The Riemann-Stieltjes Integral

Riemann-Stieltjes integrals
7.1 Prove that ]
a
b
do = ob oa, directly from Definition 7.1.
Proof: Let f = 1 on a, b, then given any partition P = a = x
0
, . . . , x
n
= b, then we
have
SP, 1, o =
_
k=1
n
ft
k
Ao
k
, where t
k
x
k1
, x
k

=
_
k=1
n
Ao
k
= ob oa.
So, we know that ]
a
b
do = ob oa.
7.2 If f Ro on a, b and if ]
a
b
fdo = 0 for every f which is monotonic on a, b,
prove that o must be constant on a, b.
Proof: Use integration by parts, and thus we have
]
a
b
odf = fbob faoa
Given any point c a, b, we may choose a monotonic function f defined as follows.
f =
0 if x _ c
1 if x > c.
So, we have
]
a
b
odf = oc = ob.
So, we know that o is constant on a, b.
7.3 The following definition of a Riemann-Stieltjes integral is often used in the
literature: We say that f is integrable with respect to o if there exists a real number A
having the property that for every c > 0, there exists a o > 0 such that for every partition
P of a, b with norm ]P] < o and for every choice of t
k
in x
k1
, x
k
, we have
|SP, f, o A| < c.
(a) Show that if ]
a
b
fdo exists according to this definition, then it is also exists according
to Definition 7.1 and the two integrals are equal.
Proof: Since refinement will decrease the norm, we know that if there exists a real
number A having the property that for every c > 0, there exists a o > 0 such that for every
partition P of a, b with norm ]P] < o and for every choice of t
k
in x
k1
, x
k
, we have
|SP, f, o A| < c. Then choosing a P
c
with ]P
c
] < o, then for P P
c
]P] < o. So,
we have
|SP, f, o A| < c.
That is, ]
a
b
fdo exists according to this definition, then it is also exists according to
Definition 7.1 and the two integrals are equal.
(b) Let fx = ox = 0 for a _ x < c, fx = ox = 1 for c < x _ b,
fc = 0, oc = 1. Show that ]
a
b
fdo exists according to Definition 7.1 but does not exist
by this second definition.
Proof: Note that ]
a
b
fdo exists and equals 0 according to Definition 7.1If ]
a
b
fdo exists
according to this definition, then given c = 1, there exists a o > 0 such that for every
partition P of a, b with norm ]P] < o and for every choice of t
k
in x
k1
, x
k
, we have
|SP, f, o| < 1. We may choose a partition P = a = x
0
, . . . , x
n
= b with ]P] < o and
c x
j
, x
j+1
, where j = 0, . . . , n 1. Then
SP, f, o = fxox
j+1
ox
j
= 1, where x c, x
j+1

which contradicts to |SP, f, o| < 1.

7.4 If f R according to Definition 7.1, prove that ]
a
b
fxdx also exists according to
definition of Exercise 7.3. [Contrast with Exercise 7.3 (b).]
Hint: Let I = ]
a
b
fxdx, M = sup|fx| : x a, b. Given c > 0, choose P
c
so that
UP
c
, f < I + c/2 (notation of section 7.11). Let N be the number of subdivision points in
P
c
and let o = c/2MN. If ]P] < o, write
UP, f =
_
M
k
fAx
k
= S
1
+ S
2
,
where S
1
is the sum of terms arising from those subintervals of P containing no points of
P
c
and S
2
is the sum of remaining terms. Then
S
1
_ UP
c
, f < I + c/2 and S
2
_ NM]P] < NMo = c/2,
and hence UP
c
, f < I + c. Similarly,
LP, f > I c if ]P] < o
'
for some o
'
.
Hence |SP, f I| < c if ]P] < mino, o
'
.
Proof: The hint has proved it.
Remark: There are some exercises related with Riemann integrals, we write thme as
references.
(1) Suppose that f _ 0 and f is continuous on a, b, and ]
a
b
fxdx = 0. Prove that
fx = 0 on a, b.
Proof: Assume that there is a point c a, b such that fc > 0. Then by continuity of
f, we know that given c =
fc
2
> 0, there is a o > 0 such that as |x c| < o, x a, b, we
have
|fx fc| <
fc
2
which implies that
fc
2
< fx if x c o, c + o , a, b := I
So, we have
0 <
fc
2
|I| _ ]
I
fxdx _ ]
a
b
fxdx = 0, where 0 < |I|, the length of I
which is absurb. Hence, we obtain that fx = 0 on a, b.
(2) Let f be a continuous function defined on a, b. Suppose that for every continuous
function g defined on a, b which satisfies that
]
a
b
gxdx = 0,
we always have
]
a
b
fxgxdx = 0.
Show that f is a constant function on a, b.
Proof: Let ]
a
b
fxdx = I, and define gx = fx
I
ba
, then we have
]
a
b
gxdx = 0,
which implies that, by hypothesis,
]
a
b
fxgxdx = 0
which implies that
]
a
b
fx cgxdx = 0 for any real c.
So, we have
]
a
b
gx
2
dx = 0 if letting c =
I
b a
which implies that gx = 0 forall x a, b by (1). That is, fx =
I
ba
on a, b.
(3) Define
hx =
0 if x 0, 1 Q
1
n
if x is the rational number m/n (in lowest terms)
1 if x = 0.
Then h R0, 1.
Proof: Note that we have shown that h is continuous only at irrational numbers on
0, 1 Q. We use it to show that h is Riemann integrable, i.e., h R0, 1. Consider the
upper sum UP, f as follows.
Given c > 0, there exists finitely many points x such that fx _ c/2. Consider a
partition P
c
= x
0
= a, . . . , x
n
= b so that its subintervals I
j
= x
j1
, x
j
for some j
containing those points and _|I
j
| < c/2. So, we have
UP, f =
_
k=1
n
M
k
Ax
k
=
_
1
+
_
2
< c/2 + c/2
= c
where _
1
= _
1
M
j
I
j
, and _
2
, is the sum of others.
So, we have shown that f satisfies the Riemann condition with respect to ox = x.
Note: (1) The reader can show this by Theorem 7.48 (Lebesgues Criterion for
Riemann Integrability). Also, compare Exercise 7.32 and Exercise 4.16 with this.
(2) In Theorem 7.19, if we can make sure that there is a partition P
c
such that
UP
c
, f, o LP
c
, f, o < c,
then we automatically have, for any finer P P
c
,
UP, f, o LP, f, o < c
since the refinement makes U increase and L decrease.
(4) Assume that the function fx is differentiable on a, b, but not a constant and that
fa = fb = 0. Then there exists at least one point on a, b for which
|f
'
| >
4
b a
2
]
a
b
fxdx.
Proof: Consider sup
xa,b
|f
'
x| := M as follows.
(i) If M = +, then it is clear.
(ii) We may assume that M < +.
Let x a,
a+b
2
, then
fx = fx fa = f
'
yx a _ Mx a, where y a, x. *
and let x
a+b
2
, b, then
fx = fx fb = f
'
zx b _ Mb x, where z x, b. **
So, by (*) and (**), we know that
]
a
b
fxdx = ]
a
a+b
2
fxdx + ]
a+b
2
b
fxdx
_ M]
a
a+b
2
x adx + M]
a+b
2
b
b xdx
= M
a b
2
2
which implies that
M _
4
b a
2
]
a
b
fxdx.
Note that by (*) and (**), the equality does NOT hold since if it was, then we had
f
'
x = M on a, b which implies that f is a constant function. So, we have
M >
4
b a
2
]
a
b
fxdx.
By definition of supremum, we know that there exists at least one point on a, b for
which
|f
'
| >
4
b a
2
]
a
b
fxdx.
(5) Gronwall Lemma: Let f and g be continuous non-negative function defined on
a, b, and c _ 0. If
fx _ c + ]
a
x
gtftdt for all x a, b,
then
fx _ ce
]
a
x
gtdt
.
In particular, as c = 0, we have f = 0 on a, b.
Proof: Let c > 0 and define
Fx = c + ]
a
x
gtftdt,
then we have
(i). Fa = c > 0.
(ii). F
'
x = gxfx _ 0 F is increasing on a, b by Mean Value Theorem
(iii). Fx _ fx on a, b F
'
x _ gxFx by (ii).
So, from (iii), we know that
Fx _ Fae
]
a
x
gtdt
= ce
]
a
x
gtdt
by (i).
For c = 0, we choose c
n
= 1/n 0, then by preceding result,
fx _
1
n
e
]
a
x
gtdt
0 as n .
So, we have proved all.
(6) Define
fx = ]
x
x+1
sint
2
dt.
(a) Prove that |fx| < 1/x if x > 0.
Proof: Let x > 0, then we have, by change of variable(u = t
2
), and integration by
parts,
fx =
1
2
]
x
2
x+1
2
sinu
u
du
=
1
2
]
x
2
x+1
2
dcos u
u
=
1
2
cos u
u
x
2
x+1
2
+ ]
x
2
x+1
2
cos u
2u
3/2
du
=
cosx
2

2x

cos x + 1
2
2x + 1
]
x
2
x+1
2
cos u
4u
3/2
du
which implies that,
|fx| _
cosx
2

2x
+
cos x + 1
2
2x + 1
+ ]
x
2
x+1
2
cos u
4u
3/2
du
<
1
2x
+
1
2x + 1
+
1
4
]
x
2
x+1
2
du
u
3/2
=
1
2x
+
1
2x + 1

1
2x + 1
+
1
2x
= 1/x.
Note: There is another proof by Second Mean Value Theorem to show above as
follows. Since
fx =
1
2
]
x
2
x+1
2
sinu
u
du by (a),
we know that, by Second Mean Value Theorem,
fx =
1
2
1
x
]
x
2
y
sinudu +
1
x + 1
]
y
x+1
2
sinudu
=
1
2
1
x
cosx
2
cos y +
1
x + 1
cos y cos x + 1
2
=
1
2

1
x
+
1
x + 1
cosy +
1
x
cosx
2

1
x + 1
cos x + 1
2
which implies that
|fx| _
1
2

1
x
+
1
x + 1
|cos y| +
cosx
2

x
+
cos x + 1
2
x + 1
_
1
2
1
x

1
x + 1
+
cosx
2

x
+
cos x + 1
2
x + 1
<
1
2
1
x

1
x + 1
+
1
x
+
1
x + 1
since no x makes |cosx
2
| = cos x + 1
2
= 1
= 1/x.
(b) Prove that 2xfx = cosx
2
cos x + 1
2
+ rx, where |rx| < c/x and c is a
constant.
Proof: By (a), we have
fx =
cosx
2

2x

cos x + 1
2
2x + 1
]
x
2
x+1
2
cos u
4u
3/2
du
which implies that
2xfx = cosx
2

x
x + 1
cos x + 1
2
+ x ]
x
2
x+1
2
cos u
2u
3/2
du
= cosx
2
cos x + 1
2
+
1
x + 1
cos x + 1
2
+ x ]
x
2
x+1
2
cos u
2u
3/2
du
where
rx =
1
x + 1
cos x + 1
2
+
x
2
]
x
2
x+1
2
cos u
u
3/2
du
which implies that
|rx| _
1
x + 1
+
x
2
]
x
2
x+1
2
|cos u|
u
3/2
du
<
1
x + 1
+
x
2
]
x
2
x+1
2
u
3/2
du
=
1
x + 1
+
1
x + 1
<
2
x
.
Note: Of course, we can use the note in (a) to show it. We write it as follows.
Proof: Since
fx =
1
2

1
x
+
1
x + 1
cosy +
1
x
cosx
2

1
x + 1
cos x + 1
2
which implies that
2xfx =
x
x + 1
1 cosy + cosx
2

x
x + 1
cos x + 1
2
= cosx
2
cos x + 1
2
+
1
x + 1
cos x + 1
2
+
x
x + 1
1 cosy
where
rx =
1
x + 1
cos x + 1
2
+
x
x + 1
1 cosy
which implies that
|rx| _
1
x + 1
+ 1
x
x + 1
=
2
x + 1
< 2/x.
(c) Find the upper and lower limits of xfx, as x .
Proof: Claim that lim sup
x
cosx
2
cos x + 1
2
= 2 as follows. Taking
x = n 2m , where n Z, then
cosx
2
cos x + 1
2
= cos n 8m + 1 . 1
If we can show that n 8m is dense in 0, 2m modulus 2m. It is equivalent to show that
n
2
m
is dense in 0, 1 modulus 1. So, by lemma ar : a Z, where r Q
c
is dense
in 0, 1 modulus 1, we have proved the claim. In other words, we have proved the claim.
Note: We use the lemma as follows. ar + b : a Z, b Z, where r Q
c
is dense in
R. It is equivalent to ar : a Z, where r Q
c
is dense in 0, 1 modulus 1.
Proof: Say ar + b : a Z, b Z = S, and since r Q
c
, then by Exercise 1.16,
there are infinitely many rational numbers h/k with k > 0 such that |kr h| <
1
k
. Consider
x o, x + o := I, where o > 0, and thus choosing k
0
large enough so that 1/k
0
< o.
Define L = |k
0
r h
0
|, then we have sL I for some s Z. So,
sL = sk
0
r sh
0
S. That is, we have proved that S is dense in R.
(d) Does ]
0

sint
2
dt converge?
Proof: Yes,
]
x
x
'
sin
2
tdt =
1
2
]
x
x
'
sinu
u
du by the process of (a)
=
1
2
1
x
]
x
y
sinudu +
1
x
'
]
y
x
'
sinudu by Second Mean Value Theorem
=
1
2
2
x
+
2
x
'
<
2
x
which implies that the integral exists.
Note: (i) We can show it without Second Mean Value Theorem by the method of (a).
However Second Mean Value Theorem is more powerful for this exercise.
(ii) Here is the famous Integral named Dirichlet Integral used widely in the STUDY
of Fourier Series. We write it as follows. Show that the Dirichlet Integral
]
0

sinx
x
dx
converges but not absolutely converges. In other words, the Dirichlet Integral converges
conditionally.
Proof: Consider
]
x
x
'
sinx
x
dx =
1
x
]
x
y
sinxdx +
1
x
'
]
y
x
'
sinxdx by Second Mean Value Theorem;
we have
]
x
x
'
sinx
x
dx _
2
x
+
2
x
'
<
4
x
.
So, we know that Dirichlet Integral converges.
Define I
n
=
m
4
+ 2nm,
m
2
+ 2nm, then
]
0

sinx
x
dx _ ]
In
sinx
x
dx
_ ]
In
2
2
m
4
+ 2nm
dx
_
_
n=0

2
2

m
4

m
4
+ 2nm
.
So, we know that Dirichlet Integral does NOT converges absolutely.
(7) Deal similarity with
fx = ]
x
x+1
sine
t
dt.
Show that
e
x
|fx| < 2
and that
e
x
fx = cose
x
e
1
cose
x+1
+ rx,
where |rx| < min1, Ce
x
, for all x and
Proof: Since
fx = ]
x
x+1
sine
t
dt
= ]
e
x
e
x+1
sinu
u
du by Change of Variable (let u = e
t
)
=
cos e
x
e
x

cos e
x+1
e
x+1
]
e
x
e
x+1
cos u
u
2
du by Integration by parts *
which implies that
|fx| <
cos e
x
e
x
+
cos e
x+1
e
x+1
+ ]
e
x
e
x+1
du
u
2
since cos u is not constant 1
_
1
e
x
+
1
e
x+1
+
1
e
x
1
1
e
which implies that
e
x
|fx| < 2.
In addition, by (*), we have
e
x
fx = cose
x
e
1
cose
x+1
+ rx,
where
rx = e
x
]
e
x
e
x+1
cos u
u
2
du
which implies that
|rx| = 1 e
1
< 1 for all x **
or which implies that, by Integration by parts,
|rx| = e
x
]
e
x
e
x+1
cos u
u
2
du
= e
x sine
x+1
e
2x+1

sine
x
e
x
+ 2 ]
e
x
e
x+1
sinu
u
3
du
< e
x 1
e
2x+1
+
1
e
2x
+ 2 ]
e
x
e
x+1
du
u
3
since sinu is not constant 1
= 2e
x
for all x. ***
By (**) and (***), we have proved that |rx| < min1, Ce
x
for all x, where C = 2.
Note: We give another proof on (7) by Second Mean Value Theorem as follows.
Proof: Since
fx = ]
e
x
e
x+1
sinu
u
du
=
1
e
x
]
e
x
y
sinudu +
1
e
x+1
]
y
e
x+1
sinudu by Second Mean Value Theorem
=
1
e
x
cos e
x
cos y +
1
e
x+1
cos y cos e
x+1
*
which implies that
e
x
|fx| = |cos e
x
cos y + e
1
cos y cos e
x+1
|
= |cos e
x
e
1
cos e
x+1
+ cos y1 e
1
|
_ |cos e
x
e
1
cos e
x+1
| + 1 e
1

< 1 + e
1
+ 1 e
1
since no x makes |cos e
x
| = |cos e
x+1
| = 1.
= 2.
In addition, by (*), we know that
e
x
fx = cos e
x
e
1
cos e
x+1
+ rx
where
rx = e
1
cos y cos y
which implies that
|rx| _ 1 e
1
< 1 for all x. **
In addition, from the proof of the process in (7), we know that
|rx| = e
x
]
e
x
e
x+1
cos u
u
2
du
= e
x 1
e
2x
]
e
x
y
cos udu +
1
e
2x+1
]
y
e
x+1
cos udu
= e
x
|siny sine
x
+ e
2
sine
x+1
siny|
= e
x
|siny1 e
2
+ e
2
sine
x+1
sine
x
|
_ e
x
1 e
2
+ e
x
|e
2
sine
x+1
| + e
x
|sine
x
|
< e
x
1 e
2
+ e
x
1 + e
2
since no x makes |sine
x+1
| = |sine
x
| = 1
= 2e
x
for all x. ***
So, by (**) and (***), we have proved that |rx| < min1, Ce
x
, where C = 2.
(8) Suppose that f is real, continuously differentiable function on a, b,
fa = fb = 0, and
]
a
b
f
2
xdx = 1.
Prove that
]
a
b
xfxf
'
xdx =
1
2
and that
]
a
b
f
'
x
2
dx ]
a
b
x
2
f
2
xdx >
1
4
.
Proof: Consider
]
a
b
xfxf
'
xdx = ]
a
b
xfxdfx
= xf
2
x|
a
b
]
a
b
fxdxfx
= ]
a
b
f
2
xdx + ]
a
b
xfxf
'
xdx since fa = fb = 0,
so we have
]
a
b
xfxf
'
xdx =
1
2
.
In addition, by Cauchy-Schwarz Inequality, we know that
]
a
b
f
'
x
2
dx ]
a
b
x
2
f
2
xdx _ ]
a
b
xfxf
'
xdx
2
=
1
4
.
Note that the equality does NOT hold since if it was, then we have f
'
x = kxfx. It
implies that
f
'
x kxfxe
kx
2
2
= 0
which implies that
fe
kx
2
2
'
= 0
which implies that
fx = Ce
kx
2
2
, a constant
which implies that
C = 0 since fa = 0.
That is, fx = 0 on a, b which is absurb.
7.5 Let a
n
be a sequence of real numbers. For x _ 0, define
Ax =
_
n_x
a
n
=
_
n=1
x
a
n
,
where x is the largest integer in x and empty sums are interpreted as zero. Let f have a
continuous derivative in the interval 1 _ x _ a. Use Stieltjes integrals to derive the
following formula:
_
n_a
a
n
fn = ]
1
a
Axf
'
xdx + Aafa.
Proof: Since
]
1
a
Axf
'
xdx = ]
1
a
Axdfx since f has a continous derivative on 1, a
= ]
1
a
fxdAx + Aafa A1f1 by integration by parts
=
_
n_a
a
n
fn + Aafa by ]
1
a
fxdAx =
_
n=2
a
a
n
fn and A1 = a
1
,
we know that
_
n_a
a
n
fn = ]
1
a
Axf
'
xdx + Aafa.
7.6 Use Eulers summation formula, integration by parts in a Stieltjes integral, to
derive the following identities:
(a) _
k=1
n
1
k
s
=
1
n
s1
+ s ]
1
n
x
x
s+1
dx if s = 1.
Proof:
_
k=1
n
1
k
s
= ]
1
n
x
s
dx + 1
= ]
1
n
xdx
s
+ n
s
n 1
s
1 + 1
= s ]
1
n
x
x
s+1
dx + n
1s
=
1
n
s1
+ s ]
1
n
x
x
s+1
dx if s = 1.
(b) _
k=1
n
1
k
= logn ]
1
n
xx
x
2
+ 1.
Proof:
_
k=1
n
1
k
= ]
1
n
1
x
dx + 1
= ]
1
n
xdx
1
+ n
1
n 1
1
1 + 1
= ]
1
n
x
1
dx ]
1
n
x
1
dx + ]
1
n
x
x
2
dx + 1
= logn ]
1
n
x x
x
2
+ 1.
7.7 Assume that f
'
is continuous on 1, 2n and use Eulers summation formula or
integration by parts to prove that
_
k=1
2n
1
k
fk = ]
1
2n
f
'
xx 2x/2dx.
Proof:
_
k=1
2n
1
k
fk =
_
k=1
2n
fk + 2
_
k=1
n
f2k
= ]
1
2n
fxdx + f1 + 2 ]
1
2n
fxdx/2
= ]
1
2n
xdfx + 2nf2n + 2 ]
1
2n
x/2dfx + f2n2n/2 f11/2
since f
'
is continuous on 1, 2n
= ]
1
2n
f
'
xxdx 2nf2n ]
1
2n
f
'
xx/2dx + 2nf2n
= ]
1
2n
f
'
xx 2x/2dx.
7.8 Let
1
= x x
1
2
if x = integer, and let
1
= 0 if x = integer. Also, let

2
= ]
0
x

1
tdt. If f
''
is continuous on 1, n prove that Eulers summation formula implies
that
_
k=1
n
fk = ]
1
n
fxdx ]
1
n

2
xf
''
xdx +
f1 + fn
2
.
Proof: Using Theorem 7.13, then we have
_
k=1
n
fk = ]
1
n
fxdx + ]
1
n
f
'
x
1
xdx +
f1 + fn
2
= ]
1
n
fxdx + ]
1
n
f
'
xd
2
x +
f1 + fn
2
= ]
1
n
fxdx + ]
1
n

2
xdf
'
x + f
'
n
2
n f
'
1
2
1 +
f1 + fn
2
= ]
1
n
fxdx ]
1
n

2
xdf
'
x +
f1 + fn
2
= ]
1
n
fxdx ]
1
n

2
xf
''
xdx +
f1 + fn
2
since f
''
is continuous on 1, n.
7.9 Take fx = logx in Exercise 7.8 and prove that
logn! = n +
1
2
logn n + 1 + ]
1
n

2
t
t
2
dt.
Proof: Let fx = logx, then by Exercise 7.8, it is clear. So, we omit the proof.
Remark: By Eulers summation formula, we can show that
_
1<k_n
logk = ]
1
n
logxdx + ]
1
n
x x
1
2
dx
x
+
logn
2
. *
Since
x x
1
2
_ 1/2
and
]
a
a+1
x x
1
2
dx = 0 for all real a, **
we thus have the convergence of the improper integral
]
1

x x
1
2
dx by Second Mean Value Theorem.
So, by (*), we have
logn! = n +
1
2
logn n + C + ,
n
where
C = 1 + ]
1

x x
1
2
dx
x
,
and
,
n
= ]
n

x x
1
2
dx
x
.
So,
lim
n
n!
e
n
n
n+1/2
= e
C
:= C
1
. ***
Now, using Wallis formula, we have
lim
n
2 2 4 4 2n2n
1 3 3 5 5 2n 12n + 1
= m/2
which implies that
2
n
n!
4
2n!
2
2n + 1
1 + o1 = m/2
which implies that, by (***),
C
1
4
2
n
n
n+1/2
e
n

4
C
1
2
2n
2n+1/2
e
2n
2n + 1
1 + o1 = m/2
which implies that
C
1
2
n
22n + 1
1 + o1 = m/2.
Let n , we have C
1
= 2m , and ]
1

x x
1
2
dx =
1
2
log2m 1.
Note: In (***), the formula is called Stirling formula. The reader should be noted that
Wallis formula is equivalent to Stirling formula.
7.10 If x _ 1, let mx denote the number of primes p _ x, that is,
mx =
_
p_x
1,
where the sum is extended over all primes p _ x. The prime number theorem states that
lim
x
mx
logx
x
= 1.
This is usually proved by studying a related function 0 given by
0x =
_
p_x
logp,
where again the sum is extended over all primes p _ x. Both function m and 0 are step
functions with jumps at the primes. This exercise shows how the Riemann-Stieltjes integral
can be used to relate these two functions.
(a) If x _ 2, prove that mx and 0x can be expressed as the following
Riemann-Stieltjes integrals:
0x = ]
3/2
x
logtdmt, mx = ]
3/2
x
1
logt
d0t.
Note. The lower limit can be replaced by any number in the open interval 1, 2.
Proof: Since 0x = _
p_x
logp, we know that by Theorem 7.9,
0x = ]
3/2
x
logtdmt,
and mx = _
p_x
1, we know that by Theorem 7.9,
mx = ]
3/2
x
1
logt
d0t.
(b) If x _ 2, use integration by parts to show that
0x = mx logx ]
2
x
mt
t
dt,
mx =
0x
logx
+ ]
2
x
0t
t log
2
t
dt.
These equations can be used to prove that the prime number theorem is equivalent to
the relation lim
x
0x
x
= 1.
Proof: Use integration by parts, we know that
0x = ]
3/2
x
logtdmt = ]
3/2
x
mt
t
dt + logxmx log3/2m3/2
= ]
3/2
x
mt
t
dt + logxmx since m3/2 = 0
= mx logx ]
2
x
mt
t
dt since ]
3/2
2
mt
t
dt = 0 by mx = 0 on 0, 2,
and
mx = ]
3/2
x
1
logt
d0t = ]
3/2
x
0t
t log
2
t
dt +
0x
logx

03/2
log3/2
= ]
3/2
x
0t
t log
2
t
dt +
0x
logx
since 03/2 = 0
=
0x
logx
+ ]
2
x
0t
t log
2
t
dt since ]
3/2
2
0t
t log
2
t
dt = 0 by 0x = 0 on 0, 2.
7.11 If o on a, b, prove that
(a) ]
a
b

fdo = ]
a
c
fdo + ]
c
b

fdo, (a < c < b)

Proof: Given c > 0, there is a partition P such that
UP, f, o < I

a, b + c. 1
Let P
'
= c P = P
1
P
2
, where P
1
= a = x
0
, . . . , x
n
1
= c and
P
2
= x
n
1
= c, . . . , x
n
2
= bthen we have
I

a, c + I

c, b _ UP
1
, f, o + UP
2
, f, o = UP
'
, f, o _ UP, f, o. 2
So, by (1) and (2), we have
I

a, c + I

c, b _ I

a, b *
since c is arbitrary.
On the other hand, given c > 0, there is a partition P
1
and P
2
such that
UP
1
, f, o + UP
2
, f, o _ I

a, c + I

c, b + c
which implies that, let P = P
1
P
2
UP, f, o = UP
1
, f, o + UP
2
, f, o _ I

a, c + I

c, b + c. 3
Also,
I

a, b _ UP, f, o. 4
By (3) and (4), we have
I

a, b _ I

a, c + I

c, b **
since c is arbitrary.
So, by (*) and (**), we have proved it.
(b) ]
a
b

f + gdo _ ]
a
b

fdo + ]
a
b

gdo.
Proof: In any compact interval J, we have
sup
xJ
f + g _ sup
xJ
f + sup
xJ
g. 1
So, given c > 0, there is a partition P
f
and P
g
such that
_
k=1
n
1
M
k
fAo
k
_ ]
a
b

fdo + c/2 2
and
_
k=1
n
2
M
k
gAo
k
_ ]
a
b

gdo + c/2. 3
So, consider P = P
f
P
g
, then we have, by (1),
UP, f + g, o _ UP, f, o + UP, g, o
along with
UP, f, o _
_
k=1
n
1
M
k
fAo
k
and UP, g, o _
_
k=1
n
2
M
k
gAo
k
which implies that, by (2) and (3),
UP, f + g, o _ ]
a
b

fdo + ]
a
b

gdo + c
which implies that
]
a
b

f + gdo _ ]
a
b

fdo + ]
a
b

gdo
since c is arbitrary.
(c) ]
a
b
f + gdo _ ]
a
b
fdo + ]
a
b
gdo
Proof: Similarly by (b), so we omit the proof.
7.12 Give an example of bounded function f and an increasing function o defined on
a, b such that |f| Ro but for which ]
a
b
fdo does not exist.
Solution: Let
fx =
1 if x 0, 1 , Q
1 if x 0, 1 , Q
c
and ox = x on 0, 1. Then it is clear that f Ro on a, b and |f| Ro on a, b.
7.13 Let o be a continuous function of bounded variation on a, b. Assume that
g Ro on a, b and define [x = ]
a
x
gtdot if x a, b. Show that:
(a) If f on a, b, there exists a point x
0
in a, b such that
]
a
b
fd[ = fa ]
a
x
0
gdo + fb ]
x
0
b
gdo.
Proof: Since o is a continuous function of bounded variation on a, b, and g Ro
on a, b, we know that [x is a continuous function of bounded variation on a, b, by
Theorem 7.32. Hence, by Second Mean-Value Theorem for Riemann-Stieltjes
integrals, we know that
]
a
b
fd[ = fa ]
a
x
0
d[x + fb ]
x
0
b
d[x
which implies that, by Theorem 7.26,
]
a
b
fd[ = fa ]
a
x
0
gdo + fb ]
x
0
b
gdo.
(b) If, in addition, f is continuous on a, b, we also have
]
a
b
fxgxdx = fa ]
a
x
0
gdo + fb ]
x
0
b
gdo.
Proof: Since
]
a
b
fd[ = ]
a
b
fxgxdx by Theorem 7.26,
we know that, by (a),
]
a
b
fxgxdx = fa ]
a
x
0
gdo + fb ]
x
0
b
gdo.
Remark: We do NOT need the hypothesis that f is continuous on a, b.
7.14 Assume that f Ro on a, b, where o is of bounded variation on a, b. Let
Vx denote the total variation of o on a, x for each x in a, b, and let Va = 0. Show
that
]
a
b
fdo _ ]
a
b
|f|dV _ MVb,
where M is an upper bound for |f| on a, b. In particular, when ox = x, the inequality
becomes
]
a
b
fdo _ Mb a.
Proof: Given c > 0, there is a partition P = a = x
0
, . . . , x
n
= b such that
]
a
b
fdo c < SP, f, o
=
_
k=1
n
ft
k
Ao
k
, where t
k
x
k1
, x
k

_
_
k=1
n
|ft
k
||ox
k
ox
k1
|
_
_
k=1
n
|ft
k
|Vx
k
Vx
k1

= SP, |f|, V
_ UP, |f|, V since V is increasing on a, b
which implies that, taking infimum,
]
a
b
fdo c _ ]
a
b
|f|dV
since |f| RV on a, b.
So, we have
]
a
b
fdo _ ]
a
b
|f|dV *
since ]
a
b
|f|dV is clear non-negative. If M is an upper bound for |f| on a, b, then (*) implies
that
]
a
b
fdo _ ]
a
b
|f|dV _ MVb
which implies that
]
a
b
fdo _ Mb a
if ox = x.
7.15 Let o
n
be a sequence of functions of bounded variation on a, b. Suppose
there exists a function o defined on a, b such that the total variation of o o
n
on a, b
tends to 0 as n . Assume also that oa = o
n
a = 0 for each n = 1, 2, . . . . If f is
continuous on a, b, prove that
lim
n
]
a
b
fxdo
n
x = ]
a
b
fxdox.
Proof: Use Exercise 7.14, we then have
]
a
b
fxdo o
n
x _ MV
n
b 0 as n
where V
n
is the total variation of o o
n
, and M = sup
xa,b
|fx|.
So, we have
lim
n
]
a
b
fxdo
n
x = ]
a
b
fxdox.
Remark: We do NOT need the hypothesis oa = o
n
a = 0 for each n = 1, 2, . . . .
7.16 If f Ro, f
2
Ro, g Ro, and g
2
Ro on a, b, prove that
1
2
]
a
b
]
a
b
fx gx
fy gy
2
doy dox
= ]
a
b
f
2
xdox ]
a
b
g
2
xdox ]
a
b
fxgxdox
2
.
When o on a, b, deduce the Cauchy-Schwarz inequality
]
a
b
fxgxdox
2
_ ]
a
b
f
2
xdox ]
a
b
g
2
xdox .
(Compare with Exercise 1.23.)
Proof: Consider
1
2
]
a
b
]
a
b
fx gx
fy gy
2
doy dox
=
1
2
]
a
b
]
a
b
fxgy fygx
2
doy dox
=
1
2
]
a
b
]
a
b
f
2
xg
2
y 2fxgyfygx + f
2
yg
2
xdoy dox
=
1
2
]
a
b
f
2
xdox ]
a
b
g
2
ydoy
]
a
b
fxgxdox ]
a
b
fygydoy
+ ]
a
b
g
2
xdox ]
a
b
f
2
ydoy
= ]
a
b
f
2
xdox ]
a
b
g
2
ydoy ]
a
b
fxgxdox
2
,
if o on a, b, then we have
0 _
1
2
]
a
b
]
a
b
fx gx
fy gy
2
doy dox
= ]
a
b
f
2
xdox ]
a
b
g
2
ydoy ]
a
b
fxgxdox
2
which implies that
]
a
b
fxgxdox
2
_ ]
a
b
f
2
xdox ]
a
b
g
2
xdox .
Remark: (1) Here is another proof: Let A = ]
a
b
f
2
xdox, B = ]
a
b
fxgxdox, and
C = ]
a
b
g
2
xdox. From the fact,
0 _ ]
a
b
fxz + gx
2
dx for any real z
= Az
2
+ 2Bz + C.
It implies that
B
2
_ AC.
That is,
]
a
b
fxgxdox
2
_ ]
a
b
f
2
xdox ]
a
b
g
2
xdox .
Note: (1) The reader may recall the inner product in Linear Algebra. We often
consider Riemann Integral by defining
< f, g >:= ]
a
b
fxgxdx
where f and g are real continuous functions defined on a, b. This definition is a real case.
For complex case, we need to preserve its positive definite. So, we define
< f, g >:= ]
a
b
fxgxdx
where f and g are complex continuous functions defined on a, b, and g means its
conjugate. In addition, in this sense, we have the triangular inequality:
]f g] _ ]f h] + ]f h], where ]f] = < f, f > .
(2) Suppose that f Ro on a, b where o on a, b and given c > 0, then there
exists a continuous function g on a, b such that
]f g] < c.
Proof: Let K = sup
xa,b
|fx|, and given c > 0, we want to show that
]f g] < c.
Since f Ro on a, b where o on a, b, given 1 >c
'
> 0, there is a partition
P = x
0
= a, . . . , x
n
= b such that
UP, f, o LP, f, o =
_
j=1
n
M
j
f m
j
fAo
j
< c
'

2
. 1
Write P = A B, where A = x
j
: M
j
f m
j
f < c
'
and
B = x
j
: M
j
f m
j
f _ c
'
, then
c
_
B
Ao
j
_
_
B
M
j
f m
j
fAo
j
< c
'

2
by (1)
which implies that
_
B
Ao
j
< c
'
. 2
For this partition P, we define the function g as follows.
gt =
x
j
t
x
j
x
j1
fx
j1
+
t x
j1
x
j
x
j1
fx
j
, where x
j1
_ t _ x
j
.
So, it is clear that g is continuous on a, b. In every subinterval x
j1
, x
j

|ft gt| =
x
j
t
x
j
x
j1
ft fx
j1
+
t x
j1
x
j
x
j1
ft fx
j

_ |ft fx
j1
| + |ft fx
j
|
_ 2M
j
f m
j
f 3
Consider
_
A
]
x
j1
,x
j

|ft gt|
2
do _
_
A
4M
j
f m
j
f
2
Ao
j
by (3)
_ 4
_
A
c
'
M
j
f m
j
fAo
j
by definition of A
_ 4c
'
_
A
Ao
j
by c
'
< 1
= 4c
'
ob oa
and
_
B
]
x
j1
,x
j

|ft gt|
2
do _
_
B
4K
2
Ao
j
_ 4K
2
c
'
by (2).
Hence,
]
a
b
|ft gt|
2
do _ 4c
'
ob oa + 4K
2
c
'
< c
2
if we choose c
'
is small enough so that 4c
'
ob oa + 4K
2
c
'
< c
2
. That is, we have
proved that
]f g] < c.
P.S.: The exercise tells us a Riemann-Stieltjes integrable function can be approximated
(approached) by continuous functions.
(3)There is another important result called Holders inequality. It is useful in Analysis
and more general than Cauchy-Schwarz inequality. In fact, it is the case p = q = 2 in
Holders inequality. We consider the following results.
Let p and q be positive real numbers such that
1
p
+
1
q
= 1.
Prove that the following statements.
(a) If u _ 0 and v _ 0, then
uv _
u
p
p
+
v
p
q
.
Equality holds if and only if u
p
= v
q
.
Proof: Let fu =
u
p
p
+
v
p
q
uv be a function defined on 0, +, where
1
p
+
1
q
= 1,
p > 0, q > 0 and v _ 0, then f
'
u = u
p1
v. So, we know that
f
'
u < 0 if u 0, v
1
p1
and f
'
u > 0 if u v
1
p1
, +
which implies that, by f v
1
p1
= 0, fu _ 0. Hence, we know that fu _ 0 for all u _ 0.
That is, uv _
u
p
p
+
v
p
q
. In addition, fu = 0 if and only if u = v
1
p1
if and only if u
p
= v
q
.
So, Equality holds if and only if u
p
= v
q
.
Note: (1) Here is another good proof by using Youngs Inequality, let fx be an
strictly increasing and continuous function defined on x : x _ 0, with f0 = 0. Then
we have, let a > 0 and b > 0,
ab _ ]
0
a
fxdx + ]
0
b
f
1
xdx, where f
1
is the inverse function of f.
And the equality holds if and only if fa = b.
Proof: The proof is easy by drawing the function f on x y plane. So, we omit it.
So, by Youngs Inequality, let fx = x
o
, where o > 0, we have the Holders
inequality.
(2) The reader should be noted that there are many proofs of (a), for example, using the
concept of convex function, or using A. P. _ G. P. along with continuity.
(b) If f, g Ro on a, b where o on a, b, f, g _ 0 on a, b, and
]
a
b
f
p
do = 1 = ]
a
b
g
q
do,
then
]
a
b
fgdo _ 1.
Proof: By Holders inequality, we have
fg _
f
p
p
+
g
q
q
which implies that, by o on a, b, and ]
a
b
f
p
do = 1 = ]
a
b
g
q
do,
]
a
b
fgdo _ ]
a
b
f
p
p
do + ]
a
b
g
q
q
do =
1
p
+
1
q
= 1.
(c) If f and g are complex functions in Ro, where o on a, b, then
]
a
b
fgdo _ ]
a
b
|f|
p
do
1/p
]
a
b
|g|
q
do
1/q
. *
Proof: First, we note that
]
a
b
fgdo _ ]
a
b
|fg|do.
Also,
]
a
b
|f|
p
do = M
p
]
a
b
|f|
M
p
do = 1
and
]
a
b
|g|
q
do = N
q
]
a
b
|g|
N
q
do = 1.
Then we have by (b),
]
a
b
|f|
M
|g|
N
do _ 1
which implies that, by (*)
]
a
b
fgdo _ MN = ]
a
b
|f|
p
do
1/p
]
a
b
|g|
q
do
1/q
.
(d) Show that Holders inequality is also true for the improper integrals.
Proof: It is clear by (c), so we omit the proof.
7.17 Assume that f Ro, g Ro, and f g Ro on a, b. Show that
1
2
]
a
b
]
a
b
fy fxgy gxdoy dox
= ob oa ]
a
b
fxgxdox ]
a
b
fxdox ]
a
b
gxdox .
If o on a, b, deduce the inequality
]
a
b
fxdox ]
a
b
gxdox _ ob oa ]
a
b
fxgxdox
when both f and g are increasing (or both are decreasing) on a, b. Show that the reverse
inequality holds if f increases and g decreases on a, b.
Proof: Since
1
2
]
a
b
]
a
b
fy fxgy gxdoy dox
=
1
2
]
a
b
]
a
b
fygy fygx fxgy + fxgxdoy dox
= ob oa ]
a
b
fygydoy ]
a
b
fxdox ]
a
b
gxdox
which implies that, (let o, f, and g on a, b),
0 _
1
2
]
a
b
]
a
b
fy fxgy gxdoy dox
and (let o, and f on a, b, g on a, b),
0 _
1
2
]
a
b
]
a
b
fy fxgy gxdoy dox,
we know that, (let o, f, and g on a, b)
]
a
b
fxdox ]
a
b
gxdox _ ob oa ]
a
b
fxgxdox
and (let o, and f on a, b, g on a, b)
]
a
b
fxdox ]
a
b
gxdox _ ob oa ]
a
b
fxgxdox.
Riemann integrals
7.18 Assume f Ro on a, b. Use Exercise 7.4 to prove that the limit
lim
n
b a
n
_
k=1
n
f a + k
b a
n
exists and has the value ]
a
b
fxdx. Deduce that
lim
n
_
k=1
n
n
k
2
+ n
2
=
m
4
, lim
n
_
k=1
n
n
2
+ k
2

1/2
= log 1 + 2 .
Proof: Since f Ro on a, b, given c > 0, there exists a o > 0 such that as
]P] < o, we have
SP, f ]
a
b
fxdx < c.
For this o, we choose n large enough so that
ba
n
< o, that is, as n _ N, we have
ba
n
< o.
So,
SP, f ]
a
b
fxdx < c
which implies that
b a
n
_
k=1
n
f a + k
b a
n
]
a
b
fxdx < c.
That is,
lim
n
b a
n
_
k=1
n
f a + k
b a
n
exists and has the value ]
a
b
fxdx.
Since _
k=1
n
n
k
2
+n
2
=
1
n
_
k=1
n
1
k
n
2
+1
, we know that by above result,
lim
n
_
k=1
n
n
k
2
+ n
2
= lim
n
1
n
_
k=1
n
1

k
n

2
+ 1
= ]
0
1
dx
1 + x
2
= arctan1 arctan0
= m/4.
Since _
k=1
n
n
2
+ k
2

1/2
=
1
n
_
k=1
n
1
1+
k
n
2
1/2
, we know that by above result,
lim
n
_
k=1
n
n
2
+ k
2

1/2
= lim
n
1
n
_
k=1
n
1
1 +
k
n

2
1/2
= ]
0
1
dx
1 + x
2

1/2
= ]
0
m/4
sec0d0, let x = tan0
= ]
0
m/4
sec0
sec0 + tan0
sec0 + tan0
d0
= ]
1
1+ 2
du
u
, let sec0 + tan0 = u
= log 1 + 2 .
7.19 Define
fx = ]
0
x
e
t
2
dt
2
, gx = ]
0
1
e
x
2
t
2
+1
t
2
+ 1
dt.
(a) Show that g
'
x + f
'
x = 0 for all x and deduce that fx + gx = m/4.
Proof: Since
f
'
x = 2 ]
0
x
e
t
2
dt e
x
2
and note that if hx, t =
e
x
2
t
2
+1
t
2
+1
, we know that h is continuous on 0, a 0, 1 for any
real a > 0, and h
x
= 2xe
x
2
t
2
+1
is continuous on 0, a 0, 1 for any real a > 0,
g
'
x = ]
0
1
h
x
dt
= ]
0
1
2xe
x
2
t
2
+1
dt
= 2e
x
2
]
0
1
xe
xt
2
dt
= 2e
x
2
]
0
x
e
u
2
du,
we know that g
'
x + f
'
x = 0 for all x. Hence, we have fx + gx = C for all x,
constant. Since C = f0 + g0 = ]
0
1
dt
1+t
2
= m/4, fx + gx = m/4.
Remark: The reader should think it twice on how to find the auxiliary function g.
(b) Use (a) to prove that
lim
x
]
0
x
e
t
2
dt =
1
2
m .
Proof: Note that
hx, t =
e
x
2
t
2
+1
t
2
+ 1
_ |e
x
2
t
2
+1
| _
1
x
2
t
2
+ 1
for all x > 0;
we know that
]
0
1
e
x
2
t
2
+1
t
2
+ 1
dt _
1
x
2
]
0
1
dt
1 + t
2
0 as x .
So, by (a), we get
lim
x
fx = m/4
which implies that
lim
x
]
0
x
e
t
2
dt =
1
2
m
since lim
x
]
0
x
e
t
2
dt exists by ]
0
x
e
t
2
dt _ ]
0
x
dt
1+t
2
= arctanx m/2 as x .
Remark: (1) There are many methods to show this. But here is an elementary proof
with help of Taylor series and Wallis formula. We prove it as follows. In addition, the
reader will learn some beautiful and useful methods in the future. For example, use the
application of Gamma function, and so on.
Proof: Note that two inequalities,
1 + x
2
_ e
x
2
=
_
k=0

x
2k
k!
for all x
and
_
k=0

x
2k
k!
_
_
k=0

x
2k
=
1
1 x
2
if |x| < 1
which implies that
1 x
2
_ e
x
2
if 0 _ x _ 1 1 x
2

n
_ e
nx
2
1
and
e
x
2
_
1
1 + x
2
if x _ 0 e
nx
2
_
1
1 + x
2
n
. 2
So, we have, by (1) and (2),
]
0
1
1 x
2

n
dx _ ]
0
1
e
nx
2
dx _ ]
0

e
nx
2
dx _ ]
0

1
1 + x
2
n
dx. 3
Note that
]
0

e
nx
2
dx =
1
n
]
0

e
x
2
dx :=
K
n
.
Also,
]
0
1
1 x
2

n
dx = ]
0
m/2
sin
2n+1
tdt =
2 4 6 2n 22n
1 3 5 2n + 1
and
]
0

1
1 + x
2
n
dx = ]
0
m/2
sin
2n2
tdt =
1 3 5 2n 3
2 4 6 2n 2
m
2
,
so
n
2 4 6 2n 22n
1 3 5 2n + 1
_ K _ n
1 3 5 2n 3
2 4 6 2n 2
m
2
which implies that
n
2n + 1
2 4 6 2n 22n
2
1 3 5 2n 1
2
2n + 1
_ K
2
_
n
2n 1
1 3 5 2n 3
2
2n 1
2 4 6 2n 2
2
m
2
2
4
By Wallis formula, we know that, by (4)
K =
m
2
.
That is, we have proved that Euler-Possion Integral
]
0

e
x
2
dx =
m
2
.
Note: (Wallis formula)
lim
n
2 4 6 2n 22n
2
1 3 5 2n 1
2
2n + 1
=
m
2
.
Proof: As 0 _ x _ m/2, we have
sin
2n+1
t _ sin
2n
t _ sin
2n1
t, where n N.
So, we know that
]
0
m/2
sin
2n+1
tdt _ ]
0
m/2
sin
2n
tdt _ ]
0
m/2
sin
2n1
tdt
which implies that
2n2n 2 4 2
2n + 12n 1 3 1
_
2n 12n 3 3 1
2n2n 2 4 2
m
2
_
2n 22n 4 4 2
2n 12n 3 3 1
.
So,
2n2n 2 4 2
2n 12n 3 3 1
2
1
2n + 1
_
m
2
_
2n2n 2 4 2
2n 12n 3 3 1
2
1
2n
.
Hence, from
2n2n 2 4 2
2n 12n 3 3 1
2
1
2n

1
2n + 1
_
1
2n
m
2
0,
we know that
lim
n
2 4 6 2n 22n
2
1 3 5 2n 1
2
2n + 1
=
m
2
.
(2) Here is another exercise from Hadamards result. We Write it as follows. Let
f C
k
R with f0 = 0. Prove that there exists an unique function g C
k1
R such that
f = xgx on R.
Proof: Consider
fx = fx f0
= ]
0
1
dfxt
= ]
0
1
xf
'
xtdt
= x ]
0
1
f
'
xtdt;
we know that if gx := ]
0
1
f
'
xtdt, then we have prove it.
Note: In fact, we can do this job by rountine work. Define
gx =
fx
x
if x = 0
0 if x = 0.
However, it is too long to write. The trouble is to make sure that g C
k1
R.
7.20 Assume g R on a, b and define fx = ]
a
x
gtdt if t a, b. Prove that the
integral ]
a
x
|gt|dt gives the total variation of f on a, x.
Proof: Since ]
a
x
|gt|dt exists, given c > 0, there exists a partition
P
1
= x
0
= a, . . . , x
n
= x such that
LP, |g| > ]
a
x
|gt|dt c. 1
So, for this P
1
, we have
_
k=1
n
|fx
k
fx
k1
| =
_
k=1
n
]
x
k1
x
k
gtdt =
_
k=1
n
|c
k
x
k
x
k1
| by Mean Value Theorem 2
where inf
xx
k1
,x
k

|gx| _ c
k
_ sup
xx
k1
,x
k

|gx|.
Hence, we know that, by (1) and (2),
_
k=1
n
|fx
k
fx
k1
| > ]
a
x
|gt|dt c
which implies that
V
f
a, b _ ]
a
x
|gt|dt
since c is arbitrary.
Conversely, since ]
a
x
|gt|dt exists, given c > 0, there exists a partition P
2
such that
UP
2
, |g| < ]
a
x
|gt|dt + c/2. 3
Also, for the same c, there exists a partition P
3
= t
0
= a, . . . , t
m
= x such that
V
f
a, b c/2 <
_
k=1
m
|ft
k
ft
k1
|. 4
Let P = P
2
P
3
= s
0
= a, . . . , s
p
= x, then by (3) and (4), we have
UP, |g| < ]
a
x
|gt|dt + c/2
and
V
f
a, b c/2 <
_
k=1
p
|fs
k
fs
k1
|
=
_
k=1
p
]
s
k1
s
k
gtdt
=
_
k=1
p
|c
k
x
k
x
k1
|
_ UP, |g|
which imply that
V
f
a, b _ ]
a
x
|gt|dt
since c is arbitrary.
Therefore, from above discussion, we have proved that
V
f
a, b = ]
a
x
|gt|dt.
7.21 If f = f
1
, . . . , f
n
be a vector-valued function with a continuous derivative f
'
on
a, b. Prove that the curve described by f has length
A
f
a, b = ]
a
b
]f
'
t]dt.
Proof: Since f
'
= f
1
'
, . . . , f
n
'
is continuous on a, b, we know that
_
j=1
n
f
j
'

2
t
1/2
= ]f
'
t] is uniformly continuous on a, b. So, given c > 0, there exists
a o
1
> 0 such that as |x y| < o
1
, where x, y a, b, we have
|]f
'
x] ]f
'
y]| <
c
3b a
. 1
Since ]f
'
t] R on a, b, for the same c, there exists o
2
> 0 such that as ]P
1
] < o
2
,
where P
1
= x
0
= a, . . . , x
n
= bwe have
SP
1
, ]f
'
] ]
a
b
]f
'
t]dt < c/3, where SP
1
, ]f
'
] =
_
j=1
n
]f
'
t
j
]Ax
j
2
and A
f
a, b exists by Theorem 6.17, for the same c, there exists a partition
P
2
= s
0
= a, . . . , s
m
= b such that
A
f
a, b c/3 <
_
k=1
m
]fs
k
fs
k1
]
=
_
k=1
m
_
j=1
n
f
j
s
k
f
j
s
k1

2
1/2
. 3
Let o = mino
1
, o
2
and P P
2
so that ]P] < o, where P = y
0
= a, . . . , y
q
= bthen by
(1)-(3), we have
(i) As |x y| < o, where x, y a, b, we have
|]f
'
x] ]f
'
y]| <
c
3b a
. 4
(ii) As ]P] < o, we have
SP, ]f
'
] ]
a
b
]f
'
t]dt < c/3, where SP, ]f
'
] =
_
j=1
q
]f
'
t

j
]Ay
j
5
(iii) As ]P] < o, we have
A
f
a, b c/3 _
_
k=1
m
_
j=1
n
f
j
s
k
f
j
s
k1

2
1/2
_
_
k=1
q
_
j=1
n
f
j
y
k
f
j
y
k1

2
1/2
=
_
k=1
q
_
j=1
n
f
j
'
z
k

2
Ay
j
, by Mean Value Theorem
=
_
k=1
q
]f
'
z
k
]Ay
j
_ A
f
a, b 6
By (ii) and (iii), we have
_
k=1
q
gz
k
Ay
j
SP, g =
_
k=1
q
gz
k
Ay
j

_
j=1
q
gt

j
Ay
j
_
_
k=1
q
|gz
k
gt

j
|Ay
j
<
_
k=1
q
c
3b a
Ay
j
= c/3. 7
Hence, (5)-(7) implies that
]
a
b
]f
'
t]dt A
f
a, b < c.
Since c is arbitrary, we have proved that
A
f
a, b = ]
a
b
]f
'
t]dt.
7.22 If f
n+1
is continuous on a, x, define
I
n
x =
1
n!
]
a
x
x t
n
f
n+1
tdt.
(a) Show that
I
k1
x I
k
x =
f
k
ax a
k
k!
, k = 1, 2, . . . , n.
Proof: Since, for k = 1, 2, . . . , n,
I
k
x =
1
k!
]
a
x
x t
k
f
k+1
tdt
=
1
k!
]
a
x
x t
k
df
k
t
=
1
k!
x t
k
f
k
t
a
x
+ k ]
a
x
x t
k1
f
k
tdt
=
f
k
ax a
k
k!
+
1
k 1!
]
a
x
x t
k1
f
k
tdt
=
f
k
ax a
k
k!
+ I
k1
x,
we know that
I
k1
x I
k
x =
f
k
ax a
k
k!
, for k = 1, 2, . . . , n.
(b) Use (a) to express the remainder in Taylors formula (Theorem 5.19) as an integral.
Proof: Since fx fa = I
0
x, we know that
fx = fa + I
0
x
= fa +
_
k=1
n
I
k1
x I
k
x + I
n
x
=
_
k=0
n
f
k
ax a
k
k!
+
1
n!
]
a
x
x t
n
f
n+1
tdt by (a).
So, by Taylors formula, we know that
R
n
x =
f
n+1
cx a
n+1
n + 1!
=
1
n!
]
a
x
x t
n
f
n+1
tdt, for some c a, x.
where R
n
x is the remainder term.
Remark: 1. The reader should be noted that with help of Mean Value Theorem, we
have
1
n!
]
a
x
x t
n
f
n+1
tdt =
f
n+1
cx a
n+1
n + 1!
. *
2. Use Integration by parts repeatedly; we can show (*). Of course, there is other
proofs such as Mathematical Induction.
Proof: Since
] uv
n+1
dt = uv
n
u
'
v
n1
+ u
''
v
n2
. . . +1
n
u
n
v + 1
n+1
] u
n+1
vdt,
letting vt = x t
n
and ut = ft, then
fx =
_
k=0
n
f
k
a
k!
x a
k
+
1
n!
]
a
x
x t
n
f
n+1
tdt.
Note: The reader should give it a try to show it. Since it is not hard, we omit the detail.
3. The remainder term as an integral is useful; the reader should see the textbook in
Ch9, pp242-244.
4. There is a good exercise related with an application of Taylors Remainder. We write
it as a reference.
Let u
''
t + ftut = 0, where ft is continuous and non-negative on 0, c If u is
defined and not a zero function on 0, c and
]
a
b
b ta tft < b a for all a, b 0, c, where a < b. *
Then u at most has one zero on 0, c.
Proof: First, we note that u has at most finitely many zeros in the interval 0, c by
uniqueness theorem on O.D.E. So, let ua = ub = 0, where a, b 0, c with a < b,
and no point y a, b such that uy = 0. Consider a, b and by Taylors Theorem with
Remainder Term as an integral, we have
ux = ua + u
'
ax a + ]
a
x
x tu
''
tdt
= u
'
ax a + ]
a
x
x tu
''
tdt
= u
'
ax a ]
a
x
x tutftdt. **
Note that ux is positive on a, b ( Or, ux is negative on a, b ) So, we have
|ux| _ |u
'
a|x a. ***
By (**),
0 = ub = u
'
ab a ]
a
b
b tutftdt
which implies that
u
'
ab a = ]
a
b
b tutft
which implies that by (***), and note that u
'
a = 0,
b a _ ]
a
b
b tt aftdt
which contradicts to (*). So, u at most has one zero on 0, c.
Note: (i) In particular, let ft = e
t
, we have (*) holds.
Proof: Since
]
a
b
b tt ae
t
dt = e
a
2 + b a + e
b
2 + b a
by integration by parts twice, we have, (let b a = x),
e
a
2 + b a + e
b
2 + b a b a
= e
a
2 + x + e
xa
2 + x x
= xe
a
1 + e
ax
2e
x
+ x + 2
< 0 since a < b and e
x
> 1 + x.
(ii) In the proof of exercise, we use the uniqueness theorem: If px and qx are
continuous on 0, a, then
y
''
+ pxy
'
+ qxy = 0, where y0 = y
0
, and y
'
0 = y
0
'
has one and only one solution. In particular, if y0 = y
'
0 = 0, then y = 0 on 0, a is the
only solution. We do NOT give a proof; the reader can see the book, Theory of Ordinary
Differential Equation by Ince, section 3.32, or Theory of Ordinary Differential
Equation by Coddington and Levison, Chapter 6.
However, we need use the uniqueness theorem to show that u (in the exercise) has at
most finitely many zeros in 0, c.
Proof: Let S = x : ux = 0, x 0, c. If #S = , then by Bolzano-Weierstrass
Theorem, S has an accumulation point p in 0, c. Then up = 0 by continuity of u. In
addition, let r
n
p, and ur
n
= 0, then
u
'
p = lim
xp
ux up
x p
= lim
n
ur
n
up
r
n
p
= 0.
(Note that if p is the endpoint of 0, c, we may consider x p
+
or x p

). So, by
uniqueness theorem, we then have u = 0 on 0, c which contradicts to the hypothesis, u
is not a zero function on 0, c. So, #S < .
7.23 Let f be continuous on 0, a. If x 0, a, define f
0
x = fx and let
f
n+1
x =
1
n!
]
0
x
x t
n
ftdt, n = 0, 1, 2, . . .
(a) Show that the nth derivative of f
n
exists and equals f.
Proof: Consider, by Chain Rule,
f
n
'
=
1
n 1!
]
0
x
x t
n1
ftdt = f
n1
for all n N,
we have
f
n
n
= f.
That is, nth derivative of f
n
exists and equals f.
Remark: (1) There is another proof by Mathematical Induction and Integration by
parts. It is not hard; we omit the proof.
(2) The reader should note that the exercise tells us that given any continuous function f
on a, b, there exists a function g
n
on a, b such that g
n
n
= f, where n N. In fact, the
function
g
n
=
1
n!
]
a
x
x t
n
ftdt, n = 0, 1, 2, . . .
(3) The reader should compare the exercise with 7.22. At the same time, look at two
integrands in both exercises.
(b) Prove the following theorem of M. Fekete: The number of changes in sign of f in
0, a is not less than the number of changes in sign in the ordered set of numbers
fa, f
1
a, . . . , f
n
a.
Hint: Use mathematical induction.
Proof: Let Tf denote the number of changes in sign of f on 0, a and S
n
f the
number of changes in sign in the ordered set of numbers
fa, f
1
a, . . . , f
n
a.
We prove Tf _ S
n
f for each n by Mathematical Induction as follows. Note that
S
n
f _ n.
As n = 1, if S
1
f = 0, then there is nothing to prove it. If S
1
f = 1, it means that
faf
1
a < 0. Without loss of generality, we may assume that fa > 0, so f
1
a < 0
which implies that
0 > f
1
a =
1
0!
]
0
a
ftdt
which implies that there exists a point y 0, a such that fy < 0. Hence, Tf _ S
1
f
holds for any continuous functions defined on 0, a.
Assume that n = k holds for any continuous functions defined on 0, a, As n = k + 1, .
we consider the ordered set of numbers
fa, f
1
a, . . . , f
k
a, f
k+1
a.
Note that
f
n+1
a = f
1

n
a for all n N,
so by induction hypothesis,
Tf
1
_ S
k
f
1

Suppose S
k
f
1
= p, and f
1
0 = 0, then f
1
'
= f at least has p zeros by Rolles
Theorem. Hence,
Tf _ Tf
1
_ S
k
f
1
= p *
We consider two cases as follows.
(i) faf
1
a _ 0 :With help of (*),
Tf _ S
k
f
1
= S
k+1
f.
(ii) faf
1
a < 0 :Claim that
Tf > S
k
f
1
= p
as follows. Suppose NOT, it means that Tf = Tf
1
= p by (*). Say
fa
1
= fa
2
=. . . = fa
p
= 0, where 0 < a
1
< a
2
<. . . < a
p
< 1.
and
f
1
b
1
= f
1
b
2
=. . . = f
1
b
p
= 0, where 0 < b
1
< b
2
<. . . < b
p
< 1.
By faf
1
a < 0, we know that
fxf
1
x < 0 where x 0, c, c = mina
1
, b
1

which is impossible since

fxf
1
x = fxf
1
x f
1
0 by f
1
0 = 0
= fxf
1
'
y, where y 0, x 0, c
= fxfy
> 0 since fx and fy both positive or negative.
So, we obtain that Tf > S
k
f
1
= p. That is, Tf _ S
k
f
1
+ 1 = S
k+1
f.
From above results, we have proved it by Mathmatical Induction.
(c) Use (b) to prove the following theorem of Fejer: The number of changes in sign of
f in 0, a is not less than the number of changes in sign in the ordered set
f0, ]
0
a
ftdt, ]
0
a
tftdt, . . . , ]
0
a
t
n
ftdt.
Proof: Let gx = fa x, then, define g
0
x = gx, and for n = 0, 1, 2, . . . ,
g
n+1
a =
1
n!
]
0
a
a t
n
gtdt
=
1
n!
]
0
a
u
n
fudu by change of variable (u = a t).
So, by (b), the number of changes in sign of g in 0, a is not less than the number of
changes in sign in the ordered set
ga, g
1
a, . . . , g
n+1
a.
That is, the number of changes in sign of g in 0, a is not less than the number of changes
in sign in the ordered set
f0, ]
0
a
ftdt, ]
0
a
tftdt, . . . , ]
0
a
t
n
ftdt.
Note that the number of changes in sign of g in 0, a equals the number of changes in
sign of f in 0, a, so we have proved the Fejer Theorem.
7.24 Let f be a positive continuous function in a, b. Let M denote the maximum
value of f on a, b. Show that
lim
n
]
a
b
fx
n
dx
1/n
= M.
Proof: Since f is a positive continuous function in a, b, there exists a point c a, b
such that fc = M = sup
xa,b
fx > 0. Then given M >c > 0, there is a o > 0 such
that as x Bc, o , a, b := I, we have
0 <M c < fx < M + c.
Hence, we have
|I|
1/n
M c _ ]
I
f
n
xdx
1/n
_ ]
a
b
fx
n
dx
1/n
_ b a
1/n
M
which implies that
M c _ lim
n
inf ]
a
b
fx
n
dx
1/n
_ M.
So, lim
n
inf ]
a
b
fx
n
dx
1/n
= M since c is arbitrary. Similarly, we can show that
lim
n
sup ]
a
b
fx
n
dx
1/n
= M. So, we have proved that lim
n
]
a
b
fx
n
dx
1/n
= M.
Remark: There is good exercise; we write it as a reference. Let fx and gx are
continuous and non-negative function defined on a, b. Then
lim
n
]
a
b
fx
n
gxdx
1/n
= max
xa,b
fx.
Since the proof is similar, we omit it. (The reader may let ox = ]
a
x
gtdt).
7.25 A function f of two real variables is defined for each point x, y in the unit square
0 _ x _ 1, 0 _ y _ 1 as follows:
fx, y =
1 if x is rational,
2y if x is irrational.
(a) Compute ]
0
1

fx, ydx and ]

0_
1
fx, ydx in terms of y.
Proof: Consider two cases for upper and lower Riemann-Stieltjes integrals as
follows.
(i) As y 0, 1/2 : Given any partition P = x
0
= 0, . . . , x
n
= 1, we have
sup
xx
j1
,x
j

fx, y = 1, and inf

xx
j1
,x
j

fx, y = 2y.
Hence, ]
0
1

fx, ydx = 1, and ]

0_
1
fx, ydx = 2y.
(ii) As y 1/2, 1 : Given any partition P = x
0
= 0, . . . , x
n
= 1, we have
sup
xx
j1
,x
j

fx, y = 2y, and inf

xx
j1
,x
j

fx, y = 1.
Hence, ]
0
1

fx, ydx = 2y, and ]

0_
1
fx, ydx = 1.
(b) Show that ]
0
1
fx, ydy exists for each fixed x and compute ]
0
t
fx, ydy in terms of x
and t for 0 _ x _ 1, 0 _ t _ 1.
Proof: If x Q , 0, 1, then fx, y = 1. And if x Q
c
, 0, 1, then fx, y = 2y. So,
for each fixed x, we have
]
0
1
fx, ydy = ]
0
1
1dy = 1 if x Q , 0, 1
and
]
0
1
fx, ydy = ]
0
1
2ydy = 1 if x Q
c
, 0, 1.
In addition,
]
0
t
fx, ydy = ]
0
t
1dy = t if x Q , 0, 1
and
]
0
t
fx, ydy = ]
0
t
2ydy = t
2
if x Q
c
, 0, 1.
(c) Let Fx = ]
0
1
fx, ydy. Show that ]
0
1
Fxdx exists and find its value.
Proof: By (b), we have
Fx = 1 on 0, 1.
So, ]
0
1
Fxdx exists and
]
0
1
Fxdx = 1.
7.26 Let f be defined on 0, 1 as follows: f0 = 0; if 2
n1
< x _ 2
n
, then
fx = 2
n
for n = 0, 1, 2, . . .
(a) Give two reasons why ]
0
1
fxdx exists.
Proof: (i) fx is monotonic decreasing on 0, 1. (ii) x : f is discontinuous at x has
measure zero.
Remark: We compute the value of the integral as follows.
Solution: Consider the interval I
n
= 2
n
, 1 where n N, then we have f R on I
n
for
each n, and
]
2
n
1
fxdx =
_
k=1
n
]
2
k
2
k+1
fxdx
=
_
k=1
n
2
k+1
]
2
k
2
k+1
dx
=
_
k=1
n
2
k+1
2
k

= 2
1
4
1
1
4

n
1
1
4
=
2
3
1
1
4
n

2
3
as n .
So, th integral ]
0
1
fxdx =
2
3
.
Note: In the remark, we use the following fact. If f R on a, b, then
]
a
b
fxdx = lim
n
]
an
b
fxdx
where a
n
is a sequence with a
n
a, and a
n
_ a for all n.
Proof: Since a
n
a, given c > 0, there is a positive integer N such that as n _ N, we
have
|a
n
a| < c/M, where M = sup
xa,b
|fx|
So,
]
a
b
fxdx ]
an
b
fxdx = ]
a
an
fxdx
_ M|a
n
a|
< c.
That is, ]
a
b
fxdx = lim
n
]
an
b
fxdx.
(b) Let Fx = ]
0
x
ftdt. Show that for 0 < x _ 1 we have
Fx = xAx
1
3
Ax
2
,
where Ax = 2
logx/ log2
and where y is the greatest integer in y.
Proof: First, we note that Fx = ]
0
1
ftdt ]
x
1
ftdt =
2
3
]
x
1
ftdt. So, it suffices to
consider the value of the integral
]
x
1
ftdt.
Given any x 0, 1, then there exists a positive N such that 2
N1
< x _ 2
N
. So,
]
x
1
ftdt = ]
2
N
1
ftdt + ]
x
2
N
ftdt
=
2
3
1
1
4
N
+ 2
N
2
N
x by Remark in (a)
=
2
3
+
1
3
1
4
N

1
2
N
x.
So,
Fx =
1
2
N
x
1
3
1
4
N
= 2
N
x
1
3
2
2N
= xAx
1
3
Ax
2
where Ax = 2
N
. Note that 2
N1
< x _ 2
N
, we have
N = log
1/2
x , where y is the Gauss symbol.
Hence,
Ax = 2
logx/ log2
.
Remark: (1) The reader should give it a try to show it directly by considering 0, x,
where 0 _ x _ 1.
(2) Here is a good exercise. We write it as a reference. Suppose that f is defined on
0, 1 by the following
fx =
1
2
n
if x =
j
2
n
where j is an odd integer and 0 < j < 2
n
, n = 1, 2, . . . ,
0 otherwise.
Show that f R on 0, 1 and has the value of the integral 0.
Proof: In order to show this, we consider the Riemanns condition with respect to
ox = x as follows. Given a partition
P = x
0
= 0 =
0
2
n
, x
1
=
1
2
n
, x
2
=
2
2
n
, . . . , x
j
=
j
2
n
, . . . , x
2
n =
2
n
2
n
= 1 , then the upper
sum
UP, f =
_
k=1
2
n
M
k
Ax
k
=
1
2
n
_
k=1
2
n
M
k
=
1
2
n
2
2
n
+ n 1 0 as n .
So, f satisfies the Riemanns condition on 0, 1.
Note: (1) The reader should give it a try to show that the set of discontinuities of f has
measure zero. Thus by Theorem 7.48 (Lebesgues Criterion for Riemann Integral), we
know that f R on 0, 1. In addition, by the fact, the lower Riemann integral equals the
Riemann integral, we know that its integral is zero.
(2) For the existence of Riemann integral, we summarize to be the theorem: Let f be a
bounded function on a, b. Then the following statements are equivalent:
(i) f R on a, b.
(ii) f satisfies Riemanns condition on a, b.
(iii) ]
a
b

fxdx = ]
a
b
fxdx
(iv) the set of discontinuities of f on I has measure zero.
P.S.: The reader should see the textbook, pp 391; we have the general discussion.
7.27 Assume f has a derivative which is monotonic decreasing and satisfies
f
'
x _ m > 0 for all x in a, b. Prove that
]
a
b
cos fxdx _
2
m
.
Hint: Multiply and divide the integrand by f
'
x and use Theorem 7.37(ii).
Proof: Since f
'
x _ m > 0, and
1
f
'
is monotonic increasing on a, b, we consider
]
a
b
cos fxdx = ]
a
b
cos fx
f
'
x
f
'
xdx
=
1
f
'
b
]
c
b
cos fxf
'
xdx, by Theorem 7.37(ii)
=
1
f
'
b
]
fc
fb
cos udu, by Change of Variable
=
sinfb sinfc
f
'
b
which implies that
]
a
b
cos fxdx _
2
m
.
7.28 Given a decreasing seq uence of real numbers Gn such that Gn 0 as
n . Define a function f on 0, 1 in terms of Gn as follows: f0 = 1; if x is
irrational, then fx = 0; if x is rational m/n (in lowest terms), then fm/n = Gn.
Compute the oscillation o
f
x at each x in 0, 1 and show that f R on 0, 1.
Proof: Let x
0
Q
c
, 0, 1. Since lim
n
Gn = 0, given c > 0, there exists a
positive integer K such that as n _ K, we have |Gn| < c. So, there exists a finite number
of positive integers n such that Gn _ c. Denote S = x : |fx| _ c, then #S < .
Choose a o > 0 such that x
0
o, x
0
+ o 0, 1 does NOT contain all points of S. Note
that fx
0
= 0. Hence, we know that f is continuous at x
0
. That is, o
f
x = 0 for all
x Q
c
, 0, 1.
Let x
0
= 0, then it is clear that o
f
0 = 1= f0 > 0. So, f is not continuous at
0. Q , 0, 1.
Let x
0
Q , 0, 1, say x
0
=
M
N
(in lowest terms). Since Gn is monotonic
decreasing, there exists a finite number of positive integers n such that Gn _ GN.
Denote T = x : |fx| _ GN, then #T < . Choose a o > 0 such that
x
0
o, x
0
+ o , 0, 1 does NOT contain all points of T. Let h 0, o, then
supfx fy : x, y x
0
h, x
0
+ h , 0, 1 = fx
0
= GN.
So, o
f
x
0
= GN. That is, o
f
x = fx for all x Q , 0, 1.
Remark: (1) If we have proved f is continuous on Q
c
, 0, 1, then f is automatically
Riemann integrable on 0, 1 since D Q , 0, 1 Q, the set of discontinuities of f has
measure zero.
(2) Here is a good exercise. We write it as a reference. Given a function f defined on
a, b, then the set of continuities of f on a, b is G
o
set.
Proof: Let C denote the set of continuities of f on a, b, then
C = x : o
fx
= 0
= ,
k=1

x : o
f
x < 1/k
and x : o
fx
< 1/k is open. We know that C is a G
o
set.
Note: (i) We call S a G
o
set if S = ,
n=1

O
n
, where O
n
is open for each n.
(ii) Given y x a, b : o
f
x < 1/k := I, then o
f
y < 1/k. Hence, there exists a
d > 0, such that
O
f
By, d < 1/k, where By, d a, b
For z By, d, consider a smaller o so that Bz, o By, d. Hence,
O
f
Bz, o < 1/k
which implies that
o
f
z < 1/k.
Hence, By, d I. That is, y is an interior point of I. That is, I is open since every point
of I is interior.
7.29 Let f be defined as in Exercise 7.28 with Gn = 1/n. Let gx = 1 if
0 < x _ 1, g0 = 0. Show that the composite function h defined by hx = gfx is not
Riemann-integrable on 0, 1, although both f R and g R on 0, 1.
Proof: By Exercise 7.28, we know that
hx =
0 if x Q
c
, 0, 1
1 if x Q , 0, 1
which is discontinuous everywhere on 0, 1. Hence, the function h (Dirichlet Function) is
not Riemann-integrable on 0, 1.
7.30 Use Lebesgues theorem to prove Theorem 7.49.
(a) If f is of bounded variation on a, b, then f R on a, b.
Proof: Since f is of bounded variation on a, b, by Theorem 6.13 (Jordan Theorem),
f = f
1
f
2
, where f
1
and f
2
are increasing on a, b. Let D
i
denote the set of discontinuities
of f
i
on a, b, i = 1, 2. Hence, D, the set of discontinuities of f on a, b is
D D
1
D
2
.
Since |D
1
| = |D
2
| = 0, we know that |D| = 0. In addition, f is of bounded on a, b since f
is bounded variation on a, b. So, by Theorem 7.48, f R on a, b.
(b) If f R on a, b, then f R on c, d for every subinterval c, d a, b, |f| R
and f
2
R on a, b. Also, f g R on a, b whenever g R on a, b.
Proof: (i) Let D
a,b
and D
c,d
denote the set of discontinuities of f on a, b and c, d,
respectively. Then
D
c,d
D
a,b
.
Since f R on a, b, and use Theorem 7.48, |D
a,b
| = 0 which implies that |D
c,d
| = 0.
In addition, since f is bounded on a, b, f is automatically is bounded on c, d for every
compact subinterval c, d. So, by Theorem 7.48, f R on c, d.
(ii) Let D
f
and D
|f|
denote the set of discontinuities of f and |f| on a, b, respectively,
then
D
|f|
D
f
.
Since f R on a, b, and use Theorem 7.48, |D
f
| = 0 which implies that |D
|f|
| = 0. In
addition, since f is bounded on a, b, it is clear that |f| is bounded on a, b. So, by
Theorem 7.48, |f| R on a, b.
(iii) Let D
f
and D
f
2 denote the set of discontinuities of f and f
2
on a, b, respectively,
then
D
f
2 D
f
.
Since f R on a, b, and use Theorem 7.48, |D
f
| = 0 which implies that D
f
2 = 0. In
addition, since f is bounded on a, b, it is clear that f
2
is bounded on a, b. So, by
Theorem 7.48, f
2
R on a, b.
(iv) Let D
f
and D
g
denote the set of discontinuities of f and g on a, b, respectively.
Let D
fg
denote the set of discontinuities of fg on a, b, then
D
fg
D
f
D
g
Since f, g R on a, b, and use Theorem 7.48, |D
f
| = |D
g| = 0 which implies that
|D
fg
| = 0. In addition, since f and g are bounded on a, b, it is clear that fg is bounded on
a, b. So, by Theorem 7.48, fg R on a, b.
(c) If f R and g R on a, b, then f/g R on a, b whenever g is bounded away
from 0.
Proof: Let D
f
and D
g
denote the set of discontinuities of f and g on a, b, respectively.
Since g is bounded away from 0, we know that f/g is well-defined and f/g is also bounded
on a, b. Consider D
f/g
, the set of discontinuities of f/g on a, b is
D
f/g
D
f
D
g
.
Since f R and g R on a, b, by Theorem 7.48, |D
f
| = |D
g| = 0 which implies that
|D
f/g
| = 0. Since f/g is bounded on a, b with |D
f/g
| = 0, f/g R on a, b by Theorem
7.48.
Remark: The condition that the function g is bounded away from 0 CANNOT omit.
For example, say gx = x on 0, 1 and g0 = 1. Then it is clear that g R on 0, 1, but
1/g R on 0, 1. In addition, the reader should note that when we ask whether a function
is Riemann-integrable or not, we always assume that f is BOUNDED on a COMPACT
INTERVAL a, b.
(d) If f and g are bounded functions having the same discontinuities on a, b, then
f R on a, b if, and only if, g R on a, b.
Proof: () Suppose that f R on a, b, then D
f
, the set of discontinuities of f on a, b
has measure zero by Theorem 7.48. From hypothesis, D
f
= D
g
, the set of discontinuities
of g on a, b, we know that g R on a, b by Theorem 7.48.
() If we change the roles of f and g, we have proved it.
(e) Let g R on a, b and assume that m _ gx _ M for all x a, b. If f is
continuous on m, M, the composite function h defined by hx = fgx is
Riemann-integrable on a, b.
Proof: Note that h is bounded on a, b. Let D
h
and D
g
denote the set of dsicontinuities
of h and g on a, b, respectively. Then
D
h
D
g
.
Since g R on a, b, then |D
g| = 0 by Theorem 7.48. Hence, |D
h
| = 0 which implies that
h R on a, b by Theorem 7.48.
Remark: (1) There has a more general theorem related with Riemann-Stieltjes
Integral. We write it as a reference.
(Theorem)Suppose g Ro on a, b, m _ gx _ M for all x a, b. If f is
continuous on m, M, the composite function h defined by hx = fgx Ro on
a, b.
Proof: It suffices to consider the case that o is increasing on a, b. If oa = ob,
there is nothing to prove it. So, we assume that oa < ob. In addition, let
K = sup
xa,b
|hx|. We claim that h satisfies Riemann condition with respect to o on
a, b. That is, given c > 0, we want to find a partition P such that
UP, h, o LP, h, o =
_
k=1
n
M
k
h m
k
hAo
k
< c.
Since f is uniformly continuous on m, M, for this c > 0, there is a
c
2K+1
> o > 0 such
that as |x y| < o where x, y m, M, we have
|fx fy| <
c
2ob oa
. 1
Since g Ro on a, b, for this o > 0, there exists a partition P such that
UP, g, o LP, g, o =
_
k=1
n
M
k
g m
k
gAo
k
< o
2
. 2
Let P = A B, where A = x
j
: M
k
g m
k
g < o and
B = x
j
: M
k
g m
k
g _ o, then
o
_
B
Ao
k
_
_
B
M
k
h m
k
hAo
k
_ o
2
by (2)
which implies that
_
B
Ao
k
_ o.
So, we have
UA, h, o LA, h, o _
_
A
M
k
h m
k
hAo
k
_ c/2 by (1)
and
UB, h, o LB, h, o _
_
B
M
k
h m
k
hAo
k
_ 2K
_
B
Ao
k
_ 2Ko
< c/2.
It implies that
UP, h, o LP, h, o = UA, h, o LA, h, o + UB, h, o LB, h, o
< c.
That is, we have proved that h satisfies the Riemann condition with respect to o on a, b.
So, h Ro on a, b.
Note: We mention that if we change the roles of f and g, then the conclusion does
NOT hold. Since the counterexample is constructed by some conclusions that we will learn
in Real Analysis, we do NOT give it a proof. Let C be the standard Cantor set in 0, 1 and
C
'
the Cantor set with positive measure in 0, 1. Use similar method on defining Cantor
Lebesgue Function, then there is a continuous function f : 0, 1 0, 1 such that
fC = C
'
. And Choose g = X
C
on 0, 1. Then
h = g f = X
C
'
which is NOT Riemann integrable on 0, 1.
(2) The reader should note the followings. Since these proofs use the exercise 7.30 and
Theorem 7.49, we omit it.
(i) If f R on a, b, then |f| R on a, b, and f
r
R on a, b, where r 0, .
(ii) If |f| R on a, b, it does NOT implies f R on a, b. And if f
2
R on a, b, it
does NOT implies f R on a, b. For example,
f =
1 if x Q , a, b
1 Q
c
, a, b
(iii) If f
3
R on a, b, then f R on a, b.
7.31 Use Lebesgues theorem to prove that if f R and g R on a, b and if
fx _ m > 0 for all x in a, b, then the function h defined by
hx = fx
gx
is Riemann-integrable on a, b.
Proof: Consider
hx = exphlogf,
then by Theorem 7.49,
f R logf R hlogf R exphlogf = h R.
7.32 Let I = 0, 1 and let A
1
= I
1
3
,
2
3
be the subset of I obtained by removing
those points which lie in the open middle third of I; that is, A
1
= 0,
1
3

2
3
, 1. Let A
2
be the subset of A
1
obtained by removing he open middle third of 0,
1
3
and of
2
3
, 1.
Continue this process and define A
3
, A
4
, . . . . The set C = ,
n=1

A
n
is called the Cantor set.
Prove that
(a) C is compact set having measure zero.
Proof: Write C = ,
n=1

A
n
. Note that every A
n
is closed, so C is closed. Since A
1
=
is closed and bounded, A
1
is compact and C A
1
, we know that C is compact by
Theorem 3.39.
In addition, it is clear that |A
n| =
2
3

n
for each n. Hence, |C| _ lim
n|A
n| = 0, which
implies that C has a measure zero.
(b) x C if, and only if, x = _
n=1

a
n
3
n
, where each a
n
is either 0 or 2.
Proof: ()Let x C = ,
n=1

A
n
, then x A
n
for all n. Consider the followings.
(i) Since x A
1
= 0,
1
3

2
3
, 1, then it implies that a
1
= 0 or 2.
(ii) Since x A
2
= 0,
1
9

2
9
,
3
9

6
9
,
7
9

8
9
,
9
9
, then it implies that a
2
= 0
or 2.
Inductively, we have a
n
= 0 or 2. So, x = _
n=1

a
n
3
n
, where each a
n
is either 0 or 2.
() If x = _
n=1

a
n
3
n
, where each a
n
is either 0 or 2, then it is clear that x A
n
for
each n. Hence, x C.
(c) C is uncountable.
Proof: Suppose that C is countable, write C = x
1
, x
2
, . . . . We consider unique ternary
expansion: if x = _
n=1

a
n
3
n
, then x := a
1
, . . . , a
n
, . . . . From this definition, by (b), we
have
x
k
= x
k1
, x
k2
, . . . x
kk,
. . . where each component is 0 or 2.
Choose y = y
1
, y
2
, . . . where
y
j
=
2 if x
jj
= 0
0 if x
jj
= 2.
By (b), y C. It implies that y = x
k
for some k which contradicts to the choice of y.
Hence, C is uncountable.
Remark: (1) In fact, C = C
'
means that C is a perfect set. Hence, C is uncountable.
The reader can see the book, Principles of Mathematical Analysis by Walter Rudin, pp
41-42.
(2) Let C = x : x = _
n=1

a
n
3
n
, where each a
n
is either 0 or 2 . Define a new
function : C 0, 1 by
x =
_
n=1

a
n
/2
2
n
,
then it is clear that is 1-1 and onto. So, C is equivalent to 0, 1. That is, C is
uncountable.
(d) Let fx = 1 if x C, fx = 0 if x C. Prove that f R on 0, 1.
Proof: In order to show that f R on 0, 1, it suffices to show that, by Theorem 7.48,
f is continuous on 0, 1 C since it implies that D C, where D is the set of
discontinuities of f on 0, 1.
Let x
0
0, 1 C, and note that C = C
'
, so there is a o > 0 such that
x
0
o, x
0
+ o , C = , where x
0
o, x
0
+ o 0, 1. Then given c > 0, there is a
o > 0 such that as x x
0
o, x
0
+ o, we have
|fx fx
0
| = 0 < c.
Remark: (1) C = C
'
:Given x C = ,
n=1

A
n
, and note that every endpoints of A
n
belong to C. So, x is an accumulation point of the set y : y is the endpoints of A
n
. So,
C C
'
. In addition, C
'
C since C is closed. Hence, C = C
'
.
(2) In fact, we have
f is continuous on 0, 1 C and f is not continuous on C.
Proof: In (d), we have proved that f is continuous on 0, 1 C, so it remains to show
that f is not continuous on C. Let x
0
C, if f is continuous at x
0
, then given c = 1/2, there
is a o > 0 such that as x x
0
o, x
0
+ o , 0, 1, we have
|fx fx
0
| < 1/2
which is absurb since we can choose y x
0
o, x
0
+ o , 0, 1 and y C by the fact C
does NOT contain an open interval since C has measure zero. So, we have proved that f is
not continuous on C.
Note: In a metric space M, a set S M is called nonwhere dense if intclS = .
Hence, we know that C is a nonwhere dense set.
Supplement on Cantor set.
From the exercise 7.32, we have learned what the Cantor set is. We write some
conclusions as a reference.
(1) The Cantor set C is compact and perfect.
(2) The Cantor set C is uncountable. In fact, #C = #R.
(3) The Cantor set C has measure zero.
(4) The Cantor set C is nonwhere dense.
(5) Every point x in C can be expressed as x = _
n=1

a
n
3
n
, where each a
n
is either 0 or
2.
(6) X
C
: 0, 1 0, 1 the characteristic function of C on 0, 1 is Riemann integrable.
The reader should be noted that Cantor set C in the exercise is 1 dimensional case. We
can use the same method to construct a n dimensional Cantor set in the set
x
1
, . . . , x
n
: 0 _ x
j
_ 1, j = 1, 2, . . n. In addition, there are many researches on Cantor
set. For example, we will learn so called Space-Filling Curve on the textbook, Ch9, pp
224-225.
In addition, there is an important function called Cantor-Lebesgue Function related
with Cantor set. The reader can see the book, Measure and Integral (An Introduction to
Real Analysis) written by Richard L. Wheeden and Antoni Zygmund, pp 35.
7.33 The exercise outlines a proof (due to Ivan Niven) that m
2
is irrational. Let
fx = x
n
1 x
n
/n!. Prove that:
(a) 0 < fx < 1/n! if 0 < x < 1.
Proof: It is clear.
(b) Each kth derivative f
k
0 and f
k
1 is an integer.
Proof: By Leibnitz Rule,
f
k
x =
1
n!
_
j=0
k

j
k
n n j + 1x
nj
1
kj
n n k + j + 11 x
nk+j
which implies that
f
k
0 =
0 if k < n
1 if k = n

n
k
1
kn
n 2n k + 1 if k > n
.
So, f
k
0 Z for each k N. Similarly, f
k
1 Z for each k Z.
Now assume that m
2
= a/b, where a and b are positive integers, and let
Fx = b
n
_
k=0
n
1
k
f
2k
xm
2n2k
.
Prove that:
(c) F0 and F1 are integers.
Proof: By (b), it is clear.
(d) m
2
a
n
fx sinmx =
d
dx
F
'
x sinmx mFx cos mx
Proof: Note that
F
''
x + m
2
Fx = b
n
_
k=0
n
1
k
f
2k+2
xm
2n2k
+ m
2
b
n
_
k=0
n
1
k
f
2k
xm
2n2k
= b
n
_
k=0
n1
1
k
f
2k+2
xm
2n2k
+ b
n
1
n
f
2n+2
x
+ m
2
b
n
_
k=1
n
1
k
f
2k
xm
2n2k
+ m
2
b
n
fxm
2n
= b
n
_
k=0
n1
1
k
f
2k+2
xm
2n2k
+ 1
k+1
f
2k+2
xm
2n2k
+ b
n
1
n
f
2n+2
x + m
2
b
n
fxm
2n
= m
2
a
n
fx since f is a polynomial of degree 2n.
So,
d
dx
F
'
x sinmx mFx cos mx
= sinmxF
''
x + m
2
Fx
= m
2
a
n
fx sinmx.
(e) F1 + F0 = ma
n
]
0
1
fx sinmxdx.
Proof: By (d), we have
m
2
a
n
]
0
1
fx sinmxdx = F
'
x sinmx mFx cos mx|
0
1
= F
'
1 sinm mF1 cos m F
'
0 sin0 mF0 cos 0
= mF1 + F0
which implies that
F1 + F0 = ma
n
]
0
1
fx sinmxdx.
(f) Use (a) in (e) to deduce that 0 < F1 + F0 < 1 if n is sufficiently large. This
contradicts (c) and show that m
2
(and hence m) is irrational.
Proof: By (a), and sinx 0, 1 on 0, m, we have
0 < ma
n
]
0
1
fx sinmxdx _
ma
n
n!
]
0
1
sinmxdx =
2a
n
n!
0 as n .
So, as n is sufficiently large, we have, by (d),
0 < F1 + F0 < 1
which contradicts (c). So, we have proved that m
2
(and hence m) is irrational.
Remark: The reader should know that m is a transcendental number. (Also, so is e). It
is well-known that a transcendental number must be an irrational number.
In 1900, David Hilbert asked 23 problems, the 7th problem is that, if o= 0, 1 is an
algebraic number and [ is an algebraic number but not rational, then is it true that o
[
is a
transcendental number. The problem is completely solved by Israil Moiseevic Gelfand in
1934. There are many open problem now on algebraic and transcendental numbers. For
example, It is an open problem: Is the Euler Constant
, = lim 1 +
1
2
+. . . +
1
n
logn
a transcendental number.
7.34 Given a real-valued function o, continuous on the interval a, b and having a
finite bounded derivative o
'
on a, b. Let f be defined and bounded on a, b and assume
that both integrals
]
a
b
fxdox and ]
a
b
fxo
'
xdx
exists. Prove that these integrals are equal. (It is not assumed that o
'
is continuous.)
Proof: Since both integrals exist, given c > 0, there exists a partition
P = x
0
= a, . . . , x
n
= b such that
SP, f, o ]
a
b
fxdox < c/2
where
SP, f, o =
_
j=1
n
ft
j
Ao
j
for t
j
x
j1
, x
j

=
_
j=1
n
ft
j
o
'
s
j
Ax
j
by Mean Value Theorem, where s
j
x
j1
, x
j
*
and
SP, fo
'
]
a
b
fxo
'
xdx < c/2
where
SP, fo
'
=
_
j=1
n
ft
j
o
'
t
j
Ax
j
for t
j
x
j1
, x
j
**
So, let t
j
= s
j
, then we have
SP, f, o = SP, fo
'
.
Hence,
]
a
b
fxdox ]
a
b
fxo
'
xdx
_ SP, f, o ]
a
b
fxdox + SP, fo
'
]
a
b
fxo
'
xdx
< c.
So, we have proved that both integrals are equal.
7.35 Prove the following theorem, which implies that a function with a positive
integral must itself be positive on some interval. Assume that f R on a, b and that
0 _ fx _ M on a, b, where M > 0. Let I = ]
a
b
fxdx, let h =
1
2
I/M + b a, and
assume that I > 0. Then the set T = x : fx _ h contains a finite number of intervals,
the sum of whose lengths is at least h.
Hint. Let P be a partition of a, b such that every Riemann sum SP, f = _
k=1
n
ft
k
Ax
k
satisfies SP, f > I/2. Split SP, f into two parts, SP, f = _
kA
+_
kB
, where
A = k : x
k1
, x
k
T, and B = k : k A.
If k A, use the inequality ft
k
_ M; if k B, choose t
k
so that ft
k
< h. Deduce that
_
kA
Ax
k
> h.
Proof: It is clear by Hint, so we omit the proof.
Remark: There is another proof about that a function with a positive integral must
itself be positive on some interval.
Proof: Suppose NOT, it means that in every subinterval, there is a point p such that
fp _ 0. So,
LP, f =
_
j=1
n
m
j
Ax
j
_ 0 since m
j
_ 0
for any partition P. Then it implies that
sup
P
LP, f = ]
a
b
fxdx _ 0
which contradicts to a function with a positive integral. Hence, we have proved that a
function with a positive integral must itself be positive on some interval.
Supplement on integration of vector-valued functions.
(Definition) Given f
1
, . . . , f
n
real valued functions defined on a, b, and let
f = f
1
, . . . , f
n
: a, b R
n
. If o on a, b. We say that f Ro on a, b means that
f
j
Ro on a, b for j = 1, 2, . . . , n. If this is the case, we define
]
a
b
fdo = ]
a
b
f
1
do, . . . , ]
a
b
f
n
do .
From the definition, the reader should find that the definition is NOT stranger for us. When
we talk f = f
1
, . . . , f
n
Ro on a, b, it suffices to consider each f
j
Ro on a, b for
j = 1, 2, . . . , n.
For example, if f Ro on a, b where o on a, b, then ]f] Ro on a, b.
Proof: Since f Ro on a, b, we know that f
j
Ro on a, b for j = 1, 2, . . . , n.
Hence,
_
k=1
n
f
j
2
Ro on a, b
which implies that, by Remark (1) in Exercise 7.30,
]f] =
_
k=1
n
f
j
2
Ro on a, b
1/2
Ro on a, b.
Remark: In the case above, we have
]
a
b
fdo _ ]
a
b
]f]do.
Proof: Consider
]y]
2
= < ]
a
b
f
1
do, . . . , ]
a
b
f
n
do, ]
a
b
f
1
do, . . . , ]
a
b
f
n
do >
=
_
j=1
n
]
a
b
f
j
do ]
a
b
f
j
do ,
which implies that, (let y
j
= ]
a
b
f
j
do, y y
1
, . . . , y
n
),
]y]
2
=
_
j=1
n
y
j ]
a
b
f
j
do
=
_
j=1
n
]
a
b
f
j
y
j
do
= ]
a
b
_
j=1
n
f
j
y
j
do
_ ]
a
b
]f]]y]do
= ]y] ]
a
b
]f]do
which implies that
]y] _ ]
a
b
]f]do.
Note: The equality holds if, and only if, ft = kty.
Existence theorems for integral and differential equations
The following exercises illustrate how the fixed-point theorem for contractions.
(Theorem 4.48) is used to prove existence theorems for solutions of certain integral and
differential equations. We denote by Ca, b the metric space of all real continuous
functions on a, b with the metric
df, g = ]f g] = max
xa,b
|fx gx|,
and recall the Ca, b is a complete metric space.
7.36 Given a function g in Ca, b, and a function K is continuous on the rectangle
Q = a, b a, b, consider the function T defined on Ca, b by the equation
Tx = gx + z ]
a
b
Kx, ttdt,
where z is a given constant.
(a) Prove that T maps Ca, b into itself.
Proof: Since K is continuous on the rectangle Q = a, b a, b, and x Ca, b,
we know that
]
a
b
Kx, ttdt Ca, b.
Hence, we prove that Tx Ca, b. That is, T maps Ca, b into itself.
(b) If |Kx, y| _ M on Q, where M > 0, and if |z| < M
1
b a
1
, prove that T is a
contraction of Ca, b and hence has a fixed point which is a solution of the integral
equation x = gx + z]
a
b
Kx, ttdt.
Proof: Consider
]T
1
x T
2
x] = z ]
a
b
Kx, t
1
t 2tdt
_ |z| ]
a
b
|Kx, t
1
t 2t|dt
_ |z|M]
a
b
|
1
t 2t|dt
_ |z|Mb a]
1
t 2t]. *
Since |z| < M
1
b a
1
, then there exists c such that |z| < c < M
1
b a
1
. Hence, by
(*), we know that
]T
1
x T
2
x] < ,]
1
t 2t]
where 0 < cMb a := , < 1. So, T is a contraction of Ca, b and hence has a fixed
point which is a solution of the integral equation x = gx + z]
a
b
Kx, ttdt.
7.37 Assume f is continuous on a rectangle Q = a h, a + h b k, b + k, where
h > 0, k > 0.
(a) Let be a function, continuous on a h, a + h, such that x, x Q for all x in
a h, a + h. If 0 < c _ h, prove that satisfies the differential equation y
'
= fx, y on
a c, a + c and the initial condition a = b if, and only if, satisfies the integral
equation
x = b + ]
a
x
ft, tdt on a c, a + c.
Proof: ()Since
'
t = ft, t on a c, a + c and a = b, we have,
x a c, a + c
x = a + ]
a
x

'
tdt
= a + ]
a
x
ft, tdt on a c, a + c.
()Assume
x = b + ]
a
x
ft, tdt on a c, a + c,
then

'
x = ft, x on a c, a + c.
(b) Assume that |fx, y| _ M on Q, where M > 0, and let c = minh, k/M. Let S
denote the metric subspace of Ca c, a + c consisting of all such that |x b| _ Mc
on a c, a + c. Prove that S is closed subspace of Ca c, a + c and hence that S is itself
a complete metric space.
Proof: Since Ca c, a + c is complete, if we can show that S is closed, then S is
complete. Hence, it remains to show that S is closed.
Given f S
'
, then there exists a sequence of functions f
n
such that f
n
f under the
sup norm ]. ]. So, given c > 0, there exists a positive integer N such that as n _ N, we
have
max
xac,a+c
|f
n
x fx| < c.
Consider
|fx b| _ |fx f
N
x| + |f
N
x b|
_ ]fx f
N
x] + ]f
N
x b]
< c + Mc
which implies that
|fx b| _ Mc for all x
since c is arbitrary. So, f S. It means that S is closed.
(c) Prove that the function T defined on S by the equation
Tx = b + ]
a
x
ft, tdt
maps S into itself.
Proof: Since
|Tx b| = ]
a
x
ft, tdt
_ ]
a
x
|ft, t|dt
_ x aM
_ Mc
we know that Tx S. That is, T maps S into itself.
(d) Now assume that f satisfies a Lipschitz condition of the form
|fx, y fx, z| _ A|y z|
for every pair of points x, y and x, z in Q, where A > 0. Prove that T is a contraction of
S if h < 1/A. Deduce that for h < 1/A the differential equation y
'
= fx, y has exactly one
solution y = x on a c, a + c such that a = b.
Proof: Note that h < 1/A, there exists z such that h < z < 1/A. Since
]T
1
x T
2
x]
_ ]
a
x
|ft,
1
t ft,
2
t|dt
_ A ]
a
x
|
1
t
2
t|dt by |fx, y fx, z| _ A|y z|
_ Ax a]
1
t
2
t]
_ Ah]
1
t
2
t]
< ,]
1
t
2
t]
where 0 < zA := , < 1. Hence, T is a contraction of S. It implies that there exists one and
only one S such that
x = b + ]
a
x
ft, tdt
which implies that

'
x = fx, x.
That is, the differential equation y
'
= fx, y has exactly one solution y = x on
a c, a + c such that a = b.
Supplement on Riemann Integrals
1. The reader should be noted that the metric space Ra, b, d is NOT complete,
where
df, g = ]
a
b
|fx gx|dx.
We do NOT give it a proof. The reader can see the book, Measure and Integral (An
Introduction to Real Analysis) by Richard L. Wheeden and Antoni Zygmund, Ch5.
2. The reader may recall the Mean Value Theorem: Let f be a continuous function on
a, b. Then
]
a
b
fxdx = fx
0
b a
where x
0
a, b. In fact, the point x
0
can be chosen to be interior of a, b. That is,
x
0
a, b.
Proof: Let M = sup
xa,b
fx, and m = inf
xa,b
fx. If M = m, then it is clear. So, we
may assume that M = m as follows. Suppose NOT, it means that x
0
= a or b. Note that,
fx
1
= m _ fx
0
:= r _ M = fx
2

by continuity of f on a, b. Then we claim that

fx
0
= m or M.
If NOT, i.e.,
fx
1
< r < fx
2

it means that there exists a point p x

1
, x
2
such that fp = r by Intermediate Value
Theorem. It contradicts to p = a or b. So, we have proved the claim. If fa = m, then
]
a
b
fxdx = mb a 0 = ]
a
b
fx mdx
which implies that, by fx m _ 0 on a, b,
fx = m forall x a, b.
So, it is impopssible. Similarly for other cases.
Remark: (1) The reader can give it a try to consider the Riemann-Stieltjes Integral as
follows. Let o be a continuous and increasing function on a, b. If f is continuous on
a, b, then
]
a
b
fxdox = fcob oa
where c a, b.
Note: We do NOT omit the continuity of o on a, b since
fx = x on 0, 1; ox =
0 if x = 0
1 if x 0, 1
.
(2) The reader can see the textbook, exercise 14.13 pp 404.
Exercise: Show that
m
2
< ]
0
m/2
dx
1
1
2
sin
2
x
<
m
2
.
Proof: It is clear by the choice of x
0
0, m/2.
3. Application on Integration by parts for Riemann-integrable function. It is
well-known that
] fxdx = xfx ] xdfx. *
If fx has the inverse function gy = x, then (*) implies that
] fxdx = xfx ] gydy.
For example,
] arcsinxdx = xarcsinx ] sinydy.
4. Here is an observation on Series, Differentiation and Integration. We write it as a
table to make the reader think it twice.
Series : Summation by parts Cesaro Sum ?
Differentiation :
fg
'
= f
'
g + fg
'
Mean Value Theorem Chain Rule
Integration : Integration by parts Mean Value Theorem Change of Variable .