A bounded derivative that is not Riemann

integrable
Russ Gordon
Whitman College

January 6, 2016

1. There exist functions that are continuous at each irrational

number but discontinuous at each rational number.

1. There exist functions that are continuous at each irrational

number but discontinuous at each rational number.

2. There exist continuous nowhere differentiable functions.

1. There exist functions that are continuous at each irrational

number but discontinuous at each rational number.

2. There exist continuous nowhere differentiable functions.

3. There exist bounded derivatives that are not Riemann

integrable.

A fangtooth fish: photo from National Geographic.

If f is Riemann integrable on [a, b] and F is any antiderivative of

Rb
f , then a f = F (b) F (a).

If f is Riemann integrable on [a, b] and F is any antiderivative of

Rb
f , then a f = F (b) F (a).
If F is differentiable on [a, b] and F 0 is Riemann integrable on
Rb
[a, b], then a F 0 = F (b) F (a).

If f is Riemann integrable on [a, b] and F is any antiderivative of

Rb
f , then a f = F (b) F (a).
If F is differentiable on [a, b] and F 0 is Riemann integrable on
Rb
[a, b], then a F 0 = F (b) F (a).

(
x 2 sin(1/x 2 ),
F (x) =
0,

if x 6= 0;
if x = 0.

If f is Riemann integrable on [a, b] and F is any antiderivative of

Rb
f , then a f = F (b) F (a).
If F is differentiable on [a, b] and F 0 is Riemann integrable on
Rb
[a, b], then a F 0 = F (b) F (a).

(
x 2 sin(1/x 2 ),
F (x) =
0,

if x 6= 0;
if x = 0.

If F is differentiable on [a, b] and F 0 is bounded on [a, b], then

Rb 0
FALSE!
a F = F (b) F (a).

In 1881, Volterra constructed a bounded derivative that was not

Riemann integrable. The existence of such functions later
convinced Lebesgue that a better integration process needed to be
devised.

In 1881, Volterra constructed a bounded derivative that was not

Riemann integrable. The existence of such functions later
convinced Lebesgue that a better integration process needed to be
devised.

(
x 2 sin(1/x),
F (x) =
0,

if x 6= 0;
if x = 0.

In 1881, Volterra constructed a bounded derivative that was not

Riemann integrable. The existence of such functions later
convinced Lebesgue that a better integration process needed to be
devised.

(
x 2 sin(1/x),
F (x) =
0,

if x 6= 0;
if x = 0.

We will present an example related to a collection of functions that

Casper Goffman used in a March 1977 Monthly note.

Let O =

(an , bn ) be an open dense set in the interval (0, 1),

n=1

where the intervals are disjoint and the inequality

(bn an )

n=1

1
2

is satisfied. Without loss of generality, we may assume that ai 6= bj

for all positive integers i and j.

Let O =

(an , bn ) be an open dense set in the interval (0, 1),

n=1

where the intervals are disjoint and the inequality

(bn an )

n=1

1
2

is satisfied. Without loss of generality, we may assume that ai 6= bj

for all positive integers i and j.

There are various ways to construct such a set, each of which

involve some prerequisite knowledge. The existence of such sets is
far from obvious for students.

Let O =

(an , bn ) be an open dense set in the interval (0, 1),

n=1

where the intervals are disjoint and the inequality

(bn an )

n=1

1
2

is satisfied. Without loss of generality, we may assume that ai 6= bj

for all positive integers i and j.

There are various ways to construct such a set, each of which

involve some prerequisite knowledge. The existence of such sets is
far from obvious for students.
For later reference, let E = [0, 1] \ O. Note that E is a perfect
nowhere dense set of positive measure.

For each positive integer n, choose points cn and dn so that

an < cn < dn < bn ,

an + bn
cn + dn
=
,
2
2

and dn cn = (bn an )2 .

For each positive integer n, choose points cn and dn so that

an + bn
cn + dn
=
,
2
2

an < cn < dn < bn ,

an

cn

dn

and dn cn = (bn an )2 .

bn

For each positive integer n, choose points cn and dn so that

an + bn
cn + dn
=
,
2
2

an < cn < dn < bn ,

an

cn

dn

and dn cn = (bn an )2 .

bn

Note that
cn an =

 b a
(bn an ) (bn an )2
bn an 
n
n
=
1(bn an ) >
,
2
2
4

For each positive integer n, choose points cn and dn so that

an + bn
cn + dn
=
,
2
2

an < cn < dn < bn ,

an

cn

dn

and dn cn = (bn an )2 .

bn

Note that
cn an =

 b a
(bn an ) (bn an )2
bn an 
n
n
=
1(bn an ) >
,
2
2
4

and thus
(bn an )2 < 16(cn an )2 .

For each positive integer n, define a continuous function fn on the

interval [0, 1] with the following properties:

For each positive integer n, define a continuous function fn on the

interval [0, 1] with the following properties:
(i) 1 fn (x) 1 for all x;

For each positive integer n, define a continuous function fn on the

interval [0, 1] with the following properties:
(i) 1 fn (x) 1 for all x;
(ii) fn is 0 on [0, cn ] and [dn , 1];

For each positive integer n, define a continuous function fn on the

interval [0, 1] with the following properties:
(i) 1 fn (x) 1 for all x;
(ii) fn is 0 on [0, cn ] and [dn , 1];
(iii) there is a point tn (cn , dn ) such that |f (tn )| = 1;

For each positive integer n, define a continuous function fn on the

interval [0, 1] with the following properties:
(i) 1 fn (x) 1 for all x;
(ii) fn is 0 on [0, cn ] and [dn , 1];
(iii) there is a point tn (cn , dn ) such that |f (tn )| = 1;
(iv)

R1

0 fn

R dn
cn

fn = 0.

For each positive integer n, define a continuous function fn on the

interval [0, 1] with the following properties:
(i) 1 fn (x) 1 for all x;
(ii) fn is 0 on [0, cn ] and [dn , 1];
(iii) there is a point tn (cn , dn ) such that |f (tn )| = 1;
(iv)

R1

0 fn

R dn
cn

fn = 0.

Define f : [0, 1] [1, 1] by f (x) =

P
n=1

fn (x).

y
1

y = fn (x)

an

cn

dn

bn

The oscillation of f on [a, b] is defined by

(f , [a, b]) = sup{f (x) : x [a, b]} inf{f (x) : x [a, b]}.

The oscillation of f on [a, b] is defined by

(f , [a, b]) = sup{f (x) : x [a, b]} inf{f (x) : x [a, b]}.

A bounded function f is Riemann integrable on [a, b] if and only if

for each  > 0 there exists a partition P = {xi : 0 i n} of [a, b]
n
P
such that
(f , [xi1 , xi ])(xi xi1 ) < .
i=1

Recall that O is our open dense set and E = [0, 1] \ O.

Recall that O is our open dense set and E = [0, 1] \ O.

Let {xi : 0 i q} be a partition of [0, 1], then define

Recall that O is our open dense set and E = [0, 1] \ O.

Let {xi : 0 i q} be a partition of [0, 1], then define
S = {i 1 : (xi1 , xi ) E = };
T = {i 1 : (xi1 , xi ) E 6= }.

Recall that O is our open dense set and E = [0, 1] \ O.

Let {xi : 0 i q} be a partition of [0, 1], then define
S = {i 1 : (xi1 , xi ) E = };
T = {i 1 : (xi1 , xi ) E 6= }.
If i S, then (xi1 , xi ) O.

Recall that O is our open dense set and E = [0, 1] \ O.

Let {xi : 0 i q} be a partition of [0, 1], then define
S = {i 1 : (xi1 , xi ) E = };
T = {i 1 : (xi1 , xi ) E 6= }.
If i S, then (xi1 , xi ) O.
If i T , then (an , bn ) (xi1 , xi ) for some index n and thus
(f , [xi1 , xi ]) 1.

X
iS

X
1
(xi xi1 )
(bn an )
2
n=1

so

X
iT

1
(xi xi1 ) .
2

X
iS

X
1
(xi xi1 )
(bn an )
2

q
X

n=1

(f , [xi1 , xi ])(xi xi1 )

i=1

so

X
iT

1
(xi xi1 ) .
2

(f , [xi1 , xi ])(xi xi1 )

iT

iT

1
(xi xi1 ) .
2

X
iS

X
1
(xi xi1 )
(bn an )
2

q
X

n=1

(f , [xi1 , xi ])(xi xi1 )

i=1

so

X
iT

1
(xi xi1 ) .
2

(f , [xi1 , xi ])(xi xi1 )

iT

iT

1
(xi xi1 ) .
2

Hence, the function f is not Riemann integrable on [0, 1].

Now define a function F on [0, 1] by

F (x) =

Z
X
k=1 0

fk .

Z
F (x) 6=

f
0

Now define a function F on [0, 1] by

F (x) =

Z
X
k=1 0

fk .

Z
F (x) 6=

The function F is well-defined on [0, 1] since for each x, at most

one term of the series is nonzero.

Now define a function F on [0, 1] by

F (x) =

Z
X
k=1 0

fk .

Z
F (x) 6=

The function F is well-defined on [0, 1] since for each x, at most

one term of the series is nonzero.
We claim that F 0 (x) = f (x) for all x [0, 1]. It then follows that
F 0 is a bounded derivative that is not Riemann integrable.

A fangtooth fish: photo from National Geographic.

First note that for all x (an , bn )

F (x) F (an ) =

Z
X

k=1 an

fk =

fn .
an

First note that for all x (an , bn )

F (x) F (an ) =

Z
X

k=1 an

fk =

fn .
an

Hence F 0 (x) = fn (x) = f (x) for all x (an , bn ) by the

Fundamental Theorem of Calculus (the other version).

So F 0 = f on the set O.

So F 0 = f on the set O.

photo of a yellowfish from the Christian Science Monitor

Now suppose that x E . We will prove that

lim+

y x

F (y ) F (x)
= 0.
y x

Now suppose that x E . We will prove that

lim+

y x

F (y ) F (x)
= 0.
y x

If x = ak , then the result is trivial.

Now suppose that x E . We will prove that

lim+

y x

F (y ) F (x)
= 0.
y x

If x = ak , then the result is trivial.

Otherwise fix a value of y so that x < y < 1. There are two cases
to consider.

I. If y
/

(ck , dk ), then F (y ) F (x) = 0 0 = 0.

k=1

I. If y
/

(ck , dk ), then F (y ) F (x) = 0 0 = 0.

k=1

II. Suppose that y (cp , dp ). Using the inequality mentioned when

the intervals [cn , dn ] were defined, we have

I. If y
/

(ck , dk ), then F (y ) F (x) = 0 0 = 0.

k=1

II. Suppose that y (cp , dp ). Using the inequality mentioned when

the intervals [cn , dn ] were defined, we have
Z y 


|F (y ) F (x)| = |F (y ) F (cp )| = 
fp 
cp
Z y

|fp | dp cp = (bp ap )2
cp

16(cp ap )2
16(y x)2 ,

where we have used the fact that x < ap < cp < y .

It follows easily that

lim

y x +

F (y ) F (x)
= 0.
y x

It follows easily that

lim

y x +

F (y ) F (x)
= 0.
y x

A similar argument reveals that

lim

y x

F (y ) F (x)
= 0.
y x

We have thus shown that F 0 (x) = 0.

It follows easily that

lim

y x +

F (y ) F (x)
= 0.
y x

A similar argument reveals that

lim

y x

F (y ) F (x)
= 0.
y x

We have thus shown that F 0 (x) = 0.

We conclude that F 0 (x) = f (x) for all x E . It follows that
F 0 = f on [0, 1].

Define sets V and W by

V = {x [0, 1] : f (x) 6= 0}

(cn , dn );

n=1

W = {x [0, 1] : f (x) = 0} = [0, 1] \ V .

Define sets V and W by

V = {x [0, 1] : f (x) 6= 0}

(cn , dn );

n=1

W = {x [0, 1] : f (x) = 0} = [0, 1] \ V .

Each point of E is a point of density of W or, equivalently, a point

of dispersion of V .

Define sets V and W by

V = {x [0, 1] : f (x) 6= 0}

(cn , dn );

n=1

W = {x [0, 1] : f (x) = 0} = [0, 1] \ V .

Each point of E is a point of density of W or, equivalently, a point

of dispersion of V .
The function f is approximately continuous at each point of E so f
is approximately continuous on [0, 1].

Define sets V and W by

V = {x [0, 1] : f (x) 6= 0}

(cn , dn );

n=1

W = {x [0, 1] : f (x) = 0} = [0, 1] \ V .

Each point of E is a point of density of W or, equivalently, a point

of dispersion of V .
The function f is approximately continuous at each point of E so f
is approximately continuous on [0, 1].
Rx
Since the function F equals 0 f , where the integral is a Lebesgue
integral, we find that F 0 = f on [0, 1] by a standard theorem in the
theory of Lebesgue integration.

Thanks for listening.

Thanks for listening.