Notes1 PDF
Notes1 PDF
Notes1 PDF
Michael Hutchings
1 Complex numbers
1.1 Basic definitions
Informally, the complex numbers are obtained by starting with the real num-
bers and introducing a new number i whose square is −1. More formally:
That is, we multiply by using the distributive law and replacing i2 by −1.
The set C, together with the addition and multiplication operations de-
fined above, is a field (review the definition of this if necessary). One can
verify most of the field axioms by straightforward calculations (however it
is not always obvious from the above definition that these will work out!).
The only axiom which is not straightforward to verify is that every nonzero
complex number has a multiplicative inverse. To prove this, one observes
that if x + iy 6= 0, then (x + iy)(x − iy) = x2 + y 2 which is a nonzero real
number, so
x − iy x y
2 2
= 2 2
− 2 i
x +y x +y x + y2
1
More formally, we could say that a complex number is an element of the set R2 . We
identify the ordered pair (x, y) ∈ R2 with the expression x + yi.
1
is a (the) multiplicative inverse of x + iy.
For an alternate proof that C is a field, you may recall a theorem from
algebra stating that if F is a field and if p ∈ F [t] is a nonconstant polynomial
which is irreducible over F , then the quotient ring F [t]/(p) is a field. Now
take F = R and p = t2 + 1. The polynomial t2 + 1 is irreducible over R since
−1 has no real square root, so by the above theorem R/(t2 + 1) is a field. On
the other hand you can check that every element of R/(t2 + 1) has a unique
representative of the form x + yt with x, y ∈ R. Replacing t by i, we obtain
C as defined previously.
We represent complex numbers geometrically as points in the x, y plane.
The x axis is called the real axis, and the y axis is called the imaginary
axis. Addition of complex numbers is now given by the familiar addition of
vectors.
To describe multiplication geometrically, we first make two additional
definitions. If z = x + iy is a complex number, define its absolute value by
p
|z| := x2 + y 2 .
z = r(cos θ + i sin θ)
where r = |z| and θ = arg z. The numbers (r, θ) are just the usual polar
coordinates of the point (x, y).
Now if z1 = r1 (cos θ1 + i sin θ1 ) and z2 = r2 (cos θ2 + i sin θ2 ), then we
calculate that
z1 z2 = r1 r2 ((cos θ1 cos θ2 − sin θ1 sin θ2 ) + (sin θ1 cos θ2 + cos θ1 sin θ2 )i)
= r1 r2 (cos(θ1 + θ2 ) + i sin(θ1 + θ2 ))
(1.1)
Here we have used the angle addition formulas from trigonometry; a little
later, using the exponential function, we will see an alternate proof of (1.1)
which does not use (and instead implies) the angle addition formulas. In any
case, the conclusion is that to multiply two nonzero complex numbers, you
multiply their absolute values and add their arguments. Note in particular
that
|z1 z2 | = |z1 ||z2 |.
2
(This also can be deduced without using trigonometry from (1.2) and (1.3)
below).
We have one more definition to make. To motivate it, recall that the
idea of the definition of C is to define i to be a square root of −1. One could
then ask: “Which square root? Doesn’t −1 have two square roots?” The
point is that there is a symmetry which exchanges i with −i. Namely, if
z = x + iy ∈ C, define the complex conjugate of z by
z = x − iy.
z1 + z2 = z1 + z2 ,
(1.2)
z 1 z2 = z1 z2 .
an z n + an−1 z n−1 + · · · + a0 = 0
3
We will prove this theorem later. Note that in fact there are exactly n
solutions (some of which may be repeated). The reason is that if z1 is a
solution, and if we write p(z) = an z n + · · · + a0 ∈ C[z], then we can divide
the polynomial p by z − z1 to obtain
p(z) = an (z − z1 ) · · · (z − zn )
z 2 = a.
x2 − y 2 = u,
2xy = v.
4
Let n be a positive integer and consider the equation
z n = 1.
To solve this, by taking absolute values we find that |z| = 1, so we can write
z = cos θ + i sin θ.
Then the equation to solve is
cos(nθ) + i sin(nθ) = 1.
Equating the real parts of both sides, we find that cos(nθ) = 1, so nθ is a
multiple of 2π. The solutions to this are
2π 4π 2(n − 1)π
θ = 0, , ,..., .
n n n
The resulting numbers z are evenly spaced around the unit circle; they are
called nth roots of unity. For example, when n = 3, we find that the three
cube roots of unity are √
−1 3
1, ± i.
2 2
One can also obtain these by factoring z 3 − 1 = (z − 1)(z 2 + z + 1); one can
then find the two nontrivial cube roots by solving z 2 + z + 1 = 0 using the
quadratic formula.
A similar argument shows that any nonzero complex number z has n
different nth roots, which are evenly spaced around the circle of radius |z|1/n
centered at the origin.
Now let us prove Proposition 1.3. Suppose there exists a continuous
function f : C → C such that f (z)2 = z for all z ∈ C. This is actually
already impossible if we restrict attention to the unit circle. To see this,
given θ ∈ [0, 2π), note that we can write
f (cos θ + i sin θ) = cos(ψ(θ)) + i sin(ψ(θ)) (1.5)
for a unique ψ(θ) ∈ {θ/2, θ/2 + π}. Since f is continuous, it follows that ψ :
[0, 2π) → [0, 2π) is continuous. Then θ 7→ ψ(θ)−θ/2 is a continuous function
which takes values in {0, π} and hence must be constant. This constant must
equal ψ(0), so we have ψ(θ) = ψ(0) + θ/2. Thus limθ→2π ψ(θ) = ψ(0) + π.
By (1.5) we get
lim f (cos θ + i sin θ) = −f (1).
θ→2π
Since f is continuous, we obtain f (1) = −f (1), which is impossible since
f (1) 6= 0.
It turns out that there is a way to change the setup so that a continuous
square root function exists, by “thinking outside of the box”. We will discuss
this later when we talk about Riemann surfaces.
5
1.3 C as a metric space
Recall that if (x1 , y1 ) and (x2 , y2 ) are two points in R2 , the Euclidean distance
between them is defined by
p
d((x1 , y1 ), (x2 , y2 )) := (x1 − x2 )2 + (y1 − y2 )2 .
d(z1 , z2 ) = |z1 − z2 |.
This distance function makes C into a metric space (review what this means
if necessary). In particular, we have the triangle inequality
Note that equality holds in this last step if and only if zw is real and non-
negative. Thus equality holds in the triangle inequality (1.6) if and only if
one of z, w is a positive real multiple of the other . (Recall that the trian-
gle inequality (1.6) holds more generally in any inner product space; in the
proof, Re(zw) in (1.7) and (1.8) is replaced by the inner product of z and
w, and (1.8) is replaced by the Cauchy-Schwarz inequality.)
Since C is a metric space, some basic definitions from real analysis carry
over directly to this setting. Let us state these precisely.
6
• If (z1 , z2 , . . .) is a sequence of complex numbers and L ∈ C, we say that
limn→∞ zn = L if for all ε > 0 there exists an integer N such that if
n > N then |zn − L| < ε.
B(a, r) = {z ∈ C | |z − a| ≤ r}.
7
Consider the unit sphere
S 2 := {(x, y, z) ∈ R3 | x2 + y 2 + z 2 = 1}.
φ : S 2 \ {n} → R2
φb : S 2 → C
b
by defining φ(p)
b = φ(p) for p ∈ S 2 \ {n} and φ(n)
b := ∞. The bijection φb
has the following nice property: Let f : C → C. Then composing with φb
determines a function g : S 2 \ {n} → C. And we have limz→∞ f (z) = L
if and only if limp→n g(p) = L, where the latter limit is defined using the
Euclidean distance on S 2 .
A cool, but not completely trivial fact, is that if C is a circle in S 2 (i.e.
the intersection of S 2 with a plane that is not tangent to it), then φ(C)
b is
either a circle in C (if n ∈
/ C) or the union of a line in C with ∞ (if n ∈ C).
Conversely, every line or circle in C is obtained in this way. We will talk
more later about maps that respect lines and circles.
2 Complex functions
2.1 Recollections from real analysis
Let U be an open subset of R and let f : U → R. Recall that if a ∈ U , then
the derivative of f at a, denoted by f 0 (a), is defined by
f (x) − f (a)
f 0 (a) = lim ,
x→a x−a
8
provided that this limit exists; otherwise f 0 (a) is undefined. If f 0 (a) is defined
then f is continuous at a, because
f (x) − f (a)
lim (f (x) − f (a)) = lim (x − a)
x→a x→a x−a
f (x) − f (a)
= lim (x − a) lim
x→a x→a x−a
= 0 · f 0 (a)
= 0.
with the right hand side absolutely convergent. (We will prove this later in
the complex analysis setting, and the same proof works here.)
9
Moreover, each of the above four conditions is strictly stronger than the
previous one. For example, there exists a function which is differentiable but
not continuously differentiable: define f : R → R by
2
x sin(1/x), x 6= 0,
f (x) :=
0, x = 0.
whenever |x| < r. Now the coefficients a0 , a1 , . . . cannot all equal zero, or
else f (x) would equal zero for all x ∈ (−r, r), a contradiction. So let n be
the smallest nonnegative integer such that an 6= 0. It then follows from (2.2)
that
f (n) (0) = n!an .
This contradicts our previous observation that f (n) (0) = 0.
10
Incidentally a similar argument shows that if U is a connected open set,
if f : U → R is any real analytic function, and if f vanishes on some open
subset of U , then f vanishes on all of U . This is called “unique continua-
tion”, because it implies that if f and g are two real analytic functions on
a connected open set U , and if they agree on some open subset of U , then
they are equal on all of U . That is, when a real analytic function can be
extended to larger domain, the extension is unique. The complex analysis
version of this will be important later.
where ux denotes ∂u/∂x and so forth. This follows directly from the defini-
tions. For example, any polynomial in the coordinates x, y is differentiable
on all of R2 .
A less obvious fact is that if u and v have continuous partial derivatives
near a, then f is differentiable at a. (This is less obvious, and might not
hold if the partial derivatives are defined but not continuous, because the
partial derivatives just describe how u and v change as one moves in the x
direction or the y direction, while the definition of differentiability describes
how u and v change as one moves in all directions.)
In any case, the above definition of differentiability is not what we want
for complex analysis. Instead we want the following:
Definition 2.2. The (complex) derivative of f at a is defined by
f (z) − f (a)
f 0 (a) = lim ∈C
z→a z−a
11
provided that this limit exists, in which case f is (complex) differentiable,
or holomorphic, at a; otherwise f 0 (a) is undefined. We inductively define
f (n) (a) as before.
ux = vy , uy = −vx . (2.3)
At such a point,
uy + ivy
f 0 = ux + ivx = . (2.4)
i
Proof. (⇒) Suppose f is complex differentiable at a. Then by definition,
We can also write the Cauchy-Riemann equations (2.3) a bit more suc-
cintly as
∂f ∂f
=i . (2.5)
∂y ∂x
12
When these equations are satisfied, we can write (2.4) more simply as
∂f 1 ∂f
f0 = = . (2.6)
∂x i ∂y
From now on, unless otherwise stated, we always consider differentiation
in the complex sense of Definition 2.2. To see how restrictive this is, note
for example:
Proposition 2.4. Let U ⊂ C be a connected open set and suppose f : U → R
is (complex) differentiable. Then f is constant.
Proof. Since f is real, it follows from the Cauchy-Riemann equations (2.3) or
(2.5) that ∂f /∂x = ∂f /∂y ≡ 0. One can then use line integrals to conclude
that f is constant (see also Corollary 2.8).
The chain rule for real functions carries over to the complex case:
Proposition 2.5. Let f and g be complex functions defined on open subsets
U, V ⊂ C, let a ∈ U , and suppose f (a) ∈ V . If f 0 (a) and g 0 (f (a)) are
defined, then g ◦ f is differentiable at a and
(g ◦ f )0 (a) = g 0 (f (a))f 0 (a).
Proof. The proof from the real case carries over directly, but let us do this
anyway as a review.
First note that since f is continuous at a, it follows that g ◦ f is defined
in a neighborhood of a, so that it makes sense to consider (g ◦ f )0 (a). Now
the statement that f is differentiable at a with derivative f 0 (a) is equivalent
to the statement that if a + h ∈ U then
f (a + h) = f (a) + f 0 (a)h + ϕ(h)
where limh→0 ϕ(h)/h = 0. Likewise, since g is differentiable at f (a), if
f (a) + j ∈ V then
g(f (a) + j) = g(f (a)) + g 0 (f (a))j + ψ(j)
where limj→0 ψ(j)/j = 0. Now if a + h is in the domain of g ◦ f , let j =
f 0 (a)h + ϕ(h). Then by the above two equations, we have
g(f (a + h)) = g(f (a)) + g 0 (f (a))f 0 (a)h + g 0 (f (a))ϕ(h) + ψ(f 0 (a)h + ϕ(h)) .
It follows from basic properties of limits (review these if necessary) that the
limit as h → 0 of 1/h times the expression in square brackets is 0. Thus we
have
(g ◦ f )(a + h) = (g ◦ f )(a) + g 0 (f (a))f 0 (a)h + ξ(h)
where limh→0 ξ(h)/h = 0. This is exactly what it means for g ◦ f to be
differentiable at a with derivative g 0 (f (a))f 0 (a).
13
There is also another chain rule, for the derivative of a complex derivative
function along a path. To prepare for the statement, if [a, b] is an interval in R
and f : [a, b] → C is a complex-valued function, then we define its derivative
(when defined) exactly as with real-valued functions f : [a, b] → R, i.e.
d f (t + h) − f (t)
f (t) = lim ∈C
dt h→0 h
where the limit is now over real numbers h (and if t = a or t = b we only
take the right or left limit). Equivalently, if f (t) = u(t) + iv(t), then
df du dv
= +i .
dt dt dt
Proposition 2.6. Let f : U → C be a (complex) differentiable function, and
let γ : [a, b] → C be a (real) differentiable path. Then
d dγ(t)
(f (γ(t))) = f 0 (γ(t)) .
dt dt
Proof. Write γ(t) = u(t) + iv(t). By the usual chain rule from multivariable
calculus (using complex-valued functions instead of real-valued functions,
which makes no difference, e.g. because you can apply the real chain rule
separately to the real and imaginary parts of f ),
d ∂f ∂f
(f (γ(t))) = u0 (t) (γ(t)) + v 0 (t) (γ(t))
dt ∂x ∂y
∂f
= (u0 (t) + iv 0 (t)) (γ(t))
∂x
dγ(t) 0
= f (γ(t)).
dt
Here in the second and third lines we have used equations (2.5) and (2.6)
respectively.
Remark 2.7. The two versions of the chain rule in Propositions 2.5 and
2.6 are special cases of the real chain rule in multiple variables: If f : Rn →
Rm and g : Rm → Rl are differentiable, then g ◦ f is differentiable and
d(g ◦ f )a = dgf (a) ◦ dfa , i.e. the Jacobian of the composition is the product
of the Jacobians. This is proved the same ways as Proposition 2.5. To
deduce Proposition 2.5 or 2.6, one just has to interpret the corresponding
matrix multiplication as multiplication of complex numbers in the complex
differentiable case.
We can deduce from Proposition 2.6 the following important fact:
Corollary 2.8. Suppose U ⊂ C is open and connected (review the definition)
and f : U → C is differentiable and satisfies f 0 ≡ 0. Then f is constant.
14
Proof. Since f 0 ≡ 0, it follows from the Cauchy-Riemann equations that
∂f /∂x ≡ ∂f /∂y ≡ 0. One can then use line integrals (which we will review
later) to deduce that f is constant.
To give another proof without using line integrals, let z, w, ∈ U . Since
U is open and connected, there is a differentiable path γ : [a, b] → U with
γ(a) = z and γ(b) = w (review why). By Proposition 2.6, the function
f ◦ γ : [a, b] → C has derivative 0 on all of [a, b]. We know from the mean
value theorem that a real-valued function on an interval with derivative 0 is
constant. Applying this theorem separately to the real and imaginary parts
of f ◦γ shows that f ◦γ is constant. Hence f (γ(a)) = f (γ(b)), or equivalently
f (z) = f (w). Since z, w ∈ U were arbitrary, this proves that f is constant
on U .
Note that if U is not connected, then one can only conclude that f is
constant on each connected component of U .
15
The number r is called the radius of convergence of the series (2.7).
If |z − a| = r, the series may or may not converge. (Examples...)
Definition 2.10. Let U be an open set in C and let f : U → C. The
function f is (complex) analytic if for every a ∈ U , there exist r > 0 and
a0 , a1 , . . . ∈ C such that the power series expansion (2.7) is valid (with the
right side absolutely convergent) whenever |z − a| < r.
Exercise 2.11. If f is defined by (2.7) on B(a, r), where r is less than or
equal to the radius of convergence, then f is analytic on B(a, r) (i.e. f can
be expanded as a power series around every point in B(a, r), not just a).
Proposition 2.12. If f : U → C is analytic, then f is infinitely differ-
entiable. Moreover, the coefficients of the power series expansion (2.7) are
given by
f (n) (a)
an = .
n!
Proof. This follows from the following lemma and induction.
Lemma 2.13. Define f : B(a, r) → C by the power series (2.7) and assume
that the radius of convergence is at least r > 0. Then f is differentiable on
B(a, r), and its derivative is given by differentiating term by term:
∞
X
0
f (z) = (n + 1)an+1 (z − a)n . (2.8)
n=0
expand the expression whose limit we are taking using the binomial theorem,
and (using absolute convergence to justify reordering the sums) write the
result as the right hand side of (2.8) plus h times something, which via some
estimates can be shown to converge to 0. This is a good exercise and does
not take too long. However it is not very satisfying because it leaves us
without any general understanding of when one can differentiate an infinite
sum of functions term by term. So here is a more systematic approach, using
some basic facts about line integrals which we will review later.
Lemma 2.14. Let fk : B(a, r) → C be functions for k = 0, 1, . . . such that
fk converges uniformly on closed balls to f : B(a, r) → C, and fk0 converges
uniformly on closed balls to g : B(a, r) → C. Then f is differentiable and
f 0 = g.
16
Proof. Since fk0 is the derivative of a function for each k, it follows that g
is also the derivative of a function, call it h. (This is because there is a
criterion in terms of line integrals for when a function is a derivative, and
this criterion is preserved under uniform limits of functions.) By adding a
constant we may assume that h(a) = f (a). To complete the proof we just
need to check that h(z) = f (z) for all z ∈ B(a, r).
To so, let z ∈ B(a, r) be given. It is enough to show that |f (z)−h(z)| < ε
for all ε > 0. Given ε > 0, choose k sufficiently large so that |fk (z) − f (z)| <
ε/2 and |fk0 (w) − g(w)| < ε/2r for all w on the line segment from a to w.
Then it follows (using line integrals again) that |fk (z) − h(z)| < ε/2. So by
the triangle inequality, |f (z) − h(z)| < ε as desired.
Contrast this with real analysis, where a differentiable function need not
even have a continuous derivative! We will prove Theorem 2.15 later using
complex integration.
17
2.4 The exponential function, sine and cosine
xn
Recall that for x real one has ex = ∞
P
n=0 n! . We would like to extend the
definition to allow taking the exponential of a complex number. It is natural
to simply use the same power series and for z ∈ C define
∞
z
X zn
e = . (2.9)
n!
n=0
This is sometimes also denoted by exp(z). This power series has radius of
convergence ∞ (why?), so by Exercise 2.11 it defines an analytic function
on all of C. Of course it agrees with the usual exponential function when
restricted to the real line.
It follows immediately from (2.9) that
e0 = 1. (2.10)
Also, Lemma 2.13 implies that
(ez )0 = ez . (2.11)
In fact, the general theory of ODE’s implies that the function ez satisfying
(2.10) and (2.11) is unique2 , so one could take this as an alternate definition
of ez . (To give a more elementary proof of the uniqueness, if f and g are two
functions satisfying f 0 = f and g 0 = g, and f (0) = g(0) = 1, then applying
Corollary 2.8 to h(z) = f (z)g(−z) shows that f (z)g(−z) is a constant, so
f (z)g(−z) = 1 for all z. The same argument shows that g(z)g(−z) = 1, so
f (z) = g(z).)
As in the real case, we have the law of exponents
ez+w = ez ew . (2.12)
One way to prove this is to expand the left hand side using the binomial
theorem, and rearrange the sum to obtain the right hand side (using absolute
convergence to justify reordering the sums). But one can also prove this
without doing that calculation as follows. Fix w ∈ C and define f : C → C
by f (z) = ez+w . Also define g : C → C by g(z) = ez ew . Observe that
f (0) = ew , f 0 (z) = f (z),
where the second equation uses the chain rule (Proposition 2.5). On the
other hand we also have
g(0) = ew , g 0 (z) = g(z).
2
The relevant theorem about ODE’s asserts that if f : Rn → Rn is continuous (we are
interested in the case n = 2), and if p ∈ Rn , then there exists ε > 0 and a differentiable
path γ : (−ε, ε) → Rn such that γ 0 (t) = f (γ(t)) and γ(0) = p. Moreover, for any ε > 0, if
a path γ as above exists then it is unique.
18
Since f and g both satisfy this differential equation, they must be equal,
for the same reason that the exponential function is the unique function
satisfying (2.10) and (2.11).
If z = x + iy with x, y real, then by (2.12) we have
ez = ex eiy .
Equation (2.13) has two useful corollaries. First, from (2.9) and (2.13)
we obtain the familiar power series expansions
θ2 θ4
cos θ = 1 − + − ··· ,
2! 4! (2.15)
θ3 θ5
sin θ = θ − + − ··· .
3! 5!
19
Second, if α and β are any real numbers then by the law of exponents (2.12)
we have
ei(α+β) = eiα eiβ .
If we expand both sides using (2.13) and take the real and imaginary parts
of both sides, then we obtain the angle addition formulas
cos(α + β) = cos α cos β − sin α sin β,
(2.16)
sin(α + β) = cos α sin β + sin α cos β.
In fact, one can take (2.15) as a definition of sin θ and cos θ also for complex
θ (although if θ is not real then sin θ and cos θ are usually not real either).
Equivalently,
eiz + e−iz eiz − e−iz
cos z = , sin z = .
2 2i
With this definition, a short calculation shows that the angle addition for-
mulas (2.16) also hold when α and β are complex.
2.5 Logarithms
One would like to define a function log which is the inverse of ez . Note that
the domain of log cannot include 0 because the exponential function never
vanishes (because ez e−z = 1). However the exponential function surjects
onto C \ {0}. This is because any complex number z can be written as
r(cos θ + i sin θ) with r > 0 and θ real, and then
z = elog r+iθ
where log r for r a positive real number is understood to be the usual natural
logarithm. However since θ is only defined modulo 2π, there are infinitely
many complex numbers w such that ew = z, which differ by integer multiples
of 2π. If we want to define log z, we need to pick one such w. Unfortunately,
there is no way to do so continuously for all nonzero z. The proof is similar
to that of Proposition 1.3 and we leave it as an exercise.
However we can define a continuous log function if we remove enough
points from the domain. In particular the principal branch of the loga-
rithm is a well-defined function
log : C \ {x ∈ R | x ≤ 0} −→ C.
To define this, given z which is not a nonnegative real number, we can
uniquely write z = r(cos θ + i sin θ) with r > 0 and θ ∈ (−π, π), and we
define log z = log r + iθ. This function is differentiable and satisfies
1
(log z)0 = . (2.17)
z
20
To see why, observe that the restriction of the exponential function defines
a bijection
'
exp : {z ∈ C | Im(z) ∈ (−π, π)} −→ C \ {x ∈ R | x ≤ 0}.
The principal branch of the logarithm is the inverse of this bijection. Now
the inverse function theorem implies that if U and V are open subsets of C
and if f : U → V is a holomorphic bijection with f 0 never vanishing3 , then
f −1 : V → U is differentiable, and then by the chain rule we must have
(f −1 )0 (f (z)) = 1/f 0 (z). This implies (2.17).
21
So we have no choice but to define
Z x Z y
v(x, y) = v(0, 0) − uy (t, 0)dt + ux (x, t)dt,
0 0
which is unique up to the choice of the additive constant v(0, 0). To complete
the proof, we have to check that this definition of v satisfies vx = −uy for
all (x, y) ∈ U , not just when y = 0. To do so, we can apply the fundamental
theorem of calculus to the first integral, and differentiate under the integral
sign (we will justify this a bit later) in the second integral, to get
Z y
vx (x, y) = −uy (x, 0) + uxx (x, t)dt
0
Z y
= −uy (x, 0) − uyy (x, t)dt
0
= −uy (x, 0) − (uy (x, y) − uy (x, 0))
= −uy (x, y).
(Where did we use the assumption that U is a disk? For which other U will
this work? We will see a more systematic approach to this later.)
22