THE GENERAL LEBESGUE INTEGRAL

by
Leonard R. Gardner III
Submitted in Partial Fulllment
of the Requirements for
Graduation with Honors from the
South Carolina Honors College
April 2000
Approved:
Dr. Maria K. Girardi
First Reader
Dr. James W. Roberts
Second Reader
Peter C. Sederberg, James L. Stiver, or Douglas F. Williams
for South Carolina Honors College
Contents
1. Thesis Summary ii
2. Introduction iv
3. Measurable Functions 1
4. Measures 16
5. The Integral 29
6. Integrable Functions 38
7. The Lebesgue Spaces L
p
46
8. Modes of Convergence 58
9. Conclusion 65
References 66
i
1. Thesis Summary
This paper expands upon Robert Bartles exploration of the General
Lebesgue Integral in his text Elements of Integration. Following Bartles
example, the paper opens with a discussion of the groundwork on which the
theory of the Lebesgue Integral stands. As a house sits on cement and brick
footings, the Lebesgue Integral is propped on solid mathematical concepts
such as algebras and measures.
After setting the basics, the framework or the skeleton of the Integral is
presented. Mathematicians are fond of handling the simple cases rst then
extending the results to more complicated and detailed cases. Obeying this
methodology, the paper initially establishes the Integral for a limited class
of functions, namely those measurable functions with non-negative values.
Then later the Integral is dened for all measurable functions.
Having completed the base and the frame of the Integral, the paper then
focuses on some of the beautiful and elegant theories that adorn the Gen-
eral Lebesgue Integral, like the Monotone Convergence Theorem and the
Lebesgue Dominated Convergence Theorem. Indeed, convergence is an ex-
tremely important idea in mathematics and many times mathematicians are
interested in sequences of functions and the convergence (if it does in fact
converge) of these sequences.. Sometimes these sequences of functions con-
verge or get very close to another function. And other times these sequences
diverge or dont get very close to another function. In a convergent sequence
of functions, no matter how far you go out in the sequence it still stays very
close to one unchanging function or what mathematicians call the limit of
the sequence. In a divergent sequence, as you move through the sequence
you never get close to one particular function. What mathematicians want
to know is if you take the limit of the sequence and then integrate will you
get the same value as if you integrate each function in the sequence and then
take the limit of the integrals? In other words, with two operations, does
ii
it matter in which order you perform them? It turns out that with some
precautions, the order may be switched.
Furthermore, the paper explores spaces of functions. In laymans terms,
through modications of the Lebesgue Integral mathematicians are able to
categorize functions that share similar properties. Once a denition of these
spaces has been set, then an analysis of the interactions between these cat-
egories can be performed. For example, we can determine what properties
are needed for a function to live in more than one space or what properties
are needed so that a function lives in every space.
Of course, mathematicians combine these concepts of sequences of func-
tions and spaces of functions. More specically, they want to know if given a
sequence of functions that converge and all of the functions in that sequence
live in a particular space, does the limit live in that same space?
By the way, did I mention that there are a handful of ways that se-
quences of functions can converge? Well, the last section of paper dicusses
these dierent modes of convergence. Of particular interest is the order of
implication. In simpler terms, if a sequence converges in a given manner,
does it then converge in another manner or manners.
iii
2. Introduction
The French mathematician Henri Leon Lebesgue developed the Lebesgue
integral as a consequence of the problems associated with the Riemann In-
tegral. In particular, whole classes of important functions could not be
integrated with the Riemann Integral. For example, the function on the in-
terval [0, 1] that maps all the rational numbers to zero and all the irrational
numbers to themselves cannot be integrated using Riemannian Integration.
In fact, Lebesgue provided necessary and sucient conditions for a function
to be Riemann Integrable.
Lebesgues major insight was to leave the x-axis and nd the area under
the curve by partitioning the y-axis. To see the diering approaches of
the two mathematicians, imagine that Riemann and Lebesgue were both
merchants. Lets say that both men sold six items at the following prices:
5,10,15,10,5,3. Riemann would total his sales by adding the numbers as
they appear (5+10+15+10+5+3). Lebesgue, on the other hand, would sum
them like this: 2(5) + 2(10) + 15 + 3. Of course, in the end, the values are
the same.
As expected, it will be seen that the Lebesgue integral of Riemann in-
tegrable functions equals the Riemann Integral. By approaching integra-
tion in this manner Lebesgue generalized the Riemann Integral which pro-
vided mathematicians with a gateway into modern mathematics. Indeed,
the Lebesgue Integral has been instrumental in the theory of trigonometric
series, curve rectication, calculus, and probability.
However, the most immediate consequence of the Lebesgue Integral is
that it relaxes the requirements needed for the interchange of the limit and
the integral in a sequence of functions. In the Riemann Integral, only uni-
form convergence of a sequence of a functions implies that the limit of the
sequence will be Riemann Integrable. With Lebesgue integrable functions
we nd that almost everywhere convergence and boundedness are sucient
iv
for integrability of the limit. This theorem is the Lebesgue Dominated Con-
vergence Theorem (LCDT).
This paper expands on Robert G. Bartles presentation of the Lebesgue
Integral in his book Elements of Integration. Throughout the remainder of
the paper I shall denote Elements of Integration as (EOI).The third section
of this paper deals with measurable functions and measurable spaces. The
fourth section presents the notion of a measure of a set. The fth section
applies to the General Lebesgue Integral for nonnegative functions and the
Monotone Convergence Theorem. In section six, the General Lebesgue in-
tegral is extended to functions with positive and negative values. Section
seven deals with the Lebesgue Spaces. Lastly section eight encapsulates all
the various modes of convergence.
Finally, when referring to a theorem, lemma, proposition, fact, or def-
inition from another text, I use the following convention: the number of
the theorem, lemma, proposition, fact, or denition and the texts author
enclosed in brackets. For example, if I use 3.6 Lemma from Elements of
Integration, I would write . . . 3.6 Lemma [Bartle] . . . . If I refer to a
theorem, lemma, proposition, fact, or denition from this paper, I simply
write its number. Thus, if I refer to 2.3 Lemma, I would write . . . 2.3
Lemma . . . .
v
3. Measurable Functions
In dening the Lebesgue Integral we need to rst discuss classes of real-
valued functions on a set A. This set A could be the unit interval, the
natural numbers, or the entire real line.
From A, we take a family X of subsets of A which behave nicely in a
technical sense. First, well develop the idea of behaving nicely for a nite
number of subsets, then well introduce the case for a countable number of
subsets.
Denition 3.1. A non-empty collection X of subsets of A is called an
algebra of sets or a Boolean algebra if the following hold:
1. (A B) X whenever A X and B X.
2. (A)
c
X whenever A X. Note: (A)
c
denotes the complement of A
3. (A B) X whenever A , B X
Proposition 3.2. If in Denition 3.1, a non-empty collection X of subsets
of A satises (2) and (3), then it satises (1) and thus, is a Boolean
algebra. Similarly, if it satises (1) and (2), then it satises (3).
Proof. Let A , B X. Then, (A)
c
, (B)
c
X. By (3), (A)
c
(B)
c
X.
Applying De Morgans laws, (A)
c
(B)
c
= (A B)
c
. Therefore, (A B)
c
X.
Using (2) gives, [(A B)
c
]
c
= (AB) X. In a similar fashion it is seen that
(1) and (2), imply (3).
It is clear from (2) of Denition 3.1 that the and the whole set A are in
X. Also, by taking unions two at a time, it is evident that if A
1
, . . . , A
n
are sets in X, then A
1
A
2
A
n
is also in X. Several useful theorems
concerning algebras of sets will follow.
Proposition 3.3. Given any collection ( of subsets of A, there is a smallest
algebra / which contains (; that is, there is an algebra / containing ( such
that if B is any algebra containing (, then B contains /.
1
Proof. Let F be the family of all algebras (of subsets of A) that contain (.
Let / =

B : B F. Then ( is a subcollection of / since each B F
contains (. Moreover, / is an algebra. For if A and B are in /, then for each
B T we have A B and B B. Since B is an algebra, (A B) B. Since
this is true for all B : B T, we have (A B)

B : B T. Likewise,
if A /, then (A)
c
/. From the denition of /, it follows that if B is an
algebra containing (, then B /.
The smallest algebra containing ( is called the algebra generated by (.
An algebra / of sets is called a algebra or a eld, if every union
of a countable collection of sets in / is again in /. From De Morgans laws
it follows that the intersection of a countable collection of sets in / is again
in /. Modifying Proposition 3.3 gives the following:
Proposition 3.4. Given any collection ( of subsets of A, there is a smallest
-algebra / which contains (; that is, there is a -algebra / containing (
and such that if B is any -algebra containing (, then B contains /.
Proof. Let F be the family of all -algebras (of subsets of A) that contain
(. Let / =

B : B F. Then ( is a subcollection of / since each B T

contains (. Moreover, / is an -algebra. For if A
i
) belongs to /, then for
each B T we have A
i
) B. Since B is an -algebra,

_
i=1
A
i
B.
Since this is true for all B : B F, we have

_
i=1
A
i

B : B T.
From the denition of /, it follows that if B is a -algebra containing (,
then B /.
As with algebras, the smallest -algebra containing ( is called the -
algebra generated by ( and is denoted by (().
Proposition 3.5. If / is the algebra generated by (, then / and ( generate
the same -algebra.
2
Proof. Let (() be the smallest -algebra generated by (. Let /(() be
the smallest algebra generated by (. Certainly, /(() ((). Therefore,
[/(()] [(()]. But that means [/(()] ((). Also, /(() (, since
the smallest algebra generated by ( contains (. Thus, (() [(/(()].
The next proposition develops the idea of disjunctication. Given a sequence
of sets in a -algebra the sets in the sequence can be separated such that no
two sets share a common element.
Proposition 3.6. Let X be a -algebra of subsets and /
n
a sequence
of sets in X. Then there is a sequence B
n
of subsets in X such that
B
i
B
j
= for i ,= j and

_
n=1
B
n
=

_
n=1
/
n
.
Proof. Since the theorem works similarly for /
n
nite and innite, assume
/
n
to be an innite sequence. Let B
1
= /
1
and for n N with n 2, let
B
n
= /
n
(/
1
/
2
/
n1
). Thus
B
n
= /
n

_
_
k<n
/
k
_
c
.
Clearly, from the denition of B
n
, B
n
/
n
, for each n N. Also since
complements and intersections of sets in X are in X, each B
n
X. If
i > j 1, then
B
i

_
_
k<i
/
k
_
c
=

k<i
(/
k
)
c
(/
j
)
c
(B
j
)
c
.
and so B
i
B
j
= . Now,

_
n=1
B
n

_
n=1
/
n
since B
n
/
n
for each n N.
Now take an x

_
n=1
/
n
, then there exist a smallest i N such that x
belongs to /
i
and x /
_
n<i
/
n
. Therefore,
x /
i

_
n<i
/
n
= B
i
3
and thus, x

_
n=1
B
n
. This demonstrates,

_
n=1
B
n

_
n=1
/
n
. Therefore,

_
n=1
B
n
=

_
i=1
/
n
Proposition 3.7. Let f be a function dened on a set A with values in a
set . If c is any subset of , let
f
1
(c) = x A : f(x) c.
Show that f
1
() = , f
1
() = A. If c and T are subsets of , then
f
1
(cT) = f
1
(c)f
1
(T).
If c

is any non-empty collection of subsets of , then

f
1
_
_

_
=
_

f
1
(c

) , f
1
_

_
=

f
1
(c

) .
In particular, it follows that if Y is a -algebra of subsets of , then f
1
(c) :
c Y is a -algebra of subsets of A.
Proof. First, f
1
() = x A : f(x) = . Take x f
1
() thus
x A and so f
1
() A . Taking x A and applying f gives the other
containment, therefore f
1
() = A.
Next, show that f
1
(cT) = f
1
(c)f
1
(T). By denition x f
1
(cT)
if and only if f(x) cT which is equivalent to f(x) c and f(x) / T.
Now f(x) c and f(x) / T if and only if x f
1
(c) and x / f
1
(T) or
equivalently x f
1
(c)f
1
(T). Therefore, f
1
(cT) = f
1
(c)f
1
(T).
Let c

be any collection of subsets of . By denition x f

1
_
_

_
if and only if f(x)
_

or equivalently f(x) c

for some . Now

f(x) c

for some if and only if x f

1
(c

) for some which is

4
the same as x
_

f
1
(c

). Thus, f
1
_
_

_
=
_

f
1
(c

). Likewise,
f
1
_

_
=

f
1
(c

)
Proposition 3.8. Let f be function dened on a set, A, with values in a set
. Let X be a -algebra of subsets of A and let Y = c : f
1
(c) X.
Show that Y is a -algebra.
Proof. We know that the empty set is a subset of and from Proposition
3.7, f
1
() = X. Thus, the empty set is in Y.
Let c Y. From denition of Y, c is a subset of and f
1
(c) X.
Since c is a subset of , then (c)
c
(the complement of c) is a subset of .
Furthermore, f
1
(c) X implies that
_
f
1
(c)

c
X From Proposition 3.7,
_
f
1
(c)

c
= f
1
[(c)
c
] X. Thus, (c)
c
Y.
To conclude, let c

be a countable collection of subsets of Y. For

each , c

and f
1
(c

) X. Thus
_

. Since X is an
-algebra , using Proposition 3.7 gives
f
1
_
_

_
=
_

f
1
(c

) X.
Thus,
_

Y.
Therefore, we have demonstrated that Y satises the denition of a -
algebra.
Proposition 3.9. Let a, b R with a < b. Then:
1. [a, b] =

n=1
(a 1/n, b + 1/n)
2. (a, b) =

_
n=1
[a + 1/n, b 1/n]
3. [a, b) =

n=1
(a 1/n, b)
4. (a, b] =

n=1
(a, b + 1/n)
5
5. (a, ) =

_
n=1
(a, b +n).
Proof.
1. By denition, an element x is in [a, b] if and only if a x b.
Then we certainly can say a 1/n < a x b < b + 1/n for all
n N. Therefore, x (a 1/n, b + 1/n) for all n N. And now we
can conclude x

n=1
(a 1/n, b + 1/n). Containment the other way
follows easily.
2. An element x is in (a, b) if and only if a < x < b However, there
exists n N such that a < a + 1/n x b 1/n < b. There-
fore, x [a + 1/n, b 1/n] for some n N which implies that x

_
n=1
[a + 1/n, b 1/n]. Containment the other way follows.
3. (3), (4) and (5) follow similarly.
Denition 3.10. The Borel Algebra is the -algebra generated by all open
intervals (a, b) in R.
The next corollary shows that in Denition 3.10 the open intervals can
be replaced by any one of the other types of intervals.
Corollary 3.11. The Borel Algebra, B, is also generated by all the closed
intervals [a, b], or by all the half open half closed intervals [a, b), (a, b],
or by all the half-rays (a, ).
Proof. Proposition 3.9 shows that the intervals can be rewritten as the count-
able unions or intersections of open or closed intervals. The Corollary follows
from the fact that -algebra s are closed under countable unions and inter-
sections.
In the following propositions the notion of the limit (if it exists) of sets
will be formalized.
6
Proposition 3.12. Let /
n

nN
be a sequence of subsets of a set A. If /
consists of all the x A which belong to innitely many of the sets /
n
,
then,
/ =

m=1
_

_
n=m
/
n
_
.
/ is called the limsup (/
n
).
Proof. By denition of /: x / if and only if x

_
n=m
/
n
for each m N
if and only if x

m=1
_

_
n=m
/
n
_
Proposition 3.13. Let /
n

nN
be a sequence of subsets of a set A. If B
consists of all the x A which belong to all but a nite number of sets /
n
,
then
B =

_
m=1
_

n=m
/
n
_
.
B is called the liminf (/).
Proof. By denition of B: x B if and only if x

n=m
/
n
for some m

N
if and only if x

_
m=1
_

n=m
/
n
_
.
Proposition 3.14. If c
n
is sequence of subsets of a set A which is mono-
tone increasing (that is, c
1
c
2
c
3
), show that
limsupc
n
=

_
n=1
c
n
= liminf c
n
Proof. Clearly,

m=1
_

_
n=m
c
n
_

_
n=1
c
n
. Now,

m=1
_

_
n=m
c
n
_

_
n=1
c
n
follows
from the fact if x is in the countable union of c
n
s, then there exists some
j N such that x is in c
j
. Therefore,
x

_
n=1
c
n

_
n=2
c
n

_
n=3
c
n

_
n=j
c
n
.
7
But, x an element of c
j
implies that x is in c
j
c
j+1
c
j+2
since c
n

monotone increasing. Therefore, x

_
n=j+1
c
n

_
n=j+2
c
n
. Whence,
x

m=1
_

_
n=m
c
n
_
.
Now, we can conclude

_
n=1
c
n
= limsupE.
To demonstrate

_
n=1
c
n
= liminf c
n
, we notice that in a monotone increas-
ing function,

n=m
c
n
= c
m
It follows then that

_
m=1
_

n=m
c
n
_
=

_
m=1
c
m
=

_
n=1
c
n
.
Therefore, liminf c
n
=

_
n=1
c
n
= limsup c
n
.
Proposition 3.15. If T
n
is sequence of subsets of a set A which is mono-
tone decreasing (that is, T
1
T
2
T
3
), show that
limsupT
n
=

n=1
T
n
= liminf T
n
Proof. In a monotone decreasing sequence,

_
n=m
T
n
= T
m
. Consequently,

m=1
_

_
n=m
T
n
_
=

m=1
T
m
=

n=1
T
n
.
Now to show

_
m=1
_

n=m
T
n
_
=

n=1
T
n
.
Let x

n=1
T
n
, then x

_
m=1
_

n=m
T
n
_
. Now, take x

_
m=1
_

n=m
T
n
_
.
Thus for some j N, x

n=j
T
n
. Since T
n
is monotone decreasing
8
x T
j1
T
1
. Consequently, x

n=1
T
n
. Therefore, we conclude that
limsupT
n
=

n=1
T
n
= liminf T
n
.
Proposition 3.16. If /
n
is a sequence of subsets of A, then
liminf /
n
limsup/
n
A.
Proof. Clearly, liminf /
n
and limsup/
n
A are clear. Now, let x be
in liminf /
n
. Thus, there exists a j N such that x

n=j
/
n
. Therefore,
we can say that
x

_
n=1
/
n

_
n=j
/
n

_
n=j+1
/
n
.
Therefore, x limsup/
n
. And thus, liminf /
n
limsup/
n
A.
Example 3.17. Let A = [0, 1). Dene /
n
a sequence in A as
_
0,
1
n
_
when n is odd and
_
0, 1
1
n

when n is even. Notice that for all even n,

/
n
is monotone increasing and /
n
is monotone decreasing when n is
odd. Therefore,

m=1
_

_
n=m
/
2n
_
=

_
n=1
/
2n
= [0, 1) = A
by Proposition 3.14. Similarly,

_
m=1
_

n=m
/
2n1
_
=

n=1
/
2n1
=
by Proposition 3.15. Since, /
2n
and /
2n1
are subsequences of /
n

with limits A and , respectively then limsupA = A and liminf A = .

Example 3.18. Let A = (1, 1) and let /
n
=
_

1
n
, 0

when n is even
and /
n
=
_
0,
1
n
_
when n is odd. Arguing in a similar fashion as above we
quickly see that limsup/
n
= 0 = liminf /
n
. Thus, the lim/
n
= 0.
9
Proposition 3.19. If a,b,c are real numbers and mid(a,b,c) denotes the
value in the middle, then
mid(a, b, c) = infsupa, b, supa, c, supb, c.
In addition, if f
1
, f
2
, f
3
are X measurable functions on A to R and if g
is dened for x A by
g(x) = mid (f
1
(x), f
2
(x), f
3
(x)) ,
then g is X measurable.
Proof. By pairing each number with the other numbers and taking the sup
of the three pairs, the largest number will appear twice and the next largest
number will appear once. Now, taking the inf of these three numbers gives
the second largest number of the trio which is exactly the value in the
middle.
To prove the second part, we must show that / = x A : g(x) >
X, where R. To do this we will show that / is obtained from the
intersections and unions of X-measurable sets.
x A : g(x) > = x A : mid (f
1
(x), f
2
(x), f
3
(x)) >
= x A : f
1
(x) > or f
2
(x) >
x A : f
1
(x) > or f
3
(x) >
x A : f
2
(x) > or f
3
(x) >
Now we see that
A = [x A : f
1
> x A : f
2
> ]
[x A : f
1
> x A : f
3
> ]
[x A : f
2
> x A : f
3
> ]
10
Proposition 3.20. If f is a measurable function and A > 0, then the trun-
cation f
A
dened by
f
A
(x) =
_

_
f(x), if [f(x)[ A
A, if f(x) > A
A, if f(x) < A
(3.1)
is measurable. (Show directly without using Proposition 3.19).
Proof. For A,
x A : f
A
(x) > = X.
For < A,
x A : f
A
(x) > = A X.
For A < A,
x A : f
A
(x) > = x A : f(x) > X.
Remark 1. Notice that the above proposition could have been shown by
dening f
1
, f
2
, f
3
as
f
1
(x) = A, f
2
(x) = f(x), f
3
= A
Now, f
A
= mid (f
1
, f
2
, f
3
) and thus by applying Proposition 3.19, we see
that f
A
is measurable.
Proposition 3.21. Let f be a nonnegative X-measurable function on A
which is bounded (that is, there exists a constant K such that 0 f(x) K
for all x A). Then, the sequence
n
dened in Lemma 2.11 [Bartle]
converges uniformly on A to f.
Proof. From Lemma 2.11,
n
(x) =
k
2
n
for x c
kn
where
c
kn
=
_
x A :
k
2
n
f(x) <
k + 1
2
n
_
if k = 0, 1, , n2
n
1 and
c
kn
= x A : f(x) n if k = n2
n
.
11
Since f(x) K then there exists a smallest n

N where K < n

such that
for all x A,
x
n2
n
_
k=0
c
kn
Thus, for all x A and for all n N where n n

, there exists k N such

that
k
2
n
f(x) <
k + 1
2
n
which implies that

f(x)
k
2
n

<
1
2
n
. Therefore, given > 0 there exists an
n
1
N such that
[f(x)
n
(x)[ <
1
2
n
< for all n n
1
n
0
and for all x A.
Proposition 3.22. Let (A, X) be a measurable space and f a function from
A to . Let A be a collection of subsets of such that f
1
(c) X for every
set c A, then f
1
(T) X for any set T which belongs to the algebra
generated by A.
Proof. By assumption A = c : f
1
(c) X. From Proposition 3.8,
A is a -algebra . Thus, for any T in the -algebra generated by A, T is
also in A since the -algebra generated by A is A. Therefore, f
1
(T) X
for any set T that belongs to the -algebra generated by A.
Proposition 3.23. Let (A, X) be a measurable space and f be a real valued
function dened on A. Then, f is measurable if and only if f
1
(c) X
for every Borel set c.
Proof. First, assume f
1
(c) X for every Borel set c. Let /

= x A :
f(x) > = f
1
(, ) where R. Now (, ) =

_
n=1
(, + n). But
(, + n) is a Borel set and thus by assumption f
1
(, + n) X and
12
therefore,
f
1
(, ) = f
1
_

_
n=1
(, +n)
_
=

_
n=1
f
1
(, +n) X.
Consequently, since was arbitrary, /

X for all R.
Now, assume f is measurable. Let A =
_
c R : f
1
(c) X
_
. From
Proposition 3.8, A is a -algebra . Since f measurable, f
1
(, ) X for
all R. It is seen that for all open intervals in R, O, f
1
(O) X, thus by
Proposition 3.22, O A. Let O denote the collection of all open intervals
in R, therefore O A. Now, the -algebra generated by O, B (the Borel
sets), is a subset of A. Thus, f
1
(c) X for all c B.
The following fact will be useful for the next proposition.
Fact 3.24. A continuous function from R to R guarantees that for each open
interval in its range there is a corresponding open interval in its domain. To
see this apply the denition of continuity. For each p R and for > 0,
there exists a > 0 such that f(x) (f(p) , f(p) + ) for each x
(p, p+). Now since each Borel set can be written as the countable union
of open intervals then for each Borel set in the range of a continuous function
the inverse mapping of that set is a countable union of open intervals and
thus a Borel set.
Proposition 3.25. If (A, X) is a measurable space, f is a Xmeasurable
function on A to R and is a continuous function on R to R, then the
composition ( f) dened by ( f)(x) = [f(x)], is A-measurable.
Proof. By assumption is continuous, thus
1
(c) B for each c B by
Fact 3.24. Now f is measurable therefore Proposition 3.23 gives f
1
[c]
X for all Borel sets c. Since
1
(c) is a Borel set for all Borel sets c,
f
1
[
1
(x)] X for each Borel set, c. Thus, by Proposition 3.23 (f)(x)
is X-measurable .
13
Lemma 3.26. If f is a X measurable and is a Borel measurable func-
tion, then ( f) is X measurable.
Proof. From assumption, Borel measurable therefore
1
(c) B for each
c B. By Proposition 3.23, f measurable implies that f
1
[
1
(c)] X
and therefore ( f) is X-measurable since f
1
[
1
(c)] X for every
Borel set c.
Proposition 3.27. A function f on A to R is X measurable if and
only if the set /

in Lemma 2.4(a) [Bartle] belongs to X for each rational

number .
Proof. If f is X-measurable then from the denition of an X-measurable
function, /

= x A : f(x) > belongs to X for all R.

If /

X for all Q, then for each RQ there exists a

n
Q
such that
/

= x X : f(x) > =

_
n=1
x X : f(x) >
n
.
Denition 3.28. A nonempty collection Mof subsets of a set A is called
a monotone class if, for each monotone increasing sequence c
n
in Mand
each monotone decreasing sequence T
n
in M, the sets

_
n=1
c
n
,

n=1
T
n
belong to M.
Proposition 3.29. A -algebra is a monotone class.
Proof. Let X be a -algebra of A and let c
n
be an increasing sequence
of sets in X. By denition of X,

n=1
c
n
X for all n N. Now let
T
n
be a decreasing sequence of sets in X, thus (T
n
)
c
X for each n N.
14
Therefore,

_
n=1
(T
n
)
c
X, by denition of X.
Whence
_

_
n=1
(T
n
)
c
_
c
X, by denition of X.
Applying DeMorgans Laws,

n=1
T
n
=

n=1
[(T
n
)
c
]
c
=
_

_
n=1
(T
n
)
c
_
c
X.
Therefore, X is a monotone class.
Proposition 3.30. If / is a nonempty collection of subsets of A, then there
is a smallest monotone class containing /.
Proof. Let F be the family of all monotone classes of A that contain /. Let
M =

B : B F. By denition of M, M /. Now let c
n
be an
increasing sequence of sets in M. Therefore, c
n
B for each B F.
For all B F,

n=1
c
n
B since for B F, B is a monotone class. Thus

n=1
c
n
M. Similarly

n=1
T
n
M for T
n
a decreasing sequence
in M. As a result, M is a monotone class and if B is a monotone class
containing / then B M.
15
4. Measures
Denition 4.1. A measure is an extended real-valued function dened
on a -algebra X of subsets of A such that
(i) () = 0
(ii) (c) 0 for all c X
(iii) is countably additive in the sense that if c
n
is any disjoint sequence
of sets in X, then

_

_
n=1
c
n
_
=

n=1
(c
n
) . (4.1)
Proposition 4.2. If is a measure on X and / is a xed set in X, then
the function , dened for c X by (c) = (/ c) is a measure on X.
Proof. First, (/ ) = . Therefore Condition (i) of Denition 4.1 is satis-
ed since,
() = (/ ) = () = 0.
Next, for all c X, (/ c) X since X is a -algebra . Thus, condition
(ii) of Denition 4.1 is satised since
(c) = (/ c) 0 for all c X.
Lastly, let c
n
be a disjoint sequence in X. Then (c
i
/) (c
j
/) =
for all i ,= j. Also,
_
/

_
n=1
c
n
_
=

_
n=1
(/ c
n
) .
Therefore we can say,

_

_
n=1
c
n
_
=
_

_
n=1
[/ c
n
]
_
=

n=1
(/ c
n
) =

n=1
(c
n
) .
Thus satises Conditions (i), (ii) and (iii) of Denition 4.1.
The following fact, from elementary Calculus, will be useful in the proof
of the next two propositions.
16
Fact 4.3. If

n=1
a
n
and

n=1
b
n
are convergent series of real numbers, then

n=1
(a
n
+b
n
) =

n=1
a
n
+

n=1
b
n
.
Thus, if

n=1
a
j
n
are a convergent series of real numbers for 1 j N, then

n=1
N

j=1
a
j
n
=
N

j=1

n=1
a
j
n
.
Proposition 4.4. If
1
, ,
n
are measures on X and a
1
, , a
n
are non-
negative real numbers, then the function , dened for c X by
(c) =
n

j=1
a
j

j
(c) (4.2)
is a measure on X.
Proof. Conditions (i) and (ii) follow from the fact that
1
, ,
n
are mea-
sures on X and that a
1
, , a
n
are nonnegative real numbers. By plugging

_
m=1
c
m
into (4.2) where E
m
are disjoint and by Fact 4.3 we get

_

_
m=1
c
m
_
=
n

j=1
a
j

m=1

j
(c
m
) =
n

j=1

m=1
a
j

j
(c
m
) =

m=1
n

j=1
a
j

j
(c
m
) .
Therefore, is a measure.
Proposition 4.5. If
n
is a sequence of measures on X with
n
(A) = 1
and if is dened by
(c) =

n=1
2
n

n
(c) , c X. (4.3)
then is a measure on X and (A) = 1.
Proof. Clearly (i) and (ii) hold for Denition 4.1. To show

_
_

_
j=1
c
n
_
_
=

j=1
(c
j
)
17
where c
j
are disjoint, we will show that the left hand side is greater than
the right hand side then we will show that the right hand side is greater
than the left hand side which implies equality.
Let c
j
be a disjoint sequence from X. Fix N N. Then by Fact 4.3
N

j=1
(c
j
) =
N

j=1

n=1

n
(c
j
)
2
n
=

n=1
N

j=1

n
(c
j
)
2
n
.
Now for each n,
n
is a measure, therefore,

n=1
N

j=1

n
(c
j
)
2
n
=

n=1

n
_

N
j=1
c
j
_
2
n

n=1

n
_

j=1
c
j
_
2
n
=
_
_

_
j=1
c
j
_
_
.
Letting N go to innity we see that

j=1
(c
j
)
_
_

_
j=1
c
j
_
_
.
Now x > 0. Since
n
is a measure for each n and
n
(A) = 1 for all n
there exists a N() N such that

_
_

_
j=1
c
j
_
_
=

n=1

n
_

j=1
c
j
_
2
n

N()

n=1

n
_

j=1
c
j
_
2
n
.
By Proposition 4.4,
N()

n=1

n
_

j=1
c
j
_
2
n
=

j=1
N()

n=1

n
(c
j
)
2
n

j=1

n=1

n
(c
j
)
2
n
=

j=1
(c
j
) .
Since was arbitrary

_
_

_
j=1
c
j
_
_

j=1
(c
j
) .
This concludes the proof.
Proposition 4.6. Let A = N and let X be the -algebra of all subsets of N.
If
n
is a sequence of nonnegative real numbers and if we dene : A

R
by
() = 0; (c) =

nE

n
, c ,= ,
then is a measure.
18
Proof. Conditions (i) and (ii) are obvious. Let c
m
be a disjoint sequence
in X. If
_

_
m=1
c
m
_
= then done. If this is not the case, then

m=1
Em

n
is unconditionally convergent since
n
is nonnegative for each n. (In other
words, we can rearrange the order of the series). Therefore,

_

_
m=1
c
m
_
=

m=1
Em

n
=

nE
1
E
2

n
=

nE
1

n
+

nE
2

n
+ =

m=1
(c
m
) .
Thus Condition (iii) is satised.
Proposition 4.7. If A is an uncountable set and if X is the family of all
subsets of A, then on c in X dened by
(c) = 0 if c is countable
(c) = if c is uncountable.
is a measure on X.
Proof. Condition (i) holds since the is countable, therefore () = 0. Con-
dition (ii) follows directly from the denition of . For Condition (iii), let
c
n
be a disjoint sequence in X. If for all n N, c
n
countable, then

n=1
c
n
is countable and

_

_
n=1
c
n
_
= 0 =

n=1
(c
n
) .
If for any n N, c
n
is uncountable, then

n=1
c
n
is uncountable and thus

_

_
n=1
c
n
_
= =

n=1
(c
n
) .
19
Proposition 4.8. Let A = N and let X be the family of all subsets of N.
If c is nite, let (c) = 0: if c is innite, let (c) = . Then, is not a
measure on X.
Proof. Let c
n
be a disjoint sequence in X where for all n N, c
n
is
nite. This means that

n=1
c
n
is innite and therefore (

n=1
c
n
) = .
However (c
n
) = 0 for all n N and thus

n=1
(c
n
) = 0.
Proposition 4.9. Let (A, X, ) be a measurable space and let c
n
be a
sequence in X. Then
(liminf c
n
) liminf (c
n
) . (4.4)
Proof. Let c
n
be a sequence in X. It is clear that
_

n=m
c
n
_

m=1
is an
increasing sequence. Therefore by Lemma 3.4(a) [Bartle],

_

_
m=1
_

n=m
c
n
__
= lim
m

n=m
c
n
_
. (4.5)
Moreover, for all n, m N such that n m, c
n

n=m
c
n
. Thus by
Lemma (3.3) [Bartle] (

n=m
c
n
) (c
n
) for all n m which implies that
lim
m

n=m
c
n
_
liminf (c
n
) . (4.6)
Combining (4.5) and (4.6) gives (4.4).
Proposition 4.10. If
_
c
n
_
< , then
limsup(c
n
) (limsupc
n
) . (4.7)
Proof. Let c
n
be a sequence such that

n=1
c
n
< . Notice that

n=m
c
n

m=1
is a decreasing sequence. Thus, by Lemma 3.4(b) EOI,

m=1
_

_
n=m
c
n
__
= lim
m

_

_
n=m
c
n
_
. (4.8)
Moreover, for all n, m N with n m, c
n

n=m
c
n
which implies that
limsup(c
n
) lim
m

_

_
n=m
c
n
_
. (4.9)
Combining (4.8) and (4.9) gives (4.7).
20
Proposition 4.11. If (A, X, ) a measure space and c
n
is a sequence in
X, then

_

_
n=1
c
n
_

n=1
(c
n
) (4.10)
Proof. Let T
n
= c
n

n1
j=1
c
j
when n > 1 and c
1
= T
1
(the disjunctication
of c
n
). From Proposition 3.6, we know that

n=1
c
n
=

n=1
T
n
therefore

_

_
n=1
c
n
_
=

n=1
(T
n
)

n=1
(c
n
) ,
since T
n
c
n
for each n.
Proposition 4.12. Let (A, X, ) be a measure space and let
Z = c X : (c) = 0 .
Then Z is not a -algebra .
Proof. Let c be in Z, thus Ac X. If (A) = 0, then (c)
c
Z. However,
in general (c)
c
/ Z since (Ac) = (A) (c).
Though Z is not a -algebra Proposition 4.13 will show that it does have
some important properties that will be useful in the completion of X.
Proposition 4.13.
(i) Let c X and let T Z, then c T Z
(ii) Let c
n
be in Z for n N, then (

n=1
c
n
) Z.
Proof. To see (i) let c X and let T Z, then c T X. Moreover,
(c T) (c) = 0. Therefore, c T Z.
To see (ii) let c
n
be in Z for n N then (c
n
) = 0 for each n N. Thus

n=1
(c
n
) = 0. But by Proposition 4.11,

_

_
n=1
c
n
_

n=1
(c
n
) = 0.
Consequently, (

n=1
c
n
) Z.
21
Proposition 4.14. Let A, X, , and Z be as in Proposition 4.12. Let X

be the family of all subsets of A of the form

(c Z
1
) Z
2
, c X (4.11)
where Z
1
and Z
2
are arbitrary subsets of sets belonging to Z. A set is in
X

if and only if it has the form

c

Z where

c X and

Z is a subset of
a set in Z.
Proof. Let T be set in X. Therefore, T has the form of (4.11). Let / be
the set in Z such that Z
1
/ and let B be the set in Z such that Z
2
B.
Thus,
T = (c Z
1
) Z
2
= cZ
2
Z
1
Z
2
.
where Z
1
/ X and Z
2
B X. Now cZ
2
= (cB) (c BZ
2
).
Therefore
cZ
2
Z
1
Z
2
= cB [c BZ
2
] [Z
1
Z
2
] .
Notice that cB X, therefore let cB =

c. Observing that (c BZ
2
)
B Z and (Z
1
Z
2
) / Z, we let

Z = [c BZ
2
] [Z
1
Z
2
]. Therefore,
T = (c Z
1
) Z
2
=

c

Z, where

c X and

Z is a subset of a set in Z.
To conclude,

c

Z =
_

c

Z
_
, which shows the other implication.
Corollary 4.15. As dened in Proposition 4.14, X

is a -algebra .
Proof. Obviously, is in X

. Let T be in X

. Therefore, T = (c Z
1
) Z
2
where c X, Z
1
/ Z and Z
2
B Z. Now,
(T)
c
= [(c Z
1
) Z
2
]
c
= [(c)
c
(Z
1
)
c
] Z
2
Notice that
(c)
c
(Z
1
)
c
= (c)
c
(/)
c
[/Z
1
(c)
c
] .
Therefore,
(T)
c
= [(c)
c
(/)
c
] [/Z
1
(c)
c
] Z
2
.
22
We see that (c)
c
(/)
c
X and [/Z
1
(c)
c
] Z
2
/ B Z.
Therefore, by Proposition 4.14, (T)
c
X. Finally, let T
n
be a sequence
in X, therefore T
n
= c
n
Z
n
where c
n
X and Z
n
/
n
Z. Now

_
n=1
(c
n
Z
n
) =

_
n=1
(c
n
)

_
n=1
(Z
n
) .
It is clear that

n=1
(c
n
) Xand

n=1
(Z
n
)

n=1
(/
n
) Z. Thus, X
is a -algebra .
Proposition 4.16. Let be dened on X by
(c Z) = (c) , (4.12)
where c X and Z is a subset of a set with -measure zero. Then is
well-dened and a measure on X which agrees with on X. This measure
is called the completion of .
Proof. First, we claim that if (c Z
1
) = (T Z
2
) where c, T X and
Z
i
/
i
with (/
i
) = 0 for i = 1, 2, then (c) = (T). To see this
let (c Z
1
) = (T Z
2
). Therefore c (T /
2
) and thus by Lemma 3.3
[Bartle] and Proposition 4.14, (c) (T) + (/
2
) = (T) . Similarly,
we see that, (T) (c). Therefore, (c) = (T) and our claim is
demonstrated. Thus, is well-dened.
Now, we demonstrate that (c Z) = (c) is a measure by showing that
it satises the three conditions of Denition 4.1. Condition (i) follows from
the fact that ( Z) = () = 0. Condition (ii) holds since (c Z) =
(c) 0 for all (c Z) X. And nally, let c
n
Z
n
be a sequence in
X.

_

_
n=1
(c
n
Z
n
)
_
=
_

_
n=1
c
n

_
n=1
Z
n
_
=

n=1
(c
n
) =

n=1
(c
n
Z
n
) .
23
Proposition 4.17. Let (A, X, ) be a measure space and let (A, X, ) be
its completion in the sense of Proposition 4.16. Suppose that f is an X-
measurable function on A to the extended real line. Then, there exists a
X-measurable function g on A to the extended real line which is -almost
everywhere equal to f.
Proof. For each rational number r, let /
r
= x A : f(x) > r. Since f
is X-measurable, /
r
X for each rational number. Now we may write
/
r
= c
r
Z
r
where c
r
X and Z
r
a subset of a set in Z. Let Z be in Z
such that

Z
r
is a subset of Z. Furthermore, we dene g(x) = f(x) for all
x / Z and g(x) = 0 for all x Z. Therefore, g(x) is -almost equal to f(x).
To see that g(x) is X-measurable, we show that x A : g(x) > r X
for all r Q and then use Proposition 3.27.
If r 0, then
x A : g(x) > r = x A : f(x) > r Z
= (c
r
Z
r
) Z = (c
r
Z ) X.
If r < 0, then
x A : g(x) > r = x A : g(x) > r Z
= (c
r
Z
r
) Z = (c
r
Z) X.
This completes the proof.
Proposition 4.18. If is a charge on X, then Lemma 3.4 [Bartle] holds.
Proof. Lemma 3.4 does not depend on the fact that (c
n
) 0 for all c
n

X, therefore it holds for a charge.
Proposition 4.19. If is a charge on X and is dened for c X as
(c) = sup(A) : A c, A X , (4.13)
is a measure on X.
24
Proof. Clearly, (i) and (ii) from Denition 4.1 hold. To show (iii) we will
argue like Proposition 4.5. Let c
n
be a pairwise disjoint sequence from
X. Fix /

n=1
c
n
with / X. Therefore,
(/) =
_

_
n=1
c
n
/
_
=

n=1
(/ c
n
)

n=1
(c
n
) .
So
(/)

n=1
(c
n
) . (4.14)
In (4.14), taking the sup over all such /s gives

_

_
n=1
c
n
_

n=1
(c
n
) . (4.15)
From (4.15), if
_

_
n=1
c
n
_
= then the proof is done. So, assume

_

_
n=1
c
n
_
< and x > 0. For each j N,
(c
j
)
_

_
n=1
c
n
_
< .
Therefore, there exists /
j
X with /
j
c
j
and
(c
j
)

2
j
(/
j
).
Thus,

_

_
n=1
c
n
_

_

_
n=1
/
n
_
=

n=1
(/
n
)

n=1
_
(c
n
)

2
n
_
=

n=1
[(c
n
)]
_

n=1

2
n
_
=

n=1
[(c
n
)] .
Since > 0 was arbitrary:

_

_
n=1
c
n
_

n=1
(c
n
).
This nishes the proof.
Proposition 4.20. Let denote the Lebesgue measure dened on the Borel
algebra, B of R.
25
(i) If c consists of a single point, then c B and (c) = 0.
(ii) If c is countable, then c B and (c) = 0.
(iii) The open interval (a, b), the half-open intervals (a, b], [a, b), and [a, b]
all have Lebesgue measure b a.
Proof. Let c = x where x R. Let /
n
= (x
1
n
, x +
1
n
) for each n N.
Clearly, /
n
B for each n N, thus c =

n=1
(x
1
n
, x +
1
n
) B. Now,
/
n
is decreasing therefore from Lemma 3.4(b) [Bartle]
(c) =
_

n=1
/
n
_
= lim
n
(/
n
) = lim
n
2
n
= 0.
This shows (i).
Now, to show (ii) let c = x
n

nN
and x
n
,= x
m
for n ,= m. So by (i)
and countable additivity of a measure we get (ii). For each x
n
c dene
/
nm
= (x
n

1
m
, x
n
+
1
m
). Thus,

m=1
/
nm
= x
n
B and therefore c =

_
n=1
_

m=1
/
nm
_
B.
Now, we see that since

m=1
/
im

m=1
/
jm
= when i ,= j,
(c) =
_

_
n=1
_

m=1
/
nm
__
=

n=1

m=1
/
nm
_
= 0.
This demonstrates (ii).
Lastly, by denition of , (a, b) = b a. From Proposition 3.9, (a, b] =

n=1
(a, b +
1
n
) B. Therefore,
(a, b] =
_

n=1
(a, b +
1
n
)
_
= lim
n
(a, b +
1
n
)
= lim
n
(b +
1
n
a) = b a.
Likewise, for [a, b) and [a, b]. This shows (iii).
Proposition 4.21. If denotes the Lebesgue measure and c ,= is an
open subset of R, then (c) > 0. Also if / is compact subset of R, then
(/) < .
26
Proof. From Theorem 3.1.13 [Stoll] there exits a nite or countable collection
1
n
of pairwise disjoint open intervals such that c =

n
1
n
for c an open
subset of R . Therefore, for c open c B. Also since c open for every point
p in c we can nd an > 0 such that, (p

2
, p +

2
) c. Applying Lemma
3.3 [Bartle], we get,
0 <
_
p

2
, p +

2
_
(c).
Therefore, (c) > 0.
Let / be compact. Therefore, by Heine-Borel Theorem, / is bounded,
i.e. there exists a positive constant M such that / [M, M]. Thus
(/) ([M, M] = 2M < .
This nishes the proof.
In the following example, we vary the Cantor set such that we obtain a
set of positive Lebesgue measure that contains no non-void open interval.
Example 4.22. The Fat Cantor Set,

P, is constructed like the Cantor set
except the open intervals removed at the n
th
step have length 3
n
, 0 <
< 1. Thus,

P =

n=1
R
c
n
where R
n
is the part removed on the n
th
step.
Note that

N
n=1
R
c
n
is the disjoint union of 2
N
closed intervals, each of length
less than
1
2
N
. Therefore, if E (a non-void open interval) is in

P, then it sits
in one of these intervals with length less than
1
2
n
for all n N. This implies
that (E) = 0 which is a contradiction of Proposition 4.21. Finally, it is
clear that
((

P)
c
) =
1
3

n=0
_
2
3
_
n
= .
Therefore, (

P) = 1 > 0.
In the following example the almost everywhere limit of a sequence of
measurable functions is not measurable.
27
Example 4.23. Let c be a subset of a set ^ X with (^) = 0, but
c / X. Let f
n
= 0 for all n N. Thus, lim
n
f
n
= X
E
(the characteristic
equation of c) almost everywhere. Therefore, the almost everywhere limit
of f
n
equals a non-measurable function.
28
5. The Integral
Denition 5.1. If f belongs to M
+
(A, X), we dene the integral of f with
respect to to be the extended real number
_
f d = sup
_
d,
where the supremum is extended over all simple functions in M
+
(A, X)
satisfying 0 (x) f(x) for all x A. If f belongs to M
+
(A, X) and c
belongs to X, then f1
E
belongs to M
+
(A, X) and we dene the integral of
f over c with respect to to be the extended real number
_
E
f d =
_
f1
E
d.
Proposition 5.2. If the simple function in M
+
(A, X) has the (not nec-
essarily standard representation)
=
m

k=1
b
k
X
F
k
, (5.1)
where b
k
R and T
k
, then
_
d =
m

k=1
b
k
(T
k
) . (5.2)
Proof. If has standard representation then done. Otherwise by rewriting
(5.1), we get
= b
1
X
F
1
+ +b
m
X
Fm
.
Therefore,
_
d =
_
(b
1
X
F
1
+ +b
m
X
Fm
) d
=
_
b
1
X
F
1
d + +
_
b
m
X
Fm
d (by Lemma 4.3(a) ([Bartle])
= b
1
(T
1
) + +b
m
(T
m
) (since b
k
X
F
k
is a simple function.)
=
m

k=1
b
k
(T
k
)
29
Proposition 5.3. If
1
and
2
are simple functions in M
+
(A, X), then
= sup
1
,
2
, = inf
1
,
2

are also simple functions in M

+
(A, X).
Proof. Let = sup
1
,
2
. Therefore,
=
1
1
{xX:
1

2
}
+
2
1
{xX:
2
>
1
}
Thus is a simple function since it is the sum of two simple functions.
Likewise for .
The following is a dierent proof of Corollary 4.7(a) from [Bartle]. We will
attain the same result without using the Monotone Convergence Theorem.
Proposition 5.4. If f M
+
(A, X) and c > 0, then
_
cf d = c
_
f d. (5.3)
Proof. Let M
+
(A, X) be a simple function such that f and let
M
+
(A, X) be a simple function such that cf. The map = c
is clearly a one to one function since c
1
= c
2
implies that
1
=
2
.
Therefore,
sup
cf
_
d = sup
ccf
_
c d = c sup
f
_
d.
This gives (5.3).
In problems, 4.E and 4.F from Elements of Integration, Bartle attempts
to show another proof for 4.7 Corollary [Bartle] without using the Monotone
Convergence Theorem. However, as stated Problem 4.E is false and therefore
Problem 4.F cannot be completed. I will state 4.E, show that it is false by
counterexample, then I will show 4.F assuming (incorrectly) that 4.E is true.
Proposition 5.5. Let f, g belong to M
+
(A, X), let be a simple function
in M
+
(A, X) with f, and let be a simple function in M
+
(A, X) with
f +g. Let
1
= inf , and let
2
= sup , 0. Then
30
(i) =
1
+
2
(ii)
1
f
(iii)
2
g.
Proof. First, conditions (i) and (ii) always hold.
(i) If
1
= inf , = , then
2
= sup , 0 = 0. Therefore,

1
+
2
= . If
1
= inf , = , then
2
= sup , 0 = .
Therefore,
1
+
2
= + = .
(ii) If
1
= , then
1
f. If
1
= , then f.
Now, condition (iii) holds if inf , = since inf , = implies that
sup , 0 = 0 and 0 g. However if inf , = the conclusion need
not be true. Let = 11
[0, 1]
, f = 21
[0, 1]
, = 41
[0, 1]
and f +g = 4.0011
[0, 1]
.
Therefore, f and f +g. But

2
= = 3
[0, 1]
2.0011
[0, 1]
= g.
Assuming Proposition 5.5 is true, Ill show the following proposition.
Proposition 5.6. Using Proposition 5.5, if f, g belong to M
+
(A, X), then
(i) f +g M
+
(A, X)
(ii)
_
(f +g) d =
_
f d +
_
g d.
Proof. From 2.6 Lemma [Bartle] f + g M
+
(A, X) Furthermore, by the
denition of the integral with respect to ,
sup

_
d =
_
(f +g) d where the simple function f +g.
Moreover, from Proposition 5.5, 4.5 Lemma [Bartle] and 4.3 Lemma [Bartle],
_
f d +
_
g d
_

1
d +
_

2
d =
_
(
1
+
2
) d =
_
d.
Therefore, from the denition of the integral and by taking the supremum
over all simple functions f +g.
_
f +g d =
_
f d +
_
g d.
31
Proposition 5.7. Let A = N, let X be all subsets of N, and let be the
counting measure on X. If f is a nonnegative function on N, then f
M
+
(A, X) and
_
f d =

n=1
f(n).
Proof. Let f
N
=
N

n=1
f(n)1
[n,n+1)
. Clearly, f
N
monotone increasing to f a
function on N. Applying the MCT we get,
_
f d = lim
N
_

n=1
f(n)1
[n,n+1)
= lim
N
N

n=1
f(n) =

n=1
f(n).
Example 5.8. Let A = R, X = B, and let be the Lebesgue measure on
X. Let f
n
= 1
[0, n]
. Clearly, f
n
converges to f = 1
[0, ]
. Therefore, by
Monotone Convergence we have,
_
fd =
_
f
n
d = . Note the MCT can
be applied here since f
n
M
+
(A, X) which means that
_
f
n
d will always
have a limit in the extended real line.
Example 5.9. Let A = R, X = B, and let be the Lebesgue measure on
X. Let f
n
=
1
n
1
[n, ]
. Obviously f
n
is monotone decreasing to f = 0. In
fact, f
n
converges uniformly to f = 0 since given > 0 there exists a n
0
N
such that,
[f
n
(x) f(x)[

1
n

=
1
n
<
for all n n
0
and for all x A. However,
_
fd = 0 ,= = lim
n
_
f
n
d
This example does not contradict the MCT because f
n
is a decreasing se-
quence.
32
Example 5.10. Let A = R, X = B, and let be the measure on X. Let
f
n
=
1
n
1
[0, n]
. Like Example 5.9, f
n
converges uniformly to f = 0 and
_
fd = 0 ,= 1 = lim
n
_
f
n
d.
Again, this example does not contradict the MCT since f
n
is not monotone
increasing. But we can apply Fatous Theorem which gives
_
(liminf f
n
) d = 0 1 = liminf
_
f
n
d.
Example 5.11. Let g
n
= n1
[
1
n
,
2
n
]
and let g = 0. Now, limg
n
(x) = 0 for
each x A. Therefore, g
n
converges everywhere to g but not uniformly.
However, MTC does not apply since g
n
is not monotone increasing. Fatous
Theorem though does apply since g
n
M
+
(A, X).
Proposition 5.12. If (A, X, ) is a nite measure space, and if f
n
is a
sequence of real-valued functions in M
+
(A, X) which converges uniformly
to a function f, then f belongs to M
+
(A, X) and
_
f d = lim
n
_
f
n
d. (5.4)
Proof. By Corollary 2.10 from [Bartle], f M
+
(A, X) since f
n
converges
uniformly to f. Also, f
n
converges uniformly to f and (A, X, ) nite
implies that given > 0 there exists an n
o
such that
f(x)

(A)
< f
n
(x) < f(x) +

(A)
.
for all n n
o
and for all x A. Thus, by 4.5 Lemma [Bartle] and 4.7
Corollary [Bartle],
_
f(x) d
_

(A)
d <
_
f
n
(x) d <
_
f(x) d +
_

(A)
d.
Whence,

_
f
n
(x) d
_
f(x) d

<
(A)
(A)
= .
This gives (5.4).
33
Proposition 5.13. Let A be a closed interval [a, b] in R, let be the
Lebesgue measure on X. If f is a nonnegative continuous function on A,
then
_
f d =
_
b
a
f(x) dx, (5.5)
where the right side denotes the Riemann integral of f.
Proof. Let be a nonnegative step function on [a, b]. Therefore, there
exists, T, a partition of [a, b] where a = x
0
< x
1
< < x
n
= b such that
for i N
n
takes on one value in the interval [x
i1
, x
i
). Let a
i
0 denote
this value. Thus,
(x) = a
1
1
[x
0
, x
1
)
+a
2
1
[x
1
, x
2
)
+ +a
n
1
[x
n1
, xn)
is a simple function. Therefore,
_
(x) d = a
1
([x
0
, x
1
) + +a
n
([x
n1
, x
n
)
=
n

i=1
a
i
(x
i
x
i1
) =
_
b
a
(x) dx.
Thus, we have shown the proposition for nonnegative step functions.
Now let f be continuous and nonnegative. Thus,
_
b
a
f(x) dx = sup
P
L(T, f) : T a partition of [a, b] .
By denition,
L(T, f) =
n

i=1
m
i
x
i
,
where m
i
= inf f(t) : f(x
i1
) f(t) f(x
i
) and T a partition of [a, b].
Therefore,
L(T, f) =
_
d
where (x) f(x) and (x) a simple function. Thus
_
b
a
f(x) dx = sup

_
d =
_
f d.
34
Proposition 5.14. Let A = [0, ), let X be the Borel subsets and let
be the Lebesgue measure on X. If f is a nonnegative continuous function
on A, show that
_
X
f d = lim
b
_
b
0
f(x) dx.
Proof. By Proposition 5.13,
_
[0, b]
f d =
_
b
0
f(x) dx.
Therefore,
lim
b
_
b
0
f(x) dx = lim
b
_
[0, b]
f d.
We conclude the proof by showing,
lim
b
_
[0, b]
f d =
_
X
f d.
The above is true however by the MCT since f1
[0,b]
is monotone increasing
to f.
Proposition 5.15. If f M
+
(A, X) and
_
f d < ,
then x A : f(x) = = 0.
Proof. Let c
n
= x A : f(x) n. Therefore, x A : f(x) = =

n=1
c
n
. Clearly, c
n
is decreasing. Thus, by 3.4(b) Lemma [Bartle],
x A : f(x) = =
_

n=1
c
n
_
= lim
n
(c
n
)
provided (c
1
) < . For each n N, n1
En
f. By 4.5(a) Lemma [Bartle]
_
n1
En

_
f d < for each n N. Now n(c
n
) < for each n. Thus,
(c
1
) < . To nish the proof, it will be shown that lim(c
n
) = 0. It
has already been established that n(c
n
)
_
f d for all n. Therefore,
(c
n
)
1
n
_
f d for all n. This implies that
lim(c
n
)
__
f d
_
lim
n
1
n
= 0.
35
Therefore,
_

n=1
c
n
_
= 0.
Proposition 5.16. If f M
+
(A, X) and
_
f d < ,
then the set ^ = x A : f(x) > 0 is -nite (that is, there exists a se-
quence, T
n

nN
X, such that ^

T
n
and (T
n
) < ).
Proof. Let T
n
=
_
x A : f(x) >
1
n
_
. For each n, T
n
X, since f
M
+
(A, X). Therefore,

n=1
T
n
X. Now, if x ^, then f(x) > 0.
But, there exists a n such that f(x) >
1
n
. Thus, x

n=1
T
n
. Therefore,
^

n=1
T
n
. By construction,
1
n
1
Fn
< f for each n. Thus,
_
1
n
1
Fn
d <
_
f d < .
Clearly,
1
n
(T
n
) < and therefore (T
n
) < .
Proposition 5.17. If f M
+
(A, X) and
_
f d < ,
then for any > 0 there exists a set c X such that (c) < and
_
f d
_
E
f d +.
Proof. Let c
n
be a disjoint sequence in X. Therefore
_

n=1
En
f d =

n=1
_
En
f d = lim
m
m

n=1
_
En
f d
_
f d.
The series

n=1
_
En
f d is increasing. Therefore given an > 0 there exist
a n
0
such that
_
f d
n
0

n=1
_
En
f d =
_

n
0
n=1
En
f d.
for all n n
o
. By letting c =

n
0
n=1
c
n
the proposition is proved.
36
Proposition 5.18. If f
n
M
+
(A, X), f
n
converges to f almost ev-
erywhere, and
_
f d = lim
_
f
n
d < ,
then
_
E
f d = lim
_
E
f
n
d
for each c A.
Proof. Since f
n
converges to f, f
n
1
E
converges to f1
E
. By Fatous Theorem,
_
E
(liminf f
n
) d =
_
E
f d liminf
_
E
f
n
d. (5.6)
By applying Fatous Theorem to f
n
f
n
1
E
it is seen that,
_
_
liminf (f
n
f
n
1
E
)
_
d liminf
_
(f
n
f
n
1
E
) d.
Since limf
n
and lim
_
f
n
exist, the limits can be pushed through the paren-
thesis to give,
_
(f f1
E
) d
_
f d + liminf
_

_
f
n
1
E
d
_
.
By subtracting
_
f d and applying a property of liminf,

_
f1
E
d limsup
__
f
n
1
E
d
_
.
Therefore,
_
f1
E
d limsup
__
f
n
1
E
d
_
. (5.7)
By combining (5.6) and (5.7), the proof is concluded.
37
6. Integrable Functions
In Chapter 5 from [Bartle], the General Lebesgue Integral is established
for functions with negative values. More specically, L(A, X, ) is dened
as the collection of all X-measurable real valued functions such that both
the positive and negative parts have nite integrals.
Proposition 6.1.
(a) If f L(A, X, ) and a > 0, then the set x A : [f(x)[ a has
nite measure.
(b) If f L(A, X, ) and a > 0, then the set x A : f(x) ,= 0 has
-nite measure (that is, the union of measurable sets with nite mea-
sure).
Proof. Let T
a
= x A : f
+
(x) a and let ^
a
= x A : f

(x) a.
Therefore, x A : [f(x)[ a = T
a
^
a
. Now a1
Pa
f
+
. Thus by
Lemma 4.5 (a) and since f L(A, X, ),
a(T
a
)
_
f
+
d < .
Likewise (^
a
) < . Therefore, for a > 0
(x A : [f(x)[ a) = (T
a
^
a
) < .
For (b), notice that
x A : f(x) ,= 0 =
_
x A : f
+
(x) > 0
_

_
x A : f

(x) > 0
_
.
Since f L(A, X, ), then (b) follows from Proposition 5.16
Proposition 6.2. If f is X-measurable function and if f(x) = 0 for -
almost all x X, then f L(A, X, ) and
_
f d = 0. (6.1)
Proof. Since f = 0 -almost everywhere, then f
+
= 0, f

= 0 -almost
everywhere. From Corollary 4.19 [Bartle],
_
f
+
d = 0,
_
f

d = 0. There-
fore, f L(A, X, ) and (6.1) holds.
38
Proposition 6.3. If f L(A, X, ) and g is an X-measurable real val-
ued function such that f(x) = g(x) almost everywhere on A, then g
L(A, X, ) and
_
f d =
_
g d.
Proof. Let ^ = x A : f(x) ,= g(x). Therefore, by assumption
(^) = 0. Thus
_
[g[ d =
_
X\N
[g[ d +
_
N
[g[ d
=
_
X\N
[f[ d +
_
N
[g[ d =
_
X\N
[f[ d < .
Thus, by 5.3 Theorem [Bartle] g L(A, X, ). Finally, arguing as above,
_
X
g d =
_
X\N
g d +
_
N
g d =
_
X\N
f d +
_
N
g d
=
_
X
f d
_
N
f d +
_
N
g d =
_
X
f d.
This completes the proof.
Proposition 6.4. If f L(A, X, ) and > 0, then there exists a X
measurable simple function such that
_
[f [ d < . (6.2)
Proof. By Lemma 2.11 [Bartle] and MCT there exists a
+
and

in
M
+
(A, X) such that
_

f
+

d < and
_

d < .
Dene T = x A : f(x) 0 and ^ = x A : f(x) < 0. Certainly,
T ^ = A and T ^ = . Therefore,
_
[f [ d =
_
P
[f [ d +
_
N
[f [ d
=
_

f
+

1
P
d +
_

1
N
d < .
39
Proposition 6.5. If f L(A, X, ) and g is a bounded measurable func-
tion, then the product fg also belongs to L(A, X, ).
Proof. By assumption g bounded therefore there exists a constant K such
that [g[ K. By 5.5 Theorem [Bartle], Kf is integrable. By 2.6 Lemma
[Bartle], fg is measurable. Clearly, [fg[ [Kf[, therefore by 5.4 Corollary,
fg is in L(A, X, ).
Proposition 6.6. Suppose that f is in L(A, X, ) and that its indenite
integral is
(c) =
_
E
f d, c A.
Then
(a) (c) 0 for all c X if and only if f(x) 0 for almost all x A.
(b) (c) = 0 for all c if and only if f(x) = 0 for almost all x A.
Proof.
(a) Assume (c) 0 for all c X. Let ^ = x A : f(x) < 0.
Therefore, (^) =
_
N
f
+
d
_
N
f

d. Clearly
_
N
f
+
d = 0, thus
_
N
f

d = 0 for all x ^. Since

_
N
f

d = 0 by 4.10 Corollary,
f

= 0 for almost all x X. Whence, f = f

+
for almost all x A
and thus f 0 for almost all x A.
Now suppose f 0 for almost all x A. Then f

= 0 almost
everywhere. By 4.10 Corollary [Bartle],
_
f

d = 0 For all c X,
(c) =
_
f1
E
d =
_
f
+
1
E
d 0.
(b) Dene T = x A : f(x) 0. Let ^ be as in (a). Suppose (c) = 0
for all c X then,
0 = (T) =
_
f
+
d
40
Therefore by 4.10 Corollary [Bartle], f
+
= 0 almost everywhere. Like-
wise
0 = (^) =
_
f

d.
Therefore, f

= 0 almost everywhere. So then f = 0 almost every-

where.
Assume f = 0 for almost x X. Then by Proposition 6.2 for all
c X
(c) =
_
E
f d = 0.
Proposition 6.7. Suppose that f
1
, f
2
are in L(A, X, ) and let
1
,
2
be
their indenite integrals, then
1
(c) =
2
(c) for all c X if and only if
f
1
(x) = f
2
(x) for almost all x in X.
Proof. Dene f = f
1
f
2
and (c) =
_
E
f d. Clearly
_
f d =
_
f
1
d
_
f
2
d. Assume
1
(c) =
2
(c) for all c X. Then
(c) =
_
E
f d =
_
E
f
1
d
_
E
f
2
d = 0.
By Proposition 6.6, f = 0 for almost all x in A which means that f
1
= f
2
for almost all x in A.
Let f
1
= f
2
for almost all x on A. Then f = 0 almost everywhere. Again
by Proposition 6.2 for all c X,
0 = (c) =
1
(c)
2
(c).
This completes the proof.
Proposition 6.8. Let A = N, let X be all subsets of N, and let be the
counting measure on X, then f belongs to L(A, X, ) if and only if the
series

n=1
f(n) is absolutely convergent, in which case
_
f d =

n=1
f(n). (6.3)
41
Proof. From 5.3 Theorem [Bartle] and the denition of L(A, X, ),
f L(A, X, ) if and only if
_
[f[
+
d,
_
[f[

d have nite values and

[f[
+
and [f[

are in M
+
(A, X). By Proposition 5.7,
_
[f[
+
d =

n=1
[f(n)[
+
< and
_
[f[

d =

n=1
[f(n)[

<
Therefore, combining the above equations, it is clear that

f(n) is abso-
lutely convergent, provided f L(A, X, ).
If

n=1
[f(n)[ < , then

n=1
f
+
(n) and

n=1
f

(n) have nite values.

Now f
+
and f

have nonnegative values, therefore Proposition 5.7 may be

applied to give f L(A, X, ). Lastly,
_
f d =
_
f
+
d
_
f

d =

n=1
f
+
(n)

n=1
f

(n) =

n=1
f(n).
Proposition 6.9. If f
n
is a sequence in L(A, X, ) which converges uni-
formly on A to a function f, and (A) < , then
_
f d = lim
n
_
f
n
d.
Proof. Let f = f
+
f

and f
n
= f
+
n
f

n
. By Proposition 5.12,
lim
n
_
f
+
n
d =
_
f
+
d and lim
n
_
f

n
d =
_
f

d.
Therefore,
_
f d =
_
f
+
d
_
f

d = lim
_
f
+
n
d lim
_
f

n
d = lim
_
f
n
d.
Remark 2. In general, the conclusion in Proposition 6.9 is false if the con-
dition (A) < is dropped. Let f
n
=
1
n
1
[0, n]
, let A = R and X = B and
let be the Lebesgue measure. Clearly, f
n
converges uniformly to f = 0.
Moreover,
_
f
n
d = 1. Thus,
lim
_
f
n
d = 1 ,=
_
f d = 0.
42
Remark 3. In general, the Lebesgue Dominated Convergence Theorem fails
if the condition [f
n
[ g for all n and g integrable is dropped. Let f
n
=
n1
[0,
1
n
]
, let A = R and X = B and let be the Lebesgue measure. Now
limf
n
converges almost everywhere to f = 0. However, there does not exist
an integrable function g such that [f
n
[ g for all n. Clearly,
_
f
n
d = 1.
Thus,
_
f d = 0 ,= 1 = lim
_
f
n
d.
Proposition 6.10. If f
n
L(A, X, ), and if

n=1
_
[f
n
[ d < ,
then the series

f
n
(x) converges almost everywhere to a function f in
L(A, X, ). Moreover,
_
f d =

n=1
_
f
n
d.
Proof. By 4.13 Corollary [Bartle] and since f
+
n
[f
n
[ for all n N,
_

n=1
f
+
n
d
_

n=1
[f
n
[ d < .
Therefore,

n=1
f
+
n
converges almost everywhere to some function in
L(A, X, ). Call this function f
+
. Likewise,

n=1
f

n
converges almost
everywhere to some function in L(A, X, ). Call this function f

. Com-
bining

n=1
f
+
n
and

n=1
f
+
n
, it is clear that

n=1
f
n
(x) converges almost
everywhere to f in L(A, X, ). Moreover,
_
f d =
_

n=1
f
n
d =
_
_

n=1
f
+
n

n=1
f

n
_
d
=
_

n=1
f
+
n
d
_

n=1
f

n
d
=

n=1
__
f
+
n

_
f

n
d
_
=

n=1
_
f
n
d.
43
Proposition 6.11. Let f
n
L(A, X, ), and suppose that f
n
converges
to a function f. If
lim
n
_
[f
n
f[ d = 0, then
_
[f[ d = lim
n
_
[f
n
[ d.
Proof. By hypothesis and by the triangle inequality, for > 0 there exists
N

N so that for each n N

[f
n
[ [f[

d
_
[f
n
f[ d < .
By 5.3 Theorem [Bartle] for each n N

_
([f
n
[ [f[) d

[f
n
[ [f[

d < .
Therefore,
lim
n
_
[f
n
[ d =
_
[f[ d.
Proposition 6.12. Let f be an X measurable function on A to R. For
n N, let f
n
be the sequence of truncates of f (see Proposition 3.20. If f
is integrable with respect to , then
_
f d = lim
n
_
f
n
d.
Conversely, if
sup
nN
_
[f
n
[ d < ,
then f is integrable.
Proof. Assume f is integrable with respect to . By Proposition 3.20 f
n
measurable for each n N. Since [f
n
[ [f[ for each n N and f integrable,
by 5.4 Corollary f
n
is integrable for each n N. Clearly f
n
converges
almost everywhere to f, therefore by the Lebesgue Dominated Convergence
Theorem (LDCT),
_
f d = lim
n
_
f
n
d.
44
Now assume sup
n
_
[f
n
[ d < . Then [f
n
[ is integrable for all n and [f
n
[
is monotone increasing to [f[. So by (MCT)
_
[f[ d = lim
n
_
[f
n
[ d sup
n
_
[f
n
[ d < .
Thus f is integrable.
45
7. The Lebesgue Spaces L
p
Denition 7.1. If 1 is a real linear (=vector) space, then a real valued
function N on 1 is said to be a norm for 1 in case it satises
(i) N(v) 0 for all v 1
(ii) N(v) = 0 if and only if v = 0
(iii) N(v) = [[ N(v) for all v 1 and real ;
(iv) N(u +v) N(u) +N(v) for all u, v 1
If condition (ii) is dropped, the function N is said to be a semi-norm or
a pseudo-norm for 1.
Example 7.2. Let C[0, 1] be the linear space of continuous functions on
[0, 1] to R. Dene N
0
for f in C[0, 1] by N
0
(f) = [f(0)[.Clearly, N
0
(f) =
[f(0)[ 0 for all f C[0, 1]. Moreover, N
0
(f) = [[ [f(0)[ = [[ N
0
(f).
Lastly, if f, g C[0, 1], then
N
0
(f +g) = [f(0) +g(0)[ [f(0)[ +[g(0)[ = N
0
(f) +N
0
(g).
Therefore, N
0
(f) = [f(0)[ is a semi-norm since Conditions (i), (iii), and (iv)
of Denition 7.1 are satised.
Example 7.3. Let C[0, 1] be as before and dene N
1
for f in C[0, 1] to be
the Riemann integral of [f[ over [0, 1]. Obviously, by the properties of the
Riemann integral N
1
satises the conditions of a semi-norm.
Proposition 7.4. If f
n
is dened for n 1 to be equal to 0 for 0
x (1 1/n)/2, to be equal to 1 for
1
2
x 1, and to be linear for
(1 1/n)/2 x
1
2
, then f
n
is a Cauchy sequence, but f
n
does not
converge relative in N
1
(as dened in Example 7.3) to an element of C[0, 1].
Proof. Clearly, limf
n
= f where f = 0 if 0 < x
1
2
and f = 1 if
1
2
x 1.
Certainly, f is not in C[0, 1]. Now, let m n where m, n N. From the
46
denition of N
1
,
N
1
(f
m
f
n
) =
_
1
0
[f
m
f
n
[ dx =
1
4n
_
f
n
_
1
2

1
2m
__
.
However f
n
_
1
2

1
2m
_
1. Therefore,
lim
n
1
4n
_
f
n
_
1
2

1
2m
__
lim
n
1
4n
= 0.
Hence, for > 0, there exists an M() N such that
N
1
(f
m
f
n
) =
_
1
0
[f
m
f
n
[ dx =
1
4n
_
f
n
_
1
2

1
2m
__
<
for all n, m M(). Therefore f
n
Cauchy.
Proposition 7.5. Let N be a norm on a linear space 1 and let d be dened
for u, v 1 by d(u, v) = N(u v), then d is a metric on N; that is
(i) d(u, v) 0 for all u, v 1
(ii) d(u, v) = 0 if and only if u = v
(iii) d(u, v) = d(v, u)
(iv) d(u, v) d(u, w) +d(w, v).
Proof. Condition (i) follows immediately, since N is a norm and u v 1.
Condition (ii) is satised since d(u, v) = N(u v) if and only if u v = 0
if and only if u = v. Condition (iii) follows since
d(u, v) = N(u v) = N
_
(1)(v u)
_
= [1[ N(v u) = d(v, u).
Finally let w, u, v 1
d(u, v) = N(u v) = N(u w +w v)
N(u, w) +N(w, v) = d(u, w) +d(w, v).
Thus, Condition (iv) is satised.
Proposition 7.6. Let 1 p < . If f L
p
and > 0, then there exists a
simple X measurable function such that |f |
p
< .
Proof. The case p = 1 follows from Proposition 6.4. Let 1 < p < . From
2.11 Lemma [Bartle] there exists
n
of X-measurable functions such that
47
[f
+

+
n
[ converges almost everywhere to 0 and
+
n
f
+
. Certainly, then
[f
+

+
n
[
p
converges almost everywhere to 0. Since [f
+

+
n
[
p
2
p
[f[ the
Lebesgue Dominated Convergence Theorem implies that
_

f
+

+
n

p
d = lim
n
_

f
+

+
n

p
d
Therefore, given > 0, |f
+

+
n
|
p
< for n suciently large. Likewise,
|f

n
|
p
< for n suciently large. Let lim
+
n
=
+
and let lim

n
=

.
Moreover, let T = x A : f(x) 0 and let ^ = x A : f(x) < 0.
Then,
__
[f [
p
d
_1
p
=
__

f
+

p
1
P
d +
_

p
1
N
d
_1
p
<
for n suciently large enough.
Remark 4. Proposition 7.6 holds if f L

.
Proof. For p = , |f(x) (x)|

= inf S(^) : ^ X, (^) = 0 where

S(^) = sup[f(x) (x)[ : x / ^. As seen before, for > 0 there exists
a simple X-measurable function such that [f(x) (x)[ < for all x / ^.
Therefore,
S(^) = sup[f(x) (x)[ : x / ^ < .
Thus
|f(x) (x)|

= inf S(^) : ^ X, (^) = 0 < .

Proposition 7.7. If f L
p
, 1 p < , and if c = x A : [f(x)[ , = 0,
then c is -nite.
Proof. Let c
n
=
_
x A : [f(x)[
1
n
_
. Then c =

n=1
c
n
. The proof is
complete if (c
n
) < for each n N. Clearly, [f[
1
n
1
En
for each n.
Thus, [f[
p

_
1
n
1
En
_
p
for each n. By 4.5 Lemma,
_
[f[
p
d
_ _
1
n
_
p
(c
n
)
for each n. Therefore,
>
__
[f[
p
d
_1
p

1
n
[(c
n
)]
1
p
for each n.
48
It can be concluded that (c
n
) < for each n N.
Proposition 7.8. If f L
p
, and if c
n
= x A : [f(x)[ n, then
(c
n
) 0 as n .
Proof. From the denition of c
n
, [f(x)[ n1
En
(x) for all x A. Whence,
[f(x)[
p
n
p
1
En
(x). By 4.5 Lemma and f L
p
lim
n
(c
n
) lim
n
1
n
p
_
[f[
p
d = 0.
This nishes the proof.
Proposition 7.9. Let A = N and let be the counting measure on N. If
f is dened on N by f(n) =
1
n
, then f does not belong to L
1
, but it does
belong to L
p
for 1 < p .
Proof. Let p = 1 therefore,
_
[f(n)[ d =

n=1
1
n
= .
Let p > 1 therefore,
_
[f(n)[
p
d =

n=1
1
n
p
< .
An alternative way to look at Proposition 7.9 is to let A = R, X = B, and
be the Lebesgue measure and dene g(x) = 0 for x < 1 and g(x) =
1
x
for
x 1. For p = 1,
_
[g(x)[ d =
_
[1,]
[g(x)[ d = lim
n
_
n
1
1
x
dx = lim
n
ln(n) = .
For p > 1,
_
[g(x)[
p
d =
_
[1,]
[g(x)[
p
d = lim
n
_
1
p 1
+
1
(p + 1)n
p1
_
=
1
p 1
.
Proposition 7.10. Let A = N, and let be the measure on N which has
measure
1
n
2
at the point n. (More precisely (c) =
_
1
n
2
: n c
_
.) Then
(i) (A) <
49
(ii) for f dened on A as f(n) =

n, f L
p
if and only if 1 p < 2.
Proof. Clearly,
(A) =

n=1
1
n
2
< .
Also,
_

p
d =

n=1
n
p4
2
.
For 1 p < 2,
3
2

p4
2
< 1. The result follows from the fact that

1
n
p
< when p > 1.
By letting (c) =
_
1
n
2.1
: n c
_
and letting f(n) = n
1
p
then f L
p
if and only if 1 p (1.1)p
0
. It is seen that
_
[f(n)[
p
d =

n=1
_
1
n
_
2.1
p
p
0
.
Obviously 2.1
p
p
0
> 1 for all 1 p (1.1)p
0
.
Proposition 7.11. Let (A, X, ) be a nite measure space. If f is X-
measurable function let c
n
= x A : (n 1) [f(x)[ < n. Show that f
L
1
if and only if

n=1
n(c
n
) < . (7.1)
More generally, f L
p
for 1 p < , if and only if

n=1
n
p
(c
n
) < . (7.2)
Proof. Assume (7.1) holds. From the denition of c
n
, n1
En
(x) [f(x)[ 1
En
(x)
for each n N and x A. By a now familiar argument,
_
En
[f(x)[ d n(c
n
) for each n N.
Since c
n
c
m
= when n ,= m and

c
n
= A,

n=1
_
En
[f(x)[ d =
_
X
[f(x)[ d

n=1
n(c
n
) < .
Therefore, f L
1
.
50
Assume f L
1
. For each n N and x A, (n1)1
En
(x) [f(x)[ 1
En
(x)
Arguing as above,

n=1
(n 1) (c
n
)
_
X
[f(x)[ d < .
Now

n=1
(n 1) (c
n
) =

n=1
(n) (c
n
)

n=1
(c
n
) < .
Since (A, X, ) is a nite measure space,

n=1
(c
n
) < , therefore (7.1)
holds.
As above, it is quickly seen that
_
X
[f(x)[
p
d

n=1
n
p
(c
n
) <
Thus, if (7.2) holds, then f L
p
. Now suppose f L
p
. For all x c
n
,
n 1 [f(x)[. Clearly then, n
p
([f(x)[ + 1)
p
2
p
[f(x)[
p
+ 1m for all
x c
n
. From this inequality, we get (7.2).
Proposition 7.12. If (A, X, ) is a nite measure space and f L
p
, then
f L
r
for 1 r p.
Proof. From Proposition 7.11, f L
p
implies that

n
p
(c
n
) < . For
1 r p, n
r
(c
n
) n
p
(c
n
) for each n N. Therefore,

n=1
n
r
(c
n
)

n=1
n
p
(c
n
) < .
Employing Proposition 7.11 again, gives the conclusion.
Proposition 7.13. Suppose that A = N and is the counting measure on
N. If f L
p
, then f L
s
with 1 p s < , and |f|
s
|f|
p
.
Proof. Since is the counting measure on N, f can be viewed as a sequence,
a
n
of real numbers. Therefore,
|f|
p
=
_

n=1
[a
n
[
p
_1
p
<
51
Since

[a
n
[
p
< , the lim[a
n
[
p
= 0. Thus for some n

N, [a
n
[ < 1 for all
n n

. Since 1 p s, [a
n
[
s
< [a
n
[
p
for all n n

. It can be concluded
then that
>

n=n
0
[a
n
[
p
>

n=n
[a
n
[
s
.
Clearly,
n1

n=1
[a
n
[
s
< .
Therefore, f L
s
.
Proposition 7.14. Let (A, X, ) be any measure space and let f belong to
both L
p
1
and L
p
2
, with 1 p
1
p
2
< , then f L
p
for any value of p
such that p
1
p p
2
.
Proof. Let / = x A : [f(x)[ 1. Let B = x A : [f(x)[ < 1. Now,
_
[f(x)[
p
1
A
d
_
[f(x)[
p
2
1
A
d < .
Furthermore,
_
[f(x)[
p
1
B
d
_
[f(x)[
p
1
1
B
d < .
Therefore,
_
[f(x)[
p
1
A
d +
_
[f(x)[
p
1
B
d =
_
[f(x)[
p
d < .
From Holders inequality, if 1 < p < ,
_
1
p
_
+
_
1
q
_
= 1 and f L
p
, then

_
fg d

|f|
p
for all g L
q
such that |g|
q
1. This leads to the following proposition.
Proposition 7.15. If f ,= 0, f L
p
, and g

(x) = c[signumf(x)]f(x)
p1
for x on A where c = (|f|
p
)
p/q
, then g

L
q
, |g

|
q
= 1 and

_
fg

= |f|
p
.
52
Proof. First, note that if
1
p
+
1
q
= 1, then pq q = p. Now
_
[g

[
q
d = [c[
q
_
[f[
pqq
d = [c[
q
_
[f[
p
d < .
Thus, g

L
q
. Secondly,
|g

|
q
=
__
[g

[
q
d
_1
q
= [c[
__
[f[
p
d
_1
q
= [c[ (|f|
p
)
p
q
=
(|f|
p
)
p
q
(|f|
p
)
p
q
= 1.
Finally

_
fg

= c
_
[signumf] f
p
d = c (|f|
p
)
p
=
(|f|
p
)
p
(|f|
p
)
p/q
= |f|
p
,
since p p/q = 1.
Proposition 7.16. If f L
p
, 1 p < and > 0 then there exists a
set c

X with (c

) < such that if T X and T c

= , then
|f1
F
|
p
< .
Proof. Let > 0. If f L
p
, then [f[ M
+
(A, X). Consequently, [f[
p

M
+
(A, X) since for R,
x A : [f[
p
< =
_
x A : [f[ <
1
p
_
X.
From Proposition 5.17 there exists a set in X, call it c

, such that (c

) <
and
_
[f[
p
d <
_
E
[f[
p
d +
p
.
Let T be such that T c

= . It is clear then that

_
E
[f[
p
d +
_
F
[f[
p
d
_
[f[
p
d <
_
E
[f[
p
d +
p
.
Therefore, |f1
F
|
p
< .
Proposition 7.17. Let f
n
L
p
(A, X, ), 1 p < , and let
n
be dened
for c X by

n
(c) =
__
E
[f
n
[
p
d
_1
p
.
Then [
n
(c)
m
(c)[ |f
n
f
m
|
p
53
Proof. Clearly it is sucient to show that,
[ |f
n
|
p
|f
m
|
p
[ |f
n
f
m
|
p
.
Now
|f
n
|
p
= |f
n
|
p
|f
m
|
p
+|f
m
|
p
|f
n
f
m
|
p
+|f
m
|
p
.
Therefore, |f
n
|
p
|f
m
|
p
|f
n
f
m
|
p
Arguing in the same fashion, it is
seen that
|f
m
|
p
|f
n
|
p
|f
m
f
n
|
p
= |f
n
f
m
|
p
.
Finally,
|f
n
f
m
|
p
|f
n
|
p
|f
m
| |f
n
f
m
|
p
which proves the proposition.
As a consequence of Proposition 7.17 if f
n
Cauchy sequence in L
p
, then
lim
n
exists for each c X since every Cauchy sequence in R converges.
Proposition 7.18. Let f
n
,
n
be as in Proposition 7.17. If f
n
is a
Cauchy sequence and > 0, then there exists a set c

A with (c

) <
such that if T X and T c

= , then
n
(T) < for all n N.
Proof. Let f
n
be a Cauchy sequence and let > 0, therefore we can
nd an N() such that for n, m N(), |f
n
f
m
|
p


2
. Employing
Proposition 7.16 there exist c
1
, , c
N()
such that
(i) ([c
i
]
c
) < where 1 i N()
(ii) |f
i
1
E
i
|
p
<

2
for 1 i N().
Let c

=
N()
_
i=1
c
c
i
and T X with T c

= . Clearly, (c

) < . If
1 n N(), then T c
n
. Thus, by (ii)
|f
n
1
F
|
p
|f
n
1
En
|
p
<

2
< .
54
If N() < n, then
|f
n
1
F
|
p
|f
n
1
F
f
N()
1
F
|
p
+|f
N()
1
F
|
p
< |f
n
1
F
f
N()
1
F
|
p
+

2
|f
n
f
N()
|
p
+

2
<

2
+

2
= .
This concludes the proof.
Proposition 7.19. Let f
n
,
n
be as in Proposition 7.18, and suppose f
n

is Cauchy. If > 0, then there exists a () such that if c A and (c) <
(), then
n
< for all n N.
Proof. Let
n
(c) =
_
[f
n
[
p
d. Clearly, [f
n
[
p
M
+
(A, X) for each n N.
By 4.9 Corollary [Bartle],
n
is a measure for each n. By 4.11 Corollary
[Bartle],
n
is absolutely continuous with respect to for each n. Fix > 0.
Since f
n
is Cauchy there exists an N() such that for |f
n
f
m
|
p
<

2
for n, m N(). Since
n
is absolutely continuous with respect to , for
each n N there exists an
n
< 0 such that if c X and (c) <
n
, then

n
(c) <
_

2
_
p
. Let

() = inf
_

1
, ,
N()
_
. Therefore for all n such that
1 n N(), if c X and (c) <

, then
n
(c) <
_

2
_
p
()
p
, (i.e) for
1 n N(), |f
n
1
E
| < . Since f
n
is Cauchy and |f
N()
1
E
|
p
<

2
, then
for n N(),
|f
n
1
E
|
p
|f
n
1
E
f
N()
1
E
|
p
+|f
N()
1
E
|
p
<

2
+

2
= .
Thus, if c X and (c) <

(), then

n
(c) = |f
n
1
E
|
p
< ,
for all n N.
The following remark will be useful in the next proposition.
Remark 5. If f L
p
(A, X, ), p = , then from the denition of |f|

,
[f(x)[ |f|

for almost all x

Proposition 7.20. If A < |f|

, then there exists a set c X with (c) >

0 such that [f(x)[ > A for all x c.
55
Proof. Without loss of generality, let 0 A < |f|

. Dene f
A
(x) for
x A as f
A
(x) = A for x c = x A : [f(x)[ > A and f
A
(x) = [f(x)[
if [f(x)[ A. From Proposition 3.20, c X. Moreover, [f(x)[ > A for all
x c. Now, let > 0 be such that A+ < |f|

. Clearly (A+) 1
E
> A1
E
.
Thus A(c) + (c) > A(c). Therefore (c) > 0. This completes the
proof.
Proposition 7.21. If f L
p
, 1 p , and g L

, then the product

fg L
p
and |fg|
p
|f|
p
|g|

.
Proof. From the denition of |g|

, there is an M R such that [g[ M

almost everywhere. Therefore [fg[
p
M
p
[f[
p
. By a now familiar argument,
_
[fg[
p
d M
p
_
[f[
p
d < .
Therefore, fg L
p
. Moreover from above
__
[fg[
p
d
_1
p
|g|

|f|
p
.
Proposition 7.22. The space L

is contained in L
1
if and only if (A) <
. Moreover, if (A) = 1 and f L

, then
|f|

= lim
p
|f|
p
.
Proof. From the denition of L

and 4.7 Corollary [Bartle],

_
[f[ d |f|

(A) where |f|

= inf M R : [f[ M -a.e .

Clearly then (A) < implies f L
1
. To show the other implication, we
will show the contrapositive (i.e. if (A) = , then L

is not contained
in L
1
). To that end, assume (A) = and let f = 1
X
then f L

but
f / L
1
.
Moreover if f L
p
and (A) = 1, then again employing the denition of
|f|

, it is evident that [f[

p
|f|
p

almost everywhere. Therefore

limsup
p
|f|
p
|f|

.
56
Now let 0 < < |f|

. By Proposition 7.20, the measurable set

c = x A : [f(x)[ |f|

satises (c) > 0. For 1 < p < , (|f|

)
p
[f[
p
on the set c. Thus
(|f|

) (c)
1
p
|f|
p
for 1 < p < . As p , (c)
1
p
= 1 therefore
|f|

liminf
p
|f|
p
.
Thus,
limsup
p
|f|
p
|f|

liminf
p
|f|
p
.
This completes the proof.
57
8. Modes of Convergence
In following examples (R, B, ) denotes the real line with Lebesgue mea-
sure dened on the Borel subsets of R. Also, 1 p .
Example 8.1. Let f
n
= n

1
p
1
[0,n]
Then, the sequence f
n
converges uni-
formly to the 0-function but it does not converge in L
p
(R, B, ).
Proof. Clearly n
1
p
> 0 for 1 p < . By Theorem 2.2.6 [Stoll]
lim
n
1
n
1
p
= 0.
Thus given > 0 there exists n

N such that [f
n
(x) 0[ =
1
n
1
p
< for all
x R and for all n n

. This shows that f

n
0 uniformly. However,
|f
n
0|
p
=
_
_
[0,n]

1
n
1
p

p
d
_1
p
=
_
n 0
n
_1
p
= 1
Letting = 1 it is seen that f does not converge to 0 in L
p
.
Example 8.2. Let f
n
= n1
[
1
n
,
2
n
]
. Then the sequence f
n
converges every-
where to the 0-function, but it does not converge in L
p
(R, B, ).
Proof. Clearly for each x R and for > 0 there exists an N(, x) N
such that [f
n
(x)[ < for all n N(, x) therefore f
n
converges everywhere
to the 0-function. However
|f
n
0|
p
=
_
_
[
1
n
,
2
n
]
[n[
p
d
_1
p
=
_
n
p
1
n
_1
p
= n
1
1
p
.
Since 1
1
p
0 when p 1,
lim
n
|f
n
|
p
,= 0.
So f
n
does not converge in L
p
to the 0-function.
Example 8.2 shows that convergence in measure does not imply L
p
con-
vergence even for a nite measure space since the integral of f
n
vanishes
outside of the interval
_
1
n
,
2
n

and it equals n
1
1
p
on the interval
_
1
n
,
2
n

.
58
Proposition 8.3. Both sequences in Examples 8.1 and 8.2 converge in mea-
sure to their limits.
Proof. Let > 0. For Example 8.1, it has been shown that there exists an
N N such that [f
n
(x) 0[ < for all x R. For n N,
x A : [f
n
0[ = .
Therefore,
lim
n
(x A : [f
n
0[ ) = () = 0.
For Example 8.2, provided 0 < 1,
lim(x A : [f
n
0[ ) = lim
__
1
n
,
2
n
__
= lim
_
1
n
_
= 0.
Therefore, each sequence of functions converges in measure.
Example 8.4. Let f
n
= 1
[n,n+1]
. The sequence f
n
converges everywhere
to the 0-function, but it does not converge in measure.
Proof. Clearly for each x R there exists an N(x) N such that f
n
(x) = 0
for all n N(x). Let 0 < 1. Now for all n N,
(x A : [f
n
(x) 0[ ) = 1.
Therefore, f
n
does not converge in measure to the 0-function.
Looking at 7.4 Example [Bartle], it is seen that the sequence converges in
L
p
to the 0-function and thus converges in measure to the 0-function, but
does not converge at any point of [0, 1]. However, from 7.6 Theorem [Bartle]
there exists a subsequence of f
n
which converges almost everywhere to
the 0-function. Clearly, if f
n
= 1
[0,
1
n
]
then for > 0 and for each x [0, 1]
there exists an N(, x) N such that [f
n
(x)[ except for x = 0 but
(0) = 0. However there is not a subsequence from 7.4 Example that
converges everywhere since each f
n
maps a rational to 1.
59
Proposition 8.5. If a sequence f
n
converges in measure to a function f,
then every subsequence of f
n
converges in measure to f. More generally,
if f
n
is Cauchy in measure, then every subsequence is Cauchy in measure.
Proof. Let f
n
k
be a subsequence of f
n
. From assumption, for > 0
lim
n
(x A : [f(x) f
n
(x)[ ) = 0.
Therefore for > 0 there exists an N such that
(x A : [f(x) f
n
(x)[ ) <
for all n N. Since n
k
strictly increasing, there exists K N so that
n
k
N for all k K. Therefore
(x A : [f(x) f
n
k
(x)[ ) <
for k K. Thus,
lim
k
(x A : [f(x) f
n
k
(x)[ ) = 0.
Similarly, if f
n
is Cauchy, then
lim
m,n
(x A : [f
m
(x) f
n
(x)[ ) = 0.
For > 0 there exists N N such that
(x A : [f
m
(x) f
n
(x)[ ) <
for all m n N. Since n
k
strictly increasing, there exists K N so
that n
l
n
k
N for all l k K. Therefore
lim
l,k
(x A : [f
n
l
(x) f
n
k
(x)[ ) = 0.
Proposition 8.6. If a sequence f
n
converges in L
p
to a function f, and
a subsequence of f
n
converges in L
p
to g, then f = g almost everywhere.
Proof. Since convergence in L
p
implies convergence in measure, f
n
con-
verges to f in measure and f
n
k
converges to g in measure. Moreover,
from Proposition 8.5 f
n
k
converges in measure to f. By 7.7 Corollary
60
[Bartle] f is uniquely determined almost everywhere therefore f = g almost
everywhere.
Proposition 8.7. If f
n

nN
is a sequence of characteristic functions of
sets in X, and if f
n

nN
converges to f in L
p
, then f is almost everywhere
equal to the characteristic function of a set in X.
Proof. Let f
n
= 1
An
where /
n
X for all n N and let f
n
(x) f(x) in
L
p
. Since f
n
(x) f(x) in L
p
there is a subsequence
_
f
n
j
_
jN
of f
n

nN
such that there exists c A where (c
c
) = 0 and f
n
j
(x) f(x) for all
x c. Since f
n
j
(x) = 0 or 1 on c, then f(x) = 0 or 1 on c. This nishes
the proof.
Proposition 8.8. Let f
n
be as in Example 8.2. If > 0, then f
n

converges uniformly to the 0-function on the complement of [0, ].

Proof. Let > 0. Clearly, there exists an N N such that
2
n
< for all
n N. Therefore, f
n
(x) = 0 for all x [0, 1]([0, ])
c
and for all n N.
Remark 6. There does not, however, exist a set of measure zero, on the
complement of which the sequence from Example 8.2 is uniformly convergent
to the 0-function.
Proof. Without loss of generality we can restrict ourselves to the interval
[0, 2] on the real line since for points outside of this interval, f
n
= 0 for all
n N. Let c B be such that c [0, 2] and (c) = 0. It is enough
then to show that c
c
[
1
n
,
2
n
] ,= for all n N. Clearly, (c
c
) = 2.
Therefore, if c
c
and
_
1
n
,
2
n

are disjoint, then (

_
1
n
,
2
n

) +(c
c
) > 2 which is
a contradiction.
The following propositions demonstrate that in Fatous Lemma and the
Lebesgue Dominated Convergence Theorem almost everywhere convergence
can be replaced by convergence in measure. In fact once it is shown that
61
Fatous Lemma holds for convergence in measure, the Lebesgue Dominated
Convergence Theorem for convergence in measure is immediate.
First the following fact will be useful in the proof of Fatous Lemma for
convergence in measure.
Fact 8.9. A sequence a
n

n=1
converges to a if and only if every subse-
quence of a
n

n=1
has a subsequence that converges to a.
Proposition 8.10. If f
n
is a sequence of nonnegative measurable func-
tions and f
n
(x) converges to f(x) in measure, then
_
f d liminf
_
f d.
Proof. By Fact 8.9, it will suce to show that each subsequence of f
n
has
a subsequence that converges a.e. to f. By assumption f
n
converges to f in
measure therefore from Proposition 8.5 every subsequence of f
n
converges
in measure to f. By 7.6 Theorem [Bartle] each subsequence of f
n
has a
subsequence which converges a.e. to f. So f
n
converges a.e. to f. Thus by
Fatous Theorem for a.e. convergence we are done.
Proposition 8.11. The Lebesgue Dominated Convergence holds for con-
vergence in measure.
Proof. Follows from Proposition 8.10.
Proposition 8.12. Let (A, X, ) be a nite measure space. If f is an X-
measurable function, let
r(f) =
_
[f[
1 +[f[
d.
A sequence f
n
of X-measurable functions converges in measure to f if
and only if r(f
n
f) 0.
Proof. Assume that r(f
n
f) 0. Notice
0
[f
n
f[
1 +[f
n
f[
< 1 (8.1)
62
Therefore, the limit of the left hand side of (8.1) exists. By Fatous Theorem,
0
_
liminf
[f
n
f[
1 +[f
n
f[
d liminf
_
[f
n
f[
1 +[f
n
f[
d = 0.
By 4.10 Corollary,
lim
[f
n
f[
1 +[f
n
f[
= 0
almost everywhere. Thus [f
n
f[ = 0 almost everywhere. Applying Ego-
ros Theorem, it is concluded that f
n
converges to f in measure.
Fix > 0 and assume that f
n
f in measure. Then there exists N
such that for all n N

__
x A : [f
n
f[

2(A)
__
<

2
.
Let B
n
=
_
x A : [f
n
f[

2(X)
_
. So for all n N,
r(f
n
f) =
_
Bn
[f
n
f[
1 +[f
n
f[
d +
_
X\Bn
[f
n
f[
1 +[f
n
f[
d

_
Bn
1 d +
_
X\Bn
[f
n
f[ d
(B
n
) +

2
_
(AB
n
)
(A)
_
<
e
2
+

2
= .
Thus, r(f
n
f) 0.
Proposition 8.13. If the sequence f
n
of measurable functions converges
almost everywhere to a measurable function f and is continuous on R to
R, then the sequence f
n
converges almost everywhere to f.
Proof. By Proposition 3.26 f is measurable and f
n
is measurable for
each n N. Since is continuous, then for > 0 and for each x A there
exists a > 0 such that [(f
n
(x)) (f(x))[ < where [f
n
(x) f(x)[ < .
Moreover f
n
f almost everywhere therefore, there exists a set / X
such that for every > 0 and x /
c
there exists N(, x) N such that
if n N(, x) then [f
n
(x) f(x)[ < . Thus for > 0 there exists a set
/ X such that for each X /
c
there exists N(, x) such that if n
63
N(, x) then [(f
n
(x) (f(x))[ < . In other words, f
n
converges
almost everywhere to f
In the following example, it will be shown that if has a point of disconti-
nuity, then there exists a sequence f
n
which converges almost everywhere
to f but f does not converge almost everywhere to f.
Example 8.14. Let = 1 for x 0 and let = 1 for x < 0. Let f
n
be
a sequence of negative-valued functions which converges almost everywhere
to f = 0. Therefore (f(x)) = 1. Now let / X be such that for all
x / / f
n
converges everywhere to the 0-function and let = 2. Thus for
all x / / and for all n N, [(f
n
(x)) (f(x))[ = 2. Therefore f
n

does not converge almost everywhere to f.

64
9. Conclusion
To write a conclusion to this paper is dicult and inappropriate since the
paper, in a sense, opens a can of worms. It has discussed the foundations
of the Lebesgue Integral and some of the important theorems that emerge
from the topic. Certainly, sections seven and eight provide a glimpse at
some of the higher theory that uses this integral. To be sure, the problems
presented in this paper merely scratch the surface of the theory behind the
General Lebesgue Integral, however they suggest the depth and power of
this mathematical technique.
65
References
[CAOB] Charalambos D. Aliprantis and Owen Burkinshaw, Principles of real analysis,
third ed., Academic Press Inc., San Diego, CA, 1998.
[CAOB] Charalambos D. Aliprantis and Owen Burkinshaw, Problems in real analysis,
second ed., Academic Press Inc., San Diego, CA, 1999, A workbook with solutions.
[Bartle] Robert G. Bartle, The elements of integration, John Wiley & Sons Inc., New
York, 1966.
[Lang] Serge Lang, Real and functional analysis, third ed., Springer-Verlag, New York,
1993.
[JMNW] John N. McDonald and Neil A. Weiss, A course in real analysis, Academic Press
Inc., San Diego, CA, 1999, Biographies by Carol A. Weiss.
[Royden] H. L. Royden, Real analysis, third ed., Macmillan Publishing Company, New
York, 1988.
[Rudin] Walter Rudin, Principles of mathematical analysis, third ed., McGraw-Hill Book
Co., New York, 1976, International Series in Pure and Applied Mathematics.
[Stoll] Manfred Stoll, Introduction to real analysis, Addison-Wesley, Reading Mas-
sachusetts, 1997.
66