Professional Documents
Culture Documents
Michael Singer
University College London
Contents
3
CHAPTER I
Plane topology
Exercise I.1.1. Fill in the details of the above remark: if z = x + iy with x > 0,
find an open disc centred at z, which is contained in H. Convince yourself also that
H is not a neighbourhood of any point on the imaginary axis.
This approaching in all directions idea makes open sets the natural domains of defi-
nition of continuous functions (and also differentiable, analytic, holomorphic functions).
However, if S is any subset of C, we define continuity as follows: a function f : S → C
is continuous at z0 ∈ S if given ε > 0, there exists δ > 0 such that
z ∈ S ∩ D(z0 , δ) =⇒ |f (z) − f (z0 )| < ε.
Further, f : S → C is continuous if it is continuous at each of its points.
Definition I.1.10. If X is any subset of C, the interior of X is the set
X ◦ = {z ∈ C : X is a neighbourhood of z}. (I.1.4)
Examples I.1.11. The interior of any open set U is, by definition, U itself. The
interior of D(a, r) is just D(a, r). The interior of Z ⊂ C is the empty set.
Exercise I.1.2. What is the interior of D∗ (a, r)? For which z is the set D(0, 1)
a neighbourhood of z? Same question for D(0, 1) and D∗ (0, 1).
Exercise I.1.3. *For enthusiasts: Show that if X is any subset, then X ◦ is
the union of all open sets U with U ⊂ X.
1. OPEN SETS, CLOSED SETS, NEIGHBOURHOODS 7
Exercise I.1.5. Which of D(a, r), D(a, r) and D∗ (a, r) are closed?
Exercise I.1.6. Is Z ⊂ C closed? How about {1/n : n ∈ N}?
Exercise I.1.7. Is it the case that any given subset of C is either open or closed?
Reasons?
Definition I.1.13. If S is any set, then z0 is a limit point of S if for every r > 0,
the punctured disc D∗ (z0 , r) contains points of S (in other words, D∗ (z0 , r) ∩ S 6= ∅).
Limit points are also often called accumulation points.
Proposition I.1.14. Let (z1 , z2 , . . .) be a convergent sequence of complex numbers
zn → w. Then if the set S = {zn : n ∈ N} is infinite, the set of limit points of S is {w}.
Proof. The assumption that S is infinite means that infinitely many of the zn are
different from w. The definition of convergence implies that given r > 0, there exists N
such that |zn − w| < r for all n > N . Since also infinitely many of the zn are different
from w, this means that
S ∩ D∗ (w, r) 6= ∅.
Hence w is a limit point of S.
Conversely, suppose that v is a limit point. It means that given any r > 0,
S ∩ D∗ (v, r) 6= ∅.
Thus at least a subsequence of (zn ) converges to v. Since zn converges to w, every
subsequence of zn also converges to w, so v = w.
We now show that the definition given above is equivalent to F containing all its
limit points.
Proposition I.1.15. A set F is closed if and only if it contains all its limit points.
8 I. PLANE TOPOLOGY
Proof. We shall use the definition that F is closed if and only if C \ F is open.
Suppose that F is closed and that z0 6∈ F . We shall show that z0 cannot be a limit
point; turning this around F contains all its limit points.
Another way of writing z0 6∈ F is z0 ∈ C \ F . But by definition, this is open, so for
some r > 0, D(z0 , r) ⊂ C \ F . Hence D(z0 , r) ∩ F = ∅ and so too D∗ (z0 , r) ∩ F = ∅.
Hence z0 cannot be a limit point of F . This proves ‘F closed’ ⇒ ‘F contains all its limit
points’.
For the converse, suppose that F contains all its limit points and let w ∈ C \ F .
Then w is not a limit point of F and this means that for some r > 0, D∗ (w, r) ∩ F = ∅,
or in other words D∗ (w, r) ⊂ C \ F . Since by hypothesis w ∈ C \ F , it follows that
D(w, r) ⊂ C \ F . Thus C \ F is a neighbourhood of w and since w was an arbitrary
point of C \ F , we conclude that C \ F is open.
Definition I.1.16. If X ⊂ C is any set, then the closure of X is the union of X and
all limit points of X. This is denoted by X. The boundary of X, denoted ∂X, is the set
X \ X ◦.
Exercise I.1.8.
(a) What are the closures and boundaries of D(0, 1), D(0, 1) and D∗ (0, 1)?
(b) What are the closure and boundary of the unit circle {z : |z| = 1}?
(c) If A = {a1 , a2 , . . .} is a countable sequence of points, is A sometimes, always,
or never, open? Is it sometimes, always, or never closed? Give reasons.
(d) *For enthusiasts Show that if X is any subset of C then X is the inter-
section of the family of all closed sets F which contain X.
1.3. Behaviour under union and intersection. The following is useful back-
ground knowledge and very significant for the development of continuity in the context
of abstract topological spaces.
Proposition I.1.17. Let (Uα )α∈A be some family of open sets. (The ‘index set’ A
need not be finite or even countable.) Then
[
U= Uα
α∈A
is again open. If A is finite, then
\
V = Uα
α∈A
is open.
Similarly if (Fα )α∈A is a family of closed sets, then the intersection
\
F = Fα
α∈A
2. BOUNDED SETS AND COMPACTNESS 9
which is a union of open sets, hence open by the first part. So F is closed. Similarly for
the finite union G of closed sets.
Exercise I.1.9.
(a) Look at the above proof and make sure you understand where we used that
A is finite in the proof that V is open.
(b) Find an example of a countable family of open sets whose intersection is
not open.
(c) Find a countable family of closed sets whose union is not closed.
Proposition I.2.4. A set S is bounded if and only if there exists M > 0 such that
for any z ∈ S,
− M 6 Re(z) 6 M and − M 6 Im(z) 6 M. (I.2.1)
In other words, S is bounded if and only if it can be enclosed in some large square.
Proof. Suppose that S is bounded. By definition, it is contained in some disc
D(0, R). So it is enough to show that there is some M > 0 such that (I.2.1) holds for
all z ∈ D(0, R).
See Figure I.2.5. Geometrically, the set of points satisfying (I.2.1) is a closed square
with horizontal and vertical sides, centre at 0 and of side-length 2M . It is clear that any
given disc will be contained in such a square if we are allowed to take M as large as we
like.
How about the converse? We are given that S is contained in a large square of
the above type, of side-length 2M . To show it is bounded, we need to show that it is
contained in some large disc. Again, geometrically, we just have to choose the disc so
large that it contains√the square. This can be done by taking the centre of the disc a = 0
an the radius to be 2M .
Remark I.2.5. The main idea of the proof is in Figure I.2.5
Note that in this proof, we used the definition of boundedness in terms of containment
within a disc and checked that the new condition (containment within a square) was
equivalent to this definition.
Definition I.2.6. A subset K of C is called compact if and only if it is closed and
bounded.
Remark I.2.7. This is a very important and powerful definition. We hope to per-
suade you of this with two propositions. You have seen precursors of both in 1101.
Theorem I.2.8. Let K be compact and let zn ∈ K be any sequence. Then there is a
convergent subsequence znj with znj → z0 ∈ K.
2. BOUNDED SETS AND COMPACTNESS 11
Exercise I.2.10. The set D(0, 1) is bounded but not closed, so not compact.
Find a sequence of points in D(0, 1) with no limit point in D(0, 1). The set C is
closed but unbounded. Find a sequence of points in C with no limit point in C.
*Challenge Questions
(a) A more abstract and general definition of compactness is the following. K
is compact if any cover of K by open sets has a finite subcover. In other
words, whenever we have a family {Uα }α∈A of open subsets such that
[
K⊂ Uα ,
α∈A
one can find α1 , . . . , αN such that
K ⊂ Uα1 ∪ · · · ∪ UαN .
Show that it is equivalent to K being closed and bounded.
12 I. PLANE TOPOLOGY
Proof. The line in 3-space which contains (0, 0, 1) and (p1 , p2 , p3 ) is given by
(x1 , x2 , x3 ) = (1 − t)(0, 0, 1) + t(p1 , p2 , p3 ), (t ∈ R). (I.3.3)
3. THE RIEMANN SPHERE 13
Note that as (p1 , p2 , p3 ) → (0, 0, 1), |σ(p1 , p2 , p3 )| → ∞. This makes the following
reasonable:
Definition I.3.3. We extend σ to define a bijection between S and C∞ by setting
σ(N ) = ∞.
Exercise I.3.11.
(a) Show that the equation a + ∞ = ∞ is consistent with the definition σ(N ) =
∞ by calculating σ −1 (a + z) and letting |z| → ∞.
(b) What transformation of S corresponds to the map z 7→ −1/z?
Exercise I.3.12. The bijection between C∪∞ and a sphere is sometimes defined
in terms of stereographic projection of the unit sphere S 0 which is tangent to the
(x1 , x2 ) plane at the origin, i.e. with equation
x21 + x22 + (x3 − 1)2 = 1.
If the map is geometrically defined as before by taking the straight line joining the
north pole (0, 0, 2) to a variable point P on S 0 and defining σ 0 (P ) as the intersection
of this line with the plane x3 = 0, find the formula for this mapping analogous to
that of Theorem I.3.2.
An important heuristic here is that straight lines in C are ‘ideal’ circles. From the
point of view of S , most circles map by σ to circles in C; it’s just we get a straight line
in C as the image if the original circle happens to pass through the north pole of S .
This idea is further illustrated by the following example.
Proof. For the moment denote σ(P ) = P 0 and σ(Q) = Q0 . From the definition
of σ it is a pleasant exercise to show that the triangles N P Q and N Q0 P 0 are similar.
Moreover,
|N P | p p
0
= 1 − p 3 1 − q3 . (I.3.15)
|N Q |
It follows that
|P Q| p p
= 1 − p 3 1 − q3 . (I.3.16)
|z − w|
Since
1 − p3 = 2/(1 + |z|2 ) (I.3.17)
and similarly for 1 − q3 (see (I.3.2)), the result follows.
Exercise I.3.14. Carry out the ‘pleasant exercise’ mentioned in the proof.
Proof. Suppose first that none of the zj is equal to ∞. Then if T (z2 ) = 0 and
T (z3 ) = ∞, then T must have the form
z − z2
T (z) = λ . (I.3.26)
z − z3
Then
z1 − z2
T (z1 ) = λ , (I.3.27)
z1 − z3
Setting this to 1, we obtain
z 1 − z3 z − z2
T (z) = . (I.3.28)
z1 − z2 z − z3
It is clear from the construction that T is unique.
If any one of the zj is equal to ∞, we obtain the corresponding T by taking suitable
limits of (I.3.28). For example, if z1 = ∞, we get
z − z2
T (z) = ,
z − z3
if z2 = ∞ we get
z1 − z3
T (z) =
z − z3
and if z3 = ∞, we get
z − z2
T (z) = .
z1 − z2
Corollary I.3.11. If (z1 , z2 , z3 ) and (w1 , w2 , w3 ) are two ordered triples of points
in C∞ , with no two of the zj equal to each other and no two of the wj equal to each
other, then there is a unique Möbius transformation T such that
T (zj ) = wj for j = 1, 2, 3. (I.3.29)
Proof. From the previous Theorem, let S1 map (z1 , z2 , z3 ) to (1, 0, ∞) and let S2
map (w1 , w2 , w3 ) to (1, 0, ∞). Then T = S2−1 ◦ S1 does the job.
Exercise I.3.16.
(a) What is the most general Möbius transformation which maps 0 to ∞?
(b) What is the most general Möbius transformation which maps the upper
half-plane H = {z ∈ C : Im(z) > 0} to itself?
Exercise I.3.17. *For enthusiasts: Show that any Möbius transformation
viewed as a mapping T : S → S (cf. Remark I.3.9) is continuous.
Theorem I.3.12. Let C be any circle on S and let T be any Möbius transformation.
Then T (C) is again a circle on S .
18 I. PLANE TOPOLOGY
T (z) = T2 ◦ S ◦ T1 (z).
Since each of the three transformations maps circles/straight lines to circles/straight
lines, the result is true also of their composite T .
Exercise I.3.18. Make sure you understand why, geometrically, the image of
the circle C under S(z) = 1/z is a straight line if |α|2 = r2 .
Exercise I.3.19. *For enthusiasts:
Show that if T has the form
az + b
T (z) = , |a|2 + |b|2 = 1, (I.3.31)
−bz + a
then T acts as a rotation of S .
There is another family of circles in the w-plane whose images are easy to understand:
the concentric circles |w| = r. This locus corresponds in the z-plane to a circle, and so
we have
Proposition I.4.2. For any r > 0, the set
z−α r
Br = z ∈ C : = (I.4.6)
z−β |λ|
is a circle in the z-plane (unless r = |λ|, in which case it is a straight line).
Click on this link for a geogebra picture of a family of these orthogonal circles
https://www.geogebra.org/m/z8H4mPWh
The circles Br are called circles of Apollonius. The A- and B- circles form a ‘net’ in
the z-plane. Every point z (6= α, β) lies on the intersection of precisely one A- and one
B-circle. Moreover, every A-circle and every B-circle meet orthogonally.
The first of these claims is clear because the corresponding statement is clear in the
w-plane, namely that every point w 6= 0 lies on a unique ray and a unique circle with
centre the origin. (Polar coordinates!) The second claim about orthogonality will follow
once we have proved that Möbius transformations are conformal or angle-preserving, for
in the w-plane rays through 0 cut the circles |w| = r orthogonally. For now we take it
on trust or take the time to prove it directly.
Exercise I.4.21. Prove that for any r > 0 and θ, the circles Aθ and Br meet
each other orthogonally.
Key point: If α and β are two points of C then the Möbius transformation
w = (z − α)/(z − β) maps any circle containing α and β to a straight line through
the origin in the w-plane.
C1 C2
−1 z2 R z1 1
0
4.2. Applications.
Example I.4.3. Suppose it is desired to map the ‘lens-shaped region’
√ √
R = {|z − 1| < 2} ∩ {|z + 1| < 2} (I.4.7)
(Figure 2) by a Möbius transformation to a sector of the w-plane. There are several
ways to do this: the following is perhaps the simplest.
If R is to map to a sector, then we need C1 and C2 to map to straight lines through
0. From the previous section, a Möbius of the form w = (z − α)/(z − β) will map a
circle to a straight line if α and β are any two points on the circle. If we want the same
Möbius to map both C1 and C2 to straight lines, then we had better choose α and β to
be on both of C1 and C2 . From the picture, the intersection points of C1 and C2 are i
and −i.
So our Möbius transformation will be
z−i
w = T (z) = . (I.4.8)
z+i
The images of C1 and C2 under T are straight lines through the origin. Which ones? To
find out, we compute the images of two other points, one on C1 and one on C2 : since the
images of C1 and C2 are definitely going to be straight lines through 0, knowing these
images will determine these lines. We
√ shall follow this through.
√
Convenient points are z1 = 1 − 2 on C1 and z2 = 2 − 1 on C2 . We calculate
√
√ 1− 2−i √ −1 + i
T (z1 ) = T (1 − 2) = √ = 2( 2 − 1) √ (I.4.9)
1− 2+i 1 + ( 2 − 1)2
and √
√ 2−1−i √ −1 − i
T (z2 ) = T ( 2 − 1) = √ = 2( 2 − 1) √ (I.4.10)
2−1+i 1 + ( 2 − 1)2
So w1 lies on the ray arg(w) = 3π/4 and w2 lies on the ray arg(w) = 5π/4. Thus the
images of C1 and C2 are the lines y = x and y = −x with the arcs bounding R mapping
to the parts of these lines in the left half-plane: see Figure 3.
The original circles C1 and C2 cut the plane up into 4 pieces: the region R, the
complement of R in each of the two discs, and the exterior region. These four regions
are mapped by T to the four sectors cut out by the two lines y = x and y = −x. The
image of R must be one of these sectors, and by the above discussion, the evidence is
that R will map to the region shaded in Figure 3,
T (R) = {w : 3π/4 < arg(w) < 5π/4}. (I.4.11)
This is confirmed by computing T (0) (or T of any other point of R—but 0 is always
easiest if available!). We have T (0) = (−i)/(i) = −1 which lies in (I.4.11) as expected.
Note that the angle between T (C1 ) and T (C2 ) is π/2. This is consistent with the
angle between C1 and C2 being π/2 at their two intersection points. (This can be seen
easily from the picture.)
We can map R to the standard positive quadrant 0 < arg(w) < π/2 by applying
a clockwise rotation which maps arg(w) = 3π/4 to the real axis. This is achieved by
22 I. PLANE TOPOLOGY
arg(w) = 3π/4
T (C1 )
T (0) = −1
0
T (C2 )
arg(w) = 5π/4
Exercise I.4.22. Work through this example with S(z) = z+i z−i instead of T (z).
Just to convince yourself that the order of the intersection points of the arcs doesn’t
matter too much.
Example I.4.4. Suppose it is required to map the half-space H above the line L =
{y = 2x} in the z-plane to the unit disc in the w-plane.
One possibility is to pick three points on the line y = 2x and three points on the
unit circle, and use Corollary I.3.11 to write down a Möbius transformation S which
maps the line to the circle. S will either map H to the inside or the outside of the unit
circle. If it maps to the inside, our work is done. If note, replace S(z) by 1/S(z), since
w 7→ 1/w interchanges the inside and outside of the unit circle in the w-plane.
A more elegant solution is to realise the line as the set of points equidistant from two
auxiliary points in the z-plane. For example, the line M , 2y + x = 0 cuts L orthogonally
at 0. If we parameterize M in the form
z(t) = (2 − i)t, (t ∈ R) (I.4.12)
4. MÖBIUS TRANSFORMATIONS WITH PRESCRIBED PROPERTIES 23
L : y = 2x
z2 = −2 + i
0
M : 2y + x = 0
z1 = 2 − i
then for any pair of points z(t), z(−t) (t 6= 0), L is the set of points equidistant from
z(t) and z(−t). For example if t = ±1, we obtain the points
z1 = z(1) = 2 − i, z2 = z(−1) = −2 + i (I.4.13)
and
L = {z ∈ C : |z − 2 + i| = |z + 2 − i|}. (I.4.14)
(See Figure 4.)
The two open half-spaces determined by L arise by replacing = with < or > in
(I.4.14). To see
√ which is which, note that z = i is in the half-space above the line and
|i − 2 + i| = 2 2 whereas |i + 2 − i| = 2. Thus
H = {z ∈ C : |z − 2 + i| > |z + 2 − i|}. (I.4.15)
It follows at once from this that if
z+2−i
w= (I.4.16)
z−2+i
then |w| < 1 corresponds precisely to z ∈ H. So (I.4.16) is a Möbius transformation
that does the job.
Remark I.4.5. Note that this Möbius transformation is not unique: we get one for
every choice of points z1 and z2 for which it is true that
H = {z ∈ C : |z − z1 | < |z − z2 |} (I.4.17)
Exercise I.4.23. Find the most general Möbius transformation which maps the
half-space H of Example I.4.4 onto the open unit disc in the w-plane.
24 I. PLANE TOPOLOGY
Exercise I.4.24.
(a) Find a Möbius transformation which maps the right half-plane {z : Re(z) >
0} onto the open unit disc {w : |w| < 1}.
(b) Find a Möbius transformation which maps the set
{z : Im(z) > 0, |z| < 1}
onto an open sector of the w-plane.
(c) Find a Möbius transformation which maps the half-space G above the line
x+y = 1 in the z-plane onto the open unit disc {w : |w| < 1} in the w-plane.
Remark I.4.6. When it comes to these kinds of mapping problems, the transforma-
tion is generally not unique. If you compare your answer with a friend’s, they may look
different yet still both be right.
There are also several different ways of solving any given problem.
A general principle is that if you have a region bounded by two circular arcs, find the
intersection points α and β, say. Then the image in the w-plane by w = (z − α)/(z − β)
will be a sector, with the intersection at α being mapped to w = 0. Figuring out exactly
which sector it is can be done in different ways, see the above example.
When it comes to mapping a half-plane G to the unit disc, you can proceed as in
the worked example. Another way is to map G by a combination of a rotation and a
translation, to the standard upper half-plane {z 0 : Im(z 0 ) > 0}. This can always be
0 −i
accomplished by a transformation of the form z 0 = az + b. Now w = zz 0 +i is a standard
0
Möbius transformation from {Im(z ) > 0} onto {|w| < 1}. This is an application of the
principle of solving a problem by breaking it up into manageable pieces, which is often
a good ploy.
Remark I.4.7. Using a Möbius transformation to transform figures in the z-plane to
different figures in the w-plane is an example of conformal mapping. (In mathematics,
‘conformal’ means ‘angle-preserving’.) We shall return to the topic later in the course.
It has many applications in mathematics and ‘physical applied mathematics’.
CHAPTER II
1. Analytic functions
Let Ω be an open set of C. We shall define A (Ω), the set of analytic functions in
Ω to be the set of functions f which are ‘locally expandable in power series’ about any
point of Ω.
The space A (Ω) is a vector space, and in fact a ring. Moreover, if f 6= 0 in Ω then
1/f ∈ A (Ω) also.
1.1. Power series. The power series (centred at 0) with coefficients (a0 , a1 , . . .) (a
sequence of complex numbers) is by definition
∞
X
f (z) = an z n (II.1.1)
n=0
Recall that any such power series has a radius of convergence r ∈ [0, ∞] with the
property that (II.1.1) is convergent for |z| < r and divergent for |z| > r. In fact,
Proposition II.1.1. For each s let M (s) = sup{|an |sn : n = 0, 1, . . .}, so that
M (s) ∈ [0, ∞]. Then the radius of convergence of (II.1.1) is equal to
r := sup{s : M (s) < +∞}. (II.1.2)
Examples II.1.2. The radius of convergence of each of
∞ ∞ ∞
X
n
X zn X zn
nz , , (II.1.3)
n n2
n=1 n=1 n=1
Exercise II.1.1.
(a) Calculate M (s) for each of the three series in (II.1.3). Note that M (1) is
not always finite.
(b) Discuss the convergence of these three series at points z with |z| = 1.
25
26 II. ANALYTIC FUNCTIONS AND HOLOMORPHIC FUNCTIONS
For the proof of the following important fact, we refer the reader to a previous course.
Theorem II.1.3. Let (II.1.1) have radius of convergence r > 0. Then f (z) is con-
tinuous for all z ∈ D(0, r).
Actually, this was probably only proved for real power series in a real variable x.
The proof, however, is the same, with a little care putting in | · | signs appropriately.
Remark II.1.4. If z0 is any point of C we can also consider power series centred at
z0 and the e
an are complex coefficients,
X
fe(z) = an (z − z0 )n
e (II.1.5)
centred at z0 . The radius of convergence re is defined using M (s) and the quantities
an |sn exactly as before. (II.1.5) will converge in the disc D(z0 , re), where re is the radius
|e
of convergence.
P∞ n
Exercise II.1.2. If n=0 an z has radius of convergence r, write down the disc
of convergence of
∞ n
X z−1
an .
10
n=0
Exercise II.1.3.
1. ANALYTIC FUNCTIONS 27
The next result shows that a convergent power series is analytic in its disc of con-
vergence. This may seem obvious, but is not quite trivial.
Proposition II.1.8. Let f (z) = ∞ n
P
n=0 an z be convergent in D(0, r), r > 0. Then
f ∈ A (D(0, r)).
Proof. From the definition, we need to check that for any given point z0 , f can
be expanded in a convergent power series centred at z0 . This follows from the absolute
convergence of power series in their disc of convergence. Write z = z0 + w. Then by the
binomial theorem,
∞
X
f (z) = an (z0 + w)n (II.1.7)
n=0
∞ Xn
X n n−j j
= an z w . (II.1.8)
j 0
n=0 j=0
Since the sum is absolutely convergent for |z0 + w| < r, we may rearrange at will, to
obtain
1
f (z0 + w) = f (z0 ) + f 0 (z0 )w + f 00 (z0 )w2 + · · · (II.1.9)
2!
where
∞
n(n − 1) · · · (n − j + 1)an z0n−j .
X
(j)
f (z0 ) = (II.1.10)
n=j
Here we just collected all terms multiplying wj in (II.1.9). (The notation in (II.1.10) will
be justified as the j-th derivative of f at z = z0 in due course.) Then (II.1.22) converges
for |z0 + w| < r, in particular for w ∈ D(z0 , r − |z0 |).
Of course, there is a precisely similar statement for a power series centred at some
point a rather than 0.
Examples II.1.9.
(a) Any polynomial
p(z) = a0 + a1 z + · · · + an z n
is analytic in C.
28 II. ANALYTIC FUNCTIONS AND HOLOMORPHIC FUNCTIONS
Statements (b) and (d) will follow from Theorem II.1.10 below.
For (c), recall the formulae
∞
z
X zn
exp(z) = e = (II.1.11)
n!
n=0
and
eiz − e−iz eiz + e−iz
sin z = , cos z = . (II.1.12)
2i 2
Combining (II.1.11) with (II.1.12), we have
∞ ∞
X (−1)n z 2n+1 X (−1)n z 2n
sin z = , cos z = . (II.1.13)
(2n + 1)! (2n)!
n=0 n=0
These power series, like (II.1.11), are convergent in the whole of C, showing that sin z
and cos z are analytic in C by Proposition II.1.8.
The power series expansion of tan z at z = 0 is much more subtle: the coefficients
involve the so-called Bernoulli numbers.
The complex logarithm log z, fractional powers such as z 1/2 and even more generally
complex powers z a where a ∈ C are tricky to deal with. In some sense they want to be
analytic where defined, but they are ‘multi-valued’ and this causes serious problems (or
leads to new and fascinating mathematics, depending upon your point of view).
The next theorem shows that analytic functions are plentiful and that they behave
well under the usual algebraic operations—as long as you don’t divide by zero.
Theorem II.1.10. For any open set Ω, A (Ω) is an infinite-dimensional vector space.
It is also a ring in the sense that f, g ∈ A (Ω) implies f g ∈ A (Ω) (pointwise product).
Furthermore, if f ∈ A (Ω) and f (z) 6= 0 for all z ∈ Ω, then 1/f ∈ A (Ω).
Proof.
1. ANALYTIC FUNCTIONS 29
(a) Suppose that f, g ∈ A (Ω) and let z0 ∈ Ω. We know that f and g are given by
convergent power series
∞
X ∞
X
f (z) = an (z − z0 )n , g(z) = bn (z − z0 )n (II.1.14)
n=0 n=0
in D(z0 , δ), for some δ > 0. (If the expansion for f is valid in D(z0 , δ1 ) and the
expansion for g is valid in D(z0 , δ2 ), then we take δ to be the smaller of δ1 and
δ2 .)
Then it is clear that
X∞
λf (z) + µg(z) = (λan + µbn )(z − z0 )n (II.1.15)
n=0
for any complex numbers λ and µ, and the RHS is convergent in the same disc
D(z0 , δ). Since z0 was an arbitrary point of Ω, this implies that λf +µg ∈ A (Ω).
Thus A (Ω) is a vector space. (Clearly the zero-function 0 lies in A (Ω).
(b) We have seen that every polynomial
p(z) = a0 + · · · + an z n
is analytic in the whole of C and hence in A (Ω). The monomials 1, z, z 2 , . . . are
linearly independent, and so the vector space A (Ω) is infinite-dimensional.
(c) In a previous course, we saw the Cauchy Product formula for power series. With
f and g as in (II.1.14) this says that
∞
X
f (z)g(z) = cn (z − z0 )n (II.1.16)
n=0
where
n
X
cn = aj bn−j (II.1.17)
j=0
and (II.1.15) is convergent in D(z0 , δ). It follows that f g (the pointwise product)
is in A (Ω).
(d) Finally, suppose that f ∈ A (Ω) is non-zero everywhere in Ω. As before suppose
X
f (z) = an (z − z0 )n , (II.1.18)
convergent in D(z0 , δ). We know f (z0 ) = a0 6= 0. The argument is simpler if
we replace f (z) by f (z)/f (z0 ) and z − z0 by w. This reduces us to considering
∞
X
g(w) = 1 + an w n (II.1.19)
n=1
which is supposed convergent in D(0, δ). If we can find a power series expansion
of 1/g(w) valid in some D(0, δ 0 ), then we can also find a power series expansion
of 1/f valid in D(z0 , δ 0 ). P
Suppose that h(w) = ∞ n
n=0 bn w satisfies
h(w)g(w) = 1. (II.1.20)
30 II. ANALYTIC FUNCTIONS AND HOLOMORPHIC FUNCTIONS
Then we know what the bn have to be in terms of the an by the Cauchy product
formula. Indeed we have
n
X
a0 b0 = 1, a0 b1 + a1 b0 = 0, . . . , a0 bn + aj bn−j = 0. (II.1.21)
j=1
Remark II.1.11. The proof of (d) is a little tricky but there are some important
ideas in it that I want to draw your attention to. Note first of all the use of simplifying
assumptions, passing from f (z) to g(w). These were not essential, but they save on
notational complication later in the proof.
The idea of the rest of the proof can be summarized as follows: the Cauchy product
tells us what the coefficients of a power series expansion of 1/g(w) must be, which is a
good start, but then we need to see that these coefficients give a series which actually
converges for some non-zero value of w. For this we use the convergence of g(w), which
gives the estimates (II.1.24) and feed these in to the inductive definition of the coefficients
of the expansion of 1/g.
In manipulations with power series, bounding a series by a geometric series, which
we did in (II.1.26) is a very common technical device (which goes back at least as far as
Cauchy!) Note also that we could not have reached our conclusion with s = r; taking
s < r gave us extra room to manoeuvre.
In a similar vein, note the remark that we could not take r = δ, the radius of
convergence of g. The numbers |an |rn may well be unbounded if r is the radius of
convergence (see Exercise II.1.1). One always needs to come in just a little bit from the
boundary of the disc of convergence to be safe!
At first, it may be surprising that the radius of convergence of h = 1/g may be smaller
than that of g. But note the following example: If g(w) = 1 + w, which converges
everywhere, then the radius of convergence is ∞, but the power series for 1/(1 + w)
(centred at w = 0) converges only in D(0, 1).
Exercise II.1.4. Can you construct a power series centred at 0, with f (0) = 1
and an infinite radius of convergence, but such that the power series for 1/f (z) has
radius of convergence 1/2? 1/10? Any number δ > 0? (Try making small changes
to the example just given.)
Exercise II.1.5. *For enthusiasts:
Prove Hadamard’s characterization of the radius of convergence:
Theorem II.1.12. For the power series ∞ n
P
n=0 an z , we have the formula
1
= lim sup |an |1/n
r n→∞
for the radius of convergence r.
z0 + h
z0 z0 + h z0
z0 + h
z0 + h
0 0
Notation II.2.2 (Big O, little o.). If f and g are functions, g real-valued and non-
negative, we write f = O(g) to mean that |f | is bounded by a multiple of g. More
precisely, if S is a set we write
f (z) = O(g(z)) for z ∈ S (II.2.4)
if and only if there is some constant C, independent of z, such that
|f (z)| 6 Cg(z) for all z ∈ S. (II.2.5)
The notation is often used for limits:
f (z) = O(g(z)) as z → z0 (II.2.6)
2. HOLOMORPHIC FUNCTIONS—COMPLEX DIFFERENTIABILITY 33
if and only if, for any sufficiently small δ > 0, there is a constant C, independent of z,
such that
|f (z)| 6 Cg(z) for all |z − z0 | < δ. (II.2.7)
We write f = o(g) as z → z0 if
f (z)
lim =0 (II.2.8)
z→z0 g(z)
Proof.
(a) The properties listed just before the statement of the theorem imply that H(Ω)
is a ring, for they show that if f, g ∈ H(Ω) then λf + µg and f g are again
in H(Ω). Similarly the derivative of f /g is (f 0 g − g 0 f )/g 2 at any point where
g(z) 6= 0.
(b) Applying the product rule repeatedly (or by direct calculation), f (z) = z n is
in H(Ω) for every positive integer n and f 0 (z) = nz n−1 . Thus we have an
infinite set of linearly independent elements of H(Ω), showing it is an infinite-
dimensional vector space.
(c) If f ∈ A (Ω), then for each point z0 ∈ Ω, there exists δ > 0 such that
X∞
f (z) = an (z − z0 )n for z ∈ D(z0 , δ).
n=0
The theorem about term-by-term differentiation of power series seen in a pre-
vious course implies here that f 0 (z) exists and
∞
X
f 0 (z) = nan (z − z0 )n−1 for z ∈ D(z0 , δ).
n=1
(The proof you’ve seen was probably only for real variables, but it goes through
in the complex case almost unchanged.) Thus every function in A (Ω) is also in
H(Ω).
(d) The usual real-variable proof that ‘differentiable ⇒ continuous’ goes through
here to show that H(Ω) ⊂ C(Ω).
(e) This is also a simple adaptation of the usual real-variable proof.
We shall see, after a lot of further work, that in fact H(Ω) = A (Ω).
This is remarkable, compared with the corresponding situation for functions of a
real variable! If a function is complex-differentiable, then with no further hypotheses,
it can be expanded in a convergent power series about every point!
Remark II.2.4. Note that it is much easier to check that a given function is in
H(Ω): one simply needs to check that the derivative exists at every point of Ω. This is
much better than trying to check that a given function can be locally expanded in power
series. Take, for example, the Γ function or the ζ function,
Z ∞ ∞
−t z−1
X 1
Γ(z) = e t dt, ζ(z) = . (II.2.11)
0 nz
n=1
It is relatively easy to check that these are holomorphic (complex-differentiable) (in
Re(z) > 0 and Re(z) > 1 respectively, since you asked). It would be much more arduous
to check that they are locally expandable as power series. In fact, though we did do
it in Theorem II.1.10, it is much harder to check that product and quotient of two
analytic functions are again analytic than it is to check the corresponding properties of
holomorphic functions.
2. HOLOMORPHIC FUNCTIONS—COMPLEX DIFFERENTIABILITY 35
Exercise II.2.6.
Show directly that z n is holomorphic with derivative nz n−1 .
Show that f (z) = z is not complex-differentiable.
Decide whether or not f (z) = |z|2 is complex-differentiable.
If f ∈ A (Ω), that is f is analytic, then all derivatives f (n) exist and f (n) ∈ A (Ω).
This is because when we differentiate the power series expansion term by term as in
the proof of Theorem II.2.3, the radius of convergence of the expansion of f 0 is the
same as for f . Hence f 0 can be differentiated term by term without changing the
radius of convergence and so on.
So those holomorphic functions that are in A (Ω) all are infinitely differentiable.
We have mentioned before that H(Ω) = A (Ω) (to be proved later). It follows that
if f ∈ H(Ω) then f is infinitely differentiable.
Another result that is proved in the same way as in the real-variable case is the chain
rule:
Exercise II.2.8. **For enthusiasts: Can you prove directly the statement
corresponding to (II.2.10) for analytic functions? In other words, if you have two
convergent power series and you substitute one into the other, can you prove that
the resulting power series is convergent? This may be very challenging!
Setting t = 0, we obtain
Γ01 (0) = f 0 (z0 )γ10 (0), Γ02 (0) = f 0 (z0 )γ20 (0).
Thus the tangent vectors Γ01 (0) and Γ02 (0) to the curves Γ1 and Γ2 at w0 are obtained
from the corresponding tangent vectors to γ1 and γ2 at z0 by multiplication by the
complex number f 0 (z0 ).
If f 0 (z0 ) = 0 the tangent vectors are 0, so the image curves are not regular. But if
0
f (z0 ) 6= 0 we may take args and use arg(zw) = arg(z) + arg(w) (mod 2π, for non-zero
complex numbers z and w) to obtain
arg Γ02 (0) − arg Γ01 (0) = (arg f 0 (z0 ) + arg γ20 (0)) − (arg f 0 (z0 ) + arg γ10 (0)) = φ.
So the angle between the image curves is equal to the angle between the original curves
as long as f 0 (z0 ) 6= 0.
What happens at points where f 0 (z0 ) = 0? We are not going to study this system-
atically, but let’s consider f (z) = z 2 as an example. f 0 (z) 6= 0 if z 6= 0, so it is conformal
everywhere except at z = 0. If z = reiθ then w = z 2 = r2 e2iθ . This means that the ray
{arg(z) = α} is mapped into the ray {arg(w) = 2α}. The angles between rays through
the origin in the z-plane are doubled by the mapping w = z 2 .
More generally, if w = z n , n a positive integer, then angles between rays through
the origin in the z-plane are multiplied by n.
Note that there is a subtlety here which is that in these cases, regular C 1 curves
through 0 in the z-plane are not mapped to regular C 1 curves in the w-plane. For
example, the curve
γ(t) = t, (−δ < t < δ)
in the z-plane is mapped to γ e(t) = t2 in the w-plane. As t increases, γ(t) comes in to
the origin of the z-plane from the left along the real axis, passes through and emerges
along the positive real axis.
On the other hand, γ e(t) comes in to the origin along the positive real axis of the
w-plane as t approaches 0 from below, then doubles-back at t = 0 and goes out again
along the positive real axis. The tangent vector at t = 0 is 0 so γ e is not regular at t = 0.
2.3. Real functions and complex functions. In applications (for example con-
tour integration) we often have a function φ(x), say, on the real line and we want to
find a holomorphic function f (z) defined at least near (i.e. in an open neighbourhood
of) the real line which extends φ into the complex, in other words f (x) = φ(x). This
is not always possible (I don’t want to get into the technical details here) but for the
38 II. ANALYTIC FUNCTIONS AND HOLOMORPHIC FUNCTIONS
kinds of simple functions that you learned about at school this is rather easy, simply by
substituting z for x in the formula.
Examples II.2.9.
(a) Suppose we have a real rational function of x such as
x5 − 7x3 + 1
φ(x) =
x8 + 4x4 + 1
Then simply replacing x by z = x + iy yields the holomorphic function
z 5 − 7z 3 + 1
f (z) = .
z 8 + 4z 4 + 1
Note that this is not holomorphic in the whole of C—it fails to be holomorphic
at the zeros of the denominator, but it is holomorphic in an open strip which
contains the real axis.
(b) Similarly if φ(x) involves the elementary transcendental functions exp, sin, cos
etc., we can simply replace x by z since we already have meanings for exp z,
sin z, cos z etc. For example
2
e−x
φ(x) =
5 + sin x
yields the holomorphic function
2
e−z
f (z) = .
5 + sin z
Note that even though φ(x) is continuous for all x (because | sin x| 6 1 for all
real x, so the denominator in φ is never 0), f (z) is not holomorphic in the whole
of C. There are infinitely many (non-real) values of z with sin z = −5 and f (z)
is singular at all these points.
Exercise II.2.9. Find a holomorphic function f (z) with the property that
1
f (x) = √ .
5 − cos x
What is the largest subset of C on which f (z) is defined?
Exercise II.2.10. Suppose that φ(x) is a smooth function on the real line. For
real (x, y), define
F (x + iy) = φ(x).
Make sure you understand the difference between this and the kind of extensions we
discussed in the examples above. What do you need to assume about φ for F to be
holomorphic? This problem will be easier after the next topic!
3. THE CAUCHY–RIEMANN EQUATIONS 39
0
−1 1
4. Connectedness
We have seen that if f 0 (z) = 0 for all z in some open set Ω, it may or may not be true
that f is constant. (For Ω = C it is true, for Ω the disjoint union of discs in Figure 2,
it is not.) So a natural question is: is there a useful characterization of the (say) open
sets Ω for which it is true that f 0 = 0 implies f is constant?
In the proof of Proposition II.3.5 what we used is that any two points in C can be
connected by a path. In the counterexample in Figure 2, on the other hand, there is no
path which connects z = −1 to z = 1 while remaining entirely in Ω. We therefore define
connectedness in terms of paths.
Definition II.4.1.
4. CONNECTEDNESS 43
IMPORTANT
Definition II.4.9. A subset Ω of C is called a domain if it is open and connected.
Remark II.4.10. Both Ahlfors and Priestley use the term ‘region’ instead of domain.
Domain is also standard terminology in analysis, however.
4.1. Path components. Let S ⊂ C be any set. On S we may define an equivalence
relation: z0 ∼ z1 if and only if z0 can be connected to z1 by a path in S. Let S1 , S2 , S3 , . . .
(in general infinitely many) be the equivalence classes. Any such equivalence class is
called a path-component of S. The path-components are pairwise disjoint (this is true
5. HARMONIC FUNCTIONS 45
of equivalence classes for any equivalence relation) and by definition each of the Si is
path-connected.
Example II.4.11. The two path-components of the set in Figure 2 are the two discs.
Hence:
Theorem II.4.12. Any subset S ⊂ C is a disjoint union of path-connected subsets.
Moreover, if S is open, each of the path-components of S is again open.
Proof. The statement about openness of the path-components is missing from the
above discussion. Suppose that S is open and that S1 is a path-component. Let z0 be a
point of S1 . Since S is open, some small disc D(z0 , δ) ⊂ S. But any point of z ∈ D(z0 , δ)
can clearly be connected to z0 (by a straight line segment, for example) so D(z0 , δ) ⊂ S1 .
This is what we need for S1 to be open.
Exercise II.4.14. *For enthusiasts: Check that the relation ∼ defined at the
beginning of this section is an equivalence relation.
5. Harmonic functions
We have now discussed analytic functions (those locally expandable as convergent
power series in z) and holomorphic functions (those which are complex-differentiable).
Closely related also are harmonic functions.
Definition II.5.1. Let Ω be an open set of C. A real- or complex-valued function
u in C 2 (Ω) is called harmonic if
∆u = 0, (II.5.1)
where ∆, the Laplace operator or Laplacian is defined as
∂2 ∂2
∆= + . (II.5.2)
∂x2 ∂y 2
Example II.5.2. The functions x, y, x2 − y 2 , xy are all harmonic.
Exercise II.5.16. Show that 12 log(x2 + y 2 ) and tan−1 (y/x) are harmonic where
they are defined (the latter, for example in the half space {x > 0} with the choice
of tan−1 lying in (−π/2, π/2)).
The first equality is the CR equation ux = vy , the second is the symmetry of mixed
partial derivatives, and the third is the CR equation −vx = uy . Thus the continuity of
the second partials is needed for vxy = vyx and hence for the proof to work. The proof
that v is harmonic is precisely similar.
We shall see later that f ∈ H(Ω) implies automatically that the second derivatives
of u and v exist and are continuous. So in fact the real and imaginary parts of f are
harmonic without further assumption.
Conversely, if u is a real harmonic function, then there is a v such that u + iv is
holomorphic—at least under some assumptions on the domain Ω.
Theorem II.5.4. Suppose that u is harmonic in C or in an open disc D(z0 , r) and all
second order partial derivatives are continuous. Then there exists a function v, unique
up to the addition of a constant, such that
f = u + iv
is holomorphic.
Remark II.5.5. Being the imaginary part of a holomorphic function, v is also har-
monic. If u is harmonic and u + iv is holomorphic, then v is called a harmonic conjugate
of u.
Proof. The proof for the disc is the same, so we just give the proof for C.
We have seen that if u+iv is holomorphic then u and v must satisfy the CR equations.
Given u, we define v so that the first CR equation holds, by integration. Let
Z y
w(x, y) = ux (x, t) dt. (II.5.3)
0
For this to make sense, it is good to have the continuity of ux . By the fundamental
theorem of calculus, we have wy = ux . That is the pair (u, w) satisfies the first of the
real CR equations. This pair may not satisfy the other CR equation, though. If we
define v(x, y) = w(x, y) + φ(x), then vy = wy and so the pair (u, v) also satisfies the first
CR equation.
5. HARMONIC FUNCTIONS 47
problem for ue in some simple domain Ω e (for example Ωe = H, the open upper half-space,
or the disc D(0, 1), then ‘all’ we need to do is find a holomorphic function f : Ω → H or
D(0, 1) and define u(z) = u e(f (z)) as in the Theorem.
Thus many (idealized, two-dimensional) physical problems are reduced to conformal
mapping of the domain of physical interest to a simpler one (and solving the problem in
this simple domain).
Example II.5.7. The function u e(w) = log |w| in the disc D∗ (0, 1) is harmonic. A
physical interpretation is the (steady-state) temperature distribution in D∗ (0, 1) when
w = 0 is heated to temperature −2π and the boundary |w| = 1 is held at temperature
0.
We have seen that w = f (z) = (z − i)/(z + i) maps the half-space H = {Im z > 0}
bijectively (and holomorphically) to {|w| < 1}. By the previous theorem, it follows that
z−i
u(z) = log |f (z)| = log
z+i
is harmonic in H \ {i} and vanishes on the boundary Im z = 0. Thus u(z) has the
physical interpretation of the steady-state temperature distribution in H if the point
z = i is held at temperature −2π, along with the side conditions that the real axis is
held at temperature 0, and that the temperature goes to 0 as Im z → +∞.
Exercise II.5.19. Let a be a point in the upper half-space. Find the harmonic
function u which is defined in {Im z > 0} \ {a}, which vanishes along Im z = 0 and
as Im z → +∞, and with a logarithmic singularity near a:
u(z) = log |z − a| + O(1) for |z − a| → 0.
then we obtain
Z Z
(−f gy dx + f gx dy) = (fx gx + fy gy + f ∆g) dxdy. (II.5.8)
C D
Interchanging the roles of f and g and subtracting, the fx gx + fy gy term drops out,
leaving
Z Z
(gfy − f gy ) dx + (f gx − gfx ) dy = (f ∆g − g∆f ) dxdy. (II.5.9)
C D
Cδ
Dδ = {δ 6 |z| 6 ρ}
Figure 3. The set Dδ with its inner and outer boundary circles C =
{|z| = ρ} and Cδ = {|z| = δ}.
by (II.5.10). Similarly
Z Z
f (uy dx − ux dy) = log ρ (uy dx − ux dy) = 0.
C C
50 II. ANALYTIC FUNCTIONS AND HOLOMORPHIC FUNCTIONS
for every ρ with D(z0 , ρ) ⊂ Ω. On the other hand, the integrand has a sign since z0 is
a maximum. More precisely, in (II.5.21) we are integrating a continuous non-negative
function over [0, 2π]: the only way to get zero is if this non-negative function is identically
zero. Hence if D(z0 , r0 ) ⊂ Ω,
u(z0 ) = u(z0 + ρeiθ ) for all 0 < ρ < r0 and θ ∈ [0, 2π]. (II.5.22)
Hence u(z) = u(z0 ) is identically zero in D(z0 , r0 ).
We now use the path-connectedness of Ω to conclude that u(z) = u(z0 ) for all z ∈ Ω.
If z1 ∈ Ω, choose a path γ(t) connecting z0 to z1 . Define
T = sup{t ∈ [0, 1] : u(γ(t)) = u(z0 )}. (II.5.23)
The set of t is non-empty and bounded above. So the supremum exists and is 6 1. Note
that
u(γ(T )) = lim u(γ(t)) = u(z0 ).
t%T
(The arrow in the limit means that T is approached from the left.)
Thus if T = 1 we are done, so suppose if possible that T < 1. In this case, let
z = γ(T ). We have u(z 0 ) = u(z0 ) and so z 0 is a maximum of u. Now we apply the first
0
part of the proof with z0 replaced by z 0 to conclude that u(z) = u(z 0 ) for all z in some
disc D(z 0 , r). This disc contains points of the form γ(T + δ) for sufficently small δ, and
so
u(γ(T + δ)) = u(z0 )
contradicting the definition of T . This contradiction shows that T = 1. So u(z0 ) = u(z1 )
for any given z1 in Ω, and u is constant.
Remark II.5.14. The second part of the proof exploits the path-connectedness of Ω
to prove that u is constant in a way that is very similar to the proof of Theorem II.4.5.
6. ‘Holomorphic at ∞’
In Chapter 1, we introduced the Riemann sphere S . We defined what it means for
f : S → C to be continuous. In this section we shall explain what it means for such a
function to be analytic or holomorphic. We shall give the main point with a minimum
of technical baggage.
The idea is simply to use a new variable ζ = 1/z. Then when |z| is very large, |ζ| is
close to 0, and we define f (z) to be holomorphic ‘near z = ∞’ if and only if fe(ζ) = f (1/ζ)
is holomorphic near ζ = 0.
6. ‘HOLOMORPHIC AT ∞’ 53
To be a little more precise, suppose that for some δ > 0, f is defined in {z : |z| > δ −1 }.
Suppose moreover that f (z) → L as |z| → ∞.
Define fe(ζ) = f (1/ζ) for ζ 6= 0, fe(0) = L. Then
Lemma II.6.1. So defined, fe(ζ) is continuous for ζ ∈ D(0, δ).
Definition II.6.2. f is holomorphic for z ∈ {z : |z| > δ −1 } ∪ {∞} if and only if fe(ζ)
is holomorphic for ζ ∈ D(0, δ).
Similarly, f is analytic for z ∈ {z : |z| > δ −1 }∪{∞} if and only if fe(ζ) is holomorphic
for ζ ∈ D(0, δ).
Examples II.6.3. With this definition z −1 is holomorphic in {|z| > 1} ∪ {∞} (with
value 0 at ∞). For in this case, fe(ζ) = ζ which is holomorphic in D(0, 1).
A rational function R(z) = p(z)/q(z) is holomorphic in a neighbourhood of z = ∞
if and only if deg p 6 deg q. For if
a0 + a1 z + · · · + am z m
R(z) =
b0 + b1 z + · · · + bn z n
with m 6 n, and am 6= 0, bn 6= 0, then
a0 + a1 ζ −1 + · · · + am ζ −m
R(ζ)
e = (II.6.1)
b0 + b1 ζ −1 + · · · + bn ζ −n
a0 ζ m + a1 ζ m−1 + · · · + am
= ζ n−m (II.6.2)
b0 ζ n + b1 ζ n−1 + · · · + bn
and this is holomorphic for |ζ| sufficiently small (so small as to avoid the zeros of the
denominator).
There is something to think about with this definition. Namely, we know that if f
is holomorphic in an open set Ω then it is also holomorphic in any open subset Ω0 ⊂ Ω.
Suppose that f is holomorphic in {z : |z| > δ −1 } ∪ {∞} in the sense of Definition II.6.2.
Combining with the above remark, this should imply that f is also holomorphic in the
subset {z : |z| > δ −1 }. But this set does not contain ∞ and we already have a definition
of holomorphic in this set, namely complex-differentiable at every point z with |z| > δ −1 .
The new definition says that f is holomorphic if fe(ζ) is complex differentiable for all ζ
with 0 < |ζ| < δ, where ζ = 1/z. Why are these two definitions consistent?
The answer is that if fe(ζ) = f (1/ζ) then by the chain rule, fe is holomorphic
(complex-differentiable) at ζ0 6= 0 if and only if f is holomorphic at z0 , where z0 = 1/ζ0 .
The key point is that ζ = 1/z is itself a holomorphic change of variables and this is
why Definition II.6.2 is consistent with the previous definition of holomorphic. There
are similar remarks with ‘holomorphic’ replaced by ‘analytic’ throughout.
CHAPTER III
Complex Integration
Examples III.1.4.
(i) γ(t) = Reit , t ∈ [0, 2π] is the circle centred at 0, radius R, traversed once
anticlockwise. This is an example of a simple, closed curve.
(ii) γ(t) = Reit , t ∈ [0, π] is the semi-circlular arc centred at 0, radius R, which
joins z = R to z = −R in the upper half-plane. This is an example of a
simple curve.
(iii) γ(t) = z0 + (z1 − z0 )t, t ∈ [0, 1] is the straight-line segment joining z0 at
t = 0 to z1 at t = 1. This is a simple curve.
It is also handy to denote the circular path γ(t) = z0 + Reit , 0 6 t 6 2π, simply by
|z − z0 | = R. In other words, when we write ‘the curve |z − z0 | = R’ we mean ‘the curve
|z − z0 | = R traversed once anticlockwise’.
We shall extend our definitions to allow for piecewise smooth curves in §1.4.
Note that the condition γ 0 (t) 6= 0 for all t means that the tangent vector is non-zero
at every point and that γ is locally well approximated by its tangent vector1, and there
are no sudden changes of direction along γ. If the assumption is dropped, this is not the
case. For example, 3
t , −1 6 t 6 0;
γ(t) = (III.1.2)
it3 , 0 6 t 6 1.
is C 2 , but γ 0 (0) = 0. This curve is runs along the negative real axis for −1 6 t < 0, hits
the origin z = 0 at t = 0 and then turns a sharp corner to go up the positive imaginary
axis for 0 < t 6 1, ending up at z = i.
Suppose that Ω is an open subset of C and that f : Ω → C is continuous. If
γ : [a, b] → C is a smooth curve in Ω, we define the integral of f along γ,
Z Z Z b
f = f (z) dz := f (γ(t))γ 0 (t) dt. (III.1.3)
γ γ a
γ0
Note that is continuous because of the assumption that γ is smooth. So the integrand
on the RHS is a continuous function on a closed bounded interval, so definitely exists
(as a Riemann integral).
Note that here we have the integral of a complex-valued function F (t) = f (γ(t))γ 0 (t)
of the real variable t. If F (t) = F1 (t) + iF2 (t) this is by definition
Z b Z b Z b Z b
F (t) dt = (F1 (t) + iF2 (t)) dt = F1 (t) dt + i F2 (t) dt. (III.1.4)
a a a a
Rb
Note that a is then a complex-linear mapping from continuous functions to C in
the sense that Z b Z b Z b
(λF (t) + µG(t)) dt = λ F (t) dt + µ (t) dt (III.1.5)
a a a
for all complex-valued continuous functions F and G and all complex constants λ and
µ.
More generally, if P (x, y) and Q(x, y) are continuous functions defined in Ω and γ is
the smooth curve in Ω we had before, we define the line integral
Z Z b
P (X(t), Y (t))X 0 (t) + Q(X(t), Y (t))Y 0 (t) dt.
P dx + Q dy = (III.1.6)
γ a
and so Z Z 1
z 2 dz = (z0 + (z1 − z0 )t)2 (z1 − z0 ) dt.
[z0 ,z1 ] 0
By the chain rule, the integrand is
d1
(z0 + (z1 − z0 )t)3
dt 3
so Z 1
2 1 3 1 1
z dz = (z0 + (z1 − z0 )t) = z13 − z03
[z0 ,z1 ] 3 0 3 3
Note that this is the same as what you might have guessed by extrapolating from
real integration.
Example III.1.7. Consider the integral of z −1 along the two curves
γ1 (t) = eit , t ∈ [0, π]
and
γ2 (t) = e−it , t ∈ [0, π].
Both γ1 and γ2 connect z = 1 to z = −1, γ1 in the upper half-plane, γ2 in the lower
half-plane. Then Z Z π
dz
= idt = πi.
γ1 z 0
Similarly, Z Z π
dz
= (−i)dt = −πi.
γ2 z 0
Note that the ‘naive’ answer log(−1) − log(1) is ambiguous in this case (because you
have to make a choice of arg to define log). This is reflected in the fact that while
the integrals along γ1 and γ2 are perfectly well-defined, the answer does depend
on the choice of path. We shall discuss this at greater length below.
By the previous result, this depends on the curve and not on the parameterization.
We often need to estimate integrals that we can’t evaluate exactly. Here is the main
tool.
Proposition III.1.15. (Length-sup or M L Lemma.) Let γ : [a, b] → S be a smooth
curve of length L. Let f : S → C be a continuous function such that
sup |f (γ(t))| = M (III.1.15)
a6t6b
60 III. COMPLEX INTEGRATION
Then
Z
f 6 ML (III.1.16)
γ
R
Proof. Applying the triangle inequality for γ f,
Z Z b Z b
0
f = f (γ(t))γ (t) dt 6 |f (γ(t))| |γ 0 (t)|dt. (III.1.17)
γ a a
as required.
where ΓR (t) = Reit , θ1 6 t 6 θ2 , for any given angles 0 < θ1 < θ2 < 2π.
Exercise III.1.7. Let log z = log |z| + i arg(z), where 0 < arg(z) < 2π. Show
that if u is a continuous function in {|z| < 1}, then
Z
u(z) log z dz → 0 as δ → 0 (III.1.20)
Γδ
where Γδ (t) = δeit , θ1 6 t 6 θ2 , for any given angles 0 < θ1 < θ2 < 2π.
1. PATHS AND LINE-INTEGRALS 61
1.4. Piecewise smooth paths. Consider the semicircular path Γ shown below. It
is not smooth because of the corners at −1 and 1.
γ2
0 γ1
z = −1 z=1
It is worth noting that there is a consistency check to be carried out here: namely if
γ : [a, b] → C is a smooth curve and c ∈ (a, b), suppose we choose to think of γ as the
sum γ1 + γ2 where
γ1 (t) = γ(t) for a 6 t 6 c (III.1.24)
and
γ2 (t) = γ(t) for c 6 t 6 b (III.1.25)
then we’d better have Z Z Z
f= f+ f. (III.1.26)
γ γ1 γ2
This however is easily verifed and follows from the standard rule
Z c Z b Z b
F (t) dt + F (t) dt = F (t) dt (III.1.27)
a c a
for integration of continuous functions over real intervals.
We also apply the notation Γ = γ1 + γ2 if γ1 and γ2 are closed curves. We shall also
write
Γ = γ1 − γ2 = γ1 + γ2opp . (III.1.28)
Thus
62 III. COMPLEX INTEGRATION
The integral is extended from smooth curves to piecewise smooth curves in the obvious
way: if γ = γ1 + · · · + γn , then
Z n Z
X
f (z) dz = f (z) dz.
γ j=1 γj
Exercise III.1.8. Calculate the integrals of the following functions around the
perimeter of the square with vertices at 0, 1, 1 + i and i: f (z) = z 2 , f (z) = z,
f (z) = |z|2 .
F 0 = 0.
R
In particular, if γ is a closed curve, then γ
Example III.1.21. Since the derivative of z n+1 /(n+1) is equal to z n (if n 6= −1),
it follows that if γ is a curve which connects z0 to z1 , then
z n+1 z n+1
Z
z n dz = 1 − 0 .
γ n+1 n+1
Example III.1.22. On the other hand, by direct computation,
Z
dz
= 2πi.
|z|=1 z
It follows that there does not exist a holomorphic function F (z) defined in an open
set containing the circle |z| = 1, with F 0 (z) = 1/z. In other words, there is no way
to define log z as a function holomorphic in C \ {0}.
Proof. Pick z0 ∈ Ω. For every point z ∈ Ω let γ be a smooth (or piecewise smooth)
path connecting z0 to z, and define
Z
F [γ, z] = f (w) dw. (III.1.36)
γ
and so Z
F (z1 + h) − F (z1 ) − hf (z1 ) = (f (w) − f (z1 )) dw.
[z1 ,z1 +h]
Rearranging,
F (z1 + h) − F (z1 )
Z
1
− f (z1 ) = (f (w) − f (z1 )) dw . (III.1.38)
h |h| [z1 ,z1 +h]
We need to show that the RHS tends to zero as |h| → 0. For this we use the continuity
of f at z1 . Given ε > 0, there exists δ > 0 so that D(z1 , δ) ⊂ Ω and
w ∈ D(z1 , δ) ⇒ |f (w) − f (z1 )| < ε.
If |h| < δ, |f (w) − f (z1 )| is bounded by ε over the segment [z1 , z1 + h] and so by the ML
lemma,
Z
1 1
|h| < δ =⇒ (f (w) − f (z1 )) dw < |h| × ε = ε, (III.1.39)
|h| [z1 ,z1 +h] |h|
since the length of the segment is |h|. Combining this with (III.1.38) we have proved
that F 0 (z1 ) exists and is equal to f (z1 ), as required.
Exercise III.1.9. Show that if Γ is any piecewise smooth closed curve in C, then
Z
sin z dz = 0.
Γ
Exercise III.1.10. Show that if f : D(0, R) → C is given by a convergent
power
R series in D(0, R) and Γ is any piecewise smooth closed curve in D(0, R), then
Γ f (z) dz = 0. [Can you apply Theorem III.1.20?]
2. GREEN’S THEOREM, CAUCHY’S THEOREM AND CAUCHY’S INTEGRAL FORMULA 65
We are now going to embark on Cauchy’s Theorem. This is a theorem with different
proofs and approaches, each with their advantages and disadvantages.
The first proof uses Green’s Theorem in the plane:
Theorem III.2.7. Let D be a bounded domain with piecewise smooth boundary ∂D
(see Definition III.2.1 for notation). Suppose that P and Q are two real- or complex-
valued functions which are C 1 in an open neighbourhood of D ∪ ∂D. Then we have
Z Z
∂Q ∂P
− dxdy = P dx + Q dy (III.2.2)
D ∂x ∂y ∂D
Remark III.2.8. The above theorem is true as stated, but you may not have seen
a rigorous proof of this. Such a rigorous proof would certainly need to define the LHS
(namely the integral over the domain D). We shall leave this for now. See Chapter IV
for discussion.
Theorem III.2.9. (Cauchy’s Theorem) Let D be a bounded domain with piecewise
smooth boundary ∂D. Suppose that f : Ω → C is holomorphic in an open set Ω which
contains D ∪ ∂D. Then Z
f (z) dz = 0. (III.2.3)
∂D
Here is an application which ties in with the complex fundamental theorem of cal-
culus.
Theorem III.2.11. Let Ω be convex open set and let f : Ω → C be holomorphic.
Then there exists F : Ω → C, holomorphic, such that F 0 = f in Ω.
Proof. We proceed as in the proof of Theorem III.1.23. We allow ourselves to
assume that f is C 1 as well. Select any point z0 ∈ Ω and for z ∈ Ω, define
Z
F (z) = f (w) dw. (III.2.4)
[z0 ,z]
This is identical to (III.1.37) in the proof of Theorem III.1.23. That F 0 (z1 ) = f (z1 )
follows in exactly the same way as in the proof of Theorem III.1.23.
Remark III.2.12. The main idea is that the hypothesis ‘f holomorphic’ implies that
the integral of f around the boundary of any triangle ∂T is zero, and this is sufficent to
run the argument of Theorem III.1.23.
R
Corollary III.2.13. If f is holomorphic in a convex set Ω, then γ f (z) dz = 0 for
any piecewise smooth closed curve γ contained in Ω.
Proof. By Theorem III.2.11, there is a holomorphic function F : Ω → C such that
F0= f . Then the result follows from the complex fundamental theorem of calculus,
Theorem III.1.20.
2.2. The Logarithm. Consider the cut plane Ω = C \ (−∞, 0]. Define
Z z
dw
L(z) = . (III.2.6)
1 w
where the integral is along the unique straight line which connects 1 to z. It is clear that
1/w is holomorphic in any triangle contained in Ω, so the integral around the perimeter
2. GREEN’S THEOREM, CAUCHY’S THEOREM AND CAUCHY’S INTEGRAL FORMULA 69
of such a triangle is zero, and so by the same argument as in Theorem III.1.20, L(z) is
holomorphic in Ω and L0 (z) = 1/z, L(1) = 0.
Let K(z) = z exp(−L(z)). Then K is holomorphic in Ω and
K 0 (z) = exp(−L(z)) − zL0 (z) exp(−L(z)) = 0 since zL0 (z) = 1.
Hence K(z) is constant and since K(1) = 1 this constant value is 1. So
exp(L(z)) = z
and L(z) is an inverse of the exponential function, also known as the logarithm.
Now we know that L0 (z) = 1/z in Ω, we have path-independence of the integral (as
long as the path is contained in Ω, of course). In particular, we compute
Z Z
dw dw
L(z) = +
γ1 w γ2 w
where γ1 is the segment [1, |z|] and γ2 is the circular arc from |z| to z, then we have
Z
dw
= log |z| (III.2.7)
γ1 w
and Z Z arg(z)
dw
= idt = i arg(z). (III.2.8)
γ2 w 0
Hence L(z) = log |z| + i arg(z), where
−π < arg(z) < π for z ∈ Ω
is the principal value of arg(z).
L(z) is called a holomorphic branch of log z. Any other holomorphic branch of log z
defined in Ω is obtained by adding 2nπi to L(z), where n ∈ Z. Holomorphic branches
of log z can be defined in other open sets of C. For example, if Ω0 = C \ [0, ∞), we have
a holomorphic branch of log z,
log(z)
f = log |z| + iarg(z),
f 0 < arg(z)
f < 2π.
And we can add 2nπi to any holomorphic branch to define another branch holomorphic
in the same domain.
Using the holomorphic branch L(z) of log z in Ω, we may also define a holomorphic
branch of the complex power z a , by defining
z a := exp(aL(z)) for z ∈ Ω. (III.2.9)
This is a holomorphic function in the cut plane Ω. In general, different holomorphic
branches of z a are determined by different holomorphic branches of log z. z a can only
be defined holomorphically in the whole of C if a is a non-negative integer; in this case
it is independent of the choice of holomorphic branch of log z.
Exercise III.2.13. Show that there are precisely two distinct holomorphic
branches of z 1/2 defined in the cut plane Ω. How many distinct holomorphic branches
of z 1/3 are there? Write down all holomorphic branches of z a defined in Ω.
70 III. COMPLEX INTEGRATION
Exercise III.2.14. Suggest a domain U , containing the part of the real axis with
x > 1, on which a holomorphic branch of log(z − 1) can be defined.
f (z)
Proof. Let F (z) = z−z 0
and let Dδ = D \ D(z0 , δ), where δ > 0 is so small that
D(z0 , δ) ⊂ D. Then F is holomorphic in Ω \ {z0 } which contains Dδ . Thus Cauchy’s
Theorem applies to F and the domain Dδ . We have
∂Dδ = ∂D − γ
where γ(t) = z0 + δeit , 0 6 t 6 2π. Thus Cauchy gives
Z Z
F (z) dz = F (z) dz. (III.2.11)
∂D γ
The integral on the RHS is computed explicitly as
Z 2π Z 2π
f (z0 + δeit ) it
it
iδe dt = i f (z0 + δeit ) dt.
0 δe 0
Now we argue as in the proof of Theorem II.5.8.
Z 2π Z 2π Z 2π
f (z0 + δeit ) dt = f (z0 ) dt + (f (z0 + δeit ) − f (z0 )) dt. (III.2.12)
0 0 0
Given ε > 0, there exists δ > 0 so that
|z − z0 | < 2δ ⇒ |f (z) − f (z0 )| < ε/2π.
For such δ,
Z 2π Z 2π
it
(f (z0 + δe ) − f (z0 )) dt 6 |f (z0 + δeit ) − f (z0 )| dt < ε.
0 0
R 2π
Since 0 f (z0 ) dt = 2πf (z0 ), the result follows by combining this estimate with (III.2.12).
Proposition III.2.15. If |a| =
6 R, then
Z
dz 2πi if |a| < R;
= (III.2.13)
|z|=R z − a 0 if |a| > R
Proof. If |a| > R then a is outside the disc D(0, R) and so 1/(z − a) is holomorphic
in a neighbourhood of this disc. Hence by Cauchy’s Theorem, the integral is zero.
If |a| < R, then we can apply Cauchy’s Integral formula with f (z) = 1.
2. GREEN’S THEOREM, CAUCHY’S THEOREM AND CAUCHY’S INTEGRAL FORMULA 71
2.4. Comments. Once we have Green’s Theorem for bounded domains with piece-
wise smooth boundary, the proof of Cauchy’s Theorem for a holomorphic function f is
very simple, at least if we assume that f is continuously differentiable. Here Green’s
Theorem is being used as a ‘black box’, and in the interest of rigour and conceptual
foundations of the subject it is worth thinking about how ‘easy’ a theorem Green’s
Theorem really is. I give a sketch proof of Green’s Theorem in Chapter IV, so if you
are interested, you can read about it and decide for yourself. The two main difficulties
are defining the two-dimensional integral over D, and dealing with a general piecewise
smooth boundary. It requires a fair amount of spadework to deal with these issues in
a fully rigorous fashion. I should emphasise, however, that Green’s Theorem and its
higher-dimensional cousins (the divergence theorem in 3D, for example) can be made
entirely rigorous and are fundamental tools in other areas of geometry and the analysis
of PDE (as well as applied mathematics) so I think it is good to see Cauchy’s Theorem
and the Cauchy Integral Formula derived using this tool.
Apart from the difficulties involved in giving a rigorous account of Green’s Theorem,
a possible defect in the proof which uses it is the need to assume that our holomorphic
function is C 1 . The Goursat2 or Cauchy–Goursat
R Theorem avoids this assumption.
Goursat’s Theorem establishes that ∂R f (z) dz = 0 if f is holomorphic in an open
set containing a closed rectangle3 R. For the detailed proof, see any reputable text on
complex analysis. The idea is as follows.
Suppose
Z
f (z) dz 6= 0.
∂R
By multiplying f by a non-zero complex number, suppose then that
Z
f (z) dz = 1. (III.2.14)
∂R
Also, let L = L(∂R) be the length of the perimeter of the rectangle. Divide R into four
equal rectangles by cutting R into four quarters. Denote these S1 , S2 , S3 , S4 . Then
Z X4 Z
f= f (III.2.15)
∂R j=1 ∂Sj
(there is cancellation along the ‘interior’ sides of the Sj ) and by the triangle inequality
Z X 4 Z
1= f (z) dz 6 f . (III.2.16)
∂R j=1 ∂Sj
R
It follows that for at least one of the four Sj , | ∂Sj f | > 1/4. Denote by R1 a choice of
R
Sj such that | ∂Sj f | > 1/4. Note that L(∂R1 ) = L/2.
R
Continue the argument: cut R1 into quarters, pick a quarter R2 on which | ∂R2 f | >
4−2 and so on. We obtain a nested sequence of closed rectangles
R ⊃ R1 ⊃ R2 ⊃ . . . ⊃ Rn ⊃ . . .
with
• L(∂R −n
R n ) = 2 L, −n
• | ∂Rn f (z) dz| > 4 .
Because the Rn are nested rectangles and L(∂Rn ) → 0, there is a unique point z0
common to all Rn . The proof is finished
R off by using the fact that f is holomorphic
at z0 . This allows us to estimate | ∂Rn f | for large n, and in particular to obtain an
estimate which contradicts the lower bound 4−n above.
Once we have the Cauchy–Goursat Theorem, it is possible to prove directly the
Cauchy Integral Formula in the form
Z
1 f (z) dz
f (w) = (III.2.17)
2πi |z−z0 |=r z − w
for w ∈ D(z0 , r). From this we obtain that f is analytic in D(z0 , r) and in particular
that f is C 1 . At this point we could use Green’s Theorem to obtain the more general
form of Cauchy’s Theorem stated in Theorem III.2.9. There are also variants of this line
of argument that avoid Green’s Theorem completely.
Theorem III.3.2. Let f ∈ H(Ω) and let z0 ∈ Ω. Let D(z0 , R) ⊂ Ω. Then the Taylor
expansion
∞
X
f (z) = an (z − z0 )n , (III.3.2)
n=0
where
f (n) (z0 )
an = , (III.3.3)
n!
is absolutely convergent in D(z0 , R). Moreover, for any r < R, we have Cauchy’s
Inequalities
|an | 6 r−n sup |f (z)|. (III.3.4)
|z−z0 |=r
Remark III.3.3. Formally, if (III.3.1) is true for n then we might imagine being
able to differentiate both sides with respect to z0 . Assuming it is valid to pass the
differentiation through the integral, we obtain (III.3.1) with n replaced by n + 1, since
d n! (n + 1)!
n+1
= .
dz0 (z − z0 ) (z − z0 )n+2
The base case n = 0 is the Cauchy Integral Formula, Theorem III.2.14. This argument
can be made rigorous.
We shall prove Theorem III.3.2 first. Cauchy’s formula for derivatives will follow
from this proof.
Proof. (Of Theorem III.3.2.) Pick any positive r, r < R. It is enough to show that
(III.3.2) is valid in D(0, r) for every r < R.
To simplify notation, suppose that z0 = 0.
For |z| < r, Cauchy’s Integral Formula gives
Z
1 f (w)
f (z) = dw. (III.3.5)
2πi |w|=r w − z
The idea is to expand 1/(w − z) using the binomial theorem. For |w| = r, |z| < r,
we have
∞
1 1 X z n
= , since |z/w| < 1. (III.3.6)
w−z w w
n=0
If we naively subsitute this into (III.3.5) and interchange the sum and the integral, we
obtain
X
f (z) = an z n for |z| < r (III.3.7)
where Z
1 f (w)
an = dw. (III.3.8)
2πi |w|=r wn+1
Then (III.3.7) gives f as a convergent power series.
74 III. COMPLEX INTEGRATION
In order to justify this argument, i.e. the interchange of the infinite sum and the
integral, let us replace (III.3.6) by a finite sum plus an error term. If |u| < 1, and N > 0,
we have
∞
X N
X ∞
X
un = un + un (III.3.9)
n=0 n=0 n=N +1
N
X uN +1
= un + . (III.3.10)
1−u
n=0
zn
P
Thus the power series an converges to f (z) in D(0, r).
Cauchy’s Inequalities follow by applying the ML lemma to (III.3.8).
Proof. (Of Theorem III.3.1.) In the course of the previous proof we showed that
f is analytic in A (Ω), so all derivatives exist. We already know that if f is analytic,
the coefficients in its power series expansion centred at 0 are the Taylor coefficients
f (n) (0)/n!. But we have the integral formula (III.3.8) for an , so
f (n) (0)
Z
1 f (w)
= an = dw. (III.3.15)
n! 2πi |w|=r wn+1
To prove the stated result, with the more general contour, suppose that D is as given in
the statement of Theorem III.3.1; we continue to assume z0 = 0. Pick r so small that
3. HOLOMORPHIC FUNCTIONS ARE ANALYTIC; HIGHER DERIVATIVES 75
D(0, r) ⊂ D and let Dr = D \ D(0, r) Then z 7→ z −n−1 f (z) is holomorphic inside and
on ∂Dr so Cauchy’s Theorem applies to give
Z Z
f (z) dz f (z) dz
n+1
= n+1
∂D z |z|=r z
and we have already shown that the RHS is equal to f (n) (0)/n!.
3.1. Liouville’s Theorem.
Theorem III.3.4. Suppose that f : C → C is holomorphic. If f is bounded then f
is constant.
Proof. Suppose that f is bounded, |f (z)| 6 M for all z ∈ C. We use the formula
for the derivative to prove that f 0 (z0 ) = 0 for all z0 ∈ C. Let z0 ∈ C and let R > 0.
Then Z
0 1 f (z)
f (z0 ) = dz
2πi |z−z0 |=R (z − z0 )2
and using the ML lemma,
M
|f 0 (z0 )| 6 . (III.3.16)
R
Here M is fixed, but R can be chosen as we please. Taking R to ∞, we see that
f 0 (z0 ) = 0.
Exercise IV.1.1. Show that a zero α of p(z) has order k if and only if
p(α) = p0 (α) = · · · = p(k−1) (α) = 0, p(k) (α) 6= 0. (IV.1.3)
Exercise IV.1.2. If p(z) is a polynomial of degree 6 10 and there are 17 distinct
complex numbers α1 , . . . , α17 for which p(αj ) = 0, what can be said about p(z)?
Generalize.
We claim that the above story is very nearly the same for analytic functions.
1.2. Zeros of analytic functions. Let Ω be a domain and let f ∈ A (Ω). Suppose
that z0 is a zero of f , so z0 ∈ Ω and f (z0 ) = 0.
Since f is analytic, we have a disc D(z0 , δ) such that we have a power series expansion,
∞
X
f (z) = an (z − z0 )n , for z ∈ D(z0 , δ). (IV.1.4)
n=0
Since f (z0 ) = 0, we have a0 = 0. There are precisely two possibilities:
(a) z0 is a zero of finite order k, say: this happens by definition if
a0 = · · · = ak−1 = 0, ak 6= 0;
77
78 IV. ZEROS, POLES, AND THE RESIDUE THEOREM
Proof. We have already seen that conditions (a)–(c) are equivalent. So we have to
prove that any one of them is equivalent (d). Clearly (d) is the strongest of the lot and
implies the other three. So it is enough to prove that (c)⇒(d). For this, we follow a
line which we have encountered at least twice before (Theorem II.4.5 and the Maximum
Principle, Theorem II.5.12), using the path-connectedness of Ω.
1. ZEROS OF ANALYTIC FUNCTIONS 79
Exercise IV.1.3. What are the orders of the zeros of ez −1, of sin πz, of z −sin z
(at z = 0)?
Exercise IV.1.4. Write down three further simple analytic functions and find
the order of their zeros.
Exercise IV.1.5. A function f ∈ A (C) is identically 0 on the real axis. Show
that f (z) = 0 in C.
Exercise IV.1.6. A function f (z) is holomorphic in C and is real on the real
axis. Show that g(z), where
g(z) = f (z̄)
is also holomorphic in C. Show that f (z) = g(z) for all z.
80 IV. ZEROS, POLES, AND THE RESIDUE THEOREM
Exercise IV.2.7. Show that h(z) remains bounded as z → 0 along the imaginary
axis. What happens to h(z) as z → 0 along the real axis a) from the left? and b)
from the right?
Show that for every non-zero w ∈ C, there is a sequence zn → 0 (depending on
w) such that f (zn ) = w as n → ∞.
Thus in any neighbourhood of 0, exp(1/z) takes every complex value apart from
0. Is that shocking?
We shall now give a precise and general definition of these three types of singularity.
We shall do this by the use of Laurent Series.
By an annulus A, we mean a set of the form
A = {z ∈ C : a < |z| < b}. (IV.2.1)
where a and b are real numbers. We allow the possibility that a = 0, in which case A is
the punctured disc centred at 0 of radius b, also denoted D∗ (0, b).
Laurent’s Theorem gives an expansion in positive and negative powers of z of any
function holomorphic in A.
Theorem IV.2.3. Let f : A → C be holomorphic. Then there exist coefficients
cn ∈ C such that
∞
X
f (z) = cn z n for all z ∈ A. (IV.2.2)
n=−∞
1However, it can be regarded as a holomorphic mapping from C into the Riemann sphere S
2. POLES AND LAURENT SERIES 81
where γ(t) = reit , 0 6 t 6 2π, for any fixed radius r between a and b.
Proof. For a < a0 < b0 < b, the smaller annulus A0 = {a0 < |z| < b0 } is a bounded
domain whose closure is contained in A. We have ∂A0 = γ1 − γ2 , where
γ1 (t) = b0 eit , 0 6 t 6 2π,
is the outer circle traversed anti-clockwise, and
γ2 (t) = a0 eit , 0 6 t 6 2π.
is the inner circle traversed anticlockwise. The formula ∂A0 = γ1 − γ2 takes care of the
orientation of the two components of ∂A0 . If z ∈ A0 , we have, by Cauchy’s Integral
Formula, Theorem III.2.14,
Z Z Z
1 f (w) 1 f (w) 1 f (w)
f (z) = dw = dw − dw. (IV.2.4)
2πi ∂A0 w − z 2πi γ1 w − z 2πi γ2 w − z
Now we expand (w − z)−1 in powers, noting that on γ1 , |w| = b0 > |z|, while on γ2 ,
|w| = a0 < |z|. Thus we have different expansions on the two curves. Indeed, for the
γ1 -integral, we must use
∞
1 1 z −1 X z n
= 1− = , for |z| < |w| (IV.2.5)
w−z w w wn+1
n=0
while for the γ2 - integral we must use
X wn ∞
1
− = , for |w| < |z|. (IV.2.6)
w−z z n+1
n=0
By the same argument as in the proof of CIF, we can interchange the sum and integral
in each case, leaving us with
∞
X −1
X
f (z) = Cn z n + Dn z n for z ∈ A0 (IV.2.7)
n=0 n=−∞
where Z Z
1 −n−1 1
Cn = f (z)z dz, Dn = f (z)z −n−1 dz, (IV.2.8)
2πi γ1 2πi γ2
We have now almost proved the theorem, for we have the expansion (IV.2.7) for z ∈ A0
and we have formulae for the coefficients. Note that in fact the coefficients given in
(IV.2.8) agree with the formula (IV.2.3) by Cauchy’s Theorem. Indeed f (z)z −n−1 is
holomorphic in A for each integer n. So we have
Z Z
f (z)z −n−1 dz = f (z)z −n−1 dz (IV.2.9)
|z|=r |z|=ρ
for any two radii a < r < ρ < b by Cauchy’s Theorem. Thus the coefficients in the
expansion of f , (IV.2.7) are independent of the choice of annulus A0 . Since any point z
of A is in some annulus A0 , the proof is finished.
82 IV. ZEROS, POLES, AND THE RESIDUE THEOREM
where, taking r = 1 in the formula for the cn , we have the standard formula for the
Fourier coefficients
Z 2π
1
cn = e−inθ F (θ) dθ.
2π 0
Since the Laurent coefficients are unique, any method which yields a valid expansion
in positive and negative powers must give the Laurent expansion.
Example IV.2.5. The power series expansion of the function
z3
f (z) = sin z = z − + ···
3!
is also its Laurent series expansion. So the cn for n < 0, all vanish for this function.
For the function 1/(1 − z), we have c−1 = −1 and all other terms in the expansion
in powers of (z − 1) are zero.
For exp(1/z), we have the standard power series for the exponential function, which
implies
∞
X 1 −n
exp(1/z) = z
n!
n=1
for all z 6= 0. Thus in the Laurent expansion about z = 0, we have c−n = 1/n! for all
n > 0, cn = 0 for all n > 0.
2.1. Classification of isolated singularities.
Definition IV.2.6. Let f be holomorphic in D∗ (z0 , δ) so that z0 is an isolated
singularity of f . Let the Laurent expansion of f in D∗ (z0 , δ) be
∞
X
f (z) = cn (z − z0 )n
n=−∞
Note that: the principal part is zero if the singularity is removable; it is a finite sum
of negative powers (z − z0 )−n if the singularity is a pole; and it is an infinite sum of
negative powers in the case of an essential singularity.
Proof. There can be at most one holomorphic function with this property, for fe(z0 )
must be defined as limz→z0 f (z). To see that fe exists, note that by definition, the Laurent
expansion of f near z0 is:
∞
X
f (z) = cn (z − z0 )n , 0 < |z − z0 | < δ.
n>0
Thus we define fe(z0 ) = c0 and then fe(z) is a convergent power series centred at z0 , and
this gives the needed holomorphic extension of f .
z3 z2
sin z = z − + O(z 5 ) so f (z) = 1 − + O(z 4 ).
3! 3!
This shows that z = 0 is a removable singularity if we define f (0) = 1. Having removed
the singularity, f is now holomorphic in the entire complex plane.
Proof. Suppose, for a contradiction, that there were an infinite number {z1 , z2 , . . .}
of poles in D. Now D is closed and bounded, so {zj } has a limit point w, say, in D.
Then w cannot possibly be an isolated singularity, contradicting the assumption that f
is meromorphic.
Let the (distinct) poles be {z1 , . . . , zn }.
Pick δ > 0 so small that Dj = D(zj , δ) ⊂ D for each j, and so that these discs are
mutually disjoint. Let Dδ = D \ nj=1 Dj . Then by definition f is holomorphic inside
S
and on Dδ , so Cauchy’s Theorem gives:
Z Z n Z
X
f (z) dz = 0 and so f (z) dz = f (z) dz. (IV.3.2)
∂Dδ ∂D j=1 |z−zj |=δ
)n
P
Now the sum n>0 cn (z − zz converges to a holomorphic function in a neighbourhood
of z = z1 and so the integral around |z − z1 | = δ is zero. Thus the integral is equal to
the integral of the principal part:
−1 Z
X
cn (z − z1 )n dz. (IV.3.4)
n=−k |z−z1 |=δ
and use the integral formulae for the c−n from Theorem IV.2.3 to get bounds on
these coefficients (analogous to the Cauchy Inequalities) to prove the result.]
Exercise IV.3.11. Let p(z) and q(z) be holomorphic in the disc D(z0 , δ), δ > 0.
Assume that
p(z0 ) 6= 0, q(z0 ) = q 0 (z0 ) = 0, q 00 (z0 ) 6= 0. (IV.3.8)
Show that f (z) = p(z)/q(z) has a double pole at z = z0 and that the residue is equal
to
p0 (z0 ) 2 p(z0 )q 000 (z0 )
Resz0 f = 2 00 − . (IV.3.9)
q (z0 ) 3 q 00 (z0 )2
Exercise IV.3.12. *For enthusiasts. But worth doing this problem for the
intuition you will get about the relation between zeros and poles and just how bad
essential singularities are.
Suppose that f (z) is a holomorphic function in a punctured disc D∗ (z0 , r) =
{z ∈ C : 0 < |z − z0 | < r}.
(a) Show that f (z) has a pole of order m > 0 at z = z0 if and only if there
exists a holomorphic function g : D(z0 , r) → C, with g(z0 ) 6= 0, and such
that
g(z)
f (z) = . (IV.3.10)
(z − z0 )m
(b) Show that f (z) has a pole of order m > 0 at z = z0 if and only if 1/f (z)
has a zero of order m at z = z0 .
(c) Use (b) to show that the isolated singularity of f at z0 is a pole if and only
if |f (z)| → ∞ as z → z0 .
(d) Prove the theorem of Casorati and Weierstrass: if z0 is an (isolated) es-
sential singularity of f , then given w ∈ C, there exists a sequence zn → 0
(depending upon w), such that f (zn ) → w as zn → ∞. Thus near an es-
sential singularity f comes arbitrarily close to every complex value. [Hint:
suppose that for some w, there is no such sequence. Write down carefully
what this means and then figure out what kind of isolated singularity the
function z 7→ (f (z) − w)−1 must have at z0 . If you succeed in this, you
should end up with a contradiction.]
4. Definite integrals
4.1. Rational functions.
The above method works for any rational function f (z) which is O(|z|−2 ) as
|z| → ∞ having no poles on the real axis. Integration around the above semicircular
contour gives
Z ∞
f (x) dx = 2πi × sum of residues of f in the upper half-plane.
−∞
Note that if the rational function is even, you get the integral from 0 to ∞ as well,
just by dividing the answer for the integral from −∞ to ∞ by 2.
4.2. Rational functions of cos and sin. Consider the integral
Z 2π
dθ
J=
0 a + b cos θ
where a > b > 0 (so that a + b cos θ is never zero). Such an integral can be computed
directly, but also using contour integral methods. I like to think of this as reverse en-
gineering. We interpret the integral as the parameterized version of an integral around
|z| = 1 of a holomorphic function. To do this, we set z = eiθ (the standard parameteri-
zation of the unit circle) and use the formulae
1 1
cos θ = (z + z −1 ), sin θ = (z − z −1 )
2 2i
dz
along with dθ = iz . In the case of the integral J, we obtain
Z Z Z
1 dz dz 2i dz
J= −1
= −2i 2
=−
|z|=1 a + b(z + z )/2 iz |z|=1 b + 2az + bz b |z|=1 1 + 2λz + z 2
where λ = a/b > 1 by assumption.
Let f (z) = (z 2 + 2λz + 1)−1 . Then f is holomorphic in C away from the two roots
z1 and z2 of the quadratic z 2 + λz + 1 = 0. Solving,
p p
z1 = −λ + λ2 − 1, z2 = −λ − λ2 − 1.
√
Since λ > 1, |z2 | = λ + λ2 − 1 > λ > 1 but z1 z2 = 1 so |z1 | < 1. Hence
2i
J =− × 2πi × Resz1 (f ).
b
We can use Proposition IV.3.2 to calculate this residue as the pole at z = z1 is simple.
We obtain
1 1
Resz1 (f ) = = √ .
z1 − z2 2 λ2 − 1
Hence
2π
J=√ .
a2 − b2
Integrals from 0 to π of rational functions can sometimes be obtained from the full
integral from 0 to 2π by using the usual properties
sin(θ + π) = − sin θ, cos(θ + π) = − cos θ
and/or the even/odd properties of cos and sin.
4. DEFINITE INTEGRALS 89
Exercise IV.4.15. Let a be a real number, 0 < a < 1. By evaluating the integral
Z
1 1
dz
2πi |z|=1 (z − a)(z − a−1 )
prove that Z 2π
1 2π
dt = .
0 1 + a2 − 2a cos t 1 − a2
Exercise IV.4.16.
(a) By considering a suitable complex integral, show that if 0 < r < 1,
Z π
cos nθ dθ πrn
2
= .
0 1 − 2r cos θ + r 1 − r2
(b) By considering a suitable complex integral, show that
Z 2π
cos(cos θ) cosh(sin θ) dθ = 2π.
0
For (a), you will have to use symmetry to get an integral from [0, 2π]. There are a
couple of ways to do the reverse engineering, and if you choose the right one, you
will save yourself work. For (b) we do not have a rational function of sin and cos.
The reverse-engineering idea still works, though, but may require a bit of trial and
error.
where a and b are real numbers. This one appears on workshop problem set 8 (Nov
27), so I will not discuss it in detail here. In that solution, we used the Cauchy Integral
Formula instead of the residue theorem.
The main thing to remember with these types of integrals is that the holomorphic
function to be used is not
cos az
(IV.4.3)
z 2 + b2
but
eiaz
(IV.4.4)
z 2 + b2
The reason is that whatever the sign of a, cos az blows up as Im(z) → ±∞. So if you try
to integrate (IV.4.3) around a semicircular contour, you will never get the contribution
from the curved part of the contour to go to zero as R → ∞. On the other hand, if
a > 0, eiaz is bounded in the upper half-plane and if a < 0, eiaz is bounded in the lower
half-plane. The function eiaz is also simpler than cos az or sin az!
Here is the general story of integrals of a trig function times a rational function:
90 IV. ZEROS, POLES, AND THE RESIDUE THEOREM
If f (z) is a rational function with no pole on the real axis and satisfying |f (z)| =
O(|z|−2 ) for |z| → ∞, then if a is a positive real number,
Z ∞
f (x)eiax dx = 2πi × sum of residues of f (z)eiaz in the upper half-space.
−∞
(IV.4.5)
If f is real on the real axis then the integrals of f (x) cos ax and f (x) sin ax can be
deduced by taking real and imaginary parts.
Note that f need not be a rational function as long as it has no pole on the real
axis and satisfies the given decay condition.
Exercise IV.4.17. Show that if a is real and positive and f is a rational function
as in the box, then
Z ∞
f (x)e−iax dx = −2πi × sum of residues of f (z)e−iaz in the lower half-space.
−∞
(IV.4.6)
Exercise IV.4.18. Let a be a positive real number. Calculate
Z ∞ iξx
e dx
x2 + a2
−∞
making sure your answer is valid for all real values of ξ. (This is the Fourier transform
of the function (x2 + a2 )−1 .)
We consider two extensions of the above technology. The first concerns functions
which do have a singularity on the real axis.
Example IV.4.2. Find the value of
Z ∞
1 − cos x
I= dx. (IV.4.7)
0 x2
Solution: Let
1 − cos x
u(x) = .
x2
Because 1 − cos x = x2 /2 + O(x4 ) for small x,
1
u(x) = + O(x2 ) for small x
2
so if we define u(0) = 1/2, the integrand becomes a continuous function near 0. Thus
despite the apparent singularity at x = 0, the integral
Z R
u(x) dx
0
4. DEFINITE INTEGRALS 91
Γδ
Notation IV.4.3. Here we wrote O(z 2 ) in (IV.4.10) for the sum of the higher
order terms in the expansion of eiz near z = 0. This is an extension of our previous
use of the ‘big Oh’ notation. It is convenient to allow ourselves to write O(z n ) to
mean a function of the form z n ψ(z), where ψ is holomorphic (hence bounded) in a
neighbourhood of z = 0.
The other direction in which we wish to extend (IV.4.5) is to weaken the decay
condition |f (z)| = O(|z|−2 ) for |z| → ∞. Let us note first that we can certainly relax
this condition to |f (z)| = O(|z|−a ) for a > 1 without any additional work. For if
|f (z)| 6 C|z|−a for all sufficiently large |z| (IV.4.14)
then the ML lemma gives that the integral of f along the large semicircular contour of
radius R is 6 2πR1−a and if a > 1 this still tends to 0 as R → ∞. This situation can
arise not for rational functions but for functions involving fractional powers
R ∞(or log).
−a
Note also that the condition |f (z)| = O(|z| ) with a > 1 also implies that 0 f (x) dx
is absolutely convergent.
The more delicate case is |f (z)| = O(|z|−1 ). So suppose that f (z) is a rational
function which is O(1/z) for large |z|. Even if f (z) has no poles on the real axis, it is
not clear that Z ∞ Z M
f (x)eix dx = lim f (x)eix dx (IV.4.15)
0 M →∞ 0
exists in this case. To see that such integrals do exist, let’s consider the simplest case,
f (x) = 1/x (but where we integrate from 1 to ∞ to avoid the singularity at x = 0—it is
the behaviour at ∞ that is the issue here).
Lemma IV.4.5. The integral
Z ∞ M
eix eix
Z
dx = lim dx (IV.4.16)
1 x M →∞ 1 x
is convergent.
4. DEFINITE INTEGRALS 93
provided that f has no poles on the real axis. Indeed, if f is such a rational function,
write
f (z) = a/z + g(z) where |g(z)| = O(|z|−2 ) for large |z|.
Then Z M Z 1 Z M ix Z M
ix ix e
f (x)e dx = f (x)e dx + a dx + g(x)eix dx.
0 0 1 x 1
The integral from 0 to 1 definitely exists and is independent of M . If we let M tend to
∞, then the third integral on the RHS is absolutely convergent because g(x) = O(x−2 )
for large x. The middle integral on the RHS converges by the Lemma. So
Z ∞ Z M
f (x)eix dx = lim f (x)eix dx
0 M →∞ 0
also exists.
Exercise IV.4.19. Use the same argument as in the proof of Lemma IV.4.5 to
prove the following generalization: if f is continuously differentiable on [0, ∞) and
satisfies the two conditions
f (x) → 0, f 0 (x) = O(x−2 ) as x → ∞ (IV.4.20)
then Z M
lim f (x)eix dx
M →∞ 0
exists. Can you weaken the hypotheses on f still further?
Suppose (to be definite) that f (z) is a rational function such that |f (z)|R = O(|z|−1 ) as
∞
|z| → ∞ and that f has no pole on the real axis. Then by the Proposition, −∞ f (x)eix dx
exists (as an improper integral) and it is reasonable to ask if integrating f (z)eiz around
the usual semicircular contour allows us to evaluate this integral. For this to work,
94 IV. ZEROS, POLES, AND THE RESIDUE THEOREM
we need to show that the contribution over the large semicircular arc goes to zero as
R → ∞.
This is achieved by Jordan’s Lemma
Proposition IV.4.6. Suppose that f (z) is continuous in the upper half-plane and
that |f (z)| → 0 for all z in the upper half-plane as |z| → ∞. Let γR be the contour
γR (t) = Reit , 0 6 t 6 π.
and let a > 0. Then Z
lim f (z)eiz dz = 0.
R→∞ γR
Given ε > 0, choose M as in (IV.4.21) and R > M . Then by the triangle inequality and
the given bound on |f |
Z Z π
iz
f (z)e dz 6 εR exp(−R sin t) dt (IV.4.22)
γR 0
(because sin t = sin(π − t)) and that sin t > 2t/π for 0 6 t 6 π/2. (This is the chord
which connects (0, 0) to (π/2, 1), and if you draw the graph, you’ll see that it is entirely
below sin t on the interval [0, π/2]. Hence
Z π Z π/2
π
exp(−R sin t) dt 6 2 exp(−2Rt/π) dt = (1 − e−2R )
0 0 R
Combining this with (IV.4.22) the factors of R cancel out, leaving us with
Z
eiz f (z) dz 6 πε.
γR
To sum up
4. DEFINITE INTEGRALS 95
Suppose that f (z) is a holomorphic function apart from a finite number of poles
in a neighbourhood of the closed upper half-plane and with no pole on the real axis.
Suppose that |f (z)| → 0 for z in the upper half plane as |z| → ∞. Then if a is a
positive real number,
Z ∞
f (x)eiax dx = 2πi × sum of residues of f (z)eiaz in the upper half-space.
−∞
(IV.4.23)
If f is real on the real axis then the integrals of f (x) cos ax and f (x) sin ax can be
deduced by taking real and imaginary parts.
The hypotheses apply in particular if f is a rational function p/q with deg q >
deg p.
R∞ R∞
Exercise IV.4.20. Calculate −∞ f (x) cos x dx and −∞ f (x) sin x dx for the fol-
lowing functions f :
1
(a) f (x) = x2 +2x+5 ;
1
(b) f (x) = (x2 +2x+5)2 ;
(c) f (x) = xx+1
2 +1 .
and define
z a = exp(a log z) = |z|a × exp(ia arg z)
for the choice of log and arg given in (IV.4.27) that if z = x+i0 with x > 0, then z a = xa
is real and positive, but if x < 0, we have x = |x|eπi so xa = |x|a eπia . So defined, f (z)
is holomorphic inside and on the indented semicircular contour Γδ . See Figure 2.
arg(z) = π/2
Γδ
arg(z) = π arg(z) = 0
0
and so
∞
xa dx πeπia/2
Z
π
2
= πia
= . (IV.4.33)
0 1+x e +1 2 cos(πa/2)
As a sanity check, the answer is positive for −1 < a < 1, which is good because the
function being integrated is everywhere positive. Also as a → ±1, the answer blows up,
which is consistent with the failure of convergence of the original integral if a 6 −1 or
a > 1.
In this example it was essential that the function multiplying xa was even. If this
had not been the case, we would not have been able to relate the contribution from the
interval [−R, −δ] to the contribution from [δ, R]. If we have to calculate
Z ∞
K= f (t)ta dt (IV.4.35)
0
To use this approach, z a or log z are defined with 0 < arg z < 2π, yielding a holo-
morphic branch that is holomorphic inside and on the contour.
As part of the limiting process, not only does the radius of the small circle go to zero
and that of the large circle to ∞, but also the angle between the two straight segments
and the positive x-axis has to be taken to zero as well.
On balance, the squaring trick has a lot to recommend it!
CHAPTER V
Proof. This follows from the residue theorem. Let F (z) = f 0 (z)/f (z). Then F is
holomorphic apart from at the poles and zeros of f . If a is a zero of order m, then we
have seen that we can write
f (z) = (z − a)m g(z), g(a) 6= 0
where g(z) is holomorphic in a small disc centred at a. Then
f 0 (z) = m(z − a)m−1 g(z) + (z − a)m g 0 (z)
and so
m g 0 (z)
F (z) = +
z−a g(z)
Thus F has a simple pole with residue equal to m at z = a. Similarly if b is a pole of f
order k, then a precisely similar calculation shows that F has a simple pole at z = b of
order −k.
99
100 V. THE ARGUMENT PRINCIPLE
Example V.2.2. Let a = 0, let γ(t) = eit , 0 6 t 6 2π, and let f (z) = z n . Then
Γ(t) = eint . This has the interpretation of the unit circle, traversed n times anticlockwise
if n > 0 and clockwise if n < 0. So the total change of argument ∆Γ arg(z) = 2πn in
this case, which is 2π times the number of times Γ(t) winds around 0.
Proposition V.2.3. With the above definition, n(Γ, a) is an integer.
Proof. For simplicity, suppose that Γ is smooth.
Define Z
dz
L(t) = (V.2.4)
Γt z
where Γt is the restriction of Γ to the interval [t0 , t]. Then
Z t 0
Γ (s) ds
L(t) = (V.2.5)
t0 Γ(s) − a
Recalling that log z = log |z| + i arg z, it follows that t 7→ Im L(t) is a continuous
choice of arg(Γ(t) − a) − arg(Γ(t0 ) − a) and so
L(t1 ) = i × total change in the argument of z − a as z goes around Γ. (V.2.12)
So
Z
1 dz
= ∆Γ arg(z − a) = total change in arg(z − a) along Γ. (V.2.13)
i Γ z−a
Remark V.2.5. Note that this proof is ‘quantitative’: we learn not only that p(z)
has n zeros in C, but that they are all in D(0, R), where R is any radius which makes
the above argument work. In fact, Rouché’s Theorem, proved in the next section, makes
this even more concrete.
3. Rouché’s Theorem
The essential idea of the proof we have just given of the FTA is generalized and
made more precise in Rouché’s Theorem:
Theorem V.3.1. Suppose that Ω ⊂ C is an open set, that f : Ω → C is holomorphic
and that D = D(z0 , R) is a disc such that D ∪ ∂D ⊂ Ω, as before. Suppose further that
g is holomorphic in Ω and
|f (z)| > |g(z)| for all z ∈ ∂D. (V.3.1)
Then f and f + g have the same number of zeros (counted with multiplicity) in D.
We give first a geometric proof:
Proof. Let h(z) = f (z) + g(z). Note first that the hypothesis (V.3.1) implies that
f has no zeros on ∂D (because of the inequality being strict). It also implies that h has
no zero, for
|h(z)| = |f (z) + g(z)| > |f (z)| − |g(z)| > 0
also by (V.3.1).
Let us write h = f + g and let γ : [a, b] → C be any piecewise smooth closed curve.
Let
Γ(t) = f (γ(t)), ∆(t) = h(γ(t)). (V.3.2)
If we assume that |f | > |g| on γ, then we have
|Γ(t) − ∆(t)| < |Γ(t)| (V.3.3)
for t ∈ [a, b]. From Figure 2, this means that ∆(t) is in the open disc centred at Γ(t),
with radius |Γ(t)| and so the angle at 0 between Γ(t) and ∆(t) lies in (−π/2, π/2).
Γ(t)
∆(t)
3.1. Analytic proof of Rouché. Let N be the total number of zeros of f and let
N 0 be the total number of zeros of f + g in D.
By the argument principle, Theorem V.1.1,
f 0 (z) f 0 (z) + g 0 (z)
Z Z
1 0 1
N= dz, N = dz. (V.3.8)
2πi ∂D f (z) 2πi ∂D f (z) + g(z)
Now
f 0 + g0 f 0 f g 0 − gf 0 d
− = = log(1 + g/f ). (V.3.9)
f +g f f (f + g) dz
The point is that the hypothesis of Rouché means that Re(1 + g(z)/f (z)) > 0 for z ∈ ∂D
and so log(1+g/f ) can be defined uniquely by taking the argument to be in (−π/2, π/2).
By the complex FTC, Z
d
log(1 + g/f ) dz = 0 (V.3.10)
∂D dz
and this completes our second proof of Rouché.
Exercise V.3.1. Let C be the unit circle |z| = 1 traversed anticlockwise. De-
termine the variation of the argument ∆C arg f (z) for the functions
z3 + 2
(a) f (z) = z 2 , (b) f (z) =.
z
Exercise V.3.2. Let λ be real and λ > 1, Show that the equation
zeλ−z = 1
has exactly one solution in the disc |z| = 1, which is real and positive.
Exercise V.3.3. (a) How many roots does the polynomial f (z) = z 4 −2z−2
have inside the annulus
1 3
< |z| < ?
2 2
Explain your answer.
(b) Find the number of the roots of the equation
z 6 − 5z 4 + 8z − 1 = 0
in the annulus {z : 1 < |z| < 2}.
(c) How many roots does the polynomial z 6 −5z 4 +8z −1 have inside the square
with vertices ±1/2 ± i1/2?
Exercise V.3.4. Suppose that p(z) = a0 + a1 z + · · · + an−1 z n−1 + z n is a
polynomial with |aj | 6 1 for all j. Show that all zeros of p(z) lie in D(0, 2).
CHAPTER VI
Green’s Theorem
1An essay on the application of mathematical analysis to the theories of electricity and magnetism
107
108 VI. GREEN’S THEOREM
As in the one-variable case, R being compact (closed and bounded) implies that f is
actually uniformly continuous in R, so given ε > 0, the same δ > 0 will guaranteee
(VI.1.2) for all z0 ∈ R. R
It is pretty plausible that one can make a definition of R f (x, y) dxdy in terms of
2D partitions of R: that is, we cover R by a grid of tiny rectangles of the form
Rjk = {sj < x < sj+1 , tk < y < tk+1 }
where
a = s0 < · · · < sN = b, c = t0 < · · · < tM = d.
and try to define the integral as the limit of sums of the form
X
fjk (sj+1 − sj )(tk+1 − tk ) (VI.1.3)
j,k
where we take
fjk = sup f or fjk = inf f
Rjk Rjk
for the ‘upper’ and ‘lower’ sums respectively. It is not hard to adapt the one-variable
proof to prove that if f is continuous then the limit over finer and finer partitions of R
of both upper and
R lower sums exist and that they are equal. Thus such a limit serves as
a definition of R f (x, y) dxdy.
From this definition one can prove various properties, in particular that the integral
over R of f can be computed by integration first with respect to x and then with respect
to y.
Proposition VI.1.1. If f is continuous on the closed rectangle R of (VI.1.1), then
Z d
F (x) := f (x, y) dy (VI.1.4)
c
is continuous for x ∈ [a, b] and
Z b
G(y) := f (x, y) dx (VI.1.5)
a
is continuous for y ∈ [c, d]. Moreover,
Z Z b Z d
f (x, y) dxdy = F (x) dx = G(y) dy. (VI.1.6)
R a c
One can also prove versions of the fundamental theorem of calculus, such as
Theorem VI.1.2. Let f (x, y) be continuous in R. Then if we define
Z x
F (x, y) = f (t, y) dt,
a
the partial derivative Fx exists in R and is equal to f . Conversely, if g is continuous
and gx exists and is also continuous, then
Z b
g(b, y) − g(a, y) = gx (x, y) dx.
a
There are analogous statements with the roles of x and y switched.
2. GREEN’S THEOREM FOR A REGION BOUNDED BY A GRAPH 109
With all this in hand, we can prove Green’s Theorem for a rectangle.
Theorem VI.1.3. Let R be our closed rectangle (VI.1.1) and let P and Q be two C 1
functions in R. Then
Z Z
(Qx − Py ) dxdy = P dx + Q dy. (VI.1.7)
R ∂R
Here, as always, ∂R is traversed anticlockwise.
Proof. It is clearly enough to prove the separate identities
Z Z Z Z
Qx dxdy = Q dy, Py dxdy = − P dx. (VI.1.8)
R ∂R R ∂R
Consider the first of these. By the results above, we can calculate the LHS by doing the
integral with respect to x first, then the integral with respect to y. For the first integral,
we can use the FTC, giving
Z Z d
Qx dxdy = (Q(b, y) − Q(a, y)) dy (VI.1.9)
R y=c
On the other hand, the RHS of the first of (VI.1.8) is the sum of the integrals along the
line segments from (a, c) to (b, c), from (b, c) to (b, d), from (b, d) to (a, d) and finally
from (a, d) to (a, c). Only the contibutions along the vertical segments are non-zero
for the integral of Qdy (for dy = 0 along the horizontal segments) and the sum of the
integrals along the segments is precisely the RHS of (VI.1.9). This proves the first of
(VI.1.8). The second is proved in precisely the same way, switching the roles of x and y
(and in particular computing the LHS by integrating first with respect to y).
(b, φ(b))
(a, φ(a))
D
(a, c) (b, c)
If φ(a) > 0, our region D is bounded on the left-hand side by the segment from (a, c)
to (a, φ(a)). If φ(b) > 0, then it is bounded on the right-hand side by the segment from
110 VI. GREEN’S THEOREM
(b, c) to (b, φ(b)). Either or both of these segments may collapse to points if φ(a) = 0 or
φ(b) = 0. So D is bounded by one, two, or three straight-line segments and the graph
of y = φ(x). R
One can prove that there is a good definition (via partitions) of D f (x, y) dx for
such domains D, and moreover,
Z Z b Z φ(x) !
f (x, y) dx = f (x, y) dy dx. (VI.2.2)
D x=a y=c
Proof. Consider terms in P and Q separately. The P -term is the easier. Starting
from (VI.2.2), putting f = Py , we have
Z Z b
− Py dxdy = (−P (x, φ(x)) dx + P (x, c)) dx. (VI.2.4)
D x=a
Now we turn to the integral over ∂D of P dx. The integrals over the vertical sides (if
they are present) of ∂D vanish because dx = 0 along any vertical line. Thus the integral
over ∂D is reduced to the integral along the segment from (a, c) to (b, c), together with
the integral along the upper, curved part of the boundary. The integral along the bottom
segment from (a, c) to (b, c) is
Z b
P (x, c) dx. (VI.2.5)
a
The integral along the boundary y = φ(x) is computed via the parameterization
γ(x) = (x, φ(x)) for a 6 x 6 b (VI.2.6)
and changing the sign, since this curve is traversed from right to left! Thus the contri-
bution from this curve is Z b
− P (x, φ(x))dx. (VI.2.7)
a
Combining (VI.2.4), (VI.2.5) and(VI.2.7), we obtain the Green’s theorem for D in the
case that Q = 0.
Now let’s do the case that P = 0. It is convenient to introduce a function Q(x,
e y)
whose derivative with respect to y is equal to Q. By the FTC, we may take
Z y
Q(x, y) =
e Q(x, t) dt. (VI.2.8)
c
Then Q
e y = Q, Q
e xy = Qx and we have
Z Z Z b
Qx dxdy = Qxy dxdy =
e e x (x, φ(x)) − Q(x,
(Q e c)) dx (VI.2.9)
D D x=a
2. GREEN’S THEOREM FOR A REGION BOUNDED BY A GRAPH 111
doing the integral with respect to y first. Now by the FTC yet again, we have
Z b
− Q(x, e c) − Q(b,
e c) dx = Q(a, e c). (VI.2.10)
a
The other term, Qe x (x, φ(x)) is trickier. We’d like it to be the derivative with respect to
x of something, so that we can use the FTC to simplify the integral. Now
d e e y (x, φ(x))φ0 (x) = Qe x (x, φ(x)) + Q(x, φ(x))φ0 (x).
Q(x, φ(x)) = Q e x (x, φ(x)) + Q
dx
(VI.2.11)
Hence
Z b Z b
Qx (x, φ(x)) dx = Q(b, φ(b)) − Q(a, φ(a)) −
e e e Q(x, φ(x))φ0 (x) dx. (VI.2.12)
a a
Combining, we obtain
Z
e c) − Q(a,
Qx (x, y) dxdy = Q(a, e φ(a)) (VI.2.13)
D
e φ(b)) − Q(a,
+ Q(b, e c) (VI.2.14)
Z b
− Q(x, φ(x))φ0 (x) dx. (VI.2.15)
a
By the FTC, (VI.2.13) is equal to the integral of Q dy along the left-hand edge, the
segment joining (a, φ(a)) to (a, c). Similarly, (VI.2.14) is the integral of Q dy along the
right-hand vertical segment and (VI.2.15) is the integral of Q dy along the edge y = φ(x),
traversed from right to left. Since the integral of Q dy along the lower horizontal edge
is zero, (VI.2.13–VI.2.15) are equivalent to
Z Z
Qx (x, y) dxdy = Q(x, y) dy.
D ∂D
A precisely similar result is true for regions bounded below by a graph, that is
D = D = {z = x + iy : a 6 x 6 b, φ(x) 6 y 6 c} (VI.2.16)
(where φ(x) 6 c for x ∈ (a, b)) and for regions bounded on the right or left by a graph:
D = {z = x + iy : a 6 x 6 ψ(y), c 6 y 6 d} or (VI.2.17)
D = {z = x + iy : ψ(y) 6 x 6 b, c 6 y 6 d} (VI.2.18)
It shows two regions with a common vertical edge. Let the one on the left be D1 ,
the one on the right D2 . Each of D1 and D2 is a region bounded by a graph, so Green’s
Theorem holds for them. Write
∂D1 = γ + ∂D10 , ∂D2 = −γ + ∂D20 .
Here γ is the vertical segment, traversed upwards; ∂D10 and ∂D20 denote the ‘rest’ of the
boundaries respectively of D1 and D2 .. Let D = D1 ∪ D2 . Then clearly
Z Z Z
(Qx − Py ) dxdy = (Qx − Py ) dxdy + (Qx − Py ) dxdy.
D D1 D2
Furthermore,
Z Z Z
P dx + Q dy = P dx + Q dy + P dx + Q dy
∂D1 ∂D10 γ
and Z Z Z
P dx + Q dy = P dx + Q dy − P dx + Q dy
∂D2 ∂D20 γ
Thus Z Z Z
P dx + Q dy = P dx + Q dy + P dx + Q dy.
∂(D1 ∪D2 ) ∂D1 ∂D2
It follows that Green’s Theorem holds for the larger domain D = D1 ∪D2 . This argument
is simple, but the key point is that the common part of the boundary is traversed once
in each direction, and the contributions to the boundary integral cancel.
Now cover the entire complex plane by a grid of squares with side δ > 0 and sides
parallel to the x- and y-axes. (The grid can be defined by the vertical lines x = x0 + mδ
and the horizontal lines y = y0 + nδ, where (x0 , y0 ) is some given point and m and n
run over the integers.) See Figure 3.
The squares are supposed to be so small that for any square S, either the interior
of S does not meet the boundary at all, or it does so in a simple arc whose tangent
doesn’t vary too much over its intersection with S. Examples of the types of possible
intersections of D with a square are shown in Figure 4.
3. GREEN’S THEOREM FOR BOUNDED DOMAINS WITH SMOOTH BOUNDARY 113
A rigorous proof that this is possible would use fact that the tangent vector is
continuously varying (and never zero). Combined with the fact that the boundary is
compact, this means that we can choose the squares so small that the tangent vector
varies only by, say, 10◦ over the square. If the square meets the boundary in two or more
pieces, it is pretty plausible that by repeated subdivision, we can reduce to the case that
the boundary meets each square in a single arc. A key point is the compactness of the
boundary, which implies that it meets at most finitely many squares.
Granted that this is possible, in Figure 3, Green’s Green’s Theorem is true for each
of the pale grey squares which meet ∂D and for all the dark grey squares which are
interior to D. Thus D is a finite union of abutting regions, each of which is either a
square or a region bounded by a graph. Since Green’s Theorem is true for such regions,
and we have seen that if it is true separately on abutting regions, it is also true for the
union, it follows that it is true for D itself.
114 VI. GREEN’S THEOREM
Figure 5. Corners
Essentially the same argument can be used to prove Green’s Theorem for bounded
domains with piecewise smooth boundary as well.
Suppose that the non-smooth points of ∂D are z1 , . . . , zN . Cover D with a grid of
rectangles as before, by drawing a large number of horizontal and vertical lines, including
the lines passing through the zj ,
Re(z) = Re(zj ), Im(z) = Im(zj ).
Once again we make this mesh so small that when the interior of a rectangle meets
the boundary, it does so as the graph of a function. Since we have arranged for each
corner to be on the intersection of two grid-lines, near each corner we shall see figures
like the ones shown in Figure 5. In the example on the left in Figure 5, there are two
squares, each of which contains one boundary arc. And we already know that Green’s
Theorem applies to such domains. Thus if this is the only thing that ever happens with
the corners, the previous argument goes through as before. Unfortunately, the picture
on the right can also happen, where the two boundary arcs lie in a single rectangle. We
do not yet know Green’s Theorem for regions with the shape of the one on the right.
However, denote by S the top-right square and by W1 and W2 the two white parts
of the square. Denote by G the region shaded grey. Then
S = G ∪ W1 ∪ W2 and ∂G = ∂S − ∂W1 − ∂W2 . (VI.4.1)
And the point is that we know Green’s Theorem for S and for W1 and W2 because W1
and W2 are regions bounded by a graph. So if P and Q are defined over the whole of
S, we obtain Green’s Theorem for G from Green’s Theorem for G, W1 and W2 by use
of (VI.4.1) and subtraction! In this argument, we needed to assume that P and Q are
defined and C 1 over the whole square. This is OK, because the assumption about P
and Q is that they are defined and C 1 in a slightly larger open set Ω containing D and
its boundary. Thus we can suppose that the squares in the above pictures are so small
that the closure of the top right square is contained in Ω, so that P and Q are defined
there.
Thus the idea of covering a domain with piecewise smooth boundary by very small
rectangles and obtaining Green’s Theorem for the complicated domain by summing over
the small rectangles can be made to work in this case as well.
This completes our discusssion of Green’s Theorem in the plane.