AN INTRODUCTION TO THE LEBESGUE INTEGRAL

CHARLES DOSS
The history of integration began with Archimedes around the 2nd
century B.C., but did not start to gain rigor until the 17th century.
Bonaventura Cavalieri began working with indivisibles in the compu-
tation of areas by the early 1700s. Leibniz and Newton toward the
end of the century discovered the relationship between the integral and
the derivative, and really invented integral calculus. It was not un-
til Cauchy that the integral was actually dened rigorously. But it
was Riemann who developed the integral frequently used today. This
integral is constructed by taking the limit of approximations, called
Riemann sums, which are based on partitioning the domain of the
function.
However, it was realized quickly that it is not too dicult to nd
functions which are not Riemann integrable. In particular, the study of
fourier series requires an integral such that
_
[

k
f
k
(x)]dx =

k
_
f
k
(x)dx,
but it is rather easy for this to fail for the Riemann integral. For in-
stance, consider A = Q

[0, 1] and 1
A
, the indicator function for A. It
is clear that 1
A
is not Riemann integrable, since for any interval con-
tained in [0, 1], the maximum of 1
A
is 1 and the minimum is 0 (since
both the rationals and irrationals are dense in R), so the upper sum will
be 1 and the lower will be 0, always. But since A is countable (since Q
is countable), its elements can be enumerated by a
k
, and we can dene
functions g
k
(x) =
_
1 x = a
k
0 otherwise
such that 1
A
=

k=1
g
k
. Note
that each partial sum on the right has only nitely many discontinu-
ities, and is thus Riemann integrable, whereas, as noted earlier, the left
side is not.
Thus, in his thesis in 1902, Henri Lebesgue developed a new inte-
gral in which the focus was on the range of the function, instead of
on the domain. The distinction between the two approaches can be
seen by envisioning the graph of a real function f whose range is the
reals. Whereas Riemann focuses on partitioning the x-axis, Lebesgues
integral partitions the y-axis instead. That is, Riemann partitions the
domain of f into a nite number of intervals and on each interval ap-
proximates the values that f takes. Using the rectangles generated by
1
2 CHARLES DOSS
the product of the value of the function on each interval and the length
of that interval, Riemann approximates the area under the function.
Lebesgue, on the other hand, partitions the range of the function into
a nite number of intervals, and for each partition chooses a value to
represent the function for that partition on his approximation (call
it s) (so for all x in the partition, s(x) equals the representative). s is
called a simple function, which means it has a nite range.
Now, it should be clear that Lebesgue needs to construct some sort of
manner of meausuring the area of very complicated sets. Whereas Rie-
mann sums approximate by just using rectangles, a set which Lebesgue
needs to approximate, such as {x|f(x) [a, b]} for some a,b, may be
much more complicated than a rectangle. So initially lets dene m, on
rectangles in R
n
: let m(

i
[a
i
, b
i
]) =

i
(b
i
a
i
). Then it is easy to see
that this measure extends onto elementary sets (nite unions of inter-
vals), and with some work Lebesgue shows that in fact this extends to
a countably additive, regular, nonnegative function on a subset of the
powerset of R
n
(I will still refer to this extension as m.) This subset
is actually closed under taking countable intersections, complements,
and unions. and it is actually these closures that give the measure
its characteristics. We also need to dene which functions are nice,
where a nice function f is called a measurable function. This means
it has the characteristic that the set {x|f(x) > a} is measurable for
all a. Also, > can be replaced by <, , or in the above
condition.
The denition of the integral is rst given for simple functions:
Denition Let g be a simple measurable function on a m-measurable
set E, where E =

E
i
and g(x) = a
i
x E
i
. Then dene the
integral of g over E:
_
E
g =

i
a
i
m(E
i
)
Then for f 0, a measurable function on a measurable set E,
_
E
f = sup
0sf
_
E
s
For general f,
_
f =
_
x|f(x)0
f
_
x|f(x)<0
(f). (Recall that f measur-
able {x|f(x) 0} and {x|f(x) < 0} are both measurable sets).
Note that from now on, the symbol
_
f refers to the Lebesgue integral,
not the Riemann integral.
Now note that by the countable additivity of m, any countable subset
must have measure 0, since a point has measure 0. This means that Q
AN INTRODUCTION TO THE LEBESGUE INTEGRAL 3
is of measure zero. It is not too hard to show that if A = B

C, A
measurable, then for measurable f
_
A
f =
_
B
f +
_
C
f. In particular,
if C is of measure zero, then
_
A
f =
_
B
f
This new Lebesgue integral has very important properties. It turns
out that the set of Lebesgue integrable functions is actually a superset
of the set of Riemann integrable functions, and when a function is both
Lebesgue integrable and Riemann integrable, then the two integrals
have the same value. Furthermore, Lebesgues convergence theorems
about the niceness of being able to pass limits through the integral
in certain conditions make this integral much more useful to the theory
of fourier series. Here are the convergence results, without proof:
Theorem 0.1. (Lebesgues Monotone Convergence Theorem) Let {f
i
}
be a sequence of measurable functions such that 0 f
1
f
2
. . . Then
lim
i
_
f
i
=
_
lim
i
f
i
(note: f will be measurable).
Theorem 0.2. (Lebesgues Bounded Convergence Theorem) Let {f
i
}
be a sequence of measurable functions on a measurable set E, such that
a Lebesgue integrable function g on E s.t. |f
i
(x)| g(x)x E.
Then
lim
i
_
f
i
=
_
lim
i
f
i
As stated above, the set of Riemann integrable functions is a subset
of the set of Lebesgue integrable functions. Also, the Lebesgue integral
actually provides a neat proof of necessary and sucient conditions
for a function to be Riemann integrable. These two theorems will be
proven below. First a lemma will be proved:
Lemma 0.1. If f(x) 0 and
_
E
fdm = 0 Then f(x) = 0 almost
everywhere on E.
Proof. Dene E
n
= {x E|f(x) < 1/n}, and let E
0
=

n
E
n
=
{x|f(x) > 0}. If n s.t m(E
n
) > 0 then to arrive at a contradiction,
dene
s(x) =
_
1
2n
x E
0 otherwise
so 0 s f, so
_
E
sdm
_
E
fdm but
_
E
sdm = m(E
n
)
1
2n
> 0,
since m(E
n
) > 0
_
E
fdm > 0, a contradiction. Thus m(E
0
) =
0 since m(E
n
) = 0 n by the additivity of measure.
4 CHARLES DOSS
For these next theorems, it will be expedient to label the Riemann
integral as
_
f.
Theorem 0.3. (I) A function f is Riemann integrable over E = [a, b]
then it is Lesbegue integrable, and
_
b
a
f =
_
b
a
f
(II)Further, f is Riemann integrable if and only if it is continous almost
everywhere on [a, b].
Proof. Assume f is bounded on E. Now we will dene a simple func-
tion associated with each partition whose behavior will mimic that of
the Riemann sums. Let U
i
(a) = f(a) and let U
i
(x) = M
i
for x
[x
i
, x
i+1
] where M
i
is the maximum value of f on [x
i
, x
i+1
]. Dene
similarly L
i
(x), but replace M
i
with m
i
, the minimum value of f on
the same domain. Notice
_
b
a
U
k
= U(P
k
, f) and
_
b
a
L
k
= L(P
k
, f) Also,
note L
1
L
2
. . . f . . . U
2
U
1
. So let lim
k
U
k
(x) =
U(x) and lim
k
L
k
(x) = L(x). Both U and L are dened nitely by
the above string of inequalities. Note that they are both bounded, and
they are measurable since they are the limits of measurable functions
(this result was not proved, but is not hard to see). So by the monotone
convergence theorem,
(1) T = lim
k
U(P
k
, f) = lim
k
_
U
k
=
_
U
B = lim
k
L(P
k
, f) = lim
k
_
L
k
=
_
L
Now, if f is Riemann integrable, then we know there exist partitions
P
k
such that lim
k
L(P
k
, f) = B = T = lim
k
U(P
k
, f) By equation
(1) , this means that
_
U =
_
L. Note that L f U, but by lemma
0.1, U(x) = f(x) = L(x) almost everywhere (by regarding U L 0).
Thus f is measurable and by equation (1) the two integrals are equal.
So (I) of the theorem is proved.
So, as noted above, f Riemann integrable
_
U =
_
L
U(x) = L(x) except on a set F of measure zero. f continuous al-
most everywhere on E

k
P
k
f continuous almost everywhere,
since

k
P
k
is countable, so of measure zero. This is true since for
x P
k
, U(x) = L(x) f is continuous at x. And f continuous
almost everywhere U(x) = L(x) almost everywhere f Riemann
integrable. So f Riemann integrable f continuous almost every-
where. So (II) is proved.