Vous êtes sur la page 1sur 23

THE H

ORMANDER COMPANION
JENNIFER HALFPAP
As the title suggests, these notes are designed as a companion to Hormanders
now-classic book An Introduction to Complex Analysis in Several Variables [Hor90].
This book has survived for many decades because it is concise and ecient. On the
ip side, these characteristics can make it hard for the student to read. It assumes
considerable background and tends to leave details to the reader.
Thus, in addition to being a graduate course in the analysis of functions of one
and several complex variables, this is a course on learning how to learn mathematics.
Sometimes it is tempting to abandon a hard book or paper because one doesnt
have all the background. This book, for example, uses the language of dierential
forms and the wedge product on the second page. If you havent met these ideas
before and hear that they are taught in a course on manifolds, you might think
you need to wait until after such a course to read this book. Similarly, this book
uses theorems commonly rst encountered in a course on measure and integration
theory. You might think that therefore such a course is also a prerequisite.
My perspective is that if you wait to read a book or paper until you have mas-
tered all the prerequisites, you wont progress. Instead, when you come up against
material assumed as background with which you are unfamiliar, aim to develop
enough of a working knowledge to be able to proceed with your main goal. As your
interests develop, it will become clear which auxiliary areas require more attention
on your part and which are peripheral.
These notes are designed to help you learn to do this. They include some expos-
itory material that may make reading Hormanders exposition easier, they include
exercises that allow you to practice using denitions and results, and they help you
ll in some of the details left to the reader.
1. Analytic Functions of One Complex Variable
Let be an open subset of C and consider f : C. Write f in terms of its
real and imaginary parts, i.e., if z = x+iy, f(z) = u(x, y) +iv(x, y) for real-valued
functions u and v of the two real variables x and y. Hormander denes f to be
analytic on if f C
1
() and
f
z
= 0, where

z
:=
1
2
(

x
+i

y
). It then follows
that for f analytic, df =
f
z
dz for

z
:=
1
2
(

x
i

y
).
Exercise 1. Using this denition of analyticity, determine the system of (real)
partial dierential equations satised by u and v.
In most undergraduate texts on complex analysis, analyticity is dened as fol-
lows:
Denition 1. f : C is analytic at z
0
if
(1) lim
zz
0
f(z) f(z
0
)
z z
0
1
2 JENNIFER HALFPAP
exists, in which case the value is denoted by f

(z
0
). f is said to be analytic on the
open set if it is analytic at every point of .
Exercise 2. Consider a function f = u + iv for which the above limit exists for
some z
0
= x
0
+ iy
0
. Show that at this point, u and v satisfy the same system of
partial dierential equations identied in the previous exercise. (Hint: Consider
restricted approaches to z
0
.) Then show that this denition of f

agrees with what


Hormander would write as
f
z
.
The point of Hormanders approach is that it reects the intuition that analytic
functions are those functions on C which are independent of z. Such intuition is
reinforced by the following elementary exercises:
Exercise 3. Dene e
z
:= e
x
cos y + ie
x
sin y. Show that this denes an analytic
function on all of C. (Such functions are called entire.)
Exercise 4. The modulus of a complex number z = x +iy is its usual Euclidean
distance from the origin and is denoted by |z|. Show that |z|
2
= z z, and determine
all points of C at which f(z) = |z|
2
is analytic.
Exercise 5. Show that f(z) = z
n
is entire for n N. This can be done in a
number of ways. In particular, either denition of analyticity can be used. Try it
both ways.
2. Stokes/Greens Theorem, the Cauchy Integral Formula, and
Consequences
Most undergraduate books in complex analysis take a long time working up to
the Cauchy Integral Formula, and then usually only give partial proofs for rather
special domains. Few of these presentations do what Hormander does and obtain
the result from a version of Stokes/Greens Theorem using the language of dier-
ential forms and the exterior derivative. Furthermore, these latter topics are often
treated in a similarly incomplete manner in the undergraduate curriculum. Such
pedagogical choices are appropriate and understandable; a thorough treatment of
Stokes theorem is usually left to a course in dierential topology or smooth mani-
folds.
We will commit similar sins here and neither prove Stokes theorem nor develop
the underlying ideas fully. Such a development would be inappropriate here because
what is really being used is the more elementary special case of Stokes Theorem
known as Greens Theorem. On the other hand, consistent with the spirit of this
exposition, we aim to move the reader one step closer to an understanding of this
deep result by introducing, albeit informally, notions of manifold, tangent space,
and dierential form, followed by two statements of Greens theorem and the proof
of the Cauchy Integral Formula given in Hormanders Section 1.2.
2.1. What is a manifold? Roughly speaking, a k-dimensional manifold is a set
which near every point resembles R
k
. For example, consider the curve in the plane
with equation y = x
2
. This set is most certainly not a vector subspace of R
2
.
However, this curve is locally linear; at every point along the curve there is a
well-dened tangent line. We mention two other (trivial) classes of examples of
manifolds: open subsets U of R
2
, and any nite set of points in R
2
.
For our purposes in this section, the three examples above are about all we need;
we will consider curves in the plane which have a well-dened tangent line at all
THE H

ORMANDER COMPANION 3
but nitely many points (e.g., the unit circle or the boundary of some polygonal
region in the plane), open sets in the plane whose boundaries are these sorts of
curves, and nite collections of points in the plane.
For any manifold, there is an associated notion of dimension, corresponding to
the dimension of the Euclidean space it resembles locally. Thus an open set U in R
2
is a manifold of dimension 2, the unit circle in R
2
is a manifold of dimension 1, and
a point or collection of points is a manifold of dimension 0. Also associated with
any manifold is its tangent space at a point. In the case in which the manifold
is also a subset of some R
n
, this notion of tangent space corresponds closely to the
notion developed in calculus, except that because we want to think of the tangent
space to a manifold at a point as a vector space, we always think of our tangent
spaces as going through the origin.
Let us consider the three examples above. In all cases, since the manifold under
consideration is a subset of R
2
, we will also think of the tangent space as a subspace
of R
2
. If p is a point of an open subset U of R
2
, the tangent space to U at p, denoted
T
p
(U), is just a copy of R
2
. For the point p = (1/

2, 1/

2) on the unit circle C,


since the equation of the tangent line is y1/

2 = (x1/

2), the tangent space


is given by T
p
(C) = { (x, y) : y = x}. Finally, the tangent space to the manifold
consisting of the single point p is T
p
({p}) = {(0, 0)}.
Exercise 6. Look up the denitions of a dieomorphism, a smooth k-dimensional
manifold M R
n
, and the tangent space T
p
(M) to M at p. Using these denitions,
show that the unit circle C in R
2
is indeed a smooth 1-dimensional manifold and
that (as claimed above) T
(1/

2,1/

2)
(C) = { (x, y) : y = x}. A good source is
[GP74], p.3 and p.9.
2.2. Tangent vectors and forms on R
n
. Consider R
n
as an n-dimensional man-
ifold, or, more generally, consider an open subset U of R
n
. As discussed in the
previous subsection, at any point p, the tangent space T
p
(R
n
) (resp. T
p
(U)) is just
R
n
itself. We could denote the standard basis for this vector space by {e
1
, . . . , e
n
},
but to stress that this copy of R
n
arises as the tangent space to a manifold at a
point, a more common notation for this basis is
{

x
1
, . . . ,

x
n
}.
Denition 2. A 1-form at p R
n
(or U) is a linear map : T
p
(R
n
) R. The
set of all 1-forms at p is the dual space to T
p
(R
n
) and is denoted T

p
(R
n
).
The above is a special case of a standard construction in linear algebra:
Denition 3. Let V be a normed vector space over R. A linear functional is a
linear map : V R. The set of all continuous linear functionals on V is called
its dual space and is denoted by V

.
It is a standard result from linear algebra that the space dual to a vector space
(equipped with the obvious operations) is itself a vector space, and when the original
space has nite dimension n, so does the dual space. We denote by {dx
1
, . . . , dx
n
}
the basis dual to the standard basis, so that the following relations hold:
dx
i
(

x
j
) =
i,j
.
More generally, we dene k-forms on R
n
.
4 JENNIFER HALFPAP
Denition 4. A k-form on R
n
(thought of as T
p
(R
n
) or T
p
(U)) is a function
: (T
p
(R
n
))
k
R that is multi-linear and anti-symmetric, i.e.,
(a) if L
1
, . . . , L
k
are vectors and L
i
= M
1
+ M
2
, then (L
1
, . . . , M
1
+
M
2
, . . . , L
k
) = (L
1
, . . . , M
1
, . . . , L
k
) +(L
1
, . . . , M
2
, . . . , L
k
).
(b) (L
1
, . . . , L
i
, . . . , L
j
, . . . , L
k
) = (L
1
, . . . , L
j
, . . . , L
i
, . . . , L
k
).
We denote the space of k-forms by
k
(T

p
(R
n
)).
The rst essential remark is that you are already quite familiar with one example
of the above.
Example 5. The determinant function det denes an n-form on R
n
. Indeed, to
every collection of n vectors in R
n
, it assigns a real number. This function is linear
in each entry, and if two vectors are interchanged, the determinant changes sign.
In fact, det is the unique n-form on R
n
for which det(

x
1
, . . . ,

x
n
) = 1.
Not only does the determinant provide a familiar example of a form, but also in
some sense all k-forms on R
n
can be thought of as generalizations of the determi-
nant.
We dene an operation that allows us to combine forms to obtain a new form.
Denition 6. Let be a k-form on R
n
and an l-form. We dene :
(T
p
(R
n
))
k+l
R by
(2)
(L
1
, . . . , L
k
, L
k+1
, . . . , L
k+l
) =

S
k+l
sgn()(L
(1)
, . . . , L
(k)
)(L
(k+1)
, . . . , L
(k+l)
)
where the set S
k+l
consists of all permutations of {1, . . . , k +l}.
For example, if we consider the 1-forms dx
1
and dx
2
on R
2
, their wedge product
dx
1
dx
2
is a 2-form on R
2
. If L
1
= a

x
1
+c

x
2
and L
2
= b

x
1
+d

x
2
, then
dx
1
dx
2
(L
1
, L
2
) = +dx
1
(L
1
)dx
2
(L
2
) dx
1
(L
2
)dx
2
(L
1
) = ad bc,
which we recognize as the determinant of the 2 2 matrix
_
a b
c d
_
.
The following proposition contains properties of this wedge product.
Proposition 7. Let , be k-forms, let be an l-form, let be an r-form, and
let , be scalars.
(1) is a k +l-form (i.e., it is multi-linear and antisymmetric).
(2) ( +) = ( ) +( ).
(3) ( ) = () . We may thus use the notation for either.
Exercise 7. Use the above denition to describe
(1) dx
1
dx
1
(2) the relationship between dx
1
dx
2
and dx
2
dx
1
.
(3)
1

2

3
where each
i
is a 1-form on R
2
.
State propositions generalizing the above for wedge products of forms on R
n
.
We make a few nal comments in light of general results suggested by the above
exercise. The only interesting spaces
k
(T

p
(R
n
)) are for k n, so henceforth when
we talk about k-forms on R
n
, we will be thinking of k n. Furthermore, there is an
obvious vector space structure on
k
(T

p
(R
n
)), and { dx
I
:= dx
i
1
dx
i
2
. . . dx
i
k
:
1 i
1
< i
2
< . . . < i
k
n} is a basis.
THE H

ORMANDER COMPANION 5
2.3. Dierential forms and the exterior derivative. For an open subset of
R
n
, let E() be the space of innitely-dierentiable complex-valued functions on
.
Denition 8. A dierential 1-form or smooth 1-form on is an object
(3) =
n

j=1

j
dx
j
,
where
j
E().
One of the most familiar examples is the exterior derivative df of a function
f E().
Denition 9. For f E(R
n
), p R
n
, and L T
p
(R
n
), dene df by
df(L) := L(f).
Of course it would be nice to know how to express df in the form (3). Taking L
to be

x
j
, one sees that
j
=
f
x
j
, i.e.,
df =
n

j=1
f
x
j
dx
j
.
More generally,
Denition 10. A smooth dierential k-form on (or, more succinctly, a
smooth k-form) is an object
(4) =

|I|=k
f
I
dx
I
where each f
I
E(), each I is an increasing k-tuple of elements from the set
{ i : 1 i n}, and the sum is taken over all increasing k-tuples. We denote by
E
k
() the set of all smooth k-forms on . Since smooth 0-forms are identied with
smooth functions, the notations E
0
() and E() may both be used for this space.
Denition 11. We dene the exterior derivative d : E
k
() E
k+1
() by
d
_
_

|I|=k
f
I
dx
I
_
_
:=

|I|=k
df
I
dx
I
.
Exercise 8. (1) Let = x
2
y dx + xy
3
dy and = e
x
(cos y + i sin y) dx
e
x
(i sin y + cos y) dy. Compute d and d.
(2) Prove that for any smooth k-form , d
2
= d(d) = 0.
(3) In a multivariable calculus course, one often learns about exact dierential
forms. Look up this denition and the connection between exact dierential
1-forms and line integrals. Discuss how these ideas relate to the earlier
pasts of this exercise.
(4) Look up the denition of a closed dierential form, and discuss the con-
nection between closed and exact forms.
6 JENNIFER HALFPAP
2.4. The theorems. The result you know as Greens Theorem is probably some-
thing close to the following:
Theorem 12 (Greens Theorem, First Version). Let R
2
be an open, connected,
and bounded set whose boundary is piece-wise smooth and positively-oriented
(i.e., as one traverses the boundary in the positive sense, lies to ones left). If P
and Q are smooth throughout some open set containing , then
_

Pdx +Qdy =
__

_
Q
x

P
y
_
dxdy.
This can be expressed in a dierent way using forms and exterior derivatives.
For the 1-form = Pdx +Qdy (using subscripts to denote partial derivatives)
d = dP dx +dQ dy
= (P
x
dx +P
y
dy) dx + (Q
x
dx +Q
y
dy) dy
= P
x
dx dx +P
y
dy dx +Q
x
dx dy +Q
y
dy dy
= P
y
dx dy +Q
x
dx dy.
We may thus restate Greens Theorem:
Theorem 13 (Greens Theorem, Second Version). Let R
2
be an open, con-
nected, and bounded set whose boundary is piece-wise smooth and positively-
oriented. If is a smooth 1-form on a neighborhood of ,
_

=
_

d.
Greens theorem is a special case of Stokes theorem, which says more generally
that for an oriented k-manifold M with boundary M having the induced orienta-
tion, the integral of a k 1 form on the boundary equals the integral of its exterior
derivative over the manifold. Stating it more precisely would require us to extend
our denitions of vector, form, smooth dierential form, and exterior derivative to
functions dened on manifolds. This is beyond the scope of this course.
We now return to the complex setting. Think of as a connected, open, bounded
subset of C with piece-wise smooth boundary. Consider = fdz for f C
1
().
Then
d(fdz) = df dz
=
_
f
x
dx +
f
y
dy
_
(dx +idy)
= i
f
x
dx dy
f
y
dx dy
= 2i
f
z
dx dy
=
f
z
d z dz.
Greens Theorem and this calculation thus give the following:
Corollary 14 (Cauchy Integral Theorem). If and satisfy the same hypotheses
as in Greens Theorem and if f is analytic in a neighborhood of , then
_

f(z) dz = 0.
THE H

ORMANDER COMPANION 7
Exercise 9. Prove the identity 2idx dy = d z dz used above.
Exercise 10. Let C denote the positively-oriented unit circle. For each n Z,
evaluate (with justications)
_
C
z
n
dz. Look up the denition of a simple closed
curve in C. What happens if C is replaced by a simple, closed, positively-oriented
curve having the origin in its interior?
We can now state the Cauchy Integral Formula.
Theorem 15 (Cauchy Integral Formula). If f C
1
(),
(5) 2if() =
_

f(z)
z
dz +
_

f
z
z
dz d z.
Exercise 11. Fill in the details of Hormanders proof of this theorem. In particular,
include complete justication of the statements
lim
0
_
2
0
f( +e
i
) d = 2f()
and
lim
0
__

f
z
z
dz d z =
__

f
z
z
dz d z.
2.5. Hormanders Theorem 1.2.2.
Theorem 16. Let be a measure with compact support K in C. The integral
(6) f() =
_
1
z
d(z)
denes an analytic C

function on C \ K. Furthermore, in any open set where


d = (2i)
1
dz d z for C
k
(), we have f C
k
() with
f
z
= if k 1.
To summarize, this theorem tells us that (1) for not in the support of ,
integrating against 1/(z ) produces an analytic function and (2) for in the
support of , if equals (2i)
1
dzd z, then f has the same degree of smoothness
as and satises
f

= .
Understanding the theorem and its proof requires us to understand what a mea-
sure is and how to dierentiate a function dened as an integral.
Measures. The theory of measure and integration is a course in itself. In this section,
we give a denition of measure which should be enough to allow us to proceed.
Denition 17. Let X be a compact subset of R
n
and let C(X) denote the vector
space of complex-valued continuous functions on X, equipped with the sup-norm
||f|| := sup
xX
|f(x)|. A complex measure on X is a continuous complex-valued
linear functional on C(X).
If is a complex measure on X, we could denote its value on an f C(X) by
(f), but it is more traditional to denote it by
_
f(x) d(x). The following exercise
helps to explain why such notation is used.
Exercise 12. Suppose is itself a continuous function on the compact subset X
of R
n
. Show that
f
_
f(x)(x) dx
denes a complex measure on X. A common notation for this measure would be
dx.
8 JENNIFER HALFPAP
In Theorem 1.2.2, Hormander assumes that is compactly supported, meaning
that there is a compact set K such that
_
f d =
_
K
f d. When d = dz d z
for continuous, this just means has compact support.
Dierentiating functions dened by integrals. This is another common situation in
analysis arising in this book for the rst time in the proof of Theorem 1.2.2. We
have a function dened by an integral:
(7) f() :=
_
1
z
d(z),
where has compact support K. We are interested in whether f is dierentiable or
analytic as a function of on C\K. Since for any in C\K, z does not vanish
for any z in K, viewed as a function of , our integrand has the desired properties.
Is the following legitimate?
f

=
_

_
1
z
_
d(z) =
_
0 d(z) = 0
Fortunately, in many situations it is, though of course some justication is required.
We discuss here several common arguments that may establish the legitimacy of
dierentiating under the integral sign.
The most obvious approach uses the denition of the derivative. For simplicity,
suppose f : R
2
R. Dene a new function F : R R by F(t) :=
_
f(x, t) dx. If
we wish to show that, under some hypotheses on f, F

(t) =
_
f
t
(x, t) dx, we must
show that for every > 0, there exists > 0 such that for all 0 < |h| < ,

F(t +h) F(t)


h

_
f
t
(x, t) dx

< .
Furthermore, the left-hand-side of this inequality
=

_
f(x, t +h) f(x, t)
h

f
t
(x, t) dx

_
1
h
_
_
t+h
t
f
t
(x, s)
f
t
(x, t) ds
_
dx

.
Exercise 13. State hypotheses on f under which the above argument can be com-
pleted to show that F

(t) =
_
f
t
(x, t) dx. Give the complete argument in this situ-
ation.
Another approach makes use of limit theorems. After all, we are trying to
determine whether
lim
h0
F(t +h) F(t)
h
= lim
h0
_
f(x, t +h) f(x, t)
h
dx
=
_
lim
h0
f(x, t +h) f(x, t)
h
dx.
Exercise 14. Many theorems exist giving hypotheses under which the limit of the
integrals is the integral of the limit. The simplest such is usually encountered in
an undergraduate analysis course. It says that the conclusion follows if f
n
f
uniformly. State the denition of uniform convergence for a sequence of functions
on a set X, and then give a precise statement and proof of this theorem.
Another more sophisticated limit theorem is the Dominated Convergence Theo-
rem:
THE H

ORMANDER COMPANION 9
Theorem 18. Let be a complex measure and let {f
n
} and f be integrable. Sup-
pose for (almost) every x, lim
n
f
n
(x) = f(x) and that there exists a function
g with
_
|g| d < such that |f
n
(x)| |g(x)| for all n and for (almost) every x.
Then lim
n
_
f
n
(x) d =
_
f d.
In other words, this theorem says that the limit of a sequence of integrals is the
integral of the limit if the integrands converge pointwise (almost everywhere) to
the limit and if there is a single integrable function dominating all the functions
in the sequence.
We use the Dominated Convergence Theorem to prove the following theorem on
the continuity and dierentiability of functions dened by integrals.
Theorem 19 ([Fol99], Theorem 2.27). Suppose f : X (a, b) C ( < a <
b < ) and that f(, t) : X C is integrable for all t (a, b). Consider F(t) :=
_
X
f(x, t) d(x).
(a) Suppose there exists g : X C such that
_
X
|g(x)| d(x) < and for
all x, t, |f(x, t)| |g(x)|. Then if lim
tt
0
f(x, t) = f(x, t
0
) for every x,
lim
tt
0
F(t) = F(t
0
). In particular, continuity of f(x, ) for each x implies
continuity of F.
(b) Suppose
f
t
exists and there exists g : X C such that
_
X
|g(x)| d(x) <
and for all x, t, |f/t(x, t)| |g(x)|. Then F is dierentiable and
F

(t) =
_
f
t
d(x).
Proof. For part (a), let {t
n
} be any sequence in [a, b] such that t
n
t
0
. Set
g
n
(x) := f(x, t
n
). Then each g
n
is integrable in x, g
n
(x) f(x, t
0
) for every x,
and |g
n
(x)| |g(x)| for all x. Thus by the Dominated Convergence Theorem,
lim
n
F(t
n
) = lim
n
_
X
f(x, t
n
) d(x)
= lim
n
_
X
g
n
(x) d(x)
=
_
X
f(x, t
0
) d(x)
= F(t
0
).
Next, consider part (b). Fix t. Let {h
n
} be a sequence of real numbers such that
h
n
0. Set
g
n
(x) :=
f(x, t +h
n
) f(x, t)
h
n
.
Then each g
n
is integrable in x, lim
n
g
n
(x) =
f
t
(x, t) for all x, and |g
n
(x)|
|g(x)| for all n and x. Thus by the Dominated Convergence Theorem,
lim
n
F(t +h
n
) F(t)
h
n
= lim
n
_
X
f(x, t +h
n
) f(x, t)
h
n
d(x)
= lim
n
_
X
g
n
(x) d(x)
=
_
X
f
t
(x, t) d(x).
Since this is true for all sequences h
n
0, F

(t) exists and equals the integral of


the derivative.
10 JENNIFER HALFPAP
Exercise 15. Return to Hormanders Theorem 1.2.2. If = t + is, show that
f
t
and
f
s
can be obtained by dierentiating under the integral sign and that
f

= 0 on
C\K. (If the presence of the measure makes you uncomfortable, assume throughout
that d = (2i)
1
dz d z.)
Completing the Proof of Theorem 1.2.2. The above discussion and exercise establish
the rst statement of the Theorem, that f() =
_
(z)
1
d(z) is C

and analytic
on C \ K. We thus consider the second assertion.
Furthermore, we treat only the case = R
2
, so that for all z, d(z) = (2i)
1
(z)dz
d z for in C
k
(R
2
) with compact support K. Then
f() =
1
2i
_
(z)
z
dz d z
=
1
2i
_
(z +)
z
dz d z.
For every = t+is, the integrand is the product of a compactly-supported function
and a locally-integrable function and is thus integrable. Since C
k
with compact
support K, the integrand is a continuous function of t and s with the integrable
upper bound z
1
||||
K
(z). By Theorem 19, f is continuous. If k 1, since any
derivative of up to order k is continuous and compactly supported, any derivative
of the integrand in t and s up to order k is bounded by an integrable function of z
alone. It follows from Theorem 19 again that f C
k
with derivatives of f obtained
by dierentiating under the integral sign.
Thus if k 1,
f

() =
1
2i
_

(z +)
z
dz d z
=
1
2i
_

z
(z +)
z
dz d z
=
1
2i
_

z
(z)
z
dz d z.
The last integral is over all of R
2
, but since has compact support K, we obtain
the same value if we integrate instead over an open disc B(0, R) large enough that
it contains K. Since is identically zero on B(0, R),
f

() =
1
2i
_
B(0,R)

z
(z)
z
dz d z
=
1
2i
_
_
B(0,R)
(z)
z
dz +
_
B(0,R)

z
(z)
z
dz d z
_
= (),
where for the last equality we have used the Cauchy Integral Formula. Theorem
1.2.2 is thus established when = R
2
.
2.6. Consequences of the Cauchy Integral Formula. It is not immediately
obvious that the condition that an f in C
1
() is also analytic is a particularly
strong one. After all, we have seen that this is merely equivalent to complex dif-
ferentiability of f, i.e., the existence of lim
zz
0
f(z)f(z
0
)
zz
0
for every z
0
. Just
THE H

ORMANDER COMPANION 11
as real-valued functions can be once-dierentiable but not twice-dierentiable, it
seems there must exist functions that are once complex-dierentiable but not twice
complex-dierentiable.
This is, in fact, not the case. It turns out that the condition of complex dif-
ferentiability is quite strong; analytic functions are remarkably well-behaved. As
evidence, we establish a number of consequences of Hormanders Theorems 1.2.1
and 1.2.2.
Corollary 20. If f A(), f C

() and f

A().
Proof. Fix . There exists r > 0 and
0
such that B(
0
, r) and
B(
0
, r) . By the Cauchy Integral Formula,
f() =
1
2i
_
B(
0
,r)
f(z)
z
dz
=
_
1
z
(2i)
1
f(z)
B(
0
,r)
(z)dz d z.
Since d(z) := (2i)
1
f(z)
B(
0
,r)
(z)dz d z is a compactly-supported measure
(supported on B(
0
, r)), by Theorem 1.2.2, f is C

and analytic on C\B(


0
, r),
in particular at . Consider g := f

. Then g C
1
(B(
0
, r)) and
g

=

2
f

= 0
for all B(
0
, r). Thus f

is analytic at and the result follows.


Thus Hormander uses his quite general Theorem 1.2.2 to conclude that an ana-
lytic f is in fact innitely-dierentiable and that all of its complex derivatives are
analytic. One can obtain the result more directly:
Exercise 16 (Cauchy Integral Formula for Derivatives). Let f be analytic on .
For and B(, R) , show without appealing to Theorem 1.2.2 that for any
j N,
(8) f
(j)
() =
j!
2i
_
B(,R)
f(z)
(z )
j+1
dz.
The next exercise contains number of familiar and beautiful consequences of the
above exercise. Of course any undergraduate text in complex analysis will have the
proofs; try to prove them on your own.
Exercise 17. Let f : C be analytic. Suppose
0
and that B(
0
, R) .
(1) (Cauchy Estimates)
(9) |f
(j)
(
0
)|
j!
R
j
sup
|
0
|R
|f()|.
(2) (Liouvilles Theorem) If f is bounded and entire, f is constant.
(3) (Fundamental Theorem of Algebra) If f is a non-constant polynomial
over C, there exists z
0
C such that f(z
0
) = 0.
It is also natural to ask for what notion of limit is the limit of a sequence of
analytic functions analytic? We will prove the following:
Proposition 21. If f
n
A() and f
n
f uniformly on compact subsets of ,
then f A().
12 JENNIFER HALFPAP
Exercise 18. You may wish to consult an undergraduate analysis text such as
[Str00].
(1) Let {f
n
} be a sequence of continuous functions from [a, b] to R. For what
mode of convergence is the limit function f guaranteed to be continuous?
State such a theorem and give an example showing that if the hypothesis is
violated, the conclusion may fail.
(2) Let {f
n
} be a sequence of dierentiable functions from [a, b] to R. Suppose
f
n
(x) f(x) for all x. Under what additional hypotheses is f dierentiable
with f

(x) = lim
n
f

n
(x)? State such a theorem and give an example
showing that if the hypothesis is violated, the conclusion may fail.
The proof of the proposition requires the next theorem on the size of derivatives
of an analytic function.
Theorem 22 (Hormanders Theorem 1.2.4). For every compact subset K of
and every open neighborhood of K in , there exist constants C
j
such that for
all f A(),
(10) sup
zK
|f
(j)
(z)| C
j
_

|f(z)| dxdy.
Proof. Fix a compact subset K and a bounded open neighborhood of K with
.
Let C

0
() denote the set of all innitely-dierentiable functions with compact
support contained in . Let C

0
() with the additional property that 1
on a neighborhood

of K. Consider f A() and dene g := f. We want to


apply the Cauchy Integral Formula to g on . Since g(z) 0 on and
g
z
= f

z
,
g() = f()()
=
1
2i
_

f(z)

z
(z)
z
dz d z.
Since is a smooth function with compact support, its rst partial derivatives are
bounded functions. Thus there exists M satisfying |/ z| M on . Further-
more, since / z 0 on

, for K there exists d > 0 such that |z | d on


the support of the integrand. Thus for K
|f()| = |g()|
=
_

f(z)

z
(z)
z

dxdy

M
d
_

|f(z)| dxdy.
The result follows for j = 0.
Exercise 19. Finish the proof by rst establishing that for j N and K,
f
(j)
() =
a
j
2i
_

f(z)

z
(z)
(z )
j+1
dz d z.
We now prove Proposition 21.
THE H

ORMANDER COMPANION 13
Proof. Fix a compact set K in and an open neighborhood of K with compact
closure in . (This is expressed succinctly as K .) Applying the previous
theorem to f
n
f
m
gives
sup
zK
|f

n
(z) f

m
(z)| C
1
_

|f
n
(z) f
m
(z)| dxdy.
Since the sequence {f
n
} is uniformly convergent, it is uniformly Cauchy, and hence
by the Dominated Convergence Theorem, the right-hand side of the above inequal-
ity tends to zero. The sequence {f
n
/z} is thus uniformly convergent on K.
Since the f
n
are analytic,
f
n
z
=
1
2
_
f
n
x
+i
f
n
y
_
= 0,
i.e.,
f
n
x
= i
f
n
y
. Thus
f
n
z
= i
f
n
y
=
f
n
x
and the uniform convergence of
{f
n
/z} on K implies the uniform convergence of the sequences {f
n
/x} and
{f
n
/y}. We now invoke the real-variable result to conclude that
f
x
= lim
n
f
n
x
and
f
y
= lim
n
f
n
y
, so that
1
2
(
f
x
+i
f
y
) = lim
n
1
2
(
f
n
x
+i
f
n
y
) = 0.
As a consequence, we obtain
Corollary 23. Suppose the power series f(z) =

j=0
a
j
z
j
converges on B(0, R).
Then f A(B(0, R)).
Proof. We will show that for every 0 < r < R, the series

j=0
a
j
z
j
converges
uniformly and absolutely on B(0, r). Recall that to say that this series converges
uniformly on B(0, r) means that the sequence {f
n
} of partial sums, dened by
f
n
(z) :=

n
j=0
a
j
z
j
, converges uniformly to f on B(0, r). Uniform convergence on
every closed ball B(0, r) in B(0, R) would then imply that f
n
f uniformly on
compact subsets of B(0, R). Since the f
n
are polynomials in z, they are analytic,
and so Proposition 21 would give the analyticity of the limit function f.
The proof that a power series whose convergence set includes B(0, R) converges
uniformly and absolutely on every B(0, r) B(0, R) is left to the reader as an
exercise.
Exercise 20. Go to it, reader! You may begin by recalling or proving that if we
have a series

u
j
of functions all dened on a common ball B, and if

M
j
is
a convergent series of non-negative numbers, then if |u
j
| M
j
throughout B, the
series

u
j
converges uniformly and absolutely on B. This is sometimes called the
Weierstrass M-test.
We are now ready to prove the converse of the above, which is yet another result
illustrating our point in this section that the condition of analyticity is remarkably
strong.
Theorem 24 (Hormanders Theorem 1.2.8). If f A(B(0, R)), we have
(11) f(z) =

j=0
f
(j)
(0)
j!
z
j
,
with uniform convergence on every compact subset of B(0, R).
14 JENNIFER HALFPAP
Proof. For 0 < r < < R, by the Cauchy Integral Formula (with a change in
notation), for |z| r,
f(z) =
1
2i
_
||=
f()
z
d.
Observe,
1
z
=
1

1
(1
1
z)
=
1

j=0

j
z
j
,
where we have used the formula for the sum of a convergent geometric series.
Dene g
n
(z, ) :=
f()

n
j=0

j
z
j
and g(z, ) :=
f()
z
. Since
|g
n
(z, )|
sup
||=
|f()|

j=0

j
r
j
=
sup
||=
|f()|

1
1
r

,
g
n
g uniformly. The following is therefore legitimate:
1
2i
_
||=
f()
z
d =
1
2i
_
||=
lim
n
g
n
(z, ) d
= lim
n
1
2i
_
||=
g
n
(z, ) d
= lim
n
n

j=0
_
1
2i
_
||=
f()

j+1
d
_
z
j
=

j=0
f
(j)
(0)
j!
z
j
.

For functions that are merely in C

(), knowing the function in a neighborhood


of a single point tells us nothing about its behavior in other parts of the domain.
The situation is radically dierent for those functions represented by power series.
Corollary 25 (Uniqueness of analytic continuation). As usual, let C be a
connected open set. If f A() and if there exists z
0
such that f
(j)
(z
0
) = 0
for all j 0, then f 0 on .
Proof. Let Z := { z : f
(j)
(z) = 0, j = 0, 1, 2, . . . }. Since Z =

j=0
(f
(j)
)
1
({0})
and each function f
(j)
is continuous, Z is a closed subset of .
On the other hand, by hypothesis, Z is not empty since z
0
Z. Fix Z and
consider g dened by g(z) := f(z +). g is analytic in some neighborhood B(0, R)
of the origin, and on this disc it is represented by the series

f
(j)
(z
0
)z
j
/j!. Thus
g 0 on B(0, R), i.e., f 0 on B(, R). This proves that Z is open in . Since Z
is both open and closed in and is connected, Z = .
THE H

ORMANDER COMPANION 15
Exercise 21 (Laurent series). Let 0 r < R and dene the annulus a(r, R) :=
{ z : r < |z| < R}. If f is analytic on this annulus, then
f(z) =

j=
a
j
z
j
, a
j
=
1
2i
_
||=
f()

j+1
d,
where r < < R. (Suggestion: Apply the Cauchy Integral Formula to a(r

, R

) for
r < r

< R

< R and use the idea of the proof of Theorem 24.)


2.7. Mean-Value and Maximum Modulus Properties. Whereas the conse-
quences of the Cauchy Integral formula discussed in the previous subsection largely
concern regularity properties of analytic functions, the results of this section begin
to explore the extent to which, for f A(), we can relate its behavior on to its
values on .
To begin with, if f A() and B(, R) , then
f() =
1
2i
_
B(,R)
f(z)
z
dz
=
1
2i
_
2
0
f( +Re
i
)
Re
i
iRe
i
d
=
1
2
_
2
0
f( +Re
i
) d.
This is the mean value property for analytic functions: for a function analytic
on B(, R), its value at the center of the disc is equal to the average of its values
on the boundary. From this, the maximum modulus principle will follow.
Theorem 26 (Maximum Principle - local version). Suppose f is analytic on
B(, R). If for all z B(, R), |f(z)| |f()|, then f is constant on B(, R).
Proof. If f() = 0, then |f(z)| |f()| on B(, R) gives f 0 on B(, R). Thus in
the remainder of the proof we may suppose f() = 0. Fix 0 < r < R. By the mean
value property,
f() =
1
2
_
2
0
f( +re
i
) d.
Equivalently,
0 =
1
2
_
2
0
f() d
1
2
f( +re
i
) d,
or, dividing by f() and writing the right-hand side as a single integral,
0 =
1
2
_
2
0
1
f( +re
i
)
f()
d.
It follows that
(12) 0 =
1
2
_
2
0
Re
_
1
f( +re
i
)
f()
_
d.
16 JENNIFER HALFPAP
Write z = +re
i
and observe
2Re
_
1
f(z)
f()
_
= 1
f(z)
f()
+ 1
f(z)
f()
= 2
f(z)f() f(z)f()
|f()|
2
=
2|f()|
2
f(z)f() f(z)f()
|f()|
2

|f()|
2
f(z)f() f(z)f() +|f(z)|
2
|f()|
2
=
|f() f(z)|
2
|f()|
2
0.
Return to equation (12). The integrand is a continuous function of . Furthermore,
the last calculation shows it to be non-negative. Thus it can only have integral zero
if it is zero for all , i.e., f( + re
i
) = f() for all . Since r < R was arbitrary,
the result follows.
Theorem 27 (Maximum Principle). Suppose f is analytic on a domain C
and f extends continuously to . Then f attains its maximum modulus on .
Exercise 22. Prove Schwarzs Lemma: Let f be analytic on B(0, 1) with f(0) =
0 and |f(z)| 1 for all z B(0, 1). Then |f

(0)| 1 and |f(z)| |z| for all


z B(0, 1). If |f

(0)| = 1 or if there exists z


0
B(0, 1) such that |f(z
0
)| = |z
0
|,
then there exists C with || = 1 such that f(z) = z for all z B(0, 1).
3. Runges Theorem
This theorem is often skipped in a rst course in complex variables. If you
would like to read a more elementary and more extensive discussion of the theorem
than that given by Hormander, see Conways book [Con78]. In fact, we will follow
Conway to some degree here.
Look again at Theorem 24. One way to restate it is to say that every analytic
function on B(0, R) is the uniform limit on compact subsets of a sequence of poly-
nomials. We consider generalizations. For instance, if we replace B(0, R) by an
arbitrary open set G, is it true that for every f A(G) and every compact K G,
f can be uniformly approximated on K by polynomials? As the next example
shows, the answer is no.
Example 28. Take G := { z : 0 < |z| < 2 }. Then f(z) = z
1
is analytic on G.
Suppose we had a sequence {p
n
} of polynomials such that p
n
f uniformly on
compact subsets of G. In particular, since K := B(0, 1) is a compact subset of G,
THE H

ORMANDER COMPANION 17
p
n
f uniformly on B(0, 1). Thus (if the circle is oriented counter-clockwise)
2i =
_
B(0,1)
z
1
dz
=
_
B(0,1)
lim
n
p
n
(z) dz
= lim
n
_
B(0,1)
p
n
(z) dz
= 0.
Since this is a contradiction, no such sequence of polynomials exists.
We make two remarks concerning this example. First, neither G
c
nor K
c
is
connected. Second, in the notation of Exercise 21, G = a(0, 2), and so every
analytic function on G is a limit (uniform on compact subsets) of a sequence of
rational functions with poles outside of G.
In light of the above, we might try to characterize those G with the property that
every analytic function on G can be approximated uniformly on compact subsets by
polynomials. We might also look for a more general theorem on the approximation
of analytic functions on a set G by rational functions with poles outside G.
Theorems of this sort vary in their precise formulation but tend to go by the name
of Runges Theorem. In many texts such as Conways [Con78] Runges Theorem is
very clearly about the uniform approximation of analytic functions on a compact
set by polynomials, whereas in Hormanders text, no explicit mention is made of
rational functions. We include both statements here so that the reader may compare
them. We will only prove the version stated by Hormander.
Denition 29. Consider C

:= C{}. We put a topology on this set by taking


as a basis for the topology all open discs B(a, r) in C together with sets of the form
{z : |z| > r}{}. Because C

with this topology is homeomorphic to the 2-sphere


S
2
, we visualize C

this way and call it the Riemann sphere.


Theorem 30 (Runges Theorem [Con78]). Let K be a compact subset of C and let
E be an open subset of C

\ K that meets each component of C

\ K. Let be
an open set containing K. If f A() and if > 0 is given, then there exists a
rational function R, all of whose poles lie in E, such that sup
zK
|f(z) R(z)| < .
Theorem 31 (Runges Theorem [Hor90]). Let be an open set in C and let K be
a compact subset of . The following conditions on and K are equivalent:
(a) Every function analytic in a neighborhood of K can be approximated uni-
formly on K by functions in A().
(b) The open set \ K has no component which is relatively compact in .
(c) For every \ K there exists f A() such that
(13) |f()| > sup
K
|f|.
If we take = C, we obtain a theorem on the uniform approximation of analytic
functions by polynomials.
Corollary 32. Let K be a compact subset of C. Every function analytic in a
neighborhood G of K can be uniformly approximated by polynomials on K if and
only if K
c
is connected, or, equivalently, if and only if for every K
c
there exists
a polynomial p in z such that |p()| > sup
K
|p|.
18 JENNIFER HALFPAP
Upon a rst reading, you may nd the statement of this theorem somewhat
intimidating. This is a normal response. It is also likely that jumping right into
the proof will not help. Something useful to do at this point is to look at some
examples. Examples dont prove theorems, but they are the key way we build
intuition.
For simplicity, let us take to be all of C. Dene two compact subsets: K
1
:=
a(1/2, 3/2) and K
2
:= B(0, 1).
Consider rst \ K
1
. This has two components, namely B(0, 1/2) and {z :
|z| > 3/2}. The rst component, call it O, has closure {z : |z| 1/2}, which is a
compact subset of . Therefor the pair (, K
1
) violates (b). If the theorem is true,
we should be able to see that (a) and (c) are violated as well. Indeed, the example
preceding the statements of the theorems essentially addresses (a); the function
f(z) = z
1
is analytic on C\ {0}, which is a neighborhood of K
1
, but this function
is not the uniform limit on K
1
of a sequence of entire functions.
To see that (c) is also violated, we must nd \K
1
such that for all f A(),
|f()| sup
K
1
|f|. Take = 0 and any entire f (since = C). Then since f is
analytic on any B(0, r), by the maximum principle,
|f(0)| sup
|z|=r
|f(z)|,
and in particular
|f(0)| sup
1/2|z|3/2
|f(z)| = sup
K
1
|f|.
On the other hand, \ K
2
= {z : |z| > 1}. This set is connected and its closure
is {z : |z| 1}, which is certainly not compact in . Thus (b) is satised, and
if the theorem is true, (a) and (c) must hold as well. Concerning (a), we have
already proved that if G is an open set containing B(0, 1), any function analytic on
G is the uniform limit on the compact subset B(0, 1) of the sequence of polynomials
consisting of the partial sums of the power series representing the analytic function.
To see that (c) holds, note that if \ K
2
, then || > 1. Consider f(z) = z.
This is entire and
|| = |f()| > 1 = sup
K
2
|f|.
With these examples in mind, we turn to the proof of the theorem.
Exercise 23. Hormander begins by proving that (c) implies (b). Read this part of
his proof, and write out a version of it for yourself in which you ll in any details
you need. Focus especially on the statement that, if O is a component of \K with
compact closure in , then for every f A(),
(14) sup
O
|f| sup
K
|f|.
Proof of (a) (b). This will be a proof by contradiction. Suppose (b) fails, and let
O denote a component of \ K with compact closure in . Since (a) is valid, for
any f analytic in a neighborhood of K, there exists a sequence {f
n
} of elements of
A() such that f
n
f uniformly on K. Inequality (14) yields
sup
O
|f
n
f
m
| sup
K
|f
n
f
m
|.
Thus {f
n
} is also uniformly convergent on O. Let F denote the limit function.
Since F is the uniform limit on compact subsets of O of a sequence of analytic
THE H

ORMANDER COMPANION 19
functions, F is analytic on O. As a uniform limit of continuous functions, it is also
continuous on O. Furthermore, since F and f are both continuous on O and are
the limit of the same sequence there, F = f on O.
The above argument applies to any f analytic in a neighborhood of K. Thus if
O, f(z) :=
1
z
is analytic in the neighborhood C \ {} of K, and we obtain
an associated F analytic in O and continuous on O with F(z) =
1
z
on O. Set
g(z) := (z )F(z). This function is analytic in O and is identically 1 on the
boundary. Thus g 1 on O. Since O and g() = 0, this is a contradiction,
proving that (a) implies (b).
Exercise 24. Prove the assertion made in the penultimate sentence of the preceding
paragraph: If O is a bounded open set and if g is analytic on O and continuous on
O, then if g c on O, then g c on O.
Proof of (b) (a). Hormander intends to apply the Hahn-Banach Theorem in this
part of the proof. See Rudins book [Rud66] for a statement of this theorem as well
as its application in the proof of Runges Theorem. We follow that development
here. Recall,
Theorem 33 (Hahn-Banach, [Rud66]). If M is a subspace of a normed vector
space X and if is a bounded linear functional on M, then can be extended
to a bounded linear functional on all of X without increasing the norm, i.e.,
|||| = ||||.
The following is a consequence:
Proposition 34. Let M be a subspace of a normed vector space X and let x
0
X.
Then x
0
is in the closure M of M if and only if there exists no bounded linear
functional on X with (x) = 0 for all x in M but (x
0
) = 0.
Exercise 25. Prove Proposition 34.
We are now prepared to understand Hormanders application of the Hahn-
Banach theorem. For K and as in part (b), consider C(K), the vector space
of continuous functions on K equipped with the sup-norm. Thus C(K) plays the
role of the normed vector space X above. Let M be the subspace consisting of the
restrictions of elements of A(). Let f be analytic on a neighborhood of K. Then
f X, and the question of whether or not it is the uniform limit on K of elements
of A() is equivalent to the question of whether or not it is in M. By Proposition
34, to prove that f M, it suces to prove that for every continuous linear func-
tional on X with (g) = 0 for every g A(), (f) = 0 as well. Since the set of
all continuous linear functionals on K is precisely the set of complex measures, we
must show that if is a measure such that
_
g d = 0 for all g A(),
then
_
f d = 0.
Exercise 26. With this background, work through the rest of Hormanders proof
of the Runge approximation theorem. Write a summary of the key steps of the
arguments that (b) implies (a) and that (b) implies (c).
We restate Runges Theorem to motivate the next denition: Given a domain
, some compact subsets have the right topological properties that guarantee that
20 JENNIFER HALFPAP
analytic functions on them are close to analytic functions on whereas other com-
pact subsets do not have this property. Given a compact subset K of without
this property, it would be nice to be able to enlarge it to satisfy the hypotheses
of Runges Theorem. Moreover, condition (c) in Runges Theorem gives us an-
other way to describe those points we must add to K. We thus make the following
denition.
Denition 35. Let K be a compact subset of . We dene

K, the A()-hull of
K by
(15)

K =

K

:= { z : |f(z)| sup
K
|f| for every f A() }.
This set has the following properties:
Proposition 36. Let K be a compact subset of ,

Kits A()-hull.
(a) K

K.
(b)

K satises the hypotheses of Runges theorem.
(c) d(K,
c
) = d(

K,
c
).
(d) If ch(K) denotes the convex hull of K,

K ch(K).
Exercise 27. Prove part (c) of Proposition 36. One direction is immediate. For
the other, follow Hormanders suggestion: For
c
, consider the function f(z) =
(z )
1
.
Proof of (b). This simply requires us to unwind the denitions. To show that

K
satises the hypotheses of Runges Theorem, we must show that for all \

K,
there exists f A() with |f()| > sup
K
|f|. If this were not the case, we could
nd
0
\

K such that for all f A(), |f(
0
)| sup
K
|f|. But then such a
0
would be an element of

K. This contradiction establishes the claim.
Proof of (d). We discuss part (d) because it presents us with an opportunity to begin
our exploration of notions related to convexity. We recall a number of denitions.
Denition 37. A subset E of R
n
is convex if for all x, y E, the line segment
joining x and y is contained in E, i.e., (1 t)x +ty E for all t [0, 1].
Denition 38. If E R
n
, the convex hull of E, denoted ch(E), is the intersec-
tion of all convex sets containing E.
We can describe the convex hull of a set E in another way. We focus here on E
closed since this is the situation we have in mind, though one can formulate similar
results for more general E. A hyperplane in R
n
is given by { x :

a
j
x
j
= c }, a
R
n
. Such a hyperplane divides R
n
into two closed half-spaces H
+
:= {

a
j
x
j
c}
and H

:= {

a
j
x
j
c} and is said to be a support hyperplane for a set E if E is
contained in one of the half-spaces and if there exists y E such that

a
j
y
j
= c.
Exercise 28. Let E R
n
be closed.
(1) If y is exterior to ch(E), then there exists a hyperplane

a
j
x
j
= c such
that E H

and

a
j
y
j
> c. Such a hyperplane separates E and {y}.
(2) If ch(E) is not all of R
n
, then ch(E) is the intersection of all support closed
half-spaces containing it.
THE H

ORMANDER COMPANION 21
With this background, we are ready to prove part (d) of Proposition 36. Take
z
0
= x
0
+iy
0


K and suppose that z
0
is not in ch(K). Then there exists (a, b) R
2
and c R such that ax + by c for all z = x + iy K but ax
0
+ by
0
> c. Since
Re[(a ib)(x +iy)] = ax +by, if we set = a +ib, we have
Re( z) c but Re( z
0
) > c.
Now, for any complex , z e
z
is entire, hence in A() for any . Furthermore,
for all z K
|e
z
| = e
Re( z)
e
c
< e
Re( z
0
)
= |e
z
0
|.
This contradicts the assumption that z
0


K, hence

K ch(K). This completes
the proof of part (d) of the proposition.
The topological description of

K. We close this section with a dierent description
of the set

K which matches with the intuition that

K lls in the holes of K.
Theorem 39 (Hormander, Theorem 1.3.3).

K

is the union of K with the com-


ponents of \ K that are relatively compact in .
Proof. Suppose that O is a component of \ K that is relatively compact in .
Then by inequality (14), O

K. Thus if K
1
is the union of K with such rela-
tively compact subsets of \ K, then K
1


K. We wish to establish the reverse
containment.
Since K
1
is the union of K with components of \ K, \ K
1
is open. It follows
that K
1
is closed and in fact compact. Furthermore, \ K
1
has no components
which are relatively compact in . Thus K
1
satises (b) in Runges theorem, and
hence (c). Thus we conclude K
1
=

K
1
. Since

K

K
1
, we conclude

K K
1
, as
desired.
3.1. Meromorphic Functions and the Mittag-Leer Theorem. This is an-
other context in which Hormanders treatment diers substantially from what is
typically done in a course on functions of one complex variable. Thus in this section
we begin by giving the more elementary denitions.
Here, we follow Section 1.4.2 of [Var11]. We begin by dening three kinds of
singularities for functions analytic on a punctured disc.
Denition 40. Let f be analytic on a punctured disc B

(, r) := { z : 0 < |z | <
r } about , with Laurent series

j=
a
j
(z )
j
about .
(1) f has a pole of order n 1 if a
n
= 0 but a
k
= 0 for all k n.
(2) f has a removable singularity at if a
j
= 0 for all j < 0.
(3) f has an essential singularity at if there are innitely many j < 0 for
which a
j
= 0.
Exercise 29. Find all poles of f(z) =
1
z(z
2
+1)
2
and determine the order of each.
Then determine
_

f(z) dz where is a positively-oriented circle of radius R >


1. (You probably know a result called the Residue Theorem that allows you to do
this quickly. If you want to use that theorem, prove it rst. If you dont recall
22 JENNIFER HALFPAP
that theorem, approach this problem by considering the integral over a contour that
includes the circle of radius R together with circles of radius about the poles.)
Exercise 30. Suppose f is analytic on B

(, r).
(1) If f is bounded in a neighborhood of , show that the Laurent series for f
about has no non-zero coecients a
j
for j < 0.
(2) Prove that f has a pole at if and only if lim
z
|f(z)| = .
This last exercise shows that, for a function with a pole at , it makes sense to
assign the value at , so that such a function could be viewed as a function into
C{}. The same can not be said of a function with an essential singularity at .
With this background, we have the following standard denition of a mero-
morphic function.
Denition 41. Let C. f is meromorphic at if there is a punctured
disc centered at on which f is analytic and if is not an essential singularity of
f. f is meromorphic on if it is meromorphic at every point of .
Since an analytic function h can only vanish to at most nite order at a point,
it is clear that the ratio of analytic functions dened on a neighborhood of is
meromorphic at . We will eventually address the question of whether every mero-
morphic functions is a ratio of analytic functions.
Now that we have reviewed the classical treatment of meromorphic functions,
we discuss Hormanders formulation of the concept.
Fix z C. We consider the set of all functions which are analytic on some
neighborhood of z. We dene an equivalence relation on this set so that f g if
there is a neighborhood of z on which f is identically equal to g. We let A
z
be
the set of equivalence classes under this relation, and denote the equivalence class
determined by f by f
z
. With addition and multiplication dened in the obvious
way, A
z
is a ring. Furthermore, it has no zero divisors, and we may form the
quotient eld M
z
:= { f
z
/g
z
: f
z
, g
z
A
z
, g
z
= 0
z
}.
Exercise 31. Let A
z
be as dened above.
(1) How should we dene f
z
+ g
z
? Show that (according to your denition)
f
z
+g
z
is in fact well-dened.
(2) Give all the details of the argument that A
z
has no zero divisors.
We can now state Hormanders denition of a meromorphic function.
Denition 42. Let be an open subset of C. A meromorphic function on
is a mapping :
z
M
z
such that for all z in , (z) M
z
and such that
for all points z
0
in there is a neighborhood and elements f, g A() such that
(z) = f
z
/g
z
for all z . We denote the set of all meromorphic functions on
by M() and write
z
instead of (z) for M().
The aspect of this denition that matches our intuition for meromorphic func-
tions is its explicit reference to quotients of (equivalence classes of) analytic func-
tions. On the other hand, its formulation in terms of equivalence classes may make
you wonder what connection these objects have with analytic functions.
The rst essential remark, then, is that A() can be identied with a subset
of M() in a natural way. Take F A(). For any z , since is open,
F is indeed analytic in a neighborhood of z and hence determines an equivalence
THE H

ORMANDER COMPANION 23
class F
z
. Since the ring A
z
is naturally embedded in M
z
, F
z
is identied with an
element of M
z
. Consider the map dened by z F
z
. In order for this to be a
meromorphic function, there is another condition it must satisfy. Let z
0
be a point
of and consider a neighborhood of z
0
in . Since F A() as well, it is indeed
the case that there is a single pair of analytic functions F and 1 on for which

z
:= F
z
/1
z
for all z . We conclude that the dened in this way is indeed a
meromorphic function.
In order for A() to be identied with a subset of M(), we need to also know
that distinct elements of A() are identied with dierent elements of M(). Thus
suppose F, G A() and that these are not identically equal. Then there exists z
0

such that F(z
0
) = G(z
0
). With respect to the equivalence relation associated with
functions analytic in a neighborhood of z
0
, these two functions are not equivalent,
and so they determine dierent equivalence classes. That is, F
z
0
= G
z
0
in A
z
0
, and
so F
z
0
/1
z
0
= G
z
0
/1
z
0
in M
z
0
.
References
[Con78] John B. Conway. Functions of One Complex Variable, second edition. Grauduate Texts
in Mathematics. Springer, 1978.
[Fol99] Gerald Folland. Real Analysis: Modern Techniques and Their Applications. John Wiley
& Sons, Inc., 1999.
[GP74] Victor Guillemin and Alan Pollack. Dierential Topology. Prentice Hall, 1974.
[Hor90] Lars Hormander. An Introduction to Complex Analysis in Several Variables, third edi-
tion. North Holland, 1990.
[Rud66] Walter Rudin. Real and Complex Analysis. McGraw-Hill, 1966.
[Str00] Robert Strichartz. The Way of Analysis. Jones and Bartlett, 2000.
[Var11] Dror Varolin. Riemann Surfaces by Way of Complex Analytic Geometry. Graduate Stud-
ies in Mathematics. American Mathematical Society, 2011.

Vous aimerez peut-être aussi