Chapter 4

Measurable Functions

4.1 Measurable Functions

4.1.1 Measurable Mappings
We begin by looking at a very general situation: consider two measurable spaces (Ω, Σ)
and (Λ, Θ) and we propose to study the properties of mappings f : Ω −→ Λ in relation
to the preservation of Σ, Θ.
4.1 Proposition. Let (Ω, Σ) and (Λ, Θ) be two measurable spaces and let f : Ω −→ Λ.
Given (En )n∈N∗ ∈ Σ, (Fn )n∈N∗ ∈ Θ, then

(i) f (∪n En ) = ∪n f (En ),

(ii) f (∩n En ) ⊆ ∩n f (En ) and if f is 0ne-to-one, then f (∩n En ) = ∩n f (En ),

(iii) f (E1 \E2 ) ⊃ f (E1 )\f (E2 ) and if f is one-to-one, then f (E1 \E2 ) = f (E1 )\f (E2 ),

(iv) f −1 (∪n Fn ) = ∪n f −1 (Fn ),

(v) f −1 (∩n Fn ) = ∩n f −1 (Fn ),

(vi) f −1 (E1 \ E2 ) = f −1 (E1 ) \ f −1 (E2 ).

4.2 Corollary. Let (Ω, Σ) and (Λ, Θ) be two measurable spaces and let f : Ω −→ Λ.
Then

(i) The family

f −1 (Θ) = { f −1 (G) : G ∈ Θ }
is a σ-algebra.

(ii) The family

{ G ∈ Θ : f −1 (G) ∈ Σ }
is a σ-algebra.

4.3 Definition. Let (Ω, Σ) and (Λ, Θ) be two measurable spaces and let f : Ω −→ Λ.
We say that f is a measurable mapping if Θf ⊂ Σ. When we need to emphasise
the roles of Σ, Θ we say that f is (Σ, Θ)-measurable.
4.4 Proposition. Let (Ω, Σ) and (Λ, Θ) be two measurable spaces and let f : Ω −→ Λ.
Let B be a gnerating family for Θ, then f is measurable iff f −1 (B) ∈ Σ for every b ∈ B.
Proof. — Let
E = { G ∈ Θ : f −1 (G) ∈ Σ }
Measurable Functions 61

It’s easy (using the above proposition) to show that E is a σ-algebra. Since B ⊂ E by
assumption, then E = Θ. Hence, f −1 (G) ∈ Σ for all G ∈ Θ, which means that f is
measurable.
4.5 Proposition. Let (Ω, Σ), (Λ, Θ) and (P, Γ) be three measurable spaces and let
f : Ω −→ Λ and g : Λ −→ P . If f is (Σ, Θ)-measurable and g is (Θ, Γ)-measurable,
then g ◦ f is (Σ, Γ)-measurable.
Proof. — Let H ∈ Γ. Since g is (Θ, Γ)-measurable, hence g −1 (H) ∈ Θ. Since f
is (Σ, Θ)-measurable, hence f −1 (g −1 (H)) ∈ Σ. But f −1 (g −1 (H)) = (g ◦ f )−1 (H).

4.1.2 Real-Valued Measurable Functions

Real-valued functions play a major role in mathematics, so we would like to study
functions f : Ω −→ R where (Ω, Σ) is a measurable space.
To begin such a study in the framework of measure theory, we should consider some
σ-algebra on R. We have previously encountered two such σ-algebras: The first one
is the σ-algebra of all Borel sets B(R), which is generated by the family of all open
intervals, or the family of all closed intervals · · · etc.
The second possible choice for a σ-algebra on R is the σ-algebra M(R) of all
Lebesgue measurable sets. We know that B(R) ⊂ M(R).
4.6 Definition. Let (Ω, Σ) be a measurable space and let f : Ω −→ R. We say that
f is measurable or Σ-measurable, if f −1 (B) ∈ Σ for any B ∈ B(R).
4.7 Proposition. Let (Ω, Σ) be a measurable space and let f : Ω −→ R. The
following conditions are equivalent :

(i) f is Σ-measurable,

(ii) f −1 (] − ∞, b[) = {x ∈ Ω : f (x) < b} ∈ Σ for every b ∈ R,

(iii) f −1 (] − ∞, b]) = {x ∈ Ω : f (x) ≤ b} ∈ Σ for every b ∈ R,

(iv) f −1 (]a, +∞[) = {x ∈ Ω : f (x) > a} ∈ Σ for every a ∈ R,

(v) f −1 ([a, +∞[) = {x ∈ Ω : f (x) ≥ a} ∈ Σ for every a ∈ R,

(vi) f −1 (I) ∈ Σ for every open interval I ⊂ R,

(vii) f −1 (I) ∈ Σ for every closed interval I ⊂ R,

(viii) f −1 (I) ∈ Σ for every interval I ⊂ R.

Proof. —

(i)=⇒ (ii) is clear, since every interval is a Borel set.

(ii)=⇒ (iii) We note that ] − ∞, b] = ∩∞

n=1 ] − ∞, b + n [, hence
1

1
f −1 (] − ∞, b]) = ∩∞
n=1 f
−1
(] − ∞, b + [)
n
and the result follows since Σ is a σ-algebra.
62 Real Analysis - Part 2

(iii)=⇒ (iv) This follows also because Σ is a σ-algebra and

f −1 (]a, +∞[) = f −1 (R\] − ∞, a]) = Ω \ f −1 (] − ∞, a]).
The rest of the proofs of (iv)=⇒ (v), (v)=⇒ (vi), (vi)=⇒ (vii) and (vii)=⇒
(viii) are similar.
(viii)=⇒ (i) Let E = {A ∈ B(R) : f −1 (A) ∈ Σ}. (viii) means that E contains the
family of all intervals. And by Corollary [2] : E is a σ-algebra, hence it contains
all Borel sets.

For Ω = R we have two important types of measurability:

4.8 Definition. Let f : A −→ R where A is a Borel subset of R. f is said to be a
Borel function if it is (B(A), B(R))-measurable. i.e.if f −1 (B) is a Borel subset of A
for any Borel subset B of R.
f is said to be Lebesgue measurable if it is (M(A), B(R))-measurable, i.e.if the
inverse image f −1 (B) of a Borel subset of R is a Lebesgue measurable subset of A.
Note that a subset E of A is Lebesgue measurable iff E = F ∩ A where F is a Lebesgue
measurable subset of R.
Since continuous real-valued functions defined on a subset A of R play a fundamen-
tal role in analysis, we prove the following
4.9 Theorem. Let A ⊂ R be a Borel set. Every continuous function f : A −→ R is a
Borel function.
Proof. — Continuity implies that the inverse image of every open interval is an
open subset of A, i.e. there for every open interval I of R there exists an open set
O ⊂ of R such that f −1 (I) = O ∩ A. This is a Borel set since A itself is a Borel set.
Since open intervals generate the Borel σ-algebra, then the inverse image of every Borel
set in R is a Borel set in A.
We turn now to the study of general measurable functions defined on an abstract
space.
4.10 Theorem. [Properties of Real-Valued Measurable Functions]
Let (Ω, Σ) be a measurable space.
(i) If f, g : Ω −→ R are measurable, then f + g is measurable,
(ii) If f : Ω −→ R is measurable and α ∈ R, then αf is measurable,
(iii) If f, g : Ω −→ R are measurable, then f g is measurable,
(iv) If f, g : Ω −→ R are measurable, then |f |, f ∨g = max(f, g) and f ∧g = min(f, g)
are also measurable.
Proof. —
(i) Let a, b ∈ R, the measurability of f, g implies that the set
Sa,b = { x ∈ Ω : g(x) < b − a } ∩ { x : f (x) < a }
belongs to Σ. Now, we have
{ x : f (x) + g(x) < b } = ∪r∈Q Sr,b .
Since Q is countable, then { x : f (x) + g(x) < b } ∈ Σ for every b ∈ R.
Measurable Functions 63

(ii) This is easy since (αf )(x) < b ⇐⇒ f (x) < b

α
when α > 0 and (αf )(x) < b ⇐⇒
f (x) > αb when α < 0.

(iii) The measurability of q is an easy consequence of



∅ if b ≤ 0,
−1
For b ∈ R : q (] − ∞, b[) = {0} if b = 0,
 √ √

] − b, b[ if b > 0.

(iv) We note that f g = 14 [q(f + g) − q(f − g)]. Then use the previous result with the
fact that q(t) = t2 is a Borel function (since it is continuous).

(v) We have for any b ∈ R:

(f ∨ g)−1 (] − ∞, b[) = f −1 (] − ∞, b[) ∩ g −1 (] − ∞, b[),

(f ∧ g)−1 (]b, +∞[) = f −1 (]b, +∞[) ∩ g −1 (]b, +∞[).

One last result is worth mentionning :

4.11 Proposition. Let Ω, Σ) be a measurable space and let f, g : Ω −→ R, then the
two sets
{ x ∈ Ω : f (x) = g(x) } and { x ∈ Ω : f (x) ̸= g(x) }
are both in Σ.
Proof. — This follows because

{ x ∈ Ω : f (x) ̸= g(x) } = (g − f )−1 (] − ∞, 0[) ∪ (g − f )−1 (]0, +∞[).

4.1.3 Simple Functions

The simplest example of a real-valued measurable function is a characteristic func-
tion of a set A in Σ. This is the function χA : Ω −→ R defined by
{
1 if x ∈ A,
χA (x) =
0 if x ∈
/ A.

To see that this function is measurable, we just compute



 ∅ if 0, 1 ∈
/ B,

A if 1 ∈ B, 0 ∈/ B,
χ−1
A (B) =

 Ω\A if 0 ∈ B, 1 ∈/ B,


Ω if 0, 1 ∈ B.

a measurable space. A Σ-simple function or a sim-

4.12 Definition. Let (Ω, Σ) be∑
ple function is a function g = ni=1 ai χAi , where A1 , · · · , An ∈ Σ and a1 , · · · , an ∈ R.
64 Real Analysis - Part 2

We remark that every simple function, g, can be represented in the form

∑
n
g= bk χBk , with B1 , · · · , Bn ∈ Σ and pairwise disjoint.
k=1

So, a simple function is a linear combination of characteristic functions of measurable

sets. This is enough to deduce that the family of all simple functions is a vector space.
Also, since the sum of two measurable functions is measurable and characteristic
functions of sets from Σ are measurable, then every simple function is measurable.
Another interesting fact is that the product of simple functions is a simple function,
because χA .χB = χA∩B .
The most important fact about simple functions is
4.13 Theorem. Let (Ω, Σ) be a measurable space. Every bounded measurable func-
tion f : Ω −→ R can be arbitrarily approximated by simple functions.
i.e. for each ε > 0 there exists a simple function g such that

sup |f (x) − g(x)| < ε.

x∈Ω

M −m
Proof. — Let range(f ) ⊂ [m, M ]. and choose n ∈ N∗ such that < ε. Put
n
k(M − m)
bk = m + , k = 0, · · · , n and
n
Bk = f −1 ([bk−1 , bk [), k = 1, · · · , n − 1, Bn = f −1 [bn−1 , bn ]).

Since f is measurable, hence B1 , · · · , Bn ∈ Σ. Let

∑
n
g= bk−1 χBk ,
k=1

then for each x ∈ Ω there exist k; 1 ≤ k ≤ n such that x ∈ Bk . So,

|f (x) − g(x)| = |f (x) − bk | ≤ |bk−1 − bk | < ε.

For an unbounded measurable function, we can construct a sequence (gn )n∈N∗ of

simple functions that converges to f pointwise. We can even prove the stronger version
4.14 Theorem. Let (Ω, Σ) be a measurable space. Every Σ-measurable positive
function is the limit of an increasing sequence of simple functions.
Proof. — Let N ∈ N∗ and let Bk,N = f −1 ([bk−1,N , bk,N [), k = 1, · · · , 2N −
2k ∑N
1, B2N ,N = f −1 [bN 2N −1,N , bN 2N ,N ]), where bk,N = −N + N . Put gN = 2k=1 bk−1,N χBk,N ,
2
then as in the previous proposition
2
|f (x) − gN (x)| ≤ .
2N
We can easily see that
{
bk,N if j = 2k, k ≤ N 2N ,
bj,N +1 = 1 , Bk,N = B2k−1,N +1 ∪ B2k,N +1 .
(b + bk−1,N ) if j = 2k − 1, k ≤ N 2N ,
2 k,N
Measurable Functions 65

Now, for x ∈ Ω, f (x) ∈ [−N, N ] there exists k, 1 ≤ k ≤ N 2N such that x ∈ Bk,N :

gN (x) = bk−1,N ≤ min(b2k−2,N +1 , b2k−1,N +1 ) ≤ gN +1 (x).

For x ∈ Ω; f (x) ∈ [−(N + 1), −N [∪]N, N + 1] we have gN (x) = 0 ≤ gN +1 (x). Finally,

for x ∈ Ω : f (x) ∈
/ [−(N + 1), (N + 1)] we have gN (x) = gN +1 (x) = 0.
Hence gN ≤ gN +1 for all N .

4.1.4 Measurable Functions with Values in R

Some times we need to consider functions that may attain infinite values. For this we
need to extend our definition of measurable functions, or to extend the notion of Borel
sets.
4.15 Definition. Let (Ω, Σ) be a measurable space and let f : Ω −→ R, where
R = R ∪ {−∞, +∞}. We say that f is Σ-measurable if

{ x ∈ Ω : a ≤ f (x) ≤ b } ∈ Σ ∀a, b ∈ R, a ≤ b.

The algebra of exteneded real-valued measurable functions is not too different from
that of the usual measurable functions. We have just to remember that the sum a + b
of two elements of R is defined only when a, b ∈ R or a ∈ R, b = ±∞ or a = ±∞, b ∈ R
or a = b = ±∞. In other words the sum a + b is defined unless one of a, b is +∞ and
the other is −∞. For the product: a.b is defined unless one of a, b is 0 and the other
is ±∞. We then have
4.16 Proposition. The sum, the product, the maximum and the minimum of two
measurable functions with values in R is measurable if it’s well-defined.
66 Real Analysis - Part 2

4.2 Sequences of Measurable Functions

Let’s consider a sequence (fn )n∈N∗ of measurable functions defined on the same mea-
surable space (Ω, Σ). For each fixed x ∈ Ω the sequence (fn (x))n∈N∗ is a sequence of
real numbers (or ±∞), so we may ask whether this sequence converges (in R)?

4.2.1 General Considerations About Convergence

It may happen that for some x’s the sequence (fn (x))n∈N∗ converges, while for other
x’s it does noy. But we have
4.17 Theorem. Let (Ω, Σ) be a measurable space and let fn : Ω −→ R be a sequence
of measurable functions, then the set
L = { x ∈ Ω : (fn (x))n∈N∗ has a limit in R }
is in Σ.
Proof. — First consider L∞ = { x ∈ Ω : limn→∞ fn (x) = +∞ }. We have
L∞ = ∩M ∈N∗ ∪n∈N∗ ∩∞
k=n+1 { x ∈ Ω : fk (x) ≥ M }.

Since each set { x ∈ Ω : fk (x) ≥ M } is in Σ and Σ is a σ-algebra, hence L∞ ∈ Σ.

The same argument works for L−∞ = { x ∈ Ω : limn→∞ fn (x) = −∞ }.
We consider next the set LR = L∞ = { x ∈ Ω : limn→∞ fn (x) = c ∈ R }. Since
every convergent sequence in R is a Cauchy sequence and vice versa, then
1 1
LR = ∩m∈N∗ ∪n∈N∗ ∩∞ k,ℓ=n+1 { x ∈ Ω : < fk (x) − fℓ (x) < },
m m
the measurability of fk − fℓ implies that { x ∈ Ω : m1 < fk (x) − fℓ (x) < m1 } ∈ Σ for
each k, ℓ, m. Hence we may deduce that LR ∈ Σ.
The second important result about sequences of measurable functions is
4.18 Theorem. Let (Ω, Σ) be a measurable space and let fn : Ω −→ R be a sequence
of measurable functions. Let
f ∗ (x) = sup fk (x), f∗ (x) = inf∗ fk (x).
k∈N∗ k∈N

Then both f ∗ , f∗ are measurable functions.

Moreover, the functions
lim sup fk , lim inf fk
k→∞ k→∞

are also measurable.

Proof. — The proof is similar to the case of the max and min of two measurable
functions.
Let a, b ∈ R, then
f ∗ (x) ≤ b ⇐⇒ ∀k ∈ N∗ : fk (x) ≤ b, f ∗ (x) ≥ a ⇐⇒ ∃k ∈ N∗ : fk (x) ≥ a.
Also
f∗ (x) ≤ b ⇐⇒ ∃k ∈ N∗ : fk (x) ≤ b, f∗ (x) ≥ a ⇐⇒ ∀k ∈ N∗ : fk (x) ≥ a.
The second part of the theorem results from the fact that
lim sup fk (x) = inf∗ sup fk (x), lim inf fk (x) = sup inf fk (x).
k→∞ n∈N k≥n k→∞ n∈N∗ k≥n
Measurable Functions 67

4.2.2 Almost Everywhere Convergence

To understand convergence of sequences of measurable functions we move further to
consider a measure µ and ask questions about the case when the sequence (fn (x))n∈N∗
converges on the whole of Ω exept for a very small set N with µ(N ) = 0.
4.19 Definition. Let (Ω, Σ, µ) be a measure space and let f, g : Ω −→ R be two
measurable functions. We say that f and g are equal µ almost everywhere if

µ({ x ∈ Ω : f (x) ̸= g(x) } = 0.

If (fn )n∈N∗ is a sequence of measurable functions, then we say that (fn )n∈N∗ converges
µ-almost everywhere if

µ({ x ∈ Ω : lim fn (x) does not exist } = 0.

n→∞

This is equivalent to saying that the sequence is Cauchy µ-almost everywhere, i.e.

µ({ x ∈ Ω : lim fn (x) is not a Cauchy sequence } = 0.

n→∞

Almost everywhere convergence is weaker that pointwise convergence, which is

weaker than uniform convergence. To see this let’s consider some examples:
4.20 Examples and Counter-Examples.
(1) Let fn : [0, 1[−→ R be defined by fn (x) = xn , n ∈ N∗ . Here [0, 1[ is considered as
a measure space endowed by its Borel σ-algebra and the Lebesgue measure re-
stricted to [0, 1[. So, all the functions fn are measurable (they are Borel functions
since they are continuous).
Let g(x) ≡ 0, then fn (x) converges to g(x) for every x ∈ [0, 1[.
However, the convergence is not uniform. Because

sup |fn (x) − g(x)| = sup |xn | = 1.

0≤x<1 0≤x<1

(2) It’s easy to construct a sequence of measurable functions that converges almost
everywhere, but not for all the points. One such example is to take the same
sequence fn (x) = xn above but to change the function g at a set A of Lebesgue
measure 0. For example, take h(x) = 0, x ∈ [0, 1[\Q and h(x) = 1, x ∈ [0, 1[∩Q,
then h = g λ a.e., because λ([0, 1[∩Q) = 0. Hence

h(x) = lim fn (x), λ − a.e.

n→∞

But of course (fn )n∈N∗ does not converge to h at the rational points.

(3) Let δ; 0 < δ < 1 be fixed, and let the rational numbers be enumerated as

Q = { rn : n ∈ N∗ }.

Attache to each rational number rn an interval In = [rn − δ

2n+1
, rn + δ
2n+1
] and
define
∑n
gn : R −→ R, gn (t) = χI (t).
k
k=1
68 Real Analysis - Part 2

We can see that (gn )n∈N∗ converges to the zero function on a very large set,
/ ∪∞
beacuse gn (t) = 0 for all t ∈ ∞
k=1 Ik and for all n. The set A = ∪k=1 Ik has
measure
∑∞ ∑∞
δ
λ(A) ≤ ℓ(Ik ) = k
= δ.
k=1 k=1
2

However, (gn )n∈N∗ does not converge to 0 λ-a.e. because gn (t) ≥ 1 for all t ∈ I1
and λ(I1 ) = δ/2 > 0.

(4) We can modifiy the previous ∑example by replacing Ik with In,k = [rk − 2n+k
δ
, , rk +
∞
δ
2n+k
]. Then lim g
n→∞ n = n χ
k=1 In,k will be zero outside ∩ A
n=1 n , where An =
∞
∪k=1 In,k . To prove this, let t ∈
/ An , then t ∈/ Am for all m ≥ n, hence gm (t) = 0
for all m ≥ n.
The measure of A is the limit of λ(An ) (since λ(A1 ) < +∞). So,

∑
∞ ∑∞
δ
λ(A) = lim λ(An ) ≤ lim ℓ(In,k ) = lim n+k
= 0.
n→∞ n→∞
k=1
n→∞
k=1
2

Hence (gn )n∈N∗ converges to zero λ-a.e.

(5) Another example of a sequence that converges almost everywhere, but not every-
where is the following more interesting example:
Let n ∈ N∗ , and let k be an integer between
∑n−1 0 and j2 −1, then k can be represented
n

as a binary number in the form k = j=0 bj × 2 with b0 , · · · , bn−1 ∈ {0, 1}. We

construct the intervals

ck ck + 1 ∑ n−1
Ik,n = [ n, n
], where c k = 2bj × 3j ,
3 3 j=0

Note that for each n: the intervals (In,k )k are pairwise disjoint, and In,k ⊃ In+1,2k ∪
In+1,2k+1 , so χIn,k (x) ≥ χIn+1,2k (x) + χIn+1,2k+1 (x). Now, define

2∑
n −1

Let fn (x) = χI (x),

n,k
x ∈ [0, 1].
k=0

Then fn+1 (x) ≤ fn (x) ∀x ∈ [0, 1], ∀n ∈ N∗ , and since fn (x) ∈ {0, 1}, then
limn→∞ fn (x) exists for each x and is either 0 or 1.
Let F = { x ∈ [0, 1] : limn→∞ fn (x) = 1 }. F is called the Cantor set. Then
the sequence (fn )n∈N∗ converges pointwise to χF .
The Cantor set has lebesgue measure 0 : To prove this note that limn→∞ fn (x) =
0 on each interval not containing any In,k for some n and all k : 0 ≤ k < 2n . So,
−1 n
F ⊂ ∪2k=0 In,k .
−1
n
2n
But λ(In,k ) = 1
3n
, so λ(∪2k=0 In,k ) = 3n
. Hence

2n
λ(F ) ≤ n ∀n ∈ N∗ ,
3
Measurable Functions 69

which implies that λ(F ) = 0.

We deduce then that (fn )n∈N∗ converges to 0, λ a.e.
Let’s now study the uniform convergence of the sequence (fn )n∈N∗ : we note that
if m > n, then
n −1
In,k ⊃ Fm = ∪2j=0−1 Im,j .
m
Fn = ∪2k=0
2n 2m
Examining the Lebesgue measures of both sets we see that λ(Fn ) = n > m =
3 3
2n 2m
λ(Fm ). So, the difference Fn \ Fm has Lebesgue measure n − m . For x in this
3 3
difference: fn (x) = 1, while fm (x) = 0. Hence

sup |fn (x) − fm (x)| = 1.

x∈[0,1]

This shows that (fn )n∈N∗ is not uniformly Cauchy, hence it does not converge
uniformly on [0, 1].
(6) Let’s now describe another example also related to the Cantor set. We will define
a sequence (gn )n∈N∗ of continuous, increasing, piece-wise linear functions, wich
converges uniformly to a function g on [0, 1]. The function g will be continuous,
increasing and constant on any interval that does not intersect F , but we will
have: g(0) = 0 and g(1) = 1. So g increases continuously from 0 to 1, while being
constant almost of the time!
To construct the sequence (gn )n∈N∗ we begin by fixing n ∈ N∗ . For an integer
k : 0 ≤ k < 2n we take as before (bj )n−1
j=0 to be the binary digits of k and put
∑n−1
ck = j=0 2bj × 3 and Ik,n = [ 3n , 3n ] as in the previous example. We also take
j ck ck +1
ck+1
dk = k−c
2n
k
and we set the interval Jk,n =] ck3+1
n , 3n [, i.e.Jk,n is an interval in the

complement of Fn . Actually,
1 2
[0, 1] = F1 ∪] , [= I1,0 ∪ J1,0 ∪ I1,1 ,
3 3
and
In,k = In+1,2k ∪ Jn+1,k ∪ In+1,2k+1 .
Jn+1,k is called the open middle third of Ik,n .
Define 
 3
 2 x, x ∈ I1,0 ,
1
g1 (x) = 2 , x ∈ J1,0 ,

3
2
x − 2 , x ∈ I1,1 .
1

Note that g1 is continuous, increasing and satisfies g(0) = 0 and g(1) = 1.

In general, we define
 n

 2n x + dk , x ∈ In,k ,
3

gn (x) = k+1 , x ∈ Jn,k



2n
gn−1 (x), x ∈ / Fn .

The reader is invited to examine gn and prove that it is continuous, increasing

and satisfies gn (0) = 0, gn (1) = 1.
70 Real Analysis - Part 2

The final step is to prove that (gn )n∈N∗ is uniformly convergent. To this end,
lets compute supx∈[0,1] |gn+1 (x) − gn (x)|: Since Fn+1 ⊂ Fn , then gn+1 (x) = gn (x)
for all x ∈
/ Fn . Now, let x ∈ Fn , then x ∈ Ik,n for some k : 0 ≤ k < 2n , and
3n
gn (x) = 2n x + dk . On the otherhand,
 n+1
 3
 2n+1 x + d2k , x ∈ In+1,2k ,
2k+1
gn+1 (x) = 2n+1 , x ∈ Jn+1,2k ,

 3n+1
2n+1
x + d2k+1 , x ∈ In+1,2k+1 .
So,
 n+1 

 3n
 2n+1 x + d2k − 2nx − dk  , x ∈ In+1,2k ,
3

|gn+1 (x) − gn (x)| =  2k+1 − 23n x − dk  ,
n



2n+1  x ∈ Jn+1,2k ,
 3n+1 x + d 3n 
2n+1 2k+1 − 2n x − dk  , x ∈ In+1,2k+1 .

We note that
ck ck + 1 2ck + 1 2ck + 1 2k − 2ck 2k + 1 ck ck + 1
gn (( + )/2) = +d k = + = = gn+1 (( + n )/2),
3n 3n 2n+1 2n+1 2n+1 2n+1 3n 3
ck k 2k ck
gn ( n
) = n = n+1 = gn+1 ( n ),
3 2 2 3
and
ck + 1 k+1 2k + 2 ck + 1
gn ( n
)= n
= n+1 = gn+1 ( n ).
3 2 2 3
Since, both gn , gn+1 are increasing functions we deduce that
sup |gn+1 (x) − gn (x)| = max (|gn+1 (x1 ) − gn (x1 )| , |gn+1 (x2 ) − gn (x2 )|) ,
x∈In,k

c2k+1
where x1 = c32kn+1
+1
, x2 = 3n+1
are the end points of Jn+1,2k . By direct substitution,
we find that
2k + 1 3k + 1 1
|gn+1 (x1 ) − gn (x1 )| = |gn+1 (x2 ) − gn (x2 )| = − = .
2n+1 3×2 n 3 × 2n+1
Clearly this value does not depend on k, hence
1
sup |gn+1 (x) − gn (x)| = .
x∈[0,1] 3 × 2n+1

A little work with geometric series will establish the desired result: (gn )n∈N∗ is
uniformly Cauchy in [0, 1], hence converges to a continuous, increasing function
on [0, 1].
Now, since each gn satisfies gn (0) = 0, gn (1) = 1, then g(0) = 0 and g(1) = 1.
Also, since for all m ≥ n : gm is constant on each interval in the complement of
Fn , then g is constant on each interval in the complement of Fn , and since this
is true for all n, then g is constant on each interval in the complement of F .
The continuity of g implies that it is surjective onto [0, 1], i.e. g : [0, 1] −→
[0, 1], continuous, increasing and surjective. We can even prove that g is strictly
increasing on F , hence has an inverse function h : [0, 1] −→ F which is continuous
and strictly increasing.
Measurable Functions 71

Our plan is to study the basic properties of almost everywhere convergence (al-
gebraic properties and other properties), then we will turn to the relation between
different types of convergence.
4.21 Theorem. Let (Ω, Σ, µ) be a measure space and let (fn )n∈N∗ be a sequence of
real-valued measurable functions on Ω, then

(i) If (fn )n∈N∗ converges µ-almost everywhere and h : Ω −→ R is any given measur-
able function, and we take
{
limn→∞ fn (x), when the limit exists,
f (x) =
h(x), otherwise.

If (Ω, Σ, µ) is a complete measure space, then f is a measurable function.

(ii) If (fn )n∈N∗ converges almost everywhere, then the limit function is unique up to
almost everywhere equality. i.e. if limn→∞ fn (x) = f (x) µ-almost everywhere,
and limn→∞ fn (x) = f˜(x) µ-almost everywhere, then f (x) = f˜(x) µ-almost ev-
erywhere.

(iii) (fn )n∈N∗ converges µ-almost everywhere ⇐⇒ (fn )n∈N∗ is Cauchy µ-almost every-
where.

Proof. —

(i) Let C = { x ∈ Ω : limn→∞ fn (x) = f (x) } and let N = { x ∈ Ω : h(x) = f (x) }.

N ∈ Σ by assumption, and since C = Ω \ N , then C ∈ Σ.
By Theorem [18] we deduce thet f |C is measurable, hence f = f |C + h is mea-
surable.

(ii) This results from the uniqueness of the limit or a sequence of real numbers.

(iii) This is also a propert of sequences of real numbers: convergence is equivalent to

the Cauchy condition.

The algebra of almost everywhere convergence is contained in the following theorem

4.22 Theorem. Let (Ω, Σ, µ) be a measure space, and let (fn )n∈N∗ , (gn )n∈N∗ be two
sequences of real-valued measurable functions defined on Ω. If (fn )n∈N∗ converges µ-a.e.
to f and (gn )n∈N∗ converges µ-a.e. to g, then

(i) (fn + gn )n∈N∗ converges µ-a.e. to f + g,

(ii) (fn .gn )n∈N∗ converges µ-a.e. to f.g,

(iii) If g(x) ̸= 0 µ-a.e., then (fn /gn )n∈N∗ converges µ-a.e. to f /g,

(iv) If there exists n0 such that fn ≤ gn µ-a.e. for all n > n0 , then f ≤ g µ-a.e.
72 Real Analysis - Part 2

Proof. — We will prove only (iii): Let Cf = { x ∈ Ω : fn (x) → f (x) }, Cg =

{ x ∈ Ω : gn (x) → g(x) }, A = { x ∈ Cg : g(x) ̸= 0 } and An = { x ∈ Cg : gn (x) ̸= 0 }.
We have
A = ∪∞n=1 ∩k≥n Ak .
fk (x)
Let x ∈ A ∩ Cf , then there exists n such that x ∈ (∩k≥n Ak ) ∩ Cf , so is defined
gk (x)
fk (x) f (x)
for each k ≥ n and limk→∞ = . By assumption we have µ(Ω \ (A ∩ Cf )) ≤
gk (x) g(x)
µ(Ω \ A) + Ω \ Cf ) = 0, hence
fk (x) f (x)
lim = µ − a,e.
k→∞ gk (x) g(x)

4.23 Corollary. The sum of an almost everywhere convergent series of measurable

functions is a measurable function.
4.24 Corollary. Let A ⊂ R be a Borel set, if (fn )n∈N∗ is a sequence of continuous
functions that converge almost everywhere (with respect to the Lebesgue measure) to
a function f , then f is a Borel function.

4.2.3 The Theorems of Egorov and Lusin

We are interested in the relation between almost everywhere convergence and uniform
convergence. We know that uniform convergence implies almost everywhere conver-
gence and we know that the converse is not true. There exist sequences (fn )n∈N∗ of
functions that are convergent a.e. but not uniformly convergent. However, given a
sequence of functions (fn )n∈N∗ on a measure space (Ω, Σ, µ) which converges µ-a.e.,
then we can cut out a little piece of our measure space Ω such that (fn )n∈N∗ converges
uniformly on the remaining part. This what asserts the famous theorem of Egoroff
4.25 Egoroff ’s Theorem. Let (Ω, Σ, µ) be a finite measure space, and let (fn )n∈N∗
be a sequence of measurable functions defined on Ω. If (fn )n∈N∗ is convergent µ-a.e.
on Ω, then for every ε > 0 there exists a set Fε ∈ Σ such that
(i) µ(Fε ) ≤ ε,
(ii) (fn )n∈N∗ converges uniformly on Ω \ Fε .
Proof. — Let E = { x ∈ Ω : limn→∞ fn (x) = f (x) }, and for each k, m ∈ N∗ let
1
Ek,m = { x ∈ E : |fk (x) − f (x)| < }, Uk,m = ∩∞
j=k Ek,m .
m
Then, for each fixed m ∈ N∗ , the sequence of sets (Uk,m )k∈N∗ is increasing and its union
is the whole of E.
Hence, the sequence of sets (E \ Uk,m )k∈N∗ is decreasing and its intersection is ∅, so

lim µ(E \ Uk,m ) = 0.

k→∞

ε
Given ε > 0, we may find km ∈ N∗ such that µ(E \ Ukm ,m ) < . We take Fε =
∞
2m
∪m=1 (Ω \ Ukm ,m ). Then we are sure that µ(Fε ) ≤ ε (because µ(Ω \ E) = 0).
Measurable Functions 73

Now, let x ∈ Ω \ Fε we have x ∈ E ∩ ∩∞ ∗

m=1 Ekm ,m , then for each m ∈ N :
1
|fk (x) − f (x)| < for all k ≥ km , hence
m
lim sup |fk (x) − f (x)| = 0
k→∞ x∈F
/ ε

i.e. (fn )n∈N∗ converges uniformly on Ω \ Fε .