MATH4050 - CUHK 2022-23T2

Lecture Ch5: Differentiation and Integration

Mar. 2023
Lecturer: Prof. Ng Kung Fu Typed after each lecture by: ZHANG Xinfang

In this lecture, we will try to establish that:

d
Rx
• dx a
f (y)dy = f (x) a.e. on [a, b] if f ∈ L[a, b].
Rx F (x0 )−F (x)
In other words, if F (x) = a f for all x ∈ [a, b], then F 0 (x) := limx0 →
− x x0 −x exists
0
a.e. on [a, b] and F (x) = f (x) a.e. on [a, b].
Rx
• (Fundamental Theorem of Calculus) F (x) − F (a) = a F 0 (y)dy for all x ∈ [a, b] holds iff
F ∈ ABC[a, b] (the class of absolutely continuous functions on [a, b]).

Definition 0.1. f : [a, b] 7→ R is said to be absolutely continuous on [a, b] if ∀ > 0, ∃δ > 0

s.t ni=1 |f (x0i ) − f (xi )| <P whenever {(x0i , xi ) : i = 1, 2, . . . , n} is a finite collection of non-
P
overlapping intervals with ni=1 |x0i − xi | < δ.

Remark 0.2. The definition of absolute continuity can be extended in a broader context. For mea-
sures µ, m on M, we use µ << m to represent that µ is ABC w.r.t m: ∀ > 0, ∃δ > 0 s.t µ(E) < 
whenever m(E) < δ.
In preparation for our goals, we will proceed from Monotone Increasing functions, and then
Bounded Variations functions, and finally to the general case.

1
1 Vitali Cover, Two Runners, Simple Change of Variables
Definition 1.1. (i) An interval I is said to be non-degenerate if its length l(I) 6= 0.

(ii) Let E ⊆ R. By a Vitali cover of E, we mean a non-empty collection I of non-degenerate

intervals satisfying the following property: ∀x ∈ E, ∀ > 0, ∃I ∈ I s.t x ∈ I and l(I) < .

Throughout this section, we fix E ⊆ R and assume that E 6= ∅ to avoid the triviality.
The major two results of this section are under the name of Vitali.

Theorem 1.2 (Vitali Cover Lemma). Suppose m∗ (E) < +∞. Let I be a Vitali cover of E and
 > 0. Then ∃ finitely many pairwise disjoint members I1 , I2 , . . . , IN from I s.t

(∗) m∗ (E \ ∪N
i=1 Ii ) < 

and so

m∗ (E ∪N ∗
T
i=1 Ii ) > m (E) − 

(the 2nd inequality follows from the 1st by the contradiction argument as m∗ is subadditive).

Proof. By Quasi-Outer-Regularity, take open G ⊇ E s.t m(G) < +∞ (thanks to m∗ (E) < +∞).
Let
I0 := I ∈ I : closure I¯ ⊆ G .


Then I0 is a ’Vitali subcover’ of I0 for E. Let

I¯0 := I¯ : I ∈ I0

(not necessarily a subfamily of I)
= I¯ : I ∈ I and I¯ ⊆ G


Since I¯0 is also a Vitali Cover of E and (∗) is equivalent to m∗ (E \ ∪N ¯

i=1 Ii ) < , we may assume
for our proof that each member of I is closed and contained in G (replace I by I¯0 if necessary).
Moreover, we assume that @ finitely many disjoint members from I covering E entirely (otherwise,
(∗) holds trivially).
Under this additional assumption, we will construct an infinite seq. {In : n ∈ N} of disjoint
members of I as follows:

• Pick I1 ∈ I and let k1 := sup l(I) : I ∈ I, disjoint from I1 . Then 0 < k1 ≤ m(G) < +∞
(recall that E * I1 ).

• Take I2 ∈ I, disjoint from I1 s.t k21 < l(I2 ) (so l(I) ≤ k1 < 2l(I2 ) ∀I ∈ I disjoint from I1 ).

• Set k2 := sup l(I) : I ∈ I, disjoint from I1 , I2 and take I3 ∈ I disjoint from I1 , I2 s.t
(l(I) ≤)k2 < 2l(I3 )∀I ∈ I disjoint from I1 , I2 .

• ......

2
 • Inductively, having disjoint I1 , I2 , . . . , In ∈ I and k1 , k2 , . . . , kn−1 with (l(I) ≤)kk <
2l(Ik+1 )∀I ∈ Idisjoint from I1 , I2 , . . . , Ik (k = 1, 2, . . . , n − 1).
Set kn := sup l(I) : I ∈ I, disjoint from I1 , . . . , In and take In+1 ∈ I disjoint from
I1 , . . . , In s.t (l(I) ≤)kn (< 2l(In+1 ))∀I ∈ I disjoint from I1 , . . . , In .
∞
P∞
Thus we have (infinite) P∞ sequence {I n }n=1 of disjoint members of I. Note that i=1 l(Ii ) ≤
m(G) < +∞ and so i=1 ki < +∞.
Let Ibi denote the closed interval with the same center as that of Ii and with length l(Ibi ) = 5l(Ii ).
Then ∀N ∈ N, one has
S∞  P∞ 
(#) E \ N
S
I
i=1 i ⊆ I
i=N +1 i
b of total length ≤ 5 i=N +1 l(Ii )
To see this, let x ∈ LHS, take I ∈ I containing x, disjoint from the closed I1 , . . . , IN . Note that
I must intersect someone from the seq. {In }n∈N
otherwise l(I) ≤ kn ∀n, against the facts that
I ∈ I is non-degenerate and kn →− 0 as n → − ∞ .
By the well-ordering property of N, it follows that ∃j ∈ N s.t
I disjoint from I1 , . . . , IN , IN +j−1 but I ∩ IN +j 6= ∅


so l(I) ≤ kN +j < 2l(IN +j ) and x ∈ I ⊆ Id N +j ,

implying that x ∈ RHS of (#) and so (#) is true.

Finally, with the any given  > 0, ∃N ∈ N s.t ∞
P
i=N +1 l(Ii ) < 
b
Then by (#), m∗ (E \ ∪N i=1 Ii ) < . So (∗) is shown.

Theorem 1.3 (Vitali Cover Theorem). Let m∗ (E) < +∞ and I be a Vitali cover of E. Then ∃
a countable (finite or infinite) pairwise disjoint members from I ’almost cover E’: ∃N0 ⊆ N and
disjoint {Ii : i ∈ N0 } ⊆ I s.t

m∗ (E \ ∪i∈N0 Ii ) = 0 and m∗ (E ∪i∈N0 Ii ) = m∗ (E).

T

Proof. If E is covered by finitely many disjoint members from I, then the asserted statement is
true, so suppose E cannot be covered by any finite disjoint subfamily, hence, we have infinite
seq. (In ) in the proof of the preceding lemma; consequently, m∗ (E \ ∪∞ I ) < ∀ > 0 and
o i=1 i
∗ ∞ ∗ ∞ ∗
T

m (E \ ∪o i=1 Ii ) = 0 so m (E ∪o i=1 Ii ) = m (E) .
Remark 1.4. With some further work, it can be shown (not straightforward) that Theorem 1.3 holds
with N0 = N, that is, one can struct an infinite sequence in the statement of Theorem 1.3. Since
the stronger version is not needed for our subsequent study, we don’t provide a proof here.
Rx Rx
Lemma 1.5 (Two Runners Lemma). Let f, g ∈ L[a, b] s.t a f = a g for all x ∈ [a, b]. Then
f = g a.e. on [a, b].

Proof. Since f, g ∈ L[a, b], we assume WLOG R xg(x) ∈R xR∀x ∈ [a, b] and note that
R x that f (x),
R(fx − g) R+x g = f on [a, b] andRhence
x
that a (f − g) + a g = a f ∀x. By the assumption
a
f = a
g ∀x, it follows that a
(f − g) = R0 ∀x. Thus, replacing f by f − g if necessary, we
x
only need to consider the case when g = 0, so a f = 0 ∀x.

3
 R R
By the countable linearity, we then have I f = 0 ∀ interval I ⊆ [a, b] and then G f =R0 ∀
open set G ⊆ [a, b] thanks to the structure
theorem for open sets and the fact that E 7→ E f
(∀E ⊆ [a,Rb] and E ∈ M ) is a ’measure’ . 
SinceR E f = 0 whenever m(E) = 0 and E ⊆ [a, b], it follows that E ⊆ [a, b] : E ∈
M and R E f = 0 ⊇ M0 is a σ-algebra containing τ R, so contains B Rand so equal the+whole M,
i.e., E f = 0 ∀E ⊆ [a, b], E ∈ M. In particular, E + f = 0 = E − f , where E := {x ∈
[a, b] : f (x) > 0} and E − := {x ∈ [a, b] : f (x) < 0}. By the partial converse of the monotonicity,
m(E + ) = 0 = m(E − ). Hence, f = 0 a.e.
Lemma 1.6 (Simple Change of Variables). Let f ∈ L[a + c, b + c] with c ∈ R. Then
R b+c Rb
(?) a+c
f (y)dy = a f (x + c)dx.
Proof. Progressively consider the following three cases for f ∈ L[a + c, b + c] where c ∈ R:
f is a simple function. i.e. f is representable in the form f = ni=1 ai XEi +c , so f (x + c) =
1o . P
P
n
i=1 ai XEi with each ai ∈ R, Ei ∈ M and Ei ⊆ [a, b].

2o . 0 ≤ f ∈ L[a + c, b + c].
3o . f ∈ L[a + c, b + c] with f = f + − f − and 0 ≤ f + , f − ∈ L[a + c, b + c].
For each of these cases, we check (?):
1o . Note that
R b+c Rb
a+c
XEi +c = m(Ei ) = a
XEi ∀i

and so
R b+c Pn Pn R b+c Pn Rb R b Pn
a+c
( i=1 a i X Ei +c ) = i=1 ai a+c
XEi +c = i=1 a i a
XEi
= a i=1 ai XEi .

2o . By Littlewood’s 2nd principle. ∃ a seq. of simple functions ϕn ∈ MF + [a + c, b + c] s.t

R b+c R b+c
0 ≤ ϕn ↑n f (and so by MCT, a+c ϕn ↑n a+c f ).
Rb Rb
Note that 0 ≤ ϕn (x+c) ↑n f (x+c) for all x ∈ [a, b] and so a ϕn (x+c)dx ↑n a f (x+c)dx,
R b+c Rb
it follows from 1o that a+c f = a f (x + c)dx. i.e. (?) holds in this case.
3o . By 2o , we have
Z b+c Z b
+
f = f + (x + c)dx
a+c a
Z b+c Z b
f− = f − (x + c)dx
a+c a

Adding up, we arrive at (?) ∀f ∈ L[a + c, b + c].

4
2 Monotone Functions
The goal of this section is: For an increasing function f defined on [a, b], we have

(i) f 0 (x) exists a.e. in R∗ := [−∞, +∞]

(ii) f 0 ∈ L[a, b] and so f 0 (x) ∈ R a.e.

Rb
(iii) a f 0 ≤ f |ba
f (x+h)−f (x)
Definition 2.1. For a function f : [a, b] 7→ R and x ∈ [a, b], f 0 (x) = limh→
−0 h
if the
limit exists. This is the case iff the following 4 limits are equal
f (x+h)−f (x)
• D+ f (x) = lim suph→
− 0+ h

f (x+h)−f (x)
• D+ f (x) = lim inf h→
− 0+ h

f (x+h)−f (x)
• D− f (x) = lim suph→
− 0− h

f (x+h)−f (x)
• D− f (x) = lim inf h→
− 0− h

In general −∞ ≤ D+ f (x) ≤ D+ f (x) ≤ +∞ and −∞ ≤ D− f (x) ≤ D− f (x) ≤ +∞. Thus

{D− f < D− f } ∪ {D− f < D+ f } ∪ {D+ f < D− f } ∪ {D+ f < D+ f }

equals to the set of all non-differentiable points of f f 0 (x) does not exists in R∗ .

Theorem 2.2. Let f be an increasing function defined on [a, b]. Then f is differentiable a.e. on
[a, b]. (i.e. f 0 (x) exists a.e. in R∗ ).

Proof. Extend f to be defined on R by f (x) = f (b) for all x > b and f (x) = f (a) for all x < a.
By similarity, we need only show that
m(E) = 0, where E := {x : D− f (x) < D+ f (x)}.
Since E is a countable union of sets of the form
Eu,v = {x : D− f (x) < v < u < D+ f (x)} for u, v ∈ Q and v < u.
We only need to show that s := m∗ (Eu,v ) = 0.
To do this, let  > 0 and take open O ⊇ Eu,v s.t m(O) < s + .
Let x ∈ Eu,v and note that D− f (x) < v, then ∃ an arbitrarily small interval [x − h, x] ⊆ O s.t

f (x) − f (x − h)
<v (1)
h

So the collection Cx = {I ⊆ O : I = [x − hxS , x] s.t f (x)−fhx(x−hx ) < v} is a Vitali cover of {x}. Do

this for all x ∈ Eu,v and it follows that C := x∈Eu,v Cx is a Vitali cover of Eu,v .
By Vitali Covering Lemma, there exists a finite disjoint collection {I1 , I2 , . . . , IN } of intervals in

5
C s.t m∗ (Eu,v \ N In ) <  and m∗ (Eu,v ∩ N ∗
S S
n=1
SN o n=1 In ) > m (Eu,v ) −  = s − .
Take A = Eu,v ∩ n=1 In , where In := [xn − hn , xn ], Ino = (xn − hn , xn ). Then m∗ (A) > s − .
For any y ∈ A, ∃ an arbitrarily small interval [y, y + k] ⊆ Ino for some n s.t
f (y + k) − f (y)
u< (2)
k
Similarly, by Vitali Covering
∗
SMLemma, there exists a finite disjoint collection {J1 , J2 , . . . , JM } of
such intervals s.t m (A ∩ m=1 Jm ) > (s − ) −  = s − 2, each of which is of the form
Ji = [yi , yi + ki ] contained in one and only one In with property (??): u < f (yi +kki )−f
i
(yi )
.
M
X M
X
Hence, u(s − 2) ≤ u ki < [f (yi + ki ) − f (yi )]
i=1 i=1
N
Xn X o
≤ [f (yi + ki ) − f (yi )]
n=1 Ji ⊆In
N
X
≤ [f (xn ) − f (xn − hn )] (since f is increasing)
n=1
XN
< vhn ≤ vm(O) < v(s + ) (by (??) and the definition of O)
n=1

Letting  →
− 0 and since v < u, it follows that s = 0.
Therefore, f 0 (x) exists a.e. in R∗ .
Theorem 2.3 (Differentiation Theorem of Monotone RFunctions). Under the same assumptions in
b
Theorem 2.2, f 0 ∈ L[a, b] and so f 0 (x) ∈ R a.e., and a f 0 ≤ f |ba .
f (x+h)−f (x)
Proof. By Theorem 2.2, g(x) := limh→ −0 h
∈ [−∞, +∞] exists a.e. on [a, b]. Thus
1
f (x+ n )−f (x)
gn →
− g a.e. on [a, b], where gn (x) := 1 for all x ∈ R.
n
Clearly, each gn is measurable and gn ≥ 0. It follows Fatou’s Lemma that
Z b Z b
0≤ g(x)dx ≤ lim inf g
a n→−∞ a n
n Z b 1
Z b o
= lim inf n[ f (x + )dx − f (x)dx]
n→−∞ a n a
1
n Z b+ n Z b o
= lim inf n f −n f (change of variable rule)
n 1
a+ n a
1 1
n Z b+ n Z a+ n o
= lim inf n f −n f ≤ f (b) − f (a) (f is increasing )
n b a

In particular, g is integrable, and hence g(x) is finite a.e.

Consequently, we have g(x) = f 0 (x) ∈ R a.e.

6
Corollary 2.4. (i) If f : [a, b] 7→ R can be written as f = f1 − f2 , where f1 , f2 are increasing
functions on [a, b], then f 0 (x) exists in R a.e. and f 0 (x) ∈ L[a, b].
Rx
(ii) Let g ∈ L[a, b] and f (x) = a g ∀x ∈ [a, b]. Then f 0 (x) exists in R a.e., and f 0 ∈ L[a, b].

Proof. (i) f 0 (x) = f10 (x) − f20 (x).

(ii) Noting g = g + − g − , apply (i).

3 Functions of Bounded Variations

Definition 3.1. f : [a, b] 7→ R is said to be bounded variation if ∃M ∈ R, M > 0 s.t
Pn
i=1 |f (xi ) − f (xi−1 )| ≤ M whenever {a = x0 < · · · < xn−1 < xn = b} is a partition of [a, b].

And we use BV F [a, b] to denote the set of functions of bounded variation defined on [a, b].
For f : [a, b] 7→ R, define

• Tab (f ) := sup ni=1 |f (xi ) − f (xi−1 )| ≤ +∞

P

• Pab (f ) := sup ni=1 {f (xi ) − f (xi−1 )}+ ≤ +∞

P

• Nab (f ) := sup ni=1 {f (xi ) − f (xi−1 )}− ≤ +∞

P

where the sup over all partitions of [a, b]. Sometimes one uses Tab , Pab , and Nab to represent
Tab (f ), Pab (f ), and Nab (f ) respectively.

Lemma 3.2. Let f : [a, b] 7→ R and a < c < b. Then with x ∈ [a, b], one has:

(i) Tax (f ) ↑x , Pax (f ) ↑x , and Nax (f ) ↑x .

(ii) Tac (f ) + Tcb (f ) = Tab (f ) ∀c ∈ [a, b] and similar results for Pab (f ), Nab (f ) .

(iii) Tab (f ) = Pab (f ) + Nab (f ).

(iv) f (x) − f (a) = Pax (f ) − Nax (f ) provided that f ∈ BV F [a, b].

(v) If f ∈ BV F [a, b], then f 0 (x) exists in R a.e., and f 0 ∈ L[a, b].
Rx Rb
(vi) If g ∈ L[a, b] and G(x) := a g ∀x ∈ [a, b], then Tab (G) ≤ a |g| and G ∈ ABC[a, b].

Proof. (i), (ii), (v), (vi) are simple exercises.

7
(iii) For any partition of [a, b], note that
n
X n
X
Tab ≥ {f (xi ) − f (xi−1 )}+ + {f (xi ) − f (xi−1 )}−
i=1 i=1
n
X n
X
+
= {f (xi ) − f (xi−1 )} + {[f (xi ) − f (xi−1 )]+ − [f (xi ) − f (xi−1 )]}
i=1 i=1
X n
=2 {f (xi ) − f (xi−1 )}+ − [f (b) − f (a)]
i=1
n
X
=2 {f (xi ) − f (xi−1 )}+ − (Pab − Nab )
i=1

Then we have Tab ≥ 2Pab − (Pab − Nab ) = Pab + Nab . On the other hand,

n
X n
X
Pab + Nab ≥ +
{f (xi ) − f (xi−1 )} + {f (xi ) − f (xi−1 )}−
i=1 i=1
n
X
= {f (xi ) − f (xi−1 )}+ + {f (xi ) − f (xi−1 )}−
i=1
Xn
= |f (xi ) − f (xi−1 )|
i=1

Then we have Pab + Nab ≥ Tab . Therefore, Tab = Pab + Nab .

(iv) It is sufficient to consider the case x = b. For any partition of [a, b],
n
X n
X
{f (xi ) − f (xi−1 )}+ − {f (xi ) − f (xi−1 )}−
i=1 i=1
n
X
= {f (xi ) − f (xi−1 )} = f (b) − f (a)
i=1
Pn
Then i=1 {f (xi ) − f (xi−1 )}+ ≤ Nab (f ) + [f (b) − f (a)] and it follows that Pab (f ) ≤ Nab (f ) +
[f (b) − f (a)].
Similarly, Pab (f ) ≥ Nab (f ) + [f (b) − f (a)] and hence f (b) − f (a) = Pab − Nab .

Lemma 3.3. ABC[a, b] ⊆ BV F [a, b]

Proof. Let f ∈ ABC[a, b], and  = 1.

Then by definition of absolute continuity, ∃δ > 0 (depends on ) s.t m 0
P
i=1 |f (xi ) − f (xi )| < 1
whenever {(x0i , xi ) : i = 1, 2, . . . , m} is a finite collection of non-overlapping intervals with total
length < δ.

8
 Take N ∈ N s.t b−a N
< δ. Divide [a, b] into N -many subintervals of equal length (= b−a N
): a =
i
y0 < y1 < · · · < yN = b, with yi = a + N (b − a), i = 1, 2, . . . , N .
Then with any fixed i, Tyyi−1
i
(f ) ≤ 1 because any partition of [yi−1 , yi ] is of total length b−a
N
< δ.
b
PN yi
Therefore, it follows from the preceding lemma that Ta (f ) = i=1 Tyi−1 ≤ N .
Theorem 3.4. A function is of bounded variation on [a, b] iff it is the difference of 2 increasing
functions on [a, b]. (ABC[a, b] ⊆ BV F [a, b] = {F1 − F2 : F1 , F2 increasing on [a, b]})
Proof. Make use of the previous Lemma.
( =⇒ part): If f ∈ BV F [a, b], then f ∈ BV F [a, x] for all x ∈ [a, b].
By the Lemma, we have f (x) − f (a) = Pax − Nax , where x 7→ Pax , Nax are clearly increasing.
( ⇐= part): BV F [a, b] is a vector space containing all increasing functions. The result is clear.
Corollary 3.5. If f is of bounded variation on [a, b], then f 0 (x) exists a.e. and f 0 ∈ L[a, b].
Exercise 3.6. a. Check (i), (ii), (v), (vi) of Lemma 3.2.
b. Apply Theorem 4.1 of the next section and show that in Lemma 3.2 (vi),
Rb
(#) Tab (G) = a |g|.
Hints.
a. (vi) Let Π := {a = x0 < x1 < · · · < xn−1 < xn = b} ∈ P ar[a, b]. Noting
Pn Pn R xi Pn R xi Rb
i=1 |G(xi ) − G(xi−1 )| = i=1 | xi−1 g| ≤ i=1 xi−1 |g| = a |g|,

Rb
one has Tab (G) ≤ a
|g|.
Rb
b. For (#), we only need to show Tab (G) ≥ a |g|.
Rb
By Lemma 3.3 (vi), G ∈ BV F [a, b] with Tab (G) ≤ a |g| < ∞ and by Lemma 3.2 (iv), we
have G(x) − R xG(a) = Pax (G) − Nax (G) ∀x ∈ [a, b]. By Theorem 4.1 of the next section,
d d d d d d
g(x) = dx a g = dx G = dx (Pax (G)) − dx (Nax (G)). Since dx (Pax (G)), dx (Nax (G)) ≥ 0
(since Pax (G), Nax (G) are increasing), it follows that
d +  d − d x
+ x x
g ≤ Pa (G) + Na (G) = P (G)
dx dx dx a
d −  d + d x
g− ≤ Pax (G) + Nax (G) = N (G)
dx dx dx a
and hence, by Theorem 2.3,
Z b Z b
+ d
g ≤ (Pax (G)) ≤ Pab (G)
a a dx
Z b Z b
d
g− ≤ (Nax (G)) ≤ Nab (G)
a a dx
Rb
implying by Lemma 3.2(iii) that a |g| ≤ Pab (G) + Nab (G) = Tab (G).

9
4 Fundamental Theorem of Calculus
d
Rx
Theorem 4.1. Let f ∈ L[a, b]. Then dx a
f (y)dy
R x = f (x) a.e. on [a, b].
0
(i.e. F (x) = f (x) a.e. on [a, b], where F (x) = a f ).
Proof. By Corollary 2.4 (ii), F 0 exists a.e. on [a, b] and F 0 ∈ L[a, b]. In particular, for a.e. x, F 0 (x)
exists in R. Then
1
F (x0 )−F (x) F (x+ n )−F (x)
F 0 (x) = limx0 →
−x x0 −x
= limn→
−∞ 1/n
= limn→
− ∞ Fn (x),
F (x+ 1 )−F (x)
where Fn (x) := n
1/n
.
By Two Runners Lemma, it suffices to show that
Z b0 Z b0
0
F = f for any b0 ∈ [a, b] (†)
a a

Case 1: f is bounded.
R x+ 1 R x+ 1
If f is bounded, say |f | ≤ M , then |Fn (x)| = |n x n f | ≤ n x n |f | ≤ M . By BCT,
Z b0 Z b0
LHS of (??) = (lim Fn (x))dx = lim Fn (x)dx
a n n a
R b0 R b0
a
F (x + n1 )dx − a F (x)dx
= lim
n 1/n
R b0 + n1 R b0
1 F (x)dx − a F (x)dx
a+ n
= lim
n 1/n
R b0 + n1 R a+ 1
b0 F (x)dx − a n F (x)dx
= lim
n 1/n
0
= F (b ) − F (a) (by continuity of F )
Z b0
= f (x)dx = RHS of (??)
a

Case 2: General case: f ∈ L[a, b].

WLOG, f ≥ 0 (because f = f + − f − ).
Take fn := f ∧ n for all n ∈ N. ThenRfn ↑n f and fn is bounded.
d x
ByRapplying Case 1 to fn , one has dx a
fn = fn (x) forRa.e. x ∈ [a, b]. R
d x d
R x d
R x d x d x Rx
dx a
f = dx a
(f − f n ) + f n = dx a
(f − f n ) + dx a
f n ≥ 0 + f
R x dx ad R xn since a
(f − fn ) is
an increasing function by non-negativity of f − fn . Hence, dx d
a
f ≥ dx a n
f , that is F 0 (x) ≥
R b0 0 R b0
fn (x) ↑n f (x) for a.e. x ∈ [a, b]. Then a F (x)dx ≥ a f (x)dx. The converse inequality also
R b0 R b0
holds because a F 0 (x)dx ≤ F (b0 ) − F (a) = a f by Theorem 2.3 as F is increasing.
Hence, F 0 (x) = f (x) a.e. on [a, b] by the partial converse of the monotonicity.
Proposition 4.2. Suppose f ∈ ABC[a, b] s.t f 0 (x) = 0 a.e. Then f is a constant function.

10
 Proof. WLOG, it suffices to show f (a) = f (b). Let , η > 0.
Note that f is ABC, for the , ∃δ > 0 s.t the variation of f <  as corresponding finitely many
non-overlapping intervals of total length < δ.
G := {x ∈ (a, b) : f 0 (x) = 0} of measure = b − a.
0
For any x ∈ G, Cx := {(x, x0 ) : ∃[x, x0 ] ⊆ [a, b] of positive length s.t | f (xx)−f 0 −x
(x)
| < η}.
∪x∈G Cx is a Vitali Cover of G. By Vitali Covering Lemma, there exist disjoint (xi , x0i ) for i =
1, 2, . . . , n s.t |f (xi ) − f (x0i )| < η|xi − x0i | for all i with [a, b] \ ∪ni=1 [xi , x0i ] is of measure < δ.
Rearrange indices if necessary, we assume that
0
a = x0 < x1 < x01 < x2 < x02 < · · · < xn < x0n < b
0 0
and note that [x0 , x1 ] ∪ [x01 , x2 ] ∪ . . . , [xn−1 , xn ] ∪ [x0n , b] is of total length < δ and so
0
|f (b) − f (x0n )| + ni=1 |f (xi−1 ) − f (xi )| < 
P

and
n n
0
X X
|f (b) − f (a)| ≤ |f (b) − f (x0n )| + |f (xi−1 ) − f (xi )| + |f (xi ) − f (x0i )|
i=1 i=1
<  + (b − a)η

which implies that f (b) = f (a) as , η > 0 are arbitrary.

Theorem 4.3 (Fundamental

Rx Theorem of Calculus). Let F : [a, b] 7→ R. Then F ∈ ABC[a, b] iff
F 0 ∈ L[a, b] and a F 0 = F (x) − F (a) ∀x ∈ [a, b].

Proof.
( ⇐= part): The absolute continuity is from Theorem 4.10 in Lecture 02.
( =⇒ part): Suppose F R∈ ABC[a, b]. Then F ∈ BV F [a, b], F 0 exists a.e. and F 0 ∈ L[a, b].
x 0 x
Define φ(x) := F (x) − a F for all x ∈ [a, b]. φ0 (x) = F 0 (x) − dx
d
F 0 = F 0 (x) − F 0 (x) = 0
R
a
a.e. on [a, b] by TheoremR4.1. Then by Proposition 4.2, φ is a constant function. Thus, φ(x) =
x
φ(a) = F (a). Therefore, a F 0 = F (x) − F (a) ∀x ∈ [a, b].
Rb Rb
Theorem 4.4 (Integration by Parts). Suppose F, G ∈ ABC[a, b]. Then a F 0 G = (F G)|ba − a F G0

Proof. Since F, G ∈ ABC[a, b], one can easily show that F G ∈ ABC[a, b].
Then by Theorem 4.3, (F G)0 exists a.e. on [a, b] and (F G)0 ∈ L[a, b]. Note that (F G)0 = F 0 G +
Rb Rb Rb Rb
F G0 and so, by Theorem 4.1, (F G)|ba = a (F G)0 = a (F 0 G + F G0 ) = a F 0 G + a F G0 .

11