You are on page 1of 22

Integral inequalities

Constantin P. Niculescu
+
Basic remark: If ) : [o, b] R is (Riemann) integrable
and nonnegative, then
_
b
o
)(t)ot 0.
Equality occurs if and only if ) = 0 almost everywhere
(a.e.)
When ) is continuous, ) = 0 a.e. if and only if ) = 0
everywhere.
Important Consequence: Monotony of integral,
) j implies
_
b
o
)(t)ot
_
b
o
j(t)ot.
+
Lecture presented on December 16, 2008, at the Abdus Salam
School of Mathematical Sciences, Lahore.
In Probability Theory, integrable functions are random
variables. Most important inequalities refer to:
A()) =
1
b o
_
b
o
)(t)ot (mean value of ))
\ ov()) = A
_
() A()))
2
_
(variance of ))
=
1
b o
_
b
o
)
2
(a)oa
_
1
b o
_
b
o
)(a) oa
_
2
.
Theorem 1 Chebyshevs inequality: If ), j : [o, b] R
have the same monotony, then
1
b o
_
b
o
)(t)j(t)ot
_
1
b o
_
b
o
)(t)ot
__
1
b o
_
b
o
j(t)ot
_
;
if ), j have opposite monotony, then the inequality should
be reversed.
Application: Let ) : [o, b] R be a dierentiable func-
tion having bounded derivative. Then
\ ov())
(b o)
2
12
sup
oab

)
t
(a)

2
.
Theorem 2 (The Mean Value Theorem). Let ) : [o, b]
R be a continuous function and j : [o, b] R be a non-
negative integrable function. Then there is c [o, b]
such that
_
b
o
)(a)j(a) oa = )(c)
_
b
o
j(a) oa.
Theorem 3 (Boundedness). If ) : [o, b] R is inte-
grable, then ) is bounded, [)[ is integrable and

1
b o
_
b
o
)(t)ot


1
b o
_
b
o
[)(t)[ ot
sup
otb
[)(t)[ .
Remark 4 If )
t
is integrable, then
0
1
b o
_
b
o
[)(a)[ oa

1
b o
_
b
o
)(a)oa

b o
3
sup
oab

)
t
(a)

.
Remark 5 Suppose that ) is continuously dierentiable
on [o, b] and )(o) = )(b) = 0. Then
sup
otb
[)(t)[
b o
2
_
b
o

)
t
(t)

ot.
Theorem 6 (Cauchy-Schwarz inequality).
If ), j : [o, b] R are integrable, then

_
b
o
)(t)j(t)ot


_
_
b
o
)
2
(t)ot
_
12
_
_
b
o
j
2
(t)ot
_
12
with equality i ) and j are proportional a.e.
Special Inequalities
Youngs inequality. Let ) : [0, o] [0, )(o)] be a
strictly increasing continuous function such that )(0) =
0. Using the denition of derivative show that
1(a) =
_
a
0
)(t) ot +
_
)(a)
0
)
1
(t) ot a)(a)
is dierentiable on [0, o] and 1
t
(a) = 0 for all a
[o, b]. Find from here that
aj
_
a
0
)(t) ot +
_
j
0
)
1
(t) ot.
for all 0 a o and 0 j )(o).
Figure 1: The geometric meaning of Youngs inequality.
Special case (corresponding for )(a) = a
j1
and )
1
(a) =
a
q1
) : For all o, b 0j, q (1, o) and 1j +
1q = 1, then
ob
o
j
j
+
b
q
q
if j, q (1, o) and
1
j
+
1
q
= 1;
ob
o
j
j
+
b
q
q
if j (o, 1)\0 and
1
j
+
1
q
= 1.
The equality holds if (and only if) o
j
= b
q
.
Theorems 7 and 8 below refer to arbitrary measure spaces
(A, , j).
Theorem 7 (The Rogers-Hlder inequality for j 1).
Let j, q (1, o) with 1j + 1q = 1, and let )
1
j
(j) and j 1
q
(j). Then )j is in 1
1
(j) and we
have

_
A
)j oj


_
A
[)j[ oj (1)
and
_
A
[)j[ oj |)|
1
j
|j|
1
q
. (2)
Thus

_
A
)j oj

|)|
1
j
|j|
1
q
. (3)
The above result extends in a straightforward manner
for the pairs j = 1, q = o and j = o, q = 1.
In the complementary domain, j (o, 1)\0 and
1j + 1q = 1, the inequality sign should be reversed.
For j = q = 2, we retrieve the Cauchy-Schwarz inequal-
ity.
Proof. If ) or j is zero j-almost everywhere, then the
second inequality is trivial. Otherwise, using the Young
inequality, we have
[)(a)[
|)|
1
j

[j(a)[
|j|
1
q

1
j

[)(a)[
j
|)|
j
1
j
+
1
q

[j(a)[
q
|j|
q
1
q
for all a in A, such that )j 1
1
(j). Thus
1
|)|
1
j
|j|
1
q
_
A
[)j[ oj 1
and this proves (2). The inequality (3) is immediate.
Remark 8 (Conditions for equality). The basic observa-
tion is the fact that
) 0 and
_
A
) oj = 0 imply ) = 0 j-almost everywhere.
Consequently we have equality in (1) if, and only if,
)(a)j(a) = c
i0
[)(a)j(a)[
for some real constant 0 and for j-almost every a.
Suppose that j, q (1, o). In order to get equality in
(2) it is necessary and sucient to have
[)(a)[
|)|
1
j

[j(a)[
|j|
1
q
=
1
j

[)(a)[
j
|)|
j
1
j
+
1
q

[j(a)[
q
|j|
q
1
q
almost everywhere. The equality case in Youngs inequal-
ity shows that this is equivalent to [)(a)[
j
|)|
j
1
j
=
[j(a)[
q
|j|
q
1
q
almost everywhere, that is,
[)(a)[
j
= 1[j(a)[
q
almost everywhere
for some nonnegative numbers and 1.
If j = 1 and q = o, we have equality in (2) if, and only
if, there is a constant A 0 such that [j(a)[ A almost
everywhere, and [j(a)[ = A for almost every point where
)(a) ,= 0.
Theorem 9 (Minkowskis inequality). For 1 j < o
and ), j 1
j
(j) we have
[[) +j[[
1
j
[[)[[
1
j
+[[j[[
1
j
. (4)
Proof. For j = 1, this follows immediately from[) +j[
[)[ +[j[. For j (1, o) we have
[) +j[
j
([)[ +[j[)
j
(2 sup [)[ , [j[)
j
2
j
([)[
j
+[j[
j
)
which shows that ) +j 1
j
(j).
According to the Rogers-Holder inequality,
[[) +j[[
j
1
j
=
_
A
[) +j[
j
oj

_
A
[) +j[
j1
[)[ oj +
_
A
[) +j[
j1
[j[ oj

__
A
[)[
j
oj
_
1j
__
A
[) +j[
(j1)q
oj
_
1q
+
+
__
A
[j[
j
oj
_
1j
__
A
[) +j[
(j1)q
oj
_
1q
= ([[)[[
1
j
+[[j[[
1
j
) [[) +j[[
jq
1
j
,
where 1j + 1q = 1, and it remains to observe that
j jq = 1.
Remark 10 If j = 1, we obtain equality in (4) if, and
only if, there is a positive measurable function , such
that
)(a),(a) = j(a)
almost everywhere on the set a : )(a)j(a) ,= 0 .
If j (1, o) and ) is not 0 almost everywhere, then
we have equality in (4) if, and only if, j = A) almost
everywhere, for some A 0.
Landaus inequality. Let ) : [0, o) R be a twice
dierentiable function. Put A
I
= sup
a0

)
(I)
(a)

for I = 0, 1, 2. If ) and )
tt
are bounded, then )
t
is
also bounded and
A
1
2
_
A
0
A
2
.
Proof. Notice that
)(a) = )(a
0
)+
_
a
a
0
_
)
t
(t) )
t
(a
0
)
_
ot+)
t
(a
0
)(aa
0
).
The case of functions on the entire real line.
Extension to the case of functions with Lipschitz deriva-
tive.
Inequalities involving convex functions
Hermite-Hadamard inequality: Let ) : [o, b] R be a
convex function. Then
)
_
o +b
2
_

1
b o
_
b
o
)(a) oa
)(o) +)(b)
2
(HH)
with equality only for ane functions.
The geometric meaning.
The case of arbitrary probability measures. See [2].
Jensens inequality: If , : [c, o] [o, b] is an integrable
function and ) : [o, b] R is a continuous convex func-
tion, then
)
_
1
o c
_
o
c
,(a) oa
_

1
o c
_
o
c
) (,(a)) oa.
(J)
The case of arbitrary probability measures.
An application of the Jensen inequality:
Hardys inequality: Suppose that ) 1
j
(0, o), ) 0,
where j (1, o). Put
1(a) =
1
a
_
a
0
)(t) ot, a 0.
Then
|1|
1
j

j
j 1
|)|
1
j
with equality if, and only if, ) = 0 almost everywhere.
The above inequality yields the norm of the averaging
operator ) 1, from 1
j
(0, o) into 1
j
(0, o).
The constant j(j 1) is best possible (though un-
tainted). The optimality can be easily checked by consid-
ering the sequence of functions )
a
(t) = t
1j

(0,a]
(t).
A more general result (also known as Hardys inequality):
If ) is a nonnegative locally integrable function on (0, o)
and j, v 1, then
_
o
0
a
jv
1
j
(a)oa
_
j
v 1
_
j
_
o
0
t
jv
)
j
(t) ot.
(5)
Moreover, if the right hand side is nite, so is the left
hand side.
This can be deduced (via rescaling) from the following
lemma (applied to & = a
j
, j 1, and I = )).
Lemma. Suppose that & : (0, o) R is convex and
increasing and I is a nonnegative locally integrable func-
tion. Then
_
o
0
&
_
1
a
_
a
0
I(t) ot
_
oa
a

_
o
0
&(I(a))
oa
a
.
Proof. In fact, by Jensens inequality,
_
o
0
&
_
1
a
_
a
0
I(t)ot
_
oa
a

_
o
0
_
1
a
_
a
0
&(I(t)) ot
_
oa
a
=
_
o
0
1
a
2
__
o
0
&(I(t))
[0,a]
(t) ot
_
oa
=
_
o
0
&(I(t))
__
o
t
1
a
2
oa
_
ot
=
_
o
0
&(I(t))
ot
t
.
Exercises
1. Prove the inequalities
1.43 <
_
1
0
c
a
2
oa <
1 +c
2
;
2c <
_
1
0
c
a
2
oa +
_
1
0
c
2a
2
oa < 1 +c
2
;
1 <
1
c
2
(c 1)
_
c
2
c
a
ln a
oa <
c
2
.
2. Let ) : [o, b] Rbe a dierentiable function having
bounded derivative. Prove that
\ ov())
(b o)
2
12
sup
oab

)
t
(a)

2
where \ ov()) represents the variance of ).
Hint: Put A = sup
oab

)
t
(a)

. Then apply the


Chebyshev inequality for the pair of functions )(a)+
Aa and )(a) Aa (having opposite monotony).
3. If )
t
is integrable, then
0
1
b o
_
b
o
[)(a)[ oa

1
b o
_
b
o
)(a)oa

b o
3
sup
oab

)
t
(a)

.
Hint: Consider the identity
(b o) )(a) =
_
b
o
)(t)ot +
_
a
o
(t o) )
t
(t)ot

_
b
a
(b t))
t
(t)ot.
4. Suppose that ) is continuously dierentiable on [0, 1].
Prove that
sup
0a1
[)(a)[
_
1
0
_
[)(t)[ +[)
t
(t)[
_
ot
and
[)(12)[
_
1
0
_
[)(t)[ +
1
2
[)
t
(t)[
_
ot.
3. Suppose that ) is continuously dierentiable on [o, b]
and )(o) = )(b) = 0. Then
sup
otb
[)(t)[
1
2
_
b
o

)
t
(t)

ot.
5. For t 1 a real number, consider the function
) : (1, o) R, )(a) = a
t
.
i) Use the Lagrange Mean Value Theorem to com-
pare )(7) )(6) with )(9) )(8);
ii) Prove the inequality 7
t
+ 8
t
< 6
t
+ 9
t
;
iii) Compute
_
2
1
7
t
ot.
iv) Conclude that
67
ln 7
+
78
ln 8
<
56
ln 6
+
89
ln 9
.
6. Consider the sequence (o
a
)
a
dened by the formula
o
a
=
_
1
0
oa
_
2 +
_
2 + +
_
2a
. .
a sqr
.
Prove that
1
2
o
a

1
_
2 +
_
2 + +
_
2
. .
a1 sqr
for all a 1
and nd the limit of the sequence (o
a
)
a
.
7. Infer from the Cauchy-Schwarz inequality that
ln(a + 1) ln a <
1
_
a(a + 1)
for a natural
and
_
2
0
sin
32
aoa <
_

3
.
8. Prove the inequalities:
_
1
0
2
a
2
oa 32;
__

0
c
sin a
oa
___

0
c
sin a
oa
_

2
.
9. Compute lim
ao
_
a+1
a
asin
1
a
oa and lim
ao
_
3a
2a
t
2
c
t
2
oa.
10. (The Bernoulli inequality). i) Prove that for all a
1 we have
(1 +a)
c
1 +ca if c (o, 0) ' (1, o)
and
(1 +a)
c
1 +ca if c [0, 1];
equality occurs only for a = 0.
ii) The substitution 1 + a aj followed by a
multiplication by j leads us to Youngs inequality
(for full range of parameters).
11. (The integral analogue of the AM-GM inequality).
Suppose that ) : [o, b] (0, o) is a continuous
function. Prove that
c
1
bo
_
b
o
ln )(a)oa

1
b o
_
b
o
)(a)oa.
12. (Ostrowskis inequality). Suppose that ) : [o, b]
R is a dierentiable function. Prove that

)(a)
1
b o
_
b
o
)(t) ot


_
_
_
1
4
+
_
a
o+b
2
_
2
(b o)
2
_
_
_(bo) sup
oab

)
t
(a)

.
13. (Z. Opial). Let ) : [0, o] R be a continuously
dierentiable function such that )(0) = 0. Prove
that
_
o
0
)(a)oa =
_
o
0
(o a))
t
(a)oa
and infer from this formula the inequalities:

_
o
0
)(a)oa


o
2
2
sup
0ao

)
t
(a)

_
o
0
[)(a)[

)
t
(a)

oa
o
2
_
o
0

)
t
(a)

2
oa.
References
[1] Constantin P. Niculescu, An Introduction to Mathe-
matical Analysis, Universitaria Press, Craiova, 2005.
[2] C. P. Niculescu and L.-E. Persson, Convex Functions
and their Applications. A Contemporary Approach.
CMS Books in Mathematics 23, Springer Verlag,
2006.

You might also like