Notes on the Lebesgue Integral

by
Francis J. Narcowich
Septemmber, 20141

1 Introduction
In the definition of the Riemann integral of a function f (x), the x-axis
is partitioned and the integral is defined in terms of limits of the Riemann
sums n−1 ∗ )∆ , where ∆ = x
P
j=0 f (x j j j j+1 −xj . The basic idea for the Lebesgue
integral is to partition the y-axis, which contains the range of f , rather than
the x-axis.
This seems like a “dumb” idea at first. Shouldn’t the two ways end
up giving the same integral? Most of time this is the case, but Lebesgue
was after integrating some functions for which the Riemann integral doesn’t
exist; for example, the Dirichlet function, which is defined on [0, 1]:

0 x rational,
χ(x) = (1)
1 x irrational.

Lebesgue’s reasoning was that there were uncountably many irrationals ver-
sus countably many rationals, so the “area” should be 1. It is is easy to show
that the Riemann integral doesn’t exist for χ. The integral Lebesgue came
up with not only integrates this function but many more. It also has the
property that every Riemann integrable function is also Lebesgue integrable.
Many of the common spaces of functions, for example the square inte-
grable functions on an interval, turn out to complete spaces – Hilbert spaces
or Banach spaces – if the Riemann integral is replaced by the Lebesgue
integral. The idea of splitting the range (y-axis) rather than the domain (x-
axis) turns out to be invaluable in dealing with integrating functions over
domains that aren’t just real numbers. Such integrals arise in many fields,
probability theory for instance. Lebesgue’s idea turns out to be a brilliant
“dumb” idea.
We now turn to the technical details involved in the Lebesgue integral,
starting with Lebesgue sums. Choose an increasing sequence of points P =
{c ≤ y0 < y1 < y2 < · · · < yn ≤ d}, where the range of f is contained
in [c, d]. As usual, we set kPk := max0≤j≤n−1 (yj+1 − yj ) Let Ej = {x ∈
[a, b] : yj ≤ f (x) < yj+1 } = f −1 ([yj , yj+1) )) and choose a point yj∗ from each
1
Revised, 9/5/2021.

1
interval [yj , yj+1 ]. (See Figure 1. Note that we have chosen yj∗ = yj+1 = t.)
The corresponding Lebesgue sum is
n−1
X
LP, Y (f ) :=
∗ yj∗ µ(Ej ) (2)
j=0

where µ(Ej ) denotes the “measure” or “length” of the set Ej and Y ∗ =

{yj∗ }n−1
j=0 For this sum to make sense, we need a concept of measure for
more sets than just intervals. For example, χ−1 ([1/2, 3/2)) is the set of all
irrational numbers between 0 and 1. It doesn’t contain any intervals at all.
This leads to the question of how to extend the concept of measure to subsets
of the real line that are much more complicated than simple intervals.

Figure 1: The region where f is defined and a typical horizontal slice.

(Downloaded on September 5, 2021: Slawomir Bialy at English Wikipedia,
CC BY-SA 4.0)

2 Measurable Sets
The ordinary idea of the measure/length of an interval can be thought of as
a function that assigns to an interval a nonnegative number. In addition, µ
satisfies standard properties; for example, if I and J are disjoint intervals,
then µ(I ∪ J) = µ(I) + µ(J).
An appropriate generalization of µ would satisfy the same properties,
but for a wider class of subsets of [a, b] than just the collection of intervals.
Suppose that Σ is a collection of subsets of [a, b] and that we have a function

2
µ : Σ → R. This is called a set function, because its domain consists
of subsets of [a, b]. To go further and require that µ satisfies the same
properties as those of length on intervals, we have to put conditions on Σ.
For example, if we want µ to satisfy µ(A ∪ B) = µ(A) + µ(B) for A ∩ B = ∅,
then A ∪ B must be in Σ. Here is a list of conditions that we will require Σ
to satisfy:

1. Σ is non-empty. It will always contain both ∅ and [a, b].

2. Σ is closed under complementation: A ∈ Σ if and only if A{ ∈ Σ.

3. Σ is closed under countable unions: {Aj ∈ Σ}∞ ∞

j=1 then ∪j=1 Aj ∈ Σ.

A collection of subsets Σ that satisfies that satisfies these conditions is called

a σ-algebra.
We can now specify the properties that a set function µ : Σ → [0, ∞)
requires so that it mimics length for intervals. We say that µ is a (σ-finite)
measure if and only if

1. µ([a, b]) = b − a and µ(∅) = 0.

2. Non-negativity: µ(A) ≥ 0 ∀A ∈ Σ.

3. Monotonicity: If A ⊆ B, then µ(A) ≤ µ(B).

4. Countable Additivity: If {Aj ∈ Σ}∞

j=1 , with Ai ∩ Aj = ∅, i 6= j, then
∞
P∞
µ(∪j=1 Aj ) = j=1 µ(Aj ).

These are fairly general conditions and they form the basis of the field of
measure theory. Our aim here is simply to construct the Lebesgue measure.
We note that the most natural choice for the class Σ would be the one com-
prising all subsets of [a, b]. Unfortunately, a theorem whose proof employs
the axiom of choice shows that there is no measure for this class. Instead,
we will use a procedure that simultaneously constructs µ and Σ.
We start with the outer measure of a set, which can be defined for
arbitrary subsets of [a, b]. Every open set G in [a, b] is the disjoint union
∗
P intervals, G = ∪i (ai , bi ) and we define the outer measure to be µ (G) =
of
i (bi − ai ). This is of course the natural generalization of length to open
sets.
We now turn to the general case. Let A ⊆ [a, b] for some finite a 6= b ∈ R.
We define the outer measure µ∗ (A) by

µ∗ (A) = inf{µ∗ (G) : A ⊂ G , G is open in [a, b]}.

3
We can also define the inner measure µ∗ (A):

µ∗ (A) = b − a − µ∗ ([a, b]\A).

It is always true that both the inner and outer measure are nonnegative and
that µ∗ (A) ≤ µ∗ (A).2 If µ∗ (A) = µ∗ (A), then we say that A is Lebesgue
measurable 3 .
The class Σ is then just defined as all the collection of all Lebesgue
measurable sets and the Lebesgue measure of A ∈ Σ is defined as µ(A) :=
µ∗ (A) = µ∗ (A). It is not hard to show that all open sets and all closed sets
are measurable, and that if A is measurable, so is its complement A{ . With
more work one can show that the measurable sets form a σ-algebra and the
Lebesgue measure is a non-negative measure defined on Σ. When we are
dealing only with intervals, the Lebesgue measure coincides with the usual
concept of length.
The inequality µ∗ (A) ≤ µ∗ (A) mentioned above is often useful. For
instance, a set has measure 0 if for every  > 0 there is an open set G ⊃ A
such that µ(G) < . Since µ∗ (A) = inf{µ∗ (G) : A ⊂ G} < ε for all ε > 0,
we have that µ∗ (A) = 0. However, we also have 0 ≤ µ∗ (A) ≤ µ∗ (A) = 0.
Therefore, A is measurable and µ(A) = µ∗ (A) = µ∗ (A) = 0. Here is an
important example.
Example 1. Every countable set4 has measure 0.
 
Proof. Let A = {x1 , x2 , . . . , xn , . . .}. Let Ij = (xj − 2j
, xj + 2j
). Then, we
see that A ⊂ ∪Ij , and hence

µ∗ (A) = inf{µ∗ (G) : A ⊂ G , G is open}

X
≤ µ∗ (∪j Ij ) ≤ µ∗ (Ij )
j
∞
X  
= ( j
+ xj − xj + j )
2 2
j=1
∞
X 
= =
2j−1
j=1

Since this holds for arbitrary  > 0, we see that A is measurable and that
µ(A) = 0.
2
See Wilcox & Myers, p. 25, Lemma 8.7.
3
The previous inequality implies that to verify A is measurable only requires showing
that µ∗ (A) ≤ µ∗ (A).
4
The cantor set is an example of an uncountable set having measure 0.

4
3 Measurable Functions
In order to form the Lebesgue sums in (2), we need to be able to find the
measure of sets of the form f −1 ([c, d)). This puts a restriction on f .

Definition 1. We say the f : [a, b] → R is (Lebesgue) measurable if and

only if for very c < d the set f −1 ([c, d)) is measurable.

There are many conditions equivalent to f −1 ([c, d)) being measurable.

In particular if any of these sets is measurable for all c < d, then f is
a measurable function: f −1 ((c, ∞)), f −1 ([c, ∞)), f −1 ((c, d)), f −1 ((c, d]),
f −1 ((−∞, c)), and other similar sets. It’s also useful to know that if, for ev-
ery open set G ⊆ [a, b], f −1 (G) is measurable, then f is measurable. In fact,
a definition of measurable functions that applies for more general measure
spaces is this: We say that f : A → R is measurable if for every measurable
E ⊂ R, f −1 (E) ⊆ A is a measurable subset of A.
Measurable functions may be combined in various ways to obtain other
measurable functions. If f and g are measurable, then so are af + bg, f · g,
f /g (g 6= 0), |f |. Every continuous function is measurable. If f is continuous
and the range of g is in the domain of f , then f ◦ g is measurable. (The
reverse composition, g ◦ f might not be mensurable. See Wilcox & Myers,
p. 56, exercise 20.12.)
In the theory of Lebesgue integration, sets of measure 0 really won’t
contribute to an integral. Consequently, when we integrate two functions
that are different only on a set of measure 0, we will find that their integrals
will be the same. This situation occurs frequently enough that we make the
definition below.

Definition 2. We say that f equals g almost everywhere (a.e.) if and only

if the set of x for which f (x) 6= g(x) has measure 0.

A nice example of this is the Dirichlet function χ(x) defined in (1). Recall
that χ(x) = 1, except on Q. Since Q has measure 0, χ = 1 a.e. Here is one
of the more important facts about measurable functions.

Proposition 1. Suppose that A is a measurable set and that fn : A → R

is a sequence of measurable functions such that for each x ∈ A we have
limn→∞ fn (x) = f (x), then f is measurable.

Definition 3. Let A be a measurable set. We say that a measurable function

s : A → R is simple if and only if the range of s has a finite number of values.

5
 The characteristic or indicator function of a measurable A is defined by
(
1 x∈A
χA (x) =
0 x∈ / A.

Since the range of this function is {0, 1}, it is simple. In fact, any simple
function can be represented as a finite linear combination of characteristic
functions.

Proposition 2. Let s : A → R be a simple function with range {dj }nj=1 ,

n < ∞, and let Ej = s−1 ({dj }). Then, s has the form
n
X
s= dj χEj .
j=1

Conversely, if s has the form above, then it is simple.

The representation for s given above is unique, provided the Ej =

s−1 ({dj }). Otherwise, one may obtain different combinations of charac-
teristic functions that result in the same s.

4 The Lebesgue Integral

We briefly introduce the concept of the Lebesgue integral in this section,
and in addition, discuss some important theorems associated with it. We
start by returning to the Lebesgue sum in (2). If f is bounded and mea-
surable
Pn−1 ∗on a measurable set A, then µ(Ej ) is defined, and so is LP, Y+ (f ) =
∗

y
j=0 j µ(E j ). At this point, we may define upper and lower sums LP (f ) =
Pn−1 − Pn−1
j=0 yj+1 µ(Ej ) and R LP (f ) = j=0 yj µ(Ej ) and then proceed to define
the Lebesgue integral A f (x)dx in roughly the same way as one defines the
Riemann integral.
Another, somewhat easier, equivalent way is to start out by defining the
Lebesgue integral of a simple function s to be
Z n
X
s(x)dµ = dj µ(Ej ), Ej = s−1 ({dj }). (3)
A j=1

Then we may define upper and lower integrals for a bounded, measurable

6
function f via
Z + Z

f (x)dµ = inf s(x)dx : s(x) ≥ f (x), s simple
A s A
Z − 
Z

f (x)dµ = sup s(x)dx : s(x) ≤ f (x), s simple
A s A
R R+ R−
As usual, if these are equal, then A f (x)dµ := A f (x)dµ = A f (x)dµ. It
turns out that for a bounded measurable function both exist and are equal:
Proposition 3. Let f be aR bounded, measurable
R− function on a bounded
+
measurable set A ⊂ R. Both A f (x)dµ and A f (x)dµ always exist and are
equal. Thus, Z Z + Z −
f (x)dµ = f (x)dµ = f (x)dµ (4)
A A A

R The Lebesgue
R integral on a measurable set A will be denoted by either
A f (x) dx or A f (x) dµ.
We now turn to unbounded functions – e.g., f (x) = x−1/2 on [0, 1]. For
a measurable function f (x) ≥ 0, whether bounded or not,
Z

sup s(x)dx : s(x) ≤ f (x), s simple ,
s A

always exists – even if it’s infinite. If the sup is finite, then we can define
the Lebesgue integral of f to be
Z Z

f (x)dµ := sup s(x)dx : s(x) ≤ f (x), s simple , (5)
A s A

Remark: One can’t use the upper bound formulation because if f is un-
bounded, one can get below the function with a simple function, but not
above it.
We can write f as the difference of two non-negative functions, f =
f+ −f− , where f+ = 21 (f +|f |) and f− = 12 (|f |−f ). Also, |f | = f+ +f− . We
sayR that a measurable function f , whetherRbounded or not, is integrableR on A
if A R|f |dµ < ∞. It’s easy to show that if A |f |dµ exists,
R then both R A f+ dµ
and
R A −f dµ exist. Consequently, we may define RA f dµ = A f+ dµ −
A f− dµ. Note that both integrals must be finite for A f dµ to exist.
This definition causes problems for certain Riemann integrablel R ∞ func-
tions. Here’s an example for an unbounded interval. The integral 0 sin(x) x dx =
R ∞ | sin(x)|
π/2, in the Riemann sense. However, 0 x dx = ∞, so the integral does
not exist in the Lebesgue sense.

7
 Let A, B denote measurable sets, with µ(A), µ(B) < ∞. The properties
and theorems below are satisfied by the Lebesgue integral.
1. If f is a bounded and measurable function, then the integral exists.

2. Integration is a linear
R operation. That Ris, if f and gR are both integrable
and a, b ∈ R, then A (af + bg) dµ = a A f dµ + b A g dµ.

3. RIf A f (x)2 dx and A g(x)2 dx both exist, then A f (x)g(x) dx and

R R R
2
A (f (x) + g(x)) dx exist.
R R R
4. A∪B f dµ = A f dµ + B f dµ, where µ(A ∩ B) = 0.
R R
5. If f = g almost everywhere, then A f dµ = A g dµ.

6. If the Riemann integral of f exists, then the Lebesgue integral exists

and the integrals are equal.
Theorem 1. (Monotone Convergence Theorem): Let {fj } be a collection
of measurable functions on A that satisfy 0 ≤ f1 ≤ f2 ≤ · · · ≤ fn ≤ . . .
almost everywhere. Define f as the pointwise limit f (x) = lim fj (x). Then,
f is measurable and Z Z
lim fj dµ = f dµ.
A A

Theorem 2. (Dominated Convergence Theorem): Let {fj } be a set of mea-

surable functions that converges pointwise to a function f and assume there
exists an integrable function g such that |fj (x)| ≤ g(x) almost everywhere.
Then, f is integrable and
Z Z
lim fj dµ = f dµ.
A A

Theorem 3. (Fubini’s Theorem): Let f be measurable on A × B. If

Z
|f (x, y)| dµ(x, y) < ∞
A×B

then A B f (x, y) dµ(x)dµ0 (y) exists and the order of integration may be
R R

switched.
We define the space Lp ([a, b]) to be the space of p integrable functions
for 1 ≤ p < ∞. That is,
Z b
p
L ([a, b]) = {f : |f (x)|p < ∞}.
a

8
The space Lp ([a, b]) is a normed space when it is equipped with the norm
Z b
1
kf kp = ( |f (x)|p dµ(x)) p .
a

For the case p = ∞, we define

kf k∞ = ess sup|f | = inf{a ∈ R : µ({x : |f |(x) > a}) = 0}

and we define
L∞ ([a, b]) = {f : kf k∞ < ∞}.

Example 2. Just because a function is in L∞ doesn’t necessarily mean that

it is bounded in the usual sense. Here is an example of this.
(
1 t ∈ R\Q
f (t) = (6)
q t = pq

where pq is reduced to lowest terms and q > 0. This function is unbounded,

but kf k∞ = ess sup|f (x)| = 1, because the set of rational numbers has
measure zero. This implies that, although f is unbounded, f ∈ L∞ .

Theorem 4. The space Lp ([a, b]) is a complete space for 1 ≤ p ≤ ∞.

This theorem implies that Lp ([a, b]) is a Banach space. In the special
case of p = 2, L2 ([a, b]) is a Hilbert space.

5 Examples
Example 3. Suppose fn (x) = q1 for x ∈ [0, 1] and n ≥ 1. Let f denote
1
x+ n
the pointwise limit of fn . Show that f is integrable.

Proof. We see that fn+1 (x) = q 1 ≥ q1 = fn (x) holds for all n and
1 1
x+ n+1 x+ n
limn→∞ fn (x) = √1 almost everywhere. Then, by the monotone conver-
x
R1 R 1 dx
gence theorem, √1 is integrable and limn→∞ 0 q dx 1 = 0 √ .
x x
x+ n

R∞
Example 4. Let f ∈ L1 (R) and define fˆ(ω) = −∞ f (t)e
−iωt dt. Show that
fˆ is a continuous function.

9
Proof. By a straightforward computation, we see that
Z ∞
ˆ ˆ
f (ω + h) − f (ω) = f (t)(e−i(ω+h)t − e−iωt ) dt
−∞
Z ∞
= f (t)e−iωt (e−iht − 1)dt
−∞

The integrand is bounded above by 2|f (t)| almost everywhere, and since
f ∈ L1 , 2|f (t)| is integrable. Applying the dominated convergence theorem
results in limh→0 fˆ(ω + h) = fˆ(ω), thus fˆ is continuous.

Previous: Banach spaces and Hilbert spaces

Next: Orthonormal sets & expansions

10