Exercise 4.

1 n(n 1)
n1
i11
= 2
=
.
n n
n 2
2n

Pn i2 1
1
n(n + 1)(2n + 1)
n(n + 1)
(n + 1)(2n + 1)
2
(2) S(Pn , x ) = i=1 2 = 3 4
4
+n =
.
n n
4n
6
2
6n2
P (2i 1)2 1
1 n(n + 1)(2n + 1)
4n2 + n 4
=
=
.
(3) S(Pn , x2 ) = ni=1
(2n)2 n
n3
6
12n2
P

2
n1 
1
1a
i1 1
.
(4) S(Pn , ax ) = ni=1 a n
=
1 + n a + n a + + n a
=
n
n
n(1 n a)
Exercise 4.2
(1) For any partition P of [0, 2], we have only one index i satisfying xi1 1 < xi . Then
S(P, f ) = f (xi )(xi xi1 ) + 1(xn xi ) = f (xi )(xi xi1 ) + 1 xi . Since f (xi ) = 0 or 1 and
0Z < xi xi xi1 kP k, we have |S(P, f ) 1| 2kP k. Therefore f (x) is integrable with
(1) S(Pn , x) =

Pn

i=1

f dx = 1.
0

we get S(P, f ) = 0. By taking all

(2) For any partition P , by taking all xi to be irrational,


P
1
1
xi rational, we get S(P, f ) = xi (xi xi1 )
kP k , where is inequality is obtained
2 2
1
by considering only those xi . This implies that S(P, f ) does not converge as kP k 0.
2
Therefore the function is not integrable.
1
(3) For any natural number N , assume kP k <
. Then any interval in the partition
2N

 



1
1
1 1
is either contained in 1,

, 1 or contained in ,
. We may divide the
2N
2NP P
N N P
0
Riemann
+ 00 , where the intervals
in 0 are contained in

 parts S(P, f ) =
suminto two

P
1 1
1
1
, 1 and the intervals in 00 are contained in ,

. Then there are at

1,
N
N
N
N
P

such that the corresponding

most 8N intervals in 0 , P
P0 f (xi ) = 1, and we also have f (xi ) = 0
0
| 4N kP k. On the other hand, the total
. Therefore we get |
for the other intervals in
P00
P
2
2
length of the intervals in
is , and we always have f (xi ) = 0 or 1. Therefore | 00 | .
N
N
P0
P00
2
We conclude that |S(P, f )| |
|+|
| 8N kP k + . This implies that
N
1
6
= |S(P, f )| ,
kP k <
2
2N
N
Z 1
which further implies that the function is integrable, with
f (x)dx = 0.
1

(4) For any natural number N , the Riemann sum S(P, f ) can be divided into two parts
1
according to xi =
for some n < N or otherwise. The number of terms in the first part is
n
1
< N , so that the sum is < N (max |f |)kP k = N kP k. For the second part we have f (xi ) ,
N
1
1
1
so that the sum of the second part is (1 0) = . Thus 0 S(P, f ) N kP k + . Then
N
N
N

2
2
would imply that |S(P, f )| < . Therefore the
for any  > 0, choosing N = and kP k <
4
Z 1
f dx = 0.
function is integrable, with
0

Exercise 4.3
P xi + xi1
1P 2
By taking middle points, we get Sm (P, x) =
(xi xi1 ) =
(xi x2i1 ) =
2
2
1 2
1
2
(x x0 ) = .
2 n
2




 xi + xi1 


1
1

 kP k. Therefore S(P, x)  =
Then for any choice of xi , we have xi


2
2
2


P  xi + xi1 
P
 xi
 xi 1 kP k xi = 1 kP k. This implies that x is
|S(P, x) Sm (P, x)|


2
2
2
Z 1
1
integrable, with
xdx = .
2
0
Exercise 4.4
For the given choice xi , the special Riemann sum is
r
Sm (P, x2 ) =

r
Denote

xi,m

Therefore


S(P, x)


x2i + xi xi1 + x2i1

3

!2
(xi xi1 ) =

1X 3
1
1
(xi x3i1 ) = (x3n x30 ) = .
3
3
3

x2i + xi xi1 + x2i1

. Then for any choice of xi , we have |xi xi,m | kP k.
3


X
X
1 

=
|S(P,
x)

S
(P,
x)|

|x

x
|x

kP
k
xi = kP k.
m
i
i
i,m
3
2

This implies that x is integrable, with

0

1
x2 dx = .
3

Exercise 4.5
(1) For any partition P , we have only one index jPsatisfying xj1 0 < xj . Then
[xi1 ,xi ] (f ) = 0 for j 6= i and [xj1 ,x
(f ) = 1. Therefore
[xi1 ,xi ] (f )xi = xj xj1 kP k.
j]
P
Thus for any  > 0, kP k <  =
[xi1 ,xi ] (f )xi < .
2
1
(2) We have [xi1 ,xi ] (f ) = xi . If kP k < , then the total length of the intervals with xi >
3
3
P
P
1
21
is at least . Therefore [xi1 ,xi ] (f )xi = xi xi
, and the function is not integrable.
3
33
1
(3) For any natural number N , assume kP k <
. Then any interval in the partition is
2N 
 



1
1 1
1

, 1 or contained in ,
. The number of intervals of
either contained in 1,
2N
2N
N N
the first type with [xi1 ,xi ] (f ) = 1 is 8N . The total length of the second type of intervals is
P
2
2
. Moreover, each [xi1 ,xi ] (f ) must be either 0 or 1. Therefore
[xi1 ,xi ] (f ) 8N kP k + .
N
N
P
6
1
In particular, we have
[xi1 ,xi ] (f )
when kP k <
.
N
2N 2
1
(4) For any natural number N , the number of intervals with [xi1 ,xi ] (f )
is 2N .
N
Since we always have [xi1 ,xi ] (f ) 1, we get
X
X X
1
+
< 2N kP k + (1 0).
[xi1 ,xi ] (f )xi =
N
1
1
N

< N

P
1
3
implies
[xi1 ,xi ] (f )xi < .
2
N
N
Exercise 4.6
2
The integrability
( of f (x) or |f (x)| does not imply the integrability of f (x). A counterex1
if x rational
ample is given by
. The integrability of f (x)3 implies the integrability of
1 if x irrational

f (x) because f (x) is the composition of 3 y with f (x)3 , and 3 y is continuous.

Thus kP k <

Exercise 4.7
Suppose |f (x)| < M for a constant M . Then the oscillation of f on any interval is at most
2M . For any  > 0, by the integrability of f onP
[a + , b], there is > 0, such that any partition
P of [a + , b] satisfying kP k < would imply
[xi1 ,xi ] (f )xi .
Now let P be a partition of [a, b] satisfying kP k < min{, }. Let xj1 a +  < xj . Let P
be the partition a +  < xj < xj+1 < < xn = b of [a + , b]. Since the partition P satisfies
kP k kP k < , the Riemann sum of the oscillation with respect to it is
[a+,xj ] (f )(xj (a + )) +

n
X

[xi1 ,xi ] (f )xi < ,

i=j

which implies
n
X
i=j

[xi1 ,xi ] (f )xi < .

On the other hand, xj a = (xj (a + )) +  (xj xj1 ) +  kP k +  < 2. Therefore
j
X

[xi1 ,xi ] (f )xi 2M

j
X

xi = 2M (xj a) < 4M .

i=1

i=1

Thus we conclude that the Riemann sum of the oscillation with respect to P is
n
X

[xi1 ,xi ] (f )xi +

j
X

[xi1 ,xi ] (f )xi (4M + 1).

i=1

i=j

By Theorem 4.2.2, the function is integrable on [a, b].

Exercise 4.8
Since f is convex, f0 (x) and f+0 (x) exists and are increasing. Therefore the one side derivatives are integrable. By choosing xi = xi , we have
X
X
S(P, f0 (x)) =
f0 (xi )(xi xi1 )
(f (xi ) f (xi1 )) = f (b) f (a).
b

f0 (x)dx f (b) f (a). On the other hand, by choosing xi = xi1 , we have

This implies
a

S(P, f0 (x)) =

f0 (xi1 )(xi xi1 )

f+0 (xi1 )(xi xi1 )

f0 (x)dx

This implies
f (b) f (a). Thus we conclude
Z b a
f+0 (x)dx = f (b) f (a) is similar.
proof for

(f (xi )f (xi1 )) = f (b)f (a).

f0 (x)dx = f (b) f (a). The

Exercise 4.9
For x2 > x1 , we have (x2 ) (x1 ) > A(x2 x1 ) > 0, so that is strictly increasing. The
inequality 0 < (x2 ) (x1 ) < B(x2 x1 ) also implies is continuous. Therefore y = (x) is
a change of variable between x1 x x2 and (x1 ) y (x2 ), and
sup f ((x)) =
x1 xx2

sup

f (y),

(x1 )y(x2 )

inf

x1 xx2

f ((x)) =

inf

f (y).

(x1 )y(x2 )

As the difference between the supremum and infimum, the oscillations are equal:
[x1 ,x2 ] (f ) = [(x1 ),(x2 )] (f ).
Let P be a partition of [a, b]. We have [xi1 ,xi ] (f ) = [(xi1 ),(xi )] (f ) from the second
part. Then
X

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

1X
[(xi1 ),(xi )] (f )((xi ) (xi1 ))
A

Now (P ) is a partition of [(a), (b)]. By (x2 ) (x1 ) < B(x2 x1 ), we also know kP k <
BkP k.
SincePf (y) is integrable on [(a), (b)], for any  > 0, there is > 0, such that k(P )k <
implies
[(xi1 ),(xi )] (f )((xi ) (xi1 )) A. Thus
kP k <

X
1

= k(P )k < BkP k < =

[xi1 ,xi ] (f )(xi xi1 ) < A = .
B
A

This completes the proof of the integrability of f ((x)).

Finally, assume is differentiable, with A < 0 (x) < B. Then for x1 < x2 , by the mean
value theorem, we have (x2 )(x1 ) = 0 (c)(x2 x1 ). Then A < 0 (c) < B implies A(x2 x1 ) <
(x2 ) (x1 ) < B(x2 x1 ).

Exercise 4.10
1
We have max{f, g} = (|f +g|+f +g). The integrability of f and g implies the integrability
2
of f + g. Then by the continuity of the absolute value, |f + g| is integrable. Then as a linear
combination of |f + g|, f , g, max{f, g} is also integrable.
Exercise 4.11
The function h(x) = f (x) g(x) takes nonzero value only at cn and is bounded by B. We
only need to prove that h is Riemann integrable.
Let lim cn = c. For any  > 0, there is natural number N , such that n > N implies
|cn c| < . If a partition P of [a, b] satisfies kP k < N , then
!
X X X
S(P, h) =
+
+
h(xi )xi ,
1

P
P
where 1 consists of those intervalP
with xi = cn for some n N , 2 consists of those interval
with xi = cn for some n > N , and 3 consists of those interval with xi 6= any cn .
The total number of intervals in 1 is at most 2N . Therefore
X
| 2N BkP k < 2B.
|
1

P
Since n > N implies cn (c , c + ), the intervals in 2 are contained
P in (c  kP k, c +
 + kP k) (c 2, c + 2), so that the total length of the intervals in 2 is < 4. This implies
X
| 4B.
|
2

Since f (xi ) = 0 in 3 , we have

Z b
hdx = 0.
integrable, with
P

= 0. Therefore |S(P, h)| 6B. This implies that h is

Exercise 4.12
Let u(x) be the straight line connecting (a, 0) to (y, f (y)), followed by straight line connecting (y, f (y)) to (b, 0). By f (x) 0 and the concavity, we have f u on [a, b]. Then
Z b
Z b
1
f (x)dx
u(x)dx = f (y)(b a).
2
a
a
f
u

Exercise 4.13
Let L be the straight line f (a) + m(x a) followed by the straight line f (b) + M (x b).
Then
(
f (a) + m(x a) if a x c
L (x) =
,
f (b) + M (b x) if c x b
f (a) f (b) + M b ma
is where the two lines meet. We clearly have L (x) f (x)
M m
on [a, b]. Similarly, we have L+ (x) f (x) on [a, b], where
(
f (a) + M (x a) if a x c+
L+ (x) =
,
f (b) + m(b x) if c+ x b

where c =

f (a) f (b) + M b ma
. As illustrated in the picture, the graph of f lies in the
M m
parallelogram. The two lower edges of the parallelogram form the graph of L , and the two
upper edges form the graph of L+ .
and c+ =

M
P

f
B

A
M

m
Q
a

Z b



f (a) + f (b)

The difference 
f (x)dx
(b a) is no more than the area of the triangle
2
a
ABQ. The area of the triangle is half of (b a) multiplied to the length of P Q. The length of
P Q is


f (a) f (b) + M (b a) f (b) f (a)
(c a)(slope of AP slope of AQ) =
m
M m
ba
Thus the area of the triangle is

(f (a) f (b) + M (b a))(f (b) f (a) m(b a))

and the
2(M m)

inequality follows.
Exercise 4.14
1
By Example 4.3.3, we have min[a,b] f
ba

f (x)dx max f . Then by the continuity of

[a,b]
Z b
1
f (x) and the intermediate value theorem, we have
f (x)dx = f (c) for some a < c < b.
ba a
a

More generally, for g(x) 0, we have (min[a,b] f )g(x) f (x)g(x) (max[a,b] f )g(x).
Z b
Z b
Z b
g(x)dx. This is the same as
f (x)g(x)dx (max f )
g(x)dx
Therefore (min[a,b] f )
[a,b]
a
a
a
Z b
f (x)g(x)dx
a
min[a,b] f Z b
max[a,b] f . By the continuity of f and the intermediate value
g(x)dx
aZ
b
f (x)g(x)dx
a
theorem, we get Z b
= f (c) for some a < c < b.
g(x)dx
a

Exercise 4.15
For any fixed a c b, we have |f (c) f (x)| [a,b] (f ) for any x in [a, b]. Therefore

 Z b
 Z b

Z b
Z b

 
 

f (c)(b a)




f (x)dx = 
f (c)dx
f (x)dx =  (f (c) f (x))dx

a
a
a
a
Z b
Z b

|f (c) f (x)|dx
[a,b] (f )dx = [a,b] (f )(b a).
a

Exercise 4.16
The continuous function is uniformly continuous on [a h, b + h]. Thus for any  > 0, there
is > 0, such that |x y| < and a h x, y b + h imply |f (x) f (y)| < . Now if |h| < ,
then
Z b
|f (x + h) f (x)|dx (b a).
a

|f (x + h) f (x)|dx = 0.

This implies limh0

a

Exercise 4.17
First assume f (a) 6= 0, f (b) 6= 0. Then the condition
tells us
0 < m f (x) M for some

n
n
constants m, M and all x [a, b].
Since limn m
= limn M = 1,pfor any  > 0, there

n
n
is N , such that n >
N
implies
|
m

1|
<

and
|
M
1| < . Since n f (x) is sandwiched
p

between n m and n M , this further implies | n f (x) 1| <  for all x [a, b], and
Z b
 Z b
Z b
Z b
p
p


n
n


g(x) f (x)dx
g(x)dx
|g(x)|| f (x) 1|dx 
|g(x)|dx.

a

Z
This completes the proof that limn

Z b
p
n
g(x) f (x)dx =
g(x)dx.

Now assume f (a) = 0, f (b) 6= 0. There is > 0, such that f (x) < 1 on [a, a + ]. Now
Z b
Z b
p
n
for any fixed a < c a + , we have limn
g(x) f (x)dx =
g(x)dx. Thus for any
c


Z b
Z b
p


n
g(x) f (x)dx
g(x)dx < . Then by
 > 0, there is N , such that n > N implies 
c
c
0 f (x) < 1 on [a, c], n > N implies
Z b

Z b
p


n

g(x) f (x)dx
g(x)dx

a
 Z c

Za b
 Z c
Z b
p
p
 


 
n
n
g(x) f (x)dx
g(x)dx + 
g(x) f (x)dx + 
g(x)dx

c
c
a
a
Z c
|g(x)|dx.
 + 2(c a)
a

By fixing c very close to a at first, the right side can be made very small (< 2 for example).
Z b
Z b
p
n
This proves that limn
g(x) f (x)dx =
g(x)dx.
a

The proof for the other cases is similar.

Exercise 4.18
Suppose f (b) 1. For any b a >  > 0, we have
Z
0

f (x) dx =
a

b

f (x) dx

f (x) dx +
a

b

b

dx +
a

f (1)n dx

b

f (b )n (b  a) + f (1)n f (b )n (b a) + .

By 0 f (b ) < f (1) 1, there is N , such that n > N implies f (b )n < . This further
Z b
Z b
n
f (x) dx < (b a + 1). Therefore limn
f (x)n dx = 0.
implies 0
a

Suppose f (b) > 1 and f is continuous at b. Then there is a < c < b, such that f (c) > 1. By
non-negativity and the increasing property, have
Z b
Z b
Z c
Z b
n
n
n
f (x)n dx (b c)f (c)n .
f (x) dx
f (x) dx +
f (x) dx =
c

c
b

f (x)n dx = +.

By f (c) > 1, we have limn f (c) = +. Therefore limn

a

Exercise 4.19
It is clear that the first implies the second. The second implies the third because
Z d

Z b
Z c
Z d
Z b
Z d



|f (x)|dx =
|f (x)|dx +
|f (x)|dx +
|f (x)|dx
|f (x)|dx 
f (x)dx .
a

The second also implies the fourth because for M = max[a,b] |g|, we have
Z

Z
|f (x)|dx

M
a

Z b



|f (x)g(x)|dx 
f (x)g(x)dx .
a

It remains to prove that either third or the fourth implies f = 0. Equivalently, we will assume
f 6= 0 and prove that the third and the fourth are broken.
Assume f (c) 6= 0 for some c [a, b]. If f (c) > 0, then by the continuity of f (x) at c, there
f (c)
for x [c , c + ] (the interval is [a, a + ] if c = a and is
is > 0, such that f (x) >
2Z
c+
f (c)
f (x)dx 2
[b , b] if c = b). Therefore
> 0, and the third is broken. Moreover,
2
c
we construct a continuous function g(x) satisfying



g(x) 0, g(x) = 1 on c , c +
, g(x) = 0 on [a, c ] [c + , b].
2
2
Then
Z

c+

Z
f (x)g(x)dx

f (x)g(x)dx =
a

c+ 2

f (x)g(x)dx

c 2

f (c)
> 0.,
2

and the fourth is broken. The argument for the case f (c) < 0 is similar.
Exercise 4.20
If f (x) changes sign, then both |f (x)| f (x) and |f (x)| + f (x) are continuous non-negative
functions that are not constantly 0. By Exercise 4.19, we have
Z b
Z b
(|f (x)| f (x))dx > 0,
(|f (x)| + f (x))dx > 0.
a

Z b
 Z b


|f (x)|dx.
Combining the two, we get 
f (x)dx <
a

Exercise 4.21
If f (x) is never zero on (a, b), then by the continuity, we have either f (x) > 0 on (a, b) or
Z b
Z b
|f (x)|dx 6= 0.
f (x)dx =
f (x) < 0 on (a, b). Then it follows from Exercise 4.19 that
a

If f (x) has only one zero at c (a, b), then by continuity, we have either f (x) 0 on (a, b),
or f (x) 0 on (a, b), or f (x) < 0 on (a, c) and f (x) > 0 on (c, b), or f (x) > 0 on (a, c) and
f (x) < 0 on (c, b). In the first two cases, since f is not constantly zero, it follows again from
Z b
Exercise 4.19 that
f (x)dx 6= 0. In the last two cases, we have (x c)f (x) 0 on (a, b) or
a

(x c)f (x) < 0 on (a, b). Then

Z

Z
xf (x)dx c

(x c)f (x)dx 6= 0

f (x)dx =
a

by the same reason.

In all the cases, we find contradiction to the assumptions. Therefore f (x) must be zero at
two places in (a, b).
Exercise 4.22

The estimation in Exericse 4.15 implies

 X 


Z xi
Z b

 




S(P, f )
f (x)dx 
f (xi )xi
f (x)dx = 

xi1
a


Z xi
X
 X

f (xi )(xi xi1 )



f
(x)dx
[xi1 ,xi ] (f )xi .


xi1

Exercise 4.23
For any partition P , we have

 Z xi
Z xi
Z



f (xi )xi
f (x)dx
|f (xi ) f (x)|dx L

xi1

xi1
Z xi

xi

|xi x|dx

xi1

(xi x)dx + L

=L
xi1

xi

(x xi )dx = L

xi1

(xi xi1 )2 + L(xi xi )2

2

=L

(xi xi1 )
L
L(xi xi1 )(xi xi ) x2i .
2
2

Then adding together, we get



 
Z b
X Z xi


 X

S(P, f )
f (x)dx
f (x)dx = 
f (xi )xi

x
a

Z xi i1
X 
 LX

f
(x
f
(x)dx
x2i .
)x

i
i


2
xi1
Exercise 4.24

 


Z b
Z xi
Z xi
Z xi

X
 X





f
(x)g(x)dx
g(x)dx

f
(x)g(x)dx
=
g(x)dx

f
(x
)
f
(x
)
i
i


 
xi1
xi1
a
xi1


Z
Z
X  xi
 X xi

(f
(x
[xi1 ,xi ] (f ) sup |g|dx
)

f
(x))g(x)dx
i


xi1

xi1

[xi1 ,xi ] (f ) sup |g|xi sup |g|

[xi1 ,xi ]

[xi1 ,xi ]

[xi1 ,xi ] (f )xi .

[a,b]

Exercise 4.25
Since f (x) is integrable, it is bounded: We have |f (x)| < M for all x [a, b] and a constant
M.
P
Since g is integrable, for any  > 0, there is > 0, such that kP k < implies [xi1 ,xi ] (g)xi <
.
Since f and g are integrable, the product
 f g is also integrable.
Z b Therefore for any  > 0,


P
there is 0 > 0, such that kP k < 0 implies  f (xi )g(xi )xi
f (x)g(x)dx < .
a

Now for kP k < min{, 0 }, we have



Z b

X


f (xi )g(xi )xi
f (x)g(x)dx

a

Z b
 X
X
X






f
(x)g(x)dx
f
(x
)g(x
)x

f (xi )g(x
)x

f
(x
)g(x
)x
+


i
i
i
i
i
i
i
i


a
X
X

<
|f (xi )(g(x
[xi1 ,xi ] (g)xi +  < (M + 1).
i ) g(xi ))|xi +  M