Lecture Notes in Complex Analysis

Eric T. Sawyer
McMaster University, Hamilton, Ontario
E-mail address: sawyer@mcmaster.ca
Contents
Preface v
Part 1. Complex Dierentiation 1
Chapter 1. The Real Field 3
1. The derivative of a real-valued function on R 3
Chapter 2. The Complex Field 7
1. Two derivatives 8
2. The exponential function 9
3. Complex partial derivatives 11
4. The Cauchy-Goursat Theorem 13
5. Cauchys representation formula 15
Chapter 3. Properties of holomorphic functions 19
1. Zeroes of analytic functions 19
2. Isolated singularities of analytic functions 20
3. Cauchys estimates, maximum principle, and uniform convergence 22
4. Normal families 25
5. The open mapping theorem 29
Chapter 4. The Riemann Mapping Theorem 33
1. Automorphisms of the disk 33
2. Simply connected subsets of the plane 36
3. Proof of the Riemann Mapping Theorem 38
Chapter 5. Contour integrals and the Prime Number Theorem 43
1. Analytic continuation of (:) 46
2. The residue theorem 50
3. Proof of the Prime Number Theorem 52
Part 2. Boundary behaviour of Riemann maps 63
Chapter 6. The Poisson representation 65
1. Harmonic functions 69
2. Fatous Theorem 73
Chapter 7. Extending Riemann maps 77
1. Lindelfs theorem 77
2. Proof of Carathodorys Theorem 79
Appendix A. Topology 83
iii
iv CONTENTS
1. Homotopy and index 83
2. The Jordan Curve Theorem 85
Appendix B. Lebesgue measure theory 97
1. Lebesgue measure on the real line 99
2. Measurable functions and integration 105
Appendix. Bibliography 115
Preface
These notes grew out of lectures given three times a week in a fourth year under-
graduate course in complex analysis at McMaster University January to April 2009.
Id like to thank all the students taking the course, in particular Martin Munoz and
Preston Wake, for a number of interesting observations and for correcting many
mistakes.
Analysis in the complex eld C has many seemingly magical advantages over
analysis in the real eld R. For example, if a function ) from an open subset \ C
to C has at every point . \, a derivative in the complex sense:
)
t
(.) = lim
u(C):
) (n) ) (.)
n .
,
then ) actually has a power series expansion
) (.) =
o

n=0
a
n
(. n)
n
,
valid for all . 1(n, 1) whenver 1(n, 1) \. In particular ) is innitely
dierentiable in \. This is in striking contrast to the situation for real-valued
functions ) that are dierentiable on an interval in the usual real sense - the function
) (r) = r
2
sin
1
r
is dierentiable on the real line, but its derivative )
t
fails to be even
continuous at the origin. A famous example of Weierstrass exhibits a continuously
dierentiable function ) that nowhere has a second derivative!
There is a plethora of special properties enjoyed by functions having a complex
derivative at every point in an open subset of the complex plane. These include in
particular Cauchys theorem and representation formula, the maximum modulus
principle, the open mapping theorem, uniform convergence and Montels theorem,
the residue theorem and many others as given in Rudin [5] and Boas [1]. An
application to the Prime Number Theorem is given following Stein and Shakarchi
[6] where a very important complex dierentiable function is studied - the Riemann
zeta function and Eulers formula,
(:) =

n=1
1
:
s
=

1
1
1

s
,
where the innite product is taken over all prime numbers.
On the other hand, the famous Riemann Mapping Theorem shows that there
are lots of these special functions, enough to realize any homeomorphism of the unit
disk to a proper subset of the plane by a biholomorphic function! This latter theo-
rem requires antidierentiation, hence integration over paths, in simply connected
domains. We treat this matter using taxicab paths as introduced in [1], together
v
vi PREFACE
with an elementary approach to the Jordan Curve Theorem for such paths initiated
in [6].
In the second part of these notes we turn to Carathodorys characterization
of when a Riemann map extends to a homeomorphism up to the boundary. This
introduces the study of continuous closed curves in the complex plane, including
simple boundary points and the Jordan Curve Theorem for general curves.
Moreover, we require for the rst time Lebesgues theory of the integral in
order to obtain Fatous Theorem on the existence of radial limits for bounded
holomorphic functions in the unit disk. Lebesgues theory renders the space of
square integrable functions on the circle complete, and thus allows both the Cauchy
integral and the Poisson integral to integrate boundary values on the unit circle.
The Poisson integral has two important advantages over the Cauchy integral: its
much smaller size permits the study of radial limits, and its real (positive) values
permit the development of a theory of harmonic functions. Together with the
maximal theorem from real analysis, this circle of ideas provides one of the early
triumphs of Lebesgues integral over that of Riemann, which until then had suced
for the purposes of complex analysis.
Part 1
Complex Dierentiation
We begin Part 1 with a short review chapter on the concepts of dierentiation
using eld and metric properties in R, and dierentiation using vector space and
metric properties in R
n
. In Chapter 2, we introduce complex dierentiation using
eld and metric properties in C, and compare this to the real dierentiation arising
in Chapter 1 from the identication of C and R
2
. The striking Cauchy-Goursat the-
orem on complex derivatives is proved, along with Cauchys representation formula
and the consequent power series expansions of holomorphic functions. Chapter 3
develops the elementary properties of holomorphic functions; zeroes and singular-
ities, the maximum principle, uniform convergence and normal families, and the
open mapping theorem. Chapter 4 uses the pivotal Schwarz lemma and the theory
of simply connected domains to prove the Riemann Mapping Theorem. Finally,
Chapter 5 uses the residue theorem and analytic continuation of the Riemann zeta
function to prove the Prime Number Theorem.
Recall that the eld of real numbers R can be constructed either from Dedekind
cuts of rational numbers Q, or from Weierstrass Cauchy sequences of rational
numbers. A Dedekind cut c Q is a "left innite interval open on the right" of
rational numbers that is associated with the "real number" on the number line that
marks its right hand endpoint. More precisely, a cut c is a subset of Q satisfying
(here j and denote rational numbers)
c ,= ? and c ,= Q,
j c and < j implies c,
j c implies there is c with j < .
One can dene an ordered eld structure on the set of cuts, and this is identied
as the eld R (see e.g. Chapter 1 of [4]). Alternatively, one can dene an ordered
eld structure on the set of equivalence classes of Cauchy sequences in Q, and this
produces an ordered eld isomorphic to R.
The eld of complex numbers can now be easily constructed from the real linear
space R
2
by dening the following multiplication on ordered pairs in R
2
:
(r, j) (n, ) = (rn j, r jn) , (r, j) , (n, ) R
2
.
It is a routine matter to verify that a eld structure is dened on R
2
with addition
as usual in R
2
, multiplication as above, (0, 0) the additive identity, and (1, 0) the
multiplicative identity. It is customary to introduce an abstract symbol i and write
R
2
= R(1, i) = r ij : r, j R .
Thus i represents the ordered pair (0, 1) and hence i
2
= 1. The eld constructed
above is now denoted C and an element . C has a unique representation as
. = r ij where r and j are called the real and imaginary parts of .. As there is
no square root of 1 in the ordered eld R, the number i was historically considered
"not real, but imaginary", and this accounts for the terminology used today.
CHAPTER 1
The Real Field
We take as known the real eld R and its association with the "points in a
line". In addition to the eld structure on R , we also consider the metric topology
given by the distance function
d (r, j) = [r j[ , r, j R.
With these algebraic and topological denitions in hand, we can proceed with
analysis, in particular the denition of the derivative )
t
of a function ) : R R.
1. The derivative of a real-valued function on R
Definition 1. Given a function ) : R R and a point r R, we dene
(1.1) )
t
(r) = lim
r
) (j) ) (r)
j r
= lim
|0
) (r /) ) (r)
/
,
for those r R for which the limit exists.
We can also rewrite the second form of the limit using Landaus "little oh"
notation as
(1.2) ) (r /) = ) (r) )
t
(r) / o (/) ,
where o (/) denotes a function of / satisfying
o(|)
|
0 as / 0. We say that
the function ) is dierentiable at r if (1.1) (equivalently (1.2)) holds. From (1.2)
we note that if ) is dierentiable at r then it is also continuous at r. From these
denitions it is easy to prove the standard "calculus" of derivatives. First we have
the calculus of the eld operations.
Proposition 1. Suppose that ) and q are both functions dierentiable at the
point r R, and suppose that c R represents the constant function. Then we
have
(1) () q)
t
(r) = )
t
(r) q
t
(r) ,
(2) (c))
t
(r) = c)
t
(r) ,
(3) ()q)
t
(r) = )
t
(r) q (r) ) (r) q
t
(r) ,
(4)
_
}

_
t
(r) =
}
0
(r)(r)}(r)
0
(r)
(r)
2
provided q (r) ,= 0.
3
4 1. THE REAL FIELD
For example, to prove (3) we use the formulation (1.1) of derivative and the
corresponding properties of limits to obtain
()q)
t
(r) = lim
r
()q) (j) ()q) (r)
j r
= lim
r
_
) (j) q (j) ) (r) q (j)
j r

) (r) q (j) ) (r) q (r)
j r
_
= lim
r
) (j) ) (r)
j r
lim
r
q (j) ) (r) lim
r
q (j) q (r)
j r
= )
t
(r) q (r) ) (r) q
t
(r) .
Second we have the calculus of composition of functions, the so-called "chain rule".
Proposition 2. Suppose that ) is dierentiable at r and that q is dierentiable
at j = ) (r). Then
(1) (q ))
t
(r) = q
t
(j) )
t
(r) = q
t
() (r)) )
t
(r) ,
(2)
_
)
1
_
t
(j) =
1
}
0
(r)
=
1
}
0
(}
1
())
provided ) is invertible near r.
For example, to prove (1) we use the Landau formulation (1.2) of derivative
and the corresponding properties of limits as follows. Write
) (r /
1
) = ) (r) )
t
(r) /
1
o
1
(/
1
) ,
q (j /
2
) = q (j) q
t
(j) /
2
o
2
(/
2
) ,
and then with /
2
= )
t
(r) /
1
o
1
(/
1
) we have,
(q )) (r /
1
) = q () (r /
1
))
= q () (r) )
t
(r) /
1
o
1
(/
1
))
= q (j /
2
)
= q (j) q
t
(j) /
2
o
2
(/
2
)
= (q )) (r) q
t
(j) )
t
(r) /
1
o
1
(/
1
)
o
2
()
t
(r) /
1
o
1
(/
1
))
= (q )) (r) q
t
(j) )
t
(r) /
1
o
3
(/
1
)
where it is easy to see that
o
3
(/
1
)
/
1
=
q
t
(j) o
1
(/
1
) o
2
()
t
(r) /
1
o
1
(/
1
))
/
1
0 as /
1
0.
To prove (2) we rst note that since ) is continuous and one-to-one, it follows
easily that ) is open, hence )
1
is continuous. Then with obvious notation,
)
1
(j /) )
1
(j)
/
=
/
) (r /) ) (r)

1
)
t
(r)
as / 0.
Example 1. There is a function ) : R R whose derivative )
t
: R R exists
everywhere on the real line, but the derivative function )
t
is not itself dierentiable
at 0. For example
) (r) =
_
r
2
sin
1
r
if r ,= 0
0 if r = 0
1. THE DERIVATIVE OF A REAL-VALUED FUNCTION ON R 5
has these properties. Indeed,
)
t
(r) =
_
2rsin
1
r
cos
1
r
if r ,= 0
0 if r = 0
fails even to be continuous at the origin.
1.1. Derivatives in Euclidean space. We can extend the denition of de-
rivative to functions ) : R
n
R
n
from one Euclidean space to another using the
Landau formulation (1.2) together with the vector space structure of R
n
and the
metric topology on R
n
given by the distance function
d (r, j) =

_
n

|=1
[r
|
j
|
[
2
, r = (r
|
)
n
|=1
, j = (j
|
)
n
|=1
.
More precisely, we say that a function ) : R
n
R
n
is dierentiable at r =
(r
|
)
n
|=1
R
n
if there is an :: matrix (equivalently a linear map from R
n
to
R
n
) such that
) (r /) = ) (r) / o (/) ,
for all / R
n
and where the function o (/) taking values in R
n
satises
]o(|)]
]|]
0
as / 0 = (0, ..., 0) in R
n
. When such a matrix exists, it is easily seen to be
unique and we dene the derivative 1) (r) at r to be the linear map from R
n
to
R
n
with matrix representation . If ) : R
n
R
n
is dierentiable on all of R
n
,
then 1) : R
n
/(R
n
, R
n
) where /(R
n
, R
n
) denotes the space of linear maps
from R
n
to R
n
. When 1) (r) exists, we thus have
(1.3) ) (r /) = ) (r) 1) (r) / o (/) , / R
n
,
which can be interpreted as stating that the linear map 1) (r) is the "best linear
approximation" to the nonlinear map / ) (r /) ) (r). In paricular, ) is
continuous at r if it is dierentiable at r. The calculus in Proposition 1 extends
easily with obvious modications to this setting with similar proofs.
Finally, we recall that the matrix representation [1) (r)[ of )
t
(r) has entries
given by the partial derivatives
0)

0r
|
(r) = lim
q0
)

(r
1
, ..., r
|
j, ..., r
n
) )

(r
1
, ..., r
n
)
j
,
of the components )

of ) with respect to the /

||
variable r
|
:
(1.4) [1) (r)[ =
_
0)

0r
|
(r)
_
1n
1|n
=
_
_
J}1
Jr1
(r)
J}1
Jrr
(r)
J}r
Jrr
(r)
J}r
Jrr
(r)
_
_
.
The chain rule, Proposition 2, here takes the form
1(q )) (r) = 1q (j) 1) (r) ,
where ) : R
n
R
n
is dierentiable at r and q : R
n
R

is dierentiable at
j = ) (r), and where the matix representations of the derivatives satisfy
[1(q )) (r)[ = [1q (j)[ [1) (r)[ ,
with the multiplication of [1q (j)[ and [1) (r)[ on the right side being matrix
multiplication of a j : matrix times an : : matrix. The proof is similar to
that in the case : = : = 1 given above.
CHAPTER 2
The Complex Field
We also take as known the complex eld C and its association with the "points
in the plane R
2
" where . = r ij C is associated with (r, j) R
2
. The eld
structure on C uses the multiplication rule
.n = (r ij) (n i) = (rn j) i (r jn) ,
where . = r ij and n = n i. If we associate . = r ij to the matrix
_
r j
j r
_
, then this multiplication corresponds to matrix multiplication:
[.[ [n[ =
_
r j
j r
_ _
n
n
_
(0.5)
=
_
rn j r jn
jn r j rn
_
= [.n[ .
Since the matrix
_
r j
j r
_
= r
_
cos 0 sin0
sin0 cos 0
_
is dilation by the nonnegative number r =
_
r
2
j
2
and rotation by the angle
0 = lan
1
r
in the counterclockwise direction, we see that if . has polar coordinates
(r, 0) and n has polar coordinates (:, c), then .n has polar coordinates (r:, 0 c).
We also consider the metric topology on C given by the distance function
d (., n) =
_
[r n[
2
[j [
2
, . = r ij, n = n i,
which coincides with the usual distance function in the plane R
2
. With these
algebraic and topological denitions in hand, we can dene derivatives )
t
(r) of
functions ) : C C at points r just as in (1.1) and (1.2), but where r, j, / now
denote complex numbers. Then the analogues of the two calculus propositions
above hold for the complex eld as well.
Definition 2. Given a function ) : C C and a point . C, we dene
)
t
(.) = lim
u:
) (n) ) (.)
n .
= lim
|0
) (. /) ) (.)
/
,
for those . C for which the limit exists. Equivalently, using Landaus "little oh"
notation this is
) (. /) = ) (.) )
t
(.) / o (/) ,
where o (/) denotes a function of / satisfying
o(|)
|
0 as / 0.
Analogues of Propositions 1 and 2 hold for complex derivatives. In particular,
since a constant function ) (.) = c has derivative 0, and the identity function ) (.) =
7
8 2. THE COMPLEX FIELD
. has derivative 1, we see that the polynomial ) (.) =

n=0
a
n
.
n
is holomorphic
in C and has derivative )
t
(.) =

n=1
:a
n
.
n1
.
1. Two derivatives
We now have two dierent denitions of derivative of a function ) : C C,
namely the complex derivative )
t
(.) using the eld structure of C, and also the
real derivative 1) (r, j) given in (1.3) of Chapter 1 with . = r ij using the
identication of C with R
2
. Note that if ) has a complex derivative )
t
(.) at
. = r ij, then (1.3) holds with the linear map 1) (r, j) /
_
R
2
, R
2
_
given by
(1.1) / = c i, )
t
(.) /, / C = R
2
.
If we write the real and imaginary components of ) as n and , i.e. ) (n) =
n(n) i (n), then (1.4) shows that
(1.2) [1) (r, j)[ =
_
Ju
Jr
(r, j)
Ju
J
(r, j)
Ju
Jr
(r, j)
Ju
J
(r, j)
_
.
On the other hand, (0.5) shows that
(1.3) [)
t
(.)[ =
_
a /
/ a
_
,
for some real numbers a and /. Equating (1.2) and (1.3) gives the Cauchy-Riemann
equations:
0n
0r
(.) =
0
0j
(.) , (1.4)
0n
0j
(.) =
0
0r
(.) .
Conversely, it is easy to see that if 1) (r, j) exists and the Cauchy-Riemann equa-
tions (1.4) hold, then ) = n i has a complex derivative at . = r ij. We
summarize this discussion in the following proposition.
Proposition 3. Suppose that ) : C C and that the real derivative 1) (r, j)
exists at . = rij. Then ) = ni has a complex derivative )
t
(.) at . if and only
if the Cauchy-Riemann equations (1.4) hold, equivalently if and only if 1) (r, j) is
the composition of a dilation and a rotation.
However, magical properties arise from the denition of derivative in the com-
plex eld! For example, if ) : C C is complex dierentiable at all points in C,
then the derivative function )
t
: C C is also complex dierentiable at all points
in C. As a consequence, the derivative )
tt
of )
t
exists and is dierentiable in C and
in fact ) is complex dierentiable of all orders. This is a far cry from the example
given at the end of the rst chapter. Note however that the function
) (.) =
_
.
2
sin
1
:
if . ,= 0
0 if . = 0
,
obtained by replacing r with . in Example 1, fails to be bounded in any neigh-
bourhood of the origin in C. Here sin. =
t
1z
t
1z
2I
where c
:
is dened in the next
section.
Definition 3. Let \ be an open subset of the complex plane C. We say that
) : \ C is holomorphic in \ if )
t
(.) exists for all . \.
2. THE EXPONENTIAL FUNCTION 9
The proof that holomorphic functions are innitely dierentiable is not easy,
and it will occupy the next few sections. But rst we review the important expo-
nential function, and then the useful complex partial derivatives
J
J:
and
J
J:
.
2. The exponential function
Definition 4. For . C dene
c
:
= oxp. =
o

n=0
.
n
:!
.
Note that by the root test for series, the series

o
n=0
:
r
n!
converges absolutely
and uniformly in each compact subset of the complex plane. We have the exponent
formula:
(2.1) c
:+u
= c
:
c
u
, ., n C.
To see this we write
c
:+u
=
o

n=0
(. n)
n
:!
=
o

n=0
1
:!
n

|=0
_
:
/
_
.
|
n
n|
=
o

n=0
n

|=0
1
/! (: /)!
.
|
n
n|
=
o

|=0
o

=0
1
/!,!
.
|
n

=
o

|=0
1
/!
.
|
n

=0
1
,!
n

= c
:
c
u
,
where the absolute convergence of the exponential series justies the substitution
, = : / with : xed.
We also compute that
c
I0
=
o

n=0
(i0)
n
:!
(2.2)
=
_
1
0
2
2!

0
4
4!

_
i
_
0
0
3
8!

0
5
!

_
= cos 0 i sin0,
and using the exponent formula we obtain de Moivres formula:
_
c
I0
_
n
= c
In0
= cos :0 i sin:0.
Thus if . = r ij has polar coordinates (r, 0), then
. = r cos 0 ir sin0 = rc
I0
and
.
n
= r
n
c
In0
.
More generally, if n = :c
I
, then
.n = r:c
I(0+)
,
so that as we observed earlier, the moduli r, : of ., n multiply as nonnegative real
numbers, and the arguments 0, c of ., n add as real numbers modulo 2.
10 2. THE COMPLEX FIELD
Finally, we observe that if ) : C C by ) (.) = c
:
, then
(2.3)
)
t
(.) = lim
|0
c
:+|
c
:
/
= lim
|0
c
:
c
|
1
/
= c
:
lim
|0
_
1
/
2!

/
2
8!
...
_
= c
:
= ) (.) .
Thus we see that c
:
= c
r
c
I
is a nonvanishing holomorphic function on C. Further
properties of the exponential function are
c
I0
= cos 0 i sin0 = cos 0 i sin0 = cos (0) i sin(0) = c
I0
,

c
I0

2
= c
I0
c
I0
= c
I0
c
I0
= c
I0I0
= c
0
= 1,
cos
2
0 sin
2
0 = [cos 0 i sin0[
2
=

c
I0

2
= 1,
d
d0
cos 0 i
d
d0
sin0 =
d
d0
c
I0
= ic
I0
= i (cos 0 i sin0) = sin0 i cos 0,
d
d0
cos 0 = sin0 and
d
d0
sin0 = cos 0.
It now follows easily that if
t
2
is dened to be the smallest positive root of cos 0
(which exists because of the intermediate value theorem), then c
:
is 2i periodic,
maps the imaginary axis onto the unit circle, and has range C 0.
2.1. Power series. We were able to show in (2.3) that the exponential func-
tion had a complex derivative by exploiting the exponent formula (2.1). In fact
all power series

o
n=0
a
n
.
n
are holomorphic in their open disks of convergence.
Note that by the root test, the derived series

o
n=1
:a
n
.
n1
also has radius of
convergence 1.
Theorem 1. Suppose the power series

o
n=0
a
n
.
n
has radius of convergence
1 0, and dene
) (.) =
o

n=0
a
n
.
n
, . 1(0, 1) .
Then ) is holomorphic in 1(0, 1) and
(2.4) )
t
(.) =
o

n=1
:a
n
.
n1
, . 1(0, 1) .
Proof : To see (2.4), dene q (.) =

o
n=1
:a
n
.
n1
and use the identity
a
n
/
n
= (a /)
n

=1
a
n
/
1
3. COMPLEX PARTIAL DERIVATIVES 11
twice to obtain:
) (n) ) (.)
n .
q (.) =
o

n=1
a
n
_
n
n
.
n
n .
:.
n1
_
=
o

n=2
a
n
_
n

|=1
n
n|
.
|1
:.
n1
_
=
o

n=2
a
n
_
n

|=1
.
|1
_
n
n|
.
n|
_
_
=
o

n=2
a
n
_
n

|=1
.
|1
(n .)
n|

|=1
n
n||
.
|1
_
= (n .)
o

n=2
a
n
_
n1

|=1
n|

|=1
n
n||
.
|+|2
_
= (n .)
o

n=2
a
n
_
_
_
n

=2
(, 1) n
n
.
2
_
_
_
.
Thus if [.[ , [n[ < j < 1, then

) (n) ) (.)
n .
q (.)

_ [n .[
o

n=2
[a
n
[
:(: 1)
2
j
n2
0
as n . since

o
n=2
[a
n
[
n(n1)
2
j
n2
< . This latter series is nite since the twice
derived series

o
n=2
a
n
:(: 1) .
n2
has radius of convergence 1 and is absolutely
convergent in its disk of convergence.
Of course the above applies equally well to a power series

o
n=0
a
n
(. a)
n
in
the disk 1(a, 1) where 1 is the radius of convergence. We will see later that every
holomorphic function in a disk 1(a, 1) has a power series representation in that
disk.
3. Complex partial derivatives
First we recall a special case of Stokes theorem in the plane. Suppose that \
is a convex open subset of the plane whose boundary 0\ is an oriented piecewise
continuously dierentiable simple closed curve in the plane. Then if
1 = (r, j) dr 1(r, j) dj
is a continuously dierentiable one-form on \, we have
(3.1)
_
J
1 =
_

d1
12 2. THE COMPLEX FIELD
where
d1 = d(r, j) . dr d1(r, j) . dj
=
_
0
0r
(r, j) dr
0
0j
(r, j) dj
_
. dr

_
01
0r
(r, j) dr
01
0j
(r, j) dj
_
. dj
=
_
01
0r
(r, j)
0
0j
(r, j) dj
_
drdj
is the exterior derivative of 1. This version (3.1) of Stokes theorem is usually
called Greens theorem.
However, when dealing with complex dierentiable functions, it is more conve-
nient to use variables intimately connected with complex derivatives, namely
. = r ij and . = r ij,
r =
1
2
(. .) and j =
1
2i
(. .) .
We also use the associated partial derivatives dened by
0
0.
=
1
2
_
0
0r

1
i
0
0j
_
and
0
0.
=
1
2
_
0
0r

1
i
0
0j
_
.
Note that these denitions are suggested by a formal application of the chain rule:
0
0.
=
0r
0.
0
0r

0j
0.
0
0j
=
1
2
0
0r

1
2i
0
0j
,
0
0.
=
0r
0.
0
0r

0j
0.
0
0j
=
1
2
0
0r

1
2i
0
0j
.
Now we compute that
d. . d. = d (r ij) . d (r ij) = 2idrdj,
and so Stokes theorem (3.1) for the one-form (.) d. (the general one-form is given
in this notation by (.) d. 1(.) d.) becomes
_
J
(.) d. =
_

d (.) d. =
_

_
0
0.
(.) d.
0
0.
(.) d.
_
. d. (3.2)
=
_

0
0.
(.) d. . d. = 2i
_

0
0.
(.) drdj.
If ) = n i, the equation
0)
0.
(.) = 0
is equivalent to the Cauchy-Riemann equations (1.4). As a consequence we obtain
from (3.2) that a special case of Cauchys theorem holds for complex dierentiable
functions with a continuous derivative (only later will we see that complex dif-
ferentiable functions ) are automatically innitely dierentiable, so that Cauchys
theorem holds without the assumption of continuity on )
t
).
Theorem 2. Let \ be a convex open subset of the plane whose boundary 0\ is
an oriented piecewise continuously dierentiable simple closed curve in the plane.
4. THE CAUCHY-GOURSAT THEOREM 13
Suppose that ) is holomorphic in \, and that ) and )
t
extend continuously to \.
Then
_
J
) (.) d. = 0.
Proof : Greens theorem yields
_
J
) (.) d. = 2i
_

0)
0.
(.) drdj = 2i
_

0drdj = 0,
since
J}
J:
(.) = 0 when ) is complex dierentiable at ..
4. The Cauchy-Goursat Theorem
In this section we obtain a variant of Theorem 2 without the assumption that
)
t
is continuous in \.
Theorem 3. Let \ be a convex open subset of the plane, and let be an oriented
piecewise continuously dierentiable simple closed curve in \. If ) is holomorphic
in \, then
(4.1)
_
~
) (.) d. = 0.
Proof : We proceed in a sequence of three steps as follows.
Step 1: If is a triangle, then (4.1) holds.
This is the main step in the proof. Set
0
1
= and let

0
1
=
_
~
0
1
) (.) d..
We divide the triangle
0
1
into four congruent triangles
1
1
,
1
2
,
1
3
,
1
4
by joining the
midpoints of the sides of
0
1
. Clearly

0
1
=
4

|=1

1
|
where

1
|
=
_
~
1
!
) (.) d., 1 _ / _ 4.
Now since

0
1

_

4
|=1

1
|

, we may assume by relabelling that

1
1

_
1
4

0
1

.
Now divide the triangle
1
1
into four congruent triangles
2
1
,
2
2
,
2
3
,
2
4
by joining the
midpoints of the sides of
1
1
. With

2
|
=
_
~
2
!
) (.) d., 1 _ / _ 4,
we may again assume by relabelling that

2
1

_
1
4

1
1

.
Continuing in this manner we obtain a sequence of triangles
0
1

1
1

2
1

...
n
1
... whose closed convex hulls H
n
1
contain a unique point .
0
\. We have
dia:(
n
1
) = 2
n
dia:
_

0
1
_
|c:qt/(
n
1
) = 2
n
|c:qt/
_

0
1
_
,
arca (H
n
1
) = 4
n
arca
_
H
0
1
_
,
[
n
1
[ _ 4
n

0
1

.
14 2. THE COMPLEX FIELD
Now from Theorem 2 applied to the function . ) (.
0
))
t
(.
0
) (. .
0
) (since
this function has an antiderivative, namely ) (.
0
) . )
t
(.
0
)
1
2
(. .
0
)
2
, we could
instead use the fundamental theorem of line integrals), we obtain
_
~
r
1
) (.
0
) )
t
(.
0
) (. .
0
) d. = 0, : _ 1.
Thus since ) is complex dierentiable at .
0
we conclude with /
n
= dia:(
n
1
) =
2
n
dia:
_

0
1
_
that
[
n
1
[ =

_
~
r
!
) (.) d.

_
~
r
1
) (.) ) (.
0
) )
t
(.
0
) (. .
0
) d.

_ o (dia:(
n
1
)) |c:qt/(
n
1
)
_ 2
n
dia:
_

0
1
_
o (/
n
)
/
n
2
n
|c:qt/
_

0
1
_
_ [
n
1
[
|c:qt/
_

0
1
_
[
0
1
[
o (/
n
)
/
n
,
which implies that [
n
1
[ = 0 as soon as
o(|r)
|r
<
[.
0
1
[
ltn||(~
0
1
)
. Thus

0
1

_ 4
n
[
n
1
[ = 0
and (4.1) holds when is a triangle.
Remark 1. In the special case where is a triangle, we need only assume
that ) is continuous in \, and holomorphic in \ j for some exceptional point
j \. Indeed, Step 1 already proves the case when j is outside . If j is a vertex
of = (a, /, c), say a = j, we consider points r and j arbitrarily close to j on
the sides [a, /[ and [a, c[ respectively of . Then (4.1) holds for replaced by either
(r, /, c) or (r, j, c), and since the perimeter of (r, j, a) is as small as we
wish, we obtain (4.1). The cases where j is on an edge or in the interior of are
now easy to establish by decomposing into appropriate subtriangles.
Step 2: There is a holomorphic antiderivative 1 of ) in \, i.e. 1
t
(.) = ) (.)
for . \.
Fix .
0
\ and dene
1 (.) =
_
[:0,:]
) (n) dn, . \,
where [.
0
, .[ denotes the curve whose image is the straight line segment joining
.
0
to ., parameterized for example by (t) = .
0
t (. .
0
), 0 _ t _ 1. By Step 1
we have
_
[:0,:]
) (n) dn
_
[:,:+|]
) (n) dn
_
[:+|,:0]
) (n) dn = 0,
and so
1 (. /) 1 (.)
/
=
1
/
_
_
[:0,:+|]
) (n) dn
_
[:0,:]
) (n) dn
_
=
1
/
_
[:,:+|]
) (n) dn
=
1
/
_
1
0
) (. t/) /dt ) (.)
5. CAUCHYS REPRESENTATION FORMULA 15
as / 0.
Step 3:
_
~
) (.) d. = 0.
Parameterize as (t), a _ t _ / (without loss of generality we may assume
that is continuously dierentiable in (a, /)). Let 1 be an antiderivative of ) as
in Step 2. The chain rule shows that
J
J|
1 ( (t)) = 1
t
( (t))
t
(t) (a formal and
nonrigorous proof of this is that
1 ( (t /)) 1 ( (t))
/
=
1 ( (t /)) 1 ( (t))
(t /) (t)
(t /) (t)
/
tends to 1
t
( (t))
t
(t) as / 0). We now compute that
_
~
) (.) d. =
_
~
1
t
(.) d. =
_
b
o
1
t
( (t))
t
(t) dt
=
_
b
o
d
dt
1 ( (t)) dt = 1 ( (/)) 1 ( (a)) = 0
since (/) = (a) when is a closed curve.
Corollary 1. Theorem 3 holds if we merely assume that ) is continuous in
\, and holomorphic in \ j for some exceptional point j \.
5. Cauchys representation formula
A remarkable property of holomorphic functions ) is that they are uniquely
determined inside a "nice" domain by their boundary values, and moreover there is
a simple formula for recovering the interior values ) (.) from the boundary values
) (n), namely
(5.1) ) (.) =
1
2i
_
J
) (n)
n .
dn, . \.
In order to state and prove a precise version of Cauchys formula (5.1), we need the
notion of the index of a curve in the complex plane relative to a point . C.
Definition 5. If is a closed piecewise continuously dierentiable curve in
the plane, and if . C
+
(we denote by
+
the image of in the plane), we dene
the index of about . to be
1:d
~
(.) =
1
2i
_
~
1
n .
dn.
Proposition 4. Suppose that is a closed piecewise continuously dierentiable
curve in the plane. Then the function 1:d
~
(.) is integer-valued, constant in each
component of C
+
, and vanishes in the unbounded component of C
+
.
Proof : Let (t), a _ t _ /, be a parameterization of and x . C
+
.
Heuristically, we expect the antiderivative of
~
0
(|)
~(|):
to be log ( (t) .) where log
is an inverse function to oxp. Then we would get
1:d
~
(.) =
_
b
o

t
(t)
(t) .
dt = log ( (t) .) [
b
o
= log
(/) .
(a) .
= log 1.
Since oxp is 2i periodic and c
0
= 1, log 1 2i:
nZ
, and this would prove the
proposition.
16 2. THE COMPLEX FIELD
Unfortunately, we have not yet dened a logarithm function, so we proceed
via a route that achieves the same goal without actually using a logarithm. We
accomplish this with the function
,(:) = oxp
__
s
o
1
(t) .

t
(t) dt
_
, a _ : _ /.
Since 1:d
~
(.) =
_
b
o
1
~(|):

t
(t) dt, we see that 1:d
~
(.) is an integer if and only if
,(/) = oxp1:d
~
(.) equals 1. To show that ,(/) = 1 it suces to show that
(5.2)
d
d:
,(:)
(:) .
= 0, a _ : _ /,
since then
,(/)
(/) .
=
,(a)
(a) .
=
1
(/) .
,
as ,(a) = 1 and (a) = (/) (since is a closed path). Actually, we can only show
(5.2) at points : where (:) is continuously dierentiable, but this is easily seen to
be enough. For this we compute
,
t
(:) = ,(:)

t
(:)
(:) .
,
so that
d
d:
,(:)
(:) .
=
( (:) .) ,
t
(:) ,(:)
t
(:)
( (:) .)
2
= 0.
Since 1:d
~
(.) is continuous in C
+
, we conclude that 1:d
~
(.) is a constant
integer in each component of C
+
, and since lim
:o
1:d
~
(.) = 0, we conclude
that 1:d
~
(.) vanishes in the unbounded component of C
+
.
Corollary 2. If is the positively oriented circle centered at a with radius r,
then
1:d
~
(.) =
_
1 if [. a[ < r
0 if [. a[ r
.
Proof : If we parameterize by (t) = c
I|
, 0 _ t _ 2, then
1:d
~
(0) =
1
2i
_
~
1
n 0
dn =
1
2i
_
2t
0
1
c
I|
0
ic
I|
dt =
1
2i
_
2t
0
idt = 1.
We can now give Cauchys representation formula. For convenience we dene
a path to be a piecewise continuously dierentiable curve in the plane.
Theorem 4. Suppose that is a closed path in a convex set \. If ) is holo-
morphic in \ and . \
+
, then
) (.) 1:d
~
(.) =
1
2i
_
~
) (n)
n .
dn.
Proof : Fix . \
+
and for n \ dene
q (n) =
_
}(u)}(:)
u:
if n ,= .
)
t
(.) if n = .
.
5. CAUCHYS REPRESENTATION FORMULA 17
Then q is continuous in \, and holomorphic in \ ., and so Corollary 1 shows
that
1
2i
_
~
) (n)
n .
dn ) (.) 1:d
~
(.) =
1
2i
_
~
) (n) ) (.)
n .
dn =
1
2i
_
~
q (n) dn = 0.
Theorem 5. Suppose that ) is holomorphic in a disk 1(a, 1). Then there is
a power series

o
n=0
a
n
(. a)
n
having radius of convergence 1 such that
) (.) =
o

n=0
a
n
(. a)
n
, . 1(a, 1) .
Moreover, ) is innitely complex dierentiable and
)
(n)
(.) =
o

n=n
a
n
:!
(: :)!
(. a)
nn
, . 1(a, 1) .
In particular,
)
(n)
(a) = :!a
n
, : = 0, 1, 2, ...
and we have Taylors formula:
) (.) =
o

n=0
)
(n)
(a)
:!
(. a)
n
, . 1(a, 1) .
Proof : We assume without loss of generality that a = 0. Then if is the
positively oriented circle of radius r < 1 about the origin, we have from Theorem
4 that for . 1(0, r):
) (.) =
1
2i
_
~
) (n)
n .
dn =
1
2i
_
2t
0
)
_
rc
I0
_
rc
I0
.
irc
I0
d0
=
1
2
_
2t
0
)
_
rc
I0
_
1
:
:t
10
d0 =
1
2
_
2t
0
o

n=0
_
.
rc
I0
_
n
)
_
rc
I0
_
d0
=
o

n=0
_
1
2
_
2t
0
)
_
rc
I0
_
r
n
c
In0
d0
_
.
n
=
o

n=0
a
n
.
n
.
The above interchange of summation and integration follows from the uniform
convergence of the geometric series in the unit disk. It follows that the power series

o
n=0
a
n
.
n
has radius of convergence at least r. Moreover, by iterating (2.4) we
see that ) is innitely dierentiable in 1(0, r) and that a
n
=
}
(r)
(o)
n!
for : _ 0.
Since r was any positive number less than 1, the proof is complete.
Definition 6. Let \ be an open set in the plane. A function ) : \ C is
analytic in \ if ) has a power series representation with radius of convergence 1
in each open disk 1(a, 1) contained in \.
Combining Theorems 1 and 5, we see that a function ) : \ C is holomorphic
in \ if and only if it is analytic in \. Example 1 shows that this fails miserably for
derivatives in the real eld.
CHAPTER 3
Properties of holomorphic functions
In this chapter we develop some of the surprising properties of holomorphic
functions, beginning with the dichotomy of zero sets, and the dichotomy of isolated
singularities, of analytic functions. Then we investigate the maximum modulus
theorem and some of its consequences, leading to the theory of normal families and
the open mapping theorem. These tools will be used in the next chapter to prove
the remarkable Riemann Mapping Theorem that characterizes the biholomorphic
images of the open unit disk as comprising all open simply connected proper subsets
of the complex plane.
1. Zeroes of analytic functions
Suppose a holomorphic function ) is dened in a disk 1(a, 1) and vanishes at
the center a. The fact that ) has a power series expansion ) (.) =

o
n=0
a
n
(. a)
n
with radius of convergence at least 1 results in a dichotomy of just two possibilites:
(1) either all of the coecients a
n
are zero and ) vanishes in the entire ball
1(a, 1),
(2) or there is a rst coecient a

that is non-zero and then

(1.1) ) (.) = (. a)

q (.)
where q is holomorphic in 1(a, 1) and nonvanishing at a: indeed
) (.) =
o

n=
a
n
(. a)
n
= (. a)

_
a

a
+1
(. a) a
+2
(. a)
2
...
_
= (. a)

q (.) .
The positive integer is uniquely determined in the second possibility and is
called the order of the zero of ) at the point a. By convention we say that ) has
a zero of order 0 at a if ) (a) ,= 0. In a connected open set, this phenomenon takes
the following form.
Theorem 6. Suppose \ is open and connected and ) H (\). Then the zero
set 7 = . \ : ) (.) = 0 of ) in \ is either all of \ or is a discrete subset of
\ (this means 7 has no limit point in \). In the latter case there is associated to
each point a 7 a unique positive integer such that ) (.) = (. a)

q (.) where
q H (\) and q (a) ,= 0.
Proof : The set 7 is closed in \ since ) is continuous. We claim that the
interior

7 of 7 is also a closed subset of \. Indeed, if a \

7 is a limit point
of

7, then a 7 and the rst possibility of the dichotomy above fails, and so the
19
20 3. PROPERTIES OF HOLOMORPHIC FUNCTIONS
second possibility holds. Then we conclude from the continuity of q in (1.1) that a
is isolated in 7, contradicting the assumption that a is a limit point of

7. Thus

7
is both open and closed in \. Since \ is connected, we conclude that either

7 = \
(in which case ) is identically zero in \) or

7 = c. In the latter case every point
a 7 is an isolated point since only the second possibililty can hold
Corollary 3. (The Coincidence Principle) Suppose that ), q H (\) where
\ is open and connected. If ) (.) = q (.) in some set 1 of points having a limit
point in \, then ) = q in all of \.
Proof : Since 1 is a subset of the zero set of ) q, the previous theorem shows
that 1 = \.
Example 2. Dene the holomorphic functions sin and cos in the complex plane
by the series for cos 0 and sin0 in (2.2), but replacing the real variable 0 by the
complex variable ., i.e.
cos . = 1
.
2
2!

.
4
4!
...
sin. = .
.
3
8!

.
5
!
...
Then
(1.2) sin(. n) = sin. cos n cos . sinn, ., n C.
To prove this we take as known that
(1.3) sin(r j) = sinrcos j cos rsinj, r, j R,
By the coincidence principle we obtain that
(1.4) sin(. j) = sin. cos j cos . sinj, . C, j R,
since for xed j R, both sides of (1.4) are holomorphic functions of . C that
coincide on the real line by (1.3). Now x . C in (1.2), and note that both sides
of (1.2) are holomorphic functions of n C that coincide on the real line by (1.4).
Another application of the coincidence principle now proves (1.2).
2. Isolated singularities of analytic functions
Suppose a holomorphic function ) is dened in a punctured disk
1
t
(a, 1) = 1(a, 1) a ,
and has a singularity at the center a in the sense that there is no analytic function
1 in the entire ball 1(a, 1) whose restriction to 1
t
(a, 1) is ). In the case that
such an 1 does exist, we say that ) has a removable singularity at a. For an actual
(nonremovable) singularity there is a dichotomy of just two possibilites:
(1) either lim
:o
[) (.)[ = and there is a positive integer such that
/(.) = (. a)

) (.) , . 1
t
(a, 1) ,
has a removable singularity at a and /(a) ,= 0,
(2) or the image ) (1
t
(a, r)) under ) of each punctured disk 1
t
(a, r), 0 <
r < 1, is dense in the complex plane C. In this case lim
:o
) (.) fails to
exist in the most spectacular way possible, namely the cluster set of ) at
a is C.
2. ISOLATED SINGULARITIES OF ANALYTIC FUNCTIONS 21
In possibility (1) we say that ) has a pole of order at a, and in possibility (2)
we say that ) has an essential singularity at a. In order to prove the dichotomy, we
begin with Riemanns theorem on removable singularities, which can be restated as
saying that if ) has an actual isolated singularity at a, then ) must be unbounded
in every neighbourhood of a.
Theorem 7. Suppose ) is holomorphic and bounded in a punctured disk 1
t
(a, 1).
Then ) has a removable singularity at a.
Proof : Let /(.) =
_
(. a)
2
) (.) if . 1
t
(a, 1)
0 if . = a
. Then /
t
(a) = 0
and so / H (1(a, 1)). By Theorem 5, / has a power series expansion in 1(a, 1):
/(.) =
o

n=0
/
(n)
(a)
:!
(. a)
n
, . 1(a, 1) .
Since /(a) = /
t
(a) = 0, we have
(. a)
2
) (.) = /(.) =
o

n=2
/
(n)
(a)
:!
(. a)
n
= (. a)
2
o

n=0
/
(n+2)
(a)
(: 2)!
(. a)
n
,
for . 1
t
(a, 1), and hence ) is the restriction to 1
t
(a, 1) of the holomorphic
function 1 (.) =

o
n=0
|
(r+2)
(o)
(n+2)!
(. a)
n
, . 1(a, 1).
Now we can prove the dichotomy of isolated singular points. Indeed, suppose
that ) has a (nonremovable) singularity at a and that possibility (2) fails. Then
there is 0 < r < 1 and a disk 1(n, -) in the plane such that
) (1
t
(a, r)) 1(n, -) = c.
It follows that
q (.) =
1
) (.) n
, . 1
t
(a, r) ,
is bounded ([q (.)[ _
1
:
) and nonvanishing in the punctured disk 1
t
(a, r). By
Riemanns theorem there is G H (1(a, r)) which restricts to q in 1
t
(a, r).
We must have G(a) = 0 since otherwise solving G(.) =
1
}(:)u
for ) shows
that
) (.) = n
1
G(.)
, . 1
t
(a, r) ,
has a removable singularity at a, a contradiction. It now follows from Theorem 6
that there is a unique positive integer such that
G(.) = (. a)

H (.) , . 1(a, r) ,
with H holomorphic and nonvanishing in all of 1(a, r). Thus we have
) (.) = n
1
(. a)

H (.)
, . 1
t
(a, r) .
It follows that
lim
:o
[) (.)[ = ,
and that
/(.) =
_
(. a)

) (.) if . 1
t
(a, 1)
n(. a)

1
1(:)
if . 1(a, r)
,
22 3. PROPERTIES OF HOLOMORPHIC FUNCTIONS
denes a holomorphic function in 1(a, 1) with /(a) =
1
1(o)
,= 0. This proves that
possibility (1) holds.
We note that when possibility (1) holds, we can write
n(. a)

1
H (.)
=
o

n=0
a
n
(. a)
n
as a power series and obtain the representation,
) (.) = (. a)

n=0
a
n
(. a)
n
=
o

n=0
a
n
(. a)
n
=
a
0
(. a)

...
a
1
. a
a

a
+1
(. a) ...
which is usually written
) (.) =
o

n=
/
n
(. a)
n
=
/

(. a)

...
/
1
. a

o

n=0
/
n
(. a)
n
,
for . 1
t
(a, 1). This motivates calling such a singularity a pole of order at a.
The polynomial in
1
:o
,
1

(.) =
/

(. a)

...
/
1
. a
,
is called the principal part of ) at the pole a.
3. Cauchys estimates, maximum principle, and uniform convergence
Theorem 8. Suppose \ is open and convex, and that is a closed path in \.
If ) H (\), then
(3.1) )
(n)
(.) 1:d
~
(.) =
:!
2i
_
~
) (n)
(n .)
n+1
dn, . \
+
, : _ 0.
Proof : For ., . / in the same component of \
+
, Cauchys formula shows
that
) (. /) ) (.)
/
1:d
~
(.) =
1
2i
_
~
1
/
_
1
n . /

1
n .
_
) (n) dn (3.2)
=
1
2i
_
~
_
1
(n . /) (n .)
_
) (n) dn

1
2i
_
~
) (n)
(n .)
2
dn
as / 0 since
1
u:|

1
u:
uniformly on
+
. Thus we have
)
t
(.) 1:d
~
(.) =
1
2i
_
~
) (n)
(n .)
2
dn, . 1(a, r) ,
and (3.1) now follows by repetition of the argument.
3. CAUCHYS ESTIMATES, MAXIMUM PRINCIPLE, AND UNIFORM CONVERGENCE 23
Corollary 4. (Cauchys estimates) If ) H (1(a, 1)), then for 0 _ : < ,

)
(n)
(a)

_
:!
r
n
_
1
2
_
2t
0

)
_
a rc
I0
_

d0
_
, 0 < r < 1, (3.3)

)
(n)
(a)

_
:!
1
n
_
sup
0<:<1
1
2
_
2t
0

)
_
a rc
I0
_

d0
_
_
:!
1
n
sup
:1(o,1)
[) (.)[ .
Proof : For 0 < r < 1 we have

)
(n)
(a)

:!
2i
_
~
) (n)
(n a)
n+1
dn

:!
2i
_
2t
0
)
_
a rc
I0
_
(rc
I0
)
n+1
irc
I0
d0

_
:!
2
_
2t
0

)
_
a rc
I0
_

r
n
d0.
The second inequality in (3.3) follows easily.
Theorem 9. (Maximum principle) Suppose that \ is open and connected. If
) H (\) then ) cannot have a strict local maximum in \. If ) has a local
maximum, it must be constant in \.
Proof : Suppose, in order to derive a contraction, that ) has a strict local
maximum at a \. Then there is 1(a, 1) \ such that
[) (a)[ [) (.)[ , . 1
t
(a, 1) .
Then for every 0 < r < 1 we have from the rst inequality in (3.3),
[) (a)[ _
1
2
_
2t
0

)
_
a rc
I0
_

d0 <
1
2
_
2t
0
[) (a)[ d0 = [) (a)[ ,
the desired contradiction. Now suppose only that ) has a local maximum at a \.
Then there is 1(a, 1) \ such that
[) (a)[ _ [) (.)[ , . 1
t
(a, 1) .
Then for every 0 < r < 1 we have from the rst inequality in (3.3),
[) (a)[ _
1
2
_
2t
0

)
_
a rc
I0
_

d0 _
1
2
_
2t
0
[) (a)[ d0 = [) (a)[ ,
and it follows from the continuity of ) that

)
_
a rc
I0
_

= [) (a)[ for all arc

I0

1(a, 1).
Thus we have [) (a)[
2
= [) (.)[
2
for . 1(a, 1). If ) (a) = 0 we have that ) is
the constant 0 in 1(a, 1), and otherwise we have
0 =
0
0.
[) (a)[
2
=
0
0.
_
) (.) ) (.)
_
= )
t
(.) ) (.), . 1(a, 1) ,
since
J
J:
) (.) =
J
J:
) (.) = 0 by the Cauchy-Riemann equations. Thus )
t
(.) = 0 in
1(a, 1) and so ) is again constant in 1(a, 1) since
) (.) ) (a) =
_
[o,:]
)
t
(n) dn = 0, . 1(a, 1) .
Finally, the coincidence principle implies that ) is constant in \.
24 3. PROPERTIES OF HOLOMORPHIC FUNCTIONS
Remark 2. Theorem 9 fails to hold with "maximum" replaced by "minimum"
as evidenced by the function ) (.) = . in \ = 1(0, 1). However, if we assume
in addition that ) is nonvanishing in \, then the theorem holds with "maximum"
replaced by "minimum" throughout (simply apply Theorem 9 to
1
}(:)
).
Corollary 5. Suppose \ is bounded, open and connected. If ) H (\)
C
_
\
_
, then
[) (.)[ _ sup
uJ
[) (n)[ , . \,
with strict inequality unless ) is constant in \. If in addition, ) is nonvanishing
in \, then
[) (.)[ _ inf
uJ
[) (n)[ , . \,
with strict inequality unless ) is constant in \.
Here is a famous application of the second inequality in (3.3).
Theorem 10. (Liouvilles theorem) If ) : C C is holomorphic and bounded,
then ) is constant.
Proof : From the second inequality in (3.3) we obtain
[)
t
(a)[ _
1
1
n
sup
:1(o,1)
[) (.)[ 0 as 1 ,
or )
t
(a) = 0 for all a C. It follows that ) is constant.
Problem 1. Suppose more generally that ) : C C is holomorphic and has
polynomial growth at innity, i.e. [) (.)[ _ C
_
1 [.[

_
for some positive constant
C and positive integer . Show that ) (.) is a polynomial of degree at most .
Hint: The second inequality in (3.3) yields

)
(n)
(a)

_
:!
1
n
sup
:1(o,1)
[) (.)[ _
:!
1
n
C
_
1 ([a[ 1)

_
.
Show that this vanishes in the limit as 1 whenever : .
Another consequence of the second inequality in (3.3) concerns uniform con-
vergence of a sequence of holomorphic functions.
We take for granted that if is a simple closed path in the plane, then C
+
has a single bounded component and 1:d
~
(.) = 1 for . 1 if surrounds 1 in
the positive direction (the same assertion for a simple closed curve is the dicult
Jordan Curve Theorem).
Theorem 11. (Uniform convergence theorem) Suppose that \ is an open con-
vex set in the plane and that is a simple closed path in \. If )
n

o
n=1
is a sequence
of holomorphic functions in \ that converge uniformly on
+
, then )
n

o
n=1
con-
verges uniformly on compact subsets of the bounded component 1of C
+
to a
holomorphic function in 1. Moreover, )
t
n

o
n=1
converges uniformly on compact
subsets of 1.
4. NORMAL FAMILIES 25
Proof : If 1 is a compact subset of 1, and c = di:t (1, 1
c
), then c 0 and
for . 1 we have
[)
n
(.) )
n
(.)[ =

1
2i
_
~
)
n
(n) )
n
(n)
n .
dn

_
1
2
_
b
o
[)
n
(n) )
n
(n)[
c
[
t
(t)[ dt
_
1
2c
|c:qt/() sup
u~

[)
n
(n) )
n
(n)[ .
Thus )
n

o
n=1
converges uniformly on 1 to a continuous function ) in 1. But since
we have
) (.) = lim
no
)
n
(.) = lim
no
1
2i
_
~
)
n
(n)
n .
dn =
1
2i
_
~
) (n)
n .
dn,
for . 1, the calculation in (3.2) shows that ) is holomorphic in 1, and moreover,
)
t
(.) =
1
2i
_
~
) (n)
(n .)
2
dn.
For convenience we repeat the argument here using 1:d
~
(.) = 1:d
~
(. /) = 1
for ., . / 1:
)
t
(.) = lim
|0
) (. /) ) (.)
/
= lim
|0
1
2i
_
~
1
/
_
1
n . /

1
n .
_
) (n) dn
=
1
2i
_
~
_
lim
|0
1
(n . /) (n .)
_
) (n) dn
=
1
2i
_
~
) (n)
(n .)
2
dn.
Since
)
t
n
(.) =
1
2i
_
~
)
n
(n)
(n .)
2
dn,
the argument above using )
n
) uniformly on
+
, shows that )
t
n
converges uni-
formly to )
t
on compact subsets of 1.
4. Normal families
Here we show that every sequence of holomorphic functions on an open set
\, that is uniformly bounded on compact subsets of \, has a subsequence that
converges uniformly on compact subsets of \ to a holomorphic function. The proof
will use the above theorem on uniform convergence together with the Arzela-Ascoli
theorem. We begin with the statement and proof of the Arzela-Ascoli theorem, one
of the most useful real-variable theorems in analysis.
Suppose 1 is a compact metric space with metric d
1
. We dene the metric
space ( (1) of continuous complex-valued functions on 1 by
( (1) = ) : 1 C : ) is continuous on 1
26 3. PROPERTIES OF HOLOMORPHIC FUNCTIONS
with metric
d
c(1)
(), q) = sup
r1
[) (r) q (r)[ , ), q ( (1) .
Recall that ( (1) is a complete metric space (every Cauchy sequence in ( (1)
converges) and that a sequence )
n

o
n=1
( (1) converges in the metric space
( (1) if and only if it converges uniformly on 1. Recall also that a continuous
function ) on a compact metric space 1 is actually uniformly continuous on 1:
for every - 0 there is c 0 such that
[) (r) ) (j)[ < -, whenever d
1
(r, j) < c.
A family T ( (1) of continuous functions on 1 is called equicontinuous if for
every - 0 there is c 0 such that
(4.1) [) (r) ) (j)[ < - whenever d
1
(r, j) < c and ) T.
Finally, we say that a family T ( (1) is pointwise bounded on 1 if
sup
}J
[) (r)[ < for each r 1.
Theorem 12. (Arzela-Ascoli theorem) Suppose that 1 is a compact metric
space and that )
n

o
n=1
( (1) is a sequence of continuous functions on 1. If the
sequence )
n

o
n=1
is both pointwise bounded and equicontinuous on 1, then:
(1) The sequence )
n

o
n=1
is uniformly bounded.
(2) There is a subsequence )
n
!

o
|=1
that converges in ( (1).
Proof : (1) We rst use equicontinuity of )
n

o
n=1
and the compactness of 1
to improve the pointwise boundedness of )
n

o
n=1
to actual boundedness in the
metric space ( (1). By equicontinuity there is c 0 so that [) (r) ) (j)[ <
1 for d
1
(r, j) < c. Now select a nite set of balls 1
1
(r
|
, 1)

|=1
that cover
the compact set 1 (the collection of all balls with unit radius covers 1). Then
'
r
!
= sup
n
[)
n
(r
|
)[ < for each r
|
by pointwise boundedness, and so ' =
max
1|
'
r
!
< . But then we have for any r 1, if r 1
1
(r
|
, 1),
[) (r)[ _ [) (r) ) (r
|
)[ [) (r
|
)[ < 1 '
r
!
_ 1 '.
Thus )
n

o
n=1
1
c(1)
(0, ' 1).
(2) We proceed in three steps.
Step 1: 1 has a countable dense subset 1.
For each : N there is a nite set of balls
_
1
1
_
r
n
|
,
1
n
__
r
|=1
that cover 1.
Clearly the set 1 =

o
n=1
r
n
|

r
|=1
is countable and dense in 1.
Step 2: There is a subsequence q
n

o
n=1
of )
n

o
n=1
that converges on 1.
Relabel 1 as 1 = c
|

o
|=1
. There is a subsequence )
n
!

o
|=1
of )
n

o
n=1
such that )
n
!
(c
1
)
o
|=1
converges in C. There is then a subsequence
_
)
n
!

_
o
=1
of )
n
!

o
|=1
such that
_
)
n
!

(c
2
)
_
o
=1
converges in C, and of course
_
)
n
!

(c
1
)
_
o
=1
converges as well. Repeating this procedure we obtain for each / N sequences :
|
=
_
)
|
n
_
o
n=1
such that :
1
is a subsequence of :
0
= )
n

o
n=1
, and :
|+1
is a subsequence
of :
|
for all / N. We also have that :
|
(c
|
) converges for 1 _ / _ /. Now Cantors
diagonal sequence q
n

o
n=1
= )
n
n

o
n=1
converges at each c
|
in 1.
Step 3: q
n

o
n=1
converges uniformly on 1.
4. NORMAL FAMILIES 27
Let - 0 and by equicontinuity choose c 0 so that
[)
n
(r) )
n
(j)[ < -
1
whenever d
1
(r, j) < c and : _ 1,
where -
1
0 will be chosen later. Now let 1
1
(j

, c)

=1
be a nite set of balls
centered at points j

in 1 and with radius c that cover 1. For each , choose

so that
[q
n
(j

) q
n
(j

)[ < -
2
, :, : _

,
where -
2
0 will be chosen later. Now set = max
1

. Then for :, : _
and r 1, with say r 1
1
(j

, c), we have
[q
n
(r) q
n
(r)[ = [q
n
(r) q
n
(j

) q
n
(j

) q
n
(j

) q
n
(j

) q
n
(r)[
_ [q
n
(r) q
n
(j

)[ [q
n
(j

) q
n
(j

)[ [q
n
(j

) q
n
(r)[
< -
1
-
2
-
1
< -
provided we choose 2-
1
-
2
< -. This shows that q
n

o
n=1
converges uniformly on
1, and since ( (1) is complete, q
n

o
n=1
converges in ( (1).
Remark 3. Using the above theorem, it can be shown that a subset T of ( (1) is
compact if and only if T is closed, bounded and equicontinuous. Indeed, a compact
set in any metric space is easily shown to be closed and bounded. To see that T
is also equicontinuous, let - 0 and select a nite collection
_
1
c(1)
()
|
, -
1
)
_

|=1
of balls in ( (1) centered at )
|
with radius -
1
that cover T. Since )
|
is uniformly
continuous there is c
|
0 such that
[)
|
(r) )
|
(j)[ < -
2
whenever d
1
(r, j) < c
|
.
Set c = min
1|
c
|
0. Then if d
1
(r, j) < c and ) T, say ) 1
c(1)
()
|
, -
1
),
then
[) (r) ) (j)[ = [) (r) )
|
(r) )
|
(r) )
|
(j) )
|
(j) ) (j)[
_ [) (r) )
|
(r)[ [)
|
(r) )
|
(j)[ [)
|
(j) ) (j)[
< -
1
-
2
-
1
< -
if we choose 2-
1
-
2
< -. Thus (4.1) holds.
Conversely, the Arzela-Ascoli theorem shows that every innite set 1 in T has a
limit point ) in ( (1), and since T is closed, ) T. Now it is a general fact that
a metric space A is compact if and only if every innite subset of A has a limit
point in A. So we are done by the "if" statement of this general fact. The proof of
this statement is a bit delicate. Using that A is contained in a nite union of balls
of radius
1
n
for each : N, one rst shows that there is a countable dense set 1 in
A. Then the collection of balls E = 1(r, r) : r 1, r Q (0, 1) is a countable
base. Now suppose that G
o

o.
is an open cover of A. For each r A there is
an index c and a ball 1
r
E such that
(4.2) r 1
r
G
o
.
Note that the axiom of choice is not needed here since E is countable, hence well-
ordered. If we can show that the cover

E = 1
r

r
has a nite subcover, then
(4.2) shows that G
o

o.
has a nite subcover as well. So it remains to show that
28 3. PROPERTIES OF HOLOMORPHIC FUNCTIONS

E has a nite subcover. Relabel the countable base

E as

E = 1
n

o
n=1
. Assume, in
order to derive a contradiction, that

E has no nite subcover. Then the sets
1

= A
_

_
|=1
1
n
_
are nonempty closed sets that are decreasing, i.e. 1
+1
1

, and that have empty

intersection. Thus if we choose r

for each , the set 1 =

o
=1
r

must
be an innite set, and so has a limit point r A. But then the fact that the 1

are closed and decreasing implies that r 1

for all , the desired contradiction.

Definition 7. A family T H (\) of holomorphic functions on an open set
\ is said to be normal if every sequence of functions from T has a subsequence that
converges uniformly on compact subsets of \ (but not necessarily to a function in
T). In other words, T is "sequentially precompact".
Theorem 13. (Montels theorem) If T H (\) is uniformly bounded on each
compact subset of \, then T is a normal family.
Proof : Let 1
n

o
n=1
be a seqence of compact subsets of \ satisfying
1
n


1
n+1
1
n+1
, : _ 1,
'
o
n=1
1
n
= \.
In view of the Arzela-Ascoli theorem and the uniform convergence theorem, it
suces to prove that the restriction of T to the compact subset 1
n
is equicontinuous
(since this will show that every sequence in T has a subsequence that converges
uniformly on 1
n
; then we apply Cantors diagonal trick).
So to prove that T [
1r
is equicontinuous, let
c
n
= di:t (1
n
, 01
n+1
) 0,
and cover 1
n
with nitely many disks
_
1
_
.
|
,
or
4
__
r
|=1
. If ., n 1
_
.
|
,
or
2
_
and
) T, we have
[) (.) ) (n)[ =

_
[u,:]
)
t
(n) dn

_ [. n[ sup
1(:
!
,
or
2
)
[)
t
()[
_ [. n[
_
1
_
or
2
_ sup
1(:
!
,or)
[) ()[
_
,
by Cauchys inequality since for each 1
_
.
|
,
or
2
_
, 1
_
,
or
2
_
1(.
|
, c
n
). Now
we note that 1(.
|
, c
n
)

1
n+1
1
n+1
and by hypothesis, there is a constant
'
n+1
such that sup
1r+1
[) ()[ _ '
n+1
for ) T. Thus we obtain
(4.3) [) (.) ) (n)[ _
2'
n+1
c
n
[. n[ , ., n 1
_
.
|
,
c
n
2
_
, ) T.
Finally, we take a pair of points ., n 1
n
such that [. n[ <
or
4
. There is /
so that . 1
_
.
|
,
or
4
_
. Then n 1
_
.
|
,
or
2
_
and so (4.3) yields
[) (.) ) (n)[ _
2'
n+1
c
n
[. n[ , ) T,
5. THE OPEN MAPPING THEOREM 29
which easily implies the equicontinuity of T [
1r
.
5. The open mapping theorem
An important topological property of a holomorphic map ) : \ C on a
connected open set is that either ) is constant or ) is an open map, i. e. ) (G) is
open for all open subests G of \. This should be compared with the Invariance of
Domain theorem of Brouwer that can use homotopy to conclude that any one-to-
one continuous map on an open subset of the plane is an open map. More generally
this can be extended to R
n
, : 2, using homology (see e.g. page 172 of [2]).
Theorem 14. (Open mapping theorem) If ) is holomorphic on an open con-
nected set \ in the complex plane, then ) (\) is either a single point or an open
set.
Proof : Suppose that ) is not constant, and x a \. We may suppose that
) (a) = 0. By the coincidence principle, there is 1(a, 1) \ such that ) (.) ,= 0
for all . 01(a, 1). By continuity of ) and compactness of 01(a, 1),
c = min
:J1(o,1)
[) (.)[ 0.
We claim that
(5.1) 1
_
0,
c
2
_
) (1(a, 1)) ) (\) ,
which clearly completes the proof that ) (\) is open.
To see (5.1), choose n 1
_
0,
o
2
_
and note that for . 01(a, 1),
c _ [) (.)[ _ [) (.) n[ [n[ < [) (.) n[
c
2
.
Thus
[) (a) n[ = [n[ <
c
2
_ min
:J1(o,1)
[) (.) n[ ,
and now the mimimum principle, Remark 2, implies that ) (.) n cannot be non-
vanishing in any open neighourhood of the closed ball 1(a, 1). It follows that
) (.
0
) = n for some .
0
1(a, 1) and this completes the proof that (5.1) holds.
5.1. Locally injective holomorphic functions. If ) is holomorphic in a
neighbourhood of a point a where )
t
(a) ,= 0, then there is 1 0 such that ) is
one-to-one in the disk 1(a, 1). Indeed, choose 1 so small that [)
t
(.) )
t
(a)[ <
1
2
[)
t
(a)[ for . 1(a, 1). Then we have for n
0
, n
1
1(a, 1),
) (n
1
) ) (n
0
) =
_
[u0,u1]
)
t
(.) d. =
_
[u0,u1]
)
t
(a) d.
_
[u0,u1]
[)
t
(.) )
t
(a)[ d..
Now
_
[u0,u1]
)
t
(a) d. = )
t
(a) (n
1
n
0
)
and

_
[u0,u1]
[)
t
(.) )
t
(a)[ d.

_
1
2
[)
t
(a)[ [n
1
n
0
[ ,
and it follows that
[) (n
1
) ) (n
0
)[
1
2
[)
t
(a)[ [n
1
n
0
[ 0.
30 3. PROPERTIES OF HOLOMORPHIC FUNCTIONS
Combining this observation with the open mapping theorem we get:
Proposition 5. If ) is holomorphic in a neighbourhood of a point a where
)
t
(a) ,= 0, then there are open sets l and \ such that a l and ) : l \ is
one-to-one and onto.
We will also need to know that if the derivative )
t
of a holomorphic function
) vanishes at a point a, then ) is not one-to-one in any neighbourhood of a. This
can be proved in many ways. We will use the fact that the Jacobian J
}
(.) of a
holomorphic function ) = n i at the point . is given by
J
}
(.) = ool
_
n
r
n

_
= n
r

r
n

= n
2
r

2
r
=

0
0r
) (.)

2
= [)
t
(.)[
2
,
where the rst equality in the second line follows from the Cauchy-Riemann equa-
tions. The idea of the proof is to show that if )
t
(a) = 0, then the integral of
the Jacobian J
}
over a small disk 1(a, 1) is too big to be the area of the image
) (1(a, 1)).
Proposition 6. Suppose that ) H (\) and that )
t
(a) = 0 for some a \.
Then ) is not one-to-one in any disk 1(a, 1) \.
Proof : Suppose without loss of generality that a = 0 and that ) is not identi-
cally zero near the origin. Then
q (.) = ) (.) ) (0) = .
n
/(.)
where / H (\) satises /(0) ,= 0. Since )
t
(0) = 0 we must have : _ 2.
Now assume in order to derive a contradiction that ) is one-to-one in some disk
1(0, 1) \, hence in every disk 1(0, r) with 0 < r _ 1. Then the change of
variable formula for q implies that
arca (q (1(0, r))) =
_
1(0,:)
J

drdj, 0 < r _ 1.
However, given - 0,
q (1(0, r)) 1(0, r
n
([/(0)[ -))
for r suciently small, and then
arca (q (1(0, r))) _ r
2n
([/(0)[ -)
2
.
On the other hand,
J

(.) = [q
t
(.)[
2
=

:.
n1
/(.) .
n
/
t
(.)

2
= [.[
2n2
[:/(.) ./
t
(.)[
2
_ [.[
2n2
:
2
([/(0)[ -)
2
for [.[ suciently small, and then
_
1(0,:)
J

drdj _ 2
_
1(0,:)
:
2n2
:
2
([/(0)[ -)
2
:d:
= 2:
2
([/(0)[ -)
2
r
2n
2:
,
5. THE OPEN MAPPING THEOREM 31
for r suciently small. Altogether then, for r suciently small,
2 _ : _
_
[/(0)[ -
[/(0)[ -
_
2
,
which is a contradiction if we take - 0 small enough.
Corollary 6. A holomorphic function ) dened in a neighbourhood of a point
a is one-to-one in some neighbourhood of a if and only if )
t
(a) ,= 0.
CHAPTER 4
The Riemann Mapping Theorem
The previous chapter established that holomorphic functions, those having a
complex derivative on an open set in the complex plane C, have many "magical"
properties compared to those functions having merely a real derivative. The pur-
pose of this chapter is to show that, on the other hand, holomorphic functions
are numerous enough to supply homeomorphisms between arbitrary nontrivial con-
nected and simply connected open subsets of the plane. We will follow for the most
part the treatment in Chapter 14 of Rudin [5].
Theorem 15. (Riemann Mapping Theorem) A subset \ of the complex plane
is the image ) (D) of a one-to-one onto holomorphic map ) : D \ if and only if
(1) \ is open
(2) \ is connected and simply connected
(3) \ ,= C
A one-to-one onto holomorphic map ) : D \ is called a Riemann mapping
for \. The open mapping theorem shows that )
1
: \ D is continuous, hence a
homeomorphism between \ and D, and Proposition 6 shows that )
t
is nonvanishing.
Thus )
1
is also holomorphic with derivative
_
)
1
_
t
(n) = lim
|0
_
)
1
_
(n /)
_
)
1
_
(n)
/
= lim
|0
/
) (. /) ) (.)
=
1
)
t
(.)
,
where ) (.) = n and ) (. /) = n/. Thus a Riemann mapping is a biholomorphic
map (meaning both the map and its inverse are holomorphic).
Now we can easily obtain the necessity of properties (1), (2) and (3) in Theorem
15. Indeed, (1) is necessary by the open mapping theorem, (2) follows from topology
and the fact that ) is a homeomorphism between D and \, and (3) now follows
from an application of Liouvilles theorem to the holomorphic inverse function )
1
:
\ D.
The question of characterizing all the Riemann maps ) for \, in terms of one
of them, is easily reduced to the special case \ = D, to which we turn in the next
section. The Riemann maps for D are the biholomorphic maps from D to itself,
usually called the automorphisms of D. These maps play an important role in the
proof of the Riemann Mapping Theorem.
1. Automorphisms of the disk
We begin by demonstrating that the only automorphisms of the disk that x
the center 0, are the rotations l
0
(.) = c
I0
., . D. This will follow from the
Schwarz lemma below. Part (2) of this lemma will prove to be a very powerful tool.
33
34 4. THE RIEMANN MAPPING THEOREM
Lemma 1. (The Schwarz Lemma) Suppose that ) : D D is holomorphic and
) (0) = 0.
(1) Then
[) (.)[ _ [.[ , . D, (1.1)
[)
t
(0)[ _ 1.
(2) If equality holds for at least one . in the rst line, or if equality holds
in the second line, then there is 0 [0, 2) such that ) is the rotation
) (.) = c
I0
., . D.
Proof : The function q (.) =
}(:)
:
has a removable singularity at the origin and
by the maximum principle,
sup
:1(0,:)
[q (.)[ _ sup
:J1(0,:)
[) (.)[
[.[
_
1
r
, 0 < r < 1.
Thus we obtain sup
:D
[q (.)[ _ 1, and hence (1.1) since q (0) = )
t
(0). If equality
holds as in part (2) of the lemma, then [q (.)[ attains its maximum inside the disk,
and so q is a constant of modulus one, hence ) is a rotation.
Corollary 7. The only automorphisms of D that x 0 are the rotations.
Proof : Let ) : D D be an automorphism. Then )
1
: D D is also an
automorphism, and by the chain rule,
(1.2) 1 = )
t
(0)
_
)
1
_
t
(0) .
Now part (1) of the Schwarz lemma implies that both [)
t
(0)[ _ 1 and

_
)
1
_
t
(0)

_
1, and it follows from (1.2) that we must have both [)
t
(0)[ = 1 and

_
)
1
_
t
(0)

= 1.
Part (2) of the Schwarz lemma now shows that ) is a rotation.
Another class of automorphisms are given by the involutions, which are special
biholomorphic maps that interchange 0 with a point n D. For n D dene
,
u
(.) =
n .
1 n.
, . C
_
n
1
_
.
Then ,
u
H
_
C
_
n
1
__
and satises
,
u
(0) = n and ,
u
(n) = 0,
,
t
u
(0) = 1 [n[
2
and ,
t
u
(n) =
1
1 [n[
2
,
and
(1.3) ,
u
,
u
(.) = . for . C
_
n
1
_
,
1. AUTOMORPHISMS OF THE DISK 35
i.e. ,
u
is its own inverse! Indeed, since [n[ < 1,
,
u
,
u
(.) =
n ,
u
(.)
1 n,
u
(.)
=
n
u:
1u:
1 n
u:
1u:
=
(1 n.) n (n .)
(1 n.) n(n .)
=
.
_
1 [n[
2
_
_
1 [n[
2
_ = ..
Claim 1. The functions ,
u
satisfy the following bijections:
,
u
: C
_
n
1
_
C
_
n
1
_
is a bijection,
,
u
: T T is a bijection,
,
u
: D D is a bijection,
,
u
: D D is a bijection.
Proof : The rst bijection follows from (1.3). If [.[ = 1, then .
1
= . since
.. = .. = [.[
2
= 1. If [n[ < 1, then 1 n. ,= 0 and we have

n .
1 n.

.
n. 1
1 n.

=
[n. 1[
[n. 1[
= 1
since the numbers n. 1 and n. 1 are complex conjugates, and hence have
the same modulus. Thus the holomorphic function ,
u
maps the unit circle T into
itself. In fact, ,
u
maps T onto T since ,
u
is its own inverse. Now ,
u
(0) = n
lies in the open unit disk D, and since ,
u
(D) is a connected subset of C T,
we have that ,
u
(D) is contained in the component of C T containing n, i.e.
,
u
(D) D (alternatively, we could appeal to the maximum principle to conclude
that ,
u
(D) D). Thus ,
u
and (,
u
)
1
= ,
u
each map D into D, and it follows
that ,
u
: D D is a bijection. Since we also showed ,
u
: T T is a bijection, we
have that ,
u
: D D is a bijection as well.
The group generated by rotations and involutions is the entire automorphism
group. Moreover, we have the following two unique representations.
Proposition 7. If , is an autormorphism of the disk such that ,(0) = a and
,(/) = 0, then there are rotations l
0
and l

such that
, = ,
o
l
0
= l

,
b
.
Proof : The automorphisms ,
o
, and , ,
b
each x 0, and so are rotations
l
0
and l

. But then ,
o
, = l
0
implies
, = ,
o
,
o
, = ,
o
l
0
,
and , ,
b
= l

implies
, = , ,
b
,
b
= l

,
b
.
36 4. THE RIEMANN MAPPING THEOREM
2. Simply connected subsets of the plane
We say that two closed curves
0
: T \ and
1
: T \ in an open set \ of
the complex plane are \-homotopic if there is a continuous map
I : T [0, 1[ C
such that
0
() = I(, 0) and
1
() = I(, 1) for T.
An open subset \ of the plane C is simply connected if every closed curve in
\ is \-homotopic to a constant curve in \.
To prove the theorems in this chapter we will use Proposition 8, the Jordan
Curve Theorem for taxicab paths, along with a standard homotopy result, Propo-
sition 9. See the appendix for a statement and proof of the general form of the
Jordan Curve Theorem. A taxicab path is a nite concatenation of line segments
that are each parallel to either the real or imaginary axis.
Proposition 8. A simple closed taxicab path divides the plane into two
connected components, one of which is bounded and in which 1:d
~
takes either the
value 1 there, or the value 1 there. More precisely, C
+
= | ' E where both
| and E are connected open subsets of C with | unbounded and E bounded, and
1:d
~
(a) is the constant 1 for all a E.
We borrow an idea from [6] to prove Proposition 8 - see after Step 2 in the
proof of Proposition 17 in the appendix. Here is the homotopy result.
Proposition 9. Let
0
and
1
be two closed paths in an open set \ of the
complex plane. If
0
and
1
are \-homotopic, then 1:d
~
0
(a) = 1:d
~
1
(a) for all
a C \.
The proof of Proposition 9 can also be found in the appendix.
We now use these two propositions to extend Cauchys theorem to closed paths
in simply connected domains \. We use the word "domain" to denote an open
connected subset of the plane, so as to avoid cumbersome expressions like "a simply
connected connected open set". Here we will use an idea in Part 6C of Chapter 1
of Boas [1].
Theorem 16. Let \ be a simply connected domain in the complex plane and
suppose that is a closed path in \. Then for ) H (\) we have
_
~
) (.) d. = 0.
Proof : By the fundamental theorem of line integrals, it suces to construct a
holomorphic antiderivative 1 of ) in \. For this we pick a reference point .
0
\
and dene
1 (.) =
_
c
) (.) d.,
where o is any simple taxicab path in \ joining .
0
to ..
We claim that the set
1 = . \ : there is a simple taxicab path joining .
0
to .
is both open and closed in \, and since \ is connected, we then have 1 = \.
Indeed, if we remove the nal segment from a simple taxicab path joining .
0
to ., and call the resulting path ,, then there is a disk 1(., r) \ ,
+
, and
2. SIMPLY CONNECTED SUBSETS OF THE PLANE 37
clearly we have 1(., r) 1. Conversely, suppose that n \ is a limit point of 1.
Choose c 0 such that 1(n, c) \ and then choose . 1 1(n, c). Let be a
simple taxicab path joining .
0
to .. If n
+
, we clearly have n 1. Otherwise,
r = di:t (
+
, n) 0 and there is a rst point .
+
such that [. n[ = r. Now it
is clear that n 1.
We claim that if o and t are each simple taxicab paths in \ joining .
0
to .,
then
(2.1)
_
c
) (.) d. =
_
r
) (.) d..
Indeed, if we denote by j = o t the (not necessarily simple) taxicab path that
follows o from .
0
to . and then continues on by following t backwards from . to
.
0
, we can write
(2.2)
_

) (.) d. =

I
_

1
) (.) d.
where each j
I
is a simple closed taxicab path in \, and the sum is nite.
Indeed, we perform the following algorithm while proceeding in the forward
direction along o. If there is an initial taxicab segment along which o and t
proceed in opposite directions, let .
1
be the rst point at which the paths diverge,
and discard the segment traversed. Otherwise, proceed along o from .
0
until the
rst time o
+
encounters a point .
1
from t
+
. Then let j
1
be the closed taxicab path
that follows o from .
0
to .
1
and continues on by following t backwards from .
1
to .
0
.
Clearly j
1
is simple. Now set aside j
1
and apply the same procedure starting at .
1
.
This produces either a taxicab segment along which o and t proceed in opposite
directions, or a simple closed taxicab path j
2
. Continue this algorithm until we
reach .. Of course we can ignore those discarded taxicab segments along which
o and t travel in opposite directions since the line integrals along these segments
cancel each other. Collecting all the simple closed taxicab paths j
I
that were set
aside results in (2.2).
Now j
I
is a simple closed taxicab path, and so by Propositions 8 and 9, the
"inside" E
I
of j
I
(the bounded component E
I
of C j
+
I
= |
I
'E
I
) is contained in \
since \ is connected and simply connected. Indeed, if there is a point a E
I
\ then
1:d

1
(a) = 1 by Proposition 8. On the other hand, since \ is simply connected,
j
I
is \-homotopic to a constant map, and so 1:d

1
(a) = 1:d
cons|on|
(a) = 0 by
Proposition 9, the desired contradiction. Since j
I
is a taxicab path, we can write
_

1
) (.) d. =

_
J1
1

) (.) d.
where 1
I

is a rectangle contained inside j

I
(hence in \), 01
I

has the same orien-

tation as j
I
, and the sum is nite for each i. Indeed, simply construct a grid of
innite lines in the plane, each passing through one of the segments in j
I
. This
creates a collection of minimal rectangles with sides that are segments of these lines.
Then the inside of j
I
is the union of all the minimal rectangles 1
I

that happen to
lie inside j
I
. Finally we know that
_
J1
1

) (.) d. = 0 by Cauchys theorem for a

rectangle in a convex set, and summing over i and , proves (2.1).
It remains to prove that 1
t
(.) exists for each . \. So x . and a simple
taxicab path joining .
0
to .. Let , denote the path obtained from by deleting
the nal segment, and let r 0 be such that 1(., r) \,
+
. For [/[ < r, let
|
be
38 4. THE RIEMANN MAPPING THEOREM
the taxicab path consisting of followed by a (possibly empty) horizontal segment
and then a (possibly empty) vertical segment ending at . / (if this procedure
results in doubling back along the nal segment of , simply remove the cancelled
portion). Using Cauchys theorem for a triangle in a convex set, we see that for /
small we have
1 (. /) 1 (.) =
_
~
!
) (n) dn
_
~
) (n) dn =
_
[:,:+|]
) (n) dn,
and so
lim
|0
1 (. /) 1 (.)
/
= lim
|0
1
/
_
[:,:+|]
) (n) dn = ) (.) .
The above proof yields the following technical consequence which will nd ap-
plication in proving the Jordan Curve Theorem below.
Porism 2: Suppose that \ is a connected open subset of the complex plane
satisfying
_
~
) (.) d. = 0, ) H (\) ,
for every simple closed taxicab path in \. Then the argument in the
proof of Theorem 16 shows that every ) H (\) has an antiderivative 1
in H (\).
The following corollary will be indispensible in our proof of the Riemann Map-
ping Theorem. It shows that nonvanishing holomorphic functions have logarithms
in simply connected domains.
Corollary 8. Suppose that \ is a simply connected domain in C. If ) H (\)
is nonvanishing in \, then there is q H (\) satisfying
) (.) = c
(:)
, . \.
Proof : Let /(.) =
}
0
(:)
}(:)
for . \. Then there is q H (\) such that
q
t
(.) = /(.) for . \. The function ) (.) c
(:)
has derivative
)
t
(.) c
(:)
) (.) c
(:)
q
t
(.) = 0,
hence is a nonzero constant c in \, so that ) (.) = cc
(:)
. We can incorporate the
nonzero constant c into q to obtain the corollary.
3. Proof of the Riemann Mapping Theorem
As motivation for the proof of the Riemann Mapping Theorem, we consider
the following extremal problem.
Problem 2. Given c, , D, calculate the maximum value
' (c, ,) = sup[)
t
(c)[ : ) : D D is holomorphic and , = ) (c) ,
and determine the extremal functions.
We use involutions to move the extremal problem to the origin. Let
q = ,
o
) ,
o
: D D.
3. PROOF OF THE RIEMANN MAPPING THEOREM 39
Then q (0) = 0 and so the Schwarz lemma and the chain rule yield
1 _ [q
t
(0)[ =

,
t
o
(,) )
t
(c) ,
t
o
(0)

=
1
1 [,[
2
[)
t
(c)[
_
1 [c[
2
_
.
Thus we have
[)
t
(c)[ _
1 [,[
2
1 [c[
2
= ' (c, ,) ,
where by assertion (2) of the Schwarz lemma, equality is achieved precisely when
,
o
) ,
o
is a rotation 1
0
, i.e.
) = ,
o
1
0
,
o
, 0 [0, 2) .
Note: It is useful to note that for xed c D, the solution to Problem 2
shows that the largest value of [)
t
(c)[ for ) : D D holomorphic occurs
when ) (c) = , = 0.
Remark 4. It is remarkable that the extremal solutions to Problem 2 are in
fact autormorphisms of the disk - rational functions taking the disk one-to-one
onto itself. We will obtain a Riemann mapping for a proper simply connected
domain \ in the complex plane by solving the analogous extremal problem for certain
holomorphic maps from \ to D. However, the extremals for this problem will not
be evident, and we will use Montels theorem on a certain normal family to obtain
the existence of an extremal, which turns out to be (the inverse of ) a Riemann map
for \.
We present the proof of the Riemann Mapping Theorem in three steps. Suppose
that \ is a proper simply connected domain in C. The only consequence of \ being
simply connected that we use is this: every nonvanishing holomorphic function
) H (\) has a holomorphic square root / H (\). This follows immediately
from Corollary 8 if we set /(.) = c
(z)
2
, . \. For future reference we record this
observation.
Porism 3: A subset \ of the complex plane is the image ) (D) of a one-to-
one onto holomorphic map ) : D \ if and only if
(1) \ is open,
(2) \ is connected and every nonvanishing holomorphic function ) H (\)
has a holomorphic square root / H (\),
(3) \ ,= C.
Let
= ) H (\) : ) is 1 1 and into D .
Our task is to show that there is ) that is onto the unit disk D.
Step 1: ,= c
Here is where we use that is a proper subset of C. Pick a C . Then
) (.) =
1
:o
is holomorphic and one-to-one in \, but not necessarily bounded in
\. To x this, note that ) is nonvanishing in \, so has a square root / H (\):
/(.)
2
= ) (.) for . \. By the open mapping theorem, there is a disk 1(n, r)
/(\) R. But then
(3.1) 1(n, r) /(\) = c.
40 4. THE RIEMANN MAPPING THEOREM
To see this we argue by contradiction. If there is n
1
1(n, r) /(\), then n
2
=
n
1
1(n, r) /(\) and n
2
,= n
1
. Thus there are distinct points .
1
and .
2
in
\ such that
/(.
1
) = n
1
and /(.
2
) = n
2
.
But then
1
.
1
a
= ) (.
1
) = /(.
1
)
2
= n
2
1
= n
2
2
= /(.
2
)
2
= ) (.
2
) =
1
.
2
a
implies that .
1
= .
2
, a contradiction.
Now we take q (.) =
1
|(:)+u
for . \. Clearly (3.1) shows that q H (\) is
bounded by
1
:
. We also have that q is one-to-one since
q (.) = q (.
t
) ==/(.) = /(.
t
) ==) (.) = ) (.
t
) ==. = .
t
,
since ) is one-to-one. Thus q .
Step 2: Fix .
0
\ and ) . If ) is not onto the unit disk D, then there
exists / such that
[/
t
(.
0
)[ [)
t
(.
0
)[ .
Suppose ) and that ) is not onto the disk, say c D ) (\). Then
,
o
) and 0 , ,
o
) (\). Thus ,
o
) has a holomorphic square root q :
q (.)
2
= ,
o
) (.) , . \.
We now compose q with an involution ,
o
chosen so that ,
o
q (.
0
) = 0, since we
expect that this choice will maximize

_
,
o
q
_
t
(.
0
)

. This choice requires that we

take , = q (.
0
), and we can now compute that the function
/ = ,
o
q
satises
,
o
o ,
o
/ = ),
where the function o is dened by o (.) = .
2
.
Now o fails to be one-to-one on D, and hence c = ,
o
o ,
o
fails to be
one-to-one on D as well. The Note following Problem 2 thus implies that

c
t
(0)

< 1.
Since /(.
0
) = 0, the chain rule now gives
[)
t
(.
0
)[ =

c
t
(0) /
t
(.
0
)

< [/
t
(.
0
)[ .
Step 3: Fix .
0
\ and let ' = sup
}
[)
t
(.
0
)[. Then ' 0 and there is
an extremal function / satisfying [/
t
(.
0
)[ = '.
Step 1 and Proposition 6 shows that ' 0. Since [) (.)[ < 1 for all )
and . \, Montels Theorem 13 shows that is a normal family. Thus there is a
sequence )
n

o
n=1
that converges uniformly on compact subsets of \ such that
lim
no
Io )
t
n
(.
0
) = lim
no
[)
t
n
(.
0
)[ = '.
Let / = lim
no
)
n
. By the Uniform Convergence Theorem 11, / H (\) and
Io /
t
(.
0
) = lim
no
Io )
t
n
(.
0
) = '. Since [/(.)[ = lim
no
[)
n
(.)[ _ 1, we have
/(\) D, and the open mapping theorem now shows that /(\) D (/ is not
constant since ' 0). Thus the only thing remaining to verify in order to conclude
that / is that / is one-to-one in \.
3. PROOF OF THE RIEMANN MAPPING THEOREM 41
To prove that / is one-to-one in \, x .
1
, .
2
\ with .
1
,= .
2
. Since / is not
constant (it has nonzero derivative at .
0
since ' 0), the coincidence principle
implies that there is r 0 such that .
1
, 1(.
2
, r) and also
inf
:J1(:2,:)
[/(.) /(.
1
)[ 0.
Now each )
n
is one-to-one on \ and it follows that )
n
(.) )
n
(.
1
) is nonvanishing
in a neighourhood of 1(.
2
, r). Thus uniform convergence of )
n
to / on 1(.
2
, r),
together with the minimum principle, yields
0 < inf
:J1(:2,:)
[/(.) /(.
1
)[ = lim
no
inf
:J1(:2,:)
[)
n
(.) )
n
(.
1
)[
_ lim
no
[)
n
(.
2
) )
n
(.
1
)[ = [/(.
2
) /(.
1
)[ .
Thus /(.
2
) ,= /(.
1
) and this completes the proof that / .
It is now easy to complete the proof of the Riemann Mapping Theorem. Indeed,
by Step 2 the extremal function / in Step 3 must be onto the unit disk D, and so
the inverse ) = /
1
: D \ of / is a Riemann map for \.
Note 1: The Riemann map ) = /
1
constructed in the proof above satises
) (0) = .
0
. Indeed, if /(.
0
) = , ,= 0, then

_
,
o
/
_
t
(.
0
)

,
t
o
(,) /
t
(.
0
)

=
[/
t
(.
0
)[
1 [,[
2
[/
t
(.
0
)[ ,
contradicting Step 3. All other Riemann maps q for \ have the form
q = ) , for some automorphism of the unit disk.
Note 2: Suppose ) (r ij) = n i maps D one-to-one and onto \, and
is continuously (real) dierentiable, but not necessarily holomorphic. If
\) = ()
r
, )

) and J) = ool
_
n
r
n

_
, then a calculation yields
1
2
[\)[
2
=

0)
0.

0)
0.

2
,
J) =

0)
0.

0)
0.

2
,
and if J) _ 0 we conclude that the Dirichlet (or energy) integral
_
D
[\)[
2
2
drdj =
_
D
J) drdj 2
_
D

0)
0.

2
drdj
= arca (\) 2
_
D

0)
0.

2
drdj
achieves its minimum value arca (\) if and only if ) is a Riemann map.
CHAPTER 5
Contour integrals and the Prime Number
Theorem
We will now develop the theory of contour integrals and the residue theorem,
and apply these results to prove the Prime Number Theorem: if
(r) equals the number of positive primes _ r,
then
(0.2) lim
ro
(r)
r
ln r
= 1.
We will follow for the most part the treatment in Chapter 7 of Stein and Shakarchi
[6].
The basic connection between complex analysis and prime numbers is the fol-
lowing beautiful identity of Euler:
(0.3)
o

n=1
1
:
s
=

1
1
1
1

s
, Io : 1,
where T = 2, 8, , 7, 11, 18, ... is the set of prime positive integers j; j is prime if it
has no positive integer factors other than 1 and itself (we exclude the multiplicative
identity 1 from T). We will often adopt the convention, common in number theory,
of simply writing

or

to denote the innite product or sum over all prime

numbers T. It is also customary in number theory to write a complex variable as
: = o it. Finally, we point out that for a 0 and : C, we dene a
s
= c
s ln o
.
At the core of the proof of Eulers identity (0.3) is the Fundamental Theorem
of Arithmetic: every positive integer : 1 has a unique factorization into a nite
product of primes,
(0.4) : = j
|1
1
j
|2
2
...j
|r
n
,
where /
I
_ 1 for 1 _ i _ :. The uniqueness refers to the positive integer :
and the numbers j
I
T and their associated powers /
I
for 1 _ i _ :. If we
insist that the primes j
I
are taken in increasing order, then the entire factorization
(0.4) is uniquely determined by :. The proof of these assertions uses the Euclidean
algorithm for division of positive integers and will not be repeated here.
Here is a formal argument that (0.3) holds:
o

n=1
1
:
s
=
o

n=1

!
1
1

!
2
2
...
!r
r
=n
1
_
j
|1
1
j
|2
2
...j
|r
n
_
s
=
o

n=1

]1,...r]1

]|1,...|r]N
n

I=1
1
j
|1s
I
43
44 5. CONTOUR INTEGRALS AND THE PRIME NUMBER THEOREM
where the second equality holds by the Fundamental Theorem of Arithmetic, and
where the notation j
1
, ...j
n
implies that the j
I
are distinct. Continuing, we
rearrange the terms in the innite sum into an innite product to obtain
o

n=1
1
:
s
=

1
_

|N
1
j
|s
_
=

1
1
1
1

s
,
upon summing the geometric series

|N
1

!s
=
1
1
1

s
.
We can make this argument rigorous when : 1 as follows: whenever 1 < <
' the Fundamental Theorem of Arithmetic shows that

n=1
1
:
s
_

_
1
1
j
s
...
1
j
s1
_
_

1
1
1

s
,
and conversely,

_
1
1
j
s
...
1
j
s1
_
_
o

n=1
1
:
s
.
Taking limits appropriately in ' and gives (0.3) for : 1.
Now we extend Eulers identity (0.3) to all complex numbers : in the half-plane
\
1
= : C : Io : 1. For this we observe that the series on the left side of (0.3)
is absolutely convergent for : \
1
, and converges uniformly on compact subsets
of \
1
. Hence by the Uniform Convergence Theorem 11, the series

o
n=1
1
n
s
denes
a holomorphic function of : in the half-plane \
1
. Moreover, we claim that the
innite product

1
1
1

s
on the right side of (0.3) is also uniformly convergent in
: on compact subsets of \
1
, and hence denes a holomorphic function of : \
1
.
This requires just a small amount of additional work involving innite products, to
which we now turn.
If 0 _ n
n
< 1 and 0 _
n
< then
o

n=1
(1 n
n
) 0 if and only if
o

n=1
n
n
< , (0.5)
o

n=1
(1
n
) < if and only if
o

n=1

n
< .
To see (0.5) we may assume 0 _ n
n
,
n
_
1
2
, so that c
ur
_ 1 n
n
_ c
2ur
and
c
1
2
ur
_ 1
n
_ c
ur
. For example, when 0 _ r _
1
2
, the alternating series estimate
yields
c
2r
_ 1 2r
(2r)
2
2!
_ 1 r,
while the geometric series estimate yields
c
1
2
r
_ 1
_
1
2
r
_
_
1 r r
2
...
_
_ 1 r.
5. CONTOUR INTEGRALS AND THE PRIME NUMBER THEOREM 45
Thus we have
oxp
_

n=1
n
n
_
_
o

n=1
(1 n
n
) _ oxp
_
2
o

n=1
n
n
_
, (0.6)
oxp
_
1
2
o

n=1

n
_
_
o

n=1
(1
n
) _ oxp
_
o

n=1

n
_
.
Now suppose that n
n
H (\) and that

o
n=1
n
n
(.) converges absolutely and
uniformly on compact subsets of \.
Claim 2. The innite product
o

n=1
(1 n
n
(.)) converges uniformly on compact
subsets of \ to a holomorphic function ) (.) in \. Furthermore, ) (.) = 0 if and
only if there is : _ 1 such that n
n
(.) = 0.
Indeed, if we expand products, cancel the 1s and take absolute values inside,
we see that
(0.7) [j
1,
(.) 1[ _ j
+
1,
(.) 1, 1 _ ' _ ,
where j
1,
(.) =

n=1
(1 n
n
(.)) and j
+
1,
(.) =

n=1
(1 [n
n
(.)[). Alterna-
tively, the case = ' is obvious and the general case follows by induction from
1
1,+1
1 = 1
1,
(1 n
+1
(.)) 1 = (1
1,
1) (1 n
+1
(.)) n
+1
(.) ,
since then
[1
1,+1
1[ _
_
1
+
1,
1
_
(1 [n
+1
(.)[) [n
+1
(.)[ = 1
+
1,+1
1.
It now follows from (0.7), the second line in (0.6), and the Uniform Conver-
gence Theorem 11, that the innite product
o

n=1
(1 n
n
(.)) converges uniformly
on compact subsets of \ to a holomophic function ) H (\):

n=1
(1 n
n
(.))
1

n=1
(1 n
n
(.))

_
11

n=1
(1 n
n
(.))
_
1
1,+1
1

_
11

n=1
(1 [n
n
(.)[) [1
1,+1
1[
_ c
P
1
r=1
]ur(:)]
_
j
+
1,
(.) 1
_
0
as ' uniformly on compact subsets of \.
From the inequality

n=1
(1 n
n
(.))

_
o

n=1
(1 [n
n
(.)[) ,
and the rst line in (0.5), we see that ) (.) vanishes at . if and only if one of the
functions 1 n
n
(.) vanishes at .. Indeed, if n
n
(.) ,= 1 for all :, then we can
discard the nitely many : for which [n
n
(.)[ _ 1 so that (0.5) applies.
46 5. CONTOUR INTEGRALS AND THE PRIME NUMBER THEOREM
If we apply the above Claim to the holomorphic functions
n
n
(:) = 1
1
1
1

s
r
=
1
j
s
n
1
, : \
1
,
where T = j
n

o
n=1
, we see that the innite product on the right side of (0.3)
converges uniformly on compact subsets of \
1
to a holomorphic function there. By
the coincidence principle, the two sides of (0.3) coincide in \
1
, so Eulers identity
holds for all : \
1
.
Definition 8. The Riemann zeta function is the holomorphic function
(:) =
o

n=1
1
:
s
, : \
1
.
Note that letting : 1 in (0.3) shows that
=
o

n=1
1
:
=

1
1
1

,
so that

_
1
1

_
= 0. Then (0.5) yields

= , which quanties the inni-

tude of primes, and begs the question of their density in the positive integers.
In the next section we will extend (:) to a holomorphic function in C 1,
and show that the extension has a simple pole at 1. Note that an extension of
a holomorphic function ) from a domain \ to a larger domain is unique by the
coincidence principle. Nevertheless, some caution must be exercised as dierent
extensions of ) need not coincide at a common point outside of \. Indeed, for any
odd integer / there is an extension of log . = ln[.[ i0 from the ball 1(1, 1) to a
domain (that winds around the origin / times) containing 1 in which log (1) =
/i. As a result, these extensions are usually referred to as an analytic continuation
of ) to a larger set.
1. Analytic continuation of (:)
We begin by comparing the series

o
n=1
1
n
s
for (:) to its corresponding integral
_
o
1
Jr
r
s
. Now the integral has a holomorphic continuation to C 1 given by
1
s1
since for Io : 1,
_
o
1
dr
r
s
=
1
: 1
.
We will show that the dierence between the series and the integral admits a holo-
morphic continuation to the larger half-plane \
0
= : C : Io : 0. This will
then give a holomorphic continuation of (:) to \
0
1 with a simple pole at 1.
Lemma 2. There is a sequence of entire functions /
n
(:)
o
n=1
which decrease
to zero at the rate
(1.1) [/
n
(:)[ _
[:[
:
c+1
, : = o it C,
and satisfy the equation
(1.2)
1

n=1
1
:
s

_

1
dr
r
s
=
1

n=1
/
n
(:) , : C,
1. ANALYTIC CONTINUATION OF (s) 47
whenever 1 is an integer. In particular, the rate of decay in (1.1) shows that
the right side of the equation
(1.3) (:) =
1
: 1

o

n=1
/
n
(:) , : \
1
,
denes a holomorphic function for : \
0
1 since the series on the right converges
uniformly on compact subsets of \
0
.
Proof : Dene
/
n
(:) =
1
:
s

_
n+1
n
dr
r
s
=
_
n+1
n
_
1
:
s

1
r
s
_
dr.
By the submean value theorem applied on the interval [:, r[, we have

1
:
s

1
r
s

:
c
s+1
n

_
[:[
:
c+1
,
for some c
n
(:, r). Estimate (1.1) and equation (1.2) now follow.
For : \
1
, the integral
_

1
Jr
r
s
converges to
1
s1
so that (1.3) holds. The
estimate (1.1) implies that the series

o
n=1
/
n
(:) on the right side of (1.3) converges
uniformly on compact subsets of \
0
.
We can continue this process and obtain an analytic continuation of (:) to
\
1
1, then to \
2
1, etc. until we have extended (:) to all of C 1.
We emphasize that for the proof of the Prime Number Theorem we only need the
analytic continuation of (:) to \
0
1 given in Lemma 2. However, for the sake
of completeness, we sketch the details of the continuation to all of C 1.
In order to continue (:) to \
1
1, we dene r = r: for : _ r < :1
to be the fractional part of r, and begin by computing for Io : 1,
:
_
o
1
r
r
s+1
dr = :
o

n=1
_
n+1
n
r :
r
s+1
dr = :
o

n=1
__
n+1
n
r
s
dr :
_
n+1
n
r
s1
dr
_
=
o

n=1
_
:
1 :
_
(: 1)
1s
:
1s
_
:
_
(: 1)
s
:
s
_
_
=
:
: 1
_
2
s
1
s

2
_
8
s
2
s

8
_
4
s
8
s

...
=
:
: 1
(:) .
Now each of the functions (:) and
s
s1
:
_
o
1
]r]
r
s+1
dr is holomorphic in \
0
1
and so the coincidence principle implies
(:) =
:
: 1
:
_
o
1
r
r
s+1
dr, : \
0
1 .
With Q(r) = r
1
2
the above identity becomes
(:) =
:
: 1

1
2
:
_
o
1
Q(r)
r
s+1
dr, : \
0
1 .
Note that Q(r) is periodic on the real line with period 1 and
_
1
0
Q(r) dr = 0. Thus
any antiderivative of Q(r) on the real line will also have period 1.
48 5. CONTOUR INTEGRALS AND THE PRIME NUMBER THEOREM
Now we set Q
0
(r) = Q(r) and dene Q
1
(r) on the real line by
Q
t
1
(r) = Q
0
(r) and
_
1
0
Q
1
(r) dr = 0.
Thus Q
1
is periodic on the real line with period 1, Q
1
(r) =
1
2
r
2

1
2
r
1
12
for
0 _ r < 1, Q
1
is continuous on [1, ), and Q
1
is continuously dierentiable on
[1, ) N. We can now integrate by parts to obtain for : \
0
1,
_
o
1
Q
t
1
(r)
r
s+1
dr =
Q
1
(r)
r
s+1
[
o
1
(: 1)
_
o
1
Q
1
(r)
r
s+2
dr
=
1
12
(: 1)
_
o
1
Q
1
(r)
r
s+2
dr.
Thus for : \
0
1 we have
(:) =
:
: 1

1
2
:
_

1
12
(: 1)
_
o
1
Q
1
(r)
r
s+2
dr
_
.
The right side denes a holomorphic function of : in \
1
1 and provides an
analytic continuation of (:) to \
1
1.
Now we recursively dene Q
|
(r) on the real line for all / _ 1 by
Q
t
|
(r) = Q
|1
(r) and
_
1
0
Q
|
(r) dr = 0.
Then by induction Q
|
is periodic on the real line with period 1, Q
|
is / 1 times
continuously dierentiable on [1, ), and Q
|
is / times continuously dierentiable
on [1, ) N. Thus we can integrate by parts / times in
(:) =
:
: 1

1
2
:
_
o
1
J
!
Jr
!
Q
|
(r)
r
s+1
dr, : \
0
1 ,
(the case / = 1 was done above), and the resulting right hand side denes an
analytic continuation of (:) to \
|
1. Indeed, the boundary terms lead to a
polynomial of degree / in :, while the integral remaining is
: (: 1) ... (: /)
_
o
1
Q
|
(r)
r
s+|+1
dr.
1.1. Nonvanishing of the zeta function. The key to the proof we give of
the Prime Number Theorem is the nonvanishing of the zeta function (:) on the
line Io : = 1. For this purpose it will be convenient to dene log (:) in the simply
connected domain \
1
. This is possible by Corollary 8 since Eulers identity (0.3)
and Claim 2 show that (:) is nonvanishing in \
1
. Indeed, Claim 2 applies since
1 n

(.) =
1
1
1

s
implies n

(.) =
1
1
s
where

1
1
s
converges absolutely and
uniformly on compact subsets of \
1
. Among the possible choices for the function
log (:) we x the one satisfying
log (:) = ln (:) 0, : 1.
Theorem 17. The zeta function (:) has no zeroes on the line Io : = 1.
Proof : Using the power series expansion for logarithm on the unit interval,
ln
_
1
1 r
_
=
o

n=1
r
n
:
, 0 _ r < 1,
1. ANALYTIC CONTINUATION OF (s) 49
together with Eulers formula (0.3), we have when : 1:
ln (:) = ln

1
1
1

s
=

ln
_
1
1
1

s
_
=

n=1
j
sn
:
,
where the double series is absolutely convergent. In fact the double series is ab-
solutely convergent for Io : 1, and hence denes a holomorphic function in \
1
.
The coincidence principle now gives the following useful formula (which we will
have occasion to use later),
(1.4) log (:) =

,n
j
sn
:
, : \
1
.
With c
n
=
_
1
n
if : = j
n
0 if not
, we can rewrite this formula as
log (:) =
o

n=1
c
n
1
:
s
, : \
1
.
We now invoke the identity
(1.5) 8 4 cos 0 cos 20 = 2 (1 cos 0)
2
_ 0,
to conclude that for : = o it \
1
, i.e. o 1 and t R,
(1.6) ln

(o)
3
(o it)
4
(o i2t)

_ 0.
Indeed, from
Io
_
1
:
s
_
= Io
_
c
(c+Ir) ln n
_
=
1
:
c
cos (t ln:) ,
we obtain
ln

(o)
3
(o it)
4
(o i2t)

= 8 ln[ (o)[ 4 ln[ (o it)[ ln[ (o i2t)[

= 8 Io log (o) 4 Io log (o it) Io log (o i2t)
=
o

n=1
c
n
1
:
c
8 4 cos (t ln:) cos (2t ln:) ,
which is nonnegative by (1.5) and c
n
_ 0.
We can now use (1.6) to derive a contradiction from the assumption that
(1 it) = 0 for some t R. Indeed, since has a simple pole at 1, we have

(o)
3

_ C (o 1)
3
, o (1, 2) .
Also, t ,= 0 and must have a zero of order at least one at 1 it, so that

(o it)
4

_ C (o 1)
4
, o (1, 2) .
Finally, since is holomorphic in a neighbourhood of the segment o it : 1 _ o _ 2,
we have
[ (o i2t)[ _ C, o (1, 2) .
Thus altogether we obtain

(o)
3
(o it)
4
(o i2t)

_ C (o 1) , o (1, 2) ,
50 5. CONTOUR INTEGRALS AND THE PRIME NUMBER THEOREM
which contradicts (1.6) when C (o 1) < 1.
2. The residue theorem
Definition 9. A function ) is said to be meromorphic in an open set \ if
there is a subset of \ such that
(1) has no limit point in \,
(2) ) is holomorphic in \ ,
(3) ) has a pole at each point of .
Recall that if ) has a pole of order N at a, then there is 1 0 so that for
. 1(a, 1),
) (.) =
o

n=
/
n
(. a)
n
(2.1)
=
_
/

(. a)

...
/
1
. a
_

n=0
/
n
(. a)
n
= 1

(.) /(.) ,
where / H (1(a, 1)), For 0 < r < 1 we compute using Cauchys theorem on /,
_
J1(o,:)
) (.) d. =
_
J1(o,:)
1

(.) d.
_
J1(o,1)
/(.) d.
=
_
J1(o,:)
/

(. a)

d. ...
_
J1(o,:)
/
1
. a
d.
=
_
J1(o,:)
/
1
. a
d. = 2i/
1
,
where we have used the fact that the function
1
(:o)
!
has antiderivative
1
1|
1
(:o)
!1
in the punctured disk 1
t
(a, 1) to conclude that
_
J1(o,:)
1
(:o)
!
d. = 0 for 2 _
/ _ . We suspect that this formula will persist with 01(a, r) replaced by any
appropriate simple path that contains a in its interior. This justies the following
denition.
Definition 10. If ) has a pole at a with principal part
1

(.) =
/

(. a)

...
/
1
. a
,
we dene the residue of ) at a to be
1c: (); a) = /
1
.
Theorem 18. Suppose that ) is meromorphic in a simply connected domain
\, and that the set of poles of ) in \ is nite. Then if is a closed path in
\ , we have
(2.2)
1
2i
_
~
) (.) d. =

o.
1c: (); a) 1:d
~
(a) .
2. THE RESIDUE THEOREM 51
Proof : Let 1
o
(.) be the principal part of ) at the pole a, and set
q (.) = ) (.)

o.
1
o
(.) .
Then q has a removable singularity at each point a , and so q H (\). Since
1
2i
_
~
1
o
(.) d. =
1
2i
_
~
1c: (); a)
. a
d. = 1c: (); a) 1:d
~
(a) ,
we have
0 =
1
2i
_
~
q (.) d. =
1
2i
_
~
) (.) d.

o.
1
2i
_
~
1
o
(.) d.
=
1
2i
_
~
) (.) d.

o.
1c: (); a) 1:d
~
(a) .
The identity (2.2) in the residue theorem can be extended to arbitrary mero-
morphic functions ) in an open set \ (it turns out that the index 1:d
~
(a) is nonzero
for only nitely many poles), but we will not pursue this here as it is not needed
in the sequel.
2.1. Calculation of residues. If ) has a simple pole at a, there is a useful
formula for computing the residue that is immediate from an inspection of the case
= 1 of (2.1):
(2.3) 1c: (); a) = lim
:o
(. a) ) (.) .
More generally, there is an analogous formula for the residue at a pole of order ,
(2.4) 1c: (); a) = lim
:o
1
( 1)!
d
1
d.
1
_
(. a)

) (.)
_
,
that follows from multiplying (2.1) by (. a)

and then using the formula from

Theorem 5 to compute the ( 1)
s|
coecient of the holomorphic function q (.) =
(. a)

) (.). Note that if we overestimate the order of the pole ) has at a point
a, formula (2.4) still gives the correct answer, i.e. (2.4) holds provided ) has a pole
of order at most at a. On the other hand, if we underestimate the order of the
pole ) has at a point a, the right side of formula (2.4) is innite, thereby alerting
us to our error.
An observation that is useful in a very special case is this: if ) is meromorphic
in a domain \ that is symmetric about the real axis R, and if ) (.) is real on \R,
then ) has a pole of order at a \ if and only if ) has a pole of order at a,
and moreover the residues are conjugate:
1c: (); a) = 1c: (); a).
This follows from the Schwarz reection principle: the function /(.) = ) (.)) (.)
is holomorphic in \ and vanishes on \R, hence vanishes in \ by the coincidence
principle.
52 5. CONTOUR INTEGRALS AND THE PRIME NUMBER THEOREM
2.2. Counting zeroes and poles. If ) has a zero of order at a then
}
0
(:)
}(:)
has a simple pole at a with
(2.5) 1c:
_
)
t
)
; a
_
= .
Indeed, if ) (.) = (. a)

/(.) with /(a) ,= 0, then for . 1(a, r) with r small,

)
t
(.)
) (.)
=
(. a)
1
/(.) (. a)

/
t
(.)
(. a)

/(.)
=

(. a)

/
t
(.)
/(.)
.
Similarly, if ) has a pole of order at a then
}
0
(:)
}(:)
has a simple pole at a with
(2.6) 1c:
_
)
t
)
; a
_
= .
Indeed, if ) (.) = (. a)

/(.) with /(a) ,= 0, then for . 1(a, r) with r small,

)
t
(.)
) (.)
=
(. a)
1
/(.) (. a)

/
t
(.)
(. a)

/(.)
=

(. a)

/
t
(.)
/(.)
.
Of course, if ) has neither a zero nor a pole at a, then
}
0
}
is holomorphic in a neigh-
bourhood of a. It now follows easily from the residue theorem that
1
2tI
_
~
}
0
(:)
}(:)
d.
counts the number of zeroes minus the number of poles inside . Here is a precise
statement that is not the most general possible.
Theorem 19. Suppose that ) is meromorphic in a simply connected domain
\, and that both the set 1 of poles of ) in \ is nite, and the set 7 of zeroes of )
in \ is nite. Then if is a simple closed path in \ (7 ' 1), we have
1
2i
_
~
)
t
(.)
) (.)
d. =

:2 that are inside ~
order of the zero at . (2.7)

1 that are inside ~
order of the pole at j.
Proof : Let = 7 ' 1. The residue theorem gives
1
2i
_
~
)
t
(.)
) (.)
d. =

o.
1c:
_
)
t
)
; a
_
1:d
~
(a) ,
and then the calculations above give (2.7).
3. Proof of the Prime Number Theorem
Now we can complete the proof of the Prime Number Theorem:
lim
ro
(r)
r
ln r
= 1.
In order to prove that (r) ~
r
ln r
, it turns out to be more convenient to consider
the asymptotic form (r) lnr ~ r. Now
(r) lnr =

r
lnr =

r
lnr
lnj
lnj,
3. PROOF OF THE PRIME NUMBER THEOREM 53
which is very close to Tchebychevs c function:
(3.1) c (r) =

r
_
lnr
lnj
_
lnj =

r
r
lnj =

1nr
A(:) ,
where
(3.2) A(:) =
_
lnj if : = j
n
for some prime j and some : _ 1
0 otherwise
.
The Prime Number Theorem is reduced to the following asymptotic estimate for
c.
Proposition 10. If lim
ro
r(r)
r
= 1, then lim
ro
t(r)
o
ln o
= 1.
Proof : We clearly have
c (r)
r
=

r
_
ln r
ln
_
lnj
r
_

r
ln r
ln
lnj
r
=
(r)
r
ln r
,
which gives
1 _ lim inf
ro
(r)
r
ln r
.
Conversely, x 0 < c < 1. Then
c (r) _

r
c
<r
lnj _ (r) (r
o
) lnr
o
,
and so
c
(r)
r
ln r
_ c
(r
o
) lnr
r

c (r)
r
_ c
r
o
lnr
r

c (r)
r
.
Taking the limit superior as r we obtain
clim sup
ro
(r)
r
ln r
_ clim sup
ro
lnr
r
1o
1 = 1.
Since 0 < c < 1 is arbitrary, we conclude that limsup
ro
t(r)
o
ln o
_ 1, and hence
lim
ro
t(r)
o
ln o
= 1.
We expect that for suitable functions c, c (r) ~ r if and only if
_
r
1
c (t) dt ~
_
r
1
tdt ~
r
2
2
. So we dene
c
1
(r) =
_
r
1
c (t) dt, r 1,
where c is Tchebychevs c function.
Proposition 11. If lim
ro
r
1
(r)
o
2
2
= 1, then lim
ro
r(r)
r
= 1.
Proof : Since c is increasing we have for 0 < c < 1 < , < ,
(3.3)
_
r
or
c (t) dt
_
r
or
dt
_ c (r) _
_
or
r
c (t) dt
_
or
r
dt
.
54 5. CONTOUR INTEGRALS AND THE PRIME NUMBER THEOREM
The second inequality in (3.3) yields
c (r)
r
_
1
r
1
(, 1) r
c
1
(,r) c
1
(r) =
1
(, 1)
_
,
2
c
1
(,r)
(,r)
2

c
1
(r)
r
2
_
,
and taking limit superior as r we obtain
lim sup
ro
c (r)
r
_
1
(, 1)
_
1
2
,
2

1
2
_
=
, 1
2
.
Letting , 1 we get limsup
ro
r(r)
r
_ 1. Arguing in similar fashion with the
rst inequality in (3.3), we get 1 _ liminf
ro
r(r)
r
, and hence lim
ro
r(r)
r
= 1.
The connection between c
1
(r) and (:) is given in the following identity, which
will ultimately yield the asymptotic c
1
(r) ~
r
2
2
.
Proposition 12. For all c 1,
(3.4) c
1
(r) =
1
2i
_
c+Io
cIo
r
s+1
: (: 1)
_

t
(:)
(:)
_
d:,
where
_
c+Io
cIo
denotes integration along the vertical line Io : = c in the upward
direction.
Proof : We rst claim that the integral on the right side of (3.4) is absolutely
convergent. Indeed, if we dierentiate (1.4),
log (:) =

,n
j
sn
:
, : \
1
,
we obtain
(3.5)

t
(:)
(:)
=

,n
(lnj) j
sn
=
o

n=1
A(:)
1
:
s
, Io : 1,
where A(:) is dened in (3.2). In particular this yields

t
(:)
(:)

_
o

n=1
ln:
:
Re s
< , : \
1
.
Since we also have

r
s+1
: (: 1)

_
r
c+1
(1 [t[)
2
, : = c it \
1
,
we see that the modulus of the integrand on the right side of (3.4) is dominated by
C (1 [t[)
2
, hence integrable.
Next we claim that the following identity holds:
(3.6)
1
2i
_
c+Io
cIo
j
s
: (: 1)
d: =
_
1
1
j
_
+
=
_
0 if 0 < j _ 1
1
1

if 1 _ j <
.
Note that the integral on the right side of (3.6) is absolutely convergent since
[j
s
[ = j
c
is constant on the path of integration. Assuming (3.6) we can quickly
3. PROOF OF THE PRIME NUMBER THEOREM 55
complete the proof of Proposition 12. Using (3.1) we have
c
1
(r) =
_
r
1
c (t) dt =
_
r
1
_
_

1n|
A(:)
_
_
dt
=

1nr
A(:) (r :) = r
o

n=1
A(:)
_
1
:
r
_
+
,
and so using (3.6) with j =
r
n
we conclude that
1
2i
_
c+Io
cIo
r
s+1
: (: 1)
_

t
(:)
(:)
_
d: =
1
2i
_
c+Io
cIo
r
s+1
: (: 1)
_
o

n=1
A(:)
1
:
s
_
d:
=
o

n=1
A(:)
1
2i
_
c+Io
cIo
r
_
r
n
_
s
: (: 1)
d:
= r
o

n=1
A(:)
_
1
:
r
_
+
= c
1
(r) .
Finally we prove the identity (3.6) using the residue theorem. Suppose rst
that 1 _ j < . For 1 0, denote by [c i1, c i1[ the line segment joining
c i1 to c i1 that is directed upward. Also denote by C
lt}|
(1) the half circle
of radius 1 that joins c i1 to c i1 by travelling down the left half of the circle
centered at c with radius 1. Now dene
) (:) =
j
s
: (: 1)
=
c
s ln
: (: 1)
, : C.
Note that ) is meromorphic in the plane with simple poles at 0 and 1. If 1 is
chosen large enough that these two poles lie inside the closed path [c i1, c i1[ '
C
lt}|
(1), then the residue theorem gives
1
2i
_
[cI1,c+I1]
) (:) d:
1
2i
_
c
cjt
(1)
) (:) d:
= 1c: (); 0) 1c: (); 1)
=
j
0
(0 1)

j
1
(1)
= 1
1
j
.
Now let 1 to obtain
lim
1o
1
2i
_
c
cjt
(1)
) (:) d: = 0,
and hence
lim
1o
1
2i
_
[cI1,c+I1]
) (:) d: = 1
1
j
.
Indeed, for 1 2 (c 1) and
t
2
_ 0 _
3t
2
,

)
_
c 1c
I0
_

_
c
(c+1cos 0) ln
(1 c) (1 c 1)
_
4c
c ln
1
2
56 5. CONTOUR INTEGRALS AND THE PRIME NUMBER THEOREM
since cos 0 _ 0 and lnj _ 0. Now the length of the path C
lt}|
(1) is 1 and so we
obtain

_
c
cjt
(1)
) (:) d:

_ 1
4c
c ln
1
2
0 as 1 .
To prove (3.6) when 0 < j _ 1, we consider instead the closed path [c i1, c i1['
C
:I||
(1) and apply Cauchys theorem (which is just the residue theorem when
there are no poles inside the path) to the holomorphic function ) (:) in the half-
plane Io : 1. We obtain
1
2i
_
[cI1,c+I1]
) (:) d:
1
2i
_
c
r1!t
(1)
) (:) d: = 0,
and then use the argument above with cos 0 _ 0 and lnj _ 0 to show that
lim
1o
1
2i
_
c
r1!t
(1)
) (:) d: = 0.
This completes the proof of (3.6), and hence that of Proposition 12.
3.1. Proof of the asymptotic for c
1
(r). Here we prove the asymptotic
formula
(3.7) lim
ro
c
1
(r)
1
2
r
2
= 1
in three steps. Propositions 10 and 11 then complete the proof of the Prime Number
Theorem (0.2).
We begin with the identity
(3.8) c
1
(r) =
1
2i
_
c+Io
cIo
r
s+1
: (: 1)
_

t
(:)
(:)
_
d:
in Proposition 12, and try to move the path of integration to [1 i, 1 i[ where

r
s+1

= r
2
is at least comparable to the asymptotic estimate
1
2
r
2
. Of course this
is impossible since (:) has a pole at 1, but we can come close enough using the
residue theorem. Let 1 _ 8 and c 1. For o 1 and 0 < j < 1, dene the closed
taxicab path
1,S,c,q
to consist of the eight segments
[c io, c io[ , [c io, 1 io[ , [1 io, 1 i1[ , [1 i1, 1 j i1[ ,
[1 j i1, 1 j i1[ , [1 j i1, 1 i1[ , [1 i1, 1 io[ , [1 io, c io[
concatenated with the directions and in the sequence given. Since (:) has a pole
at 1 and is nonvanishing on the rest of the line Io : = 1, there exists 0 < j (1) < 1
depending on 1 such that (:) has no zero on or inside the rectangle with sides
[1 j i1, 1 i1[ , [1 i1, 1 i1[ , [1 i1, 1 j i1[ , [1 j i1, 1 j i1[ .
We will always assume j is so small that 0 < j < j (1).
Step 1: For o 1 _ 8 and 0 < j < j (1) < 1 < c we have
(3.9)
1
2i
_
~
T:c_
r
s+1
: (: 1)
_

t
(:)
(:)
_
d: =
r
2
2
.
3. PROOF OF THE PRIME NUMBER THEOREM 57
Since the integrand
(3.10) q
r
(:) =
r
s+1
: (: 1)
_

t
(:)
(:)
_
is meromorphic in \
0
with a single pole at : = 1 inside the simple path
1,S,c,q
,
the residue theorem implies
1
2i
_
~
T:c_
q
r
(:) d: = 1c: (q
r
; 1) 1:d
~
T:c_
(1) = lim
s1
(: 1) q
r
(:) .
Now by Lemma 2, (:) =
1
s1
/(:), / H (\
0
), and so

t
(:) =
1
(: 1)
2
/
t
(:) ,
hence by (2.3),
(: 1) q
r
(:) = (: 1)
r
s+1
: (: 1)
_

t
(:)
(:)
_
= (: 1)
r
s+1
: (: 1)
_

1
(s1)
2
/
t
(:)
1
s1
/(:)
_
=
r
s+1
: (: 1)
_
1 (: 1) /
t
(:)
1 (: 1) /(:)
_

r
2
2
as : 1.
Alternatively, (2.6) shows that
1c:
_

t
(:)
(:)
; 1
_
= 1,
and then as above, (2.3) implies that
1c:
_
r
s+1
: (: 1)

t
(:)
(:)
; 1
_
=
_
r
s+1
: (: 1)
[
s=1
_
(1) =
r
2
2
.
Let ,
1,q
be the innite taxicab path consisting of the ve segments
[1 i, 1 i1[ , [1 i1, 1 j i1[ , [1 j i1, 1 j i1[ ,
[1 j i1, 1 i1[ , [1 i1, 1 i[ ,
concatenated with the directions and in the sequence given. Thus ,
1,q
is the line
[1 i, 1 i[ but with a rectangular jog to the left around the pole at 1.
Step 2: For 1 _ 8 and 0 < j < j (1) < 1 < c we have,
1
2i
_
c+Io
cIo
r
s+1
: (: 1)
_

t
(:)
(:)
_
d: =
r
2
2

1
2i
_
o
T_
r
s+1
: (: 1)
_

t
(:)
(:)
_
d:.
Let o tend to innity in (3.9). With q
r
as in (3.10) we claim that the integrals
over the top and bottom horizontal segments,
(3.11)
_
[c+IS,1+IS]
q
r
(:) d: and
_
[1IS,cIS]
q
r
(:) d:,
each tend to zero as o . For this we will need estimates on the growth of the
function t

0
(1+Ir)
(1+Ir)

. Unfortunately the formula (3.5) for

0
(s)
(s)
is not much help
in this regard since the formula doesnt take into account cancellations as : nears 1.
58 5. CONTOUR INTEGRALS AND THE PRIME NUMBER THEOREM
Instead we will estimate

t
(1 it)

and

1
(1+Ir)

separately using Lemma 2 and

a renement of the argument in the proof of Theorem 17.
In order to estimate

t
(1 it)

, recall from Lemma 2 that

(:) =
1
: 1

o

n=1
/
n
(:) , : \
1
,
where the entire function
/
n
(:) =
_
n+1
n
_
1
:
s

1
r
s
_
dr
satises (1.1),
[/
n
(:)[ _
[:[
:
c+1
, : = o it C,
as well as the crude estimate
[/
n
(:)[ _
2
:
c
, : = o it C.
Let 0 < c, - < 1. Taking an --skewed geometric mean of these estimates we have
[/
n
(:)[ _
_
[:[
:
c+1
_
:
_
2
:
c
_
1:
_ C
[:[
:
:
c+:
, : = o it C,
and then for o - _ 1 c, we have
[ (:)[ _
1
[: 1[

o

n=1
[/
n
(:)[ _
1
[t[
C
o

n=1
(o [t[)
:
:
c+:
(3.12)
_
1
[t[
C
(o [t[)
:
o - 1
_ C
o
[t[
:
,
for : = o it, o - _ 1 c, and [t[ _ 1.
Now with - = 2c (which forces c <
1
2
), apply Cauchys estimates to (:) on
the disk 1(1 it, c) using (3.12) to conclude that for any 0 < c <
1
2
(and with a
constant C
o
that may change from one occurence to the next),
(3.13)

t
(1 it)

_
C
o
[t[
:
c
_ C
o
[t[
2o
, [t[ _ 1.
In order to estimate

1
(1+Ir)

we recall from the proof of Theorem 17 that

(o)
3
(o it)
4
(o i2t)

_ 1, o _ 1, t R 0 .
Thus

(o it)
4

_ [ (o)[
3
[ (o i2t)[
1
, o _ 1,
and from (3.12) we now conclude that for o _ 1,
[ (o it)[
4
_ c (o 1)
3
C
1
o
[t[
:
, : = o it, o - _ 1 c, [t[ _ 1.
Rewriting the inequality with 0 < - = c < 1 we get
[ (o it)[ _ C (o 1)
3
4
[t[

o
4
.
3. PROOF OF THE PRIME NUMBER THEOREM 59
Now if o
r
= 1
1
]r]
13o
(the number 18 is convenient for numerology but any
number bigger than 0 would work here), then we have
[ (o
r
it)[ _ C
_
[t[
13o
_3
4
[t[

o
4
= C [t[
10o
,
and using (3.13), which requires 0 < c <
1
2
, we get
[ (o
r
it) (1 it)[ =

_
c:
1

t
(n it) dn

_
_
c:
1

t
(n it)

dn
_
_
1+
1
j:j
13o
1
C
o
[t[
2o
dn = C
o
[t[
11o
.
Thus for [t[ suciently large we have
[ (o
r
it) (1 it)[ _
1
2
[ (o
r
it)[ ,
since C
o
[t[
11o
_
1
2
C [t[
10o
for [t[ suciently large. Consequently, we obtain
[ (1 it)[ _
1
2
[ (o
r
it)[ _
C
2
[t[
10o
, 0 < c <
1
2
,
for [t[ suciently large.
Altogether we have
(3.14)

t
(1 it)
(1 it)

_ c
_
C
o
[t[
2o
__
C
o
[t[
10o
_
= C
o
[t[
12o
, 0 < c <
1
2
.
It now follows from (3.14) with any 0 < c <
1
2
that
[q
r
(:)[ =

r
s+1
: (: 1)
_

t
(:)
(:)
_

_ C
o
r
c+1
[t[
2
[t[
12o
= C
o
r
c+1
[t[
12o2
,
for [t[ suciently large. Thus the integrals in (3.11) are dominated by (c 1) o
12o2
for o suciently large, which tends to 0 as o provided 0 < c <
1
6
.
Step 2 now follows immediately from Step 1.
Step3: Given - 0 there is 8 _ 1 < and 0 < j < j (1) < 1 so that for
all r suciently large,

1
2i
_
o
T_
r
s+1
: (: 1)
_

t
(:)
(:)
_
d:

< -
r
2
2
.
We may assume that r _ 1. First we x 1 so large that

_
1I1
1Io
r
s+1
: (: 1)
_

t
(:)
(:)
_
d:

_
1+Io
1+I1
r
s+1
: (: 1)
_

t
(:)
(:)
_
d:

_
1I1
1qI1
r
s+1
: (: 1)
_

t
(:)
(:)
_
d:

_
1+I1
1q+I1
r
s+1
: (: 1)
_

t
(:)
(:)
_
d:

< -
r
2
4
, 0 < j < 1,
uniformly in r _ 1. This is possible since (3.14) implies both

_
1+Io
1+I1
r
s+1
: (: 1)
_

t
(:)
(:)
_
d:

_
_
o
1
r
2
t
2
C
o
t
12o
dt _
C
o
1 12c
1
12o1
r
2
,
60 5. CONTOUR INTEGRALS AND THE PRIME NUMBER THEOREM
and

_
1I1
1qI1
r
s+1
: (: 1)
_

t
(:)
(:)
_
d:

_
1+I1
1q+I1
r
s+1
: (: 1)
_

t
(:)
(:)
_
d:

_ 2
_
1
1q
r
2
1
2
C
o
1
12o
do _ C
o
r
2
1
12o2
,
where both 1
12o1
and 1
12o2
tend to 0 as 1 if 0 < c <
1
12
.
Recall that for 0 < j < j (1), (:) has no zero on or inside the rectangle with
sides
[1 j i1, 1 i1[ , [1 i1, 1 i1[ , [1 i1, 1 j i1[ , [1 j i1, 1 j i1[ .
Thus we estimate

_
1q+I1
1qI1
r
s+1
: (: 1)
_

t
(:)
(:)
_
d:

_ r
2q
_
1
1
1
[1 j it[ [2 j it[

t
(:)
(:)

dt = C
1,q
r
2q
.
Altogether we then have

1
2i
_
o
T_
r
s+1
: (: 1)
_

t
(:)
(:)
_
d:

< -
r
2
4
C
1,q
r
2q
=
_
-
2

C
1,q
r
q
_
r
2
2
,
which completes the proof of Step 3.
The asymptotic estimate (3.7),
lim
ro
c
1
(r)
1
2
r
2
= 1,
now follows immediately. Indeed, from the identity (3.8),
c
1
(r) =
1
2i
_
c+Io
cIo
r
s+1
: (: 1)
_

t
(:)
(:)
_
d:,
and Steps 2 and 3 above, we have for any - 0,

c
1
(r)
1
2
r
2
1

c
1
(r)
1
2
r
2
1
2
r
2

<

-
r
2
2
1
2
r
2

= -,
for all suciently large r.
This completes the proof of the Prime Number Theorem.
Corollary 9. If j
n

o
n=1
is the sequence of primes in increasing order, then
lim
no
j
n
:ln:
= 1.
Proof : Since lnr is continuous, the Prime Number Theorem yields
0 = ln1 = lim
ro
ln
(r)
r
ln r
= lim
ro
ln (r) lnlnr lnr .
If we now divide by lnr and use lHspitals rule to get
lim
ro
lnlnr
lnr
= lim
ro
1
r ln r
1
r
= lim
ro
1
lnr
= 0,
3. PROOF OF THE PRIME NUMBER THEOREM 61
we obtain lim
no
ln t(r)
ln r
= 1. This and another application of the Prime Number
Theorem yield
lim
ro
(r) ln (r)
r
= lim
ro
(r)
r
ln r
lim
ro
ln (r)
lnr
= 1.
If we now replace r by j
n
and use (j
n
) = :, we obtain lim
no
nln n
r
= 1.
Part 2
Boundary behaviour of Riemann
maps
In Part 2 of these notes we address Carathodorys theorem on extending a
Riemann map to a homeomorphism up to the boundary. Chapter 6 introduces
the theory of the Poisson integral and establishes Fatous theorem on radial lim-
its, which is the key ingredient in extending a Riemann map to a simple bound-
ary point. In Chapter 7 we use Lindelfs theorem to help complete the proof of
Carathodorys theorem. Finally, Appendices A and B provide those aspects of
topology and measure theory needed in Part 2.
Suppose that ) : D \ is a Riemann map onto a bounded simply connected
domain \. The question we consider here is this: when does the map ) extend
to a homeomorphism from the closure D to the closure \? An obvious necessary
condition is that 0\ is a Jordan curve since ) : T 0\ is continuous, one-to-one
and onto. Carathodorys Theorem shows that the converse holds. Moreover, we
also give a topological test for 0\ to be a Jordan curve involving the following
notion of a simple boundary point.
Definition 11. A boundary point n of a simply connected domain \ in the
plane is called simple if for every sequence .
n

o
n=1
\ with lim
no
.
n
= n, there
is a (continuous) curve , : [0, 1) \ and an increasing sequence of parameter
points t
n

o
n=1
[0, 1) with lim
no
t
n
= 1 such that , (t
n
) = .
n
for all : _ 1 and
lim
|1
, (t) = n.
Roughly speaking the point n 0\ is simple if for every sequence in \ that
approaches n, there is a curve passing in order through the sequence that has limit
n. For example, every boundary point of the unit disk D is simple, and we will
show below that that every point in a Jordan curve is a simple boundary point
of its bounded component. An example of a simply connected domain \ that has
nonsimple boundary points is the open square (0, 1)
2
with the vertical line segments
1
n
=
_
1
2
r
_

_
0, 1
1
2
r

removed for each : _ 1. Each point on 1

n
other than the
endpoint
_
1
2
r
, 1
1
2
r
_
is a nonsimple boundary point (consider a sequence that
alternates each side of 1
n
) and each point (0, j) with 0 < j < 1 is an especially
"nonsimple" boundary point in that there is not a single curve in \ with limit
(0, j).
Theorem 20. (Carathodorys Theorem) Suppose that \ is a bounded simply
connected domain \ in the plane. Then the following conditions are equivalent:
(1) Every Riemann map ) : D \ extends to a homeomorphism from D to
\,
(2) The boundary 0\ of \ is a Jordan curve,
(3) Every boundary point of \ is simple.
The most dicult implication is that (3) implies (1). For this we will need the
Poisson representation of holomorphic functions along with Fatous Theorem and
Lindelfs Theorem, which we take up in the next chapter.
CHAPTER 6
The Poisson representation
We begin by introducing one of the most famous Hilbert spaces in analysis,
the Hardy space H
2
(D) on the unit disk D. For ) (.) =

o
n=0
a
n
.
n
, . D, the
orthogonality relations
(0.15)
1
2
_
2t
0
c
I(nn)0
d0 =
_
1 if : = :
0 if : ,= :
yield
1
2
_
2t
0

n=0
a
n
_
rc
I0
_
n

2
d0 =
1
2
_
2t
0
o

n,n=0
a
n
_
rc
I0
_
n
a
n
(rc
I0
)
n
d0
=
o

n,n=0
a
n
a
n
r
n+n
1
2
_
2t
0
c
I(nn)0
d0 =
o

n=0
[a
n
r
n
[
2
for 0 < r < 1 by absolute convergence. Thus we have
(0.16)
o

n=0
[a
n
[
2
= sup
0<:<1
1
2
_
2t
0

n=0
a
n
_
rc
I0
_
n

2
d0 = sup
0<:<1
1
2
_
2t
0

)
_
rc
I0
_

2
d0,
which we use to dene the norm squared |)|
2
1
2
(D)
for the Hardy space
H
2
(D) =
_
) H (D) : sup
0<:<1
_
1
2
_
2t
0

)
_
rc
I0
_

2
d0
_
1
2
<
_
.
We digress for a moment to provide the proof of completeness of the Hilbert
space 1
2
(T) of square integrable Lebesgue measurable functions on the circle T =
0D, i.e. those measurable )
_
c
I0
_
satisfying
|)|
2
=
_
1
2
_
2t
0

)
_
c
I0
_

2
d0
_
1
2
< .
We equip 1
2
(T) with the metric d (), q) = |) q|
2
, ), q 1
2
(T). It is here
that Lebesgue integration is required, as the next result would fail with Riemann
integration in place of Lebesgue integration. See the appendix for this and the basic
theory of Lebesgue integration.
Proposition 13. The metric space 1
2
(T) is complete.
Proof : Suppose that )
n

o
n=1
is a Cauchy sequence in 1
2
(T). Choose a rapidly
converging subsequence )
n
!

o
|=1
, by which we mean

o
|=1
_
_
)
n
!+1
)
n
!
_
_
2
< .
65
66 6. THE POISSON REPRESENTATION
This is easily accomplished inductively by choosing for example :
|

o
|=1
strictly
increasing such that
|)
n
)
n
!
|
2
<
1
2
|
, : _ :
|+1
.
Then set
q = [)
n1
[
o

|=1

)
n
!+1
)
n
!

.
By Minkowskis inequality we have
|q|
2
_ |)
n1
|
2

o

|=1
_
_
)
n
!+1
)
n
!
_
_
2
< ,
and it follows that
0 _ q
_
c
I0
_
=

)
n1
_
c
I0
_

|=1

)
n
!+1
_
c
I0
_
)
n
!
_
c
I0
_

<
for almost every 0 [0, 2). Thus the series
)
n1
_
c
I0
_

|=1
_
)
n
!+1
_
c
I0
_
)
n
!
_
c
I0
__
converges absolutely for almost every 0 [0, 2) to a Lebesgue measurable function
)
_
c
I0
_
.
We claim that ) 1
2
(T) and that lim
no
)
n
= ) in 1
2
(T). Indeed, Fatous
lemma gives
1
2
_
2t
0

)
_
c
I0
_
)
n

_
c
I0
_

2
d0 =
1
2
_
2t
0
lim inf
|o

)
n
!
_
c
I0
_
)
n

_
c
I0
_

2
d0
_ lim inf
|o
1
2
_
2t
0

)
n
!
_
c
I0
_
)
n

_
c
I0
_

2
d0
= lim inf
|o
|)
n
!
)
n

|
2
2
0
as / by the Cauchy condition. This shows that ) )
n

1
2
(T), hence
) 1
2
(T), and also that )
n

) in 1
2
(T) as / . Finally, this together with
the fact that )
n

o
n=1
is a Cauchy sequence, easily shows that )
n
) in 1
2
(T) as
: .
Porism 1: If )
n

o
n=1
is a rapidly converging sequence in 1
2
(T),
o

n=1
|)
n+1
)
n
|
2
< ,
then
lim
no
)
n
_
c
I0
_
= )
1
_
c
I0
_

n=1
_
)
n+1
_
c
I0
_
)
n
_
c
I0
__
exists for almost every 0 [0, 2).
6. THE POISSON REPRESENTATION 67
Returning to the Hardy space H
2
(D) we note that (0.15) shows that )
:

0<:<1
is Cauchy in 1
2
(T) as r 1 where )
:
(.) = ) (r.) for r < 1 and . D:
|)
:
)
s
|
2
J
2
(T)
=
1
2
_
2t
0

n=0
(r
n
:
n
) a
n
c
In0

2
d0 =
o

n=0
[r
n
:
n
[
2
[a
n
[
2
0
as r, : 1 by the dominated convergence theorem for series since

o
n=0
[a
n
[
2
<
by (0.16). By Proposition 13, the completeness of 1
2
(T), there is )
+
1
2
(T) such
that )
+
= lim
:1
)
:
in 1
2
(T). We easily compute by taking limits that the Fourier
coecients of )
+
satisfy

)
+
(:) =
1
2
_
2t
0
)
+
_
c
I0
_
c
In0
d0 = lim
:1
1
2
_
2t
0
)
:
_
c
I0
_
c
In0
d0
=
_
a
n
if : _ 0
0 if : < 0
,
and that the inner product , on H
2
(D) satises
), q =
o

n=0
a
n
/
n
=
1
2
_
2t
0
)
+
_
c
I0
_
q
+
(c
I0
)d0
where ) (.) =

o
n=0
a
n
.
n
and q (.) =

o
n=0
/
n
.
n
. We also have by taking limits
the Cauchy formula
(0.17) ) (.) = lim
:1
)
:
(.) = lim
:1
1
2i
_
T
)
:
(n)
n .
dn =
1
2i
_
T
)
+
(n)
n .
dn,
which can also be expressed in terms of the inner product as
(0.18) ) (.) =
1
2i
_
T
)
+
_
c
I0
_
c
I0
.
ic
I0
d0 =
1
2
_
T
)
+
_
c
I0
_
1 c
I0
.
d0 = ), /
:
,
for . D where
(0.19) /
:
(n) =
1
1 .n
=
o

n=0
.
n
n
n
H
2
(D)
(since |/
:
|
1
2
(D)
=
_

o
n=0
[.[
2n
=
1
_
1]:]
2
) is the so-called reproducing kernel for
H
2
(D).
We would now like to obtain a representation formula of Cauchy type, such as
in (0.17), for the real part of ) (.). So let ) (.) = n(.) i (.) where n and are
real-valued functions in the disk. Unfortunately we cannot just take the real part
of each side in the Cauchy representation
(0.20) ) (.) =
1
2i
_
T
)
+
(n)
n .
dn =
1
2
_
2t
0
)
+
_
c
I|
_
c
I|
c
I|
.
dt,
since both the function )
+
_
c
I|
_
and the kernel
t
1t
t
1t
:
are complex-valued. If we
could replace the kernel
t
1t
t
1t
:
in (0.20) with a kernel that was real-valued, then
we could indeed take real parts of both sides of (0.20) to obtain a representation
formula for n(.) = Io ) (.).
For this purpose we have at our disposal Cauchys theorem which says that
1
2i
_
T
)
+
(n)
n
dn = lim
:1
1
2i
_
T
)
:
(n)
n
dn = lim
:1
0 = 0
68 6. THE POISSON REPRESENTATION
for any C D (since
}r(u)
u
is holomorphic in a neighbourhood of the closed disk
D). Motivated by the fact that . . is real where . is the reection of . across the
real line R, we consider the reection .
+
of . across the circle T given by
.
+
=
.
[.[
2
=
1
.
, . ,= 0.
If we let = .
+
we obtain
) (.) 0 =
1
2
_
2t
0
)
+
_
c
I|
_
_
c
I|
c
I|
.

c
I|
c
I|
.
+
_
dt.
Now we compute that
c
I|
c
I|
.

c
I|
c
I|
.
+
=
c
I|
c
I|
.

c
I|
c
I|

1
:
=
c
I|
c
I|
.

.
. c
I|
=
c
I|
c
I|
.

.
c
I|
.
=
1 [.[
2
[. c
I|
[
2
,
which yields the Poisson representation of ):
(0.21) ) (.) =
1
2
_
2t
0
)
+
_
c
I|
_
1 [.[
2
[. c
I|
[
2
dt, . D.
Finally, we can take real parts of each side of this latter equation to obtain
(0.22) n(.) =
1
2
_
2t
0
n
+
_
c
I|
_
1 [.[
2
[. c
I|
[
2
dt, . D,
where we write )
+
= n
+
i
+
with n
+
and
+
real-valued functions on T. This
formula is valid for n = Io ) where ) H
2
(D).
Remark 5. An important side benet of the Poisson kernel is its small size in
comparison to the Cauchy kernel:
1
2
_
2t
0
1 [.[
2
[. c
I|
[
2
dt = 1 while
1
2
_
2t
0
1
[c
I|
.[
dt - ln
_
1
1
1 [.[
_
for . D. This small size will play a crucial role in the proof of Fatous Theorem
22 below.
Remark 6. The Poisson representation (0.21) can also be quickly obtained
from the Cauchy representation (0.20) via the following trick. For ) H
2
(D) and
. D, let q (n) =
}(u)
1u:
. Then q H
2
(D), q
+
_
c
I|
_
=
}

(t
1t
)
1t
1t
:
and (0.20) yields
) (.)
1 [.[
2
= q (.) =
1
2
_
2t
0
)
+
_
c
I|
_
1 c
I|
.
c
I|
c
I|
.
dt =
1
2
_
2t
0
)
+
_
c
I|
_
[. c
I|
[
2
dt.
We note in passing the following consequence of Porism 6: if lim
no
r
n
= 1
and )
:r

o
n=1
is a rapidly converging sequence in 1
2
(T), then
(0.23) )
+
_
c
I|
_
= lim
no
)
:r
_
c
I|
_
, a.c. c
I|
T,
and an analogous statement holds for n
+
_
c
I|
_
. For the proof of Carathodorys
Theorem 20 we will need the much stronger theorem of Fatou that says )
+
_
c
I|
_
=
lim
:1
)
:
_
c
I|
_
for a.e. c
I|
T. This will be proved below after a short detour to
consider harmonic functions and the Dirichlet problem in the disk. But rst we
1. HARMONIC FUNCTIONS 69
note the following uniqueness result for bounded holomorphic functions that will
play a signicant role in the proof of Theorem 20.
Lemma 3. If ) H
o
(D) and )
+
_
c
I|
_
= 0 for almost every c
I|
in some arc 1
of positive length in T, then ) is identically zero.
Proof : Choose an integer so that [1[
2t

and consider the function 1

H
2
(D) given by
1 (.) =

|=1
)
_
c
I
2!
1
.
_
=

|=1
)
|
(.) , . D.
The boundedness of ) is used in concluding that the product 1 is actually a bounded
holomophic function in the disk D, hence is in H
2
(D). Next choose a sequence
r
n

o
n=1
with lim
no
r
n
= 1 for which 1
:r
and
_
)
|
:r
_
o
n=1
, 1 _ / _ , are each
rapidly convergent sequences in 1
2
(T). Then for almost every t [0, 2) we have
1
+
_
c
I|
_
= lim
no
1
:r
_
c
I|
_
= lim
no

|=1
_
)
|
_
:r
_
c
I|
_
=

|=1
lim
no
_
)
|
_
:r
_
c
I|
_
=

|=1
_
)
|
_
+
_
c
I|
_
.
Since
_
)
|
_
+
_
c
I|
_
= 0 for a.e. t c
I
2!
1
1, and since

|=1
c
I
2!
1
1 = T, we conclude
that 1
+
_
c
I|
_
= 0 for a.e. t T. Now either the Cauchy representation (0.17) or
the Poisson representation (0.21) shows that 1 is identically zero in the disk D.
Thus the zero set 7 ()) of ) must be uncountable since D = 7 (1) =

|=1
7
_
)
|
_
,
and hence ) is identically zero in the disk D by Theorem 6.
1. Harmonic functions
We pause at this point to note that we only use the Poisson representation for
holomorphic functions (0.21) in our proof of Carathodorys Theorem 20. However,
since we used the real part representation (0.22) to motivate our calculation of the
Poisson kernel
1]:]
2
]:t
1t
]
2
, we will spend some time using (0.22) to develop the most
elementary facts in the theory of harmonic functions n, i.e. twice continuously
dierentiable solutions n to Laplaces equation
n =
0
2
n
0r
2

0
2
n
0j
2
= 0.
In particular we will prove formula (0.22) with n
+
_
c
I|
_
= n
_
c
I|
_
for any n
C
_
D
_
C
2
(D) that is harmonic in D.
We rst observe that if ) = n i H (D), then the functions n and are
harmonic in D:
(1.1) n(.) = (.) = 0, . D.
Indeed, if ) = n i is a holomorphic function of . = r ij then
J
J:
) = 0 and so
0 =
0
0.
0 =
0
0.
0
0.
) (.) =
1
2
_
0
0r

1
i
0
0j
_
1
2
_
0
0r

1
i
0
0j
_
) (.)
=
1
4
_
0
2
0r
2

0
2
0j
2
_
(n(.) i (.)) =
1
4
n(.)
i
4
(.) .
70 6. THE POISSON REPRESENTATION
We next note the many faces of the Poisson kernel
1]:]
2
]:t
1t
]
2
.
Lemma 4. For . = rc
I0
D and c
I|
T we have
(1.2)
1 r
2
1 2r cos (0 t) r
2
=
1 [.[
2
[. c
I|
[
2
= Io
_
c
I|
.
c
I|
.
_
=

nZ
r
]n]
c
In(0|)
.
Proof : We have
Io
_
c
I|
.
c
I|
.
_
= Io
_
1 2
.
c
I|
.
_
= 1 2 Io
.
c
I|
.
= 1
.
c
I|
.

.
c
I|
.
=
c
I|
c
I|
.

.
c
I|
.
=
1 [.[
2
[. c
I|
[
2
,
and also
1 2 Io
.
c
I|
.
= 1 2 Io
.c
I|
1 .c
I|
= 1 2 Io
o

n=1
_
.c
I|
_
n
=

nZ
r
]n]
c
In(0|)
.
An inspection of the representation formula (0.22) together with the lemma
reveals that (0.22) can be rewritten
n(.) =
1
2
_
2t
0
n
+
_
c
I|
_
1 r
2
1 2r cos (0 t) r
2
dt
=
1
2
_
2t
0
n
+
_
c
I|
_
1
:
(0 t) dt
= 1
:
+ n
+
(0) , . = rc
I0
D,
where
1
:
(c) =
1 r
2
1 2r cos (c) r
2
, 0 _ r < 1, 0 _ c < 2,
and
() + q) () =
_
T
)
_
j
1
_
q (j) do (j)
denotes convolution of ) and q on the compact group T with Haar measure do
_
c
I|
_
=
1
2t
dt.
A crucial observation at this juncture is that the function 1
:
+,(0) is a harmonic
function of . = rc
I0
D for any integrable function , on T. Indeed, if , is real-
valued, then
1
:
+ ,(0) =
1
2
_
2t
0
,
_
c
I|
_
1
:
(0 t) dt
=
1
2
_
2t
0
,
_
c
I|
_
Io
_
c
I|
.
c
I|
.
_
dt
= Io
1
2
_
2t
0
,
_
c
I|
_
_
c
I|
.
c
I|
.
_
dt.
1. HARMONIC FUNCTIONS 71
Now the calculation in (3.2) shows that the integral
_
2t
0
,
_
c
I|
_
_
c
I|
.
c
I|
.
_
dt
denes a holomorphic function of . in the disk D, and hence its real part 1
:
+ ,(0)
is harmonic by (1.1). In general we write , as a sum of its real and imaginary parts
and apply the above result to each part. This completes the proof that 1
:
+ ,(0)
is harmonic for integrable ,.
The above observation justies the following denition: for , 1
1
(T), the
space of complex-valued Lebesgue integrable functions on T, we set
P,(.) = 1
:
+ ,(0) , . = rc
I0
D,
and refer to P, as the Poisson integral of ,. Note that , is an integrable function
dened on T = 0D the boundary of the disk D, while P, is a harmonic function
dened on the disk D itself. A natural question now arises: if , is continuous on
the circle, does the Poisson integral P, have limit , at the boundary, i.e. is the
function
(1.3) ,(.) =
_
P,(.) for . D
,(.) for . T
,
dened for . in the closed disk D, actually continuous on D? The answer turns out
to be yes, and this justies writing P, for , on the closed disk D, and referring to
P, as the Poisson extension of , to the closed disk D.
Proposition 14. If , C (T), then , C
_
D
_
where , is the Poisson exten-
sion of , to D dened in (1.3).
Proof : We have 1
:
(c) =
1:
2
]:t
10
1]
2
by (1.2) with 0 = c and t = 0. We now
claim that 1
:
has the following three properties of an approximate identity:
(1) 1
:
(c) 0 for all rc
I
D,
(2)
_
2t
0
1
:
(c) dc = 1 for all rc
I
D,
(3) max
]t
1
1]o
1
:
(c) tends to 0 as r 1 for each xed 0 < c < 2.
The rst property is obvious, the second property follows from (0.21) with
) = 1, and the third property follows from

rc
I
1

c
I
1

_ c.
We now use a standard paradigm to show that P,
_
rc
I0
_
,
_
c
I
_
as rc
I0
c
I
.
By rotation invariance we may assume that c = 0, i.e. c
I
= 1, and by property (2)
we may subtract the constant ,(1) which means we may assume that ,(1) = 0.
Thus we must prove
lim
:t
10
1
P,
_
rc
I0
_
= 0.
Let - 0 be given. By the continuity of , there is 0 < c < 2 so that
(1.4)

,
_
c
I|
_

<
-
2
, whenever

c
I|
1

< c.
72 6. THE POISSON REPRESENTATION
So using properties (1) and (2) and inequality (1.4) we estimate

P,
_
rc
I0
_

1
2
_
2t
0
,
_
c
I|
_
1
:
(0 t) dt

_
1
2
_
_
]t
1t
1]<o

_
]t
1t
1]o
_

,
_
c
I|
_

1
:
(0 t) dt
<
-
2
max
T
[,()[ max
]t
1
1]
o
2
1
:
(c) ,
provided

c
I0
1

<
o
2
, since then

c
I(0|)
1

_
c
I(0|)
c
I0
_

_
c
I0
1
_

c
I|
1

c
I0
1

_ c
c
2
=
c
2
.
But now property (3) shows that for all r suciently close to 1 and

c
I0
1

<
o
2
we have

P,
_
rc
I0
_

<
-
2

-
2
= -.
Since this inequality holds also for r = 1 and

c
I0
1

<
o
2
, the proof of Proposition
14 is complete.
Proposition 14 solves the Dirichlet problem for the Laplace operator in the unit
disk. Namely, given continuous boundary data , on T, there is n C
_
D
_
C
2
(D)
satisfying the boundary value problem:
(1.5)
_
n = 0 in D
n = , on T = 0D
.
Uniqueness of the solution n to the Dirichlet problem (1.5) will follow from the
maximum principle for harmonic functions.
Proposition 15. (Maximum principle for harmonic functions) Let \ be a
bounded domain in the plane. If n C
_
\
_
C
2
(\) is harmonic in \, then n
achieves its maximum on the boundary:
(1.6) sup
:
n(.) _ sup
:J
n(.) .
Proof : For - 0 consider the function
n
:
(.) = n(.) - [.[
2
,
for which we have
n
:
(.) = n(.) -
_
r
2
j
2
_
= 4- 0.
By Fermats theorem, we have both
J
2
uz
Jr
2
(.) _ 0 and
J
2
uz
J
2
(.) _ 0 at a relative
maximum ., and it follows that n
:
cannot have a relative maximum in \, and so
must achieve its maximum on the boundary 0\:
sup
:
n(.) _ sup
:
n
:
(.) _ sup
:J
n
:
(.) _ sup
:J
n(.) - sup
:J
[.[
2
.
If we let - 0 we obtain (1.6).
Now we prove the uniqueness of the solution n to the Dirichlet problem (1.5).
If is another solution, then n = n is harmonic in D and vanishes on 0D. By
2. FATOUS THEOREM 73
the maximum principle Proposition 15, we conclude that n _ 0 in D. But n is
also harmonic and vanishes on the boundary, so n _ 0 in D as well. Thus n = .
Suppose now that n C
_
D
_
C
2
(D) is harmonic in D. Let Pn be the Poisson
integral of the restriction n [
T
of n to the circle T. Then both n and Pn satisfy the
Dirichlet problem (1.5) with , = n [
T
. By uniqueness we then have
n(.) = Pn(.) =
1
2
_
2t
0
n
_
c
I|
_
1
:
(0 t) dt, . = rc
I0
D.
Thus n equals its Poisson integral for any n C
_
D
_
C
2
(D) that is harmonic in
D. This proves (0.22) for such functions and moreover, by translating and rescal-
ing formula (0.22) to arbitrary disks we obtain the following characterization of
harmonic functions.
Theorem 21. Let \ be an open subset of the plane. Then n C
2
(\) is
harmonic in \ if and only if n is locally the real part of a holomorphic function, i.e.
for any disk 1(a, 1) \, there is ) H (1(a, 1)) such that n = Io ) in 1(a, 1).
Moreover, n equals its Poisson integral in any disk 1(a, 1) \:
n
_
a 1rc
I0
_
=
1
2
_
2t
0
n
_
a 1c
I|
_
1
:
(0 t) dt.
Remark 7. The above theorem is an example of a regularity theorem for a
partial dierential equation. It says that any C
2
solution n to Laplaces equation
_
J
2
Jr
2

J
2
J
2
_
n = 0 must be innitely dierentiable, in fact real-analytic. This type
of phenomenon persists more generally for elliptic partial dierential equations.
2. Fatous Theorem
Our purpose now is to extend the pointwise limit on certain radii in (0.23) to
the full radial limit.
Theorem 22. (Fatous Theorem) For ) H
2
(D) we have
lim
:1
)
_
rc
I|
_
= )
+
_
c
I|
_
, a.c. c
I|
T.
Proof : Due to (0.23), we see that if the radial limit lim
:1
)
_
rc
I|
_
exists, then
it must equal )
+
_
c
I|
_
almost everywhere. So it suces to prove that lim
:1
)
_
rc
I|
_
exists almost everywhere, equivalently that the oscillation .
_
); c
I|
_
given by
max
_
lim sup
:1
n
_
rc
I|
_
lim inf
:1
n
_
rc
I|
_
, lim sup
:1

_
rc
I|
_
lim inf
:1

_
rc
I|
_
_
,
where ) = n i, vanishes for almost every c
I|
T. The two pieces of information
at hand that we can exploit for this purpose are
.
_
); c
I|
_
= 0 if ) C
_
D
_
H (D),
)
s
C
_
D
_
H (D) and )
+
s
)
+
in 1
2
(T) if ) H
2
(D).
In order to take advantage of the second bullet item, we will need a way of
controlling the oscillation of ) )
s
in terms of the 1
2
norm |) )
s
|
2
. This is
accomplished with the aid of the maximal function: for / 1
1
(T) we dene
//
_
c
I0
_
= sup
t
10
1T
1
[1[
_
1

/
_
c
I|
_

dt, c
I0
T,
74 6. THE POISSON REPRESENTATION
where the notation sup
t
10
1T
means that the supremum is taken over all arcs 1
in the circle T that contain the point c
I0
. It turns out to be easy to show that
the maximal function dominates the Poisson integral in the sense that there is a
positive constant C such that
(2.1)

P/
_
rc
I0
_

_ C//
_
c
I0
_
, 0 _ r < 1.
Indeed, this follows readily from the following inequality with c = 1 r and C a
positive constant;
1
:
(0 t) =
1 r
2
[rc
I0
c
I|
[
2
_ C
1
2c

[0o,0+o]
(t) C
o

|=1
2
|
1
2
1|
c

[02
!
o,0+2
!
o]
,
since we conclude from this that

P/
_
rc
I0
_

1
2
_
t
t
1
:
(0 t) /
_
c
I|
_
dt

_ C
1
2c
_
0+o
0o

/
_
c
I|
_

dt C
o

|=1
2
|
1
2
1|
c
_
0+2
!
o
02
!
o

/
_
c
I|
_

dt
_ C//
_
c
I0
_
C
o

|=1
2
|
//
_
c
I0
_
_ C//
_
c
I0
_
.
The following Maximal Theorem is then decisive. We denote the rotation invariant
probability measure o of a subset 1 of T = [0, 2) by [1[
c
=
_
J
J|
2t
.
Theorem 23. (Maximal Theorem) For / 1
1
(T) and ` 0, we have
(2.2) `

_
c
I0
T : //
_
c
I0
_
`
_

c
_
2

_
2t
0

/
_
c
I|
_

dt.
Let us assume the Maximal Theorem for the moment, and show how it and the
above bullet items prove Fatous Theorem. Fix ) H
2
(D) and note that by the
subadditivity of the oscillation ., the bullet items and (2.1), we have
.
_
); c
I0
_
_ .
_
) )
s
; c
I0
_
.
_
)
s
; c
I0
_
= .
_
) )
s
; c
I0
_
_ 2 lim sup
:1

) )
s
_
rc
I0
_

= 2 lim sup
:1

P() )
s
)
+
_
rc
I0
_

_ 2C/() )
s
)
+
_
c
I0
_
= 2C/()
+
)
+
s
)
_
c
I0
_
.
Applying the Maximal Theorem we have for any ` 0,

_
c
I0
T : .
_
); c
I0
_
`
_

c
_

_
c
I0
T : /()
+
)
+
s
)
_
c
I0
_

`
2C
_

c
_
4C
`
_
2t
0

)
+
_
c
I|
_
)
+
s
_
c
I|
_

dt
_
4C
_
2
`
__
2t
0

)
+
_
c
I|
_
)
+
s
_
c
I|
_

2
dt
_
1
2
,
which tends to zero as : 1 by the second bullet item above. Thus

_
c
I0
T : .
_
); c
I0
_
`
_

c
= 0
2. FATOUS THEOREM 75
for all ` 0 and so

_
c
I0
T : .
_
); c
I0
_
,= 0
_

c
=

o
_
n=1
_
c
I0
T : .
_
); c
I0
_

1
:
_

c
_
o

n=1

_
c
I0
T : .
_
); c
I0
_

1
:
_

c
=
o

n=1
0 = 0.
This completes the proof that .
_
); c
I0
_
= 0 for almost every 0 T.
Proof of the Maximal Theorem 23: For _ 1, let c

=
_

2
1+2
2
_
12
1+2
and denote by
T

= [a, /) : a, / c

the collection of arcs in the circle having endpoints in the set c

. The point of
introducing the collections T

is that each one is a nite collection of arcs and the

following swallowing property holds: given any arc 1 T with [1[ _
1
2
1
2, there is
an arc J T

satisfying
(2.3) 1 J and [J[ _ 2 [1[ .
This allows us to reduce the proof of (2.2) to the same inequality with / replaced
by /

where
/

/
_
c
I0
_
= sup
t
10
11
1
1
[1[
_
1

/
_
c
I|
_

dt, c
I0
T.
Indeed, if we introduce the intermediate operator
/
+

/
_
c
I0
_
= sup
t
10
1T:]1]2
1
1
[1[
_
1

/
_
c
I|
_

dt, c
I0
T,
then since the measure of an increasing union of sets is the limit of the measures of
the sets, we have

_
c
I0
T : //
_
c
I0
_
`
_

c
= lim
o

_
c
I0
T : /
+

/
_
c
I0
_
`
_

c
.
Moreover, the swallowing property (2.3) of T

shows that
_
c
I0
T : /
+

/
_
c
I0
_
`
_

_
c
I0
T : /

/
_
c
I0
_

`
2
_
,
since if
1
]1]
_
1

/
_
c
I|
_

dt ` where [1[ _ 2

, then there is J T

satisfying (2.3),
so that
1
[J[
_

/
_
c
I|
_

dt _
[1[
[J[
1
[1[
_
1

/
_
c
I|
_

dt
1
2
`.
Altogether then it suces to prove
(2.4) `

_
c
I0
T : /

/
_
c
I0
_
`
_

c
_
1

_
2t
0

/
_
c
I|
_

dt.
76 6. THE POISSON REPRESENTATION
The inequality (2.4) is proved by a simple "covering lemma": if 1 =

1
n=1
1
n
,
1
n
T

, then there exists a subcollection 1

no

.
o=1
satisfying
(2.5) 1 =
.
_
o=1
1
no
and
.

o=1

1ro
_ 2,
i.e. the subcollection 1
no

.
o=1
covers 1 with overlap at most 2. To see this let
1 = [a, /) and pick 1
n1
= [a, /
n1
) to be a largest interval with endpoint a. Then
choose 1
n2
= [a
n2
, /
n2
) to be an interval containing /
n1
with largest endpoint /
n2
.
Inductively choose 1
no+1
=
_
a
no+1
, /
no+1
_
to be an interval containing /
no
with
largest endpoint /
no+1
. This procedure ends in nitely many steps and it is clear
that (2.5) holds with overlap 2 since if 1
no1
1
no+1
,= c, then we would have
chosen 1
no+1
in place of 1
no
at the a
||
inductive step.
Now we note that
_
c
I0
T : /

/
_
c
I0
_
`
_
=
_
_
1 T

:
1
[1[
_
1

/
_
c
I|
_

dt `
_
=

_
=1
1

,
where 1

= [a

, /

) with /

< a
+1
and
1

=
1
_
n=1
1

n
where
1

_
1

/
_
c
I|
_

dt `.
Apply the covering lemma to 1

to obtain 1

.
o=1
1

no
satisfying (2.5). Then we
have

_
c
I0
T : /

/
_
c
I0
_
`
_

c
=
1
2

=1

_
1
2

=1
.

o=1

no

<
1
2

=1
.

o=1
1
`
_
1

ro

/
_
c
I|
_

dt
=
1
2

=1
1
`
_
_
_
.

o=1

ro
(t)
_
_

/
_
c
I|
_

dt
_
1
2

=1
2
`
_
1

/
_
c
I|
_

dt _
1
`
_
2t
0

/
_
c
I|
_

dt.
This completes the proof of the Maximal Theorem 23 and with it, the proof of
Fatous Theorem 22.
CHAPTER 7
Extending Riemann maps
Fatous theorem plays a crucial role below in extending a Riemann map ) : \
D to a simple boundary point n 0\. However, we will also need to know that such
extensions to distinct boundary points n
1
, n
2
have distinct images ) (n
1
) , ) (n
2
).
For this we need one nal result concerning radial limits, namely Lindelfs theorem.
1. Lindelf s theorem
The next theorem shows that if a bounded holomorphic function q in the disk
D has limit 1 along some curve ending at c
I0
T, then we can conclude that q has
radial limit 1 at c
I0
.
Theorem 24. (Lindelf s Theorem) Suppose I : [0, 1[ D ' 1 is a con-
tinuous curve such that [I(t)[ < 1 if t < 1 and I(1) = 1. Then if q H
o
(D)
satises
lim
|1
q (I(t)) = 1,
it follows that q has radial limit 1 at 1:
lim
:1
q (r) = 1.
Proof : Without loss of generality we assume that 1 = 0 and |q|
o
< 1. Let
- 0 be given. We will show there is r
0
< 1 such that
[q (r)[ _
4
_
-, r
0
< r < 1.
First we note that since lim
|1
q (I(t)) = 0, there is a < 1 such that Io I(a)
1
2
and [q (I(t))[ < - whenever a < t < 1. Now let r
0
= Io I(a) and
t
0
= sup0 _ t < 1 : Io I(t) = r
0
,
so that we have
[q (I(t))[ < - and Io I(t) r
0

1
2
, t
0
< t < 1.
Fix r (r
0
, 1) for the moment and dene a lens-shaped region \ centered at r
by
\ = D D 2r = 1(0, 1) 1(2r, 1) ,
and a holomorphic function / H (\) by
/(.) = q (.) q (.)q (2r .)q (2r .) .
Note that / is a product of four holomorphic functions
/ = q
1
q
2
q
3
q
4
,
namely q
1
(.) = q (.), a "reection" q
2
(.) = q (.) of q
1
(.) across the real axis in the
disk D, a "reection" q
3
(.) = q
2
(2r .) of q
2
(.) across the vertical axis Io . = r
77
78 7. EXTENDING RIEMANN MAPS
to the disk D 2r, and a "reection" q
4
(.) = q
3
(.) of q
3
(.) across the real axis
in the disk D 2r. Clearly we have [/(.)[ < 1 for . \. Since /(r) = [q (r)[
4
we
will be done once we show that
(1.1) /(r) _ -.
We will use the maximum modulus principle, together with the geometry of
the curve I and its reections and translations, to obtain (1.1). Let
t
1
= sup0 _ t < 1 : Io I(t) = r (t
0
, 1) ,
and dene 1
1
= I([t
1
, 1[) to be the closed arc of I
+
that connects I(t
1
) to 1 and
has the property that 1
1
minus its endpoints lies in the open right half of \. Let
1
2
be the reection of 1
1
across the real axis, 1
3
be the reection of 1
2
across
the vertical axis Io . = r of symmetry of \, and nally let 1
4
be the reection of
1
3
across the real axis, which coincides with the reection of 1
1
across the vertical
axis Io . = r. Note that 1
1
and 1
2
can have a complicated intersection, but that
1
1
' 1
2
and 1
3
' 1
4
have only the points I(t
1
) and I(t
1
) in common. Then set
1 = 1
1
' 1
2
' 1
3
' 1
4
.
We have 1 \
+
= \'1, 2r 1. A crucial property of the function / on the
set 1 is
(1.2) [/(.)[ < -, . 1.
Indeed, for . 1, we must have . 1
I
for some i, and then the corresponding
factor q
I
satises [q
I
(.)[ < - while the other factors q

satisfy [q

(.)[ < 1. Now

pick a small c 0 and dene
/
o
(.) =
_
/(.) (1 .)
o
(2r 1 .)
o
for . \
0 for . = 1, 2r 1
.
The point of /
o
is that it is holomorphic in \ and continuous on \
+
= \'1, 2r 1
for c 0.
Now let 1 be the union of the compact set 1 and the bounded components of
the open set C 1. Then 1 is compact and /
o
satises the following properties on
1:
(1) /
o
is continuous on 1,
(2) /
o
is holomorphic in the interior

1 of 1,
(3) [/
o
[ < - on the boundary 01 of 1 by (1.2).
The maximum principle now shows that [/
o
[ < - on 1 since the boundary of
any connected component of C 1 lies in 1.
We claim that r 1 by the construction of 1. Indeed, if r 1 we are done, so
we assume r lies in the open set C1. Now consider the closed curve , obtained by
concatenating the curve I
1
: [t
1
, 1[ \
+
with its three reections I
2
, I
3
and I
4
,
taken in the appropriate direction, about the real axis and the axis of symmetry
Io . = r. Then ,
+
= 1 and we assume, in order to derive a contradiction, that
r lies in the unbounded component of C ,
+
= C 1. Then there is a path o in
C 1 joining r to the number 8, which is well outside the convex set \. Let c 0
satisfy both
1(r, 2c) C 1 and c <
1
2
di:t (o
+
, 1) .
2. PROOF OF CARATHODORYS THEOREM 79
Now choose a closed polygonal path (t) that joins consecutive (suciently close)
points on ,
+
with line segments, so that
(1.3) [, (t) (t)[ <
c
10
.
Then o
+
is disjoint from
+
and r lies in the unbounded component of C
+
.
Proposition 4 shows that
(1.4) 1:d
~
(r) = 0.
On the other hand, it is not hard to see that is (C r)-homotopic to
01(r, c), taken in the positive direction, simply by following in sequence those
portions
1
(t),
2
(t),
3
(t) and
4
(t) of (t) when , (t) is given by I
1
(t), I
2
(t),
I
3
(t) and I
4
(t) respectively. In fact, if c
1
(t), c
2
(t), c
3
(t) and c
4
(t) denote the
corresponding quarter arcs of 01(r, c) taken in the same sequence, then one can
use the homotopies H
I
(t, 0) = (1 0)
I
(t) 0c
I
(t) which avoid the point r since
both
+
I
and c
+
I
lie in a common half plane that doesnt contain r. Thus Proposition
9 shows that
1:d
~
(r) = 1:d
J1(:,o)
(r) = 1,
which contradicts (1.4) and shows that r must lie in a bounded component of C1,
hence in 1.
We thus conclude that [/
o
(r)[ < -. Now let c 0 to obtain [/(r)[ _ -, which
is (1.1), and this completes the proof of Lindelfs Theorem 24.
2. Proof of Carathodorys Theorem
We can now prove Carathodorys Theorem 20. First we will use Fatous The-
orem 22 to prove Lemma 5 below, and then we will use Lindelfs Theorem 24 to
prove Lemma 6 below.
Lemma 5. Let \ be a bounded simply connected domain in the complex plane,
and let ) : \ D be holomorphic, one-to-one and onto. Suppose that n is a simple
boundary point of \. Then ) has a continuous extension ) : \ ' n D and
) (n) T.
Lemma 6. Let \ be a bounded simply connected domain in the complex plane,
and let ) : \ D be holomorphic, one-to-one and onto. Suppose that n
1
and n
2
are distinct simple boundary points of \, and that ) : \ ' n
1
, n
2
D is as in
Lemma 5. Then ) (n
1
) ,= ) (n
2
).
Proof of Lemma 5: Let q = )
1
so that q : D \ is one-to-one and onto,
and q H
o
(D). Let n
n

o
n=0
\ be a sequence with limit n and such that
lim
no
) (n
2n
) =
0
and lim
no
) (n
2n+1
) =
1
.
Suppose, in order to derive a contradiction, that
0
,=
1
. Since n is a simple
boundary point of \, there is a curve : [0, 1[ \ ' n as in Denition 11 that
passes through the sequence n
n

o
n=0
and ends at n. Set
I(t) = ) ( (t)) , 0 _ t < 1.
Now for 0 < r < 1, q
_
rD
_
is a compact subset of \ disjoint from n. Thus
there is t
:
< 1 depending on r such that (t) , q
_
rD
_
if t
:
< t < 1. It follows that
80 7. EXTENDING RIEMANN MAPS
[I(t)[ r for t
:
< t < 1, and so lim
|1
[I(t)[ = 1. In particular, if n
n
= ) (t
n
),
then both
(2.1) [
0
[ = lim
|1
[I(t
2n
)[ = 1 and [
1
[ = lim
|1
[I(t
2n+1
)[ = 1.
Let 1
1
and 1
2
be the two open arcs of T whose union is T
0
,
1
. At least
one of these arcs, call it J, has the property that every segment o
0
from the origin
to a point c
I0
in J intersects the range of I in an innite subset of o
0
that has c
I0
as a limit point. Indeed, if not, there would be two segments o
01
and o
02
ending in
1
1
and 1
2
respectively such that for a suciently large T < 1, I(t) is disjoint from
o
01
' o
02
for all T < t < 1. But this contradicts the connectedness of I((T, 1))
since both
0
and
1
lie in the closure of I((T, 1)).
Thus for every c
I0
J at which q has a radial limit, we have
lim
:1
q
_
rc
I0
_
= n.
Now q H
o
(D) and so by Fatous Theorem 22, q
+
_
c
I0
_
= n for almost every
c
I0
J. The uniqueness result Lemma 3 now shows that q n H
o
(D) is
identically zero, the desired contradiction since q is one-to-one on D.
Thus
0
=
1
T, and we conclude that ) has a continuous extension to
\ ' n, and that ) (n) T.
Proof of Lemma 6: We prove the contrapositive: n
1
= n
2
if ) (n
1
) = ) (n
2
).
We may suppose that ) (n
1
) = ) (n
2
) = 1. Since n
I
is a simple boundary point of
\, there is a curve
I
: [0, 1[ \ ' n
I
with
I
([0, 1)) \ and
I
(1) = n
I
. Set
I
I
(t) = ) (
I
(t)) , 0 _ t _ 1.
Then I
I
([0, 1)) D and I
I
(1) = 1. Since q (I
I
(t)) =
I
(t) for 0 _ t < 1 we have
lim
|1
q (I
I
(t)) = lim
|1

I
(t) = n
I
.
Thus Lindelofs Theorem 24 implies that
n
I
= lim
:1
q (r) ,
for both i = 1 and i = 2, hence n
1
= n
2
.
The proof that (3) implies (1) in Theorem 20 is now easily accomplished in a
few lines. Suppose (3) holds and that ) : \ D is holomorphic and one-to-one.
Lemma 5 shows that there is an extension ) : \ D such that ) (n
n
) ) (n)
whenever n
n

o
n=1
\ is a sequence in \ converging to n. If .
n

o
n=1
\ is a
sequence in \ converging to n, then there exist points n
n
\ such that
[.
n
n
n
[ <
1
:
and [) (.
n
) ) (n
n
)[ <
1
:
.
Thus n
n

o
n=1
\ is a sequence in \ converging to n, and so ) (n
n
) ) (n).
But then ) (.
n
) ) (n) as well. This proves that ) : \ D is continuous.
Since D )
_
\
_
D and )
_
\
_
is compact, hence closed, we have )
_
\
_
= D
and so ) : \ D is onto.
Lemma 6 shows that ) : \ D is one-to-one.
Finally, it is a standard result that ) : \ D is now a homeomorphism.
Indeed, )
1
: D \ is continuous since if G is open in \, then \ G is compact,
so )
_
\ G
_
= D ) (G) is compact and hence closed, so ) (G) =
_
)
1
_
1
(G) is
open.
2. PROOF OF CARATHODORYS THEOREM 81
2.1. Simple boundary points and Jordan curves. Now it is obvious that
(1) implies (2) in Carathodorys Theorem 20, and so we turn our attention to the
remaining implication, namely that if 0\ is a Jordan curve, then every boundary
point of \ is a simple boundary point. Our proof of this fact is largely topological,
in contrast to the converse implication that we proved using the full force of the
Riemann mapping theorem and its boundary behaviour.
So we suppose that 0\ = ,
+
where , : T C is continuous and one-to-
one. Since \ is bounded and connected, we must have \ = E
o
where E
o
is the
bounded component of C,
+
= C0\. Indeed, x .
0
\ and suppose that :
[0, 1[ C0\ is a simple taxicab path joining .
0
to .. If . \
c
0\, and
t
1
= inf t [0, 1[ : (t) \
c
0\, then (t
1
) 0\, a contradiction. So . \.
It now follows that \ is a connected component of C,
+
, and hence must be the
bounded component E
o
. Now it follows from Proposition 18 in the appendix below
that every boundary point of 0\ is simple.
APPENDIX A
Topology
We collect here some background material requiring contributions from topol-
ogy.
1. Homotopy and index
Here we prove Proposition 9 that says the index is unchanged by homotopy.
We begin with a short lemma that gives a condition on two paths under which one
of them cannot wrap around a point more often than the other. The reader who
has walked a dog in a park will recognize this condition as shortening the leash near
a pole or tree just enough to prevent the dog from winding around it.
Lemma 7. (Dog leash lemma) Suppose that
0
: T C and
1
: T C are
closed paths in the complex plane, and that a is a complex number such that
[
1
()
0
()[ < [a
0
()[ , T.
Then 1:d
~
0
(a) = 1:d
~
1
(a).
Proof : Let T = [0, 2[ with 0 and 2 identied, and dene
(t) =

1
(t) a

0
(t) a
, 0 _ t _ 2.
Then
[ (t) 1[ =

1
(t) a

0
(t) a
1

1
(t)
0
(t)

0
(t) a

< 1
implies that
+
1(1, 1) and so 1:d
~
(0) = 0 by Cauchys theorem. Thus we have
0 = 1:d
~
(0) =
1
2i
_
~
1
. 0
d. =
1
2i
_
2t
0
1
(t)

t
(t) dt
=
1
2i
_
2t
0

0
(t) a

1
(t) a
(
0
(t) a)
t
1
(t) (
1
(t) a)
t
0
(t)
(
0
(t) a)
2
dt
=
1
2i
_
2t
0

t
1
(t)

1
(t) a
dt
1
2i
_
2t
0

t
0
(t)

0
(t) a
dt
= 1:d
~
1
(a) 1:d
~
0
(a) .
Now we recall Proposition 9.
Proposition 16. Let
0
and
1
be two closed paths in an open set \ of the
complex plane. If
0
and
1
are \-homotopic, then 1:d
~
0
(a) = 1:d
~
1
(a) for all
a C \.
83
84 A. TOPOLOGY
Proof : There is a continuous map I : T [0, 1[ \ such that I(, 0) =
0
and I(, 1) =
1
. Choose
(1.1) 0 < - <
1
4
di:t (a, \) .
By uniform continuity of I on the compact set T [0, 1[ there is c 0 such that
[I(
1
, t
1
) I(
2
, t
2
)[ < - whenever [(
1
, t
1
) (
2
, t
2
)[ < c.
If we now choose points 0 = t
1
< t
2
... < t

= 1 with
t
|
= t
|
t
|1
< c,
and if we dene closed curves ,
|
: T \ by
,
|
() = I(, t
|
) , 1 _ / _ ,
then ,
1
is
0
and ,

is
1
; moreover,
(1.2)

,
|+1
() ,
|
()

= [I(, t
|+1
) I(, t
|
)[ < -,
since [(, t
|+1
) (, t
|
)[ = [t
|+1
t
|
[ < c. Using (1.1) we then have
(1.3)

,
|+1
() ,
|
()

<
1
4
di:t (a, \) <
1
4
[a ,
|
()[ .
If it were the case that the curves ,
|
were actually paths, then Lemma 7 and
inequality (1.3) (even without the factor
1
4
) would yield
1:d
o
!+1
(a) = 1:d
o
!
(a) , 1 _ / < ,
which would prove that
1:d
~
1
(a) = 1:d
o
1
(a) = 1:d
o
11
(a) = ... = 1:d
o
1
(a) = 1:d
~
0
(a) .
Of course there is no reason to assume that the ,
|
are paths, so we must make
a further approximation. Write T = [0, 2[ with 0 and 2 identied, and choose
0 = 0
0
< 0
1
< ... < 0
1
= 2
so that
0
|
= 0
|
0
|1
< c.
For 1 _ / _ we dene c
|
: T \ to be the polygonal path obtained from ,
|
by joining each point ,
|
(0
|1
) to ,
|
(0
|
) by a straight line segment with the usual
parameterization. Then for 0 [0
|1
, 0
|
[ and 1 _ / _ we have
[c
|
(0) ,
|
(0)[ (1.4)
=

0
|
0
0
|
,
|
(0
|1
)
0 0
|1
0
|
,
|
(0
|
) ,
|
(0)

_
0
|
0
0
|
[,
|
(0
|1
) ,
|
(0)[
0 0
|1
0
|
[,
|
(0
|
) ,
|
(0)[
=
0
|
0
0
|
[I(0
|1
, t
|
) I(0, t
|
)[
0 0
|1
0
|
[I(0
|
, t
|
) I(0, t
|
)[
<
0
|
0
0
|
-
0 0
|1
0
|
- = -.
2. THE JORDAN CURVE THEOREM 85
Altogether, from (1.2) and (1.4) we obtain for 1 _ / < ,
[c
|+1
(0) c
|
(0)[ _

c
|+1
(0) ,
|+1
(0)

,
|+1
(0) ,
|
(0)

[,
|
(0) c
|
(0)[
< - - - = 8- < di:t (a, \) -
< [a ,
|
(0)[ - _ [a c
|
(0)[ [c
|
(0) ,
|
(0)[ -
_ [a c
|
(0)[ .
Thus Lemma 7 applies to show that
1:d
o
!+1
(a) = 1:d
o
!
(a) , 1 _ / < ,
as well as
1:d
o1
(a) = 1:d
o
1
(a) and 1:d
o
1
(a) = 1:d
o
1
(a) .
We conclude that
1:d
~
1
(a) = 1:d
o
1
(a)
= 1:d
o
1
(a) = 1:d
o
11
(a) = ... = 1:d
o1
(a)
= 1:d
o
1
(a) = 1:d
~
0
(a) .
2. The Jordan Curve Theorem
If , : T C is a simple closed curve, by which we mean that , is continuous
and one-to-one, the Jordan Curve Theorem says that the complement C ,
+
of
the image ,
+
has exactly two connected components, one unbounded and the other
bounded and simply connected.
Theorem 25. Suppose , : T C is continuous and one-to-one. Then C,
+
=
| ' E where
(1) | is unbounded and connected,
(2) E is bounded, connected and simply connected,
(3) | E = c.
We rst establish this theorem for closed taxicab paths, where we follow the
arguments for paths in [6], but with some simplications permitted by the restric-
tion to taxicab paths. The reader is referred to Maehara [3] for a dierent proof
using the Brouwer xed point theorem in the plane (for which see e.g. page 31 in
[2]). However, our proof also yields that every point on a Jordan curve is a simple
boundary point of each connected component of the complement. See Proposition
18 below.
Proposition 17. Theorem 25 holds if in addition , is assumed to be a taxicab
path, i.e. ,
+
is a nite concatenation of line segments parallel to either the real axis
or the imaginary axis. In this case 1:d
o
is either 1 or 1 throughout the bounded
component E.
Proof : Let o
1
, o
2
, ..., o
n
be a list of the distinct line segments in ,
+
where
o
|
= [a
|
, /
|
[ is the closed line segment joining a
|
to /
|
, and suppose that
o
|
o
|+1
= /
|
= a
|+1
, 1 _ / _ :,
where we dene o
n+1
= o
1
. We proceed in three steps.
Step 1: C ,
+
is not connected.
Let a be the midpoint of the segment o
1
and choose r 0 so small that
1(a, r)

n
|=2
o
|
= c. Now
86 A. TOPOLOGY
01(a, r) o
1
consists of two points / and c,
1(a, r) o
1
consists of an open line segment 1 endpoints / and c,
01(a, r) o
1
consists of two open semicircles C
1
and C
2
each of whose
endpoints are / and c.
For , = 1, 2 dene the path ,

to be the path , but with the open line segment

1 replaced by the open semicircle C

. Then ,

is still a closed path since the

endpoints of 1 and C

coincide (the direction on C

is chosen to match that on 1).

We then have
(2.1) 1:d
o
1
(a) 1:d
o
2
(a) = 1:d
J1(o,:)
(a) = 1,
where the sign is determined by the direction on 1.
It now follows from Cauchys theorem that 1:d
o
has the same value as 1:d
o
1
on the semicircle C
2
, and that 1:d
o
has the same value as 1:d
o
2
on the semicircle
C
1
. Since these two values dier by exactly 1 or 1, it now follows from Proposition
4 that
(2.2) the semicircles C
1
and C
2
lie in dierent components of C ,
+
,
and thus that there are at least two connected components in C ,
+
.
Step 2: C ,
+
has exactly two connected components.
Let
c = mindi:t (o
I
, o

) : o
I
and o

are not consecutive segments in ,

+
.
Since , is continuous and one-to-one, we have c 0. Now take 0 < t <
o
4
and
consider the set
1
|
= . C : d
|orI
(., ,
+
) = t ,
where d
|orI
(., n) = max [.
1
n
1
[ , [.
2
n
2
[ is the taxicab distance between .
and n.
To x notation let us parameterize , by arc length : (this is particularly easy
for a taxicab path). Starting at the point a in Step 1, we (where we identify with
the point , (:) on the path ,) begin travelling along o
1
in the positive direction
(as determined by the map ,) and note that, at least initially since t <
o
4
and
c _ |c:qt/(o
1
), there are exactly two points in 1
|
that are at (Euclidean) distance
t from , (:), one to the "right" of , (:), call it ,
|
:I||
(:), and one to the "left"
of , (:), call it ,
|
lt}|
(:). As we approach to within distance t of the segment o
2
,
which we may assume veers to the right from the endpoint /
1
of the segment o
1
,
we halt the point ,
|
:I||
(:), keeping it constant as we go around the right angle
at /
1
traversing a distance t along o
1
to /
1
followed by a distance t from /
1
along
o
2
. As for the point ,
|
lt}|
(:), we continue on until , (:) reaches /
1
, and then we
pause momentarily to let ,
|
lt}|
continue on for a distance t and then turn right for
a distance t, allowing it to "catch up" to ,
|
lt}|
(:) as , (:) begins traversing o
2
. A
picture makes this quite transparent.
In this way we continue to construct ,
|
:I||
and ,
|
lt}|
until , (:) returns to the
point , (a) on the curve , at which we started. By construction both ,
|
lt}|
and
,
|
:I||
are taxicab paths lying in the set 1
|
. If we parameterize each of these paths
by arc length, and denote by 1
lt}|
and 1
:I||
their respective lengths, then we have
,
|
lt}|
: [0, 1
lt}|
[ 1
|
C ,
+
,
,
|
:I||
: [0, 1
:I||
[ 1
|
C ,
+
.
2. THE JORDAN CURVE THEOREM 87
It is clear that
(2.3)
_
,
|
lt}|
_
+
'
_
,
|
:I||
_
+
= 1
|
,
and furthermore that only the rst of the two possibilities listed below can occur
when the paths ,
|
lt}|
and ,
|
:I||
return:
Either: (railway track matchup) ,
|
lt}|
(1
lt}|
) = ,
|
lt}|
(0) and ,
|
:I||
(1
:I||
) =
,
|
:I||
(0),
Or: (Mbius band matchup) ,
|
lt}|
(1
lt}|
) = ,
|
:I||
(0) and ,
|
:I||
(1
:I||
) =
,
|
lt}|
(0).
The reason the rst possibility occurs is that the path ,
|
:I||
always stays to
the "right" of , (:) as : moves in the positive direction. Alternatively, the Mbius
band matchup cannot occur in the plane since otherwise it would follow that the
two semicircles C
1
and C
2
in Step 1 (with r = t) would lie in the same connected
component of C ,
+
, contradicting (2.2).
Now we claim that C,
+
is the union of the component containing the semicircle
C
1
(in Step 1) and the component containing the semicircle C
2
(in Step 1). Indeed,
if . C,
+
, pick a in one of the segments comprising ,
+
such that the line segment
[., a[ is neither horizontal nor vertical. Let n (., a[ ,
+
be the rst point in ,
+
encountered by [., a[ as it travels from . to a. It is now clear that [., n[ must
intersect 1
|
for a suciently small positive t. Indeed, if we zoom in at a, we just
need to know that two non-parallel lines must intersect. This is one of Euclids
axioms. It is interesting to note that in the proof of the Jordan Curve Theorem
in [3], the Brouwer xed point theorem is used to prove the analogue of Euclids
axiom with curves in place of lines, a much more dicult task.
Now let [., n) 1
|
be the rst time the line segment [., n) intersects 1
|
.
By (2.3) and the railway track matchup, we must have either
_
,
|
lt}|
_
+
or

_
,
|
:I||
_
+
. If we take r = t in Step 1, then . can be connected to either C
1
or C
2
by a path that lies entirely in C ,
+
, namely [., [ followed by a portion of
either ,
|
lt}|
or ,
|
:I||
. In particular this shows that there are exactly two connected
components in C ,
+
, and completes the proof of Step 2.
Now if 1(0, 1) is a large disk containing ,
+
, then the connected set 1(0, 1)
c
is contained in one of the two components of C ,
+
, namely the unbounded one
|. The other connected component E of C ,
+
is contained in 1(0, 1), hence
is bounded. Moreover, the argument in Step 1 shows that the value of 1:d
o
on
opposite sides of ,
+
diers by 1. Thus 1:d
o
takes either the value 1 or 1 on the
bounded component since 1:d
o
= 0 on the unbounded component by Proposition
4. At this point the proof of Proposition 8 is complete, and so both the Riemann
Mapping Theorem 15 and its Porism 3 are at our disposal. We will use these
observations in the proof of the next step.
Step 3: The bounded component E of C ,
+
is simply connected.
We will prove this using Porism 3 to the Riemann Mapping Theorem. Note
that there is no circularity here since our proof of the Riemann Mapping Theorem
and Porism 3 used only Propositions 9 and 8, the rst of which was proved in
the previous section, and the second of which is proved by Steps 1 and 2 above.
In fact, once we show that E satises the hypotheses of Porism 3, the Riemann
88 A. TOPOLOGY
map from D to E shows that E is homeomorphic to the unit disk D, thus simply
connected. In order to verify the hypotheses of Porism 3, it remains to show that
every nonvanishing ) H (E) has a holomorphic square root in E, and this in turn
will follow if we show that every ) H (E) has an antiderivative in H (E). We now
apply Porism 2. For this we need only show that
(2.4)
_
~
) (.) d. = 0, ) H (E) ,
for all simple closed taxicab paths in E.
Finally, to prove (2.4), we rst note that it follows easily using the proof of
Step 1 that
(2.5) C = E ' ,
+
' | = E
~
'
+
' |
~
,
where E
~
and |
~
are the bounded and unbounded components respectively of C
+
.
Now
+
E implies
+
| = c. Since | is connected and | |
~
,= c (they both
contain points near innity), it follows that | |
~
. This together with
+
,
+
= c
and (2.5) show that E
~
E. We can now follow an argument already used in the
proof of Theorem 16. Indeed, we write
_
~
) (.) d. =

_
J1
1

) (.) d.
where 1
I

is a rectangle contained inside (hence in E), 01

I

has the same orien-

tation as , and the sum is nite for each i. For this we simply construct a grid
of innite lines in the plane, each passing through one of the segments in . This
creates a collection of minimal rectangles with sides that are segments of these lines.
Then the inside E
~
of is the union of all the minimal rectangles 1
I

that happen
to lie inside . Finally we know that
_
J1
1

) (.) d. = 0 by Cauchys theorem for a

rectangle in a convex subset of E, and summing over i and , proves (2.4). This
completes the proof of Proposition 17.
2.1. Proof of the Jordan Curve Theorem. We now turn our attention to
the proof of Theorem 25, which we prove in a series of four lemmas. We begin with
an approximation lemma which makes the connection with taxicab paths.
Lemma 8. Given a simple closed curve , in the plane and - 0, there is a
simple closed taxicab path c such that
|, c|
o
= sup
T
[, () c()[ < -.
Proof : We dene upper and lower moduli of continuity for , : T C by
.
+
(c) = sup
]q]o
[, () , (j)[ , (2.6)
.

(c) = inf
]q]o
[, () , (j)[ ,
where , j range over the unit circle T. Since , is continuous and one-to-one we
have
0 < .

(c) _ .
+
(c) 0 as c 0.
Now choose c 0 such that
(2.7) .
+
(c) <
-
4
,
2. THE JORDAN CURVE THEOREM 89
and then choose a closed, but not necessarily simple, taxicab path such that
|, |
o
<
.

(c)
2
.
This is easily accomplished by rst choosing a polygonal approximation with suf-
ciently small edges, and then replacing each edge with a vertical and horizontal
segment.
We now use to construct a simple closed taxicab path c satisfying
(2.8) |, c|
o
_ 4.
+
(c) ,
which by (2.7) proves Lemma 8. First we modify the path by halting it for short
periods of time according to the following algorithm. We identify T with [0, 2)
under = c
I0
0, and set j (0
1
, 0
2
) =

c
I01
c
I02

. Let 0
1
_ 0 be the rst time
(0
1
) intersects ((0
1
, 0
1
[). Let t
1
(0
1
, 0
1
[ be the last time for which
(t
1
) = (0
1
). Then we have
[, (t
1
) , (0
1
)[ _ [, (t
1
) (t
1
)[ [ (0
1
) , (0
1
)[ < .

(c) ,
and it follows that j (t
1
, 0
1
) < c. Now we begin dening a modication of by
halting at (0
1
) for the interval [0
1
, t
1
[, i.e.
(t) =
_
(t) for 0 _ t _ 0
1
(t
1
) for 0
1
_ t _ t
1
.
Now let 0
2
_ t
1
be the rst time (0
2
) intersects ((0
2
, 0
2
[), and let t
2

(0
2
, 0
2
[ be the last time for which (t
2
) = (0
2
). It is easy to see using c <
t
2
that in fact 0
2
t
1
. Moreover we have just as above that
j (t
2
, 0
2
) < c.
Then we continue by again halting at 0
2
:
(t) =
_

_
(t) for 0 _ t _ 0
1
(0
1
) for 0
1
_ t _ t
1
(t) for t
1
_ t _ 0
2
(0
2
) for 0
2
_ t _ t
2
.
We proceed in this way for nitely many steps until has been dened for all
t [0, 2). Now is a closed taxicab path that has no self-intersections apart from
those arising from the nite number of intervals of constancy [0

, t

[ constructed in
the above algorithm. A crucial observation is that for t [0

, t

[ we have j (t, 0

) _
j (t

, 0

) < c and so
[ (t) , (t)[ = [ (0

) , (t)[ _ [ (0

) , (0

)[ [, (0

) , (t)[
< .

(c) .
+
(c) _ 2.
+
(c) .
Thus we have |, |
o
_ 2.
+
(c), and it is now a simple matter to modify the
parameterization of near the intervals of constancy so as to produce a simple
closed taxicab path c that satises (2.8). This completes the proof of Lemma 8.
The complement C ,
+
of the image of the curve , is a pairwise disjoint union
of connected open sets called components.
Lemma 9. Let , be a simple closed curve in the plane. Then C,
+
has exactly
one unbounded component and at least one bounded component.
90 A. TOPOLOGY
Proof : Clearly there is exactly one unbounded component |
o
, the one contain-
ing the complement of 1(0, 1) for any 1 chosen large enough that ,
+
1(0, 1).
We now show using Lemma 8, Proposition 17 and an elementary argument in
Maehara [3] that there is at least one bounded component. Indeed, we recall the
situation depicted in Figure 1 on page 643 of [3]. The square Q = [1, 1[ [2, 2[
contains ,
+
and ,
+
0Q consists of just two points (1, 0) , (1, 0) which we label
\ = (1, 0) and 1 = (1, 0). We also label = (0, 2) and o = (0, 2). The curve
, is divided into two closed arcs by the points \ and 1, and we label the upper
and lower arcs ,
no:||
and ,
sou||
. The upper arc ,
no:||
is determined by passing
through the highest point of the intersection of the vertical line segment

o joining
to o (such an intersection point exists by the connectedness of ,
+
). Denote by
T
no:||
and 1
no:||
the top and bottom points in ,
+
no:||

o.
We claim that ,
+
sou||
must intersect the segment

1
no:||
o. If not, let
^
T
no:||
1
no:||
denote the arc of ,
no:||
that joins T
no:||
to 1
no:||
and note that the simple closed
curve
o =

T
no:||

^
T
no:||
1
no:||

1
no:||
o

o (1, 2)

(1, 2) (1, 2)

(1, 2)
is disjoint from the compact set ,
+
sou||
provided we modify the segment

(1, 2) (1, 2)
to jut to the left around \ = (1, 0). Using Lemma 8, we can replace the curve
o with a nearby simple closed taxicab path c whose image c
+
is still disjoint from
,
+
sou||
. Moreover, for |o c|
o
small enough, there are portions of ,
+
sou||
in both
the bounded and unbounded components of C c
+
(namely portions near \ and
portions near 1 respectively), contradicting the connectedness of ,
+
sou||
. This
proves our claim. Now we denote by T
sou||
and 1
sou||
the top and bottom points
in ,
+
sou||

1
no:||
o. Let 7
0
be the midpoint of

1
no:||
T
sou||
. Thus the labeled
points running down the vertical segment

o are given in order by
, T
no:||
, 1
no:||
, 7
0
, T
sou||
, 1
sou||
, o.
Following [3] we claim that the component of C ,
+
containing 7
0
is bounded.
If not, then there is a simple taxicab path j in C ,
+
joining 7
0
to (2, 0). Let 7
1
be the rst point where j meets 0Q, and let i be the arc of j joining 7
0
to 7
1
.
Suppose that 7
1
lies in the bottom half of 0Q (otherwise we mirror the argument
given here). Now consider the closed curve
t =

T
no:||

^
T
no:||
1
no:||

1
no:||
7
0
i c,
where c is the simple path in 0Q that starts at 7
1
and proceeds along 0Q to o
without passing through \ or 1, and then continues on along 0Q through either
\ or 1 (but not both) before ending at . Once again we modify c so as to jut
around either \ or 1, whichever of these it initially passed through. The image t
+
of the curve t doesnt intersect ,
+
sou||
and we can again replace t by a simple closed
taxicab path c whose image doesnt intersect ,
+
sou||
and yet contains portions of
,
+
sou||
in both the bounded and unbounded components of C c
+
, contradicting
the connectedness of ,
+
sou||
. This completes the proof of Lemma 9.
Lemma 10. Suppose that , is a simple closed curve in the plane. Then every
bounded component of C ,
+
is simply connected.
Proof : If is a closed path in a bounded component E of C ,
+
then we can
use Lemma 8 to nd a simple closed taxicab path c such that is contained in the
2. THE JORDAN CURVE THEOREM 91
bounded component E
o
of C c
+
. By Proposition 17 we have that E
o
is simply
connected, so that is E
o
-homotopic to a point in E
o
. Since E
o
E we also have
that is E-homotopic to a point in E. This proves that E is simply connected and
completes the proof of Lemma 10.
Thus far we have shown that
(1) C ,
+
has one unbounded component |
o
,
(2) C ,
+
has at least one bounded component,
(3) every bounded component of C ,
+
is simply connected.
It remains only to show that there is exactly one bounded component in C,
+
.
Lemma 11. Suppose that , is a simple closed curve in the plane. Then C
(,
+
' |
o
) is connected.
Suppose that two disks 1(.
1
, r) and 1(.
2
, r) are contained in C (,
+
' |
o
)
for some r 0. It suces to show there is a path in C (,
+
' |
o
) joining .
1
to .
2
.
With moduli of continuity .
+
and .

as in (2.6), choose c 0 such that

(2.9) .
+
(c) <
r
2
,
and then choose a simple closed taxicab path so that both
(2.10) |, |
o
<
.

(c)
1000
and 1
_
.

,
8r
4
_
E
~
, , = 1, 2,
where E
~
is the bounded component of C
+
. Note that 1:d
~
(.

) = 1 for , = 1, 2
(if 1:d
~
(.

) = 1 we can reverse the direction of ). Now consider a very ne

grid of horizontal and vertical lines whose consecutive distances apart lie in a small
interval [j, 4j[ with j 0. With j small enough we can arrange to have all the
segments in lie in grid lines, and furthermore we can assume that j <
.(o)
1000
. Let
be the collection of all minimal rectangles with edges in the grid lines. Dene
1 =
_
_
1 : 1 E
~
and di:t (1, ) <
.

(c)
100
_
.
Then 1 is a nite union of rectangles 1 from, and 01 consists of nitely many
simple closed taxicab paths, one of which is . Now it is easy to see that exactly
one of the remaining taxicab paths, say j, includes .
1
in its bounded component E

.
We claim that .
2
is included in E

as well. To see this it suces by Proposition 9

to show that the paths and j are ^-homotopic to each other where ^ = C.
2
.
Indeed, we would then have
(2.11) 1:d

(.
2
) = 1:d
~
(.
2
) = 1,
and it would follow from Proposition 8 that .
2
lies in the bounded component E

of j as well.
So pick a sequence 1
n

n=1
of points on j that traverse j in the positive
direction and that satisfy
.

(c)
10
< [1
n
1
n+1
[ <
.

(c)

, 1 _ : _ ,
where 1
+1
= 1
1
. Then pick a sequence Q
n

n=1
of points on satisfying
(2.12) [1
n
Q
n
[ <
.

(c)
100
.
92 A. TOPOLOGY
Now we note that
[Q
n
Q
n+1
[ _ [1
n
1
n+1
[ [1
n
Q
n
[ [1
n+1
Q
n+1
[
.

(c)
20
,
and thus that is divided into two taxicab arcs by Q
n
and Q
n+1
. One of these
taxicab arcs joining Q
n
and Q
n+1
must have diameter at least
:
2
.
+
(c) in order
that 1:d
~
(.
1
) = 1 (use that every half-ray from .
1
must intersect
+
), and it then
follows from (2.6) and
[Q
n
Q
n+1
[ _ [Q
n
1
n
[ [1
n
1
n+1
[ [1
n+1
Q
n+1
[ < .

(c)
that the other taxicab arc joining Q
n
and Q
n+1
must have diameter at most
.
+
(c) <
:
2
. We denote this latter taxicab arc of by
^
Q
n
Q
n+1
. It also follows
that the circular arcs ,
1
_
^
Q
n
Q
n+1
_
have diameter at most c, have pairwise dis-
joint interiors since 1:d

(.
1
) = 1, and have union equal to T. Thus we have
(2.13) =
^
Q
1
Q
2
...
^
Q
1
Q

^
Q

Q
1
.
We will show that and j are -homotopic by exhibiting intermediate paths
which are successively -homotopic to one another by elementary homotopies such
as "growing a nger" and "shrinking of a closed curve to a point by dilation". First
we claim that is -homotopic to the path t
1
constructed as follows:
Start at 1
2
and proceed along the segment

1
2
1
1
(which will not normally
lie on the path j), then along the segment

1
1
Q
1
, then along the arc
^
Q
1
Q
2
,
then along the segment

Q
2
1
2
. This results in a closed path
o
2
=

1
2
1
1

1
1
Q
1

^
Q
1
Q
2

Q
2
1
2
through the point 1
2
that is contained in the disk 1
_
1
2
,
2:
3
_
since dia:
_
^
Q
n
Q
n+1
_
<
:
2
. Then proceed from 1
2
to 1
3
along the segment

1
2
1
3
.
Recursively, for : _ 8, start at 1
n
and follow the algorithm described in
the rst bullet item. This results in a closed path
o
n
=

1
n
1
n1

1
n1
Q
n1

^
Q
n1
Q
n

Q
n
1
n
through the point 1
n
that is contained in the disk 1
_
1
n
,
2:
3
_
, which is
then followed by the segment

1
n
1
n+1
.
In this way we construct the path
t
1
= o
2

1
2
1
3
o
3

1
3
1
4
... o

1
1
o
1

1
1
1
2
,
which is clearly -homotopic to
^
Q
1
Q
2
...
^
Q
1
Q

^
Q

Q
1
,
since

Q
n1
1
n1

1
n1
1
n

1
n
1
n1

1
n1
Q
n1
is a "nger" disjoint from .
2
.
But this latter path is by (2.13). We next claim that t
1
is -homotopic to the
path t
2
given by
t
2
=

1
1
1
2

1
2
1
3
...

1
1
1

1
1
.
Indeed, simply contract by dilation the closed path o
n
to the point 1
n
within the
disk 1
_
1
n
,
2:
3
_
. This clearly avoids the point .
2
since 1
n
lies on j and 1
_
.
2
,
3:
4
_
is disjoint from j by (2.10). Finally, it is an easy exercise to show that the taxicab
2. THE JORDAN CURVE THEOREM 93
path j is also homotopic to the path t
2
(for example, one can argue as above with
j in place of ). Thus Proposition 9 shows that (2.11) holds, and then Proposition
8 shows that .
2
E

.
Thus both .
1
and .
2
lie in E

, which is simply connected by Proposition 17, and

it follows that there is a path connecting .
1
to .
2
in E

. However, by construction
and the denition of 1, the path j is at a distance at least
.(o)
200
from , while
itself is at a distance at most
.(o)
1000
from ,. It follows that j is at a distance at
least
.(o)
300
from ,, and so
(2.14) E

C (,
+
' |
o
) = E
o
.
This completes the proof of Lemma 11.
The proof of the Jordan Curve Theorem 25 follows immediately from the four
lemmas above.
2.2. Simple boundary points. We now prove that every point n on ,
+
is
not only a boundary point of the bounded component E
o
of C ,
+
, but is actually
a simple boundary point of E
o
. We recall from Denition 11 that n is a simple
boundary point of \ if for every sequence .
n

o
n=1
\ with lim
no
.
n
= n, there
is a (continuous) curve I : [0, 1) \ and an increasing sequence of parameter
points t
n

o
n=1
[0, 1) with lim
no
t
n
= 1 such that I(t
n
) = .
n
for all : _ 1 and
lim
|1
I(t) = n.
Proposition 18. Suppose , : T C is continuous and one-to-one. If n ,
+
,
then n is a simple boundary point of the bounded component E
o
of C ,
+
.
Proof : We refer to the proof of Lemma 11 above. Consider points .
1
and .
2
as in the proof there. Select points
1
and
2
in ,
+
with
[.
I

I
[ = di:t (.
I
, ,
+
) , i = 1, 2,
and then select points 1
1
and 1
2
in j
+
with
1
I
1(.
I
, di:t (.
I
, ,
+
)) 1(
I
, .

(c)) , i = 1, 2.
Let j 0 satisfy
[
1

2
[ < .

(j) .
Then one of the two arcs of , obtained by removing
1
and
2
from ,
+
must
have diameter at most .
+
(j). It follows that one of the two arcs of j obtained by
removing 1
1
and 1
2
from j
+
, call it t, must have diameter at most
.
+
(j) .

(c) .
Now let o be the path obtained by joining .
1
to 1
1
with a line segment, then joining
1
1
to 1
2
with t, and nally joining 1
2
to .
2
with a line segment. We then have
the estimate
dia:(o
+
) < di:t (.
1
, ,
+
) .
+
(j) .

(c) di:t (.
2
, ,
+
) .
Claim 3. Given - 0, there exists ` 0 such that whenever .
1
, .
2
E
o
with
[.
1
.
2
[ < `, there is a path o in E
o
that joins .
1
to .
2
and has
dia:(o
+
) < -.
94 A. TOPOLOGY
To see this let - 0, and choose j 0 so that .
+
(j) <
:
3
. Then if .
1
and .
2
satisfy
(2.15) di:t (.
1
, ,
+
) [.
1
.
2
[ di:t (.
2
, ,
+
) < .

(j) ,
it follows that
[
1

2
[ < .

(j) ,
and we showed above that there is in this case a path o in E
o
that joins .
1
to .
2
and satises
dia:(o
+
) < di:t (.
1
, ,
+
) .
+
(j) .

(c) di:t (.
2
, ,
+
)
< 2.

(j) .
+
(j) _ 8.
+
(j) < -,
since
.

(c) _ .
+
(c) _
r
2
< di:t (.
1
, ,
+
) < .

(j) _ .
+
(j) .
On the other hand, if [.
1
.
2
[ <
1
4
.

(j) and (2.15) fails, it follows that

both di:t (.
1
, ,
+
) and di:t (.
2
, ,
+
) exceed
1
4
.

(j). Indeed, di:t (.

1
, ,
+
) _
1
4
.

(j)
implies
di:t (.
2
, ,
+
) _ [.
2
.
1
[ di:t (.
1
, ,
+
) _
1
2
.

(j) ,
which implies that (2.15) holds. It now follows that
1(.
1
, di:t (.
1
, ,
+
)) 1(.
2
, di:t (.
2
, ,
+
)) ,= c,
and so the line segment joining .
1
to .
2
lies in E
o
and has length
1
4
.

(j) <
1
4
.
+
(j) <
:
12
. Altogether this shows that the claim holds if we take ` =
1
4
.

(j).
It is now an easy matter to show that n 0E
o
,
+
is a simple boundary point
of E
o
. Let .
n

o
n=1
E
o
have limit n. By the above claim there is a decreasing
sequence `
|

o
|=1
of positive numbers with limit 0 as / such that if
(2.16) [.
n
.
n
[ < `
|
,
there is a path o
n,n
joining .
n
to .
n
in E
o
with
dia:
_
o
+
n,n
_
< 2
|
.
Denote by ] o
n,n
= o
n,n
o
n,n
the closed path that runs from .
n
to .
n
and back
to .
n
. Fix a subsequence
|

o
|=1
of positive integers satisfying
sup
n
!
[.

!
.
n
[ < `
|
, / _ 1.
Now let
|
: [0, 1) E
o
be the curve joining .

!
to .

!+1
obtained by con-
catening in order the following paths:

|
= ^ o

!
,
!
+1
^ o

!
,
!
+2
..., ^ o

!
,
!+1
1
, o

!
,
!+1
.
Note that
|
does indeed pass through the points .

!
, .

!
+1
, ....

!+1
in order (we
can ignore the fact that it also returns to .

!
and possibly other of the points
repeatedly). Now we let : [0, 1) E
o
be the curve obtained by concatening in
order the paths
|
:
=
1

2
... =
o

|=1

|
.
2. THE JORDAN CURVE THEOREM 95
By construction there is an increasing sequence t
n

o
n=1
in [0, 1) such that t
n
is a
parameter point for
|
if
|
_ : <
|+1
, and such that o (t
n
) = .
n
. Since
dia:( ([t
n
, 1))) _
o

|=|
2
|
= 2
1|
,
if : _
|
, we conclude that
lim
|1
o (t) = lim
no
o (t
n
) = lim
no
.
n
= n.
Finally, it remains to show that 0E
o
= ,
+
. We continue to refer to the proof
of Lemma 11 above. By (2.10) there is a point Q on
+
such that [n Q[ <
.(o)
1000
.
By (2.13) the point Q lies on one of the arcs
^
Q
n
Q
n+1
, which by construction has
diameter at most .
+
(c). From (2.12) it follows that
[1
n
n[ _ [1
n
Q
n
[ [Q
n
Q[ [Qn[
<
.

(c)
100
.
+
(c)
.

(c)
1000
_ 2.
+
(c) .
Now j is a simple closed taxicab path and it is clear that 1
n
is a boundary point of
the bounded component E

of Cj
+
. Since E

E
o
by (2.14), it follows that there
is a point in E
o
within distance 2.
+
(c) of the point n. Since lim
o0
.
+
(c) = 0
and c can be chosen arbitrarily small in the proof of Lemma 11, we conclude that
n E
o
.
APPENDIX B
Lebesgue measure theory
Let ) : [0, 1) [0, ') be a nonnegative bounded function on the half open
unit interval [0, 1). In Riemanns theory of integration, we partition the domain
[0, 1) of the function into nitely many disjoint subintervals
[0, 1) =

_
n=1
[r
n1
, r
n
) ,
and denote the parition by T = 0 = r
0
< r
1
< ... < r

= 1 and the length of the

subinterval [r
n1
, r
n
) by r
n
= r
n
r
n1
0. Then we dene upper and lower
Riemann sums associated with the partition T by
l (); T) =

n=1
_
sup
[rr1,rr)
)
_
r
n
,
1(); T) =

n=1
_
inf
[rr1,rr)
)
_
r
n
.
Then we dene the upper and lower Riemann integrals of ) on [0, 1) by
| ()) = inf
1
l (); T) , /()) = sup
1
1(); T) .
Thus the upper Riemann integral | ()) is the "smallest" of all the upper sums, and
the lower Riemann integral is the "largest" of all the lower sums. By considering
the renement T
1
' T
2
of two partitions T
1
and T
2
it is easy to see that
l (); T
1
) _ l (); T
1
' T
2
) _ 1(); T
1
' T
2
) _ 1(); T
2
) .
Taking the inmum over T
1
and the supremum over T
2
shows that
| ()) _ /()) .
Finally we say that ) is Riemann integrable on [0, 1), written ) [0, 1), if | ()) =
/()), and we denote the common value by
_
1
0
) or
_
1
0
) (r) dr.
This denition is simple and easy to work with and applies in particular to
bounded continuous functions ) on [0, 1) since it is not too hard to prove that
) [0, 1) for such ). However, if we consider the vector space 1
2
7
([0, 1)) of
Riemann integrable functions ) [0, 1) endowed with the metric
d (), q) =
__
1
0
[) (r) q (r)[
2
dr
_
1
2
,
it turns out that while 1
2
7
([0, 1)) can indeed be proved a metric space, it fails to be
complete. This is a serious shortfall of Riemanns theory of integration, and is our
main motivation for considering the more complicated theory of Lebesgue below.
We note that the immediate reason for the lack of completeness of 1
2
7
([0, 1)) is the
97
98 B. LEBESGUE MEASURE THEORY
inability of Riemanns theory to handle general unbounded functions. However,
even locally there are problems. For example, once we have Lebesgues theory in
hand, we can construct a famous example of a Lebesgue measurable subset 1 of
[0, 1) with the (somewhat surprising) property that
0 < [1 (a, /)[ < / a, 0 _ a < / _ 1,
where [1[ denotes the Lebesgue measure of a measurable set 1 (see Problem 5
below). It follows that the characteristic function
J
is bounded and Lebesgue
measurable, but that there is no Riemann integrable function ) such that ) =

J
almost everywhere, since such an ) would satisfy | ()) = 1 and /()) = 0.
Nevertheless, by Lusins Theorem (see page 34 in [6] or page 55 in [5]) there is a
sequence of compactly supported continuous functions (hence Riemann integrable)
converging to
J
almost everywhere.
On the other hand, in Lebesgues theory of integration, we partition the range
[0, ') of the function into a homogeneous partition,
[0, ') =

_
n=1
_
(: 1)
'

, :
'

_
=

_
n=1
1
n
,
and we consider the associated upper and lower Lebesgue sums of ) on [0, 1) dened
by
l
+
(); T) =

n=1
_
:
'

)
1
(1
n
)

,
1
+
(); T) =

n=1
_
(: 1)
'

)
1
(1
n
)

,
where of course
)
1
(1
n
) =
_
r [0, 1) : ) (r) 1
n
=
_
(: 1)
'

, :
'

__
,
and [1[ denotes the "measure" or "length" of the subset 1 of [0, 1).
Here there will be no problem obtaining that l
+
(); T) 1
+
(); T) is small
provided we can make sense of

)
1
(1
n
)

. But this is precisely the diculty with

Lebesgues approach - we need to dene a notion of "measure" or "length" for
subsets 1 of [0, 1). That this is not going to be as easy as we might hope is
evidenced by the following negative result. Let T ([0, 1)) denote the power set of
[0, 1), i.e. the set of all subsets of [0, 1). For r [0, 1) and 1 T ([0, 1)) we dene
the translation 1 r of 1 by r to be the set in T ([0, 1)) dened by
1 r = 1 r (moo1)
= . [0, 1) : there is j 1 with j r . Z .
Theorem 26. There is no map j : T ([0, 1)) [0, ) satisfying the following
three properties:
(1) j([0, 1)) = 1,
(2) j
_

o
n=1
1
n
_
=

o
n=1
j(1
n
) whenever 1
n

o
n=1
is a pairwise disjoint
sequence of sets in T ([0, 1)),
(3) j(1 r) = j(1) for all 1 T ([0, 1)).
1. LEBESGUE MEASURE ON THE REAL LINE 99
Remark 8. All three of these properties are desirable for any notion of measure
or length of subsets of [0, 1). The theorem suggests then that we should not demand
that every subset of [0, 1) be "measurable". This will then restrict the functions )
that we can integrate to those for which )
1
([a, /)) is "measurable" for all <
a < / < .
Proof : Let r
n

o
n=1
= Q [0, 1) be an enumeration of the rational numbers
in [0, 1). Dene an equivalence relation on [0, 1) by declaring that r ~ j if r
j Q. Let / be the set of equivalence classes. Use the axiom of choice to
pick a representative a = from each equivalence class in /. Finally, let
1 = : / be the set consisting of these representatives a, one from each
equivalence class in /.
Then we have
[0, 1) =

_
o
n=1
1 r
n
.
Indeed, if r [0, 1), then r for some /, and thus r ~ a = , i.e.
r a r
n

o
n=1
. If r _ a then r a Q[0, 1) and r = a r
n
where a 1 and
r
n
r
n

o
n=1
. If r < a then r a 1 Q [0, 1) and r = a (r
n
1) where
a 1 and r
n
1 r
n

o
n=1
. Finally, if a r
n
= / r
n
, then a / = r
n
r
n
Q
which implies that a ~ / and then r
n
= r
n
.
Now by properties (1), (2) and (3) in succession we have
1 = j([0, 1)) = j
_

_
o
n=1
1 r
n
_
=
o

n=1
j(1 r
n
) =
o

n=1
j(1) ,
which is impossible since the innite series

o
n=1
j(1) is either if j(1) 0 or
0 if j(1) = 0.
1. Lebesgue measure on the real line
In order to dene a "measure" satisfying the three properties in Theorem 26,
we must restrict the domain of denition of the set functional j to a "suitable"
proper subset of the power set T ([0, 1)). A good notion of "suitable" is captured
by the following denition where we expand our quest for measure to the entire
real line.
Definition 12. A collection / T (R) of subsets of real numbers R is called
a o-algebra if the following properties are satised:
(1) c /,
(2)
c
/ whenever /,
(3)

o
n=1

n
/ whenever
n
/ for all :.
Here is the theorem asserting the existence of "Lebesgue measure" on the real
line.
Theorem 27. There is a o-algebra / T (R) and a function j : / [0, [
such that
(1) [a, /) / and j([a, /)) = / a for all < a < / < ,
(2)

o
n=1
1
n
/ and j
_

o
n=1
1
n
_
=

o
n=1
j(1
n
) whenever 1
n

o
n=1
is a
pairwise disjoint sequence of sets in /,
(3) 1 r / and j(1 r) = j(1) for all 1 /,
100 B. LEBESGUE MEASURE THEORY
(4) 1 / and j(1) = 0 whenever 1 1 and 1 / with j(1) = 0.
It turns out that both the o-algebra / and the function j are uniquely deter-
mined by these four properties, but we will only need the existence of such / and
j. The sets in the o-algebra / are called Lebesgue measurable sets.
A pair (/, j) satisfying only property (2) is called a measure space. Property
(1) says that the measure j is an extension of the usual length function on intervals.
Property (3) says that the measure is translation invariant, while property (4) says
that the measure is complete.
From property (2) and the fact that j is nonnegative, we easily obtain the
following elementary consequences (where membership in / is implied by context):
c / and j(c) = 0, (1.1)
1 / for every open set 1 in R,
j(1) = / a for any interval 1 with endpoints a and /,
j(1) = sup
n
j(1
n
) = lim
no
j(1
n
) if 1
n
1,
j(1) = inf
n
j(1
n
) = lim
no
j(1
n
) if 1
n
1 and j(1
1
) < .
For example, the fourth line follows from writing
1 = 1
1

'
_

_
o
n=1
1
n+1
(1
n
)
c
_
and then using property (2) of j.
To prove Theorem 27 we follow the treatment in [6] with simplications due to
the fact that the connected open subsets of the real numbers R are just the open
intervals (a, /). Dene for any 1 T (R), the outer Lebesgue measure j
+
(1) of 1
by,
j
+
(1) = inf
_
o

n=1
(/
n
a
n
) : 1

_
o
n=1
(a
n
, /
n
) and _ a
n
< /
n
_
_
.
It is immediate that j
+
is monotone,
j
+
(1) _ j
+
(1) if 1 1.
A little less obvious is countable subadditivity of j
+
.
Lemma 12. j
+
is countably subadditive:
j
+
_
o
_
n=1
1
n
_
_
o

n=1
j
+
(1
n
) , 1
n

o
n=1
T (R) .
Proof : Given 0 < - < 1, we have 1
n

o
|=1
(a
|,n
, /
|,n
) with
o

|=1
(/
|,n
a
|,n
) < j
+
(1
n
)
-
2
n
, : _ 1.
Now let
o
_
n=1
_

_
o
|=1
(a
|,n
, /
|,n
)
_
=

_
1

n=1
(c
n
, d
n
) ,
1. LEBESGUE MEASURE ON THE REAL LINE 101
where '
+
N ' . Then dene disjoint sets of indices
J
n
= (/, :) : (a
|,n
, /
|,n
) (c
n
, d
n
) .
In the case c
n
, d
n
R, we can choose by compactness a nite subset T
n
of J
n
such that
(1.2)
_
c
n

-
2
c
n
, d
n

-
2
c
n
_

o
_
(|,n)Jr
(a
|,n
, /
|,n
) ,
where c
n
= d
n
c
n
. Fix : and arrange the left endpoints a
|,n

(|,n)Jr
in
strictly increasing order a
I

1
I=1
and denote the corresponding right endpoints by
/
I
(if there is more than one interval (a
I
, /
I
) with the same left endpoint a
I
, discard
all but one of the largest of them). From (1.2) it now follows that a
I+1
(a
I
, /
I
) for
i < 1 since otherwise /
I
would be in the left side of (1.2), but not in the right side,
a contradiction. Thus a
I+1
a
I
_ /
I
a
I
for 1 _ i < 1 and we have the inequality
(1 -) c
n
=
_
d
n

-
2
c
n
_

_
c
n

-
2
c
n
_
_ /
1
a
1
= (/
1
a
1
)
11

I=1
(a
I+1
a
I
)
_
1

I=1
(/
I
a
I
) =

(|,n)Jr
(/
|,n
a
|,n
)
_

(|,n)1r
(/
|,n
a
|,n
) .
We also observe that a similar argument shows that

(|,n)1r
(/
|,n
a
|,n
) =
if c
n
= . Then we have
j
+
(1) _
o

n=1
c
n
_
1
1 -
o

n=1

(|,n)Jr
(/
|,n
a
|,n
)
_
1
1 -

|,n
(/
|,n
a
|,n
) =
1
1 -
o

n=1
o

|=1
(/
|,n
a
|,n
)
<
1
1 -
o

n=1
_
j
+
(1
n
)
-
2
n
_
=
1
1 -
o

n=1
j
+
(1
n
)
-
1 -
.
Let - 0 to obtain the countable subadditivity of j
+
.
Now dene the subset / of T (R) to consist of all subsets of the real line such
that for every - 0, there is an open set G satisfying
(1.3) j
+
(G ) < -.
Remark 9. Condition (1.3) says that can be well approximated from the
outside by open sets. The most dicult task we will face below in using this deni-
tion of / is to prove that such sets can also be well approximated from the inside
by closed sets.
Set
j() = j
+
() , /.
102 B. LEBESGUE MEASURE THEORY
Trivially, every open set and every interval is in /. We will use the following two
claims in the proof of Theorem 27.
Claim 4. If G is open and G =

n=1
(a
n
, /
n
) (where
+
N ' ) is the
decomposition of G into its connected components (a
n
, /
n
), then
j(G) = j
+
(G) =

n=1
(/
n
a
n
) .
We rst prove Claim 4 when
+
< . If G

o
n=1
(c
n
, d
n
), then for each
1 _ : _
+
, (a
n
, /
n
) (c
n
, d
n
) for some : since (a
n
, /
n
) is connected. If
J
n
= : : (a
n
, /
n
) (c
n
, d
n
) ,
it follows upon arranging the a
n
in increasing order that

n1r
(/
n
a
n
) _ d
n
c
n
,
since the intervals (a
n
, /
n
) are pairwise disjoint. We now conclude that
j
+
(G) = inf
_
o

n=1
(d
n
c
n
) : G

_
o
n=1
(c
n
, d
n
)
_
_
o

n=1

n1r
(/
n
a
n
) =

n=1
(/
n
a
n
) ,
and hence that j
+
(G) =

n=1
(/
n
a
n
) by denition since G

n=1
(a
n
, /
n
).
Finally, if
+
= , then from what we just proved and monotonicity, we have
j
+
(G) _ j
+
_

_

n=1
(a
n
, /
n
)
_
=

n=1
(/
n
a
n
)
for each 1 _ < . Taking the supremum over gives j
+
(G) _

o
n=1
(/
n
a
n
),
and then equality follows by denition since G

o
n=1
(a
n
, /
n
).
Claim 5. If and 1 are disjoint compact subsets of R, then
j
+
() j
+
(1) = j
+
(' 1) .
Since c = di:t (, 1) 0, we can nd open sets l and \ such that
l and 1 \ and l \ = c.
For example, l =

r.
1
_
r,
o
2
_
and \ =

r1
1
_
r,
o
2
_
work. Now suppose that
' 1 G =

_
o
n=1
(a
n
, /
n
) .
Then we have
l G =

_
1

|=1
(c
|
, )
|
) and 1 \ G =

_
J

|=1
(q
|
, /
|
) ,
1. LEBESGUE MEASURE ON THE REAL LINE 103
and then from Claim 4 and monotonicity of j
+
we obtain
j
+
() j
+
(1) _
1

|=1
()
|
c
|
)
J

|=1
(/
|
q
|
)
= j
+
_
_
_
_

_
1

|=1
(c
|
, )
|
)
_
_

'
_
_

_
J

|=1
(q
|
, /
|
)
_
_
_
_
_ j
+
(G) =
o

n=1
(/
n
a
n
) .
Taking the inmum over such G gives j
+
() j
+
(1) _ j
+
(' 1), and subaddi-
tivity of j
+
now proves equality.
Proof (of Theorem 27): We now prove that / is a o-algebra and that / and j
satisfy the four properties in the statement of Theorem 27. First we establish that
/ is a o-algebra in four steps.
Step 1: / if j
+
() = 0.
Given - 0, there is an open G with j
+
(G) < -. But then j
+
(G ) _
j
+
(G) < - by monontonicity.
Step 2:

o
n=1

n
/ whenever
n
/ for all :.
Given - 0, there is an open G
n

n
with j
+
(G
n

n
) <
:
2
r
. Then
=

o
n=1

n
is contained in the open set G =

o
n=1
G
n
, and since G is
contained in

o
n=1
(G
n

n
), monotonicity and subadditivity of j
+
yield
j
+
(G ) _ j
+
_
o
_
n=1
(G
n

n
)
_
_
o

n=1
j
+
(G
n

n
) <
o

n=1
-
2
n
= -.
Step 3: / if is closed.
Suppose rst that is compact, and let - 0. Then using Claim 4 there is
G =

n=1
(a
n
, /
n
) containing with
j
+
(G) =
o

n=1
(/
n
a
n
) _ j
+
() - < .
Now G is open and so G =

n=1
(c
n
, d
n
). We want to show that
j
+
(G ) _ -. Fix a nite ' _ '
+
and
0 < j <
1
2
min
1n1
(d
n
c
n
) .
Then the compact set
1
q
=
1
_
n=1
[c
n
j, d
n
j[
is disjoint from , so by Claim 5 we have
j
+
() j
+
(1
q
) = j
+
(' 1
q
) .
104 B. LEBESGUE MEASURE THEORY
We conclude from subadditivity and ' 1
q
G that
j
+
()
1

n=1
(d
n
c
n
2j) = j
+
() j
+
_
1
_
n=1
(c
n
j, d
n
j)
_
_ j
+
() j
+
(1
q
)
= j
+
(' 1
q
)
_ j
+
(G) _ j
+
() -.
Since j
+
() < for compact, we thus have
1

n=1
(d
n
c
n
) _ - 2'j
for all 0 < j <
1
2
min
1n1
(d
n
c
n
). Hence

1
n=1
(d
n
c
n
) _ - and taking
the supremum in ' _ '
+
we obtain from Claim 4 that
j
+
(G ) =
1

n=1
(d
n
c
n
) _ -.
Finally, if is closed, it is a countable union of compact sets =

o
n=1
([:, :[ ),
and hence / by Step 2.
Step 4:
c
/ if /.
For each : _ 1 there is by Claim 4 an open set G
n
such that j
+
(G
n
) <
1
n
. Then 1
n
= G
c
n
is closed and hence 1
n
/ by Step 3. Thus
o =
o
_
n=1
1
n
/, o
c
,
and
c
o G
n
for all : implies that
j
+
(
c
o) _ j
+
(G
n
) <
1
:
, : _ 1.
Thus j
+
(
c
o) = 0 and by Step 1 we have
c
o /. Finally, Step 2 shows that

c
= o ' (
c
o) /.
Thus far we have shown that / is a o-algebra, and we now turn to proving that
/ and j satisfy the four properties in Theorem 27. Property (1) is an easy exercise.
Property (2) is the main event. Let 1
n

o
n=1
be a pairwise disjoint sequence of sets
in /, and let 1 =

o
n=1
1
n
.
We will consider rst the case where each of the sets 1
n
is bounded. Let - 0
be given. Then 1
c
n
/ and so there are open sets G
n
1
c
n
such that
j
+
(G
n
1
c
n
) <
-
2
n
, : _ 1.
Equivalently, with 1
n
= G
c
n
, we have 1
n
closed, contained in 1
n
, and
j
+
(1
n
1
n
) <
-
2
n
, : _ 1.
2. MEASURABLE FUNCTIONS AND INTEGRATION 105
Thus the sets 1
n

o
n=1
are compact and pairwise disjoint. Claim 5 and induction
shows that

n=1
j
+
(1
n
) = j
+
_

_
n=1
1
n
_
_ j
+
(1) , _ 1,
and taking the supremum over yields
o

n=1
j
+
(1
n
) _ j
+
(1) .
Thus we have
o

n=1
j
+
(1
n
) _
o

n=1
j
+
(1
n
1
n
) j
+
(1
n
)
_
o

n=1
-
2
n

o

n=1
j
+
(1
n
) _ - j
+
(1) .
Since - 0 we conclude that

o
n=1
j
+
(1
n
) _ j
+
(1), and subadditivity of j
+
then
proves equality.
In general, dene 1
n,|
= 1
n
(/ 1, /[ ' [/, / 1) for /, : _ 1 so that
1 =

_
o
n=1
1
n
=

_
o
n,|=1
1
n,|
.
Then from what we just proved we have
j
+
(1) =
o

n,|=1
j
+
(1
n,|
) =
o

n=1
_
o

|=1
j
+
(1
n,|
)
_
=
o

n=1
j
+
(1
n
) .
Finally, property (3) follows from the observation that 1

o
n=1
(a
n
, /
n
) if and
only if 1 r

o
n=1
(a
n
r, /
n
r). It is then obvious that j
+
(1 r) = j
+
(1)
and that 1 r / if 1 /. Property (4) is immediate from Step 1 above. This
completes the proof of Theorem 27.
2. Measurable functions and integration
Let [, [ = R ' , be the extended real numbers with order and
(some) algebra operations dened by
< r < , r R,
r = , r R,
r = , r R,
r = , r 0,
r = , r < 0,
0 = 0.
The nal assertion 0 = 0 is dictated by

o
n=1
a
n
= 0 if all the a
n
= 0. It turns
out that these denitions give rise to a consistent theory of measure and integration
of functions with values in the extended real number system.
Let ) : R [, [. We say that ) is (Lebesgue) measurable if
)
1
([, r)) /, r R.
106 B. LEBESGUE MEASURE THEORY
The simplest examples of measurable functions are the characteristic functions
J
of measurable sets 1. Indeed,
(
J
)
1
([, r)) =
_
_
_
c if r _ 0
1
c
if 0 < r _ 1
R if r 1
.
It is then easy to see that nite linear combinations : =

n=1
a
n

Jr
of such
characteristic functions
Jr
, called simple functions, are also measurable. Here
a
n
R and 1
n
is a measurable subset of R. It turns out that if we dene the
integral of a simple function : =

n=1
a
n

Jr
by
_
R
: =

n=1
a
n
j(1
n
) ,
the value is independent of the representation of : as a simple function. Armed
with this fact we can then extend the denition of integral
_
R
) to functions ) that
are nonnegative on R, and then to functions ) such that
_
R
[)[ < .
At each stage one establishes the relevant properties of the integral along with
the most useful theorems. For the most part these extensions are rather routine, the
cleverness inherent in the theory being in the overarching organization of the con-
cepts rather than in the details of the demonstrations. As a result, we will merely
state the main results in logical order and sketch proofs when not simply routine.
We will however give fairly detailed proofs of the three famous convergence theo-
rems, the Monotone Convergence Theorem, Fatous Lemma, and the Dominated
Convergence Theorem. The reader is referred to the excellent exposition in [6] for
the complete story including many additional fascinating insights.
2.1. Properties of measurable functions. From now on we denote the
Lebesgue measure of a measurable subset 1 of R by [1[ rather than by j(1) as in
the previous sections. We say that two measurable functions ), q : R [, [
are equal almost everywhere (often abbreviated a.c.) if
[r R : ) (r) ,= q (r)[ = 0.
We say that ) is nite-valued if ) : R R. We now collect a number of elementary
properties of measurable functions.
Lemma 13. Suppose that ), )
n
, q : R [, [ for : N.
(1) If ) is nite-valued, then ) is measurable if and only if )
1
(G) / for
all open sets G R if and only if )
1
(1) / for all closed sets 1 R.
(2) If ) is nite-valued and continuous, then ) is measurable.
(3) If ) is nite-valued and measurable and 1 : R R is continuous, then
1 ) is measurable.
(4) If )
n

o
n=1
is a sequence of measurable functions, then the following func-
tions are all measurable:
sup
n
)
n
(r) , inf
n
)
n
(r) , ... lim sup
no
)
n
(r) , lim inf
no
)
n
(r) .
(5) If )
n

o
n=1
is a sequence of measurable functions and ) (r) = lim
no
)
n
(r),
then ) is measurable.
(6) If ) is measurable, so is )
n
for : N.
(7) If ) and q are nite-valued and measurable, then so are ) q and )q.
2. MEASURABLE FUNCTIONS AND INTEGRATION 107
(8) If ) is measurable and ) = q almost everywhere, then q is measurable.
Comments: For property (1), rst show that ) is measurable if and only if
)
1
((a, /)) / for all < a < / < . For property (3) use (1 ))
1
(G) =
)
1
_
1
1
(G)
_
and note that 1
1
(G) is open if G is open. For property (7), use
) q a =
_
:Q
[) a r q r[ , a R,
)q =
1
4
_
() q)
2
() q)
2
_
.
Recall that a measurable simple function , (i.e. the range of , is nite) has
the form
, =

|=1
c
|

J
!
, c
|
R, 1
|
/.
Next we collect two approximation properties of simple functions.
Proposition 19. Let ) : R [, [ be measurable.
(1) If ) is nonnegative there is an increasing sequence of nonnegative simple
functions ,
|

o
|=1
that converges pointwise and monotonically to ):
,
|
(r) _ ,
|+1
(r) and lim
|o
,
|
(r) = ) (r) , for all r R.
(2) There is a sequence of simple functions ,
|

o
|=1
satisfying
[,
|
(r)[ _

,
|+1
(r)

and lim
|o
,
|
(r) = ) (r) , for all r R.
Comments: To prove (1) let )
1
= min), ', and for 0 _ : < ' dene
1
n,,1
=
_
r R :
:

< )
1
(r) _
: 1

_
.
Then ,
|
(r) =

2
!
|
n=1
n
2
!

J
r2
!
!
(r) works. Property (2) is routine given (1).
2.2. Properties of integration and convergence theorems. If , is a
measurable simple function (i.e. its range is a nite set), then , has a unique
canonical representation
, =

|=1
c
|

J
!
,
where the real constants c
|
are distinct and nonzero, and the measurable sets 1
|
are pairwise disjoint. We dene the Lebesgue integral of , by
_
,(r) dr =

|=1
c
|
[1
|
[ .
If 1 is a measurable subset of R and , is a measurable simple function, then so is

J
,, and we dene
_
J
,(r) dr =
_
(
J
,) (r) dr.
Lemma 14. Suppose that , and c are measurable simple functions and that
1, 1 /.
108 B. LEBESGUE MEASURE THEORY
(1) If , =

1
|=1
,
|

J
!
(not necessarily the canonical representation), then
_
,(r) dr =
1

|=1
,
|
[1
|
[ .
(2)
_
(a, /c) = a
_
, /
_
c for a, / C,
(3)
_
J|J
, =
_
J
,
_
J
, if 1 1 = c,
(4)
_
, _
_
c if , _ c,
(5)

_
,

_
_
[,[.
Properties (2) - (5) are usually referred to as linearity, additivity, monotonicity
and the triangle inequality respectively. The proofs are routine.
Now we turn to dening the integral of a nonnegative measurable function
) : R [0, [. For such ) we dene
_
) (r) dr = sup
__
q (r) dr : 0 _ , _ ) and , is simple
_
.
It is essential here that ) be permitted to take on the value , and that the
supremum may be as well. We say that ) is (Lebesgue) integrable if
_
) (r) dr <
. For 1 measurable dene
_
J
) (r) dr =
_
(
J
)) (r) dr.
Here is an analogue of Lemma 14 whose proof is again routine.
Lemma 15. Suppose that ), q : R [0, [ are nonnegative measurable func-
tions and that 1, 1 /.
(1)
_
(a) /q) = a
_
) /
_
q for a, / (0, ),
(2)
_
J|J
) =
_
J
)
_
J
) if 1 1 = c,
(3)
_
) _
_
q if 0 _ ) _ q,
(4) If
_
) < , then ) (r) < for a.e. r,
(5) If
_
) = 0, then ) (r) = 0 for a.e. r.
Note that convergence of integrals does not always follow from pointwise con-
vergence of the integrands. For example,
lim
no
_

[n,n+1]
(r) dr = 1 ,= 0 =
_
lim
no

[n,n+1]
(r) dr,
and
lim
no
_
:
(0,
1
r
)
(r) dr = 1 ,= 0 =
_
lim
no
:
[0,
1
r
[
(r) dr.
In each of these examples, the mass of the integrands "disappears" in the limit; at
"innity" in the rst example and at the origin in the second example. Here are our
rst two classical convergence theorems giving conditions under which convergence
does hold.
Theorem 28. (Monotone Convergence Theorem) Suppose that )
n

o
n=1
is an
increasing sequence of nonnegative measurable functions, i.e. )
n
(r) _ )
n+1
(r),
and let
) (r) = sup
n
)
n
(r) = lim
no
)
n
(r) .
2. MEASURABLE FUNCTIONS AND INTEGRATION 109
Then ) is nonegative and measurable and
_
) (r) dr = lim
no
_
)
n
(r) dr.
Proof : Since
_
)
n
_
_
)
n+1
we have lim
no
_
)
n
= 1 [0, [. Now ) is
measurable and )
n
_ ) implies
_
)
n
_
_
) so that
1 _
_
).
To prove the opposite inequality, momentarily x a simple function , such that
0 _ , _ ). Choose c < 1 and dene
1
n
= r R : )
n
(r) _ c,(r) , : _ 1.
Then 1
n
is an increasing sequence of measurable sets with

o
n=1
1
n
= R. We have
_
)
n
_
_
Jr
)
n
_ c
_
Jr
,, : _ 1.
Now let , =

|=1
c
|

J
!
be the canonical representation of ,. Then
_
Jr
, =

|=1
c
|
[1
n
1
|
[ ,
and since lim
no
[1
n
1
|
[ = [1
|
[ by the fourth line in (1.1), we obtain that
_
Jr
, =

|=1
c
|
[1
n
1
|
[

|=1
c
|
[1
|
[ =
_
,
as : . Altogether then we have
1 = lim
no
_
)
n
_ c
_
,
for all c < 1, which implies 1 _
_
, for all simple , with 0 _ , _ ), which implies
1 _
_
) as required.
Corollary 10. Suppose that a
|
(r) _ 0 is measurable for / _ 1. Then
_
o

|=1
a
|
(r) dr =
o

|=1
_
a
|
(r) dr.
To prove the corollary apply the Monotone Convergence Theorem to the se-
quence of partial sums )
n
(r) =

n
|=1
a
|
(r).
Lemma 16. (Fatous Lemma) If )
n

o
n=1
is a sequence of nonnegative mea-
surable functions, then
_
lim inf
no
)
n
(r) dr _ lim inf
no
_
)
n
(r) dr.
Proof : Let q
n
(r) = inf
|n
)
|
(r) so that q
n
_ )
n
and
_
q
n
_
_
)
n
. Then
q
n

o
n=1
is an increasing sequence of nonnegative measurable functions that con-
verges pointwise to liminf
no
)
n
(r). So the Monotone Convergence Theorem
yields
_
lim inf
no
)
n
(r) dr = lim
no
_
q
n
(r) dr _ lim inf
no
_
)
n
(r) dr.
110 B. LEBESGUE MEASURE THEORY
Finally, we can give an unambiguous meaning to the integral
_
) (r) dr in the
case when ) is integrable, by which we mean that ) is measurable and
_
[) (r)[ dr <
. To do this we introduce the positive and negative parts of ):
)
+
(r) = max ) (r) , 0 and )

(r) = max ) (r) , 0 .

Then both )
+
and )

are nonnegative measurable functions with nite integral.

We dene
_
) (r) dr =
_
)
+
(r) dr
_
)

(r) dr.
With this denition we have the usual elementary properties of linearity, addi-
tivity, monotonicity and the triangle inequality.
Lemma 17. Suppose that ), q are integrable and that 1, 1 /.
(1)
_
(a) /q) = a
_
) /
_
q for a, / R,
(2)
_
J|J
) =
_
J
)
_
J
) if 1 1 = c,
(3)
_
) _
_
q if ) _ q,
(4)

_
)

_
_
[)[.
Our nal convergence theorem is one of the most useful in analysis.
Theorem 29. (Dominated Convergence Theorem) Let q be a nonnegative in-
tegrable function. Suppose that )
n

o
n=1
is a sequence of measurable functions sat-
isfying
lim
no
)
n
(r) = ) (r) , a.c. r,
and
[)
n
(r)[ _ q (r) , a.c. r.
Then
lim
no
_
[) (r) )
n
(r)[ dr = 0,
and hence
_
) (r) dr = lim
no
_
)
n
(r) dr.
Proof : Since [)[ _ q and ) is measurable, ) is integrable. Since [) )
n
[ _ 2q,
Fatous Lemma can be applied to the sequence of functions 2q [) )
n
[ to obtain
_
2q _ lim inf
no
_
(2q [) )
n
[)
=
_
2q lim inf
no
_

_
[) )
n
[
_
=
_
2q lim sup
no
_
[) )
n
[ .
Since
_
2q < , we can subtract it from both sides to obtain
lim sup
no
_
[) )
n
[ _ 0,
which implies lim
no
_
[) )
n
[ = 0. Then
_
) = lim
no
_
)
n
follows from the
triangle inequality

_
() )
n
)

_
_
[) )
n
[.
2. MEASURABLE FUNCTIONS AND INTEGRATION 111
Finally, if ) (r) = n(r) i (r) is complex-valued where n(r) and (r) are
real-valued measurable functions such that
_
[) (r)[ dr =
_ _
n(r)
2
(r)
2
dr < ,
then we dene
_
) (r) dr =
_
n(r) dr i
_
(r) dr.
The usual properties of linearity, additivity, monotonicity and the triangle inequal-
ity all hold for this denition as well.
2.3. Three famous measure problems. The following three problems are
listed in order of increasing diculty.
Problem 3. Suppose that 1
1
, ..., 1
n
are : Lebesgue measurable subsets of [0, 1[
such that each point r in [0, 1[ lies in some / of these subsets. Prove that there is
at least one set 1

with [1

[ _
|
n
.
Problem 4. Suppose that 1 is a Lebesgue measurable set of positive measure.
Prove that
1 1 = r j : r, j 1
contains a nontrivial open interval.
Problem 5. Construct a Lebesgue measurable subset of the real line such that
0 <
[1 1[
[1[
< 1
for all nontrivial open intervals 1.
To solve Problem 3, note that the hypothesis implies / _

n
=1

J
(r) for
r [0, 1[. Now integrate to obtain
/ =
_
1
0
/dr _
_
1
0
_
_
n

=1

J
(r)
_
_
dr =
n

=1
_
1
0

J
(r) dr =
n

=1
[1

[ ,
which implies that [1

[ _
|
n
for some ,. The solution is much less elegant without
recourse to integration.
To solve Problem 4, choose 1 compact contained in 1 such that [1[ 0. Then
choose G open containing 1 such that [G 1[ < [1[. Let c = di:t (1, G
c
) 0. It
follows that (c, c) 1 1 1 1. Indeed, if r (c, c) then 1 r G and
1 (1 r) ,= c since otherwise we have a contradiction:
2 [1[ = [1[ [1 r[ _ [G[ _ [G 1[ [1[ < 2 [1[ .
Thus there are /
1
and /
2
in 1 such that /
1
= /
2
r and so
r = /
2
/
1
1 1.
Problem 5 is most easily solved using generalized Cantor sets 1
o
. Let 0 < c _ 1
and set 1
0
1
= [0, 1[. Remove the open interval of length
1
3
c centered in 1
0
1
and denote
the two remaining closed intervals by 1
1
1
and 1
1
2
. Then remove the open interval of
length
1
3
2
c centered in 1
1
1
and denote the two remaining closed intervals by 1
2
1
and
1
2
2
. Do the same for 1
1
2
and denote the two remaining closed intervals by 1
2
3
and 1
2
4
.
112 B. LEBESGUE MEASURE THEORY
Continuing in this way, we obtain at the /
||
generation, a collection
_
1
|

_
2
!
=1
of 2
|
pairwise disjoint closed intervals of equal length. Let
1
o
=
o

|=1
_
_
2
!
_
=1
1
|

_
_
.
Then by summing the lengths of the removed open intervals, we obtain
[[0, 1[ 1
o
[ =
1
8
c
2
8
2
c
2
2
8
3
c ... = c,
and it follows that 1
o
is compact and has Lebesgue measure 1 c. It is not hard
to show that 1
o
is also nowhere dense. The case c = 1 is particularly striking: 1
1
is a compact, perfect and uncountable subset of [0, 1[ having Lebesgue measure 0.
This is the classical Cantor set.
In order to construct the set 1 in Problem 3, it suces by taking unions of
translates by integers, to construct a subset 1 of [0, 1[ satisfying
(2.1) 0 <
[1 1[
[1[
< 1, for all intervals 1 [0, 1[ of positive length.
Fix 0 < c
1
< 1 and start by taking 1
1
= 1
o1
. It is not hard to see that
[J
1
|1[
]1]
< 1
for all 1, but the left hand inequality in (2.1) fails for 1 = 1
1
whenever 1 is a subset
of one of the component intervals in the open complement [0, 1[ 1
1
. To remedy
this x 0 < c
2
< 1 and for each component interval J of [0, 1[ 1
1
, translate and
dilate 1
o2
to t snugly in the closure J of the component, and let 1
2
be the union
of 1
1
and all these translates and dilates of 1
o2
. Then again,
[J
2
|1[
]1]
< 1 for all
1 but the left hand inequality in (2.1) fails for 1 = 1
2
whenever 1 is a subset of
one of the component intervals in the open complement [0, 1[ 1
2
. Continue this
process indenitely with a sequence of numbers c
n

o
n=1
(0, 1). We claim that
1 =

o
n=1
1
n
satises (2.1) if and only if
(2.2)
o

n=1
(1 c
n
) < .
To see this, rst note that no matter what sequence of numbers c
n
less than
one is used, we obtain that 0 <
]J|1]
]1]
for all intervals 1 of positive length. Indeed,
each set 1
n
is easily seen to be compact and nowhere dense, and each component
interval in the complement [0, 1[ 1
n
has length at most
c
1
8
c
2
8
...
c
n
8
_ 8
n
.
Thus given an interval 1 of positive length, there is : large enough such that 1 will
contain one of the component intervals J of [0, 1[ 1
n
, and hence will contain the
translated and dilated copy (
_
1
or+1
_
of 1
or+1
that is tted into J by construction.
Since the dilation factor is the length [J[ of J, we have
[1 1[ _

(
_
1
or+1
_

= [J[

1
or+1

= [J[ (1 c
n+1
) 0,
since c
n+1
< 1.
It remains to show that [1 1[ < [1[ for all intervals 1 of positive length in
[0, 1[, and it is here that we must use (2.2). Indeed, x 1 and let J be a component
interval of [0, 1[ 1
n
(with : large) that is contained in 1. Let (
_
1
or+1
_
be the
2. MEASURABLE FUNCTIONS AND INTEGRATION 113
translated and dilated copy of 1
or+1
that is tted into J by construction. We
compute that
[1 J[ =

(
_
1
or+1
_

(1 c
n+2
)

J (
_
1
or+1
_

...
= (1 c
n+1
) [J[ (1 c
n+2
) (1 (1 c
n+1
)) [J[
(1 c
n+3
) (1 (1 c
n+1
) (1 c
n+2
) (1 (1 c
n+1
))) [J[ ...
=
o

|=1
,
n
|
[J[ ,
where by induction,
,
n
|
= (1 c
n+|
) c
n+|1
...c
n+1
, / _ 1.
Then we have
[1 J[ =
_
o

|=1
,
n
|
_
[J[ < [J[ ,
and hence also
]J|1]
]1]
< 1, if we choose c
n

o
n=1
so that

o
|=1
,
n
|
< 1 for all :.
Now we have
o

|=1
,
n
|
=
o

|=1
(1 c
n+|
) c
n+|1
...c
n+1
= 1
o

|=1
c
n+|
,
and by the rst line in (0.5) of Chapter 5, this is strictly less than 1 if and only if

o
|=1
(1 c
n+|
) < for all :. Thus the set 1 constructed above satises (2.1) if
and only if (2.2) holds.
Bibliography
[1] R. P. Boas, Invitation to Complex Analysis, Random House/Birkhuser Mathematics Series,
Random House, New York, 1987.
[2] A. Hatcher, Algebraic Topology, Cambridge University Press, 2002.
[3] R. Maehara, The Jordan curve theorem via the Brouwer xed point theorem, Amer. Math.
Monthly , (1984), 641-643.
[4] W. Rudin, Principles of Mathematical Analysis, McGraw-Hill, 3rd edition, 1976.
[5] W. Rudin, Real and Complex Analysis, McGraw-Hill, 3rd edition, 1987.
[6] E. M. Stein and R. Shakarchi, Complex Analysis, Princeton Lectures in Analysis II, Prince-
ton University Press, Princeton and Oxford, 2003.
115