Note on Real Analysis

The following part is the summary of [5, Section 11]. We start with some definitions (denoted
by [D*]), remarks (denoted by •) and problems (denoted by [P*]).

(D1) Let R be a family of sets. We say that R is a ring if A ∪ B ∈ R and A \ B ∈ R for any
A, B ∈ R.

ˆ If R is a ring, then A ∩ B ∈ R, since A ∩ B = A \ (A \ B).

(D2) We say that R is a σ-ring if R is a ring and ∪∞

n=1 An ∈ R for any An ∈ R.

(D3) We say φ is a set function if φ : R → R ∪ {±∞}. In such case, φ is additive if φ(A ∪ B) =

φ(A) + φ(B) for any A, B ∈ R with A ∩ B = ∅. We say φ is countable additive if
∞
X
φ (∪∞
n=1 An ) = φ(An ), Ai ∩ Aj = ∅, i 6= j.
n=1

ˆ It is straightforward that φ(∅) = 0 and φ(A ∪ B) + φ(A ∩ B) = φ(A) + φ(B).

(D4) We say that a non-negative set function φ is monotone, if φ(A \ B) = φ(A) − φ(B) for B ⊂ A
and |φ(B)| < ∞.

ˆ If φ is non-negative, then φ(A) ≤ φ(B), if A ⊂ B.

Next, we show that incresing sequece of sets in a ring yields increasing convergent numbers
evaluated by countable additive set function.

Theorem 0.1 Let φ be a countable additive set function on a ring R. Suppose that {Ai }∞ i=1 is an
∞
increasing sequence, i.e. Ai ⊂ Ai+1 , i = 1, 2, · · · (denoteby by {Ai }i=1 %, for simplicity). Then
φ(Ai ) → φ(A) as i → ∞, where A = ∪∞ i=1 Ai .

Indeed, set B1 = A1 and Bi = Ai \ Ai−1 , i = 2, 3, · · · . Then, Bi ∩ Bj = ∅, if i 6= j. Thus,

n
X
φ(An ) = φ(Bi ) −→ φ(A).
i=1

(P1) What if the increasing sequence Ai is replaced by a decreasing sequence Ai in Theorem 0.1?

”construction of the Lebesgue measure”

(D5) Let Rn be the n-dimensional Eucledian space and we mean by an interval in Rn , denoted by
I, a set of the form

I = {(x1 , x2 , · · · , xn ) ∈ Rn : ai ≤ xi ≤ bi , i = 1, 2, · · · , n},

where ≤ can be replaced by <, i.e. ai < xi ≤ bi , ai ≤ xi < bi or ai < xi < bi . Here its size of
A is denoted by m(I) := Πni=1 (bi − ai ) (you can say it the measure of I). We say that a set
A ⊂ Rn is an elementary set if it is the union of a finite number of intervals, i.e. A = ∪ki=1 Ii ,
and we let E the collection of all elementary sets in Rn , i.e. E = {A : A = ∪ki=1 Ii }.

1
 ˆ E is a ring (but not a σ−ring). A ∈ E is the union of disjoint intervals. As a set function, m
is additive.

(D6) Let φ : E → [0, ∞] be a set function. We say that φ is regular if the following holds: For any
A ∈ E and for every  > 0 there exist sets a closed set F ∈ E and an open set G ∈ E such
that F ⊂ A ⊂ G (it is direct that φ(F ) ≤ φ(A) ≤ φ(G)) and φ(G) −  ≤ φ(A) ≤ φ(F ) + .

ˆ The set function m is regular.

(D7) (Outer measure) Let µ be a non-negative set function that is addtive, regular, and finite on
E. For any set E ⊂ Rn we define the outer measure of E corresponding to µ, denoted by
µ∗ (E) by
∞
X
µ∗ (E) = inf µ(Ai ), E ⊂ ∪∞i=1 Ai , Ai ∈ E,
i=1

where the inf is taken for all possible countable covering of E by open elementary sets.

ˆ It is direct that µ∗ (E) ≥ 0 and µ∗ (E1 ) ≤ µ∗ (E2 ) if E1 ⊂ E2 . It is also easy to see that
µ∗ (E) = µ(E) for E ∈ E, and thus µ∗ is an extension of µ.
∞
X
n
Theorem 0.2 Let E ⊂ R . If E = ∪∞
i=1 Ei ,
∗
then µ (E) ≤ µ∗ (Ei ).
i=1

Indeed. We assume that µ∗ (Ei ) < ∞, WLOG. For given  >P0 there exists a sequence of
∞
elementary sets,P{Ai,kP such that Ei ⊂ ∪∞
, k = 1, 2, · · · , P k=1 Ai,k and
∗ −i
k=1 µ(Ai,k ) ≤ µ (Ei ) + 2 .
∗ ∞ ∞ ∞ ∗
Thus, µ (E) ≤ i=1 k=1 µ(Ai,k ) ≤ i=1 µ (Ei ) + . 

(D8) Let A, B ⊂ Rn . We denote the symmetric difference of A and B by S(A, B) = (A\B)∪(B \A)
and we define d(A, B) = µ∗ (S(A, B)). We say that Ai → A if lim d(A, Ai ) = 0. Furthermore,
i→∞
if Ai ∈ E and Ai → A, we say that A is finitely µ-measurable and we denote A ∈ MF (µ). In
addition, if A is union of a countable collection of finitely µ−measurable sets, i.e. A = ∪∞
i=1 Ai
with Ai ∈ MF (µ), we say that A is µ−measurable and we denote A ∈ M(µ).

ˆ Note that S(A, B) = S(B, A), S(A, A) = ∅ and S(A, B) ⊂ S(A, C) ∪ S(C, B). Furthermore,

S(A1 ∪ A2 , B1 ∪ B2 ) 
S(A1 ∩ A2 , B1 ∩ B2 ) ⊂ S(A1 , B1 ) ∪ S(A2 , B2 ) (0.1)
S(A1 \ A2 , B1 \ B2 )


ˆ Observe that d(A, B) = d(B, A), d(A, A) = 0 and d(A, B) ≤ d(A, C) + d(C, B). In addition,
 
S(A1 ∪ A2 , B1 ∪ B2 )
d  S(A1 ∩ A2 , B1 ∩ B2 )  ≤ d(A1 , B1 ) + d(A2 , B2 ). (0.2)
S(A1 \ A2 , B1 \ B2 )

However, d(A, B) = 0 doesn’t imply that A 6= B. For example, A is a countable set and
B = ∅.

(D9) A ∼ B is equivalent if d(A, B) = 0. (This is an underlying idea of MF (µ), closure of E).

2
 ˆ We note that |µ∗ (A) − µ∗ (B)| ≤ d(A, B) if µ∗ (A) or µ∗ (B) is finite. Indeed, suppose without
loss of generality (WLOG) that 0 ≤ µ∗ (B) < µ∗ (A) and µ∗ (B) < ∞. Then,

µ∗ (A) = d(A, ∅) ≤ d(A, B) + d(B, ∅) = d(A, B) + µ∗ (B).

which implies that

|µ∗ (A) − µ∗ (B)| ≤ d(A, B). (0.3)

(P2) It is clear that A ∈ MF (µ) means A ∈ M (µ). Is there any case such that A ∈ M (µ) but
A∈ / MF (µ)?

Theorem 0.3 M(µ) is a σ−ring, and µ∗ is countably additive on M(µ).

Indeed. • (step 1 ; µ∗ is additiv on MF (µ))
Suppose that A, B ∈ MF (µ). Then, we choose An , Bn with An , Bn ∈ E such that An → A and
Bn → B . Then, due to (0.2) and (0.3), we can see that

An ∪ Bn , An ∩ Bn , An \ Bn , µ∗ (An ) −→ A ∪ B, A ∩ B, A \ B, µ∗ (A) respectively. (0.4)

Indeed, it is due to the facts that following sets are subset of S(A, An ) ∪ S(B, Bn );

S(A ∪ B, An ∪ Bn ), S(A ∩ B, An ∪ Bn ), S(A \ B, An \ Bn ),

and |µ∗ (A) − µ∗ (An )| ≤ d(A, An ). In addition, µ∗ (A) < ∞, since µ∗ (An ) < ∞ and d(An , A) → 0.
Next, suppose that A, B ∈ MF (µ) with A ∩ B = ∅. Observing that MF (µ) is a ring via first
and third convergence in (0.4), we recall that

µ(An ) + µ(Bn ) = µ(An ∪ Bn ) + µ(An ∩ Bn ),

which implies via (0.4) that since A ∩ B = ∅, as n → ∞

µ∗ (A) + µ∗ (B) = µ∗ (A ∪ B) + µ∗ (A ∩ B) = µ∗ (A ∪ B).

Here we used that

|µ∗ (An ∩ Bn ) − µ∗ (A ∩ B)| ≤ d(An ∩ Bn , A ∩ B) ≤ d(An ∩ Bn , A ∩ Bn ) + d(A ∩ Bn , A ∩ B)

= µ∗ (S(An ∩ Bn , A ∩ Bn )) + µ∗ (S(A ∩ Bn , A ∩ B)) ≤ µ∗ (S(An , A)) + µ∗ (S(Bn , B))

= d(An , A) + d(Bn , B) −→ 0 as n → 0.
• (step 2 ; A ∈ MF (µ) if A ∈ M (µ) and µ∗ (A) < ∞)
Firstly, we claim that if A ∈ M(µ), then A = ∪∞ i=1 Ai with Ai ∈ MF (µ), where Ai is mutually
n−1 0
disjoint. Truly, if A = ∪i=1 Ai , then we define A1 = A01 and An = ∪ni=1 A0i \ ∪i=1
∞ 0
Ai for n ≥ 2. Then,
∞
Ai ∈ MF (µ) becomes mutually disjoint and A = ∪ i=1 iA .
that µ∗ (A) ≤ ∞ ∗ n
P
Secondly, it is obvious P
∗ n n ∗
i=1
∗
µ (A
∗
P∞∪i=1
i ). On the other hand, since
∗
Ai ⊂ A, it
follows that µ (∪i=1 Ai ) = i=1 µ (Ai ) ≤ µ (A), which implies that µ (A) = i=1 µ (Ai ). Since
µ∗ (A) < ∞, then ∪ni=1 Ai → A as n tends to infinity, because
∞
X
∗ ∗
|µ (A) − µ (∪ni=1 Ai )| ≤ d(A, ∪ni=1 Ai ) =µ∗ (∪∞
i=n+1 Ai ) = µ∗ (Ai ).
i=n+1

3
 Therefore, we conclude that A ∈ MF (µ), since ∪ni=1 Ai is finitely µ−measurable.
• (step 3; M(µ) is a σ−ring)
Now it is clear that µ∗ is countable additive on M(µ). Truly, suppose that A = ∪Ai , where
Ai ∈ M(µ)P∞ is mutually disjoint. Then, by repeating above argument, it is easy to check that
µ (A) = i=1 µ (Ai ). To show that M(µ) is a σ−ring, suppose that A, B ∈ M(µ) and A = ∪∞
∗ ∗
i=1 Ai
and B = ∪∞ B
i=1 i , where Ai , Bi ∈ M F (µ). It remains to show that A \ B ∈ M(µ).
Note that for fixed n we see that An ∩ B = ∪∞ i=1 (An ∩ Bi ), which means An ∩ B ∈ M(µ).
Furthermore, An ∩ B ∈ MF (µ), since µ∗ (An ∩ B) ≤ µ∗ (An ) < ∞ (see step 2). Therefore, An \ B ∈
MF (µ) as well, because An \ B = ∩∞ ∗
j=1 (An \ (An ∩ Bj )) and µ (An \ B) < ∞. This yields that
A \ B ∈ M(µ) due to A \ B = ∪∞ i=1 (Ai \ B) ∈ M(µ). 

ˆ To sum up, µ is initially defined on E and is extended to µ∗ , which is a countable additive set
function (we call it a measure) on the σ−ring M(µ). From now on, µ∗ (A) can be replaced by
µ(A) if A ∈ M(µ).

ˆ If A is open, then A ∈ M(µ), since any open set is the union of a countable collection of open
intervals.

ˆ If A ∈ M(µ), then for any  > 0 there exists sets closed F and open G such that F ⊂ A ⊂ G
and µ(G \ A) + µ(A \ F ) < .

(D10) We say that E is a Borel set if E is obtained by a countable number of operations, unions,
intersections or complements starting from open sets.

ˆ The family of all Borel sets in Rn , denoted by B, is a σ−ring. In fact, B is the smallest σ−ring
containing all open sets.

ˆ If A ∈ M(µ), there exists Borel sets F and G such that F ⊂ A ⊂ G and F ⊂ A ⊂ G and
µ(G \ A) + µ(A \ F ) = 0.

ˆ Sets of measure zero form a σ−ring. There is a measure zero set that is uncoutable, such as
the Cantor set.

(P2) Prove that the Cantor set is of measure zero and uncountable, and all points of the Cantor
set are accumulate.

(D11) Suppose that X is a metric space. We say that X is a measure space, if there exists a σ−ring
M of subsets (measurable sets) of X and non-negative countably additive set function µ.

(D12) Let f : X → R ∪ {±∞} be a measurable function, where X is a measure space. We say that
f is measurable if {x : f (x) > c} is measurable for every c ∈ R.

ˆ we can see that if one of the following sets is measurable, then the other sets are measurable.

{x : f (x) > c}, {x : f (x) ≥ c}, {x : f (x) < c}, {x : f (x) ≤ c}.
∞
x : f (x) > c − 1i , {x : f (x) < c} = X \ {x : f (x) ≥ c},

Indeed, {x : f (x) ≥ c} = ∩ i=1
{x : f (x) ≤ c} = ∩∞ 1

i=1 x : f (x) < c + i , {x : f (x) > c} = X \ {x : f (x) ≤ c}.

ˆ If f is measurable, so is |f |.

4
 ˆ Let {fn } be a sequence of measurable functions. For each x ∈ X we define g(x) = supn fn (x)
and h(x) = lim supn→∞ fn (x). Then, g and h are measurable. Indeed, {x : g(x) > c} =
∪∞
n=1 {x : fn (x) > c} and h(x) = inf m gm (x), where gm (x) = supn≥m {fn (x)} for each x ∈ X.
As a consequence, if f and g are measurable, then max{f, g} and min{f, g} are measurable.
We denote f + = max{f, 0} and f − = − min{f, 0}. Thus, f + and f − are measurable. Also it
is direct that the limit of a convergent sequence of measurable funcions is measurable.
Theorem 0.4 Let f, g : X → R be measurable and F : R2 → R be continuous. We set h(x) =
F (f (x), g(x)). Then h is measurable. Thus, f + g and f g are measurable.
Indeed. Note that Gc = {(u, v) : F (u, v) > c} is open. Then, Gc = ∪∞ n=1 In , where In is an
open interval of the form In = {(u, v) : an < u < bn , cn < v < dn }. Since {an < f (x) < bn }
and {cn < g(x) < dn } are mesurable, it follows that {x : (f (x), g(x)) ∈ In } is measurable. Hence,
{h(x) > c} is measurable, because
{x : h(x) > c} = {x : (u, v) = (f (x), g(x)) ∈ Gc } = ∪∞
n=1 {x : (f (x), g(x)) ∈ In }.


ˆ The above result implies that f ◦ g is measurable if f is continuous and g is measurable.
However, f ◦ g is not necessarily mesurable, if f and g are just measurable, in fact, even if f
is measurable and g is continuous (see Appendix and compare to Riemann integral).
(D13) Let E ⊂ X. We define a characteristic function of E by

1 if x ∈ E
χE =
0 if x ∈/E
We say that s is a simple function, if s : X → R has finite range, i.e. {s(x)} is a finite set.
ˆ Suppose
Pn that s is a simple function and the range of s consists of c1 , c2 , · · · , cn . Then, s =
i=1 ci χEi , where Ei = {x : s(x) = ci }. If s is measurable, so is Ei for each i.

Theorem 0.5 If f : X → R, there is a sequence of {sn } of simple functions such that sn (x) → f (x)
as n → ∞, for every x ∈ X. If f is measurable, {sn } can be a sequence of measurable functions. If
f ≥ 0, {sn } can be a monotonicaly increasing sequence.
Indeed. Assume WLOG that f ≥ 0 (since f = f + − f − and positive and negative parts can be
treated separely). For fixed n = 1, 2, · · · , we introudce
En,i = {x : (i − 1)/2n ≤ f (x) < i/2n }, Fn = {x : f (x) ≥ n}, i = 1, 2, · · · , n2n .
n2 n
X i−1
We then define sn = n
χEn i + nχFn . If f is bounded, the converegence becomes uniform. 
i=1
2
Next, we introduce integration on a measurable space X with a σ−ring M and a measure µ.
Xn Z
(D14) Suppose that s(x) = ci χEi (x) is measurable and E ∈ M. We then define s(x)dµ =
i=1 E
n
X
IE (s) = ci µ(E ∩ Ei ). If f is measurable and nonnegative, we define
i=1
Z Z n
X
f dµ = sup s(x)dµ = sup IE (s) = sup ci µ(E ∩ Ei ), (0.5)
E 0≤s≤f E 0≤s≤f 0≤s≤f i=1

5
 n
X R
where s(x) = ci χEi (x) is any possible simple function with 0 ≤ s ≤ f . We call E
f dµ in
i=1
(0.5) the Lebesgue integral of f (comapre to notion of Riemann integral).
(D15) Let f = f + − f − and E be measurable. f or E f − is finite. Then, the
R + R
R Suppose R that E
by E f µ = E f +R− E f − . RWe say that f is integrable,
R
(Lebesgue) integral of f is defined
denoted by f ∈ L(µ) on E, if E f is finite (i.e. both E f + and E f − are finite).
R

ˆ E f can be defined, although E f isn’t integrable, i.e. E f could be ∞ or −∞.

R R R

ˆ Let f be measurable and bounded. If µ(E) < ∞,Rthen f ∈ L(µ) on E. Furthermore, if

a ≤ f (x) ≤ b on E and if µ(E) < ∞, then aµ(E) ≤ E f dµ ≤ bµ(E).
ˆ If f, g ∈ L(µ) and if f (x) ≤ g(x) for x ∈ E, then E f dµ ≤ E gdµ.
R R

ˆ If f ∈ L(µ) on E, then cf ∈ L(µ) on E and E cf dµ = c E f dµ.

R R

ˆ If µ(E) = 0, and f is measurable, then E f = 0.

R

ˆ If f ∈ L(µ) on E, A ∈ M, and A ⊂ E, then f ∈ L(µ) on A.

R
Theorem 0.6 Suppose that f is measurable on X. For A ∈ M, let φ(A) = A
f dµ. Then, φ is
countably additive on M.
Indeed. Suppose that f is non-negative. Let A = ∪∞ i=1 Ai , where Ai and Aj are mutually disjoint.
If f = χE is a characteristic function, then the countableP additivity of φ isPthe same as the countable
additivity of µ, i.e. φ(A) = A χE dµ = µ(A ∩ E) = ∞ ∞
R
i=1 µ(A i ∩ E) = i=1 φ(Ai ). If f is simple,
then RconclusionPis similar. ForPgeneral case, for any simple function s with 0 ≤ s ≤P f , we note
that A sdµ = ∞ ∞ R
R P∞ ∞
i=1 An sdµ ≤ i=1 An f dµ ≤ i=1 φ(A i ), which implies that φ(A) ≤ i=1 φ(Ai ).
For Preverse inequality,
R for givenR  > 0 there exists a simple function s such that 0 ≤ s ≤ f and
n −i
s = i=1 ci χAi with Ai f dµ ≤ Ai sdµ + 2 . It follows that
Z n Z
X n Z
X n
X
φ(∪ni=1 Ai ) ≥ sdµ = sdµ ≥ ( f dµ − 2−i ) ≤ φ(Ai ) − .
∪n
i=1 Ai i=1 Ai i=1 Ai i=1
P∞
Since  and n are arbitray, we get φ(A) ≥ φ(Ai ). In case f could be negative, we remind
i=1
f = f + − f − and treat f ± separately. 

ˆ If A ∈ M and B ⊂ A with µ(A \ B) = 0, then A f dµ = B f dµ. This means that sets of

R R

measure zero are negligible in Lebesgue integration.

(D16) We write f ∼ g on E if {x : f (x) 6= g(x)} ∩ E has measure zero. We say that f = g on E
almost everywhere (a.e.).
R R
Theorem 0.7 If f ∈ L(µ) on E, then |f | ∈ L(µ) on E and E f dµ ≤ E |f | dµ.
Indeed. Since f + and f − are integrable, E |f | dµ = E f + dµ + E f − dµ < ∞, which implies
R R R

|f | ∈ L(µ). Furthermore,
Z Z Z Z Z Z
+ − + −
f dµ = f dµ − f dµ ≤ f dµ + f dµ = |f | dµ.
E E E E E E

6
 ˆ If f is measurable on E, |f | ≤ g, and g ∈ L(µ) on E. Then, f ∈ L(µ) on E.

Theorem 0.8 (Lebesgue’s Monotone Convergence Theorem) Let E ∈ M. Suppose that {fn } be
a sequence of measurable functions such that 0 ≤ f1 (x) ≤ f2 (x) ≤ · · · and lim fn (x) = f (x) for
Z Z n→∞

x ∈ E. Then lim fn (x)dµ = f (x)dµ.

n→∞ E E
R R
Indeed.
R Suppose
R that E
f n (x)dµ tends to α as n → ∞. It is dierect that α ≤ E
f (x)dµ, since
f (x)dµ ≤ E f (x)dµ. Let s be a simple fucntion with 0 ≤ s(x) ≤ f (x). We fix c ∈ (0, 1). We
E n
set En R= {x : fn (x) ≥ cs(x)} and R we then see R that En is increasing
R and E = ∪∞ n=1 En . We note
that c E s(x)dµ ≤ α, because E fn (x)dµ ≥ En fn (x)dµ ≥ c En s(x)dµ for every n. The choice
R
Rof c is arbitraty, i.e. c could be chosen so close to 1, which implies that E s(x)dµ ≤ α, and thus
E
f (x)dµ ≤ α, since f can be approximated by simple functions. 

ˆ If {fn } be a sequence of non-ngeative measurable functions such that fi (x) ≥ fi+1 (x) and
lim fn (x) = f (x) for x ∈ E, in general, the integral and the limit may not be interchangeable.
n→∞

Theorem 0.9 (Fatou’s Theorem) Let E ∈ M. Suppose that {fn } be a sequence of non-negative Z
R
measurable functions such that lim inf fn (x) = f (x) for x ∈ E. Then E f (x)dµ ≤ lim inf fn (x)dµ.
n→∞ n→∞ E

Indeed. Let gn (x) = inf fi (x). It is direct that gn is non-negative and increasing, and gn (x) → f (x)
n≤i Z
as n goes to ∞. Due to MCT (Monotone Convergence Theorem), we have lim gn (x)dµ =
Z Z n→∞ E
R R R
f (x)dµ. Since E gn dµ ≤ E fn dµ, it is immediate that E f (x)dµ ≤ lim inf fn (x)dµ. 
E n→∞ E

Theorem 0.10 (Lebesgue’s Dominated Convergence Theorem) Let E ∈ M. Suppose that {fn } be
a sequence of measurable functions such that lim fn (x) = f (x) for x ∈ E. Assume further that
n→∞ Z Z
there is g ∈ L(µ) on E such that |fn (x)| ≤ g(x) for all n. Then lim fn (x)dµ = f (x)dµ.
n→∞ E E

Indeed.
R It is easy toR see that fn , f ∈ L(µ). Since fn + g ≥ 0, we can see via Fatou’s Theorem that
E
f dµ ≤ lim inf n→∞ E fn dµ, since
Z Z Z Z Z Z
f dµ + gdµ = (f + g)dµ ≤ lim inf (fn + g)dµ = lim inf fn dµ + gdµ.
E E E n→∞ E n→∞ E E
R R
On the other hand, again due to Fatou’s Theorem, we can obtain that E f dµ ≥ lim supn→∞ E fn dµ,
because g − fn ≥ 0 and
Z Z Z Z Z Z
gdµ − f dµ = (g − f )dµ ≤ lim inf (g − fn )dµ = gdµ − lim sup fn dµ.
E E E n→∞ E E n→∞ E
R R R
Combining estimates, we deduce that lim supn→∞ E
fn dµ ≤ E
f dµ ≤ lim inf n→∞ E
fn dµ. 
∞
X Z ∞ Z
X
ˆ If fn : E → R is non-negative and measurable, and f = fn , then f dµ = fn dµ.
n=1 E n=1 E

7
 ˆ There is an example of sequence of non-negative functions such that the inequality in Fatou’s
Theorem is really inequality, not equality. E.g. we take f = 0 on (0, 1) and fn (x) = nχ(0, 1 ) .
R1 R1 n

Then, it is direct that 0 = 0 f dµ < lim inf n→∞ 0 fn dµ = 1.

Next, we compare the Lebesgue integral with the Riemann Integral. In fact, Riemann integral
function is also Lebesgue integrable, but the converse is, in general, not true (e.g. f = χQ∩[0,1] is
not Riemann integrable but Lebesgue integrable) . Obviously, in that sense, the Lebesgue integral
covers much larger class of function. More serious issue is the many limit operations for Lebesgue
integral are available, but the Riemann integral is usually not the case. That is, the limit of Riemann
integral functions may fail to be Riemann integrable.
RFor
b
convenience, we consider only one dimensioal
Rb case and we denote the Riemann integral by
R a f dx (the Lebesgue integral is denoted by a f dx).

Theorem 0.11 If f is Riemann integrable on [a, b], f is Lebesgue integrable on [a, b].
Indeed. Let Pk be a partition of [a, b], i.e. Pk = {x0 , x1 , · · · , xn } with x0 = a and xk = b. We
define functions Lk and Uk as follows:

Uk (a) = Lk (a) = f (a), Uk (x) = Mi , Lk (x) = mi , xi−1 < x ≤ xi , i = 1, 2, · · · , n,

where Mi = max f (x) and mi = min f (x). If Pk+1 is a refinment of Pk , then, it is obvious
xi−1 <x≤xi xi−1 <x≤xi
that {Lk } is increasing and bounded above by f . Similarly, we can see that {Uk } is decreasing and
bounded below by f . Hence, we can define L(x) = lim Lk (x) and U (x) = lim Uk (x), which are
k→∞ k→∞
bounded and measurable, in addition L(x) ≤ f (x) ≤ U (x) for x ∈ [a, b]. If f is Riemann integrable,
we have by definition
Z b Z b Z b Z b Z b
lim Lk (x)dx = L(x)dx = R f (x)dx = U (x)dx = lim Uk (x)dx.
k→∞ a a a a k→∞ a

This equality means that L(x) = f (x) = U (x) for a. e. x ∈ [a, b], and thus f is measurable and
Lebesgue integrable, since L and U are Lebesgue integrable. 

Theorem 0.12 Let f be bounded on [a, b]. Then, f is Riemann integrable on [a, b] if and only if f
is continuous almost everywhere on [a, b].
Indeed. Let U and L be the functions constructed in the previous theorem. If x ∈ / Pk for any
k, it is direct that U (x) = L(x) if and only if f is continous at x. Since the union of the set of
Pk is countable, L(x) = U (x) a.e. if and only if f is continous almost everywhere. Thus, f is the
Riemann integrable if and only if f is continous almost everywhere. 
The following is a condition for the uniform convergence of a pointwise convergent sequence of
measurable functions, so called Severini–Egorov theorem (see e.g. [6]).

Theorem 0.13 Suppose that {fk } is a sequence of measurable functions in a set E with µ(E) < ∞
which converges a. e. to a finite limit f . Then, given  > 0 there is a closed subset F ⊂ E such
that µ(E \ F ) <  and {fk } converges uniformly to f on F .
Indeed. (Step 1) We first show that for any  and η there exists a closed set F ⊂ E and an integer
k such that |E \ F | < η and |f (x) − fn (x)| <  on F for n > k.
Truly, let Em = {x : |f (x) − fk (x)| < , k > m} for each m. Thus, Em is measurable, since
Em = ∩k>m {x : |f (x) − fk (x)| < }. It is direct that Em ⊂ Em+1 . Hence, Em increases to E \ Z

8
with |Z| = 0, since fk → f a. e. in E and f is finite. It is straightforward that |Em | → |E \Z| = |E|.
Since |E| < ∞, it follows that |E \ Em | → 0. Thus, |E \ Ek | < η/2 for large k and there is a closed
set F with F ⊂ Ek such that |Ek \ F | < η/2. Therefore, |E \ F | < η and |f (x) − fn (x)| <  for
x ∈ F if n > k.
(Step 2) For any  > 0 choose a closed set Fm and an integer km such that |E \ Fm | < 2−m
∞
and |f − fn | < 1/m on Fm for n > km . LetP F = ∩m=1 Fm is closed, and fn → f uniformly on F .
Furthermore, |E \ F | = | ∪m (E \ Fm )| ≤ m |E \ Fm | < . 

ˆ We note that µ(E) < ∞ is necessary, because fn = χ[ n, n + 1] converges to f = 0 a.e. but

| {|f − fk | > 1/2} | = 1 for any k. It is also necessary that f is finite a. e.

(D17) Let {fk } be a sequence of measurable and finite a. e. functons defined in E. We say that fk
converges in measure on E to f if limk→∞ |{x ∈ E : |f (x) − fk (x)| > }| = 0 for every  > 0.
We review the relation between pointwise convergence and convergence in measure (see e.g. [6]).

Theorem 0.14 (i) Suppose that {fk } is a sequence of measurable functions in a set E with
µ(E) < ∞ which converges a. e. to a measurable f a.e. on E. Furthermore, assume that fk
and f are finite a.e. Then, fk converges in measure on E to f .

(ii) If fk converges in measure on E to f , then there exists a subsequence, say fkj , such that fkj
converges a. e. to f .
Indeed. (i) Due to Egorov Theorem, for any  > 0 and for any η there is a closed set F ⊂ E such
that fk converges to f uniformly on F with µ(E \F ) < η. Therefore, {x ∈ E : |f (x) − fk (x)| > } ⊂
E \ F for sufficiently large k. Since η is arbitrary, {x ∈ E : |f (x) − fk (x)| > } converges to zero as
k tends to ∞.
(ii) Due to convergence in measure, there is a sequence of positive integer, {kj }, such that µ(Ej ) <
2 , where Ej = x ∈ E : |f − fkj | > j −1 . We set Hm = ∪∞
−j

j=m Ej . Note that Hm is decreasing and
−m+1 −1
µ(Hm ) ≤ 2 . Thus, if j ≥ m, then |f − fkj | ≤ j in E \ Hm , which implies that fkj converges
to f in E \ Hm . Since limm→∞ µ(Hm ) = 0, it is immediate that fkj converges to f a. e. 

ˆ fk converges in measure to f if and only if limi,j→∞ µ ({x ∈ E : |fi (x) − fj (x)| > }) = 0 for
any  > 0.

Theorem 0.15 (Tchebyshev’s inequality) Let f ≥ 0 be measurable on E. If α > 0, then

Z
1
µ ({x ∈ E : f (x) > α}) ≤ f dµ.
α E

Indeed. Let A := {x ∈ E : f (x) > α}. We then see that

Z Z Z
αµ(A) = α dµ ≤ f dµ ≤ f dµ,
A A E

where we used that f > α on A. 

9
1 Appendix
(A1) (Cantor set) Let I1 = (1/3, 2/3) and C1 = [0, 1] \ I1 = [0, 1/3] ∪ [2/3, 1]. Inductively, we define
Ck as follows: we divide each interval into three subintervals with the same length, and remove
the interior of middle, e.g. C2 = C1 \ ((1/9, 2/9) ∪ (7/9, 8/9)). We then define the Cantor set by
C = ∩∞ k=1 Ck .

ˆ It is direct that C is closed and of measure zero. It is uncountable and all points of C are
accumulation points (we call it a perfect set) (see e.g. [5].
(A2) (Cantor-Lebesgue function) Let Dk = [0, 1] \ Ck . Note that Dk consists of 2k − 1 open intervals
Ik,j , j = 1, 2, · · · , 2k −1, which is ordered from left to the right. Next, we define a continuous function
fk : [0, 1] → [0, 1] such that fk (0) = 0, fk (1) = 1, fk (x) = j2−k on Ik,j , and fk is linear on each
intervalPof Ck . Then it is direct that fk is increasing, fk+1 = fk on Ik,j , and |fk − fk+1 | < 2−k .
Thus, (fk − fk+1 ) converges uniformly on [0, 1], and so does fk . Set f (x) = limk→∞ fk (x). Then
f (0) = 0 and f (1) = 1 and f is increasing and continuous on [0, 1]. Furthermore, f is constant on
every interval removed in the process of construction (see e.g. [6] for the details).
ˆ (Composite function of measurable functions may not be measurable) We define g : [0, 1] →
[0, 2] by g(x) = x + f (x), where f is the Cantor-Lebesgue function. Note that g is strictly
increasing and continuous. Thus, the inverse of g, say h = g −1 , is continuous. Furthermore,
observe that |g(C)| = 1, since f (C) = [0, 1]. Hence, there is an non-measurable set A ⊂ g(C).
Let B := g −1 (A) ⊂ C. Then, B is of measure zero and thus is measurable. Let k = χB ◦ h :
[0, 2] → [0, 1]. Although χB is measurable and h is continuous, k is not measurable, because
{x : k(x) = 1} = {x : h(x) ∈ B} = A and A is non-measurable.

ˆ (Composite function of integrable functions may not be integrable) The above example is
a counter example, since composite function is not even measurable. However, if compsoite
function is measurable and bounded on a set of finite measure, it is integrable. If compsoite
function is measurable and finite a.e., it may not be integrable, e.g. f (x) = x2 and g(x) = x−1/2
on (0, 1), then f is continuous and g is integrable, but (f ◦ g)(x) = 1/x is not integrable on
(0, 1).
(A3) (non-measurable set) We recall the axiom of choice, which is stated as follows:
For every indexed family (Si )i∈I of nonempty sets there exists an indexed family (xi )i∈I of elements
such that xi ∈ Si for every i ∈ I. 1
Next, we claim that there is a non-measurable set (see [6, Theorem 3.38 (Vitali)]). Indeed, firstly
we note that if E ⊂ R is measurable and |E| > 0, then {d : d = x − y, x, y ∈ E} contains an interval
centered at the origin. For the time being, we agree with this claim. Next, for each x ∈ R we define
[x] = {x + r : r ∈ Q} and we then see that [x] = [y] or [x] ∩ [y] = ∅ for each x, y ∈ R. Due to the
axiom of choice, we choose one element from each equivalent class, and we set E the collection of
all choosen elements. Let x, y ∈ E and F = {x − y : x, y ∈ E}. Since F has no rational number,
and thus any interval cannot be contained in F . Due to the claim, if E is measurable, then |E| = 0.
On the other hand, R = ∪r∈Q E + r , which implies that |R| = 0, which is a contradiction. Thus, E
must be non-measurable.
Now it remains to show the claim. For fixed  > 0 there is an open set G with E ⊂ G and
|G| ≤ (1 + )|E|. Since G is open, G = ∪∞ j=1 Ij , where Ij is non-overlapping interval (some intervals

1
The axiom of choice was formulated in 1904 by Ernst Zermelo in order to formalize his proof of the well-ordering
theorem.

10
could be closed, not open). Set Ej = E ∩ Ij , and we then see E = ∪∞ i=1 Ei . We also note that
Ej is measurable and Ej has at most one common point with Ek . There exists Ek0 such that
|Ik0 | ≤ (1 + )|Ek0 | , because |G| ≤ (1 + )|E|. We denote F := Ek and J = Ik . We set  = 1/3. We
note that F ⊂ J and |J| < 34 |F |. Let d be the number such that |d| < 12 |J|. We define by Fd the
translated set by d of F . We claim that F ∩ Fd has a common point, i.e. F ∩ Fd 6= ∅. Suppose not
the case. Noting that F ∪ Fd ⊂ J ∪ Jd , it follows that |F ∪ Fd | ≤ |J| + d. By the hypothesis that
F ∩ Fd = ∅, we see, due to |J| < 34 |F |, that

3 1
|J| < |F | + |Fd | = |F ∪ Fd | ≤ |J| + |d| −→ |J| < |d|.
2 2
This is contrary to the assumption that |d| < 21 |J|. Therefore, (−d, d) ⊂ {d : d = x − y, x, y ∈ E}.

(A4) (Caratheodory Theorem) E is measurable if and only if |A|e = |A ∩ E|e + |A \ E|e for
any set A.
Indeed, suppose that E is measurable. For any set A there is a measurable set H with A ⊂ H and
|H| = |A|e . Since H = (H \ E) ∪ (H ∩ E), it follows that

|A|e = |H| = |H \ E| + |H ∩ E| ≥ |A \ E| + |A ∩ E| .

The opposite inequality is automatic, and thus |A|e = |A \ E| + |A ∩ E|. Conversely, firstly we
consider the case that |E|e < ∞. We choose a measurable set O such that E ⊂ O and |O| = |E|e .
By the hypothesis, |O| = |O \ E|e + |O ∩ E|e = |O \ E| + |E|e , which implies that |O \ E|e = 0.
Thus, O \ E is measurable and so is E. Now we suppose |E|e = ∞. We set Ek = E ∩ Bk , where Bk
is the ball centered at zero with its radius k. Then, |Ek |e < ∞ and we then take a mesurable set
Ok with Ek ⊂ Ok and |Ok | = |Ek |. Via the hypothesis,

|Ek |e = |Ok | = |Ok \ E|e + |Ok ∩ E|e ≥ |Ok \ E|e + |Ok ∩ Ek |e = |Ok \ E|e + |Ek |e ,

which yields |Ok \ E|e = 0 and so Ok \ E is measurable. Let O = ∪∞

k=1 Ok . Obviously, O is measur-
∞
able, and so is O \ E, because O \ E = ∪k=1 (Ok \ E). Since E = O \ (O \ E), E is measurable.

(A5) (Composite of Riemann integrable fucntions) In general, compsoite of Riemann integrable

fucntions may not be integrable. For example, f, g : [0, 1] → R, which are defined by

0 if x ∈ Qc ∩ [0, 1]
 
1 if x 6= 0
f (x) = , g(x) =
0 if x = 0 1
p
if x = pq ∈ Q ∩ [0, 1],

where Q is the set of rational numbers and (p, q) = 1. Then, it is direct that f ◦ g is not Riemann
integrable, since it becomes the Dirichlet function. However, it is known that f ◦ g is Riemann
integrable, in case that f is continiuous and g is Riemann integrable (see e.g. [3]). However, there
is the case that f ◦ g is not Riemann integrable, in case that g is continiuous and f is Riemann
integrable (see e.g. [4]).

11
References
[1] Folland, G. B., Real Analysis; Modern Techniques and Their Applications

[2] Kang, K., Lecture notes on Real Analysis.

[3] Lewin, J. and Lewin, M., An Introduction to Mathematical Analysis

[4] Lu, J., Mathematical Monthly, Vol 106, no 8, pp 763-766, 1999.

[5] Rudin, W., Principles of Mathematical Analysis

[6] Wheeden, R. L. and Zygmund, A., Measure and Integral ; An Intrroduction to Real Analysis

12