View metadata, citation and similar papers at core.ac.

uk brought to you by CORE

provided by Liberty University Digital Commons

A Short Journey Through the Riemann Integral

Jesse Keyton
April 23, 2014

Abstract
An introductory-level theory of integration was studied, focusing primarily on the well-known Rie-
mann integral and ending with the Lebesgue integral. An examination of the Riemann integral’s basic
properties and necessary conditions shows that this integral is not very strong. This conclusion leads
to Lebesgue’s necessary and sufficient condition for Riemann integrability, perhaps where the Riemann
integral ends and the Lebesgue integral begins: a function defined on a closed integral [a, b] is Riemann
integrable if and only if the function is discontinuous on a set of measure zero. The proof of this uses that
the infinite union of sets of measure zero has a measure of zero. To distinguish between the Riemann and
Lebesque integral, the classical example of the dirac-delta function displays the strength of the Lebesgue
integral over the Riemann integral. In conclusion, every function that is Riemann integrable is Lebesgue
integrable, but the converse is not true.

1 Introduction: Integration
When talking about integration, the question being asked is what the area is between a real-valued function
and the line y = 0, at least for functions defined on the real numbers. This paper seeks to understand
the theory behind integration, how to build a useful tool in studying calculus. Typically the student learns
the calculational method (specifically the anti-derivative) in pre-calculus or calculus I, but the tools the
Riemann developed in his analysis explain why those calculational methods work. A major result of his
analysis is both parts of the fundemental theorem of calculus, and his integral is useful in many other areas
of theoretical study. However, as the conclusion of this paper approaches it becomes evident that there are
many instances when Riemann’s idea cannot be used. There are several other types of integrals, some a more
advanced version of the Riemann, but others formed from a new idea completely. In regards to the latter,
Henri Lebesgue came up with an idea of integration that is much more advanced and much stronger than
the Riemann integral, covering a much wider range of functions. However, this paper does not dive too far
into what the integral really is. A proper introduction would require much background information in other
areas of mathematics. Therefore, in this paper I am focusing on the Riemann integral and answering the
question of when this integral will actually work. I will start out explaining what the integral is, provide some
properties, and the reveal an example where this integral fails. In conclusion, I will provide a theorem from
Lebesgue which is a criteria for Riemann integration and then get into some topics for further discussion.

2 The Riemann Integral

The basic idea behind the Riemann integral is to look at a bounded function on a closed interval [a, b] into
R, and to create rectangles with a base from a partition of [a, b] and the height from the max and min values
of the function within each base. The summation of the areas of the rectangles can approximate the area
under the curve given by the function.

Definition: Upper/Lower Sums. Let f be a bounded, real-valued function defined on a closed interval
[a, b]. We say the set P = {a = x0 , x1 , ..., xn−1 , xn = b} is a partition of [a, b] and for any i ∈ {1, ..., n} let

Mi = sup{f (x) : x ∈ [xi−1 , xi ]}

1
and
mi = inf {f (x) : x ∈ [xi−1 , xi ]}.
We call an Upper Sum of f with respect to P :
n
X
U (f, P ) = Mi ∆xi
i=1

and the Lower Sum of f with respect to P :

n
X
L(f, P ) = mi ∆xi
i=1

where ∆xi = (xi − xi−1 ) for each i.

Remark: The ”sup” and ”inf” used above are called the greatest lower bound and least upper bound,
respectively, and to avoid a careful, detailed and unnecessary explanation for my purposes here, they can be
thought of as the maximum and minimum values, respectively.

Definition: Upper/Lower Integrals. For a bounded function f on a closed interval [a, b], the Upper
Integral is denoted as
U (f ) = inf {U (f, P ) : P is a partition of [a, b]}
and the Lower Integral is denoted as
L(f ) = sup{L(f, P ) : P is a partition of [a, b]}
The function f is Riemann Integrable if and only if U (f ) = L(f ), and when this is true its value is
denoted by
Z b
f = U (f ) = L(f )
a

2.1 Some Immediate Observations of the Riemann Integral

Based on the defintions given, there should be some immediate questions to be asked:
1. What if a function f is defined on an open interval? A half-open interval?
2. What if a function is not bounded?
3. If the function is negative at some points, how does that affect the interpretation of the integral?
To answer the first question, without being defined on a closed interval, the Fundemental Theorem of Cal-
culus fails because we need uniform continuity of the function. Concerning the second observation, if the
function is not bounded, then there would be no way to calculate the area of at least one rectangle in the
partial sums. Hence the area of an unbounded function exceeds any real number.

Both of these issues lead to a definition of an ”improper integral:”

Let f be defined on (a, b] and integrable on [c, d] for all c ∈ (a, b]. The improper integral of f on (a, b] is
Z b Z b
f = lim f.
a c→a+ c

When the improper integral exceeds any finite number, it is said to diverge. When it doesn’t, it converges.
An interesting observation is that f does not need to be defined at a. We could also look at functions defined
on the whole real number line by a similar idea.

As for the third question, negative values in the range of f does not compromise the definitions given
at the start. Riemann integration can still be attained by such a function, but it may skew the way we look
at the area of the function under its curve, as its negative area will surely ”take away” from the positive
area.

2
2.2 Properties/Conditions for Riemann Integration
Looking at area approximations with small partitions is not very effective. The whole idea of Riemann
integration is to look at the sums as the partitions increase in size. In fact, many definitions of the Riemann
integral involve the limits of the sum, which beckons the efficiency of working with sequences as learned in
Real Analysis. For instance, consider the following proposition:

Proposition 2: Let f be a bounded function on [a, b]. Suppose that there exists a sequence {Pn } of
partitions of [a, b] such that
limn→∞ [U (f, Pn ) − L(f, Pn )] = 0
Let ε > 0. Then there exists N ∈ R such that for any n > N it follows

|U (f, Pn ) − L(f, Pn )| = U (f, Pn ) − L(f, Pn ) < ε

which means that f is Riemann integrable. Furthermore, since

L(f, Pn ) ≤ L(f ) = U (f ) ≤ U (f, Pn ) ∀n,

0 ≤ L(f ) − L(f, Pn ) ≤ U (f, Pn ) − L(f, Pn ) → 0

So by squeeze theorem, L(f ) = limn→∞ L(f, Pn ) = limn→∞ U (f, Pn ).

So we can view the integral as a limit of the area of the rectangles. As we shall see with the following
definitions and results, the whole idea of the Riemann integral is to make the partions really big so we can
get an accurate approximation of the area.

Now we give definitions of refinements and some comparitive observations between the partition and its
refinement.

Defintion: Refinements. Let f be a bounded function on a closed interval [a, b] ∈ R with a partion P of
[a, b]. A refinement Q of P is a partition of [a, b] with P ⊆ Q. Note that equality of the two partitions is
entirely useless.

An immediate consquence of this defintion is given in the following proposition:

Proposition 1: L(f, P ) ≤ L(f, Q) ≤ U (f, Q) ≤ U (f, P )

Proof: Let P = {x0 = a, x1 , ..., xn−1 , xn = b} be a partion of [a, b] and let Q1 be another partition of
[a, b] constructed by adding a point t ∈ [a, b], t 6∈ P to P . Then there exists some i ∈ {1, ..., n} such that
xi−1 < t < xi . So let
y1 = inf {f (x) : x ∈ [xi−1 , t]},
y2 = inf {f (x) : x ∈ [t, xi ]}.
Remember mi = inf {f (x) : x ∈ [xi−1 , xi ]}, which means that y1 , y2 ≥ mi (We are looking at every point in
the interval, not just the endpoints). Now
i−1
X i+1
X n
X
L(f, Q1 ) − L(f, P ) = mj ∆xj + y1 (t − xi−1 ) + y2 (xi − t) + mj ∆xj − mj ∆xj
j=1 j=1 i=j

= y1 (t − xi−1 ) + y2 (xi − t) − mi (xi − xi−1 )

= y1 (t − xi−1 ) + y2 (xi − t) − mi (t − xi−1 ) − mi (xi − t)
= (y1 − mi )(t − xi−1 ) + (y2 − mi )(xi − t)
≥0 Since each term is nonnegative.

Hence, L(f, P ) ≤ L(f, Q1 ).

A similar argument will show that U (f, Q1 ) ≤ U (f, P ), and it is easily seen by defintion that

3
L(f, Q1 ) ≤ U (f, Q1 ). If a partition Qn has n more points that P , then we can just repeat this argument n
times. ♦

Now what does this proposition actually tell us? For the lower sums, this means that the more points
we have for a partition, the sum approximation of the integral becomes more accurate and the lower rectan-
gles are becoming ”bigger,” or rather closer to the actual curve. The same is true for the upper sums, but
for these the rectangles are shrinking in area. But even these statements, although important to notice, do
not say much about a function that is Riemann integrable. This next theorem provides a sufficient condition
for such a function.

Theorem 1: Let f be a bounded function of the closed interval [a, b] ∈ R. Then f is Riemann integrable if
and only if for any ε > 0 there exists a partition P of [a, b] such that

U (f, P ) − L(f, P ) < ε.

Proof: Suppose first that f is Riemann integrable and let ε > 0. By definition, U (f ) = L(f ). Using
supremum and infimum definitions, there exists some partition P of [a, b] such that
ε
L(f, P ) > L(f ) − ,
2
and a partition Q of [a, b] such that
ε
U (f, Q) < U (f ) + .
2
Then for the refinement R = P ∪ Q we have
ε
L(f, R) ≥ L(f, P ) > L(f ) −
2
and
ε
U (f, R) ≤ U (f, Q) < U (f ) + .
2
Putting these together we obtain:
ε ε
U (f, R) − L(f, R) ≤ U (f, Q) − l(f, P ) < (U (f ) + ) + (−L(f ) + ) = ε.
2 2

For the converse, with ε > 0, suppose that there exists a partition P of [a, b] such that U (f, P ) − L(f, P ) < ε,
or U (f, P ) < L(f, P ) + ε. Then,

U (f ) ≤ U (f, P ) < L(f, P ) + ε ≤ L(f ) + ε

so that, put more visibly,

U (f ) ≤ L(f ) + ε.
But since ε is arbitrary, it follows that U (f ) ≤ L(f ), and since L(f ) ≤ U (f ), U (f ) = L(f ), which means
that f is Riemann integrable. ♦

In conclusion, if we can make the upper sum and lower sum of f with repect to some partition arbi-
trarily small, it is reasonable to say that the upper integral and lower integral are the same value, and
hence is the Riemann integral. But what kind of functions does this work for? As we shall soon see,
there are certain limitations for a function that this will not work for. Before we get to those limitations,
it will be important to discuss the idea of continuity of a function from R to R. While there are several
equivalent conditions for continuity, I will focus on one definition that will be most helpful for this discussion.

Definition: Continuous Function. A function f : D → R is continuous at a point c ∈ R provided

that for any ε > 0 there exists some δ > 0 such that for any x ∈ D satisfying |x − c| < δ it follows that

4
|f (x) − f (c)| < ε. If this holds for every c ∈ D then f is a continuous function.

What this is saying is that if we take any open interval, call it V , around a point in the range of f we can
find an open interval, call it U , in the domain of f such that f maps U to a set contained in V . The function
that is continuous is not very erratic; there are no sudden ”jumps” in the range values, nor are there certain
points of the domain that the function is not defined for. An even nicer kind of function is given by the
following definition.

Definition: Uniformly Continuous Function. A function f : D → R is uniformly continuous provided

that for any ε > 0 there exists some δ > 0 such that for any x, y ∈ D satisfying |x − y| < δ it follows that
|f (x) − f (y)| < ε.

When a function is continuous on a closed interval in R, it follows that f is uniformly continuous and that
the function attains a maximum and minimum value. Hence continuity on a closed interval is an overkill for
Riemann integrability, since we can just increase the partition large enough to make the ∆xi ’s sufficiently
small, which would keep the differences of the range values for the Upper and Lower sums sufficiently small
as well. A formal proof is given below.

Theorem 2 Let f be a continuous function on [a, b]. Then f is Riemann integrable on [a, b].
Proof: Suppose f is a continuous function of the closed interval [a, b] ∈ R. Then f is uniformly continuous.
Hence, given any ε > 0, there exists a δ > 0 such that if x, y ∈ [a, b] and |x − y| < δ it follows that
ε
|f (x) − f (y)| < b−a . Hence, as mentioned in the above comment, let P = {x0 = a, x1 , ..., xn−1 , xn = b}
be a partition of [a, b] such that ∆xi < δ for every i. As mentioned in the observations, f also attains its
maximum and minimum on [a, b], and therefore for each subinterval created by the partition. More precisely,
for any interval [xi−1 , xi ] there exists points ai , zi ∈ [xi−1 , xi ] such that mi = f (ai ), Mi = f (zi ). But since
ε
|ai − zi | < δ, it follows that |f (zi ) − f (ai )| = f (zi ) − f (ai ) < b−a . Now we have
n n n
X X X ε ε
U (f, P ) − L(f, P ) = (Mi − mi )∆xi = (f (zi ) − f (ai ))∆xi < ∆xi = (b − a) = ε
i=1 i=1 i=1
b−a b−a

so that f is Riemann integrable by Theorem 1. ♦

Now that we have a basic idea of the Riemann integral and some defining characterstics of it, we can take a
look at some of its limitations that will lead to a nice criteria for Riemann integrability.

3 Limitations of the Riemann Integral

1. Fails when a function is discontinuous on a set of nonzero measure (Lebesgue’s criteria).

Example (Dirichlet function (figure 1)):


0 :x∈Q
f (x) =
1 :x∈
/Q

We have, for any partition P of the domain,

L(f, P ) = 0 and U (f, P ) = 1,

because the minimum value in each subinterval is 0, while the maximum value for each subinterval is
1, since there will be an infinite amount of both rational and irrational numbers in each subinterval.
Therefore, Riemann integrability fails for this function. Notice also that the Dirichlet function is
discontinuous at every point in its domain. To see this, take an open set around any (of the two) point
in the range, let’s just say a neighborhood centered around f (m) = 1, where m is a rational number,
with a radius of the neighborhood strictly less than 1. Now take any open set in the domain centered

5
 Figure 1: Dirichlet function defined on [0, 1]

around m. It does not matter how small you make this open set around m, it will always contain an
irrational number, in fact it will contain an infinite amount of irrational numbers! Then when you map
the set, there will be many values that go to 0, which means that the mapping of the set will miss the
neighborhood around f (m), as this neighborhood does not contain the point 0 because of the radius
we gave it. Therefore, this function is not continuous at m, or at any other point in the domain. Hence
the Dirichlet function is nowhere continuous. There are two resoultions to this classical example: we
can redifine the function to make it only discontinuous at the rationals, or we could use a different
integral. We will get back to these later on.
2. Convergence of simple functions may not be integrable (Fourier Series/Fourier Transforms)
3. Foundationally the Riemann integral is not as strong as Lebesgue, as the Lebesgue uses an abstract con-
cept of measure to be able to extend the integration for a much richer mathematical theory. Lebesgue is
also much stronger because of the use of convergent sequences in the theory, which provides much more
rigor and therefore strength in its analysis (at least from how I, in my naivity, perceive it). Speaking
of Measure Theory, I will talk about a result from measure theory that will be useful in establishing
the criteria for Riemann integrability.

4 Sets of Measure Zero

In order to understand the culmination of the Riemann integral, a look at measurable sets is necessary. To
start the discussion, consider the interval [0, 1]. It is intuitively clear that this set takes up some space on
the real number line, and we say it has a measure of 1. How about [0, 1] ∪ [4, 5]? One result of measure
theory is that the mesaure of the union of disjoint mesaurable sets is just the summation of the individ-
ual measures, so in this case the measure µ([0, 1] ∪ [4, 5]) = µ([0, 1]) + µ([4, 5]) = 1 + 1 = 2. However, for
the Riemann intergal, our main interest lies in sets of mesaure zero, to which we shall now turn our attention.

Consider the set S1 = {1}. We can picture this point on the number line, but upon the realization that there
are an uncountably infinite amount of points in any neighborhood of 1, it seems as though just one point
does not occupy any space at all on the line. In fact, for an arbitrarily small positive number, say ε, we can
find some interval I1 = (1 − 3ε , 1 + 3ε ) which has S1 ⊆ I1 and µ(I1 ) = (1 + 3ε ) − (1 − 3ε ) = 23 ε < ε. We can
now formalize this observation in a definition.

Definition: Lebesgue Measure Zero. The set S ⊆ R has Lebesgue Measure Zero, i.e. µ(S) = 0, if
and only if for any ε > 0 there exists a family of intervals {Ik } such that:
S∞
(i) S ⊆ j=1 Ij
∞
P
(ii) µ(Ij ) < ε.
j=1

6
Remark: It is not hard to show that any denumerable set contained in R has Lebesgue measure zero. To do
this, we find intervals as we did above for every number in the set and use a geometric series to prove (ii) of
the definition. A more interesting and useful theorem is the following:

Theorem 3: If a denumerable collection of subsets of R, {S1 , S2S , ..., Sn , Sn+1 , ...}, has the property that
∞
each set in the family has Lebesgue measure zero, the the set S = j=1 Sj has Lebesgue measure zero.
Proof: For each Si ∈ {Sj }∞ i
j=1 , there exists a family of intervals {Ik } associated with Si such that

∞ ∞
[ X ε
Si ⊆ Iki and µ(Iki ) < ,
2i+2
k=1 k=1
S∞
by definition. Let the set HSi = k=1 Iki for every i. Now consider the family of sets {H 1 2 n
S∞ , H , ..., H , ...}.
∞
If there is an element x ∈ j=1 Sj , then x ∈ St for some t ∈ N. This implies x ∈ j=1 Ijt = H t . But
S∞ S∞ S∞ S∞
H t ⊆ j=1 H j which implies x ∈ j=1 H j . Hence S = j=1 Sj ⊆ j=1 H j .
Also, for any m ∈ N,
ε
µ(H m ) < m+2 .
2
Hence,
∞ ∞
X ε ε ε εX 1 ε
µ(H j ) ≤ 1+2 + 2+2 + ... + n+2 + ... = j
= < ε. ♦
j=1
2 2 2 4 j=1
2 4

Now the question of discontinuity on a set of measure zero can be understood, but there is still some
more prerequisites that must be met before getting to the Lebesgue’s theorem.

5 Lebesgue’s Criteria For Riemann Integrability

Before jumping right into the theorem, an extremely useful tool called the oscillation of a function should
be discussed.
Definition: Oscillation of f on S. Let f : D → R be a bounded function and let C ⊆ D. The oscillation
of f on S is defined as
W (f ; C) = sup{|f (x) − f (y)| : x, y ∈ C}.
Remember in the discussion of continuous functions, we were concerned about a neighborhood around a
point in the domain. In order to see how the oscillation of a function relates to the function’s continuity, we
define the oscillation at a specific point.

Definition: Oscillation of f at c. With c ∈ C, the oscillation of f at c is defined as

w(f ; c) = inf {W (f ; V (c; δ)) : δ > 0}

where the neighborhood V (c; δ) = {x ∈ D : |x − c| < δ}.

Remark: What is the second definition actually saying? We are looking at neighborhoods around a point in
the domain. W (f ; C) calculates the greatest difference in range, so w(f ; c) calculates the greatest difference
in the range when their respective values in the domain are made arbitrarily close together. So as the δ gets
smaller and smaller, the differences in the range should be getting smaller and smaller (it is certain the dif-
ferences are not increasing), at least for continuous functions as we shall prove shortly. Hence an immediate
conclusion can be that w(f ; c) ≤ W (f ; V (c; δ)). Also, this function is nonnegative, as f (c) − f (c) = 0 and
0 will be greater than any negative difference. These insights will be useful when proving the following lemma.

Lemma: If f : D → R, then f is continuous at c if and only if w(f ; c) = 0.

Proof: Suppose f is continuous at c ∈ D. Let ε > 0. Then there exists some δ > 0 such that for any x ∈ D

7
with |x − c| < δ we have |f (x) − f (c)| < 2ε . This tells us that for every x, y ∈ V (c; δ), |f (x) − f (c)| < ε
2 and
|f (y) − f (c)| < 2ε . Hence,
ε ε
|f (x) − f (y)| = |f (x) − f (c) + f (c) − f (y)| ≤ |f (x) − f (c)| + |f (y) − f (c)| < + =ε
2 2
and therefore, W (f ; V (c; δ)) = sup{|f (x) − f (y)| : x, y ∈ V (c; δ)} < . In conclusion,

0 ≤ w(f ; c) ≤ W (f ; V (c; δ)) < ε.

Since ε is arbitrary, w(f ; c) = 0.

Consider the converse, and suppose that w(f ; c) = 0 and let ε > 0. By definition, this means that
inf {W (f ; V (c; δ)) : δ > 0} = 0. But if this is true then there must be some δ > 0 such that W (f ; V (c; δ)) < ε,
which implies that for any x ∈ V (c; δ), |f (x) − f (c)| < ε. Hence f is continuous at c. ♦

Now we can gather all of our results and put it into a very important theorem for Riemann integration.
This proof will use the oscillation of a function to look at points of discontinuity, and then use the theorem
discussed from measure theory to prove the points of discontinuity are in a set of measure zero.

Theorem 4: Let f be a bounded, real-valued function on a closed interval [a, b]. Then f is Riemann inte-
grable if and only if f is only discontinuous on a set in [a, b] that has Lebesgue measure zero.

Proof: (⇒) Suppose first that f is Riemann integrable. Define a family of sets {H1 , H2 , ..., Hn , ...} each
1
in [a, b] such that Hj = {x ∈ [a, b] : w(f ; x) <
S∞2j }. If we can show each of these Hj ’s has Lebesgue measure
zero, then it will follow by Theorem 3 that i=1 Hi has Lebesgue measure zero.
Since f is Riemann integrable, there exists a partition Pj = {a = xj0 , xj1 , ..., xjn = b} of [a, b] such that
ε
U (f, Pj ) − L(f, Pj ) < .
4j
Suppose there exists an x ∈ Hj ∩ (xji−1 , xji ) for some i ∈ 0, 1, ..., n. Since (xji−1 , xji ) is open, there exists a
δ > 0 such that V (x; δ) ⊆ (xji−1 , xji ). Now for this x,

1
< w(f ; x) ≤ W (f ; V (x; δ)) = sup{|f (x) − f (y)| : x, y ∈ V (c; δ)} ≤ Mij − mji .
2j
Hence 21j ≤ Mij − mji and if we denote the set of i’s such that the intersection of Hj and (xji−1 , xji ) is
nonempty by T , then it follows that
n
1 X j j
X ε
(xi − xi−1 ) ≤ U (f, Pj ) − L(f, Pj ) = (Mij − mji )(xji − xji−1 ) < j .
2j i=1
4
i∈T

Hence X ε
(xji − xji−1 ) <
2j
i∈T
S j j
Let F = i∈T Hj ∩ (xi−1 , xi ),
and notice that Hj ⊆ F ∪ Pj . As we have already seen, µ(Pj ) = 0, so that
µ(Hj ) ≤ µ(F ∪ Pj ) < 2εj . S∞
Now let D = {x ∈ [a, b] : w(f ; x) > 0}, which is in fact the same as i=1 Hi , and we find that
∞ ∞ ∞
[ X X ε
µ(D) = µ( Hi ) = µ(Hi ) = =ε
i=1 i=1 i=1
2i

Therefore D, the set of discontinuous points in [a, b], has Lebesgue measure zero.

(⇐) Now consider the converse; suppose f is discontinuous on a set of measure zero. Let f : [a, b] → R be
a bounded function, i.e. |f (x)| ≤ M for some M ∈ R, and suppose that f is discontinuous only on a set of

8
Lebesgue measure zero. Let D denote the set of discontinuous points and let ε > 0. By definition it follows
S∞ ∞
that there is a collection of real intervals {Ik }∞ ε
P
k=1 such that D ⊆ k=1 Ik and µ(Ik ) < 2M . For each
k=1
t ∈ [a, b] there are two possibilities:
ε
1. t ∈
/ D. Then f is continuous at t, which means there is some δ(t) > 0 such that x ∈ V (t; δ(t)) ⇒ |f (x) − f (t)| < 2
. If
this is true, then
0 ≤ Mt − m t < ε

2. t ∈ D. Then choose a δ(t) > 0 such that V (t; δ(t)) ⊆ Ik for some k. Also, since f (x) is bounded by M , it follows that
0 ≤ Mt − mt ≤ 2M
Now let P = {a = x0 , x1 , ..., xn−1 , xn = b} be a partition of [a, b] such that when we choose ti ∈ [xi−1 , xi ] we have
(xi − xi−1 ) < δ(ti ).

To distinguish between discontinuous and continuous points, we define two sets:

Sc = {i : ti ∈
/ D} and Sd = {i : ti ∈ D}.

Notice that [xi−1 , xi ] ⊆ V (ti , δ(ti )) for every i. Then it follows that Mi − mi ≤ Mti − mti for all i. Moreover,
when i ∈ Sd , [xi−1 , xi ] ⊆ V (ti , δ(ti )) ⊆ Ik(i) for some k(i) and Mi − mi ≤ 2M . When i ∈ Sc , Mi − mi ≤ ε.
In conclusion:
n
X
U (f, P ) − L(f, P ) = (Mi − mi )(xi − xi−1 )
i=1
X X
= (Mi − mi )(xi − xi−1 ) + (Mi − mi )(xi − xi−1 )
i∈Sc i∈Sd
X X
< ε(xi − xi−1 ) + 2M · µ(Ik(i) )
i∈Sc i∈Sd
ε
≤ ε(b − a) + 2M ( ) = ε(b − a + 1)
2M
Since ε is arbitrary, it follows that f is Riemann integrable.

5.1 Dirichlet Function Revisited

With Lebesgue’s useful tool, consider a modification of the Dirichlet function (Figure 2)1 :
 1
n : x ∈ Q is in lowest terms
f (x) =
0 :x∈ /Q

I will prove that this modified verson, defined on [0, 1], is discontinuous on a set of measure zero:

Proof: Let t be an irrational number in [0, 1]. There exists a sequence of rationals {an } that converge to
t. Now let ε > 0. Then there exists a b ∈ N such that 1b < ε by the Archimedian Property. Notice that there
are only finite number of fractions in the interval [0, 1] whose denominator in lowest terms is less than b, and
therefore we can create a δ-neighborhood around t that will be absent of such fractions. Since lim an = t,
there exists a number N such that for any n > N we have |an − t| < δ. Then it follows that for any n > N
we have
1
|f (an ) − f (x)| = |f (an ) − 0| = |f (an )| < < ε.
b
Hence f is continuous at any irrational. At this point we can conclude that f is Riemann integrable, because
even if the functions are discontinuous on the rationals, the rationals have Lebesgue measure zero since
they are denumerable, and hence f would be discontinuous only on a set of measure zero. As it turns out,
however, the function is discontinuous on the rationals.

9
 Figure 2: Modified Dirichlet discontinuous only on rationals.

There is another way, however, to integrate the (unmodified) Dirichlet function, but we must move past
the Riemann integral. Without further adieu, we turn our attention to the Lebesgue integral for a (petty)
introduction to this beast of a machine.

6 Lebesgue Integral
The Riemann integral partitions the domain, whereas the Lebesgue integral essentially partitions the range
of a function. To picture this, one can think of rectangles side-by-side for the Riemann integral, and stacked
on top of each other for the Lebesgue integral (Figure 3).2

To get started, there are a few important concepts that must be understood. The Lebesgue integral uses
characteristic functions, step functions, and sequences along with tools from measure theory to understand
the area under the curve. This theory is very useful in probability theory, as it deals with measure and looks
at data that may or may not be continuous. We start out with some definitions and will end this short look
at the Lebesgue with another look at the Dirichlet function.

Characteristic function: 
1 :x∈A
χA (x) =
0 :x∈
/A

Lebesgue integral of the characteristic function:

Z
χA = µ(A)

Simple function: A simple function f is a function whose range is denumerable.

Notice if f is a simple function, we can rewrite f has a linear combination of characteristic functions.
To illustrate, suppose 
1 : x ∈ [0, 1)
f (x) =
2 : x ∈ [1, 2]
Then we can write f = 1 · χ[0,1) + 2 · χ[1,2] , and the Lebesgue integral would just be
Z Z
f = (1 · χ[0,1) + 2 · χ[1,2] ) = 1 · µ([0, 1)) + 2 · µ([1, 2]) = 1 · 1 + 2 · 1 = 3

1 Reference: math.feld.cvut.cz/mt/txtb/4/txe3ba4s.htm
2 Reference: wikipedia.org/wiki/LebesgueIntegration

10
 Figure 3: Riemann (top) vs. Lebesgue (bottom).

With this integral, the dirac-delta function is no problem to integrate. As we have shown, since the set of
rationals is denumerable, it’s measure is zero. The measure of the irrationals on the interval [0, 1] is just 1,
so when we integrate:
Z Z
f = (0 · χQ + 1 · χR−Q ) = 0 · µ(χQ ) + 1 · µ(χR−Q ) = 0 · 0 + 1 · 1 = 1

So again, just to reiterate, Lebesgue looks at the range values of a function and its inverse image, i.e. the
values in the domain that map to the range values, and multiplies the measure of the inverse image and
the value in the range to get the area of the rectangle. As functions become more complicated, we turn to
sequences of functions and look at limits of step functions to approximate the areas.

To conclude this section, there is a very profound theorem that I will not prove: Every function that is
Riemann integrable is Lebesgue integrable. Obviously, the converse is not true, as this can be seen by
the Dirichlet function. Therefore we can say that the Lebesgue is much stronger as it covers a much wider
class of functions. Does this mean every function is Lebesgue integrable? The answer is no (One example
if f (x) = x1 ). However, the only requirement really needed for a function to be Riemann integrable is that
there exists a sequence of step functions that converge to the given function.

7 Conclusion
The Riemann integral is a widely known and used integral in undergraduate mathematics. The concept
is pretty straightforward and is generally nice to work with (especially when the function is uniformly
continuous. This integral is certainly strong, as it is used for the fundemental theorem of calculus and can be
used (according to Dr. Skoumbourdis) in studying theoretical physics, among many other areas. However,
as can be seen by the Dirichlet function, there are some functions whose area seems easy to guess but is
actually impossible to integrate with Riemann’s partial sums. Lebesgue’s criteria for Riemann integrability
is essentially where the Riemann integral ends and more advanced techniques must be sought out. One
of those techniques is the Lebesgue integral, which partitions the range instead of the domain and uses
measure theory to approximate the area. If I continued studying this topic, I would like to gain a much
better understanding of the Lebesgue integral and the idea of a convergent sequence of step functions. Also,
measure theory has been very fascinating, and although I only looks at sets of measure zero for this project,
I hope to be able to explore more of this topic in graduate school.

11
8 References
1. Bartle, R. G. & Sherbert, D. R. (2011). Introduction to Real Analysis. Hoboken, New Jersey: Wiley.
2. Ray, S. R. (2005). Analysis: With an Introduction to Proof. Upper Saddle River, New Jersey: Pearson
Education.
3. Dr. Evangelos Skoumbourdis

12