Chapter 3.

Absolutely Continuous Functions

§1. Absolutely Continuous Functions

A function f : [a, b] → IR is said to be absolutely continuous on [a, b] if, given

ε > 0, there exists some δ > 0 such that

n
X
|f (yi ) − f (xi )| < ε,
i=1

whenever {[xi , yi ] : i = 1, . . . , n} is a finite collection of mutually disjoint subintervals of

Pn
[a, b] with i=1 |yi − xi | < δ.
Clearly, an absolutely continuous function on [a, b] is uniformly continuous. Moreover,
a Lipschitz continuous function on [a, b] is absolutely continuous. Let f and g be two
absolutely continuous functions on [a, b]. Then f +g, f −g, and f g are absolutely continuous
on [a, b]. If, in addition, there exists a constant C > 0 such that |g(x)| ≥ C for all x ∈ [a, b],
then f /g is absolutely continuous on [a, b].
If f is integrable on [a, b], then the function F defined by
Z x
F (x) := f (t) dt, a ≤ x ≤ b,
a

is absolutely continuous on [a, b].

Theorem 1.1. Let f be an absolutely continuous function on [a, b]. Then f is of bounded
variation on [a, b]. Consequently, f 0 (x) exists for almost every x ∈ [a, b].

Proof. Since f is absolutely continuous on [a, b], there exists some δ > 0 such that
Pn
i=1 |f (yi ) − f (xi )| < 1 whenever {[xi , yi ] : i = 1, . . . , n} is a finite collection of mu-
Pn
tually disjoint subintervals of [a, b] with i=1 |yi − xi | < δ. Let N be the least inte-
ger such that N > (b − a)/δ, and let aj := a + j(b − a)/N for j = 0, 1, . . . , N . Then
a
aj − aj−1 = (b − a)/N < δ. Hence, ∨ajj−1 f < 1 for j = 0, 1, . . . , N . It follows that

b aj
N _
_ X
f= f < N.
a j=1 aj−1

This shows that f is of bounded variation on [a, b]. Consequently, f 0 (x) exists for almost
every x ∈ [a, b].

1
Theorem 1.2. If f is absolutely continuous on [a, b] and f 0 (x) = 0 for almost every
x ∈ [a, b], then f is constant.

Proof. We wish to show f (a) = f (c) for every c ∈ [a, b]. Let E := {x ∈ [a, c] : f 0 (x) = 0}.
Pn
For given ε > 0, there exists some δ > 0 such that i=1 |f (yi ) − f (xi )| < ε whenever
{[xi , yi ] : i = 1, . . . , n} is a finite collection of mutually disjoint subintervals of [a, b] with
Pn 0
i=1 |yi − xi | < δ. For each x ∈ E, we have f (x) = 0; hence there exists an arbitrary
small interval [ax , cx ] such that x ∈ [ax , cx ] ⊆ [a, c] and

|f (cx ) − f (ax )| < ε(cx − ax ).

By the Vitali covering theorem we can find a finite collection {[xk , yk ] : k = 1, . . . , n} of

mutually disjoint intervals of this sort such that

λ E \ ∪nk=1 [xk , yk ] < δ.

Since λ([a, c] \ E) = 0, we have

λ [a, c] \ ∪nk=1 [xk , yk ] = λ E \ ∪nk=1 [xk , yk ] < δ.

Suppose a ≤ x1 < y1 ≤ x2 < · · · < yn ≤ c. Let y0 := a and xn+1 := c. Then

n
X
(xk+1 − yk ) = λ [a, c] \ ∪nk=1 [xk , yk ] < δ.

k=0

Consequently,
n
X
|f (xk+1 ) − f (yk )| < ε.
k=0

Furthermore,
n
X n
X
|f (yk ) − f (xk )| < ε(yk − xk ) ≤ ε(c − a).
k=1 k=1

It follows from the above inequalities that

n
X n
X
|f (c) − f (a)| ≤ |f (xk+1 ) − f (yk )| + |f (yk ) − f (xk )| < ε(c − a + 1).
k=0 k=1

This shows that |f (c) − f (a)| ≤ ε(c − a + 1) for all ε > 0. Therefore, f (c) = f (a).

2
 §2. The Fundamental Theorem of Calculus

In this section we show that absolutely continuous functions are precisely those func-
tions for which the fundamental theorem of calculus is valid.

Theorem 2.1. If f is integrable on [a, b] and

Z x
f (t) dt = 0 ∀ x ∈ [a, b],
a

then f (t) = 0 for almost every t ∈ [a, b].

Proof. By our assumption,

Z d
f (t) dt = 0
c

for all c, d with a ≤ c < d ≤ b. If O is an open subset of [a, b], then O is a countable union
of mutually disjoint open intervals (cn , dn ) (n = 1, 2, . . .); hence,
Z ∞ Z
X dn
f (t) dt = f (t) dt = 0.
O n=1 cn

It follows that for any closed subset K of [a, b],

Z Z Z
f (t) dt = f (t) dt − f (t) dt = 0.
K [a,b] [a,b]\K

Let E+ := {x ∈ [a, b] : f (x) > 0} and E− := {x ∈ [a, b] : f (x) < 0}. We wish to show
that λ(E+ ) = 0 and λ(E− ) = 0. If λ(E+ ) > 0, then there exists some closed set K ⊆ E+
R
such that λ(K) > 0. But K f (t) dt = 0. It follows that f = 0 almost everywhere on K.
This contradiction shows that λ(E+ ) = 0. Similarly, λ(E− ) = 0. Therefore, f (t) = 0 for
almost every t ∈ [a, b].

Theorem 2.2. If f is integrable on [a, b], and if F is defined by

Z x
F (x) := f (t) dt, a ≤ x ≤ b,
a

then F 0 (x) = f (x) for almost every x in [a, b].

Proof. First, we assume that f is bounded and measurable on [a, b]. For n = 1, 2, . . ., let

F (x + 1/n) − F (x)
gn (x) := , x ∈ [a, b].
1/n

3
It follows that Z x+1/n
gn (x) = n f (t) dt, x ∈ [a, b].
x

Suppose |f (x)| ≤ K for all x ∈ [a, b]. Then |gn (x)| ≤ K for all x ∈ [a, b] and n ∈ IN. Since
limn→∞ gn (x) = F 0 (x) for almost every x ∈ [a, b], by the Lebesgue dominated convergence
theorem, we see that for each c ∈ [a, b],
Z c Z c
0
F (x) dx = lim gn (x) dx.
a n→∞ a

But F is continuous; hence,

Z c Z c+1/n Z a+1/n 
lim gn (x) dx = lim n F (x) dx − F (x) dx = F (c) − F (a).
n→∞ a n→∞ c a

Consequently,
Z c Z c Z c
0
F (x) dx = lim gn (x) dx = F (c) − F (a) = f (x) dx.
a n→∞ a a

It follows that Z c
[F 0 (x) − f (x)] dx = 0
a

for every c ∈ [a, b]. By Theorem 2.1, F 0 (x) = f (x) for almost every x in [a, b].
Now let us assume that f is integrable on [a, b]. Without loss of any generality, we
may assume that f ≥ 0. For n = 1, 2, . . ., let fn be the function defined by

f (x) if 0 ≤ f (x) ≤ n,
fn (x) :=
0 if f (x) > n.

It is easily seen that F = Fn + Gn , where

Z x Z x
Fn (x) := fn (t) dt and Gn (x) := [f (t) − fn (t)] dt, a ≤ x ≤ b.
a a

Since f (t) − fn (t) ≥ 0 for all t ∈ [a, b], Gn is an increasing function on [a, b]. Moreover, by
what has been proved, Fn0 (x) = fn (x) for almost every x ∈ [a, b]. Thus, we have

F 0 (x) = Fn0 (x) + G0n (x) ≥ Fn0 (x) = fn (x) for almost every x ∈ [a, b].

Letting n → ∞ in the above inequality, we obtain F 0 (x) ≥ f (x) for almost every x ∈ [a, b].
It follows that Z Z b b
F 0 (x) dx ≥ f (x) dx = F (b) − F (a).
a a

4
On the other hand,
Z b
F 0 (x) dx ≤ F (b) − F (a).
a
Consequently,
Z b
[F 0 (x) − f (x)] dx = 0.
a
But F (x) ≥ f (x) for almost every x ∈ [a, b]. Therefore, F 0 (x) = f (x) for almost every x
0

in [a, b].

Theorem 2.3. A function F on [a, b] is absolutely continuous if and only if

Z x
F (x) = F (a) + f (t) dt
a

for some integrable function f on [a, b].

Proof. The sufficiency part has been established. To prove the necessity part, let F be an
absolutely continuous function on [a, b]. Then F is differentiable almost everywhere and
F 0 is integrable on [a, b]. Let
Z x
G(x) := F (a) + F 0 (t) dt, x ∈ [a, b].
a

By Theorem 2.2, G0 (x) = F 0 (x) for almost every x ∈ [a, b]. It follows that (F − G)0 (x) = 0
for almost every x ∈ [a, b]. By Theorem 1.2, F − G is constant. But F (a) = G(a).
Therefore, F (x) = G(x) for all x ∈ [a, b].

§3. Change of Variables for the Lebesgue Integral

Let f be an absolutely continuous function on [c, d], and let u be an absolutely con-
tinuous function on [a, b] such that u([a, b]) ⊆ [c, d]. Then the composition f ◦ u is not
necessarily absolutely continuous. However, we have the following result.

Theorem 3.1. Let f be a Lipschitz continuous function on [c, d], and let u be an absolutely
continuous function on [a, b] such that u([a, b]) ⊆ [c, d]. Then f ◦u is absolutely continuous.
Moreover,
(f ◦ u)0 (t) = f 0 (u(t))u0 (t) for almost every t ∈ [a, b],

where f 0 (u(t))u0 (t) is interpreted to be zero whenever u0 (t) = 0 (even if f is not differen-
tiable at u(t)).

Proof. Since f is a Lipschitz continuous function on [c, d], there exists some M > 0 such
that |f (x)−f (y)| ≤ M |x−y| whenever x, y ∈ [c, d]. Let ε > 0 be given. Since u is absolutely

5
 Pn
continuous on [a, b], there exists some δ > 0 such that i=1 |u(ti )−u(si )| < ε/M , whenever
{[si , ti ] : i = 1, . . . , n} is a finite collection of mutually disjoint subintervals of [a, b] with
Pn
i=1 (ti − si ) < δ. Consequently,
n
X n
X n
X
|(f ◦ u)(ti ) − (f ◦ u)(si )| = |f (u(ti )) − f (u(si ))| ≤ M |u(ti ) − u(si )| < ε.
i=1 i=1 i=1

This shows that f ◦ u is absolutely continuous on [a, b].

Since both u and f ◦ u are absolutely continuous on [a, b], there exists a measurable
subset E of [a, b] such that λ(E) = 0 and both u0 (t) and (f ◦ u)0 (t) exist for all t ∈ [a, b] \ E.
Suppose t0 ∈ [a, b] \ E. If u0 (t0 ) = 0, then for given ε > 0, there exists some h > 0 such
that |u(t) − u(t0 )| ≤ ε|t − t0 | whenever t ∈ (t0 − h, t0 + h) ∩ [a, b]. It follows that

|f ◦ u(t) − f ◦ u(t0 )| ≤ M |u(t) − u(t0 )| ≤ M ε|t − t0 |

for all t ∈ (t0 − h, t0 + h) ∩ [a, b]. This shows that

(f ◦ u)0 (t0 ) = 0 = f 0 (u(t0 ))u0 (t0 ).

Now suppose t0 ∈ [a, b] \ E and u0 (t0 ) 6= 0. Suppose u(t) 6= u(t0 ). Then we have
(f ◦ u)(t) − (f ◦ u)(t0 ) f (u(t)) − f (u(t0 )) u(t) − u(t0 )
= .
t − t0 u(t) − u(t0 ) t − t0
Since u0 (t0 ) and (f ◦ u)0 (t0 ) exist, we obtain
(f ◦ u)(t) − (f ◦ u)(t0 ) u(t) − u(t0 )
lim = (f ◦ u)0 (t0 ) and lim = u0 (t0 ) 6= 0.
t→t0 t − t0 t→t0 t − t0
Consequently,
f (u(t)) − f (u(t0 )) (f ◦ u)0 (t0 )
lim = .
t→t0 u(t) − u(t0 ) u0 (t0 )
Let r := (f ◦ u)0 (t0 )/u0 (t0 ). For given ε > 0, there exists some δ > 0 such that
f (u(t)) − f (u(t0 ))
r−ε< <r+ε ∀ t ∈ (t0 − δ, t0 + δ) ∩ [a, b].
u(t) − u(t0 )
Since u0 (t0 ) 6= 0, there exists some η > 0 such that any x ∈ (u(t0 ) − η, u(t0 ) + η) ∩ [c, d]
can be expressed as x = u(t) for some t ∈ (t0 − δ, t0 + δ) ∩ [a, b]. Therefore,
f (x) − f (u(t0 ))
r−ε< <r+ε ∀ x ∈ (u(t0 ) − η, u(t0 ) + η) ∩ [c, d].
x − u(t0 )
This shows that f 0 (u(t0 )) exists and f 0 (u(t0 )) = r = (f ◦ u)0 (t0 )/u0 (t0 ), as desired.

6
Theorem 3.2. Let g be a bounded and measurable function on [c, d], and let u be an
absolutely continuous function on [a, b] such that u([a, b]) ⊆ [c, d]. Then (g ◦ u)u0 is
integrable on [a, b]. Moreover, for any α, β ∈ [a, b],
Z u(β) Z β
g(x) dx = g(u(t))u0 (t) dt.
u(α) α

Proof. Let Z x
F (x) := g(t) dt, x ∈ [c, d].
c

Since g is bounded, F is Lipschitz continuous. Moreover, F 0 (x) = g(x) for almost every
x ∈ [a, b]. By Theorem 3.1, F ◦ u is absolutely continuous on [a, b] and, for almost every
t ∈ [a, b], (F ◦ u)0 (t) = g(u(t))u0 (t). Suppose α, β ∈ [a, b] and α < β. By Theorem 2.3, we
have
Z u(β) Z u(β)
0
(F ◦ u)(β) − (F ◦ u)(α) = F (u(β)) − F (u(α)) = F (x) dx = g(x) dx.
u(α) u(α)

On the other hand,

Z β Z β
0
(F ◦ u)(β) − (F ◦ u)(α) = (F ◦ u) (t) dt = g(u(t))u0 (t) dt.
α α

This proves the desired formula for change of variables.

Theorem 3.3. Let g be an integrable function on [c, d], and let u be an absolutely con-
tinuous function on [a, b] such that u([a, b]) ⊆ [c, d]. If (g ◦ u)u0 is integrable on [a, b],
then Z u(β) Z β
g(x) dx = g(u(t))u0 (t) dt, α, β ∈ [a, b].
u(α) α

Moreover, (g ◦ u)u0 is integrable if, in addition, u is monotone.

Proof. Suppose that g is integrable on [a, b]. Without loss of any generality, we may
assume g ≥ 0. For n = 1, 2, . . ., let gn be the function defined by

g(x) if 0 ≤ g(x) ≤ n,
gn (x) :=
0 if g(x) > n.

Then gn ≤ gn+1 for all n ∈ IN. Suppose α, β ∈ [a, b] and α < β. By Theorem 3.2 we have
Z u(β) Z β
gn (x) dx = gn (u(t))u0 (t) dt.
u(α) α

7
If u is monotone, then u0 (t) ≥ 0 for almost every t ∈ [a, b]. Letting n → ∞ in the above
equation, by the monotone convergence theorem we obtain
Z u(β) Z β
g(x) dx = g(u(t))u0 (t) dt.
u(α) α

Since g is integrable on [c, d], it follows from the above equation that (g ◦ u)u0 is integrable
on [a, b]. More generally, we assume that (g ◦ u)u0 is integrable on [a, b] but u is not
necessarily monotone. Then |gn (u(t))u0 (t)| ≤ g(u(t))|u0 (t)| for all n ∈ IN and almost every
t ∈ [a, b]. Thus, an application of the Lebesgue dominated convergence theorem gives the
desired formula for change of variables.