Measure Theory and Integration

MATH 34000
Thomas Jordan

September 2017

Copyright Notice This material is copyright of the University unless ex-

plicitly stated otherwise. It is provided exclusively for educational purposes
at the University and is to be downloaded or copied for your private study
only.

Prerequisites: MATH 20200 Metric Spaces 2

Unit aims: The aim of the unit is to introduce measure theory and the
Lebesgue integral.

Relation to Other Units: This unit is an element of a sequence of anal-

ysis courses at Levels I/5, H/6 and M/7. However it also relates to the unit
ergodic theory and dynamical systems and all units in probability.

Teaching Methods: A course of 30 lectures and additional problem classes.

Lecturer: Ivor McGillivray

Assessment Methods: The final assessment mark for the unit is calcu-
lated from a standard 2.5-hour written closed-book examination in January
2014. No calculators are permitted.

Texts:

• R. G. Bartle, The Elements of Integration and Lebesque Measure,

Wiley, New York 1995. (this is the closest text to this unit).

• G. de Barra, Introduction to Measure Theory, Van Nostrand Reinhold

Company, New York 1974.

• R. G. Bartle, D. R. Sherbert, Introduction to Real Analysis, Wiley,

New York 1992.

1
1 Motivation and Background
1.1 The Riemann Integral
In further topics in analysis you saw the Riemann integral.
Definition 1.1. Let f : [a, b] → R, a < b and let P = {x0 , x1 , · · · , xn } be a
partition of [a, b] with a = x0 < x1 < · · · < xn = b. Set
n
X
L(P, f ) = inf{f (x) : xi−1 ≤ x < xi }(xi − xi−1 ),
i=1

and likewise
n
X
U (P, f ) = sup{f (x) : xi−1 ≤ x < xi }(xi − xi−1 ).
i=1

Then f is Riemann integrable if

sup L(P, f ) = inf U (P, f ),

P P

where the supremum and infimum

R b are taken over all possible partitions of
[a, b]. The Riemann integral a f (x)dx is defined to be this common value.
However there are some features of this definition which are not ideal.
Remark. Not all functions are Riemann integrable, and in particular the
pointwise limit of a sequence of Riemann integrable functions need not be
Riemann integrable.
Example. Define f : [0, 1] → R by
(
1, x ∈ Q ∩ [0, 1]
f (x) = .
0, else

Let q1 , q2 , · · · be an enumeration of Q ∩ [0, 1]. Then f is the pointwise limit

of the sequence of Riemann integrable functions fn : [0, 1] → R,
(
1, x ∈ {q1 , · · · , qn }
fn (x) = ,
0, else

but f is not Riemann integrable.

Definition 1.2. Let X be a set and let E ⊂ X. We define the characteristic
function of E, χE : X → R by

1 if x ∈ E
χE (x) =
0 if x ∈ /E

2
 A step function on R is defined to be a finite linear combination of
characteristic functions for intervals. One way of thinking about Riemann
integration is that you first approximate the function from above and below
by step functions. The approach in this course will be to instead approximate
the function by linear combinations of more general characteristic functions
(called simple functions). This will allow more functions to be integrable,
such as the example above, and better behaviour when looking at limits at
functions.

1.2 The Extended Real Number System

Definition 1.3. R̄ := R ∪ {−∞, +∞}.
Definition 1.4. Arithmetic on R̄:
Let x ∈ R. Then

(±∞) + (±∞) = (±∞) = (±∞) + x,

(±∞)(±∞) = (+∞),
(±∞)(∓∞) = (−∞),
and 
±∞, x > 0

x(±∞) = ∓∞, x < 0 .

0, x = 0


Definition 1.5. Let (xn ) be a sequence in R̄. Then

lim inf xn := lim ( inf xr ),

n→∞ n→∞ r≥n

and
lim sup xn := lim (sup xr ).
n→∞ n→∞ r≥n

Note that these always exist for sequences in R̄.

Theorem 1.6. If (xn ) is a sequence in R̄ and Λ ∈ R then

(i) Λ = lim sup xn ⇐⇒ ∀ > 0

n→∞
(a) xn < Λ +  for all sufficiently large n, and
(b) xn > Λ −  for infinitely many n.

Likewise

(ii) λ = lim inf xn ⇐⇒ ∀ > 0

n→∞
(a) xn > λ −  for all sufficiently large n, and
(b) xn < λ +  for infinitely many n.

3
Corollary 1.7. Let (xn ) be a sequence in R̄. Then

lim xn = l ⇐⇒ lim sup xn = lim inf xn = l.

n→∞ n→∞ n→∞

2 Measurable Sets
Definition 2.1. Let X 6= ∅. A family X of subsets of X is a σ-algebra if

(i) ∅ ∈ X, X ∈ X,
(ii) A ∈ X =⇒ Ac ∈ X, and
(iii) A1 , A2 , · · · ∈ X =⇒ ∪∞
n=1 An ∈ X.

(X, X) is called a measurable space, and the sets in X are called measur-
able sets.
Remark. Let (X, X) be a measurable space, and let A1 , A2 , · · · ∈ X. Then
∩∞
n=1 An ∈ X.

Example. Let X be a set then {∅, X} is always a sigma algebra. Similarly

the power set P(X) consisting of all subsets of X is an example of a sigma
algebra.
If we take X = {a, b, c} and X = {X, {a, b}, {c}, ∅} then we can see this
is a sigma algebra (it can be thought of as the sigma algebra which does not
distinguish between a and b). However Y = {X, {a}, {b}, {c}, ∅} is not a
sigma algebra.
Definition 2.2. Let X 6= ∅, and let A be a non-empty collection of subsets
of X. Let Y be the collection of all σ-algebras containing A. Then ∩X∈Y X is
the σ-algebra generated by A. This is the smallest σ-algebra containing
A.
Definition 2.3. Let X = R, and let A = {(a, b) : a, b ∈ R̄, a < b}. The
σ-algebra generated by A is the Borel algebra B. This is the smallest
σ-algebra containing all open sets. A set B ∈ B is a Borel set.
The Borel σ-algebra is the one most commonly used throughout this
unit.
Definition 2.4. Let (X, X) be a measurable space. Then f : X → R is a
X-measurable function if for any α ∈ R, the set {x ∈ X : f (x) > α} is
in X.
Example. For any measurable space (X, X) a constant function is measur-
able. If X = P(X) then any function is measurable. If we let X = N,
O = {1, 3, 5, 7 . . .} and E = {2, 4, 6, 8, . . .} then X = {∅, O, E, N} is a sigma
algebra. A function f : N → R is measurable if and only if f (x) = f (y) for
all x, y ∈ E and for all x, y ∈ O.

4
Lemma 2.5. Let (X, X) be a measurable space, and let f : X → R. Then
the following are equivalent:

(i) ∀α ∈ R, Aα = {x ∈ X : f (x) > α} ∈ X,

(ii) ∀α ∈ R, Bα = {x ∈ X : f (x) ≤ α} ∈ X,
(iii) ∀α ∈ R, Cα = {x ∈ X : f (x) ≥ α} ∈ X,
(iv) ∀α ∈ R, Dα = {x ∈ X : f (x) < α} ∈ X.

Proof. The fact that (i) and (ii) are equivalent is immediate by taking com-
plements (similarly (iii) and (iv)). So we need to show (i) and (iii) are
equivalent. To do this let αn be a sequence which converges monotonically
to α from below. We can now show

∩n Aαn = Cα .

Similarly if αn is a sequence which converges monotonically to α from above

then
∪n Cαn = Aα .
The result now follows using the definition of a measurable function.

Lemma 2.6. Let (X, X) be a measurable space, and let f, g : X → R be

measurable. Let F : R2 → R be continuous. Then h : X → R given by
h(x) = F (f (x), g(x)) is measurable.

Proof. Fix α ∈ R and note we need to show that the set

Xα := {x : F (f (x), g(x)) < α} = {x : (f (x), g(x)) ∈ F −1 ((−∞, α))}

is measurable. We can let A = F −1 ((−∞, α)) and note that it is an open set
in R2 (F is continuous) and thus a countable union of open rectangles (sets
of the form (a, b)×(c, d)). So it suffices to show that for any set (a, b)×(c, d)
we have that

{x : (f (x), g(x)) ∈ (a, b) × (c, d)} = {x : a < f (x) < b} ∩ {x : c < g(x) < d}

is measurable. We can now use the fact that f and g are measurable to
complete the proof.

Corollary 2.7. Let f and g be measurable and let c ∈ R. Then f +

g, f g, |f |, fg (for g 6= 0), max{f, g}, min{f, g} and cf are all measur-
able.

Definition 2.8. Let (X, X) be a measurable space. Then f : X → R̄ is

X-measurable if for any α ∈ R, the set {x ∈ X : f (x) > α} is in X.
The collection of all R̄-valued, X-measurable functions on X is denoted by
M (X, X).

5
Lemma 2.9. Let (X, X) be a measurable space. A function f : X → R̄ is
measurable if and only if
(i) A = {x ∈ X : f (x) = +∞} ∈ X,
B = {x ∈ X : f (x) = −∞} ∈ X,
and
(
f (x), x ∈ (A ∪ B)c
(ii) the function f1 (x) = is measurable.
0, x ∈ A ∪ B
Proof. Suppose f1 , A, B are measurable. For α ≥ 0 we have that
{x : f (x) > α} = {x : f1 (x) > α} ∪ A
and so is measurable. Similar arguments work for α < 0.
Now suppose f is measurable. We can write
A = ∩n∈N {f (x) > n} and B = ∩n∈N {f (x) ≤ −n}.
Thus both these sets are measurable. For α ≥ 0
{x : f1 (x) > α} = {x : f (x) > α}\A
and so is measurable. For α < 0
{x : f1 (x) > α} = {x : f (x) > α} ∪ B
and is measurable.

Corollary 2.10. If f ∈ M (X, X) and c ∈ R, then f 2 , cf, |f |, f + , and f −

are all in M (X, X).
Lemma 2.11. Let fn be a sequence in M (X, X), and define
f (x) = inf fn (x), F (x) = sup fn (x),
n n
∗
f (x) = lim inf fn (x), and F ∗ (x) = lim sup fn (x).
n n

Then f, F, f ∗ , and F ∗ are in M (X, X).

Proof. We prove this for f and f ∗ the proofs for F and F ∗ are analogous.
Fix α ∈ R and consider
Aα = {x ∈ X : f (x) > α} = ∪m∈N ∩n∈N {x ∈ X : fn (x) > α + 1/m}.
Since each {x ∈ X : fn (x) > α + 1/m} ∈ X it follows that Aα ∈ X and so f
is measurable.
For the second part note that gk = inf n≥k fn is measurable. Thus for
any α ∈ R we have that
{x ∈ X : f ∗ (x) > α} = ∪∞
k=1 {x ∈ X : gk (x) > α} ∈ X.

6
Corollary 2.12. Let fn be a sequence in M (X, X), and let fn → g pointwise
on X. Then g ∈ M (X, X).

Lemma 2.13. If f, g ∈ M (X, X), then f g ∈ M (X, X).

Proof. Truncate f for k ∈ N by


 f (x) if −k < f (x) < k
fk (x) = k if f (x) ≥ k
−k if f (x) ≤ −k


It is an exercise to show this is measurable. Now do the same to g and use

Lemma 2.11 and Corollary 2.7.

Definition 2.14. A simple function is a finite linear combination of char-

acteristic functions of measurable sets.

Lemma 2.15. (Approximation by simple functions.)

Let f ∈ M (X, X), f ≥ 0. Then there exists a sequence (φn ) in M (X, X)
such that

(i) 0 ≤ φn (x) ≤ φn+1 (x) ∀x ∈ X, n ∈ N,

(ii) lim φn (x) = f (x),
n→∞
(iii) Each φn is a simple function.

Proof. We fix n ∈ N and for k ∈ {0, 1, . . . , n2n − 1} define

  
k k+1
Ek,n = x : f (x) ∈ n , n and En2n ,n = {x : f (x) ∈ [n, ∞)}.
2 2
n
Note that X = ∪2k=0 nE
k,n , the sets are disjoint and are all measurable. We
then define
k
φn (x) = n if x ∈ Ek,n .
2
P nn k
In other words φn = 2k=0 2n χEk,n is a simple function. We can easily show
φn (x) ≤ φn+1 (x) and limn→∞ φn (x) = f (x).

Definition 2.16. A function f : X → C is measurable if Re(f ) and Im(f )

are measurable.

Definition 2.17. Let (X, X) and (Y, Y) be measurable spaces. A function

f : X → Y is measurable if A ∈ Y =⇒ f −1 (A) ∈ X.

7
3 Measures
Definition 3.1. Let (X, X) be a measurable space. Then µ : X → R̄ is a
measure if

(i) µ(∅) = 0,
(ii) A ∈ X =⇒ µ(A) ≥ 0, and
(iii) if A1 , A2 , · · · is a sequence of disjoint sets in X,
X∞
∞
then µ(∪n=1 An ) = µ(An ).
n=1

(X, X, µ) is called a measure space.

Definition 3.2. A measure µ is called a finite measure if µ(X) < ∞.
It is called a σ-finite measure if there is a sequence (An ) of sets in X with
X = ∪∞ n=1 An and µ(An ) < ∞ for each n.

Remark. Let (X, X) be a measure space and let x ∈ X. For E ∈ X we

define
δx (E) = 1 if x ∈ E and δx (E) = 0 if x ∈
/ E.
We can see that δx is a measure (often named after Dirac).
Take (N, P (N)) to be our measurable set for any E ⊂ N we define µ(E)
to be the number of elements in E. µ is called the counting measure and is
not finite but is sigma finite.
Remark. Consider the measurable space (R, B). Then there is a unique
measure λ supported on B, which agrees with length for open intervals (a, b),
i.e. λ((a, b)) = b − a. This measure is called the Lebesgue measure.
We will justify this much later in the unit. If f : R → R is a strictly
increasing continuous function we can define λf (a, b) = f (b) − f (a) and this
also uniquely defines a measure.
Lemma 3.3. (Monotonicity of measures.)
Let µ be a measure on (X, X). If A, B ∈ X and A ⊂ B, then µ(A) ≤ µ(B).
If in addition µ(A) < ∞, then µ(B \ A) = µ(B) − µ(A).
Proof. We can write B = (B ∩ A) ∪ (B ∩ Ac ) and since A ⊂ B this gives
that B = A ∪ B ∩ Ac = A ∪ (B \ A). Since this union is disjoint we can
use properties (ii) and (iii) in the definition of a measure to prove both
results.

Lemma 3.4. Let µ be a measure on (X, X).

(i) If (An ) is an increasing sequence (A1 ⊂ A2 ⊂ · · · ) of measurable sets in
X, then µ(∪∞ n=1 An ) = limn→∞ µ(An ).
(ii) If (Bn ) is a decreasing sequence (B1 ⊃ B2 ⊃ · · · ) of measurable sets in
X, and µ(B1 ) < ∞, then µ(∩∞ n=1 Bn ) = limn→∞ µ(Bn ).

8
Proof. For part (i) let A0 = ∅ and write ∪∞ ∞
n=1 An = ∪k=1 Ak \ Ak−1 . Now
since this is a union of disjoint sets we can use Lemma 3.3 and property (iii)
in the definition of a measure (as long as µ(An ) < ∞ for all n). This gives
that
X∞
∞
µ (∪n=1 An ) = (µ(Ak ) − µ(Ak−1 )) = lim µ(An ).
n→∞
k=1
If µ(AN ) = ∞ for some N ∈ N then µ(An ) = ∞ for all n ≥ N and the result
follows.
For part (ii) use that B1 \ (∩∞ ∞
n=1 Bn ) = ∪k=1 Bk \Bk+1 . Since this union
is disjoint (and µ(B1 ) < ∞)) we have that
∞
X
µ((B1 ) \ (∩∞
n=1 Bn )) = (µ(Bk ) − µ(Bk+1 )) = lim µ(B1 ) − µ(Bk+1 ).
k→∞
k=1

Rearranging and using Lemma 3.3 gives the result.

Definition 3.5. Let (X, X, µ) be a measure space. A proposition P holds

µ-almost everywhere (µ-a.e.) if there is a set N in X such that µ(N ) = 0
and such that P holds on X \ N .
Definition 3.6. A measure space (X, X, µ) is complete if A ⊂ N ∈ X and
µ(N ) = 0 implies A ∈ X.

4 Integration of Non-Negative Functions

In this section we define the notion of an integral with respect to a measure
for a non-negative measurable function. We do this first by defining the
integral for simple functions and then by using the fact that non-negative
measurable functions can be approximated by simple functions to extend
the definition to non-negative measurable functions.
Definition 4.1. Let (X, X, µ) be a measure space. The collection of non-
negative R̄-valued measurable functions on X is denoted by M + (X, X).
Definition 4.2. Recall that a simple function is a function of the form
n
X
φ= ci χAi ,
i=1

where ci ∈ R and Ai ∈ X. It is in standard representation if X = ∪ni=1 Ai ,

the sets Ai are pairwise disjoint, and the numbers ci are distinct.
Definition
Pn 4.3. Let φ be a simple function with standard representation
φ = i=1 ci χAi . Then the integral of φ with respect to µ is
Z Xn
φdµ := ci µ(Ai ) ∈ R̄.
i=1

9
 We now prove a few natural results about the integral for simple func-
tions.
Lemma 4.4. Let (X, X, µ) be a measure space. Let φ, ψ be simple functions
in M + (X, X) and let c ≥ 0. Then
Z Z
(i) cφdµ = c φdµ, and
Z Z Z
(ii) (φ + ψ)dµ = φdµ + ψdµ.

Proof. Part (i) is straightforward. To prove part (ii) we need to consider the
intervals A1 , . . . , An which make up the standard representation for φ and
the intervals B1 , . . . , Bm which make up the standard representation for ψ.
We then take the intervals Ai ∩ Bj for 1 ≤ i ≤ n and 1 ≤ m ≤ m. Note
that these intervals will be disjoint and that φ + ψ will be constant on each
interval. Now group these intervals into those on which ψ + φ take the same
value and then we are in a standard form and the result will follow.

Lemma 4.5. Let (X, X, µ) be a measure space, and letR φ be a simple function
in M + (X, X). Let ν : X → R̄ be defined by ν(A) = φχA dµ. Then ν is a
measure.
Proof. This is set as an exercise.

We can now give the definition of an integral for a non-negative measur-

able function.
Definition 4.6. Let f ∈ M + (X, X). Then the integral of f with respect
to µ is
Z Z 
f dµ := sup φdµ : φ ≤ f, φ is a simple measurable function ∈ R̄.

Definition 4.7. Let f ∈ M + (X, X), and let A ∈ X. Then the integral of
f with respect to µ over A is
Z Z
f dµ := f χA dµ.
A

Lemma 4.8. Let (X, X, µ) be a measure space.

(i) Let f, g ∈ M + (X, X), and let f ≤ g on X.
Z Z
Then f dµ ≤ gdµ.

(ii) Let f ∈ M + (X, X), and let A, B ∈ X with A ⊂ B.

Z Z
Then f dµ ≤ f dµ.
A B

10
Proof. Part (i) follows since if φ ≤ f where φ is a simple function then
φ ≤ g. Part (ii) follows from part (i).

We now come to one of the most important Theorems for the unit.

Theorem 4.9. (Monotone Convergence Theorem)

Let (X, X, µ) be a measure space. Let (fn ) be a monotone increasing sequence
of functions in RM + (X, X), whichR converges point wise to f . Then f ∈
M + (X, X), and f dµ = limn→∞ fn dµ.
R
Proof.
R Since fn ≤ f for all n it follows Lemma 4.8 that lim n→∞ fn dµ ≤
f dµ. To show the inequality in the other direction we fix a simple function
φ ≤ f , 0 < α < 1 and let

An = {x : αφ(x) ≤ fn (x)}.

Note that for each n An ∈ X and An ⊆ An+1 . We also have that X =

∪∞
n=1 An . By Lemma 1 we can define a measure ν on (X, X) by
Z
ν(A) = φdµ(x)
A

for all A ∈ X. It follows from Lemma 3.4 that

Z
φdµ = ν(X) = ν (∪∞
n=1 An ) = lim ν(An ).
n→∞

However for all n ∈ N

Z Z Z
fn dµ ≥ fn dµ ≥ α φdµ = αν(An )
X An An
R R
and thus α φdµ ≤ limn→∞ fn dµ. Since this holds for all 0 < α < 1 and
simple functions φ ≤ f it follows that
Z Z
f dµ ≤ lim fn dµ.
n→∞

This is a very important result on which much of the unit is based. We

will now give several consequences of the Theorem.

Corollary 4.10.
Z Z
+ +
(i) f ∈ M (X, X), c ≥ 0 =⇒ cf ∈ M (X, X), cf dµ = c f dµ, and
Z Z Z
+ +
(ii) f, g ∈ M (X, X) =⇒ f + g ∈ M (X, X), (f + g)dµ = f dµ + gdµ.

11
Remark. The proof is based on approximation by simple functions (Lemma
2.14) and the Monotone Convergence Theorem. The proofs of the next three
results are also based on the Monotone Convergence Theorem.
Theorem 4.11. (Fatou’s Lemma)
Let (X, X, µ) be a measure space. Let (fn ) be a sequence of functions in
M + (X, X). Then
Z Z
lim inf fn dµ ≤ lim inf fn dµ.
n→∞ n→∞

Proof. We consider the functions defined gn (x) = inf{fk (x) : k ≥ n} and

use the Monotone Convergence Theorem.

Theorem 4.12. Let (X, X, µ) be a measure space, and let (fn ) be a sequence
in M + (X, X). Then
∞ ∞ Z
Z X !
X
fn dµ = fn dµ.
n=1 n=1
Pk
Proof. Note that n=1 fn is a monotone increasing sequence and apply the
Monotone Convergence Theorem.

Theorem 4.13. Let (X, X, µ) be a Rmeasure space, and let f ∈ M + (X, X).
Let ν : X → R̄ be defined by ν(A) = A f dµ. Then ν is a measure.
Proof. By definition it follows that ν(∅) = 0 and ν(A) ≥ 0 for all A ∈ X.
∞
P∞ let A1 , A2 , ... ∈ X be disjoint, A = ∪n=1 An and note that f χA =
Now
n=1 f χAn . We have that

∞ ∞ Z ∞
Z Z X !  X
X
ν(A) = f χA dµ = f χAn dµ = f χAn dµ = ν(Ai )
n=1 n=1 n=1

Theorem 4.14. Let (X, X, µ) be a measure space, and let f ∈ M + (X, X).
Then Z
f (x) = 0 µ − a.e. on X ⇐⇒ f dµ = 0.

Proof. Note that f (x) = 0 µ − a.e. on X implies that the set

E = {x ∈ X : f (x) > 0}
R
satisfies µ(E) = 0. Let fn = nχE . We have that fn dµ = 0 and lim inf n→∞ fn ≥
f . Thus by Fatou’s Lemma
Z Z Z
f dµ ≤ lim inf fn dµ ≤ lim inf fn dµ = 0.
n→∞ n→∞

12
 On the other hand if f dµ = 0 let A = {x ∈ X : f (x) = 0} and B = Ac .
R

Let Bn = {x ∈ X : f (x) ≥ n1 } and note that

Z Z
µ(Bn )
0= f dµ ≥ f dµ ≥ .
Bn n

Thus µ(Bn ) = 0 for all n and µ(B) = µ(∪∞

n=1 Bn ) = 0.

Remark. Theorem 4.14 allows us to extend the Monotone Convergence

Theorem to the case when we have convergence µ-a.e.

Theorem 4.15. (Monotone Convergence Theorem for convergence

µ-a.e.)
Let (X, X, µ) be a measure space. Let (fn ) be a monotone increasing sequence
of functions
R in M + (X,RX), which converges µ-a.e. on X to a function f .
Then f dµ = limn→∞ fn dµ.

Proof. Let A ∈ X be such that if x ∈ A then limn→∞ fn (x) = f (x) and

B := Ac satisfies µ(B) = 0. If we consider the functions fn χA then they are
a monotonically increasing sequence of function which converge to f χA . So
by the Monotone Convergence Theorem we have that
Z Z
lim fn χA dµ = f χA dµ.
n→∞

Furthermore
Z Z Z Z Z Z
f dµ = f χA dµ + f χB dµ and fn dµ = fn χA dµ + fn χB dµ
R
However
R χB (x) = 0 for µ almost every x so by our previous result f χB dµ =
fn χB dµ = 0 and the result follows.

Definition 4.16. Let µ and ν be measures on (X, X). Then ν is absolutely

continuous with respect to µ, denoted ν  µ, if A ∈ X and µ(A) = 0
implies ν(A) = 0.

Theorem
R 4.17. Let f ∈ M + (X, X), and define ν : X → R̄ by ν(A) =
A f dµ. Then ν is absolutely continuous with respect to µ.

Proof. Let A ∈ X satisfy µ(A) = 0. We then have that

Z
ν(A) = f χA dµ.

However since µ(A) = 0, f χA = 0 for µ almost all x so ν(A) = 0.

13
Theorem 4.18. Let X = [a, b] ⊂ R, and let X be the collection of Borel
sets in X. Let f : [a, b] → [0, ∞) be continuous and supported in [a,b]. The
Z Z b
f dλ = f (x)dx,
a

where the LHS denotes the Lebesgue integral and the RHS denotes the Rie-
mann integral.

Remark. The idea of the proof is to show that the result holds for step
functions, and then use approximation by these.

5 Integrable functions
Throughout this section we will let (X, X, µ) be an arbitrary measure space.
We want to extend the definition of the integral with respect to µ to all
measurable functions, f . To do this we will compute the integrals of f + and
f − and take the difference. However this can cause problems since we have
not defined ∞ − ∞ so for the definition to make sense we insist that both
these integrals are finite.

Definition 5.1. An R-valued, X-measurable function f : X → R is inte-

grable (or summable) if
Z Z
+
f dµ < +∞ and f − dµ < +∞,

where f + := max{f, 0} and f − := − min{f, 0}.

We then define Z Z Z
f dµ := f + dµ − f − dµ.

For A ∈ X we define
Z Z Z
f dµ := +
f dµ − f − dµ.
A A A

The collection of all R-valued, X-measurable functions on X is denoted by

L(X, X, µ).

Lemma 5.2. Let f, g ∈ M + (X, X) with f dµ < +∞ and gdµ < +∞.
R R

Then f − g ∈ L(X, X, µ) and

Z Z Z
(f − g)dµ = f dµ − gdµ.

14
Proof. To proof this we let h = (f − g) and note that h+ ≤ f and h− ≤ g.
So h is integrable and
Z Z Z
hdµ = h dµ − h− dµ.
+

We can then write

h+ + g = h− + f
and since the integrals of all of h+ , h− , f, g are finite we can use Corollary
4.10 to deduce the result.

We now introduce the notion of a charge which is the same as the notion
of measures except we allow negative values for the charge of sets.

Definition 5.3. Let (X, X) be a measurable space. Then ν : X → R is a

charge if

(i) ν(∅) = 0, and

(ii) If A1 , A2 , · · · is a sequence of disjoint sets in X,
X∞
then ν(∪∞ n=1 n A ) = ν(An ).
n=1
P∞
Note that the sum n=1 ν(An ) must converge unconditionally whenever
A1 , A2 , . . . ∈ X is a sequence of disjoint sets, as the union is independent of
order, and ν is R-valued.

Theorem
R 5.4. Let f ∈ L(X, X, µ), and define ν : X → R by ν(A) =
A f dµ. Then ν is a charge.

Proof. If we let ν + (A) = A f + dµ and ν − (A) = A f − dµ then ν = ν + − ν − .

R R

ν + and ν − are measures and thus charges. It is straightforward to now show

that ν is a charge.

Theorem 5.5. If (X, X, µ) is a measure space and f ∈ M (X, X) then

f ∈ L(X, X, µ) ⇐⇒ |f | ∈ L(X, X, µ).

Proof. Write |f | = Rf + + f −R . Thus if |f | ∈ L(X, X, µ) then f − , f + ∈

M + (X, X) and max{ f + dµ, f − dµ} < ∞. This means that f ∈ L(X, X, µ).
− + +
R +handR if f− ∈ L(X, X, µ) then by definition f , f ∈ M (X, X)
On the other
and max{ f dµ, f dµ} < ∞. This means that |f | ∈ L(X, X, µ).

We make use of this with the following simple corollary.

Corollary 5.6. Let f be measurable, let g ∈ L(X, X, µ), and suppose |f | ≤

|g|. Then f ∈ L(X, X, µ).

15
 R R
Proof. It immediately follows that |f |dµ ≤ |g|dµ < ∞ and thus |f | ∈
L(X, X, µ). We can now apply Theorem 5.5 to deduce the result.

Theorem 5.7. L(X, X, µ) is a vector space over R.

Proof. We need to show that for all α ∈ R and f, g ∈ L(X, X, µ) that

αf, f + g ∈ L(X, X, µ). Firstly 0 · f = 0 ∈ L(X, X, µ). Now let α 6= 0 and
f = f + − f − . If α > 0 then we can immediately see that αf ∈ L(X, X, µ).
If α < 0 then (αf )+ = |α|f − and (αf )− = |α|f + and again it follows that
αf ∈ L(X, X, µ).
For the second part we use the triangle inequality and the previous two
results. We have
|f + g| ≤ |f | + |g|.
By Theorem 5.5 and corollary 4.10 it follows that |f | + |g| R∈ L(X, X, µ).
Thus |f +
R g| ∈ L(X, X, µ) by Corollary 5.6. (NB to see that (f + g)dµ =
f dµ + gdµ write f + g = f + + g + − (f − + g − ) and use Lemma 5.2.)
R

Theorem 5.8. (Lebesgue’s Dominated Convergence Theorem)

Let (fn ) be a sequence of functions in M (X, X), which converges µ-a.e. on
X to a measurable function f : X → R. Suppose there exists a function
g ∈ L(X, X, µ) such that |fn | ≤ g for all n ∈ N. Then fn ∈ L(X, X, µ) for
each n ∈ N, f ∈ L(X, X, µ) and
Z Z
f dµ = lim fn dµ.
n→∞

Proof. We will first prove the result in the case when convergence is for all
of X. By corollary 5.6 it follows that each function fn ∈ L(X, X, µ), and
since |f | ≤ g it also follows that f ∈ L(X, X, µ). We now prove the result by
using Fatou’s lemma twice. Note that fn + g ≥ 0 for all n so we can apply
Fatou. This gives
Z Z Z Z 
f + gdµ = (lim inf fn + g)dµ ≤ lim inf fn dµ + gdµ
n→∞ n→∞
R R
and so f dµ ≤ lim inf n→∞ fn dµ. On the other hand g − fn > 0 and we
can apply Fatou again to see that
Z Z
−f dµ ≤ lim inf −fn dµ.
n→∞
R R
Multiplying throughR by −1 gives Rf dµ ≥ lim supn→∞ f dµ. So putting
this together gives f dµ = limn→∞ fn dµ.

16
 We now need to consider the case when the convergence is only for µ-a.e
x ∈ X. In this case we know there exists A ∈ X where for all x ∈ A we have
that limn→∞ fn (x) = f (x) and where µ(Ac ) = 0. So
Z Z Z
f dµ = f χA dµ + f χAc dµ

and Z Z Z
fn dµ = fn χA dµ + fn χAc dµ.
R R
We can deduce from above that limn→∞ fn χA dµ = f χA dµ and use The-
orem 4.14 to show that the integrals over Ac are 0. The result follows.
−nx
Example. Take (R+ , B, λ) as our measure space Let fn (x) = e√x and
R
In = fn dλ. We have that limn→∞ fn (x) = 0 and |fn (x)| ≤ g(x) where

x−1/2

if 0<x<1
g(x) =
e−x if x≥1

We can use that for non-negative continuous functions on closed bounded

intervals that the Riemann integral and the Lebesgue integral agree to show
that Z
g(x)dλ < ∞.

So we can apply Lebesgue’s dominated convergence theorem to show that

limn→∞ In = 0 (without actually having to integrate fn ). We can easily see
that the convergence is not uniform (for each n, limx→0 fn (x) = ∞), so we
could not show this so easily working just with the Riemann integral.

Remark. Lebesgue’s Dominated Convergence Theorem allows us to deduce

several results about the behaviour of limits inside integrals The following
results follow from Lebesgue’s Dominated Convergence Theorem. Through-
out the rest of this section let f : X × [a, b] ⊂ R −→ R be a function such
that f (., t) is X-measurable for a ≤ t ≤ b.

Theorem 5.9. (Convergence for a family of functions parametrized

by t)
Suppose f (x, t0 ) = limt→t0 f (x, t) for all x ∈ X, and suppose there exists
g ∈ L(X, X, µ) such that |f (x, t)| ≤ g(x) for all t ∈ [a, b]. Then
Z Z
f (x, t0 )dµ(x) = lim f (x, t)dµ(x).
t→t0

Proof. Take a sequence tn , with elements in [a, b], such that limn→∞ tn =
t0 . Let fn (x) = f (x, tn ) and apply the Lebesgue dominated convergence
theorem.

17
Corollary 5.10. Suppose t −→ f (x, t) is continuous on [a, b] for each x ∈
X, and suppose there exists g ∈ L(X, X, µ) such that |f (x, t)| ≤ g(x) for all
t ∈ [a, b]. Then F defined by
Z
F (t) = f (x, t)dµ(x)

is continuous on [a, b].

Proof. This is an immediate corollary of Theorem 5.9.

Corollary 5.11. (Interchanging differentiation and integration)

Suppose the function x −→ f (x, t0 ) is integrable on X for some t0 ∈ [a, b] ⊂
R. Also suppose that ∂f ∂t exists on X × [a, b], and that there exists g ∈
L(X, X, µ) such that | ∂f
∂t (x, t)| ≤ g(x) for all t ∈ [a, b]. Then
Z Z
∂F ∂ ∂f
(t) := f (x, t)dµ(x) = (x, t)dµ(x).
∂t ∂t ∂t

Proof. First of all we show that f (x, t) is integrable for all t ∈ [a, b]. Fix
x ∈ X, t ∈ [a, b] and let {tn }n∈N be a sequence in [a,b] which satisfies
limn→∞ tn = t then
∂f f (x, t) − f (x, tn )
(x, t) = lim
∂t n→∞ t − tn
which is measurable. Moreover we can write

|f (x, t)| = |f (x, t0 ) + f (x, t) − f (x, t0 )| ≤ |f (x, t0 )| + |f (x, t) − f (x, t0 )|

and by the mean value theorem we have that there exists y ∈ (a, b) such
that  
 ∂df 
|f (x, t) − f (x, t0 )| =  (x, y) |t − t0 |. ≤ |g(x)|(b − a).
∂dt
So |f (x, t)| ≤ |f (x, t0 )| + |g(x)|(b − a) and thus f (x, t) ∈ L(X, X, µ). So
f (x, t) is integrable for all t ∈ [a, b].
We now consider fn (x) = f (x,tntn)−f−t
(x,t)
. We know that by the mean value
theorem there exists z ∈ [a, b] such that fn (x) = ∂F ∂t (x, z) and so |fn (x)| ≤
g(x) for all n. So we can apply the Lebesgue dominated convergence theorem
to get Z Z
∂f
(x, t)dµ(x) = lim fn dµ.
∂t n→∞

We then have that since f (x, t) is integrable for all t that

R R
f (x, tn )dµ − f (x, t)dµ
Z
lim fn (x)dµ(x) = lim .
n→∞ n→∞ tn − t

18
This holds for any sequence tn → t, with elements, in [a, b] and we can
deduce that ∂F
∂t (t) exists and
Z Z
dF ∂ ∂f
(t) = f (x, t)dµ(x) = (x, t)dµ(x).
∂t ∂t ∂t

Corollary 5.12. (Interchanging Lebesgue integration with Riemann

integration)
Suppose t −→ f (x, t) is continuous on [a, b] for each x ∈ X, and suppose
there exists g ∈ L(X, X, µ) such that |f (x, t)| ≤ g(x) for all t ∈ [a, b]. Then
Z b Z bZ Z Z b
F (t)dt := f (x, t)dµ(x)dt = f (x, t)dtdµ(x),
a a a

Rwhere the integral with respect to t is the Riemann integral and F (t) =
f (x, t)dµ(x).

Proof. We’ll make use of the fundamental Theorem of calculus R t to relate the
Riemann integral to derivatives. First of all let h(x, t) = a f (x, s)ds. We
know that ∂h∂t (x, t) = f (x, t) and that for a fixed t the function x → h(x, t)
is measurable. Moreover |h(x, t)| ≤ (b −Ra)g(x) so h(x, t) is integrable with
respect to µ and we can define H(t) = h(x, t)dµ. We now use Corollary
5.11 to see that
Z Z
dH ∂h
(t) = (x, t)dµ = f (x, t)dµ(x) = F (t).
dt ∂t
So Z b Z
F (t)dt = H(b) − H(a) = h(x, b) − h(x, a)dµ(x).
a
We know that h(x, a) = 0 and so
Z b Z Z Z b
F (t)dt = h(x, b)dµ(x) = f (x, t)dtdµ(x).
a a

6 Lp spaces
Throughout this section we will again let (X, X, µ) be an arbitrary measure
space.

19
Definition 6.1. Let V be a vector space over R. Then N : V → R is a
norm if for all u, v ∈ V and for all α ∈ R

(i) 0 ≤ N (v) < ∞,

(ii) N (v) = 0 ⇐⇒ v = 0,
(iii) N (αv) = |α|N (v), and
(iv) N (u + v) ≤ N (u) + N (v).

Remark. Normed spaces are metric spaces with the usual metric being
d(u, v) = N (u − v).
Example. Take the vector space over R,

C([0, 1]) = {f : [0, 1] → R : f is continuous}

R1
and N (f ) = 0|f (x)|dx. This is a normed space. However consider
1 1

 0
if x≤ 2 − 2n
1 1 1 1 1 1


fn (x) = n x − 2 + 2n if x∈ 2 − 2n , 2 + 2n
1 1
1 if x> 2 + 2n


This is a Cauchy sequence, however fn can not converge to a continuous

function in the metric induced by this norm. This means the norm is not
complete.
Definition 6.2. N0 : V → R is a seminorm if it satisfies (i), (iii) and
(iv).
R
Remark. Nµ (f ) = |f |dµ R is a seminorm on L(X, X, µ). However it is not
necessarily a norm since |f − g|dµ = 0 does not always imply f = g. For
example take (R, B, λ), f (x) = 0 for all x ∈ R.

1 if x ∈ Q
f (x) =
0 if x ∈ /Q
R
We have that f 6= g but |f − g|dλ = 0.
Definition 6.3. Two functions f, g ∈ L(X, X, µ) are µ-equivalent if f =
g µ-almost everywhere.
Remark. The relation µ-equivalent is an equivalence relation.
Definition 6.4. The equivalence class containing an element f ∈ L(X, X, µ)
is denoted by [f ]. The Lebesgue space L1 (X, X, µ) consists of all µ-equivalence
classes of functions in L(X, X, µ). For [f ] ∈ L1 (X, X, µ) we define
Z
k[f ]k1 := |f |dµ,

where f is any representative of [f ].

20
Theorem 6.5. L1 (X, X, µ) with kk1 is a normed vector space over R.
Proof. The quickest way to see this is a vector space is to consider the
subspace
N = {f ∈ L(X, X, µ) : f (x) = 0 for µ a.e. x}
and observe that L1 (X, X, µ) = L(X, X, µ)/N , i.e. that L1 (X, X, µ) is a
quotient space. If you are not familiar with quotient spaces then just check
that you get a vector space with the operations [f ] + [g] = [f + g] and
α[f ] = [αf ].
To see that kf k is indeed a norm we first have to check it is well defined.
Let f, g ∈ L(X,
R X, µ) with
R [g] = [f ] then f = g for µ almost every x and
so k[f ]k = |f |dµ = |g|dµ = k[g]k. We can then see that for [f ], [g] ∈
L1 (X, X, µ)
R
1. k[f ]k = |f |dµ ≥ 0.
R
2. k[0]k = 0 and if 0 = k[f ]k = |f |dµ then f (x) = 0 for µ almost every
x and [f ] = [0].
R R
3. For α ∈ R, k[αf ]k = |αf |dµ = |α| |f |dµ = αk[f ]k.
R R R
4. k[f + g]k = |f + g|dµ ≤ |f |dµ + |g|dµ = k[f ]k + k[g]k.

Remark. The elements of L1 (X, X, µ) are equivalence classes of functions

in L(X, X, µ), but these classes are represented by functions which are equal
µ-a.e.. We will usually write f rather than [f ].
Definition 6.6. For 1 ≤ p < ∞, Lp (X, X, µ) consists of all µ-equivalence
classes of X-measurable, R-valued functions f for which
Z
|f |p dµ < ∞.

We define
Z 1
p
p
kf kp := |f | dµ .

When there is no risk of confusion we will write Lp (X) or Lp instead of

Lp (X, X, µ).
We’ll now prove several inequalities, the final one of which will show the
triangle inequality for Lp .
Lemma 6.7. [Youngs’ inequality] If A, B, p, q ∈ R+ where p, q ≥ 1 and
1 = p1 + 1q then
Ap B q
AB ≤ + .
p q

21
Proof. Fix A, B as positive real numbers and p, q ≥ 1 where p1 + 1q = 1.
We also let a = Ap , b = B q and 0 < α = p1 < 1. We define a function
φ : (0, ∞) → R by
φ(t) = αt − tα .
We can see that φ0 (t) < 0 for t < 1, φ0 (1) = 0 and φ0 (t) > 0 for t ≥ 1. Thus
by the mean value theorem φ has a unique minimum value at t = 1. Thus
for all positive t we have that

αt − tα ≥ α − 1.
a
We now substitute in t = b which gives that
aα a
α
≤ (1 − α) + α
b b
and multiplying through by b gives

aα b1−α ≤ (1 − α)b + αa.

1 1
Recall that q =1− p = 1 − α and substitute this in to get

b a
a1/p b1/q ≤ + .
q p
We can now use that a = Ap and b = B q to get
Ap B q
AB ≤ + .
p q

Theorem 6.8. [Hölder’s Inequality]

Let p, q > 1 with p1 + 1q = 1, and suppose f ∈ Lp (X) and g ∈ Lq (X). Then
f g ∈ L1 (X), and
kf gk1 ≤ kf kp kgkq .
Proof. Let f ∈ Lp and g ∈ Lq . If either |f |p = 0 or |g|q = 0 then f g = 0 µ
almost everywhere and the inequality follows. So we assume |f |p 6= 0 and
|g|q 6= 0. We now let x ∈ X, A = |fkf(x)| |g(x)|
kp and B = kgkq . We can now apply
Young’s inequality for a and b to get that
|f (x)||g(x)| |f (x)|p |g(x)|q
≤ + .
kf kp kgkq pkf kpp qkqkqq
Since f p and g q are integrable it follows that f g is integrable and so f g ∈
L1 (X). If we integrate both sides with respect to µ we obtain that
Z
1 1 1
|f (x)g(x)|dµ(x) ≤ + = 1
kf kp kgkq p q

22
and we can rearrange to get

kf gk1 ≤ kf kp kgkq .

Corollary 6.9. (Cauchy-Schwartz Inequality)

Suppose f, g ∈ L2 (X). Then f g ∈ L1 (X), and

kf gk1 ≤ kf k2 kgk2 .

Proof. This is Hölder’s inequality with p = q = 2.

Corollary 6.10. Let p, q, r > 1 with p1 + 1

q = 1r , and suppose f ∈ Lp (X)
and g ∈ Lq (X). Then f g ∈ Lr (X), and

kf gkr ≤ kf kp kgkq .
r r
Proof. We use that |f |r ∈ Lp/r , |g|r ∈ Lq/r and that p + q = 1 to allow us
to apply Hölder’s inequality.

Corollary 6.11. Let 1 ≤ p < r < q, and suppose f ∈ Lp (X) ∩ Lq (X). Then
f ∈ Lr (X) with
kf kr ≤ kf kαp kf k1−α
q ,
α 1−α
where α ∈ (0, 1) is such that p + q = 1r .

Proof. We use that |f |α ∈ Lp/α and |f |1−α ∈ Lq/(1−α) to apply the previous
inequality.

Remark. The next result is the triangle inequality for the Lp -norms.

Theorem 6.12. (Minkowski’s Inequality)

Suppose f, g ∈ Lp (X) for some p ≥ 1. Then f + g ∈ Lp (X), and

kf + gkp ≤ kf kp + kgkp .

Proof. We have already proved this for p = 1 so we suppose p > 1 and let
q satisfy that p1 + 1q = 1 (so p = (p − 1)q). To avoid trivialities we assume
kf + gkp > 0 and we also know straight away that f + g is measurable. We
can write for all x that

|f + g(x)|p ≤ (2 max{|f (x)|, |g(x)|})p ≤ 2p (|f (x)|p + |g(x)|p )

and we can conclude that by Corollary 5.6 and Theorem 5.7 that |f + g|p is
integrable and so f + g ∈ Lp .
We now write

|(f +g)(x)|p = |(f +g)(x)||(f +g)(x)|p−1 | ≤ |f (x)||(f +g)(x)|p−1 +|g(x)||(f +g)(x)|p−1 .

23
Since p = (p − 1)q we know that (f + g)p−1 ∈ Lq and so we can apply
Holder’s inequality to get

kf |f + g|p−1 k1 ≤ kf kp k(f + g)p−1 kq = kf kp kf + gkp/q

p .

and
kg|f + g|p−1 k1 ≤ kgkp kf + gkp/q
p .

So

kf + gkpp ≤ kf kp kf + gkp/q p/q

p + kgkp kf + gkp

≤ (kf kp + kgkp )kf + gkp/q

p .

p/q
Dividing through by kf + gkp > 0 and using that p − p/q = p(1 − 1/q) = 1
gives that
kf + gkp−p/q
p = kf + gkp ≤ (kf kp + kgkp ).

Definition 6.13. A sequence (fn ) in Lp is a Cauchy sequence in Lp if

∀ > 0, ∃N ∈ N such that m, n ≥ N =⇒ kfn − fm kp < .

A sequence (fn ) in Lp converges to f in Lp if

∀ > 0, ∃N ∈ N such that n ≥ N =⇒ kfn − f kp < .

A normed space is complete if every Cauchy sequence converges in the

metric induced by the norm.
A Banach space is a complete normed space.

Theorem 6.14. Let 1 ≤ p < ∞. Then Lp is a Banach space.

Proof. Using Minkowski’s inequality we can see that Lp is a normed space

so we just need to show the metric induced by the norm is complete. Let
{fn } be a Cauchy sequence in Lp . Thus for all  > 0 there exists N such
that for all n, m ≥ N kfn − fm kp < . We can pick a subsequence gk of fn
such that for all k we have that kgk+1 − gk kp ≤ 2−k . Let
∞
X
g(x) = |g1 (x)| + |gk+1 (x) − gk (x)|
k=1

which is measurable. Thus by Fatou’s Lemma

Z Z n
!p
X
p
g dµ ≤ lim inf |g1 (x)| + |gk+1 (x) − gk (x)| dµ.
n→∞
k=1

24
However by Minkowski’s inequality
Z n
!p !1/p n
X X
|g1 (x)| + |gk+1 (x) − gk (x)| dµ ≤ kg1 kp + kgk+1 (x)−gk (x)kp ≤ kg1 kp +1
k=1 k=1

g p dµ < ∞.
R
which means that
Thus if we let
E = {x : g(x) < ∞}
then µ(E c ) = 0 and gχE ∈ Lp . If we let

g1 (x) + ∞
 P
f (x) := k=1 (gk+1 (x) − gk (x)) if x∈E
0 if x∈
/E

then since |f (x)|p ≤ g(x)p we have that f ∈ Lp . Furthermore we have that

for µ almost all x f (x) = limk→∞ gk (x) and
  
 g(x) + g1 (x) + k−1 (gn+1 (x) − gn (x)) ≤ 2g(x) if x ∈ E
P
n=1
|f −gk (x)| ≤  P k−1

 g1 (x) + g (x) − g (x) ≤ g(x) if x ∈
/ E.
 
n=1 n+1 n

R
Thus by the Lebesgue Dominated Convergence Theorem limk→∞ |f −
gk (x)|p = 0 and so limk→∞ kf − gk kp = 0.
So we have shown convergence of fn to f in Lp along a subsequence.
Now let  > 0 choose M such that for n, m ≥ N that kfn − fm kp ≤ /2
and choose k such that kf − gk kp ≤ /2 and where k is sufficiently large (i.e
gk = fl where l ≥ N ). Now for n ≥ N we can use Minkowski to get that

kfn − f kp = kfn − gk + gk − f kp ≤ kfn − gk kp + kgk − f kp ≤ .

Definition 6.15. L∞ (X, X, µ) is the set of all µ-equivalence classes of X-

measurable, R-valued functions which are bounded µ-almost everywhere. We
define
kf k∞ := inf{M ∈ R : |f | ≤ M µ − a.e.}.
This is called the essential supremum. When there is no risk of confusion
we will write L∞ (X) or L∞ instead of L∞ (X, X, µ).

Theorem 6.16. kk∞ is a norm on L∞ (X, X, µ).

Proof. Exercise

Theorem 6.17. L∞ is a Banach space.

25
Proof. Again we let fn be a Cauchy sequence in L∞ . Let E ∈ X satisfy
that µ(E c ) = 0 and x ∈ E if |fn (x) ≤ kfn k∞ and |fn − fm | ≤ kfn − fm k∞
for all n, m ∈ N. For all x ∈ E we have that fn (x) is a Cauchy sequence in
R so limn→∞ fn (x) exists. Set

limn→∞ fn (x) if x ∈ E
f (x) =
0 if x ∈ /E
For any  > 0 we can find N such that for all n, m > N we have that
kfn − fm k∞ < . Moreover for any x ∈ E there exists k > N such that
|f (x) − fk (x)| < . Thus for x ∈ E and n ≥ N

|f (x) − fn (x)| ≤ |f (x) − fk (x)| + |fk (x) − f (x)| ≤ 2.

So since µ(E c ) = 0 we have that f ∈ L∞ and limn→∞ kf − fn k∞ = 0.

7 Modes of Convergence
As usual we let (X, X, µ) be any measure space.
Remark. We are already familiar with uniform convergence, pointwise con-
vergence, convergence µ-a.e. and convergence in Lp (1 ≤ p ≤ ∞). Now we
will introduce several new modes of convergence, and study the links between
the different types of convergence.
Theorem 7.1. Suppose µ(X) < ∞. Let (fn ) be a sequence in Lp , 1 ≤ p <
∞, which converges uniformly to a function f . Then f ∈ Lp and fn → f in
Lp .
Proof. Let  > 0 and choose N such that for all n ≥ N we have that
supx∈X |fn (x) − f (x)| < . We then have that
Z Z
|fn − f | dµ ≤ p dµ = p µ(X)
p

and the result follows.

Example. Uniform convergence does not in general imply convergence in

Lp . For example take (R, B, λ) as our measure space and define fn (x) =
n−1/p χ[0,n] . We have that

sup |fn (x) − 0| = n−1/p

x∈R

and thus fn converges uniformly to 0. However for all n

Z
|fn |p dλ = n−1 n = 1

so fn does not converge to 0 in Lp .

26
Remark. Let (fn ) be a sequence in L∞ , which converges uniformly to a
function f . Then f ∈ L∞ , and fn → f in L∞ .
Theorem 7.2. Let (fn ) be a sequence in Lp , 1 ≤ p < ∞, which converges
µ-a.e. to a measurable function f . Suppose there exists g ∈ Lp such that
|fn | ≤ g for all n ∈ N. Then f ∈ Lp , and fn → f in Lp .
Proof. This is an application of the Lebesgue Dominated Convergence The-
orem. See exercise sheet 5 question 8.

Corollary 7.3. Suppose µ(X) < ∞. Let (fn ) be a sequence in Lp , 1 ≤ p <

∞, which converges µ-a.e. to a function f . Suppose there exists K ∈ R such
that |fn | ≤ K for all n ∈ N. Then f ∈ Lp and fn → f in Lp .
Remark. Convergence in Lp does not imply pointwise convergence. For
example for (R,R B, λ) take fn (x) = 1 if x ∈ Q and fn (x) = n1 if x ∈
/ Q.
p 1
We then have |fn | dλ = np and so limn→∞ ||fn |||p = 0 but fn does not
converge to f everywhere.
Definition 7.4. A sequence (fn ) of measurable functions converges to f
in measure if for any α > 0

lim µ({x ∈ X : |fn (x) − f (x)| ≥ α}) = 0.

n→∞

A sequence (fn ) of measurable functions is Cauchy in measure if for any

α,  > 0, there exists M ∈ R such that

m, n ≥ M =⇒ µ({x ∈ X : |fn (x) − fm (x)| ≥ α}) < .

Remark. Uniform convergence implies convergence in measure.

Remark. Convergence in Lp implies convergence in measure. To see this
fix fn ∈ Lp such that fn converges to f in Lp . Let a > 0 and let An = {x :
|fn (x) − f (x)| > a}. Thus
Z
|fn − f |p ≥ µ(An )ap

|fn − f |p dµ = 0 we must have that limn→∞ µ(An ) = 0.

R
and since limn→∞
Remark. Pointwise convergence need not imply convergence in measure.
Again take (R, B, λ) and let fn : R → R be defined by fn = χ[n,∞] . We then
have that limn→∞ fn (x) = 0 but

λ{x : |fn (x) ≥ 1} = ∞ for all n ∈ N.

Theorem 7.5. Let (fn ) be a sequence of R-valued, measurable functions,

which is Cauchy in measure. Then there is a subsequence of fn , which
converges µ-a.e. and in measure to a R-valued, measurable function f .

27
Proof. Since fn is Cauchy in measure if we take k ∈ N then there exists
N (k) such that for all n, m ≥ N (k) we have that

µ({x : |fn (x) − fm (x)| ≥ 2−k } ≤ 2−k

and we can take N (1) < N (2) < · · · . Thus we can choose a subsequence
fnk = gk such that nk ≥ N (k) and thus

µ({x : |gk+1 (x) − gk (x)| ≥ 2−k } ≤ 2−k .

Let Ek = {x : |gk+1 (x) − gk (x)| ≥ 2−k }, Fk = ∪∞ j=k Ek and note that

µ(Fk ) ≤ 2 −k+1 . If x ∈/ Fk and i > j > m > k then we have that
 j−1  ∞
X  X
|gi (x) − gj (x)| =  gz (x) − gz+1 (x) ≤ |gz (x) − gz+1 (x)| ≤ 2−m+1 .
 
  z=m
z=i

Thus gn (x) is a Cauchy sequence in R and so convergent. Now let F =

∩∞k=1 Fk and note that µ(F ) = limk→∞ µ(Fk ) = 0 and that for x ∈ F we
c

will have that gn (x) is convergent.

Let f (x) = limn→∞ gn (x) if x ∈ / F and set f (x) = 0 if x ∈ F . We have
that gn converges µ a.e. to f so the first part is done. For the second part
let  > 0 and let a > 0. Fix k such that 2−k+1 < min{a, }. Thus

{x : |gk (x) − f (x)| ≥ a} ⊂ {x : |gk (x) − f (x)| ≥ 2−k+1 } ⊂ Fk

So
µ({x : |gk (x) − f (x)| ≥ a}) ≤ µ(Fk ) ≤ 2−k+1 < .

Corollary 7.6. Let (fn ) be a sequence of R-valued, measurable functions,

which is Cauchy in measure. Then (fn ) converges in measure to a R-valued,
measurable function f , which is uniquely determined µ-a.e.
Proof. We know from Theorem 7.5 that there exists a real measurable func-
tion f and a subsequence gk of fn such that gk converges in measure to f .
Let a,  > 0 we can choose N such that for all n, m > N

µ({x : |fn (x) − fm (x)| ≥ a/2}) ≤ /2

and k such that gk = fl with l > N and

µ({x : |gk (x) − f (x)| ≥ a/2}) ≤ /2.

Thus by considering the union of these two sets and using the triangle in-
equality we have that for all n > N

µ({x : |f (x) − fn (x)| ≥ a}) ≤ .

28
Thus fn converges to f in measure. We now need to show f is uniquely
determined µ a.e.. So suppose fn converges to some real valued function g
in measure. Thus for any a,  > 0 we can find N such that for any n ≥ N
the sets

E1 = {x : |f − fn (x)| ≤ a/2} and E2 = {x : |g − fn (x)| ≤ a/2}

satisfy µ((E1 ∩ E2 )c ) ≤ . By the triangle inequality it follows that for any

x ∈ (E1 ∩ E2 ) we have |f (x) − g(x)| ≤ a. Thus for any n ∈ N we have that
µ({x : |f (x) − g(x)| > 1/n} = 0 and so µ({x : f (x) 6= g(x)}) = 0.

Remark. Convergence in measure does not imply convergence in Lp .

Theorem 7.7. Let (fn ) be a sequence of functions in Lp , which converges

to f in measure. Suppose there exists g ∈ Lp such that |fn | ≤ g for all
n ∈ N. Then f ∈ Lp , and fn → f in Lp .

Proof. We prove this by contradiction and by applying Theorem 7.2. Let

 > 0 and suppose there exists a subsequence gk of fn such that kgk −f kp > 
for all k ∈ N. We know that gk converges to f in measure and is thus
Cauchy in measure. So by Theorem 7.5 and Corollary 7.5 it follows that
there exists a subsequence of hl which converges almost everywhere to f .
Thus by Theorem 7.2 it follows that f ∈ Lp and fn → f in Lp .

Definition 7.8. A sequence (fn ) of measurable functions is almost uni-

formly convergent to f ∈ M if for any δ > 0, there exists Eδ ∈ X with
µ(Eδ ) < δ, such that (fn ) is uniformly convergent to f on X \ Eδ .
A sequence (fn ) of measurable functions is almost uniformly Cauchy
if for any δ > 0, there exists Eδ ∈ X with µ(Eδ ) < δ, such that (fn ) is
uniformly convergent on X \ Eδ .

Remark. Almost uniform convergence implies convergence almost every-

where.

Theorem 7.9. Let (fn ) be almost uniformly Cauchy. Then (fn ) is almost
uniformly convergent to a measurable function f .

Proof. For any k ∈ N we can find a set Ek ∈ X for which fn converges

uniformly to some function g on Ekc and µ(Ek ) < 21k . We let Fk = ∪∞
n=k En
c c
and note that since Fk ⊂ Ek we know that fn will converge uniformly to g
1
on Fkc and µ(Fk ) < 2k−1 . Thus we can define a measurable function by

x ∈ Fkc

limn→∞ fn (x) if
gk (x) =
0 if x ∈ Fk

Moreover if we let F = ∩∞ k=1 Fk then µ(F ) = 0. Thus if x ∈

/ F then for k
sufficiently large gk (x) = limn→∞ fn (x). So limk→∞ gk (x) = limn→∞ fn (x)

29
if x ∈ F c and limk→∞ gk (x) = 0 if x ∈ F . So g = limk→∞ gk is a measurable
function for which fn (x) converges to g for µ a.e. x. We need to show this
convergence is almost uniform.
1
Fix δ > 0 and choose k such that 2k−1 < δ. We take the set Fk and note
that µ(Fk ) <  and fn (x) converges uniformly to gk (x) = g(x) on Fkc . The
proof is complete.

Theorem 7.10. Let (fn ) be almost uniformly convergent to f . Then (fn )

converges to f in measure.

Proof. Fix a,  > 0. We can find a set A ∈ X such that fn is uniformly

convergent to f on Ac and µ(A ) < . Thus we can choose N such that for
n≥N
{x ∈ X : |fn (x) − f (x)| ≥ a} ⊂ A
and thus
µ({x ∈ X : |fn (x) − f (x)| ≥ a}) < .

Theorem 7.11. Let (fn ) converge to f in measure. Then there is a subse-

quence that converges almost uniformly to f .

Proof. This can be deduced as an exercise from the proof of Theorem 7.5(Note
that converging in measure implies Cauchy in measure).

Corollary 7.12. Let (fn ) converge to f in Lp . Then there is a subsequence

that converges almost uniformly to f .

Proof. We’ve seen that Lp convergence implies convergence in measure and

so this follows from Theorem 7.11.

Remark. Convergence almost everywhere can be replaced by convergence

in measure in the statements of the Monotone Convergence Theorem and
Lebesgue’s Dominated Convergence Theorem. Similarly, a version of Fatou’s
Lemma holds for convergence in measure. (See Exercise Sheet 6 for details.)

Remark. In general convergence almost everywhere does not imply almost

uniform convergence. The next result shows that this holds if µ(X) < ∞.

Theorem 7.13. (Egorov’s Theorem)

Suppose µ(X) < ∞. Let (fn ) be a sequence of measurable, R-valued func-
tions which converges almost everywhere to a measurable, R-valued function
f . Then (fn ) converges almost uniformly to f .

30
Proof. Fix m, k ∈ N and let

Em,k = {x ∈ X : |fk (x) − f (x)| ≥ m−1 }.

Let Fm,k = ∪∞ ∞
n=k Em,n and Fm = ∩k=1 Fm,k . We know that µ(Fm ) = 0 and
since µ(X) < ∞ we have that limk→∞ µ(Fm,k ) = 0. Now let δ > 0 and
choose km such that µ(Fm,km ) < δ/2k . Let F = ∪∞m=1 Fm,km and observe
that µ(F ) < δ. Moreover if we let  > 0 and choose m > 1 then for all
n ≥ km and x ∈ F c we know that |fn (x) − f (x)| ≤ m−1 < . Thus fn
converges uniformly to f on F c .

8 Radon-Nikodým Theorem
Throughout this section we will let (X, X) be a measurable space.
Recall that ν : X → R is a charge on (X, X) if

(i) ν(∅) = 0, and

(ii) If A1 , A2 , · · · is a sequence of disjoint sets in X,
X∞
then ν(∪∞ n=1 n A ) = ν(An ).
n=1

Definition 8.1. Let ν : X → R be a charge.

• A set P ∈ X is positive with respect to ν if ν(A ∩ P ) ≥ 0, ∀A ∈ X.

• A set N ∈ X is negative with respect to ν if ν(A ∩ N ) ≤ 0, ∀A ∈ X.

• A set M ∈ X is a null set with respect to ν if ν(A ∩ M ) = 0, ∀A ∈ X.

Lemma 8.2. (i) A measurable subset of a positive set is positive.
(ii) The union of two positive sets is positive.
Theorem 8.3. (The Hahn Decomposition Theorem)
Let ν : X → R be a charge. Then there exists positive P ∈ X and negative
N ∈ X such that P ∪ N = X and P ∩ N = ∅.
Proof. We let P ⊂ X denote the set of all positive measurable subsets of X.
Since ∅ ∈ P we know that P is nonempty and we can define

α = sup{ν(A) : A ∈ P}.

Thus we can find a sequence of sets An ∈ P where An ⊂ An+1 for all

n ∈ N and limn→∞ ν(An ) = α. We now let P = ∪∞ n=1 An and note that
ν(P ) = limn→∞ ν(An ) = α which means that α < ∞. To see that P is
positive we let E ∈ X and calculate

ν (E ∩ P ) = ν (E ∩ ∪∞ ∞
n=1 An ) = ν (∪n=1 E ∩ An ) = lim ν(E ∩ An ) ≥ 0.
n→∞

31
 We now let N = X\P and show that this is a negative set by con-
tradiction. Suppose N is not negative which means we can find a set
E ∈ X such that E ⊂ N and ν(E) > 0. E cannot be a positive set since
ν(E ∪ P ) = ν(E) + ν(P ) > α. Therefore we can let

n1 = inf{n ∈ N : there exists A ⊂ E with ν(A) ≤ −n−1 }

and choose E1 ⊂ E such that ν(E1 ) ≤ −n−1 −1

1 . Thus ν(E\E1 ) > ν(E)+n1 >
0. We now proceed inductively by letting
 
nk = inf{n ∈ N : there exists A ⊂ E\ ∪k−1 E
m=1 m with ν(A) ≤ −n−1 }
 
−1
and then choosing Ek ⊂ E\ ∪k−1
m=1 m such that ν(E1 ) ≤ −nk .
E
We will now let F = ∪∞
k=1 Ek and note that since the sets Ek are disjoint

∞
X
ν(F ) = ν (∪∞
k=1 Ek ) ≤− n−1
k < 0.
k=1

Since ν(F ) > −∞ this implies that limk→∞ n−1 k = 0. We know that
ν(E\F ) = ν(E) − ν(F ) > 0 and since ν(E\F ) ∪ P ) = ν(E\F ) + ν(P ) > α
E\F cannot be positive. Therefore we can find A such that A ⊂ E\F
−1 and note that
and ν(A)  < 0. Let k be such that ν(A) ≤ −(nk − 1)
k−1
A ⊂ E\ ∪m=1 Em with ν(A) ≤ −(nk − 1)−1 which contradicts the defini-
tion of nk . Thus N is negative and we have that X = P ∪ N where P is
positive, N is negative and P ∩ N = ∅.

We now show that the Hahn decomposition is essentially unique.

Lemma 8.4. Let P1 , N1 and P2 , N2 be Hahn Decompositions for ν. Then

for any A ∈ X

ν(A ∩ P1 ) = ν(A ∩ P2 ), and

ν(A ∩ N1 ) = ν(A ∩ N2 ).

Proof. We have that

ν(A ∩ (P1 ∪ P2 )) = ν(A ∩ P1 ) + ν(A ∩ (P2 \P1 )) = ν(A ∩ P2 ) + ν(A ∩ (P1 \P2 )).

However P2 \P1 ⊂ N1 and P1 \P2 ⊂ N2 and so ν(A ∩ (P1 \P2 )) = ν(A ∩

(P2 \P1 )) = 0. Therefore

ν(A ∩ P1 ) = ν(A ∩ (P1 ∪ P2 )) = ν(A ∩ P2 ).

32
Definition 8.5. Let ν : X → R be a charge and let P, N be the Hahn
Decomposition for ν. The positive variation of ν is the finite measure ν +
on X defined by ν + (A) = ν(A ∩ P ). The negative variation of ν is the
finite measure ν − defined by ν − (A) = −ν(A ∩ N ). The total variation of
ν is the finite measure |ν| = ν + + ν − .

Remark. Note that ν = ν + − ν − .

Theorem 8.6. (The Jordan Decomposition Theorem)

Let ν : X → R be a charge. If ν = ν1 − ν2 , where ν1 , ν2 are finite measures,
then ν1 (A) ≥ ν + (A) and ν2 (A) ≥ ν − (A) for any A ∈ X.

Proof. We let X = N ∪ P be the Hahn decomposition. We then have for

any A ∈ X
ν1 (A) ≥ ν1 (A ∩ P ) ≥ ν(A ∩ P ) = ν + (A)
and
ν2 (A) ≥ ν2 (A ∩ N ) ≥ −ν(A ∩ N ) = ν − (A).

Theorem 8.7. LetR f ∈ L(X, X, µ), and consider the charge ν : RX → R

given by ν(A)R= A f dµ. Then ν and ν − are given by ν + (A) = A f + dµ
+

and ν − (A) = A f − dµ.

Proof. See exercise sheet.

Remark.
R Let f ∈ M + (X, X). Recall that λ : X → R̄ defined by λ(A) =
A f dµ is a measure, and is absolutely continuous with respect to µ. A
partial converse to this is given by the Radon-Nikodym Theorem. To see
that in general there is no converse let µ be counting measure on (R, B) and
λ Lebesgue measure. We know that if µ(A) = 0 then A = ∅ so λ(A) = 0
which means that λ is absolutely continuous with respect to µ. In this case
we can easily see thatR no such f exists, since if such an f exists then for
any x, 0 = λ({x}) = {0} f dµ and thus f (x) = 0. However this would then
mean that every set has Lebesgue measure 0 which is a contradiction. The
problem here is that counting measure on R is not sigma finite.

Theorem 8.8. (Radon-Nikodým Theorem)

Let λ, µ be σ-finite Rmeasures with λ  µ. Then there exists f ∈ M + (X, X)
such that λ(A) = A f dµ for all A ∈ X. This f is uniquely determined
µ−almost everywhere.

Proof. We first prove this in the case when µ and λ are finite and then
extend to case when they are sigma-finite. So we suppose µ and λ are finite
measures where λ is absolutely continuous with respect to µ. Our aim is to
consider sets A on which for all subsets B ⊂ A λ(A)/µ(A) is close to some
common value. to do this we use the Hahn decomposition.

33
 For c > 0 consider the charge λ − cµ and let P (c) and N (c) be the
positive and negative parts of the Hahn decomposition for this charge. We
now let A(1) = N (c) and define inductively Ak+1 = N (c(k + 1))\(∪ki=1 Ai ).
We can see that the sets Ak are disjoint and

Ak+1 = N (c(k + 1))\(∪kj=1 N (jc)) = N (c(k + 1)) ∩ (∩kj=1 P (cj)).

We now let B := (∪∞ c ∞

k=1 Ak ) = ∩j=1 P (jc). Thus we have that for all k

0 ≤ ckµ(B) ≤ λ(B) < ∞

which means that µ(B) = 0 and thus λ(B) = 0 since λ is absolutely contin-
uous with respect to µ.
For any measurable E ⊂ Ak we have that

λ(E) − ckµ(E) ≤ 0 ≤ λ(E) − c(k − 1)µ(E).

which means if we define fc : X → R by fc (x) = c(k − 1) if x ∈ Ak and

fc (x) = 0 when x ∈ B then for any A ∈ X
Z Z Z
fc dµ ≤ λ(A) ≤ (fc + c)dµ = fc dµ + cµ(A).
A A A

We now use this construction with c = 2−k to create a sequence of functions

fk ∈ M + such that
Z Z Z
−k
fk dµ ≤ λ(A) ≤ (fk + 2 )dµ = fc dµ + 2−k µ(A). (1)
A

Thus for positive integers m ≥ n we have that for any A ∈ X

Z Z 
fm dµ ≤ 2−n µ(A).
 
 fn dµ −

A A

By taking sets {x : fn (x) ≥ fm (x)} and {x : fm (x) ≥ fn (x)} respectively we

can see that Z
|fn − fm |dµ ≤ 2−n+1 µ(X).

Note that each function fn ∈ L1 (µ) and so fn converges in L1 (µ) to some

function f ∈ L1 (µ) and thus fn also converges in measure to f . Since each
fn ∈ M + and by Theorem 7.5 fn must converge to f almost everywhere
along a subsequence we can deduce that f ∈ M + . Furthermore for any
E ∈ X we have that
Z Z  Z Z
 
 fn dµ − f dµ ≤
 |fn − f |dµ ≤ |fn − f |dµ → 0 as n → ∞.

E E E
R
Thus
R by equation (1) we can conclude that λ(E) = limk→∞ E fk dµ =
E f dµ. So we’ve shown the existence of f in the case when µ and λ are
finite.

34
 To show that f is uniquely defined a.e. we suppose there exists g such
that for all E ∈ X Z
λ(E) = gdµ.
E
If we let A = {x : g(x) > f (x)}
Z Z Z
|f − g|dµ = (f − g)dµ + (g − f )dµ = 0
Ac A
and therefore f = g µ a.e..
Finally we turn to the case where µ and λ are sigma finite. Let {An }n∈N
be a sequence of sets such that for each n, λ(An ) < ∞, µ(An ) < ∞ and
An ⊂ An+1 . For each n we can useRthe finite case result to find fn ∈ M +
such that for all E ⊂ An , λ(E) = E fn dµ and f (x) = 0 for x ∈ Acn . So
fn is a monotone increasing sequence of functions so we can let f (x) =
limn→∞ fn (x). By the monotone convergence theorem for all A ∈ X
Z
∞
λ(A) = λ(∪n=1 A ∩ An ) = lim λ(A ∩ An ) = lim fn dµ.
n→∞ n→∞ A

Remark. f is called the Radon-Nikodym derivative of λ with respect to µ

dλ
and is denoted by dµ .
Definition 8.9. Two measures λ and µ on X are mutually singular if
∃A, B ∈ X with A ∪ B = X, A ∩ B = ∅ and λ(A) = µ(B) = 0. We
write λ ⊥ µ.
Theorem 8.10. (Lebesgue Decomposition Theorem)
Let λ, µ be σ-finite measures on X. Then there exists measures λ1 , λ2 such
that λ1 ⊥ µ, λ2  µ and λ = λ1 + λ2 . The measures λ1 , λ2 are unique.
Proof. Let ν = λ+µ. Thus ν is sigma finite and both λ and µ are absolutely
continuous with respect to ν. We can now use the Radon Nikodym Theorem
to see there exists f, g ∈ M + such that for all A ∈ X
Z Z
µ(A) = f dν and λ(A) = gdν.
A A

Take the set B = {x : f (x) = 0} and note that µ(B) = 0. We define λ1

and λ2 to be the measures such that
λ1 (A) = λ(A ∩ B) and λ2 (A) = λ(A ∩ B c ).
λ1 (X\B) = 0 so we can see it is singular with respect to µ. To show λ2 is
absolutely continuous with respect to µ we let A ∈ X satisfy µ(A) = 0. We
then have that f (x) = 0 for ν almost all x ∈ A and thus for λ almost all
x. Therefore λ1 (A) = λ(A ∩ B) = 0 and so λ1 is absolutely continuous with
respect to µ.

35
8.1 Linear Functionals
Definition 8.11. A linear functional on Lp (X, X, µ) is a map
G : Lp (X, X, µ) → R such that ∀a, b ∈ R, ∀f, g ∈ Lp ,
G(af + bg) = aG(f ) + bG(g).
A linear functional G is bounded if ∃M ∈ R such that ∀f ∈ Lp ,
|G(f )| ≤ M kf kp .
The norm of G is the real number kGk := sup{|G(f )| : f ∈ Lp , kf kp ≤ 1}.

Theorem 8.12. (1) Let q ≥ 1 and let gR ∈ Lq (X, X, µ).

q
Define G : Lp (X, X, µ) → R by G(f ) = f gdµ for p = q−1 > 1. Then G is
a bounded linear functional, and kGk = kgkq .
(2) Suppose µ is σ-finite and let q = ∞,R p = 1.
Define G : L1 (X, X, µ) → R by G(f ) = f gdµ. Then G is a bounded linear
functional, and kGk = kgk∞ .

Proof. (1) It is straightforward to see that G is linear. Furthermore for any

f ∈ Lp we have that by Hölder’s inequality
Z 
 
|G(f )| =  f gdµ ≤ kf kp kgkq .

So G is bounded and kGk ≤ kgkq . Now let h(x) = sgn(g(x))|g(x)|q−1 and

q/p
observe that since p(q − 1) = q, h ∈ Lp , khkp = kgkq = kgkqq−1 and
Z Z
G(h) = hgdµ = |g|q dµ = kgkqq = kgkq khkp .

Thus kGk = kgkq .

(2) Again it is straightforward to see that G is a bounded linear functional
and kGk ≤ kgk∞ . We now fix  > 0 and choose X = {x : g(x) ≥ kgk∞ − }.
We now µ(X ) > 0 and since µ is sigma finite we can find Y ⊂ X such
that 0 < µ(Y ) < ∞. We now let h = χY sgn(g(x)) which is in L1 and
khk1 = µ(Y ). We then have
Z
G(h) = hgdµ ≥ (kgk∞ − )µ(Y ) = (kgk∞ − )khk1 .

Definition 8.13. A linear functional G is positive if f ≥ 0 =⇒ G(f ) ≥ 0.

Lemma 8.14. Let G be a bounded linear functional on Lp (X, X, µ). Then

there exists bounded, positive linear functionals G+ and G− on Lp (X, X, µ)
such that G(f ) = G+ (f ) + G− (f ), ∀f ∈ Lp (X, X, µ).

36
Theorem 8.15. (Riesz Representation Theorem for L1 )
Let (X, X, µ) be a σ-finite measure space, and let G be a bounded linear
R on L1 (X, X, µ). Then there exists g ∈ L∞ (X, X, µ) such that
functional
G(f ) = f gdµ, ∀f ∈ L1 (X, X, µ).
We have kGk = kgk∞ . Moreover, if G is positive, then g ≥ 0.

Proof. ‘Outline the details will be filled in via a problem sheet’. We start
with the finite case, assume that G is a positive bounded linear functional
and define λ : X → R by λ(E) = G(χE ). We then show that λ is a
finite measure and is absolutely continuous. We can then apply the R Radon-
+
Nikodym derivative to find g ∈ M such that G(χE ) = λ(E) = gdµ. R
Thus it follows for all simple functions φ ∈ L1 that G(φ) = φgdµ.
We then use the Monotone convergence theorem to extend this to all non-
negative functions in L1 and finally use linearity to extend it to all functions
in L1 .
To extend this to σ measure spaces we work on sets An where ∪∞ n=1 An =
X, µ(An ) < ∞ and An ⊂ An+1 for all n ∈ N. This can be readily seen to
yield the result for non-negative functions (Monotone Convergence Theo-
rem) and positive functionals. Finally for general bounded linear functionals
we use Lemma 8.14.

For the case when 1 < p < ∞ we can relax the condition that the
measure space is sigma finite.

Theorem 8.16. (Riesz Representation Theorem for Lp )

Let (X, X, µ) be any measure space, and let G be a bounded linear functional
p
on Lp (X, X, µ), 1 < p < ∞. Then there exists g ∈ Lq (X, X, µ), q = p−1 ,
R
such that G(f ) = f gdµ, ∀f ∈ Lp (X, X, µ).
We have kGk = kgkq .

Proof. For a sigma-finite measure the proof for L1 can be adapted without
difficulty. To see how it may be extended to the non-sigma finite case please
consult Bartle (NB: However nearly all measure theory is done with the
assumption of the measure space being sigma finite and in my opinion it is
the sigma finite case which is important to understand.)

9 Constructing measures and unmeasurable sets

In this section we show how Lebesgue measure can properly be defined and
in doing this give a method which can be used to generate many different
measures. We’ll also show at the end how to construct sets which are not
Lebesgue measurable.

Definition 9.1. Let X be a space we define an algebra to be a set A of

subsets of X which satisfies that

37
 1. ∅, X ∈ A

2. If Y ∈ A then Y c ∈ A.

3. If A1 , . . . , An ∈ A then ∪nk=1 Ak ∈ A

Definition 9.2. A premeasure on an algebra of subsets of X is a function

µ : X → R such that

1. µ(∅) = 0

2. µ(A) ≥ 0 for all A ∈ A

3. If A1 , A2 , . . . ∈ A, the sets Ai are pairwise disjoint and ∪∞

i=1 Ai ∈ A
then
∞
X
∞
µ (∪i=1 Ai ) = µ(Ai ).
i=1

Lemma 9.3. The collection F of all finite unions of sets of the form (a, b],(−∞, b),
[b, ∞) and (−∞, ∞) is an algebra on R and the length gives a premeasure
on these sets.

Proof. To show that these sets form an algebra and that length satisfies the
first two properties to be a premeasure is straight forward. So we need to
show that length is countably subadditive on F. To prove (iii) we just need
to show that if one of the sets of the form (a, b], (a, ∞), (−∞, b], (−∞, ∞)
is a disjoint countable union of other sets of this form then length adds up
countably. We do this for (a, b] and note that other sets can be dealt with
similarly.
Suppose we have a sequence of disjoint intervals (a1 , b1 ], (a2 , b2 ], . . . such
that (a, b] = ∪∞n=1 (an , bn ] and an < bn < an+1 for all n ∈ N. For any k ∈ N
we have that ∪kn=1 (an , bn ] ⊂ (a, b] and thus

  Xk
l ∪kn=1 (an , bn ] = bn − an ≤ b − a.
n=1

This holds for any k so we must have

l (∪∞
n=1 (an , bn ]) < b − a.
P∞
On the other hand let  > 0 and j > 0 satisfy that j=1 j < /2. We
then have that
[a, b] ⊂ ∪∞
j=1 (aj − /2, bj + /2)

and since [a, b] is compact we can do this with finitely many intervals. So
we can find k such that

[a, b) ⊂ ∪kj=1 (aj − j /, bj + j )

38
and thus
k
X
b−a≤ (bj − aj ) + 
j=1

and the result follows.

We now want to show that we can extend premeasures on an algebra to

give a measure on a suitable sigma algebra.
Definition 9.4. If µ is a premeasure defined on an algebra A we can define
µ∗ : P (X) → R by
 
X∞ 
µ∗ (B) = inf µ(Aj ) : Aj ∈ A and B ⊆ ∪∞ A
j=1 j .
 
j=1

It can be shown that µ∗ has several properties similar to that of a measure

and it is called an outer measure, however it is not in general a measure.
Lemma 9.5. (Properties of the outer measure:)
(a) µ∗ (∅) = 0,
(b) µ∗ (B) ≥ 0, ∀B ⊆ X,
(c) A ⊆ B =⇒ µ∗ (A) ≤ µ∗ (B),
(d) B ∈ A =⇒ µ∗ (B) = µ(B), and
∞
X
(e) If B1 , B2 , · · · is a sequence of subsets of X, then µ∗ (∪∞
k=1 Bk ) ≤ µ∗ (Bk ).
k=1

Proof. Parts (a), (b) and (c) all follow immediately from the definition. For
part d for B ∈ A write B = B ∪ ∅ ∪ ∅ ∪ · · · to see that µ∗ (B) ≤ µ(B).
P∞ let Aj ∈ A be a family of
Now sets such that B ⊆ ∪∞j=1 Aj . We know that
∗
j=1 µ(Aj ) ≥ µ(B) and so µ (B) ≥ µ(B).
Finally to establish (e), we let  > 0 and note that for
∞
P∞each∗ n, k ∈ N
we can find An,k ∈ A such that Bn ⊆ ∪k=1 An,k and k=1 µ (An,k ) ≤
µ∗ (Bn ) + /2n . So ∪∞
n=1 B n ⊆ ∪n,k∈N A n,k and
X ∞
X
∗
µ (An,k ) ≤ µ∗ (Bn ) + .
n,k∈N n=1
P∞
Therefore it follows that µ∗ (∪∞
k=1 Bk ) ≤ k=1 µ
∗ (B
k ).

Definition 9.6. For an outer measure µ∗ we say that a set E is µ∗ mea-

surable if for all A ⊂ X we have that
µ∗ (A) = µ∗ (A ∩ E) + µ∗ (A ∩ E c ).
We call the set of all µ∗ measurable sets A∗ .

39
Theorem 9.7. (Caratheodory’s Extension Theorem)
(i) The collection A∗ of all µ∗ −measurable sets is a σ−algebra containing
A.
∗ ∗ ∞
(ii)
P∞If A 1 , A2 , · · · is a sequence of disjoint sets in A , then µ (∪n=1 An ) =
∗
n=1 µ (An ).

Proof. ’Outline’. Step 1 We first need to show that A∗ is an algebra and

that µ∗ is additive on A∗ . It is easy to show X, ∅ ∈ A∗ and to show that A∗
is closed under taking complements. We then show it is closed under taking
finite intersections which is enough to show it is an algebra. Finally taking
E, F ∈ A∗ with E ∩ F = ∅ it can be shown that µ∗ (E ∪ F ) = µ∗ (E) + µ∗ (F ).
Step 2 We show that A∗ is in fact a sigma algebra and µ∗ is a measure
on A∗ . Let {En }n∈N be a disjoint collection of sets in A∗ . If we let Fn =
∪nk=1 Ek then it follows from step 1. that Fn ∈ A∗ . Now fix A ⊂ X and
let E = ∪∞ k=1 Ek . By parts (c) and (e) of Lemma 9.5 it follows that for any
n∈N
∞
X ∞
X
∗ ∗ ∗ ∗ ∗
µ (A∩E)+µ (A∩E ) ≤ c
µ (A∩Ek )+µ (A∩Fnc ) = µ (A)+ µ∗ (A∩Ek )
k=1 k=n+1
(2)
where we have used that µ∗ (A) = µ∗ (A∩Fn )+µ∗ (A∩Fnc ). Thus if µ∗ (A) < ∞
then by letting n → ∞ we can see that µ∗ (A ∩ E) + µ∗ (A ∩ E c ) ≤ µ∗ (A) and
if µ ∗ (A) = ∞ then the result is trivial. On the other hand we have that by
part (e) of Lemma 9.5
µ∗ (A) ≤ µ∗ (A ∩ E) + µ∗ (A ∩ E c ).
Thus we have shown that E ∈ A∗ . To see that µ∗ is countably additive on
A∗ take A = E in (2) to get
∞
X ∞
X
µ∗ (E) ≤ µ∗ (Ek ) ≤ µ∗ (E) + µ∗ (Ek )
k=1 k=n+1

and letting n → ∞ yields the result.

Step 3 We show that A ⊂ A∗ . Let A ∈ A (we need to show that E ∈ A∗ )
and E ⊂ X. By part (e) of Lemma 9.5 we have that
µ∗ (E) ≤ µ∗ (E ∩ A) + µ∗ (E ∩ Ac ).
On the
P∞ other hand let  > 0 and Fn ∈ A be a sequence such that E ⊂ ∪∞
n=1 Fn
and n=1 µ∗ (Fn ) ≤ µ∗ (E) + . Since each Fn ∈ A we have that
µ∗ (Fn ) = µ∗ (Fn ∩ A) + µ∗ (Fn ∩ Ac ).
So
∞
X ∞
X
µ∗ (E) ≥ µ∗ (Fn )+ = µ∗ (Fn ∩A)+µ∗ (Fn ∩Ac )+ ≥ µ∗ (E∩A)+µ∗ (E∩Ac )+.
n=1 n=1

40
Thus letting  → 0 yields that

µ∗ (A) = µ∗ (A ∩ E) + µ∗ (A ∩ E c )

and the proof is complete.

Remark. (X, A∗ , µ∗ ) is a complete measure space. Recall that this means

for any A ∈ A∗ with µ∗ (A) = 0 all subsets of A are in A∗ .

Theorem 9.8. (Hahn’s Uniqueness Theorem)

Let µ be a σ−finite measure on A. Then the extension of µ to a measure on
A∗ is unique.

Proof. We first suppose ν is finite a measure on A∗ such that ν(A) = µ(A)

for all A ∈ A∗ . We use the definition of µ∗ to show for all E ⊂ A∗ that
ν(E) = µ∗ (E). We then extend this to the sigma finite case. In the case
where µ is finite, let E ⊂ AP∗ , for  > 0 we can find sets E ∈ A such that
n
∞
E ⊆ ∪∞ n=1 En and µ ∗ (E) ≥
n=1 µ ∗ (E ) − . Thus
n

∞
X ∞
X
ν(E) ≤ ν(En ) = µ∗ (En ) ≤ µ∗ (E) + .
n=1 n=1

This is using that ν and µ∗ must agree on elements of A and since it holds
for any  > 0 it follows that ν(E) ≤ µ∗ (E). However this argument works
equally well for E c and we have since X ∈ A

µ∗ (X) = µ∗ (E) + µ∗ (E C ) ≥ ν(E) + ν(E c ) = ν(X).

Since µ(X) < ∞ this must mean ν(E) = µ∗ (E).

We can extend this to the sigma finite case in the usual way.

Definition 9.9. F∗ is the σ−algebra of Lebesgue measurable sets, and

l∗ is the Lebesgue measure.

Remark. Recall that B is the smallest σ−algebra containing F, and hence

B ⊆ F∗ . The restriction of l∗ to B is the Borel measure (also called
Lebesgue measure.) It is possible to find a set which is not Lebesgue mea-
surable.

Remark. (i) B is a proper subset of F∗ .

(ii) Not every subset of R is Lebesgue measurable.

The following procedure constructs a subset of R which is not Lebesgue

measurable (It is due to Giuseppe Vitali and was done in 1905). For x, y ∈ R
we say x ∼ y if and only if x − y ∈ Q. This gives an equivalence relation.
We then consider the equivalence classes x + Q and take V ⊂ (0, 1) to be a
set which contains exactly one element in each equivalence class x + Q (we

41
can do this by the axiom of choice, if you haven’t heard about this before
look it up).
We can then take rn to be an enumeration of the rationals in (−1, 1) and
consider the sets V + rn for each n ∈ N. since Lebesgue measure is invariant
under translation if V was measurable we would have that λ(V +rn ) = λ(V ).
However we can see that (0, 1) ⊂ ∪n∈N (V + rn ) ⊂ (−1, 2) which would
contradict either λ(V ) = 0 or λ(V ) > 0 so V cannot be measurable.

Exercise sheet 1

1. Show that if f : R2 → R is a continuous function and A ⊂ R is an

open set then f −1 (A) = {(x, y) ∈ R2 : f (x, y) ∈ A} is an open set in
R2 (take the usual Euclidean metric on R2 ).

2. Show that
T∞
(i) [a, b] = n=1 (a − 1/n, b + 1/n);
S∞
(ii) (a, b) = n=1 [a + 1/n, b − 1/n].

Conclude that any σ-algebra of subsets of R which contains all open

(resp. closed) intervals also contains all closed (resp. open) intervals.

3. Show that every open set A ⊂ R can be written as a countable union

of open intervals.

4. Show that the Borel σ-algebra B is generated by the collection of half-

open intervals (a, b]. Show that it is also generated by the collection
of half-rays (a, ∞), a ∈ R.

5. Let f : (X, X) → R (or R). Show that the following are equivalent.

(i) f is X-measurable;
(ii) {f > q} ∈ X for each q ∈ Q;
(iii) {f ≤ q} ∈ X for each q ∈ Q.

6. Give an example of an R-valued function f on some measurable space

(X, X) which is not X-measurable, but is such that f 2 is X-measurable.

7. Consider the Borel σ-algebra (R, B). Show that any monotone function
f : R → R is Borel measurable.

8. Let x ∈ (0, 1] have the expansion x = 0.x1 x2 x3 . . . in base 2, the non-

terminating expansion being used in cases of ambiguity. Show that
fn (x) = xn is a Borel measurable function of x for each n.

42
 9. Let f : X → R be a measurable function on (X, X) and A > 0. The
truncation fA of f is defined by

 f (ω), if |f (ω)| ≤ A,
fA (ω) = A, if f (ω) > A,
−A, if f (ω) < −A.


Show that fA is measurable.

10. Let f be a non-negative measurable function on (X, X) which is bounded

(so 0 ≤ f (ω) ≤ K for all ω ∈ X). Show that the sequence of sim-
ple measurable functions ϕn constructed in the lecture converges to f
uniformly on X.

11. Let (X, X) be a measurable space, and let f : X → R be a function.

Show that f is measurable if and only if f −1 (G) is measurable for
each open set G ⊂ R.

C1 Show that the set of all x ∈ [0, 1] which are normal base 2 is Borel
measurable (if you don’t know what normal base 2 means look it up).

Exercise sheet 2

1. Consider the measure space (R, B, λ) where λ is Lebesgue measure.

For any x ∈ R show that λ({x}) = 0. Thus show that any countable
set X ⊂ R satisfies λ(X) = 0.

2. Again let λ denote Lebesgue measure. Show the following.

a) If E is a non-empty open subset of R then λ(E) > 0.

a) If K is a compact subset of R then λ(K) < +∞. (Use the Heine-
Borel theorem)

3. Find a set with positive Lebesgue measure but which contains no non-
void open interval.

4. Let (X, X, µ) be a measure space. Show that if for any A ∈ X we

define µA : X → R by µA (B) = µ(A ∩ B) then µA is a measure.

5. Let (X, X) be a measurable space, µ1 , µ2 , . . . , µn be measuresPon (X, X)

and a1 , a2 , . . . , an be positive real numbers. Show that ν = nk=1 ak µk
is a measure on (X, X). Thus prove Lemma 4.5 from the lectures. That
+
is that for any measure µ on (X, X) and R simple function φ ∈ M (X, X)
we can define a measure by ν(A) = A φdµ for all A ∈ X.

6. Let X = N and X be the family of all subsets of X. If E is finite, let

µ(E) = 0; if E is infinite, let µ(E) = +∞. Is µ a measure on X?

43
 7. Let (An ) be a sequence of sets in X. Put
∞ [
\ ∞ ∞ \
[ ∞
A= [ An ], B= [ An ].
m=1 n=m m=1 n=m

The set A is called the limit superior of the sets (An ) and denoted
by lim sup An . The set B is called the limit inferior of the sets (An )
and denoted by lim inf An . A consists of all those x ∈ X that belong
to infinitely many of the sets An while B consists of all those x ∈ X
that belong to all but finitely many of the sets An .
Let (X, X, µ) be a measure space and let (An ) be a sequence in X.
Show that
a) µ(lim inf An ) ≤ lim inf µ(An );
and, if µ(∪An ) < +∞,
b) lim sup µ(An ) ≤ µ(lim sup An ).
Show that the inequality in b) may fail if µ(∪An ) = +∞.
C2 Let (X, X, µ) be a measure space where µ(X) = 1. Show that
a If A1 , A2 , . . . ∈ X and ∞
P
n=1 µ(An ) < ∞ then

µ({x ∈ X : x ∈ An for infinitely many n ∈ N} = 0.

Q
. ∈ X, µ(∩i∈I Ai ) = i∈I µ(Ai ) for all finite subsets
b If A1 , A2 , . .P
I ⊂ N and ∞ n=1 µ(An ) = ∞ then

µ({x ∈ X : x ∈ An for infinitely many n ∈ N} = 1.

It may help to use the inequality that for x ≥ 0, 1 − x ≤ e−x .
C3 Show that the set of numbers which are weakly normal base 2 in [0, 1]
(i.e the numbers for which the digit 0 occurs with frequency 1/2 in the
binary expansion) has Lebesgue measure 1. It’ll help to use question
C2. and Stirling’s formula (if you have not seen this before look it up).
Exercise sheet 3
Notation: (X, X, µ) will denote a measure space and (R, B, λ) will be the
specific measure space of the real numbers with the Borel sigma algebra and
Lebesgue measure.
1. Working on (R, B, λ) let fn = (1/n)χ[0,n] and f = 0. Show that the
sequence (fn ) converges uniformly to f , but that
Z Z
f dλ 6= lim fn dλ.

Why does this not contradict the Monotone Convergence Theorem?

Does Fatou’s Lemma apply?

44
 2. Show that if (X, X, µ) is a finite measure space (i.e. µ(X) < ∞), and if
(fn ) is a real-valued sequence in M + (X, X) which converges uniformly
to a function f , then f belongs to M + (X, X) and
Z Z
f dµ = lim fn dµ.

3. Let An ∈ X and Bn = Acn . suppose that for Rall n An ⊆ An+1 and

∪∞ +
n=1 An = X.R If f ∈ M (X, X) satisfies that f dµ < ∞ then show
that limn→∞ Bn f dµ = 0 (Hint: Use Theorem 4.13). Is this true if we
R
don’t assume that f dµ < ∞.

4. Let X be a finite closed interval [a, b] in R, B the collection of Borel

sets in X and λ be Lebesgue measure on X. Let f be a non-negative
continuous function on X. Show that
Z Z b
f dλ = f (x)dx
a

where the right side denotes the Riemann integral.

5. If f ∈ M + and f dµ < +∞ then µ({f = +∞}) = 0.

R

6. Let fn ∈ M + , lim fn = f and fn ≤ f . Show that f dµ = lim fn dµ.

R R

7. Let (fn ) be a sequence of non-negative R-valued measurable functions

such that fn ↓ f .R R R
(a) Show that ifR fk dµ < +∞ for some k, then lim f
R n dµ = f dµ.
(b) Show that fk dµ = +∞ for all k does not imply f dµ = ∞.

c4 Suppose that (fn ) ⊆ M + , that (fn ) converges to f , and that

Z Z
f dµ = lim fn dµ < +∞.

Prove that Z Z
f dµ = lim fn dµ
E E
for each E ∈ X. Hint. It mayThelp to consider An := {fn ≥ αf } (for
some 1 < α < ∞) and Bn := m≥n Acn .

Exercise sheet 4

NB: Throughout this sheet (X, X, µ) refers to a measure space.

1. Let f ∈ L(X, X, µ) and a > 0. Show that {x : |f (x)| ≥ a} has finite
measure. Show that {x : f (x) 6= 0} has σ-finite measure (i.e for each
n ∈ N there exists An ∈ X such that ∪n∈N An = {x ∈ X : f (x) 6= 0}
and µ(An ) < ∞.)

45
2. Let f ∈ L(X, X, µ) and ε > 0. Show that there exists an X-measurable
simple function ϕ such that
Z
|f − ϕ| dµ < ε.

3. Let fn ∈ L such that fn → f ∈ L. Suppose that

Z
lim |fn − f | dµ = 0.
n→∞

Show that Z Z
|f | dµ = lim |fn | dµ.

4. Let f ∈ L(X, X, µ) and g : X → R is a measurable function such

that
R f =R g for µ-almost all x. Show that g ∈ L(X, X, µ) and that
gdµ = f dµ.
√
R n x
5. Show that limn→∞ [0,1] fn dλ = 0 where fn (x) := 1+n2 x2 .

1
R
6. Show that for the function f (x) := 1+x2
you have Rf dλ = π.

7. Let n ≥ 2, x ≥ 0. Prove that (1 + x/n)n ≥ (1/4)x2 . and deduce that

x → (1 + x/n)−n is Lebesgue-integrable over [1, ∞) and that
Z
x
lim (1 + )−n dλ = e−1 .
n→∞ [1,∞) n
Ry
c5 Let f : (0, ∞) → R be a continuous function such that limy→∞ 0 f (x)dx
exists and is finite. Show that it is possible that f ∈
/ L((0, ∞), B, λ).

Exercise sheet 5

1. Let (X, X, µ) be a finite measure space. If f is X-measurable, let

En = {x : (n − 1) ≤ |f (x)| < n}. Show that f ∈ L1 if and only if
∞
X
nµ(En ) < +∞.
n=1

2. Let X = N and let X be the collection of all subsets of N. Let λ be

defined by X
λ(E) = (1/n2 ), E ∈ X.
n∈E
√
(a) Show that f : X → R, f (n) = n satisfies f ∈ Lp if and only if
1 ≤ p < 2.

46
 (b) Find a function f such that f ∈ Lp if and only if 1 ≤ p ≤ p0 .

3. Let (X, X, µ) be a finite measure space and let f ∈ Lp . Show that

f ∈ Lr for any 1 ≤ r < p with
1
− p1
kf kr ≤ µ(X) r kf kp .

4. Suppose that X = N and µ is the counting measure on X. Show that

f ∈ Lp implies f ∈ Ls for 1 ≤ p ≤ s < ∞.

5. Let (X, X, µ) be a measure space and suppose f ∈ Lp1 and f ∈ Lp2

with 1 ≤ p1 < p2 < ∞. Prove that f ∈ Lp for any p with p1 ≤ p ≤ p2 .
(Hint: Prove Corollary 6.11 from the lecture notes)

6. Let f ∈ Lp , 1 ≤ p ≤ ∞ and g ∈ L∞ . Show that f g ∈ Lp and

kf gkp ≤ kf kp kgk∞ .

7. (a) Prove that the space L∞ (X, X, µ) is contained in L1 (X, X, µ) if

and only if µ(X) < ∞.
(b) Show that if µ(X) = 1 and f ∈ L∞ , then

kf k∞ = lim kf kp .
p→∞

8. Let 1 ≤ p < ∞, f : X → R be a measurable function and fn : X → R

be a sequence of measurable functions such that limn→∞ fn (x) = f (x)
for every x ∈ X and there exists g ∈ Lp such that |fn (x)| ≤ g(x) for
all N ∈ N and x ∈ X. Show that f ∈ LP and limn→∞ kfn − f kp = 0
(i.e fn tends to f in Lp .)

9 This is an additional question on the Dominated Convergence Theo-

rem. In the measure space ((0, ∞), B, λ) show that

cos(x5 )
Z
lim dλ = 0.
n→∞ 1 + nx2

c6 A function φ : R → R is called convex if for every x1 , x2 ∈ R and

t ∈ [0, 1] we have that φ(tx1 + (1 − t)x2 ) ≤ tφ(x1 ) + (1 − t)φ(x2 ). Prove

(a) If φ is convex then for all x ∈ R

φ(x) = sup{ψ(x) : ψ is affine and ψ(y) ≤ φ(y) for all y ∈ R}.

(b) Jensen’s inequality that if (X, X, µ) is a probability space, f ∈

L(X, X, µ) and φ : R → R is convex then
Z  Z
φ f dµ ≤ φ(f )dµ.

47
 (c) Hölders inequality using Jensen’s inequality. For g ∈ Lq it may
help to assume
R that g > 0, consider the measure ν defined by
1
ν(A) = kgkq A g dµ and h = |f |/g q−1 .
q
q

Exercise sheet 6

1. Let (X, X, P) be a probability space (i.e. P(X) = 1). Let Y : X → R

be an X-measurable random variable. Assume that Y ∈ L2 (X, X, P).
Define Z
µ = E[Y ] = Y dP (expectation of Y ),

σ 2 = E[(Y − µ)2 ] (variance of Y )

Prove Chebychev’s inequality,

P({x ∈ X : |Y (x) − µ| ≥ α}) ≤ (σ/α)2

for any α > 0.

2. Consider the measure space ([0, 1], B, λ). Give an example of a se-
quence of functions (fn ) in L1 and a function f ∈ L1 where fn con-
verges to f in L1 but for which fn is not almost everywhere convergent
to f .

3. Let (fn ), (gn ) be R-valued X-measurable functions which converge in

measure to X-measurable functions f and g respectively. Prove that

a) (a fn + b gn ) converges in measure to a f + b g for any a, b ∈ R;

b) (|fn |) converges to |f | in measure.

4. Prove the following versions of Fatou’s Lemma. Assume fn , f ∈ M+

(n ∈ N).
R R
i) If fn → f µ-a.e. then f dµ ≤ lim inf fn dµ.
R R
ii) If fn → f in measure then f dµ ≤ lim inf fn dµ.

5. Show that Lebesgue’s Dominated Convergence Theorem holds if al-

most everywhere convergence is replaced by convergence in measure.

6. Let (X, X, µ) be a finite measure space. If f is an X-measurable func-

tion, put
|f |
Z
r(f ) = dµ.
1 + |f |
Show that a sequence (fn ) of X-measurable R-valued functions con-
verges in measure to f if and only if r(fn − f ) → 0.

48
7. Let (fn ) be a sequences of R-valued X-measurable functions which
converge in measure to a X-measurable function f . Show that (fn ) is
Cauchy in measure.

8. Give an example to show that convergence a.e does not always imply
almost uniform convergence (i.e. show that the assumption µ(X) < ∞
is necessary in Egorov’s Theorem).

c7 Prove Vitali’s Convergence Theorem. This states that for 1 ≤ p < ∞

a sequence {fn }n∈N in Lp is convergent if and only if the following
three criteria are satisfied.

(a) fn is Cauchy in measure.

(b) For each  > 0 there exists E ∈ X such that µ(E ) < ∞ such
that if F ∈ X and F ∩ E = ∅ then
Z
|fn |p dµ < p for all n ∈ N.
F

(c) For each  > 0 there is a δ > 0 such that if E ∈ X and µ(E) < δ()
then Z
|fn |p dµ < p for all n ∈ N.
E

Exercise sheet 7

1. Let ν be a charge on (X, X).

a) Let X 3 En ↑ E (this means for all n, En ⊂ En+1 ). Show that

ν(E) = lim ν(En ).
b) Let X 3 Fn ↓ F (this means for all n, Fn ⊃ Fn+1 ). Show that
ν(F ) = lim ν(Fn ).

2. Let ν be a charge on (X, X). Show that

a) ν + (E) = sup{ν(F ) : X 3 F ⊆ E},

b) ν − (E) = − inf{ν(F ) : X 3 F ⊆ E}.

3. Let (X, X, µ) be a measure spaceR and f ∈ L(X, X, µ). Let ν : X → R

be the charge given by ν(A) = A f dµ. Show that a set E is null with
respect to ν if and only if µ(E ∩ {x ∈ X : f (x) 6= 0}) = 0.

4. Prove Theorem 8.7 from the lecture notes. That isR if f ∈ L(X, X, µ)
and ν : X → R is the charge given by ν(A) = A f dµ then show
that
R +the positive and
R negative variations of ν, are given by ν + (A) =
− −
A f dµ and ν = A f dµ respectively.

49
 2
5. Let ν(E) = E xe−x dλ (E ∈ B, λ is Lebesgue measure). Give a Hahn
R

decomposition of R with respect to ν.

dν
6. Let ν, µ be σ-finite measures on (X, X) with ν << µ. Let f = dµ ∈
M + . Show that for any g ∈ M + ,
Z Z
g dν = gf dµ.

Hint: Apply Monotone Convergence Theorem to simple functions.

7. Let ν, λ, µ be σ-finite measures on (X, X) with ν << λ and λ << µ.
Show that ν << µ and
dν dν dλ
= µ-a.e.
dµ dλ dµ

8. Let  √
1 − x, x ≤ 1
f (x) =
0, x>1
and
x2 , x ≤ 0

g(x) =
0, x > 0
Let Z Z
ν(E) = f dλ and µ(E) = g dλ (E ∈ B)
E E
Find the Lebesgue decomposition of ν with respect to µ.
9. Let (X, X0 , µ) be a probability space, f : X → R an integrable function
and X ⊂ X0 a sigma algebra. Show that there exists a function g which
is integrable with respect
R to the
R measurable space (X, X) and for which
any A ∈ X satisfies A f dµ = A gdµ. (This function (random variable)
is known as the conditional expectation).

Exercise sheet 8

This sheet goes through the proof of the Riesz representation theorem for
p = 1. Suppose that (X, X, µ) is a measure space where µ is sigma finite. For
questions 1,2 and 3 suppose that µ is finite. For questions 1 to 4 G : L1 → R
is a positive bounded linear functional.
1. Show that if we define λ : X → R by
λ(A) = G(χA )
then λ is a measure which is absolutely continuous with respect to µ.
(Hint: The hardest Ppart is to show that for disjoint sets An ∈ X we
∞
have λ(∪∞ A
n=1 n ) = n=1 λ(An ) use Corollary 7.3 and the functions
χ∪k An ).
n=1

50
2. Use the Radon-Nikodým Theorem to show that for all measurable
simple functions φ : X → R there exists a nonnegative measurable
function g : X → R such that
Z
G(φ) = φgdµ.

3. Let f ∈ L1 be nonnegative and φn a monotone increasing sequences

of simple functions which converge almost everywhere and in L1 to f .
Show that Z
G(f ) = f gdµ

and that this can be extended to all L1 functions by the linearity of

G.

4. Show that the previous result is still true if we suppose that µ is a

sigma finite measure.

5. Use Lemma 8.14 and the previous results to extend these results to
show that for all bounded
R linear functionals G : L1 → R there exists
g ∈ L∞ with G(f ) = gf dµ for all f ∈ L1 and kGk = kgk∞ .

51