Chapter 19

Team 1 Measure theory and integration

Problem 19.2
Let X = [a, b] and (X, L, µ) be the induced Lebesgue measure space. Let ϕ : X → R
be a step function with respect to some partition P of [a, b] as in Definition 15.1.4.
Show that:

1. ϕ is a measure function.
R Rb
2. X
ϕdµ = a
ϕ(x)dx where the former is the Lebesgue integral and the latter is
the Riemann integral.

Solution
1. Let P = {x0 , x1 , ..., xn }. Then the collection of intervals defined by P is {Ij =
(xj−1 , xj ] : j = 1, 2, ..., n}. So, by definition
n
X
ϕP (x) = cj 1Ij (x)
j=1

where cj ∈ R. In the section 2 of chapter 19, J = {(c, d] ⊂ [a, b] : a ≤ c < d ≤ b} ∪ {∅}.

In the preposition 18.7.1 is proof that R(J ) ⊂ L, then J ⊂ L, which implies that
Ij ∈ L for all j. In the example 18.9.6 is shown 1Ij is L-measurable and the sum and
scalar multiplication of measurable functions is measurable by proposition 18.9.9. So,
ϕP is L-measurable, and since the P is arbitrary ϕ is L-measurable.

2. Since ϕ is a step function, lets consider the case that ϕ ≥ 0. Since 1) implies the ϕ is a
measurable function, by Lebesgue integral definition
Z Xn
ϕdµ = I(ϕ) = cj µ(Ij )
[a,b] j=1
Sn
where {Ij } ⊂ L is a collection disjoint pairwise sets that i=1 Ij = X. Even more,
by proposition 18.7.1, µ|R(J ) = m, where m is the content defined (in 18.2.4) for J ,
m((c, d]) = c − d for c < d. In Riemann integral language, exists a Q = {x0 , x1 , ..., xn }
which (xj , xj+1 ] = Ij for all j, then |xj+1 − xj | = m(Ij ). For a tagged partition Qτ ,
since ϕ is a step function for all qj ∈ Ij , ϕ(qj ) = cj . Then,
n
X
I(ϕ) = Rϕ,Qτ = cj |xj − xj−1 |
j=1

1
for a particular partition.
Before going to the ε − δ definition of Riemann integral, let’s think a little bit. Been
pessimistic, what would happen if (xj , xj+1 ] there exists a jump, i.e. the tag has to
options. So in order to absolve this possible change, will need to make this interval as
tiny such that the difference between the Rf,Pτ and I(ϕ) don’t exceed ε. So we give δ
as ε dived by this difference. Been overwhelmed by this cases, we need to define δ as ε
dived by the sum of all possible differences. So...
Pn−1 ε
Now let ε > 0. Let S = i=1 |ci − ci+1 | and defined δ = S
> 0. Let Pτ =
{x′0 , x′1 , ..., x′m } be tagged partition such that ∥Pτ ∥ < δ. Then, we have max{|x′i −
x′i−1 |, i = 1, 2, ..., m} < δ. In particular,

|x′i − x′i−1 | < δ, ∀i = 1, 2, ..., m (19.2.2.1)

Then, the Riemann sum with respect the partition Pτ

m
X
Rf,Pτ = ϕ(pi )|x′i − x′i−1 |
i=1

where pi ∈ (yi−1 , yi ], ∀i. and, since ϕ ≥ 0, ϕ(pi ) ≥ 0. Then,

|Rϕ,Pτ − I(ϕ)| = |Rϕ,Pτ − Rϕ,Qτ |

Note that the refinement H = {y1 , y2 , ..., yd } of P and Q satisfies that ∥H∥ < δ and in
this way
Rϕ,Pτ = Rϕ,Hτ

if ar ∈ (yi , yi+1 ] ⊂ (x′j , x′j+1 ] is chosen as ϕ(ar ) = ϕ(pj ) for all i ≤ d, j ≤ m. And

Rϕ,Qτ = Rϕ,Hτ ′

if br ∈ (yi , yi+1 ] ⊂ (xj , xj+1 ] is chosen as ϕ(br ) = ϕ(qj ) = cj for all i ≤ d, j ≤ n. Then,
d
X d
X
|Rϕ,Pτ − Rϕ,Qτ | = |Rϕ,Hτ − Rϕ,Hτ ′ | = ϕ(ar )|yi − yi−1 | − ϕ(br )|yi − yi−1 |
i=1 i=1

d
X d
X
= [ϕ(ar ) − ϕ(br )]|yi − yi−1 | ≤ |ϕ(ar ) − ϕ(br )||yi − yi−1 |
i=1 i=1

Since ϕ is a step function, if ϕ(ar ) − ϕ(br ) ̸= 0 implies that ϕ(ar ) = cl and ϕ(br ) = cl+1
or vice versa. Then, using 19.2.2.1
X X
|Rϕ,Pτ − Rϕ,Qτ | ≤ |ϕ(ar ) − ϕ(br )||yi − yi−1 | < δ |cl − cl+1 |
ϕ(ar )̸=ϕ(br ) ϕ(ar )̸=ϕ(br )

2
 |cl − cl+1 | ≤ S by definition of S. Finally, remembering δ = Sε ,
P
Note that ϕ(ar )̸=ϕ(br )

|Rϕ,Pτ − I(ϕ)| = |Rϕ,Pτ − Rϕ,Qτ | < ε

Rb R
Then, a
ϕ(x)dx = X
ϕdµ. ■
For the case that ϕ : X → R, by definition
Z Z Z
ϕdµ = +
ϕ dµ − ϕ− dµ (19.2.2.2)
X X X

Since ϕ+ , ϕ− : X → [0, ∞] are simple functions and µ(X) < ∞, we know that
R +
X
ϕ dµ < ∞; so 19.2.2.2 exists. So applying the case of non-negative simple functions,
Z Z b Z b Z b Z b
+ − + −
ϕdµ = ϕ (x)dx − ϕ (x)dx = (ϕ (x) − ϕ (x))dx = ϕ(x)dx
[a,b] a a a a

Problem 19.10
Prove the theorem 19.5.9 namely:
Let (R, L, µ) be the Lebesgue measure space and f ∈ L1 (R). Prove that

(a) For every ϵ > 0 there exists a, b ∈ R with a < b, a partition of P of [a, b], and a
R
step function ϕP : R → R adapted to P such that R |f − ϕP |dµ < ϵ.

(b) For every ϵ > 0 there exists a, b ∈ R and a continuous function g : R → R which
R
vanishes outside of [a, b] such that R |f − g|dµ < ϵ.

Solution

a) Since f ∈ L1 (R), it’s know that

Z Z Z
|f |dµ = f dµ + f − dµ < ∞
+
R R R

By Lebesgue definition of non-negative L-measure functions, there exists simple func-

tions ϕ+ , ϕ− : R → [0, ∞) such that ϕ+ ≤ f + , ϕ− ≤ f − and
Z Z Z Z Z Z
ε − − ε
+
f dµ − ϕ dµ =+ + +
f − ϕ dµ < , f dµ − ϕ dµ = f − − ϕ− dµ <
R R R 4 R R R 4

So we can defined the step function ϕ = ϕ+ − ϕ− which satisfies

Z Z Z
+ + − − ε
f − ϕ dµ + f − ϕ dµ = |f − ϕ|dµ < (19.10.1)
R R R 2

3
 By corollary 19.5.8 there exists a, b ∈ R with a < b, a partition P of [a, b], and a step
function ϕP : R → R adapted to P such that R |ϕ − ϕP | < 2ε . Adding with 19.10.1, we
R

get
Z Z Z Z
|f − ϕP |dµ ≤ |f − ϕ| + |ϕ − ϕP |dµ = |f − ϕ|dµ + |ϕ − ϕP |dµ < ε
R R R R

b) Let ε > 0 for (a) exists a, b ∈ R with a < b, a partition P = {x0 , x1 , ..., xn } of
[a, b] (note that x0 = a and xn = b) and a step function ϕP : R → R such that
|f − ϕP |dµ < 2ε . By definition, ϕP (x) = nj=1 cj 1Ij (x) where Ij = (xj−1 , xj ]. Let
R P
R
0 < δ < m = min{|xj − xj−1 | : j = 1, 2, ..., n}, we can defined a function g : R → R as



 0 if x ∈ (−∞, a)
 h(x) = 2δ ϕP (a + 2δ )(x − a) if x ∈ a, a + 2δ

 



if x ∈ xj−1 + 2δ , xj − 2δ , j = 1, ..., n

 ϕ (x)

P
g(x) =
if x ∈ xj − 2δ , xj + 2δ , j = 1, ..., n − 1
 


 Lj (x)
2 δ δ
 



 k(x) = δ
ϕ P (b + 2
)(x − b) if x ∈ b − 2
, b

 0 if x ∈ (b, ∞)

where Lj (x) = 1δ ϕP xj + 2δ − ϕP xj − 2δ x − xj + 2δ + ϕP xj − 2δ . Note the





since δ < m, all intervals (xj−1 + 2δ , xj − 2δ ) are not empty. Then g is well defined.
By definition of g, for each x ∈ (−∞, a) ∪ (b, ∞), g(x) = 0. Namely, g vanishes outside
[a, b].
To show the continuity of g on R, we can proceed by parts. For (−∞, a) and (b, ∞)
is continuous because is constant. Also, for each j = 1, ..., n, in the interval (xj−1 +
δ
2
, xj − 2δ ) ⊂ (xj−1 , xj ] in which ϕP (x) is constant.
Since h and k are linear functions, are continuous in (a, a+ 2δ ) and (b− 2δ , b), respectively.
Then g is continuous in (a, a + 2δ ) ∪ (b − 2δ , b). Also, for a fix j = 1, ..., n − 1, Lj is a
linear function then Lj is continuous in (xj − 2δ , xj + 2δ ). Subsequently, g is continuous
in ∪n−1 δ δ
j=1 (xj − 2 , xj + 2 ). Recapping, g is continuous in R − ({a, b} ∪ {xj +
δ
2
: j =
δ
0, 1, ..., n − 1} ∪ {xj − 2
: j = 1, 2, ..., n}). Using that the limit of a function exist and
is L iff the left limit is L and right limit is L. Is clear that:
For a, we have
 
2 δ
lim g(x) = lim− 0 = 0 lim g(x) = lim+ ϕP a + (x − a) = 0
x→a− x→a x→a+ x→a δ 2
Then limx→a g(x) = 0 = g(a), namely g is continuous in a. Similarly, for b
 
2 δ
lim g(x) = lim− ϕP b + (x − b) = 0 lim+ g(x) = lim+ 0 = 0
x→b− x→b δ 2 x→b x→b

4
Then, limx→b g(x) = 0 = g(b), namely g is continuous in b.
For a fix j = 0, 1, ..., n − 1, we have
 
δ
lim g(x) = lim Lj (x) = ϕP xj +
x→(xj + 2δ )− x→(xj + 2δ )− 2
 
δ
lim g(x) = lim ϕP (x) = ϕP xj +
x→(xj + 2δ )+ x→(xj + 2δ )+ 2
Then,    
δ δ
lim g(x) = ϕP xj + = g xj +
x→xj + 2δ 2 2
δ
So g is continuous in {xj + 2
: j = 0, 1, ..., n − 1}.
For another j = 1, ..., n, we have
 
δ
lim g(x) = lim ϕP (x) = ϕP xj −
x→(xj − 2δ )− x→(xj − 2δ )− 2
 
δ
lim g(x) = lim Lj (x) = ϕP xj −
x→(xj + 2δ )+ x→(xj − 2δ )+ 2
Then,    
δ δ
lim g(x) = ϕP xj − = g xj −
x→xj − 2δ 2 2
δ
So g is continuous in {xj − 2
: j = 0, 1, ..., n − 1}. Finally, g is continuous on R.
Since ϕP is a step function, ∃M > 0 such that |ϕP (x)| ≤ M for all x ∈ [a, b]. It will be
useful to calculate

Z Z
|ϕP − g|dµ = |ϕP − g|dµ
R [a,b]
Z b
= |ϕP (x) − g(x)|dx
a

Even more, by definition of g, ϕP is different to g where h, Lj ’s and k are defined. So

Z Z a+ 2δ Xn−1 Z xj + 2δ Z b
|ϕP −g|dµ = |h(x)−ϕP (x)|dx+ |Lj (x)−ϕP (x)|dx+ |k(x)−ϕP (x)|dx
R a j=1 xj − 2δ b− 2δ

5
For the left handed term, we have the next inequality
Z a+ 2δ 
Z a+ 2δ

2 δ
|h(x) − ϕP (x)|dx = ϕP a + (x − a) − ϕP (x) dx
a a δ 2
Z a+ δ   Z a+ δ
2 2 δ 2
≤ ϕP a + |x − a|dx + |ϕP (x)|dx
a δ 2 a
Z a+ δ Z a+ δ
2 2 2
≤ M (x − a)dx + M dx
δ a a
2 δ2 δ δ
≤ M + M < M δ + M < 2M δ, if δ < 1
δ 2 2 2
Similarly, for the right handed term
Z b
|k(x) − ϕP (x)|dx < M δ
b− 2δ

For the middle term, for a fix j = 1, ..., n − 1

Z xj + 2δ Z xj + 2δ Z xj
|Lj (x) − ϕP (x)|dx = |Lj (x) − ϕP (x)|dx + |Lj (x) − ϕP (x)|dx
xj − 2δ xj xj − 2δ

Solving the left handed integral

Z xj + δ δ

δ 


− −
  
2 ϕP xj + ϕP x j δ δ
2 2
IL = x − xj + + ϕP xj − − ϕP (x) dx
xj δ 2 2

Since, ϕP (x) is constant in (xj , xj + 2δ ) ⊂ (xj , xj+1 ]. Then,

δ δ xj + 2δ Z xj + δ


Z
|ϕP xj + − ϕP xj − |
     
2 2 δ 2 δ δ
IL ≤ x − xj + dx+ ϕP xj − − ϕP xj + dx
δ xj 2 xj 2 2
δ δ



We have |ϕP xj + 2
− ϕP xj − 2
| < 2M

2M δ 2 δ
IL < + 2M = M δ + M δ = 2M δ
δ 2 2
And, in a similar way for right handed integral,

IR < 2M δ

Therefore,
Z xj + 2δ
|Lj (x) − ϕP (x)|dx < 4M δ
xj − 2δ

6
 ε
If we choose a δ < min{1, m, 8nM } we get
Z a+ δ n−1 Z xj + δ Z b n−1
2 X 2 X
|h(x)−ϕP (x)|dx+ |Lj (x)−ϕP (x)|dx+ |k(x)−ϕP (x)|dx < 2M δ+ (4M δ)+2M δ
a i=1 xj − 2δ b− 2δ j=1

= 4M δ + (n − 1)(4M δ) = 4nM δ
So Z
ε
|ϕP − g|dµ <
R 2
Finally, by the triangle inequality
Z Z Z
ε ε
|f − g|dµ ≤ |g − ϕP |dµ + |ϕP − g|dµ < + = ε
R R R 2 2
□

Problem 19.14
Suppose that (X, F, µ) is a finite measure space. Let f : X → R be an F-measurable
function such that: Z Z Z
2 3
f dµ = f dµ = f 4 dµ
X X X
Show that f = 1E µ-a.e. for some E ∈ F.

Solution Given the condition, we can form

Z Z Z
4 3
f dµ − 2 f dµ + f 2 dµ = 0
X X X
Using linearity Z Z
4 3 2
f − 2f + f dµ = f 2 (f − 1)2 dµ = 0
X X
Since f 2 (f − 1)2 : X → [0, ∞]. Using proposition 19.3.6, we got f 2 (f − 1)2 = 0 µ−a.e. in X.
Moreover, since x2 = 0 if only if x = 0, then f (f − 1) = 0 µ−a.e. in X. Let E ∈ F be the set
that satisfies µ-a.e. condition, namely f (x)[f (x) − 1] = 0 if x ∈ E. Since f is F-measurable
and {1} ∈ B, then f ∗ (1) ∈ F. So the indicator function 1f ∗ (1) is F-measurable and satisfies
1f ∗ (1) (x)(1f ∗ (1) (x) − 1) = 0 for all x ∈ E. Then f = 1f ∗ (1) µ-a.e. in X.
□

Problem 19.20
Let X = [0, ∞), (X, L, µ) be the induced Lebesgue measure space and f : X →
R∞
[0, ∞) be a non-negative function. If the improper Riemann integral 0 f (x)dx =
Rk R
limk→∞ 0 f (x)dx exists, then the Lebesgue integral X f dµ exists and coincides with
the improper Riemann integral.

7
 Solution By the theorem 19.7.4, for each n ∈ N
Z n Z
f (x)dx = f dµ
0 [0,n]

Even more, Z Z
f dµ = f 1[0,n] dµ
[0,n] X

Since f is non-negative, the sequence {f 1[0,n] }n∈N satisfies f 1[0,n] ≤ f 1[0,n+1] and f 1[0,n] → f
when n → ∞, this means f 1[0,n] ↑ f . Then, by definition of Riemann improper integral and
using the Monotone Convergence Theorem,
Z ∞ Z n Z Z
f (x)dx = lim f (x)dx = lim f 1[0,n] dµ = f dµ
0 n→∞ 0 n→∞ X X