Are you sure?
This action might not be possible to undo. Are you sure you want to continue?
on Complex Analysis
Scribed by Oleg Ivrii during Spring 2007
Preface
These notes of my lectures on Complex Analysis at the University of Toronto were written
by Oleg Ivrii on his own initiative, after he took my course in the spring, 2007. The course
is crosslisted as MAT1001S, one of the Core Courses in our Graduate Program, and as
MAT454S, a fourthyear undergraduate course which is the second part of a twoterm
sequence in complex analysis in our Specialist Program in Mathematics. Oleg took this
course as a secondyear undergraduate.
My lectures were based on the classical textbooks of Lars Ahlfors (Complex Analy
sis, Third Edition, McGrawHill 1979) and Henri Cartan (Elementary Theory of Analytic
Functions of One or Several Variables, Dover 1994). My exposition closely follows these
texts. Oleg has included the problems that I assigned as homework; most of these can be
found in Ahlfors, Cartan or other textbooks. Oleg has added a concluding chapter as an
introduction to several fundamental further results. I have included some of these topics
when giving the course on earlier occasions, and they improve my lectures as presented
here!
During my thirtyﬁve years of teaching at the University of Toronto, I have been truly
fortunate to have had so many inspiring students like Oleg. They are passionate about
mathematics and they care for each other and the world around them. I am grateful to
Oleg that he thought my lectures were interesting enough to want to write them up, and
I give him my warmest wishes for a very happy and fruitful mathematical career.
March 2008 Edward Bierstone
2
Contents
1 Space of Holomorphic Functions 5
1.1 Topology on the Space of Holomorphic Functions . . . . . . . . . . . . . . . 5
1.2 Series of Meromorphic Functions . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 The Weierstrass pfunction . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 The Weierstrass pfunction II . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.5 Complex Projective Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.6 The Elliptic Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.7 Implicit Function Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2 Interpolation 16
2.1 Meromorphic functions with prescribed singularities . . . . . . . . . . . . . 16
2.2 Inﬁnite products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.3 Entire functions with prescribed zeros . . . . . . . . . . . . . . . . . . . . . 19
2.4 The Gamma Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.5 Normal families and compact subsets of C(Ω) . . . . . . . . . . . . . . . . . 22
3 Automorphisms 25
3.1 Local Geometry of Holomorphic functions . . . . . . . . . . . . . . . . . . . 25
3.2 Automorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.3 Riemann Mapping Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.4 Riemann Mapping Theorem II . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.5 ChristoﬀelSchwarz Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3
4 Complex Geometry 33
4.1 Complex Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.2 Complex Manifolds II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.3 Examples of Riemann Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.4 Example: y =
_
P(x) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.5 Abel’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.6 Abel’s Theorem II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
5 A Modular Function 42
5.1 Analytic Continuation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
5.2 Monodromy Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.3 Modular Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.4 Modular Function II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
5.5 Little Picard Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
6 Problems 49
7 Further Results 55
7.1 Big Picard Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
7.2 Runge’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
7.3 Prescribing Zeros in Arbitrary Domains . . . . . . . . . . . . . . . . . . . . 59
7.4 Prescribing Singularities in Arbitrary Domains . . . . . . . . . . . . . . . . 60
4
Chapter 1
Space of Holomorphic Functions
1.1 Topology on the Space of Holomorphic Functions
Suppose Ω is open in C, let C(Ω) be the ring of continuous C–valued functions on Ω and
let H(Ω) be the subring of holomorphic functions.
We can topologize C(Ω): a sequence of functions {f
n
} converges uniformly on compact
sets if for every compact set K ⊂ Ω, {f
n

K
} converges uniformly.
Uniform convergence on compact sets deﬁnes the socalled compactopen topology.
The fundamental system of open neighbourhoods at 0 is
V (K, ) = {f : f(z) < , z ∈ K}.
This is because f
n
→f uniformly on compact sets if and only if f −f
n
lies in V (K, ) for
n large enough for any given K, .
In fact, C(Ω) is metrizable (even by a translationinvariant metric).
Write Ω =
i
K
i
and let d(f) =
i
1
2
i
· min(1, M
i
(f)) where M
i
(f) = max
z∈K
i
f(z).
C(Ω) is complete: if {f
n
} is a sequence of continuous functions which converge uniformly
on compact sets, f = limf
n
is continuous.
We give H(Ω) the subspace topology.
Theorem. (1). H(Ω) is a closed subspace. (2). Furthemore, the map f → f
is continu
ous.
5
(1) means that if {f
n
} ⊂ H(Ω) converges uniformly on compact sets, f = limf
n
∈
H(Ω). (2) means that if {f
n
} ⊂ H(Ω) converges uniformly on compact sets and if f =
limf
n
, then {f
n
} converges uniformly on compact sets to f
.
Proof of (1). We have to show that f(z)dz is closed. By Morera’s theorem, this is
equivalent to showing that
_
γ
f(z)dz = 0 for any small loop γ. As f
n
(z) are holomorphic,
_
γ
f
n
(z)dz = 0 holds for all n. Then,
_
γ
f(z) = lim
n→∞
f
n
(z)dz because f
n
converges to f
uniformly on the image of γ.
Proof of (2). It is enough to show that f
n
converges to f
uniformly on a closed disk
D(z, r) in Ω. Let γ be a positively oriented loop around a disk of slightly larger radius.
By Cauchy’s integral formula,
lim
n→∞
f
n
(z) = lim
n→∞
1
2πi
_
γ
f
n
(ζ)
(ζ −z)
2
dζ =
1
2πi
_
γ
f(ζ)
(ζ −z)
2
dζ = f
(z).
The limit above converges uniformly as ζ −z is bounded away from 0.
Corollary. If a series of holomorphic functions
∞
n=1
f
n
(z) converges uniformly on com
pact sets of Ω, the sum is holomorphic and can be diﬀerentiated termbyterm.
Proposition (Hurwitz). If Ω is a domain (i.e connected), if {f
n
} ⊂ H(Ω) converges
uniformly on compact sets and each f
n
vanishes nowhere in Ω, then f = limf
n
is never 0
or identically 0.
Suppose f was not identically 0. Then its zeros are isolated. In particular around any
point z, there is a small loop γ around z for which f does not vanish on the image of γ.
But as the image of γ is compact, f is bounded away from 0. Then,
lim
n→∞
1
2πi
_
γ
f
n
(z)
f
n
(z)
dz =
1
2πi
_
γ
f
(z)
f(z)
dz.
The left hand side counts the number of zeros of f
n
enclosed within γ, which is 0. Hence,
f(z) also does not possess any zeros inside γ, i.e f(z) = 0. As z was arbitrary, f(z) vanishes
nowhere in Ω.
Proposition. If Ω is a domain and {f
n
} ⊂ H(Ω) are onetoone and converges uniformly
on compact sets, then f = limf
n
is either onetoone or a constant.
Suppose f was not identically a constant, but nevertheless f(z
0
) = f(z
1
) = a. Choose
disjoint neighbourhoods U and V of z
0
and z
1
respectively. Then f(z) − a vanishes at a
6
point in U, so we can extract a subsequence of {f
n
} for which f
n
(z) −a vanishes at a point
in U. From this subsequence, we extract another subsequence for which f
n
(z) −a vanishes
at a point in V . But then, f
n
(z) would fail to be injective.
1.2 Series of Meromorphic Functions
A series of meromorphic functions
∞
n=0
f
n
on an open set Ω ⊂ C converges uniformly on
compact subsets of Ω if for every compact K ⊂ Ω, all but ﬁnitely many terms have no poles
in K and form a uniform convergent series.
We deﬁne the sum on any relatively compact (precompact) open U as
n<p
f
n
+
n≥p
f
n
with each function in the second sum having no pole in U.
Theorem. If
∞
n=0
f
n
is a series of meromorphic functions that converges on compact
subsets of Ω, then the sum f is a meromorphic function on Ω and
∞
n=0
f
n
converges
uniformly on comapct subsets of U.
The poles of f, P(f) are contained are in
n
P(f
n
).
Example 1
The series f(z) =
∞
n=−∞
1
(z−n)
2
converges uniformly and absolutely on compact subsets of
C. In fact the series converges uniformly and absolutely in each vertical strip x
0
< x < x
1
(we write z = x +iy).
To see this, note that in
n<x
0
1
(z−n)
2
, each term is bounded by
1
(x
0
−n)
; while in
n>x
1
1
(z−n)
2
, each term is bounded by
1
(x
1
−n)
2
.
The function f(z) is meromorphic on C and enjoys three properties: (1) it is periodic,
f(z + 1) = f(z), (2) it has poles z = n, each with principal part
1
(z−n)
2
and (3) f(z) → 0
as y → ∞ uniformly with repect to x, i.e for all > 0, there is a b such that f(z) <
whenever y > b.
The function g(z) =
_
π
sin πz
_
2
also enjoys these properties. The ﬁrst property is obvious.
To check the second property, we ﬁrst check that sin πz vanishes only on the integers.
By periodicity, it suﬃces to check that the principal part of g(z) at the origin is
1
z
2
; for a
proof, see the calculation below.
7
Easiest way to see property (3) is by means of the identity  sin πz
2
= sin
2
πx+sinh πy,
as y goes to ∞, sin πz →∞ and hence g(z) →0.
Hence, f − g is holomorphic in C because f and g have the same poles and same
principal parts; f −g is bounded, so f −g is a constant by Liouville’s theorem, and by the
third property, the constant is 0.
We derive the formula
∞
n=1
1
n
2
=
π
2
6
.
Consider the function
_
π
sin πz
_
2
−
1
z
2
=
n=0
1
(z−n)
2
. It is holomorphic in a neighbour
hood of 0 and as z →0, it tends to
π
2
(πz −
1
6
π
3
z
3
+. . . )
2
−
1
z
2
=
1
z
2
(1 −
1
3
πz
2
+. . . )
−
1
z
2
=
π
2
3
.
But when z = 0, the function is 2
∞
n=1
1
n
2
.
Example 2
The series f(z) =
1
z
+
n=0
_
1
z−n
+
1
n
_
=
1
z
+
n=0
z
n(z−n)
converges uniformly and
absolutely on compact subsets of C by comparison with
n=0
1
n
2
.
Thus f(z) is a meromorphic function with poles z = n and principal parts
1
z−n
(i.e
simple poles with residue 1).
We notice:
f
(z) = −
1
z
2
−
n=0
1
(z −n)
2
= −
_
π
sin πz
_
2
=
d
dz
(π cot πz).
Hence, f(z) − π · cot πz is constant, which is 0 as both functions are odd (to see that
f(z) is odd, group terms with +n and −n). So,
1
z
+
∞
n=1
2z
z
2
−n
2
= π cot πz.
1.3 The Weierstrass pfunction
Let Γ be a discrete subgroup of C with generators e
1
, e
2
linearly independent over R, i.e
Γ = {n
1
e
1
+n
2
e
2
: n
1
, n
2
∈ Z}.
Vectors e
1
, e
2
are another set of generators for Γ if and only if e
1
, e
2
are are linear
combination of e
1
, e
2
with determinant of matrix of coeﬃcients invertible in the ring of
integers, i.e ±1.
8
A function f(z) has Γ as a group of periods if f(z+n
1
e
1
+n
2
e
2
) = f(z) for all n
1
, n
2
∈ Z.
We deﬁne
p(z) =
1
z
2
+
w∈Γ,w=0
_
1
(z −w)
2
−
1
w
2
_
.
Note that the Weierstrass p function depends on the choice of Γ.
We check that it converges uniformly and absolutely on compact subsets, for this we
estimate:
¸
¸
¸
1
(z −w)
2
−
1
w
2
¸
¸
¸ =
¸
¸
¸
2zw −z
2
w
2
(z −w)
¸
¸
¸ = . . .
We can restrict our attention to the compact subset z ≤ r. As ﬁnitely many terms don’t
matter, we only need to estimate when w is suﬃciently large, say w ≥ 2r. We continue:
· · · =
¸
¸
¸
z(2 −
z
w
)
w
3
(1 −
z
w
)
2
¸
¸
¸ ≤
¸
¸
¸
r ·
5
2
w
3
·
1
4
¸
¸
¸ =
10r
w
3
.
We want to show that
w∈Γ,w=0
1
w
3
< ∞. We deﬁne
P
n
= {n
1
e
1
+n
2
e
2
: max(n
1
, n
2
)}.
We let k be the distance between the origin to the closest point of P
1
. As the number of
points in P
n
is 8n, we see that
w∈Γ,w=0
1
w
3
=
∞
n=1
w∈Pn
1
w
3
<
∞
n=1
8n
k
3
n
3
=
8
k
3
∞
n=1
1
n
2
< ∞.
We have thus shown that p(z) is a meromorphic function with pole
1
(z−w)
2
at w ∈ Γ. Also,
p(z) is even.
The derivative p
(z) = −2
w∈Γ
1
(z−w)
3
is doubly periodic. To show that p(z+e
i
) = p(z)
for i = 1, 2, we consider p
(z +e
i
) −p
(z) = 0, so p(z +e
i
) −p(z) is constant.
But as p(z) is even, plugging in z = −
e
i
2
, we see that the constant is 0, so p(z) also has
Γ as a group of periods.
As p(z) −
1
z
2
= g(z) is a holomorphic even function in a neighbourhood of the origin
with g(0) = 0, the Laurent expansion at 0 of p(z) is
1
z
2
+a
2
z
2
+a
4
z
4
+. . .
The 2nth derivative of
1
(z−w)
2
is (2n+1)!
1
(z−w)
2n+2
. Hence a
2n
= (2n+1)
w=0
1
w
2n+2
,
in particular a
2
= 3
1
w
4
, a
4
= 5
1
w
6
.
9
We now ﬁnd a diﬀerential equation for p
(z):
p(z) =
1
z
2
+a
2
z
2
+a
4
z
4
+. . .
p
(z) = −
2
z
3
+ 2a
2
z + 4a
4
z
3
+. . .
p
(z)
2
=
4
z
6
−
8a
2
z
2
−16a
4
+. . .
p(z)
3
=
1
z
6
+
3a
2
z
2
+ 3a
4
+. . .
So, p
(z)
2
−4p(z)
3
= −
20a
2
z
2
−28a
4
+. . .
Hence, p
(z)
2
− 4p(z)
3
+ 20a
2
p(z) + 28a
4
is a holomorphic function vanishing at the
origin. As it is periodic, it is bounded and by Liouville’s theorem is constant, so it must
be identically 0.
Thus, (x, y) = (p(z), p
(z)) parametrizes the algebraic equation
y
2
= 4x
3
−20a
2
x −28a
4
.
1.4 The Weierstrass pfunction II
Suppose Γ is a discrete subgroup of C given by vectors e
1
, e
2
(linearly dependent over R).
Let X ⊂ C
2
be the collection of pairs (x, y) satisfying the polynomial equation
y
2
= 4x
3
−20a
2
x −28a
4
with a
2
= 3
1
w
4
and a
4
= 5
1
w
6
. The goal of this lecture it to show that the RHS has
three distinct roots and any (x, y) ∈ X is given by x = p(z), y = p
(z) for a unique C/Γ.
Proposition. If f(z) is a nonconstant meromorphic function having Γ as a group of
periods, then the number of zeros of f(z) in a period parallelogram is equal to the number
of poles (counting with multiplicity).
Let γ be the positively oriented loop around the period parallelogram. The num
ber of zeros minus the number of poles enclosed with in γ is calculated by the integral
1
2πi
_
γ
f
(z)
f(z)
dz. However, as γ is doublyperiodic, the opposite sides of γ cancel each other
and the integral reduces to 0.
Proposition. Suppose f(z) is above. Let its zeros inside the fundamental parallelogram
be {α
i
} and its poles be {β
j
}. Then,
i
α
i
=
j
β
j
(mod Γ).
10
The value
i
α
i
−
j
β
j
is counted by the integral
1
2πi
_
γ
zf
(z)
f(z)
dz. If γ
1
and γ
2
are two
sides of γ along the vectors e
1
and e
2
starting at the origin respectively, then
i
α
i
−
j
β
j
=
1
2πi
_
γ
zf
(z)
f(z)
dz = −
1
2πi
_
γ
1
e
1
f
(z)
f(z)
dz +
1
2πi
_
γ
2
e
2
f
(z)
f(z)
dz.
However,
1
2πi
f
(z)
f(z)
dz is an integer because it is a diﬀerence between two determinations of
log f(z) with same value of f(z).
Remark. The integral formulae used above may be obtained from the Residue theorem as
follows: suppose a
i
is a zero; we write f(z) = (z−α
i
)
k
(c+. . . ), f
(z) = k(z−α
i
)
k−1
(c+. . . )
and z = α
i
+ (z −α
i
). Then
f
(z)
f(z)
=
k
z −α
i
(1 +. . . ),
zf
(z)
f(z)
=
k
z −α
i
(1 +. . . )(α
i
+. . . ).
The considerations for poles {β
j
} are analogous.
The zeros of p
(z) are
e
1
2
,
e
2
2
and
e
1
+e
2
2
(the points z such that 2z ∈ Γ but z / ∈ Γ). The
fact that they actually are zeros follows from p
(z) being doublyperiodic and odd. But as
p
(z) has a unique pole of order 3 inside a period parallelogram, these zeros are simple and
the only ones there.
Now we examine the zeros of p(z) − a (i.e roots of p(z) = a). For any value of a ∈ C,
there are exactly two roots in a period parallelogram. Write a = p(z
0
). Clearly, z
0
/ ∈ Γ as
Γ is the set of poles of p. We have two possibilities:
(1) If 2z
0
∈ Γ, then p(z) −p(z
0
) has a unique zero of order 2 as p
(z
0
) vanishes.
(2) If however 2z
0
∈ Γ, then p(z
0
) = p(−z
0
) are the two noncongruent zeros.
With these preparations, we can return back to our equation y
2
= 4x
3
−20a
2
x −28a
4
.
If y = 0, we may choose x = p(
e
1
2
), p(
e
2
2
), p(
e
1
+e
2
2
). These are distinct as the values they
take are unique by (1).
If (x, y) ∈ X, but y = 0, then x = p(z), y = p
(z) and x = p(−z), y = p
(−z) are the
two points which map to the same value of x by (2), but they map to diﬀerent values of y
as −p
(z) = p
(−z).
11
1.5 Complex Projective Space
The set X ⊂ C
2
: y
2
= 4x
3
− 20a
2
x − 28a
4
is a smooth curve as it is locally a graph
of a function y = f(x) or x = g(y). In other words, X is a onedimensional complex
submanifold.
If (x
0
, y
0
) ∈ X, y
0
= 0, then y = f(x) near (x
0
, y
0
), i.e x is a local coordinate.
On the other hand, if y
0
= 0, then p
(x
0
) = 0 since p(x) has 3 distinct roots, so X = g(y)
near (x
0
, y
0
), i.e y is a local coordinate.
Remark. These statements follow from the implicit function theorem, but in classical form,
the implicit function theorem only gives that locally y = f(x) and x = g(y) are diﬀerentiable
as functions of two real variables, whereas we want to imply that the mappings are actually
holomorphic. We return to this subtle point in a later lecture.
The ndimensional complex projective space P
n
(C) is deﬁned to be C
n+1
\{0} modulo
the equivalence relation (x
0
, . . . x
n
) ∼ (x
0
, . . . x
n
) if (x
0
, . . . x
n
) = (λx
0
, . . . λx
n
) for some
λ = 0 in C.
We write [x
0
, . . . x
n
] for the equivalence class of (x
0
, . . . x
n
). These are called homoge
nous coordinates for we are using n+1 “coordinates” to represent an ndimensional object.
The complex projective space may be covered by coordinate charts
U
i
= {[x
0
, . . . x
n
] ∈ P
n
(C) : x
i
= 0}.
Each chart U
i
is homeomorphic to C
n
by map [x
0
, . . . x
n
] →(
x
0
x
i
, . . .
ˆ x
i
x
i
, . . .
xn
x
i
) where the hat
notation as usual means that the variable is omitted. The inverse map is (y
1
, y
2
, . . . y
n
) →
[y
1
, . . . y
i−1
, 1, y
i
, . . . y
n
].
We thus get a complex (algebraic) manifold structure (actually, we have to check that
the transition maps are analytic, they even turn out to be rational, this is left as an
exercise).
We can think of P
n
(C) as the chart U
0
∼
= C
n
together with the set {x
0
= 0}
∼
= P
n−1
(C)
which is “the hyperplane at inﬁnity”.
It is easy to see that P
1
(C) is just the Riemann sphere S
2
. Right now, we are interested
in P
2
(C) = [x, y, t]. We compactify X ⊂ C
2
to X
⊂ P
2
(C) by rewiting y
2
= 4x
3
−20a
2
x−
12
28a
4
in homogeneous coordinates
_
y
t
_
2
= 4
_
x
t
_
3
−20a
2
_
x
t
_
−28a
4
and allowing t to be 0. The equation simpliﬁes to y
2
t = 4x
3
−20a
2
xt
2
−28a
4
t
3
.
Thus, X
is X together with an added point at inﬁnity: if t = 0, the equation reduces
to 4x
3
= 0 from which we see x = 0. The variable y can be any complex number with the
exception of 0, but all these options are equivalent to the point [0, 1, 0].
1.6 The Elliptic Integral
Let Γ ⊂ C be a discrete subgroup and X ⊂ C
2
be the set of points y
2
= 4x
3
−20a
2
x−28a
4
.
Recall that X
⊂ P
2
(C) is the set of points y
2
t = 4x
3
− 20a
2
xt
2
− 28a
4
t
3
, in which X
naturally embeds by (x, y) →[x, y, 1]. Then, X
= X ∪ [0, 1, 0].
The point [0, 1, 0] lies in the chart {[x, y, t] : y = 0}. Let (x
, t
) = (
x
y
,
t
y
) be the aﬃne
coordinates in this chart. In these coordinates, the equation of X
is
t
= 4x
3
−20a
2
x
t
2
−28a
4
t
3
.
The implicit function theorem tells us that in some neighbourhood of (x
, t
) = (0, 0), the
point at inﬁnity, t
is a holomorphic function of x
.
If we let t
= c
0
+c
1
x
+c
2
x
2
+. . . then t
= 4x
3
−320a
2
x
7
+. . .
Indeed, c
0
= 0 as t
(0) = 0. By comparing coeﬃcients, we see that c
1
= c
2
= 0 as
well and c
3
= 4x
3
. The next nonzero coeﬃcient comes from the term −20a
2
x
t
2
and is
−20a
2
x
(4x
3
)
2
= −320a
2
x
7
.
In a neighbourhood of the point at inﬁnity, we can take x
as a local coordinate; this
concludes the proof that X
has complex manifold structure.
In Section 1.4, we have seen that the map C/Γ →X
given by
z → [p(z), p
(z), 1]
. ¸¸ .
in usual coordinates
=
_
p(z)
p
(z)
, 1,
1
p
(z)
_
. ¸¸ .
in coords at inﬁnity
.
is a bijection of sets, but more can be said. The map is holomorphic: if z = 0, the coordinate
functions p(z), p
(z), 1 are holomorphic; while near z = 0, the functions
p(z)
p
(z)
, 1,
1
p
(z)
are
holomorphic.
13
Thus C/Γ →X
is a holomorphic bijection from a compact space and a Hausdorﬀ space;
so in particular, a homeomorphism. In fact, the inverse function theorem guarantees us
that the inverse is also holomorphic.
From the parametrization x = p(z), y = p
(z), we see that dx = ydz; so dz =
dx
y
whenever y = 0. Diﬀerentiating the equation y
2
= 4x
3
− 20a
2
x − 28a
4
, we ﬁnd that
(12x
2
−20a
2
)dx = 2ydy so
dx
y
=
dy
6x
2
−10a
2
.
By “dz” we mean the holomorphic diﬀerential form on X
which is
dz =
dx
√
4x
3
−20a
2
x −28a
4
on y = 0 and
dy
6x
2
−10a
2
on y
= 0.
Remark. Thus dz is welldeﬁned on all of X
: functions y(x) and y
(x) cannot vanish
simultaneously, otherwise y(x) would have multiple roots but we have seen that it is not the
case.
Inverse deﬁnes z as a multivalued holomorphic function on X
where 2 branches diﬀer
by a constant in Γ:
z =
_
z
0
dz
. ¸¸ .
in C
=
_
p(z)
[0,1,0]
dx
√
4x
3
−20a
2
x −28a
4
.
The form dz is closed but not exact, so a particular value of z not only determines p(z)
but also the choice of path of integration from [0, 1, 0] to p(z).
We say that the Weierstrass p function is given by “inversion” of the elliptic integral.
Trigonometric Functions
For convenience, we repeat our considerations with X ⊂ C
2
given by x
2
+y
2
= 1. We can
parametrize X by (x, y) = (cos θ, sin θ).
We can diﬀerentiate the equation x
2
+ y
2
= 1 to obtain xdx + ydy = 0; so,
dy
x
= −
dx
y
.
But, we can also diﬀerentiate the parametrization: cos θdθ = d(sin θ) or xdθ = dy to obtain
dθ =
dy
x
=
dy
√
1−y
2
.
Thus, by “dθ”, we mean the form which is −
dx
y
when y = 0 and
dy
√
1−y
2
when y = ±1.
We then invert the integral
_
dy
x
=
_
dy
√
1−y
2
by deﬁning sin θ by the formula
θ =
_
(cos θ,sin θ)
(1,0)
dy
x
=
_
sin θ
0
dy
_
1 −y
2
.
14
The form dθ is closed but not exact, so θ is a multivalued function with branches whose dif
ference is a multiple of 2π. Put another way, a particular value of θ determines (cos θ, sin θ)
as well as the winding number of the curve from (1, 0) to (cos θ, sin θ).
1.7 Implicit Function Theorem
We say that a function f(z
1
, z
n
, . . . z
n
) in an open set Ω ⊂ C
n
is holomorphic if f ∈ C
1
and
df =
n
i=1
∂f
∂z
i
dz
i
. This is equivalent to:
(1) f is continuous and holomorphic in each variable separately,
(2) f is analytic (locally represented by convergent power series).
Actually, it is even suﬃcient just to say “holomorphic in each variable separately,” but
showing that this condition is equivalent to the previous conditions is rather diﬃcult.
We discuss the implicit function theorem for 2 variables: suppose f(x, y) is a holomor
phic function with
∂f
∂y
(x
0
, y
0
) = 0 and we wish to solve the equations z = f(x, y), z = f(x, y)
for y, y near (x
0
, y
0
).
Write x = x
1
+ ix
2
, y = y
1
+ iy
2
and z = z
1
+ iz
2
. A simple calculation shows that
dx ∧ dx = −2idx
1
∧ dx
2
(we will use this equation for variables y and z in place of x).
For a ﬁxed x, dz =
∂f
∂y
dy and dz =
∂f
∂y
dy =
∂f
∂y
dy, so dz ∧ dz =
¸
¸
¸
∂f
∂y
¸
¸
¸
2
dy ∧ dy. By the
above equation, this is the same as dz
1
∧dz
2
=
¸
¸
¸
∂f
∂y
¸
¸
¸
2
dy
1
∧dy
2
, i.e the Jacobian determinant
det
∂(z
1
, z
2
)
∂(y
1
, y
2
)
=
¸
¸
¸
∂f
∂y
¸
¸
¸
2
> 0.
By the implicit function theorem (for realvalued functions), we see that we can solve
y
1
, y
2
as C
1
functions in terms of z
1
, z
2
; x
1
, x
2
near (x
0
, y
0
, 0).
The equation dz =
∂f
∂x
dx +
∂f
∂y
dy shows that that dy is a linear combination of dx, dz
with holomorphic coeﬃcients (here we are using that
∂f
∂y
= 0). This means the solution is
holomorphic.
15
Chapter 2
Interpolation
2.1 Meromorphic functions with prescribed singularities
We now shift gears to constructing meromorphic functions with prescribed singularities,
i.e with given poles and their singular parts.
Over the Riemann sphere, the only meromorphic functions are rational functions; but
over the complex plane, we have a wealth of meromorphic functions such as tan z or sec z
but inﬁnity is a limit of poles.
Theorem (MittagLeﬄer). Suppose {b
k
} ∈ C are given and distinct with lim
k→∞
b
k
= ∞
and P
k
(z) be polynomials without constant term. Then there is a meromorphic function
in C with poles b
k
and principal parts P
k
(
1
z−b
k
). The most general meromorphic function
with these poles and primitives parts is
f(z) =
∞
k=1
_
P
k
_
1
z −b
k
_
−p
k
(z)
_
. ¸¸ .
♦
+g(z)
where p
k
(z) are polynomials chosen such that the series convergences and g(z) is an entire
function.
We can assume that no b
k
= 0. The function P
k
(
1
z−b
k
) is holmorphic in z < b
k
so we
can expand it as a convergent power series in that disk.
Let p
k
(z) be the sum of the terms of order up to m
k
where m
k
is chosen such that
¸
¸
¸P
k
_
1
z −b
k
_
−p
k
(z)
¸
¸
¸ ≤
1
2
k
on z ≤
b
k

2
16
We show ♦ converges uniformly and absolutely on z ≤ r for any r. Chosen n such that
b
k
 ≥ 2r if k > n. It suﬃces to compare
∞
k=n+1
P
k
_
1
z −b
k
_
−p
k
(z)
with
1
2
k
. Thus ♦ is meromorphic in C with poles {b
k
} and principle parts P
k
(
1
z−b
k
).
Given any other function with prescribed singularities, the diﬀerence of it and ♦ is holo
morphic and conversely, given any holomorphic function g(z), the function ♦ + g(z) also
has the desired singularities.
2.2 Inﬁnite products
We say that
∞
n=1
b
n
converges to p ∈ C if no b
n
are 0 and the partial products p
n
=
b
1
b
2
. . . b
n
have a nonzero limit p. Actually, this is too restrictive: we should allow a ﬁnite
number of terms to be 0, but after removing these few terms, the partial products should
have a nonzero limit.
A necessary condition for convergence is b
n
=
pn
p
n−1
→ 1. We will instead write the
product as (P)
∞
n=1
(1 +a
n
), so the condition requires that a
n
→0. To the product, we
shall associate the sum (S)
∞
n=1
log(1 + a
n
) where we take the principal branch of log,
deﬁned on the complement of the negative real ray with log 1 = 0.
Proposition. The product (P) and the sum (S) converge or diverge simultaneously. Just
as p
n
is the nth partial product of (P), we will write s
n
for the nth partial sum of (S).
If s
n
converges, then so does p
n
as p
n
= e
sn
. Conversely, suppose p
n
converges to
some p = 0. Fix a determination of log p = log p + i · arg p. Choose determinations of
log p
n
= log p
n
 +i · arg p
n
such that arg p −π < arg p
n
≤ arg p +π.
Then s
n
= log p
n
+ 2πi · k
n
for integers k
n
. For s
n
to converge, we need to know that
k
n
stabilize. Indeed,
2πi(k
n+1
−k
n
) = log(1 +a
n+1
) −log p
n+1
+ log p
n
,
= log(1 +a
n+1
) −(log p
n+1
−log p) + (log p
n
−log p).
The three terms log(1 + a
n+1
), log p
n+1
− log p, log p
n
− log p are small when n is large.
As the left hand side can take upon a discrete set of values, it must be 0.
17
We say that the product (P) converges absolutely if (S) does, but the latter is equivalent
to (S
)
∞
n=1
a
n
converging absolutely as lim
z→0
log(1+z)
z
= 1.
Now, suppose that f
n
(z) are continuous complexvalued functions deﬁned on an open
set Ω ⊂ C. We say that the product
∞
n=1
f
n
(z) converges uniformly (or absolutely) on a
compact set K ⊂ Ω if
log f
n
(z) does.
Theorem. Suppose f
n
(z) are holomorphic in Ω and
∞
n=1
f
n
(z) converges uniformly and
absolutely on compact subsets of Ω. Then f(z) =
∞
n=1
f
n
(z) is holomorphic on Ω and
f = f
1
f
2
. . . f
q
n>q
f
n
(z).
If Z(f) denotes the zeros of function f, then Z(f) =
Z(f
n
). Additionally, the
multiplicity of a zero of f is the sum of that of f
n
.
Corollary. Under the same hypothesis, the series of meromorphic functions
∞
n=1
f
n
fn
con
verges uniformly on compact subsets of Ω and limit is
f
f
.
Let U be a relatively compact subset of Ω and g
q
(z) = exp
_
n>q
log f
n
(z)
_
. By the
product rule:
f
(z)
f(z)
=
q
n=1
f
n
(z)
f
n
(z)
+
g
q
(z)
g
q
(z)
=
q
n=1
f
n
(z)
f
n
(z)
+
n>q
f
n
(z)
f
n
(z)
.
The latter equality is justiﬁed as log
n>q
log f
n
converges uniformly to a (branch of) log g
q
and thus can be diﬀerentiated termbyterm (take q large enough so that f
n
have no zeros
in U for n > q).
Example
We develop an inﬁnite product for sin πz. The product f(z) = z
∞
n=1
(1 −
z
2
n
2
) converges
uniformly and absolutely on compact subsets of C (because
z
2
n
2
does).
Diﬀerentiating logarithmically, we ﬁnd that
1
z
+
∞
n=1
−
2z
n
2
1 −
z
2
n
2
=
1
z
+
2z
z
2
−n
2
= π cot πz =
sin
πz
sin πz
by Example 2 in Section 1.2. It follows that f(z) = c sin πz. But as lim
z→0
f(z)
z
= 1 and
lim
z→0
sin πz
z
= π, we see that c =
1
π
. Hence, sin πz = πz
∞
n=1
(1 −
z
2
n
2
).
18
2.3 Entire functions with prescribed zeros
An entire function f(z) with no zeros is of the form e
g(z)
for some entire function g(z).
Indeed, the form
f
(z)
f(z)
dz counts the number of zeros, but since there aren’t any, it is exact.
So, it is the derivative of an entire function g(z). Then
d
dz
_
f(z) · e
−g(z)
_
= f
(z) · e
−g(z)
−f(z) ·
f
(z)
f(z)
e
−g(z)
= 0,
so f(z) = const · e
−g(z)
but the constant can be absorbed into g(z).
An entire function with ﬁnitely many zeros {a
k
} (these not be necessarily distinct) can
be written as
f(z) = z
m
e
g(z)
n
k=1
_
1 −
z
a
k
_
.
Here, we put the zeros at z = 0 separately in the factor z
m
. Now, we examine the case
when an entire function has inﬁnitely many zeros.
Theorem (Weierstrass). Suppose {a
k
} ∈ C, lim
k→∞
a
k
= ∞, a
k
not necessarily distinct.
There exists an entire function with prescisely {a
k
} as zeros. Any such function is of the
form
f(z) = z
m
e
g(z)
∞
k=1
_
_
1 −
z
a
k
_
· p
k
(z)
_
.
Here, p
k
(z) are factors which guarantee that the product converges. The product
converges if the sum of logarithms of its terms converge, so we look at the expansion
log
_
1 −
z
a
k
_
= −
z
a
k
−
1
2
·
z
2
a
2
k
−
1
3
·
z
3
a
3
k
−. . .
Thus, a good form for
p
k
(z) = exp
_
z
a
k
+
1
2
·
z
2
a
2
k
+· · · +
1
m
k
·
z
m
k
a
m
k
k
_
.
We need to show we can choose integers m
k
to make that the product converge, i.e we
need to make
k
_
log
_
1 −
z
a
k
_
+
z
a
k
+
1
2
·
z
2
a
2
k
+· · · +
1
m
k
·
z
m
k
a
m
k
k
_
converge. We denote the
general term of this sum by g
k
(z).
Actually, we would need that −π < Img
k
(z) < π, so we know that we took the principal
branch of logarithm and not some other branch (this would be automatic because we need
g
k
to tend to 0 for convergence).
19
Suppose r > 0, for z ≤ r, we consider terms with a
k
 > r:
g
k
(z) = −
1
m
k
+ 1
_
z
a
k
_
m
k
+1
−
1
m
k
+ 2
_
z
a
k
_
m
k
+2
−. . .
Thus,
g
k
(z) ≤
1
m
k
+ 1
_
r
a
k

_
m
k
+1
_
1 −
r
a
k

_
−1
It suﬃces to choose {m
k
} so that
(
r
a
k

)
m
k
+1
converges for every r > 0 (note; however,
that the sum has less terms with greater r); then the product converges uniformly and
absolutely in z ≤ r. Clearly, m
k
= k suﬃces.
Corollary. Every meromorphic function F(z) deﬁned on the entire complex plane is a
quotient of entire functions.
Indeed, let g(z) be the entire function with the poles of F(z) as its zeroes. Then,
F(z)g(z) is an entire function f(z), so F(z) =
f(z)
g(z)
.
Corollary. Suppose a
k
∈ C, lim
k→∞
a
k
= ∞, b
k
∈ C, m
k
∈ N. Then there exists an entire
function f(z) such that a
k
is a root of order m
k
of f(z) = b
k
.
Let g(z) be an entire function with zero of order m
k
at a
k
. Write:
f(z) = g(z)h(z) = b
k
+g(z) ·
_
h(z) −
b
k
g(z)
_
Now choose h(z) to be a meromorphic function with poles {a
k
} such that at each a
k
, the
principal part of h(z) was the principal part of
b
k
g(z)
.
At z = a
k
, g(z) has a zero of order m
k
, h(z) −
b
k
g(z)
is holomorphic which shows that
f(z) = b
k
has a root of order at least m
k
at a
k
.
To make sure that f(z) = b
k
has a root of order exactly m
k
at a
k
, we make two changes:
(1) we change the order of g(z) at a
k
from m
k
to m
k
+ 1 and (2) we change the principal
part of h(z) at a
k
to be
b
k
g(z)
+
1
z−a
k
.
20
2.4 The Gamma Function
Consider the inﬁnite product:
g(z) = z(1 +z)
. ¸¸ .
f
1
(z)
∞
n=2
_
_
1 +
z
n
__
1 −
1
n
_
z
. ¸¸ .
fn(z)
_
.
We want to show it converges uniformly and absolutely on compact subsets of C. For this
purpose, we examine the sum of
log f
n
(z) = log
_
1 +
z
n
_
+z log
_
1 −
1
n
_
.
=
_
z
n
−
z
2
2n
2
+
z
3
3n
3
−. . .
_
+z(−
1
n
−
1
2n
2
−
1
3n
3
−. . .
_
If z ≤ r and n > 2r, we have the estimate
¸
¸
¸log f
n
(z)
¸
¸
¸ ≤ 2 ·
r
2
n
2
·
_
1 −
r
n
_
−1
< 4r
2
·
1
n
2
.
Hence the associated sum converges by comparison with
1
n
2
.
Let g
n
(z) =
n
k=1
f
k
(z). As
_
1 −
1
2
__
1 −
1
3
_
. . .
_
1 −
1
n
_
=
1
2
·
2
3
· · ·
n−1
n
=
1
n
, we see
g
n
(z) = z(1 +z)
_
1 +
z
2
_
. . .
_
1 +
z
n
_
· n
−z
=
z(z + 1) . . . (z +n)
n!
n
−z
The function g(z) has zeros at 0, −1, −2, . . . , all of which are simple.
Also,
g(z)
g(z+1)
= z. Indeed, g
n
(z + 1) =
(z+1)···(z+n+1)
n!
n
−z−1
, so
gn(z)
gn(z+1)
=
zn
z+n+1
→ z as
n →∞. Additionally, g(1) = lim
n→∞
g
n
(1) = lim
n→∞
n+1
n
= 1.
Deﬁne the Euler gamma function Γ(z) as
1
g(z)
. It is meromorphic with simple poles at
0, −1, −2, . . . and satisﬁes Γ(z+1) = zΓ(z), Γ(1) = 1, from which we can see Γ(n+1) = n!.
We now prove the identity Γ(z)Γ(1 −z) =
π
sin πz
:
g
n
(z)g
n
(1 −z) =
_
z(z + 1) . . . (z +n)
n!
· n
−z
__
(1 −z)(2 −z) . . . (n + 1 −z)
n!
· n
z−1
_
=
n + 1 −z
n
· z
n
k=1
_
1 −
z
2
k
2
_
.
By the Example in Section 2.2, as n → ∞, the inﬁnite product converges to
sin πz
π
and
n+1−z
n
→1, hence g(z)g(1 −z) =
sin πz
z
.
21
We now give an alternate inﬁnite product for
1
Γ(z)
. We write
g
n
(z) = z
n
k=1
_
_
1 +
z
k
_
e
−
z
k
_
e
z(1+
1
2
+···+
1
n
−log n)
.
This suggests that
g(z) = z
∞
k=1
_
_
1 +
z
k
_
e
−
z
k
_
· e
zγ
.
The convergence of the inﬁnite product can be established as before (we can again compare
with
1
n
2
). This alternate expansion shows that 1+
1
2
+· · · +
1
n
−log n converges; its limit
is γ, the Euler gamma constant.
The logarithmic derivative of Γ(z) is
Γ
(z)
Γ(z)
= −
1
z
−γ +
∞
n=1
_
1
n
−
1
z +n
_
.
Taking derivatives again, we ﬁnd
d
dz
_
Γ
(z)
Γ(z)
_
=
∞
n=0
1
(z+n)
2
. This is positive when z is real
and positive, so log Γ(z) is convex.
A theorem of Bohr and Mollerup says that Γ is the only meromorphic function which
is logarithmically convex on the positive real axis which satisﬁes the functional equation
Γ(z + 1) = zΓ(z) and Γ(1) = 1.
2.5 Normal families and compact subsets of C(Ω)
Suppose Ω ⊂ C is open. We have seen that C(Ω) is metrizable and that H(Ω) is a closed
subset of C(Ω) with the induced metric.
Recall that a metric space is compact if and only if every sequence has a convergent
subsequence.
We say that S ⊂ H(Ω) is a “normal family” if every sequence in S has a subsequence
that converges uniformly on compact subsets of Ω; in other words, S is normal if its closure
is compact.
A set S ⊂ H(Ω) or of C(Ω) is bounded uniformly on compact subsets of Ω if for every
compact K ⊂ Ω, there is a constant M(K) such that f(z) ≤ M(K) when z ∈ K for all
f ∈ S.
22
Proposition. If S ⊂ H(Ω) or of C(Ω) is compact, then S is closed and bounded uniformly
on compact sets.
The fact that S is closed follows from H(Ω) being Hausdorﬀ. For a ﬁxed compact set
K, the map H(Ω) → R : f → max
z∈K
f(z) is continuous and hence it takes K to some
bounded set of R.
The converse is only true in H(Ω). It follows from the following theorem:
Theorem (Montel). A set of functions S ⊂ H(Ω) is a normal family if and only if it is
bounded uniformly on compact sets.
We prove it using two lemmas.
Lemma. The mapping H(Ω) → H(Ω) : f → f
takes a subset that is bounded uniformly
on compact sets to another subset bounded uniformly on compact sets.
It suﬃces to show that for any z
0
∈ Ω, there is a small disk D ⊂ Ω centered at z
0
such that S
is bounded uniformly on D. Choose D with radius r/2 such that the disk
˜
D
centered at z
0
with radius r is contained in Ω. Let γ be the positively oriented boundary
of
˜
D.
Suppose z ∈ D. By Cauchy’s integral formula, f
(z) =
1
2πi
_
γ
f(ζ)
(ζ−z)
2
dζ. Suppose M is
such that f(z) ≤ M on γ for all f ∈ S. As ζ −z ≥ r/2, f
(z) ≤
M
2π
·
4
r
2
· 2πr =
4M
r
.
Lemma. Let D be the open disk with center z
0
and radius R and S ⊂ H(Ω) be bounded
uniformly on compact subsets of D. Then a sequence {f
n
} ⊂ S converges uniformly on
compact sets if and only if for every k, the sequence of derivatives at z
0
, {f
(k)
n
(z
0
)} con
verges.
For the “only if” direction, we simply need to know that the map f →f
is continuous
(we don’t need the hypothesis).
Now we prove the “if” direction. It suﬃces to show that for any z
0
∈ Ω, there is a
small disk D centered at z
0
such that for any f ∈ S, {f
n
} converges uniformly on D. Let
D(z
0
, R) be a closed disk in Ω. We pick D = D(z
0
, r) with radius r < R.
Let r < r
0
< R. Choose M so that f
n
(z) ≤ M for all n whenever z − z
0
 ≤ r
0
. We
can write f
n
(z) as
∞
k=0
f
(k)
n
(z
0
)
k!
· (z −z
0
)
k
.
23
If we let the coeﬃcients
f
(k)
n
(z
0
)
k!
be a
nk
, then by Cauchy’s estimate, a
nk
 ≤
M
r
k
0
. It follows
that
f
n
(z) −f
m
(z) ≤
l
k=0
a
nk
−a
mk
 · r
k
+ 2M
k>l
_
r
r
0
_
k
.
As the geometric series converges, we can choose l large enough so the second term is small.
We then choose n
0
large so that when m, n > n
0
, a
nk
−a
mk
 are simultaneously small for
all 0 ≤ k ≤ l. Then {f
n
(z)} converges uniformly as desired.
We are now ready to prove Montel’s theorem. Suppose S ⊂ H(Ω) is bounded uniformly
on compact sets and {f
n
} ⊂ S. We wish to show {f
n
} has a convergent subsequence.
Take a countable cover of Ω by disks D(z
i
, r
i
) ⊂ Ω. Let λ
j
i
: H(Ω) → C take f →
f
(k)
(z
i
). It suﬃces to ﬁnd a subseqence N ⊂ N such that lim
n∈N
λ
k
i
(f
n
) exists for all k, i.
We construct N as follows: We reindex λ
k
i
as µ
l
. By the ﬁrst lemma, for each l, µ
l
(f
n
)
is bounded. Then we diagonalize, i.e we ﬁrst choose N
1
⊂ N so that the lim
n∈N
1
µ
1
(f
n
)
converges; next we choose N
2
⊂ N
1
so that lim
n∈N
2
µ
2
(f
n
) converges, then N
3
⊂ N
2
so
lim
n∈N
3
µ
3
(f
n
) converges and so on.
We then take N to be the diagonal sequence: N
11
, N
22
, N
33
, . . .
24
Chapter 3
Automorphisms
3.1 Local Geometry of Holomorphic functions
We say that a function f is conformal (biholomorphic) if f is holomorphic and has a
holomorphic inverse. If f is a holomorphic function not identically constant, either of the
two possibilities hold:
(1) Suppose w
0
= f(z
0
) and f
(z
0
) = 0. By the inverse function theorem, f
−1
exists
and is holomorphic in a neighbourhood of w
0
and so f is conformal at z
0
.
(2) Suppose instead f
(z
0
) = 0. We may assume z
0
= 0, f(z
0
) = 0; otherwise we work
with f(z + z
0
) − f(z
0
). We may write w = z
p
f
1
(z) with f
1
(0) = 0 for some p ≥ 2. Then,
w = (zg(z))
p
where g(z) is some determination of the pth root of f
1
. Then w = ζ
p
where
ζ = zg(z) is a holomorphic coordinate change, i.e a map of type (1).
Thus up to a conformal change of coordinates, every holomorphic function is of form
ζ
p
+f(z
0
). The number p −1 is called the ramiﬁcation index of f at z
0
.
Lemma. A nonconstant holomorphic mapping is open.
We know that in a small neighbourhood of f(z
0
), every value suﬃciently close to f(z
0
)
appears exactly p times (as this is true for ζ
p
+f(z
0
)), so f is open.
Corollary. If f ∈ H(Ω) and injective, then f is a homeomorphism. Furthermore, f
−1
is
holomorphic.
Since (2) cannot be injective, (1) must always be the case, hence f is a local homeo
morphism. But as f is injective, it is a genuine homeomorphism.
25
Remark. Even if f
(z) = 0 for all z ∈ Ω, this not guarantee that f is injective. For
instance, f(z) = e
z
is onetoone on any strip a < Im z < b with b −a < 2π.
Remark. While the complex plane C and the unit disk D = {z : z < 1} are homeomor
phic; they are not biholomorphic, for any function f : C → D is constant by Liouville’s
theorem.
Notice that if f, g : Ω → Ω
are two biholomorphic maps, then g
−1
◦ f : Ω → Ω is
biholomorphic. Hence all biholomorphic maps Ω → Ω
are given by composing a ﬁxed
holomorphic map with an automorphism of Ω.
Examples
The automorphisms of the Riemann sphere are the fractional linear transformations z →
az+b
cz+d
. The automorphisms of the complex plane are aﬃne maps z →az +b.
The automorphisms of the upper half plane are fractional linear transformations with
real coeﬃcients and ad −bc > 0.
3.2 Automorphisms
We classify automorphism groups of some domains in C.
Theorem. Aut C are aﬃne maps w = az +b with a = 0.
Suppose w = f(z) is an automorphism of C. Either, (1) f(z) has an essential singularity
at inﬁnity, or (2) it is a polynomial (being a rational function with only pole at inﬁnity).
The ﬁrst case cannot happen: on one hand, Im{z < 1} and Im{z > 1} should be
disjoint (as f(z) is injective), but as f is open, the ﬁrst set is open; but the essential
singularity at inﬁnity implies the second set is dense which is impossible.
We investigate the second case further. Suppose f is a polynomial of degree n. Then,
f(z) = w has n distinct roots except for special values of w (namely, when f
(z) = 0). As
f is injective, we see that n = 1.
We conclude f(z) = az + b. Obviously, a = 0. If a = 1, f(z) is a translation; if a = 1,
f has a unique ﬁxed point z =
b
1−a
.
26
Theorem. Aut S
2
are fractional linear transformations w =
az+b
cz+d
, ad−bc = 0 (the inverse
is z =
dw−b
−cw+a
). (Note: w takes inﬁnity to
a
c
if c = 0 and to inﬁnity if c = 0).
Obviously, the fractional linear transformations form a subgroup of G ⊂ Aut S
2
. We
want to show that actually G = Aut S
2
.
The subgroup of Aut S
2
which ﬁxes the point at inﬁnity is precisely Aut C. As fractional
linear transformations act transitively on the Riemann sphere, for T ∈ Aut S
2
, we could
consider a fractional linear transformation S which takes T(∞) → ∞. Then S ◦ T must
be aﬃne which shows that T is a fractional linear transformation.
Let H
+
be the upper halfplane and D the unit disk. The two are biholomorphic:
indeed, the map z →
z−i
z+i
which takes H
+
to D: it maps i to the origin and points 0, 1, ∞
to −1, −i, 1 respectively.
Theorem. Aut D are fractional linear transformations of the form
w = e
iθ
·
z −z
0
1 −z
0
z
with z
0
 < 1.
The above transformations are automorphisms, they take the unit disk into itself and
their inverse is of the same form. We now show that there are no other automorphisms.
Suppose T ∈ Aut D. Let S = e
iθ
·
z−z
0
1−z
0
z
with z
0
= T
−1
(0) and θ = arg T
(z
0
). Let
f(z) = S ◦ T
−1
(z). Then f(0) = 0 and f(z) < 1 when z < 1. By Schwarz’s lemma,
f(z) ≤ z.
Applying the same argument to the inverse of f, we see that z ≤ f(z), so f(z) = z.
By the equality condition of Schwarz’s lemma, we see that f(z) = e
iα
·z, so S(z) = e
iα
·T(z).
We now diﬀerentiate and set z = z
0
: S
(z
0
) = e
iα
· T
(z
0
); from which follows that α = 0
or S = T as desired.
Theorem. Aut H
+
are fractional linear transformations with real coeﬃcients and positive
determinant.
Conjugation by z →
z−i
z+i
shows that Aut H
+
is composed solely of fractional linear
transformations. If a fractional linear transformation w(z) is to take the real axis to itself,
it takes 1, 0, ∞ to three real numbers z
1
, z
0
, z
∞
; so w(z) =
z−z
0
z−z∞
·
z
1
−z∞
z
1
−z
0
.
27
3.3 Riemann Mapping Theorem
Our next goal is to show that any simplyconnected subset Ω of C (except for C itself) is
biholomorphic to the unit disk D = {z < 1}.
We ﬁrst reduce to the case of bounded domain. If Ω isn’t the entire complex plane,
then there is some point a ∈ Ω. As Ω is simplyconnected,
1
z−a
has a primitive g(z) in Ω.
Then, z −a = e
g(z)
.
As G(Ω) is open, it contains a disk E centered at g(z
0
). As e
g(z)
is injective, E and
its translate E + 2πi are disjoint. Then,
1
g(z)−[g(z
0
)+2πi]
is a biholomorphism of Ω onto a
bounded domain.
By translating and scaling if necessary, we can assume that Ω contains the origin and
lies inside the unit disk.
Let A = {f ∈ H(Ω) : injective, f(0) = 0, f(z) < 1}.
Lemma. Then g(Ω) = D if and only if g
(0) = sup
f∈A
f
(0).
Assuming the lemma, it suﬃces to show that sup
f∈A
f
(0) is attained. For this pur
pose, we introduce B = {f ∈ A : f
(0) ≥ 1}. B is nonempty as f(z) = z is in it. It
suﬃces to show that B is compact.
By Montel’s theorem, B is normal as it is bounded uniformly on compact sets (its
actually bounded everywhere by 1). To show that B is closed, suppose f = limf
n
with
f
n
∈ B (as usual, we assume that the sequence converges uniformly on compact sets).
We know that f(0) = limf
n
(0) = 0, clearly f
(0) ≤ 1. Injectivity follows from
Hurwitz’s theorem (see Section 1.1), the functions are nonconstant as f
n
(0) ≥ 1 (from
which follows that f
(0) ≥ 1, recall that the map f →f(0) is continuous).
It is easy to see that f(z) ≤ 1, but if the value 1 is attained, by the maximummodulus
principle, f reduces to a constant which is impossible.
The remaining lemma follows from Schwarz lemma which says that if f ∈ H(D), f(0) =
0, f(z) < 1 then f(z) ≤ z for all z ∈ D. Indeed, applying the maximummodulus
principle to g(z) =
f(z)
z
, we see g(z) ≤
1
z
≤ 1 and f
(0) ≤ 1.
If additionally f(z
0
) = z
0
 for some z
0
∈ D, then f(z) = λz with λ = 1: by the
maximummodulus principle, g must be constant; and at z
0
, λ has absolute value 1.
28
3.4 Riemann Mapping Theorem II
We deﬁned family A = {f ∈ H(Ω) : injective, f(0) = 0, f(z) < 1}. We want to show
that g(Ω) = D if and only if g
(0) = sup
f∈A
f
(0).
The “only if” direction is quite clear: suppose g ∈ A with g(Ω) = D and f ∈ A is
some other holomorphic map. Then h = f ◦ g
−1
takes the disk to f(Ω) and takes the
origin to itself. By Schwarz’s lemma, h
(0) ≤ 1, diﬀerentiating f = g ◦ h, we see that
f
(0) ≤ g
(0).
Suppose f ∈ A and a is some point in the disk but not in f(Ω). We want to show that
there is some g ∈ A with g
(0) > f
(0).
Let ϕ(ζ) =
ζ−a
1−aζ
be an automorphism of the disk which sends a to 0. Then, ϕ◦ f(z) =
f(z)−a
1−af(z)
is nonvanishing. Since Ω is simplyconnected, it has a singlevalued holomorphic
square root F(z), i.e if θ(w) = w
2
, we may write ϕ ◦ f(z) = θ ◦ F(z).
Then
f = ϕ
−1
◦ θ ◦ F = ϕ
−1
◦ θ ◦ ψ
−1
. ¸¸ .
h(z)
◦ ψ ◦ F
. ¸¸ .
g(z)
where ψ(η) =
η−F(0)
1−F(0)η
. Then h take the disk to itself and ﬁxes the origin.
This allows us to apply Schwarz’s lemma: h
(0) ≤ 1. We don’t win unless we see that
h
(0) is actually less than 1. But if h
(0) = 1, the equality condition of Schwarz would
imply that h is a rotation, but it isn’t.
Boundary Behaviour
Having shown that any two simply connected domains in the complex plane are biholo
morphic, we wish to see if the mapping extends a homeomorphism of closed domains. We
will do this in the special case of a polygon (although a theorem of Carath´eodory shows
that this is true for any domains with Jordan curve boundary).
Suppose Ω is a simplyconnected open set whose boundary is a polygonal curve with
vertices z
1
, z
2
, . . . z
n
; z
n+1
= z
1
. Let α
k
be the inner angle at z
k
, so 0 < α
k
< 2 (more
precisely, α
k
is the value of the argument of
z
k−1
−z
k
z
k+1
−z
k
between 0 and 2).
Let β
k
π be the outer angle at z
k
, so β
k
= 1−α
k
, i.e −1 < β
k
< 1 and β
1
+· · · +β
n
= 2.
It is easy to see that Ω is convex if and only if all β
k
> 0.
29
We will need a nonstandard version of the Schwarz reﬂection principle:
Lemma (Schwarz). Suppose Ω = Ω
+
∪Ω
−
is a domain symmetric about the real axis and
f(z) is a holomorphic function deﬁned on Ω
+
whose imaginary part tends to zero as z
approaches the real axis. Then f(z) can be completed to a holomorphic function deﬁned on
all of Ω by setting f(z) = f(z).
The diﬃculty here is that a priori the real part of f(z) may not extend continuously to
the real axis. We prove that this is indeed the case. For this purpose, we assume that Ω is
simplyconnected (this is a local question, if x ∈ R, we can restrict to a small disk centered
at x).
The function v(z) = Imf(z) is harmonic. We can extend it to a function on Ω by
setting v(z) = −v(z). It is harmonic as it satisﬁes the mean value principle. We take
u(z) to be the “harmonic conjugate” which agrees with Re f(z) on Ω
+
. Then u + iv is
an extension of f(z). But if f(z) is to admit any extension beyond the real axis, it must
extend continuously up to the real axis.
We now know that the formula f(z) = f(z) deﬁnes a function which is holomorphic
on Ω
+
∪ Ω
−
and is continuous on the real axis. By a corollary of Morera’s theorem, f(z)
must be holomorphic on the real axis as well.
We now return to the question of boundary behaviour of conformal mapping of a
polygon Ω. Suppose f : Ω → D is conformal. We ﬁrst show that f extends continuously
at the sides of Ω. We even show more: namely that f extends beyond each side as a
holomorphic function.
This is easy. Simply compose f with the fractional linear transformation from ϕ : D →
H
+
given by ϕ(w) =
w−i
w+i
and apply the Schwartz reﬂection principle (exercise: ϕ does
extend to a continuous map from D →H).
At the angles, the situation is a little more tricky. We straighten the angle at vertex
z
k
of Ω. For this purpose, we consider the function g(ζ) = f(z
k
+ ζ
α
k
) deﬁned in the
neighbourhood of the origin in the upper half plane (see picture).
By arguments with the Schwartz reﬂection principle, g(ζ) extends to a holomorphic
function in small disk centered at the origin and as f
∂Ω
= g
R
, f extends continuously at
the angles as well.
30
z
k
z
w
0
ζ
α
k
π
z
k
+ζ
α
k
f
F
g
π
w
k
Thus we have shown that the map f : ∂Ω → ∂D is continuous. The above reasoning
also allows us to conclude that f is locally onetoone. However, we really want to see that
f is genuinely onetoone on the boundary.
For this purpose, take a positively oriented curve γ ⊂ Ω which is close and homotopic
to ∂Ω. Then f(γ) is a simple closed curve which goes around the origin with winding
number +1. So f(∂Ω) must wind around once counter clockwise around the origin (details
are left as an exercise).
3.5 ChristoﬀelSchwarz Formula
We can even ﬁnd an explicit formula of a Riemann mapping from a polygon to the unit
disk. Actually, the formula is not for f(z), but rather its inverse, which we denote F.
Theorem (ChristoﬀelSchwarz). F is of the form F(w) = c
_
w
0
n
k=1
(w − w
k
)
−β
k
dw + c
where are c, c
are constants.
We denote the image of z
k
under f by w
k
. Any 3 of the w
k
’s can be picked arbitrarily
but the remainder are uniquely determined (exercise).
31
Remark. The formula always takes {z = 1} to a closed polygonal curve, but the curve
may have self intersections. If the curve does not have selfintersections, then F takes the
unit disk biholomorphically onto the interior of the curve.
Fix a vertex z
k
of Ω and consider its associated function g(ζ) constructed previously.
As w = g(ζ) is locally onetoone at the origin, it is invertible and so exists a power series
expansion ζ =
∞
m=1
b
m
(w −w
k
)
m
with b
1
= 0.
Raising to α
k
th power we see that ζ
α
k
= F(w) − z
k
= (w − w
k
)
α
k
G(w) where G(w)
is holomorphic and nonzero near w
k
. Diﬀerentiating, we ﬁnd that F
(w)(w − w
k
)
β
k
=
α
k
G(w) + (w − w
k
)G
(w) is holomorphic and nonzero at w
k
. We can now forget the
auxiliary functions g and G.
Set H(w) = F
(w)
n
k=1
(w −w
k
)
β
k
. By the discussion above, it is holomorphic in the
unit disk, extends continuously to the boundary and does not vanish in the closed unit
disk. We claim that in fact H(w) is constant.
As H(w) does not vanish, it suﬃces to show that arg H(w) is constant on {w = 1}
(recall arg is the imaginary part of log). To this end, we show that arg H(w) is constant
for w = e
iθ
with arg w
k
< θ < arg w
k+1
and by continuity, arg H(w) will be constant on
the whole unit circle.
Taking arguments, we see that
arg H(w) = arg F
(w) +β
k
arg(w −w
k
)
The chain rule tells us that arg F
(w) = arg(F ◦ w)
(θ) − arg w
(θ). The argument of
(F ◦ w)
(θ) measures the direction of the tangent to the graph of F which is constant as
F traces a side of Ω. On the other hand, w
(θ) represents the direction of the tangent to
circle at w = e
iθ
, i.e w
(θ) = θ +π/2 = θ + const.
By elementary geometry, we see that arg(w − w
k
) = θ/2 + const (exercise, hint: by
rotation, we can assume that w
k
= 1). Continuing our calculation, we see that
arg H(w) = −θ +
_
β
k
_
θ
2
+ const = const.
If H(w) = c (identically), then F
(w) = c
n
k=1
(w − w
k
)
−β
k
and we obtain the Schwarz
Christoﬀel formula by integration.
32
Chapter 4
Complex Geometry
4.1 Complex Manifolds
A manifold is a paracompact Hausdorﬀ topological space locally homeomorphic to R
n
, i.e
we can cover M by open sets U
i
for which exist homeomorphisms (charts) ϕ
i
: U
i
→ V
i
with V
i
open in R
n
.
For complex structure, we replace R
n
with C
n
and insist that the transition maps
ϕ
j
◦ ϕ
−1
i
: V
i
→ V
j
be holomorphic for all i, j. A complex manifold is a manifold with
complex structure.
A function on a complex manifold M is holomorphic if it is holomorphic in local coor
dinates (that is, for any coordinate chart ϕ : U
i
→V
i
, f ◦ ϕ
−1
i
is holomorphic on V
i
).
A onedimensional complex manifold is called a complex curve or an abstract Riemann
surface.
Example: the Riemann sphere. As charts, we take U = S
2
\ {N} with ϕ
U
: (x, y, t) →
x+iy
1−t
(stereographic projection) and V = S
2
\ {S} with ϕ
V
: (x, y, t) →
x−iy
1+t
(stereographic
projection followed by conjugation).
The transition map ϕ
U
◦ ϕ
−1
V
: C\{0} → C\{0} takes z to
1
z
. In local coordinates,
z
U
= z ◦ φ
U
, z
V
= z ◦ φ
V
, we have z
U
z
V
= 1. It is an exercise to the reader to verify that
S
2
is biholomorphic to P
1
(C).
Given two complex manifolds M and N, a function f : M → N is holomorphic if it is
in local coordinates, i.e if ψ
j
◦ f ◦ ϕ
−1
i
is holomorphic for all coordinates charts ϕ
i
of M
and ψ
j
of N.
33
Given manifolds M and N with complex structures we say that f : M →N is an “iso
morphism” (or a biholomorphism) if f is a homeomorphism of the underlying topological
spaces and f, f
−1
are holomorphic.
Two complex structures on M are equivalent if the identity mapping is an isomorphism.
Thus a complex manifold is really a paracompact Hausdorﬀ space with an equivalence class
of complex structures.
Properties of holomorphic functions extend to manifolds
(1) If M and N are complex manifolds with M connected and f, g : M →N are holomor
phic and coincide on a set with a limit point, then f = g on M.
Consider the set of points in which f, g coincide in a neighbourhood. It is open (auto
matic). It is closed (given a sequence {z
k
}, its tail lies in one chart). It is not empty, for
it contains the limit point; so f, g must coincide everywhere on M.
(2) Suppose M is connected and f is holomorphic on M, if f has a relative maximum,
it is constant.
If f has a relative maximum, in a neighbourhood, it coincides with the constant
function, use part (1). In fact, any holomorphic function on a compact connected manifold
is constant.
4.2 Complex Manifolds II
We have seen that holomorphic functions on compact complex manifolds are constant.
We are more interested in meromorphic functions; these are holomorphic maps into the
Riemann sphere.
For example, p : C → C/Γ : f → f ◦ p is a bijection between meromorphic functions
on C/Γ and meromorphic functions on C with Γ as a group of periods.
Suppose f : M → N is holomorphic, a ∈ M and b = f(a). Choose local coordinates
z, w vanishing at a, b respectively. We deﬁne the ramiﬁcation index to be the multiplicity
of the root of w ◦ f ◦ z
−1
.
The multiplicity does not depend on the choice of local coordinates, for coordinates
charts are homeomorphisms; in particular, they are locally onetoone.
34
If the ramiﬁcation index is k, we can choose local coordinates in which w = z
k
.
A holomorphic diﬀerential form ω on manifold M with atlas (U
i
, z
i
) is a set of forms
ω = f
i
(z
i
)dz
i
which transform properly, i.e dz
i
= f
ij
(z
j
)dz
j
.
In a neighbourhood of any point, a holomorphic diﬀerential form has a primitive g, i.e
a holomorphic function for which ω = dg (by deﬁnition dg is the form g
(z
i
)dz
i
). If M is
simplyconnected, then ω has a global primitive.
Complex manifolds have a natural orientation as the transition maps are orientation
preserving. This allows us to consider the oriented boundary of compact set. In particular,
if ω is a holomorphic diﬀerential form and Γ is the oriented boundary of a compact set
then
_
Γ
w = 0.
Let ω be a diﬀerential form holomorphic in the complement of a discrete set E. If a ∈ E
and z is a local coordinate in a neighbourhood of a, we may write ω = ω
1
+(
c
1
z
+
c
2
z
2
+. . . )dz
where ω
1
is a diﬀerential form holomorphic near a.
We deﬁne Res(ω, a) =
1
2πi
_
γ
ω where γ is a small closed curved around a with winding
number 1. The Residue theorem carries over to manifolds (with same proof):
Theorem. Given a compact set K ⊆ M with piecewise smooth oriented boundary Γ, with
Γ containing no points of E, we have
1
2πi
_
Γ
ω =
a∈K
Res(ω, a).
Let Y be a complex curve. By a Riemann surface over Y , we mean a connected complex
curve X together with a nonconstant holomorphic map ϕ : X → Y whose ramiﬁcation
points are isolated and the ﬁbers ϕ
−1
(y) are discrete.
35
4.3 Examples of Riemann Surfaces
Example 1: y = log x
Let Y = C
∗
be the complex plane without the origin. The map log z : Y → C is multi
valued. We consider X = C, ϕ : X → Y given by ϕ(t) = e
t
. Then t is a holomorphic
function on X which is locally a branch of log, i.e t = log ◦ϕ.
We made the function log z singlevalued by introducing the Riemann surface.
Here ϕ : X → Y is a covering space or an unramiﬁed covering space, i.e any point of
Y has an open neighbourhood V such that ϕ
−1
(V ) is a disjoint union of open sets U
i
each
mapped homeomorphically onto V .
Example 2: y = x
1/2
We now construct a Riemann surface for the function y = x
1/2
over Y = C. Let X ⊂ C
2
be the points (x, y) for which y
2
= x.
The origin is a ramiﬁcation point: X lies in two sheets over Y everywhere except at
the origin.
To construct a model of X, take two copies of the complex plane, cut out their positive
real rays and glue the top of one to bottom of the other and vica versa.
Example 3: y = (1 −x
3
)
1/3
Let X ⊂ C
2
be the set of points (x, y) for which x
3
+ y
3
= 1. It is a complex manifold,
near (x
0
, y
0
) with y
0
= 0, we may take x as a local coordinate, x →(x,
3
√
1 −x
3
) with the
branch of cube root corresponding to (x
0
, y
0
). Similarly, at x
0
= 0, y → (
3
_
1 −y
3
, y) is a
local coordinate.
To see that the two charts are compatible, if x
0
= 0 and y
0
= 0,
3
√
1 −x
3
has a branch
equal to y
0
at x = x
0
and
3
√
1 −x
3
has a branch equal to x
0
at y = y
0
.
As functions on X, both x and y are holomorphic: e.g to see that x is holomorphic,
near (x
0
, y
0
) with y
0
= 0, x is the local coordinate, x = x, the identity map; near (x
0
, y
0
)
with x
0
= 0, y is the local coordinate, x =
3
_
1 −y
3
.
X has 3 points over every point of C except at the cube roots of unity, 1, j, j
2
where
j = e
2πi/3
.
36
Consider the diﬀerential form ω =
dx
(1−x
3
)
1/3
=
dx
y
on X. The equation deﬁnes ω when
y = 0. Diﬀerentiating the equation x
3
+ y
3
− 1 = 0, we see that 3x
2
dx + 3y
3
dy = 0 or
dx
y
= −
y
x
2
dy when both x = 0 and y = 0. Thus, ω is implicitly deﬁned to be −
y
x
2
dy when
x = 0.
We made the diﬀerential form
dx
√
1−x
3
singlevalued by introducing the Riemann surface.
We can extend X to a Riemann surface over S
2
= P
1
(C). Let X
⊂ P
2
(C) be the set
of points [x, y, z] given by x
3
+y
3
= z
3
formed by homogenizing (
x
z
)
3
+ (
y
z
)
3
= 1.
X
is composed of X together with 3 points at inﬁnity (i.e where z = 0), namely
[1, −1, 0], [j, −1, 0] and [j
2
, −1, 0].
4.4 Example: y =
_
P(x)
Let P(x) = (x − a
1
)(x − a
2
) . . . (x − a
d
) be a polynomial with distinct roots. Suppose
X ⊂ C
2
is the set of points (x, y) for which y
2
= P(x). We complete it to a Riemann
surface X
⊂ P
2
(C) over P
1
(C) by homogenizing the equation:
y
2
z
d−2
= (x −a
1
z)(x −a
2
z) . . . (x −a
n
z).
We now investigate this in detail, paying attention to the branch points and behaviour at
inﬁnity.
Case d = 1:
The set X : {y
2
= x −a} is a parabola and hence a manifold; so X
: {y
2
= xz −a
1
z
2
} is
smooth at least in the ﬁnite part of P
2
(C). At inﬁnity, X
has one point: if we plug z = 0
into the equation for X
, we get y
2
= 0 yielding the point [1, 0, 0].
The equation for X
has multiple roots at a and inﬁnity, which are the branch points.
One can form a model of X
by taking two copies of the Riemann sphere, making cuts in
each of them between a and ∞, gluing the top bank of the ﬁrst sphere to the bottom bank
of the second sphere and vica versa. This construction shows that X
is topologically a
sphere.
37
Case d = 2:
The equation for X is {y
2
= (x − a)(x − b)}. Homogenizing, we obtain the equation for
X
: {y
2
= (x −az)(x −bz)}. If we set z = 0, we get the equation y
2
= x
2
, so X
has two
points at inﬁnity, namely [1, 1, 0] and [1, −1, 0]. In coordinates (y, z) at inﬁnity,
y = ±(1 −az)
1/2
(1 −bz)
1/2
.
So, X
is a manifold and has branch points are a and b. Actually, this case is pretty much
the same as the previous case: if we take the equation y
2
= (x −az)(x −bz) and perform
a linear change of coordinates to make it y
2
= uv, after we dehomogenize with respect to
v, we are left with the equation y
2
= u.
Case d = 3:
Now, the equation for X is y
2
= (x−a)(x−b)(x−c) and for X
, y
2
z = (x−az)(x−bz)(x−cz).
If we set z = 0, we obtain the equation x
3
= 0, so there is one point at inﬁnity, namely
[0, 1, 0].
X
has four branch points: a, b, c, ∞. To form a model for X
, we take two copies of
the Riemann sphere, pair up the points, say a, b and c, d, make cuts between a and b and
between c and d in both spheres and glue the banks of a −b and c −d to each other. This
shows that X
is topologically a torus.
Case d ≥ 4:
The equation for X
is y
2
z
d−2
=
d
i=1
(x−a
i
z). There is only one point at inﬁnity, [0, 1, 0].
We must choose (x, z) to be the coodinates at inﬁnity:
z
d−2
=
(x −a
i
z).
Unfortunately, this is not smooth at the origin, for it is not true that either
∂
∂x
= 0 or
∂
∂z
= 0, so the implicit function theorem does not apply.
38
4.5 Abel’s Theorem
Let X ⊂ C
2
be given by the equation y
2
= P(x) with P(x) = 4x
3
−20a
2
x −28a
4
having 3
distinct roots. Let X
⊂ P
2
(C) be given by the equation y
2
t = 4x
3
−20a
2
xt −28a
4
. Then,
X embeds into X
⊂ P
2
(C) by taking (x, y) to [x, y, 1].
We have seen that X
has one point at inﬁnity, namely [0, 1, 0]. If (x
, t
) → [x
, 1, t
]
are the coordinates at [0, 1, 0], then t
= 4x
3
− 20a
2
x
t
2
− 28a
4
t
3
. This tells us that
t
= 4x
3
−320a
2
x
7
+... (see Section 1.6 for details).
The projection map ϕ : X → C given by ϕ(x, y) → x gives X the structure of a
Riemann surface over C. We extend it to ϕ
which makes X
a Riemann surface over
P
1
(C)
∼
= S
2
:
ϕ
: [x, y, 1] →[x, 1], ϕ
: [x
, 1, t
] →[x
, t
] in P
1
(C),
i.e to
x
t
∈ C in coordinates at 0 or to
t
x
in coordinates at inﬁnity.
The form ω =
dx
y
=
dy
6x
2
−10a
2
has a primitive in a neighbourhood of each point of X;
globally the primitive is a multivalued holomorphic function z = z(x, y). Each branch of
z in the neighbourhood of any point (x
0
, y
0
) ∈ X is a local coordinate:
(1) if y
0
= 0, using dz =
dx
y
, we see that
dz
dx
=
1
y
= 0,
(2) if y
0
= 0, using dz =
dy
6x
2
−10a
2
, we see that
dz
dy
=
1
6x
2
−10a
2
= 0.
Near the point at inﬁnity in X
, [0, 1, 0], we have [x, y, 1] = [
x
t
,
1
t
, 1], so
ω =
dx
y
= t
· d
_
x
t
_
= dx
−
x
t
dt
= dx
−
12x
2
−2240a
2
x
6
+. . .
4x
2
−320a
2
x
6
+. . .
· dx
= −2dx
(1 +g(x
)).
with g(x
) holomorphic near x
= 0 and g(0) = 0.
Suppose that a
2
, a
4
come from a discrete subgroup Γ, i.e a
2
= 3
w=0
1
w
4
, a
4
=
5
w=0
1
w
6
. We have seen that the meromorphic transformation x = p(z), y = p
(z) deﬁnes
a biholomorphism from C/Γ → X
. Recall that the inverse map is z(p) =
_
p
[0,1,0]
ω where
the integral is determined only modulo Γ.
We are interested in the converse:
Theorem (Abel). Given a
2
, a
4
such that p(x) = 4x
3
−20a
2
x −28a
4
has 3 distinct roots,
there is a discrete subgroup Γ of C generated by e
1
, e
2
linearly independent over R such
that a
2
= 3
w=0
1
w
4
, a
4
= 5
w=0
1
w
6
.
39
We will prove it using two lemmas:
Lemma (1). The diﬀerence of two branches of z belongs to some discrete group Γ.
We can then consider the elliptic curve associated to Γ, i.e X
given by equation
y
2
= 4x
3
−20b
2
x −28b
4
where b
2
= 3
w=0
1
w
4
and b
4
= 5
w=0
1
w
6
. We can consider the
composition:
X
?
//
C/Γ
∼
=
//
X
[x, y, 1] z [(p(z), p
(z), 1]
Abel’s theorem says that this composition is in fact the “identity mapping” , i.e [p(z), p
(z), 1]
is nothing but the coordinate mapping [x, y, 1]. And therefore we must have a
2
= b
2
and
a
4
= b
4
. It will turn out necessary to show ﬁrst the weaker statement ﬁrst:
Lemma (2). The map z is a biholomorphism.
4.6 Abel’s Theorem II
Lemma 1
Let p
0
= [0, 1, 0] be the point at inﬁnity, then z(p) =
_
p
p
0
ω is globally welldeﬁned up to a
period of ω, i.e to π(γ) =
_
γ
ω where γ is a loop.
We can extend π to 1cycles by linearity. By Stokes theorem, the map π vanishes on
boundaries:
_
∂σ
ω =
_
σ
dω = 0, so π descends to a map from H
1
(X
, Z) →C.
We now need two facts not from this course: ﬁrst, X
is a torus, say by the degree
genus formula g =
(d−1)(d−2)
2
=
(3−2)(3−1)
2
= 1 and secondly, H
1
(X
, Z) = Z ⊕Z.
We claim that Γ is lattice. If not, Γ is contained within a straight line. Take α so that
(απ(γ)) = 0. But then (αz) deﬁnes a singlevalued function harmonic on X
. As X
is compact, by maximummodulus, we see that it is constant. But z is not constant as ω
does not vanish identically.
Lemma 2
First, we establish that z is surjective: as branches of z are local coordinates at each point
of X
, z is a local homeomorphism, in particular open, so has open image. But the image
is also closed as X
is compact.
40
Thus X
is a covering space of C/Γ and so it must be biholomorphic to C/Γ
for some
subgroup Γ
of Γ (and z is just the quotient map). We show that in fact Γ = Γ
, i.e we
need to prove the other inclusion.
The universal covering space of C/Γ is
˜
X
∼
= C. The map z : X
→C/Γ lifts to a map
˜ z between the covering space
˜
X
and C/Γ. We have the following commutative diagram:
˜
X
∼
= C
p
˜ z
//
C
q
Γ
X
∼
= C/Γ
z
//
C/Γ
We explain the map ˜ z. Let q
0
∈
˜
X
be a point above p
0
. Given a point q ∈
˜
X
, let ˜ γ be
an arbitrary path joining q
0
to q. If γ = p(˜ γ) then ˜ z(q
0
) =
_
γ
ω. Note that it is no longer
necessary to quotient out by Γ to make the map welldeﬁned.
Clearly ˜ z takes Γ
into Γ for if q ∈ Γ
, when we project the path connecting q
0
to q on
X
, it becomes a loop and we know periods of ω lie in Γ. Denote the restriction ˜ z
Γ
by I.
As every element of Γ is a periodic, the map I is surjective. This forces I to be injective
(its a map from Z
2
→ Z
2
) and therefore bijective so the covering map z : C/Γ
→ C/Γ is
trivial, i.e z is a biholomorphism.
Conclusion
The function x is meromorphic on X
with a pole of order 2 at inﬁnity; indeed
x =
x
t
=
x
4x
3
−320a
2
x
7
+. . .
.
By lemma 2, we can think of x as a meromorphic function on C of z with Γ being the
group of periods and having a double pole at z = 0. Write:
x(z) =
c
z
2
+
d
z
+e +fz +. . .
x
(z) = −
2c
z
3
−
d
z
2
+f +. . .
As dz =
dx
y
, x
(z) = y i.e x
(z)
2
= y
2
= 4x
3
−20a
2
x −28a
4
. Equating coeﬃcients, we see
that 4c
2
= 4c
3
, 4cd = 12c
2
d and d
2
= 12c
2
e + 12cd. It follows c = 1, d = 0, e = 0.
Hence x(z) = p(z) as the diﬀerence is holomorphic and doubly periodic and vanishing at
the origin. Diﬀerentiating, we also see that y(z) = p
(z). Abel’s theorem is now completely
proved.
41
Chapter 5
A Modular Function
5.1 Analytic Continuation
Given a holomorphic function f deﬁned on a domain U, we wish to ﬁnd the largest con
nected open set V ⊇ U to which we can extend f.
Suppose (f
0
, D
0
) is a function element, that is f
0
is a holomorphic function deﬁned on
the disk D
0
. We say that (f
0
, D
0
) may be analytically continued along a chain of disks
C = {D
0
, D
1
, . . . D
n
} (D
i
∩ D
i+1
= ∅) if there exist function elements (f
i
, D
i
) such that
f
i
= f
i+1
on D
i
∩ D
i+1
.
We say that (f
n
, D
n
) is an analytic continuation of (f
0
, D
0
) along C and that (f
i
, D
i
),
(f
i+1
, D
i+1
) are direct analytic continuations of each other.
Suppose that f ∈ H(D) where D is a disk centered at a. Suppose γ : [0, 1] → C is a
curve with f(0) = a. We say that (f, D) can be analytically continued along γ if it can be
continued along a chain of open disks {D = D
0
, D
1
, . . . D
n
} which covers γ.
This means we have a partition 0 = t
0
< t
1
< · · · < t
n+1
= n such that γ([t
i
, t
i+1
]) ⊂ D
i
and γ(1) being the center of the last disk.
Lemma. A function element (f, D) has at most one analytic continuation along γ.
Suppose we can continue (f, D) along chains C
1
= {D = D
0
, D
1
, . . . D
m
} and C
2
=
{D = E
0
, E
1
, . . . , E
n
} covering γ to (g
m
, D
m
) and (h
n
, E
n
) respectively, we need to show
that g
i
= h
j
in D
i
∩ E
j
whenever D
i
and E
j
intersect.
42
Suppose not, consider the indices (i, j) with the smallest i + j for which this fails.
Without loss of generality, we can assume that s
i
≥ t
j
.
The functions g
i−1
and g
i
agree in D
i−1
∩ D
i
while g
i−1
and h
j
agree on D
i−1
∩ E
j
. It
follows that g
i
= h
j
in D
i−1
∩ D
i
∩ E
j
which is not empty, so then g
i
= h
j
continues to
hold in D
i
∩ E
j
.
More generally, suppose U ⊂ C is a domain and f ∈ H(U). We wish to ﬁnd an
unramiﬁed Riemann surface (X, ) over C and a biholomorphing mapping σ which takes
U onto an open subset of X such that:
(1) ◦ σ is the inclusion map of U into C,
(2) f extends to a holomorphic function g on X (i.e g ◦ σ = f),
(3) (X, ) is the “largest” Riemann surface satisfying (1) and (2).
We explain condition (3). If (X
,
, σ
) also satisfy (1) and (2), we require that there
is a unique holomorphic map h : X
→X such that
= ◦ h and σ
= σ ◦ h. In this case,
g
= g ◦ h as they coincide on σ(U) = σ
(U) (g
◦ σ
= f = g ◦ h ◦ σ).
We shall not address this more general question.
5.2 Monodromy Theorem
Theorem (Monodromy). Suppose (f, D) can be analytically continued along any curve in
Ω starting at a (the center of D). Then (1) If γ
0
, γ
1
are homotopic curves in Ω from a
to b, then continuations of (f, D) along γ
0
and γ
1
coincide. (2) If Ω is simplyconnected,
then there is a holomorphic function g on Ω such that g = f on D.
Proof of (1). Let γ
s
, s ∈ [0, 1] be a homotopy with endpoints a, b. Then (f, D) may
analytically continued along each γ
s
to (γ
s
, D
s
). Need to show that g
0
= g
1
(in D
0
∩ D
1
).
Fix s. Let C = {D = E
0
, E
1
, . . . E
n
} be a chain of disks covering γ
s
. This means that
there is a partition 0 = t
0
< t
1
< · · · < t
n+1
= 1 with γ([t
i
, t
i+1
]) ⊂ E
i
for i = 0, 1, . . . n.
We will ﬁnd a δ > 0 such that if s
−s < δ, then C covers γ
s
as well.
By the uniqueness of analytic continuation, it would follow that g
s
= g
s
for s −s
 < δ.
By compactness of [0, 1], we would have g
0
= g
1
.
Choose to be less than the distance from γ
s
[0, 1] to C\E
i
. Since γ is uniformly
continuous, we have a δ > 0 such that γ
s
(t) − γ
s
(t) < for all t ∈ [0, 1] whenever
s −s
 < δ.
43
Proof of (2). To deﬁne g near b ∈ Ω, choose any chain of disks C = {D = E
0
, E
1
, . . . E
n
}
with E
n
containing b and analytically continue (f, D) to obtain the function element
(g
b
, E
n
). By part (1), the construction of the function element (g
b
, E
n
) does not depend
on the choice of C.
From the construction, it is clear various g
b
’s patch together: we have to check that if
b
∈ E
n
is some other point, then the analytic continuation of (f, D) near b
matches up
with (g
b
, E
n
). But we can just use precisely the same chain of disks!
5.3 Modular Function
We will introduce the modular function to show the following results:
Theorem (Little Picard). A nonconstant entire function f(z) omits at most one value
(it is possible that one value is omitted, for instance f(z) = e
z
omits 0).
Theorem (Big Picard). If f has an essential singularity at z = a, then in any neigh
bourhood of a, f takes every complex value, with one possible exception, inﬁnitely many
times.
The Little Picard theorem is a special case of the Big Picard theorem: if f is entire,
at inﬁnity, either f has a pole (in which case, it is a polynomial) or it has an essential
singularity (so f(
1
z
) has an essential singularity at the origin).
The modular group G = SL(2, Z) acts on the upper halfplane H
+
: if ϕ ∈ G given by
the matrix (a, b; c, d) with ad −bc = 1 then ϕ(z) =
az+b
cz+d
.
Exercise: G is generated by z + 1, −
1
z
.
Losely speaking, a modular function is a holomorphic or a meromorphic function deﬁned
on H
+
which is invariant under the action of G (or a subgroup of G).
To prove the Little Picard Theorem, we will construct a modular function invariant
under the subgroup Γ ⊂ G generated by σ(z) =
z
2z+1
and τ(z) = z + 2.
Suppose z = x + iy. Let ∆ ⊂ C be the region: y > 0, −1 ≤ x ≤ 1, 2z + 1 ≥ 1,
2z−1 > 1 (draw this region carefully, note that some boundary components are contained
and some are not).
44
1 0 −1
∆
Theorem. (1) If ϕ
1
, ϕ
2
∈ Γ with ϕ
1
= ϕ
2
, then the images ϕ
1
(∆) and ϕ
2
(∆) are disjoint.
(2) The union
ϕ∈Γ
ϕ(∆) = H
+
(∆ is the “fundamental domain” for the action of Γ).
(3) Let Γ
1
be the subgroup of G of elements ϕ(z) =
az+b
cz+d
with a, d odd and b, c even. Then
Γ = Γ
1
.
Clearly, Γ is a subgroup of Γ
1
as σ, τ are in Γ
1
. Let (1
) be the statement (1) with Γ
1
in place of Γ. It thus suﬃces to prove (1
) and (2).
Proof of (1
)
If ϕ(z) =
az+b
cz+d
∈ G and z ∈ H
+
then Imϕ(z) = Im
az+b
cz+d
·
cz+d
cz+d
= Im
z(ad−bc)
cz+d
2
=
Imz
cz+d
2
.
To show (1
), it is suﬃces to show that ∆ and ϕ(∆) are disjoint for id = ϕ ∈ Γ
1
. If
c = 0, ad = 1, so a = d = ±1 and thus ϕ(z) = z + 2n in which case the conclusion is
obvious.
Suppose instead c = 0. We claim that cz + d ≥ 1 for z ∈ ∆. If not, ∆ intersects the
interior of the circle D = {w : cw +d ≤ 1} centered at −
d
c
with radius
1
c
.
If a circle with center on the real axis is to intersect ∆, it must contain at least one of
the points −1, 0, 1. So, at least one of the points −1, 0, 1 lies in the interior of D. But if
w = −1, 0, 1, cw + d is an odd integer, so cw + d cannot be less than 1 (i.e these three
points are possibly on the boundary, but deﬁnitely are not in the interior of D).
So Imϕ(z) ≤ Imz for z ∈ ∆. Suppose both z
0
and ϕ(z
0
) are in ∆ (this is where we
need to ﬁnd a contradiction). Applying the above to ϕ
−1
, we obtain the reverse inequality
Imz
0
≤ Imϕ(z
0
), so cz
0
+d = 1, i.e z
0
lies on the boundary of D.
45
Then D must be the circle centered at −
1
2
with radius
1
2
, i.e c = 2, d = 1, so
ϕ(z) =
az +b
cz +d
=
(1 + 2b)z +b
2z + 1
=
z
2z + 1
+b = τ
n
σ(z).
But then ϕ(z
0
) cannot possibly be in D (σ takes the semicircle {w ∈ H
+
: 2w + 1 = 1}
to the semicircle {w ∈ H
+
: 2w −1 = 1}; more precisely σ(−
1
2
+
1
2
e
iθ
) →
1
2
+
1
2
e
i(π−θ)
).
Proof of (2)
Let Σ be the union
ϕ∈Γ
ϕ(∆) ⊂ H
+
. Claim: Σ includes all z ∈ H
+
such that 2z −(2m+
1) ≥ 1 for all m ∈ Z. This is almost trivial, up to some boundary components, this region
is just
n∈Z
τ
n
∆ – but these components belong to
n∈Z
τ
n
σ∆.
Fix w ∈ H
+
. Choose a ϕ
0
∈ Γ which minimizes cw + d. Such a ϕ
0
exists: the
imaginary part of cw + d comes from cw, so c cannot be too big. Then d cannot be too
big either for otherwise the real part cw+d is large; so we really take minimum over ﬁnite
amount of terms.
Let z = ϕ
0
(w). Thus z has the largest imaginary part out of all {Γw}. Noting that
σ
−1
(z) =
z
−2z+1
, we consider:
(στ
−n
)z = σ(z −2n) =
z −2n
2z −4n + 1
, (σ
−1
τ
−n
)z = σ
−1
(z −2n) =
z −2n
−2z + 4n + 1
Hence 2z − 4n + 1 ≥ 1, 2z − 4n − 1 ≥ 1 for all integers n, i.e 2z − (2m + 1) ≥ 1 for
integral m. So z ∈ Σ implies w ∈ Σ.
5.4 Modular Function II
Last time, we considered a group Γ acting on H
+
generated by σ(z) =
z
2z+1
and τ(z) = z+2.
Theorem. The exists a holomorphic function λ(z) deﬁned on the upper halfplane such
that (1) λ(z) is invariant under the action of Γ, (2) λ is injective on the fundamental
domain ∆, (3) λ(∆) = C\{0, 1} and (4) R is the natural boundary of λ.
Condition (4) means that λ cannot be extended to a holomorphic function beyond the
real axis.
Let ∆
0
= ∆ ∩ {y > 0}. The Riemann mapping theorem tells us that there exists a
biholomorphism h : ∆
0
→ H
+
which extends continuously to a homeomorphism of closed
domains such that h(0) = 0, h(1) = 1, h(∞) = ∞.
46
We could try to use methods of Section 3.4 to argue that h extends continuously to
∆
0
(although inventiveness is needed to argue continuously of h near 0, 1, ∞). Its best
just to note that ∆
0
is simplyconnected and has Jordan curve boundary and appeal to
Carath´eodory’s theorem.
The Schwartz reﬂection principle allows us to continue h to all of ∆ by h(−x + iy) =
h(x +iy). Then h extends continuously to the ∂∆ and takes all possible values exactly
once on ∆.
Suppose z ∈ H
+
and λ(z), let ϕ ∈ Γ be the automorphism for which ϕ(z) ∈ ∆. Deﬁne
λ(z) = h(ϕ(z)).
As boundary values of h on ∂∆ are real, we have h(−1+iy) = h(1+iy) = h(τ(−1+iy))
and h(−
1
2
+
1
2
e
iθ
) = h(
1
2
+
1
2
e
i(π−θ)
) = h(σ(h(−
1
2
+
1
2
e
iθ
)). This tells us λ is continuous
on H
+
and holomorphic possibly except on
ϕ∈Γ
ϕ(∂G). A corollary to Morera’s theorem
tells us that actually λ is holomorphic on
ϕ∈Γ
ϕ(∂G) and thus holomorphic on H
+
.
To show (4), recall h(ϕ(0)) = h(0) = 0 for any ϕ ∈ Γ. If ϕ(z) =
az+b
cz+d
, we see that
λ(
a
c
) = h(0) = 0 for all even a and odd b. Hence zeros of λ are dense on the real line and
thus no holomorphic extension past the real line can be made.
5.5 Little Picard Theorem
We ﬁnally have all the tools to prove Picard’s Little Theorem. Suppose f omits the values
a = b. We may assume that a = 0 and b = 1 by replacing f by
f−a
f−b
if necessary.
Last time we have constructed the modular function λ : H
+
→ Ω = C\{0, 1} and
showed that it is a covering map.
Suppose D
0
⊂ Ω is a disk with center f(0). We can choose a disk E
0
⊂ C such that
f(E
0
) ⊆ D
0
. Next choose V
0
∈ H
+
so that λ
V
0
is a biholomorphism onto D
0
.
Let g
0
= (λ
V
0
)
−1
◦ f
E
0
. We will show that (g
0
, E
0
) has unrestricted analytic contin
uation to C. Indeed if γ : [0, 1] is any curve starting at 0, cover γ by a chain of disks
E
0
, E
1
, . . . E
n
so that f(E
j
) lies within some disk D
j
⊂ Ω.
For each disk D
j
, choose a V
j
so that that λ
V
j
is a biholomorphism onto D
j
and V
j
intersects V
j−1
. Then (λ
V
j
)
−1
◦ f
E
j
: D
j
→H
+
is the desired analytic continuation.
47
The Monodromy theorem gives us a holomorphic function g : C → H
+
. But such a
function must be constant (composing with
z−i
z+i
, we get a map from the complex plane to
the unit disk). Hence f itself must be constant.
Corollary. If f(z) is an entire function, then f◦f has a ﬁxed point unless f is a translation
f(z) = z +a (f need not have a ﬁxed point itself, e.g f(z) = z +e
z
).
Suppose f ◦ f has no ﬁxed points. Then f has no ﬁxed points either. Deﬁne g(z) =
f(f(z))−z
f(z)−z
. It is entire and omits values 0, 1 and hence constant, so g(z) = c with c = 0, 1,
i.e f(f(z)) = c(f(z) −z).
Diﬀerentiating, we see that f
(f(z))f
(z)−1 = c(f
(z)−1) so f
(z)(f
(f(z)−c) = 1−c.
As c = 1, f
(z) = 0 hence f
(f(z)) omits the values 0, c and hence is constant, i.e f
(z) is
constant, so f(z) = az +b. For f not to have ﬁxed points, a must equal 1.
Corollary. Any entire periodic function has a ﬁxed point.
Suppose f has period p, i.e f(z) = f(z + p). If f has no ﬁxed points then f(z) − z
omits the value 0. But it also omits the value p, otherwise f(z + p) = f(z) = z + p. By
Picard’s theorem, f(z) = z +c but such functions are not periodic.
Lemma. Let f, g be entire functions such that f
2
+g
2
= 1. There exists an entire function
h(z) so that f = cos(h(z)) and g = sin(h(z)).
The equation factors (f +ig)(f −ig) = 1. The function f +ig, f −ig are entire without
zeros, so f + ig = e
ϕ(z)
and f − ig = e
−ϕ(z)
, i.e f =
e
ϕ
+e
−ϕ
2
= cos ◦h with h =
ϕ
i
and
similarly, g = sin ◦h.
48
Chapter 6
Problems
1. Show the identities
(a)
∞
n=−∞
(−1)
n
(z −n)
2
=
π
2
(sin πz)(tan πz)
.
(b)
1
z
+
∞
n=1
(−1)
n
2z
z
2
−n
2
=
π
sin πz
.
(c)
∞
n=−∞
1
(z +n)
2
+a
2
=
π
a
·
sinh 2πa
cosh 2πa −cos 2πz
.
2. Prove the identity by taking logarithmic derivatives of both sides:
πx
∞
n=1
_
1 +
x
2
n
2
_
= sinh(πx).
3. Let F be the family of holomorphic functions in the unit disk satisfying f
(n)
 ≤ n!
for all n ≥ 0. Show that F is a normal family.
4. Suppose f is an entire function. Show that f has an nth root if and only if all zeros
of f have multiplicity divisible by n.
49
5. Suppose f(z) is a holomorphic function deﬁned on the unit disk with f(z) ≤ M.
Let the zeros of f are a
1
, a
2
, . . . a
n
counted with their multiplicities. Show that
f(z) ≤ M
¸
¸
¸
n
k=1
a
k
−z
1 −a
k
z
¸
¸
¸.
In particular, f(0) ≤ M
n
k=1
a
k
 and if f(0) = 0 then f(z) ≤ Mz.
6. (Blaschke product) Suppose f(z) is a holomorphic function deﬁned on the unit disk
with a zero at 0 of order s and the other zeros {a
k
} satisfying
k
(1 −a
k
) < ∞ (or
equivalently
k
log a
k
 > −∞). It then admits a nice factorization f = BG where
B is a product of
B(z) = z
s
∞
k=1
a
k

a
k
·
a
k
−z
1 −a
k
z
and G(z) is a holomorphic functions without zeros. Show that B(z) is holomorphic.
7. Show that a bounded holomorphic function admits a Blaschke product, i.e that the
sum
k
log a
k
 converges (Hint: use the ﬁrst corollary of Problem 5).
8. (Jensen’s formula) Suppose f(z) is a holomorphic function in the unit disk without
zeros. As log f(z) is harmonic, we have log f(0) =
1
2π
_
2π
0
log f(re
iθ
)dθ for 0 <
r < 1. But what if f(z) has zeros in the unit disk? Suppose that f(0) = 0 and denote
its zeros by {a
k
}. In this case,
log f(0) =
1
2π
_
2π
0
log f(re
iθ
)dθ −
a
k
<r
log
_
r
a
k

_
Here log f(0) is not equal to, but is actually less than the mean value. Such functions
are called subharmonic. Prove the above formula (Hint: ﬁrst consider r for which no
{a
k
} lie on z = r, also show that RHS is continuous).
9. (Canonical Product) Look back at the proof of Weierstrass’ theorem (Section 2.3) and
observe the following: Suppose f(z) is an entire function with zeros {a
k
} satisfying
k
1
a
k

m+1
< ∞ for some integer m. Then it is possible to choose all m
k
= m.
50
10. Prove that
sin πz
πz
=
∞
n=1
_
1 −
z
2
n
2
_
Hint: The Weierstrass theorem implies that
sin πz = ze
g(z)
n=0
_
1 −
z
n
_
e
z/n
To ﬁnd g(z), take the logarithmic derivative and use Example 2 from Section 1.2.
11. (Wedderburn’s lemma) Suppose f, g are entire functions without common zeros.
Show that there exists entire functions a, b such that af +bg = 1.
12. Find the residues at the poles of the Γ function (Hint: use the functional equation).
13. (BohrMollerup theorem) Suppose f : R
+
→R
+
is a logarithmically convex function
satisfying f(x+1) = f(x) and f(1) = 1. Then necessarily, f(x) = Γ(x) for all x > 0.
Hint: Show that for all natural n ≥ 2 and positive x
(n −1)
x
(n −1)! ≤ f(x +n) ≤ n
x
(n −1)!
and
n
x
n!
x(x + 1) · · · (x +n)
≤ f(x) ≤
n
x
n!
x(x + 1) · · · (x +n)
·
x +n
n
.
14. Show that the automorphisms of the upper halfplane which preserve i are given by
w(z) =
z + tan(θ/2)
1 −z tan(θ/2)
.
Write the formula for θ = π.
15. Find the automorphism group of C\{0}.
16. Show that two annuli are biholomorphic if and only if the ratio of their radii are the
same (complete the proofs below).
(a) Use the previous problem.
(b) Use the uniqueness of solution of the Dirichlet problem: suppose that f :
A(0; 1, r) → A(0; 1, R) is the biholomorphism and that f takes {z : z = 1}
to itself. Then log f(z) =
log R
log r
· log z. Treating it as an equation in variables
z, z, diﬀerentiate with respect to z.
51
(c) Given an annulus, A
r
= A(0; 1, r), we can consider its universal covering space
f
r
: H
+
→A
r
with f
r
given by
f(z) = exp
_
−2πi ·
log z
log λ
r
_
where λ
r
> 0 depends on r. The fundamental group of the annulus (which is
Z) acts on H
+
, with generator acting by multiplication by λ
r
. If two annuli A
r
and A
R
are biholomorphic, then these actions must be conjugate.
17. Show that two tori C/Γ
1
(with Γ
1
generated by e
1
, e
2
) and C/Γ
2
(with Γ
2
generated
by f
1
, f
2
) are biholomorphic if and only if there is a fractional linear transformation
with integer coeﬃcients and determinant 1 which takes (e
1
, e
2
) to (f
1
, f
2
).
18. (SchwarzChristoﬀel formula) Show that the mapping F(w) : H
+
→Ω (where Ω is a
polygon) given by the formula
F(w) = C
_
w
0
n−1
k=1
(w −w
k
)
−β
k
dw +C
is conformal for some distinct real numbers w
k
and
β
k
= 2.
19. Show that F(w) =
_
w
0
(1−w
n
)
−2/n
dw maps the unit disk onto the interior of a regular
polygon with n sides.
20. Find the image of the unit disk under the mapping F(z) =
1
z
n
k=1
(z −a
k
)
λ
k
where
λ
k
are positive with
n
k=1
λ
k
= 2.
21. Prove the addition theorem:
p(z
1
+z
2
) = −p(z
1
) −p(z
2
) +
1
4
_
p
(z
1
) −p
(z
2
)
p(z
1
) −p(z
2
)
_
2
.
22. Another form of the addition theorem: if u +v +w = 0 in C/Γ then
det
¸
¸
¸
¸
¸
¸
¸
1 1 1
p(u) p(v) p(w)
p
(u) p
(v) p
(w)
¸
¸
¸
¸
¸
¸
¸
= 0.
52
23. Suppose that f(z) is a even doubly periodic function (let the group of periods be Γ).
There exists points a
1
, a
2
, . . . a
n
; b
1
, b
2
, . . . b
n
∈ C/Γ such that
f(z) = c
n
k=1
p(z) −p(a
k
)
p(z) −p(b
k
)
.
24. Show that a doubly periodic function is a rational function of the p and p
(Hint: use
the previous problem).
25. Show that while the function f(z) =
n
z
2
n
is holomorphic in the unit disk, it does
not extend holomorphically to any larger open set (Hint: f(z
2
) = f(z) −z).
Remark. This problem is a special case of the Hadamard Gap theorem.
26. Find the radius of convergence of f(z) =
k≥1
z
2k
k+1
. Find a maximal domain of
existence (a maximal open set in C to which f(z) may be analytically continued).
27. The set of solutions (z, w) of w
2
− 2wz + 1 = 0 can be completed to a compact
Riemann surface over P
1
(C). Find the residues of the diﬀerential form
dz
√
z
2
−1
points
at inﬁnity.
28. Prove
_
1
0
dx
3
√
1 −x
3
=
2π
3
√
3
.
by integrating over a lift of γ in the Riemann Surface X ⊆ P
2
(C) given by the
equation x
3
+ y
3
= z
3
(show that the form being integrated is meromorphic and
apply the Residue theorem).
γ
1
j
j
2
0
53
29. Prove Picard’s theorem for meromorphic functions: if a meromorphic function deﬁned
on the entire complex plane omits three values, it is necessarily constant.
30. Show that a nonconstant holomorphic function deﬁned on C\{0} omits at most 1
value.
31. Show that the function f(z) = ze
z
attains every value from C.
32. Suppose f, g are meromorphic functions such that f
3
+ g
3
= 1. Show that actually
f and g are constant functions. Is result still true if 3 is replaced by a larger positive
integer?
33. Suppose f, g are entire functions satisfying e
f
+ e
g
= 1. Show that f, g are actually
constant functions.
54
Chapter 7
Further Results
7.1 Big Picard Theorem
To prove Big Picard Theorem, we need the concept of normal families of meromorphic
functions. A family of meromorphic functions S ⊂ M(Ω) is normal if it is every sequence
in S has a subsequence that converges uniformly in the spherical metric on compact subsets
of Ω.
Instead of using geodesic spherical distance, it is easier to use the equivalent metric
z −z

S
2 = min(z −z
, 
1
z
−
1
z
). Notation: We will always write S
2
as a subscript when
dealing with spherical distance.
Theorem. Suppose {f
n
} ⊂ M(Ω) which converges uniformly in the spherical metric to f.
(1) Then either f is meromorphic or identically inﬁnite. (2) Furthermore, if each f
n
is
holomorphic, the limit function f is also holomorphic or identically inﬁnite.
Remark. The functions f
n
(z) = n converge to the identically inﬁnite function, so the
holomorphic functions do not form a closed subspace of continuous functions C(Ω, S
2
)
taken with the compactopen topology. But if we do add the identically inﬁnite function,
the above theorem implies that we really do get a closed subspace.
Proof of (1): Case I. Suppose a ∈ Ω is such that f(a) = ∞. There is an for which
f(a) −f
n
(a) < when n > N is suﬃciently large. In particular, f
n
(a) < f(a) + = M
for some constant M.
As {f
n
}
n>N
∪{f} is compact, by the ArzelaAscoli theorem, we see that it is uniformly
S
2
equicontinuous. Let δ < d
S
2(M, ∞). There exists an r such that for z ∈ D(a, r),
55
f
n
(z) − f
n
(a)
S
2 < δ simultaneously for all n > N. It follows that {f
n
(z)} is bounded
uniformly in D(a, r). This means that near a, f(z) is a holomorphic function.
Case II. Now consider the case a ∈ Ω where f(a) = ∞. Let g
n
(z) = f
n
(
1
z
) and
g(z) = f(
1
z
). Then g(a) = 0 and g
n
converges in C(Ω, S
2
) to g. By Case I, we see
that there exists a ball D(a, r) on which g
n
are holomorphic (for n suﬃciently large) and
converge uniformly on compact subsets to g.
We have two cases. Either g is identically 0, in which case f is identically inﬁnite or
the zeros of g are isolated and so
1
g
= f is a meromorphic function with a pole at a.
Proof of (2): Now suppose that {f
n
} are actually holomorphic functions. To see that
the limit f is holomorphic, we must show that Case II implies that f is identically inﬁnite.
Suppose a ∈ Ω is such that f
n
(a) tends to inﬁnity. In this case, g
n
(z) never vanishes but
g(a) = 0, so by Hurwitz’s theorem, g(z) must vanish identically, i.e f is identically inﬁnite.
Theorem (MontelCarath´eodory). If a family of functions S ⊂ H(Ω) avoids two values
(say 0 and 1), it is normal.
Notation: D will denote the unit disk and λ : D → C\{0, 1} will be a holomorphic
covering map (compose the modular function constructed in Sections 5.3 and 5.4 with a
fractional linear transformation).
As normality is a local property, it suﬃces to treat the case when Ω is a disk (let the
center be a). Suppose {f
n
} ⊂ S. Passing to a subsequence, we may assume f
n
(a) → α.
First suppose that α = 1, 0, ∞. This allows us to consider U, a small ball centered at α
within C\{0, 1}.
As λ is a holomorphic covering map, we can lift each f
n
: Ω →C\{0, 1} to g
n
: Ω →D
so that g
n
(a) ∈ U. Since g
n
(z) ≤ 1, we can extract a subsequence g
n
k
which converges
uniformly on compact sets to a function g : Ω →D.
But actually g is a function from Ω to D – otherwise, the maximummodulus prin
ciple implies that g(z) = c is constant (of absolute value 1). But this would imply that
(λ
U
)
−1
(α) = lim
n
(λ
U
)
−1
(f
n
(a)) = lim
n
g
n
(a) = c which is absurd.
Suppose K ⊂ Ω is compact. On K, g(z) is bounded by some constant M < 1. As g
n
k
converges to g uniformly, there is a constant M
< 1 such that g
n
k
(z) < M
on K. But
the set λ(D(0, M
)) is bounded, so {f
n
k
} bounded uniformly on K. Since this is true of all
K, by Montel’s theorem, {f
n
k
} is normal.
56
Now we deal with the three special values of α:
Suppose α = 1. As f
n
don’t vanish, we choose g
n
=
√
f
n
: Ω →C so that g
n
(a) →−1.
But then g
n
(Ω) ⊂ C\{0, 1} and normality of f
n
is equivalent to normality of g
n
; so we
have reduced the case α = 1 to the general case.
If α = 0, let g
n
= 1 −f
n
; this reduces the case when α = 1.
If α = ∞, let g
n
=
1
fn
. Again g
n
are analytic and functions into C\{0, 1}; so have a
subsequence g
n
k
which converges uniformly on compact sets of Ω to some function g. But
g
n
k
have no zeros, so by Hurwitz’s theorem, g is identically 0; and thus f is identically
inﬁnite.
Proof of the Big Picard Theorem
Without loss of generality, we may assume that f has an essential singularity at the origin.
Suppose in some disk D(0, R), f(z) omits values 0, 1. Let Ω = D(0, R) \ {0}. Deﬁne
f
n
(z) = f(z/n). The MontelCarath´eodory theorem implies that f
n
is a normal family of
meromorphic functions.
This means that some subsequence {f
n
k
} converges to g uniformly on z =
R
2
. If
{f
n
k
} genuinely converges uniformly, then f(z/n) ≤ M for all integers n
k
. Now consider
the annulus A(0;
1
n
k+1
,
1
n
k
). On its boundary, f ≤ M, by maximummodulus, f must be
bounded by M in its interior. Hence f is bounded near 0, so we actually have a removable
singularity.
The other possibility is that g is identically inﬁnite. But this means that lim
z→0
f(z) =
∞ meaning that f really has a pole at 0.
7.2 Runge’s Theorem
Proposition. Suppose K ⊂ Ω is compact. Then there exists line segments σ
1
, σ
2
, . . . σ
n
lying in Ω\K such that f(z) =
1
2πi
n
k=1
_
σ
k
f(ζ)
ζ−z
dζ for all z ∈ K. Furthemore, the σ
k
may
taken of the same length and parallel to the coordinate axis.
For this purpose, we can cover K by a grid of squares of side length less than d(K, ∂Ω)
– see diagram. Let {Q
j
} be the squares which intersect K (then by hypothesis, the {Q
j
}
lie inside Ω).
57
K
Consider the sum
S(z) =
1
2πi
j
_
∂Q
j
f(ζ)
ζ −z
dζ
If z ∈ K, we are tempted to evaluate S(z) =
1
2πi
j
f(z)χ
Q
j
(z) = f(z). This calculation
has the slight problem that z may lie on a boundary of one of the Q
j
’s.
However, the sum in S(z) has a great many cancellations: if two squares in {Q
j
} are
adjacent then their common side will cancel out (due to opposite orientation). In fact, the
only segments of ∂Q
j
that will remain will lie outside of K. These are our σ
k
.
Exercise: One can arrange the σ
k
into ﬁnitely many disjoint cycles.
Lemma. Suppose K ⊂ C is compact. We can approximate any function f holomorphic
on a neighbourhood K uniformly by rational functions with poles lying outside K.
Suppose f is a holomorphic function deﬁned on Ω ⊃ K. By the above proposition, we
can choose segments {σ
k
} such that f(z) =
1
2πi
n
k=1
_
σ
k
f(ζ)
ζ−z
dζ holds for all z ∈ K.
Each integral
_
σ
f(ζ)
ζ−z
dζ may be approximated by its Riemann sum
1
n
n
j=1
f(ζ
j
)
ζ
j
−z
(here
ζ
k
lie on σ). By compactness of K and σ, this approximation is in fact uniform on K.
More precisely, as σ is compact, for a ﬁxed z ∈ K and any > 0, if the partition in
the Riemann sum is ﬁne enough, then the integral diﬀers from the Riemann sum by less
than . But the integrand is continuous in z (at least while z ∈ K) so by compactness of
K, we can make the partition so ﬁne to make Riemann sum approximate the integral
simultaneously for all z ∈ K.
58
Theorem (Runge). Suppose K ⊂ C is compact and E is a set containing a point from
every bounded component of C\K. Then any function f holomorphic on a neighbourhood
of K can be uniformly approximated by rational functions having poles in E.
To obtain the theorem from the lemma, it suﬃces to prove the following: if a, b are in
the same component of C\K then
1
z−a
can be approximated uniformly by polynomials of
1
z−b
. If we prove this claim we can “push” poles from bounded components of C\K to E
and poles lying in the unbounded component to inﬁnity.
By transitivity of uniform approximation by polynomials, it suﬃces to show if b ∈
D(a, r) then
1
z−b
can be uniformly approximated by polynomials in
1
z−a
on C\D(a, r).
Proof is left as an exercise for the reader.
7.3 Prescribing Zeros in Arbitrary Domains
Recall that to construct an entire function with zeros {a
k
} we took the product of G
m
k
_
z
a
k
_
where G
m
(w) = (1 −w) · exp
_
w+
1
2
w
2
+· · · +
1
m
z
m
_
. Let Ω ⊂ C be a domain and let {a
k
}
be points accumulating only at the boundary of Ω.
Suppose {b
k
} is a sequence of points outside Ω such that b
k
− a
k
 → 0 (a satellite
sequence). Exercise: Such a sequence exists if {a
k
} is bounded.
We propose the product f(z) =
G
m
k
_
a
k
−b
k
z
k
−b
k
_
. Suppose r > 0, for we work in
Ω
1/r
= {z ∈ Ω : d(z, ∂Ω) > 1/r and consider terms with ra
k
−b
k
 < 1:
g
k
(z) = −
1
m
k
+ 1
_
a
k
−b
k
z
k
−b
k
_
m
k
+1
−
1
m
k
+ 2
_
a
k
−b
k
z
k
−b
k
_
m
k
+2
−. . .
Thus,
g
k
(z) ≤
1
m
k
+ 1
_
ra
k
−b
k

_
m
k
+1
_
1 −ra
k
−b
k

_
−1
Just like in the proof of the regular Weierstrass theorem, we need to choose m
k
so that
(ra
k
−b
k
)
m
k
+1
converges for every r > 0. Again, m
k
= k suﬃces.
Remark. Unfortunately, if {a
k
} is unbounded, it is not always possible to choose a satellite
sequence. For instance, let Ω = H
+
and a
k
= ik. For this {a
k
} a satellite sequence does
not exist.
59
We continue the construction when {a
k
} happens to be unbounded. Pick a point a ∈ C
which happens not to be one of the designated a
k
, nor is one of their accumulation points.
The mapping h(z) =
1
z−a
will ensure that {h(a
k
)} is bounded.
Let
˜
f(z) be a holomorphic function with {h(a
k
)} as zeros. By construction f(z) =
˜
f(h(a)) has the desired zeros; trouble is that it may not be a holomorphic at a (and it may
not be possible to choose a outside of Ω). But actually, it is holomorphic at a!
But lim
z→∞
G
m
_
a
k
−b
k
z
k
−b
k
_
= 1 and as product converges absolutely and uniformly near
inﬁnity, we see that actually
˜
f(∞) = 1, i.e f(a) = 1.
Corollary. For any domain Ω ⊆ C, there exists a holomorphic function f(z) ∈ H(Ω)
which cannot be continued analytically to any larger domain.
Let f(z) be a function whose zeros accumulate to every boundary point of Ω (actually,
this is easier said than done).
It is even true that any domain Ω ⊆ C is a domain of holomorphy which means that
there is a holomorphic function f(z) ∈ H(Ω) such that for any point z
0
, the power series
expansion of f(z) converges only inside Ω.
For instance, the principal branch of log z (which is deﬁned on C without the negative
real ray) cannot be extended anywhere on this ray, but power series expansions do converge
beyond this ray.
7.4 Prescribing Singularities in Arbitrary Domains
We wish to construct a function f(z) meromorphic in a domain Ω with poles {b
k
} (accu
mulating only at the boundary of Ω) and principal parts P
k
(
1
z−b
k
).
Plan: write Ω =
i
K
i
as the increasing union of compact sets. Let L
i
= K
i+1
\K
i
and
B
i
be the b
k
’s which lie inside L
i
. By Runge’s theorem, exists a rational function p
k
(z)
with poles lying outside K
i+1
such that P
k
(
1
z−b
k
) −p
k
(z) < /2
i
on K
i
. We want to take
f(z) =
k
1
#B
i
_
P
k
_
1
z −b
k
_
−p
k
(z)
_
.
The contribution from the b
k
’s on L
j
to K
i
with j > i is less than /2
j
, hence the series
converges uniformly and absolutely implying that f(z) is meromorphic.
60
We have to insure that application of Runge’s theorem is valid; for this we need to
take K
1
“large enough” so that each bounded component of C\K
1
contains a bounded
component of C\Ω (then this will also be true of K
j
, j ≥ 2).
Lemma. Given a compact subset K ⊂ Ω, it is possible to choose a compact K
1
⊂ Ω
containing K such that every hole (compact component) of K
1
contains a hole of Ω.
If Ω = C, we can let K
1
be giant ball, so in what follows we consider Ω = C.
Let Ω
r
= {z ∈ Ω : d(z, Ω) ≤ r} where r is some number less than d(K, Ω). Alternatively,
we may describe the complement of Ω
r
:
C\Ω
r
=
_
z∈C\Ω
D(z, r).
In particular, Ω
r
is closed. We would take K
1
= Ω
r
, but Ω
r
may not be compact; so we
take be a giant ball B containing K and set K
1
= Ω
r
∩ B.
Let Z be a bounded component of C\K
1
. As C\B is a connected and unbounded
chunk of C\K
1
, it cannot intersect Z; for if Z has common points with C\B, Z would
have to contain C\B which is absurd. Hence Z ⊂
_
C\K
1
_
\
_
C\B
_
= C\Ω
r
.
Each D(z, r) which makes up C\Ω
r
is connected; so is either contained or disjoint from
Z (that is why we have shown Z ⊂ C\Ω
r
), this tells us that Z is the union of disks D(z, r)
which have center in Z, i.e
Z =
_
z∈Z∩(C\Ω)
D(z, r) =
_
z∈Z\Ω
D(z, r).
As Z is nonempty, the above representation tells us that Z intersects C\Ω, but then we
are done: suppose Z intersects C\Ω at a point p and if S is the connected component of
p in C\Ω, then as C\K
1
⊃ C\Ω, then Z would have to contain all of S as desired.
61
Preface
These notes of my lectures on Complex Analysis at the University of Toronto were written by Oleg Ivrii on his own initiative, after he took my course in the spring, 2007. The course is crosslisted as MAT1001S, one of the Core Courses in our Graduate Program, and as MAT454S, a fourthyear undergraduate course which is the second part of a twoterm sequence in complex analysis in our Specialist Program in Mathematics. Oleg took this course as a secondyear undergraduate. My lectures were based on the classical textbooks of Lars Ahlfors (Complex Analysis, Third Edition, McGrawHill 1979) and Henri Cartan (Elementary Theory of Analytic Functions of One or Several Variables, Dover 1994). My exposition closely follows these texts. Oleg has included the problems that I assigned as homework; most of these can be found in Ahlfors, Cartan or other textbooks. Oleg has added a concluding chapter as an introduction to several fundamental further results. I have included some of these topics when giving the course on earlier occasions, and they improve my lectures as presented here! During my thirtyﬁve years of teaching at the University of Toronto, I have been truly fortunate to have had so many inspiring students like Oleg. They are passionate about mathematics and they care for each other and the world around them. I am grateful to Oleg that he thought my lectures were interesting enough to want to write them up, and I give him my warmest wishes for a very happy and fruitful mathematical career.
March 2008
Edward Bierstone
2
Contents
1 Space of Holomorphic Functions 1.1 Topology on the Space of Holomorphic 1.2 Series of Meromorphic Functions . . . 1.3 The Weierstrass pfunction . . . . . . 1.4 The Weierstrass pfunction II . . . . . 1.5 Complex Projective Space . . . . . . . 1.6 The Elliptic Integral . . . . . . . . . . 1.7 Implicit Function Theorem . . . . . . 5 5 7 8 10 12 13 15 16 16 17 19 21 22 25 25 26 28 29 31
Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
2 Interpolation 2.1 Meromorphic functions with prescribed singularities 2.2 Inﬁnite products . . . . . . . . . . . . . . . . . . . . 2.3 Entire functions with prescribed zeros . . . . . . . . 2.4 The Gamma Function . . . . . . . . . . . . . . . . . 2.5 Normal families and compact subsets of C(Ω) . . . . 3 Automorphisms 3.1 Local Geometry of Holomorphic functions 3.2 Automorphisms . . . . . . . . . . . . . . . 3.3 Riemann Mapping Theorem . . . . . . . . 3.4 Riemann Mapping Theorem II . . . . . . 3.5 ChristoﬀelSchwarz Formula . . . . . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
. . . . .
3
. . . . . . . . . . . . . . .1 Analytic Continuation 5. . . . . . . . . . . . . .1 Complex Manifolds .5 Abel’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Further Results 7. . . . . . . . . . . . . . .4 Complex Geometry 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4 Prescribing Singularities in Arbitrary Domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5 Little Picard Theorem 6 Problems . . . .2 Complex Manifolds II . . . . . . . . . . . . . . . .6 Abel’s Theorem II . . . . . . . . . . . . 4. . . . . . . . . . . . .1 Big Picard Theorem . . . . . . . . . . . . .2 Monodromy Theorem 5. . . . . .4 Example: y = P (x) . . . . . . . . . . . . . . . . . . . . . . . . . . 7. . . . . . 7. . . . . . . . . 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5. . . . . . . . . . . . .3 Examples of Riemann Surfaces 4. . . . . . . . . . . . . . . . . . . . . 5. . . . .2 Runge’s Theorem . . . 4 . . . . . . . . . .3 Modular Function . . . . . . . . . . . . . . . . . . . 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7. . .3 Prescribing Zeros in Arbitrary Domains . . . . 5 A Modular Function 5. 4. .4 Modular Function II . . . . 33 33 34 36 37 39 40 42 42 43 44 46 47 49 55 55 57 59 60 . . . . . . . . . . . . . . . . . . . . . . . . . .
(2). f = lim fn is continuous. C(Ω) is metrizable (even by a translationinvariant metric). We can topologize C(Ω): a sequence of functions {fn } converges uniformly on compact sets if for every compact set K ⊂ Ω. Mi (f )) where Mi (f ) = maxz∈Ki f (z). . ) for n large enough for any given K. We give H(Ω) the subspace topology. z ∈ K}. C(Ω) is complete: if {fn } is a sequence of continuous functions which converge uniformly on compact sets. This is because fn → f uniformly on compact sets if and only if f − fn lies in V (K. Theorem. {fn K } converges uniformly. Furthemore. Write Ω = i Ki and let d(f ) = 1 i 2i · min(1. H(Ω) is a closed subspace.Chapter 1 Space of Holomorphic Functions 1. The fundamental system of open neighbourhoods at 0 is V (K. (1). In fact. Uniform convergence on compact sets deﬁnes the socalled compactopen topology.1 Topology on the Space of Holomorphic Functions Suppose Ω is open in C. the map f → f is continuous. 5 . ) = {f : f (z) < . let C(Ω) be the ring of continuous C–valued functions on Ω and let H(Ω) be the subring of holomorphic functions.
this is equivalent to showing that γ f (z)dz = 0 for any small loop γ. Choose disjoint neighbourhoods U and V of z0 and z1 respectively. Then its zeros are isolated. i. lim f (z) n→∞ n = lim 1 2πi fn (ζ) 1 dζ = (ζ − z)2 2πi f (ζ) dζ = f (z). In particular around any point z. As fn (z) are holomorphic. then {fn } converges uniformly on compact sets to f . It is enough to show that fn converges to f uniformly on a closed disk D(z. By Morera’s theorem. 1 n→∞ 2πi lim fn (z) 1 dz = fn (z) 2πi f (z) dz. Proposition (Hurwitz).(1) means that if {fn } ⊂ H(Ω) converges uniformly on compact sets. Corollary. f (z) vanishes nowhere in Ω. if {fn } ⊂ H(Ω) converges uniformly on compact sets and each fn vanishes nowhere in Ω.e connected). which is 0. Then. f (z) γ γ The left hand side counts the number of zeros of fn enclosed within γ. f (z) also does not possess any zeros inside γ. Proof of (2). Then. the sum is holomorphic and can be diﬀerentiated termbyterm. If a series of holomorphic functions ∞ fn (z) converges uniformly on comn=1 pact sets of Ω. Proof of (1). Let γ be a positively oriented loop around a disk of slightly larger radius. Then f (z) − a vanishes at a 6 . Hence. If Ω is a domain (i. But as the image of γ is compact. If Ω is a domain and {fn } ⊂ H(Ω) are onetoone and converges uniformly on compact sets. (ζ − z)2 n→∞ γ γ The limit above converges uniformly as ζ − z is bounded away from 0. Suppose f was not identically a constant. Proposition. then f = lim fn is never 0 or identically 0. f = lim fn ∈ H(Ω).e f (z) = 0. We have to show that f (z)dz is closed. Suppose f was not identically 0. f is bounded away from 0. By Cauchy’s integral formula. there is a small loop γ around z for which f does not vanish on the image of γ. γ f (z) = limn→∞ fn (z)dz because fn converges to f uniformly on the image of γ. but nevertheless f (z0 ) = f (z1 ) = a. then f = lim fn is either onetoone or a constant. γ fn (z)dz = 0 holds for all n. r) in Ω. (2) means that if {fn } ⊂ H(Ω) converges uniformly on compact sets and if f = lim fn . As z was arbitrary.
n<p fn + ∞ Theorem. each term is bounded by (x1 −n)2 .point in U . In fact the series converges uniformly and absolutely in each vertical strip x0 < x < x1 (we write z = x + iy). But then. 7 . 1 f (z + 1) = f (z). each term is bounded by 1 1 n>x1 (z−n)2 . 1 To see this. We deﬁne the sum on any relatively compact (precompact) open U as n≥p fn with each function in the second sum having no pole in U . If n=0 fn is a series of meromorphic functions that converges on compact ∞ subsets of Ω. then the sum f is a meromorphic function on Ω and n=0 fn converges uniformly on comapct subsets of U . The poles of f . To check the second property. so we can extract a subsequence of {fn } for which fn (z) − a vanishes at a point in U . see the calculation below. for a proof. Example 1 1 The series f (z) = ∞ n=−∞ (z−n)2 converges uniformly and absolutely on compact subsets of C. 1. note that in n<x0 (z−n)2 . there is a b such that f (z) < whenever y > b. By periodicity. we ﬁrst check that sin πz vanishes only on the integers. while in The function f (z) is meromorphic on C and enjoys three properties: (1) it is periodic. it suﬃces to check that the principal part of g(z) at the origin is z12 . each with principal part (z−n)2 and (3) f (z) → 0 as y → ∞ uniformly with repect to x. we extract another subsequence for which fn (z) − a vanishes at a point in V . From this subsequence. P (f ) are contained are in n P (fn ).e for all > 0. all but ﬁnitely many terms have no poles in K and form a uniform convergent series. 1 (x0 −n) . fn (z) would fail to be injective. The function g(z) = π sin πz 2 also enjoys these properties. i. The ﬁrst property is obvious. (2) it has poles z = n.2 Series of Meromorphic Functions A series of meromorphic functions ∞ fn on an open set Ω ⊂ C converges uniformly on n=0 compact subsets of Ω if for every compact K ⊂ Ω.
) z 6 3 But when z = 0. f − g is holomorphic in C because f and g have the same poles and same principal parts. . the function is 2 ∞ 1 n=1 n2 . Example 2 The series f (z) = 1 z + n=0 1 z−n + 1 n = 1 z + z n=0 n(z−n) converges uniformly and absolutely on compact subsets of C by comparison with 1 n=0 n2 . group terms with +n and −n). It is holomorphic in a neighbour 1 π2 1 1 π2 − 2 = 2 − 2 = . f (z) − π · cot πz is constant. dz Hence. e2 are are linear combination of e1 . We notice: f (z) = − 1 − z2 1 π =− 2 (z − n) sin πz 2 (i. i. 8 . .Easiest way to see property (3) is by means of the identity  sin πz2 = sin2 πx + sinh πy.e = n=0 d (π cot πz). We derive the formula ∞ 1 π2 n=1 n2 = 6 . sin πz → ∞ and hence g(z) → 0. as y goes to ∞. z + ∞ z 22z 2 = π cot πz. n=1 −n 1. e2 with determinant of matrix of coeﬃcients invertible in the ring of integers. 2 π − z12 = sin πz Consider the function hood of 0 and as z → 0. 1 z−n Thus f (z) is a meromorphic function with poles z = n and principal parts simple poles with residue 1). 3 (πz − 1 π 3 z 3 + . Hence. . i. so f − g is a constant by Liouville’s theorem. and by the third property. e2 are another set of generators for Γ if and only if e1 . .e ±1. which is 0 as both functions are odd (to see that 1 f (z) is odd.3 The Weierstrass pfunction Let Γ be a discrete subgroup of C with generators e1 . it tends to 1 n=0 (z−n)2 . the constant is 0. So. n2 ∈ Z}. f − g is bounded. e2 linearly independent over R.e Γ = {n1 e1 + n2 e2 : n1 . )2 z z (1 − 1 πz 2 + . Vectors e1 .
for this we estimate: 1 2zw − z 2 1 − 2 = 2 = .A function f (z) has Γ as a group of periods if f (z+n1 e1 +n2 e2 ) = f (z) for all n1 .. 1 The derivative p (z) = −2 w∈Γ (z−w)3 is doubly periodic. we see that 1 = w3 ∞ n=1 w∈Pn w∈Γ. But as p(z) is even. n2 ∈ Z. (z − w)2 w2 w∈Γ. plugging in z = − ei . n2 )}. we only need to estimate when w is suﬃciently large. a4 = 5 w6 . at w ∈ Γ. 9 . Also. We deﬁne p(z) = 1 + z2 1 1 − . . 2. We check that it converges uniformly and absolutely on compact subsets. .w=0 w3 < ∞. so p(z + ei ) − p(z) is constant. 2 (z − w) w w (z − w) We can restrict our attention to the compact subset z ≤ r. the Laurent expansion at 0 of p(z) is z12 + a2 z 2 + a4 z4 + . so p(z) also has 2 Γ as a group of periods. 3 (1 − z )2 w w3 w ·4 w 1 w∈Γ. As p(z) − z12 = g(z) is a holomorphic even function in a neighbourhood of the origin with g(0) = 0. Hence a2n = (2n + 1) 1 1 in particular a2 = 3 w4 . We deﬁne Pn = {n1 e1 + n2 e2 : max(n1 . 1 w=0 w2n+2 . As ﬁnitely many terms don’t matter. We continue: ··· = We want to show that z z(2 − w ) r· 5 10r ≤ 3 21 = . say w ≥ 2r.w=0 1 < w3 ∞ n=1 8n 8 = 3 3 n3 k k ∞ n=1 1 < ∞. 1 1 The 2nth derivative of (z−w)2 is (2n + 1)! (z−w)2n+2 . As the number of points in Pn is 8n. We let k be the distance between the origin to the closest point of P1 . we see that the constant is 0. To show that p(z+ei ) = p(z) for i = 1. we consider p (z + ei ) − p (z) = 0. n2 1 (z−w)2 We have thus shown that p(z) is a meromorphic function with pole p(z) is even..w=0 Note that the Weierstrass p function depends on the choice of Γ.
Let γ be the positively oriented loop around the period parallelogram. . If f (z) is a nonconstant meromorphic function having Γ as a group of periods. 10 . y) satisfying the polynomial equation y 2 = 4x3 − 20a2 x − 28a4 1 1 with a2 = 3 w4 and a4 = 5 w6 . z2 2 p (z) = − 3 + 2a2 z + 4a4 z 3 + . 1. z6 z p(z) = So. . Suppose f (z) is above. so it must be identically 0. . Then. . p (z)2 − 4p(z)3 + 20a2 p(z) + 28a4 is a holomorphic function vanishing at the origin. y = p (z) for a unique C /Γ. . the opposite sides of γ cancel each other and the integral reduces to 0. p (z)) parametrizes the algebraic equation y 2 = 4x3 − 20a2 x − 28a4 . Let its zeros inside the fundamental parallelogram be {αi } and its poles be {βj }. Proposition. . then the number of zeros of f (z) in a period parallelogram is equal to the number of poles (counting with multiplicity). e2 (linearly dependent over R). p (z) = z6 z 1 3a2 p(z)3 = + 2 + 3a4 + . Let X ⊂ C2 be the collection of pairs (x. . (x. i αi = j βj (mod Γ). The number of zeros minus the number of poles enclosed with in γ is calculated by the integral f (z) 1 2πi γ f (z) dz. as γ is doublyperiodic. However. The goal of this lecture it to show that the RHS has three distinct roots and any (x. p (z)2 − 4p(z)3 = − 20a2 − 28a4 + . Proposition.We now ﬁnd a diﬀerential equation for p (z): 1 + a2 z 2 + a4 z 4 + . z 8a2 4 2 − 2 − 16a4 + . . z2 Hence. As it is periodic. Thus. . it is bounded and by Liouville’s theorem is constant. y) ∈ X is given by x = p(z). . y) = (p(z).4 The Weierstrass pfunction II Suppose Γ is a discrete subgroup of C given by vectors e1 .
. then x = p(z). we write f (z) = (z−αi )k (c+. f (z) z − αi k zf (z) = (1 + . If (x. f (z) (z) 1 However. 2πi f (z) dz is an integer because it is a diﬀerence between two determinations of f log f (z) with same value of f (z). . . ). . y = p (z) and x = p(−z). Remark. .(z) 1 The value i αi − j βj is counted by the integral 2πi γ zf(z) dz. . y) ∈ X. Then f (z) k = (1 + . there are exactly two roots in a period parallelogram. f (z) = k(z−αi )k−1 (c+. ) and z = αi + (z − αi ). )(αi + . . p( e1 +e2 ). we can return back to our equation y 2 = 4x3 − 20a2 x − 28a4 . e2 and e1 +e2 (the points z such that 2z ∈ Γ but z ∈ Γ). but y = 0. ). If γ1 and γ2 are two f sides of γ along the vectors e1 and e2 starting at the origin respectively. Write a = p(z0 ). (2) If however 2z0 ∈ Γ.e roots of p(z) = a). For any value of a ∈ C. With these preparations. then p(z0 ) = p(−z0 ) are the two noncongruent zeros. . . . f (z) z − αi The considerations for poles {βj } are analogous. y = p (−z) are the two points which map to the same value of x by (2). these zeros are simple and the only ones there. z0 ∈ Γ as / Γ is the set of poles of p. 2 1 / The zeros of p (z) are e2 . 11 . we may choose x = p( e2 ). But as p (z) has a unique pole of order 3 inside a period parallelogram. Clearly. then p(z) − p(z0 ) has a unique zero of order 2 as p (z0 ) vanishes. The integral formulae used above may be obtained from the Residue theorem as follows: suppose ai is a zero. The 2 fact that they actually are zeros follows from p (z) being doublyperiodic and odd. then αi − i j βj = 1 2πi γ zf (z) 1 dz = − f (z) 2πi γ1 e1 f (z) 1 dz + f (z) 2πi γ2 e2 f (z) dz. but they map to diﬀerent values of y as −p (z) = p (−z). 2 1 If y = 0. We have two possibilities: (1) If 2z0 ∈ Γ. Now we examine the zeros of p(z) − a (i. ). These are distinct as the values they 2 take are unique by (1). p( e2 ).
but in classical form. .e y is a local coordinate. .5 Complex Projective Space The set X ⊂ C2 : y 2 = 4x3 − 20a2 x − 28a4 is a smooth curve as it is locally a graph of a function y = f (x) or x = g(y). xn ) where the hat xi xi notation as usual means that the variable is omitted. Right now. . if y0 = 0. . . . These are called homogenous coordinates for we are using n + 1 “coordinates” to represent an ndimensional object. . . yn ) → [y1 . so X = g(y) near (x0 . yn ]. y0 ). It is easy to see that P 1 (C) is just the Riemann sphere S 2 . . . y. . On the other hand. In other words. . λxn ) for some λ = 0 in C. . . this is left as an exercise). The complex projective space may be covered by coordinate charts Ui = {[x0 . . If (x0 . . . xn ] for the equivalence class of (x0 .e x is a local coordinate.1. . We compactify X ⊂ C2 to X ⊂ P 2 (C) by rewiting y 2 = 4x3 − 20a2 x − 12 . y0 = 0. 1. xn ] → ( x0 . . xn ). . Remark. The inverse map is (y1 . . xn ) if (x0 . xn ) = (λx0 . y2 . the implicit function theorem only gives that locally y = f (x) and x = g(y) are diﬀerentiable as functions of two real variables. . . . We write [x0 . . . i. . y0 ). We can think of P n (C) as the chart U0 ∼ Cn together with the set {x0 = 0} ∼ P n−1 (C) = = which is “the hyperplane at inﬁnity”. We return to this subtle point in a later lecture. then y = f (x) near (x0 . . . X is a onedimensional complex submanifold. . xn ] ∈ P n (C) : xi = 0}. we have to check that the transition maps are analytic. . t]. then p (x0 ) = 0 since p(x) has 3 distinct roots. . We thus get a complex (algebraic) manifold structure (actually. y0 ) ∈ X. yi−1 . These statements follow from the implicit function theorem. . they even turn out to be rational. . . xn ) ∼ (x0 . xi . . . i. whereas we want to imply that the mappings are actually holomorphic. . yi . . The ndimensional complex projective space P n (C) is deﬁned to be Cn+1 \{0} modulo the equivalence relation (x0 . we are interested in P 2 (C) = [x. xi ˆ Each chart Ui is homeomorphic to Cn by map [x0 .
By comparing coeﬃcients.4. In these coordinates. while near z = 0. . 1. . y) → [x. but all these options are equivalent to the point [0. t ) = (0. t] : y = 0}. we can take x as a local coordinate. 0] lies in the chart {[x. the point at inﬁnity. 1. 0]. c0 = 0 as t (0) = 0. we have seen that the map C /Γ → X given by z → [p(z). In Section 1. 1. X is X together with an added point at inﬁnity: if t = 0. this concludes the proof that X has complex manifold structure. 1. Indeed. p (z). in which X naturally embeds by (x. the coordinate p(z) functions p(z). y ) be the aﬃne y coordinates in this chart. y. Recall that X ⊂ P 2 (C) is the set of points y 2 t = 4x3 − 20a2 xt2 − 28a4 t3 . The variable y can be any complex number with the exception of 0. the functions p (z) .28a4 in homogeneous coordinates y t 2 =4 x t 3 − 20a2 x − 28a4 t and allowing t to be 0. . the equation of X is t = 4x 3 − 20a2 x t 2 − 28a4 t 3 . The next nonzero coeﬃcient comes from the term −20a2 x t 2 and is −20a2 x (4x 3 )2 = −320a2 x 7 . The map is holomorphic: if z = 0. 0). we see that c1 = c2 = 0 as well and c3 = 4x 3 . 1] = in usual coordinates in coords at inﬁnity 1 p(z) . . p (z). p 1 are (z) holomorphic.6 The Elliptic Integral Let Γ ⊂ C be a discrete subgroup and X ⊂ C2 be the set of points y 2 = 4x3 − 20a2 x − 28a4 . 0]. X = X ∪ [0. 1]. 1. the equation reduces to 4x3 = 0 from which we see x = 0. . 1. p (z) p (z) is a bijection of sets. Then. In a neighbourhood of the point at inﬁnity. t The point [0. 13 . If we let t = c0 + c1 x + c2 x 2 + . Thus. y. then t = 4x 3 − 320a2 x 7 + . t is a holomorphic function of x . but more can be said. The implicit function theorem tells us that in some neighbourhood of (x . The equation simpliﬁes to y 2 t = 4x3 − 20a2 xt2 − 28a4 t3 . t ) = ( x . Let (x . 1 are holomorphic.
0] 0 2 4 in C The form dz is closed but not exact. y = p (z). we ﬁnd that dy (12x2 − 20a2 )dx = 2ydy so dx = 6x2 −10a2 . the inverse function theorem guarantees us that the inverse is also holomorphic. so. . y) = (cos θ. 0] to p(z).Thus C /Γ → X is a holomorphic bijection from a compact space and a Hausdorﬀ space. so dz = dx y whenever y = 0.0) dy x 1−y 2 when y = ±1. 1. Trigonometric Functions For convenience. Remark. we mean the form which is − dx when y = 0 and √ dy y We then invert the integral θ= (1. = √ dy 1−y 2 by deﬁning sin θ by the formula sin θ 0 (cos θ. sin θ). From the parametrization x = p(z). Diﬀerentiating the equation y 2 = 4x3 − 20a2 x − 28a4 . we can also diﬀerentiate the parametrization: cos θdθ = d(sin θ) or xdθ = dy to obtain dθ = dy = √ dy 2 . x y But. we see that dx = ydz. We can parametrize X by (x. We say that the Weierstrass p function is given by “inversion” of the elliptic integral. Inverse deﬁnes z as a multivalued holomorphic function on X where 2 branches diﬀer by a constant in Γ: p(z) z dx √ . z= dz = 3 − 20a x − 28a 4x [0.sin θ) dy = x 14 dy 1 − y2 . In fact. x 1−y Thus. y By “dz” we mean the holomorphic diﬀerential form on X which is dx dz = √ 3 4x − 20a2 x − 28a4 on y = 0 and 6x2 dy − 10a2 on y = 0. Thus dz is welldeﬁned on all of X : functions y(x) and y (x) cannot vanish simultaneously. we repeat our considerations with X ⊂ C2 given by x2 + y 2 = 1. We can diﬀerentiate the equation x2 + y 2 = 1 to obtain xdx + ydy = 0. otherwise y(x) would have multiple roots but we have seen that it is not the case. by “dθ”. so a particular value of z not only determines p(z) but also the choice of path of integration from [0. so in particular. a homeomorphism.1. dy = − dx .
y).The form dθ is closed but not exact. For a ﬁxed x. By the Jacobian determinant ∂(z1 . Write x = x1 + ix2 . 0) to (cos θ. dz ∂x ∂y with holomorphic coeﬃcients (here we are using that ∂f = 0). zn ) in an open set Ω ⊂ Cn is holomorphic if f ∈ C 1 and ∂f df = n ∂zi dzi . zn . . The equation dz = ∂f dx + ∂f dy shows that that dy is a linear combination of dx.7 Implicit Function Theorem We say that a function f (z1 . so dz ∧ dz = 2 ∂f ∂y dy1 ∧ dy2 . sin θ) as well as the winding number of the curve from (1. y0 . a particular value of θ determines (cos θ. y) ∂y for y. this is the same as dz1 ∧ dz2 = det ∂f ∂y dy. z2 .” but showing that this condition is equivalent to the previous conditions is rather diﬃcult. x2 near (x0 . Actually. (2) f is analytic (locally represented by convergent power series). z = f (x. x1 . By the implicit function theorem (for realvalued functions). This is equivalent to: i=1 (1) f is continuous and holomorphic in each variable separately. y2 as C 1 functions in terms of z1 . . Put another way. i. y near (x0 . z2 ) ∂f = ∂(y1 . so θ is a multivalued function with branches whose difference is a multiple of 2π. This means the solution is ∂y holomorphic. 0). it is even suﬃcient just to say “holomorphic in each variable separately. y0 ). dz = ∂f ∂y dy and dz = ∂f ∂y dy = above equation. y = y1 + iy2 and z = z1 + iz2 . y) is a holomorphic function with ∂f (x0 . y2 ) ∂y > 0. y0 ) = 0 and we wish to solve the equations z = f (x. 1. sin θ). We discuss the implicit function theorem for 2 variables: suppose f (x. . A simple calculation shows that dx ∧ dx = −2idx1 ∧ dx2 (we will use this equation for variables y and z in place of x).e the 2 ∂f ∂y 2 dy ∧ dy. we see that we can solve y1 . 15 .
Over the Riemann sphere. 1 We can assume that no bk = 0. Let pk (z) be the sum of the terms of order up to mk where mk is chosen such that Pk 1 1 − pk (z) ≤ k z − bk 2 16 on z ≤ bk  2 . i.e with given poles and their singular parts. but over the complex plane. Suppose {bk } ∈ C are given and distinct with limk→∞ bk = ∞ and Pk (z) be polynomials without constant term.Chapter 2 Interpolation 2. we have a wealth of meromorphic functions such as tan z or sec z but inﬁnity is a limit of poles. Then there is a meromorphic function 1 in C with poles bk and principal parts Pk ( z−bk ). the only meromorphic functions are rational functions. Theorem (MittagLeﬄer). The function Pk ( z−bk ) is holmorphic in z < bk so we can expand it as a convergent power series in that disk. The most general meromorphic function with these poles and primitives parts is ∞ f (z) = k=1 Pk 1 − pk (z) +g(z) z − bk ♦ where pk (z) are polynomials chosen such that the series convergences and g(z) is an entire function.1 Meromorphic functions with prescribed singularities We now shift gears to constructing meromorphic functions with prescribed singularities.
Conversely. suppose pn converges to some p = 0. 17 . the diﬀerence of it and ♦ is holomorphic and conversely. 2. we ∞ shall associate the sum (S) n=1 log(1 + an ) where we take the principal branch of log. Actually. = log(1 + an+1 ) − (log pn+1 − log p) + (log pn − log p). 2πi(kn+1 − kn ) = log(1 + an+1 ) − log pn+1 + log pn . log pn+1 − log p. For sn to converge. If sn converges. The three terms log(1 + an+1 ). We will instead write the ∞ product as (P ) n=1 (1 + an ). The product (P ) and the sum (S) converge or diverge simultaneously. To the product. . the function ♦ + g(z) also has the desired singularities. Indeed.We show ♦ converges uniformly and absolutely on z ≤ r for any r. bn have a nonzero limit p. we need to know that kn stabilize. then so does pn as pn = esn . but after removing these few terms. we will write sn for the nth partial sum of (S). the partial products should have a nonzero limit. It suﬃces to compare ∞ Pk k=n+1 1 − pk (z) z − bk 1 1 . Fix a determination of log p = log p + i · arg p. pn A necessary condition for convergence is bn = pn−1 → 1. this is too restrictive: we should allow a ﬁnite number of terms to be 0. given any holomorphic function g(z). As the left hand side can take upon a discrete set of values. Choose determinations of log pn = log pn  + i · arg pn such that arg p − π < arg pn ≤ arg p + π. Thus ♦ is meromorphic in C with poles {bk } and principle parts Pk ( z−bk ). Chosen n such that bk  ≥ 2r if k > n. with 2k Given any other function with prescribed singularities. deﬁned on the complement of the negative real ray with log 1 = 0. it must be 0. . so the condition requires that an → 0. log pn − log p are small when n is large.2 Inﬁnite products We say that ∞ bn converges to p ∈ C if no bn are 0 and the partial products pn = n=1 b1 b2 . Then sn = log pn + 2πi · kn for integers kn . Proposition. Just as pn is the nth partial product of (P ).
Under the same hypothesis. Then f (z) = f = f1 f2 .2. The product f (z) = z ∞ (1 − n=1 z2 uniformly and absolutely on compact subsets of C (because does). Let U be a relatively compact subset of Ω and gq (z) = exp n>q log fn (z) . Hence. but the latter is equivalent log(1+z) ∞ to (S ) = 1. . sin πz = πz ∞ (1 − n2 ).We say that the product (P ) converges absolutely if (S) does. Example We develop an inﬁnite product for sin πz. Z(fn ). Additionally. Theorem. we see that c = π . Suppose fn (z) are holomorphic in Ω and absolutely on compact subsets of Ω. It follows that f (z) = c sin πz. n=1 = 1 and 18 . fq n>q fn (z). the Corollary. f (z) fn (z) gq (z) fn (z) n>q fn (z) n=1 n=1 The latter equality is justiﬁed as logn>q log fn converges uniformly to a (branch of) log gq and thus can be diﬀerentiated termbyterm (take q large enough so that fn have no zeros in U for n > q). n=1 an converging absolutely as limz→0 z Now. then Z(f ) = multiplicity of a zero of f is the sum of that of fn . suppose that fn (z) are continuous complexvalued functions deﬁned on an open set Ω ⊂ C. the series of meromorphic functions verges uniformly on compact subsets of Ω and limit is f f ∞ fn n=1 fn con . ∞ n=1 fn (z) converges uniformly ∞ n=1 fn (z) is holomorphic on Ω and and If Z(f ) denotes the zeros of function f . But as limz→0 1 z2 limz→0 sinzπz = π. we ﬁnd that 1 + z ∞ n=1 2z − n2 z2 ) n2 converges 1− z2 n2 = 1 + z z2 2z sin πz = π cot πz = 2 −n sin πz f (z) z by Example 2 in Section 1. n2 Diﬀerentiating logarithmically. We say that the product ∞ fn (z) converges uniformly (or absolutely) on a n=1 compact set K ⊂ Ω if log fn (z) does. . By the product rule: q q fn (z) gq (z) fn (z) fn (z) f (z) = + = + .
the form f (z) dz counts the number of zeros. There exists an entire function with prescisely {ak } as zeros. Suppose {ak } ∈ C. ak not necessarily distinct. 19 . a good form for pk (z) = exp z 1 z2 1 z mk + · 2 + ··· + · . The product converges if the sum of logarithms of its terms converge. 1− ak k=1 Here. i. f (z) = z m eg(z) 1− ak k=1 Here. we put the zeros at z = 0 separately in the factor z m . Theorem (Weierstrass). f So. dz f (z) so f (z) = const · e−g(z) but the constant can be absorbed into g(z). ak 2 ak mk amk k z ak z ak =− z 1 z2 1 z3 − · 2 − · 3 − . Any such function is of the form ∞ z · pk (z) . Now.e we need to make k log 1 − z ak + + 1 2 · z2 a2 k + ··· + 1 mk · z mk m ak k converge. so we know that we took the principal branch of logarithm and not some other branch (this would be automatic because we need gk to tend to 0 for convergence). we would need that −π < Im gk (z) < π.. it is exact.2. so we look at the expansion log 1 − Thus. Actually.. Then d f (z) −g(z) f (z) · e−g(z) = f (z) · e−g(z) − f (z) · e = 0. (z) Indeed. limk→∞ ak = ∞. but since there aren’t any. ak 2 ak 3 ak We need to show we can choose integers mk to make that the product converge. it is the derivative of an entire function g(z). we examine the case when an entire function has inﬁnitely many zeros. pk (z) are factors which guarantee that the product converges. An entire function with ﬁnitely many zeros {ak } (these not be necessarily distinct) can be written as n z f (z) = z m eg(z) .3 Entire functions with prescribed zeros An entire function f (z) with no zeros is of the form eg(z) for some entire function g(z). We denote the general term of this sum by gk (z).
F (z)g(z) is an entire function f (z). limk→∞ ak = ∞. gk (z) ≤ 1 r mk + 1 ak  mk +1 1 z m k + 1 ak mk +1 − 1 z mk + 2 ak mk +2 − . bk ∈ C. Every meromorphic function F (z) deﬁned on the entire complex plane is a quotient of entire functions.. we make two changes: (1) we change the order of g(z) at ak from mk to mk + 1 and (2) we change the principal bk 1 part of h(z) at ak to be g(z) + z−ak . for z ≤ r. h(z) − f (z) = bk has a root of order at least mk at ak .. 20 . g(z) Corollary. Then there exists an entire function f (z) such that ak is a root of order mk of f (z) = bk . Then. we consider terms with ak  > r: gk (z) = − Thus. then the product converges uniformly and absolutely in z ≤ r. mk ∈ N. the bk principal part of h(z) was the principal part of g(z) . g(z) has a zero of order mk . let g(z) be the entire function with the poles of F (z) as its zeroes. Corollary. that the sum has less terms with greater r). mk = k suﬃces. Clearly. Let g(z) be an entire function with zero of order mk at ak . 1− r ak  −1 It suﬃces to choose {mk } so that ( ark  )mk +1 converges for every r > 0 (note.Suppose r > 0. so F (z) = f (z) . bk g(z) is holomorphic which shows that To make sure that f (z) = bk has a root of order exactly mk at ak . Indeed. however. At z = ak . Suppose ak ∈ C. Write: f (z) = g(z)h(z) = bk + g(z) · h(z) − bk g(z) Now choose h(z) to be a meromorphic function with poles {ak } such that at each ak .
(z + n) −z ·n n! n n+1−z z2 ·z 1− 2 . hence g(z)g(1 − z) = sinzπz . We now prove the identity Γ(z)Γ(1 − z) = gn (z)gn (1 − z) = = π sin πz : z(z + 1) .. . n 21 sin πz π and .2. we have the estimate log fn (z) ≤ 2 · r2 r · 1− 2 n n −1 < 4r2 · 1 .4 The Gamma Function ∞ Consider the inﬁnite product: g(z) = z(1 + z) f1 (z) n=2 1+ z n 1− fn (z) 1 n z . n 2n 3n n 2n 3n If z ≤ r and n > 2r. and satisﬁes Γ(z +1) = zΓ(z).. Additionally.. n n z z2 z3 1 1 1 − 2 + 3 − . Indeed.. . from which we can see Γ(n+1) = n!. all of which are simple. (z + n) −z . . as n → ∞. (n + 1 − z) z−1 ·n n! By the Example in Section 2. . n2 1 n As 1 − 1 2 1− 1 3 . For this purpose.2. g(1) = limn→∞ gn (1) = limn→∞ n+1 = 1. −2. . n k k=1 (1 − z)(2 − z) . . 1 + · n−z = n 2 n n! gn (z) gn (z+1) The function g(z) has zeros at 0. Γ(1) = 1. n2 Hence the associated sum converges by comparison with Let gn (z) = n k=1 fk (z). . We want to show it converges uniformly and absolutely on compact subsets of C. −2. . . the inﬁnite product converges to n+1−z → 1. . . . + z(− − 2 − 3 − . . gn (z + 1) = (z+1)···(z+n+1) n−z−1 . −1. so n! n → ∞. we see 3 n gn (z) = z(1 + z) 1 + z z z(z + 1) . It is meromorphic with simple poles at 0. 1 . . we examine the sum of log fn (z) = log 1 + = z 1 + z log 1 − . g(z) Also. −1. n = zn z+n+1 → z as 1 Deﬁne the Euler gamma function Γ(z) as g(z) . . . 1 − = 1 2 1 · 2 · · · n−1 = n . g(z+1) = z. .
A set S ⊂ H(Ω) or of C(Ω) is bounded uniformly on compact subsets of Ω if for every compact K ⊂ Ω. n z+n This is positive when z is real = ∞ 1 n=0 (z+n)2 . k This suggests that g(z) = z ∞ 1+ k=1 z −z e k k · ezγ . Γ (z) Γ(z) ∞ n=1 1 1 − . S is normal if its closure is compact. A theorem of Bohr and Mollerup says that Γ is the only meromorphic function which is logarithmically convex on the positive real axis which satisﬁes the functional equation Γ(z + 1) = zΓ(z) and Γ(1) = 1. The logarithmic derivative of Γ(z) is 1 Γ (z) =− −γ+ Γ(z) z d Taking derivatives again. The convergence of the inﬁnite product can be established as before (we can again compare 1 1 1 with ). Recall that a metric space is compact if and only if every sequence has a convergent subsequence. We write gn (z) = z k=1 1+ z − z z(1+ 1 +···+ 1 −log n) 2 n e k e . 22 . its limit n2 is γ. This alternate expansion shows that 1 + 2 + · · · + n − log n converges. 2. so log Γ(z) is convex.We now give an alternate inﬁnite product for n 1 Γ(z) . We have seen that C(Ω) is metrizable and that H(Ω) is a closed subset of C(Ω) with the induced metric. we ﬁnd dz and positive. in other words.5 Normal families and compact subsets of C(Ω) Suppose Ω ⊂ C is open. the Euler gamma constant. there is a constant M (K) such that f (z) ≤ M (K) when z ∈ K for all f ∈ S. We say that S ⊂ H(Ω) is a “normal family” if every sequence in S has a subsequence that converges uniformly on compact subsets of Ω.
f (ζ) 1 Suppose z ∈ D. Let D(z0 . A set of functions S ⊂ H(Ω) is a normal family if and only if it is bounded uniformly on compact sets.Proposition. Let γ be the positively oriented boundary ˜ of D. {fn } converges uniformly on D. Let D be the open disk with center z0 and radius R and S ⊂ H(Ω) be bounded uniformly on compact subsets of D. We prove it using two lemmas. It suﬃces to show that for any z0 ∈ Ω. We pick D = D(z0 . f (z) = 2πi γ (ζ−z)2 dζ. Then a sequence {fn } ⊂ S converges uniformly on (k) compact sets if and only if for every k. It follows from the following theorem: Theorem (Montel). Lemma. Choose D with radius r/2 such that the disk D centered at z0 with radius r is contained in Ω. r) with radius r < R. the map H(Ω) → R : f → maxz∈K f (z) is continuous and hence it takes K to some bounded set of R. Let r < r0 < R. For a ﬁxed compact set K. For the “only if” direction. there is a small disk D centered at z0 such that for any f ∈ S. If S ⊂ H(Ω) or of C(Ω) is compact. The converse is only true in H(Ω). The fact that S is closed follows from H(Ω) being Hausdorﬀ. The mapping H(Ω) → H(Ω) : f → f takes a subset that is bounded uniformly on compact sets to another subset bounded uniformly on compact sets. We can write fn (z) as ∞ (k) fn (z0 ) · (z − z0 )k . Suppose M is M such that f (z) ≤ M on γ for all f ∈ S. It suﬃces to show that for any z0 ∈ Ω. As ζ − z ≥ r/2. By Cauchy’s integral formula. R) be a closed disk in Ω. k! k=0 23 . we simply need to know that the map f → f is continuous (we don’t need the hypothesis). {fn (z0 )} converges. Choose M so that fn (z) ≤ M for all n whenever z − z0  ≤ r0 . Now we prove the “if” direction. there is a small disk D ⊂ Ω centered at z0 ˜ such that S is bounded uniformly on D. the sequence of derivatives at z0 . r Lemma. then S is closed and bounded uniformly on compact sets. f (z) ≤ 2π · r42 · 2πr = 4M .
We are now ready to prove Montel’s theorem. then by Cauchy’s estimate. We then take N to be the diagonal sequence: N11 . . for each l. It suﬃces to ﬁnd a subseqence N ⊂ N such that limn∈N λk (fn ) exists for all k. Suppose S ⊂ H(Ω) is bounded uniformly on compact sets and {fn } ⊂ S. N22 . Then we diagonalize. i.e we ﬁrst choose N1 ⊂ N so that the limn∈N1 µ1 (fn ) converges. Let λj : H(Ω) → C take f → i f (k) (zi ). ank  ≤ l M k. As the geometric series converges. µl (fn ) i is bounded. r0 It follows fn (z) − fm (z) ≤ k=0 ank − amk  · rk + 2M k>l r r0 k . N33 . 24 . We then choose n0 large so that when m. . i. By the ﬁrst lemma.If we let the coeﬃcients that fn (z0 ) k! (k) be ank . ank − amk  are simultaneously small for all 0 ≤ k ≤ l. next we choose N2 ⊂ N1 so that limn∈N2 µ2 (fn ) converges. i We construct N as follows: We reindex λk as µl . ri ) ⊂ Ω. n > n0 . Then {fn (z)} converges uniformly as desired. Take a countable cover of Ω by disks D(zi . then N3 ⊂ N2 so limn∈N3 µ3 (fn ) converges and so on. we can choose l large enough so the second term is small. We wish to show {fn } has a convergent subsequence. .
25 . otherwise we work with f (z + z0 ) − f (z0 ). (2) Suppose instead f (z0 ) = 0. Then w = ζ p where ζ = zg(z) is a holomorphic coordinate change. Thus up to a conformal change of coordinates. every holomorphic function is of form ζ p + f (z0 ). (1) must always be the case. Lemma. The number p − 1 is called the ramiﬁcation index of f at z0 . We may assume z0 = 0. either of the two possibilities hold: (1) Suppose w0 = f (z0 ) and f (z0 ) = 0. Furthermore. i. But as f is injective. it is a genuine homeomorphism. every value suﬃciently close to f (z0 ) appears exactly p times (as this is true for ζ p + f (z0 )). f (z0 ) = 0. Then. so f is open. hence f is a local homeomorphism. By the inverse function theorem. Corollary. We may write w = z p f1 (z) with f1 (0) = 0 for some p ≥ 2. f −1 exists and is holomorphic in a neighbourhood of w0 and so f is conformal at z0 . w = (zg(z))p where g(z) is some determination of the pth root of f1 . A nonconstant holomorphic mapping is open. f −1 is holomorphic.1 Local Geometry of Holomorphic functions We say that a function f is conformal (biholomorphic) if f is holomorphic and has a holomorphic inverse. Since (2) cannot be injective.Chapter 3 Automorphisms 3. We know that in a small neighbourhood of f (z0 ). If f is a holomorphic function not identically constant.e a map of type (1). then f is a homeomorphism. If f ∈ H(Ω) and injective.
Then. Either. Theorem. b f has a unique ﬁxed point z = 1−a . this not guarantee that f is injective. Suppose f is a polynomial of degree n. a = 0. the ﬁrst set is open. Even if f (z) = 0 for all z ∈ Ω. We conclude f (z) = az + b. if a = 1. they are not biholomorphic. for any function f : C → D is constant by Liouville’s theorem. or (2) it is a polynomial (being a rational function with only pole at inﬁnity).Remark. We investigate the second case further. The ﬁrst case cannot happen: on one hand. Im{z < 1} and Im{z > 1} should be disjoint (as f (z) is injective). f (z) = w has n distinct roots except for special values of w (namely. Hence all biholomorphic maps Ω → Ω are given by composing a ﬁxed holomorphic map with an automorphism of Ω. (1) f (z) has an essential singularity at inﬁnity. but the essential singularity at inﬁnity implies the second set is dense which is impossible. then g −1 ◦ f : Ω → Ω is biholomorphic. 26 . but as f is open. f (z) is a translation. The automorphisms of the upper half plane are fractional linear transformations with real coeﬃcients and ad − bc > 0. g : Ω → Ω are two biholomorphic maps. we see that n = 1. Suppose w = f (z) is an automorphism of C. 3. f (z) = ez is onetoone on any strip a < Im z < b with b − a < 2π. The automorphisms of the complex plane are aﬃne maps z → az + b. Obviously. If a = 1. As f is injective. Notice that if f. Aut C are aﬃne maps w = az + b with a = 0. when f (z) = 0). While the complex plane C and the unit disk D = {z : z < 1} are homeomorphic. Examples The automorphisms of the Riemann sphere are the fractional linear transformations z → az+b cz+d .2 Automorphisms We classify automorphism groups of some domains in C. Remark. For instance.
ad−bc = 0 (the inverse cz+d dw−b is z = −cw+a ). the map z → z−i which takes H+ to D: it maps i to the origin and points 0. for T ∈ Aut S 2 . z − z0 1 − z0 z Applying the same argument to the inverse of f . The two are biholomorphic: indeed. Aut D are fractional linear transformations of the form w = eiθ · with z0  < 1. Then S ◦ T must be aﬃne which shows that T is a fractional linear transformation. so S(z) = eiα ·T (z). c Obviously. Let f (z) = S ◦ T −1 (z). Conjugation by z → z−i shows that Aut H+ is composed solely of fractional linear z+i transformations. Aut S 2 are fractional linear transformations w = az+b . 1. Then f (0) = 0 and f (z) < 1 when z < 1. (Note: w takes inﬁnity to a if c = 0 and to inﬁnity if c = 0). they take the unit disk into itself and their inverse is of the same form. ∞ to three real numbers z1 . By Schwarz’s lemma. By the equality condition of Schwarz’s lemma. 0. z−z0 it takes 1. z∞ . We want to show that actually G = Aut S 2 . f (z) ≤ z. The subgroup of Aut S 2 which ﬁxes the point at inﬁnity is precisely Aut C. we see that z ≤ f (z). If a fractional linear transformation w(z) is to take the real axis to itself. from which follows that α = 0 or S = T as desired. Let H+ be the upper halfplane and D the unit disk. z0 . ∞ z+i to −1. Let S = eiθ · 1−z00z with z0 = T −1 (0) and θ = arg T (z0 ). we see that f (z) = eiα ·z. so f (z) = z. z−z Suppose T ∈ Aut D. −z0 27 . so w(z) = z−z∞ · zz11−z∞ . The above transformations are automorphisms. we could consider a fractional linear transformation S which takes T (∞) → ∞. Aut H+ are fractional linear transformations with real coeﬃcients and positive determinant. −i. the fractional linear transformations form a subgroup of G ⊂ Aut S 2 . We now show that there are no other automorphisms. Theorem. Theorem.Theorem. As fractional linear transformations act transitively on the Riemann sphere. We now diﬀerentiate and set z = z0 : S (z0 ) = eiα · T (z0 ). 1 respectively.
we see g(z) ≤ z ≤ 1 and f (0) ≤ 1.1). and at z0 . Assuming the lemma. Then.3. If Ω isn’t the entire complex plane. z If additionally f (z0 ) = z0  for some z0 ∈ D. it contains a disk E centered at g(z0 ). recall that the map f → f (0) is continuous). To show that B is closed.3 Riemann Mapping Theorem Our next goal is to show that any simplyconnected subset Ω of C (except for C itself) is biholomorphic to the unit disk D = {z < 1}. By translating and scaling if necessary. z−a has a primitive g(z) in Ω. E and 1 its translate E + 2πi are disjoint. f (0) = 0. suppose f = lim fn with fn ∈ B (as usual. Injectivity follows from Hurwitz’s theorem (see Section 1. We ﬁrst reduce to the case of bounded domain. g(z)−[g(z0 )+2πi] is a biholomorphism of Ω onto a bounded domain. Let A = {f ∈ H(Ω) : injective. By Montel’s theorem. the functions are nonconstant as fn (0) ≥ 1 (from which follows that f (0) ≥ 1. f reduces to a constant which is impossible. For this purpose. As eg(z) is injective. we introduce B = {f ∈ A : f (0) ≥ 1}. The remaining lemma follows from Schwarz lemma which says that if f ∈ H(D). g must be constant. Then g(Ω) = D if and only if g (0) = supf ∈A f (0). We know that f (0) = lim fn (0) = 0. but if the value 1 is attained. it suﬃces to show that supf ∈A f (0) is attained. 28 . f (0) = 0. 1 then there is some point a ∈ Ω. we can assume that Ω contains the origin and lies inside the unit disk. As Ω is simplyconnected. It suﬃces to show that B is compact. clearly f (0) ≤ 1. Lemma. Indeed. z − a = eg(z) . It is easy to see that f (z) ≤ 1. As G(Ω) is open. then f (z) = λz with λ = 1: by the maximummodulus principle. λ has absolute value 1. f (z) < 1 then f (z) ≤ z for all z ∈ D. applying the maximummodulus 1 principle to g(z) = f (z) . by the maximummodulus principle. B is nonempty as f (z) = z is in it. we assume that the sequence converges uniformly on compact sets). Then. f (z) < 1}. B is normal as it is bounded uniformly on compact sets (its actually bounded everywhere by 1).
but it isn’t.e −1 < βk < 1 and β1 + · · · + βn = 2. Since Ω is simplyconnected. Boundary Behaviour Having shown that any two simply connected domains in the complex plane are biholomorphic. Let αk be the inner angle at zk . we see that f (0) ≤ g (0). 1−F (0)η where ψ(η) = Then h take the disk to itself and ﬁxes the origin. Then f = ϕ−1 ◦ θ ◦ F = ϕ−1 ◦ θ ◦ ψ −1 ◦ ψ ◦ F h(z) g(z) η−F (0) . Then. diﬀerentiating f = g ◦ h. i. f (z) < 1}. so 0 < αk < 2 (more z −zk precisely. Suppose Ω is a simplyconnected open set whose boundary is a polygonal curve with vertices z1 .4 Riemann Mapping Theorem II We deﬁned family A = {f ∈ H(Ω) : injective. We don’t win unless we see that h (0) is actually less than 1. We will do this in the special case of a polygon (although a theorem of Carath´odory shows e that this is true for any domains with Jordan curve boundary). zn+1 = z1 . αk is the value of the argument of zk−1 −zk between 0 and 2). the equality condition of Schwarz would imply that h is a rotation. i. Then h = f ◦ g −1 takes the disk to f (Ω) and takes the origin to itself. . The “only if” direction is quite clear: suppose g ∈ A with g(Ω) = D and f ∈ A is some other holomorphic map. We want to show that g(Ω) = D if and only if g (0) = supf ∈A f (0).e if θ(w) = w2 . It is easy to see that Ω is convex if and only if all βk > 0. f (0) = 0. we wish to see if the mapping extends a homeomorphism of closed domains. We want to show that there is some g ∈ A with g (0) > f (0). k+1 Let βk π be the outer angle at zk . 29 . . . it has a singlevalued holomorphic square root F (z). z2 . But if h (0) = 1. Suppose f ∈ A and a is some point in the disk but not in f (Ω). This allows us to apply Schwarz’s lemma: h (0) ≤ 1. ϕ ◦ f (z) = is nonvanishing. zn . Let ϕ(ζ) = f (z)−a 1−af (z) ζ−a 1−aζ be an automorphism of the disk which sends a to 0. so βk = 1 − αk . h (0) ≤ 1.3. By Schwarz’s lemma. we may write ϕ ◦ f (z) = θ ◦ F (z).
By arguments with the Schwartz reﬂection principle. f extends continuously at the angles as well. We can extend it to a function on Ω by setting v(z) = −v(z). We even show more: namely that f extends beyond each side as a holomorphic function. it must extend continuously up to the real axis. We ﬁrst show that f extends continuously at the sides of Ω. We take u(z) to be the “harmonic conjugate” which agrees with Re f (z) on Ω+ . we consider the function g(ζ) = f (zk + ζ αk ) deﬁned in the neighbourhood of the origin in the upper half plane (see picture). We straighten the angle at vertex zk of Ω. For this purpose. 30 . The function v(z) = Im f (z) is harmonic. But if f (z) is to admit any extension beyond the real axis. By a corollary of Morera’s theorem. Then u + iv is an extension of f (z). f (z) must be holomorphic on the real axis as well. if x ∈ R. Then f (z) can be completed to a holomorphic function deﬁned on all of Ω by setting f (z) = f (z). We prove that this is indeed the case. Suppose Ω = Ω+ ∪ Ω− is a domain symmetric about the real axis and f (z) is a holomorphic function deﬁned on Ω+ whose imaginary part tends to zero as z approaches the real axis. H+ At the angles.We will need a nonstandard version of the Schwarz reﬂection principle: Lemma (Schwarz). For this purpose. We now know that the formula f (z) = f (z) deﬁnes a function which is holomorphic on Ω+ ∪ Ω− and is continuous on the real axis. we can restrict to a small disk centered at x). The diﬃculty here is that a priori the real part of f (z) may not extend continuously to the real axis. This is easy. the situation is a little more tricky. It is harmonic as it satisﬁes the mean value principle. we assume that Ω is simplyconnected (this is a local question. g(ζ) extends to a holomorphic function in small disk centered at the origin and as f ∂Ω = gR . Simply compose f with the fractional linear transformation from ϕ : D → w−i given by ϕ(w) = w+i and apply the Schwartz reﬂection principle (exercise: ϕ does extend to a continuous map from D → H). We now return to the question of boundary behaviour of conformal mapping of a polygon Ω. Suppose f : Ω → D is conformal.
5 ChristoﬀelSchwarz Formula We can even ﬁnd an explicit formula of a Riemann mapping from a polygon to the unit disk. 31 . So f (∂Ω) must wind around once counter clockwise around the origin (details are left as an exercise). 3. but rather its inverse.z f w αk π F zk g zk + ζ αk ζ π 0 wk Thus we have shown that the map f : ∂Ω → ∂D is continuous. take a positively oriented curve γ ⊂ Ω which is close and homotopic to ∂Ω. Actually. we really want to see that f is genuinely onetoone on the boundary. which we denote F . Then f (γ) is a simple closed curve which goes around the origin with winding number +1. Any 3 of the wk ’s can be picked arbitrarily but the remainder are uniquely determined (exercise). the formula is not for f (z). However. Theorem (ChristoﬀelSchwarz). The above reasoning also allows us to conclude that f is locally onetoone. F is of the form F (w) = c where are c. For this purpose. w 0 n k=1 (w − wk )−βk dw + c We denote the image of zk under f by wk . c are constants.
As H(w) does not vanish. Taking arguments.Remark. we ﬁnd that F (w)(w − wk )βk = αk G(w) + (w − wk )G (w) is holomorphic and nonzero at wk . To this end. we see that arg H(w) = −θ + If H(w) = c (identically). we see that arg(w − wk ) = θ/2 + const (exercise. We claim that in fact H(w) is constant. we see that arg H(w) = arg F (w) + βk arg(w − wk ) The chain rule tells us that arg F (w) = arg(F ◦ w) (θ) − arg w (θ). it is invertible and so exists a power series expansion ζ = ∞ bm (w − wk )m with b1 = 0. w (θ) represents the direction of the tangent to circle at w = eiθ . it suﬃces to show that arg H(w) is constant on {w = 1} (recall arg is the imaginary part of log).e w (θ) = θ + π/2 = θ + const. extends continuously to the boundary and does not vanish in the closed unit disk. we can assume that wk = 1). By the discussion above. we show that arg H(w) is constant for w = eiθ with arg wk < θ < arg wk+1 and by continuity. βk θ + const = const. The formula always takes {z = 1} to a closed polygonal curve. Set H(w) = F (w) n (w − wk )βk . hint: by rotation. it is holomorphic in the k=1 unit disk. By elementary geometry. As w = g(ζ) is locally onetoone at the origin. then F (w) = c Christoﬀel formula by integration. Fix a vertex zk of Ω and consider its associated function g(ζ) constructed previously. The argument of (F ◦ w) (θ) measures the direction of the tangent to the graph of F which is constant as F traces a side of Ω. On the other hand. arg H(w) will be constant on the whole unit circle. We can now forget the auxiliary functions g and G. If the curve does not have selfintersections. Diﬀerentiating. 2 − wk )−βk and we obtain the Schwarz n k=1 (w 32 . then F takes the unit disk biholomorphically onto the interior of the curve. i. but the curve may have self intersections. Continuing our calculation. m=1 Raising to αk th power we see that ζ αk = F (w) − zk = (w − wk )αk G(w) where G(w) is holomorphic and nonzero near wk .
For complex structure. It is an exercise to the reader to verify that S 2 is biholomorphic to P 1 (C). A function on a complex manifold M is holomorphic if it is holomorphic in local coordinates (that is. As charts. In local coordinates. A complex manifold is a manifold with i complex structure.e if ψj ◦ f ◦ ϕ−1 is holomorphic for all coordinates charts ϕi of M i and ψj of N .1 Complex Manifolds A manifold is a paracompact Hausdorﬀ topological space locally homeomorphic to Rn . 33 . t) → 1+t (stereographic projection followed by conjugation). i A onedimensional complex manifold is called a complex curve or an abstract Riemann surface. V zU = z ◦ φU . a function f : M → N is holomorphic if it is in local coordinates. zV = z ◦ φV . we replace Rn with Cn and insist that the transition maps ϕj ◦ ϕ−1 : Vi → Vj be holomorphic for all i. Example: the Riemann sphere. Given two complex manifolds M and N . t) → x−iy x+iy 2 1−t (stereographic projection) and V = S \ {S} with ϕV : (x. for any coordinate chart ϕ : Ui → Vi . i. 1 The transition map ϕU ◦ ϕ−1 : C \{0} → C \{0} takes z to z . i. y. f ◦ ϕ−1 is holomorphic on Vi ).e we can cover M by open sets Ui for which exist homeomorphisms (charts) ϕi : Ui → Vi with Vi open in Rn . y.Chapter 4 Complex Geometry 4. we have zU zV = 1. we take U = S 2 \ {N } with ϕU : (x. j.
w vanishing at a. it is constant. g : M → N are holomorphic and coincide on a set with a limit point. It is not empty. for coordinates charts are homeomorphisms. a ∈ M and b = f (a). f −1 are holomorphic. then f = g on M . in a neighbourhood. (2) Suppose M is connected and f is holomorphic on M . these are holomorphic maps into the Riemann sphere. It is closed (given a sequence {zk }. Consider the set of points in which f. p : C → C /Γ : f → f ◦ p is a bijection between meromorphic functions on C /Γ and meromorphic functions on C with Γ as a group of periods. Properties of holomorphic functions extend to manifolds (1) If M and N are complex manifolds with M connected and f. b respectively. so f. Choose local coordinates z. In fact. g coincide in a neighbourhood. We are more interested in meromorphic functions. if f  has a relative maximum. g must coincide everywhere on M . We deﬁne the ramiﬁcation index to be the multiplicity of the root of w ◦ f ◦ z −1 . Thus a complex manifold is really a paracompact Hausdorﬀ space with an equivalence class of complex structures. If f  has a relative maximum. its tail lies in one chart). in particular. 4. It is open (automatic). The multiplicity does not depend on the choice of local coordinates. Two complex structures on M are equivalent if the identity mapping is an isomorphism. Suppose f : M → N is holomorphic.Given manifolds M and N with complex structures we say that f : M → N is an “isomorphism” (or a biholomorphism) if f is a homeomorphism of the underlying topological spaces and f. they are locally onetoone. it coincides with the constant function. use part (1). any holomorphic function on a compact connected manifold is constant. 34 . For example.2 Complex Manifolds II We have seen that holomorphic functions on compact complex manifolds are constant. for it contains the limit point.
Let Y be a complex curve. with 1 Γ containing no points of E. zi ) is a set of forms ω = fi (zi )dzi which transform properly. The Residue theorem carries over to manifolds (with same proof): Theorem. i.If the ramiﬁcation index is k.e dzi = fij (zj )dzj . a) = 2πi γ ω where γ is a small closed curved around a with winding number 1. then ω has a global primitive. we can choose local coordinates in which w = z k . If a ∈ E c and z is a local coordinate in a neighbourhood of a. )dz 2 where ω1 is a diﬀerential form holomorphic near a. . If M is simplyconnected. Given a compact set K ⊆ M with piecewise smooth oriented boundary Γ. if ω is a holomorphic diﬀerential form and Γ is the oriented boundary of a compact set then Γ w = 0. we have 2πi Γ ω = a∈K Res(ω. 35 . . we mean a connected complex curve X together with a nonconstant holomorphic map ϕ : X → Y whose ramiﬁcation points are isolated and the ﬁbers ϕ−1 (y) are discrete. i. a). This allows us to consider the oriented boundary of compact set. we may write ω = ω1 +( cz1 + z2 +. In particular.e a holomorphic function for which ω = dg (by deﬁnition dg is the form g (zi )dzi ). a holomorphic diﬀerential form has a primitive g. 1 We deﬁne Res(ω. A holomorphic diﬀerential form ω on manifold M with atlas (Ui . By a Riemann surface over Y . In a neighbourhood of any point. Complex manifolds have a natural orientation as the transition maps are orientationpreserving. Let ω be a diﬀerential form holomorphic in the complement of a discrete set E.
x = x. To construct a model of X. √ near (x0 . cut out their positive real rays and glue the top of one to bottom of the other and vica versa.e any point of Y has an open neighbourhood V such that ϕ−1 (V ) is a disjoint union of open sets Ui each mapped homeomorphically onto V . Similarly. y → ( 3 1 − y 3 . Here ϕ : X → Y is a covering space or an unramiﬁed covering space. take two copies of the complex plane.g to see that x is holomorphic. Let X ⊂ C2 be the points (x. i. x = 3 1 − y 3 . y0 ). Example 2: y = x1/2 We now construct a Riemann surface for the function y = x1/2 over Y = C.e t = log ◦ϕ. 36 . y) for which y 2 = x.4. Then t is a holomorphic function on X which is locally a branch of log. we may take x as a local coordinate. ϕ : X → Y given by ϕ(t) = et . The origin is a ramiﬁcation point: X lies in two sheets over Y everywhere except at the origin. We consider X = C. The map log z : Y → C is multivalued. y0 ) with y0 = 0. y0 ) with y0 = 0. x is the local coordinate. 3 1 − x3 ) with the branch of cube root corresponding to (x0 . the identity map. It is a complex manifold. near (x0 . X has 3 points over every point of C except at the cube roots of unity. We made the function log z singlevalued by introducing the Riemann surface. i. at x0 = 0. As functions on X. j 2 where j = e2πi/3 . Example 3: y = (1 − x3 )1/3 Let X ⊂ C2 be the set of points (x. 3 1 − x3 has a branch √ equal to y0 at x = x0 and 3 1 − x3 has a branch equal to x0 at y = y0 . y) is a local coordinate. 1. y is the local coordinate. x → (x. y) for which x3 + y 3 = 1. j. y0 ) with x0 = 0. if x0 = 0 and y0 = 0. √ To see that the two charts are compatible. near (x0 . both x and y are holomorphic: e.3 Examples of Riemann Surfaces Example 1: y = log x Let Y = C∗ be the complex plane without the origin.
The equation for X has multiple roots at a and inﬁnity. X has one point: if we plug z = 0 into the equation for X . . The equation deﬁnes ω when y y = 0. paying attention to the branch points and behaviour at inﬁnity.e where z = 0). We now investigate this in detail.4 Example: y = P (x) Let P (x) = (x − a1 )(x − a2 ) . 0. At inﬁnity. ω is implicitly deﬁned to be − x2 dy when x = 0. We can extend X to a Riemann surface over S 2 = P 1 (C). which are the branch points. 0] and [j 2 . . . We made the diﬀerential form √ dx 1−x3 singlevalued by introducing the Riemann surface. −1. [j. Diﬀerentiating the equation x3 + y 3 − 1 = 0. we see that 3x2 dx + 3y 3 dy = 0 or y y dx y = − x2 dy when both x = 0 and y = 0. y) for which y 2 = P (x). making cuts in each of them between a and ∞. . −1. −1. (x − an z). This construction shows that X is topologically a sphere. z z X is composed of X together with 3 points at inﬁnity (i. We complete it to a Riemann surface X ⊂ P 2 (C) over P 1 (C) by homogenizing the equation: y 2 z d−2 = (x − a1 z)(x − a2 z) . 0]. One can form a model of X by taking two copies of the Riemann sphere. Let X ⊂ P 2 (C) be the set of points [x. 37 . (x − ad ) be a polynomial with distinct roots. 4. so X : {y 2 = xz − a1 z 2 } is smooth at least in the ﬁnite part of P 2 (C).dx Consider the diﬀerential form ω = (1−x3 )1/3 = dx on X. 0]. y. Thus. namely [1. 0]. we get y 2 = 0 yielding the point [1. Case d = 1: The set X : {y 2 = x − a} is a parabola and hence a manifold. gluing the top bank of the ﬁrst sphere to the bottom bank of the second sphere and vica versa. z] given by x3 + y 3 = z 3 formed by homogenizing ( x )3 + ( y )3 = 1. Suppose X ⊂ C2 is the set of points (x.
= 0 or 38 . −1. ∞. after we dehomogenize with respect to v. we obtain the equation for X : {y 2 = (x − az)(x − bz)}. 0]. say a. There is only one point at inﬁnity. z) at inﬁnity. this is not smooth at the origin. 0]. we get the equation y 2 = x2 . we obtain the equation x3 = 0. [0. b and c. X has four branch points: a. In coordinates (y. d. i=1 We must choose (x. so X has two points at inﬁnity. ∂ ∂x Unfortunately. 0] and [1. To form a model for X . the equation for X is y 2 = (x−a)(x−b)(x−c) and for X . Case d = 3: Now. this case is pretty much the same as the previous case: if we take the equation y 2 = (x − az)(x − bz) and perform a linear change of coordinates to make it y 2 = uv. This shows that X is topologically a torus. make cuts between a and b and between c and d in both spheres and glue the banks of a − b and c − d to each other. y = ±(1 − az)1/2 (1 − bz)1/2 . b. 1.Case d = 2: The equation for X is {y 2 = (x − a)(x − b)}. If we set z = 0. for it is not true that either ∂ ∂z = 0. z) to be the coodinates at inﬁnity: z d−2 = (x − ai z). we take two copies of the Riemann sphere. X is a manifold and has branch points are a and b. Case d ≥ 4: The equation for X is y 2 z d−2 = d (x − ai z). c. namely [0. y 2 z = (x−az)(x−bz)(x−cz). Actually. 0]. So. If we set z = 0. namely [1. 1. so the implicit function theorem does not apply. pair up the points. we are left with the equation y 2 = u. so there is one point at inﬁnity. Homogenizing. 1.
4.5
Abel’s Theorem
Let X ⊂ C2 be given by the equation y 2 = P (x) with P (x) = 4x3 − 20a2 x − 28a4 having 3 distinct roots. Let X ⊂ P 2 (C) be given by the equation y 2 t = 4x3 − 20a2 xt − 28a4 . Then, X embeds into X ⊂ P 2 (C) by taking (x, y) to [x, y, 1]. We have seen that X has one point at inﬁnity, namely [0, 1, 0]. If (x , t ) → [x , 1, t ] are the coordinates at [0, 1, 0], then t = 4x 3 − 20a2 x t 2 − 28a4 t 3 . This tells us that t = 4x 3 − 320a2 x 7 + ... (see Section 1.6 for details). The projection map ϕ : X → C given by ϕ(x, y) → x gives X the structure of a Riemann surface over C. We extend it to ϕ which makes X a Riemann surface over P 1 (C) ∼ S 2 : = ϕ : [x, y, 1] → [x, 1], i.e to
x t
ϕ : [x , 1, t ] → [x , t ] in P 1 (C),
t x
∈ C in coordinates at 0 or to
in coordinates at inﬁnity.
dy The form ω = dx = 6x2 −10a2 has a primitive in a neighbourhood of each point of X; y globally the primitive is a multivalued holomorphic function z = z(x, y). Each branch of z in the neighbourhood of any point (x0 , y0 ) ∈ X is a local coordinate:
(1) if y0 = 0, using dz = (2) if y0 = 0, using dz =
dx y ,
dz 1 we see that dx = y = 0, dy 1 dz , we see that dy = 6x2 −10a2 = 0. 6x2 −10a2
1 Near the point at inﬁnity in X , [0, 1, 0], we have [x, y, 1] = [ x , t , 1], so t
ω=
dx x = t ·d y t
= dx −
x 12x 2 − 2240a2 x 6 + . . . dt = dx − · dx = −2dx (1 + g(x )). t 4x 2 − 320a2 x 6 + . . .
with g(x ) holomorphic near x = 0 and g(0) = 0.
1 Suppose that a2 , a4 come from a discrete subgroup Γ, i.e a2 = 3 w=0 w4 , a4 = 1 5 w=0 w6 . We have seen that the meromorphic transformation x = p(z), y = p (z) deﬁnes p a biholomorphism from C /Γ → X . Recall that the inverse map is z(p) = [0,1,0] ω where the integral is determined only modulo Γ.
We are interested in the converse: Theorem (Abel). Given a2 , a4 such that p(x) = 4x3 − 20a2 x − 28a4 has 3 distinct roots, there is a discrete subgroup Γ of C generated by e1 , e2 linearly independent over R such 1 1 that a2 = 3 w=0 w4 , a4 = 5 w=0 w6 . 39
We will prove it using two lemmas: Lemma (1). The diﬀerence of two branches of z belongs to some discrete group Γ. We can then consider the elliptic curve associated to Γ, i.e X given by equation 1 1 = 4x3 − 20b2 x − 28b4 where b2 = 3 w=0 w4 and b4 = 5 w=0 w6 . We can consider the composition: y2 X [x, y, 1]
?
/ C/Γ
∼ =
/X
z
[(p(z), p (z), 1]
Abel’s theorem says that this composition is in fact the “identity mapping” , i.e [p(z), p (z), 1] is nothing but the coordinate mapping [x, y, 1]. And therefore we must have a2 = b2 and a4 = b4 . It will turn out necessary to show ﬁrst the weaker statement ﬁrst: Lemma (2). The map z is a biholomorphism.
4.6
Abel’s Theorem II
Lemma 1
Let p0 = [0, 1, 0] be the point at inﬁnity, then z(p) = period of ω, i.e to π(γ) = γ ω where γ is a loop.
p p0
ω is globally welldeﬁned up to a
We can extend π to 1cycles by linearity. By Stokes theorem, the map π vanishes on boundaries: ∂σ ω = σ dω = 0, so π descends to a map from H1 (X , Z) → C. We now need two facts not from this course: ﬁrst, X is a torus, say by the degree genus formula g = (d−1)(d−2) = (3−2)(3−1) = 1 and secondly, H1 (X , Z) = Z ⊕ Z. 2 2 We claim that Γ is lattice. If not, Γ is contained within a straight line. Take α so that (απ(γ)) = 0. But then (αz) deﬁnes a singlevalued function harmonic on X . As X is compact, by maximummodulus, we see that it is constant. But z is not constant as ω does not vanish identically.
Lemma 2
First, we establish that z is surjective: as branches of z are local coordinates at each point of X , z is a local homeomorphism, in particular open, so has open image. But the image is also closed as X is compact. 40
Thus X is a covering space of C /Γ and so it must be biholomorphic to C /Γ for some subgroup Γ of Γ (and z is just the quotient map). We show that in fact Γ = Γ , i.e we need to prove the other inclusion. ˜ = The universal covering space of C /Γ is X ∼ C. The map z : X → C /Γ lifts to a map ˜ z between the covering space X and C /Γ. We have the following commutative diagram: ˜ ˜ = X ∼C
p z ˜
/C / C /Γ
qΓ
X ∼ C /Γ =
z
˜ ˜ We explain the map z . Let q0 ∈ X be a point above p0 . Given a point q ∈ X , let γ be ˜ ˜ an arbitrary path joining q0 to q. If γ = p(˜ ) then z (q0 ) = γ ω. Note that it is no longer γ ˜ necessary to quotient out by Γ to make the map welldeﬁned. Clearly z takes Γ into Γ for if q ∈ Γ , when we project the path connecting q0 to q on ˜ X , it becomes a loop and we know periods of ω lie in Γ. Denote the restriction z Γ by I. ˜ As every element of Γ is a periodic, the map I is surjective. This forces I to be injective (its a map from Z2 → Z2 ) and therefore bijective so the covering map z : C/Γ → C/Γ is trivial, i.e z is a biholomorphism.
Conclusion
The function x is meromorphic on X with a pole of order 2 at inﬁnity; indeed x x = . 3 − 320a x 7 + . . . t 4x 2 By lemma 2, we can think of x as a meromorphic function on C of z with Γ being the group of periods and having a double pole at z = 0. Write: c d x(z) = + + e + fz + . . . z2 z 2c d x (z) = − 3 − 2 + f + . . . z z dx 2 = y 2 = 4x3 − 20a x − 28a . Equating coeﬃcients, we see As dz = y , x (z) = y i.e x (z) 2 4 2 = 4c3 , 4cd = 12c2 d and d2 = 12c2 e + 12cd. It follows c = 1, d = 0, e = 0. that 4c x= Hence x(z) = p(z) as the diﬀerence is holomorphic and doubly periodic and vanishing at the origin. Diﬀerentiating, we also see that y(z) = p (z). Abel’s theorem is now completely proved. 41
D) along chains C1 = {D = D0 . Dn } which covers γ. . We say that (fn . D0 ) is a function element. Suppose that f ∈ H(D) where D is a disk centered at a. Suppose γ : [0. D0 ) along C and that (fi . . .Chapter 5 A Modular Function 5. . . We say that (f. Dn ) is an analytic continuation of (f0 . E1 . D1 . Di ) such that fi = fi+1 on Di ∩ Di+1 . Di ). ti+1 ]) ⊂ Di and γ(1) being the center of the last disk. Dn } (Di ∩ Di+1 = ∅) if there exist function elements (fi . Di+1 ) are direct analytic continuations of each other. En } covering γ to (gm . This means we have a partition 0 = t0 < t1 < · · · < tn+1 = n such that γ([ti . D) can be analytically continued along γ if it can be continued along a chain of open disks {D = D0 . . . Dm } and C2 = {D = E0 .1 Analytic Continuation Given a holomorphic function f deﬁned on a domain U . Lemma. Suppose (f0 . . D1 . we wish to ﬁnd the largest connected open set V ⊇ U to which we can extend f . (fi+1 . Dm ) and (hn . . 42 . A function element (f. D) has at most one analytic continuation along γ. that is f0 is a holomorphic function deﬁned on the disk D0 . . D0 ) may be analytically continued along a chain of disks C = {D0 . Suppose we can continue (f. D1 . . we need to show that gi = hj in Di ∩ Ej whenever Di and Ej intersect. . We say that (f0 . 1] → C is a curve with f (0) = a. En ) respectively. .
5. More generally. Proof of (1). (2) If Ω is simplyconnected. . D) may analytically continued along each γs to (γs . Need to show that g0 = g1 (in D0 ∩ D1 ). D) along γ0 and γ1 coincide. g = g ◦ h as they coincide on σ(U ) = σ (U ) (g ◦ σ = f = g ◦ h ◦ σ). E1 . By the uniqueness of analytic continuation.e g ◦ σ = f ). . Without loss of generality. so then gi = hj continues to hold in Di ∩ Ej . Since γ is uniformly continuous. It follows that gi = hj in Di−1 ∩ Di ∩ Ej which is not empty. 1] to C \Ei . Suppose (f. . Ds ).2 Monodromy Theorem Theorem (Monodromy). Choose to be less than the distance from γs [0. suppose U ⊂ C is a domain and f ∈ H(U ). 43 . (3) (X.Suppose not. 1] be a homotopy with endpoints a. If (X . Let C = {D = E0 . b. n. We explain condition (3). ti+1 ]) ⊂ Ei for i = 0. This means that there is a partition 0 = t0 < t1 < · · · < tn+1 = 1 with γ([ti . consider the indices (i. s ∈ [0. We will ﬁnd a δ > 0 such that if s − s < δ. Fix s. . We wish to ﬁnd an unramiﬁed Riemann surface (X. 1]. (2) f extends to a holomorphic function g on X (i. In this case. we require that there is a unique holomorphic map h : X → X such that = ◦ h and σ = σ ◦ h. γ1 are homotopic curves in Ω from a to b. Then (f. 1. Then (1) If γ0 . The functions gi−1 and gi agree in Di−1 ∩ Di while gi−1 and hj agree on Di−1 ∩ Ej . then there is a holomorphic function g on Ω such that g = f on D. then C covers γs as well. En } be a chain of disks covering γs . By compactness of [0. 1] whenever s − s  < δ. ) is the “largest” Riemann surface satisfying (1) and (2). Let γs . σ ) also satisfy (1) and (2). . it would follow that gs = gs for s − s  < δ. then continuations of (f. j) with the smallest i + j for which this fails. we can assume that si ≥ tj . we would have g0 = g1 . . D) can be analytically continued along any curve in Ω starting at a (the center of D). We shall not address this more general question. we have a δ > 0 such that γs (t) − γs (t) < for all t ∈ [0. . ) over C and a biholomorphing mapping σ which takes U onto an open subset of X such that: (1) ◦ σ is the inclusion map of U into C.
From the construction. En } with En containing b and analytically continue (f. − z . we will construct a modular function invariant z under the subgroup Γ ⊂ G generated by σ(z) = 2z+1 and τ (z) = z + 2. The Little Picard theorem is a special case of the Big Picard theorem: if f is entire. Z) acts on the upper halfplane H+ : if ϕ ∈ G given by the matrix (a. But we can just use precisely the same chain of disks! 5. either f has a pole (in which case. Let ∆ ⊂ C be the region: y > 0. d) with ad − bc = 1 then ϕ(z) = az+b . then in any neighbourhood of a. 2z + 1 ≥ 1. note that some boundary components are contained and some are not). inﬁnitely many times. The modular group G = SL(2. Suppose z = x + iy. If f has an essential singularity at z = a. D) to obtain the function element (gb . then the analytic continuation of (f. D) near b matches up with (gb . . cz+d 1 Exercise: G is generated by z + 1. with one possible exception. the construction of the function element (gb . En ). for instance f (z) = ez omits 0). a modular function is a holomorphic or a meromorphic function deﬁned on H+ which is invariant under the action of G (or a subgroup of G). −1 ≤ x ≤ 1. . 44 . c. To prove the Little Picard Theorem. To deﬁne g near b ∈ Ω. E1 .Proof of (2). Theorem (Big Picard). b. f takes every complex value. . at inﬁnity. choose any chain of disks C = {D = E0 .3 Modular Function We will introduce the modular function to show the following results: Theorem (Little Picard). Losely speaking. En ) does not depend on the choice of C. 2z −1 > 1 (draw this region carefully. it is clear various gb ’s patch together: we have to check that if b ∈ En is some other point. By part (1). it is a polynomial) or it has an essential 1 singularity (so f ( z ) has an essential singularity at the origin). En ). A nonconstant entire function f (z) omits at most one value (it is possible that one value is omitted.
∆ −1 0 1 Theorem. 0. we obtain the reverse inequality Im z0 ≤ Im ϕ(z0 ). Proof of (1 ) If ϕ(z) = az+b cz+d ∈ G and z ∈ H+ then Im ϕ(z) = Im az+b · cz+d cz+d cz+d = Im z(ad−bc) = cz+d2 Im z . Suppose both z0 and ϕ(z0 ) are in ∆ (this is where we need to ﬁnd a contradiction). (1) If ϕ1 . (2) The union ϕ∈Γ ϕ(∆) = H+ (∆ is the “fundamental domain” for the action of Γ). cw + d is an odd integer. 1. ad = 1. it is suﬃces to show that ∆ and ϕ(∆) are disjoint for id = ϕ ∈ Γ1 . it must contain at least one of the points −1. So Im ϕ(z) ≤ Im z for z ∈ ∆. so a = d = ±1 and thus ϕ(z) = z + 2n in which case the conclusion is obvious. 0. c If a circle with center on the real axis is to intersect ∆. so cw + d cannot be less than 1 (i. 0. i.e z0 lies on the boundary of D. So. We claim that cz + d ≥ 1 for z ∈ ∆. But if w = −1. It thus suﬃces to prove (1 ) and (2). at least one of the points −1. c even. cz+d2 To show (1 ). Let (1 ) be the statement (1) with Γ1 in place of Γ. 45 . but deﬁnitely are not in the interior of D). Clearly. d odd and b. Suppose instead c = 0. τ are in Γ1 . 1. (3) Let Γ1 be the subgroup of G of elements ϕ(z) = az+b with a. 1 lies in the interior of D. then the images ϕ1 (∆) and ϕ2 (∆) are disjoint. so cz0 + d = 1. ∆ intersects the 1 interior of the circle D = {w : cw + d ≤ 1} centered at − d with radius c . If not. ϕ2 ∈ Γ with ϕ1 = ϕ2 . If c = 0. Then cz+d Γ = Γ1 .e these three points are possibly on the boundary. Applying the above to ϕ−1 . Γ is a subgroup of Γ1 as σ.
Thus z has the largest imaginary part out of all {Γw}. 2z − 4n + 1 (σ −1 τ −n )z = σ −1 (z − 2n) = z − 2n −2z + 4n + 1 (στ −n )z = σ(z − 2n) = Hence 2z − 4n + 1 ≥ 1. Condition (4) means that λ cannot be extended to a holomorphic function beyond the real axis.Then D must be the circle centered at − 1 with radius 1 . σ −1 (z) Let z = ϕ0 (w). we consider: z − 2n . so 2 2 ϕ(z) = az + b (1 + 2b)z + b z = = + b = τ n σ(z). 5. so we really take minimum over ﬁnite amount of terms. Such a ϕ0 exists: the imaginary part of cw + d comes from cw. h(∞) = ∞. h(1) = 1. Claim: Σ includes all z ∈ H+ such that 2z − (2m + 1) ≥ 1 for all m ∈ Z. Noting that z = −2z+1 .e c = 2. so c cannot be too big. The Riemann mapping theorem tells us that there exists a biholomorphism h : ∆0 → H+ which extends continuously to a homeomorphism of closed domains such that h(0) = 0. d = 1. 2 2 2 Proof of (2) Let Σ be the union ϕ∈Γ ϕ(∆) ⊂ H+ . Let ∆0 = ∆ ∩ {y > 0}. we considered a group Γ acting on H+ generated by σ(z) = and τ (z) = z+2. this region is just n∈Z τ n ∆ – but these components belong to n∈Z τ n σ∆. So z ∈ Σ implies w ∈ Σ. cz + d 2z + 1 2z + 1 But then ϕ(z0 ) cannot possibly be in D (σ takes the semicircle {w ∈ H+ : 2w + 1 = 1} 1 to the semicircle {w ∈ H+ : 2w − 1 = 1}. i.e 2z − (2m + 1) ≥ 1 for integral m. (3) λ(∆) = C \{0. (2) λ is injective on the fundamental domain ∆. The exists a holomorphic function λ(z) deﬁned on the upper halfplane such that (1) λ(z) is invariant under the action of Γ. This is almost trivial. Theorem. more precisely σ(− 2 + 1 eiθ ) → 1 + 1 ei(π−θ) ). 46 . up to some boundary components. Then d cannot be too big either for otherwise the real part cw + d is large. i. 2z − 4n − 1 ≥ 1 for all integers n. 1} and (4) R is the natural boundary of λ.4 Modular Function II z 2z+1 Last time. Choose a ϕ0 ∈ Γ which minimizes cw + d. Fix w ∈ H+ .
For each disk Dj . If ϕ(z) = az+b . Then (λVj )−1 ◦ f Ej : Dj → H+ is the desired analytic continuation. . E0 ) has unrestricted analytic continuation to C. let ϕ ∈ Γ be the automorphism for which ϕ(z) ∈ ∆. Its best just to note that ∆0 is simplyconnected and has Jordan curve boundary and appeal to Carath´odory’s theorem. f −b Last time we have constructed the modular function λ : H+ → Ω = C \{0. 1. we see that cz+d λ( a ) = h(0) = 0 for all even a and odd b. e The Schwartz reﬂection principle allows us to continue h to all of ∆ by h(−x + iy) = h(x + iy). Let g0 = (λV0 )−1 ◦ f E0 . Then h extends continuously to the ∂∆ and takes all possible values exactly once on ∆. As boundary values of h on ∂∆ are real. En so that f (Ej ) lies within some disk Dj ⊂ Ω. 1} and showed that it is a covering map. Hence zeros of λ are dense on the real line and c thus no holomorphic extension past the real line can be made. Indeed if γ : [0. Deﬁne λ(z) = h(ϕ(z)). Next choose V0 ∈ H+ so that λV0 is a biholomorphism onto D0 . A corollary to Morera’s theorem tells us that actually λ is holomorphic on ϕ∈Γ ϕ(∂G) and thus holomorphic on H+ . Suppose D0 ⊂ Ω is a disk with center f (0). We can choose a disk E0 ⊂ C such that f (E0 ) ⊆ D0 . 5. 1] is any curve starting at 0. ∞). E1 . 47 . To show (4). choose a Vj so that that λVj is a biholomorphism onto Dj and Vj intersects Vj−1 . Suppose f omits the values a = b. recall h(ϕ(0)) = h(0) = 0 for any ϕ ∈ Γ. This tells us λ is continuous 2 on H+ and holomorphic possibly except on ϕ∈Γ ϕ(∂G).4 to argue that h extends continuously to ∆0 (although inventiveness is needed to argue continuously of h near 0. cover γ by a chain of disks E0 . We may assume that a = 0 and b = 1 by replacing f by f −a if necessary. .We could try to use methods of Section 3. . Suppose z ∈ H+ and λ(z). we have h(−1+iy) = h(1+iy) = h(τ (−1+iy)) 1 1 1 1 1 and h(− 2 + 2 eiθ ) = h( 2 + 1 ei(π−θ) ) = h(σ(h(− 2 + 2 eiθ )). We will show that (g0 .5 Little Picard Theorem We ﬁnally have all the tools to prove Picard’s Little Theorem.
Suppose f ◦ f has no ﬁxed points. we get a map from the complex plane to z+i the unit disk). Diﬀerentiating. But such a function must be constant (composing with z−i . f (z) = 0 hence f (f (z)) omits the values 0. otherwise f (z + p) = f (z) = z + p. 48 . 1. f (f (z))−z f (z)−z . e. Then f has no ﬁxed points either. then f ◦f has a ﬁxed point unless f is a translation f (z) = z + a (f need not have a ﬁxed point itself. Hence f itself must be constant. we see that f (f (z))f (z) − 1 = c(f (z) − 1) so f (z)(f (f (z) − c) = 1 − c. so g(z) = c with c = 0. If f has no ﬁxed points then f (z) − z omits the value 0. The function f + ig. For f not to have ﬁxed points. f (z) = z + c but such functions are not periodic. Lemma. Corollary. Corollary. If f (z) is an entire function. 1 and hence constant. g be entire functions such that f 2 + g 2 = 1. g = sin ◦h. Any entire periodic function has a ﬁxed point. i.g f (z) = z + ez ). f − ig are entire without ϕ −ϕ zeros. But it also omits the value p. i.The Monodromy theorem gives us a holomorphic function g : C → H+ . so f (z) = az + b.e f (z) is constant.e f = e +e = cos ◦h with h = ϕ and 2 i similarly. i. i. Suppose f has period p. so f + ig = eϕ(z) and f − ig = e−ϕ(z) .e f (f (z)) = c(f (z) − z). The equation factors (f + ig)(f − ig) = 1. Let f. c and hence is constant. a must equal 1. By Picard’s theorem. There exists an entire function h(z) so that f = cos(h(z)) and g = sin(h(z)).e f (z) = f (z + p). As c = 1. Deﬁne g(z) = It is entire and omits values 0.
4. 49 . Suppose f is an entire function. Show that f has an nth root if and only if all zeros of f have multiplicity divisible by n. 3. Show that F is a normal family. 2 −n sin πz πx n=1 1+ x2 n2 = sinh(πx).Chapter 6 Problems 1. Prove the identity by taking logarithmic derivatives of both sides: ∞ ∞ ∞ ∞ (−1)n n=1 z2 2z π = . Show the identities (a) (−1)n π2 = . Let F be the family of holomorphic functions in the unit disk satisfying f (n)  ≤ n! for all n ≥ 0. 2 + a2 (z + n) a cosh 2πa − cos 2πz n=−∞ 2. (z − n)2 (sin πz)(tan πz) n=−∞ (b) 1 + z (c) π sinh 2πa 1 = · .
As log f (z) is harmonic. i. a2 . 1 − ak z In particular.5. Let the zeros of f are a1 .3) and observe the following: Suppose f (z) is an entire function with zeros {ak } satisfying 1 k a m+1 < ∞ for some integer m. Show that n f (z) ≤ M k=1 ak − z . f (0) ≤ M n k=1 ak  and if f (0) = 0 then f (z) ≤ M z. . (Canonical Product) Look back at the proof of Weierstrass’ theorem (Section 2. Suppose f (z) is a holomorphic function deﬁned on the unit disk with f (z) ≤ M . also show that RHS is continuous). (Jensen’s formula) Suppose f (z) is a holomorphic function in the unit disk without 2π 1 zeros. Show that B(z) is holomorphic. In this case.e that the sum k log ak  converges (Hint: use the ﬁrst corollary of Problem 5). But what if f (z) has zeros in the unit disk? Suppose that f (0) = 0 and denote its zeros by {ak }. 9. 7. Prove the above formula (Hint: ﬁrst consider r for which no {ak } lie on z = r. It then admits a nice factorization f = BG where B is a product of ∞ ak  ak − z s B(z) = z · ak 1 − ak z k=1 and G(z) is a holomorphic functions without zeros. Show that a bounded holomorphic function admits a Blaschke product. . log f (0) = 1 2π 2π log f (reiθ )dθ − 0 ak <r log r ak  Here log f (0) is not equal to. k 50 . Such functions are called subharmonic. 6. . Then it is possible to choose all mk = m. we have log f (0) = 2π 0 log f (reiθ )dθ for 0 < r < 1. (Blaschke product) Suppose f (z) is a holomorphic function deﬁned on the unit disk with a zero at 0 of order s and the other zeros {ak } satisfying k (1 − ak ) < ∞ (or equivalently k log ak  > −∞). but is actually less than the mean value. 8. an counted with their multiplicities.
13.10. (a) Use the previous problem. Hint: Show that for all natural n ≥ 2 and positive x (n − 1)x (n − 1)! ≤ f (x + n) ≤ nx (n − 1)! and nx n! x+n nx n! ≤ f (x) ≤ · .2. 12. Treating it as an equation in variables log r z. diﬀerentiate with respect to z. f (x) = Γ(x) for all x > 0. Show that the automorphisms of the upper halfplane which preserve i are given by w(z) = Write the formula for θ = π. (BohrMollerup theorem) Suppose f : R+ → R+ is a logarithmically convex function satisfying f (x + 1) = f (x) and f (1) = 1. 1 − z tan(θ/2) . 1. b such that af + bg = 1. 1. take the logarithmic derivative and use Example 2 from Section 1. Then log f (z) = log R · log z. 51 z + tan(θ/2) . (b) Use the uniqueness of solution of the Dirichlet problem: suppose that f : A(0. Show that there exists entire functions a. Find the automorphism group of C \{0}. Then necessarily. R) is the biholomorphism and that f takes {z : z = 1} to itself. 16. r) → A(0. g are entire functions without common zeros. Show that two annuli are biholomorphic if and only if the ratio of their radii are the same (complete the proofs below). x(x + 1) · · · (x + n) x(x + 1) · · · (x + n) n 14. 11. (Wedderburn’s lemma) Suppose f. z. Prove that sin πz = πz ∞ 1− n=1 z2 n2 Hint: The Weierstrass theorem implies that sin πz = zeg(z) n=0 1− z z/n e n To ﬁnd g(z). Find the residues at the poles of the Γ function (Hint: use the functional equation). 15.
(c) Given an annulus. k=1 21. 1. f2 ). The fundamental group of the annulus (which is Z) acts on H+ . then these actions must be conjugate. 18. Ar = A(0. Show that two tori C /Γ1 (with Γ1 generated by e1 . Show that F (w) = 0 (1−wn )−2/n dw maps the unit disk onto the interior of a regular polygon with n sides. 22. 52 . r). (SchwarzChristoﬀel formula) Show that the mapping F (w) : H+ → Ω (where Ω is a polygon) given by the formula w n−1 F (w) = C 0 k=1 (w − wk )−βk dw + C βk = 2. e2 ) and C /Γ2 (with Γ2 generated by f1 . If two annuli Ar and AR are biholomorphic. with generator acting by multiplication by λr . 17. Find the image of the unit disk under the mapping F (z) = λk are positive with n λk = 2. f2 ) are biholomorphic if and only if there is a fractional linear transformation with integer coeﬃcients and determinant 1 which takes (e1 . Prove the addition theorem: p(z1 + z2 ) = −p(z1 ) − p(z2 ) + 1 p (z1 ) − p (z2 ) 4 p(z1 ) − p(z2 ) 2 1 z n k=1 (z − ak )λk where . 20. we can consider its universal covering space fr : H+ → Ar with fr given by f (z) = exp −2πi · log z log λr where λr > 0 depends on r. is conformal for some distinct real numbers wk and w 19. Another form of the addition theorem: if u + v + w = 0 in C /Γ then 1 1 1 p(u) p(v) p(w) p (u) p (v) p (w) det = 0. e2 ) to (f1 .
b2 . . z 26. 25. . Find a maximal domain of existence (a maximal open set in C to which f (z) may be analytically continued). 28. Suppose that f (z) is a even doubly periodic function (let the group of periods be Γ). 3 3 3 1−x by integrating over a lift of γ in the Riemann Surface X ⊆ P 2 (C) given by the equation x3 + y 3 = z 3 (show that the form being integrated is meromorphic and apply the Residue theorem). . it does not extend holomorphically to any larger open set (Hint: f (z 2 ) = f (z) − z). Remark. . j γ 0 1 j2 53 . w) of w2 − 2wz + 1 = 0 can be completed to a compact Riemann surface over P 1 (C).23. Show that while the function f (z) = n z 2 is holomorphic in the unit disk. There exists points a1 . bn ∈ C /Γ such that n f (z) = c k=1 p(z) − p(ak ) . This problem is a special case of the Hadamard Gap theorem. b1 . Find the residues of the diﬀerential form √zdz−1 points 2 at inﬁnity. 2k n 27. Prove 0 1 √ 3 2π dx = √ . a2 . . an . The set of solutions (z. Find the radius of convergence of f (z) = k≥1 k+1 . Show that a doubly periodic function is a rational function of the p and p (Hint: use the previous problem). p(z) − p(bk ) 24. .
it is necessarily constant. Prove Picard’s theorem for meromorphic functions: if a meromorphic function deﬁned on the entire complex plane omits three values. g are entire functions satisfying ef + eg = 1. 31. 30. Suppose f. Is result still true if 3 is replaced by a larger positive integer? 33. Show that f. Show that actually f and g are constant functions. Show that the function f (z) = zez attains every value from C. 54 . g are actually constant functions. 32. g are meromorphic functions such that f 3 + g 3 = 1. Suppose f.29. Show that a nonconstant holomorphic function deﬁned on C \{0} omits at most 1 value.
Theorem.Chapter 7 Further Results 7.  z − z ).1 Big Picard Theorem To prove Big Picard Theorem. by the ArzelaAscoli theorem. As {fn }n>N ∪ {f } is compact. Suppose a ∈ Ω is such that f (a) = ∞. Notation: We will always write S 2 as a subscript when dealing with spherical distance. Remark. ∞). There is an for which f (a) − fn (a) < when n > N is suﬃciently large. Instead of using geodesic spherical distance. Let δ < d (M. the limit function f is also holomorphic or identically inﬁnite. r). A family of meromorphic functions S ⊂ M (Ω) is normal if it is every sequence in S has a subsequence that converges uniformly in the spherical metric on compact subsets of Ω. But if we do add the identically inﬁnite function. (1) Then either f is meromorphic or identically inﬁnite. In particular. There exists an r such that for z ∈ D(a. S S2 55 . S 2 ) taken with the compactopen topology. fn (a) < f (a) + = M for some constant M . the above theorem implies that we really do get a closed subspace. Suppose {fn } ⊂ M (Ω) which converges uniformly in the spherical metric to f . we need the concept of normal families of meromorphic functions. so the holomorphic functions do not form a closed subspace of continuous functions C(Ω. we see that it is uniformly 2 equicontinuous. The functions fn (z) = n converge to the identically inﬁnite function. if each fn is holomorphic. Proof of (1): Case I. it is easier to use the equivalent metric 1 1 z − z S 2 = min(z − z . (2) Furthermore.
so {fnk } bounded uniformly on K.4 with a fractional linear transformation). 1} will be a holomorphic covering map (compose the modular function constructed in Sections 5. there is a constant M < 1 such that gnk (z) < M on K. As gnk converges to g uniformly. Theorem (MontelCarath´odory). it suﬃces to treat the case when Ω is a disk (let the center be a). But the set λ(D(0. Proof of (2): Now suppose that {fn } are actually holomorphic functions. This allows us to consider U . This means that near a. we may assume fn (a) → α. Since gn (z) ≤ 1. 0. r) on which gn are holomorphic (for n suﬃciently large) and converge uniformly on compact subsets to g. by Montel’s theorem. i. But actually g is a function from Ω to D – otherwise. In this case. g(z) is bounded by some constant M < 1. By Case I. we can lift each fn : Ω → C \{0. a small ball centered at α within C \{0. the maximummodulus principle implies that g(z) = c is constant (of absolute value 1). we can extract a subsequence gnk which converges uniformly on compact sets to a function g : Ω → D.3 and 5. Either g is identically 0. so by Hurwitz’s theorem.e f is identically inﬁnite. {fnk } is normal. Now consider the case a ∈ Ω where f (a) = ∞. To see that the limit f is holomorphic. Since this is true of all K. r). ∞. As λ is a holomorphic covering map. 1} to gn : Ω → D so that gn (a) ∈ U . S 2 ) to g. As normality is a local property. M )) is bounded. in which case f is identically inﬁnite or 1 the zeros of g are isolated and so g = f is a meromorphic function with a pole at a. 56 . it is normal. Then g(a) = 0 and gn converges in C(Ω. gn (z) never vanishes but g(a) = 0. Let gn (z) = fn ( z ) and 1 g(z) = f ( z ). f (z) is a holomorphic function. It follows that {fn (z)} is bounded uniformly in D(a. We have two cases. Passing to a subsequence. But this would imply that (λU )−1 (α) = limn (λU )−1 (fn (a)) = limn gn (a) = c which is absurd. Suppose {fn } ⊂ S. we see that there exists a ball D(a.fn (z) − fn (a)S 2 < δ simultaneously for all n > N . Notation: D will denote the unit disk and λ : D → C \{0. we must show that Case II implies that f is identically inﬁnite. Suppose K ⊂ Ω is compact. Suppose a ∈ Ω is such that fn (a) tends to inﬁnity. First suppose that α = 1. 1 Case II. On K. g(z) must vanish identically. 1}. If a family of functions S ⊂ H(Ω) avoids two values e (say 0 and 1).
If α = ∞. we can cover K by a grid of squares of side length less than d(K. Let {Qj } be the squares which intersect K (then by hypothesis. and thus f is identically inﬁnite. Hence f is bounded near 0. R). ∂Ω) – see diagram.2 Runge’s Theorem Proposition. If 2 {fnk } genuinely converges uniformly. 1. R) \ {0}.Now we deal with the three special values of α: √ Suppose α = 1. the {Qj } lie inside Ω). This means that some subsequence {fnk } converges to g uniformly on z = R . we may assume that f has an essential singularity at the origin. 57 . so have a n subsequence gnk which converges uniformly on compact sets of Ω to some function g. Again gn are analytic and functions into C \{0. If α = 0. The MontelCarath´odory theorem implies that fn is a normal family of e meromorphic functions. this reduces the case when α = 1. the σk may taken of the same length and parallel to the coordinate axis. . so by Hurwitz’s theorem. 1}. let gn = f1 . f must be bounded by M in its interior. g is identically 0. As fn don’t vanish. nk+1 . . Now consider 1 1 the annulus A(0. But then gn (Ω) ⊂ C \{0. On its boundary. 1} and normality of fn is equivalent to normality of gn . σn f (ζ) 1 lying in Ω \ K such that f (z) = 2πi n k=1 σk ζ−z dζ for all z ∈ K. 7. f ≤ M . then f (z/n) ≤ M for all integers nk . Let Ω = D(0. we choose gn = fn : Ω → C so that gn (a) → −1. . let gn = 1 − fn . Furthemore. Proof of the Big Picard Theorem Without loss of generality. Suppose in some disk D(0. The other possibility is that g is identically inﬁnite. so we actually have a removable singularity. by maximummodulus. nk ). Suppose K ⊂ Ω is compact. Deﬁne fn (z) = f (z/n). Then there exists line segments σ1 . But gnk have no zeros. σ2 . But this means that limz→0 f (z) = ∞ meaning that f really has a pole at 0. For this purpose. so we have reduced the case α = 1 to the general case. f (z) omits values 0.
then the integral diﬀers from the Riemann sum by less than . we can make the partition so ﬁne to make Riemann sum approximate the integral simultaneously for all z ∈ K. as σ is compact. Suppose K ⊂ C is compact. However. In fact. These are our σk . we are tempted to evaluate S(z) = 2πi j f (z)χQj (z) = f (z). By the above proposition. Lemma. if the partition in the Riemann sum is ﬁne enough. But the integrand is continuous in z (at least while z ∈ K) so by compactness of K. for a ﬁxed z ∈ K and any > 0. Suppose f is a holomorphic function deﬁned on Ω ⊃ K. the sum in S(z) has a great many cancellations: if two squares in {Qj } are adjacent then their common side will cancel out (due to opposite orientation). Exercise: One can arrange the σk into ﬁnitely many disjoint cycles. j 1 Each integral σ f (ζ) dζ may be approximated by its Riemann sum n n ζj −z (here j=1 ζ−z ζk lie on σ). 58 . This calculation has the slight problem that z may lie on a boundary of one of the Qj ’s. the only segments of ∂Qj that will remain will lie outside of K. We can approximate any function f holomorphic on a neighbourhood K uniformly by rational functions with poles lying outside K. f (ζ ) More precisely.K Consider the sum S(z) = 1 2πi j ∂Qj f (ζ) dζ ζ −z 1 If z ∈ K. By compactness of K and σ. this approximation is in fact uniform on K. we f (ζ) 1 can choose segments {σk } such that f (z) = 2πi n k=1 σk ζ−z dζ holds for all z ∈ K.
Again. let Ω = H+ and ak = ik. To obtain the theorem from the lemma. it is not always possible to choose a satellite sequence. b are in 1 the same component of C \K then z−a can be approximated uniformly by polynomials of 1 z−b . For this {ak } a satellite sequence does not exist. Then any function f holomorphic on a neighbourhood of K can be uniformly approximated by rational functions having poles in E. for we work in z Ω1/r = {z ∈ Ω : d(z. Suppose K ⊂ C is compact and E is a set containing a point from every bounded component of C \K. r) then z−b can be uniformly approximated by polynomials in z−a on C \D(a. Unfortunately. it suﬃces to prove the following: if a. gk (z) ≤ 1 rak − bk  mk + 1 mk +1 1 − rak − bk  −1 Just like in the proof of the regular Weierstrass theorem. Let Ω ⊂ C be a domain and let {ak } 2 be points accumulating only at the boundary of Ω. ∂Ω) > 1/r and consider terms with rak − bk  < 1: gk (z) = − Thus.. if {ak } is unbounded. 59 . Suppose {bk } is a sequence of points outside Ω such that bk − ak  → 0 (a satellite sequence). r). For instance. we need to choose mk so that (rak − bk )mk +1 converges for every r > 0.. 1 ak − bk mk + 1 zk − bk mk +1 − 1 ak − bk mk + 2 zk − bk mk +2 − . Proof is left as an exercise for the reader. If we prove this claim we can “push” poles from bounded components of C \K to E and poles lying in the unbounded component to inﬁnity.Theorem (Runge). Remark. k −bk We propose the product f (z) = Gmk ak −bk . Exercise: Such a sequence exists if {ak } is bounded. Suppose r > 0. mk = k suﬃces. 7. it suﬃces to show if b ∈ 1 1 D(a. By transitivity of uniform approximation by polynomials.3 Prescribing Zeros in Arbitrary Domains z ak Recall that to construct an entire function with zeros {ak } we took the product of Gmk 1 where Gm (w) = (1 − w) · exp w + 1 w2 + · · · + m z m .
the power series expansion of f (z) converges only inside Ω. it is holomorphic at a! = 1 and as product converges absolutely and uniformly near ˜ inﬁnity. #Bi z − bk The contribution from the bk ’s on Lj to Ki with j > i is less than /2j . For any domain Ω ⊆ C. trouble is that it may not be a holomorphic at a (and it may not be possible to choose a outside of Ω). But limz→∞ Gm Corollary. the principal branch of log z (which is deﬁned on C without the negative real ray) cannot be extended anywhere on this ray. exists a rational function pk (z) 1 with poles lying outside Ki+1 such that Pk ( z−bk ) − pk (z) < /2i on Ki . hence the series converges uniformly and absolutely implying that f (z) is meromorphic.4 Prescribing Singularities in Arbitrary Domains We wish to construct a function f (z) meromorphic in a domain Ω with poles {bk } (accu1 mulating only at the boundary of Ω) and principal parts Pk ( z−bk ). ˜ Let f (z) be a holomorphic function with {h(ak )} as zeros. By Runge’s theorem. 1 The mapping h(z) = z−a will ensure that {h(ak )} is bounded. By construction f (z) = ˜ f (h(a)) has the desired zeros. Let f (z) be a function whose zeros accumulate to every boundary point of Ω (actually. nor is one of their accumulation points. It is even true that any domain Ω ⊆ C is a domain of holomorphy which means that there is a holomorphic function f (z) ∈ H(Ω) such that for any point z0 . this is easier said than done).We continue the construction when {ak } happens to be unbounded.e f (a) = 1. 60 . ak −bk zk −bk 7. We want to take f (z) = k 1 1 Pk − pk (z) . But actually. Let Li = Ki+1 \ Ki and Bi be the bk ’s which lie inside Li . Pick a point a ∈ C which happens not to be one of the designated ak . but power series expansions do converge beyond this ray. i. Plan: write Ω = i Ki as the increasing union of compact sets. there exists a holomorphic function f (z) ∈ H(Ω) which cannot be continued analytically to any larger domain. For instance. we see that actually f (∞) = 1.
Each D(z. In particular. so in what follows we consider Ω = C. then Z would have to contain all of S as desired. Ω). it is possible to choose a compact K1 ⊂ Ω containing K such that every hole (compact component) of K1 contains a hole of Ω. we may describe the complement of Ωr : C \Ωr = z∈C \Ω D(z. If Ω = C. Ωr is closed. Let Z be a bounded component of C \K1 . As Z is nonempty. Alternatively. As C \B is a connected and unbounded chunk of C \K1 . r) = z∈Z\Ω D(z. Given a compact subset K ⊂ Ω. Let Ωr = {z ∈ Ω : d(z. We would take K1 = Ωr . so is either contained or disjoint from Z (that is why we have shown Z ⊂ C \ Ωr ). so we take be a giant ball B containing K and set K1 = Ωr ∩ B. j ≥ 2). then as C \K1 ⊃ C \Ω. but then we are done: suppose Z intersects C \Ω at a point p and if S is the connected component of p in C \Ω. Hence Z ⊂ C \K1 \ C \B = C \Ωr . it cannot intersect Z.We have to insure that application of Runge’s theorem is valid. the above representation tells us that Z intersects C \Ω. r). Ω) ≤ r} where r is some number less than d(K. this tells us that Z is the union of disks D(z. r) which makes up C \Ωr is connected. Z would have to contain C \B which is absurd. for if Z has common points with C \B. r). r) which have center in Z. Lemma. for this we need to take K1 “large enough” so that each bounded component of C \K1 contains a bounded component of C \Ω (then this will also be true of Kj . 61 . i. but Ωr may not be compact. we can let K1 be giant ball.e Z= z∈Z∩(C \Ω) D(z.
This action might not be possible to undo. Are you sure you want to continue?
Use one of your book credits to continue reading from where you left off, or restart the preview.