MA30056: Complex Analysis: Lecture Notes
MA30056: Complex Analysis: Lecture Notes
MA30056: Complex Analysis: Lecture Notes
Lecture Notes∗
∗ by Bernd Sing, but (in large parts!) based on lecture notes by Udo Hertrich-Jeromin
Manifesto
Our Mission
Study functions f : C ⊃ D → C which are differentiable, i.e., which have, for all z ∈ D, a
derivative
f (z + h) − f (z)
f 0 (z) = lim .
h→0 h
This does not look very different from M11, does it?
WRONG!
Consequences
(i) If f is differentiable in the above sense, then it is C ∞ ! Indeed, it is even C ω , meaning
that it has a power series expansion.
Rx
Not so in R: take f (x) = 0 |ξ| dξ; then f 0 (x) = |x| is not differentiable at x = 0.
(ii) A differentiable f (in the above sense) is uniquely determined inside a circle by its
values on the circle.
Completely false in R: a differentiable function is certainly not determined by its
values on the endpoints of an interval (the “circle” in R).
(iii) If f : C → C is differentiable and |f | is bounded then f is constant.
False in R: consider, for example, sin : R → R.
Applications
(i) An easy proof of the Fundamental Theorem of Algebra: Any nonconstant polynomial
(with complex coefficients) has a root in C.
(ii) We shall learn quick ways to evaluate tricky series/integrals, like
∞ ∞
π2
Z
1 π X 1
4
dx = √ or = ,
0 1+x 2 n=1
n2 6
(iii) In M9, differential equations are solved using the (one-sided) Laplace transform
Z ∞
ˆ
f (s) = L{f }(s) = f (t) e−s·t dt,
0
where f : [0, ∞) → R and s is a complex variable. It was noted there that “not
all functions possess Laplace transforms; furthermore, the Laplace transform of a
2
function may generally only be defined for certain values of the variable s ∈ C, e.g.,
for a function f of exponential order α, the Laplace transform fˆ(s) exists for all
s with Re(s) > α.” These statements are actually statements about the complex
differentiability of fˆ(s)! Moreover, with the knowledge of complex analysis, one does
not have to “guess” the inverse Laplace transform (as in M9), but one can use (and
make meaning of) the formula
Z c+iy
1
f (t) = L {fˆ}(t) =
−1
lim fˆ(s) es·t ds
2πi y→∞ c−iy
where the path of integration is along the vertical line ω → s = c + i ω (and c can be
any real constant greater than α).
(iv) We can use the theory to study PDE’s, like 4
-2
-4
-4 -2 0 2 4
It also lies the foundation stone for many other applications, we mention the following two:
(v) Combinatorics: We enumerate binary trees with n = 0, 1, 2, 3, . . . binary nodes (more
precisely, we enumerate ordered rooted binary trees):
n=0: ◦ c0 = 1
n=1: •// c1 = 1
◦ ◦
n=2: •/ •/ c2 = 2
/ /
•/ ◦ ◦ •/
/ /
◦ ◦ ◦ ◦
3
n=3: •/ •/ •?
/ / ?
•/ ◦ •/ ◦ •/ •/
/ / / /
•/ ◦ ◦ •// ◦ ◦ ◦ ◦
/
◦ ◦ ◦ ◦
•/ •/ c3 = 5
/ /
◦ •/ ◦ •//
/
◦ •/ •// ◦
/
◦ ◦ ◦ ◦
The number cn (of binary trees with n nodes) is the nth Catalan number. One can
show that √
X 1 − 1 − 4z
cn · z n = .
n≥0
2z
Thus, a simple series expansion of the function on the right yields the Catalan num-
bers:
√
1 − 1 − 4z
= 1 + z + 2 z 2 + 5 z 3 + 14 z 4 + 42 z 5 + 132 z 6 + 429 z 7 + 1430 z 8 + . . .
2z
Question: How fast does cn grow as n → ∞?
√
1− 1−4z
This can be deduced from the (complex) function f (z) = 2z
using the following
“principles of coefficient asymptotics”:
• The location of a function’s singularities (i.e., where f is not differentiable)
dictates the exponential growth An of the coeffictions cn .
• The nature of a function’s singularities determines the associated subexponential
factor Θ(n).
The function f in question here, is not differentiable at z = 14 (already as real function
−n
we have limx→ 1 − f 0 (x) = +∞) – this yields the exponential growth 14 = 4n –
4
and this singularity is a so-called “square-root singularity” (see the graph of this
function), which yields a subexponential factor Θ(n) = n−3/2 . Thus, asymptotically
one has1
1 4n
cn ∼ √ · 4n · n−3/2 = √ .
π π · n3
√
1− 1−4x
And here the graphs of the real function f (x) = 2x
(left) and of |f (z)| over the
1 g(n)
Here, g(n) ∼ h(n) means that h(n) → 1 as n → ∞.
4
complex plane (right):
1
0
-1
2.0
2.5
1.5
2.0
1.0
1.5
0.5
1.0
0.0
0.5
1
0
-1.0 -0.5 0.0 0.5 1.0 -1
(vi) Number Theory: Riemann’s Zeta Function ζ(s) is defined for Re(s) > 1 by
X 1
ζ(s) = .
n∈N
ns
Since any natural number has a unique prime factorization, one easily can establish
Euler’s product formula
X 1 Y 1
ζ(s) = = ,
n∈N
n s
p∈P
1 − p−s
where P denotes the set of primes. This yields an intriguing connection between
Riemann’s Zeta Function and the primes.
Now, there is a problem: The above infinite sum (and, similarly, the infinite product)
does only converge (absolutely) for Re(s) > 1. However, there exists exactly one
complex differentiable function (again called) ζ(s) defined on C \ {1} that coincides
with the above function on the half-plane Re(s) > 1.
What does this all have to do with the primes? The statement that ζ(s) has a
singularity (or, more precisely, “a pole of order one”) at s = 1 implies the statement
that there are infinitely many primes. And then there are the zeros of ζ(s) (or
singularities of 1/ζ(s)): While there are none in the half-plane Re(s) > 1, there
are the so-called “trivial zeros” at s = −2, −4, −6, −8, −10, . . ., while the remaining
(“non-trivial” and therefore interesting) zeros lie in the strip 0 ≤ Re(s) ≤ 1. Their
location is symmetrical with respect to the critical line 21 + it. Furthermore, we note:
• Essentially by showing that there are no zeros on the line 1 + it, Hadamard
proved the Prime Number Theorem in the year 1896 (de la Vallée Poussin gave
independently also a proof in the same year): “The number of primes less than
or equal to x is approximately x/ log(x) where the relative error of this approx-
imation approaches 0 as x → ∞.”
• The Riemann Hypothesis states that all non-trivial zeros of ζ(s) lie on the critical
strip! The Riemann Hypothesis is mentioned in the 8th (of the 23) Hilbert’s
problem and one of the Clay Mathematics Institute’s Millenium Prize Problems
(and thus its proof (not a counterexample!) is worth at least $1 million).
5
• The first 1013 (and at least 40% of the) non-trivial zeros lie on the critical line –
needless to say that no counterexample has been found so far. The correctness of
the Riemann Hypothesis, for example, has implications for the error term in the
Prime Number Theorem, the distribution of primes would be “quite regular”.
3.0
2.5
2.0
1.5
1.0
0.5
-40 -20 20 40
Plot of |ζ( 12 + it)|, i.e., of the modulus of Riemann’s Zeta Function along the critical line.
6
I. The Complex Number Plane
I.1. Algebra
References: [DET, Section 1.2] and [ST, Chapter 1]
There are various ways to think about C:
√
(i) C = {x + iy | x, y ∈ R} as a field extension of R, with an “imaginary” unit i = −1
(a solution of x2 + 1 = 0).
(ii) C = {(x, y) ∈ R2 } as vector space with multiplication
z = x + iy = x − iy.
7
Real and imaginary parts. If z = x + iy ∈ C these are defined as
Re z = x and Im z = y.
z 1 z
Note that, if z 6= 0, then 1 = z · |z|2
, that is, z has an inverse z
= |z|2
.
ex Let z, w ∈ C. Verify that
(i) | Re z|, | Im z| ≤ |z|,
(ii) |z| = |z|, i.e., C 3 z 7→ z ∈ C is an isometry,
(iii) |z · w| = |z| · |w|, i.e., | . | : (C, ·) → (R, ·) is a homomorphism,
(iv) |z| ≥ 0 and |z| = 0 iff z = 0.
Conclude that the triangle inequality |z + w| ≤ |z| + |w| holds.
Hint: compute |z + w|2 .
ex Prove that | . | : C → R is continuous.
ex Show that C is not(!) an ordered field.
Note that an ordering of a field K is a subset P ⊂ K having the following properties:
(O1) Given x ∈ K, we have either x ∈ P , or x = 0 or −x ∈ P , and these three possibilities
are mutually exclusive. In other words, K is the disjoint union of P , {0} and −P .
(O2) If x, y ∈ P , then x + y ∈ P and x · y ∈ P .
We shall also say that K is ordered by P , and we call P the set of positive elements.
Hint: Proof by contradiction.
I.2. Geometry
References: [DET, Section 1.3] and [ST, Chapter 1]
When we think of C ∼
= R2 with a multiplication then:
(i) addition is vector addition,
(ii) conjugation is reflection in the real axis Im z = 0,
(iii) the modulus is the Euclidean length of the vector z,
(iv) multiplication of z by a real number w, Im w = 0, is scalar multiplication.
Question: How to think about multiplication by a complex number?
8
Polar form of a complex number. This is given by
then
z · w = rs (cos(ϑ + ϕ) + i sin(ϑ + ϕ)).
I.3. Topology
References: [DET, Section 1.4] and [ST, Section 2.1]
(Bolzano-Weierstrass: [DET, Theorem 1.4.1]; Heine-Borel: [DET, Theorem 1.4.3])
Overall theme: C is homeomorphic (even isometric) to R2 (both have the “same” topology
(i.e., open/closed sets), (Cauchy/convergent) sequences etc.).
Notation:
• open disk/ball Br (z) = {w ∈ C | |w − z| < r},
• closed disk/ball B r (z) = {w ∈ C | |w − z| ≤ r}.
If z = 0, it is sometimes dropped, i.e., we write Br and B r instead of Br (0) respectively
B r (0)
Recall. A subset A ⊂ C (∼
= R2 ) is called
• open if ∀z ∈ A ∃ε > 0 : Bε (z) ⊂ A (or if A = ∅),
• closed if C \ A is open.
ex Show that open/closed discs are open/closed.
9
Convergence. (zn )n∈N converges to z ∈ C if |zn − z| → 0 as n → ∞.
ex Show that zn → z iff ∀ε > 0 ∃N ∈ N s.t. ∀n ≥ N : zn ∈ Bε (z).
ex Prove that M ⊂ A ⊂ C is closed in A iff it contains the limit of every sequence
(zn )n∈N ⊂ M that converges in A.
Remark. If A = C the previous exercise yields the usual sequential characterization of
closed subsets M ⊂ C.
Lemma I.3.1. If zn = xn + iyn and z = x + iy then
zn → z ⇔ xn → x and yn → y.
ex Prove Lemma I.3.1
Hint: Use |xn − x|, |yn − y| ≤ |zn − z|.
10
I.4. Stereographic Projection (Not examinable!)
Reference: [DET, Section 1.5] and [ST, Section 11.4]
A complex number z = x + iy ∈ C can be represented as point (x, y) in the plane R2 . One
can also associate a point (u, v, w) on the unit sphere S = {(u, v, w) ∈ R3 |u2 +v 2 +w2 = 1},
called the Riemann sphere in this context, with a given point (x, y) in the plane1 . The
associated mapping is called stereographic projection.
• north pole
◦???(u, v, w)
??
??
??
??
??
??
??
• (x, y)
We note:
• A point (u, v, w) on the sphere corresponds to (x, y) if the north pole (0, 0, 1), (u, v, w)
and (x, y, 0) are on a line.
• The equator of the sphere corresponds to the unit circle in the plane.
• The south pole (0, 0, −1) corresponds to the origin (0, 0).
1
We embed the plane into R3 by (x, y) 7→ (x, y, 0).
11
Definition. C b = C ∪ {∞} is called the extended complex plane, where ∞ denotes the point
at infinity (its image point on the Riemann sphere is the north pole).
Remark. • A sequence (zn )n∈N is unbounded in C iff there exists a subsequence
(znk )k∈N s.t. znk → ∞ in C.
b
1.0
Remark. Path connectedness implies connectedness in the
topological sense (A ⊂ C is connected if it is not the union 0.5
Theorem I.5.1. Let A ⊂ C be open. Then A is connected iff it is path connected (iff it is
polygonally/step connected2 ).
Proof. Let A be connected and let z0 ∈ A and define A1 = {z ∈ A|∃ a path joining z and z0 }
and A2 = A \ A1 (A1 is the connected component of the point z0 ). Then, one can show
that A1 , A2 are open
ex , therefore they are open and closed in A. They are also disjoint
and their union is all of A. So, since A is connected and A1 is nonempty, A2 must be the
empty set. Thus, A = A1 and A is path connected.
2
A set A is polygonally connected if for any two points z, z 0 ∈ A there is a polygonal line joining them
which is contained in A. If, furthermore, each straight line segment in this polygonal line is parallel to
either the real or the imaginary axis, then the set is called step connected, see [ST, Section 2.6].
12
The last part in the previous proof is actually an alternative characterisation of connect-
edness: Let A ⊂ C be connected and A1 ⊂ A open and closed in A. Then A1 = A or
A1 = ∅. A similar statement holds if we replace “connected” by “path connected”.
Lemma I.5.2. Let A ⊂ C be path connected and A1 ⊂ A open and closed in A. Then
A1 = A or A1 = ∅.
ex Prove Lemma I.5.2.
Hint: Proof by contradiction.
Restriction of paths. If γ : [a, b] → C is a path and [c, d] ⊂ [a, b] then γ|[c,d] : [c, d] → C
is clearly a path.
Composition of paths. If γ1 : [a, b] → C and γ2 : [c, d] → C are paths with γ1 (b) = γ2 (c),
then (
γ1 (t) if t ∈ [a, b]
[a, b + d − c] 3 t 7→ γ(t) =
γ2 (t − b + c) if t ∈ [b, b + d − c]
is the path γ = γ1 + γ2 .
3
A region is a domain plus none/any/all of its boundary points.
4
If you know what “homotopic” means: D is simply connected iff every closed path in D is homotopic
to a “null path”/point curve (i.e., a single point) in D.
13
Inverse of a path. If γ : [a, b] → C is a path then the “reversed”/opposite path −γ is
ex Prove that the composition of two (piecewise smooth) paths is a (piecewise smooth)
path.
Length of a contour. If γ : [a, b] → C is a regular path then the length of Γ = γ([a, b])
Rb
is given by Γ ds = a |γ 0 (t)| dt.
R
The length of a contour is the sum of the lengths of its regular pieces.
14
II. Functions of a Complex Variable
Example. f (z) = z 2 maps the vertical and horizontal lines to parabolas (or rays, if c = 0):
2 2
-8 -6 -4 -2 2 -2 2 4 6 8
-2 -2
-4 -4
http://www.youtube.com/watch?v=JX3VmDgiFnY&NR=1
15
ex Show that
x2 y
C \ {0} 3 z = x + iy 7→ f (z) = ∈R⊂C
x4 + y 2
has no continuous extension to the whole plane.
Remark. It makes no sense any more to talk about a maximum of f (note that C is not
ordered ). So, the M11 proof has to be modified.
ex Let f : C ⊃ D → C be continuous and K ⊂ D compact. Show that f (K) ⊂ C is
compact and, in particular, bounded.
Proof. . . . as in M11
ex .
II.2. Differentiability
References: [DET, Section 2.2] and [ST, Section 4.1]
Here comes the main definition:
(iv) Similarly, f (z) = Re z, f (z) = Im z and f (z) = |z| are not differentiable anywhere in
C.
16
ex Show that C 3 z 7→ f (z) = |z|2 is differentiable at z = 0 only.
Proof. . . . as in M11
ex .
Proof. . . . as in M11
ex .
Remark. Using the above formula, we see that every polynomial is holomorphic on C,
and every rational function (the quotient of two polynomials) is holomorphic outside the
roots of its denominator.
Proof. . . . as in M11
ex .
Given the standard basis {e1 , . . . , en } of Rn , the limits (if they exist)
∂f f (p + tei ) − f (p)
(p) = lim
∂xi t→0 t
are the partial derivatives of f at p.
17
∂f
Remark. • ∃ dfp ⇒ ∃ ∂x i
(p): if f is differentiable at p, then all partial derivatives
exist and dfp is given by the Jacobi matrix (whose columns are the partial deriva-
tives).
∂f
• ∃ ∂xi
(p) 6⇒ ∃ dfp : the existence of the partial derivatives at p is not sufficient for f
to be differentiable at p (even though its Jacobi matrix exists); example
ex :
xy
R2 3 (x, y) 7→ p ∈ R.
x2 + y 2
∂f
• ∃ ∂xi
∈ C 0 (U ) ⇒ ∀p ∈ U ∃ dfp : if all partial derivatives exist and are continuous
then f is (continuously) differentiable.
∂f
• ∃ dfp 6⇒ ∃ ∂x i
∈ C 0 (U ): If f is differentiable, then the partial derivatives might be
discontinuous3 .
Informally, we see that real analysis is “complicated”!
Remark and Examples. We interpret the (real) differentiation geometrically:
• Let f : R → R. Then, for small x − x0 , we have4 f (x) ≈ f (x0 ) + f 0 (x0 ) · (x − x0 ).
Here, we can interpret f 0 (x0 ) either as a “local scale factor” (by which (x − x0 ) is
scaled), or we use the more familiar interpretation that f (x0 ) + f 0 (x0 ) · (x − x0 ) is
the “best possible linear approximation” of f (x) in x0 .
• Let f : R2 → R2 . Then we have f (x, y) ≈ f (x0 , y0 ) + df(x ,y ) (x − x0 , y − y0 )t .
0 0
Here, the Jacobi matrix df(x ,y ) ∈ M (2 × 2, R) wherefore the derivative is locally
0 0
an affine map, i.e., f (x0 , y0 ) + df(x ,y ) (x − x0 , y − y0 )t is the “best possible linear
0 0
approximation” of f (x, y) in (x0 , y0 ).
Now , consider a linear transformation (x, y) 7→ A(x, y)t given by some matrix A ∈ M (2 ×
2, R) has the following effect:
• The standard basis vectors e1 and e2 are mapped to the first and the second column
of A.
• A circle is mapped to an ellipse.
We would like to look at some examples of derivatives of functions R2 → R2 . These,
however, are 2 × 2-matrices! So, to get a geometric intuition what the derivative is, we
look at its effect on the basis vectors and a unit circle.
We consider the following three maps fk : R2 → R2 : f1 (x, y) = (x2 − xy, 2xy), f2 (x, y) =
(x2 − y 2 , 2xy) and f3 (x, y) = (x2 − xy, −2xy).
3
Consider
1 1
2 2
x sin x + y sin y
if x · y 6= 0,
x2 sin 1
if x 6= 0 and y = 0,
f (x, y) = x
y 2 sin y1
if x = 0 and y 6= 0,
0 if x = 0 = y.
Then, the partial derivatives fx , fy are discontinuous at the orgin, but f is differentiable everywhere
ex .
4
To line it up with the definition of differentiability at the beginning of this section, we have f (x) =
f (x0 ) + f 0 (x0 ) · (x − x0 ) + r(x − x0 ) with limh→0 r(h)
h = 0, so
18
We calculate the partial derivatives of f1 : ∂x f1 is given by
((x0 + t)2 − (x0 + t)y0 ) − (x20 − x0 y0 )
(2(x0 + t)y0 ) − (2x0 y0 )
∂f1 2x0 − y0
(x0 , y0 ) = lim = .
∂x t→0 t 2y0
∂f1
A similar calculation yields ∂y
(x0 , y0 ) and we obtain the following Jacobi matrix:
2x0 − y0 −x0
(df1 )(x0 ,y0 ) =
2y0 2x0
Note that both partial derivatives are continuous, thus by the previous remark this Jacobi
matrix is indeed the derivative of f1 .
yO
“Geometry” of df1 .
For f2 and f3 we calculate the following Jacobians:
2x0 −2y0 2x0 2y0
(df2 )(x0 ,y0 ) = and (df3 )(x0 ,y0 ) =
2y0 2x0 −2y0 2x0
yO yO
/x /x
19
For holomorphic functions, pictures like that for df1 (with ellipses and varying angles be-
tween the images of the basis vectors) or for df3 (where the “handiness” of the basis vectors
changes when considering their image) cannot occur (accidently, f2 (z) = z 2 is holomor-
phic), as the following statement shows which connectes real and complex diffeerentiation.
Theorem II.3.1 (Necessary Cauchy-Riemann conditions). If f : C ⊃ D → C is holo-
morphic then all first partial derivatives ux , uy , vx , vy of u = Re f and v = Im f exist and
satisfy the so-called Cauchy-Riemann equations
ux = v y and uy = −vx .
Moreover, f 0 = ux + ivx = vy − iuy .
f (z+h)−f (z)
ex Compute f 0 (z) = limh→0
Proof.
h
twice, for h = t ∈ R and for h = it purely
imaginary.
z 5
ex Show that (the continuous extension of) f (z) = |z| 4 is not (complex) differentiable at
z = 0 but satisfies the Cauchy-Riemann equations there. Can you find more functions with
this property?
ex (i) Define f : R → R by
(
x2 sin(1/x) if x 6= 0,
f (x) =
0 if x = 0.
Show: f is differentiable everywhere but f 0 is not continuous.
(ii) Define f : C → C by f (z) = z 2 sin(1/z) for z 6= 0, f (0) = 0. The solution to part (i)
almost seems to give a proof that f is (complex) differentiable everywhere but f 0 is
not continuous at the origin. This is impossible – where does the “proof” fail?
20
Corollary II.3.2. A holomorphic map f : D → C is conformal, i.e., preserves the inter-
section angle of curves in D.
to see that the intersection angle of the f ◦ γi is the same as the one of the γi .
Image of vertical (solid) and horizontal (dotted) Image of a few vertical (solid) and horizontal (dot-
lines under the map z 7→ z 2 . These parabolas in- ted) lines under the map f (z) = z 3 . A vertical line
tersect each other perpendicular. t 7→ x0 + i t intersects a horizontal line t 7→ t + i y0
perpendicularly at x0 + i y0 , and so do their images
at f (x0 + i y0 ). However, they might intersect at
other places arbitrarily.
ex Let f (x + iy) = u(x, y) + i v(x, y) be holomorphic in a domain D, and let (x0 , y0 ) ∈ D
where the gradient vectors of u and v do not vanish. Set u0 = u(x0 , y0 ) and v0 = v(x0 , y0 ).
Show that the level curves u(x, y) = u0 and v(x, y) = v0 intersect perpendicularly at
(x0 , y0 ).
Show a similar result if one uses f (x + iy) = R(x, y) eiΨ (x,y) .
In plain words, this says that level curves of constant real and imaginary part (or of constant
modulus and argument) are perpendicular to each other.
21
Some “real-world” level curves, the lines of the same height above sea-level, as seen on
http://maps.google.co.uk. We are looking at a region in the Lake District here.
For f (z) = z 2 , level curves of constant real (dot- For f (z) = z 2 , level curves of constant modulus
ted) and imaginary (solid) part intersect perpen- (dotted) and argument (solid) intersect perpendic-
dicularly. ularly.
For f (z) = z 3 , level curves of constant real (dot- For f (z) = z 3 , level curves of constant modulus
ted) and imaginary (solid) part intersect perpen- (dotted) and argument (solid) intersect perpendic-
dicularly. ularly.
22
Corollary II.3.3. Let f : C ⊃ D → C be holomorphic with f 0 ≡ 0. Then f is constant.
f (x1 + iy1 ) = u(x1 , y1 ) + i v(x1 , y1 ) = u(x1 , y0 ) + i v(x1 , y0 ) = f (x1 + iy0 ) = f (x0 + iy0 ).
23
Remark. Harmonic functions are studied in physics when considering fluid flows or elec-
trical fields. We will have a brief look at this later in this course (see Section III.4.3).
C-R 1
= lim |f (z + h) − f (z) − (ux h1 − vx h2 , vx h1 + ux h2 )|
h→0 |h|
f (z + h) − f (z) 6 h
= lim − (ux + ivx ) ,
h→0 h 6h
ux uy
where dfz = is the differential of f at z.
vx vy
Hence, f is (complex) differentiable at z, with f 0 (z) = ux (z) + ivx (z). This holds for every
z ∈ D, so f is holomorphic in D, as required.
II.4. Examples
Reference: [DET, Section 2.5] and [ST, Chapter 5]
We look at the following examples:
(i) The exponential function f : C → C, z 7→ ez = ex (cos y + i sin y) is holomorphic
ex with f 0 = f and f (z1 + z2 ) = f (z1 ) f (z2 ) (later, we will define the exponential
function via its series expansion). Note that it is periodic with period 2πi (since
cos, sin : R → R are 2π-periodic).
Note that ez = ex (cos y +i sin y) maps horizontal lines to rays from w = 0 and vertical
lines to concentric circles.
(ii) The (complex) trigonometric functions sin and cos are defined by
eiz + e−iz eiz − e−iz
cos z = and sin z =
2 2i
for all z ∈ C. They are holomorphic and have the expected properties (derivatives,
cos2 z + sin2 = 1, 2π-periodicity, etc.).
24
(iii) The (complex) hyperbolic functions sinh and cosh are defined by
ez + e−z ez − e−z
cosh z = = cos(iz) and sinh z = = −i sin(iz)
2 2
for all z ∈ C. They are also holomorphic.
(iv) Riemann’s Zeta Function7 is holomorphic on C \ {0} as remarked on p. 5.
We will later also look at the complex logarithm and arbitrary powers.
We want to get some geometric intuition of these functions. Two methods we have already
mentioned are to investigate the images/pre-images of “simple” curves (like vertical and
horizontal lines) under the function in ques-
tion, or the level curves of the function. If
we now would like to see their “graph” in
four-dimensional(!) space, we can refine the
level curve method as follows: We use a
colour palette to “visualise” two of the four
dimensions. We colour the origin 0 white
and colours get darker the greater the mod-
ulus of the complex number is (the point at
infinity ∞ is black). A colour corresponds8
to the argument of the complex number,
e.g., the positive real numbers are red-ish.
In the following, for a given function f , a
point z ∈ C is coloured according to its
value f (z) by the above palette, e.g., if we
have f (z) = 1 for a point z then that point
z is coloured red.
7
An article that uses level curves to study Riemann’s Zeta Function in the critical strip can be found at
http://arxiv.org/abs/math.NT/0309433
8
Compare this with the earlier discussed level sets/curves: the level curve of constant modulus corre-
sponds to the line of the same brightness/darkness, while the level curve of constant argument corre-
sponds to the line of constant colour (which might, however, become brighter or darker).
25
The function z → z 2 :
The image of the vertical (solid) and horizontal Vector field (Re z 2 , Im z 2 ) and colouring of the com-
(dotted) lines under f (z) = z 2 ; they intersect per- plex plane for z 7→ z 2 .
pendicularly.
Colouring and level curves of constant real (dot- Colouring and level curves of constant modulus
ted) and imaginary (solid) part for z 7→ z 2 ; they (dotted) and argument (solid) for z 7→ z 2 ; they
intersect perpendicularly. intersect perpendicularly.
26
The function z 7→ z 3 :
The image of the vertical (solid) and horizontal Vector field (Re z 3 , Im z 3 ) and colouring of the com-
(dotted) lines under f (z) = z 3 ; they intersect per- plex plane for z 7→ z 3 .
pendicularly.
Colouring and level curves of constant real (dot- Colouring and level curves of constant modulus
ted) and imaginary (solid) part for z 7→ z 3 ; they (dotted) and argument (solid) for z 7→ z 3 ; they
intersect perpendicularly. intersect perpendicularly.
27
The complex exponential:
The image of the vertical (solid) and horizontal Vector field (Re exp z, Im exp z) and colouring of
(dotted) lines under f (z) = exp z; they intersect the complex plane for exp z.
perpendicularly.
Colouring and level curves of constant real (dotted) Colouring and level curves of constant modulus
and imaginary (solid) part for exp z; they intersect (dotted) and argument (solid) for exp z; they in-
perpendicularly. tersect perpendicularly.
28
The complex cosine:
The image of the vertical (solid) and horizontal Vector field (Re cos z, Im cos z) and colouring of the
(dotted) lines under f (z) = cos z; they intersect complex plane for cos z.
perpendicularly.
Colouring and level curves of constant real (dotted) Colouring and level curves of constant modulus
and imaginary (solid) part for cos z; they intersect (dotted) and argument (solid) for cos z; they in-
perpendicularly. tersect perpendicularly.
29
The complex sine:
The image of the vertical (solid) and horizontal Vector field (Re sin z, Im sin z) and colouring of the
(dotted) lines under f (z) = sin z; they intersect complex plane for sin z.
perpendicularly.
Colouring and level curves of constant real (dotted) Colouring and level curves of constant modulus
and imaginary (solid) part for sin z; they intersect (dotted) and argument (solid) for sin z; they in-
perpendicularly. tersect perpendicularly.
30
The complex hyperbolic cosine:
The image of the vertical (solid) and horizontal Vector field (Re cosh z, Im cosh z) and colouring of
(dotted) lines under f (z) = cosh z; they intersect the complex plane for cosh z.
perpendicularly.
Colouring and level curves of constant real (dotted) Colouring and level curves of constant modulus
and imaginary (solid) part for cosh z; they intersect (dotted) and argument (solid) for cosh z; they in-
perpendicularly. tersect perpendicularly.
31
The complex hyperbolic sine:
The image of the vertical (solid) and horizontal Vector field (Re sinh z, Im sinh z) and colouring of
(dotted) lines under f (z) = sinh z; they intersect the complex plane for sinh z.
perpendicularly.
Colouring and level curves of constant real (dotted) Colouring and level curves of constant modulus
and imaginary (solid) part for sinh z; they intersect (dotted) and argument (solid) for sinh z; they in-
perpendicularly. tersect perpendicularly.
32
Riemann’s Zeta Function:
The image of the vertical (solid) and horizontal Vector field (Re ζ(z), Im ζ(z)) and colouring of the
(dotted) lines under f (z) = ζ(z); they intersect complex plane for ζ(z).
perpendicularly.
Colouring and level curves of constant real (dotted) Colouring and level curves of constant modulus
and imaginary (solid) part for ζ(z); they intersect (dotted) and argument (solid) for ζ(z); they inter-
perpendicularly. sect perpendicularly.
33
9
II.5. Cauchy-Riemann Revisited (Not examinable!)
We will later see that a holomorphic function f : D → C has derivatives of all orders; in
particular, it thus has continuous partial derivatives ux , uy , vx , vy . Using Theorems II.3.1
& II.3.5 (necessary & sufficienten Cauchy-Riemann conditions), we can therefore state:
f : D → C is holomorphic iff f = u + iv satiesfies the Cauchy-Riemann equations at every
point of the domain D. In this case, f 0 = ux + i vx .
We can state all this using an alternate notation which is more suggestive, but needs some
interpretation. We start with defining two operators ∂/∂z and ∂/∂z by:
∂f 1 ∂f ∂f 1 ∂f ∂f
= −i = +
∂z 2 ∂x ∂y 2 ∂x ∂(iy)
∂f 1 ∂f ∂f 1 ∂f ∂f
= +i = +
∂z 2 ∂x ∂y 2 ∂x ∂(−iy)
34
Theorem II.5.1. Suppose that F is a holomorphic function of two variables and f (z) =
F (z, z). Then,
∂f ∂f
(z) = F1 (z, z), (z) = F2 (z, z),
∂z ∂z
and hence:
f is holomorphic iff F2 = 0, in which case f 0 (z) = F1 (z, z).
35
III. Integration in the Complex Plane
where U (t) = Re (f (γ(t)) · γ 0 (t)) and V (t) = Im (f (γ(t)) · γ 0 (t)) with U, V : [a, b] → R.
Remark. U, V : [a, b] → R are piecewise continuous since f is continuous and γ is piecewise
regular; hence the right-hand side integral exists.
Remark. If we use Leibniz notation, there is an almost irresistible temptation to write
Z Z b Z b
dz
f (z) dz = f (z) dt = f (γ(t)) · γ 0 (t) dt.
γ a dt a
36
Comparison with the real line integral of a vector field. If ~v = (v1 , v2 ) : D → R2 is a
continuous vector field then (see M10)
Z Z Z b
~v · d~s = v1 dx + v2 dy = ~v (γ(t)) · γ 0 (t) dt. (III.2)
γ γ a
ex Verify that the complex path integral can be written in terms of two real path integrals
Z Z Z
f dz = u dx − v dy + i v dx + u dy,
γ γ γ
where f = u + iv.
ex Recall that for a continuous vector field ~v = (v1 , v2 ) : D → R2 and a smooth path
γ : [a, b] → D, γ(t) = (x(t), y(t)) the line integral is defined by Eq. (III.2). One can show
(we take it as a definition here!) that for a simple closed path γ
Z
F (γ) = x dy
γ
is
R the area enclosed by γ (i.e., the area of Iγ ). How can we interpret the path integral
γ
z dz for a simple closed smooth path γ in C?
and, for c ∈ C, Z Z
cf dz = c f dz.
γ γ
37
(ii) Path-additivity of the integral: If γ : [a, b] → D and γ̃ : [ã, b̃] → D with γ(b) = γ̃(ã)
then Z Z Z
f dz = f dz + f dz,
γ+γ̃ γ γ̃
Proof. As h : [ã, b̃] → [a, b] is regular, h0 does not change sign and h(ã) = a, h(b̃) = b
if h0 > 0 and h(ã) = b, h(b̃) = a if h0 < 0. Now compute
Z Z b̃ Z b
0 0
f dz = ((f ◦ γ) · γ ) (h(t)) · h (t) dt = ± ((f ◦ γ) · γ 0 ) (t) dt,
γ̃ ã a
where
• γ1 is the straight line segment from z = 0 to z = 1 + i, and
• γ2 is the polygonal path from z = 0 to z = 1 + i via z = 1.
What do you observe?
R
Orientation of a contour. If Γ ⊂ C is a simple contour then the integral γ
f dz, where
γ is a parametrization of Γ , is determined up to sign:
• Any two regular parametrizations of a (regular) piece of Γ are related by γ̃ = γ ◦ h
so that the integral is unique up to sign, by (iii), and the sign is picked by the choice
of initial point.
• The first choice of initial point determines the initial point of any subsequent regular
piece (after a partition has been fixed).
• The final integral does not depend on possibly different partitions into pieces, by (ii).
38
The same arguments justify the assertion for simple closed contours.
The assertion is not true for non-simple contours: for example, for each loop of a figure-
∞ contour there are two possible orientations so that, in general, there are four different
values for the integral.
We refer to the sense of parametrization of a simple (closed) contour as its orientation;
for a simple closed contour we will (unless stated otherwise) take the counter-clockwise
orientation when writing the contour integral
Z
f dz.
Γ
Much of complex analysis depends on the fact that the following integral for k = −1 does
not vanish.
Example. (
2πi if k = −1,
Z
(z − z0 )k dz =
|z−z0 |=r 0 otherwise,
for k ∈ Z.
Note:
H If one wants to emphasise
H that one integrates around a closed contour, the notations
|z−z |=r
(z − z0 ) dz and ∂B (z ) (z − z0 )k dz are used.
k
0 r 0
Proof. We parametrize the circle ∂Br (z0 ) by [0, 2π] 3 t 7→ γ(t) = z0 + r eit . Note that this
is a counterclockwise parametrization of the circle.
Thus we compute:
Z Z 2π Z 2π
k k 0
(z − z0 ) dz = (γ(t) − z0 ) γ (t) dt = r ek it ir eit dt
∂Br (z0 ) 0 0
Z 2π
= irk+1 cos((k + 1)t) + i sin((k + 1)t) dt
0
(
2πi if k = −1,
=
0 otherwise,
Lemma III.1.1 (M L-inequality). Let f : γ([a, b]) → C be continuous (where γ([a, b]) is a
contour) and denote M = max |f (z)|, and let L be the length of γ([a, b]); then
z∈γ([a,b])
Z
f dz ≤ M L.
γ
39
Remark. Note that |f ◦ γ| : [a, b] → R is continuous and hence
M = max |f (z)| = max |f ◦ γ(t)|
z∈γ([a,b]) t∈[a,b]
Proof. It is enough to prove1 this for a regular path γ (instead of a piecewise regular path).
Let
g(t) = f (γ(t)) · γ 0 (t) = U (t) + iV (t)
R
and let γ f (z) dz = r eiϕ ∈ C. Then, we have (using only real analysis)
Z Z b Z b
−iϕ
f (z) dz = g(t) dt = r = Re r = Re e
g(t) dt
γ a a
Z b Z b Z b
Re e−iϕ g(t) dt ≤
−iϕ
= e g(t) dt = |g(t)| dt
a a a
Z b Z b
0
= |f (γ(t))| · |γ (t)| dt ≤ M · |γ 0 (t)| dt = M · L.
a a
Remark. This result is not hard to prove2 but it is of immense theoretical importance:
Most of the major results of complex analysis involve at least one use of it.
ex Let Γ ⊂ C be a simple closed contour and z0 ∈ IΓ a point in its interior. Prove that
there is a % > 0 so that Z
dz 1
(z − z )k < %k L,
Γ 0
where k ∈ N and L is the length of Γ .
1
Alternative proofs can be found in [ST, Section 6.6] (using a slighly different method to obtain integrals
over real functions only) or, using Riemann sums, in [DET, pp. 85–86] and [ST, Section 6.6]. We give
the proof of [ST, Section 6.6] here:
Rb Rb Rb
First we show | a U (t) + iV (t) dt| ≤ a |U (t) + iV (t)| dt: thus we write X + iY = a U (t) + iV (t) dt;
then
Z b
X2 + Y 2 = (XU (t) + Y V (t)) dt
a
Z b p p
≤ X 2 + Y 2 U 2 (t) + V 2 (t) dt
a
p Z b p
= X2 + Y 2 U 2 (t) + V 2 (t) dt.
a
Now we find Z Z b Z b
0
M |γ 0 (t)| dt = M L
f dz ≤ |f (γ(t))| |γ (t)| dt ≤
γ a a
from real analysis. R R
2 b b
Also, compare this estimate with the corresponding estimate in real analysis a f (x) dx ≤ a |f (x)| dx.
For this, we look at the following example:
it 0
Let f (z) = 1/z and γ : [0, 2π] 3 t 7→ e . Then |f (γ(t))| = 1 and |γ (t)| = 1 (and L = 2π). The
R
M L-estimate is γ f (z) dz ≤ 2π.
R R
However, observe that γ |f (z)| dz = γ 1 dz = 0; so the M L-estimate is not the above estimate from
real analysis with z replacing x naively (compare, however, to the previous footnote)!
40
ex Let γ : [a, b] → C be a closed regular path. For z 6∈ γ([a, b]) we define the winding
number of γ around z by Z
1 dζ
w(γ, z) = .
2πi γ ζ − z
Prove that
(i) w(γ, z) ∈ Z
R t 0 ) dτ
Hint: write ϕ(t) = a γγ(τ(τ)−z and show that (γ(t) − z) e−ϕ(t) is constant – remember
that ex+iy = ex (cos y + i sin y).
(ii) z 7→ w(γ, z) is constant on each connected set A ⊂ C \ γ([a, b])
Hint: first prove that z 7→ w(γ, z) is continuous by using the M L-inequality.
by the chain rule. For a piecewise regular path γ1 + · · · + γn : [a, b] → D with regular
γk : [tk−1 , tk ] → D, where t0 = a and tn = b, we therefore find
Z n
X
f dz = F (γ(tk )) − F (γ(tk−1 )) = F (γ(b)) − F (γ(a)).
γ1 +···+γn k=1
41
III.2. Cauchy’s Theorem
References: [DET, Sections 3.3 & 3.4] and [ST, Chapters 8 & 9]
(Cauchy’s Theorem: [DET, Theorem 3.3.1] and [ST, Section 8.5]; Cauchy’s Theorem for
simply connected domains: [DET, Theorem 3.4.3] and [ST, Theorem 8.8]; Cauchy-Goursat
Theorem: [DET, Lemma 3.3.1] and [ST, Theorem 8.1]; Homotopy version of Cauchy’s
Theorem: [DET, Theorem 3.4.5] and [ST, Theorem 9.4])
The following lemma is the converse to Lemma III.1.2 and provides the ultimate answer
to our initial question: Under
R z which conditions does f have an anti-derivative, i.e., when
does it make sense to write z0 f (z) dz?”.
Lemma
R III.2.1 (Criterion for the FTC). If f : D → C is continuous on a domain D and
Γ
f dz = 0 for every closed contour Γ ⊂ D then f has an anti-derivative F : D → C.
To see that F is holomorphic with F 0 = f , fix z ∈ D and take % > 0 so that B% (z) ⊂ D.
Then, for |h| < %, let γ1 be a path joining z0 and z and γ2 a path joining z0 and z+h and cal-
culate
Z Z
•O. z + h
|F (z + h) − F (z) − f (z)h| = f dz −
f dz − f (z)h
γ
Z 21
γ1
γ2 •H z
?
= f (z + th)h dt − f (z)h B% (z)
γ1
Z0 1
z0 •
= (f (z + th) − f (z))h dt
0
M L-ineq.
≤ max |f (z + th) − f (z)| |h|,
t∈[0,1]
R
where in the step “?” we used that Γ
f dz = 0 for any closed contour.
F (z+h)−F (z)
Thus h
→ f (z) as |h| → 0 since f is continuous at z.
The
R following is the main theorem of this course(!), it puts forward conditions under which
Γ
f dz = 0 when there is no initial reason for f to have an anti-derivative.
Theorem III.2.2 (Cauchy’s Theorem). Let f : D → C be holomorphic in Ra domain D
and let Γ ⊂ D be a simple closed contour so that its interior IΓ ⊂ D. Then Γ f dz = 0.
42
Corollary III.2.3 (Cauchy’s Theorem for simply connected domains). Let f : D → C be
holomorphic in a simply connected domain D. Then
Z
f dz = 0
Γ
Remark. Cauchy’s Theorem directly implies the Corollary for simple closed contours Γ
but it can be proven to hold for any closed contour Γ , see for example [DET, Proof of
Theorem 3.4.3].
The previous Corollary III.2.3 together with the Criterion for the FTC for path integrals
(Lemma III.2.1) yields the following corollary.
Corollary III.2.4 (Existence of anti-derivatives). If f : D → C is holomorphic in a simply
connected domain D, then it has an anti-derivative F : D → C.
There are various versions3 of Cauchy’s Theorem. Our proofs, however, will only work for
simple closed contours Γ , or for rather special domains D, respectively.
We shall prove the theorem in two special versions:
(i) when additionally f 0 is continuous in D, or
(ii) when Γ is a triangle.
43
Theorem III.2.5 (Cauchy’s Theorem (weak version I)). Let f : D → C be holomorphic
in a simply connected domain D, with continuous f 0 , and let γ1 , γ2 : [0, 1] → D be two
smooth paths which have common endpoints and can be smoothly deformed into each other
within D. Then Z Z
f dz = f dz. (III.3)
γ1 γ2
Proof. Set Z Z 1
∂C(s, t)
I(s) = f (z) dz = f (C(s, t)) dt.
C(s,t) 0 ∂t
Then Eq. (III.3) is equivalent to I(0) = I(1), which will be proved if we can show that
I 0 (s) = 0 for 0 < s < 1. Reversing – using Lemma III.2.6 – the order of differentiation and
integration, we obtain
Z 1
0 d ∂ ∂C(s, t)
I (s) = I(s) = f (C(s, t)) dt
ds 0 ∂s ∂t
Z 1
∂2C
0 ∂C ∂C
= f (C(s, t)) + f (C(s, t)) dt
0 ∂s ∂t ∂s∂t
Z 1
∂ ∂C(s, t)
= f (C(s, t)) dt
0 ∂t ∂s
t=1
∂C(s, t)
= f (C(s, t)) ,
∂s t=0
∂C(s,0) ∂C(s,1)
which is zero because C(s, 1) and C(s, 0) are constants (thus ∂s
=0= ∂s
).
The following lemma was used to reverse the order of differentiation and integration.
Lemma III.2.6. Let f (z, w) and6Rfz (z, w) be continuous for z in a domain D and w on
a simple contour Γ . Then F (z) = Γ f (z, w) dw is holomorphic in D, and
Z
0
F (z) = fz (z, w) dw.
Γ
ex Fill in the details (all “ ”) in the following proof of Lemma III.2.6.
5
Even these conditions need only hold piecewise, in the sense that there exists a partition 0 = t0 < t1 <
. . . < tn = 1 such that the conditions above hold on each strip [0, 1] × [tk−1 , tk ].
6
So, f (z, w) is complex differentiable with respect to z, i.e.,
f (z, w) − f (z0 , w)
fz (z0 , w) = lim .
z→z0 z − z0
44
Proof. Let F (z) = U (x, y) + iV (x, y) and f (z, w) = u (x, y, ξ(t), η(t)) + iv (x, y, ξ(t), η(t)),
where γ(t) = ξ(t) + iη(t) is a parametrisation of Γ with t ∈ [a, b]. Then
Z b Z b
U (x, y) = and V (x, y) = .
a a
U (x0 + h, y0 ) − U (x0 , y0 ) 1 b
Z
Ux (x0 , y0 ) = lim = lim
h→0 h h→0 h a
Z b Z b
MVT
= lim = [ux (x0 , y0 , ξ, η) ξ 0 − vx (x0 , y0 , ξ, η) η 0 ] dt.
h→0 a a
The justification for taking the limit under the integral sign in the last step is supplied by
and the following lemma (for a proof see [DET, Theorem 3.5.2]):
Lemma. Let f (z,R w) be continuous for z in a domain D and w on a simple contour
Γ . Then F (z) = Γ f (z, w) dw is continuous in D.
Similarly,
Z b Z b
Uy (x0 , y0 ) = , Vx (x0 , y0 ) = ,
a a
Z b
Vy (x0 , y0 ) = .
a
Ux (x0 , y0 ) = and = .
Alternatively, one can also prove a weak version using Green’s Theorem (cf. M10 or [DET,
Theorem 3.1.2]).
Theorem (Green’s Theorem). Let α, β : R2 ⊃ D → R be continuously differentiable and
Ω ⊂ D bounded with piecewise smooth boundary ∂Ω. Then
Z ZZ
α dx + β dy = (βx − αy ) dx dy,
∂Ω Ω
45
ex Let f : D → C be holomorphic in a domain D, with continuous f 0 , and let Γ ⊂ D be
aR simple closed contour so that its interior IΓ ⊂ D. Use Green’s theorem to show that
Γ
f dz = 0.
Remark. (i) Our proofs require continuity of f 0 but the more general statement of
Cauchy’s Theorem III.2.2 above yields the corollary without that assumption (cf. [DET,
Theorem 3.3.1]).
(ii) The proof using smooth deformations requires Cs , Ct etc. to be piecewise continuous,
but it can be formulated and holds with C continuous (cf. [ST, Theorem 9.3]).
(iii) The proof using Green’s Theorem requires Γ to be simple closed but (in contrast
to the above version of Cauchy’s Theorem III.2.2) it can be formulated and holds
without it (cf. [DET, Proof of Theorem 3.4.3]).
z3 /
Definition. A triangle is a simple closed polygonal •/
///
contour consisting of three line segments, ∆ = [z2 , z3 ]∪ //
//
[z3 , z1 ]∪[z1 , z2 ] (where [z, z̃] = {z +t (z̃ −z)|t ∈ [0, 1]}).
z1• •z
2
46
R
For the proof, let L denote the length of (the perimeter of) ∆ and I = | ∆
f dz|.
Let ∆1 = ∆, and suppose that ∆n with (III.4) has just been constructed.
Join the midpoints of the sides of ∆n to produce four
similar triangles Γj , j = 1, 2, 3, 4. Then, by the path-
additivity of the integral
•/
///
4 Z
4 Z
W/
////// W/
X X
In = f dz ≤ f dz .
// /
j=1 Γj Γ4 ////////
Γj
j=1
/ //
/ // //
• // o • / /
R
Let k be such that Γk f dz is the largest of W////// W///////
G
//
///////// Γ2 ///////
R R
f dz
, . . . ,
f dz
and set ∆ n+1 = Γ k . Then
/// // ///
Γ1 // Γ3 //
Γ1 Γ4
// //
/ /
I≤4 n−1
In ≤ 4 n−1 n
· 4In+1 = 4 In+1 . • • / •
Denote by Tn = ∆n ∪ I∆n the triangle ∆n with its interior I∆n (so, Tn is the filled
triangle). All Tn are compact, Tn+1 ⊂ Tn , and diam Tn ≤ 12 Ln → 0 as n → ∞. By
Cantor’s Intersection Theorem (see p. 10) there is a point z0 ∈ C so that
\
{z0 } = Tn .
n∈N
47
differentiablity means that
L2n
∀ε > 0 ∃N ∈ N s.t. ∀n ≥ N : In ≤ ε .
2
Writing
f (z) = f (z0 ) + f 0 (z0 )(z − z0 ) + η(z)(z − z0 ),
2
we note that f (z0 ) + f 0 (z0 )(z − z0 ) is the derivative of z 7→ z f (z0 ) + f 0 (z0 ) (z−z2 0 ) . So,
by the FTC for path integrals, the integral ∆n (f (z0 ) + f 0 (z0 )(z − z0 )) dz vanishes.
R
zz
zz
Z Z
z
In = f (z) dz = 0 + η(z)(z − z0 ) dz zz
∆n ∆n zzz•
zz z0
1 ε L2 zz
≤ Ln · (ε · Ln ) = ∆N
2 2 · 4n−1
for all n ≥ N by the M L-inequality.
Conclusion: Fix ε > 0 and let N ∈ N be as in (vi); then, for n ≥ N ,
ε L2 L2
I ≤ 4n−1 In ≤ 4n−1 = ε ,
2 · 4n−1 2
and we find I = 0, as desired, since ε > 0 was arbitrary.
Remark. This constitutes the first part of a proof for the above more general version of
Cauchy’s Theorem. The next step would be to show it for (convex) polygons, then arbitrary
closed polygonal contours (both by triangulation), and finally using that arbitrary simple
closed contours can be well-approximated by a polygonal contour; compare [DET, Section
3.3 & Proof of Theorem 3.4.3] and [ST, Section 8.5]. The result will then be Theorem
III.2.2.
Example. Any disk Br (z) is a star-domain and, more generally, any open convex set C
is a star-domain with any z0 ∈ C as a centre (C ⊂ C is convex if, for any pair of points
z, z̃ ∈ C, the line segment [z, z̃] ⊂ C).
48
We can now prove the existence of anti-derivatives on star-domains from the Cauchy-
Goursat theorem (without using the general version of Cauchy’s theorem) by replacing the
general paths in the above proof by line segments:
Corollary III.2.9 (Existence of anti-derivatives). If f : D → C is holomorphic in a
star-domain D then it has an anti-derivative F : D → C.
Proof. This follows from the existence of an anti-derivative and the FTC for path integrals
(Lemma III.1.2).
Remark. Note that we do not need to assume Γ to be simple closed.
ex Prove that the function f (z) = z1 has an anti-derivative F in the cut plane C \ R≤0 =
{z = r eiϕ |r > 0, −π < ϕ < π}. Set F (1) = 0, what do you get for F (z) with z ∈ C \ R≤0 ?
Hint: For the second part, find a suitable path joining 1 and z = r eiϕ .
Remark: This anti-derivative with F (1) = 0 is called the principal value of the (complex)
logarithm and denoted Log(z).
Proof. We take two (piecewise) regular simple and disjoint paths that “join” Γ1 and Γ2 ,
49
Now we denote:
• γj the arc on Γj from α1 (j) to α2 (j) and
γ1
o
• γ̃j the arc on Γj from α2 (j) to α1 (j).
Then, by Cauchy’s Theorem,
Z _
Γ1
γ2
0= f dz,
γ1 +α2 −γ2 −α1
Z
0= f dz. α2 α1
γ̃1 +α1 −γ̃2 −α2 • •/ •o •
Adding up, we obtain
Z Z Z Z γ̃/ 2
0= f dz + f dz + f dz + f dz Γ2
γ1 γ̃1 −γ2 −γ̃2
Z Z
= f dz − f dz
γ̃1
Γ1 Γ2
as desired.
50
(an ellipse with half-axes radii 2 and 7).
ex We again explore winding numbers (compare exercise on p. 41).
(i) Let γ : [a, b] → C be a simple closed regular path, so that Γ = γ([a, b]) is oriented
counter-clockwise as usual. Prove that, for z 6∈ Γ ,
(
1 if z ∈ IΓ ,
w(γ, z) =
0 otherwise.
o
? _
_ ?
/
/ o
51
Since IΓ is open and f continuous at z0 , we can fix δ > 0 s.t.
dz
R
and, since |z−z0 |=δ z−z 0
= 2πi (used in the step “?”) and by the M L-inequality,
−
Z Z
1 f (z) ? 1 f (z) f (z0 )
dz − f (z0 ) = dz
Γ z − z0 z − z0
2πi 2πi
|z−z0 |=δ
M L-ineq. 1 f (z) − f (z0 )
≤ max 2πδ
2π |z−z0 |=δ z − z0
|z−z0 |=δ
= max |f (z) − f (z0 )|
|z−z0 |=δ
∂Bδ (z0 )⊂B2δ (z0 )
< ε.
1
R f (z)
Thus | 2πi Γ z−z0
dz − f (z0 )| = 0.
If we now knew that we could differentiate under the integral we would obtain the formulae
Z
(n) n! f (w)
f (z) = dw
2πi Γ (w − z)n+1
for the derivatives of f . However, can we really interchange differentiation and integration?
The answer is “yes” by Lemma III.2.6 (that we have already used in Theorem III.2.5).
Remark. (Not examinable!)
Even without Lemma III.2.6 we are able to prove Theorem III.3.2 by taking the following different
route:
Lemma. Let γ : [a, b] → C be a piecewise regular path and let Γ = γ([a, b]); suppose
ϕ : Γ → C is continuous. For n ∈ N define
Z
ϕ(w)
fn : C \ Γ → C, fn (z) = n
dw.
γ (w − z)
In particular,
52
Proof. Consider
τ 7→ g(τ ) = ψ(z) − ψ(τ ) − ψ 0 (τ )(z − τ ).
This function is differentiable with
g 0 (τ ) = −ψ 00 (τ )(z − τ );
From this
≤ max |ψ 00 (τ )| |z − z0 |2
τ ∈[z0 ,z]
by the M L-inequality.
auxiliary lemma,
ψ(z) − ψ(z0 ) − ψ 0 (z0 )(z − z0 )
≤ max n(n + 1) |z − z0 |
z − z0 τ ∈[z0 ,z] (w − τ )
n+2
n(n + 1)
≤ |z − z0 |
%n+2
for any z ∈ B% (z0 ) and all w ∈ Γ .
Now take z ∈ B% (z0 ); then, using the M L-inequality again,
fn (z) − fn (z0 )
− nfn+1 (z0 )
z − z0
Z
1 1 1
= ϕ(w)
−
Γ (z − z0 ) (w − z)n (w − z0 )n
n
− n+1
dw
(w − z0 )
0
Z
ψ(z) − ψ(z0 ) − ψ (z0 )(z − z0 )
= ϕ(w)
dw
Γ z − z0
0 (z )(z − z )
ψ(z) − ψ(z 0 ) − ψ 0 0
≤ max ϕ(w) L
w∈Γ z − z0
n(n + 1)L
≤ max |ϕ(w)| |z − z0 | → 0 as z → z0 .
w∈Γ %n+2
Thus fn is (complex) differentiable at z0 with fn0 (z0 ) = nfn+1 (z0 ).
53
Proof. This follows now immediately from Lemma III.2.6 (or the lemma in the previous
remark).
Remark. This shows that Corollary II.3.4 (real and imaginary parts of a holomorphic
function are harmonic) holds without the additional assumption of f being twice continu-
ously differentiable.
R
Corollary III.3.4 (Morera’s Theorem). Let f : D → C be continuous so that Γ f (z) dz =
0 for any closed contour Γ ⊂ D. Then f is holomorphic.
R
Proof. Let f : D → C be continuous and Γ
f (z) dz = 0 for every closed contour Γ ⊂ D.
By the criterion for the FTC (Lemma III.2.1), f has an anti-derivative F : D → C. In
particular, F is holomorphic with F 0 = f .
But, since F is holomorphic, it has derivatives F (n) of any order in D, in particular, f 0 = F 00
exists.
III.4. Applications
References: [DET, Section 3.6] and [ST, Sections 10.4 & 10.7]
(Liouville’s Theorem: [DET, Theorem 3.6.4] and [ST, Theorem 10.6]; Gauss’ Fundamental
Theorem of Algebra: [DET, Theorem 3.6.5] and [ST, Theorem 10.7]; Local Maximum
Modulus Theorem: [ST, Proposition 10.12]; Potential theory (physics): [DET, Chapter
6] and [ST, Section 13.4]; Logarithms and complex powers: [DET, Section 2.4] and [ST,
Sections 7.3 & 14.5 & 14.6])
54
III.4.1. Theoretical Applications
ex Prove Gauss’ Fundamental Theorem of Algebra.
1
Hint: suppose that a polynomial p(z) has no zeros and conclude that f (z) = p(z)
is
bounded.
p(z) = c(z − z1 ) · · · (z − zn )
By the triangle inequality (for real integrals) and since |f (z)| ≤ |f (z0 )| for all z with
|z − z0 | = r we find
Z 2π Z 2π
1 it 1
|f (z0 )| ≤ |f (z0 + r e )| dt ≤ |f (z0 )| dt = |f (z0 )|.
2π 0 2π 0
9
Recalling that the extended complex plane Ĉ = C ∪ {∞} is compact (and that the continuous image of
a compact set is compact, and thus bounded in C), Liouville’s Theorem has the following consequence:
The only holomorphic functions f : Ĉ → C are the constants.
55
Thus, equality holds throughout the previous line; this gives
Z 2π
|f (z0 )| − |f (z0 + r eit )| dt = 0 ⇒ ∀t ∈ [0, 2π] : |f (z0 )| − |f (z0 + r eit )| = 0
0
Rb
since |f (z0 )| − |f (z0 + r eit )| ≥ 0 and continuous (noting that if a
g(x) dx = 0 for a
nonnegative real continuous function g then g ≡ 0
ex ).
Since this holds for all r ∈ (0, R) we conclude that |f |, hence f (see p. 23), is constant on
BR (z0 ).
We look at harmonic functions again (compare Corollary II.3.4), and first show why they
are studied in physics (fluid dynamics/electrostatics):
A complex function f˜(z) defines a two-
4
dimensional vector field (u, v, 0) which
represents a steady-state (there is no
time-dependence), laminar (stratified into 2
10
If you want to see what the effect of “laminar” is, look at the following physics video:
http://www.youtube.com/watch?v=p08 KlTKP50
56
On p. 25, we have already compared the vector fields for for f (z) = cos(z) (see below on
the left) and f (z) = cos(z) (see below on the right). Here, we note that the vector field on
the right satisfies the conditions just stated (e.g., it has no “sinks” and “springs” where
the incompressibility assumptions would certainly not hold).
by Poincaré’s Lemma since the partial derivatives of u and v are continuous and D is a
star-domain.
Finally, U and V are by their gradients determined up to a real constant each. These
constants combine to determine F up to a complex constant.
In physics, a vector field that possesses a potential is called conservative or exact. In that
case, the work exerted to move a particle along a path, only depends on the endpoints of
the path (the difference of the potential at the endpoints).
57
Also note that the functions U and V in 4
-4 -2 0 2 4
Proof. We let α = −uy and β = ux . Then α and β are continuously differentiable with
βy − αx = ∆u = 0 so that, by Poincaré’s Lemma, there is v : D → R with vx = α = −uy
and vy = β = ux .
Now, as u and v are both continuously differentiable and satisfy the Cauchy-Riemann
equations f = u + i v is holomorphic by the sufficient Cauchy-Riemann conditions.
ex Let u(x, y) = xy on D = C. Show that u is harmonic and find a holomorphic function
f with Re f = u.
2
Solution: Clearly ∆u = uxx + uyy = 0 + 0 = 0 and observe that u(x, y) = − Re( iz2 − 42i).
Together with the exponential function, we must also study its inverse function, the (or,
more precisely, a) logarithm.
Definition. Let D be a simply connected domain with 0 6∈ D. A holomorphic function
F : D → C is a logarithm if eF (z) = z for all z ∈ D.
On the cut plane C \ R≤0 , the function Log z = log |z| + i arg(z) with arg(z) ∈ (−π, π)
(and where log denotes the real logarithm) is a logarithm, called the principal value of the
logarithm.
Lemma III.4.5. Let D be a simply connected domain with 0 6∈ D and F be a logarithm
on D. Then F̃ is a logarithm on D iff F̃ = F + 2πi k for some k ∈ Z.
58
Lemma III.4.6. Let D be a simply connected domain with 0 6∈ D. F is a logarithm on D
iff F 0 (z) = z1 on D and eF (a) = a for at least one a ∈ D.
Proof. “⇒”: From eF (z) = z and the chain rule follows 1 = F 0 (z) eF (z) = F 0 (z) · z, and so
F 0 (z) = z1 .
“⇐”: Let F be an anti-derivative of z 7→ z1 with eF (a) = a for one a ∈ D. Then, the
function g(z) = z e−F (z) is holomorphic on D and satisfies g 0 ≡ 0 on D. Thus g is constant
on D, in fact g ≡ 1, and it follows that eF (z) = z for all z ∈ D.
Remark. The principal value of the logarithm Log z coincides on R>0 ⊂ C \ R≤0 with the
usual real logarithm.
Remark. Defining the argument of z by arg z = {ϕ ∈ R|z = |z| eiϕ }, we see that arg is also
a multifunction, its principal value Arg is given by the above restriction arg(z) ∈ (−π, π].
ex Noting that for the multifunction log on the cut plane C \ R≤0 we have log z = log |z| +
i Arg z + 2k πi and log(−z) = log |z| + i Arg z + (2k + 1) πi where k ∈ Z, find the mistake
in the following argument:
We can now also define arbitrary (complex) powers of arbitrary (complex) numbers.
Definition. If a is a complex number, we define for z 6= 0,
z a = ea log z .
59
Example. There are two branches of the square root in C \ R≤0 (i.e., two functions f such
that (f (z))2 = z):
1 1 1 1 1
log z
z 2 = e2 = e 2 (Log z+2πik) = e 2 Log z
eπi k) = ± e 2 Log z
.
Remark. (Not examinable!) Another way out of the above dilemma is to define the log-
arithm (and then also other multifunctions) on a more complicated surface, the so-called
Riemann surface, consisting of infinitely many planes joined together so that the function
varies continuously as one passes from one plane to the next, see [DET, Section 2.6] and
[ST, Sections 14.5 & 14.6].
√
Riemann surface for the cubic root z 7→ 3 z: The
cubic root has three branches that are pasted to-
gether in such a way that the principal branch D0
connects to the second branch D1 , the second to
the third branch D2 and the third to the principal
Riemann surface for the logarithm: countably in- branch again. This picture is taken from R.A. Sil-
finitely many copies Ck (where k ∈ Z, see above) of verman: Introductory Complex Analysis; Dover,
the cut plane (below) are pasted together “at the NY (1972); library: 513.317 SIL (Fig. 16.6.).
respective cut” to allow the logarithm to vary con-
tinuously. This figure is taken from [ST, Fig. 14.8].
60
The principal value of the logarithm Log(z):
The image of the vertical (solid) and horizontal Vector field (Re Log(z), Im Log(z)) and colouring of
(dotted) lines under f (z) = Log(z); they are con- the complex plane for Log(z).
tained in the strip {z ∈ C | Im z ∈ (−π, π]}.
Colouring and level curves of constant real (dotted) Colouring and level curves of constant modulus
and imaginary (solid) part for Log(z); they inter- (dotted) and argument (solid) for Log(z); they in-
sect perpendicularly. tersect perpendicularly.
61
√ 1 1
The principal value of the square root z = z 2 = e 2 Log(z) :
√ √
The image of the vertical (solid) and horizontal Vector field (Re z, Im z) and colouring of the
√ √
(dotted) lines under f (z) = z; they are contained complex plane for z.
in the half plane {z ∈ C | Re z ≥ 0}.
Colouring and level curves of constant real (dotted) Colouring and level curves of constant modulus
√ √
and imaginary (solid) part for z; they intersect (dotted) and argument (solid) for z; they inter-
perpendicularly. sect perpendicularly.
62
IV. Sequences and Series
Proof. . . . as in M11
ex .
Remark. If you are familiar with metric spaces (see M41), then the previous statements
can be reformulated as: The set of all holomorphic functions on D is a complete metric
space where the metric d is given by the supremum metric d(f, g) = supz∈D |f (z) − g(z)|.
Indeed, the Cauchy criterion for uniform convergence is a statement about Cauchy se-
quences in this metric space!
ex Prove that fn → f locally uniformly on D iff fn → f on every compact subset K ⊂ D.
Hint: Use the topological definition of compactness.
63
Proof. . . . as in M11.
ex Prove that the limit function of a locally uniformly convergent sequence of continuous
functions is continuous.
We will need the following technical lemma later (see proof of Lemma IV.6.1).
ex Prove Lemma IV.1.3.
Rb
Proof. Take ε > 0 and let L = a
|γ 0 (t)| dt > 0 be the length of γ. Since fn → f on Γ
there is an N ∈ N so that
ε
|fn (z) − f (z)| <
L
for all n ≥ N and z ∈ Γ . Thus
Z Z
fn (z) dz − f (z) dz ≤ max |fn (z) − f (z)| L < ε
γ γ
z∈Γ
R R
or n ≥ N by the M L-inequality, that is, f (z) dz →
γ n γ
f (z) dz.
1
Z lim fn (w)
n→∞
= dw
2πi |w−z0 |=2% w − z
Z
1 f (w)
= dw
2πi |w−z0 |=2% w − z
64
lim fn (w)
ex that f 0 (z) = 1
dw = lim fn0 (z).
R
A calculation as before shows
2πi |w−z0 |=2%
n→∞
(w−z)2 n→∞
<ε
n+1
ex Let fn : B1 (0) → C, z 7→ fn (z) = zn+1 . Prove that fn → 0 on B1 (0) and fn0 → 0 locally
uniformly but not uniformly on B1 (0).
If ∞
P P∞
k=0 wk is absolutely convergent, i.e., if the series k=0 |wk | is convergent, then it is
convergent.
Proof. . . . as in M11.
65
ex Let fk : D → C, k ∈ N andP∞D ⊂ C any subset of C, satisfy
P∞|fk (z)| ≤ Mk for all k ∈ N
and z ∈ D and assume that k=0 Mk converges. Prove that k=0 fk converges absolutely
and uniformly on D (i.e., the Weierstrass M -test).
Hint: Use the Cauchy criterion.
(Note: Another way to test for absolute convergence, is the ratio test 2 ; if limk→∞ |ak+1 /ak |
exists, then R = 1/ limk→∞ |ak+1 /ak |.)
Theorem IV.2.2. Functions defined by power series are holomorphic; the derivative is
obtained by differentiating term by term.
Proof. Suppose, w.l.o.g., z0 = 0 and take z ∈ BR (0); now choose r so that |z| < r < R.
1 1/n P∞ P∞
Let M = lim supk→∞ |wk | . If M < 1, the series k=1 wk converges; if M >P 1, the series k=1 wk
∞
diverges; if M = 1, then no conclusion can be reached about the convergence of k=1 wk . P∞
2
Let M = lim supk→∞ |wk+1 /wkP | and m = lim inf k→∞ |wk+1 /wk |. Then, if M < 1, the series k=1 wk
∞
converges; if m > 1, the series
P∞ k=1 wk diverges; if m ≤ 1 ≤ M , then no conclusion can be reached
about the convergence of k=1 wk .
66
Then
Z
f (w)
2πi f (z) = dw
|w|=r w−z
Z
f (w) 1
= dw
|w|=r w 1 − wz
Z ∞
f (w) X z k
= dw
|w|=r w k=0
w
∞ Z
(?) X f (w)
= dw zk
k=0 |w|=r wk+1
∞
X f (k) (0) k
= 2πi z
k=0
k!
for wz ≡ |z|
r
< 1 (see Lemma IV.1.4).
Remark. In the course of the proof we learnt that
(i) the coefficients in the Taylor series expansion (around z0 ) are equivalently given by
Z
1 f (w)
ak = dw.
2πi |w−z0 |=r (w − z0 )k+1
(ii) the radius of convergence R of the Taylor series is at least the radius of the largest
disk that is entirely contained in the domain D of f : if B% (z0 ) ⊂ D, then R ≥ %.
ex Let f be an entire function and z0 arbitrary. Show that the Taylor series expansion of
f at z0 has radius of convergence R = ∞.
Definition. We say that a function f : D → C is analytic if
∞
X
∀z0 ∈ D ∃% > 0 s.t. ∀z ∈ B% (z0 ) : f (z) = ak (z − z0 )k
k=0
orem IV.4.1).
In R not even C ∞ -functions
0.4
67
Example.
∞
z
X zk
e =
k=0
k!
∞
ei z − e−i z X z 2 k+1
sin z = = (−1)k
2i k=0
(2 k + 1)!
∞
ei z + e−i z X z2 k
cos z = = (−1)k
2 k=0
(2 k)!
be a power series with radius of convergence R > 0. Suppose f (zn ) = 0 for a sequence
(zn )n∈N ⊂ BR∗ (z0 ) = BR (z0 ) \ {z0 } with zn → z0 . Then ak = 0 for all k = 0, 1, . . . , that is,
f ≡ 0.
Proof. . . . by induction:
• k = 0: First note that, since z 7→ f (z) is continuous,
a0 = f (z0 ) = lim f (zn ) = 0.
n→∞
68
Remark. Thus a power series expansion of a function is unique: if
∞
X ∞
X
k
ak (z − z0 ) = bk (z − z0 )k
k=0 k=0
∀n ∈ N : B 1 (z) ∩ N 0 6= ∅,
n
i.e., we have zn ∈ N 0 ⊂ N with |zn − z| < n1 – this contradicts the assumption z 6∈ N 0 since
zn → z as n → ∞. So, N 0 contains all its limit points and is therefore closed.
Also, N 0 ⊂ D is open in D: by the Cauchy-Taylor Theorem
∞
X
0
∀z0 ∈ N ⊂ D ∃% > 0 s.t. ∀z ∈ B% (z0 ) we have f (z) = ak (z − z0 )k
k=0
Since f is continuous at z0
f (z0 ) = lim f (zn ) = 0.
n→∞
69
Note that, if 0 ≤ t < τ , then there is t̃ ∈ [t, τ ) so that f (γ(s)) = 0 for all s ∈ [0, t̃].
Consequently, t ∈ T .
By the Cauchy-Taylor Theorem
∞
X
∃% > 0 ∀z ∈ B% (z0 ) : f (z) = ak (z − z0 )k
k=0
and by the Identity Theorem for power series f ≡ 0 on B% (z0 ) since f (zn ) = 0 for a sequence
with zn → z0 , zn 6= z0 . And, since γ is continuous,
70
ex Show (without calculations): Given the complex sine and cosine, we have cos2 z +
sin2 z = 1.
Also argue that the compound angle formulae (e.g., cos(z + w) = cos z cos w − sin z sin w
with z, w ∈ C) hold for the complex sine and cosine.
Theorem IV.3.3 (The Maximum Modulus Theorem). Let f : D → C be holomorphic in
a domain and suppose that |f | : D → R has a maximum at z0 ∈ D. Then f is constant.
Proof. Since D is open there is % > 0 so that B% (z0 ) ⊂ D. Then, by the Local Maximum
Modulus Theorem (Theorem III.4.4), f ≡ c ∈ C is constant on B% (z0 ).
And, by the Identity Theorem f ≡ c in D since g(z) = c defines a holomorphic function
on D with f (z) = g(z) for z ∈ B% (z0 ).
ex Suppose that x2 + y 2 ≤ 1. Prove that (x2 − y 2 − 1)2 + 4x2 y 2 attains its maximum value
when x = 0, y = ±1.
ex In this exercise we show that there is no holomorphic function f : B1 (0) → B1 (0) with
f ( 21 ) = 43 and f 0 ( 12 ) = 35 .
(i) Consider the two Möbius transformations ϕ and ψ given by
z + 12 z − 34
ϕ(z) = and ψ(z) = .
1 + 21 z 1 − 43 z
Here, B1∗ (0) = B1 (0) \ {0} denotes the punctured unit disk.
(iii) Now show that there is no holomorphic function f : B1 (0) → B1 (0) with f ( 12 ) = 3
4
and f 0 ( 21 ) = 35 .
Hint: Suppose such a function exists and consider Φ = ψ ◦ f ◦ ϕ.
4
Not to be confused with the earlier statement by the same name that the second mixed partial derivatives
commute.
71
IV.4. Characterisation of Holomorphicity
We summarise the characterisation of holomorphic functions we have obtained:
Theorem IV.4.1. For a continuous function f on a domain D the following statements
are equivalent:
(i) f is holomophic in D.
(ii) f (x + iy) = u(x, y) + iv(x, y) with C 1 -functions u, v : D → R which satisfy the
Cauchy-Riemann equations ux = vy and vx = −uy .
(iii) f (z) = F (z, z) where F is a holomorphic function of two variables such that F2 = 0
(i.e., ∂f
∂z
(z) = 0).
R
(iv) Γ f dz = 0 for every simple closed contour Γ with IΓ ⊂ D.
(v) f is analytic in D.
1
The function f (z) = 1+z 2 on the left, and two partial sums of its Taylor series around 0, namely up to
order 6 in the middle and up to order 32 on the right. Observe that the Taylor series converges in the unit
disk to f , but not outside.
72
We now turn this observation around and ask: Given a holomorphic function on some
“small domain”, can we find a holomorphic function (and if so, how many5 such functions)
on a “bigger domain” (or at least a domain that has some overlap with the former domain)
that extends the function in question.
Definition. If f1 is holomorphic on a domain D1 and f2 is holomorphic on a domain D2 ,
where D1 ∩ D2 6= ∅ and f1 (z) = f2 (z) for all z ∈ D1 ∩ D2 , then we say that f2 is a direct
analytic continuation of f1 to the domain D2 .
The identity theorem has the following consequences (see [DET, Theorem 4.5.1 & Corollary
4.5.1] for a proof):
Theorem IV.5.1. If f2 is holomorphic on a domain D2 and f3 is holomorphic on D3 and
both functions are direct analytic continuations of a holomorphic function f1 on a domain
D1 such that D2 ∩ D3 is connected and not empty and D1 ∩ D2 ∩ D3 6= ∅, then f2 = f3 in
D2 ∩ D3 .
In particular, if f1 is holomorphic on D1 and D2 is domain such that D1 ∩ D2 6= ∅, then
a direct analytic continuation of f1 into D2 is unique if it exists.
5
In real analysis, such a question would “not make sense”: there are many ways to extend a (real)
differentiable function (even if it is C ∞ )!
73
In 8 steps to Riemann’s Zeta Function 7
Riemann’s Zeta function is defined by an analytic continuation. One way to establish the
Riemann’s Zeta function on all of C \ {1} is by considering (and proving) the following
steps:
(1) We begin with the function defined by an infinite series:
∞
X
ζ(z) = n−z .
n=1
This series is convergent iff Re z > 1 (and we always use the principal value n−z =
e−z Log(n) here).
(2) We consider the function ζ1 defined by
Z ∞
[t] − 1 z
ζ1 (z) = z z+1
dt +
1 t z−1
(where [t] denotes the integer part of the real number t). We can show that ζ1 is
holomorphic on {z ∈ C | Re z > 0, z 6= 1} and that ζ1 = ζ for Re z > 1. Thus, ζ1 is a
direct analytic continuation of ζ.
(3) We consider the function ζ2 defined by
Z ∞ 1
[t] − t + 2 1 1
ζ2 (z) = z dt + + .
1 tz+1 z−1 2
One can show that ζ2 = ζ1 on {z ∈ C | Re z > 0, z 6= 1}.
(4) One can show that ζ2 is holomorphic on {z ∈ C | Re z > −1, z 6= 1} and thus ζ2 is a
direct analytic continuation of ζ1 (and ζ).
(5) We consider the function ζ3 defined by
∞ 1
[t] − t +
Z
2
ζ3 (z) = z dt.
0 tz+1
One can show that ζ3 is holomorphic and ζ3 = ζ2 on {z ∈ C | −1 < Re z < 0}. Hence,
ζ3 is an analytic but not a direct analytic contnuation of ζ.
(6) One establishes that
z z−1 1
ζ3 (z) = −2 π z sin πz Γ(−z) ζ(1 − z)
2
if −1 < Re z < 0. Here, Γ denotes the (complex) gamma function (here is another
analytic continuation hidden!).
6
In the Riemann surface picture, we “go up the helix” if we go round the origin counter-clockwise and
“down the helix” if we go round the origin in clockwise direction, compare p. 60.
7
This is adapted from Open University Complex Analysis Course Team: “Course M332 Unit 15, Complex
Analysis: Number Theory”, The Open University Press, Milton Keynes (1975); library: 513.317 OPE.
74
(7) One shows that the function
z z−1 1
ζ4 (z) = −2 π z sin πz Γ(−z) ζ(1 − z)
2
Im
O Im O Im O
···················································································· ···························································································································· ····································································································································································
······························································· ····························································································· ···························································································································
······························································· ····························································································· ···························································································································
······························································· ····························································································· ···························································································································
······························································· ····························································································· ···························································································································
······························································· ····························································································· ···························································································································
−1
◦·
······························································/ Re
1······························································· −1
······························◦·······························································/ Re
······························1······························································· −1
····························································◦·······························································/ Re
····························································1·······························································
···················································································· ···························································································································· ····································································································································································
······························································· ····························································································· ···························································································································
······························································· ····························································································· ···························································································································
······························································· ····························································································· ···························································································································
······························································· ····························································································· ···························································································································
The domain on which ζ (see step (1)) is defined on the left, the domain on which ζ1 (see step (2)) is
holomorphic in the middle, and the domain on which ζ2 (see step (4)) is holomorphic on the right.
Im O Im O
········································ ················································································
······························ ····························································
······························ ····························································
······························ ····························································
······························ ····························································
······························ ····························································
······························ ◦ / Re ···························································· ◦ / Re
−1 ········································ ····································−1 ······················
··················································································
1 1
······························
······························ ····························································
······························ ····························································
······························ ····························································
······························ ····························································
The domain on which ζ3 (see step (5)) is holomorphic on the left and the domain on which ζ4 (see step
(7)) is holomorphic on the right.
75
3
So, the series z1 − 2!z + z4! − . . . converges to (cos z)/z for every nonzero z; we have a series
expansion for this function, and the fact that 0 is a “singularity” of the function explains,
or is explained by, the presence of the term 1/z in the series.
Definition. A Laurent series is a series of the form
∞
X ∞
X
−k
a−k (z − z0 ) + ak (z − z0 )k ;
k=1 k=0
We call ∞ −k
the principal part (or singular part) and ∞ k
P P
k=1 a−k (z − z0 ) k=0 ak (z − z0 ) the
regular part (or power series part) of the Laurent series.
Remark. Given a series ∞ −k
P
k=1 a−k (z − z0 ) there is an R ∈ [0, ∞] so that the series
(i) diverges for |z − z0 | < R and
(ii) converges (absolutely) for |z − z0 | > R.
Proof. W.l.o.g. z0 = 0. Let % ∈ [0, ∞] denote the radius of convergence of the power series
∞
X
a−k wk .
k=1
1
Then, with w = z
and R = %1 , the series
(i) converges (absolutely) when
1 1
|w| = <% ⇔ |z| > = R;
|z| %
76
Example. The annulus of convergence of the Laurent series
∞
X
zk
k=−∞
P∞ k
is empty:
P∞ −k its regular part k=0 z has radius of convergence R2 = 1 and the singular part
k=1 z converges for |z| > R1 = 1.
(i) 0 < |z| < 1, (ii) 1 < |z| < 2, (iii) 2 < |z|.
(i) 0 < |z| < 1, (ii) 0 < |z − 1| < 1, (iii) 0 < |z − 2| < 1.
A = {z ∈ C | R1 < |z − z0 | < R2 }.
We already know (from the Weierstrass M -test, as in the proof of Theorem IV.2.2) that
f2 (z) converges absolutely and locally uniformly on BR2 (0) for some R2 ∈ [0, ∞].
77
Similarly, we know that there is an R1 ∈ [0, ∞] so that the power series
X ∞
1
f1 = a−k wk
w k=1
1
C \ {0} 3 z 7→ w = ∈ C \ {0}
z
is continuous, f1 (z) also converges locally uniformly for |z| > R1 by Lemma IV.1.3.
Consequently the Laurent series f1 (z) + f2 (z) converges absolutely and locally uniformly
when |z| > R1 and |z| < R2 .
Theorem IV.6.2. Functions defined by Laurent series are holomorphic in their annuli of
convergence; the derivative is obtained by differentiating term-by-term.
A = {z ∈ C | R1 < |z − z0 | < R2 } ⊂ D.
Then, for z ∈ A,
∞ Z
X
k 1 f (w) dw
f (z) = ak (z − z0 ) with ak =
k=−∞
2πi |w−z0 |=r (w − z0 )k+1
R2
R1 < r1 < |z − z0 | = |z| < r2 < R2 r2
and % > 0 so that B2% (z) ⊂ {w ∈ C | r1 < |w| <
r2 } (it is actually enough to have B% (z) ⊂ {w ∈ r1 •z
C | r1 < |w| < r2 }).
R1
•z0
78
By Cauchy’s formula and the Homotopy version
?
of Cauchy’s theorem8
? •z
Z
1 f (w)
f (z) = dw
2πi |w−z|=% w − z •z0
−1
Z Z
f (w) 1 f (w)
= dw + dw
2πi |w|=r1 w − z 2πi |w|=r2 w − z Γ
= f1 (z) + f2 (z).
w k
< 1 so that w 7→ ∞ f (w)
r
since wz = |z|
P
1
k=0 w z
converges uniformly on {w ∈ C | |w| = r1 }
(by Lemma IV.1.4).
Then we note that Z Z
f (w) f (w)
dw = dw
|w|=ri wk+1 |w|=r wk+1
for any r ∈ (R1 , R2 ) by the Homotopy version of Cauchy’s theorem since w 7→ wf (w) k+1 is
8 1
R f (w)
For the step from the first to the second line, observe that 2πi Γ w−z
dw = 0 by Cauchy’s Theorem
where Γ is the grey contour in the picture to the right and note that the direction is chosen such that
the paths between ∂Br1 (z0 ) and ∂Br2 (z0 ) cancel with ∂B% (z).
79
V. Residue Calculus
80
sin z
(i) z →
7 z
= 1 − 61 z 2 1 4
+ 120 z − . . . has a removable singularity at the point z0 = 0;
z
e −1
z→ 7 z
= 1 + 12 z 1 2
+ 6 z + . . . has a removable singularity at the point z0 = 0;
7 sin
(ii) z → z2
z
= z1 − 16 z + 120
1 3
z − . . . has a “simple pole” (a pole of order n = 1) at z0 = 0;
z→ 7 1/z 2 has a pole of order 2 at z0 = 0;
(iii) z 7→ sin z1 = z1 − 6z13 + 120z
1
5 − . . . has an essential singularity at z0 = 0;
−1/z 2
z 7→ e = 1 − z2 + 2 z4 − 6 1z6 + . . . has an essential singularity at z0 = 0;
1 1
1
(iv) z 7→ sin1 1 has singularities for z = kπ , k ∈ Z and z = 0; therefore z0 = 0 is not an
z
isolated singularity and the above classification does not apply to z0 = 0.
We use the “colouring method” introduced in Section II.4 to depict these functions on the
next two pages.
ex Let p be a polynomial. Show: |p(z)| → ∞ as |z| → ∞.
(Not examinable!) Regarding p as a function from the extended complex plane Ĉ = C∪{∞}
to C, what are we showing here in terms of singularities? Do other entire functions like
exp or sin also have this property?
ex Prove that z0 ∈ C is a pole of order n ∈ N of f iff there is a holomorphic function
g : BR (z0 ) → C, defined on some disk about z0 , with g(z0 ) 6= 0 and g(z) = (z − z0 )n f (z)
for 0 < |z − z0 | < R.
ex Let f : D → C be holomorphic. We say that z0 ∈ D is a zero of order m ∈ N of f if
the Taylor series expansion of f at z0
∞
X
f (z) = ak (z − z0 )k , where am 6= 0.
k=m
ex Each of the following functions f has an isolated singularity at z = 0. Determine its
nature; if it is a removable singularity define f (0) so that f is holomorphic at z = 0; if it
is a pole find the singular part; if it is an essential singularity just state it.
cos(z)−1
(i) f (z) = z
1/z
(ii) f (z) = e
cos(1/z)
(iii) f (z) = 1/z
1
(iv) f (z) = 1−ez
1
The e-book-link of the library leads to the Italian edition of that book!
81
z −1
z7→ sinz z z7→ e z
sin z ez −1
The map z 7→ z
has a removable singularity at 0. One can The map z 7→ z
has a removable singularity at 0. One
holomorphically extend this function to an entire function can holomorphically extend this function to an entire func-
with value 1 at 0. Indeed, the neighbourhood of 0 is coloured tion with value 1 at 0. Indeed, the neighbourhood of 0 is
red. coloured red.
sin z 1
The map z 7→ z2
has a pole at 0. Colours in the neigh- The map z 7→ z2
has a pole at 0. Colours in the neighbour-
bourhood of 0 are dark. hood of 0 are dark.
82
1 2
z7→sin z
z7→e−1/z
1 2
The map z 7→ sin z
has an essential singularity at 0. In the The map z 7→ e−1/z has an essential singularity at 0. In
neighbourhood of 0, colours from the imaginary side(s) are the neighbourhood of 0, bright (along the imaginary axis)
dark, while they are bright along the real axis. and dark (along the real axis) can be found. Compare this
complex function with the corresponding (very smooth) real
function on p. 67.
1
z7→ 1 z7→ζ(z)
sin z
1
The map z 7→ 1/ sin z
has no isolated singularity at 0. Poles Riemann’s Zeta Function has a simple pole at z0 = 1 and is
1
can be found at z = kπ
, k ∈ Z, which is reflected in the holomorphic on C \ {1}, compare Section IV.5. This can be
black spots along the real axis (with an accumulation point used to show that there are infinitely many primes.
at 0).
83
Theorem V.1.1 (Characterization of singularities). Suppose z0 ∈ C is an isolated singu-
larity of f , i.e., f : BR∗ (z0 ) → C is holomorphic. Then
(i) z0 is removable iff lim supz→z0 |f (z)| < ∞, i.e., iff f is bounded in a neighbourhood
of z0 ;
(ii) z0 is a pole iff |f (z)| → ∞ as z → z0 ;
(iii) z0 is an essential singularity iff, for every c ∈ C, there is a sequence (zn )n∈N with
zn → z0 and f (zn ) → c for n → ∞.
Remark. Part (iii) is the remarkable part – its essence can be rephrased as the
In fact, more is true and we state without proof (cf. [DET, Theorem 4.6.5]):
Note that we therefore have a sequence zn → z0 so that f (zn ) = c for every but possibly
one given c ∈ C. Compare this statement with the Identity Theorem for holomorphic
functions (Theorem IV.3.2)!
84
so that
1
g : Br∗ (z0 ) → C, z 7→ g(z) =
f (z)
is a well defined holomorphic function which is bounded and therefore, by (i),
extends holomorphically to Br (z0 ) with g(z0 ) = 0.
Since g 6≡ 0 we have, for z ∈ Br (z0 ),
∞
X
g(z) = bk (z − z0 )k = (z − z0 )n h(z)
k=n
1
ex Prove that g has a pole of order n ∈ N at z0 if and only if f (z) = g(z)
has a zero of
order n.
ex Prove the Casorati-Weierstrass Theorem.
1
Hint: Proof by contradiction; fix c and consider g = f −c .
85
Remark. By Laurent’s Theorem
Z
1
Res(f, z0 ) = a−1 = f (w) dw
2πi |w−z0 |=r
a−1
We have f (z) = z−z0
+ g(z) with g holomorphic in some disk BR (z0 ), so that
ex Prove the p/q 0 -rule: Suppose p, q : BR (z0 ) → C are holomorphic and q has a zero of
order n = 1 at z0 , i.e., q(z0 ) = 0 and q 0 (z0 ) 6= 0. Then f = pq has Res(f, z0 ) = qp(z 0)
0 (z ) .
0
g(z)
ex Let z0 be a pole of order n of f , i.e., f (z) = (z−z0 )n
, where g is holomorphic in some
BR (z0 ) with g(z0 ) 6= 0. Prove that
g (n−1) (z0 )
Res(f, z0 ) = .
(n − 1)!
Hint: Use Cauchy’s formulae for the derivatives.
86
At all other points, z 6= z1 , . . . , zm , f is holomorphic and therefore Res(f, z) = 0 so that
X m
X
Res(f, z) = Res(f, zi ).
z∈IΓ i=1 Γ
Now, since the zi are isolated there are %i > 0 so that •
•
B %i (zi ) ∩ B %j (zj ) = ∅ for i 6= j.
Then
•
Z m Z
X m
X
f dz = f dz = 2πi Res(f, zi ) •
Γ i=1 |z−zk |=%k i=1
87
has no limit as R, r → ∞ but the limit r = R → ∞ does exist.
R∞
However, if −∞ f (x) dx does exist, then the principal value integral exists and
Z ∞ Z ∞
PV f (x) dx = f (x) dx.
−∞ −∞
R∞ dx
Example. Integrate 0 1+x4
.
(i) We
R associate with the given real integral a related contour integral, of the form
Γ
f (z) dz.
We observe that Z R
Z R
dx dx
2 4
= 4
.
0 1+x −R 1 + x
R 1
So, we consider the contour integral Γ 1+z 4 dz where Γ = [−R, R] ∪ Γ (R) (with
it
Γ (R) = {R e |t ∈ [0, π]}) is a semicircular contour.
(ii) We use the Residue Theorem to evaluate the contour integral. O
1 4
= − ξ8k , since ξ8k = −1.
4
Then the Residue Theorem yields (for R > 1)
Z
1 3
dz = 2πi Res(f, ξ8 ) + Res(f, ξ8 )
Γ 1 + z4
√
√
1 3
1 2
= 2πi − · ξ8 + ξ8 = − πi · 2i = π.
4 2 2
(iii) We split the contour integral into two parts: a real integral we are interested in, and
a complex integral we want to get rid of.
Obviously,
Z R Z R Z π
iR eit
Z Z
1 1 1 1
4
dz = 4
dx+ 4
dz = 4
dx+ 4 4it
dt.
Γ 1+z −R 1 + x Γ (R) 1 + z −R 1 + x 0 1+R e
88
(iv) We use the M L-inequality to show that this complex integral becomes arbitrarily small
in modulus if we take R to be large enough.
Using the M L-inequality, we have (by |z 4 + 1| ≥ |z|4 − 1)
Z
1 1
4
dz ≤ max 1 + R4 e4it · π R
Γ (R) 1 + z t∈[0,π]
1 R
≤ 4 · πR = π 4 → 0 as R → ∞.
R −1 R −1
R∞ dx
Hence the integral 0 1+x4
exists and
Z ∞ Z ∞
dx 1 dx π
= πi Res(f, ξ8 ) + Res(f, ξ83 ) = √ .
4
= PV 4
0 1+x 2 −∞ 1+x 2 2
R∞ cos x dx
ex Evaluate 0 (1+x2 )(4+x2 )
using the Theorem of Residues.
eiz
Hint: Consider f (z) = (z 2 +1)(z 2 +4)
.
dz
R
ex Compute |z|=2 z 3 −1
.
R∞
ex Let α > 0. Show that −∞ cos
αx
1+x2
dx = π e−α .
Remark: This is a question from the 2006 exam.
ex For n ∈ N, evaluate
n
e(z )
Z
dz.
|z|=1 z
Hence show that Z 2π
ecos(nϑ) cos (sin(nϑ)) dϑ = 2π
0
and Z 2π
ecos(nϑ) sin (sin(nϑ)) dϑ = 0.
0
Remark: This is a question from the 2008 exam.
R∞
ex Evaluate 0 sin xx dx .
Remark: This is a hard one!
89
Consider f (z) = π cot πz
. We intend O
z2
to apply the Residue Theorem to the (N + 12 )i
o
integral
Z
f dz
max{| Re z|,| Im z|}=N + 21
O /
• •• • • • • • • • • • • • • • • • •
and then take the limit N → ∞. −(N + 12 ) N + 21
1 π2 π4 2π 6 5
g(z) = − z − z3 − z − ...
z 3 45 945
So, g has simple poles of residue 1 at every integer.
(ii) Then f (z) = g(z)/z 2 has simple poles for z = n, n ∈ Z \ {0}, and a triple pole at
z = 0. Thus
h · g(n + h) h · g(h) 1
Res(f, n) = lim hf (n + h) = lim = lim =
h→0 h→0 (n + h)2 h→0 (n + h)2 n2
for n 6= 0 by the periodicity of g. Moreover, the above Laurent-series for g also yields
2
Res(f, 0) = − π3 .
Now we use the M L-inequality for the above contour integral: Obviously, the length of the
contour is L = 8 · (N + 12 ). For M we observe3 :
cos(πx) sin(πx) − i·cosh(πy) sinh(πy)
(i) g(x + iy) = π cosh2 (πy) sin2 (πx)+cos2 (πx) sinh2 (πy)
(iii) we have for the “horizontal parts” of the contour (using | cos x|, | sin x| ≤ 1, sinh y <
cosh y etc.)
1 1
g x ± i N + 1
≤π 1 + cosh π N + 2
sinh π N + 2
N ≥1
≤ 1.28 π
sinh2 π N + 12
2
2 πz
To obtain the Laurent series, we observe that πz · cot(πz) = sin(πz) · cos(πz) has a removable singularity
at 0. Using the Taylor series for sine and cosine in the relationship πz · cot(πz) · sin(πz) = πz · cos(πz),
one can sucessively obain the coefficients of the Laurent-series for the cotanget.
3
Using sin(x + iy) = sin x cosh y + i cos x sinh y and cos(x + iy) = cos x cosh y − i sin x sinh y.
90
Therefore
|g(z)| 1.28 π
|f (z)| ≤ ≤
|z|2 (N + 21 )2
for z ∈ ΓN = {x + iy | max{|x|, |y|} = N + 12 }. The M L-inequality then gives
Z
1.28 π 1
f dz ≤
·8· N + →0 as N → ∞.
ΓN (N + 12 )2 2
as N → ∞, that is,
∞
X 1 π2
2
= .
k=1
k 6
π
P∞ (−1)n 2
and,
ex using the function z2 sin(πz)
, also n=1 n2
= − π12 .
Remark. (Not examinable!) One can use the functional equation of Riemann’s Zeta Func-
tion (compare p. 75) to calculate its values at the (odd) negative integers: Since
z z−1 1
ζ(z) = −2 π z sin π z Γ(−z) ζ(1 − z)
2
we have4
1 1
ζ(−1) = − 2
· ζ(2) = − .
2π 12
observe that one cannot use the above method to calculate ζ(3) = ∞ 1
P
Remark. P∞Also k=1 k3 ,
1
ζ(5) = k=1 k5 etc.; in fact, little is know about these values of Riemann’s Zeta Function.
Note that ζ(3) is also known as Apéry’s Constant, but that is almost all one knows about
it, see
http://en.wikipedia.org/wiki/Apéry’s constant
http://mathworld.wolfram.com/AperysConstant.html
Remark. The above method is, of course, not the only possibility to establish that ζ(2) =
π2
6
. It is the one usually used in lectures on complex analysis. For a list of more proofs
(the above one is listed as “Proof 9”), see
http://www.secamlocal.ex.ac.uk/people/staff/rjchapma/etc/zeta2.pdf
4
P∞ 1 1
However, note that clearly we do not have “ n=1 n−1 = 1 + 2 + 3 + 4 + 5 + . . . = − 12 ”. At z = −1, we
are talking about the analytic continuation of ζ, the definition via the infinite sum does not make sense
there (it is not convergent).
91
Remark. Riemann’ Zeta Function is closely related to properties of primes and natu-
ral numbers (“Riemann Hypothesis”), e.g., 1/ζ(2) ≈ 0.6079 is the probability that two
arbitrarily chosen natural numbers are relatively prime.
A nice interpretation of this number uses Euclid’s Orchard : Suppose you stand in “the
middle” of a forest where the trees are planted on the square lattice Z2 \ {0} (well, you
are standing on the origin). What fraction of trees do you see? You can convince yourself
that a tree/lattice point is visible from the origin iff its coordinates are relatively prime.
Thus, one can see almost 61% of all trees from the origin!
92
Bibliography
[DET] J.W. Dettman: Applied Complex Variables; Dover, NY (1984); library: 513.317
DET
[ST] I. Stewart & D. Tall: Complex Analysis (The Hitchhiker’s Guide to the Plane);
Cambridge UP, Cambridge (1985); library: 513.317 STE
Good wi:
Bernd Sin
93
Contents
Manifesto 2
V. Residue Calculus 80
V.1. Isolated Singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
V.2. The Residue Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
V.3. Evaluation of Real Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . 87
V.4. Evaluation of Infinite Sums . . . . . . . . . . . . . . . . . . . . . . . . . . 89
Bibliography 93
94