You are on page 1of 132

BVR/CMI Complex Analysis 01-02-2022

In the social and cultural milieu we all live now, there are too many issues
to overcome in order to learn.
We seem to be in the age of internet surfing and if your page does not open
in a jiffy, you get impatient and start cursing. If readymade recipe/software
is not available you get frustrated.
Think about it.
In our course some concepts may take several minutes/hours to reveal them-
selves. There are no readymade things, you need to take pen and paper and
work your way through.
We seem to be compelled by social media (whatsapp/facebook etc) to
quickly react without having to think. Many programs instil in you a sense
of dichotomy: I know, I answer or I do not know, I can not answer.
Think about it.
No need to respond quickly to anything. There is a third possibility: I may
not know, but I can think and answer. I am as good as any one else.
Society at large seems to be ‘majoritarian society’ and a majority can
force anything they like. Facts have no relevance, opinions seem to rule –
Unfortunate.
Think about it.
You need to be disciplined, obey the rules set for you and follow logical
reasoning in your arguments. In final analysis do not assume hypotheses not
granted to you.
In society at large discussions are discouraged and questioning is taboo,
even regarded as serious punishable offences – a degrading and demeaning
path the country is taking.
Think about it.
Questioning and discussions are the only way to understand matters. Listen
carefully; read the material; question if not convinced, then think about it
and assimilate.

Obeying rules does not mean you should curtail your imagination. You
should not. Harish-Chandra says: ‘I have often pondered over the roles of
knowledge or experience on the one hand and imagination or intuition on the
other, in the process of discovery. I believe that there is a certain fundamental
conflict between the two; and knowledge, by advocating caution, tends to inhibit
the flight of imagination. Therefore a certain naiveté, unburdened by conven-
tional wisdom, can sometimes be a positive asset.’

1
1. Alap

We first learnt natural numbers

N = {0, 1, 2, 3, . . .}.

Then we invented negative integers (may be in connection with borrowing,


or accounting etc, but now we say: to make a group)

Z = {. . . , −2, −1, 0, +1, +2, . . .}

Then we invented fractions, rational numbers (may be when we wanted to


divide cake or a piece of land, but now we say: to make it a field)
a
Q = { : a, b ∈ Z, b > 0}.
b
Now there are two thought processes. First is geometric/algebraic: We
realized that there is no way to measure hypotenuse of a right angled triangle
with side lengths one. So we introduced square roots, cube roots etc. Second
path is geometric/analytic: We realized that there are sequences in Q whose
terms are getting closer and closer but there is nothing to which they are
getting close to! In other words many Cauchy sequences were not converging.
So we completed and got R, the real number system.
Let us return to the first path. We saw that there is no rational number
whose square is 2. So attach a new object called θ to Q, with the purpose of
having θ2 = 2 (and θ > 0). Make new additive group Q∗ = {a+θb : a, b ∈ Q}.
This is a group.

(a + θb) + (c + θd) = (a + c) + θ(b + d)

We can also multiply – usual multiplication with the understanding θ2 = 2

(a + θb)(c + θd) = (ac + 2bd) + θ(bc + ad)

again a number in Q∗ . But the important thing is:


This is already a field, has inverses too!. (♠)
In fact for a non-zero number
1 a −b
= 2 2
+θ 2 .
a + θb a − 2b a − 2b2
we shall write 3+θ(−4) as 3−θ4. The above expression makes sense because
a2 6= 2b2 . Unfortunately

2
√ √
Q∗ does not have 3 2 or 7 (♦).
That is Q∗ has no element x such that x3 = 2 and it has no element y
such that y 2 = 7. In other words other square roots and even cube root of 2
are not in Q∗ .
suppose there is a number in Q∗ whose cube root equals 2, say (a+θb)3 = 2
with a, b ∈ Q. Note that b 6= 0 because then a3 = 2 and we know there is
no such rational number. Similarly a 6= 0 because then 2b3 θ = 2 and θ ∈ Q
which is not true. Thus a3 + 6ab2 + (2b3 + 3a2 b)θ = 2. If (2b3 + 3a2 b) = 0,
that is (2b2 + 3a2 ) = 0 or a = 0 = b which is not the case. If (2b3 + 3a2 b) 6= 0,
then θ = (2 − a3 − 6ab2 )/(2b3 + 3a2 b) ∈ Q which is again not possible.
Similarly, if (a + θb)2 = 7, then you first argue that a 6= 0, b 6= 0 and then
θ = (7 − a2 − 2b2 )/(2ab) ∈ Q to show the impossibility.

Also unfortunately
Given c ∈ Q∗ , there may not be a z ∈ Q∗ with z 2 = c (♥)

In other words Q is not closed under square roots of its own elements.
For example, it is impossible to have (a + θb)2 = θ where a, b ∈ Q. In
other words a2 + 2b2 + 2abθ = θ. If 2ab = 1, then this equation implies
that a2 + 2b2 = 0 or a = 0 = b which is not possible. If 2ab 6= 1 then, this
equation implies θ = (a2 + 2b2 )/(1 − 2ab) a rational number which is again
not possible.
Since we knew that x2 ≥ 0 for any x ∈ Q, we were adding only square
roots of positive nunbers.
Now let us return to R. Of course for every x ∈ R we still have x2 ≥ 0.
Thus there is no θ such that θ2 = −1. Let us imagine that indeed there is an
object called square root of negative one. Let us apply the same technique
as above and add an object θ with the understanding θ2 = −1 and make the
additive group as above: C = {a + θb : a, b ∈ R}. As above
(a + θb) + (c + θd) = (a + c) + θ(b + d).
Again as above you can multiply too – usual multiplication with the under-
standing θ2 = −1 so that
(a + θb)(c + θd) = ac − bd + θ(ad + bc).
Again as above this is a field: For non-zero number,
1 a −b
= 2 2
+θ 2
a + θb a +b a + b2

Thus conceptually
√ there is no difference between adding −1 to R and
adding 2 for Q. But this is just superficial!

3
MAGIC: Unlike the previous case, with the introduction of just one ‘imag-
ined number’ square root of negative one to the real number system, we have
any root of any number, even for the new numbers there are square roots,
tenth roots etc!
The Panaroma of complex analysis does not stop here.
If f : C → C, is differentiable once then it differentiable as many times as
you want, that is, differentiability implies C ∞ . Remember we have functions
f : R → R differentiable just as many times as you want and no more.

If f : C → C and g : C → C are differentiable and if they agree on a


tiny line segment then they agree everywhere! Remember We have functions
f : R → R and g : R → R which are both C ∞ , distinct but agree on any
preassigned large (proper) subinterval of R.

Integration over cantours is, many times, reduced to just calculating cer-
tain quantities called ‘residues’. This cantour integration will help evaluate
some real integrals too which are otherwise difficult.

We shall see some of these soon. C has rich structure and plays funda-
mental role in Physics, Group representations and so on. It is customary
to denote by i what we denoted above by θ. Thus complex numbers are of
the form a + ib where a, b are real numbers. i, stands for imaginary(?). The
number a is called the real part and b is called the imaginary part of the
complex number z = a + ib, notation <(z) and =(z) or some times simply
<z and =z. (remember, both real part and imaginary part are real numbers).

Before formally starting the complex number system a little history. You
should not get the wrong impression that some one sat down and imagined
adding square root of negative one. Complex numbers entered naturally and
very quietly!

1: We had the real number system R. People were solving equations.


Consider quadratic x2 + bx + c = 0. it is easy to √ show that it has NO root if
2 2 −b± b2 −4c
b − 4c < 0. When b − 4c ≥ 0 it has roots 2
. What about cubic?
Again there is no loss in taking leading coefficient to be one. y 3 +by 2 +cy+d =
0. If you put y = x − 3b , then this equation reduces to: x3 = 3px + 2q where
p, q ∈ R. del Ferro found (as stated by Cardano in 1545) the solution:
√ √ p
x = 3 q + w + 3 q − w; w = q 2 − p3
interestingly enough if q 2 < p3 then w is a question mark. But you can
blindly manipulate and show that x indeed does satisfy the equation:

4
x3 = (q + w) + 3(q + w)2/3 (q − w)1/3 + 3(q + w)1/3 (q − w)2/3 + (q − w)

= 2q + 3(q + w)1/3 (q − w)1/3 [x] = 2q + x(q 2 − w2 )1/3 = 2q + 3px


Bombelli (1547) did something
√ interesting. Consider x3 = 15x + 4. Thus
p = 5, q = 2 so that w = −121
√ √
q q
3 3
x = 2 + 11 −1 + 2 − 11 −1

He Wondered if for some number c;


√ √ √ √
q q
3 3
2 + 11 −1 = 2 + c −1; 2 − 11 −1 = 2 − c −1

so that you get the expected solution x = 4, the imaginary parts cancel. he
discovered √ √
(2 + i)3 = 2 + 11 −1; (2 − i)3 = 2 − 11 i
In other words you do get the ‘real’ solution passing through complex num-
bers!

2: d’Alembert was studying the theory of fluids in 1746. He ended up with


the following problem: When are both M dx+N dy and N dx−M dy complete
differentials? In other words, we have two functions M (x, y) and N (x, y)
given to us and we want the stated expressions to be complete differemntials.
This means we want two functions u(x, y) and v(x, y) such that

∂ ∂
du(x, y) = M dx + N dy i.e. u = M; u=N
∂x ∂y
∂ ∂
dv(x, y) = N dx − M dy i.e. v = N; u = −M
∂x ∂y

He arrived at Cauchy Riemann equations, actually, he did use −1 and gave
expressions for u and v.

3: Power series were understood (afterall they are infinite degree poly-
nomials!). We have power series for the exponential function: et . Euler
(1707-1783) discovered complex exponential series. For an ordinary differ-
ential equations with constant coefficients, you pretend that erx (where r is
a number) is a solution of your equation. You end up with a polynomial
equation to be satisfied by r. Even if a root r is complex, the function erx

5
is indeed a solution for the differential equation. Abraham De’Moivre (1667-
1754) (possibly Euler too) discovered: eit = cos t + i sin t for real numbers t.

How do you represent complex nunbers? You represent as points in the


plane. Norwegian Surveyor and Cartographer Casper Wessel in 1797; French-
Swiss Mathematician Jean Robert Argand in 1806, and Karl Frederich Gauss
in 1811 (implicitly used in 1799) discovered this representation. In honour of
Gauss, numbers of the form a + ib where a, b are integers are called Gaussian
integers.

There are too many good books:


Ahlfors (complex analysis);
Nevanlinna and Paatero (introduction to complex Analysis);
Hille (analytic function theory);
Alpay (a complex analysis problem book);
Elias Stein and Rami Sakarchi (complex Analysis),
and many more. We cover only basics. The only way to learn is to do it
yourself, so try the exercises. (I shall also try!)

Shall have midsem (40 marks) and final (60 marks).

2 complex numbers

Method 1: C as symbols x + iy
C = {x + iy, x, y ∈ R}

Method 2: C as Plane R2
point z = (x, y) is identified with z = x + iy.

Method 3: C as quotient field R(t)/[t2 + 1]


R[t] = polynomials (with real coefficients) in one variable t
I = ideal generated by t2 + 1
C = R[t]/I, the quotient {x + yt, x, y ∈ R.} identified with x + iy.

Points in C denoted by z, w etc.


We use method 1. Easy to translate to other methods
z = x + iy then x = <z and y = =z Real part and Imaginary part.

s Imaginary part IS a real number.

(a + ib) + (c + id) = (a + c) + i(b + d)

6
(a + ib)(c + id) = (ac − bd) + i(ad + bc).

1 a − ib
= 2
a + ib a + b2
write x + i0 as simply x
Say this complex number x + i0 is the real number x.
Thus identify R: {a + ib : b = 0} ⊂ C.
z ≥ 0 Means z is real and non-negative.

s do NOT compare two complex numbers unless both are real.

The reason I stress the above fact is the following: you may intuitively
feel 4 + 3i < 1 + 3i because, after all I can cancel 3i from both sides and
indeed 4 < 1! NO, you can NOT. There is no order among complex numbers
that satisfies reasonable conditions.
write 0 + iy as simply iy
Say this complex number is purely imaginary.
The real and imaginary parts are x-axis and y-axis in planar representa-
tion.
z = x + iy DEFINE p
conjugate z = x − iy. modulus, |z| = + x2 + y 2 .

7
BVR/CMI Complex Analysis 02-02-2022

TA:
Aadrita Laha
Aniruddhan Ganesaraman
have kindly agreed.
*****************

Field Automorphisms of C: Have to think how many.



(2ℵ0 ↔ 22 0 ?)
subject to continuity two: identity/conjugation.

Using one goes to one etc, you see real rational goes to itself. Now where
does i go?
if it goes to θ, then θ2 = −1 so that θ = ±i. Thus on complex rationals
(that is x + iy, x, y real rationals) you see your map is one of the two: z 7→ z
or z 7→ z. if continuity is granted then you see that on all of C it must be
identity or conjugation.
Without continuity? using transcendence degree, one can manufacture
many, but I have to think how many.
****************

z = x + iy DEFINE p
conjugate z = x − iy. modulus, |z| = + x2 + y 2 .

We observe a large number of easy/simple facts, some you probably know.


This is just to make us comfortable with complex numbers.

FACTS: z + w = z + w and zw = z w and z = z

z = (x, y), w = (u, v), then


xu − yv + i(xv + yu) = (x − iy)(u − iv)?

z is real iff z = z = <z;

z is purely imaginary iff z = −z = −i=z


zero is real/purely imaginary! It is the only such number.

1 z
<z = (z + z)/2 and =z = (z − z)/2i; and z
= |z|2
.

8
|z| = |z|; |z|2 = zz; |zw| = |z||w| |1/z| = 1/|z|.

In particular, |z n | = |z|n . And |z/w| = |z|/|w|.

When is |z| = z? iff z ≥ 0 That is


iff z is real and z ≥ 0.

When is |z| = <z? iff z ≥ 0

|z| ≥ 0; |z| = 0 iff z = 0; |z + w| ≤ |z| + |w|;

This last is ‘triangle inequality’.

usual proof: z = (x, y), w = (u, v) then reduces to


p
(x + u)2 + (y + v)2 ≤ (x2 + y 2 ) + (u2 + v 2 ) + 2 (x2 + y 2 )(u2 + v 2 )?

Use Cauchy-Schwarz? OR without expanding into x, y, u, v etc


problem reduces to <(zw) ≤ |zw| = |z||w|?

When is |z + w| = |z| + |w| iff either w = 0 or wz ≥ 0


Asume w 6= 0. If wz = c ≥ 0
|z + w| = (1 + c)|w| = |w| + |cw| = |z| + |w|.
Converse
If |z + w| = |z| + |w|; then (z + w)(z + w) = (|z| + |w|)2
<(zw) = |z||w| = |zw| So zw ≥ 0
2z z
|w| w ≥ 0 So w ≥ 0

A useful fact
If P (z) is a polynomial with real coefficients,
then P (z) = P (z).
roots of P occur in conjugate pairs.

|z + w|2 = |z|2 + |w|2 + 2<(zw) |z − w|2 = |z|2 + |w|2 − 2<(zw)

|z + w|2 + |z − w|2 = 2(|z|2 + |w|2 ), Parallelogram equality.

−|z| ≤ <(z) ≤ |z| and −|z| ≤ =(z) ≤ |z|

|z − w| ≥ | |z| − |w| |.

9
This follows from |z| ≤ |z − w| + |w| and |w| ≤ |w − z| + |z|

z−w
If either |z| = 1 or |w| = 1, then | 1−zw | = 1.
For ex. |w| = 1 implies
|z − w| = |z − w||w| = |zw − ww| = |1 − zw|
OR
|z − w|2 = (z − w)(z − w) = zz − wz − zw + 1

= 1 + zw zw − zw − zw = (1 − zw)(1 − zw)

z−w
If |z| < 1 and |w| < 1, then | 1−zw | < 1. because

zz + ww − zw − zw < 1 + zz ww − zw − zw?

What is the imnportance of the above two? If you fix w in the unit disc,
z−w
that is, with |w| < 1, then the map z 7→ 1−zw takes unit disc to tself. Further
when z approaches the boundary (that is |z| = 1) then the values approach
boundary. Theseare nice maps ans shall see later again.

Lagrange identity:
X X X X
| zj wj |2 = ( |zj |2 )( |wj |2 ) − |zj wk − zk wj |2
1≤j<k≤n

First
Pterm of RHSP minus LHS equals
zj zj wk wk − zj wj zk wk
j,k P j,k
= [zj zj wk wk + zk zk wj wj − zj wj zk wk − zk wk zj wj ]
1≤j<k≤n
P
= [zj wk − zk wj ][zj wk − zk wj ]
1≤j<k≤n

zk wk |2 ≤ ( |zk |2 )( |wk |2 )
P P P
Cauchy’s Inequality: |

Immediate from above. Another proof: for any complex λ

|zj − λwj |2 = |zj |2 + |λ|2 |wj |2 − 2<(λ


P P P P
zj wj ) ≥ 0
P
z w
Take λ = P j j2 .
|wj |

z w |2 z w |2
P P
| |
|zj |2 +
P
P j j2 − 2< P j j2 ≥0
|wj | |wj |

10
In the planar representation:

Addition= vector space addition, Multiplication:

(x, y) · (u, v) = (xu − yv, xv + yu)

R corresponds to the x-axis.


Purely imaginary numbers: y-axis.

The point (0, 1) represents i. √


From 1 go to −1 with 1800 rotation. Reasonable to call 900 rotation −1
because it applied twice takes to −1.

(x, y) = (x, −y) Reflection in x-axis.


p
|(x, y)| = x2 + y 2 , Euclidean distance from origin.

Why Triangle inequality called so? Consider Triangle O, z, (−w), distance


of the side (z → −w) is at most of the distances of the other two sides:
(O → z) (O → −w).
Why parallelogram law or parallelogram equality called so?.
Consider the parallelogram: O, z, z + w, w (Imagine all in first quadrant
for nice picture) has sides (0z), (z, z + w), (z + w, w), (w, O). (O, z + w) is
a diagonal and other diagonal joins (z, w) is length z −w. In a parallelogram:

sum of squares of sides = sum of squares of diagonals.

See the parallelogram ABCD next page


ABCD : AB 2 + BC 2 − 2AB · BC cos B = AC 2 (Law of cosines)
AB 2 + DA2 − 2AB · DA cos A = BD2
Use cos A = cos(180 − B) = − cos B and add.

If you do not know law of cosines


See the triangle ABC (of the parallelogram next page)
Triangle ABC: Draw AD ⊥ BC. cos B = BD/AB, sin B = AD/AB
AC 2 = AD2 + DC 2 = (AB sin B)2 + (BC − AB cos B)2 .

multiplication has interpretation? Yes, wait.

11
BVR/CMI Complex Analysis 04-02-2022

Method 4: C as plane with ‘polar coordinates’

C = {reiθ : r > 0, 0 ≤ θ < 2π} ∪ {0}


= {(r cos θ, r sin θ) : r > 0, 0 ≤ θ < 2π} ∪ {0}
θ is polar angle or argument or amplitude of the point z.
r is modulus: |z|
Suited for products. modulus multiplies/argument adds up.

z = reiθ , w = seiψ zw = rsei(θ+ψ) .


[Shall do exponenetial function soon. WAIT]
Take: θ + ψ modulo 2π.

z = reiθ and n ≥ 2. Then n-th roots:


 
θ j
√ i + 2π
zj = n r e n n 0≤j ≤n−1
We Shall return to roots soon.
Shall see:
polynomial P (z) of degree n has n roots (may be repeated) and
P (z) = a0 (z − z1 )(z − z2 ) · · · (z − zn )

Method 5: C as 2 × 2 matrices

  
x −y
C= : x, y ∈ R
y x
Above matrix identified with z = x + iy.
matrix addition/multiplication corresponds to complex addition and mul-
tiplication.
    
x −y u −v xu − yv −xv − yu
=
y x v u xv + yu xu − yv
 
x 0
R corresponds to diagonal matrices:
    0 x
1 0 0 −1
1: i:
0 1 1 0

13
    
0 −1 0 −1 −1 0
Note = = −I.
1 0 1 0 0 −1
conjugate z corresponds to changing signs off-diagonal.
 
2 x −y
|z| is determinant: det = x2 + y 2
y x
complex numbers of modulus one are rotation matrices:
 
cos θ − sin θ
sin θ cos θ
0 ≤ θ < 2π (or −π ≤ θ < π)

Method 6: C as Riemann Sphere, Stereographic projection.

Very useful representation:


C = {(x1 , x2 , x3 ) : x21 + x22 + x23 = 1, x1 , x2 , x3 ∈ R} \ {(0, 0, 1)} = S1
(Actually S1 is the entire surface of the sphere including the north pole
(0, 0, 1) too.)
Warning: NOT x, y, z-axes; but x1 , x2 , x3 -axes
{x3 = 0} identified with complex plane: (x, y, 0) ↔ z = x + iy
x1
Line: (x1 , x2 , x3 ), (0, 0, 1) meets horizontal plane at (x, y, 0) = ( 1−x , x2 , 0).
3 1−x3

x1 + ix2
(x1 , x2 , x3 ) ←→ z =
1 − x3
line joining (a1 , a2 , a3 ) and (b1 , b2 , b3 ) (running coordinates (u, v, w))
u − a1 v − a2 w − a3
= =
b 1 − a1 b 2 − a2 b 3 − a3
In our case (a1 , a2 , a3 ) = (0, 0, 1) (b1 , b2 , b3 ) = (x1 , x2 , x3 )
u v w−1
= =
x1 x2 x3 − 1
If the line cuts horizontal plane at (x, y, 0), then
x y −1
= =
x1 x2 x3 − 1
x1 x2
x= y=
1 − x3 1 − x3
So
x1 + ix2
z=
1 − x3

14
Thus given (x1 , x2 , x3 ) ∈ S1 (except north pole) we produced z. We shall
now explain how to recover the point from z. In other words the correspon-
dence is one-one.

x21 + x22 1 − x23 1 + x3


|z|2 = 2
= 2
=
(1 − x3 ) (1 − x3 ) 1 − x3

|z|2 − x3 |z|2 = 1 + x3 ; |z|2 − 1 = x3 (|z|2 + 1)

|z|2 − 1
x3 =
|z|2 + 1
2x1 2
z+z = ; 1 − x3 =
1 − x3 1 + |z|2
z+z z−z
x1 = x2 =
1 + |z|2 i(1 + |z|2 )
If
(x1 , x2 , x3 ) ←→ z = x + iy
x : y : −1 = x1 : x2 : x3 − 1
Stereographic projection of z ∈ C is (x1 , x2 , x3 ) ∈ S1 .
OR
Stereographic projection of (x1 , x2 , x3 ) ∈ S1 is z ∈ C

How do distances compare?


z ↔ s(z) = (x1 , x2 , x3 ) z 0 ↔ s(z 0 ) = (x01 , x02 , x03 )

x1 x01 + x2 x02 + x3 x03 =

(z + z)(z 0 + z 0 ) − (z − z)(z 0 − z 0 ) + (|z|2 − 1)(|z 0 |2 − 1)


(1 + |z|2 )(1 + |z 0 |2 )
(1 + |z|2 )(1 + |z 0 |2 ) − 2|z − z 0 |2
=
(1 + |z|2 )(1 + |z 0 |2 )
X 4|z − z 0 |2
ks(z) − s(z 0 )k2 = 2 − 2 xj x0j =
(1 + |z|2 )(1 + |z 0 |2 )
2|z − z 0 |
d(s(z), s(z 0 )) = p
(1 + |z|2 )(1 + |z 0 |2 )
very interesting;

16
d(s(z), s(z 0 )) = d(s(z), s(z 0 )) = d(s(1/z), s1/(z 0 ))
using 1/z implies we are taking z 6= 0. The above is direct verification.

The distance between s(z) and s(z 0 ) is also called the ‘chordal distance’
between z, z 0 – because it si the length of the chord joining the ‘image points’
on the sphere. What the aboe says is that the chordal distance remains
invariant under conjugation. It is invariant under (multiplicative) inverses
too of course, It is invariant under additive inverses too:

d(s(z), s(z 0 )) = d(s(−z), s(−z 0 )).

We shall not use the stereographic projection much in our course, but is
important when you think of ‘complex numbers’ as a ‘surface’. (Why should
one think of it as a surface?)

17
BVR/CMI Complex Analysis 08-02-2022

Stereographic Projection:

2|z − z 0 |
d(s(z), s(z 0 )) = p
(1 + |z|2 )(1 + |z 0 |2 )

d(s(z), s(z 0 )) = d(s(z), s(z 0 )) = d(s(1/z), s(1/z 0 )) = d(s(−z), s(−z 0 )).

= d(s(−z), s(−z 0 )) = d(s(−1/z), s(−1/z 0 )) = d(s(−1/z), s(−1/z 0 )).


[using 1/z implies we are taking z 6= 0.]

2<z 2=z |z|2 − 1


x1 = ; x 2 = ; x 3 =
1 + |z|2 1 + |z|2 |z|2 + 1

Upper hemisphere ←→ (|z| > 1)


lower hemisphere ←→ (|z| < 1)
Equator ←→ (|z| = 1) (z = x + iy ←→ (x, y, 0))
1+d
circle of Latitude (north) x3 = d > 0 ←→ circle |z|2 = 1−d >1
(Thus 0 < d < 1)
Circle of Latitude (south) x3 = d < 0 ←→ circle |z|2 = 1+d
1−d
<1
(Thus −1 ≤ d < 0)
Meridian (full) Circle of longitude ←→ straight lines pasing through zero
IF You let in ∞ into C; and correspond to (0, 0, 1) ∈ S1
2
d(s(z), s(∞)) = p
(1 + |z|2 )
4|z|2 4
x21 + x22 + (1 − x3 )2 = 2 2
+
(1 + |z| ) (1 + |z|2 )2
C ∪ {∞} is called extended complex plane.
Useful but to be handled with care.
****************************

OUR PICTURE of C as set R2


We write z = x + iy
C ∪ {∞} enters too = S1
*******************

Before proceeding:

18
Several Algebraic formulae you know in R are true in C. But must check.
EXample: Binomial theorem:

For complex z, w and for n = 1, 2, 3, . . .


     
n n n n−1 n n−2 2 n n
(z + w) = z + z w+ z w + ··· + w
1 2 n
z n − wn = (z − w)(z n−1 + wz n−2 + w2 z n−3 + . . . + wn−1 )
√ √ √
s Discipline: −1 −1 = −1 × −1 = 1?
In case of reals; we talked about ONLY square roots of non-negative
numbers; Further we made convention that squareroot means non-negative
square √root.
√ [To√put it differently, if you ever felt that it is your birth right
to say 2 3 = 6, think again, this needs proof!]

3. Sequences; series; open/closed sets


Sequences: Sequence is function {1, 2, 3, . . .} → C OR
{0, 1, 2, 3, . . .} → C. Instead of writing as a function we use {zn }.

DEF: zn → z means |zn − z| → 0

Thus zn → z if zn − z gets ‘closer and closer’ to zero, equivalently, zn gets


‘closer and closer’ to z.

D(z, r) = {w : |w − z| < r} for z ∈ C and r > 0. open disc of radius r


centered at z. This is USUAL High School disc in the plane. Thus zn → z
iff given any  > 0, after some stagethe sequence lies in -disc centered at z.

FACTS:
(i) zn → z iff <zn → <z and =zn → =z.
This is because |<zn − <z| = |<(zn − z)| ≤ |zn − z|. Thus if |zn − z| → 0
then |<zn − <z| → 0. Similarly |=zn − =z| → 0. Conversely if both these
happen, then use |zn − z| ≤ |<zn − <z| + |=zn − =z|.
(ii) If zn → z then |zn | → |z|
Use definition of |z| and convergence of real and imaginary parts.
(iii) zn → z, wn → w; then zn + wn → z + w
You can prove it either by saying that <zn → <z and <wn → <w and
so from proiperties of real sequences <(zn + wn ) → <(z + w). Similarly
=(zn + wn ) → =(z + w). OR you can repeat the proof of the real case:
|(zn + wn ) − (z + w)| ≤ |zn − z| + |wn − w| → 0

19
(iv) zn → z iff (zn − z) → 0
(v) zn → z then {zn } is bounded: (∃M )(∀n)(|zn | ≤ M )
This is because after some stage |zn − z| < 1, that is, after some stage
|zn | ≤ |z| + 1, now take M to be larger than: maximum of the first few |zj |
and |z| + 1
(vi) zn → z, wn → w; then zn wn → zw
You can use real and imaginary parts or use |zn wn −zw| ≤ |zn wn −zn w|+
|zn w − zw|.
(vii) zn → z then czn → cz for c ∈ C.
(viii) zn → z and z 6= 0 then (∃N )(∀n ≥ N )(zn 6= 0). In fact there is N
and a > 0 such that for all n ≥ N we have |zn | ≥ a.
If |z| = r > 0, the for every w ∈ D(z, r/3) we have |w| ≥ r/3 > 0
(otherwise |z| ≤ |w| + |z − w| ≤ 2r/3). For large n, zn is in this disc
D(z, r/3).
(viii) zn → z 6= 0 and zn 6= 0 for all n, Then 1/zn → 1/z.

DEF: (zn ) Cauchy sequence means


(∀ > 0)(∃N )(∀n, m ≥ N )|zm − zn | < 

Facts:
(ix) zn Cauchy iff (<zn ) and (=zn ) are Cauchy.
(x) {zn } converges iff it is Cauchy sequence.
(xi) If zn → z, changing/adding/removing finitely many terms,
results in a sequence which again converges to z.
You should not take on faith but verify. Once you do, you can use the
adjective ‘easy’.

For complex sequences NO lim sup and lim inf.


Recall: for real sequence {an }

lim sup an = lim[sup an ]


k n≥k

= sup[set of limit points of (an )]


P P
Series: symbol zn OR zn
P n≥1 n≥0
DEF:
P zn converges means sequence
P of partial sums converges. Put
sn = k≤n zk . If sn → s, then put zn = s.
P
Remember we can never add infinitely many numbers, thus if zn = s
then it is NOT because we have actually added and found the sum to be s
but because partial sums converge to s.

20
P
(i) If zn = s, then
the P + · · · convergesPto s − s30 .
series z31 + z32P
(ii) Pzn = s and wn = t thenP (zn + wn ) = s + t
(iii) zn = s and a ∈ C then (azn ) = as.
(iv) Replacing finitely many terms of a convergent series
resultsP in again a convergent series.
(v) If zn converges then zn → 0.
Warning s Converse is NOT true.
P P
DEF: zn converges absolutely means: |zn | converges.
P P
(vi) If zn converges absolutely, then zn converges.
WarningP sP Converse is NOT true. P
(vii) zn , wn converge absolutely, then (zn + wn ) converges abso-
lutely, P P P
(viii) zn converges absolutely, then | zn | ≤ |zn |
We have proved this only for P finite sums earlier. to prove
P it for infinite
sums, denote the partial sums of zn by sn and those of |zn | by tn . Now
the result for finite sums tells you |sn | ≤ tn . Since sn → s we have |sn | → |s|.
Thus |s| ≤ t. Remember, for real sequences if an → a and bn → b and an ≤ bn
for all n, then a ≤ b.

DEF: Convolution: (an , n ≥ 0) (bn , n ≥ 0) two sequences


Their convolution is the sequence (cn , n ≥ 0):
X
cn = a0 bn + a1 bn−1 + · · · + an b0 = ak bn−k

This looks like a funny definition. But you already know this.
P When you
100 n 100 n
P
have two polynomials P (t) = n = 0 a n t and Q(t) = n = 0 b nt
their product is also a polynomial. What are the coefficients? Precisely the
{cn } above.
Another place where you came across is in probability. Suppose X
is a random variable taking values {0, 1, 2, . . .} with respective probabili-
ties {an , n ≥ 0}. Similarly suppose Y is a random variable taking values
{0, 1, 2, . . .} with respective probabilities {bn , n ≥ 0}. What is the distribu-
tion of X+Y . Obviously X+Y takes again values {0, 1, 2, . . .}. Unfortunately
you can not decide probabilities. But if you are tild that the random variables
are independent, then you can calculate. The event (X + Y = 2) is union
of diajoint events (X = 0, Y = 2), (X = 1, Y = 1) and (X = 2, Y = 0) and
using independence you can calculate this: c2 . In general P (X +Y = n) = cn .

21
Here is an important theorem.
P P
THEOREM (CAUCHY): If P an = A and bn = B AND one of the
series absolutely converges then cn = AB.P P P
Proof:
P s n , t n and u n : partial sums of a n , b n and cn . Let us
assume an is absolutely convergent.
sn → A and tn → B. un → AB?
Yes sn B → AB. Shall show un − sn B → 0

un = a0 b0 + (a0 b1 + a1 b0 ) + · · · + (a0 bn + a1 bn−1 + · · · an b0 ).

un = a0 tn + a1 tn−1 + a2 tn−2 + · · · + an t0 .
sn B = a0 B + a1 B + a2 B + · · · an B,
u n − sn B =
a0 (tn − B) + a1 (tn−1 − B) + a2 (tn−2 − B) + · · · + an−1 (t1 − B) + an (t0 − B).
X
= an−k (tk − B)
P P
Fix  > 0. Using absolute convergence of ( an ), Let |an | = α > 0.
(α = 0 is easy!). Choose n0 such that |tn − B| < /(2α) for n ≥ n0 .
Thus n > n0 implies
n
X X
|un − sn B| = | an−k (tk − B)| + | an−k (tk − B)|
k=n0 k<n0

X X
≤ |an−k ||tk − B| + | an−k (tk − B)|
k≥n0 k<n0
 X X
≤ |an−k | + | an−k (tk − B)|
2α k≥n k≤n0
0

 X
≤ +| an−k (tk − B)|
2 k≤n 0

IMPORTANT: n0 fixed. an−k → 0 for k = 1, 2, · · · n0 −1. (Do you understand


what sequence are we talking about?) choose n1 > n0 so that in the above
display, second sum ≤ /2 for all n ≥ n1 .
Now For n > n1 |un − sn B| <  as claimed.

Warning s absolute convergence of one sequence important.

22
BVR/CMI Complex Analysis 09-02-2022

EXAMPLE: Take any complex number z,


∞ n
z
P
n!
absolutely convergent So converges:
0

z z2 z3
E(z) = 1 + + + + ···
1! 2! 3!
E(0) = 1, E(z + w) = E(z)E(w), if z is real then E(z) real.
ak = z k /k! bn−k = wn−k /(n − k)!
X X 1 n (z + w)n
ak bn−k = z k wn−k = .
n! k n!
Shall return soon but here is a general fact about ‘such’ series.

Power Series

an z n = a0 + a1 z + a2 z 2 + · · · is called a
P
DEF: A series of the form
power series. Here {an } are fixed complex numbers. z is a variable.
The series may converge for some complex numbers z, and may not con-
verge for some others. But interestingly enough there is a certain regular
behaviour.

Let
1
R= p
lim sup n
|an |
(if the limsup is zero take R = ∞ and if the limsup is infinity take R = 0.)

Theorem: If |z| < R then the above series converges absolutely and hence
converges. If |z| > R the above series does not converge. If |z| = R we can
not decide. For example for some z, with |z| = R the series may converge
and for some other z, with |z| = R the series may not converge.

R is called the radius of convergence of the power series.


p
Proof is simple. Suppose p |z| > R. Then lim sup n
|an ||z| > 1. So for
n
infinitely many n,we have n
|an ||z| > 1 Or |an ||z | > 1. Thus the series
n
P
an z can not converge. p
Suppose
p |z| < R. Then lim sup n
|an ||z| < c < 1. So after some stage we
n n
|an z n | converges.
P
have |an ||z| < c Or |an ||z | < c . Thus the series
n

This completes the proof.

23
1 − 11 z + 12 z 2 − 13 z 3 + · · ·
R = 1; z = 1 it converges; z = −1 does not converge.
1 + z + z2 + z3 + · · ·
n+1
Here again R = 1. partial sum sn (z) = 1−z 1−z
and when |z| = 1 this
does not converge. We shall see this soon. of course if z = e2πiθ for rational
θ, you can see {z n } is periodic and does not converge. Actually it does not
converge for other values too.
1 + z + 2 2 z 2 + 33 z 3 + 44 z 4 + · · ·
Here R = 0
1 + z + 2!z 2 + 3!z 3 + 4!z 4 + · · ·
Here again R = 0. (Actually ratio test will help.)

More general power series: Fix z0 ∈ C.


a0 + a1 (z − z0 ) + a2 (z − z0 )2 + a3 (z − z0 )3 + · · ·
Its nature of convergence follows from above discussion: It converges if
|z − z0 | < R and fails to converge for |z − z0 | > R.

DEF: U ⊂ C is open if z ∈ U ⇒ (∃r > 0)[D(z, r) ⊂ U ]


DEF: F ⊂ C closed if [(∀n)zn ∈ F, zn → z] ⇒ [z ∈ F ].

Fact: F is closed iff C \ F is open.


Open disc D(z, r) = {w : |w − z| < r} is open.
Closed disc D(z, r) = {w : |w − z| ≤ r} is closed.

DEF: K ⊂ C is compact if every open cover of K has finite subcover.

i.e. {Uα } open sets, K ⊂ ∪Uα implies


K ⊂ Uα1 ∪ · · · ∪ Uαm for some finitely many.
Fact: K compact ←→ K closed and bounded (in Euclidean metric).
←→
every sequence in K has a subsequence converging to a point in K.

4. Continuity and Differentiability

DEF: A ⊂ C, and f : A → C then f is continuous if


(zn ∈ A) for all n, zn → z and z ∈ A imply f (zn ) → f (z).

Facts: f on A is continuous iff both <f and =f are so.


Here: (<f )(z) = <(f (z)) and (=f )(z) = =(f (z))
Usual notation: <f = u and =f = v

24
f = u + iv, u, v real valued.
f + g, f g, cf , f /g (provided g 6= 0) continuous if f, g are.
f (z) = z is conitnuous on all of C.
So z 2 , z 3 , . . . are continuous functions.
Every polynomial (in z) is continuous on C.
P zn
E(z) = n!
is continuous on C. To see this Fix z0 and  > 0. Fix N ,
∞ −1 n
NP
P (|z0 |+1)n z
n!
< /4. P (z) = n!
Polynomial and hence continuous
N 0
Fix 0 < δ < 1, |z0 − w| < δ ⇒ |P (z0 ) − P (w)| < /4.
Claim: |z0 − w| < δ ⇒ |E(z0 ) − E(w)| < .
Simply use: |E(z0 ) − P (z0 )| + |P (z0 ) − P (w)| + |P (w) − E(w)| < 3/4
and also note |w| ≤ |z0 | + |w − z0 | < |z0 | + 1.
Same proof gives general result about functions defined by power series
in the disc of convergence:

an z n power series, radius of convergence R > 0. A


P
Theorem: A(z) =
is continuous on D(0, R), open disc.
Proof: Take z0 , |z0 | < R, and  > 0. Fix η > 0; |z0 | + η < R .
∞ NP−1
an (|z0 | + η)n < /4. P (z) = an z n Polynomial and hence
P
Fix N ,
N 0
continuous.
Fix 0 < δ < η, |z0 − w| < δ → |P (z0 ) − P (w)| < /4.
Claim: |z0 − w| < δ → |A(z0 ) − A(w)| < .
The proof is exactly as above.
Here is Another proof that has somne useful ideas (essentially what was
used above)

DEF: A sequence of functions {fn } converges pointwise to a functions f


on A if fn (z) → f (z) for every z ∈ A.

Thus given a point z ∈ A and  > 0 there is a N (possibly depending on


the given z such that |fn (z) − f (z)| <  for n ≥ N .)

DEF: fn → f uniformly on A if
(∀ > 0)(∃N )(∀n > N )(∀z ∈ A)|fn (z) − f (z)| < 

similar definition for series of functions:


P P
DEF: Say fn = f pointwise on A if fn (z) converges to f (z) for all

25
z ∈ A.
P n
P
DEF: Say fn = f uniformly on A if fm (z) converges to f (z) uni-
m=0
formly.

THEOREM: each fn continuous; fn → f uniformly Then f is continuous.


Entire action taking on a set A.
P
THEOREM:
P If there are numbers (Mn ) such that |fn | ≤ Mn and Mn <
∞ Then fn converges uniformly.
This is kmnown as Weierstrass M -test.

THEOREM: A(z) power series; R > 0 radius of convergence. For any


n
0 < r < R, fn (z) = an z n converges uniformly to A(z) on D(0, r).
P
0
Proof is immediate.
This proves that A(z) is continuous in the open disc of convergence.

26
BVR/CMI Complex Analysis 11-02-2022

Can localize the concept of continuity: continuous at a point z. Fix z.

DEF: Say f continuous at z if zn ∈ A for all n; zn → z then f (zn ) → f (z).

Facts: f continuous at z iff


(∀ > 0)(∃δ > 0)[|w − z| < δ, w ∈ A → |f (z) − f (w)| < .]
Remember z fixed, so δ could depend on this z.
(f (x) = x2 on R to R.)
Fact: f continuous on a compact set K, then f is bounded. (that is
(∃M )(∀z ∈ K)(|f (z)| ≤ M ))
Fact: f continuous on a compact set K, then f is uniformly continuous.
That is (∀ > 0)(∃δ > 0)(z, w ∈ K, |z − w| < δ → |f (z) − f (w)| < ). In
other words the δ in the definition of continuity does not depend on any
point, works for all.

DEF: f defined on an open set U and z0 ∈ U . SAY f differentiable at z0


if lim f (z0 +h)−f
h
(z0 )
exists as h → 0.

This means there is α such that (∀n)hn 6= 0, (∀n)z0 + hn ∈ U , hn → 0


imply f (z0 +hhnn)−f (z0 ) → α.
can there be two such α? Convince yourself that there can not be.
Equivalently:
(∀n)hn 6= 0, z0 + hn ∈ U implies lim f (z0 +hhnn)−f (z0 ) exists.
This appears weaker than the earlier statement. It appears can the limit
depend on the sequence. But it is not so. If there are two sequences (hn )
and (h0n ) for which the ratios converge two different limits then you can see
the following. Consider the sequence

(ηn ) = (h1 , h01 , h2 , h02 , h3 , h03 , . . .)

Then the ratios for this sequence can not converge. This new sequence is the
‘interlacing’ of the sequences (hn ) and (h0n ).
Equivalently:
there is α such that
(∀ > 0)(∃δ > 0)[0 < |h| < δ → | f (z0 +h)−f
h
(z0 )
− α| < .]

When the above happens, we write f 0 (z0 ) = α.


DEF: f differentiable in U if so at every point of U .

27
EXAMPLE: f (z) ≡ 3 + 29i Then it is differentiable on all of C and f 0 ≡ 0
Example: f (z) = z This is differentiable on all of C and f 0 ≡ 1
Example: f (z) = z n (n ≥ 1 integer) is differentiable and f 0 (z) = nz n−1 .
This is routine.
(z + h)n − z n
   
n−1 n : n n−3 2
= nz + z n − 2h + z h + ...
h 2 3

and hence converges to nz n−1 .


Non-example: f (z) = z is NOT differentiable. If we take hn = 1/n, we
see that the limit is 1; for hn = i/n, we see the limit is −1

FACTS:
(i) f differentiable, then it is continuous.
(ii) (f + g)0 = f 0 + g 0 ,
(iii) (af )0 = af 0 for a ∈ C.
(iv) (f.g)0 = f g 0 + f 0 g. Use

f (z0 + h)g(z0 + h) − f (z0 + h)g(z0 ) f (z0 + h)g(z0 ) − f (z0 )g(z0 )


+ ]
h h
(v) (1/g)0 = −g 0 /g 2 if g never zero in U . Use

1/g(z0 + h) − 1/g(z0 ) g(z0 ) − g(z0 + h)


=
h g(z0 )g(z0 + h)h

(vi) g : U → V , f : V → C differentiable then so is the composition f og.


And (f og)0 (z) = f 0 (g(z)) g 0 (z) on U . Recall: (f og)(z) = f (g(z)); z ∈ U
Use
f (g(z0 + h)) − f (g(z0 )) g(z0 + h) − g(z0 )
×
g(z0 + h) − g(z0 ) h
There is just one problem with above argument. What if g(z0 +h)−g(z0 ) = 0?
You can overcome it as in real variable calculus. Remember, you can not say
this till you have done it once and convinced yourself.

EXAMPLE: P (z) = a0 + a1 z + · · · + an z n Then P 0 (z) = a1 + 2a2 z +


3a3 z 2 + · · · + nan z n−1

STORY UNFOLDS:

Is f = u + iv differentiable same as saying u, v as functions of two real


variables ‘differentiable’ ? FAR FROM IT! See:

28
f (x + iy) = x + iy; Then u(x, y) = x v(x, y) = y. If g(x + iy) = x − iy;
then u(x, y) = x v(x, y) = −y. Clearly the functions u, v are very nice
as functions of two variables. Thus complex differentiability is more than
simply two variable-function differentiability.

Theorem (Cauchy-Riemann equations):


If f differentiable (complex) then so are the two variable real functions
u, v. further
∂u ∂v ∂u ∂v
= ; =−
∂x ∂y ∂y ∂x
Proof of C-R equations:

f (z0 + h) − f (z0 )
→ a + ib
h
Take cn 6= 0 real, cn → 0 hn = cn + i0

u(x0 + cn , y0 ) − u(x0 , y0 ) v(x0 + cn , y0 ) − v(x0 , y0 )


+i → a + ib
cn cn
So real parts and imaginary parts converge to the respective quantities. This
gives,
∂u ∂v
(x0 , y0 ) = a (x0 , y0 ) = b
∂x ∂x
Take dn 6= 0 real, dn → 0 hn = 0 + idn

u(x0 , y0 + dn ) − u(x0 , y0 ) v(x0 , y0 + dn ) − v(x0 , y0 )


+i → a + ib
idn idn
That is
u(x0 , y0 + dn ) − u(x0 , y0 ) v(x0 , y0 + dn ) − v(x0 , y0 )
−i + → a + ib
dn dn
Comparing real and imaginary parts, we have , as above
∂v ∂u
(x0 , y0 ) = a (x0 , y0 ) = −b
∂y ∂y
This proves the result. Now let us show that u, v are differentiable.
as (h, k) → (0, 0)

u(x0 + h, y0 + k) − u(x0 , y0 ) + i[v(x0 + h, y0 + k) − v(x0 , y0 )]


→ a + ib
h + ik

29
u(x0 + h, y0 + k) − u(x0 , y0 ) − (ah − bk) +
i[v(x0 + h, y0 + k) − v(x0 , y0 ) − (bh + ak)]
h + ik
→0
So modulus square converges to zero, remember |z/w|2 = |z|2 /|w|2 . This
leads us to
|u(x0 + h, y0 + k) − u(x0 , y0 ) − (a, −b) · (h, k)|
→0
k(h, k)|

|v(x0 + h, y0 + k) − v(x0 , y0 ) − (b, a) · (h, k)|


→0
k(h, k)k
Same as

ku(x0 + h, y0 + k) − u(x0 , y0 ) − L(h, k)k


→0
k(h, k)k
L linear map: L(x, y) = (a, −b) · (x, y)

k|v(x0 + h, y0 + k) − v(x0 , y0 ) − M (h, k)k


→0
k(h, k)k

M linear map: M (x, y) = (b, a) · (x, y).


This shows u, v as functions of two variables differentiable.

Every power series defines differentiable function in the disc of conver-


gence. In fact you can differentiate term by term.
THEOREM: A(z) = a0 + a1 z + a2 z 2 + a3 z 3 + · · · ; R radius of convergence.
Then A is differentiable in the D(0, R) and

A0 (z) = a1 + 2a2 z + 3a3 z 2 + · · ·

Proof: Same idea as showing continuity. Cut off the infinite sum at a
finite stage to get a polynomial and show ‘error’ is small. In other words
Choose suitable N , Put
N
X ∞
X
j
P (z) = aj z R(z) = aj z j
0 N +1

Thus A = P + R. For h 6= 0,

A(z0 + h) − A(z0 ) 0

− A (z0 ) ≤
h

30

P (z0 + h) − P (z0 ) 0
R(z0 + h) − R(z0 )
− P (z0 ) +
+ |P 0 (z0 ) − A0 (z0 )|
h h
= (i) + (ii) + (iii)
We are already using A0 for the ‘alleged derivative’ ! You may just rename it
as B(z) and show the result and then conclude that B is derivative of A.
Execution:
Fix  > 0. To show δ > 0

A(z0 + h) − A(z0 ) 0

0 < |h| < δ ⇒ − A (z0 ) ≤ 
h

Step 1: A0 has also R, radius of conv. This is easy, Convergence of a1 +


2a2 z + 3a3 z 2 + · · · is same as that of 0 + a1 z + 2a2 z 2 + 3a3 z 3 + · · · which has
radius of convergence 1/ lim sup |nan |1/n . Since n1/n → 1 we see that this is
same as 1/ lim sup |an |1/n =R.
Step 2: |z0 | < R, fix η > 0 such that |z0 | + η < R. Fix N so that

X
n|an |(|z0 | + η)n−1 < /4
N +1

set
N
X ∞
X
j
P (z) = aj z R(z) = aj z j
0 N +1

Now P being a polynomial, we can choose 0 < δ < η such that



P (z0 + h) − P (z0 ) 0

− P (z0 ) < /4
h

We show this δ does the job.


By Choice of δ we have (i) ≤ /4.
By Choice of N we have (iii) < /4.
n−1
(z0 + h)n − z0n X j
| |= z0 (z0 + h)n−1−j ≤ n(|z0 | + η)n−1 .
h 0

R(z0 + h) − R(z0 ) X (z0 + h)n − z0n
= an
h N +1
h

X
≤ |an |n(|z0 | + η)n−1 < /4
N +1

31
again by choice of N .
This completes the proof.

TERMINOLOGY: U open subset of C, f : U → C, differentiable. Wesay


f is Holomorphic on U ; or analytic on U ; or regular on U .

32
BVR/CMI Complex Analysis 15-02-2022

If A(z) is a power series,

A(z) = a0 + a1 z + a2 z 2 + a3 z 3 + a4 z 4 + · · ·

then formula for its derivative A0 (z) is also a power series (with same radius
of convergence).

A0 (z) = a1 + 2a2 z + 3a3 z 2 + 4a4 z 3 + 5a5 z 4 + · · ·

In particular you can apply the theorem again to say that A0 is differentiable
and

A00 (z) = 2a2 + (3 × 2)a3 z + (4 × 3)a4 z 2 + (5 × 4)a5 z 3 + (6 × 5)a6 z 4 + · · ·

In particular a function given by a power series is infinitely differentiable,


that is, differentiable any finite number of times.
Suppose we have

A(z) = a0 + a1 (z − z0 ) + a2 (z − z0 )2 + a3 (z − z0 )3 + a4 (z − z0 )4 + · · ·

with Radius of convergence R. Then this defines an analytic function in


D(z0 , R). you can prove this by translation (or repeat the earlier proof,
bad). Further

A0 (z) = a1 + 2a2 (z − z0 ) + 3a3 (z − z0 )2 + 4a4 (z − z0 )3 + 5a5 (z − z0 )4 + · · ·

In fact A(z) is differentiable any number of times.


Before discussing some examples, an amplification of differentiability. Let
f = u + iv usual representation of f on an open set U . Let z0 ∈ U and f
be differentiable with f 0 continuous and f 0 (z0 ) 6= 0. Actually you need not
assume continuously differentiable, later you see derivative is again differ-
entiable too, but now assume. Shall think of u, v as functions of two real
variables. Thus (u, v) : U ⊂ R2 → R2 . We keep writing row, but actually R2
is column vectors and thus the map we are talking about is
   
x u(x, y)
7→
y v(x, y)

Let us see the jacobian at the point z0 = (x0 , y0 ),


 ∂u
(x0 , y0 ) ∂u
  ∂u ∂v

∂x ∂y
(x0 , y0 ) ∂x
(x 0 , y0 ) − ∂x
(x 0 , y0 )
J= ∂v ∂v = ∂v
∂x
(x0 , y0 ) ∂y (x0 , y0 ) ∂x
(x0 , y0 ) ∂u∂x
(x0 , y0 )

33
 
a −b ∂u ∂v
= ; a= (x0 , y0 ), b = (x0 , y0 )
b a ∂x ∂x
Notice that J is just f 0 (z0 ), when this later complex number is viewed as a
2 × 2 matrix. You can also see that J, as a map of R2 , is composition of
‘multiplication’ with rotation (Remember f 0 (z0 ) 6= 0). In fact multiplication
by (a2 + b2 ) and rotation by the matrix
!
√ a √ −b
a2 +b2 a2 +b2
√ b √ a
a2 +b2 a2 +b2

This shows that f preserves angles between curves at the point z0 .

34
What does this mean? First remember that if you take two smooth curves
passing through z0 then their angle at z0 is the angle between the tangents to
the curves at z0 . Also if you take two curves C1 , C2 passing through z0 then
their images (under f ) give you two curves passing through w0 = f (z0 ). The
statement is: angle between C1 , C2 at z0 equals the angle between the image
curves at w0 . This is expressed by saying that the map f is ‘conformal’ or
‘angle preserving’. Remember that our conditions on f assure us that it is
1-1 in a neighbourhood of z0 and the image curves are genuine curves.
There are several ways of proving this conformality. If the curve is C(t) =
x(t)+iy(t) with C(t0 ) = z0 then slope of the tangent at t0 equals y 0 (t0 )/x0 (t0 )
or (avoid zero in denominator) it is tan θ where θ = arg C 0 (t0 ). Its image
curve is f (C(t)) which has slope tan ψ where ψ = arg f 0 (z0 ) + arg C 0 (t0 ) =
arg f 0 (z0 ) + θ. We have used formula for derivative of composition: f oC
(what is the formula?). If now we have another curve C ∗ with C ∗ (t0 ) = z0
(you can take s0 but why complicate life, anyway then you will look at the
tangent at s0 , that is all). Then slope of its tangent is tan θ∗ , say. Its image
would have slope tan ψ ∗ where ψ ∗ = arg f 0 (z0 ) + θ∗ . Thus ψ − ψ ∗ = θ − θ∗
as stated.
You can also prove it the hard way too, namely calculate the slopes m1 , m2
for the two curves C, C ∗ and use the formula that if α is the angle between
the curves, then tan α = (m1 − m2 )/1 + m1 m2 . Calculate the same quantity
for the image curves and show they are same. Here it is.
curves (a(t), b(t)) and (c(t), d(t)) z0 at t0 . slopes m1 = b0 /a0 and
m2 = d0 /c0 . EVERYTHING evaluated at t0 .
tan(Angle between tangents) =
m1 − m2 b 0 c 0 − a0 d 0
= 0 0 (♠)
1 + m1 m2 a c + b0 d0
For image curves corresponding quantities:
v1 a0 + v2 b0 −u2 a0 + u1 b0
m1 = =
u1 a0 + u2 b0 u1 a0 + u2 b0
suffixes partial derivatives w.r.t. those variables
−u2 c0 + u1 d0
m2 =
u 1 c0 + u 2 d 0
For the image curves
m1 − m2
=
1 + m1 m2
−u2 a0 +u1 b0 0 0
u1 a0 +u2 b0
− −u 2 c +u1 d
u1 c0 +u2 d0
= 0 0 0 0
1 + −u 2 a +u1 b −u2 c +u1 d
u1 a0 +u2 b0 u1 c0 +u2 d0

36
(−u2 a0 + u1 b0 )(u1 c0 + u2 d0 ) − (u1 a0 + u2 b0 )(−u2 c0 + u1 d0 )
=
(−u2 c0 + u1 d0 )(u1 c0 + u2 d0 ) + (−u2 a0 + u1 b0 )(u1 a0 + u2 b0 )
−a0 d0 (u22 + u21 ) + b0 c0 (u21 + u22 )
= 0 0 2 = (♠)
a c (u1 + u22 ) + b0 d0 (u22 + u21 )
Differentiable functions are called analytic because if f is differentiable in
U , then given z0 ∈ U , there is a small disc in which it has a power series exp-
nasion, in powers of (z − z0 ). Differentiable functions are called holomorphic
because they preserve ‘everything’, well this just means if f is non-constant
then it preserves the ‘local angles beteen curves’ as explained above.

5. EXP and all that


(A): Exponential/Trigonometric functions

z2 z3
E(z) = 1 + z + + + ···
2! 3!
is called exponential function. This is defined for all complex numbers z.
This is a continuous function. In fact.

(i) E(z) is a holomorphic function and E 0 (z) = E(z).


This follows from theorem on power series.

(ii) E(0) = 1 and E(z) is real if z is real.


direct calculation.

(iii) E(z + w) = E(z)E(w) for all z, w.


In particular E(−z) = 1/E(z) and E(z)is never zero.
This was seen already as a consequence of Cauchy product theorem. Also
1 = E(0) = E(z − z) = E(z)E(−z).

(iv) E(x) is increasing on the real axis and agrees with the definition you
know on R. lim E(x) = ∞ as x → ∞ and lim E(x) = 0 as x → −∞.
For x > 0 we see E(x) > x > 0. In particular it takes non-negative values
and limit as x → ∞ is infinity. E(−x) = 1/E(x) implies that for x < 0, we
have E(x) > 0. Further limit as x → −∞ is zero. Finally E 0 (x) = E(x) says
that the functiuon is indeed increasing.

DEFINITION:

z2 z4 X (−1)n z 2n
cos z = 1 − + + ··· =
2! 4! 0
(2n)!

37

z3 z5 X (−1)n z 2n+1
sin z = z − + + ··· =
3! 5! 0
(2n + 1)!
These two power series converge for every z.

(v)
E(iz) = cos z + i sin z, E(−iz) = cos z − i sin z
1 1
cos z = [E(iz) + E(−iz)] sin z = − i[E(iz) − E(−iz)]
2 2
These are Euler’s formulae.
For example,
X in z n ∞ 2k 2k ∞ 2k+1 2k+1
X i z X i z
E(iz) = = +
n! k=0
(2k)! k=0
(2k + 1)!

By absolute convergence this rearrangement is permitted.

(vi) In particular, For any real t, E(it) = cos t+i sin t. This is DeMoivre’s
formula.

(vii)

cos2 z + sin2 z = 1 cos(z ± w) = cos z cos w ∓ sin z sin w

sin(z ± w) = sin z cos w ± cos z sin w


For example
1 1
cos2 z + sin2 z = [E(iz) + E(−iz)]2 + (−i)2 [E(iz) − E(−iz)]2
4 4
(viii) sin z and cos z are differentiable and

(cos z)0 = − sin z (sin z)0 = cos z

Remember z 7→ iz 7→ E(iz) has derivative iE(iz).


Clearly cos 0 = 1
22 24 26 28 210
cos 2 = 1 − + − + − + ···
2! 4! 6! 8! 10!
16 26 28 210 212
≤ (1 − 2 + )−[ − ]−[ − ] − ···
24 6! 8! 10! 12!
1
≤− .
3
38
DEFINITION: Least number t > 0 with cos t = 0 denoted π/2.
Observe that there is such a least number.

(ix) cos t strictly decreasing in [0, π/2) from 1 to zero. sin t strictly in-
creasing in [0, π/2) from zero to 1.
This is because cos 0 = 1 and π/2 is first zero implies that cos t > 0 on
[0, π/2) Now sin0 t = cos t implies sin t is strictly increasing on this interval.
Now (cos t)0 = − sin t tells cos t is strictly decreasing in this interval. Further
cos π/2 = 0 implies sin π/2 = 1, in view of cos2 t + sin2 t = 1.

(x)

cos(π/2) = 0 sin(π/2) = 1 cos π = −1 sin π = 0

cos(3π/2) = 0 sin(3π/2) = −1 cos(2π) = 1 sin(2π) = 0


cos(z + π/2) = − sin z sin(z + π/2) = cos z
cos(z + π) = − cos z sin(z + π) = − sin z
cos(z + 2π) = cos z sin(z + 2π) = sin z
First is definition and second was observed above already. Others follow
from formulae for cos(z + w), sin(z + w)

(xi) cos z and sin z are periodic with period 2π in C.

(xii) If a, b real numbers a2 + b2 = 1, there is exactly one θ ∈ (−π, π] with


cos θ = a and sin θ = b. Given any z = x + iy 6= 0 there is unique θ as above
z = |z|E(iθ).
Follows from (x).
Above θ is called principal argument/amplitude of z. Any θ satsifying
z = |z|E(iθ) is called an argument/amplitude of z. Thus there are infinitely
many arguments and any two differ by a multiple of 2π.
However zero is not assigned any amplitude.

(xiii) argument is continuous function on C \ {x + i0 : x ≤ 0}.


Direct verification.

(xiv) E(z) is periodic with period 2πi: E(z + 2πi) = E(z). Further it
assumes all non-zero values.
Periodicity follows from that of sin z and cos z functions. Given any com-
plex number z 6= 0, we know there is a real number such that E(x) = |z|.

39
You can use the facts: limits of E(x) on real axis, continuity and intermedi-
ate value theorem. Thus z = |z|E(iθ) = E(x)E(iθ) = E(x + iθ). Here θ is
from previous observation (use z/|z|).

Here is the upshot of this: On each vertical line, the values repeat with
period 2π. Each of the horizontal strips R × [2kπ, (2k + 2)π) is mapped by
exponential function onto C − {0}, (k ∈ Z). Each of these horizontal strips
are called ‘ a Fundamental domain’ for the exponential function. Of course
if you consider any horizontal strip of height 2π (include only one boundary
line) the exponential function is 1-1. Further on each vertical line it has
constant modulus. On the vertical line {x + iy : y ∈ R} (x fixed) the val-
ues of the exponential function go round and round the circle; center zero,
radius E(x). On each horizontal line {x + iy : x ∈ R} (y fixed) the values
of the exponential function trace the ‘radial line’ from the origin (excluding
the origin) which makes angle y with horizontal axis.

One can now define


sin z 1 E(iz) − E(−iz)
tan z = =
cos z i E(iz + E(−iz))
and
cos z E(iz) + E(−iz)
cot z = =i
sin z E(iz − E(−iz))
One can define sinh z and cosh z etc.
before leaving this discussion, let us attend to one question. You might say
that the definition of π/2 must be reconciled with your high school definition,
namely, that it is area of unit disc. Yes. This is simple integration. The part
of disc in first quadrant has area
Z 1√ Z π/2
2
π 1
1 − x dx = cos θ cos θdθ =
0 0 2 2
No need to evaluate any integral, use integral of cos2 t and sin2 t are same
and add to length of the interval, so each equals 1/2.
Or you may say that 2π is defined as the circumference√ of the unit circle.
Yes. Again the curve in the first quadrant is y(x) = 1 − x2 so that the
length of the curve equals
Z 1r Z 1 Z π/2
dy 2 1 1
1 + ( ) dx = √ dx = dθ = π
0 dx 0 1 − x2 0 2
as stated.

40
BVR/CMI Complex Analysis 15-02-2022

(B). logarithm/Logarithm

Given any complex number w 6= 0 there is a complex number z, such


that E(z) = w. In fact if z = x + iy we must have E(x)E(iy) = w. Since
|E(iy)| = 1 we conclude that we must have E(x) = |w|. Clearly there is only
one such real number x, it is the usual logarithm of |w| as we know on the real
line. Just remember that x is the real part of z. Remember We showed E(x)
is a 1-1 continuous function on R onto (0, ∞). Thus z = log |w| + i arg w.
Of course arg w is determined only upto multiples of 2π. Thus every non-
zero complex number has infinitely many logarithms, any two differing by a
multiple of 2πi. Thus all z with E(z) = w are given by

z = log |w| + i arg w + k2πi k∈Z

Let us take the principal argument: usually [0, 2π) but some do take [−π, π).
The principal branch of the logarithm is given by

Logz = log |w| + i arg w

one uses log z to denote ‘any one’ or ‘the set of all’ w with E(w) = z. Of
course this set is nothing but the principle value plus integer multiples of
2πi.
CONVENTION: If z is real and positive, say a then log a means the real
logarithm (of course, it is the principal value.)
Using logarithm one defines ‘exponentiation’: For complex numbers a, b
with a 6= 0
ab = E(b log a)
If a is restricted to real numbers (non-zero), then log a is real and ab is single
valued. Otherwise ab has infinitely many values which differ by factors e2πnb .
There will be a single value iff b is an integer, say n. Then ab is interpreted as
power of a√or of a−1 . If b = p/q (reduced form) rational then ab has exactly
q values: q ap . In particular, denoting e = E(1), we have

ez = E(z log e) = E(z)

because log e = 1. We denote E(z) by ez from now on.

Clearly
Log(w1 w2 ) = Log(w1 ) + Log(w2 )
Remember both sides infinite sets of complex numbers.

41
Both logarithm and exponentiation are tricky. We shall return to log-
arithm later but not much to exponentiation. Here are some interesting
calculations. Since |i| = 1,
π 1
log i = log 1 + i( + 2nπ) = i(2n + )π, n = 0, ±1, ±2, . . . .
2 2
so that
i−2i = e−2i log i = e[(4n+1)π] ; n∈Z
The principal value of (−i)i is given
π
P V (−i)i = ei log(−i) = ei(log 1−i 2 ) = eπ/2

principal argument of −i is taken −π/2. Of course if our convention is


[0, 2π) we could take 3π/2.
Should learn to deal with Multiple valued functions.

(C). z n and n
z

The function w = z n is holomorphic.

|w| = |z|n , arg w = n arg z

need to take argument above mod 2π. Every circle in the z-plane is mapped
to a circle in the w-plane. As z goes around the circle (|z| = r) once, the
image goes around the circle (|w| = rn ), n times. For one value of w = |w|eiθ
the n values of z are the equally spaced points
p
zk = n |w| ei(θ+2kπ)/n k = 0, 1, . . . n − 1

The sector (z : 0 ≤ arg z ≤ 2π/n) is mapped onto the full w-plane. There
are n such sectors.
The n many n-th roots of unity form a cyclic group under multiplication.

(D). Polynomials

We have already used this term. Polynomial is a function of the form

P (z) = a0 + a1 z + · · · + an z n

for some integer n ≥ 0, and an 6= 0. It is polynomial of degree n. It is


n
nan z n−1 . Later we see that
P
a holomorphic function and its derivative is
1
when n ≥ 1, there is at least one value z0 such that P (z0 ) = 0.

42
From now on n ≥ 1.
Then we can, after dividing by (z −z0 ) express as P (z) = (z −z0 )Q(z)+R
and using P (z0 ) = 0 deduce R = 0 so that P (z) = (z − z0 )Q(z) where Q is
of degree n − 1. You can use induction (on what?) to see we can write

P (z) = an (z − z0 )(z − z1 ) · · · (z − zn )

Of course some of the zj may be equal. Thus if the distinct ‘roots’ are
z1 , z2 , . . . , zk appearing n1 , n2 , . . . , nk times then we have
X
P (z) = an (z − z1 )n1 · · · (z − zk )nk ; nj ≥ 1, nj = n.

if P (z) has arepresentation P (z) = (z − z0 )m Q(z) with Q(z0 ) 6= 0, we say


z0 is a ‘zero of order m’. Thus in the above representation, (since the zj are
distinct), we have zj is a zero of order nj for 1 ≤ j ≤ k. Note that P is NOT
zero for any other value, except the zj .Thus a polynomial of order n ≥ 1
has exactly n roots (counted with multiplicities). This also implies that it
assumes every value exactly n time (again counting with multiplicities). This
is clear from considering P (z) − w = 0 for a given w. the expression

n
an−1 an−2
|Pn (z)| = |z | an + + 2 + ···

z z
shows that lim |Pn (z)| = ∞ as |z| → ∞.

Pn (z)
lim n = |an | =
6 0
z

This is expressed by saying that Pn has a ‘pole of order n at infinity’.

all these things we shall use and appreciate in the cointext of holomorphic
functions. Yes, the context of continuous functions is too general. you can
speak of fractional orders. But in the context of holomorphic functions our
wishfuyl thinking is indeed true and need only integer powers.
zeros and poles:

In general when a function (always continuity is assumed now) is not


a polynomial, the above divisions are not possible (at least superficially).
Notice that a is a zero of order k for P (z) means P (z) = (z − a)k Q(z) where
Q(a) 6= 0. In general let f be a function defined in a neighbourhood of
a. Suppose f (a) = 0. We say a is a zero of order k ≥ 1 for f if f (z) =
(z − a)k f1 (z) in a neighbourhood of a, where f1 (a) 6= 0. Note that such a
k, if exists is unique. For example, if f (z) = (z − a)k f1 (z) and also f (z) =

43
(z − a)k+1 fz (z) then (z − a)k f1 (z) = (z − a)k+1 fz (z) so that for z 6= a, we
have f1 (z) = (z − a)f2 (z) But since both sides are continuous functions, this
holds for z = a also, but then f1 (a) = 0.
Thus z = 0 is a zero of order k ≥ 1 if in a neighbourhood of zero, we have
f (z) = z k f1 (z) with f1 (0) 6= 0.
Let now suppose f is defined in a deleted neighbourhood of a, that is,
a neighbourhood of a excluding the point a. Suppose that lim |f (z)| = ∞
(same as lim f (z) = ∞) as z → a. We say a is a pole of order k ≥ 1
if f (z) = f1 (z)/(z − a)k where f1 (z) has finite non-zero limit at a. Again
note that such a k, when exists, is unique. Indeed if f (z) = f1 (z)/(z − a)k
and f (z) = f2 (z)/(z − a)k+1 in a deleted neighbourhood of a, then f2 (z) =
f1 (z)(z − a). Since f1 has finite limit at a, we see f2 has limit zero.

Thus z = 0 is a pole of order k ≥ 1 if in a deleted neighbourhood of zero,


we have f (z) = f1 (z)/z k with f1 having a finite non-zero limit as z → 0.
Wishful thinking:
Suppose f has a power series expansion in a neighbourhood of a,

f (z) = c0 + c1 (z − a) + c2 (z − a)2 + · · ·

if f (a) = 0, then c0 = 0. It is quite possible c1 may also be zero. Take the


first k such that ck 6= 0. Clearly k is the order of zero at a for f .
Suppose that f defined in a deleted neighbourhood of a has an expansion
in powers of (z − a) with some negative powers too:
c−k c−k+1 c−1
f (z) = k
+ k−1
+· · ·+ + c0 + c1 (z − a) + c2 (z − a)2 + · · ·
(z − a) (z − a) (z − a)

for z 6= a with c−k 6= 0. First of all lim f (z) as z → a is infinity. Further


(z − a)k f (z) has a non-zero finite limit as z → a. Thus a is a pole of order k.
In particular near z = 0 if

f (z) = ck z k + ck+1 z k+1 + · · · ck 6= 0

then z = 0 is a zero of order k. Suppose in a deleted neighbourhood of z = 0


c−k c−k+1 c−1
f (z) = k
+ k−1 + · · · + + c0 + c1 z + · · · c−k 6= 0
z z z
Then z = 0 is a pole of order k.
Suppose we have a function defined in a deleted neighbourhood of ∞. If
lim f (z) = 0 it is reasonable to regard f has zero at the point z = ∞(?). we
z→∞
say it is a zero of order k ≥ 1 in case ‘the same happens at z = 0 when the

44
drama is shifted near zero’. More precisely, define in a deleted neighbourhood
of z = 0, the function g(z) = f (1/z). If z = 0 is a zero of order k for g,
we say infinity is a zero of order k for f . That is g(z) = z k g1 (z) with
g1 (0) 6= 0. equivalently lim g(z)/z k 6= 0. Equivalewntly lim f (1/z)/z k 6= 0.
z→0 z→0
Equivalewntly lim z k f (z) 6= 0 (we have substituted w for 1/z thus as z → 0
z→∞
we have w → ∞ and then replaced the dummy variable w by z). Thus
infinity is a zero of order k for f if lim f (z) = 0 and lim z k f (z) 6= 0.
z→∞ z→∞
Exactly the same nomenclature for pole too. Suppose we have a function
defined in a deleted neighbourhood of ∞. Suppose f (z) → ∞ as z → ∞,
we say it is a pole of order k ≥ 1 in case ‘the same happens at z = 0 when
the drama is shifted near zero’. More precisely, define in a neighbourhood of
z = 0, the function g(z) = f (1/z) (g(0) undefined). If z = 0 is a pole of order
k for g, we say infinity is a pole of order k for f . That is g(z) = g1 (z)/z k
has non-zero finite limit as z → 0. equivalently lim z k g(z) 6= 0. Equivalently
z→0
lim z k f (1/z) is finite and non-zero. Equivalewntly lim f (z)/z k is finite non-
z→0 z→∞
zero. we have substituted w for 1/z thus as z → 0 we have w → ∞ and then
replaced the dummy variable w by z. Thus infinity is a pole of order k for
f if lim f (z) is infinity as z → ∞ and lim f (z)/z k is non-zero finite as z → ∞.

This formalizes the intuitive feeling: If f (a) = 0 how many times can we
divide f by z − a so that we get a finite non-zero limit as z approaches a.
In some sense ‘order of smallness’. (If the function is already small near a,
no use multiplying by (z − a), which is also small near a). Similarly if f
approaches infinity as z aproaches a, how many times I need to multiply by
(z − a) to get a finite non-zero limit as z approaches a. In some sense ‘order
of Largeness’. (If the function is Large near a, there is no use multiplying by
1/(z − a) which is also Large near a).

Even though I started with continuous f , all this will be put to use in the
context of holomorphic functions. In the context of holomorphic functions,
our wishful thinking is true and you have only integer powers of (z − a) when
you expand a holomorphic function near a in terms of powers of (z − a). In
the context of continuous functions, you can think of f (z) = |z − a|1/2 , and
feel a is a zero with fractional order 1/2. Unfortunately, this is not holomor-
phic and has no expansion in powers of (z − a). Of course there are functions
having very very high order of zero, but again in the context of holomorphic
functions, every zero has a specific order – unless the function is the zero
function. We are just looking at some examples, when we start analysis of
holomorphic functions some of these points will be clear.

45
In the context of continuous functions, it is interesting to see that you
can not choose fractional power of (z − a) in a continuous manner in a
neighbourhood of a. Just to get a feel: is there a continuous fucntion f (z)
defined on the unit circle such√that for every z, (f (z))2 = z, that is, can you
choose continuous version of z on unit circle. NO. This is because z = eit
(0 ≤ t < 2π) has two distinct square roots: eit/2 and ei(t+2π)/2 = −eit/2 . If
there is a continuous choice, you have a function s(t) taking only two values
+1, −1 such that f (z) = s(t)eit/2 . But then f and t 7→ eit/2 being continuous
(non-vanishing), you see s(t) is continuosu too. But there is no continuous
function on the the unit circle to {+1, −1} unless it is constant. You can
see that the functions f (t) ≡ eit/2 is not continuous and so is the function
f (t) ≡ ei(t+2π)/2 .

46
BVR/CMI Complex Analysis 22-02-2022

Wishful thinking:
Suppose f has a power series expansion in a neighbourhood of a,

f (z) = c0 + c1 (z − a) + c2 (z − a)2 + · · ·

if f (a) = 0, then c0 = 0. It is quite possible c1 may also be zero. Take the


first k such that ck 6= 0.

f (z) = ck (z − a)k + ck+1 (z − a)k+1 + · · ·

Clearly k is the order of zero at a for f .


Suppose that f defined in a deleted neighbourhood of a has an expansion
in powers of (z − a) with some negative powers too:
c−k c−k+1 c−1
f (z) = k
+ k−1
+· · ·+ + c0 + c1 (z − a) + c2 (z − a)2 + · · ·
(z − a) (z − a) (z − a)
for z 6= a with c−k 6= 0. Here k ≥ 1. First of all; lim f (z) as z → a is infinity.
Further (z − a)k f (z) has a non-zero finite limit as z → a. Thus a is a pole
of order k.
In particular near z = 0 if

f (z) = ck z k + ck+1 z k+1 + · · · ck 6= 0

then z = 0 is a zero of order k. Suppose in a deleted neighbourhood of z = 0


c−k c−k+1 c−1
f (z) = k
+ k−1 + · · · + + c0 + c1 z + · · · c−k 6= 0
z z z
Then z = 0 is a pole of order k.
Suppose we have a function defined in a deleted neighbourhood of ∞. If
lim f (z) = 0 it is reasonable to regard f has zero at the point z = ∞(?).
z→∞
we say the point infinity is a zero of order k ≥ 1 in case ‘the same happens
at z = 0 when the drama is shifted near zero’. More precisely, define in a
deleted neighbourhood of z = 0, the function g(z) = f (1/z). If z = 0 is
a zero of order k for g, we say infinity is a zero of order k for f . That is
lim g(z)/z k 6= 0. Equivalewntly lim f (1/z)/z k 6= 0, finite. Equivalewntly
z→0 z→0
lim z k f (z) 6= 0, finite. (we have substituted w for 1/z thus as z → 0 we
z→∞
have w → ∞ and then replaced the dummy variable w by z).

Thus the point infinity is a zero of order k for f if lim f (z) = 0 and
z→∞
lim z k f (z) 6= 0, finite.
z→∞

47
Exactly the same nomenclature for pole too. Suppose we have a function
defined in a deleted neighbourhood of ∞. Suppose f (z) → ∞ as z → ∞,
we say the point infinity is a pole of order k ≥ 1 in case ‘the same happens
at z = 0 when the drama is shifted near zero’. More precisely, define in a
deleted neighbourhood of z = 0, the function g(z) = f (1/z) (g(0) undefined).
If z = 0 is a pole of order k for g, we say infinity is a pole of order k for
f . That is lim z k g(z) 6= 0, finite. Equivalently lim z k f (1/z) is finite and
z→0 z→0
non-zero. Equivalewntly lim f (z)/z k is finite non-zero. we have substituted
z→∞
w for 1/z thus as z → 0 we have w → ∞ and then replaced the dummy
variable w by z.

Thus the point infinity is a pole of order k for f if lim f (z) is infinity as
z → ∞ and lim f (z)/z k is non-zero finite as z → ∞.

This formalizes the intuitive feeling: If f (a) = 0 how many times can
we divide f by z − a so that we get a finite non-zero limit as z approaches
a. In some sense ‘order of smallness’. (If the function is already small near
a, no use multiplying by (z − a), which is also small near a). Similarly if
f approaches infinity as z aproaches a, how many times I need to multiply
by (z − a) to get a finite non-zero limit as z approaches a. In some sense
‘order of Largeness’. (If the function is Large near a, there is no use mul-
tiplying by 1/(z − a) which is also Large near a). I am saying all this in
words because if you do not understand the meaning and spirit of defini-
tion and if you pay attention to superficial words; you are sure to feel it is
complicated because sometimes I multiply and some times divide by (z −a)k .

Even though I started with continuous f , all this will be put to use in
the context of holomorphic functions. In the context of continuous functions,
you can think of f (z) = |z − a|1/2 , and feel a is a zero with fractional order
1/2. Unfortunately, this is not holomorphic and has no expansion in powers
of (z − a). In the context of holomorphic functions, our wishful thinking is
indeed true and you have only integer powers of (z − a) when you expand f
near a in terms of powers of (z − a). Thus every point a which is a zero of
f has an order, unless the function is identically zero.
Similarly for poles too. For a to be a pole first of all f is defined in a
deleted neighbourhood of a and lim f (z) = ∞ as z → a. Then the point a is
indeed a pole of some order. This is because of the following reason. In a nbhd
of a, f is never zero. Thus g(z) = 1/f (z) is holomorphic. Clearly, the point a
is zero of g, get k, lim g(z)/(z − a)k is nonzero finite Thus lim 1/[f (z)(z − a)k ]
is nonzero finite. Equivalently, lim f (z)(z − a)k is nonzero finite.

48
In the context of continuous functions, it is interesting and useful to see
that you can not choose fractional power of (z − a) in a continuous manner in
a neighbourhood of a. Just to get a feel: is there a continuous fucntion f (z)
defined on the unit circle such√that for every z, (f (z))2 = z, that is, can you
choose continuous version of z on unit circle. NO. This is because z = eit
(0 ≤ t < 2π) has two distinct square roots: eit/2 and ei(t+2π)/2 = −eit/2 . If
there is a continuous choice, you have a function s(t) taking only two values
+1, −1 such that f (z) = s(t)eit/2 . But then f and t 7→ eit/2 being continuous
(non-vanishing), you see that their ratio, s(t), is continuosu too. But there
is no continuous function on the the unit circle to {+1, −1} unless it is con-
stant. You can see that the functions f (t) ≡ eit/2 is not continuous and so is
the function f (t) ≡ ei(t+2π)/2 .

(E). Rational functions

These are functions of the form


P (z)
R(z) =
Q(z)

where P, Q are polynomials and have NO common zeros. Say P is of degree


m and Q is of degree n (both at least one). Of course, the above expression
defines a function only for those z with Q(z) 6= 0. Further it is holomorphic
on this open set.
Let us now assume
X
P (z) = α(z − c1 )m1 · · · (z − ck )mk ; mj ≥ 1, mj = m.
X
Q(z) = β(z − d1 )n1 · · · (z − dl )nl ; nj ≥ 1, nj = n.
where α 6= 0 and β 6= 0. Denoting α/β = k 6= 0 we have

(z − c1 )m1 · · · (z − ck )mk
R(z) = k
(z − d1 )n1 · · · (z − dl )nl

Clearly we can express R(z) = (z − cj )mj R1 (z) where R1 has a finite non-
zero value for z = cj . Thus cj is a zero of order mj for R. similarly R(z) =
R1 (z)/(z − dj )nj where R1 has a finite non-zero value at dj . In other words,
(z − dj )nj R(j) has a finite non-zero limit as z approaches dj . Thus R has a
pole of order nj at dj .

49
How does the function behave as |z| → ∞? Writing in the usual form
(not as factors) we see
am−1 am−2
m−n am + z
+ z2
+ ···
R(z) = z bn−1 bn−2
bn + z
+ z2
+ ···

For m > n we see lim R(z) = ∞ as |z| → ∞ but R(z)/z m−n has finite non-
zero limit. Thus R has a pole of order m − n at infinity. Thus R has m zeros
(mj at cj , 1 ≤ j ≤ k) and m poles (nj at dj , 1 ≤ j ≤ l and m − n at infinity).
For n = m, as |z| → ∞ the function has a finite non-zero value. Thus a
finite non-zero value can be assigned for R at z = ∞. The function has n
zeros and n poles, all finite.
For m < n, we have lim R(z) = 0 as |z| → ∞, but lim z n−m R(z) is finite
and non-zero. In other words R has a zero of order n − m at ∞. Thus R has
n zeros (mj at cj , 1 ≤ j ≤ k and (n − m) at infinity) and n poles (nj at dj ,
1 ≤ j ≤ l).
This neat result that number of zeros equals the number of poles for
rational functions was made possible because we allowed the point infinity
also as part of the family.

50
BVR/CMI Complex Analysis 25-02-2022

(F). fractional linear transformations

These are also called Linear transformations (these are NOT linear trans-
formations). These are also called Moebius maps.
Fix complex numbers a, b, c, d.
az + b
S(z) =
cz + d
To make sense both c, d can NOT be zero. We assume this from now on.
What if ad − bc = 0? Then this is constant map! Why? If c = 0 then
ad = 0 so a = 0 hence S = b/d. If c 6= 0 then az+b cz+d
= db again. We do
not want constant maps. So assume ad − bc 6= 0. So we NORMALIZE:
ad − bc = 1. Can we? Yes. take k ∈ C so that k 2 = ad − bc Replace a, b, c, d,
by a/k; b/k, c/k, d/k. Note k 6= 0. Will this give the same transformation?
Yes. Note that ad − bc = 1 automatically assures you that not both c, d can
be zero. Of course one of them can be zero.
We denote this transformation by S(a,b,c,d) .

Will two different such quadruples give different transformations? No


You can show

S(a,b,c,d) = S(a∗ ,b∗ ,c∗ ,d∗ ) ←→ (a∗ , b∗ , c∗ , d∗ ) = (a, b, c, d) or = (−a, −b, −c, −d)

Convention:
S(∞) = a/c; S(−d/c) = ∞
NO finite complex number z gives the value a/c.

c = 0 then S(∞) = ∞

Thus S maps Riemann sphere to itself. This is Invertible. if Sz = w, then


dw − b
S −1 (w) =
−cw + a
−1
S(a,b,c,d) = S(d,−b,−c,a)
Thus these transformationsare one-one maps of the Riemann sphere or ex-
tended complex plane onto itself. (another advantage of allowing infinity into
our family). Actually this set of transformations is a group.
Think z: z/1 and w: w1 /w2

51
Sz = w or S(z/1) = w1 /w2
    
w1 a b z
=
w2 c d 1
What is the inverse of the above matrix? Note ad − bc = 1
 −1  
a b d −b
=
c d −c a

How do you compose? matrix multiplication. Say we have S with (a, b, c, d)


and T with (a∗ , b∗ , c∗ , d∗ ) then in the above thinking
   ∗ ∗ 
a b a b
S∼ ; T =
c d c∗ d ∗
then
a∗ b ∗ a∗ a + b∗ c a∗ b + b∗ d
    
a b
T oS ∼ = ?
c∗ d∗ c d c∗ a + d ∗ c c ∗ d + d ∗ d
Yes
a∗ [(az + b)/(cz + d)] + b∗ a∗ az + a∗ b + b∗ cz + b∗ d
T oS(z) = =
c∗ [(az + b)/(cz + d)] + d∗ c∗ az + c∗ b + d∗ cz + d∗ d
Thus with the matrix identification, this group of transformations is like 2×2
matrices of determinant one.
Here are some Special cases: I am not normalizing. If I did, you will have
denominators and feel things are complicated. But if you want normalized
quadruple, you should replace, for example (k, 0, 0, 1) by (k ∗ , 0, 0, 1/k ∗ ) where
(k ∗ )2 = k.
(1, θ, 0, 1). Sz = z + θ parallel translations.
In particular, (1, 0, 0, 1) identity map.
(k, 0, 0, 1) with |k| = 1. Sz = kz rotation.
(k, 0, 0, 1) with k > 0. Sz = kz homothety
k
(k, 0, 0, 1). Sz = kz = |k|. |k| z homothety + rotation.
(0, 1, 1, 0). Sz = 1/z. inversion.
Any fractional linear transformation is composition of these.
if c 6= 0, then
1
z 7−→ (z + d/c) 7−→ c2 (z + d/c) 7−→
c2 (z + d/c)
bc − ad bc − ad a az + b
7−→ 7−→ 2 + = .
c2 (z+ d/c) c (z + d/c) c cz + d

52
If c = 0
az + b
z 7−→ (a/d)z 7−→ (a/d)z + (b/d) =
d
az + b
Sz =
cz + d
if S leaves the three points 1.0, ∞ invariant then S must be identity.
Given three distinct points z2 , z3 , z4 put

z − z3 z2 − z3 z − z3 z2 − z4
Sz = =
z − z4 z2 − z4 z2 − z3 z − z4

Then (z2 , z3 , z4 ) go to (1, 0, ∞).


when z2 = ∞ then
z − z3
(∞, z3 , z4 ) 7→ (1, 0, ∞) Sz =
z − z4
when z3 = ∞ then
z2 − z4
(z2 , ∞, z4 ) 7→ (1, 0, ∞) Sz =
z − z4
when z4 = ∞ then
z − z3
(z2 , z3 , ∞) 7→ (1, 0, ∞) Sz =
z2 − z3
Is there other ransformation that does this? NO. If T were another, then
ST −1 leaves 1, 0, ∞ invariant and hence must be identity. That is S = T .
Cross ratio (z1 , z2 , z3 , z4 ) between four DISTINCT complex numbers (no
infinity) is by definition
z1 − z3 z2 − z4
(z1 , z2 , z3 , z4 ) =
z2 − z3 z1 − z4
equivalently it is the image of z1 under the map (Moebius map) that carries
(z2 , z3 , z4 ) to (1, 0, ∞). The cross ratio is invariant under Moebius maps.

6. Functions R to C:

continuity

Suppose you have complex valued function f defined on an interval [a, b] ⊂


R. We say that it is continuous if tn → t implies f (tn ) → f (t). Since we
can write f (t) = x(t) + iy(t) where x, y are real valued, the above definition

53
amounts to saying simply that both the real valued functions x and y are
continuous. Usual rules apply, sum, product etc: cf is conitnuous even if c
is complex.

Differentiation

We say that f is differentiable at a point t0 if

f (t0 + h) − f (t0 )
lim
h6=0;h→0 h

exists (and is finite). Then it is denoted by f 0 (t). Note that in the expres-
sion above h varies over only reals such that t0 + h is in the interval again.
Equivalently, whenever tn → t0 and for all n, tn 6= t0 , then

f (tn ) − f (t0 )
lim
tn − t0
exists (and finite). The function f is differentiable if it is so at every point.
Then we have derivative function t 7→ f 0 (t). If f is defined on a closed
interval [a, b] this means that the left derivative at b and right derivative at
a exist. When we consider differentiable functions the domain is always an
interval (or finite union of intervals).
We say f is smooth if it is differentiable and f 0 is continuous. This is same
as saying that the two real valued functions <f = x, =f = y are differentiable
and x0 and y 0 are continuous functions. Then f 0 (t) = x0 (t)+iy 0 (t) This is also
same as saying that the R2 valued function t 7→ (x(t), y(t)) is differentiable
and the ‘derivative’ is continuous. [If f is column R2 valued then derivative
is row R2 valued].
Usual rules hold sums, products etc: for example (cf )0 = cf 0 even if c is
complex.
If f is defined on an interval [a, b] then it is said to be piecewise smooth,
if it is continuous and there is a finite partition: a = a0 < a1 < · · · < am = b
such that on each [aj , aj+1 ] it is smooth. Of course this last condition already
implies that the function is continuous. Thus a piecewise function need not
be smooth, but on each [aj , aj+1 ] it is smooth. This in particular implies that
the right/left derivatives at each of the intermediate points exist. Of course
at the end points the corresponding derivatives exiet too.

integration

54
Let f : [a, b] → C be continuous. We define
Z b Z b Z b
f (t)dt = <f (t)dt + i =f (t)dt
a a a

Note that <f and =f are real valued and thus on the right side we have only
integrals of real valued functions.
Example: f (t) = eit on [a, b]. Then
b
eib − eia
Z
eit dt =
a i
This is because
1
[sin b − sin a] − i[cos b − cos a] = [i sin b − i sin a + cos b − cos a]
i
Example: f (t) = ezt on [a, b]. Here z is fixed complex number. Of course
when z = 0, then f ≡ 1 and integral is just (b − a). Assume z 6= 0.
b
ezb − eza
Z
ezt dt =
a z

Reason: Fundamental theorem of calculus! if F (t) is such that F 0 (t) = f (t),


then
Zb
f (t)dt = F (b) − F (a)
a

Thisa is because if f (t) = x(t) + iy(t) and F (t) = u(t) + iv(t), then we know
F 0 (t) = f (t) implies u0 (t) = x(t) and v 0 (t) = y(t) so that the real variable
fundamental theorem of calculus gives
Z b Z b
x(t)dt = u(b) − u(a); y(t)dt = v(b) − v(a)
a a

Now definition of integral gives the earlier claim. In the present case we
only need to show derivative of ezt /z (function of t) equals ezt . But this is
immediate from definition of derivative.

55
BVR/CMI Complex Analysis 01-03-2022

Having evaluated the integral without splitting the integrand into real
and imaginary parts, we now split and get two equations about real integrals.
Take z = x + iy 6= 0.

ezt = ext cos(yt) + iext sin(yt)


ezt [ext cos(yt) + iext sin(yt)][x − iy]
=
z x2 + y 2
ext [x cos(yt) + y sin(yt)] ext [−y cos(yt) + x sin(yt)]
= +
x2 + y 2 x2 + y 2
Thus
ext [x cos(yt) + y sin(yt)]
Z
ext cos(yt)dt = ;
x2 + y 2
ext [−y cos(yt) + x sin(yt)]
Z
ext sin(yt)dt =
x2 + y 2
in the sense; to evaluate the integral on left from a to b, you take difference
of values of the function on the right. Of course these can be evaluated using
real variable theory and integration by parts. But see how painless is the use
of complex integration.
Example: f (t) = cos(zt) on [a, b]. Here z is fixed complex number. Of
course when z = 0, then f ≡ 1 and integral is just (b − a). Assume z 6= 0.
Then Z b
sin(zb) − sin(za)
cos(zt)dt =
a z
R R
Usual rules hold: sum, difference, etc. cf = c f even for complex
numbers. You only need to identify <(cf ) and =(cf ) and proceed.
f : [a, b] → C continuous; ϕ : [c, d] → [a, b] strictly increasing smooth,
and g on [c, d] is defined by g(t) = f (ϕ(t)) then g is continuous on [c, d] and
Z d Z b
0
f (ϕ(t))ϕ (t)dt = f (t)dt
c a

In other words ‘change of variable formula’ holds. This is immediate by


considering real and imaginary parts. Even if ϕ is strictly increasing piecewise
smooth, the formula remains true - just observe the same for real integrals.
You may get a doubt: why did we separate real and imaginary parts
and
P use real calculus? Why not define the integral as limit of Riemann sums:
f (tj )[tj+1 −tj ] – usual notation and the limit is as the norm of the partition
goes to zero. Sure, can be done and we get the same answer.

56
Let us observe Z Z Z
f dt = f dt = f dt

dt = dt because t is real. Of course, one needs to prove the above equation.


But This is immediate from definition of integral. I have written the middle
expression just to draw your attention to the fact that what is involved is:
z + w = z + w and zw = z w.

curves/paths

A curve (in C) is a continuous function t 7→ z(t) = (x(t), y(t)) ∈ C of


some compact interval [a, b] ⊂ R. We consider only curves in C.
Peano observed that this is too general a definition. there is a continuous
function on [0, 1] onto the unit square. Thus [0, 1] × [0, 1] is a curve.
A curve is smooth if the map t 7→ z(t) is so . A curve is ‘piecewise smooth’
if the map t 7→ z(t) is so .
From now on we consider only piecewise smooth curves. A curve z(t) is
said to be simple if z(s) 6= z(t) whenever s 6= t (in the domain of z). A curve
defined on [a, b] is closed if z(a) = z(b). A curve defined on [a, b] is said to
simple closed curve if it is closed and z(s) = z(t) only when s, t ∈ {a, b}.
Thus a simple closed curve is one-one function on [a, b) and z(b) = z(a).
Simple closed curves are called Jordan curves. Of course we are restricting
to piecewise smooth curves but the general definition of Jordan curve does
not do so.
z1 (t) = c + reit for 0 ≤ t ≤ 2π. Here c ∈ C and r > 0. Clearly
this is circle with center c and radius r – travelled in the counter clockwise
direction. z2 (t) = c + re−it for 0 ≤ t ≤ 2π. Here c ∈ C and r > 0. Clearly
this is also circle with center c and radius r as above – but travelled in the
clockwise direction. Consider z3 (t) = c + re2ti for 0 ≤ t ≤ π. This is also
the same curve, direction just as z1 but travelled with ‘double the speed’.
z4 (t) = c + re2ti for 0 ≤ t ≤ 2π. The image of this curve is same as the
previous ones, circle center c and radius r; however this curve goes around
the circle twice. Clearly, this is not simple curve, though closed.
Here is a curve defined on the interval [0, 4]: z(t) = (t, 0) for 0 ≤ t ≤ 1
and z(t) = (1, t − 1) for 1 ≤ t ≤ 2 and z(t) = (3 − t, 1) for 2 ≤ t ≤ 3
and z(t) = (0, 4 − t) for 3 ≤ t ≤ 4. This is the boundary of the unit
square travelled in the anticlockwise direction. You can think of the square
travelled in the clockwise direction too. This is not smooth, but piecewise
smooth. Such curves as this which are made up of straight line pieces are
called polygonal lines or polygonal curves.

57
So what exactly is the curve: is it the function or range of the function?
well, it is both! We denote the image by Γ or by C etc. But you should
keep in mind the function too which tells you the direction of the curve.
For example the two circles described by z1 and z2 above are same but the
directions are different. As long as the direction is understood, you can keep
the image in mind. If the curve is defined on [a, b], then z(a) is the initial
point and z(b) is the terminal point of the curve.
Allowing piecewise smooth curves makes it possible to develop some arith-
metic of curves. Suppose C1 is a curve and C2 is a curve. if the end point
of C1 is same as the initial point of C2 , you can define the curve C1 + C2 .
Thus if C1 is defined on [a, b], by z1 (t) and C2 is defined on [c, d] by z2 (t)
then C1 + C2 is the curve defined on [a, b + d − c] by z(t) = z1 (t) for a ≤ t ≤ b
and z(t) = z2 (c + t − b) for b ≤ t ≤ b + d − c. Note thatthe first part of
the curve traces C1 and from b onwards it traces the curve C2 . This is a
good definition because of the assumption that z1 (b) = z2 (c). Similarly you
can define C1 + C2 + · · · + C100 provided the terminal point of Cj equals the
initial point of Cj+1 . when we write C1 + C2 + · · · + C100 it is assumed that
the condition for concatenation is satisfied.
You can also define −C for a curve C, just reverse the direction. If C
is given by z(t) for a ≤ t ≤ b, then −C is the curve z ∗ (t) = z(−t) for
−b ≤ t ≤ −a. Our idea is not to develop theory of curves and their arith-
metic, but to put them to use. Some more aspects we shall develop as we go
along.

open connected sets are path connected.

Let Ω be an open connected set. Then given any two points in Ω there
is a path in Ω connecting the two points. We can choose path so that it is a
polygonal line made of pieces each parallel to either x-axis or y-axis.
Proof is simple. Let us start with the observation that any two points in
an open disc can be connected by polygonal path with each piece parallel to
one of the axes.
Let Ω be an open connected set. Fix a ∈ Ω. consider the set A of all
points in Ω which can be connected by a special polygonal line (that is, sides
parallel to axes) to a. And B those points which can not be. Shall show
both are open, so that one must be empty. A 6= ∅ because there is r > 0
such that D(a, r) ⊂ Ω and clearly D(a, r) ⊂ A. Suppose b ∈ A. Fix r > 0
such that D(b, r) ⊂ Ω. we claim D(b, r) ⊂ A. Indeed any point in D(b, r)
can be connected to the center by special polygonal line. and hence can be
connected to a too. Suppose b ∈ B, then D(b, r) ⊂ B by the same reason.
This completes the proof.

58
DEF: A region in C is an open connected non-mepty set.
Every nonempty open set is union of countably many regions. In fact
given Ω an open set, consider z ∈ Ω. The connected component containing z
be denoted by Ωz . Clearly, Ωz is open (use path connected). Also if z 6= w,
then either Ωz = Ωw or Ωz ∩ Ωw = ∅. Thus if you take all the distinct Ωz you
get disjoint open sets with union Ω. This is countable by separability of the
space.
This is just analogous to the fact you must have proved in real analy-
sis: Any non-empty open subset of R is a countable disjoint union of open
intervals. In R, the only non-empty connected sets are intervals (including
singletons, semi-open etc.)

59
BVR/CMI Complex Analysis 04-03-2022

Integration along curves:

suppose we have a curve C defined by z(t) on [a, b]. Let f be a continuous


function whose domain includes the curve (image). We define
Z Z b
f (z)dz = f (z(t))z 0 (t)dt.
C a

Note that the integral is well defined though z 0 is probably not defined at
finitely many points. Notation:
Z Z Z
f f dz f (z)dz
C C C

Does the integral depends on how you ‘parametrize’ the path? For example
you have a path C given by z : [0, 1] → C. Define curve C ∗ by z ∗ : [10, 100] →
C by z ∗ (t) = z( t−10
9
). You see that the two paths are the same. RSo it is
reasonable to ask if we have a nice function f on the image then is f dz =
C
f dz ∗ ? Yes.
R
C∗
Change of variable formula holds. Have a curve C defined by z : [a, b] →
C. Have a continuous; ϕ : [c, d] → [a, b] strictly increasing smooth. Define a
curve C ∗ given by z ∗ with domain [c, d] and z ∗ (t) = z(ϕ(t)). Then
Z Z Z b Z d
∗ 0
f dz = f dz ; i.e. f (z(t))z (t)dt = f (z ∗ (t))z ∗0 (t)dt (♠)
C C∗ a c

this is true because z ∗0 (t) = z 0 (ϕ(t))ϕ0 (t). and usual change of variable for-
mula tells
Z d Z d
∗ ∗0
RHS(♠) = f (z (t))z (t)dt = f (z(ϕ(t)))z 0 (ϕ(t))ϕ0 (t)dt
c c
Z b
= f (z(u))z 0 (u)du = LHS(♠).
a

by the change of variable u = ϕ(t). Even if ϕ is strictly increasing piecewise


smooth, the formula remains true - just observe the same for real integrals.
We treat two curves C on [a, b] given by z and C ∗ on [a∗ , b∗ ] given by z ∗ same
if there is a piecewise smooth strictly increasing ϕ : [a∗ , b∗ ] → [a, b] such that
z ∗ (t) = z(ϕ(t)).

60
Integral is additive in both its arguments: path and function. Let us first
fix the curve C, then
Z Z Z
(af + bg) = a f + b g
C C C
R
In usual calculus if |f | ≤ M then | f | ≤ M L(I) where L(I) is length of the
I R
interval I. The same is true here too; | f | ≤ M L(C). We need to define
C
length of C. Let the curve be: z(t) = (x(t), y(t)) : [a, b] → C. Take a finite
partition of [a, b] : π = {a = t0 < t1 < · · · < tk = b}. Consider
X
L(π) = |z(tj+1 ) − z(tj )| L = sup L(π)
π

Observe that the length of the curve, no matter how you define, must be at
least as much as L(π) – straight line being curve of shortest length between
two points. The supremum over all π is called the length of the curve. note
that if η is a refinement of π then L(η) ≥ L(π).
If L is finite we say that the curve is rectifiable.

Fact: L(π) → L as kπk → 0.


Here kπk is norm of the partition: max |tj+1 − tj |.

Here is the proof. Fix  > 0. Take a partition π such that L(π) > L − .
Let N be the number of points in the partition. Using uniform continuity of
z fix δ > 0 such that |z(s) − z(t)| < /2N whenever |s − t| < δ. Now take
ant partition π 0 with kπ 0 k < δ. We show L(π 0 ) ≥ L − 2. Let η = π ∨ π 0 .
Calculation of L(η): Look at all subintervals of η. Those that have one end
point from π will be at most 2N and each such interval conbtributes at most
/2N . Thus these intervals together contribute at most . The other intervals
are already intervals of π 0 and hence their total contribution is at most L(π 0 ).
Thus L(η) ≤ L(π 0 ) + . Thus using L(η) ≥ L(π),

L(π 0 ) ≥ L(η) −  ≥ L(π) −  ≥ L − 2

Since we always have L(π 0 ) ≤ L, this completes the proof.


You must understand importance of such theorems. Afterall L is defined
as sup of something. But that necessitates considering all finite partitions.
This theorem tell you, you can take any sequence opf partitions convenient
to you with norm going to zero. It is enough ot calculate for these and take
limit. You need nopt worry about other partitions.

61
If the curve is smooth then it is rectifiable and L = |z 0 (t)|dt. In fact
R

any L(π) is an appropriate Riemann sum for this integral and it converges
to the integral as norm goes R to zero. Even if the curve is piecewise smooth,
0
it is rectifiable and L = |z (t)|dt. Thus restricting ourselves to piecewise
smooth curves we are considering, in particular, rectifiable curves.
This circle of ideas lead to the definition of integral over C as usual limit
of Riemann sums.
P In fact take a partition π of [a, b]. Define the Riemann
sum R(π) = f (z(tj ))[z(tj+1 ) − z(tj )]. Then we can show R(π) converges
Rb
to f (z(t))z 0 (t)dt = f .
R
a c
Returning to integral: Suppose |f | ≤ M on the curve C. Then
Z
| f | ≤ M L(C)
C

This is because
Z Z b Z b
0
| f| ≤ |f (z(t))z (t)|dt ≤ M |z 0 (t)|dt = M L(C)
C a a

if the curve is z(t) = x(t) + iy(t) and f = u + iv then


Z Z Z Z Z Z Z
f dz = f dx + i f dy = ( udx − vdy) + i( udy + vdx) (•)

This is also written as


Z Z Z
f dz = (udx − vdy) + i (vdx + udy)
R
which identifies the real and imaginary
R parts of the integral f dz. This also
allows the possibility of defining f dz. If dz = dx + idy then symbolically
dz = dx − idy. Thus we can define
Z Z Z Z Z
f dz = f dx − i f dy = (udx + vdy) + i (−udy + vdx) (••)

You should carefully notice that we are using dz and not d z. It is quite true
that when z is a path then then z(t) = z(t) is another path and it makes
sense to integrate on this path. But the trouble is that f may not be defined
on this new path and even if Rit is defined integrating over this new path,
would not (in general) lead to f dz. We see from (•) and (••)
Z Z Z Z
1 dz + dz
f dx = [ f dz + f dz] = f
2 2

62
and
dz − dz
Z Z Z Z
1
f dy = [ f dz − f dz] = f
2 2i
Thus dx = (dz + dz)/2 and dy = (dz − dz)/2i reflecting x = (z + z)/2 and
y = (z − z)/2i. These definitions also lead to
Z Z
f dz = f dz
R R R
The integrals of the form udx, vdy are called line integrals. udx is noth-
Rb Rb
ing but the one variable integral u(x(t), y(t))dx(t) = u(x(t), y(t))x0 (t)dt
a a
where [a, b] is the interval on which the curve (x(t), y(t)) is defined and u is a
function whose domain includes the curve (image). These appear in physics
in calcualting work done while moving a particle along a path. The main
point is; if a particle has to go from a point P to a point Q how does it go,
which path does it take? For each path from P to Q it calculates the work
needed to go along that path and selects the path that minimizes the work;
very clever indeed!

If you fix the function f , then


Z Z Z
f= f+ f
C1 +C2 C1 C2

This simply reflects for a < b < c


Z Z b Z c
0 0
f (z(t))z (t)dt = f (z(t))z (t)dt + f (z(t))z 0 (t)dt
abc a b

we also have Z Z
f =− f
−C C
This reflects the fact: if z ∗ on [−b, −a] is defined by z ∗ (t) = z(−t), then
Z −a Z b
∗ ∗0
f (z (t))z (t)dt = − f (z(t))z 0 (t)dt
−b a

Is there analogue of fundamental theorem of calculus? Given f (z) in a


region (connected non-empty open set) Ω. firstly, Is there a primitive?; is
there a F such that for two points w1 , w2 in the region and for any curve C
within the region joining w1 → w2
Z
f = F (w2 ) − F (w1 )??
C

63
Observe that when this happens the above integral of f is ‘independent’ of
the path, as long as it lies in Ω. This is sufficient too.
Theorem: Ω is a region. p, q real continuous functions defined on Ω.
following are equivalent: R
(i) The line integral (pdx + qdy) depends only on the end points of C
C
for every path in Ω.
(ii) pdx + qdy is an exact differential, that is, there is a function U such
∂ ∂
that ∂x U = p and ∂y U = q.
R
(iii) (pdx + qdy) = 0 for any closed path C in Ω.
C
The theorem holds even if p, q are complex.
Just keep in mind we are here talking about partial derivatives and not
holomorphy.

64
BVR/CMI Complex Analysis 08-03-2022
R
In proving, | f dt| ≤ M L(C), we used the following: Let f be a complex
C
continuous function on an interval [a, b]. Then
Z b Z b
| f (t)dt| ≤ |f (t)|dt
a a

This is immediate by considering


P PRiemann sums for the two integrals and
using the inequality | aj | ≤ |aj |. Here is another proof. Put α =
Rb
f (t)dt and θ = α/|α|. Then |θ| = 1 and |α| = αθ.
a
Z b Z b Z b
| f (t)dt| = θ f (t)dt = θf (t)dt
a a a

by linearity of integral Z b
= <(θf )dt
a
because the integral is real, look at the initial term in the string of equalities;
and so integral of imaginary part is zero.
Z b
≤ |<(θf |dt
a
R R
for real functions we know h≤ |h|
Z b
≤ |θf |dt
a

because |<(θf )| ≤ |θf | and for real functions ϕ ≤ ψ implies integrals obey
the same, Z b
= |f |dt
a
because |θ| = 1. This completes the proof.
Why do we need two proofs? When the first does not work, then second
helps. Further if you want to know as to when equality holds in the stated
inequality, then the second proof is helpful.

Thus if z is the curve defined on [a, b] and f is defined on the curve with
|f | ≤ M then
Z Z b Z b
0
| f| = | f (z(t))z )(t)dt| ≤ |f (z(t))z 0 (t)|dt
C a a

65
equality is definition of integral over C, inequality is what we saw now
Z b
≤M |z 0 (t)|dt
a
0 0
because |f (z(t))z (t)| ≤ M |z (t)| and for real functions we know, ϕ ≤ ψ
implies integrals obey the same,
= M L(C)
because we identified length of the curve with |z 0 |.
R

Returning to discussion of primitive here is the theorem.

Theorem 0: Ω is a region. p, q real continuous functions defined on Ω.


following are equivalent: R
(i) The line integral (pdx + qdy) depends only on the end points of C
C
for every path in Ω.
(ii) pdx + qdy is an exact differential, that is, there is a function U such
∂ ∂
that ∂x U = p and ∂y U = q.
R
(iii) (pdx + qdy) = 0 for any closed path C in Ω.
C
The theorem holds even if p, q are complex.
Just keep in mind we are here talking about partial derivatives and not
holomorphy. Proof is simple: (i) and (iii) are clearly equivalent. Given two
curves C1 , C2 joining w1 , w2 , then C1 + (−C2 ) is a closed curve (starts and
ends at w1 ). Thus if you are given (iii) then
Z Z Z
− f= f =0
C1 C2 C1 +(−C2 )

Conversely given (i) and any closed curve C, take two points w1 , w2 and
consider the part of the curve from w1 → w2 as C1 and the remaining part
w2 → w1 as −C2 and use integrals over C1 and −C2 are equal to see integral
over C1 + C2 = C is zero.
[How do you explain without using words like ‘remaining part’ etc? Let
C be defined on [a, b] with z(a) = z(b). Fix a < c < b. Let C1 be the curve
defined on [a, c] by z1 (t) = z(t). Let D be the curve defined on [c, b] by
η(t) = z(t). Let C2 = −D.]

Given (ii) and a curve C = (x(t), y(t)) defined on [a, b] observe by rules of
calculus; the composed function t 7→ (x(t), y(t)) 7→ U (x(t), y(t)) has deriva-
tive,
d ∂U ∂U
U (x(t), y(t)) = (x(t), y(t)) x0 (t) + (x(t), y(t)) y 0 (t)
dt ∂x ∂y

66
= p(x(t), y(t))x0 (t) + q(x(t), y(t))y 0 (t)
so that Z Z
pdx + qdy = (px0 + qy 0 )dt
C
Z b
d
= U (x(t), y(t))dt = U (x(b), y(b)) − U (x(a), y(a))
a dt
Thus (i) holds.
Conversely if (i) holds, fix a point w ∈ Ω. Define
Z
U (z) = pdx + qdy; z = (x, y)
C(w,z)

where C(w, z) is any path w → z. This is good definition because the value
of the integral does not depend on the path joining w to z.
∂ ∂
Shall show ∂x U = p and ∂y U = q Fix (x0 , y0 ). Fix h0 > 0 such that the
horizontal line segment (x0 − h0 , y0 ) → (x0 + h0 , y0 ) is within Ω. We consider
h with |h < h0 | from now on. To calculate U (x0 + h, y0 ) − U (x0 , y0 ) we
consider any path w → (x0 , y0 ) as C(x0 , y0 ). Having done this we consider
C(x0 + h, y0 ) to be the previous path extended by the horizonal line segment
(x0 , y0 ) → (x0 + h, y0 ); the extension parametrized on the interval [0, h] in
the obvious way. This allows us to compute

U (x0 + h, y0 ) − U (x0 , y0 ) 1 h
Z
= p(x0 + t, y0 )dt
h h 0
Note that integral upto x0 gets cancelled and R only from x0 to x0 + h remains;
further, the curve being horizontal line the qdy part is then zero (y 0 = 0 in
this part). Now the usual one variable theorem of calculus shows that this
limit equals p(x0 , y0 ) as required. The theorem used is: If ϕ is a continuous
Rh ∂
function on [0, h0 ] then ϕ(t)dt/h converges to ϕ(0) as h → 0. Thus ∂x U=
0

p. similarly we can show ∂y U = q.
The case p, q complex is obtained by considering the real and imaginary
parts separately to get U and V and then taking U + iV .

67
An interesting special case arises if we take a continuous complex function
f in a region Ω and take p(z) = f (z) and q(z) = if (z). Thus we want
f dx + if dy to be exact differential. This is so iff there is an F such that
∂ ∂
F = f; F = if
∂x ∂y
But this means, in particular
∂ ∂
F = −i F
∂x ∂y
but this is is precisely C-R equations. Indeed if F = U + iV then the above
equation means
∂U ∂V ∂U ∂V ∂V ∂U
+i = −i[ +i ]= −i
∂x ∂x ∂y ∂y ∂y ∂y
But then f + if = F 0 (z) equating real and imaginary parts we have the C-R
equations. Thus the condition for f dx + if dy to be exact differential is that
it is derivative F 0 (z) of an analytic function F (z).
Writing f dx + if dy = f (dx + i dy) = f dz we could
R have expressed
this as: f (z)dz is an exact differential (equivalently f depends only on
C
endpoints of C) iff f is the derivative of an analytic function. Later after
local power series expansion, we see derivative of an analytic function is itself
analytic.
Here are further special cases of the above. If n ≥ 0 is an integer, then
(z − a)n is indeed derivative of (z − a)n+1 /(n + 1) so that
Z
(z − a)n dz = 0 n≥0 C closed curve.
C

Even if n ≤ −2 the same holds for any curve not passing through a, take the
region {a}c .
Z
(z − a)n dz = 0 n ≤ −2 C closed curve a 6∈ C.
C
R
In particular, taking a = 0, we see we have C z n dz = 0 for n 6= −1; wiht the
additional hypothesis thst for n negative we should have 0 6∈ C.
what about n = −1? If C is circle with center a, say, z(t) = a + reit for
t ∈ [0, 2π]; then Z Z 2π
1 1
dz = it
ireit = 2πi.
C z − a 0 re

69
On the other hand if C ∗ : z(t) = a + re−it for t ∈ [0, 2π] – clockwise circle;
then Z Z 2π
1 1
dz = it
− ireit = −2πi
C ∗ z − a 0 re
. as expected.
Before getting to the study of holomorphic functions, two more things
about curves and length you should know. Did not mention this earlier
because you already need
R to get used to dz, dz, d z, dx and dy-integrals!
(i) we can define f |dz|. C : z : [a, b] → C and f defined on the curve.
C
Then we define Z Z
f (z)|dz| = f (z(t))|z 0 (t)|dt

This can also be defined using Riemann sums too. The increments in usual
dt-integral are (tj+1 − tj ) and in dz-integral they are z(tj+1 ) − z(tj ) and so
on. Now the increments are |z(tj+1 ) − z(tj )|. Thus,
Z X
f (z)|dz| = lim f (z(tj ))|z(tj+1 ) − z(tj )|
kπk→0

In other words, One can show that the above two displays give the same
value.
(ii) ‘arc-length parametrization’: Suppose z is a smooth curve defined on
[a, b] which is non-constant on any interval. Suppose its length is L. Then
you can parametrize it on [0, L] so that length of the curve upto t is t for
Rt
t ∈ [0, L]. You simply consider the function s(t) = |z 0 (u)|du. It is one-one
a
(remember z non-constant on any interval) strictly increasing on [a, b] onto
[0, L] and is smooth Thus ϕ = s−1 : [0, L] → [a, b] is smooth and strictly
increasing. Define z ∗ on [0, L] by z ∗ (t) = z(ϕ(t)). This does. Verification is
easy.
If you pretend particle travelling along the curve, then the parametriza-
tion z ∗ gives unit speed to the particle. With this parametrization on [0, L],
the length of the curve upto t equals t. Thus total length is L. This is called
‘arclength parametrization’ and is many times useful.

What did we do so far? 1 motivation: something about origins of Com-


plex analysis. 2 complex numbers: Several avatars of complex numbers and
working with them. 3. Sequences/Series open/closed sets: beginning of anal-
ysis of numbers; 4. continutiy and differentiation: beginning of analysis of
functions; 5. Exp and all that: some examples like polynomials, exponential,
sine, cosine, logarithm, rational functions, poles and zeros and their orders;

70
Moebius maps; and 6. Functions R → C: about curves and integration along
curves and properties. Even though the drama of integration along curves is
played in C or R2 , the integrals are just high school one dimensional integrals
over a closed bounded interval.
The actual flavour of complex analysis started appearing in our last the-
orem about primitive and fundamental theorem of calculus. Now we shall
enter study of holomorphic functions.

7. Cauchy

We shall now prove the first basic theorem about holomorphic functions,
namely, Cauchy theorem or Cauchy-Goursat Theorem: if a holomorphic func-
tion is defined on a region that includes a simple closed curve and ‘its interior’
then integral of f over the curve is zero. But of course, you need to explain
what is interior for the simple closed curve. Though the curve is simple,
explaining its interior is not simple. However this can be done for rectangle,
disc, polygon etc, it is difficult to define for general curves. This is Jordan
curve theorem. However, much theory can be developed without getting into
this aspect nd considering only ‘nice’ curves and ‘nice’ regions (and without
loosing power of the theory).

THEOREM 1: (Cauchy-Goursat): Let Ω be a convex region; f holo-


morphic on Ω; R ⊂ Ω a closed rectangle. Then
Z
f =0
∂R

Note that the region being convex, if we assume that ∂R ⊂ Ω then the
interior of the rectangle is in Ω.
For a rectangle interior is easy to precisely define. (Do it.)

71
BVR/CMI Complex Analysis 11-03-2022

Proof of the theorem is by contradiction. Let us assume


R that it is not
true. For any rectangle in this convex region we name f = η(R). Thus we
∂R
have an R for which |η(R)| = α > 0. Fix one such. Divide this rectangle into
four rectangles by considering the midpoints of the sides. A simple algebra
shows that integral of f over the four smaller rectangles equals integral over
R. Thus there must be one of these smaller rectangles, say R1 such that
|η(R1 )| ≥ α/4. Do the same with R1 to get R2 with |η(R2 )| ≥ α/42 . Thus
you can get a decreasing sequence of rectangles Rn such that Rn is one of the
four smaller rectangles when Rn−1 is subdivided at midpoints of sides and

|η(Rn )| ≥ α/4n ; i.e. 4n |η(Rn )| ≥ α (♠)

Let us note two facts. If length of diagonal of R equals d then length of


diagonal of Rn equals dn = d/2n . If perimeter of R equals L then perimeter
of Rn equals Ln = L/2n .
By Cantor intersection theorem, there is a point common to all the rect-
angles, say w. Keep in mind, no matter what open set you take around w,
all rectangles Rn after some stage are contained in the open set.
Let us now take  = α/(2dL), towards showing contradiction.
(whenever you see such choices do not get puzzled. Initially, being ignorant,
we take  > 0 anything; then do rough calculations and see if a clever choice
of  leads to contradiction, Then come back and make that choice.)
By definition of derivative we can choose δ such that 0 < |z − w| < δ then

f (z) − f (w)
| − f 0 (w)| < .
z−w
Thus
|z − w| < δ ⇒ |f (z) − f (w) − (z − w)f 0 (w)| ≤ |z − w|
Let us now consider n so large that dn < δ. Thus for any point z ∈ Rn , we
have |z − w| < δ. Now, observe (remember w, f (w) and f 0 (w) are numbers,
integration below is dz)
Z Z
f (w)dz = 0; (z − w)f 0 (w)dz = 0
∂Rn ∂Rn

So that we have
Z
η(Rn ) = [f (z) − f (w) − f 0 (w)(z − w)] dz
∂Rn

73
By our choice of n, the integrand is bounded by |z − w| ≤ dn , because z
and w are in Rn , and the distance is at most length of the diagonal. Also
length of the curve ∂Rn equals Ln . Hence

|η(Rn )| ≤ dn Ln = dL/4n

Thus for all large n we have

4n |η(Rn )| ≤ dL ≤ α/2 (♣)

since (♠) holds for all n; and (♣) holds for all large n; we have a contradic-
tion. Proof is complete.

THEOREM 2 (C-G): Let Ω be a convex region; F ⊂ Ω a finite subset;


f is defined and holomorphic on Ω \ F . Assume that lim (z − ζ)f (z) = 0 if
z→ζ
ζ ∈ F . Let R ⊂ Ω a closed rectangle ∂R ∩ F = ∅. Then
Z
f =0
∂R

Remember Ω \ F is also a region, not convex.


Proof is simple. Firstly it is enough to consider when F is a singleton.
This is because you can divide R into rectangles so that the boundaries do not
contain any point of F and each of these rectangles contains at most one point
of F . Remember integral over original ∂R is sum of these integrals over the
boundaries of smaller rectangles. If a smaller rectangle does not contain any
point of F , then the integral over its boundary is zero by previous theorem
– see (*) before end of proof. Thus we need only handle other rectangles .
Thus you can assume F = {ζ} and ζ ∈ R – see (**) before end of proof. Let
 > 0. We show integral over ∂R is at most .
Using hypothesis, choose δ > 0 so small that

|z − ζ| < δ ⇒ |(z − ζ)f (z)| < /8.

Now divide your rectangle into nine subrectangles so that the part R0 con-
taining ζ is a square with ζ at the center, side is δ/2. Actually you first fix a
small square as stated around ζ and extend its sides to meety the rectangle
R. This divides R into nine subrectangles. Since the integral over boundaries
of other rectangles is zero. It suffices to show integral over R0 is at most .
Since side of the square R0 is δ/2, and ζ is at the center of the square, if
we take z on the boundary of R0 then |z − ζ| ≥ δ/4. we have by choice of δ,
 1 4 
z ∈ ∂R0 ⇒ |f (z)| ≤ ≤ =
8 |z − ζ| 8δ 2δ

74
Also length of ∂R0 is exactly 2δ. As a result
Z Z

| f dz| ≤ 2δ =  | f dz| < 
R0 2δ R

as stated. This completes the proof.

(*) one needs to be careful. Just because one of the smaller rectangles R0
does not include any point of F you can not right away say that integral over
∂R0 is zero. Because the previous theorem needs a convex region Ω where
f is defined and a rectangle in that. Eventhough present Ω is convex, We
have f defined only on Ω \ F . Thus application of previous theorem is not
justified. what you do is the following. Since F is finite, and each point of F
is outside the compact set R0 , its distance from R0 is positive. Take minimum
of all the distances, say α. Thus every point of F is at a distance larger than
α from R0 . Now enlarge this R0 to a open rectangle R∗ enlarging each side
by α/8. Thus you have a convex region R∗ on which f is holomorphic and
R0 is a rectangle in the region. Previous theorem applies.
(**) It is alright to assume that our rectangle R contains only one point of
F . But can we assume that F itself consists of just one point and that it
is in our rectangle? Yes, argue as in previous para, get a little bigger open
rectangle R∗ ⊃ R that does not contain any other point of F . Now R∗ is a
convex region and restrict drama to this and note f is defined on R∗ \ {ζ}.
These are simple matters but should be well understood.

75
BVR/CMI Complex Analysis 22-03-2022

In the previous theorems we had integral over rectangles. Can we not do


for closed curves. Yes, but then we restrict to closed curves within a disc.
THEOREM 3 C-G:
f is holomorphic in a disk D and C ⊂ Ω closed curve; then
Z
f =0
C

Proof: We use the theorem above for rectangles to show f (z)dz is exact
differential. Then the basic theorem on primitives allows us to conclude that
integral over C is zero.
Fix w ∈ D. For every z ∈ D we define two curves Cz : go from w vertically
to match y-coordinate of z, then horizontally to z. Similarly Cz∗ is the curve:
go from w horizontally to match x-coordinate of z, then vertically to z. Note
that the fact that we have a disc allows us to say that these curves are in
D. Since integral of f over a rectangle is zero by previous theorem and since
Cz + (−Cz∗ ) is a rectangle starting and ending at w we see
Z Z
f= f = F (z) say.
Cz Cz∗

Then as in the basic theorem, we can show using curve Cz that ∂F ∂x


= f and
using the curve Cz∗ that ∂F
∂y
= if . Thus f + if is exact differential and F is
dF
holomorphic and dz = f . Now the basic theorem (Theorem 0) gives us the
stated result.

Do we need a disk for the above proof? For example upper half plane is
also fine. What we needed was that some rectangles needed should be here.
In the same vein we shall have one more theorem that allows exceptional
points. Just as theorem 2 is a strengthening of Theorem 1 with some ex-
ceptional points, we gave next theorem which is strengthening of Theorem 3
where some exceptional points are allowed.

76
THEOREM 4 C-G:
D is an open disk and F ⊂ Ω a finite subset; f is defined and holomorphic
on D \ F . Assume that lim (z − ζ)f (z) = 0 if ζ ∈ F . Let C ⊂ D a closed
z→ζ
curve not passing through any point of F . Then
Z
f =0
C

Proof similar to that of Theorem 2 using theorem 1. WE show that f is


an exact differential, using Theorem 3 and apply basic theorem (Theorem 0).
the only thing is that instead of two paths making one rectangle (see proof of
theorem 3), you may need to take two paths making two or three rectangles.
Remember the path should not pass through the points of F . Not difficult,
shall prove later. Too many C-G theorems without knowing their use would
be boring. So let us proceed to see the splender of holomorphic functions
and then return to this theorem.
After so many C-G theorems you would wonder what exactly is the na-
ture of the result? What is its final form? As mentioned earlier: If Ω is
simply connected and f holomorphic in Ω and C is a closed path in Ω; then
integral of f over C is zero. You would wonder what are the refinements
involving exceptional points. Actually those are not exceptional points!

THEOREM 5: winding number


C a piecewise smooth curve given by z(t). a 6∈ C. Then
Z
dz
is an integer multiple of 2πi.
C z −a

Proof: Remember that if z is defined on [α, β] then


Z β
z 0 (t)
Z
dz
= dt
C z −a α z(t) − a

While reading such an equation you should pause and make it clear to yourself
that we are dealing with piecewise smooth curves and so z 0 is undefined at a
few points, corners (left and right derivatives are defined).
Let us define the function
Z t
z 0 (u)
h(t) = du, α ≤ t ≤ β.
α z(u) − a

This is a continuous function on [α, β]. It is differentiable, except at the


corners of z and
z 0 (t)
h0 (t) = ; z 0 (t) = h0 (t)[z(t) − a] (•)
z(t) − a

78
Set
ϕ(t) = e−h(t) [z(t) − a]
so that
ϕ0 (t) = −e−h(t) h0 (t)[z(t) − a] + e−h(t) z 0 (t) = 0
in view of (•). Keep in mind this happens at points other than corners. Thus
ϕ(t) is a constant between corners. But ϕ being continuous, it must be a
constant on all of [α, β]. But h(α) = 0 so that ϕ(α) = z(α) − a. Thus

e−h(β) [z(β) − a] = ϕ(β) = [z(α) − a].

But z(β) = z(α) and a is not on the curve, so that we have

e−h(β) = 1; h(β) a multiple of 2πi

as stated.
Did you see how instead of dealing with number stated in the theorem,
we defined a function and studied its properties and concluded about our
number.

1 dz
R
DEF: 2πi z−a
= n(C, a) is called the index or winding number of the
C
point a w.r.t. the curve C or winding number of the curve C w.r.t. the point
a.
Here are some examples explaining why it is called winding number.
Example: z(t) = eit for 0 ≤ t ≤ 2π a = 0 (any point inside circle would
do) Then winding number equals 1.
Example: z(t) = e2it for 0 ≤ t ≤ 2π a = 0. Then winding number equals
2
Example: z(t) = eit for 0 ≤ t ≤ 2π and = e−i(t−2π) for 2π < t ≤ 4π and
= ei(t−4π) for 4π < t ≤ 6π. a = 0 Then winding number equals 1
Example : z(t) = e2it for 0 ≤ t ≤ 2π and = e−i(t−2π) for 2π < t ≤ 4π and
= ei(t−4π) for 4π < t ≤ 6π. a = 0. Then winding number equals 1

Here are some properties of winding numbers.


1. n(−C, a) = −n(C, a).
This immediate from change of variable formula.
2. constant in each region of C c .
Remember that C c is an open set and is a union of countably many
disjoint connedcted open sets, regions. Shall show that n(C, a) in each region
continuous. Fix one region. Take a disc D contained in the region. get γ

79
with d(z, w) ≥ γ for z ∈ C, w ∈ D. This is possible becasusewe have a closed
set and compact set. For any a, b in this disk,
Z β
z 0 (t)(a − b)
Z Z
dz dz
| − |=| |
z−a z−b α (z(t) − a)(z(t) − b)

L(C)
≤ |a − b|
γ2
This does.
3 For unbounded region winding number is zero. This is because you can
take large a in the denominator and argue.
4. If C is circle and a outside the circle, then winding number is zero.
5 This is invariant if you translate the point and curve by same number.

80
BVR/CMI Complex Analysis 25-03-2022

Here now is a BASIC Theorem:

THEOREM 6 Cauchy Integral formula:


Ω be an open disc; f be holomorphic in Ω; C be a closed curve in Ω,
a ∈ Ω \ C, then Z
1 f (z)
dz = n(C, a) f (a)
2πi C z − a
In particular, if n(C, a) = 1, then we have
Z
1 f (z)
f (a) = dz
2πi C z − a

In particular, if a closed disc D ⊂ Ω and a ∈ D, then


Z
1 f (z)
f (a) = dz
2πi ∂D z − a

In particular, if D = D(a, r) is center of the disc, then


Z 2π
1
f (a) = f (a + reit )dt
2π 0
This remains true if there is a finite subset F ⊂ Ω such that f is holomorphic
on Ω \ F ; lim (z − ζ)f (z) = 0 for ζ ∈ F ; C a closed curve in Ω \ F , a 6∈ C,
z→ζ
Remenber if C is a circle anticlockwise, travelled once then for a within
the circle n(C, a) = 1 and so the representation formula applies. We are
assuming that the circle does not have points from F . Also the last formula
says that value at the center equals its average on the circle.
Proof is simple. Take

f (z) − f (a)
F (z) =
z−a
This is not defined at a, but T4 applies. You only have to notice that as
z → a, we have (z − a)F (z) = f (z) − f (a) → 0. Thus

f (z) − f (a)
Z
dz = 0.
C z−a
Z Z
1 f (z) 1 f (a)
dz = dz = n(a, C)f (a)
2πi C z − a 2πi C z − a

81
This completes proof of first equation. The second one follows from n(a, C) =
1. Third follows from n(a, ∂D) = 1.

Theorem 7:
Representation formula/ Cauchy’s integral formula
If f is analytic in a disc Ω C ⊂ Ω closed curve. For every z ∈ Ω \ C
with n(z, C) = 1 we have
Z
1 f (ζ)dζ
f (z) =
2πi C ζ − z
This is just the previous formula with change of notation. Instead of a
we have now z and instead of the dz we have dζ. The onlyyhing you need to
notice is that C being compact, there is a smaller disc D so that C ⊂ D ⊂ Ω.
Why change of notation? Now we want to study the function f itself,
not the number f (a), using the integral formula. Since the above formula is
true for any a in the disc.

THEOREM 8: differention under integral sign


Let C be a curve and n ≥ 1 integer. Suppose that ϕ is a continuous
function on C. Put
Z
ϕ(ζ)
Fn (z) = n
dζ; z ∈C\C
C (ζ − z)

Then FN is analytic in each region of C c ; and Fn0 (z) = nFn+1 (z)


Proof: Fix a region Ω as stated. Note that C c is a countable union of
regions. Note that we did not demand that C be a closed curve.
First we discuss the case n = 1, that is F1 . Fix a disc D such that D ⊂ Ω.
K = d(D, C) = min{d(u, v) : u ∈ D, v ∈ C}. Note that K > 0. Let M be a
bound for |ϕ|. for z, w ∈ D
Z Z
ϕ(ζ) ϕ(ζ)
|F1 (z) − F1 (w)| = | dζ − dζ|
C (ζ − z) C (ζ − w)
Z
ϕ(ζ) M
=| (z − w)| ≤ 2 |z − w|L(C)
C (ζ − z)(ζ − w) K
Thus F1 is lipschitz and so continuous. This is true for every disc contained
in Ω and so F1 is continuous on Ω.
To prove differentiability, Fix z and disc as above. For w 6= z in the disc

F1 (z) − F1 (w)
Z
ϕ(ζ)
= → F2 (z)
z−w C (ζ − z)(ζ − w)

82
The last implication above uses the following two facts.
(i) continuous functions ϕj → ϕ uniformly on C, then integrals over C
converge. This is because,
Z Z
| ϕj − ϕ| ≤ max |ϕj − ϕ| L(C)
C C

(ii) If wj → z

ϕ(ζ) ϕ(ζ) M
| − 2
| ≤ 3 |wj − z|
(ζ − z)(ζ − wj ) (ζ − z) K

Proof for General n:


Fix r so that |u| ≤ r for u ∈ D ∪ C, this is possible because this is a
compact set.
Z n
ϕ(ζ) X
|Fn (z) − Fn (w)|g| (z − w) (ζ − z)j−1 (ζ − w)n−j |
C (ζ − z)n (ζ − w)n j=1

≤ n(2r)n−1 |z − w|/(K)2n
Since r, K, n are all fixed this shows Lipschitz and hence continuity. To show
differentiability, Fix z, for w 6= z (in the small disc)
n
Fn (z) − Fn (w)
Z
ϕ(ζ) X
= (ζ − z)j−1 (ζ − w)n−j
z−w C (ζ − z)n (ζ − w)n j=1
Z
ϕ(ζ)
→ 2n
n(ζ − z)n−1 = nFn+1 (z)
C (ζ − z)
Again the implication above uses the two facts mentioned above. Note also
the following two facts:
if ϕj → ϕ uniformly and ψj → ψ uniformly and all functions here are
bounded by a number, then ϕj ψj → ϕψ.
if ϕj → ϕ uniformly and ψj → ψ uniformly then ϕj + ψj → ϕ + ψ uni-
formly.

THEOREM 9: Cauchy formula for derivatives


f analytic in a region Ω; D(z, r) ⊂ Ω
C ⊂ D(z, r) any curve with n(z, C) = 1 Then
Z
(k) k! f (ζ)
f (z) = dζ.
2πi C (ζ − z)k+1

83
This follows immediately from Cauchy integral formula and the above
theorem.

In particular, derivative of analytic function is analytic too. Analytic


function is differentiable as many times as you wish and gives you an ana-
lytic function.

Theorem 10: Morera’s theorem R


Suppose that f is continuous in a region Ω and f dz = 0 for every closed
C
curve C ⊂ Ω. Then f is analytic.
We have already seen after the basic theorem 0, that if f satisfies the
above hypothesis, then it is derivative of an analytic fuunction. hence it is
analytic too.

Theorem 11: Cauchy’s estimate for derivatives


Let f be analytic in a region Ω; D(z, r) ⊂ Ω. Let M be bound for f on
∂D. Then
|f (k) (z)| ≤ M k!/rk .
Proof follows from the Cauchy’s formula for the derivative.

Theorem 12: Liouville’s theorem


An analytic function which is defined on all of C and bounded must be
constant.
Cauchy’s estimate shows f 0 ≡ 0.
Functions defined on all of C and analytic are called ‘entire’ functions.

Theorem 13: Fundamental theorem of algebra


If P (z) is a polynomial of degree at least one, then there is w with P (w) =
0
If P is of degree n, then we can express

P (z) = c(z − w1 )(z − w2 ) · · · (z − wn )

Proof: we have already seen the second part. To prove the first part,
suppose there is no such w, then 1/P (z) is an analytic function on C. In our
discussion on rational functions we saw that |P (z)| → ∞ as |z| → ∞, so that
1/P (z) → 0. Thus there is an α such that |z| ≥ α implies |1/P (z)| < 1. But
the set (|z| ≤ α) is a compact set and 1/P (z) is bounded on this set. Thus
1/P (z) is bounded on all of C. By Liouville, it must be a constant, thus P
is constant, a contradiction. This completes the proof.

84
BVR/CMI Complex Analysis 29-03-2022

Have a region Ω and a ∈ Ω. have an analytic function f on Ω \ {a}.


When can I remove this ‘singularity’, that is, when can I define it at a also
so that it is still analytic. equivalently, when can I get an analytic function
on all of Ω extending f ? equivalently when can I get an analytic function g
on Ω such that g = f on Ω \ {a}. Note that if this is possible then there is
only one such extension, due to continuity.

Theorem 14: Extension


Ω is a region; a ∈ Ω; f analytic on Ω \ {a}. Then there is an extension of
f to all of Ω as an analytic function iff lim (z − a)f (z) = 0.
z→a
Proof is simple. if there is such an extension, then clearly f is in particular
bounded near a and so the stated limit is zero. Conversely, if the condition
is satisfied, then take a disc D = D(a, r) with center a so that D ⊂ Ω. By
Cauchy formula
Z
1 f (ζ)
f (z) = dz; z ∈ D \ {a}
2πi ∂D ζ − z
The integral on right is an analytic function in all of D as well, by an earlier
theorem. This gives required extension because the right side agrees with f
etc.
You should be careful when you read such sentences as the last one. Obvi-
ously, I have shown you extension only on D, did I show on Ω? Should you
accept it?
Compare with Theorem 4 where we had exceptional points. This theorem
says that they are not really exceptional, we were ignorant! But of course
that better version was needed to prove Cauchy formula and now we know
reality. So remember that Theorem 4 needs to be proved as stated.

Theorem 15: Finite Taylor series


Ω is a region, a ∈ Ω, f analytic in Ω, n ≥ 1 integer. Then there is an
analytic function fn in Ω such that

f 0 (a) f 00 (a) f (n−1) (a)


f (z) ≡ f (a)+ (z−a)+ (z−a)2 +· · ·+ (z−a)n−1 +fn (z)(z−a)n .
1! 2! (n − 1)!

Proof is simple. Consider

f (z) − f (a)
F (z) = z ∈ Ω \ {a}
z−a

85
and undefined for z = a. This satisfies lim (z − a)F (z) = 0. Thus there is an
z→a
analytic function f1 on Ω extending F . In other words

f (z) = f (a) + (z − a)f1 (z)

where f1 is analytic on all of Ω. Note that this equality holds for z = a also.
Repeat the same argument as above with f1 to see

f1 (z) = f1 (a) + (z − a)f2 (z).

continue
fn−1 (z) = fn−1 (a) + (z − a)fn (z).
Thus
f (z) ≡ f (a) + f1 (a)(z − a) + f2 (a)(z − a)2
+ · · · + fn−1 (a)(z − a)n−1 + fn (z)(z − a)n . (♠)
Differentiate and put z = a to see f1 (a) = f 0 (a). Differentiate twice and
put z = a to see f2 (a) = f 00 (a)/2!. And finally fn−1 (a) = f (n−1) (a)/(n − 1)!.
substitute in (♠) to get the stated result. Of course we also get fn (a) =
f (n) (a)/n!.
Let us understand the function fn above. Continuing with the same set
up, Ω, a, D as in the earlier theorem, we see that for z ∈ D, z 6= a
f (z) c1 c2 c3 cn−1 cn
fn (z) ≡ − − − +· · ·− − .
(z − a)n (z − a)n (z − a)n−1 (z − a)n−2 (z − a)2 (z − a)

we see by Cauchy [Pause, check notation before confusion sets in]


Z Z
1 fn (ζ) 1 f (ζ)
fn (z) = dζ = dζ
2πi ∂D ζ − z 2πi ∂D (ζ − a)n (ζ − z)
Z Z
c1 c2
− n
dζ − n−1

∂D (ζ − a) (ζ − z) ∂D (ζ − a) (ζ − z)
Z Z
c3 cn−1
− n−2
dζ + · · · − 2

∂D (ζ − a) (ζ − z) ∂D (ζ − a) (ζ − z)
Z 
cn
− dζ .
∂D (ζ − a)(ζ − z)

NOW consider a fixed z ∈ D. Last term above is zero for all a because
a, z ∈ D and the integral is, except for a factor, equals difference of winding
numbers each of which is one. Note that this is true even when a = z by
direct calcualtion (or use continuity of the integral). By earlier theorem, each

86
term as a function of a, equals derivative of the next term and hence they
are all zero except the first term.
Do not get confused with the fact that a is center of the disc etc, it is
irrelevant. These terms that we are handling, have no f in them and they
are fucntions of a on the disc (z being fixed). We apply the differentiation
theorem. Further, the above equation holds for z 6= a is also irrelevant
because we are discussing the integrals and not the equality. Yes, when we
return to the equality, it is for z 6= a. Thus we have
Z
1 f (ζ)
fn (z) = dζ; z ∈ D (♣)
2πi ∂D (ζ − a)n (ζ − z)
This is true for z = a too because both sides are analytic and equal for all
z 6= a in the disc. This expression leads to interesting consequences.

Theorem 16: all derivatives zero at one point


Ω is a region, a ∈ Ω. If f (a) = 0 and all derivatives of f at a equal zero
then f ≡ 0 in Ω.
First we show f (z) = 0 for z ∈ D. So fix z ∈ D, z 6= a. Let f be bounded
by M on ∂D. Observe: for ζ ∈ ∂D we have |ζ − a| = r so that

r = |ζ − a| ≤ |ζ − z| + |z − a| or |ζ − z| ≥ (r − |z − a|)

and length of the curve ∂D is 2πr. Thus


M
|fn (z)| ≤
rn−1 (r − |z − a|)

Note that by hypothesis we have f (z) = fn (z) (z − a)n . Thus


 n
M n |z − a| Mr
|f (z)| ≤ n−1 |z − a| =
r (r − |z − a|) r (r − |z − a|)

Since this is true for every n ≥ 1 and since (|z − a|/r) < 1 for z ∈ D we see
that f (z) = 0. This is true for all z ∈ D. Thus f ≡ 0 on D.
In other words if f (z0 ) = 0 and all derivatives at z0 are aso zero then f ≡ 0
in a disc around z0 .
Is it zero on all of Ω? Yes, by connectedness. A is the set of points where
value and all derivatives vanish; and B is the set where either value or some
derivative is nonzero. Since f and derivatives are all continuous functions,
clearly B is open. From the argument above, we see A is open. From hy-
pothesis a ∈ A and so A 6= ∅. We conclude A = Ω completing the proof.

87
Theorem 17: zeros are isolated
Let f be analytic in a region Ω and not the zero function. If f (a) = 0,
then there is a disc D(a, r) ⊂ Ω such that the only zero of f within the disc
is a.
Since f is not the zero function, in view of the previous theorem, there
is a k ≥ 1 so that f (k) (a) 6= 0 and f (j) (a) = 0 for j < k. Consider the
finite Taylor expansion f (z) = (z − a)k fk (z). Differentiating k times we see
fk (a) = f (k) (a)/k! 6= 0 and thus fk is an analytic function with fk (a) 6= 0.
There is a disc D(a, r) in which fk is not zero. Clearly in this disk, a is the
only zero of f .

Theorem 18: uniqueness from values on a sequence


Let Ω be a region and f, g analytic on Ω. {an } a sequence of distinct
points in Ω converging to a point in Ω. Suppose f (an ) = g(an ) for all n.
Then f = g on all of Ω.
Let Ω be a region and f, g analytic on Ω. Suppose f = g on an arc which is
not a singleton. Then f = g on all of Ω.
Since the each zero must be an isolated zero, and lim an is not an isolated
zero (Remember an are distinct) we conclude that f ≡ 0. The next sentence
follows because the arc is not a singleton, being connected, must be uncount-
able compact set, so has a sequence as above.

This is very interesting. Suppose we have an analytic function on all of


C. Suppose we know its values on a tiny arc near zero, say on the interval
[0, 0.0001] on the x-axis. Then its values on all of C are determined!

Theorem 19: Every zero has an order


Suppose that f is analytic in a region and f (a) = 0 and f is not the zero
function. Then there is a unique k ≥ 1 such that f (z) = (z − a)k g(z), where
g is analytic and g(a) 6= 0.
Take the first k such that f (k) (a) 6= 0, thus for j < k we have f (j) (a) = 0.
Consider the finite Taylor expansion. f (z) = (z − a)k g(z) with g analytic
and g(a) 6= 0. Uniqueness is routine as in case of rational function.

This k is called the order of the zero at a.

Theorem 20: Every pole has an order


Suppose Ω is a region and a ∈ Ω. Suppose f is an analytic in Ω \ {a} and
lim |f (z)| = ∞. Then there is a unique k ≥ 1 such that g(z) = (z − a)k f (z)
z→a
is analytic on Ω with g(a) 6= 0.

88
This means g is an analytic function on Ω and g(a) 6= 0 and g(z) =
(z − a)k f (z) on Ω \ {a}. Afterall f is defined only on Ω \ {a}.
The proof is obtained by considering a (deleted) neighbourhood of a where
|f | > 1, exists by hypothesis. Consider h = 1/f in this neighbourhood. Note
that h(z) → 0 as z → a. Apply extension theorem and the previous theorem
to write h(z) = (z − a)k ψ(z) for some ψ non-zero, defined on all of the neigh-
bourhood. Thus taking g(z) = 1/ψ(z) we see g(z) = (z − a)k f (z) is analytic
and non-zero. Uniqueness is routine (but please do it).

Such an a is called a pole of f and k is the order of the pole. In particular,


we can define (z − a)k f (z) at a also so that it is analytic on all of Ω.
You should recall our discussion of rational functions. The notion of
zero and pole we defined there is same as what we defined here. These
two theorems say that, not only for rational functions, but for every analytic
function there is a specific order for zero and a specific order for pole. We are
using the word analytic in a loose sense because the function is not defined at
a. Our rational functions were also undefined at the zeros of the denominator
polynomial.
You can view the two theorems above in a nice fashion if you consider the
Riemann sphere. f (z) → 0 as z → a means values approach the ‘South pole’.
|f (z)| → ∞ as z → a means values approach the ‘North pole’. Thus roughly
speaking, when vaklues approach zero they must approach like (z − a) or
(z − a)2 · · · or (z − a)k for some k ≥ 1. Similarly when the values approach
‘infinity’ then they should approach like 1/(z−a) or 1/(z−a)2 · · · or 1/(z−a)k
for some k ≥ 1.
Even if f is not defined at a, but defined in a neighbourhood of a; in case
lim f (z) = 0, then you can extend the analytic function to a be defining
z→a
f (a) = 0. It remains analytic. In the same fashion, if f is not defined at
a, but defined in a neighbourhood of a; in case lim |f (z)| = ∞, then you
z→a
can extend the analytic function to a be defining f (a) = ∞ or North pole.
It remains analytic. Of course you have to allow ∞ ot north pole into our
system.
There is a perfect symmetry between the ‘poles’ or between ‘zero/infinity’.

89
BVR/CMI Complex Analysis 01-04-2022

Theorem 21: uniform limits, sequences


fn analytic in Ωn ; fn → f uniformly on each compact subset of region Ω.
Then f is analytic on Ω. Further fn0 → f 0 uniformly on compact subsets of
Ω. In fact, for any k ≥ 1, the k-th derivatives converge uniformly on compact
subsets of Ω.
The hypothesis means the following: Given any compact K ⊂ Ω, there
is an N such that all the fn for n > N are defined on K and the sequence
{fn : n > N } converges uniformly to f on K.
If you feel uncomfortable taking different regions for different functions,
just take all Ωn = Ω. But is it worth taking different Ωn . Yes, many times
while studying conformal maps you have to deal with different regions. But
for now just the the example from Ahlfors will help.
z
fn (z) = ; Ωn = (|z| < 2−1/n ); Ω = (|z| < 1).
2z n+1
The analyticity is immediate from Morera, because hypothesis implies
that for any closed curve C
Z Z
f = lim fn = 0.
C C

To show fn0 converges uniformly on compact subsets of C, fix a point a ∈ Ω


and a disc D = D(a, 2r) such that D ⊂ Ω. Let us now take any z ∈ D(a, r)
Note that for any ζ ∈ ∂D we have |ζ − z| ≥ r. Indeed,

2r = |ζ − a| ≤ |ζ − z| + |z − a| ≤ |ζ − z| + r

Now By Cauchy, we have for z ∈ D(a, r)

fn (ζ) − f (ζ)
Z
0 0 1
|fn (z) − f (z)| = | dζ|
2πi ∂D (ζ − z)2
1
≤ sup |fn (ζ) − f (ζ)| 2r → 0.
ζ∈∂D r2
Thus every point has a disc around it in which the derivatives converge uni-
formly; note that the above estimate does not depend on z. This is enough
to show uniform convergence on compatc sets.
Repeated application of what is proved yields for all derivatives.

Theorem 22: uniform limits, series

90
P
Suppose Ω is a region, each fn analytic in Ω and fn converges to f ,
uniformly on compact subsets of Ω, then fPis analytic and the series can be
differentiated term by term. That is f 0 = fn0 . More generally, for the k-th
P (k)
derivatives we have f (k) = fn .
n
This is immediate by previous theorem applied to the partial sums se-
quence.

Theorem 23: number of zeros


f analytic in a disc Ω and not the zero function. {zj } are the zeros of f ,
each zero counted as many times as its order. Then for every closed curve
C ⊂ Ω such that no zero belongs to C; we have
Z 0
X 1 f (z)
n(zj , C) = dz
2πi C f (z)
In particular if C is a circle, then the integral gives the number of zeros of f
(do not forget to count with multiplicity) inside the circle.
Note that there are only finitely many zeros in the compact region: inside
the curve; and hence the sum on left side has only finitly many terms. We
are not using Jordan curve theorem. Since C is compact, it is at a positive
distance from the boundary of Ω. So it is possible to draw smaller disc Ω∗
such that C ⊂ Ω∗ ⊂ Ω∗ ⊂ Ω. In Ω∗ there are finitely many zeros. This round
about way is to avoid using ‘inside’ C; we can say ‘inside’ a disc.
For a zero z1 , using finite Taylor (only one term, a = z1 ) we can write,
since f (z1 ) = 0;
f (z) = (z − z1 )f1 (z)
continue with f1 now. Do as many times as there are zeros in the smaller
disc mentioned above; to write

f (z) = (z − z1 )(z − z2 ) · · · (z − zk )g(z)

where g has no zeros in the disc Ω∗ . Remember the zj need not be distinct.
In fact each zero appears as many times as its order. Using product rule for
derivatives we see
f 0 (z) 1 1 1 g 0 (z)
= + + ··· + + z∈C
f (z) z − z1 z − z2 z − zk g(z)

Remember g has no zeros in Ω∗ and g 0 is analytic too and so is the ratio g 0 /g.
Thus by Cauchy Z 0
g (z)
dz = 0
C g(z)

91
OR
k
f 0 (z)
Z
1 X
dz = n(zj , C)
2πi C f (z) 1

where the sum runs over the zeros in Ω∗ . But, instead of taking the ‘inter-
mediate’ disc Ω∗ that we made up, it is better to take the sum over zeros in
all of Ω because for other points (if any) the number n is zero.
The last statement of the theorem is immediate from the fact that n(zj , C)
is one for points inside the circle and zero for points outside the circle.

This theorem has interesting consequences.

Theorem 24: Hurwitz Theorem, dichotomy for limits


For each n, fn is analytic and never zero in a region Ω. fn → f uniformly
on each compact subset of Ω. Then: either f is never zero OR f ≡ 0.
Suppose f is not the zero function. Take, any point say z0 ∈ Ω. Since
zeros are isolated take a disk D(z0 , r) so that there is no zero in D \ {z0 };
of course z0 may or may not be zero at this stage. Thus |f | has a positive
minimum on ∂D. Observe 1/fn → 1/f uniformly on ∂D. In fact if |f | ≥ c > 0
on ∂D, then by uniform convergence |fn | > c/2 on ∂D for all n after some
stage. Thus after that stage

1 1 |fn − f | 2 1
| − |= ≤ |fn − f |
fn f |fn ||f | c c

and so converges uniformly on ∂D. Also fn0 → f 0 uniformly on ∂D. Hence


fn0 /fn converges uniformly to f 0 /f on ∂D. [if un → u uniformly; vn → v
uniformly; all bounded by a fixed number; then un vn → uv uniformly]. Thus

fn0 (z) f 0 (z)


Z Z
1 1
lim dz = dz
2πi ∂D fn (z) 2πi ∂D f (z)

By previous theorem, Integrals on left are all zero, hence so is the right side.
Again by the previous theorem we see f (z0 ) 6= 0.

Theorem 25: infinite Taylor expansion


f analytic in a region Ω and z0 ∈ Ω. For every D = D(z0 , r) ⊂ Ω we have
the representation as an infinite series:

f 0 (z0 ) f (n) (z0 )


f (z) = f (z0 ) + (z − z0 ) + · · · + (z − z0 )n + · · · ; z∈D
1! n!

92
Proof is simple. Assume that D ⊂ Ω. The finite Taylor expansion tells
us
f 0 (z0 ) f (n) (z0 )
f (z) = f (z0 ) + (z − z0 ) + · · · + (z − z0 )n + fn+1 (z)(z − z0 )n+1
1! n!
with Z
1 f (ζ)
fn+1 (z) = dζ
2πi ∂D (ζ − z0 )n+1 (ζ − z)
If M = max{|f (w)| : w ∈ ∂D}, then as seen earlier

M |z − z0 |n+1
|fn+1 (z)(z − z0 )n+1 | ≤ → 0.
rn (r − |z − z0 |)

In other words
0 (n)
 
f (z) − f (z0 ) + f (z0 ) (z − z0 ) + · · · + f (z0 ) (z − z0 )n → 0

1! n!

Thus the result is true for all z ∈ D. We assumed that D ⊂ Ω. But given
any D(z0 , r) ⊂ Ω and given any z ∈ D(z0 , r) we can select 0 < r∗ < r such
that
z ∈ D(z0 , r∗ ) ⊂ D(z0 , r∗ ) ⊂ D(z0 , r) ⊂ Ω.
and the above argument applies for D(z0 , r∗ ) and hence equality holds at z.

Theorem 26: infinite Taylor expansion, at a pole


f analytic in a region Ω \ {a} and a ∈ Ω. Suppose that a is a pole
of order k ≥ 1. Then we have numbers {cj , j ≥ −k} such that For every
D = D(a, r) ⊂ Ω we have the representation, for z ∈ D(a, r) \ {a}. as an
infinite series:
f (z) =
c−k c−k+1 c−1
k
+ k−1
+···+ + c0 + c1 (z − a) + c2 (z − a)2 + · · · + · · · ;
(z − a) (z − a) (z − a)
for z ∈ D(a, r) \ {a}.
Proof follows from Taylor expansion of (z − a)k f (z). More precsiely, let
g(z) = (z − a)k f (z). We know that g can be defined at a as well so that g is
analytic in Ω and further g(a) 6= 0. Now expand g into Taylor series around
a (theorem above). We have numbers αi such that for any D(a, r) ⊂ Ω we
have

g(z) = α0 + α1 (z − a) + α2 (z − a)2 + · · · + · · · ; z ∈ D(a, r)

93
Thus, remembering that f is defined only on D(a, r) \ {a}

(z − a)k f (z) = α0 + α1 (z − a) + α2 (z − a)2 + · · · + · · · ; z ∈ D(a, r) \ {a}

equivalently,
α0 α1 αk−1
f (z) = k
+ k−1
+···+ + αk + αk+1 (z − a) + · · · + · · · ;
(z − a) (z − a) (z − a)

for z ∈ D(a, r) \ {a}. As you see the coefficients α are the value and deriva-
tives of g = (z − a)k f at a.

94
BVR/CMI Complex Analysis 05-04-2022

[Please note that final exams will be on campus and not on-line. Those
from BSC III, who are taking the course have already been advised several
weeks ago that if they have to write any admission tests, they should opt
Chennai center.]
Initially I thought that I shall do some more important theorems like
maximum modulus principle/Schwarz lemma, reflection principle. But it
appears to me that we should do some applications to keep your attention.
So we shall discuss the Residue calculus: using Cauchy theorem and the
above theorem to evaluate some integrals.
Returning to the Taylor-Laurent expansion, Let us carefully understand
the coefficients in the above expansion because they play a role later in
computations. Let us start with the Taylor expansion. If f is analytic in a
region Ω containing a point a and if D(a, r) ⊂ then
f 0 (a) f (n) (a)
f (z) = f (a) + (z − a) + · · · + (z − a)n + · · · ; z ∈ D(a, r)
1! n!
or equivalently we have a power series expansion at a:

X
n
f (z) = c0 + c1 (z − a) + · · · + cn (z − a) + · · · = cn (z − a)n ; z ∈ D(a, r)
n=0

where the coefficients are given by


c0 = f (a); c1 = f 0 (a)/1!; cn = f (n) (a)/n!
If f is analytic in Ω \ {a} with a, pole of order k ≥ 1 we have
α0 α1 αk−1
f (z) = + + · · · + + αk + αk+1 (z − a) + · · · + · · · ;
(z − a)k (z − a)k−1 (z − a)
for z ∈ D(a, r) \ {a}. Since the above expansion is obtained from the Taylor
of g(z) = (z − a)k f (z) we see the the coefficients are given by
αn = g (n) (a)/n! = [(z − a)k f (z)](n) /n!
that is

c−k c−k+1 c−1 X
f (z) = + + · · · + + cn (z − a)n
(z − a)k (z − a)k−1 (z − a) n=0

where we have
cn = αk+n = [(z − a)k f (z)](n+k) /(n + k)!; n ≥ −k

95
Note that the coefficients have nothing to do with r, thus this expansion
is valid in any (punctured) disc centered at a as long as the closed disc is
contained in Ω. Also we changed the notation to ‘match’ power of (z − a).
The coeficient for power of (z − a)n is cn . In Taylor expansion, the series
starts with powers from zero on. In case of pole of order k, it starts with
powers (z − a)n for n ≥ −k. The expresion
c−k c−k+1 c−1
k
+ k−1
+ ··· +
(z − a) (z − a) (z − a)

is called the principal part of the expansion. The number c−1 is called the
residue of f at the pole a, denoted Res(f, a). The pole of order one is called
simple pole. Thus for simple pole, the residue is nothing but lim (z − a)f (z).
z→a
In the general case, if a is a pole of order k, then residue at a is given by
 k−1
1 d
Res(f, a) = [(z − a)k f (z)]
(k − 1)! dz z=a

When k = 1, there is no derivative, just limit as z → a.


In stead of saying f is analytic in Ω \ {a} with pole at a, we shall from
now on say: f is analytic in Ω with pole at a; it means f is actually analytic
in Ω \ {a} and the point a is a pole.
Residue is very important and we shall next state Cauchy Residue the-
orem. Let us start observing the following. Even though f is undefined at
a it makes sense on the circle (positively oriented, as always) C of center a,
radius
R r, when D(a, r) ⊂ Ω. And of course continuous on C. So the integral
f dz makes sense. What is it?
C
Z
f dz = 2πi Res(f, a)
C

Proof is simple. f is sum of two functions,the principal part and the remain-
ing. But the remaining part is analytic and so its integral is zero. Even in
the principal part, which is sum of finitely many terms, the only term that
remains is the c−1 /(z −a) term because others are zero, they have a primitive
(where?) or do it directly. Thus
Z Z
c−1
f dz = dz = 2πi n(a, C) Res(f, a) = 2πi Res(f, a)
C C z −a

Thus we have
Theorem 27: Cauchy Residue theorem

96
Let a ∈ Ω, a region and f an analytic function in Ω with pole at a. If
D(a, r) ⊂ Ω, then Z
f (z)dz = 2πi Res(f, a)
∂D
F More generally, suppose f is analytic in Ω except for poles at finitely
many points a1 , · · · , aN . C ⊂ Ω a nice contour like rectangle or circle etc
which along with its interior is contained in Ω and the points a1 , . . . , aN are
inside C, then
Z N
X
f (z)dz = 2πi Res(f, aj ) (♠)
C j=1

The equation (♠) is called Residue formula.


There are two things that should be made clear. Firstly we use the word
cantour for a simple closed curve travelled in the anticlockwise direction.
It is just a curve as we are used to. The difference is that a curve could
be, for example, a circle travelled twice. But when we consider cantours, we
specify a simple closed curve and you travel it once in anticlockwise direction.
Secondly what is ‘etc’ ? in the above theorem? It is – roughly speaking –
such curves for which we can define interior without appeal to the Jordan
curve theorem. Of course, it is difficult to ‘quantify/describe precisely’ what
exactly are such curves. Stein-Sakarchi avoid this using ‘Toy cantours’. After
all in applications, we come across only a few easily describable cantours. For
toy cantours we can, with patience, easily define interior.
If we have an analytic functionR f in an open set that includes the toy
cantour C and its interior, then f dz = 0. The proof, only a little more
C
complicated than Cauchy’s theorem 3. Idea is the following: First take a
region Ω with ∂Ω a toy cantour. This is possible by enlarging the original
cantour in ‘all directions’ a little bit. It suffices to show that f has a primitive
in Ω, then integral over any closed curve is zero. This is done by explicitly
constructing F . Recall the proof of C − G3. We used the fact integral over a
rectangle is zero to define it. In C − G4 (not yet proved) we exhibit primitive
by the same procedure; fixing a point w and taking a path to z which consists
of horizontal and vertical lines; and F (z) is integral of f over this path. In
case of disc we used two sides of a rectangle. In this case we need to use sides
of ‘many’ rectangles. We postpone the argument.
Here are some Toy cantours. In each case, given descriptions (equa-
tions) of the circles/semicircles/lines; we can easily describe what is inte-
rior.

97
We now see applications of these theorems to evaluate some integrals.
You realize that the Residue theorem says that certain integrals are nothing
but sum of residues (multiple) and calcualting residues is easier. This body
of techniques goes by the name of Residue Calculus. First we use Cauchy
theorem itself, so without poles and without residues!
(1) normal cf
Z
1 2 2
√ e−x /2 eitx dx = e−t /2 ; t∈R

Proof: Z Z
1 itx −x2 /2 1 2 2
√ e e dx = √ e−(x−it) /2 e−t /2 dx
2π 2π
put x − it = y and dx = dy
Z
1 2 2 2
√ e−y /2 e−t /2 dy = e−t /2 ???

Shoeuld acept this proof? No because the substitution is illegal (at least we
do not know) because it uses complex number ‘it’.
But the conclusion is correct and Cauchy formula justifies it. Fix R > 0.
2
Let us integrate e−z /2 over the contour

(−R + i0) −→ (R + i0) −→ (R − it) −→ (−R − it) −→ (−R + i0)

Cauchy tells us
Z R Z −t Z −R Z 0
−x2 /2 −(R+iy)2 /2 (x−it)2 /2 2 /2
e dx+ e idy + e dx+ e−(−R+iy) idy = 0
−R 0 R −t

t being fixed, as R → ∞
Z −t
2 2
| e−(R+iy) /2 idy| ≤ e−R /2 (constant) → 0
0

Thus Z R Z −R
−x2 /2 2 /2
lim e dx + e(x−it) dx = 0
R→∞ −R R
or Z R Z R
−x2 /2 2 /2
lim e dx = e(x−it) dx
R→∞ −R −R
or Z ∞ Z ∞
−x2 /2 2 /2
e dx = e(x−it) dx
−∞ −∞

100
The functions we have; are easily seen to be integrable and so instead of limit
we used integral on all of R.
If you are curious whether the integral can be evaluated using only real
variables, here is the answer. Let g(x) be standard normal density. Since
‘sin tx g(x)’ is integrable odd function its integral is zero. The cos part being
even function, require dintegral equals
Z ∞
1 2
ϕ(t) = 2 cos(tx) √ e−x /2
0 2π
One can justify differentiation (w.r.t. t) under integral sign
Z ∞
0
ϕ (t) = 2 [−x sin tx] g(x)dx
0

Noting that −xg(x) = g 0 (x), do integration by parts to see


Z ∞

= 2g(x) sin tx|0 − 2 t cos tx g(x)dx
0

ϕ0 (t) = −tϕ(t); ϕ(0) = 1


The onl;y solution of thsi is
2 /2
ϕ(t) = e−t .

101
BVR/CMI Complex Analysis 08-04-2022

2 Trigonometric integral
Z ∞
1 − cos x π
2
dx =
0 x 2
iz 2
Take f (z) = (1 − e )/z and integrate on the Cantour: indented semi circle:
Fix R >  > 0

(−R + i0) −→ (− + i0) −→ (semicircle upper half plane)(+ + i0)

−→ (R + i0) −→ (semicircle, upper half plane)(−R + i0)


made of C1 , C2 , C3 , C4
1 − eiz 1 − eiz 1 − eiz 1 − eiz
Z Z Z Z
dx + dz + dz + dz = 0
C1 z2 C2 z2 C3 z2 C4 z2
Z − Z R
1 − eix 1 − eiz 1 − eix 1 − eiz
Z Z
dx + dz + dx + dz = 0 (♠)
−R x2 C2 z2  x2 C4 z2
The curve C4 is given by: z(θ) = Reiθ for 0 ≤ θ ≤ π. Note that sin θ ≥ 0
for these values of θ.

|eiRe | = |eiR cos θ ||e−R sin θ | ≤ 1
1 − eiz 2
| 2
|≤ 2 lnegth of curve = πR
z R
So integral over C4 , which is, in modulus, at most R22 πR goes to zero as
R → ∞.
The curve C2 is given by: z(θ) = eiθ on [π, 0]. If you do not like my
writing interval [π, 0], you can take it as z(θ) = ei(π−θ) on [0, π] and conti
nue the calculation. Observe that
1 + iz − eiz
r(z) =
z2
is bounded as z → 0, in fact it converges to (−1/2) as z → 0. Note
−iz
f (z) = + r(z)
z2
So integral over C2 is
Z π
−iei(π−θ)
[ 2 2i(π−θ) + r(ei(π−θ) )] [−iei(π−θ) ]dθ
0 e

102
sum of two integrals, the first is
Z π
(−1)dθ = −π
0

second is at most (bound of r)×π ×  and so goes to zero as  → 0. Thus if


you take real parts in (♠) on bioth sides and then make  → 0 and R → ∞
in (♠), you see
Z 0 Z ∞
1 − cos x 1 − cos x
2
dx + n(−π) + dx = 0
−∞ x 0 x2

Equivalently,

1 − eix
Z
dx = π
−∞ x2
This is what we wanted. Incidentally, note thsat our original function is
integrable and hence we can write integral over all of real line.
Just one subtle point: you can not take limits in (♠) and write theb
limiting equation and take real parts. The point is even though
Z − Z R
sin x sin x
2
dx + dx = 0
−R x  x2

it does not make sense to say


Z 0 Z ∞
sin x sin x
2
dx + dx = 0
−∞ x 0 x2

(3) Another triginometric integral


Z ∞
sin x π
dx =
0 x 2
We consider the function eiz /z. Fix R >  > 0 and consider the Cantour
C:
−R −→ − −→ +(semicircle upper half plane)
−→ R −→ −R(semicircle upper half plane)
 Z R ix Z π iReiθ
eix eiz
Z Z
e e
dx + dz + dx + iθ
iReiθ dθ = 0 (♣)
−R x C2 z  x 0 Re
where C2 is the  semicircle.

103
(i) By (proof of the) Cauchy Residue theorem, the second term converges
to −πi as  → 0. Indeed, f has a simple pole at z = 0 and we have the
expansion f = z1 + g(z) where g is holomorphic.
Z Z Z
1
f (z)dz = dz + gdz.
C2 C2 z C2

g is bounded and length of the curve goes to zero as  → 0. The first integral,
byut direct evaluation equals −πi.
(ii) The last term goes to zero as R → ∞. You only need to notice that
the integrand term

|eiRe | = |eiR cos θ eiRi sin θ | = e−R sin θ
In the interval π to zero sin θ is non-negative and so the above quantity is
bounded by one and converges to zero as R → ∞ (except at zero and π) and
we have bounded range of integration, so that the integral converges to zero.
(iii) For the first and third terms, (cos x)/x being odd that cancels. Also
(sin x) being even we get sum of these two terms to be
Z R
i sin x
2 dx
 x
Thus letting  → 0 and then R → ∞ in (♣) we get the desired result.
Z ∞
sin x
2i dx − πi = 0
0 x
proving the result.

(4) well known Cauchy distribution


Z ∞
1
2

−∞ 1 + x
You must have done this already, by putting x = tan θ on (−π/2, +π/2) to
see this is true.
1
Let us see how Cauchy helps. Consider the function f (z) = 1+z 2 . This is

holomorphic, except for simple poles at ±i. Consider the Cantour


−R −→ +R −→ −R(semicircle)
Made up of two curves: line C1 and semicircle C2 . Inside the cantour there
is one simple pole +i — let us consider from now on R > 2. The residue at
this pole equals
1
lim = 1/2i
z→i z + i

104
Thus Z Z
1
f (z)dz + f (z)dz = 2πi =π
C1 C2 2i
Or Z R Z
1
dx + f (z)dz = π. (♣)
−R 1 + x2 C2

Regarding Second integral, first observe that


z2
lim =1
|z|→∞ 1 + z 2

Hence for all z with |z| > R0 we have |1/(1+z 2 )| ≤ 2/|z|2 Thus when R > R0 ,
we have Z
1 2
| 2
dz| ≤ πR → 0
C2 1 + z R2
as R → ∞. Thus taking limit in (♣) we get the stated result.

(4) exponential integral


If 0 < a < 1 then Z ∞
eax π
x
=
−∞ 1+e sin πa
consider f (z) = e /(1 + ez ) and the rectangular Cantour C:
az

−R −→ +R −→ R + 2πi −→ −R + 2πi −→ −R

In this rectangle the only singularity is when eZ = −1 or z = πi. It is a


simple pole.
z − πi
(z − πi)f (z) = eaz z
e − eπi
which converges to eaπi × 1/eπi Thus Res(f, πi) = −eaπi So
Z
f (z)dz = −2πieaπi (♣)
C

C is made up of four lines:


(i)
R
eax
Z
dx
−R 1 + ex
(ii) here z(t) = R + it for 0 ≤ t ≤ 2π so that integral equals
Z 2π
ea(R+it)
idt
0 1 + e( R + it)

105
Integrand, in modulus, equals (remember that a, t are real)

ea(R+it) eaR 1 1
| | = | −R | ≤ e(a−1)R it
1+e (R+it) R
e e +e it |e | − |e−R |

Used |a + b| ≥ |b| − |a| For R ≥ R0 , we have e−R ≤ 1/2 so that for R > R0
we have integrand, in modulus, at most 2e(a−1)R . Thus integral is at most
2π2e(a−1)R which converges to zero as R → ∞.
(iii) here z(x) = x + 2πi for x from R to −R (if you do not like this
z(x) = −x + 2πi for x ∈ [−R, R]). Thus integral equals
R R
ea(−x+2πi) eax
Z Z
(−dx) = −e2πia
−R 1 + e(−x+2πi) −R 1 + ex

(iv) is similar to (ii) and converges to zero as R → ∞.


Thus taking limit in (♣) we get
Z ∞
2πia eax
(1 − e ) x
dx = −2πieaπi
−∞ 1 + e

so that

eax eπia
Z
π π
x
dx = −2πi 2πia
= πia −πia
=
−∞ 1+e 1−e (e − e )/2i sin πa

(6) Another ‘Fourier transform’. For t ∈ R,


Z ∞ −2πixt
e 1
dx =
−∞ cosh πx cosh πt

Remember
ez + e−z
cosh z =
2
Fix t ∈ R. Consider the function
e−2πizt
f (z) =
cosh πz
and the rectangular Cantour

−R −→ R −→ R + 2i −→ −R + 2i −→ −R

The singularities of f in this rectangle are when eπz = −e−πz or e2πz = −1


or a = i/2 and b = 3i/2. These are actually simple poles.

106
2e−2πizt −2πizt (z − i/2)
(z − a)f (z) = (z − i/2) = 2e eπz
eπz + e−πz e2πz − e2πi/2
The derivative of e2πz at i/2 equals 2πe2πi/2 = −2π Hence
1 eπt
lim (z − i/2)f (z) = 2eπt (i) =
z→a=i/2 −2π iπ
Similarly b = 3i/2 is a simple pole and
e3πt
lim (z − b)f (z) = −
z→b=i/2 ıπ
Thus
eπt e3πt
Res(f, i/2) = ; Res(f, 3i/2) = −
iπ iπ
Now let us see the integral over the four lines:
(i) The curve is z(x) = x for x between −R and R. So integral equals
Z R −2πixt
e
dx
−R cosh πx

(ii) The curve is z(y) = R + iy for 0 ≤ y ≤ 2. So integral equals


Z 2
e−2πi(R+iy)t
dx
0 cosh π(R + iy)

But
|e−2πi(R+iy)t | = e2πy|t| ≤ e4π|t| (•)
Use |a + b| ≥ | |a| − |b| | to see
1 1
| cosh πz| = |eπz + e−πz | ≥ | |eπz | − |e−πz | |
2 2
Thus on the line z(y) = R + iy under consideration
1
| cosh πz| ≥ |eπR − e−πR | → ∞ (†)
2
as R → ∞. Thus from (•) and (†) we see that the integral converges to zero
as R → ∞
(iii) The curve is z(x) = x + 2i for x between R and −R. So integral
equals Z −R Z R −2πixt
e−2πi(x+2i)t 4πt e
dx = −e dx
R cosh π(x + 2i) −R cosh πx

107
(iv) Just as in (ii), this integral converges to zero. Thus (i)+(iii) equals
R
e−2πixt
Z
4πt
(1 − −e ) dx
−R cosh πx

Hence by Residue theorem


Z R −2πixt
4πt e eπt e3πt
(1 − e ) dx + (ii) + (iv) = 2πi( − )
−R cosh πx iπ iπ

= 2(eπt − e3πt )
Taking limit as R → ∞ we have
Z ∞ −2πixt
e 2(eπt − e3πt ) eπt (1 − e2πt )
dx = = 2
−∞ cosh πx (1 − e4πt ) (1 − e2πt )(1 + e2πt )

2 1
= −πt
=
eπt +e cosh πt

108
BVR/CMI Complex Analysis 12-04-2022

back to theory:
Recall if f analytic in a disc Ω; and not the zero function, then for any
closed curve C in the disc such that no zero of f is on the curve we have
Z 0
X 1 f (z)
n(zj , c) = dz
2πi C f (z)
where (zj ) are the zeros of f counted with multiplicity.
The quantity f 0 /f is called the ‘logarithmic derivative’. In case f : R →
(0, ∞) and differentiable, then this is exactly derivative of log f . But even in
the complex situation, this operation transforms product to addition.
(f g)0 f 0 g0
= +
fg f g
There is a better theorem that goes by the name of ‘argument principle’,
but before that some consequences of above theorem.

Theorem 28: Rouche’s theorem


f , g are analytic in a region Ω. C is a circle/rectangle/(simple closed
curve) which along with its interior is contained in Ω. Suppose that |f | > |g|
on C. Then f and f + g have the same number of zeros inside C.
Proof: We want to move from f towards f + g. For 0 ≤ t ≤ 1, define
ft = f + tg. Thus at time zero we have f and at time one f + g. Observe ft
has no zero on C. Indeed, if it has then at such a point z ∈ C, f (z) = −tg(z)
so that |f (z)| = t|g(z)| ≤ |g(z)| contradicting our assumption. If nt is the
number of zeros of ft within C, then Theorem 23 tells us
Z 0
1 ft (z)
nt = dz
2πi C ft (z)
Since both ft (z) and ft0 (z) are continuouis functions of (t, z), the above in-
tegral is a continuous function of t. But it is integer valued and hence must
be constant. Thus n0 = n1 .

Theorem 29: analytic maps are open maps


A non-constant analytic function maps open sets to open sets.
This is expressed by saying that analytic functions are open maps (if
non-constant).
Proof: Take open U in domain f and w0 ∈ f (U ). Need to show a disc
around w which is contained in f (U ). Let z0 ∈ U with f (z0 ) = w0 . Since f is

109
non-constant, so is f − w0 and its zeros are isolated. Get a disc D = D(z0 , r)
such that D ⊂ U and f (z) 6= w0 for all z ∈ D \ {z0 }. In particular, the point
w0 is outside the compact set {f (z) : z ∈ ∂D}, get  > 0 such that for z ∈ ∂D
we have |f (z) − w0 | > . We show D(w0 , ) ⊂ f (U ). Let w ∈ D(w0 , ). then
for every z ∈ ∂D we have |f (z) − w0 | >  > |w0 − w|. Thus applying Rouche
(g is constant function) we see that f − w0 and f − w = (f − w0 ) + (w0 − w)
have the same number of zeros. Since f − w0 has at least one zero (namely,
z0 ) in D, we conclude that f − w also has at least one zero in D. In other
words there is z ∈ D with f (z) = w. Proof is complete.

Theorem 30: local conformality


If f is analytic and f 0 (z0 ) 6= 0, then there is a neighbourhood of z0 which
is conformally mapped onto a region.
Do exactly as above. get D(z0 , r) as above and also such that f 0 is non
zero in D. Remember that f 0 is also continuous (actually holomorphic). Let
Image of D(z0 , r) be U . Then first of all it is open by previous theorem.
Further, in the previous proof we have used that f − w0 is a zero of order at
least one. Now you use that it is of order exactly one (because f 0 (z0 ) 6= 0)
to see for any w ∈ U f − w also has a zero of order exactly one. In other
words f is one-one on D onto U . Further f is opne tells you that f is a
homeomorphism from D to U . The inverse is differentiable. In fact, let the
inverse function be g : U → D. Fix w ∈ U . Let wn → w and wn 6= w for all
n. Let f (zn ) = wn and w = f (z). The fact that f is homeomorphism tells
zn → z. Since (f (zn ) − f (z))/(zn − z) converges to f 0 (z) 6= 0, we see that
g(wn ) − g(w) zn − z 1
= = 0 .
wn − w f (zn ) − f (z) f (z)
In other words f is a conformal map: D to U .

Theorem 31: Maximum principle 1


f is a non-constant analytic function in a region Ω, Let D be a disc with
D ⊂ Ω. then

max{|f (z)| : z ∈ D} = max{|f (z)| : z ∈ ∂D}

(max is attained on the boundary)


proof: (Landau) We only need to show that for z ∈ D, |f (z)| ≤ M =
max{|f (z)| : z ∈ ∂D} For any integer k ≥ 1, considering the analytic func-
tion f k , Cauchy gives
[f (ζ)]k
Z
k 1
[f (z)] = dζ; z ∈ D
2πi ∂D ζ − z

110
Let us fix a z ∈ D and say d(z, ∂D) = α and r is radius of the disc. Then
using the fact that integral, in modulus, is at most max of integrand times
length of curve, we get
p
|f (z)|k ≤ M k /(rα); |f (z)| ≤ M k 1/(rα)

This being true for all k ≥ 1 we get the result.

Something stronger is true, namely for every z ∈ D, actually |f (z)| < M ,


unless f is a constant.

Theorem 32: Maximum principle 2


f is an analytic function in a region Ω and |f (z)| attains maximum value
at a point in Ω. Then f is a constant function.
(max can not be attained inside, unless constant function.)
Proof: Recall if D = D(a, r) and D ⊂ Ω, then
Z 2π
1
f (a) = f (a + reiθ )dθ
2π 0

If possible let sup{|f (z)| : z ∈ Ω} = M = |f (a)| where a ∈ Ω. Note that sup


is maximum because f is finite in Ω. Then
Z 2π Z 2π
1 iθ 1
M = |f (a)| ≤ |f (a + re )|dθ ≤ M dθ = M
2π 0 2π 0
Thus Z 2π
1
|f (a + reiθ )|dθ = M
2π 0
But integrand is non-negative and continuous and at most M . Thus wecon-
clude that |f | = M on ∂D. This being true for any disc as above (in particular
for 0 < s < r) we conclude that |f | = M on D. But an analytic function
of constant modulus must be a constant. Thus there is a complex number c
with |c| = M such that f ≡ c in D. Clearly then f ≡ c on all of Ω.

Another proof: a nonconstant analytic function maps open sets to open


sets. Thus if a ∈ D, then range of f on D is an open set containing f (a).
But clearly if an open set U contains a point c, then there are w ∈ U with
|w| > |c|. Thus f (a) can not be the maximum in modulus. This is so what-
ever be the point a ∈ Ω.

Theorem 33: Maximum principle 3

111
Ω is a bounded region; f is a continuous on Ω and is analytic in Ω, then
the maximum of f is attained on the boundary ∂Ω. if it is assumed in the
interior, then the function is constant on Ω.
Proof: Since the set is compact and f is continuous, we conclude that
M < ∞ and is attained. If it is attained at a point in Ω, then previous
theorem already tells you that it must be a constant on Ω and hence on Ω.

Thus for example if f is a continuous function on the closed unit disc and
analytic in the open unit disc then it attains maximum ONLY on the unit
circle unless the function is a constant on all of the closed disc.
∂ ∂
Digression: ∂z and ∂z :
I was not planning to do this because we really do not need this. Also
needs a conceptual understanding. But since you asked, let us understand.
Except to make sense of the statement ‘an analytic function depends only
on z and not on z’; I do not see immediate use.
Let us recall astory in a known territory. Consider R2 , with points (x, y)
and f : R2 → R or f : R2 → C. Sometime it is good to change your
coordinate system. This you have seen in matrices how for a symmetric
matrix, eigen vector basis helps you understand action of the matrix in a
simple way. Even if you forgot the spectral representation, here is the idea.
Consider the function f (x, y) = x3 + 3x2 y + 3xy 2 + y 3 . If you had chosen
basis u = (x + y) and v = (x − y), then this function is simple to calculate:
just cube of the first coordinate. But now suppose we want to stick to these
new coordinates and try to understand ‘rate changes’ relative to this system.
Your original function f on this new stage is f ( u+v
2
, u−v
2
). Thus if you want
∂f ∂f
to understand ∂u and ∂v what would you do? You consider

u+v u−v u+v u−v


(u, v) −→ ( , ) −→ f ( , )
2 2 2 2
with derivatives    
1/2 1/2 f1
;
1/2 −1/2 f2
∂f ∂f
Here f1 and f2 are respectively ∂x
and ∂y
. Your composition rules tell you

∂f
!   
∂u 1/2 1/2 f1
=
∂f 1/2 −1/2 f2
∂v

112
Or
1 ∂f ∂f
∂f
! !
∂u
[
2 ∂x
+ ∂y
]
∂f
= 1 ∂f ∂f
∂v
[
2 ∂x
− ∂y
]
Or
1 ∂ ∂

! !
∂u
f [
2 ∂x
+ ∂y
]f

= 1 ∂ ∂
∂v
f [
2 ∂x
− ∂y
]f
which can symbolically be stated as
∂ 1 ∂ ∂ ∂ 1 ∂ ∂
= [ + ]; = [ − ] (♣)
∂u 2 ∂x ∂y ∂v 2 ∂x ∂y
We now want to do similar thing with C. So far we were thinking of C
as pairs (x, y) of real numbers. Instead of real number pairs,I want to think
of C as pairs (z, z). Obviously, if you know the real numbers x, y, you know
the complex numbers z, z. Conversely if you know the two complex numbers
z, z, you know the real numbers x, y. You face a conceptual issue already: if
I know z, do I not know its real and imaginary parts. No. Just as when x + y
is given you do not know what exactly x, y are you do not know here the real
and imaginary parts. In a sense x + iy is tied and given to you – difficult
to imagine, because you have been brainwashed that a complex number is
x + iy and giving a complex number means telling x and y. Think about it.
Now let us think of C as not (x, y) but as pairs (z, z). Thus we have
exactly (not really!) the situation that we discussed earlier. Suppose you
have a function f : R2 → R or f : R2 → C.
z+z z−z z+z z−z
(z, z) −→ ( , ) −→ f ( , )
2 2i 2 2i
We are now thinking of z and z as the new variables (just like, u, v earlier)
The derivatives, if you extend your imagination, of these maps are
   
1/2 1/2i f1
;
1/2 −1/2i f2

Of course, as earlier Here f1 and f2 are respectively ∂f


∂x
and ∂f
∂y
. Your com-
position rules ‘should’ lead you, carrying the same steps as above, to the
symbolic relation – analogue of ♣.
∂ 1 ∂ 1 ∂ ∂ 1 ∂ 1 ∂
= [ + ]; = [ − ] (♠)
∂z 2 ∂x i ∂y ∂z 2 ∂x i ∂y
Of course, earlier calculation was justified because we had precise definitions
of partial derivatives etc. Now there are no definitions of partial derivatives

113
w.r.t. z and z. So we take the above defining equations. Why are there
no definitions? earlier the ‘full’ (x, y) plane is identified with the ‘full’ (u, v)
plane. Now NOT so! It is foolish to think we identified C with C×C. We only
identified with complex one dimensional something (manifold) consisting of
all pairs (z, z). Thus for each z, there is only one point on the horizontal
line and also for each z there is only one point on the vertical line. Can you
define partial derivatives? Think.
∂ ∂
Thus (♠) is taken as ‘the definition’ of the quantities ∂z , ∂z .
If you consider an analytic function f = u + iv, where u = <(f ) and
v = =(f ), then f1 = u1 + iv1 and f2 = u2 + iv2 . The Cauchy-Riemann
equations do indeed tell you, after simplification (please do) that for an
analytic function we have

f ≡ 0.
∂z
In this sense an analytic function depends ‘only’ on z.
Having gotten these symbols, you can use them too. For example, Laplacian
can be described as ‘mixed derivative’, that is
∂ ∂
4∆ = .
∂z ∂z

114
BVR/CMI Complex Analysis 19-04-2022

Suppose you have an analytic function in a deleted neighbourhood of


a. Then there are three possibilities: Either the point is good but you did
not know; that is, you can define at ‘a0 too so that you have a holomorphic
function through out. Or it is a bad point but after a little modification you
can correct; that is the function is heading to infinity at a certain rate and if
you compensate, you can define at a as well to get a holomorphic function.
Or finally it is an ugly unpredictable point and no modification can make it
holomorphic at the point a.
To give you a carricature in the real context, consider, on R, the function
f (x) = (sin x)2 /x. you can not put x = 0 here, however, you can define at
x = 0 so that the function is nice. Consider g(x) = sin x/x3 . Again you
can not put x = 0, and you can not define this at zero. however this is
preddictable, approaching infinity at a rate x2 and if you compensate for this
you can modify. Thus x2 g(x) is a nice function that can be defined at x = 0.
The function h(x) = sin(1/x) is so bad, it is unpredictable near x = 0, you
can not easily modify it. However, in the context of complex functions, this
unpredictability is spectacular (and universal).

Theorem 34: Classification of singularities


Ω is a region and a ∈ Ω. Let f be analytic in Ω \ {a}. Then There are
exactly three possibilities:
(i) we can extend f to all of Ω as an analytic function. Either f (a) 6= 0 OR
f (a) = 0. In the later case, either f ≡ 0 or for some k ≥ 1, f (z) = (z−a)k g(z)
where g is analytic and g(a) 6= 0.
(ii) lim |f (z)| = ∞. Then there is a unique integer k ≥ 1 such that,
z→a
(z − a)k f (z) is analytic on all of Ω and takes non-zero value at a.
(iii) On every neighbourhood of a, range of f is dense in C; that is given
any neighbourhood of a and any complex number c and any  > 0, there is
z in the given neighbourhood of a with |f (z) − c| < .

In case (i) the point a is called a removable singularity. when f is of the


form (z − a)k g with g(a) 6= 0, we say a is a zero of order k. There is only one
such k.
In case (ii) it is called a pole. It is pole of order k. In this case again
There is only one such k.
In case (iii) it is called an essential singularity.
A function which is analytic in a region except for poles is said to be
meromorphic in that region.

115
Proof: Step 1: Suppose

lim |z − a|r |f (z)| = 0 for some real number r (♠)


z→a

Then we show (i) or (ii) must occur. Also conversely. if (i) or (ii) occurs
then (♠) holds.
If (♠) holds for one r then it holds for larger r too; so take an integer m
(positive or negative) such that lim (z − a)m f (z) = 0. By extension theorem,
z→a
you can extend (z − a)m f (z) to an analytic function on all of Ω. Naturally
the extension takes value zero at a. Thus a is a zero of the extension. If the
extension is identically zero, then clearly f ≡ 0 as well. Suppose that the
extension is not the zero function. So by finite Taylor, we can express

(z − a)m f (z) = (z − a)l g(z); g(a) 6= 0

for some integer l ≥ 1. Or

f (z) = (z − a)l−m g(z); g(a) 6= 0

If l = m: we have an extension of f to all of Ω with f (a) = g(a) 6= 0. Case


(i) of the theorem holds.
If l > m: we have f (z) = (z − a)k g(z) with k = l − m ≥ 1; g is analytic and
g(a) 6= 0. a is zero of order k. case (i) of the theorem holds.
If l < m: with k = m − l ≥ 1, we have f (z) = (z − a)−k g(z) with g analytic
and g(a) 6= 0. case (ii) of the theorem holds.
For the converse, suppose (i) holds. The existence of extension already tells
lim |z−a||f (z)| = 0 so that ♠ holds. If (ii) holds and a is a pole of order k ≥ 1,
z→a
let f (z) = (z − a)−k g(z) with g analytic. Clearly lim |z − a|k+1 |f (z)| = 0
z→a

Step 2: Let us now consider analogous to (♠)

lim |z − a|r |f (z)| = ∞ for some real number r (♣)


z→a

We claim that (♣) implies either (i) or (ii) must occur. Conversely, if (i) or
(ii) occurs then (♣) must hold. [Did you see symmetry between zero/infinity
or south/north poles]
Proof exactly same as above. If (♣) holds for some r, it holds for
smaller r too, so take integer m for which it holds. From our discussion
on poles, (z − a)m f (z) has a pole at the point a, say of order l ≥ 1.
Thus (z − a)m f (z) = (z − a)−l g(z) with g holomorphic and g(a) 6= 0. Or
f (z) = (z − a)−l−m g(z) with g holomorphic and g(a) 6= 0.
If −m − l = 0, then f has an extension with f (a) 6= 0 and (i) of theorem

116
holds
if −m − l ≥ 1, then f has an extension and a is a zero of order −(m + l) ≥ 1
and case (i) holds.
If −m − l ≤ −1 then a is a pole of f of order m + l ≥ 1.
Conversely, if (i) holds, in case for the extension f (a) 6= 0, then clearly
lim |z − a|−1 |f (z)| = ∞; where as if a is a zero of order k for the extension,
z→a
f (z) = (z − a)k g(z) with g(a) 6= 0, then lim |z − a|−k−1 |f (z)| = ∞. Thus in
z→a
any case (♣) holds. If (ii) holds, then clearly lim |f (z)| = ∞.
z→a

To complete the proof we now show that if (iii) fails then (i) or (ii) must
occur. So let us fix a neighbouthood of a, say D(a, δ) and a complex number
c and an  > 0 such that for every z ∈ D(a, δ) \ {a} we have |f (z) − c| ≥ .
Then for every r < 0 we have lim |z − a|r |f (z) − c| = ∞. In view of step 2,
z→a
(i) or (ii) must hold for f (z) − c. In view of step 1, we must have for some s
lim |z − a|s |f (z) − c| = 0. Clearly s > 0. But then
z→a

lim |z − a|s |f (z)| ≤ lim |z − a|s |f (z) − c| + lim |z − a|s |c| = 0


z→a z→a z→a

In other words ♠ holds for f and so (i) or (ii) must occur for f .

Theorem 35: Argument principle:


f meromorphic in a region Ω. C is a circle/rectangle/simple closed curve
which along with its interior is contained in Ω. assume that no zero or pole
of f is on the curve C. Then
Z 0
1 f (z)
dz = number of zeros inside C − number of poles inside C
2πi C f (z)
Here everything is counted with multiplicities.
Proof: Exactly same as the earlier theorem on number of zeros, direct
consequence of Residue theorem. The integral is just sum of residues. f 0 /f
has poles if f has a pole or f has a zero. At this stage you may suspect that
a zero of f may nit be a pole, in case the numerator f 0 also has zero there
and cancels the sero of f . But you are sure that there are nmo other poles.
If a is a zero of f of order k, then we have in a neighbourhood of a,
f (z) = (z − a)k g(z); g(a) 6= 0
f 0 (z) k g 0 (z)
= +
f (z) z−a g(z)
0
Since g(a) 6= 0, we see that g /g is holomorphic at a (meaning, in a neigh-
bourhood of a), so that Res(f 0 /f, a) = k, the order of the zero. If b is

117
a pole of f of order k, then f (z) = (z − b)−k g with g(b) 6= 0. Thus
f 0 = −k(z − b)−k−1 + (z − k)−k g 0 so that

f0 −k g0
= +
f z−b g
Since g(b) 6= 0, g 0 /g is holomorphic near b so that Res(f 0 /f, b) = −k. The
proof is complete.

You can state the above theorem for a simple closed curve within a disc,
On the right side you take n(a, C) for each zero and pole. But we shall not
enter into this generality.

Theorem 36: Schwarz’s reflection principle


Suppose Ω is a region symmetric about x-axis and Ω+ = {z ∈ Ω : =z >
0}, Ω− = {z ∈ Ω : =z < 0} and Ω0 = {z ∈ Ω : =z = 0}. Suppose that f is
a continuous function on Ω+ ∪ Ω0 such that it is analytic in Ω+ and it takes
real values on Ω0 . then you can extend f as a analytic function on all of Ω.
In fact, if we define f (z) = f (z) for z ∈ Ω− , then it is analytic in Ω.
Proof: Note that Ω being connected non-empty open set, all the sets
Ω , Ω− and Ω0 are non-empty. First observe that f as defined above on
+

Ω is continuous. We only need to show that it is continuous at points of


Ω0 . By hypothesis, if zn ∈ Ω+ ∪ Ω0 and zn → z ∈ Ω0 , then f (zn ) → f (z).
So let zn ∈ Ω− and zn → z ∈ Ω0 . then zn → z = z and zn ∈ Ω+ . Thus
f (zn ) → f (z). Thus f (zn ) = f (z n ) → f (z) = f (z) as f (z) is real.
To show f is analytic, we use Morera’s theorem. Take any R triangle ∆
which along with its interior is contained in Ω. Shall show f = 0. If

∆ ⊂ Ω+ , then hypothesis does. If ∆ ⊂ Ω+ ∪ Ω0 and meets Ω0 , there are
two possibilities. Either a vertex is on Ω0 or one side is on Ω0 . In the first
case, you can express ∆ as sum of two curves, one a small trangle ∆1 with
this vertex and another C a simple closed curve (quadrilateral) C ⊂ Ω+ . ∆1
can be made as small as we please so that its length is very small, hence
integral over that is very small and integral over C is zero. In the second
case raising the base a litle bit we can express ∆ as sun of two curves a
triangle ∆1 ⊂ Ω+ and a narrow strip quadrilateral C with one side on the
x-axis. The integral over ∆1 is zero. The integral over C can be made very
small because the integral over the two sides x-axis and its parallel side are
integrated in opposite directions and by uniform continuity this can be made
small; integral over the other two sides is small because lengths are small.
If ∆ ⊂ Ω− ∪ Ω0 , then by taking its reflection we can argue integral is zero.
For general triangle ∆, split it into two parts, above and below, these parts

118
may not be triangles but quadrilaterals, but they can be split into triangles.
This completes the proof.
R
[is it enough to show f = 0 for ∆ which along with interior is contained

in Ω? Yes, because, differentiability being a local property, we restrict the
entire drama to a disc contained in Ω]

119
BVR/CMI Complex Analysis 22-04-2022

Theorem 37: Schwarz’s Lemma


D open unit disc. f : D → D is analytic and f (0) = 0; Then
(i) |f (z)| ≤ |z| and If |f (z)| = |z| for some z 6= 0 then f (z) = cz for some
number c with |c| = 1.
(ii) |f 0 (0)| ≤ 1 and if |f 0 (0)| = 1 then f (z) = cz forsome number c with
|c| = 1.
Proof: First we claim that f (z)/z is analytic, that is, has removable
singularity at zero. This is easy, f (0) = 0 tells that the Taylor series around
zero is
f (z) = a1 z + a2 z 2 + · · ·
so that
f (z)
= a1 + a2 z + · · · ; z ∈ D \ {0}
z
Thus f (z)/z is an analytic function on D.
Take any 0 < r < 1. On the circle of radius r we have |f (z)/z| ≤ 1/|z| =
1/r so that by Maximum modulus principle we conclude that if |z| ≤ r,
then |f (z)/z| ≤ r. Thus if the disc of radius r is denoted by Dr , then for
every r < 1, we have |f (z)/z| ≤ r on Dr . If you now take any z ∈ D then
z ∈ Dr for all r with |z| ≤ r < 1 and hence |f (z)/z| ≤ 1/r for all these r
and hence taking limit we get |f (z)/z| ≤ 1. If for some non-zero z0 ∈ D, we
have |f (z0 )| = |z0 |, that is |f (z0 )/z0 | = 1 then the maximum is attained at
an interior point and thus f (z)/z is a constant, say α. Thus f (z) ≡ αz. In
particular we have

|z0 | = |f (z0 )| = |αz0 | = |α||z0 |; |α| = 1

because z0 6= 0. Thus f (z) = eiθ z for some θ.


The above tells that |f (z)/z| ≤ 1 for all z 6= 0 and so by taking limit
as z → 0 we get |f 0 (0)| ≤ 1. Finally, if g(z) = f (z)/z then g is analytic in
D (remember g(0) = f 0 (0)) and |g(z)| ≤ 1. If |f 0 (0)| = 1, then g attains
maximum, in modulus, at an interior point and so must be a constant, that
is f (z) = αz again as above. This completes the proof.

As usual D denotes the open unit disc. Any biholomorphic (one-one, onto,
holomorphic, inverse also holomorphic) of a region is called an automorphism.
Theorem 38: Automorphisms of the disc:
For every θ and for every a ∈ D, the map
a−z
f (z) = eiθ
1 − az

121
is an automorphism of the disc. These are all the automorphisms of the disc.
Further,the only automorphisms of the disc that fix zero are rotations.
Proof: Rotations are automorphisms. Rotation means a map of the form
z 7→ eiθ z for some θ. It is called rotation by θ. Its inverse is rotation by −θ.
Here are some others. Fix a ∈ D and consider
a−z
ϕa (z) =
1 − az
This is an automorphism. In fact it interchanges a and zero. It is its own
inverse. Since composition of autos is again an auto we see that maps of the
form
a−z
f (z) = eiθ
1 − az
for some θ and some a ∈ D are autos.
Conversely suppose that f is an automorphism of the disc. Get a such
that f (a) = 0. Consider the map g(z) = f oϕa .Then this is an auto of D.
Also g(0) = 0. Schwarz tells us

|g(z)| ≤ |z|; ∀z ∈ D (•)

Applying Schwarz to g −1 we get

|g −1 (z)| ≤ |z|; ∀z ∈ D

Applying this to g(z) ∈ D we get |z| ≤ |g(z)|. This with (•) gives us

|g(z)| = |z|; ∀z ∈ D

Now Schwarz tells that we must have g(z) = eiθ z. In particular,

g(ϕa (z)) = eiθ ϕ(z).

But notice
goϕa = f oϕa oϕa = f
Thus we have
f (z) = eiθ ϕ(z).
The last statement is immediate because, then a = 0. this completes the
proof.

Theorem 39: Automorphisms of upper half plane:


Let
   
a b
SL2 (R) = M = : a, b, c, d ∈ R; ad − bc = 1
c d

122
The for every M ∈ SL2 (R), the map
az + b
fM (z) = z∈H
cz + d
is an auto of H. These are the only autos of H.
Proof:
This is where conformality helps. Onec you solved the problem for D;
you have solved it for H too because they are comformally equivalent. Of
course there are several conformal maps. The stnadard one will do. Let us
recall.
If F : H → D and G : D → H are given by
i−z 1−z
F (z) = ; G(z) = i
i+z 1+z
then G = F −1 and establishes a conformal equivalence between H and D.
Towards proof of the theorem, first observe that For M as above, if z ∈ H
(ad − bc)
=fM (z) = =z > 0
|cz + d|2
showing fM maps H into itself. Since fM ofN = fM N and fI is the identity
map, we conclude that fM is one-one onto with inverse fM −1 . Differentiabil-
ity is clear. (note cz + d is never zero because c, d are real while =z > 0).
To
 prove the converse,
 first let us observe interesting matrices. If ρ =
cos θ − sin θ
then this is a rotation of the disc when points in disc are
sin θ cos θ  
x
viewed as x+iy = . But this is a matrix in SL2 as well. What is fρ and
y
how is it related to the rotation of the disc? We claim that F ofρ = e−2iθ F
so that F ofρ oF −1 is rotation by −2θ in the disc. This is easy to verify.
Towards the converse let us first show that an auto f of H that fixes i
is given by some M . What is special about i? Under the above conformal
equivalence i corresponds to zero and we know conformal maps of disc fixing
zero! Clearly F of oF −1 is an auto of the disc fixing the origin and hence must
be rotation. But from earlier observation any rotation is F ofρ oF −1 for some
ρ. Hence F of oF −1 = F ofρ oF −1 for some ρ, that is f = fρ as wanted.
We claim that given any z ∈ H there is an N such that fN (z) = i. This
2 2
is easy.  with c = =z/|z| (makes sense because =z > 0. Take
 Take c real
0 −1/c
M = to see ={fM (z)} = 1. Now translate by a matrix of the
c 0 
1 b
form M1 = . Clearly N = M1 M will do the job. Also notice that
0 1

123
given any z, there is a matrix N such that fN (i) = z.
Returning to the proof of the theorem, let f be any auto. Say, f (b) = i. Fix
N so that fN (i) = b. Set g = f ofN so that g is auto of H with g(i) = i. From
what has been proved above we see that there is an M such that f fN = fM
or f = fM fN −1 . Thus M N −1 achieves f . Proof is complete.

Since M and −M give the same auto, the automorphism group is not
SL2 (R) but P SL2 (R), the quotient. [SL is special linear group; PSL is pro-
jective special linear group.]

124
BVR/CMI Complex Analysis 26-04-2022

You see how the biholomorphic map of D onto H helped in analyzing the
auto group of one using the other. A conformal map f : Ω1 → Ω2 means a
1-1 holomorphic map of Ω1 onto Ω2 with inverse also holomorphic. Here is
the main theorem about regions conformal to D.

Theorem 40: Riemann Mapping Theorem (statement):


If Ω is a simply connected proper subset of C and z0 ∈ Ω, then there is a
conformal map f : Ω → D such that f (z0 ) = 0.
A connected open set Ω ⊂ C is simply connected if its complement in
the extended plane is connected, that is, C ∪ {∞} \ Ω1 is connected. For
example D, H are simply connected. Of course, their complements in C
are connected too. If you take a strip, say, Ω = {−1 < <z < 1} then
this is simply connected. However notice that its complement in C is not
connected. For example punctured disc D \ {0}; annulus {1 < |z| < 2} are
not simply connected These sets have holes in them and not having holes
characterizes this concept. But how do you make the concept precise. By
saying any simple closed curve can be deformed to a point. Since you learnt
this concept in topology using homotopy, we shall not discuss in detail. Of
course the equivalence of these two definitions needs proof which we shall
not do.
Rrturning to the Riemann mapping theorem, the idea of the proof is to
consider all 1-1 holomorphic maps of Ω into D with f (z0 ) = 0. Show there
is one with range all of D. This is done by making f expand as large as
possible, or |f 0 (0)| as large as possible. We shall not prove this.
Note that C is not conformal to D by Liouville theorem.

As already said, conformality allows transfer of problems from one re-


gion to a known region. Here is one specific example, known as Dirichlet
problem. Consider open unit disc D, its boundary unit circle C. Given a
real continuous function f on C; wanted a harmonic function u on D so that
the function as defined is continuous on the closure of the disc D. In other
words a continuous extension of the given f to all of D so that it is har-
monic in D. Such problems are called boundary value problems and arise in
physics. You know ‘something’ on the boundary; want to fill ‘inside’. These
problems necessitate understanding not only conformal equivalence but be-
havious of the map at the boundary. For example we have a conformal map
from D to H; how does the boundary of D gets transformed to boundary of H.

Before proceeding further, let us make a brief execution (or pretend so)

125
of proofs of Theorem 4 and Theorem 27: Cauchy Residue theorem.
Theorem 4 is executed using theorem 3; in the same manner as Theorem 2
was executed using Theorem 1. One shows that f has a primitive. One Rdoes
this by exhibiting F . Fix a point w in the domain and define F (z) = Cz f
where C is a path from w to z consisting of horizontal/vertical segments.
If one shows that this is well defined, does not depend on the path, then
its differentiability follows as in basic Theorem 0. In theorem 2 this was
achieved by prescribing specific path: horizontal/vertical. To show partial
derivatives, we used two paths: one ending with horizontal segment and other
ending with vertical segment. That these paths give same values is precisely
theorem 1.
To prove theorem 4, one shows, as in theorem 2, a primitive. Now because
there are exceptional points you need to take horizontal/vertical paths that
avoid the exceptional points. If you take two such paths they give the same
value is again achieved with not one rectangle, but many rectangles and
theorem 3. We shall rest the matter here.
Cauchy Residue theorem is executed as follows. Imagine there is only one
pole a and you are integrating over ∂D(a, r). Expand around a, if a is pole
of order k ≥ 1, we have
−2 ∞
X cj c−1 X
f (z) = + + cj (z − a)j
−k
(z − a)j (z − a) 0

The last sum is holomorphic and so integral over ∂D is zero. The first sum
has primitives and so integral is again zero. Only the middle term remains
and by winding number consideration, it equals 2πic−1 .

126
Assuming that there is only one pole, suppose the contour C is a toy
contour.
Just to fix ideas, let us consider the Contour shown in the picture:
ABCDEF GHIA. Our function is defined in an open set containing this
Contour and interior, except at the point a which is a pole. Want to show
Z
f (z)dz = 2πi Res(f, a)
C

We take a small disc D around the pole as shown. Consider the two contours:
C1 = P ABSR(↓)QP and C2 = P Q(↑)RSCDEF GHIP . None of these
Contours contain the point a. By splitting each into rectangles, triangles,
semicircles etc and (using  s near Q, R) we can show that integral over each
is zero. Thus Z Z
f (z)dz + f (z)dz = 0
C1 C2

But a careful analysis shows that the integrals over P Q and QP cancel, so
do RS and SR. Thus the above equation reduces to
Z Z
f (z)dz − f (z)dz = 0
C ∂D

The negative sign for second term is because ∂D appears in clockwise direc-
tion in C1 + C2 . The first rem in the above display is of our interest and
second term is known from above argument. If there are more (finite number
of) poles, we take disjoint discs around each and take connecting lines etc.
Try to do with three poles.
One reason why we do not execute all the details is because we are con-
vinced (but that is not good enough, we are many times convinced of wrong
things too!) and more over there is a general theorem that we see below.

Theorem 41: Big Cauchy Theorem (shall not prove):


Let Ω be a simply connected
R region and f be analytic on Ω. Then for
every closed curve C ⊂ Ω f (z)dz = 0.
C
remember that in proof of the first C-G theorem, we divided the rectangle
into smaller and smaller parts and finally ended with ‘a point in the rectangle
which is in our region’. This led to a contradiction. If it so happens that
there is a hole and this final point is not in our region then there would be
no contradiction. you can see the winding number theorem.

We had Taylor expansion around a for analytic function, expansion in-


volving only non-negative powers of (z − a). If a is a pole for f , then we

128
had Taylor-Laurent expansion which had again powers of (z − a) but now
finitely many negative powers appear. Can we expand if a is an essential
singularity? Yes, but now many negative powers appear.
Theorem 42: Laurent expansion, at a singularity
Suppose Ω contains an annular region A = {r ≤ |z − a| ≤ R} and f is
analytic in Ω. Then we have an expansion valid in the annular region

X
f (z) = ck (z − a)k
−∞

that is, there are numbers {cj } such that


∞ ∞
X c−n X
f (z) = + cn (z − a)n ; z∈A
n=1
(z − a)n n=0

This is called Laurent expansion of f . Proof is not difficult, given all the
work we have done so far (see HA).

129
BVR/CMI Complex Analysis 29-04-2022

Gamma/Zeta functions
For real numbers s > 0 the following integral converges
Z ∞
e−t ts−1 dt
0

For s = 1, integrand is e−t , its integral from zero to A can be calculated


and as A → ∞ it converges to one. For t > 1 the fucntion is a bounded
continuous function on [0, 1] taking value zero at zero. So that this integral
is finite. On the range [1, ∞), if n is an integer larger than s − 1, then using

(t/2)n
tn = 2n n! ≤ 2n n!et/2
n!
we see
e−t ts−1 ≤ e−t tn ≤ Ce−t et/2 = Ce−t/2
and hence integral over [1, ∞) is finite. Let now 0 < s < 1. On the interval
(0, 1) we have
e−t ts−1 ≤ ts−1
which is integrable. On [1, ∞) we have e−t ts−1 ≤ e−t and is hence integrable.
Thus the above integral is finite for s > 0.
The value of the integral is denoted by Γ(s). Only for special s, you can
explicitly evaluate the Gamma integral. For example for integers n ≥ 1,
Γ(n) = (n − 1)! (convention 0! = 1). In fact the desire to define factorial for
non-integers is supposed to be a reason for Euler to discover Gamma integral.
Of course, it appears in many contexts: Probability, statistics, physics and
so on. In fact the volume of the unit ball in n-dimenaional Euclidean space
is
π n/2
.
Γ( n2 + 1)

We also know Γ(1/2) = π.
It is easy to see, using integration by parts, that for s > 0,
1
Γ(s + 1) = sΓ(s); Γ(s) = Γ(s + 1) (♠)
s
This simple equation, unfolds a beautiful story. For example you can define
Γ(s) for all s different from {0, −1, −2, −3 · · · }. Use the formula (♠) to define
for −1 < s < 0, then for −2 < s < −1 and so on. But there is more to the
story.

130
First let us observe that this function can be defined for s ∈ C, <s > 0
this is simply because
|e−t ts−1 | = |e−t t<s−1 |
Keep in mind that for positive real numbers t, by definition, log t is the usual
one you know, and no need to confuse with multivalued nature of complex
logarithm. Thus

|ts−1 | = |e(s−1) log t | = |e(<s−1) log t | = |t<s−1 |.

You can extend the Gamma function for more complex numbers s, in fact
for all s except {0, −1, −2, −3 · · · } as follows:
If <s > −1 put Γ(s) = (1/s)Γ(s + 1) then this is defined for all s with
<s > −1, s 6= 0. Carefully observe that even if <s = 0 the formula defines
Γ(s) as long as s 6= 0.Now define for <s > −2 and so on. Notice that this
definition does not alter the values of Γ(s) for <s > 0. It is also clear thast
the function so defined is holomorphic. For example
1 √
Γ(−1/2) = Γ(1/2) = −2 π
−1/2

This process of extending a function to a larger domain from a smaller


domain is called ‘analytic continuation’.
For analytic continuation of the Gamma function, we used the functional
equation (♠). Many a times such functional equations are not available.

f (z) = 1 + z + z 2 + z 3 + · · · |z| < 1

is a holomorphic function on unit disc D. Note that at every point on the


unit circle, the above series fails to converge. Now let us coinsider another
function
 
1 z+1 z+1 2 z+1 3
g(z) = 1+( )+( ) +( ) + ··· |z + 1| < 2
2 2 2 2

. This is defined on disc of radius two centered at −1. This extends f and
can be regarded as analytic continuationn of f . Of course, on close scrutiny,
you realize that there was 1/(1 − z) in the background and f is nothing but
its Taylor series at zero while g is Taylor expansion at −1.
It is interesting to note that we are using s for the argument of the func-
tion, rather than the customary z. Well, this is the standard notation for
these functions. Nothing would go wrong if you used z.

131
The Riemann zeta function is defined as follows. First observe that for
real numbers s > 1 the series
1 1 1
s
+ s + s + ···
1 2 3
converges. Just as in case of Gamma function this series converges even
for complex numbers s provided <s > 1. Further it defines a holomorphic
function in that region.
1 1 1
ζ(s) = s
+ s + s + ··· <s > 1
1 2 3
This can be continued analytically for all complex values except for the two
values n = 0, 1, though not simple to prove. Here is a nice relatiion (for
s 6= 0, 1):
 
−s/2
s
−(1−s)/2 1−s
π Γ ζ(s) = π Γ ζ(1 − s).
2 2

In other words, if you denote the function on the left side by χ(s), then what
we said is this: χ(s) = χ(1 − s). If we take s = 2 in the above equation and
simplify we get
ζ(−1) = −1/12.
This is usually (and dramatically) stated as
1
1 + 2 + 3 + ··· = −
12
Such equations you should understand carefully.

Actually the True Complex Analysis starts now. No doubt what we have
done is very important but is a mere beginning, it sets the stage for several
profound dramas to unfold.

Final: May 10 Tuesday 9:30 am to 12 noon in cmi campus.


It is closed book exam.

132

You might also like