1.

The Lebesgue Measure

Now we will discuss the Lebesgue measure. The reference for this part is the book Real Variables, by
A.Torchinsky, which is on reserve at the library.
A closed interval in R
n
is a set of the form
I = [a
1
, b
1
] ... [a
n
, b
n
], < a
j
< b
j
< , 1 j n.
Similarly an open interval is a set of the form
I = (a
1
, b
1
) ... (a
n
, b
n
), < a
j
< b
j
< , 1 j n.
The volume of either the open or closed interval is
v(I) =
n

j=1
(b
j
a
j
).
The following lemma is an important part in the construction of the Lebesgue measure.
Lemma 1.1. Let O R
n
be a nonempty open subset. Then there exist closed intervals {I
j
, j N}
such that

Ij

Ik
= , j = k, and O =

j=1
I
j
.
Proof. For k N, let P
k
be the set of points of R
n
whose coordinates are integral multiples of 2
k
. For
each p = (p
1
, p
2
, ..., p
n
) P
k
let
I(p, 2
k
) = {x R
n
: p
j
x
j
< p
j
+ 2
k
, 1 j n}.
Let
k
= {I(p, 2
k
) : p P
k
}.
These sets satisfy the following properties:
1) Fixed k, then for each point x R
n
there exists a unique interval I(p, 2
k
) in
k
such that x I(p, 2
k
).
2) Let k

> k. If I(p
1
, 2
k
)
k
, I(p
2
, 2
k

)
k
then either I(p
2
, 2
k

) I(p
1
, 2
k
) or I(p
2
, 2
k

)
I(p
1
, 2
k
) = .
To deduce property 1, just notice that any q R can be written as q = m2
k
+m
1
where m N and
|m
1
| < 2
k
.
To prove property 2, notice that it follows from property 1 that there exists a unique box I(q, 2
k
)
k
such that the corner p
2
I(q, 2
k
). Then of course, I(p
2
, 2
k

) I(q, 2
k
). So either q = p
1
, and in that
case I(p
2
, 2
k

) I(p
1
, 2
k
) or q = p
1
and I(p
2
, 2
k

) I(p
1
, 2
k
) = .
Now we can prove the lemma. Since O is open, then for every x O there exists I(p, 2
k
) such that
x I(p, 2
k
) O. Thus if F = {I(p, 2
k
)
k
: I(p, 2
k
) O},
O =
_
I(p,2
k
)F
I(p, 2
k
).
Now we just lter the collection, i.e. take those intervals in
1
and throw out all the intervals in
j
,
j > 1, which are contained in one of the intervals in
1
which are contained in O. From the remaining
collection select the intervals in
2
which are contained in O and remove the ones in
j
, j > 2, which
are contained in any of these. Repeat the process for n = 3, 4, ... The remaining intervals {I
j
, j N}
are disjoint and satisfy O =

j=1
I
j
. To nish the proof of the lemma just take the closures of these
intervals.
Given a set A R
n
we dene its outer measure by
|A|
e
= inf{

k
v(I
k
) : A

_
k=1
I
k
}.
1
2
Here the inmum is taken over the family of countable covers of A by closed intervals.
Theorem 1.1. The outer measure has the following properties:
P.1) If A B, then |A|
e
|B|
e
.
P.2) If I
j
, j = 1, 2, ..., N be either closed intervals with

Ik

Ij
= if k = j, or pairwise disjoint open
intervals, then

N
_
j=1
I
j

e
=
N

j=1
v(I
j
).
P.3) Let E
k
R
n
, k = 1, 2, .... Then

_
k=1
E
k

k=1
|E
k
|
e
.
P.4) For any E R
n
,
|E|
e
= inf{|O|
e
: E O and O is open }.
P.5) For any E R
n
there exists a G

set H such that E H and |E|

e
= |H|
e
.
Proof. To prove property P.1 just notice that any family of closed intervals that cover B also cover A.
So the inequality just follows from the denition of inmum.
We rst prove P.2 for closed intervals, and we begin by observing that the intervals I
k
, 1 k N
form a cover of the set A =

N
j=1
I
j
. Therefore, by denition
|A|
e

N

j=1
v(I
j
).
Now we want to prove that |A|
e

N
j=1
v(I
j
). Given > 0, it follows from the denition of inmum that
there exist closed intervals J
m
, m = 1, 2, ..., covering A, such that
|A|
e

m=1
v(J
m
) < |A|
e
(1 +). (1.1)
Let us denote J
m
= [a
m
1
, b
m
1
] .... [a
m
n
, b
m
n
]. Then v(J
m
) =

n
i=1
(b
m
i
a
m
i
). Let
m
> 0 and let J

m
be the
open interval J

m
= (a
m
1

m
, b
m
1
+
m
) .... (a
m
n

m
, b
m
n
+
m
). Then v(J

m
) =

n
i=1
(b
m
i
a
m
i
+2
m
).
Since < a
m
i
< b
m
i
< , i = 1, 2, ..., n, one can pick
m
such that
v(J

m
) < (1 +)v(J
m
). (1.2)
Since A

m=1
J

m
, J

m
is open, for all m, and A is compact, there exist a nite collection of the J

m
,
1 m M, covering A.
A
M
_
m=1
J

m
.
This implies that if
A
and
J

m
are the characteristic functions of A and J

m
respectively,

A

M

m=1

m
. (1.3)
3
Now we appeal to the fact that the volume of an interval is the Riemann integral of its characteristic
function. Since A is the union of closed intervals I
k
, 1 k N, with non-intersecting interiors, this
implies that
N

j=1
v(I
k
)
M

m=1
v(J

m
). (1.4)
Putting together equations (1.1), (1.2), (1.3) and (1.4) we get
N

j=1
v(I
k
)
M

m=1
v(J

m
)
M

m=1
(1 +)v(J
m
) (1 +)

m=1
v(J
m
) (1 +)
2
|A|
e
.
Since this holds for every , it shows that |A|
e


N
j=1
v(I
j
). This proves P.2 when the intervals are
closed. To prove P.2 for pairwise disjoint open intervals, we rst observe that
N
_
j=1
I
j

N
_
j=1
I
j
.
By P.1 and the rst part of P.2,

N
_
j=1
I
j

N
_
j=1
I
j

j=1
v(I
j
). (1.5)
On the other hand, if I
j
= (a
j
1
, b
j
1
) ... (a
j
n
, b
j
n
), then for 0 < < b
j
k
a
j
k
, k = 1, ..., n, j = 1, ..., N, et
I

j
= [a
j
1
+, b
j
1
] ... [a
j
n
+, b
j
n
] I
j
, and since the I
j
are paiwise disjoint, the I

j
are disjoint.
Therefore by P.1, the rst case of P.2, and (1.5)
N

j=1
v(I

j
) =

N
_
j=1
I

N
_
j=1
I
j

j=1
v(I
j
).
Since this holds for every , P.2 also holds for open intervals.
Property P.3 is obvious if |E
k
|
e
= for some k. So we may assume |E
k
|
e
< for every k. Let > 0
then for each k we can pick a collection {I
j,k
, j N} of closed intervals such that
E
k

_
j=1
I
j,k
, and |E
k
|
e

j=1
v(I
jk
) +/2
k
, k = 1, 2, ... (1.6)
Clearly,

k=1
E
k

j,k=1
I
j,k
, and hence

_
k=1
E
k

j,k
v(I
j,k
)
Since this is a series of positive terms, we have from (1.6)

_
k=1
E
k

j,k
v(I
j,k
) =

k=1

j=1
v(I
j,k
)

k=1
(|E
k
|
e
+/2
k
) +

k=1
|E
k
|
e
.
Since is arbitrary, P.3 holds.
Since |E|
e
|O|
e
for every E O, P.4 is obvious if |E|
e
= . So we may assume that |E|
e
< .
4
For > 0, let I
k
, k = 1, 2, ... be closed intervals such that
E

_
k=1
I
k
, and

k=1
v(I
k
) |E|
e
+/2.
For each k, pick an open interval I

k
such that
I
k
I

k
and v(I

k
) v(I
k
) +/2
k+1
. (1.7)
Let O =

k=1
I

k
. Then O is open and E O. By property P.1 |E|
e
|O|
e
. By property P.3, property
P.2, applied to a single interval, and (1.7)
|O|
e

k=1
|I

k
|
e
=

k
v(I

k
)

k
(v(I
k
) +/2
k+1
) /2 +

k
v(I
k
) |E|
e
+.
Thus for every > 0 there exists an open subset O such that E O and that |E|
e
|O|
e
|E| +. This
proves P.4.
Now we prove P.5. If |E|
e
= , just take H = R
n
. So we may assume that |E|
e
< . By property
P.4, there exists a sequence of open subsets O
k
, k = 1, 2, ..., such that E O
k
and
|O
k
|
e
|E|
e
+ 1/k.
Let H

k=1
O
k
. Then H is a G

, E H O
k
, k = 1, 2, ... Therefore, by property P.1
|E|
e
|H|
e
|O
k
| |E
e
| + 1/k, k = 1, 2, ...
Thus |H|
e
= |E|
e
.
This ends the proof of the theorem.
The following is an important generalization of Property P.2
Proposition 1.1. Let E
1
R
n
and E
2
R
n
be such that d(E
1
, E
2
) > 0, where
d(E
1
, E
2
) = inf{|x x

| : x E
1
, x

E
2
}.
Then |E
1
E
2
|
e
= |E
1
|
e
+|E
2
|
e
.
Proof. If either |E
1
|
e
= or |E
2
|
e
= , the statement is true. So we may assume that |E
1
|
e
< and
|E
2
|
e
< . From property P.3 we know that
|E
1
E
2
|
e
|E
1
|
e
+|E
2
|
e
.
So we only need to prove the opposite inequality. Since |E
1

E
2
|
e
< , for any > 0 there exists a
cover of E
1

E
2
by closed intervals {I
k
, k N} such that

k=1
v(I
k
)

E
1
_
E
2

e
+.
We assume that for every interval I
k
in this cover I
k
(E
1
E
2
) = , otherwise we can just remove it
from the collection. We then divide this family into three parts:
{I
k
, k N} = F
1
F
2
F
3
, where
I
k
F
1
if and only if I
k
E
1
= , I
k
E
2
= ,
I
k
F
2
if and only if I
k
E
1
= , I
k
E
2
= ,
I
k
F
3
if and only if I
k
E
1
= , I
k
E
2
= .
5
Each interval in I
k
F
3
can be divided into a nite number of intervals I
j,k
with diameter less than
d(E
1
, E
2
). And we have
v(I
k
) =

v(I
j,k
). (1.8)
But each interval I
j,k
falls into the family F
1
or F
2
, or it does not intersect E
1
E
2
. We throw out the
intervals that do not intersect E
1
E
2
. Then we replace each of the intervals I
k
F
3
by the intervals
in {I
j,k
} which are in F
1
or F
2
. We have now a family of closed intervals {I

k
} = F
1
F
2
which cover
E
1
E
2
, and consists of the intervals {I
k
} which are in either F
1
or F
2
, and the intervals in {I
jk
} which
are also in F
1
F
2
. In view of (1.8)

k=1
v(I

k
)

E
1
_
E
2

e
+. (1.9)
Therefore, by denition of outer measure
|E
1
|
e
+|E
2
|
e

k
F1
v(I

k
) +

k
F2
v(I

k
) =

k
v(I

k
) |E
1
E
2
|
e
+.
Since > 0 is arbitrary, this implies that
|E
1
|
e
+|E
2
|
e

E
1
_
E
2

e
.

So far we have constructed the outer measure, which is dened for all subsets of R
n
. But we do not
know whether the outer measure is in fact a measure, i.e. if it is -additive. In fact it is not.
Proposition 1.2. There exists a family {A
k
R : k = 0, 1, ...} such that A
k
A
j
= and |

k=1
A
k
|
e
<

k=1
|A
k
|
e
.
Proof. The rst thing we need to observe is that if A R
n
and w R
n
, then |A+w|
e
= |A|
e
. Indeed, if
{I
k
, k = 1, ...} are closed intervals that cover A, then the translates {I
k
+w} will cover A+w. Reciprocally,
if {J
k
, k N}, cover of A +w then {J
k
w, k N} cover A. So it follows from the denition of outer
measure, and the invariance of the volume of an interval under translation, that |A|
e
= |A +w|
e
.
Next we dene the following relation for elements of I = [1/2, 1/2] : we say tat x y if x y Q
[1, 1]. It is easy to see that is an equivalence relation. Let A = [1/2, 1/2]/ = {[x] : x [1/2, 1/2]},
where [x] denotes the class of equivalence of the point x. Let r
k
, k N be an enumeration of the rational
numbers in [1, 1], and for each k, let A
k
= A+r
k
. Then A
k
A
j
= if k = j. Otherwise [x] +r
j
= [y] +r
k
and then [x] [y] = r
k
r
j
Q. This implies that [x] = [y] but then r
j
= r
k
.
One one hand A
k
[3/2, 3/2] for k = 1, 2, ..., therefore

k=1
A
k
[3/2, 3/2]. On the other hand,
by the denition of the equivalence relation, [1/2, 1/2]

k=1
A
k
. So we conclude that
[1/2, 1/2]

_
k=1
A
k
[3/2, 3/2].
Therefore
1

_
k=1
A
k

e
3. (1.10)
But we also know that |A
k
|
e
= |A|
e
. If |

k=1
A
k
|
e
=

k=1
|A
k
|
e
, it would follow that |A
k
|
e
= 0, but
then, by property P.3, |

k=1
A
k
|
e
= 0, which contradicts (1.10).
To make the outer measure -additive, we have to restrict the -algebra where it is dened.
6
Denition 1.1. Let L be the family of subsets E R
n
such that for any > 0 there exists an open
subset O R
n
such that
E O and |O \ E|
e
< .
If E L we say that E is Lebesgue measurable.
This is the class of sets we will work with.
Theorem 1.2. The family L is a -algebra and the function
: L [0, ]
(E) = |E|
e
is a positive measure on L.
The proof of this theorem will be divided into several lemmas.
Lemma 1.2. If E R
n
is an open subset, then E L. If E R
n
is such that |E|
e
= 0, then E L.
Proof. Of course, if E is open one can take O = E and thus E satises the requirement of the denition.
If |E|
e
= 0, then it follows from property P.4 that for any > 0, there exists an open set O such that
E O and |O|
e
< . Since O \ E O, property P.1 implies that |O \ E|
e
< .
Lemma 1.3. Let E
k
L, k = 1, 2, ... then E =

k=1
E
K
L.
Proof. Let > 0. Since E
k
L, there exists O
k
, open, such that
E
k
O
k
and |O
k
\ E
k
|
e
< /2
k
, k = 1, 2, ...
Hence O =

k=1
O
k
E. Moreover,
O \ E =

_
k=1
(O
k
\ E)

_
k=1
(O
k
\ E
k
) .
Therefore, in view of property P.3,
|O \ E|
e

k=1
|O
k
\ E
k
|
e
< .
This ends the proof of the Lemma.
Lemma 1.4. If F R
n
is closed, then F L.
Proof. We will show that if F is compact then F L. For a general closed set F, we write F =

k=1
F
k
,
F
k
= F {x : |x| k}, which is compact. Then by Lemma 1.3, F L.
So let F be compact. Then for > 0 there exists an open set O such that F O and |O|
e
|F|
e
+.
We claim that in fact |O \ F|
e
< .
Since O \ F is open, then by Lemma 1.1,
O \ F =

_
j=1
I
j
, I
j
non-overlapping closed intervals.
Hence
|O \ F|
e

j=1
v(I
j
).
7
On the other hand
O = F
_
_

_
j=1
I
j
_
_
F
_
_
N
_
j=1
I
j
_
_
, N N.
Hence
|O|
e

F
_
_
N
_
j=1
I
j
_
_

e
, N N. (1.11)
Since F and

N
j=1
I
j
are compact disjoint subsets, the distance between them is positive, and then
Proposition 1.1 guarantees that

F
N
_
j=1
I
j

e
= |F|
e
+

N
_
j=1
I
j

e
.
However, since the intervals I
k
are non-overlapping property P.2 implies that

N
j=1
I
j

e
=

N
k=1
v(I
k
).
Hence

F
N
_
j=1
I
j

e
= |F|
e
+
N

k=1
v(I
k
).
So we conclude from (1.11) that
|F|
e
+
N

k=1
v(I
k
) |O|
e
, N = 1, 2, ...
Therefore
N

k=1
v(I
k
) |O|
e
|F|
e
< , N = 1, 2, ...
So
|O \ F|
e

k=1
v(I
k
) .
This concludes the proof of the lemma.
Lemma 1.5. Suppose E L then R
n
\ E L.
Proof. Let O
k
be open subsets of R
n
, k = 1, 2, ... with E O
k
and
|O
k
\ E|
e
1/k, k = 1, 2, ...
Now R
n
\ O
k
is closed, and therefore in L. Moreover
R
n
\ O
k
R
n
\ E, k = 1, 2, ...
Therefore
H =

_
k=1
(R
n
\ O
k
) R
n
\ E
H is an F

, and thus H L. Let A = (R

n
\ E) \ H. Then
R
n
\ E = H A.
We claim that |A|
e
= 0. In view of Lemma 1.2, this shows that R
n
\ E L.
8
To prove our claim, notice that
A = (R
n
\ E) \
_

_
k=1
(R
n
\ O
k
)
_
(R
n
\ E) \ (R
n
\ O
k
) = O
k
\ E.
Hence A O
k
\ E, k = 1, 2, ... Then
|A|
e
|O
k
\ E|
e
1/k, k = 1, 2, ...
Hence |A|
e
= 0.
This Lemma has an important consequence:
Corollary 1.1. If E L, then E = H A, H an F

and |A|
e
= 0.
Proof. The proof of the Lemma 1.5 shows that if E L, then R
n
\ E = H A, where H is an F

and
|A|
e
= 0. Just apply this result to E = R
n
\ (R
n
\ E)).
Another important consequence of this is
Corollary 1.2. If E L then for every > 0, there exists a closed set F R
n
such that F E and
|E \ F|
e
< .
Finally we prove that the outer measure restricted to L gives a measure
Theorem 1.3. The function
: L [0, ]
(E) = |E|
e
is a measure.
Proof. Let E
j
L, j = 1, 2, ... be such that E
j
E
k
= if j = k. We know from property P.3 that

_
_

_
j=1
E
j
_
_

j=1
(E
j
) .
We want to show that

j=1
(E
j
)
_
_

_
j=1
E
j
_
_
. (1.12)
Assume that E
j
is bounded j = 1, 2, ... Given > 0, by Corollary 1.2 there exists a closed set F
k
E
k
such that (E
k
\ F
k
) < /2
k
. Since
E
k
= F
k
(E
k
\ F
k
),
we have that
(E
k
) (F
k
) +/2
k
. (1.13)
The family {F
k
} consists of pairwise disjoint compact subsets. Therefore, for any N N,

_
N
_
k=1
F
k
_
=
N

k=1
(F
k
) (E), N = 1, 2, ...
Hence

k=1
(F
k
) (E). (1.14)
9
So we conclude from (1.13) and (1.14) that

k=1
(E
k
)

k=1
(F
k
) +/2
k
(E) +.
Since this holds for every (1.12) must hold.
In general, if E
k
is not bounded, let I
0
= and I
j
= [j, j] ... [j, j], j = 1, 2, ... Let S
j
= I
j
\I
j1
,
j = 1, 2, ... Then E
k,j
= E
k
S
j
are measurable, pairwise disjoint and bounded. Therefore
(E
k
) =

j=1
(E
j,k
).
Moreover

_
k=1
E
k
=

_
k=1
E
j,k
So

_
k=1
E
k
_
=
_
_

_
j,k=1
E
j,k
_
_
=

j,k=1
(E
j,k
) =

k=1
(E
k
).
Therefore is -additive, and we are done.
We end this part with a very important characterization of the Lebesgue measurable sets.
Theorem 1.4. (Caratheodory) A subset E R
n
is Lebesgue measurable if and only if, for every A R
n
|A|
e
= |A E|
e
+|A \ E|
e
(1.15)
Proof. Suppose rst that E L and we will prove (1.15). Let A R
n
. Since A = A E (A \ E), we
have
|A|
e
|A E|
e
+|A \ E|
e
. (1.16)
We know that there exists a G

set H such that H A and |H|

e
= |A|
e
. Since H L, and H =
H E (H \ E), is a union of disjoint measurable sets. Then
|A|
e
= |H|
e
= |H E|
e
+|H \ E|
e
But H E A E and H \ E A \ E therefore
|A|
e
= |H|
e
= |H E|
e
+|H \ E|
e
|A E|
e
+|A\ E|
e
. (1.17)
Then (1.16) and (1.17) show (1.15).
Now suppose (1.15) holds for every A. Let us rst make the extra assumption that |E|
e
< . In that
case, there exists a G

set H such that

E H and |H|
e
= |E|
e
.
But H L and using (1.15) with A = H we nd that
|H|
e
= |H E|
e
+|H \ E|
e
= |E|
e
+|H \ E|.
Hence |H \ E|
e
= 0. Therefore H \ E L Since E = H \ (H \ E) it follows that E L.
In the case where |E|
e
= , write E =

k=1
(E {x : |x| < k}) .
Let H
k
be a G

set such that

E
k
H
k
, |H
k
|
e
= |E
k
|
e
.
10
Applying (1.15) to H
k
we have
|E
k
|
e
= |H
k
|
e
= |H
k
E|
e
+|H
k
\ E|
e
But H
k
E H
k
E
k
= E
k
and H
k
\ E H
k
\ E
k
. So
|H
k
E|
e
|E
k
|
e
, |H
k
\ E|
e
= 0.
Therefore H
k
\ E L and hence, if H =

k=1
H
k
, H L, and E H. But H \ E =

k=1
(H
k
\ E) L
and therefore E = H \ (H \ E) L.