Supplementary Lecture Notes on Integration

Math 414
Spring 2004

Integrable Functions with Many Discontinuities

We give several examples of integrable functions with many discontinuities.
We showed for homework that every nondecreasing function f : [a, b]
R is integrable. It is not hard to construct a nondecreasing function with
countably many discontinuities. For example, let f : [0, 1] R be the
function

1
if x = 1
f (x) =
1 .
1 n1 if 1 n1 x < 1 n+1
ppp
s c
s c
s c
s
s
s

Then f is nondecreasing and hence integrable. Clearly f is discontinuous

at 1 n1 for all n N.
Next consider the Thomae function t : [0, 1] R

1 if x = 0
t(x) = 0 x 6 Q
.
1 if x = m Q where m 6= 0 and n are relatively prime
n
n

We have shown that t is continuous at every irrational number but discontinuous at every rational number.
R1
Proposition 1 The Thomae function is integrable and 0 t = 0.

Proof The irrational numbers are dense. Thus for any partition P =
{x0 . . . , xn } there is always an irrational in every interval [xi1 , xi ]. Thus
L(t, P ) = 0. To prove that t is integrable it is enough to show that for every
> 0 there is a partition P with U (t, P ) < .
Let An = {x : t(x) n1 }. If x An , then x = i/j where i, j n. In
particular An is finite.
Suppose > 0. Choose n such that n1 < 2 . We will choose a partition P
such that each point of An is in an interval [xi1 , xi ] where
xi = xi xi1 <

.
2|An |

Let B = {i : An [xi1 , xi ] = }. Note that |B| |An |. If i Bi then

Mi < n1 < 2 , while if i 6 Bi , then Mi = 1. Thus
X
X
U (t, P ) =
Mi xi +
Mi xi
iB

i6B

X
X
xi +
xi
<
2
iB
i6B

<
+ |An |
2
2|An |
< .

Since U (t, P ) L(t, P ) < , t is integrable. Since

Z 1
0 = L(t, P )
t U (t, P ) <
0

for all ,

R1
0

t = 0.

The Thomae function has countably many discontinuities. We next give

an example of an integrable function with uncountably many discontinuities.
Let C be the Cantor set and let f : [0, 1] R. Be the function

1 if x C
f (x) =
.
0 if x 6 C
R1
Proposition 2 f is integrable and 0 f = 0.

Recall the construction of the Cantor set from Quick Tour of the Topology
of R 11 or section 3.1 of Abbott. We build a sequence of closed sets C0
C1 C2 . . . as follows:
C0 = [0, 1],

we throw out the middle third and are left with

1
2
C1 = [0, ] [ , 1],
3
3
we throw out the middle third of each interval and get
1
2 1
2 7
8
C2 = [0, ] [ , ] [ , ] [ , 1] . . .
9
9 3
3 9
9
In general Cn is a union of
2n closed intervals of length 31n .
S
The Cantor set C = n=0 Cn is an uncountable, closed, nowhere dense
set.
Let fn : [0, 1] R be the function

1 if x Cn
fn (x) =
.
0 if x 6 Cn
Note that f (x) fn (x) for all x [0, 1]. The function fn is easy to integrate.
fn (x) = 1 on 2n intervals of length 31n and is 0 everywhere else. Thus
n
Z 1
2
.
fn =
3
0

Suppose > 0. Choose n such that (2/3)n < 2 and choose a partition P
such that U (fn , P ) L(fn , P ) < 2 . Then U (fn , P ) < . But f (x) fn (x)
for all x [0, 1]. Thus
U (f, P ) U (fn , P ) < .
3

Clearly L(f, P ) 0 (indeed since C is nowhere dense L(f, P ) = 0). Thus

U (f, P ) L(f, P ) < . Hence f is integrable.
R 1 Since for any > 0 there is a partition P with U (f, P ) < we must have
f = 0.
0
Exercise 3 Show that f is continuous at x if and only if x 6 C. Thus f
has uncountably many discontinuities.

Approximating Integrable Functions

Lemma 4 Suppose f : [a, b] R is bounded, P is a partition of [a, b] and
> 0. There is a continuous function h : [a, b] R such that f (x) h(x)
for all x [a, b] and
Z b
h U (f, P ) < .
a

Proof We begin by finding a step function b

h such that f (x) b
h(x) for all
Rb
b
x [a, b] and a h(x) = U (f, P ). Suppose P = {x0 , . . . , xn }.
Let Mi = sup{f (x) : x [xi1 , xi ]} and let

Mi if xi1 < x < xi

b
h(x) =
.
Mn if x = b

Then f (x) b
h(x) for all x [a, b] and
Z b
n
X
b
h=
Mi xi = U (f, P ).
a

i=1

Rb
We next find a continuous function h b
h with a (h b
h) < . We
b
do this by modifying h near the points x1 , . . . , xn1 where there may be a
discontinuity.
The idea of the proof is easy but the notation can get messy. Rather than
giving a detailed proof we give an example.
Suppose P = {a, x1 , x2 , b} and
(c a x < x
1
1
b
h(x) = c2 x1 x < x2
c 3 x2 x b

where c1 < c2 and c2 > c3 .

c2

c3

c
s

c1 s

x1

x2

We want a continuous function h b

h that looks like this:
c2
c3

c1

u x1

x2v
b
Rb
We will choose u and v so that a (h b
h) < . We do this by making sure
both of the triangles have area less than 2 .

and v x2 < c2 c
. Then each triangle has area
Suppose x1 u < c2 c
1
3
R
b

less than 2 and a (h b

h) < .
In this case

c1
axu

c2 c1

c1 + x1 u (x u) u < x < x1
x1 x x2 .
h(x) = c2

c3 c2

c2 + vx2 (v x) x2 < x v
c3
v<xb
Rb
It is clear that h(x) b
h(x) for all x [a, b] and a (h b
h) < . Thus
Rb
h(x) f (x) for all x [a, b] and a h U (f, P ) < .
5

Similarly we can find a continuous g f such that L(f, P )

Rb
a

g < .

Corollary 5 If f : [a, b] R is Riemannintegrable, then for any > 0

there are continuous functions g, h : [a, b] R such that g(x) f (x) h(x)
Rb
for all x [a, b] and a (h g) < .

Proof Since f is integrable we can find a partition P such that

U (f, P ) L(f, P ) < .

3

By the Lemma we can find continuous functions g and h such that g(x)
f (x) h(x) for all x [a, b],

Then

Rb
a

b
a

h U (f, P ) < and L(f, P )

3

b
a

g< .
3

h g < .

Another Characteriztation of Integrable Functions

We prove Theorem 8.1.2 from Abbotts Understanding Analysis
Theorem 6 A bounded function f : [a, b] R is Riemann-integrable with
Z

f =A
a

if and only if for every > 0, there is a > 0 such that

|R(f, P ) A| <
for any -fine tagged partition P .
Proof
() We first prove this in case f is continuous. We begin as in the proof
that continuous functions are integrable.
Let > 0. By Uniform Continuity there is > 0 such that if x, y [a, b]
and |x y| < , then

.
|f (x) f (y)| <
ba
6

Choose a partition P = {x0 , . . . , xn } and tags z1 , . . . , zn such that

xi = xi xi1 <
for i = 1, . . . , n. Since f is continuous there are u, v [xi1 , xi ] with f (u) =

. Thus
mi and f (v) = Mi . By choice of , |Mi mi | < ba
n
X

X
(Mi mi )xi <
xi = .
U (f, P ) L(f, P ) =
b

a
i=1
i=1
But
L(f, P ) R(f, P ) U (f, P )
and
L(f, P )
Thus

f U (f, P ).
a

Z b

R(f, P )
< .
f

We now consider the general case where f may not be continuous. Let
> 0. We know that there are continuous functions g, h : [a, b] R such
that g(x) f (x) h(x) for all x [a, b] and
Z b
Z b

h
g < .

2
a
a

By the argument above we can find > 0 such that for any tagged -fine
partition P

Z b
Z b

R(h, P )

< .
h
<
and
R(g,
P
)

2
a
a

Since

R(g, P ) R(f, P ) R(h, P )

and

R(g, P )

g
a

f
a

h R(f, P )
a

h,
a

f R(h, P )
a

g
a

and

Z b
Z b
Z b

R(f, P )
max R(g, P )
, R(h, P )
.
f
h
g

But

and

Z b
Z b Z b
Z b

R(g, P )

+ =
h

R(g,
P
)

g
+
h

2 2
a
a
a
a

Z b
Z b Z b
Z b

R(h, P )
g + = .
h
h +
g R(h, P )

2 2
a
a
a
a

Thus

Z b

R(g, P )
h

as desired.

() We first need one claim.

Claim For any bounded function f : [a, b] R, any partition P and any
> 0 we can find a tagging such that
U (f, P ) R(f, P ) < .
Note: if f is continous we can do better, we could choose zi [xi1 , xi ]
such that zi = Mi and then R(f, P ) = U (f, P ). If f is not continuous, it is
possible that there is no zi [xi1 , xi ] with f (zi ) = Mi . Even if this is not
possible, we can find zi [xi1 , xi ] with f (zi ) as close as wed like to Mi .
For i = 1, . . . , n choose zi [xi1 , xi ] with

Mi f (zi ) <
.
n(xi xi1 )
Then
U (f, P ) R(f, P ) =
<

n
X

i=1
n
X
i=1

(Mi f (zi ))xi

xi
n(xi xi1 )

n
X

=
= .
n
i=1

as desired. This proves the claim.

R b We can now finish the proof. We want to show that f is integrable and
f = A.
a
Let > 0. There is > 0 such that
|R(f, P ) A| <

for any tagged -fine partition P .

By the claim we can find a tagging P1 of P such that

U (f, P ) R(f, P1 ) < .

4
Similarly, we can find a tagging P2 of P such that

R(f, P2 ) L(f, P ) < .

4
Note that
|R(f, P1 ) R(f, P2 )| |R(f, P1 ) A| + |R(f, R2 ) A| <

and
U (f, P )L(f, P ) U (f, P )R(f, P1 )+R(f, P2 )L(f, P )+|R(f, P1 )R(f, P2 )| < .
Thus f is integrable.
These arguments also show that for any > 0 we can find a partiion P
with
|U (f, P ) A| < .
Rb
Thus a f = A.

Lebesgues Characterization of Riemmann Integrable

Functions
Measure Zero Sets
Read
Abbott 7.6 pg 203207
Chapter 2 1 & 2 of Quick Tour of the Topology of R on measure zero
sets.

Sets of Discontinuity
In Math 413 we proved that for any f : R R the set of points where f is
discontinuous is an F -set. Lets review the key steps of that proof.
Definition 7 For f : [a, b] R let
D = {x [a, b] : f is discontinuous at x}.
If > 0, let
D = {x : for all > 0 there are y, z (x , x + ) with |f (y) f (z)| }.

Lemma 8 D =

D1 .
n

n=1

Proof Suppose x D. Then there is > 0 such that for any > 0 there
is a y such that |x y| < and |f (x) f (y)| . If n1 < , then x D 1 .
n
S
1.
Thus D
n=1 D n
On the other hand, suppose x D 1 . For all n choose
m

1
1
,x+ )
n
n
1
such that |f (yn ) f (zn )| m . Note that lim yn = lim zn = x. If f were
continuous at x, then lim f (yn ) = lim f (zn ) = x and there would be an N
such that |f (yn ) f (xn )| < m1 for all n > N , a contradiction. Thus each
D 1 D and
m

[
D=
D1.
yn , zn (x

n=1

10

Lemma 9 Each D is closed.

Proof Suppose x 6 D . Then there is > 0 such that |f (y) f (z)| < for
all y, z (x , x + ). It is easy to see that if y (x , x + ) then y 6 D .
Thus Dc is open and D is closed.
Corollary 10 D is an F -set.
Theorem 11 (Lebesgues Theorem) Suppose f : [a, b] R is bounded.
Then f is Riemann-integrable if and only if D has measure zero.
Proof
() Suppose f is integrable. To prove that D has measure zero it suffices
to prove that each D 1 has measure zero.
m

.
Let P = {x0 , . . . , xn } be a partition such that U (f, P ) L(f, P ) < 2m

Let Oi be the interval (xi 4(n+2) , xi + 4(n+2) ). Then |Oi | = 2(n+1) .

Let B = {i : (xi1 , xi ) D 1 6= }. If i B there are x, y (xi1 , xi )
m
with |f (x) f (y)| m1 . Thus Mi mi m1 . Thus
X 1

xi U (f, P ) L(f, P ) <

m
2m
iB

P
and iB xi < 2 . If x D 1 , then either x (xi1 , xi ) for some i B, or
m
x = xi for some i = 0, . . . , n. Thus
[
D 1 O0 . . . On
(xi1 , xi )
m

iB

and
|O0 | + + |On | +

xi < (n + 1)

iB

+ = .
2(n + 1) 2

Thus D 1 has measure zero and D has measure zero.

m

() Suppose D has measure zero. Suppose |f (x)| < M for all x [a, b].
Let > 0. Choose n such that ba
< 2 . The set D 1 D has measure
n
n
zero. Thus we can find a countable collection of open intervals O1 , O2 , . . .
such that

[
X

D1
Oj and
|Oj | <
.
n
4M
j=1
j=1
11

Since D 1 is a closed subset of [a, b] it is compact. Thus, by the HeineBorel

n
Theorem, we can find j1 , . . . , jm such that
D 1 Oj1 . . . Ojm .
n

Let O = Oj1 . . . Ojm . If x [a, b] \ O, then x 6 D 1 . Thus there is an open

n
interval Ix with x Ix such that |f (y) f (z)| < n1 for all y, z Ix . Note
that
[
[a, b] Oj1 . . .
Ix .
x6O

By one more application of compactness, we can find x1 , . . . , xk such that

[a, b] Oj1 . . . Ojm Ix1 . . . Ixk .

We choose a partition P that contains all of the endpoints of Oj1 , . . . , Ojm , Ix1 , . . . , Ixk .
Each interval [xi1 , xi ] is either contained in O or |f (z) f (y)| < n1 for all
y, z [[xi1 , xi ]. In the later case Mi mi n1 . Let B = {i : [xi1 , xi ] O}.
Then
X
X
U (f, P ) L(f, P ) =
(Mi mi )xi +
(Mi mi )xi
iB

i6B

X1

2M xi +
xi
n
iB
i6B

1
2M (|Oj1 + . . . + Ojm |) + (b a)
n

< 2M
+
< .
4M
2

Thus f is integrable.

12