Complex Variables

020701 F. Porter
Revision 091006

Introduction

This note is intended as a review and reference for the basic theory of complex
variables. For further material, and more rigor, Whittaker and Watson is
recommended, though there are very many sources available, including a
brief review appendix in Matthews and Walker.

Complex Numbers

Let z be a complex number, which may be written in the forms:

z = x + iy
= rei ,

(1)
(2)

where x, y, r, and are real numbers. The quantities x and y are referred
to as the real and imaginary parts of z, respectively:
x = (z),
y = (z).

(3)
(4)

The quantity r is referred to as the modulus or absolute value of z,

r = |z| =

x2 + y 2 ,

(5)

and is called the argument, = arg(z), or the phase, or simply the angle
of z. We have the transformation between these two representations:

and nally also

x = r cos ,
y = r sin ,

(6)
(7)

= tan1 (y/x),

(8)

with due attention to quadrant. Noticing that ei = ei(+2n) , where n is any

integer, we say that the principal value of arg z is in the range:
< arg z .

(9)

y
z
r

z*

Figure 1: Complex number and its complex conjugate.

The complex conjugate, z , of z is obtained from z by changing the

sign of the imaginary part:
and
The product of two complex
z1 i
z = numbers,
x iy = re
. z2 , is given by:
z1 z2 = r1 ei1 r2 ei2 = r1 r2 ei(1 +2 )
= (x1 + iy1 )(x2 + iy2 ) = (x1 x2 y1 y2 ) + i(x1 y2 + x2 y1 ).
Notice that

zz = x2 + y 2 = |z|2 .

(10)

(11)

(12)

It is also interesting to notice that in the product:

z1 z2 = (x1 x2 + y1 y2 ) i(x1 y2 x2 y1 ),

(13)

the real part looks something like a scalar product of two vectors, and the
imaginary part resembles a cross product.

Complex Functions of a Complex Variable

We are interested in (complex-valued) functions of a complex variable z. In

particular, we are especially interested in functions which are single-valued,
continuous, and possess a derivative in some region.

Dening a suitable derivative requires some care. Start with the denition
for real functions of a real number:
f (x + x) f (x)
df
(x) = lim
.
(14)
x0
dx
x
But in the complex case we have real and imaginary parts to worry about.
First, dene what we mean by a limit. Let f (z) be a single-valued function
dened at all points in a neighborhood of z0 (except possibly at z0 ). Then
we say that f (z) w0 as z z0 , or limzz0 f (z) = w0 , if, for every  > 0,
there exists a > 0 such that (Fig. 2):
f  (x) =

|f (z) w0 | <  z

satisfying 0 < |z z0 | < .

(15)

Note that we have not required f (z0 ) to be dened, in order to dene the
limit (Fig. 3).
y
d
z0

Figure 2: Circle of radius about z0 .

"f(z)"

z0

"z"

Figure 3: Function not dened at z0 .

However, in order to dene the derivative at z0 , we require f (z0 ) to be
dened. If limzz0 = f (z0 ), where the limit exists, then we say that f (z) is
continuous at z0 . In general f (z) is complex, and we may write:
f (z) = u(x, y) + iv(x, y),

(16)

where u and v are real. Then limzz0 = f (z0 ) implies

lim

u(x, y) = u(x0 , y0 ),

(17)

lim

v(x, y) = v(x0 , y0 ),

(18)

xx0 , yy0
xx0 , yy0

where the path of approach to the limit point must lie within the region of
denition. We may thus dene continuity to the boundary of a closed region,
if the path is within the region.
Now, in our denition of f  (z), we note that there are an innite number
of possible paths along which we can make z = x + iy 0.

x
Figure 4: Various paths along which to approach a point.
For our derivative to be well-dened, we demand that the value of f  (z)
be independent of the way in which z 0. Thus, if we approach along the
path x = 0:
f (z + iy) f (z)
y0
iy


u(x, y + y) u(x, y) i [v(x, y + y) v(x, y)]
+
= lim
y0
iy
iy
u v
+
.
(19)
= i
y y

f  (z) =

lim

If instead we make our approach along the path y = 0, we obtain:

f (z + x) f (z)
x0
x
v
u
+i
=
x
x

f  (z) =

lim

(20)

The two expressions are equal if and only if the real and imaginary parts are
separately equal:
v
u
=
x
y
u
v
= .
y
x

(21)
(22)

These important conditions are known as the Cauchy Riemann equations,

or C-R equations, for short. We may state this in the following theorem:
Theorem: If u, v possess rst derivatives throughout a neighborhood of z0 ,
which are continuous at z0 , then the Cauchy Riemann equations, if
df
satised, guarantee the existence of dz
(z0 ).
Proof: Write:
u
x +
x
v
v =
x +
x

u =

u
y + ux x + uy y
y
v
y + vx x + vy y,
y

(23)
(24)

where the correction terms for non-linearities, ij , approach zero as

x, y 0.
Using the Cauchy Riemann equations, we obtain:
u
x
x
v
v =
x +
x

u =

v
y + ux x + uy y
x
u
y + vx x + vy y.
x

(25)
(26)

Thus,
f
z

u + iv
z
u
v
(x
+ iy) + i x
(x + iy) + x x + y y
, (27)
= x
x + iy
=

where x ux + ivx 0, y uy + ivy 0 as x, y 0.

Furthermore,




x




 x + iy 

1,





y




 x + iy 

1.

(28)

Therefore,
df
u
v
u + iv
= lim
=
+i ,
z0
dz
z
x
x
independent of path. This completes the proof.

(29)

We have the following equivalent ways of expressing the derivative:

df
u
v
v
v
v
u
u
u
=
+i
=
+i
=
i
=
i .
dz
x
x
y
x
y
y
x
y

(30)

It is of interest to also consider this discussion in terms of the polar form.

In this case, we may consider the r = 0 path:
y
z+D z

Dz

Dq z
q
x

Figure 5: Polar path description.

f (zei ) f (z)
0
z(ei 1)


u(r, + ) u(r, ) i [v(r, + ) v(r, )]
+
= lim
0
iz
iz
i u 1 v
+
.
(31)
=
z z
Similarly, for the = 0 path:
f  (z) =

lim

f  (z) =

lim

f (r + r)ei f (z)

ei r

u(r + r, ) u(r, ) i [v(r + r, ) v(r, )]
= lim
+
r0
ei r
ei r
i v
1 u
+ i .
(32)
= i
e r e r
r0

Hence,
u
v
i u 1 v
+i
=
+
.
r
r
r r
We have thus obtained the Cauchy-Riemann relations in polar form:
ei f  (z) =

1 v
u
=
r
r
1 u
v
=
.
r
r
6

(33)

(34)
(35)

A function f (z) of complex variable z is called analytic at the point z0 if

it is single-valued and possesses a derivative at every point in a neighborhood
of z0 . Otherwise, z0 is a singular point of f (z). If f (z) is analytic at every
point in a simply connected open region (domain) D, then it is referred
to as analytic throughout D. Other terms that are often used for this (with
some variation of meaning) are regular and holomorphic. Sometimes the
term analytic is not required to be single-valued, that is, single-valuedness
in a domain D means that, after following any closed path in D, the function
f (z) returns to its initial value. If f (z) is analytic for all nite z, then f (z)
is an entire function.
Examples:
f (z) = z 3 is an entire function.
f (z) = 1/z 2 is analytic everywhere except at z = 0, where it is not
dened. We note that for this function,
u=

x2 y 2
,
(x2 + y 2 )2

v=

2xy
.
(x2 + y 2)2

(36)

f (z) = z 3/2 is analytic everywhere except at z = 0. Lets look at why

this is the case in some detail. We may write
z 3/2 = r 3/2 e3i/2 = r 3/2 (cos 3/2 + i sin 3/2).

(37)

Hence,
u
r
v
r
1 u
r
1 v
r

3 1/2
r cos 3/2
2
3 1/2
=
r sin 3/2
2
3
= r 1/2 sin 3/2
2
3 1/2
=
r cos 3/2.
2
=

(38)
(39)
(40)
(41)

Comparison with Eqns. 34 and 35 shows that the C-R conditions are
satised everywhere. Now consider a path containing the origin as an
interior point (see Fig. 6). Well start at z = ei0 , with  real. Table 1
shows the values of f (z) as we traverse the path once around the origin.
We see that f (z) = z 3/2 is multi-valued in any neighborhood of the
origin, and hence is not analytic at z = 0.

e
x

Figure 6: Circular path around origin, radius .

Table 1: Evaluation of the function z 3/2 at various points on a circle.

0
/2

3/2
2

z

i

i


f (z) = z 3/2
3/2
3/2 ei3/4
3/2 ei3/2 = i3/2
3/2 ei9/4
3/2 ei3 = 3/2

Riemann Surfaces

Let us continue to think about the interesting f (z) = z 3/2 example. Note
that if = 4 and r = , then z 3/2 = 3/2 ei6 = 3/2 , so we come back to
the = 0 value after two circuits. Thus z 3/2 is a double-valued function. We
may visualize this behavior via the use of Riemann surfaces, or sheets.
For z 3/2 , we have two sheets (Fig. 7).
The point z = 0 is called a branch point. Since there are only a nite
number of branches (2) for this function, the origin is called an algebraic
branch point.
For another example, the function f (z) = z 1/4 will have four branches,
see Table 2 and Fig. 8.
Now consider the function f (z) = ln z, dened by ef (z) = z:
f (z) = ln z = ln r + i.

(42)

This function has an innite number of branches. In this case, the point
z = 0 is called a logarithmic branch point.
We note a couple of things about branches:
There may be many branch points for a function.

Figure 7: Two Riemann sheets for the double-valued function z 3/2 . The
lower sheet is for 0 < 2, 4 < 6, etc., and the top sheet is for
2 < 4, etc. The branch cuts are indicated by the cuts in the planes.

Table 2: The function f (z) = z 1/4 , evaluated at multiples of 2.

0
2
4
6
8

ei/4
1
ei/2 = i
ei = 1
ei3/2 = i
ei2 = 1

There are many ways to make branch cuts, but they can only terminate
at a branch point, they cannot intersect themselves, and they must have
the same form on all sheets.
For a slightly more complicated example illustrating these ideas, consider
the function:

f (z) = 1 z 2 = (1 z)1/2 (1 + z)1/2 .

(43)
This function has singularities (branch points) at z = 1. There are two
sheets, and various possible ways of choosing the branch cuts, as illustrated
in Fig. 9.
There is a choice in how to take branch cuts one makes cuts that are
convenient to the problem at hand (for example, when we integrate along a
path, we arrange it so that the path does not cross a cut). Branch cuts are
used in eect to make multi-valued functions single-valued if you dont
cross a branch, you stay on the same sheet.

Sheet
4

0,8

Figure 8: Four Riemann sheets for the quadruple-valued function z 1/4 . The
view is edge-on, with the branch cut at the transitions among the sheets.
y

-1


+1


-1


+1


-1


+1

Figure 9: Some possible choices of branch cuts for the function

1 z2 .

Integration of Complex Functions

As with the derivative, we must face the problem of forming an integral that
makes sense in some correspondence with the integral for real functions. For
real functions, the indenite integral may be dened as the inverse of differentiation (i.e., as the limit of a sum, rather than the limit of a dierence).
For a complex function, such an indenite integral may not always exist.
Consider f (z) = z = x iy. Suppose


F (z) =

f (z) dz = U + iV,

(44)

and (if integration is inverse of dierentiation)

dF
= f (z) = x iy.
dz

(45)

If the derivative exists, we must be able to use the Cauchy-Riemann equations, hence,
dF
U
U
=
i
= x iy.
(46)
dz
x
y
Thus,
U
U
= x,
= y,
(47)
x
y
10

and

2U
2U
+
= 2.
x2
y 2

(48)

Let us see what the Cauchy-Riemann equations imply for this quantity:
 U
x x
 U
y y
Therefore:

V 
y
V 
.
=
x
=

(49)
(50)

2U
2U
+
= 0,
x2
y 2

(51)

which may be recognized as Laplaces Equation in two dimensions. Thus,

F (z) cannot be an analytic function; it does not possess a derivative, and
there exists no function with derivative x iy. This suggests that we should
restrict consideration to functions which are analytic in the region of interest.
Referring to Fig. 10, let us consider the denite integral:


f (z)dz.

(52)

Figure 10: Possible paths of integration from to .

There are an innite number of possible paths to integrate along. In general,
we must specify the path, e.g.,


f (z) dz.

(53)

To dene this integral, rst divide path C into n intervals by points

z0 = , z1 , z2 , . . . , zn = , as in Fig. 11. Let j z zj zj1 , and let zj be

11

a point on C between zj1 and zj . Then we dene the line integral along C
as:

n

f (z)dz = lim
f (zj )j z,
(54)
n

j=1

C
where we require the intervals to satisfy:
n

lim max |j z| = 0,

(55)

n j=1

and the limit must exist, of course. Note that this denition is compatible
with the usual denition for real variables.

= zn

z 'j

zj
z j-1

= z0
Figure 11: Dividing a path into intervals to obtain an approximate integral.

We list some immediate consequences of our denition:

1. Considering j z j z, we have the path-reversed integral:


f (z)dz =

f (z)dz.

(56)

f (z)dz.

(57)

2. If k is any complex constant, then



kf (z)dz = k

3. If the integrals of f and g separately exist, then the integral of their

sum exists, and:


[f (z) + g(z)] dz =

12

f (z)dz +

g(z)dz.

(58)

4. If is a point on C (between and ), then



f (z)dz +

f (z)dz =

f (z)dz.

(59)

Toward proving this, note that we can always arrange our subintervals
such that is a dividing point.
5. If M = max |f (z)| (including the endpoints) then:
C





f (z)dz 




= n
lim

lim




f (zj )j z 

j=1

n 




f (zj )j z 

(follows from triangle inequality)

j=1

|dz| = MLC ,

(60)

where LC is the length of the integration path (in the usual Euclidean
sense).

Cauchys Theorem

If a function f (z) is analytic at all points on and inside a contour C, then



f (z)dz = 0.

(61)

Note that by
contour, we mean a simple closed curve. We could also use
the notation to stress this. Our assertion is known as Cauchys Theorem.
Let us prove the theorem: Assume f (z) is analytic as stated. Write
f (z) = u(x, y) + iv(x, y). Then:


f (z) dz =
C

(u + iv)(dx + idy)
C

=
C

(udx vdy) + i

(udy + vdx)

(62)

Let S stand for the region enclosed by contour C. Greens theorem states
that, for functions and :
 

dx + dy =
C

13

dxdy.
x
y

(63)

Figure 12: Contour and surface of integration in Greens theorem.

Therefore,


v u
u v
f (z) dz =
dxdy + i
dxdy
+

x y
x y
C
S
S
= 0, by the Cauchy-Riemann relations.

(64)

Cauchys theorem tells us that the integral of an analytic function is

path-independent in a domain of analyticity:


f (z) dz =

C


f (z) dz

f (z) dz,

(65)

where the latter equality is without ambiguity, due to Cauchys theorem.

y

C'
C
x

Figure 13: Equivalent paths of integration in a region of analyticity.

Note that the way we have stated Cauchys theorem, it holds for functions which have singularities, provided our contours do not encircle the
singularities:

14

C2

(a)

C1

(b)

Figure 14: (a) C1 +C2 f (z) dz = 0, where C2 encircles a singularity, but the
contour C1 + C2 does not. Integrals along the portions joining C1 and C2
cancel out. (b) A branch cut may be chosen for convenience, so that the
contour does not cross it.

Indefinite Integral of an Analytic Function

Let f (z) be analytic in simply connected domain D. Then

F (z) =

 z

f (z  ) dz 

(66)

z0

depends only on z and z0 (and not the path), as long as the path is entirely
in D.
What is F  (z) = dF
(z) (for z D)?
dz


z+z
z
1
F (z + z) F (z)
=
f (z  )dz 
f (z  ) dz 
z
z z0
z0
 z+z
1
=
f (z  )dz 
z z
 z+z
1
= f (z) +
[f (z  ) f (z)] dz  .
z z

(67)

The last integral is path independent in D, so chose for path the straight
line segment joining z and z + z (noting that, for z small enough, such
a path must exist). Thus, by the continuity of f (z), given an  > 0, we can
always nd |z| small enough such that |f (z  ) f (z)| <  for any z  on the
path. Thus,


 z+z



[f (z  ) f (z)] dz  

 z


< |z|.

(68)

Given any  > 0 then, we can nd a z > 0 such that


 F (z





+ z) F (z)

f (z) < .
z

15

(69)

Therefore,

d z
F (z) =
f (z  ) dz  = f (z).
(70)
dz z0
The indenite integral of an analytic function is an analytic function.
It is important, when performing integrations, to be careful about singularities and regions of non-analyticity. For example, consider the integral
1 1
1 z dz. We might try an integration path along a semi-circle in the positive
y plane 1/z is analytic there.


y
y
e

-1

+1

+1 x

-1

(a)

(b)

Figure 15: Two possible semi-circular paths from 1 to +1.

We let z = ei , and hence dz = iei d. Alternatively, we could choose to
integrate along a semicircle in the negative y plane 1/z is analytic there as
well. The two choices yield:
I+ = i
I = i

 0

ei ei d = i

(71)

ei ei d = i.

(72)

 2

The two answers are dierent! The path-dependence is a result of the fact
that we have chosen paths which lie in dierent simply-connected domains
of analyticity. There is a branch cut from the origin, a singular point. Note
that, while 1/z is not multi-valued, its integral (ln z) is.

Cauchy Integral Formula

Suppose f (z) is analytic everywhere in some domain D. Consider the integral:


f (z)
dz,
(73)
C z z0
16

f (z)
where C is contained in D, and z0 is interior to C. Thus, zz
is analytic
0
everywhere on and inside C, except at the point z = z0 . The integral is
unchanged if we deform the contour to the circle C0 with center at z0 :

C0

z0
D
Figure 16: Domain D, contour C and deformed contour C0 about point z0 .


C

f (z)
dz =
z z0


C0

=
C0

f (z)
dz
z z0

f (z0 )
f (z) f (z0 )
dz +
dz.
z z0
z z0
C0

(74)

Consider the second of the two integrals in the above expression:





 C0

f (z) f (z0 ) 
dz 

z z0


C0

|f (z) f (z0 )|
|dz|.
|z z0 |

(75)

Since f (z) is analytic at z0 , it must be continuous there. Hence, given any

 > 0, there exists a > 0 such that |f (z) f (z0 )| <  whenever |z z0 | < .
We pick an , and let = |z z0 |, i.e., we pick a circle of small enough radius
such that |f (z) f (z0 )| <  on the circle. Remember that the value of the

17

integral does not depend on the radius of the circle. Thus,





 C0


f (z) f (z0 ) 
dz 
|dz|

z z0
C
2.

(76)

The integral is smaller than any positive number, i.e., is equal to zero. Therefore,

C0


f (z)
dz
dz = f (z0 )
z z0
C0 z z0
 2
irei d
= f (z0 )
rei
0
= 2if (z0 ).

(letting z z0 = rei )
(77)

We have derived Cauchys Integral Formula: For any function f (z) which
is analytic on and inside the contour C,
f (z0 ) =

1  f (z)
dz.
2i C z z0

(78)

Note that the Cauchy integral formula tells us that if we know the value of a
function everywhere along a closed contour, then we know its value at every
point inside the contour, provided the function is analytic on and inside the
contour.

8.1

Cauchy Integral Formula and Derivatives of an Analytic Function

Start with Cauchys integral formula (assuming f (z) appropriately analytic),

and take derivatives:


f (z)
df
dz
d2 f
dz 2
dn f
dz n

=
=
=

1
2i C

1
2i C

2
2i C
n!
2i


C

f (z  ) 
dz
z z
f (z  )
dz 

2
(z z)
f (z  )
dz 
(z  z)3
f (z  )
dz  .
(z  z)n+1

(79)
(80)
(81)

(82)

Is this procedure justied? If so, then we have evidently shown that the
derivative of an analytic function is analytic, at least at all points inside C.

18

If f (z) is analytic, we know its derivative exists:

f (z + h) f (z)
h0
h


1
f (z  ) dz 
f (z  ) dz 
= lim

h0 2ih
C z z h
C z z


f (z  ) dz 
1
lim
=
2i h0 C (z  z h)(z  z)

f  (z) = lim

(83)

Adding and subtracting f (z  )/(z  z)2 to the integrand, we obtain:

1
f (z) =
2i



C

f (z  ) dz 
h
+ lim

2
h0 2i
(z z)


C

f (z  ) dz 
.
(z  z h)(z  z)2

(84)

By assumption, f (z  ) is continuous on C, hence it is bounded. Likewise, (z 

z)2 is bounded on C. Furthermore, take h < minC 12 |z  z|, guaranteeing
that |z  z h| > 0. Therefore,



 
 (z

f (z  )
K < ,
z)2 (z  z h)

(85)

i.e., the integrand is bounded for z  on C by some nite number K. Then,




 lim
h0

h
2i


C


f (z  ) dz 
K

lim |h|LC = 0,



2

(z z h)(z z)
2 h0

(86)

where LC is the length of contour C. Hence,

1 
f (z  )
dz  ,
f (z) =
2i C (z  z)2


(87)

as desired.
Then we may similarly consider:


f  (z + h) f  (z)
1
1
1
= lim
lim
f (z  ) dz 


2
h0
h0
h
2ih C
(z z h)
(z z)2

1
2(z  z h/2)
= lim
f (z  ) dz  
h0 2i C
(z z)2 (z  z h)2

f (z  )
2
=
dz  + lim hAh ,
(88)
h0
2i C (z  z)3
where Ah is a bounded function of z when h < 12 |z  z|. Hence, f  exists,
and
2 
f (z  )

dz  .
(89)
f =
2i C (z  z)3
19

The same argument may be continued indenitely, since the integral representation has f (z), which we know is continuous, hence bounded. Thus, we
have established the result for the nth derivative:
f

(n)

n!
=
2i


C

f (z  )
dz  ,

n+1
(z z)

(90)

as hoped.

8.2

Mean Value Theorem from the Cauchy Integral

Formula

If f (z) is analytic on and within contour C, we know that

1
f (z) =
2i


C

f (z  ) dz 
.
z z

(91)

Consider contour C that is a circle of radius r with center at z0 :



f (z  ) dz 
1
2i C z  z0
 2
 2
f (z  )irei d
1
1
=
=
f (z  )rd
2i 0
rei
2r 0

1
f (z  ) ds,
=
2r C

f (z0 ) =

(92)

where ds is an element of circular arc. Thus, f (z0 ) is given by the average value of f (z) on a circle centered at z0 (entirely within the domain of
analyticity).

Taylor Series

Let f (z) be analytic in domain D with z0 D, and circle C D centered

at z0 [hence, f (z) is analytic within and on C]. Let z = z0 + h be interior to
C. Then, use Cauchys integral:
1  f (z  ) dz 
2i C z  z0 h


1
1
h
=
f (z  ) dz  
+ 
+
2i C
z z0 (z z0 )2

hn
hn+1
+ 
+
,
(z z0 )n+1 (z  z0 )n+1 (z  z0 h)

f (z) = f (z0 + h) =

20

(93)

R
z0 z
C
D
Figure 17: Illustration for Taylor series discussion.

where we have used

1
1
h
=
.
+
(z  z0 h)
(z  z0 (z  z0 )(z  z0 h)

(94)

We also know that:

f

(n)

n!
=
2i


C

f (z  )
dz  ,

n+1
(z z)

(95)

Comparing, we have:
h2 2
hn (n)
f (z) = f (z0 ) + hf (z0 ) + f (z0 ) + + f (z0 )
2
n!

f (z  ) dz 
hn+1
.
+
2i C (z  z0 )n+1 (z  z0 h)
(1)

Thus, we have
f (z) =

(z z0 )k

k=0

21

f ( k)(z0 )
+ Rn ,
k!

(96)

(97)

where
Rn =

(z z0 )n+1
2i


C

f (z  ) dz 
.
(z  z0 )n+1 (z  z)

(98)

We see that term by term this is the same form as the Taylor series expansion
for a real function of a real variable.
Let us investigate the remainder term, Rn . In particular, how big is it?
We rst notice that f (z  ) and 1/|z  z| are continuous, hence bounded, on
C:


 f (z  ) 


 
 M, z  C, z inside C.
(99)
z z 
Let R be the radius of C. Then:



f (z  ) dz 
1 

n+1
(z z0 )

|Rn | =

n+1


2
(z z) 
C (z z0 )
M
1

|z z0 |n+1 n+1 2R
2
R


 z z0 n+1

MR 
.
R 



(100)




0
 < 1, we can approximate f (z) to any desired accuracy with our
Since  zz
R
nite Taylor series expansion.

10

Bolzano-Weierstrass Theorem

We wish to consider innite series next, which means we must concern ourselves with issues of convergence. Let us begin with sequences. Given any
sequence of complex numbers, z1 , z2 , . . . {zn }, we say that the sequence
{zn } tends to the limit L as n :
lim zn = L,

(101)

if, for every  > 0, there exists N such that |zN +k L| <  for all positive
integers k. If {zn } is such that for any real number G, we can nd N so that
|zN +k | > G for all positive intergers k, then we say that |zn | tends to as
n .
Finally, if a sequence does not tend to a unique limit, and does not tend
to plus or minus innity, then the sequence is said to oscillate.
Definition: A limit point of a set S is a point such that there are an
unlimited number of elements of S which are arbitrarily close to the
limit point.

22

For example, 1 is a limit point for the sequence 1 + 1/n (even though 1 is
not an element of the sequence). For another example, 1 is a limit point for
the sequence 1, 2, 1, 2, 1, 2, 1, 2, . . .
Theorem: (Bolzano-Weierstrass) If {xn } is an innite sequence of real numbers, and there exists a, b such that a xn b for all n (where a and
b are independent of n), then {xn } has at least one limit point.
Proof: Let G be a real number such that G > |a|, G > |b|. Then, G > |xn |
for all n. Consider the interval I0 = (G, G). Cut it in half (say, to
(G, 0) and [0, G)): At least one subinterval must contain an innite
number of members of the sequence {xn }. Call the rightmost such
interval I1 . Now cut I1 in half. Again, at least one subinterval must
contain an innite number of members of the sequence {xn }. Call the
rightmost such interval I2 . We may continue this interval subdivision
indenitely, making our interval as small as we please. In the nested set
of intervals I1 , I2 , I3 , . . . there exists a point L which belongs to all the
intervals of the nest. Choose k suciently large such that the length
of Ik is less than any given  > 0. Then if {xn }k is the innite set of
members of {xn } which lies in Ik , we have that |xn L| <  for all
members of {xn }k . Hence L is a limit point of the sequence.

11

Cauchys Condition for the Existence of a

Limit, or, Cauchys Principle of Convergence

Theorem: A sequence of complex numbers z1 , z2 , . . . has a limiting value if

and only if, given any  > 0 there is an N such that |zN +k zN | < 
for all positive integers k.
This convergence condition is referred to as Cauchys condition. Note the
distinction between this theorem and the denition of the limiting value. To
apply this test, one does not need to know, a priori, what the limit is.
Proof: Necessity: We suppose a limit, L, exists. Then, given any  > 0,
there exists an N such that |zN L| < /2, and |zN +k L| < /2 for
all positive integers k. By the triangle inequality:
|zN +k zN | |zN +k L| + |zN L| < .

23

(102)

Suciency: We suppose that given an  > 0 there exists an N such

that |zN +k zN | <  for all positive integers k. But the hypoteneuse
of a triangle is longer than either other leg, and hence:
 > |zN +k zN | |xN +k xN |
|yN +k yN |.

(103)
(104)

Thus, we may consider a real sequence {xn } which satises the Cauchy
condition. Consider  = 1, and pick an N = M such that:
|xM +k xM | < 1 k = 1, 2, 3, . . .

(105)

Let a1 , b1 be the least and greatest values, respectively, of the nite

sequence x1 , x2 , . . . , xM . Let a = a1 1 and b = b1 +1. Then a < xn < b
for all n. By the Bolzano-Weierstrass theorem, {xn } has at least one
limit point, G.
Now we must demonstrate that there is only one limit point: Suppose
there are at least two, G and H. Then, given  > 0, there exists an
n such that |xn+p xn | < , by hypothesis, and there exists positive
integers q and r such that |G xn+q | <  and |H xn+r | < , since G
and H are limit points. Thus,
|G H| = |G xn+q + xn+q xn + xn xn+r + xn+r H|
|G xn+q | + |xn+q xn | + |xn+r xn | + |H xn+r |
< 4.
(106)
Hence G=H, and there is only one limit point. Thus, given > 0,
there are at most a nite number of terms of the sequence outside the
interval (G , G + ), so G is the limit of {xn }.
Similarly, the imaginary part sequence has a limit, hence {zn } has
a limit [noting that if limn zn = L, and limn zn = L , then
limn (zn + zn ) = L + L ].

12

Infinite Series

Given a sequence {un }, we can construct a sequence:

S0 = u 0
S1 = u 0 + u 1
..
.

(107)
(108)

Sn =

(109)

k=0

24

uk .

These are the partial sums of the infinite series:

S=

uk .

(110)

k=0

The innite series is said to converge if, given  > 0 there exists S and
n0 such that:
|S Sn | < , n > n0 .
(111)


If the series
n=0 un converges: It is said to be absolutely convergent if

n=0 |un | converges; otherwise it is conditionally convergent. Note that an
absolutely convergent series may be rearranged at will, with identical results,
but this doesnt hold for a conditionally convergent series.
We give some tests for convergence, leaving the proofs to the reader:
1. Cauchy Integral test for convergence: If f (x) is a positive, real,

decreasing function of x for real x 1, then the series S =
n=1 f (n)
converges or diverges, depending on whether the integral
lim
n

 n

f (x)dx

(112)

converges or diverges.
2. Comparison test for absolute convergence: S =
convergent if
|un | < c|vn |, n > N,
c is independent of n, and

n=0 vn

n=0

un is absolutely
(113)

is known to be absolutely convergent.



3. dAlemberts ratio test for absolute convergence:

n=0 un converges
absolutely if


 un+1 

 < 1,
limn 
(114)
un 
where lim is the limit superior, or least upper bound of all convergent
subsequences of {un }. The sum diverges if


 un+1 


limn 


un

> 1.

4. Raabes test for absolute convergence: If



 un+1 

 = 1,
lim

n  u
n
and




 un+1 
1

limn n 


un

then n=0 un converges absolutely.

25

(115)

(116)
< 1,

(117)

5. Cauchys test for absolute convergence: If

limn |un |1/n < 1,
then

13

n=0

(118)

un converges absolutely.

Series of Functions

If the terms of an innite series are functions of complex variable z, then the
series may converge or not, depending on the value of z. We are interested
in the region of convergence of such a series. We are also interested in
continuity, integrability, and dierentiability of such a series (especially of
analytic functions, including power series).

N
If S(z) =
n=0 un (z) and SN (z) =
n=0 un (z), then S(z) is said to be
uniformly convergent over the set of points {z|z R} = R if, given any
 > 0, there exists an N such that:
|S(z) SN +k (z)| < , k = 0, 1, 2, . . . , and z R.

(119)

Note that the condition of uniform convergence is in a sense stronger than

simple convergence S(z) may converge for all z R, without being uniformly convergent. As an example, consider f (z) = 1/(1 z), for R = {z :
|z| < 1}.
A necessary and sucient condition for uniform convergence is Cauchys

principle for uniform convergence: Given S(z) =
n=0 un (z) which converges
for all z R, where R is a closed region, and any  > 0, then S(z) converges
uniformly in R if there exists an N such that
|SN (z) SN +k (z)| < , k = 0, 1, 2, . . . , and z R.

(120)

It is left to the reader to prove this, using techniques similar to methods

already encountered.
Another, sucient, test for uniform convergence is the Weierstrass M
test: If |un (z)| Mn , where Mn is a positive real number, independent

of z R, and if
n=0 Mn converges, then S(z) is uniformly convergent on
z R.
Let us consider the following example: Suppose we have the real series
x2
x2
+
+
1 + x2 (1 + x2 )2

x2
=
.
2 n
n=0 (1 + x )

S(x) = x2 +

26

(121)
(122)

We see that S(x) converges absolutely for all real x, since:

x2
2 n
n=0 (1 + x )

0
=
1 + x2 (1+x12 )N

SN (x) =

(123)
if x = 0,
if x =
0.

(124)

Thus, S(x) converges absolutely for all possible real x values.

But does this series converge uniformly? We suspect trouble because of
the peculiar behavior at x = 0:

0
at x = 0,
S(x) =
(125)
2
1 + x for x = 0.
That is, S(x) is discontinuous at x = 0. For uniform convergence, we must
have the case that, given any  > 0 there exists an N, independent of x, such
that
|SN (x) SN +k (x)| < , k = 0, 1, 2, . . . , and x.
(126)
Assume x > 0. Then:
|SN (x) SN +k (x)| =

1
.
(1 + x2 )N +k

(127)

Lets choose  = 1/2. Notice that for any xed N, and any chosen k, we can
always pick x > 0 small enough so that
1
1
(128)
> = .
2
N
+k
(1 + x )
2
Hence the convergence is not uniform near x = 0.
The following theorem addresses the question of continuity:


Theorem: If S(z) =
n=0 un (z) is a uniformly convergent series of continuous functions un (z) for all z R, where R is a closed region, then S(z)
is a continuous function of z, for all z R.


Proof: Write S(z) = Sn (z)+Rn (z), where Rn (z) =

k=1 un+k (z). Sn (z) is a
nite sum of continuous functions and hence is continuous throughout
R. By uniform convergence, given any  > 0, we can nd N such that
RN (z) < 13 , for all z R. Furthermore, since SN (z) is continuous for
all z R, there exists a > 0 such that |SN (z + ) SN (z)| < 13 
whenever || < and z + R. Therefore:
|S(z + ) S(z)| = |SN (z + ) SN (z) + RN (z + ) RN (z)|
|SN (z + ) SN (z)| + |RN (z + )| + |RN (z)|
< .
(129)
Hence, S(z) is continuous for all z R.

27

We may also be concerned with the question of multiplication of series. If

two series are absolutely convergent, then the series formed of product terms
is absolutely convergent independent of order, and the product series is equal
to the product of the individual series.
Next, let us consider the integration of a series:


Theorem: Let S(z) =

n=0 be a uniformly convergent series of continuous
functions in a domain D. Then, if C D, where C is a nite path in
D, we have:


S(z)dz =
C

n=0 C

un (z)dz.

(130)

The order of integration and summation may be interchanged for a

series of continuous functions in its domain of uniform convergence.
Proof: Write S(z) =
Then


n

k=0

uk (z) + Rn (z), where Rn (z) =

S(z)dz =
C

n 

k=1 un+k (z).

uk (z)dz +

k=0 C

Rn (z)dz.

(131)

Since S(z) is uniformly convergent, for any given  > 0, there exists
an N such that Rn (z) <  for all n N and all z D. Now, if

LC = C |dz| < is the path length, then





Hence,





 C



Rn (z)dz 

S(z)dz

< LC , n N.




uk (z)dz 

C

n 

k=0

< LC , n N,

(132)

(133)

which can be made arbitrarily small.

Next, we investigate the dierentiation of a series.


Theorem: If S(z) =
n=0 un (z) is a series of functions which are analytic
on and inside a contour C, and if S(z) converges uniformly on C, then
S(z) is analytic everywhere inside C, with derivative:

dS
dun
(z) =
(z).
dz
n=0 dz

(134)

That is, The order of dierentiation and summation may be reversed.

28

Proof: Let z0 be a point inside C.

1 
dz
1  S(z)dz
=
un (z)
2i C z z0
2i C n=0
z z0




n
uk (z)
Rn (z)
1 
=
dz +
dz
2i C k=0 z z0
C z z0

1  Rn (z)
=
uk (z0 ) +
dz.
2i C z z0
k=0
n

(135)

The series converges uniformly on C, so given any  > 0 there exists

an N such that |Rk (z)| <  for all k n and for all z C. Thus,




 C




Rn (z) 

dz  <  


z z0

dz 
 < 2.
z z0 

(136)

Therefore, the series converges, and we have:

1
2i


C

S(z)
dz =
un (z0 ) S(z0 ),
z z0
n=0

(137)

where we take the latter as the denition of S(z0 ) interior to C.



Thus, S(z) =
n=0 un (z) is dened on and inside C. To prove analyticity inside C, we show that the derivative exists:
S(z0 + h) S(z0 )
h0
h

 
S(z)
1 1
S(z)
= lim

dz
h0 2i h C z z0 h
z z0

1
S(z)
dz
= lim
h0 2i C (z z0 h)(z z0 )
 n 


1
uk (z)
Rn (z)
=
dz +
dz .
2i k=0 C (z z0 )2
C (z z0 )2

S  (z0 ) = lim

(138)

The rst term is the form of the derivative of an analytic function we

saw earlier. The second term can be made arbitrarily small by taking
n large enough, by the uniform convergence of S on C. Hence,
S  (z) =

dun
(z).
n=0 dz

(139)

Let us now turn to the special case of power series, of which the Taylor
series is an important example.

29

n
Theorem: If S(z) =
n=0 an z converges for z = z1 , then it is absolutely
convergent for all |z| < |z1 |.

Proof: Since the series converges for z = z1 , an z1n must be bounded: |an z1n | <
M for all n. Pick any z such that |z| < |z1 |. Let r = |z|/|z1 | < 1. Then


|an z | =
n

and


|an z1n | 

|an z n | <

n=0

n=0

z n
n
 < Mr ,
z1

Mr n = M

(140)

rn .

(141)

n=0

Since r < 1, this is convergent, hence S(z) is absolutely convergent for

all |z| < |z1 |.


n
A similar argument can be used to show that, if S(z) =
n=0 an z diverges for z = z1 , then it diverges for all |z| > |z1 |. Thus, the region of
convergence of a power series is a circle: Inside the circle there is absolute
convergence, and outside there is divergence. On the circle, we cannot say
in general. For example,

S(z) =

zn
n=1

diverges for |z| > 1

absolutely converges for |z| < 1
converges for z = 1
diverges for z = +1.

(142)

We state and leave it for the reader to prove the following:

Theorem: A power series is uniformly convergent in any closed region inside
the circle of convergence.
We have the following uniqueness theorem for power series:


n
Theorem: If S(z) =
n=0 an (z z0 ) converges for all points inside the
circle |z z0 | = r0 , then the series is the Taylor series for S(z) (about
z0 ).

Proof: The proof consists in dierentiating k times, and showing that an =

S (n) (z0 )/n!.

30

C1

z0

C2

Figure 18: Illustration for Laurent series discussion.

14

Laurent Series

We now introduce a generalizaton of the Taylor series, the Laurent series.

Consider a function f (z) which is analytic in a region containing two concentric circles (but not necessarily in the interior of the smaller circle).
Contour C2 C1 (Fig. 18 represents a closed path in a simply-connected
domain, so we can use the Cauchy Integral Formula (for z in the annulus):
f (z) =

1
2i


C2

1
f (z  ) 
dz

z z
2i


C1

f (z  ) 
dz .
z z

(143)

Now,
z

1
1
1
= 
.

z
z z0 1 (z z0 )/(z  z0 )

(144)

For z  on C2 , z in the annulus, and with z0 the center of the circles, |(z
z0 )/(z  z0 )| < 1. We may thus write, for z  on C2 :

(z z0 )n
1
=
.
z  z n=0 (z  z0 )n+1

(145)

Similarly, for z  on C1 :

(z  z0 )n
1
=
.
z  z n=0 (z z0 )n+1

31

(146)

Putting this back into 143:

f (z) =

n=0

(z z0 )n
2i


C2

1
f (z  )
1
dz 

n+1
(z z0 )
2i (z z0 )n+1

C1

(147)

Thus, we can write:

f (z) =

an (z z0 ) +
n

n=0

n=1

bn

1
,
(z z0 )n

(148)

where:


f (z  )
1
dz  , n = 0, 1, 2, . . .

n+1
2i C2 (z z0 )

f (z  )
1
=
dz  , n = 1, 2, . . .

n+1
2i C1 (z z0 )

an =

(149)

bn

(150)

Or, we may combine the series:

f (z) =

An (z z0 )n ,

(151)

n=

where,
An =

f (z  )
1 
dz  ,
2i C (z  z0 )n+1

(152)

where C is any contour which makes one counter-clockwise passage around

z0 , and lies in the region bounded by C1 and C2 . This is called the Laurent
series.
If we express f (z) = (z) + (z), where
(z) =
(z) =

n=0

An (z z0 )n ,

(153)

An (z z0 )n ,

(154)

n=1

then (z) is called the principal part of f (z). Note that (z) converges
uniformly in any closed region interior to the outer edge of the annulus.
Hence, f (z) = (z) + (z) converges uniformly in any closed region within
the annulus.
If z = z0 is a singularity of f (z), and there exists a neighborhood of z0
which contains no other singularity, then z0 is called an isolated singularity
of f (z). For example, z = 1 is an isolated singularity of f (z) = 1/(z 1). If

32

(z z0 ) f (z ) dz .
n

all the coecients of the principal part vanish, then an isolated singularity z0
is called a removable singularity. For example, the origin is a removable
singularity of f (z) = sin z/z. The singularity in this case may be removed
by dening
sin z
f (0) lim
= 1.
(155)
z0 z
If the principal part terminates after a nite number of terms, say
Am = 0,
A(m+k) = 0, k = 1, 2, 3, . . . ,

(156)
(157)

then f (z) is said to have a pole of order m at z0 . For example, f (z) =

1/(z z0 )2 has a pole of order 2 at z0 .
If the principal part has an innite number of non-vanishing coeecients,
then z0 is called an essential singularity of f (z). An essential singularity
need not be isolated. For example, z = 0 is an essential singularity of f (z) =
1/ sin(1/z). It is also the limit point of a sequence of poles, and hence is
not an isolated singularity. On the other hand, z = 0 is an isolated essential
singularity of f (z) = e1/z .
If the Laurent series is not known, the order of a pole may be determined
by examining limits. Consider the limits:
lim (z z0 )n f (z), n = 1, 2, 3, . . .

zz0

(158)

The lowest n for which the limit exists is the order of the pole at z0 .

15

Residues

Consider the integral:

In =

(z z0 )n dz,

(159)

where C is a closed contour surrounding z = z0 and n is an integer. Since

(z z0 )n is analytic, except possibly at z0 , we may deform the contour into a
circle centered at z0 without aecting In . Then we may write z z0 = Rei ,
and hence
In = iR


n+1

 2
0

ei(n+1) d.

0
n = 1
.
2i n = 1

33

(160)
(161)

Now suppose we have a function, f (z), which is analytic in a region except

at the point z0 in the region. Then we can make the Laurent expansion about
z0 :

f (z) =

An (z z0 )n .

(162)

n=

Take a contour C around z0 :

1
2i

1
2i

f (z) dz =
C

C n=

An (z z0 )n dz

= A1 .

(163)
(164)

Thus
the coecient of 1/(z z0 ) in the Laurent series is given by A1 =
1
f
(z)dz. This coecient is called the residue of f (z) at z0 . Notice
2i C
that the residue is zero if f (z) is analytic at z0 , or if the coecient A1 is
zero (even if z0 is a pole or isolated essential singularity).
We now come to the important and useful residue theorem. Consider
contour C in a region where f (z) is analytic except at isolated singularities
(poles or essential singularities).

Cb

Ca

Cc

Figure 19: Contours to illustrate residue theorem. Singularities are at a, b,

and c.

We want to determine C f (z) dz. We can write this in terms of the sum
of the integrals around each singularity:


f (z) dz =
C

f (z) dz +
Ca

Cb

34

f (z) dz +

= 2i(a1 + b1 + c1 + )

= 2i
R,
singularities

(165)


where a1 , b1 , c1 , . . . are the residues at a, b, c, . . ., respectively, and R is

the sum of the residues of f (z) interior to the countour C.
The computation of the residues is thus often an important part of evaluating integrals. At a simple pole, the Laurent series is

A1
f (z) =
+
An (z z0 )n ,
z z0 n=0

(166)

A1 = lim [(z z0 )f (z)] .

(167)

and hence
zz0

For a pole of order m,

f (z) =

Am+1
A1
Am
+
+

+
+
An (z z0 )n .
(z z0 )m (z z0 )m1
z z0 n=0

(168)

If we multiply both sides by (z z0 )m , we have:

(zz0 )m f (z) = Am +Am+1 (zz0 )+ +A1 (zz0 )m1 +

An (zz0 )n+m .

n=0

(169)

Now dierentiate m 1 times and evaluate at z = z0 :

dm1 [(z z0 )m f (z)] 

= (m 1)!A1 ,
z=z0
dz m1

(170)

and hence,

dm1 [(z z0 )m f (z)] 

1

.
(171)
z=z0
(m 1)!
dz m1
But sometimes it is easier to just carry out the expansion suciently to nd
A1 directly.
A1 =

16

Cauchy Principal Value Integral

Suppose f (z) has a simple pole at z = z0 = x0 + i0 on the real axis. We may

dene an integral along the real axis through this pole according to:
P

f (z)dz lim
0


x0 

f (x)dx +


x0 +

f (x)dx ,

(172)

where < x0 < . This is known as the Cauchy Principal Value Integral.

35