Lecture Notes on Measure Theory and Integration

4 — Integration

Joel Feldman

University of British Columbia

November 5, 2019

1
4 Integration
4.1 Definitions and Basic Properties
Finally we are ready for

Definition 4.1 (Integral). Let (X, M, µ) be a measure space and E ∈ M.


(a) L+ (X, M) = f : X → [0, ∞] f is M-measurable . If f ∈ L+ (X, M), then
Z n P n
P
f (x) dµ(x) = sup ai µ(Ei ∩ E) ai χEi (x) ≤ f (x) for all x ∈ X
E 1≤i≤n i=1 o
ai 6=0
0 ≤ ai < ∞, Ei ∈ M ∀ 1 ≤ i ≤ n ∈ N
∈ [0, ∞]
 R
(b) L1 (E, X, M, µ) = f : X → R f is M-measurable and E |f (x)| dµ(x) < ∞
(c) If f ∈ L1 (E, X, M, µ), then
Z Z Z
f (x) dµ(x) = max{f (x), 0} dµ(x) − max{−f (x), 0} dµ(x)
E E E

Remark 4.2.
(a) In part (a) of Definition 4.1, think of


−1
ai = F + i−1
N
E i = f F+ i−1
N
,F + i
N
F = inf f (x)
x∈E

with N big.
(a) In part (a) of Definition 4.1, the purpose of the condition ai 6= 0 is to make clear
that if f is identically zero on part, E ′ , of the domain of integration, then E ′
contributes zero to the integral, even if µ(E ′ ) = ∞.

(c) In part (b), |f (x)| = max f (x), −f (x) , max{f (x), 0} and max{−f (x), 0} are
all automatically measurable, because f is measurable.
R
(d) The reason for separating the definition of f (x) dµ(x) for f ≥ 0 from the
Rdefinition for f of general sign Ris that ∞ − ∞ doesn’t make sense. That
E
max{f (x), 0} dµ(x) < ∞ and E max{−f (x), 0} dµ(x) < ∞ in part (c) follow
from Theorem 4.5 below.

Integration 2 November 5, 2019

Theorem 4.3 (Integration Toolbox — Algebra). Let (X, M, µ) be a measure space
and E ∈ M. Let either f, g ∈ L+ (X, M) or f, g ∈ L1 (E, X, M, µ). Then
(a) If ai ≥ 0 and Ei ∈ M for each 1 ≤ i ≤ n, then
Z h n i Xn
P
ai χEi (x) dµ(x) = ai µ(Ei ∩ E)
E i=1 i=1

The same formula applies if ai ∈ R and µ(Ei ∩ E) < ∞ for each 1 ≤ i ≤ n.

(b) In the case f, g ∈ L1 , we have f + g ∈ L1 . In both cases
Z Z Z
(f + g) dµ = f dµ + g dµ
E E E

(c) If f ∈ L , let λ ∈ R. If f ∈ L , let λ > 0. In the case f ∈ L1 , we have λf ∈ L1 .

1 +

In both cases Z Z
(λf ) dµ = λ f dµ
E E
R R
(d) We have X f χE dµ = E f dµ and, if f (x) = g(x) for all x ∈ E, then
Z Z
f dµ = g dµ
E E

Proof.
(a) Case ai ≥ 0 and Ei ∈ M for each 1 ≤ i ≤ n:
Recall that
Z h n i n P
m o
P P P
ai χEi (x) dµ(x) = sup bj µ(Fj ∩ E) bj χFj ≤ ai χEi
E i=1 j=1 j i

R h Pn i P
The inequality E i=1 ai χE i
(x) dµ(x) ≥ ni=1 ai µ(Ei ∩ E) is trivial — choose
bi = ai , Fi = Ei .
So it suffices to prove that
m
P n
P m
P n
P
bj χFj ≤ ai χEi =⇒ bj µ(Fj ∩ E) ≤ ai µ(Ei ∩ E)
j=1 i=1 j=1 i=1

Given any finite collection of sets G1 , · · · , Gp (think E1 , · · · , En , F1 , · · · , Fm ) we

can always find (by induction on p) a finite collection of pairwise disjoint sets H1 ,
· · · , Hq such that each [
Gk = Hj
1≤j≤q
Hj ⊂Gk

Integration 3 November 5, 2019

 Furthermore

p
P P q
P 
ck χGk = ck χHj = dj χHj 


k=1 j,k j=1
 X
Hj ⊂Gk
with dj = ck
p
P q
P 
 1≤k≤p
ck µ(E ∩ Gk ) = dj µ(E ∩ Hj ) 

 Hj ⊂Gk
k=1 j=1

So we can always assume that Ei = Fi , 1 ≤ i ≤ n = m and that the Ei ’s are

disjoint and non empty. So
n
P n
P x ∈ Ej
bi χEi (x) ≤ ai χEi (x) =⇒ bj ≤ aj for all 1 ≤ j ≤ n
i=1 i=1
n
X n
X
=⇒ bj µ(E ∩ Ej ) ≤ aj µ(E ∩ Ej )
j=1 j=1

(a) Case ai ∈ R and µ(Ei ∩ E) < ∞ for each 1 ≤ i ≤ n:

We may again assume that the Ei ’s are disjoint. Then
n
X
f (x) = ai χEi (x)
i=1
Z Z h n i n
X
P
=⇒ |f (x)| dµ(x) = |ai | χEi (x) dµ(x) = |ai | µ(Ei ∩ E) < ∞
E E i=1 i=1
=⇒ f ∈ L1
Z Z Z
=⇒ f (x) dµ(x) = max{f (x), 0} dµ(x) − max{−f (x), 0} dµ(x)
E E
Z h E Z h
P i P i
= ai χEi (x) dµ(x) − −ai χEi (x) dµ(x)
E 1≤i≤n E 1≤i≤n
ai >0 ai <0
X X
= ai µ(Ei ∩ E) − (−ai ) µ(Ei ∩ E)
1≤i≤n 1≤i≤n
ai >0 ai <0
n
X
= ai µ(Ei ∩ E)
i=1

(b) will be proven later (following the proof of Theorem 4.10).

Integration 4 November 5, 2019

(c) Case λ > 0 and f ≥ 0:
Z n P o
P
(λf ) dµ = sup ai µ(Ei ∩ E) ai χEi ≤ λf
E
ni P i
P o
ai =λãi
= sup λãi µ(Ei ∩ E) λãi χEi ≤ λf
i i
n P P o
= λ sup ãi µ(Ei ∩ E) ãi χEi ≤ f
i i
Z
= λ f dµ
E

(c) For general signs: The case λ = 0 is trivial. So assume that λ 6= 0. Since
|λf (x)| = |λ| |f (x)|, we have, by the previous case,
Z Z
1
f ∈ L =⇒ |λf | dµ = |λ| |f | dµ < ∞ =⇒ λf ∈ L1
E E

and then
Z Z Z
λf (x) dµ(x) = max{λf (x), 0} dµ(x) − max{−λf (x), 0} dµ(x)
E ZE E Z
= max{|λ| sgnλ f (x), 0} dµ(x) − max{−|λ| sgnλ f (x), 0} dµ(x)
EZ EZ

= |λ| max{sgnλ f (x), 0} dµ(x) − |λ| max{−sgnλ f (x), 0} dµ(x)

Z E Z E

= λ max{f (x), 0} dµ(x) − λ max{−f (x), 0} dµ(x)

E E
(just look at the cases λ > 0 and λ < 0 separately)
Z
=λ f (x) dµ(x)
E
R R
(d) It suffices to prove that E f dµ = X f χE dµ. The case f ∈ L1 (E, X, M, µ)
follows from the case f ∈ L+ (X, M), so assume that f ∈ L+ (X, M). We first show
that the two sets of numbers
n P P n o
S1 = ai µ(Ei ∩ E) ai χEi (x) ≤ f (x) ∀x ∈ X, 0 ≤ ai < ∞, Ei ∈ M
1≤i≤n i=1
ai 6=0

and
n P n
P o
S2 = ai µ(Ei ) ai χEi (x) ≤ f (x)χE (x) ∀x ∈ X, 0 ≤ ai < ∞, Ei ∈ M
1≤i≤n i=1
ai 6=0

Integration 5 November 5, 2019

 P
are identical. The condition ni=1 ai χEi (x) ≤ f (x)χE (x) for all x ∈ X forces that,
for each 1 ≤ i ≤ n, either Ei ⊂ E or ai = 0. Consequently
n P n
P o
S2 = ai µ(Ei ∩E) ai χEi (x) ≤ f (x)χE (x) ∀x ∈ X, 0 ≤ ai < ∞, Ei ∈ M
1≤i≤n i=1
ai 6=0

P
That S2 ⊂ S1 is now trivial. To see that S1 ⊂ S2 pick any number 1≤i≤n ai µ(Ei ∩E)
ai 6=0
n
P
in S1 . Then ai χEi (x) ≤ f (x) for all x ∈ X, and, setting Ẽi = Ei ∩ E, we have
i=1
n
P
ai χẼi (x) ≤ f (x)χE (x) for all x ∈ X and
i=1

P P
ai µ(Ei ∩ E) = ai µ(Ẽi ∩ E) ∈ S2
1≤i≤n 1≤i≤n
ai 6=0 ai 6=0

This completes the proof that S1 = S2 . As S1 = S2 ,

Z Z
f dµ = sup S1 = sup S2 = f χE dµ
E X

Warning 4.4. We still haven’t proved part (b) of Theorem 4.3. We have to be careful
not to use it before we have proved it.

Theorem 4.5 (Integration Toolbox — Bounds). Let (X, M, µ) be a measure space

and E ∈ M. Let f, g, h : X → R be measurable with f, g ∈ L1 (E, X, M, µ). Then
R R
(a) If |h| ≤ f on E, then h ∈ L1 (E, X, M, µ) and E |h| dµ ≤ E f dµ.
R R
(b) E f dµ ≤ E |f | dµ
R R
(c) If f ≤ g on E, then E f dµ ≤ E g dµ
(d) If h is bounded and µ(E) < ∞, then h ∈ L1 (E, X, M, µ) and
Z
h dµ ≤ µ(E) sup |h(x)|
E x∈E

Proof.

Integration 6 November 5, 2019

(a) If |h| ≤ f on E and f ∈ L1 (E, X, M, µ), then
Z Z
|h| dµ = |h|χE dµ
E X
nR o
= sup X
ϕ dµ 0 ≤ simple ϕ ≤ |h|χE
nR o
≤ sup X
ϕ dµ 0 ≤ simple ϕ ≤ f χE
Z Z
= f χE dµ = f dµ < ∞
X E

(b) By part (a) with h = max{±f, 0} and f replaced by |f | = max{−f, 0} +

max{−f, 0},
Z Z Z
f dµ = max{f (x), 0} dµ(x) − max{−f (x), 0} dµ(x)
E E E
Z Z 
≤ max max{f (x), 0} dµ(x) , max{−f (x), 0} dµ(x)
E E
|f (x)|
Z z }| {

≤ max{f (x), 0} + max{−f (x), 0} dµ(x) (by part (a))
E

(c) Once we have proved part (b) of Theorem 4.3, we will be able to argue that
Z Z Z
f ≤ g =⇒ g − f ≥ 0 =⇒ (g − f ) dµ ≥ 0 =⇒ g dµ − f dµ ≥ 0
E E E
Z Z
=⇒ f dµ ≤ g dµ
E E

(d) Set f (x) = supx∈E |h(x)| χE (x). By part (a) of Theorem 4.3,
Z
f (x) dµ(x) = µ(E) sup |h(x)| < ∞
E x∈E

Then, by part (a), h ∈ L1 , and

Z Z
|h| dµ ≤ f (x) dµ = µ(E) sup |h(x)|
E E x∈E

and finally, by part (b),

Z Z
h dµ ≤ |h| dµ ≤ µ(E) sup |h(x)|
E E x∈E

Integration 7 November 5, 2019

Warning 4.6. R We still haven’t R proved Rpart (b) of Theorem 4.3. So we can’t yet
use it (i.e. R E (f + g) Rdµ = E f dµ + E g dµ), or part (c) of Theorem 4.5 (i.e.
f ≤ g =⇒ E f dµ ≤ E g dµ). On the other hand, we can use that
Z Z
0 ≤ f ≤ g =⇒ f dµ ≤ g dµ
E E

because
Z ≤g
nR z}|{ o
f dµ = sup E
ϕ dµ simple ϕ ≤ f
E
nR o
≤ sup E
ϕ dµ simple ϕ ≤ g
Z
= g dµ
E

4.2 Limit Theorems

Theorem 4.7 (Fatou’s1 Lemma). Let (X, M, µ) be a measure space and E ∈ M.
Let for each n ∈ N, fn : X → [0, ∞] be measurable. Then
Z Z
lim inf fn (x) dµ(x) ≤ lim inf fn (x) dµ(x)
E n→∞ n→∞ E

Remark 4.8.
(a) In Problem Set 6, you defined R = R ∪ {−∞, ∞}

BR = E ⊂ R E ∩ R ∈ BR
and proved that BR is
a σ-algebra and that
f : X → R is (M, BR )-measureable if
and only if f −1 {∞} ∈ M, f −1 {−∞} ∈ M and f is measurable on f −1 (R).
For g : X → [0, ∞],


g is measurable ⇐⇒ g −1 (a, ∞] ∈ M for all a ∈ R
T
(because the (a, ∞]’s generate all (a, b] = (a, ∞]∩(b, ∞]c ’s and {∞} = (n, ∞]
 S c n∈N

and {−∞} = (−n, ∞] ) and

n∈N
Z n n
P P
g(x) dµ(x) = sup ai µ(Ei ∩ E) ai χEi ≤ g, n ∈ N,
E 1≤i≤n i=1 o
ai 6=0
0 ≤ ai < ∞, Ei ∈ M ∀ 1 ≤ i ≤ n
1
Pierre Fatou (1878–1929) was a French mathematician and astronomer.

Integration 8 November 5, 2019

 n o
(b) lim inf n→∞ fn (x) = sup inf fm (x) always exists and is measurable by part (f)
n m≥n
of Theorem 3.6.
(c) It is possible to have strict inequality in Fatou’s Lemma. For example, if
(
1 if |x| > n
fn (x) =
0 if |x| ≤ n
then
Z
lim fn (x) = 0 but lim fn (x) dx = ∞
n→∞ n→∞ R

Proof of Fatou’s Lemma, Theorem 4.7. We must show that

Xm Z Z
ϕ(x) = ai χEi (x) ≤ lim inf fn (x) =⇒ ϕ dµ ≤ lim inf fn dµ
n→∞ E n→∞ E
i=1
R
Case E ϕ dµ = ∞:
In this case there is an 1 ≤ i ≤ m such that ai > 0 and µ(E ∩ Ei ) = ∞. For each
x ∈ Ei ∩ E,
lim inf fk (x) ≥ ai =⇒ ∀ ε > 0 ∃ n such that k ≥ n =⇒ fk (x) ≥ ai − ε
k→∞
=⇒ ∃ n such that k ≥ n =⇒ fk (x) ≥ 21 ai (∗)
This need not be uniform in x. That is, the n in (∗) may depend on x.
Define
 \

An = x ∈ E fk (x) ≥ 21 ai for all k ≥ n = fk−1 [ai/2, ∞] ∩ E ∈ M
k≥n

Under that definition,

◦ An ⊂ An+1 , for all n ∈ N
S
◦ Ei ∩ E ⊂ n∈N An by (∗)
R R
◦ E fn dµ ≥ E a2i χAn dµ = ai
2
µ(An ), for all n ∈ N
Consequently
Z
lim inf fn dµ ≥ lim inf a2i µ(An )
n→∞ E n→∞
S 
ai
= 2 µ n∈N An (by continuity from below)
ai
≥ 2
µ(E ∩ Ei ) = ∞

Integration 9 November 5, 2019

 R
Case E ϕ dµ < ∞:
Let ε > 0, c = 1 − ε and set

A = x ∈ E ϕ(x) > 0

An = x ∈ A fk (x) ≥ c ϕ(x) for all k ≥ n
Then
[
An ⊂ An+1 A= An µ(A) < ∞
n∈N

Now µ(A) = µ(An ) + µ(A\ An ) and, by continuity from below, lim µ(An ) = µ(A),
n→∞
so that
there exists an N ∈ N such that µ(A \ An ) < ε for all n ≥ N
and, consequently,
Z Z
fn dµ ≥ fn dµ
E An
Z
≥ c ϕ dµ
An
<ε for n≥N
Z m z }|
X
{
= c ϕ dµ − c ai µ Ei ∩ (A \ An ) (by Theorem 4.3(a))
A i=1
Z Z m
X
≥ ϕ dµ − ε ϕ dµ − ε ai for all n ≥ N
A A i=1
Z Z Z m
X 
=⇒ lim inf fn dµ ≥ ϕ dµ − ε ϕ dµ + ai for all ε > 0
n→∞ E A A i=1
Z Z Z
=⇒ lim inf fn dµ ≥ ϕ dµ = ϕ dµ
n→∞ E A E

As usual, we can put in “almost everywhere” almost everywhere.

Theorem 4.9 (Another Fatou’s Lemma). Let (X, M, µ) be a measure space and
E ∈ M. Let, for each n ∈ N, fn ∈ L+ (X, M), and let f : X → [0, ∞] obey
f (x) = lim inf fn (x) a.e. on E
n→∞

Integration 10 November 5, 2019

Assume that f is measurable2 . Then
Z Z
f (x) dµ(x) ≤ lim inf fn (x) dµ(x)
E n→∞ E

Proof. Let Ω be a set of measure zero such that

f (x) = lim inf fn (x) for all x ∈ E \ Ω

n→∞

Define

g(x) = f (x) χE\Ω (x)

gn (x) = fn (x) χE\Ω (x) for each n ∈ N

By Theorem 4.3(d) and Problem Set 7, #1(a),

Z Z Z Z
f dµ = g dµ fn dµ = gn dµ for all n ∈ N
E E E E

and we can apply the old Fatou’s Lemma, Theorem 4.7, to g = lim inf n→∞ gn .
THEOREM 4.10 (The Monotone Convergence Theorem). Let (X, M, µ) be a mea-
sure space and E ∈ M. Let, for each n ∈ N, fn ∈ L+ (X, M), and let f : X → [0, ∞]
obey

f (x) = lim fn (x) a.e. on E

n→∞
fn ≤ f a.e. on E for all n ∈ N

Assume that f is measurable2 . Then

Z Z
f (x) dµ(x) = lim fn (x) dµ(x)
E n→∞ E

Remark 4.11. The reason that Theorem 4.10 is called the monotone convergence
theorem is that if fn ∈ L+ (X, M) for all n ∈ N and
)
fn+1 ≥ fn ∀ n ∈ N
i.e. fn ր f
f = lim fn
n→∞

then the hypotheses of Theorem 4.10 are satisfied.

2
This is not needed if (X, M, µ) is complete. See Theorem 3.6, Lemma 3.10 and Example 3.11.

Integration 11 November 5, 2019

 R R
Proof of Theorem 4.10. Since 0 ≤ fn ≤ f a.e. on E, we have E fn dµ ≤ E f dµ.
So, by Fatou’s Lemma, Theorem 4.9,
Z Z Z Z
f dµ ≤ lim inf fn dµ ≤ lim sup fn dµ ≤ f dµ
E n→∞ E n→∞ E E

We can now go back and fill in the

Proof of Theorem 4.3(b). Recall that we are to prove that
Z Z Z
(f + g) dµ = f dµ + g dµ
E E E

Case 1: Positive, simple measurable functions:

Pn
If f and g are both of the
R form i=1 ai χREi , with each
R ai ≥ 0 and each Ei ∈ M,
then the conclusion that E (f + g) dµ = E f dµ + E g dµ follows directly from
Z h n i n
X
P
ai χEi (x) dµ(x) = ai µ(Ei ∩ E)
E i=1 i=1

which is part (a) of Theorem 4.3.

Case 2: f, g ∈ L+ (X, M):


Define Ii,n = 2in , i+1
2n and
4 n
X i
fn (x) = n
χf −1 (Ii,n ) (x) + 2n χf −1 ({∞}) (x)
i=0
2
4 n
X i
gn (x) = n
χg−1 (Ii,n ) (x) + 2n χg−1 ({∞}) (x)
i=0
2

Then fn and gn are measurable and simple and

0 ≤ fn (x) ≤ f (x) 0 ≤ gn (x) ≤ g(x)

and
f (x) = lim fn (x) g(x) = lim gn (x)
n→∞ n→∞

Integration 12 November 5, 2019

 So, by the monotone convergence theorem, Theorem 4.10 (three times),
Z Z
(f + g) dµ = lim (fn + gn ) dµ
E n→∞ E
Z Z 
= lim fn dµ + gn dµ (by Case 1)
n→∞ E E
Z Z
= f dµ + g dµ
E E
1
Case 3: f, g ∈ L , f + g ≥ 0, f ≥ 0:
By Case 2,
Z Z Z


|f | + |g| dµ = |f | dµ + |g| dµ < ∞ =⇒ |f | + |g| ∈ L1
E E E

As |f + g| ≤ |f | + |g| we have that f + g ∈ L1 by part (a) of Theorem 4.5.

Define g+ = max{g, 0} and g− = max{−g, 0}. Then
Z Z Z
g+ dµ + f dµ = (f + g+ ) dµ (by Case 2)
E E E =f +g≥0
Z ≥0
z }| { z}|{
= (f + g+ − g− ) + g− dµ
ZE Z
= (f + g) dµ + g− dµ (by Case 2)
E E
Z Z Z Z Z Z
=⇒ (f + g) dµ = f dµ +
g+ dµ − g− dµ = f dµ + g dµ
E E E E E E
R
by the definition (in Definition 4.1) of E g dµ.
Case 4: f, g ∈ L1 :
We have that f + g ∈ L1 as in Case 3. As
Z ≥0 Z Z z ≥0
z }| {
}| {

|f | + |g| dµ + (f + g) dµ = |f | + |g| + f + g dµ (by Case 3)
E E ≥0 E
Z z }| {
Z
= |f | + |g| + f dµ + g dµ (by Case 3)
E ≥0 E
Z z }| { Z Z


= |f | + |g| dµ + f dµ + g dµ (by Case 3)
E E E
R

Now just subtract E |f | + |g| dµ from both sides.

Integration 13 November 5, 2019

THEOREM 4.12 ((Lebesgue) Dominated Convergence Theorem). Let (X, M, µ)
be a measure space and E ∈ M. Let f, g : X → R be measurable and let, for each
n ∈ N, fn : X → R be measurable. Assume that, g ∈ L1 (E, X, M, µ) and
|fn | ≤ g a.e. on E ∀ n ∈ N and lim fn = f a.e. on E
n→∞

Then f ∈ L1 (E, X, M, µ) and

Z Z
f (x) dµ(x) = lim fn (x) dµ(x)
E n→∞ E
Remark 4.13. The hypothesis, in Theorem 4.12, that the sequence {fn }n∈N is
bounded by an L1 function, g, uniformly in n, is crucial. For example, if
(
1
1/n
if |x| ≤ n/2
fn (x) = n
0 if |x| > n/2 −n n
2 2
then
f (x) = lim fn (x) = 0
n→∞

and yet
Z Z
f (x) dx = 0 6= 1 = fn (x) dx for all n ∈ N
R R
Proof of Theorem 4.12. By Fatou’s lemma,
<∞
Z z≥0}|a.e.{ Z z≥0}|a.e.{ zZ }| { Z
(g + f ) dµ ≤ lim inf (g + fn ) dµ = g dµ + lim inf fn dµ
E n→∞ E E n→∞ E
Z Z
=⇒ f dµ ≤ lim inf fn dµ
E n→∞ E
and
<∞
Z z≥0}|a.e.{ Z z≥0}|a.e.{ zZ }| { Z
(g − f ) dµ ≤ lim inf (g − fn ) dµ = g dµ − lim sup fn dµ
E n→∞ E E n→∞ E
Z Z
=⇒ lim sup fn dµ ≤ f dµ
n→∞ E E
So
Z Z Z Z
f dµ ≤ lim inf fn dµ ≤ lim sup fn dµ ≤ f dµ
E n→∞ E n→∞ E E

Integration 14 November 5, 2019

4.3 The Connection to Riemann Integrals
In this section we see that the Lebesgue integral is an extension of the Riemann
integral. We start with a review of the definition of the Riemann integral.

Definition 4.14 (Riemann Integral). Let a < b be real numbers and f : [a, b] → R
be bounded3 .
(a) The upper Riemann integral of f is
Z b nP o
n

f (x) dx = inf ti − ti−1 sup f (x) n ∈ N, a = t0 < t1 < · · · < tn = b
a i=1 ti−1 ≤x≤ti

(b) The lower Riemann integral of f is

Z b nP
n o


f (x) dx = sup ti − ti−1 inf f (x) n ∈ N, a = t0 < t1 < · · · < tn = b
i=1 ti−1 ≤x≤ti
a

(c) The function f is Riemann integrable if

Z b Z b
f (x) dx = f (x) dx
a a
Z a
We shall denote the common value R f (x) dx.
b

Theorem 4.15 (Riemann integrable =⇒ Lebesgue integrable). Let a < b be real

numbers and let f : [a, b] → R be bounded.
(a) If f is Riemann integrable, then f is Lebesgue measurable (and hence L1 , since
f is bounded and the interval [a, b] is finite) and
Z a Z
R f (x) dx = f (x) dm(x)
b [a,b]

(b) f is Riemann integrable if and only if x ∈ [a, b] f is not continuous at x
has Lebesgue measure zero.

Proof. (a) Here is the strategy of the proof. We will

3
IfP
f is not bounded,

it cannot be Riemann integrable. For example, if f is not bounded above,
n
then i=1 ti − ti−1 supti−1 ≤x≤ti f (x) = ∞ for every partition.

Integration 15 November 5, 2019

• use the definition of the upper Riemann integral to show that
Z b Z
f (x) dx = ḡ(x) dm(x)
a [a,b]

with ḡ ≥ f constructed as a limit of a decreasing sequence of step functions, and

• use the definition of the lower Riemann integral to show that
Z b Z
f (x) dx = g(x) dm(x)
a [a,b]

with ḡ ≤ f constructed as a limit of an increasing sequence of step functions.

R  
• As f is, by hypothesis, Riemann integrable, this will imply that [a,b] ḡ − g dm =
R R
[a,b]
ḡ dm − [a,b] g dm = 0 and hence that ḡ = g = f a.e..
• That, in turn, will give that f is measurable and that
Z a Z Z
R f (x) dx = ḡ(x) dm(x) = f (x) dm(x)
b [a,b] [a,b]

We’ll start by setting up some notation. For a partition

P = {a = t0 < t1 < · · · < tn = b}

set

Mi = sup f (x) mi = inf f (x)

ti−1 ≤x≤ti ti−1 ≤x≤ti

X n n
X
S̄P = Mi (ti − ti−1 ) SP = mi (ti − ti−1 )
i=1 i=1
n
X n
X
ḡP = Mi χ[ti−1 ,ti ) gP = mi χ[ti−1 ,ti )
i=1 i=1

so that
Z Z
S̄P = ḡP dm SP = g P dm (1)
[a,b] [a,b]

and
g P ≤ f ≤ ḡP (2)

Integration 16 November 5, 2019

Note that, if P and P′ are partitions with P ⊂ P′ , then

SP ≤ SP′ ≤ S̄P′ ≤ S̄P (3)

g P ≤ g P′ ≤ ḡP′ ≤ ḡP (4)

Here is a figure that gives the basis for (3) and (4)

y = f (x)

y = f (t′j )
g P′

y = f (ti−1 ) = f (t′j−1 )
gP

t′j−1 t′j t′j+1

ti−1 ti
In this figure
◦ ti−1 ≤ x ≤ ti is one subinterval determined by the partition P and
◦ t′j−1 ≤ x ≤ t′j and t′j ≤ x ≤ t′j+1 are two subintervals determined by the finer
partition P′ and
 
◦ because, in this example, f happens to be increasing on the interval ti−1 , ti ,
we have that mi = m′j = f (ti−1 ) and that m′j+1 = f (t′j ) > mi .


◦ Consquently g P = g P′ on the subinterval t′j−1, t′j but g P < g P′ on the subinterval
′ ′

tj , tj+1 .
Since f is Riemann integrable
Z b Z b
f dx = inf S̄P = sup SP = f dx (5)
a P P a

There is a sequence of partitions Q1 , Q2 , Q3 , · · · with

Z b
lim S̄Qn = f dx (6)
n→∞ a

Integration 17 November 5, 2019

and a sequence of partitions Q′1 , Q′2 , Q′3 , · · · with
Z b
lim SQ′n = f dx (7)
n→∞
a
Setting Pn = Q1 ∪ Q′1 ∪ Q2 ∪ Q′2 ∪ · · · ∪ Qn ∪ Q′n we have P1 ⊂ P2 ⊂ P3 ⊂ · · · and
Z Z b
(1)
lim ḡPn dm = lim S̄Pn = f dx (8)
n→∞ [a,b] n→∞ a

since Z b Z b
(5) (3)
n→∞
f dx ≤ S̄Pn ≤ S̄Qn −−−→ f dx by (6)
a a
Similarly,
Z Z b
(1)
lim g P dm = lim SPn = f dx by (5), (3), (7) (9)
n→∞ [a,b]
n n→∞
a
Now, by (2) and (4), the ḡPn ’s form a decreasing sequence of functions that are
bounded below by f . So they converge pointwise to some ḡ ≥ f . Similarly, by
(2) and (4), the g P ’s form an increasing sequence of functions that are bounded
n
above by f . So they converge pointwise to some g ≤ f . By the Lebesgue dominated
convergence theorem,
Z Z Z b
(8)
ḡ dm = lim ḡPn dm = f dx (10)
[a,b] n→∞ [a,b] a
Z Z Z b
(9)
g dm = lim g P dm = f dx
[a,b] n→∞ [a,b]
n
a
Since f is Riemann integrable
Z
  (5)
ḡ − g dm = 0
[a,b]

As the function ḡ − g is nonnegative (by the limit of (2)), ḡ − g = 0 a.e. and

ḡ ≥ f ≥ g =⇒ ḡ = g = f a.e.
Consequently f is measurable (m is complete) and
Z Z Z b Z a
(10)
f dm = ḡ dm = f dx = R f dx
[a,b] [a,b] a b

(b) is Exercise #23 on page 59 of Folland.

Integration 18 November 5, 2019