − z0 ) converges uniformly over the disk D(z0 , r), in other words, the
N n
sequence of polynomials n= 0 cn (z − z0 ) (N ≥ 0) converges uniformly to some function
f (z) over D(z0 , r)
∞ n
Consider the power series n= 0 z /n!, with cn = 1/n!. For any R > 0, we have
limn→ ∞ Rn /n! = 0 and hence the sequence {Rn /n!}n≥0 is bounded. By Theorem the
above fact, this power series convegers uniformly on the disk Dr = {z ∈ C : |z| ≤ r} as
∞ n
long as r < R. Since R > 0 is arbitrary, we see that the series n= 0 z /n! converges to
some f (z) at every point z in the complex plane. Moreover, the convergence is uniform
over any disk. The sum of this series turns out to be ez :
∞
z n /n! = ez
n= 0

Recall that, for real t, eit = cos t + i sin t and e−it = cos t − i sin t. We can extend these
identities by letting t to be the complex z:

eiz = cos z + i sin z, e−iz = cos z − i sin z

An standard algebraic procedure converts these two identities into

eiz + e−iz eiz − e−iz


cos z = , sin z =
2 2i
which should be treated as the definition of cos z and sin z for complex z. By using
the power series expansion for ez given above, we can deduce

∞ z 2k ∞ z 2k+ 1
cos z = (−1)k , sin z = (−1)k .
k= 0 (2k)! k= 0 (2k + 1)!

Both series converge at any point in the complex plane.

Now we come to a basic definition (that must be read slowly and be understood
thoroughly)): a complex function f (z) with its domain D in the complex plane C is
said to be analytic if, for each point z0 in D, there is some r > 0 such that the disk
D(z0 , r) is contained in D and there is a sequence of polynomials {pn (z)}n≥1 converges
uniformly to f (z) over D(z0 , r), that is, there is a sequence of positive numbers εn

2
(n ≥ 1) converging to zero such that |f (z) − pn (z)| < εn for all z in D(z0 , r) and for all
n. In short, we define analytic functions to be those complex functions which can be locally
approximated by polynomials in the uniform sense. For example, under the assumption
∞ n
of the previous theorem, the limit function f (z) of the series n= 0 cn (z − z0 ) is an
analytic function, according to the conclusion of the theorem.

Now we suppose that f (z) is an analytic function on D and let D(z0 , r), pn (z),
and εn be those described in the above definition. The Cauchy’s formula for polynomials
tells us that 
1 pn (ζ)
pn (z) = dζ for all z with |z − z0 | < r,
2πi C ζ − z
where C is the boundary circle of the disk D(z0 , r), given by the parametric equation
z = z0 + reit (0 ≤ t ≤ 2π). Letting n → ∞, we get
 
1 pn (ζ) 1 f (ζ)
f (z) = lim pn (z) = lim dζ = dζ. (9.2)
n→ ∞ 2πi C n→ ∞ ζ −z 2πi C ζ − z
We have arrived at Cauchy’s formula for an analytic function f (z):

1 f (ζ)
f (z) = dζ for all z with |z − z0 | < r. (9.3)
2πi C ζ − z
Exchanging the integral sign and the limit sign in (9.2) needs to be justified. To justify
(9.2), it is enough to check that
 
1 pn (ζ) 1 f (ζ)
In = dζ − dζ
2πi C ζ − z 2πi C ζ − z
converges to zero as n → ∞. Now
   
1  2π pn (z0 + reit ) − f (z0 + reit )

 1 pn (ζ) − f (ζ)  it

|In | = 
 dζ  = ire dt
2πi C ζ −z 2π  0 (z0 + reit ) − z 
 2π  2π
1 |pn (z0 + reit ) − f (z0 + reit )| 1 εn εn
≤ it
dt ≤ dt = →0
2π 0 |re − (z − z0 )| 2π 0 δ δ

as n → ∞. Here δ = r − |z − z0 | > 0. Notice that |reit − (z − z0 )| ≥ |reit | − |z − z0 | = δ, in


view of the following general inequality for complex numbers ζ and η: |ζ − η| ≥ |ζ| − |η|.
Thus (9.3) is jusified.

We return to Cauchy’s formula (9.3). First, we manipulate the Cauchy kernel


1/(ζ − z) as follows:
 −1
1 −1 −1 z − z0
= ((ζ − z0 ) − (z − z0 )) = (ζ − z0 ) 1− . (9.4)
ζ −z ζ − z0

3
Since z is a point in the interior of the disk D(z0 , r) and ζ is a point on its circumference
C, we have |z − z0 | < r = |ζ − z0 | and hence |(z − z0 )/(ζ − z0 )| < 1. Thus we are allowed
to expand the right hand side of (9.4) into a geometric series:
 n 
1 1 ∞ z − z0 ∞ (z − z0 )n
= = .
ζ −z ζ − z0 n= 0 ζ − z0 n= 0 (ζ − z0 )n+ 1

Substituting this into (9.3) and exchanging the integral and summation signs, we arrive at
the Taylor expansion for f at z0 :
∞
f (z) = cn (z − z0 )n , (9.5)
n= 0

where the nth Taylor coefficient cn is given by



1 f (ζ)
cn = dζ. (9.6)
2πi C (ζ − z0 )n+ 1

The justification of exchanging the summation sign and the integral sign is left to the
reader as an exercise:
 ∞ ∞ 
 (z − z0 )n  (z − z0 )n
Problem 9.1. Prove: f (ζ) dζ = f (ζ) dζ.
C n= 0 (ζ − z0 )n+ 1
n= 0 C (ζ − z0 )n+ 1

The Taylor expansion above tells us that an analytic function has a power series expansion
at each point. Earlier we have indicated that the converse is also true. Thus analytic
functions are those complex functions which have local power series expansion.

Problem 9.2. (a) Use (9.6) to show that, if |f (ζ)| ≤ M for all ζ on the circle C, then

M
|cn | ≤ (9.7)
rn
for all n. (b) An analytic function defined on the wole complex plane C is called an entire
function. Use the estimate (9.7) to prove Liouville’s theorem: bounded entire functions
are constant.

According to Chapter 5, the Taylor coefficients cn in p(z) = ∞n= 0 cn (z − z0 )n s for




a polynomial function p(z) is given by cn = p′ (z)/n!. In order to prove this formula for
analytic functions, we need to study complex differentiation. The derivative of a complex
function f (z) at a point z0 is defined to be the following limit of a difference quotient

f (w) − f (z)
f ′ (z) = lim
w→ z w−z

4
provid that this limit exists. If f ′ (z) exists for every point z in the domain D (assumed
to be an open set), then we say that f is complex differentiable. When f (z) is a polynomial
function, say p(z), then f (z) is complex differentiable and its derivative f ′ (z) is the
same as the derivative p′ (z) of the polynomial function described in Chapter 5. The
following example shows that not all nice functions are complex differentiable.

Example 9.1. The function f (z) = z is nowhere differentiable in the complex sense.
Indeed, writing w = z + reiθ , we have

f (w) − f (z) w−z re−iθ


= = = e−2iθ .
w−z w−z reiθ

Notice that ‘w → z’ means ‘ρ → 0′ , but θ can take any value. So the difference
quotient (f (w) − f (z))/(w − z) does not “settle down” to a definite value as w → z.

In the future we will give more precise conditions for a function to be complex differentiable
so that we can easily see why functins like f (z) = z are not so. At present we remark that
the usual product rule and quotient rule still hold for complex differentiation. Our main
goal next is to prove that analytic functions are complex differentiable an we are going to
relate their derivatives to Cauchy’s formula. The basic technical side of our argument has
been covered in some analogous situations from the last and the present chapters and its
details are left to the reader as exercises. Our starting point is Cauchy’s formula (9.3),
from which it follows that
   
f (w) − f (z) 1 1 f (ζ) f (ζ) 1 f (ζ)
= − dζ = dζ,
w−z 2πi C w − z ζ −w ζ −z 2πi C (ζ − w)(ζ − z)

Hence we have
 
′ f (w) − f (z) 1 f (ζ) 1 f (ζ)
f (z) = lim = lim dζ = dζ.
w→ z w−z w→ z 2πi C (ζ − w)(ζ − z) 2πi C (ζ − z)2

Here the reader is asked to fill in the detail of the above argument:

 2π  2π
Problem 9.3. Verify the step “limw→ z 0 = 0 limw→ z” in the above argument.

By means of induction, we can show


 n   n
(n) ∂ 1 ∂ f (ζ)
f (z) = f (z) = dζ (9.8)
∂z 2πi C ∂z ζ −z

5
with (∂/∂z)n (ζ − z)−1 = n!(ζ −z)−n−1 . Evaluating at z = z0 , we have

(n) n! f (ζ)
f (z0 ) = dζ = n!cn .
2πi C (ζ − z0 )n+ 1

Hence the Taylor coefficient cn in the power series expansion (5.3) is also given by

f (n) (z0 )
cn = , (9.9)
n!
which is often easier to apply for computational purposes. Thus, given an analytic function
f and a point z0 in its domain, we have the following Taylor’s formula:
∞ 

n f (n) (z0 ) 1 f (ζ)
f (z) = cn (z − z0 ) , where cn = = dζ.
n= 0
n! 2πi ∂D (ζ − z0 )n+ 1

where D is any (closed) disk contained in the domain of f with z0 as its center.

Let us mention some “bonus results” form the above discussion. First, all analytic
functions have complex derivatives of all orders. Second, if a seqeunce of analytic functions
fn conveges uniformly to a function f , then f is also analytic and the kth derivatives of
fn converges to the kth derivative of f at each point. This follows from (9.8), but the
detailed argument is left to the reader:

Problem 9.4. Prove the last statement.


∞ n
Consider a power series n= 0 cn (z − z0 ) , assumed to be uniformly convergent to f (z)
N
on a disk D centered at z0 . Then its partial sum pN (z) = n= 0 cn (z − z0 )n (N ≥ 0)
converges uniformly to f (z) as N → ∞ and hence its kth derivative, which is

N N
(k)

n−k
 n!
pN (z) = cn n(n − 1) · · · (n − k + 1)(z − z0 ) ≡ ck (z − z0 )n−k
(n − k)!
n= k n= k

for N ≥ k, converges to f (k) (z) for each point z in this disk. Thus we have
∞ n!
f (k) (z) = ck (z − z0 )n−k
n= k (n − k)!
(9.10)
2 (k + 2)! 2
≡ k!ck + (k + 1)!ck+ 1 (z − z0 ) + ck+ 2 (z − z0 ) + · · · .
2!

In particular, this identity is true for z = z0 : f (k) (z0 ) = k!ck , which gives ck =
∞
f (k) (z0 )/k!. We have proved that if a power series n= 0 cn (z − z0 )
n
converges uniformly

6
to a function f (z) on a disk center at z0 , then this power series is necessarily the Taylor
series expansion of f (z) at z0 .

Example 9.2. Let f (z) = 1/(1 − z). It is well-known that this function can be
reprsented as a geometric series

1
f (z) = = 1 + z + z2 + z3 + · · ·
1−z

if |z| < 1. From the above discussion we know this is also the Taylor expansion of f (z) at
z = 0. The nth Taylor coefficient is given by f (n) (0)/n! = 1 and hence f (n) (0) = n!. You
can obtain f (n) (0) = n! by direct computation but it takes longer in that way. Indeed,
by induction, we can show
n!
f (n) (z) =
(1 − z)n+ 1
which gives f (n) (0) = n!.
∞
Recall the power series ez = n
n= 0 z /n! which convegers uniformly on any disk
Dr = {z ∈ C : |z| ≤ r}. By (9.10), we have, for each z,
∞ n! 1 n−k ∞ 1 ∞ 1 n
f (k) (z) = z = z n−k = z = f (z),
n= k (n − k)! n! n= k (n − k)! n= 0 n!

showing that all derivatives of f (z) is equal to f (z) itself. The Taylor expansion of f (z)
at z = z0 is given by
∞ ∞ ∞
 f (n) (z0 ) n
 f (z0 ) n

f (z) = (z − z0 ) = (z − z0 ) = f (z0 ) (z − z0 )n = f (z0 )f (z − z0 ).
n= 0
n! n= 0
n! n= 0

Replace z0 by v and z − z0 by w so that z becomes v + w. Then we have


f (v+w) = f (v)f (w). The function f (z) is called the (natural) exponential function and is
denoted by ez . Then f (n) (z) = f (z) becomes (d/dz)n ez = ez and f (v + w) = f (v)f (w)
becomes ev+ w = ev ew .

You might also like