Mathematical Database

Page 1 of 15
INEQUALITIES
UNIT 1 CLASSICAL INEQUALITIES

1. Inequality of the Means

To motivate our discussion, lets look at several situations.

(A) A man drove for 2 hours. In the first hour he travelled 16 km, and in the second hour he
travelled 32 km. What is his average speed for the whole journey?

(B) A man drove from City P to City Q at a speed of 16 km/h and returned at 32 km/h. What
is his average speed for the whole journey?

(C) In the traditional Chinese game mahjong, the value (in order not to encourage gambling
we use the term value in place of payoff) of a hand is determined by the number of
points of the hand and usually grows in a geometric sequence. For instance,
Points 0 1 2 3 4 5 6
Value 4 8 16 32 64 128 256
But in this case the value grows too fast. Some people want to modify it so that the value
doubles every 2 points rather than every point. For instance,
Points 0 1 2 3 4 5 6
Value 4 x 8 y 16 z 32
What should be x, y and z? Some people tend to simply put 6 x = , 12 y = and 24 z = , but
then the numbers in the second row will no longer form a geometric sequence. Can we
choose x, y, z so that 4, x, 8, y, 16, z, 32 form a geometric sequence?

These questions should be fairly easy. For (A), the average speed is
16 32
24 km/h
2
+
= .

Mathematical Database
Page 2 of 15
For (B), suppose the distance between P and Q is d km. Then the time taken for the journey
from P to Q and the return journey are
16
d
and
32
d
hours, respectively. Hence the average speed is
1 1
16 32 16 32
2 2 64
km/h
3
d d
d
= =
+ +
.

For (C), in order that the numbers form a geometric sequence, we must have
8
4
x
x
= , i.e.
4 8 4 2 x = = . Similarly, we get 8 16 8 2 y = = and 16 32 16 2 z = = .

The above three situations illustrate the three means that we are going to study. In (A), the
mean
16 32
24
2
+
= , the usual mean, is known as the arithmetic mean (A.M.). The mean in (B),
1 1
16 32
2 64
3
=
+
, is known as the harmonic mean (H.M.). The mean in (C), 16 32 16 2 = , is known
as the geometric mean (G.M.).

In general, given n positive numbers
1
a ,
2
a ,,
n
a we define the three means as follows:
1 2
1 2
1 1 1
1 2
A.M.
G.M.
H.M.
n
n
n
a a a
n
a a a
n
a a a
n
+ + +
=
=
=
+ + +

From the previous three examples, we see that for the particular case 2 n = ,
1
16 a = and
2
32 a = , A.M. G.M. H.M.. Indeed, for 2 n = this is not hard to prove:
( )
( ) ( ) ( )
1 2 1 2 1 2
1 1 2
1 2 1 2
1 2 1 2 1 2
1 2
2 2 2
1 2 1 2 1 2
1 2
2 2 2 2
2
2
4 4
2
a a
a a a a a a
a a a a
a a a a
a a
a a a a a a
a a
= =
+ +
+
=
+ +
=
+
=


In fact, this is true for the n-variable case in general. We have
Mathematical Database
Page 3 of 15
Theorem 1.1. (AM-GM-HM inequality)
Let
1
a ,
2
a , ,
n
a be positive numbers. Then A.M. G.M. H.M. for these numbers, i.e.
1 2
1 2
1 2
1 1 1
n
n
n
n
a a a
a a a n
a a a
n
+ + +

+ + +

.
Equality holds if and only if
1 2 n
a a a = = = .

Example 1.1.
Prove that for positive real numbers a, b, c,
( )
1 1 1
9 a b c
a b c
| |
+ + + +
|
\ .
.

Solution.
By the AM-HM inequality,
1 1 1
3
3
a b c
a b c + +

+ +
.
Hence
( )
1 1 1
9 a b c
a b c
| |
+ + + +
|
\ .
.

Example 1.2.
Let a and b be real numbers such that 0 a b > > . Determine the least possible value of
1
( )
a
b a b
+

.

Solution.
By the AM-GM inequality,
3
1 1 1
( ) 3 ( ) 3
( ) ( ) ( )
a a b b a b b
b a b b a b b a b
+ = + + =

.
Equality holds if
1
( )
a b b
b a b
= =

, i.e. 2 a = and 1 b = .
Hence the answer is 3.

Mathematical Database
Page 4 of 15
2. The Cauchy-Schwarz Inequality

Before formally presenting the inequality, lets consider an example.

Example 2.1.
Let a, b, c, d be real numbers. Prove that
2 2 2 2 2
( )( ) ( ) a c b d ab cd + + + .
Determine when equality holds.

Solution.
If any one of a, b, c, d is zero, the inequality is trivial.
In particular, if 0 a b = = or 0 c d = = , then equality holds.
Assume a, b, c, d are non-zero.
Then
2 2
a d and
2 2
b c are positive. By the AM-GM inequality, we have
2 2 2 2 2 2 2 2
2 ( )( ) 2 2 a d b c a d b c abcd abcd + = .
Adding
2 2 2 2
a b c d + on both sides, we have
2 2 2 2 2 2 2 2 2 2 2 2
2 a b c d a d b c a b c d abcd + + + + + ,
i.e.
2 2 2 2 2
( )( ) ( ) a c b d ab cd + + + .
Equality holds if and only if
2 2 2 2
a d b c = and abcd abcd = , i.e. ad bc = .
Combining all cases, equality holds if and only if ad bc = .

Basically, the above inequality says that the product of the sum of squares is greater than the
square of the sum of products. Indeed, this generalizes to 2n variables. We have

Theorem 2.1. (Cauchy-Schwarz inequality)
Let
1
a ,
2
a , ,
n
a ,
1
b ,
2
b , ,
n
b be real numbers. Then
2 2 2 2 2 2 2
1 2 1 2 1 1 1 2
( )( ) ( )
n n n n
a a a b b b a b a b a b + + + + + + + + + .
Equality holds if and only if
i j j i
a b a b = for all i, j, i.e.
Mathematical Database
Page 5 of 15
1 2
1 2
n
n
a a a
b b b
= = = or 0
2 1
= = = =
n
b b b .
Illustration. Take
1
2 a = ,
2
3 a = ,
3
5 a = ,
1
1 b = ,
2
0 b = ,
3
4 b = . Then
2 2 3 2 2 2
1 2 3 1 2 3
2
1 1 2 2 3 3
( )( ) 646
( ) 324
a a a b b b
a b a b a b
+ + + + =
+ + =


Remark 1. An easy way to memorise the Cauchy-Schwarz inequality is
( )
2
2 2
a b ab


which actually means
2
2 2
1 1 1
n n n
i i i i
i i i
a b a b
= = =
| |

|
\ .

.

Remark 2. By a slight modification, the Cauchy-Schwarz inequality holds for complex numbers as
well. In fact, if
1
a ,
2
a , ,
n
a ,
1
b ,
2
b , ,
n
b are complex numbers, then
( )( )
2 2 2 2 2 2 2
1 2 1 2 1 1 1 2 n n n n
a a a b b b a b a b a b + + + + + + + + + .
Equality holds if and only if
i j j i
a b a b = for all i, j.

Example 2.2.
Prove that for positive real numbers a, b, c,
( )
1 1 1
9 a b c
a b c
| |
+ + + +
|
\ .
.

Solutions.
Note that this is the same inequality as in Example 1.1.
Here we apply the Cauchy-Schwarz inequality. We have
( ) ( ) ( )
2 2 2
2 2 2
2
1 1 1 1 1 1
( )
1 1 1
9
a b c a b c
a b c a b c
a b c
a b c
(
| | | | | | | |
(
+ + + + = + + + + (
| | | |
(

\ . \ . \ . \ . (

| |
+ +
|
\ .
=

Mathematical Database
Page 6 of 15

Remark. Equality holds if and only of
1 1 1
a b c
a b c
= = , i.e. a b c = = .
Example 2.3.
Let x and y be real numbers such that
2 2
1 1 1 x y y x + = .
Prove that
2 2
1 x y + = .

Solution.
By the Cauchy-Schwarz inequality (with
1
a x = ,
2
2
1 a x = ,
2
1
1 b y = ,
2
b y = ), we have
( ) ( )
2 2
2 2 2 2 2 2
1 1 1 1 1 x y y x x x y y
( (
+ + + =
( (

.
Equality holds if and only if
2
2
1
1
x x
y
y

=

,
i.e.
2 2 2 2
(1 )(1 ) x y x y = , or
2 2
1 x y + = .
Now
2 2
1 1 1 x y y x + = , so we must have
2 2
1 x y + = .

The Cauchy-Schwarz inequality is unique in its special condition for equality case. Most other
inequalities have equality case when the variables are equal. Indeed, observing the equality case is
one important technique in proving inequalities, as we shall see in Unit 2.


3. The rearrangement inequality

A class of 40 students is split into four groups, A, B, C and D, with 7, 8, 12 and 13 students
respectively. Each group is to choose a type of shirts to represent their group. These are four types
of shirts, and their unit prices are $10, $15, $20 and $25 respectively. Of course, each student will
buy a shirt representing his group and each group will choose a different type of shirts. What
matching between groups and shirts will yield a minimum total expenses of the class on shirts?
What matching will yield a maximum?

Mathematical Database
Page 7 of 15
It is quite intuitive that the total expenses will be minimum when Group D, with the largest
number of students, chooses the least expensive type of shirts, Group C chooses the second least
expensive type of shirts, and so on. By contrast, maximality occurs when Group D chooses the most
expensive type of shirts, Group C the second most expensive type of shirts and so on.
The above situation illustrates the following:

Theorem 3.1. (Rearrangement Inequality)
If
i
a and
i
b (i = 1, 2, , n) are real numbers such that
1 2 n
a a a and
1 2 n
b b b ,
then
1 1 2 2 1 1 2 2 1 2 1 1 n n j j n jn n n n
a b a b a b a b a b a b a b a b a b

+ + + + + + + + + ,
where
1 2
{ , , , }
n
j j j is a permutation of {1, 2, , } n . In other words,
Direct Sum Random Sum Reverse Sum.
Equality holds if and only if
1 2 n
a a a = = = or
1 2 n
b b b = = = .

Illustration. In the above example, , 13
1
= a , 12
2
= a , 8
3
= a , 7
4
= a , 25
1
= b , 20
2
= b , 15
3
= b
. 10
4
= b We have
Direct sum 13(25) 12(20) 8(15) 7(10) 755
A random sum 13(20) 12(15) 8(25) 7(10) 710
Reverse sum 13(10) 12(15) 8(20) 7(25) 645
= + + + =
= + + + =
= + + + =


Example 3.1.
Let
1
a ,
2
a , ,
n
a be positive real numbers and
1 2
{ , , , }
n
b b b be a permutation of
1 2
{ , , , }
n
a a a .
Prove that
1 2
1 2
n
n
a a a
n
b b b
+ + + .

Solution.
Without loss of generality, assume
1 2 n
a a a . Then
1 1
1 1 1
n n
a a a

.
By the fact that Random Sum Reverse Sum, we have
Mathematical Database
Page 8 of 15
1 2
1 2
1 2 1 2
1 2
1 2
1 1 1
1 1 1
n
n
n n
n
n
a a a
a a a
b b b b b b
a a a
a a a
n
| | | | | |
+ + + = + + +
| | |
\ . \ . \ .
| | | | | |
+ + +
| | |
\ . \ . \ .
=


as desired.


4. Other Inequalities

In this section we will look at generalizations and variations to the three inequalities we have
learned.

We first look at two generalizations of the AM-GM inequality. For any real number s and
positive real numbers
1
a ,
2
a , ,
n
a , we define
1
1 2
s s s
s
n
s
a a a
M
n
| | + + +
=
|
\ .

to be the s-power mean of
1
a ,
2
a , ,
n
a . It is clear from this definition that
1
M is the familiar AM
and
1
M

is the HM. A little knowledge in calculus would enable us to define

= which is the GM. Hence Theorem 1.1 says

. The following
theorem generalizes this to any
s
M .

Theorem 4.1. (Power Mean Inequality)
Let
1
a ,
2
a , ,
n
a be positive real numbers and s, t be real numbers such that s t > . Then
s t
M M .
Equality holds if and only if
1 2 n
a a a = = = .

Remark. The number
2 2 2
1 2
2
n
a a a
M
n
+ + +
=


Mathematical Database
Page 9 of 15
is usually known as the root mean square (RMS) or quadratic mean (QM) of
1
a ,
2
a , ,
n
a .
Also, it is customary to define
} {
} {
1 2
1 2
lim max , ,...,
lim min , ,...,
s n
s
s n
s
M M a a a
M M a a a
+
+

= =
= =

This implies that all means lie between the maximum and minimum of the group of numbers.

Example 4.1.
Given a set of data
1
x ,
2
x , ,
n
x with mean (the usual arithmetic mean) x , we define the mean
deviation d and the standard deviation as
( )
1
2
1
1
1
n
i
i
n
i
i
d x x
n
x x
n

=
=
=
=

Show that d .

Solution.
Consider the n non-negative numbers
1
x x ,
2
x x , ,
n
x x .
If all of these are zeros the result is trivial.
So we assume that exactly k of these numbers are non-zero, and without loss of generality we may
assume that the first k of them are non-zero.
Applying the RMS-AM inequality (
1 2
M M ) on these k numbers, we have
2
1 1
1 1
n n
i i
i i
x x x x
k k
= =


.
Note that 1
k
n
, so
k k
n n
. Multiplying this to the above inequality, we get
( )
2
1 1
1 1
n n
i i
i i
x x x x
n n
= =


,
i.e.
d .

Mathematical Database
Page 10 of 15
Again let
1
a ,
2
a , ,
n
a be positive real numbers. We define
k
P to be the average of all
products of k of the
i
a s, i.e.
1 2
1
1 2 1 3 1 2 3 1
2 1
2
1 2
( 1)
n
n n n
n n
a a a
P
n
a a a a a a a a a a
P
n n
P a a a

+ + +
=
+ + + + + +
=

Hence the AM is simply
1
P and the GM is
1/ n
n
P . Theorem 1.1 says
1/
1
n
n
P P . The following
theorem generalizes this.

Theorem 4.2. (Maclaurins Symmetric Mean Inequality)
For positive real numbers
1
a ,
2
a , ,
n
a ,
1/ 2 1/
1 2
n
n
P P P .
Equality holds if and only if
1 2 n
a a a = = = .

Example 4.2.
Prove that for positive real numbers a, b, c,
8 8 8
3 3 3
1 1 1 a b c
a b c a b c
+ +
+ +

Solution.
Since
8 1
M M , we have
1
8 8 8
8
3 3
a b c a b c | | + + + +

|
\ .
.
By the symmetric mean inequality,
( ) ( )
6 2
8 6 2 1/ 3 1/ 2
1 1 1 3 2
P P P P P = ,
i.e.
8
2
( ) ( )
3 3
a b c abc ab bc ca + + + + | |

|
\ .
.
Combining these results, we have
Mathematical Database
Page 11 of 15
8
8 8 8 2
3 3 3 3 3
1 ( ) ( ) 1 1 1
3
3 ( ) ( )
a b c a b c abc ab bc ca
a b c abc abc a b c
+ + + + + + | |
= + +
|
\ .
.
Mathematical Database
Page 12 of 15
Now we proceed to look at a generalization of the Cauchy-Schwarz inequality.

Theorem 4.3. (Hlders Inequality)
Let
1
a ,
2
a , ,
n
a ,
1
b ,
2
b , ,
n
b be complex numbers, p, q be real numbers greater than 1 such that
1 1
1
p q
+ = . Then
( ) ( )
1 1
1 1 1 2 1 2 1 2
p p p q q q
p q
n n n n
a b a b a b a a a b b b + + + + + + + .
Equality holds if and only if
1 2
1 2
n
n
a a a
b b b
= = = .

Remark. The Cauchy-Schwarz inequality is a special case of Hlders inequality with 2 p q = = .

Example 4.3.
Prove that for positive real numbers a, b, c,
3 3 3 2 2 2
a b c a b b c c a + + + + .

Solution.
Setting
3
2
p = , 3 q = in Hlders inequality, we have
( ) ( ) ( ) ( )
2
3 3 3 1
3
2 2 2 2 2 2 3 3 3
2 2 2 3
3 3 3
a b b c c a a b c a b c
a b c
(
+ + + + + +
(

= + +


Alternative Solution.
Without loss of generality, assume a b c . Then
2 2 2
a b c .
Since Direct Sum Random Sum, we have
2 2 2 2 2 2
( ) ( ) ( ) ( ) ( ) ( ) a a b b c c a b b c c a + + + + ,
i.e.
3 3 3 2 2 2
a b c a b b c c a + + + + .
Mathematical Database
Page 13 of 15
Finally, we will look at a variation of the rearrangement inequality. The rearrangement
inequality says
Direct Sum Random Sum Reverse Sum.
Now instead of the random sum, we put another expression in the middle. This is given by the
following theorem.

Theorem 4.4. (Chebyschevs Inequality)
Let
1
a ,
2
a , ,
n
a ,
1
b ,
2
b , ,
n
b be real numbers,
1 2 n
a a a ,
1 2 n
b b b . Then
1 2 1 2
1 1 2 2 1 2 1 1
( )( )
n n
n n n n n
a a a b b b
a b a b a b a b a b a b
n

+ + + + + +
+ + + + + +

.
Equality holds if and only if
1 2 n
a a a = = = or
1 2 n
b b b = = = .

Example 4.4.
Let
1
a ,
2
a , ,
n
a be positive real number less than 1,
1 2 n
a a a S + + + = . Prove that
1 2
1 2
1 1 1
n
n
a a a nS
a a a n S
+ + +

.

Solution.
Without loss of generality assume
1 2 n
a a a .
Then
1 1
1 1 1
n n
a a a

and so
1 2
1 2
1 1 1
n
n
a a a
a a a


.
By Chebyschevs inequality,
| |
1 2
1 2
1 2
1 2
1 2
1 2
1 2
1 2
1 2
(1 ) (1 ) (1 )
1 1 1
1
(1 ) (1 ) (1 )
1 1 1
1
( )
1 1 1
n
n
n
n
n
n
n
n
n
S a a a
a a a
a a a
a a a
a a a
a a a
n a a a
a a a
n S
n a a a
= + + +
= + + +

(
+ + + + + +
(


(
= + + +
(

Hence
Mathematical Database
Page 14 of 15
1 2
1 2
1 1 1
n
n
a a a nS
a a a n S
+ + +

.
Mathematical Database
Page 15 of 15
5. Exercises

In the following, all variables denote positive real numbers.

1. Prove that if
1 2
(1 )(1 ) (1 ) 2
n
n
a a a + + + = ,
then
1 2
1
n
a a a .

2. Prove that
2 2 2
1 2
1 2
2 3 1
n
n
a a a
a a a
a a a
+ + + + + + .

3. Prove that if a, b, c are less than 1 and 2 a b c + + = , then
8
(1 )(1 )(1 )
abc
a b c


.

4. Prove that
3
2
a b c
b c c a a b
+ +
+ + +
.

5. (IMO 1978) Let
1
a ,
2
a , ,
n
a be distinct positive integers. Prove that
2
1
2 2
1 1
1
2 2
n
a a
a
n n
+ + + + + + .