Вы находитесь на странице: 1из 523

B. J.

Venkatachala

Inequalities
An Approach Through Problems

Second Edition

123
B. J. Venkatachala
(emeritus) Homi Bhabha Centre
for Science Education
Mumbai, Maharashtra
India

ISSN 2366-8725 (electronic)


Texts and Readings in Mathematics
ISBN 978-981-10-8732-5 (eBook)
https://doi.org/10.1007/978-981-10-8732-5
Library of Congress Control Number: 2018941811

1st edition: © Hindustan Book Agency (lndia) 2009


Preface(First edition)
The International Mathematical Olympiad(IMO), which started as a simple
contest among seven communist block countries in Europe in 1959, has now
encompassed the whole world. With nearly 100 countries participating in this
mega event, this has acquired a true international character. Mathematical
olympiad has effused new enthusiasm in the last few generations of young stu-
dents and really talented young minds have started getting attracted to the rare
beauty of mathematics. Along with it, new ideas have emerged and many intri-
cate problems woven around these ideas have naturally been discovered. This
has enriched basic mathematics, strengthened the foundations of elementary
mathematics and has posed challenging problems to the younger generation.
In turn, high school mathematics has undergone a profound change.
Even though the concept of inequalities is old, the mathematical olympiad
movement has generated new problems based on these inequalities and newer
applications of these old inequalities. There are classical inequalities like the
Arithmetic mean-Geometric mean inequality and the Cauchy-Schwarz inequal-
ity which are very old and which have innumerable applications. The main
purpose of this book is to give a comprehensive presentation of inequalities
and their use in mathematical olympiad problems. This book is also intended
for those students who would like to participate in mathematical olympiad.
Hopefully, this will fill a little vacuum that exists in the world of books. I
have not touched on integral inequalities; they are not a part of olympiad
mathematics.
The book is divided into six chapters. The first chapter describes all the
classical inequalities which are useful for students who are interested in mathe-
matical olympiad and similar contests. I have tried to avoid too much of theory;
many results are taken for granted whenever a need for some advanced math-
ematics is required(especially results from calculus). Rather, I have put more
emphasis on problems. The second chapter gives many useful techniques for
deriving more inequalities. Here again the stress is on the application of differ-
ent methods to problems. An important class of inequalities, called geometric
inequalities, is based on geometric structures, like, triangles and quadrilaterals.
Some important geometric inequalities are derived in the third chapter. In
the fourth chapter, the problems(mainly taken from olympiad contests) whose
solution(s) involve application of inequalities are discussed. The fifth chapter
is simply a large collection of problems from various contests around the world
and some problems are also taken from several problem journals. The reader
is advised to try these problems on his own before looking into their possible
solutions, which are discussed in the sixth chapter.
As I said earlier, the problems have been taken from various sources. I have
tried to give reference to them where ever possible and whenever I had one.
I deeply regret and apologise for any inadvertent omission in mentioning the
source.
This whole exercise of writing a book on inequalities is the outcome of my
discussion with the bright students who have attended training camps and
with my colleagues from the training camp. I am really grateful to all of
them. Special thanks go to my colleagues, Prof. C.R.Pranesachar, from the
Mathematical Olympiad cell and Prof. R.B.Bapat from the Indian Statistical
Institute, New Delhi, for their encouragement. I also wish to acknowledge the
support given by Prof. Arvindkumar, Director of the Homi Bhabha Centre
for Science Education(TIFR), Mumbai. I would like to thank the referees for
giving their valuable comments and invaluable tips to improve the quality of
the material in this book.
Finally, the moral support I got from my family throughout this work is
something I never forget.
My thanks are also due to the Homi Bhabha Centre for Science Educa-
tion(TIFR), Mumbai/the Tata Institute of Fundamental Research, Mumbai,
my parent institutions.
I also acknowledge the support of: the National Board for Higher Math-
ematics, Department of Atomic Energy; Department of Mathematics, Indian
Institute of Science, Bangalore.
B J Venkatachala,
MO Cell, HBCSE(TIFR),
Department of Mathematics,
Indian Institute of Science,
Bangalore-560012, INDIA.
Preface(Second edition)

Eight years have elapsed after the first edition of this book has come out.
More inequalities have been generated as a need for various national math-
ematical olympiads and International Mathematical Olympiads. New books
have emerged and newer results have been added during these years. This has
prompted me to revise the book adding more material to the existing one. New
problems based on the old methods, new solutions to the old problems and new
methods have been added to the book to make it richer.
A section on proving symmetric inequalities has been added to chapter
2. More problems have been solved based on these methods. More than 70
problems have been added to chapter 5 and their solutions have been given
in chapter 6. It is my sincere wish that the new edition will help all those
who aspire to study inequalities for mathematics competitions and for general
interest.
I thank all my colleagues and students who have enriched me with their
lively conversations and made this revision possible.
B J Venkatachala,
Professor (Retired),
HBCSE, TIFR,
MUMBAI-400088, INDIA.
Contents

1 Some basic inequalities ...................................... 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Arithmetic mean-Geometric mean inequality . . . . . . . . . . . 2

1.3 Cauchy-Schwarz inequality . . . . . . . . . . . . . . . . . . . . 11

1.4 Chebyshev’s inequality . . . . . . . . . . . . . . . . . . . . . . 17

1.5 Rearrangement inequality . . . . . . . . . . . . . . . . . . . . . 21

1.6 Hölder’s and Minkowski’s inequalities . . . . . . . . . . . . . . 27

1.7 Convex and Concave functions, Jensen’s inequality . . . . . . . 34

1.8 Inequalities for symmetric functions . . . . . . . . . . . . . . . 47

2 Techniques for proving inequalities . . . . . . . . . . . . . . . . . . . . . . 51


2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

2.2 Use of induction . . . . . . . . . . . . . . . . . . . . . . . . . . 51

2.3 Application of known inequalities . . . . . . . . . . . . . . . . . 54

2.4 Use of calculus in inequalities . . . . . . . . . . . . . . . . . . . 65

2.5 Trigonometric substitutions . . . . . . . . . . . . . . . . . . . . 73

2.6 Properties of quadratic polynomials . . . . . . . . . . . . . . . 75

2.7 A useful transformation . . . . . . . . . . . . . . . . . . . . . . 78

2.8 Schur’s inequality . . . . . . . . . . . . . . . . . . . . . . . . . 80

2.9 Majorisation technique . . . . . . . . . . . . . . . . . . . . . . 84

2.10 Muirhead’s theorem . . . . . . . . . . . . . . . . . . . . . . . . 86

2.11 Homogenisation . . . . . . . . . . . . . . . . . . . . . . . . . . 88

2.12 Normalisation . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
2.13 Stolarsky’s theorem . . . . . . . . . . . . . . . . . . . . . . . . 93

2.14 Methods for symmetric inequalities . . . . . . . . . . . . . . . 95

3 Geometric inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

3.2 Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

3.3 Some identities involving elements of a triangle . . . . . . . . . 113

3.4 Some geometric inequalities . . . . . . . . . . . . . . . . . . . 115

3.5 Two triangles one inscribed in the other . . . . . . . . . . . . . 141

3.6 Let P be a point . . . . . . . . . . . . . . . . . . . . . . . . . . 147

4 Applications involving inequalities . . . . . . . . . . . . . . . . . . . . . . 173

5 Problems on inequalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205

6 Solutions to problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 245


7 Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515

8 Abbreviation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 516

9 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 517
Chapter 1
Some basic inequalities

1.1 Introduction
As we all know, one of the important properties of real numbers is comparability.
We can compare two distinct real numbers and say that one is smaller or larger
than the other. There is an inherent ordering < on the real number system
R which helps us to compare two real numbers. The basic properties of this
ordering on R are:
(i) Given any two real numbers a and b, one and only one of the following
three relations is true:

a < b or a = b or a > b;

(law of trichotomy.)
(ii) a > 0 and b > 0 imply a + b > 0;
(iii) a > 0 and b > 0 imply ab > 0.
Any new inequality we derive is totally dependent on these basic properties.
These properties are used to derive the arithmetic mean-geometric inequality,
the Cauchy-Schwarz inequality, Chebyshev’s inequality, the rearrangement in-
equality, Hölder’s and Minkowski’s inequalities, and are also used in the study
of convex and concave functions. Here are some easy consequences of these
fundamental properties of ordering on R:
(1) a < b, then a + c < b + c, for any real c;
(2) a < b and c > 0 give ac < bc; if c < 0, we have ac > bc;
(3) 0 < a < b implies 0 < 1/b < 1/a;
(4) a < 0 and b < 0, then ab > 0; a < 0 and b > 0 imply ab < 0;
(5) a < b and b < c together imply a < c (transitivity);
(6) if ac < bc and c > 0, we have a < b;
(7) 0 < a < 1 implies a2 < a; if a > 1, we have a2 > a;
(8) for any real a, a2 ≥ 0;
(9) if a and b are positive and a2 < b2 , we have a < b.
2 Inequalities

We emphasise here that the subtraction of inequalities is generally not al-


lowed. If a > b and c > d, we cannot guarantee either a − c > b − d or
c − a > d − b. The reason is obvious: x > y implies −x < −y. Similarly,
we cannot divide an inequality by another one. If a > b and c > d, none of
a b c d
a, b, c, d equal 0, neither > is true nor is > . Again the reason is
c d a b
1 1
simple: x > y gives < (x, y not equal to 0). On the other hand, we may
x y
add any two inequalities. If all the terms in two inequalities are positive, we
may also multiply them : if a, b, c, d are positive and a > b, c > d holds, the
inequality ac > bd also holds.
For any real number x, we define its absolute value by

x if x ≥ 0
|x| =
−x if x < 0.

Note that |x| ≥ 0 and |x| = 0 if and only if x = 0. The absolute value function
x → |x| has some nice properties:

(1) | − x| = |x|; |x|2 = x2 ; |xy| = |x| |y|; |x/y| = |x|/|y| for y = 0.

(2) |x + y| ≤ |x| + |y| and equality holds if and only if x and y have the same
sign. This is known as the triangle inequality.
 
 
(3) |x| − |y| ≤ |x − y|.

On the other hand, there is no useful ordering on the complex number


system C; there is no natural ordering on C as enjoyed by R. (This is the
inherent universal principle that you have to pay some thing if you want to
get some thing.) However, we can come down to the real number system using
the absolute value of a complex number. For any complex number z = a + ib,
√ associate its complex2 conjugate by z = a − ib and absolute value by |z| =
we
a2 + b2 . Note that |z| = z · z. It is easy to check the following properties of
|z|:

(1) |z| ≥ 0; and |z| = 0 if and only if z = 0;

(2) |z1 z2 | = |z1 | |z2 |;

(3) |z1 + z2 | ≤ |z1 | + |z2 | and equality holds if and only if z1 = λz2 for some
positive real number λ, or one of z1 , z2 is zero.

1.2 Arithmetic mean-Geometric mean inequality


The basic inequality in the real number system is x2 ≥ 0 for any real number
x. This is so fundamental that any other inequality for real numbers is a
consequence of this. Consider any two non-negative real numbers a and b.
Basic Inequalities 3

√ √
Then we know that a and b are meaningful as real numbers. Since the
√ √ 2
square of a real number is always non-negative, a − b ≥ 0. This may be
rewritten in the form
a+b √
≥ ab. (1.1)
2
a+b √
The real number is called the arithmetic mean of a and b; similarly ab
2
is known as the geometric mean of a and b. Thus the property that x2 ≥ 0
for a real number x implies that the arithmetic mean of two non-negative
real numbers cannot be smaller than their geometric mean. Moreover, the
derivation also shows that equality in (1.1) holds if and only if a = b.
This may be considered in a general setting. Starting with n non-negative
real numbers a1 , a2 , . . . , an , we define their arithmetic mean A(a1 , a2 , . . . , an )
and geometric mean G(a1 , a2 , . . . , an ) by

  a1 + a2 + · · · + an
A a 1 , a2 , . . . , an = ,
n
   1/n
G a 1 , a2 , . . . , an = a1 a2 · · · an .

As in the case of two numbers, a comparison between these two means leads
to the AM-GM inequality.

Theorem 1. Given any n non-negative real numbers a1 , a2 , . . . , an , they satisfy


the inequality
a1 + a2 + · · · + an  1/n
≥ a1 a2 · · · an . (1.2)
n
Equality holds in the inequality if and only if a1 = a2 = · · · = an .

Proof: There are several proofs of this classical theorem. We give here a clever
induction proof which was originated by Cauchy. It also illustrates how one
can go ahead in an induction proof leaving out the validity of induction for
some numbers and subsequently prove the validity for missing numbers using
an interpolation argument.
We have already obtained the result for n = 2;

a1 + a2  1/2
≥ a1 a2 . (1.3)
2

Here equality holds if and only if a1 = a2 . Consider 4 non-negative real numbers


a1 , a2 , a3 , a4 . Divide them in to two groups; {a1 , a2 } and {a3 , a4 }. Applying
the known inequality for each group, we obtain

a1 + a2  1/2 a3 + a4  1/2
≥ a1 a2 , ≥ a3 a4 .
2 2
4 Inequalities

This leads to
   
a1 + a2 a3 + a4
+
a1 + a2 + a3 + a4 2 2
=
4 2
 1/2 1/2

a1 a2 + a3 a4

2

 1/2  1/2 1/2
≥ a1 a2 a3 a4
 1/4
= a1 a2 a3 a4 .

Thus the inequality (1.2) is obtained for n = 4. It may again be observed that
equality holds here if and only if a1 = a2 , a3 = a4 and a1 a2 = a3 a4 . Since all
numbers are non-negative, it follows that a1 = a2 = a3 = a4 . Now this can be
used to prove (1.2) for n = 8. Now induction shows that (1.2) holds for n = 2k
for all k ∈ N. Here again the condition for equality is that all the 2k numbers
be equal.
Now consider any n non-negative real numbers a1 , a2 , . . . , an . We choose a
natural number k such that 2k−1 ≤ n < 2k . Put
a1 + a2 + · · · + an
A= ,
n

and consider the set {a1 , a2 , . . . , an , A, A, . . . A} of 2k numbers; here A appears


2k −n times. We apply the AM-GM inequality for these 2k numbers; its validity
has already been ascertained for such a collection. We thus obtain
 1/2k
a1 + a2 + · · · + an + A + A + · · · + A ≥ 2 k
a1 a2 · · · an AA · · · A

.


2k −n 2k −n

This reduces to
 1/2k
2k −n
A≥ a1 a2 · · · an A .

A simple manipulation now yields An ≥ a1 a2 · · · an which is equivalent to the


inequality (1.2).
We also observe that equality holds if and only if all the numbers are equal.

Example 1.1. Let a1 , a2 , . . . , an be n positive real numbers whose product is 1.


Prove that     
1 + a 1 1 + a 2 · · · 1 + a n ≥ 2n .
Basic Inequalities 5
  √
Solution: Using the inequality 1 + aj ≥ 2 aj for 1 ≤ j ≤ n, we obtain
      1/2
1 + a1 1 + a2 · · · 1 + an ≥ 2n a1 a2 · · · an = 2n .

Example 1.2. Show that for any natural number n > 1, the inequality
 n
(2n)! < n(n + 1) ,
holds good.

Solution: We split (2n)! as two products:


  
(2n)! = 1 · 3 · 5 · · · (2n − 1) 2 · 4 · 6 · · · (2n) .
Using the AM-GM inequality, we get
 
2 · 4 · 6 · · · (2n) = 2n 1 · 2 · 3 · · · n
 n
1 + 2 + 3 + ··· + n
< 2n = (n + 1)n ,
n
and
 n
1 + 3 + 5 + · · · + (2n − 1)
1 · 3 · 5 · · · (2n − 1) <
n
n
= n .
Combining these two inequalities, one can obtain
 n
(2n)! < (n + 1)n nn = n(n + 1) ,
as desired.

Example 1.3. If a, b, c are the sides of a triangle, prove that


3 a b c
≤ + + < 2. (1.4)
2 b+c c+a a+b

Solution: The first part of the above inequality is equivalent to


9 a b c
≤ +1+ +1+ +1
2 b+c c+a a+b
  
1 1 1
= a+b+c + + .
b+c c+a a+b
If we introduce a + b = x, b + c = y and c + a = z, this reduces to
  
1 1 1
9≤ x+y+z + + ,
x z z
6 Inequalities

which is a consequence of the AM-GM inequality. (We observe that, all we


need here is the positivity of a, b, c; full force of the hypothesis that a, b, c are
the sides of a triangle is not needed in this part.)
Suppose c is the largest among a, b, c. By the symmetry, we may assume
a ≤ b ≤ c. In this case
a b c a c c
+ + ≤ + +
b+c c+a a+b a+c c+a a+b
c
= 1+
a+b
< 2,

since c < a + b by the triangle inequality.

If a1 , a2 , . . . , an are n positive real numbers, we define their harmonic mean


by
  n
H a 1 , a2 , . . . , an = .
1 1 1
+ + ···
a1 a2 an
Thus the harmonic mean of n positive real numbers is equal to the reciprocal
of the arithmetic mean of the reciprocals of the given numbers. Since
 1/n
1 1 1 1
+ + ··· + ≥n ,
a1 a2 an a1 a2 · · · an

it follows that    
G a 1 , a2 , . . . , a n ≥ H a 1 , a2 , . . . , an .
Thus we have for n positive real numbers a1 , a2 , . . . , an , the inequality
     
A a 1 , a 2 , . . . , a n ≥ G a 1 , a 2 , . . . , a n ≥ H a 1 , a2 , . . . , a n ,

or briefly AM ≥ GM ≥HM. This is often referred to as the AM-GM-HM in-


equality. Here again equality holds if and only if all the numbers are equal.

Example 1.4. For any four positive real numbers a1 , a2 , a3 , a4 , prove the inequal-
ity

a1 a2 a3 a4
+ + +
a1 + a2 a2 + a3 a3 + a4 a4 + a1
a1 a2 a3 a4
≤ + + + .
a2 + a3 a3 + a4 a4 + a1 a1 + a2

Solution: Adding
a2 a3 a4 a1
+ + +
a1 + a2 a2 + a3 a3 + a4 a4 + a1
Basic Inequalities 7

to both sides, the inequality can be written in the following equivalent form:
   
1 1 1 1
a2 + + a3 +
a3 + a4 a1 + a2 a4 + a1 a2 + a3
   
1 1 1 1
+ a4 + + a1 + ≥ 4.
a1 + a2 a3 + a4 a4 + a1 a2 + a3

However, using the AM-HM inequality, we know that

1 1 4
+ ≥ ,
a3 + a4 a1 + a2 a1 + a2 + a3 + a4
1 1 4
+ ≥ .
a4 + a1 a2 + a3 a1 + a2 + a3 + a4

Hence, the required inequality follows.

Example 1.5. Let a, b, c be the sides of a triangle such that

bc ca ab
+ + = s,
b+c c+a a+b
where s is the semi-perimeter of the triangle. Prove that the triangle is equilateral.

Solution: Observe that


a+b b+c c+a
2s = + +
2 2 2
2 2 2
≥ + +
1 1 1 1 1 1
+ + +
a b b c c a
2ab 2bc 2ca
= + +
a+b b+c c+a
= 2s.

Here the AM-HM inequality has been used. Thus equality holds in the AM-HM
inequality and hence a = b = c.

Example 1.6. Suppose a, b, c are positive real numbers. Prove the inequality
      2/3
a+b b+c c+a a+b+c
≥ abc . (1.5)
2 2 2 3

Solution: Introduce new variables x, y, z by

a b c
x= , y= , z= ,
a+b+c a+b+c a+b+c
8 Inequalities

so that x + y + z = 1. Now the inequality (1.5) is equivalent to


   
x+y y+z z+x 1 2/3
≥ xyz . (1.6)
2 2 2 3
This can be rewritten using x + y + z = 1 in the following form:
    8 2/3
1 − x 1 − y 1 − z ≥ xyz . (1.7)
3
Expanding the left hand side and using once again x + y + z = 1, we may write
(1.7) in the form  
 1/3 1 1 1 8
xyz + + −1 ≥ . (1.8)
x y z 3
Now the AM-HM inequality gives
 1/3 3
xyz ≥ ,
1 1 1
+ +
x y z
so that  
 1/3 1 1 1  1/3
xyz + + − 1 ≥ 3 − xyz .
x y z
Hence it is sufficient to prove that
 1/3 1
xyz ≤ .
3
Since x + y + z = 1, this follows from the AM-GM inequality.
Now let us see how one may proceed to generalise the AM-GM inequality.
Consider n non-negative real numbers a1 , a2 , a3 , . . . , an and n positive integers
k1 , k2 , k3 , . . . , kn . Form a set in which each aj appears exactly kj times, 1 ≤
j ≤ n. This set contains k1 + k2 + k3 + · · · + kn non-negative real numbers.
The arithmetic mean and the geometric mean of these numbers are
 k 1
k1 a1 + k2 a2 + · · · + kn an 1 +k2 +···+kn
, ak11 ak22 · · · aknn
k1 + k2 + · · · + kn
respectively. Now the AM-GM inequality gives
 k 1
k1 a1 + k2 a2 + · · · + kn an 1 +k2 +···+kn
≥ ak11 ak22 · · · aknn .
k1 + k2 + · · · + kn
Suppose we have positive rational numbers r1 , r2 , . . . , rn . Then we can
reduce all of them to a common denominator:
kj
rj = , 1 ≤ j ≤ n,
m
Basic Inequalities 9

where m and kj , 1 ≤ j ≤ n are natural numbers. It is not hard to see that


r 1 a 1 + r2 a 2 + · · · + r n a n k1 a1 + k2 a2 + · · · + k n an
= .
r1 + r 2 + · · · + r n k1 + k2 + · · · + k n
Moreover, we have
 r 1  k 1
1 +r2 +···+rn 1 +k2 +···+kn
ar11 ar22 · · · arnn = ak11 ak22 · · · aknn .

It follows that
 r 1
r1 a 1 + r2 a 2 + · · · + rn a n 1 +r2 +···+rn
≥ ar11 ar22 · · · arnn .
r1 + r2 + · · · + r n
It is possible to further extend this inequality using a continuity argument. If
λ1 , λ2 , . . . , λn are positive reals and a1 , a2 , . . . , an are non-negative reals, then
 λ 1
λ1 a 1 + λ2 a 2 + · · · + λn a n 1 +λ2 +···+λn
≥ aλ1 1 aλ2 2 · · · aλnn . (1.9)
λ1 + λ2 + · · · + λn
This is known as the generalised or the weighted arithmetic mean- geometric
mean inequality. With obvious modifications, this can be further carried to
the weighted arithmetic mean- geometric mean-harmonic mean inequality.

Example 1.7. Let a, b, c be positive real numbers. Prove that

aa bb cc ≥ (abc)(a+b+c)/3 .

Solution: Assigning the weights a, b, c to the numbers a, b, c respectively, the


weighted GM-HM inequality may be invoked. This leads to
a+b+c
√ a+b+c a+b+c √
3
a a bb c c ≥ a b c
= ≥ abc,
a+b+c 3

which gives the required inequality.

Example 1.8. Let a1 ≥ a2 ≥ · · · ≥ an and b1 , b2 , . . . , bn be two sequences


of non-negative real numbers such that for each k, 1 ≤ k ≤ n, the inequality
b1 b2 · · · bk ≥ a1 a2 · · · ak holds. Prove that

b1 + b 2 + · · · + b n ≥ a 1 + a 2 + · · · + a n . (1.10)

Solution: We may assume that aj > 0 for all j; otherwise delete those par-
ticular aj ’s which are equal to 0. This does not affect the inequality. Consider
the set 
b1 b2 bn
, ,... , ,
a1 a2 an
10 Inequalities

where each bj /aj is attached with the weight aj . The weighted arithmetic
mean is
     
a1 a1 + a2 a2 + · · · + an abnn
b1 b2
b1 + b 2 + · · · + b n
=
a1 + a2 + · · · + an a1 + a2 + · · · + an
and the weighted geometric mean is
 a 1  a 2  an  a1 +a2 +···+a
1
n
b1 b2 bn
··· .
a1 a2 an

Since aj ≥ aj+1 and b1 b2 · · · bj ≥ a1 a2 · · · aj for 1 ≤ j ≤ n, we have


 aj −aj+1
b 1 b2 · · · b j
≥ 1,
a1 a2 · · · aj

for 1 ≤ j ≤ n. Since
 a 1  a 2  a n
b1 b2 bn
···
a1 a2 an
 a n  an −an−1  a2 −a1
b1 b2 · · · bn b1 b2 · · · bn−1 b1
= ··· ,
a1 a2 · · · an a1 a2 · · · an−1 a1

we conclude that  a 1  a 2  a n
b1 b2 bn
··· ≥ 1.
a1 a2 an
It follows from the weighted AM-GM inequality that

b1 + b 2 + · · · + b n ≥ a 1 + a 2 + · · · + a n .

Alternate Solution:
Put λj = bj /aj , 1 ≤ j ≤ n. We have to show that


n
 
aj λj − 1 ≥ 0.
j=1

k  
If Lk = j=1 λj − 1 then

Lk = λ 1 + λ2 + · · · + λk − k
b1 b2 bk
= + + ··· + −k
a1 a2 ak
 1/k
b 1 b2 · · · bk
≥ k − k ≥ 0,
a1 a2 · · · ak
Basic Inequalities 11

for all k. On the other hand, we observe that


   k
n
  n
    
Lk ak − ak+1 = ak − ak+1 λj − 1
k=1 k=1 j=1
     
= λ1 − 1 a 1 + λ2 − 1 a 2 + · · · + λ n − 1 a n .
(Here we have set an+1 = 0.) Since Lk ≥ 0 and ak ≥ ak+1 for all k, it follows
that      
λ1 − 1 a1 + λ2 − 1 a2 + · · · + λn − 1 an ≥ 0,
which is to be proved.

1.3 Cauchy-Schwarz inequality


Consider the real numbers a, b, c, d. We have
(ac + bd)2 = a2 c2 + 2acbd + b2 d2 ≤ a 2 c 2 + a 2 d 2 + b 2 c 2 + b2 d 2
 2  
= a + b2 c 2 + d 2 ,
2 2 2 2
where we have used 2abcd ≤ a d +b c . This is a consequence of the inequality
(ad − bc)2 ≥ 0. Thus we obtain
   
ac + bd ≤ a2 + b2 c2 + d2 .

This is the simplest form of the Cauchy-Schwarz inequality. Geometrically, if


we consider two vectors (a, b) and (c, d) in the plane (here we represent vectors
by coordinates), then ac + bd gives the scalar product (or the dot product) of
→ →
two vectors. If we represent (a, b) by v1 and (c, d) by v2 , then
→ →  →  → 
ac + bd =v1 · v2 =  v1   v2  cos θ,
→ →
where θ is the angle between v1 and v2 . This shows that
      
ac + bd ≤  v→1   v→2  = a2 + b2 c2 + d2 .

Thus the Cauchy-Schwarz inequality here represents the fact that the absolute
value of the dot product of two vectors does not exceed the product of their
lengths. This property extends to higher dimensions giving us the general form
of the Cauchy-Schwarz inequality.
   
Theorem 2. Let a1 , a2 , a3 , . . . , an and b1 , b2 , b3 , . . . , bn be any two sets
of real numbers. Then
   1/2   1/2
 n  n n
 a b  ≤ a 2
b 2
. (1.11)
 j j j j
j=1 j=1 j=1

Equality holds if and only if there exists a constant λ such that aj = λbj , for
1 ≤ j ≤ n.
12 Inequalities

Proof: Let us write



n 
n 
n
A= a2j , B= b2j , C= a j bj .
j=1 j=1 j=1

The inequality to be proved is C 2 ≤ AB. If B = 0, then bj = 0 for all j so that


C = 0. Hence the inequality trivially holds. We may therefore assume that
B = 0. Since B is the sum of squares of real numbers, B must be positive.
Using the fact that the square of a real number is non-negative, we have

n
 2 
n 
n 
n
0≤ Baj − Cbj = B2 a2j + C 2 b2j − 2BC a j bj
j=1 j=1 j=1 j=1

= B 2 A + BC 2 − 2BC 2
 
= B BA − C 2 .
Since B > 0, it follows that C 2 ≤ AB. We also infer that equality holds if and
only if Baj − Cbj = 0 for all j, 1 ≤ j ≤ n. Taking λ = C/B, the condition for
equality that aj = λbj for 1 ≤ j ≤ n is obtained.

Remark 2.1: The Cauchy-Schwarz inequality extends to complex  numbers as 


well with obvious modifications.
 Let us consider two sets a 1 , a 2 , a 3 , . . . , an
and b1 , b2 , b3 , . . . , bn of complex numbers. Then the inequality
     n 
 n  n
 2 1/2   2 1/2
 a b ≤ a j  bj 
 j j 
j=1 j=1 j=1

holds good. Moreover equality holds here if and only if aj = λbj , 1 ≤ j ≤ n,


for some constant λ. The proof is similar to the one given above.
If we take n real numbers a1 , a2 , . . . , an , then the Cauchy-Schwarz inequal-
ity gives
√ 
a1 + a2 + · · · + an ≤ n a21 + a22 + · · · + a2n .
This may be written in the form

a1 + a2 + · · · + an a21 + a22 + · · · + a2n
≤ .
n n
Here the left hand side is the arithmetic mean of a1 , a2 , . . . , an and the right
side is the square root of the arithmetic mean of a21 , a22 , . . . , a2n . This is often
referred as the Root-Mean-Square inequality or some times called as the RMS
inequality for short.

Example 1.9. Let a, b, c be positive real numbers. Prove that


a2 b2 c2 a c b
2
+ 2+ 2 ≥ + + .
b c a c b a
Basic Inequalities 13

Solution: We write
a c b a b b c c a
+ + = · + · + · .
c b a b c c a a b
Now applying the Cauchy-Schwarz inequality to the sets
 
a b c b c a
, , and , , ,
b c a c a b
we get
 1/2  1/2
a c b a2 b2 c2 b2 c2 a2
+ + ≤ 2
+ 2+ 2 2
+ 2+ 2
c b a b c a c a b
2 2 2
a b c
= + 2 + 2.
b2 c a

Example 1.10. Let a1 , a2 , . . . , an and A be real numbers such that


 n  n 2
2 1
A+ ak < ak .
n−1
k=1 k=1
 
Prove that A < 2al am for every pair al , am , l = m.

Solution: We may take l = 1 and m = 2; the proof for any other pair of
indices is the same. Suppose A ≥ 2a1 a2 . Then we see that

n 
n
A+ a2k ≥ 2a1 a2 + a21 + a22 + a2k
k=1 k=3
 2 
n
= a1 + a2 + a2k
k=3

n 2
1
≥ ak .
n−1
k=1

(Here we haveapplied  the Cauchy-Schwarz
 inequality to the (n−1) tuples a1 +
a2 , a3 , . . . , an and 1, 1, . . . , 1 .) But this contradicts the given condition. We
conclude that A < 2a1 a2 .

Example 1.11. (Short-List, IMO-1993) Let a, b, c, d be positive real numbers.


Prove that
a b c d 2
+ + + ≥ .
b + 2c + 3d c + 2d + 3a d + 2a + 3b a + 2b + 3c 3
14 Inequalities

Solution: We write
  √  
a √ √
a= √ a b + 2c + 3d ,
cyclic cyclic
b + 2c + 3d

where the sum is taken cyclically over a, b, c, d. Now we apply Cauchy-Schwarz


inequality to the sets
 √ √ √ √
a b c d
√ ,√ ,√ ,√
b + 2c + 3d c + 2d + 3a d + 2a + 3b a + 2b + 3c
and

√ √ √ √ √ √ √ √
a b + 2c + 3d, b c + 2d + 3a, c d + 2a + 3b, d a + 2b + 3c ,

to get
  2   a
  
a ≤ a(b + 2c + 3d) .
b + 2c + 3d
cyclic cyclic cyclic

However, we have

a(b + 2c + 3d) = 4(ab + ac + ad + bc + bd + cd).
cyclic

Thus it follows that


 a (a + b + c + d)2
≥ .
b + 2c + 3d 4(ab + ac + ad + bc + bd + cd)
cyclic

On the other hand,

3(a + b + c + d)2 − 8(ab + ac + ad + bc + bd + cd)


 
= 3 a2 + b2 + c2 + d2 − 2(ab + ac + ad + bc + bd + cd)
= (a − b)2 + (a − c)2 + (a − d)2 + (b − c)2 + (b − d)2 + (c − d)2
≥ 0.

Combining these two, it follows that

(a + b + c + d)2 2

4(ab + ac + ad + bc + bd + cd) 3

and hence  a 2
≥ .
b + 2c + 3d 3
cyclic
Basic Inequalities 15

Example 1.12. Let a1 ≥ a2 ≥ · · · ≥ an ≥ 0 and b1 , b2 , . . . , bn be two sequences


of real numbers such that
 k k
aj ≤ bj ,
j=1 j=1

for each k, 1 ≤ k ≤ n. Prove that


n 
n
a2j ≤ b2j .
j=1 j=1

Solution: Take an+1 = 0. Since ak − ak+1 ≥ 0 for 1 ≤ k ≤ n, we have

 
k  
k
ak − ak+1 aj ≤ ak − ak+1 bj ,
j=1 j=1

for 1 ≤ k ≤ n. Summing this over k, we obtain

a21 + a22 + · · · + a2n ≤ a1 b1 + a2 b2 + · · · + an bn .

Applying the Cauchy-Schwarz inequality to the right side, we get


 1/2  1/2
a21 + a22 + · · · + a2n ≤ a21 + a22 + · · · + a2n b21 + b22 + · · · + b2n .

This simplifies to

a21 + a22 + · · · + a2n ≤ b21 + b22 + · · · + b2n .

Example 1.13. If a, b, c are positive real numbers, prove that


 a2 3
≥ ,
(a + b)(a + c) 4

where the summation is taken cyclically over a, b, c.

Solution: We have
⎛ ⎞2 ⎛ ⎞2
  a 
⎝ a⎠ = ⎝  (a + b)(a + c)⎠
cyclic cyclic
(a + b)(a + c)
⎛ ⎞⎛ ⎞
 a 2 
≤ ⎝ ⎠⎝ (a + b)(a + c)⎠ ,
(a + b)(a + c)
cyclic cyclic
16 Inequalities

using the Cauchy-Schwarz inequality. Thus we obtain


 a2 (a + b + c)2
≥ .
(a + b)(a + c) (a + b)(a + c) + (b + c)(b + a) + (c + a)(c + b)
cyclic

We need to show that


 
4(a + b + c)2 ≥ 3 (a + b)(a + c) + (b + c)(b + a) + (c + a)(c + b) .

This simplifies to
a2 + b2 + c2 ≥ ab + bc + ca,
which again follows from the Cauchy-Schwarz inequality.

Example 1.14. (China, 1989) Let a1 , a2 , a3 , . . . , an be n positive real numbers


which add up to 1. Prove that

n
a 1 √ n
 j ≥√ aj .
j=1
1 − aj n − 1 j=1

Solution: We may write



n
a  1
n  n
 j =  − 1 − aj .
j=1
1 − aj j=1
1 − aj j=1

Using the AM-GM inequality, we have


 1/n
n
1 1
 ≥ n n 
j=1
1 − aj j=1 1 − aj

1
= n!  1/n .
n
j=1 (1 − a j )

However,
"
n 1/n 1
n
n−1
(1 − aj ) ≤ (1 − aj ) = .
j=1
n j=1 n
Thus we obtain,

n 
1 n
 ≥n .
j=1
1 − aj n−1
On the other hand, the Cauchy-Schwarz inequality gives

n
 √ 
n 
1 − aj ≤ n! (1 − aj ) = n(n − 1).
j=1 j=1
Basic Inequalities 17

We thus get
   √
n
1
n
n  n
 − 1 − aj ≥ n − n(n − 1) = √ .
j=1
1 − aj j=1 n−1 n−1

We also have ⎛ ⎞2
n
√ 
n
⎝ aj ⎠ ≤ n aj = n.
j=1 j=1

Thus
√ 
n

n≥ aj
j=1

and hence √

n
aj n 1 n

 ≥√ ≥√ aj .
j=1
1 − aj n−1 n − 1 j=1

1.4 Chebyshev’s inequality


Let
 us consider the Cauchy-Schwarz
  inequality for two sets of real numbers
a1 , a2 , . . . , an and b1 , b2 , . . . , bn :


n 
n  21  
n  12
a j bj ≤ a2j b2j .
j=1 j=1 j=1

We may write this also in the form


    12   12
1 2 1 2
n n n
1
a j bj ≤ a b .
n j=1
n j=1 j n j=1 j

This shows that the Cauchy-Schwarz inequality relates the arithmetic means of
a21 , a22 , . . . , a2n and b21 , b22 , . . . , b2n with that of the numbers a1 b1 , a2 b2 , . . . , an bn .
How do we relate the arithmetic means of the numbers a1 , a2 , . . . , an and
b1 , b2 , . . . , bn with that of a1 b1 , a2 b2 , . . . , an bn . Chebyshev’s inequality answers
this question.
   
Theorem 3. Let a1 , a2 , . . . , an and b1 , b2 , . . . , bn be two sets of real num-
bers. Suppose either

a1 ≤ a2 ≤ · · · ≤ an and b1 ≤ b2 ≤ · · · ≤ bn ;

or
a1 ≥ a2 ≥ · · · ≥ an and b1 ≥ b2 ≥ · · · ≥ bn ;
18 Inequalities

i.e., both the sequences ak and bk are non-decreasing or both non-increasing.
Then the inequality
  
a1 + a2 + · · · + a n b1 + b2 + · · · + bn a 1 b1 + a 2 b2 + · · · + a n bn

n n n

holds. The inequality is strict unless at least one of the sequences is a constant
sequence.

Proof: We have

n  n 
  
n
  n
a j bj − a j bk = naj bj − aj bk
j=1 k=1 j=1 k=1
n 
n  
n 
= n a j bj − aj bk .
j=1 j=1 k=1

Similarly, it is easy to obtain

 n 
n   
n 
n  
n 
a k b k − a k bj =n a k bk − ak bj .
j=1 k=1 k=1 k=1 j=1

Using these two we obtain


n 
n  
n 
n a k bk − ak bj
k=1 k=1 j=1

n 
1
n
 
= a j b j − a j b k + a k bk − a k bj
2 j=1 k=1

1 
n n
  
= a j − a k b j − bk .
2 j=1 k=1

  
Since aj − ak bj − bk ≥ 0 whether both the sequences are non-decreasing or
both non-increasing, we get


n 
n  
n 
n a k bk − ak bj ≥ 0.
k=1 k=1 j=1

Here equality holds if and only if for each pair of indices j, k either aj = ak or
bj = bk . In particular taking j = 1, k = n we see that a1 = an or b1 = bn . It
follows that either aj is a constant sequence or bj is a constant sequence.
Basic Inequalities 19

Remark 3.1: If the sequences aj and bj are oppositely ordered, i.e., either
a1 ≤ a2 ≤ · · · ≤ an and b1 ≥ b2 ≥ · · · ≥ bn , or a1 ≥ a2 ≥ · · · ≥ an and
b1 ≤ b2 ≤ · · · ≤ bn , then the inequality reverses;
  
a1 + a2 + · · · + an b1 + b2 + · · · + bn a 1 b1 + a 2 b2 + · · · + a n bn
≥ .
n n n

The proof is


similar to the one given above, but with the
 crucial
 observation
 
that aj −ak bj −bk ≤ 0 holds in this case for any pairs aj , ak and bj , bk .

Remark 3.2: Chebyshev’s inequality can be generalised to three or more sets


of real numbers. Suppose a1,j , a2,j , a3,j . . . , ar,j , 1 ≤ j ≤ n be r sets
of real numbers such that

0 ≤ ak,1 ≤ ak,2 ≤ · · · ≤ ak,n ,

for 1 ≤ k ≤ r. Then the following inequality holds:


      
1 1 1
n n n n
1
a1,j a2,j a3,j · · · ar,j
n j=1 n j=1 n j=1 n j=1

1
n
≤ a1,j a2,j a3,j · · · ar,j .
n j=1

For r ≥ 3, it is essential to consider only non-negative sequences.

Remark 3.3: Suppose a1 , a2 , . . . , an and b1 , b2 , . . . , bn are two real sequences


such that either both are non-decreasing or both are non-increasing, nand let
p1 , p2 , . . . , pn be a sequence of non-negative real numbers such that j=1 pj is
positive. Then the following inequality holds:
 n   n   n 
j=1 pj aj bj j=1 pj aj j=1 pj bj
n ≥ n n .
j=1 pj j=1 pj j=1 pj

The proof runs on similar lines. This gives Chebyshev’s inequality with weights.

Example 1.15. Let a, b, c be positive real numbers and n a natural number.


Prove that
an bn cn an−1 + bn−1 + cn−1
+ + ≥ .
b+c c+a a+b 2

Solution: Since the expression is symmetric in a, b, c, we may assume that


a ≥ b ≥ c. Then we have
a b c
≥ ≥ .
b+c c+a a+b
20 Inequalities

In fact
a b
≥ ⇐⇒ ac + a2 ≥ bc + b2
b+c c+a
⇐⇒ (a − b)(c + a + b) ≥ 0
⇐⇒ a ≥ b.

Thus we have
a b c
an−1 ≥ bn−1 ≥ cn−1 and ≥ ≥ .
b+c c+a a+b
Using Chebyshev’s inequality for these numbers, we get
  a b c
  n
a bn cn

a n−1
+b n−1
+c n−1
+ + ≤3 + + .
b+c c+a a+b b+c c+a a+b
However, we have
a b c 3
+ + ≥ ,
b+c c+a a+b 2
using the left part of the inequality (1.4). Hence

an bn cc an−1 + bn−1 + cn−1


+ + ≥ .
b+c c+a a+b 2

Example 1.16. Let a1 , a2 , . . . , an be n real numbers in the interval [0, 1] such


that a1 + a2 + · · · + an = 1. Prove that
n
n − aj n2 − 1
≥ .
j=1
1 + naj 2

Solution: First we arrange aj ’s such that a1 ≤ a2 ≤ · · · ≤ an . Then we have

n − a1 ≥ n − a2 ≥ · · · ≥ n − an .

The arrangement of aj ’s shows that, 1 + na1 ≤ 1 + na2 ≤ · · · ≤ 1 + nan , and


hence
1 1 1
≥ ≥ ··· ≥ .
1 + na1 1 + na2 1 + nan
Using Chebyshev’s inequality, we obtain
  n   
1 
n n
n − aj 1
≥ (n − aj )
j=1
1 + naj n j=1 j=1
1 + naj
 
n2 − 1 
n
1
≥ .
n j=1
1 + naj
Basic Inequalities 21

Using the AM-HM inequality, we also get


n
1 n2 n2 n
≥ n = = .
1 + naj  2n 2
j=1
(1 + naj )
j=1

Combining these, we finally get

n  2  
n − aj n −1 n n2 − 1
≥ = .
j=1
1 + naj n 2 2

1.5 Rearrangement inequality


   
Consider two sets of real numbers: a1 , a2 , . . . , an and b1 , b2 , . . . , bn . We
pose the following question: among all permutations aj of aj and bj of

bj , which permutation maximises the product aj bj and which minimizes
it? The rearrangement inequality answers this question.

Theorem 4. Let a1 ≤ a2 ≤ · · · ≤ an and b1 ≤ b2 ≤ · · · ≤ bn (or a1 ≥


a2 ≥ · · · ≥ an and b1 ≥ b2 ≥ · · · ≥ bn ) be two sequences of real numbers. If
a1 , a2 , . . . , an is any permutation of a1 , a2 , . . . , an , the inequality


n 
n 
n
aj bn+1−j ≤ aj bj ≤ a j bj ,
j=1 j=1 j=1


n
holds. Thus the sum aj bj is maximum when the two sequences aj and bj
j=1
are ordered similarly (i.e., either both non-decreasing or both non-increasing). And
the sum is minimum when aj and bj are ordered in opposite manner (i.e., one
of them is increasing and the other decreasing).

Proof: We may assume that both aj ’s and bj ’s are non-decreasing; the proof
is similar in the other case. Suppose aj = aj . Let r be the largest index
such that ar = ar ; i.e., ar = ar and aj = aj for r < j ≤ n. This implies that
 
ar is from the set a1 , a2 , . . . , ar−1 and ar < ar . Further this also shows that
a1 , a2 . . . , ar is a permutation of a1 , a2 , . . . , ar . Thus we can find indices k < r
and l < r such that ak = ar and ar = al . It follows that

ak − ar = ar − al ≥ 0, br − bk ≥ 0.
22 Inequalities

We now interchange ar and ak to get a permutation a1 , a2 , . . . , an of a1 , a2 ,
. . . , an ; thus ⎧
⎪  
⎨ aj = aj , if j = r, k
 
ar = ak = ar ,

⎩ 
ak = ar = al .
Let us consider the sums

S  = a1 b1 + a2 b2 + · · · + an bn , S  = a1 b1 + a2 b2 + · · · + an bn .

Consider the difference S  − S  :



n
 
S  − S  = aj − aj bj
j=1
   
= ak − ak bk + ar − ar br
    
= ar − ak bk + ak − ar br
   
= ak − ar br − bk .

Since ak − ar ≥ 0 and br − bk ≥ 0, we conclude that S  ≥ S  . Observe


that the permutation a1 , a2 , . . . , an of a1 , a2 , . . . , an has the property that
aj = aj = aj for r < j ≤ n and ar = ak = ar . Hence the permutation aj in
place of aj may be considered and the procedure can be continued as above.
After at most n − 1 such steps, we arrive at the original permutation aj from
aj . At each step the corresponding sum is non-decreasing. Hence it follows
that
a1 b1 + a2 b2 + · · · + an bn ≤ a1 b1 + a2 b2 + · · · + an bn . (1.12)
To get the other inequality, let us put

cj = an+1−j , dj = −bn+1−j .

Then c1 , c2 , . . . , cn is a permutation of a1 , a2 , . . . , an and d1 ≤ d2 ≤ · · · ≤ dn .


Using the inequality (1.12) for the sequences cj and dj , we get

c 1 d1 + c 2 d2 + · · · + c n dn ≤ a 1 d1 + a 2 d2 + · · · + a n dn .

Substituting back cj and dj , we get


n 
n
− an+1−j bn+1−j ≤ − aj bn+1−j .
j=1 j=1

This reduces to

a1 b1 + a2 b2 + · · · + an bn ≥ a1 bn + a2 bn−1 + · · · + an b1 , (1.13)

which gives the other part.


Basic Inequalities 23

It is not difficult to find the condition under which equality holds in the
inequality. If for each pair k, l with 1 ≤ k < l ≤ n, either ak = al or ak > al
and bk = bl , then equality holds in (1.12). A similar condition is true for
equality in (1.13): for each k, l with 1 ≤ k < l ≤ n, either an+1−k = an+1−l or
an+1−k > an+1−l and bn+1−k = bn+1−l .
Many of the inequalities we have studied so far can be derived using the
rearrangement inequality.
Corollary 4.1: Let α1 , α2 , . . . , αn be n given real numbers and β1 , β2 , . . . , βn
be a permutation of α1 , α2 , . . . , αn . Then

n 
n
αj βj ≤ αj2 ,
j=1 j=1

with equality if and only if αj = βj .


Proof: Let α1 , α2 , . . . , αn be a permutation of  α1 , α2 , . . . , αn such that α1 ≤
 
α2 ≤ · · · ≤ αn . Then we can find a bijection σ of 1, 2, 3, . . . , n onto itself such 
that αj = ασ(j) , 1 ≤ j ≤ n; i.e., σ is a permutation on the set 1, 2, 3, . . . , n .
Let βj = βσ(j) . Then β1 , β2 , . . . , βn is a permutation of α1 , α2 , . . . , αn . Ap-
plying the rearrangement inequality to α1 , α2 , . . . , αn and β1 , β2 , . . . , βn , we
get
n n
  2 n
αj βj ≤ αj = αj2 .
j=1 j=1 j=1

On the other hand, we observe that



n 
n 
n
αj βj = ασ(j) βσ(j) = αj βj ,
j=1 j=1 j=1
 
since σ is a bijection on 1, 2, 3, . . . , n . It follows that


n 
n
αj βj ≤ αj2 .
j=1 j=1

Suppose equality holds in the inequality and αj =  βj . Then αj =
 βj .
Let k be the largest index such that αk = βk ; i.e., αk = βk and αj = βj for
    

k < j ≤ n. Let m be the least integer such that αk = βm 


. If m > k, then
      
βm = αm and hence αk = αm . This implies that αk = αk+1 = · · · = αm and
 
hence βk+1 = · · · = βm . But now we have a block of m + 1 − k equal elements
among α ’s and m − k elements among β  ’s. It follows that there is an m1 > m
such that αk = βm 
1
. Using m1 as pivotal, we obtain αk = αk+1 
= ··· =
    
αm = · · · = αm1 and βk+1 = · · · = βm = · · · = βm1 . This process cannot be
continued indefinitely. We conclude that αk = βl for some l < k, thus forcing
m < k.
24 Inequalities


Obviously βm = βk by our choice of k. We know that equality holds if and
only if for any two indices r = s, either αr = αs or βr = βs . Since βm 
= βk ,
    
we must have αm = αk . But then we have αm = αm+1 = · · · = αk . Using the
 
minimality of m, we see that k − m + 1 equal elements αm , αm+1 , · · · , αk must
    
be among βm , βm+1 , · · · , βn and since βk is different from αk , we must have
αk = βl for some l > k. But then using βl = αl we get
 
αm = αm+1 = · · · = αk = · · · = αl .

Thus the number of equal elements gets enlarged to l −m+1 > k −m+1. Since
this process cannot be continued indefinitely, we conclude that αj = βj . It
now follows that αj = βj .

Corollary 4.2: Let α1 , α2 , . . . , αn be positive real numbers; let β1 , β2 , . . . , βn


be a permutation of α1 , α2 , . . . , αn . Then
n
βj
≥ n,
α
j=1 j

and equality holds if and only if αj = βj .


Proof: Let α1 , α2 , . . . , αn be a permutation of α1 , α2 , . . . , αn such that α1 ≤
α2 ≤· · · ≤ αn . As in the proof of previous corollary, we can find a permutation
σ of 1, 2, 3, . . . , n such that αj = ασ(j) for 1 ≤ j ≤ n. We define βj = βσ(j) .
Then βj is a permutation of αj . Using the rearrangement theorem, we get


n   
n  
1 1
βj − ≤ αj − = −n.
j=1
αj j=1
αj

This gives the desired inequality. The condition for equality can be derived as
earlier.

Example 1.17. Prove the AM-GM-HM inequality using the rearrangement in-
equality.

Solution: Let a1 , a2 , . . . , an be n positive real numbers. Put


 1/n a1 a2 · · · ak
G = a1 a2 · · · an , and αk = , for 1 ≤ k ≤ n.
Gk
Now we set
β1 = α2 , β2 = α3 , . . . , βn−1 = αn , βn = α1 .
Now using corollary 4.2, we obtain

n
βj  αj+1
n−1
α1
n≤ = + .
α
j=1 j j=1
α j α n
Basic Inequalities 25

However,  
αj+1 a1 a2 · · · aj+1 /Gj+1 aj+1 α1 a1
=   = , = .
αj a1 a2 · · · aj /Gj G αn G
Thus we get
a1 + a2 + · · · + an
n≤ ,
G
which is same as the AM-GM inequality. Here equality holds if and only if
αj = βj . This is equivalent to α1 = α2 = · · · = αn which in turn is
equivalent to a1 = a2 = · · · = an .
Using corollary 4.2, it may also be obtained that


n
αj G G G G
n≤ = + + ··· + + .
j=1
βj a2 a3 an a1

This implies that


n  1/n
≤ G = a1 a2 · · · an .
1 1 1 1
+ + + ··· +
a1 a2 a3 an
Equality holds if and only if a1 = a2 = · · · = an .

Example 1.18. Prove the Cauchy-Schwarz inequality using the rearrangement


inequality.

Solution: Let a1 , a2 , . . . , an and b1 , b2 , . . . , bn be real numbers. We have to


show that

n 2 
n  
n 
a j bj ≤ a2j b2j .
j=1 j=1 j=1
n n
If either j=1 a2j = 0 or 2
j=1 bj = 0, the inequality is immediate. Hence, it
may be assumed that

n 1/2 
n 1/2
A= a2j and B = b2j
j=1 j=1

are both positive. Define



a
αj = Aj for 1 ≤ j ≤ n,
bj
αn+j = B for 1 ≤ j ≤ n.

We thus obtain 2n numbers, α1 , α2 , . . . , αn , αn+1 , . . . , α2n . Now consider the


permutation
βj = αn+j , βn+j = αj , 1 ≤ j ≤ n.
26 Inequalities

Using corollary 4.1, we obtain



n 
n 2n

αj αn+j + αn+j αj ≤ αj2 = 2.
j=1 j=1 j=1

Thus it follows that  n 


j=1 a j bj
2 ≤ 2.
AB
Hence

n 
n 1/2  
n 1/2
aj bj ≤ AB = a2j b2j .
j=1 j=1 j=1
Here equality holds if and only if αj = αn+j , for 1 ≤ j ≤ n. This is equivalent
to
a1 a2 an
= = ··· = = λ, a constant.
b1 b2 bn

Example 1.19. Let x, y, z be positive real numbers. Prove that



(z + x)2 (x + y)2 (y + z − x)(z − x) ≥ 0,
where the sum is taken cyclically over x, y, z.

Solution: Put z + x = 2a, x + y = 2b and y + z = 2c. Then a, b, c are the


sides of a triangle and the equivalent inequality is

a2 b2 (3c − a − b)(c − b) ≥ 0. (1.14)
We may assume that a ≤ c and b ≤ c. If a ≤ b ≤ c, then we have
s−a s−b s−c
≥ ≥ ,
a b c
where 2s = a + b + c. If b ≤ a ≤ c, then
s−b s−a s−c
≥ ≥ .
b a c
In any case the rearrangement inequality gives
(s − a)2 (s − b)2 (s − c)2 (s − b)2 (s − c)2 (s − a)2
a 2
+b 2
+c 2
≤a 2
+b 2
+c .
a b c b c a2
Simplification gives the inequality (1.14).

Example 1.20. Let a1 , a2 , . . . , an be positive real numbers and put


λ = a1 + a2 + · · · + an .
Prove that
a1 a2 an n
+ + ··· + ≥ .
λ − a1 λ − a2 λ − an n−1
Basic Inequalities 27

Solution: Note that the sum on the left side is symmetric in aj ’s and hence
we may assume that a1 ≤ a2 ≤ · · · ≤ an . This implies that
λ − a1 ≥ λ − a2 ≥ · · · ≥ λ − an ,
and hence
1 1 1
≤ ≤ ··· ≤ .
λ − a1 λ − a2 λ − an
For any k, consider the permutation of a1 , a2 , . . . , an defined by

ak+j−1 for 1 ≤ j ≤ n − k + 1,
βj =
ak+j−1−n for n − k + 2 ≤ j ≤ n.

Using the rearrangement inequality, we obtain


a1 a2 an
+ + ··· +
λ − a1 λ − a2 λ − an
ak ak+1 an
≥ + + ··· +
λ − a1 λ − a2 λ − an+k−1
a1 ak−1
+ + ··· + .
λ − an+k λ − an
This is true for every k. Now summing over k, 2 ≤ k ≤ n, we get
  a a2 an

1
n−1 + + ··· +
λ − a1 λ − a2 λ − an
 1     λ − aj
n n
≥ al = = n.
λ − aj λ − aj
j=1 l=j
j=1

This implies that



n
aj n
≥ .
j=1
λ − aj n−1

(This can also be obtained using the Cauchy-Schwarz inequality.)

1.6 Hölder’s and Minkowski’s inequalities


There is a very useful generalisation of the Cauchy-Schwarz  inequality which 
is
 known as Hölder’s
 inequality. Given any two n-tuples a1 , a2 , . . . , an and
b1 , b2 , . . . , bn of real numbers, the Cauchy-Schwarz inequality states that
n  n 1/2  
n 1/2
  2
 a j bj  ≤ aj b2j ,
j=1 j=1 j=1

with equality if and only if aj = λbj , 1 ≤ j ≤ n, for some constant λ. Here


squares and square-roots have special significance in the context of a more
general inequality.
28 Inequalities
   
Theorem 5. Let a1 , a2 , . . . , an and b1 , b2 , . . . , bn be two n-tuples of real
1 1
numbers and let p, q be two positive real numbers such that + = 1. (Such a
p q
pair of indices is called a pair of conjugate indices.) Then the inequality
 n  n
 p 1/p  n
 q 1/q
  a j  bj 
 a j bj  ≤ , (1.15)
j=1 j=1 j=1
 p  q
holds. Moreover this is an equality if and only if aj  = λbj  , 1 ≤ j ≤ n, for
some real constant λ.

Proof: We first prove an auxiliary result which is useful for the proof of the
above theorem.
If x, y are two positive real numbers, and p, q are positive
reals such that {p, q} is a pair of conjugate indices, then

xp yq
+ ≥ xy.
p q

There are many ways to prove this, but each one of them depends on some
continuity argument. We can use the generalised AM-GM inequality to get
xp yq     
1/p q 1/q
+ ≥ xp y = xy.
p q
Or one may consider the function
tp t−q
f (t) = + ,
p q
on the interval (0, ∞). Then f has a unique minimum at t = 1 and hence
tp t−q 1 1
+ ≥ + = 1,
p q p q
for all t > 0. Taking t = x1/q y −1/p , we obtain the desired inequality. It is also
easy to see that equality holds at the minimum point t = 1 which corresponds
to xp = y q .
Now taking
   
a k   bk 
x=  n ,y= n ,
  p 1/p   q 1/q
a j   bj 
j=1 j=1

we get
 p  q   
a k   bk   a k   bk 
   +    ≥     1/p   .
n n n n
 q 1/q
p  aj  p
q  bj  q
 aj  p
 bj 
j=1 j=1 j=1 j=1
Basic Inequalities 29

Now summing over k, we obtain


n
 
 a k bk 
1 1 j=1
+ ≥ n .
p q   p 1/p  
n
 q 1/q
a j  bj 
j=1 j=1

Thus we have

n
  n
 p 1/p  
n
 q 1/q
 a k bk  ≤ a j  bj  .
j=1 j=1 j=1

This proves (1.15). Equality holds if and only if


 p  q
a k   bk 
= ,

n
 p 
n
 q
a j  bj 
j=1 j=1

for 1 ≤ k ≤ n. Taking

n
 p ' 
n
 q
λ= a j  bj  ,
j=1 j=1

it may seen that equality holds if and only if


 p  
ak  = λbk q , for 1 ≤ k ≤ n.

Remark 5.1: If we take p = q = 2, Hölder’s inequality reduces to the Cauchy-


Schwarz inequality.

Remark 5.2: If either of p and q is negative, the inequality (1.15) gets re-
versed. For example, if p < 0 and q > 0 (obviously p and q both cannot be
negative simultaneously), we take

p 1
p 1 = − , q1 = .
q q

We observe that p1 > 0, p2 > 0 and

1 1 q  1
+ =− +q =q 1− = 1.
p1 q1 p p
30 Inequalities
 −q  q  q
Introducing uk = ak  , vk = ak  bk  and applying Hölder inequality to the
collections uk , vk with the conjugate indices p1 , q1 , we obtain


n
 −q  q  q   n
 −qp1  p11  
n
  qq1  q11
 a k   a k   bk  ≤ a k   a k   bk 
k=1 k=1 k=1

n
 p − pq  
n
 q
= a k  a k bk  .
k=1 k=1

Rearranging this, we get



n
  n
 p 1/p  
n
 q 1/q
 a k bk  ≥ a k  bk  ,
k=1 k=1 k=1

which is the opposite of the inequality (1.15).

Remark 5.3: We can have  Hölder’s inequality


  with weights.
  Consider two
n-tuples of real numbers a1 , a2 , . . . , an , b1 , b2 , . . . , bn ; let w1 , w2 , . . . , wn
be an n-tuple of positive real numbers. Then the following inequality holds:

n
  n
 p 1/p  
n
 q 1/q
w k  a k bk  ≤ w k a k  w k  bk  ,
k=1 k=1 k=1

for a pair p, q of positive conjugate indices. If either of p and q is negative, the


inequality gets reversed.

An inequality related to Hölder’s inequality is Minkowski’s inequality. This


is a generalisation of the well known  triangle
    inequality: if a and b are any two
arbitrary real numbers, then a+b ≤ a + b. It can easily be generalised to n-
  
tuples: if a1 , a2 , . . . , an and b1 , b2 , . . . , bn are two n-tuples of real numbers,
then
n
   n
   n
 
a j + bj  ≤ a j  +  bj  .
j=1 j=1 j=1
 
For an n-tuple a = a1 , a2 , . . . , an , we define its Euclidean norm ||a||2 by

n 1/2
||a||2 = |aj |2 .
j=1

This is precisely the Euclidean distance of a from the origin in Rn . It is an


easy consequence of the Cauchy-Schwarz inequality that

n
  1/2  
n
 2 1/2  
n
 2 1/2
a j + bj  2 ≤  aj  + bj  ,
j=1 j=1 j=1
Basic Inequalities 31
   
for any two n-tuples a1 , a2 , . . . , an and b1 , b2 , . . . , bn . This can be easily
seen by squaring on both sides. Thus we obtain

||a + b||2 ≤ ||a||2 + ||b||2 ,

for any two vectors a and b in Rn . As in the case of Hölder’s inequality, we


can replace 2 by any positive real number p, with proper inequality sign.
   
Theorem 6. Let p ≥ 1 be a real number; let a1 , a2 , . . . , an , b1 , b2 , . . . , bn
be two n-tuples. Then

n
  1/p  
n
 p 1/p  
n
 p 1/p
 a j + bj  p ≤ a j  +  bj  . (1.16)
j=1 j=1 j=1

Here equality holds if and only if aj = λbj for some constant λ, 1 ≤ j ≤ n.

Proof: We may assume p > 1, because the result for p = 1 is clear. Observe
that

n
  
n
   
a j + bj  p = aj + bj p−1 aj + bj 
j=1 j=1
n
     n
   
≤ aj + bj p−1 aj  + aj + bj p−1 bj .
j=1 j=1

Let q be the conjugate index of p. We use Hölder’s inequality to each sum on


the right hand side. Thus we have

n
     n
 p 1/p  
n
  1/q
aj + bj p−1 aj  ≤ a j  aj + bj (p−1)q .
j=1 j=1 j=1

Since p, q are conjugate indices, we get (p − 1)q = p. It follows that


n
     n
 p 1/p  
n
  1/q
aj + bj p−1 aj  ≤ a j   a j + bj  p .
j=1 j=1 j=1

Similarly,

n
     n
 p 1/p  
n
  1/q
aj + bj p−1 bj  ≤ bj  a j + bj  p .
j=1 j=1 j=1

It now follows that


  n

n
  n
 p 1/p  
n
 p 1/p     1/q
a j + bj  p ≤ a j  +  bj   a j + bj  p .
j=1 j=1 j=1 j=1
32 Inequalities

If we use 1 − (1/q) = 1/p, we finally get



n
  1/p  
n
 p 1/p  
n
 p 1/p
a j + bj  p ≤ a j  +  bj  .
j=1 j=1 j=1

Using the conditions for equality in Hölder’s inequality, we may obtain the
conditions for equality in Minkowski’s inequality: aj = λbj , for all 1 ≤ j ≤ n,
where λ is a real constant.

Remark 6.1: For 0 < p < 1, the inequality (1.16) gets reversed.
 
Remark 6.2: For any vector a = a1 , a2 , . . . , an and p > 0, we define


n 1/p
||a||p = |aj |p .
j=1

Thus Minkowski’s inequality can be put in the form

||a + b||p ≤ ||a||p + ||b||p ,

for p ≥ 1 and
||a + b||p ≥ ||a||p + ||b||p ,
for 0 < p < 1, where a and b are two vectors in Rn .

Example 1.21. Let a1 , a2 , . . . , an be real numbers. Prove that


 n   2   
  a k 3 n n
1
  ≤ |ak | 3/2
.
 k 1/3  k
k=1 k=1 k=1

Solution: We take p = 3/2 and q = 3. Then p, q are conjugate indices and


using Hölder’s inequality we obtain,
   2/3   1/3
 n ak  n n
1
 ≤ |ak | 3/2
.
 k 1/3  k
k=1 k=1 k=1

This gives the required inequality.

1 1
Example 1.22. Let p, q be two positive real numbers such that + = 1.
  p q
Suppose a1 , a2 , . . . , an is an n-tuple of real numbers. Show that


n
ak bk ≤ C, (1.17)
k=1
Basic Inequalities 33

  
n
 q
for every n-tuple b1 , b2 , . . . , bn of real numbers with bk  = 1 if and only if
k=1

n
 p 1/p
a k  ≤ C. (1.18)
k=1

 
Solution: Suppose (1.17) holds for all n-tuples b1 , b2 , . . . , bn such that
n
 q
bk  = 1. Consider the n-tuple defined by
k=1
 p−1
a k  Sgn(ak )
bk = 
n  p 1/q
,
a k 
k=1

where 
ak
|ak | if ak = 0,
Sgn(x) =
0 if ak = 0.
For this n-tuple, we have
n  (p−1)q n  p

n
 q   a k 
bk  = k=1 ak
n  p = k=1
n  p = 1.
a k 
k=1 k=1 ak k=1

Now the condition (1.17) gives


⎛ ⎞
 p−1 n  p 
 n 
⎜ ak  Sgn(ak ) ⎟ k=1 ak
n
 p 1/p
C≥ ak ⎝   ⎠ =   = a k  .
n  p 1/q n  p 1/q
k=1 ak k=1 ak
k=1 k=1

This shows that (1.18) holds.  


Conversely, suppose (1.18) holds and b1 , b2 , . . . , bn an n-tuple satisfying
n
 q
bk  = 1. Then Hölder’s inequality gives
k=1


n 
n
 p 1/p  
n
 q 1/q
a k bk ≤ a k   bk  ≤ C.
k=1 k=1 k=1

Thus (1.17) holds.


     
Example 1.23. Let a1 , a2 , . . . , an , b1 , b2 , . . . , bn , and c1 , c2 , . . . , cn be
three sets, where aj ’s, bj ’s, cj ’s are positive real numbers. If p, q, r are positive
reals such that (1/p) + (1/q) + (1/r) = 1, then
⎛ ⎞1/p ⎛ ⎞1/q ⎛ ⎞1/r
n 
n n n
a j bj c j ≤ ⎝ apj ⎠ ⎝ bqj ⎠ ⎝ crj ⎠ .
j=1 j=1 j=1 j=1
34 Inequalities

Solution: Using Hölder’s inequality with exponents p and (p − 1)/p, we get


⎛ ⎞1/p ⎛ ⎞(p−1)/p
n 
n n
a j bj c j ≤ ⎝ apj ⎠ ⎝ (bj cj )p/(p−1) ⎠ .
j=1 j=1 j=1

Now consider the exponents (p − 1)q/p and (p − 1)r/p. We observe that


 
p p p 1 1
+ = + = 1,
(p − 1)q (p − 1)r p−1 q r
because of (1/p) + (1/q) + (1/r) = 1. Thus Hölder’s inequality with these
exponents can be used to get
⎛ ⎞(p−1)/p ⎛ ⎞1/q ⎛ ⎞1/r
n n 
n
⎝ (bj cj )p/(p−1) ⎠ ≤⎝ bqj ⎠ ⎝ crj ⎠ .
j=1 j=1 j=1

Remark: This can be generalised to any number of sets with proper ex-
ponents.

1.7 Convex and Concave functions, Jensen’s inequality


Most of the inequalities, we have studied so far, are in fact consequences of
inequalities for a special class of functions, known as convex and concave func-
tions. Consider the function f (x) = x2 defined on R. If we take any two points
on the graph of f (x) = x2 , then the chord joining these points always lies above
this graph. In fact taking a < b, and the point λa + (1 − λ)b between a and b,
we see that
 2
λa + (1 − λ)b − λa2 − (1 − λ)b2 = −λ(1 − λ)(a − b)2 ≤ 0.

Thus we see that


 
f λa + (1 − λ)b ≤ λf (a) + (1 − λ)f (b),

for f (x) = x2 . This is taken as defining property of a convex function. We


shall see that the class of convex functions obey a general inequality known as
Jensen’s inequality.
Let I be an interval in R. A function f : I → R is said to be convex if for
all x, y in I and λ in the interval [0, 1], the inequality
 
f λx + (1 − λ)y ≤ λf (x) + (1 − λ)f (y), (1.19)

holds. If the inequality is strict for all x = y, we say that f is strictly convex
on I. If the inequality in (1.19) is reversed, i.e.,
 
f λx + (1 − λ)y ≥ λf (x) + (1 − λ)f (y),
Basic Inequalities 35

for all x, y ∈ I and λ ∈ [0, 1], then we say f is concave. If the inequality is
strict for all x = y, we say that f is strictly concave.
There are other equivalent characterisations of a convex function. Suppose
f : I → R is convex and x1 , x2 , x3 are in I such that x1 < x2 < x3 . Take
x3 − x2
λ= ,
x3 − x1
so that
x2 − x1
1−λ= , and x2 = λx1 + (1 − λ)x3 .
x3 − x1
We have
   
f x2 = f λx1 + (1 − λ)x3
   
≤ λf x1 + (1 − λ)f x3
x3 − x2   x2 − x1  
= f x1 + f x3 .
x3 − x1 x3 − x1
We may write this in the form
       
f x1 − f x2 f x2 − f x3
≤ ,
x1 − x2 x2 − x3
for all x1 < x2 < x3 in I. This may also be written in a more symmetric form:
     
f x1 f x2 f x3
  +  +   ≥ 0.
x1 − x2 x1 − x3 x2 − x1 x2 − x3 x3 − x1 x3 − x2

The convexity of a function has an intrinsic geometric interpretation. If we


look at the graph of a convex function, this is a subset of the plane and the
line joining any two points on the graph always lies above the graph. Suppose
z 1 = (a, f (a)) and z 2 = (b, f (b)) are two points on the graph of f . Then the
equation of the line joining these two points may be written in the form

f (b) − f (a)
L(x) = f (a) + (x − a).
b−a
Now any point between a and b is of the form x = λa + (1 − λ)b for some
λ ∈ [0, 1]. Hence
 
L(x) = L λa + (1 − λ)b
f (b) − f (a)  
= f (a) + λa + (1 − λ)b − a
b−a
  
= f (a) + 1 − λ f (b) − f (a)
= λf (a) + (1 − λ)f (b)
 
≥ f λa + (1 − λ)b = f (x).
36 Inequalities
 
Thus x, L(x) lies above (x, f (x)), a point on the graph of f .
There is still another way of looking at these things. A subset E of the
plane R2 is said to be convex if for every pair of points z 1 and z 2 in E, the
line joining z 1 and z 2 lies entirely in E. With every function f : I → R, we
associate a subset of R2 by
 
E(f ) = (x, y) : a ≤ x ≤ b, f (x) ≤ y .

Theorem 7. The function : I → R is convex if and only if E(f ) is a convex


subset of R2 .

Proof: Suppose f is convex. Let z 1 = (x1 , y1 ) and z 2 = (x2 , y2 ) be two points


of E(f ). Consider any point on the line joining z 1 and z 2 . It is of the form

z = λz 1 + (1 − λ)z 2
 
= λx1 + (1 − λ)x2 , λy1 + (1 − λ)y2 ,

for some λ ∈ [0, 1]. Observe that a ≤ λx1 + (1 − λ)x2 ≤ b. Moreover,


       
f λx1 + (1 − λ)x2 ≤ λf x1 + 1 − λ f x2
≤ λy1 + (1 − λ)y2 .

Hence it follows that z ∈ E(f ), proving that E(f ) is convex.


Conversely,
 supposeE(f ) is convex.
 Let x1 , x2 be two points in I and let
z 1 = x1 , f (x1 ) , z 2 = x2 , f (x2 ) . Then z 1 and z 2 are in E(f ). By convexity
of E(f ), the point λz 1 + (1 − λ)z 2 also lies in E(f ) for each λ ∈ [0, 1]. Thus
 
λx1 + (1 − λ)x2 , λf (x1 ) + (1 − λ)f (x2 ) ∈ E(f ),

for all λ ∈ [0, 1]. The definition of E(f ) shows that


 
f λx1 + (1 − λ)x2 ≤ λf (x1 ) + (1 − λ)f (x2 ),

for all λ ∈ [0, 1]. This shows that f is convex on the interval I.
The convexity of a function implies something about the slope of its graph.
The following theorem gives a complete description.

Theorem 8. Let f : I → R be a convex function and a ∈ I be a fixed point.


Define a function P : I \ {a} → R by

f (x) − f (a)
P (x) = .
x−a
Then P is a non-decreasing function on I \ {a}.
Basic Inequalities 37

Proof: Suppose f is convex on I and let x, y be two points in I, x = a, x = b


such that x < y. Then exactly one of following three possibilities hold:

a < x < y; x < a < y; or x < y < a.

Consider the case a < x < y; other cases may be taken care of similarly. We
write
x−a y−x
x= y+ a.
y−a y−a
The convexity of f shows that
 
x−a y−x x−a y−x
f y+ a ≤ f (y) + f (a).
y−a y−a y−a y−a

This is equivalent to
f (x) − f (a) f (y) − f (a)
≤ .
x−a y−a
Thus P (x) ≤ P (y). This shows that P (x) is a non-decreasing function for
x = a.
Interestingly, the converse is also true; if P (x) is a non-decreasing function
on I \ {a} for every a ∈ I, then f (x) is convex. The proof is not difficult. Fix
x < y in I and let a = λx + (1 − λ)y where λ ∈ (0, 1). (The cases λ = 0 or 1
are obvious.) In this case

f (x) − f (a) f (x) − f (a)


P (x) = =
x−a (1 − λ)(x − y)
f (y) − f (a) f (y) − f (a)
P (y) = = .
y−a λ(y − x)

The condition P (x) ≤ P (y) implies that f (a) ≤ λf (x) + (1 − λ)f (y). Hence
the convexity of f follows.
There is a useful, easy way of deciding whether a function is convex or
concave for twice differentiable functions. If f is convex on an interval I and
if its second derivative exists on I, then f is convex(strictly convex) on I if
f  (x) ≥ 0(> 0) for all x ∈ I. (Here f  (x) denotes the second derivative of f
at x.) Similarly f is concave(strictly concave) on I if f  (x) ≤ 0(< 0) for all
x ∈ I.
Here are some examples of convex and concave functions.

1. The function f (x) = xα is concave for 0 < α ≤ 1 and convex for 1 ≤ α <
∞ on the interval (0, ∞). We observe that

f  (x) = α(α − 1)xα−2 .

Hence f  (x) ≥ 0 on (0, ∞) if α ≥ 1 and f  (x) ≤ 0 for 0 < α ≤ 1.


38 Inequalities

2. Consider f (x) = ex . This is convex on R = (−∞, ∞). Here f  (x) =


ex > 0 for all x in (−∞, ∞) and hence ex is strictly convex.
3. The inverse of exponential function, namely the logarithmic function
f (x) = ln x is concave on (0, ∞). In this case f  (x) = −1/x2 which
is negative on (0, ∞). Thus f (x) = ln x is strictly concave on (0, ∞).
4. The function f (x) = sin x is concave on (0, π). We have f  (x) = − sin x
and hence f  (x) < 0 on (0, π).
5. If a function f is convex on I, then −f is concave on I.
The definition of a convex function involves inequality for two points x, y;
refer to (1.19). But this can be extended to any finite number of points.

Theorem 9. (Jensen’s inequality) Let f : I → R be a convex function. Suppose


x1 , x2 , . . . , xn are points in I and λ1 , λ2 , . . . , λn are real numbers in the interval
[0, 1] such that λ1 + λ2 + · · · + λn = 1. Then

n  n
 
f λj x j ≤ λj f xj . (1.20)
j=1 j=1

Proof: We use the induction on n. For n = 2, this is precisely the definition of


a convex function. Suppose the inequality (1.20) is true for all k < n; i.e., for
k < n, if x1 , x2 , . .
. , xk are k points in I and λ1 , λ2 , . . . , λk are real numbers
k
in [0, 1] such that j=1 λj = 1, then


k  k
 
f λj x j ≤ λj f xj .
j=1 j=1

Now consider any n points x1 , x2 , . . . , xn in the interval I and real numbers


λ1 , λ2 , . . . , λn in the interval [0, 1] such that λ1 + λ2 + · · · + λn = 1. Let us put
n−1
j=1 λj xj

n−1
y1 = n−1 , y2 = xn , α1 = λj , α 2 = λ n .
j=1 λj j=1

Observe that α2 = 1 − α1 , and y1 , y2 are in I. Using the convexity of f , we get


   
f α 1 y1 + α 2 y 2 = f α1 y1 + (1 − α1 )y2
   
≤ α1 f y1 + (1 − α1 )f y2
   
= α 1 f y1 + α 2 f y2 .

However, we have

n
α 1 y1 + α 2 y2 = λj x j .
j=1
Basic Inequalities 39
 
Now consider f y1 . If
λl
μl = n−1 , 1≤l ≤n−1
j=1 λj
n−1
then it can be easily verified that l=1 μl = 1. Using the induction hypothesis,
we get
 n−1
  n−1
  
f μl xl ≤ μl f xl .
l=1 l=1
Since

n−1
μ l x l = y1 ,
l=1
we get n−1  
n−1  
  l=1 λl f xl j=1 λj f x j
f y1 ≤ n−1 = .
j=1 λj
α1
Thus we obtain
 n−1  

n
   
j=1 λj f xj
f λj f x j ≤ α1 n−1 + λn f x n
j=1 j=1 λj

n
 
= λj f x j .
j=1

This completes the proof by the induction.

Remark 9.1: If f : I → R is concave, then the inequality (1.20) gets reversed.


If x1 , x2 , . . . , xn are points in I and λ1 , λ2 , . . . , λn are real numbers in the
interval [0, 1] such that λ1 + λ2 + · · · + λn = 1, then we have the inequality
  
n n
 
f λj x j ≥ λj f x j . (1.21)
j=1 j=1

Remark 9.2: Using the concavity of f (x) = ln x on (0, ∞), the AM-GM in-
equality may be proved. If x1 , x2 , . . . , xn are points in (0, ∞) and λ1 , λ2 , . . . , λn
are real numbers in the interval [0, 1] such that λ1 + λ2 + · · · + λn = 1, then we
have
 n   n
 
ln λj x j ≥ λj ln xj .
j=1 j=1

Taking λj = 1/n for all j,



n
x  1
n   1/n  n
j
ln ≥ ln xj = ln xj .
j=1
n n j=1 j=1
40 Inequalities

Using the fact that g(x) = exp(x)(= ex ) is strictly increasing on the interval
(−∞, ∞), this leads to

1
n 
n  
1/n
xj ≥ exp ln xj
n j=1 j=1
"
n   
1/n
= exp ln xj
j=1
 1/n
= x1 x2 · · · xn .

In fact, the generalised AM-GM inequality may also be proved by this


method . We have

n  n
   n
λ
ln λj xj ≥ λj ln xj = ln xj j ,
j=1 j=1 j=1

for all points x1 , x2 , . . . , xn in (0, ∞) and λ1 , λ2 , . . . , λn in the interval [0, 1]


such that λ1 + λ2 + · · · + λn = 1. Using the property of exponential function,
we obtain
n "
n
λ
λj xj ≥ xj j .
j=1 j=1

Now for any n positive real numbers α1 , α2 , . . . , αn , consider


αj
λj =  n .
k=1 αk
n
Observe that λj are in [0, 1] and j=1 λj = 1. These choices of λj give

α 1 x1 + α 2 x2 + · · · + α n xn  1/(α1 +α2 +···+αn )


≥ xα
1 x2 · · · xn
1 α2 αn
,
α1 + α2 + · · · + αn
which is the generalised AM-GM inequality.

Remark 9.3: We use the function f (x) = xp to derive Hölder’s inequality.


We know that f (x) = xp is convex for p ≥ 1 and concave  for 0 < p < 1 on
the interval (0, ∞). Hence for any n-tuple x1 , x2 , . . . , xn of real numbers and
λ1 , λ2 , . . . , λn in [0, 1], we have

n p 
n
λj x j ≤ λj xpj for p ≥ 1,
j=1 j=1

and

n p 
n
λj x j ≥ λj xpj for 0 < p < 1.
j=1 j=1
Basic Inequalities 41
   
Let a1 , a2 , . . . , an and b1 , b2 , . . . , bn be two n-tuples of real numbers and
1 1
p > 1 be a given real number. Let q be the conjugate of p; + = 1. We
p q
may assume bj = 0 for all j; otherwise we may delete all those bj which are
zero without affecting the inequality. We now set


n
|bk |q |ak |
t= |bj |q , λk = , xk = .
j=1
t |bk |q−1

We observe that λk are in [0, 1] and λ1 + λ2 + · · · + λn = 1. Using the convexity


of xp , we get
 n p n
λj x j ≤ λj xpj .
j=1 j=1

This implies that


 p  1
n n n
|bk |q |ak | |bk |q |ak |p
≤ = |ak |p ,
t |bk |q−1 t |bk | (q−1)p t
k=1 k=1 k=1

since (q − 1)p = q. Simplification gives


n 
n 1/p 
n 1/p  
n 1/q
1−(1/p)
|bk ak | ≤ |ak | p
t = |ak | p
|bk | q
.
k=1 k=1 k=1 k=1

If 0 < p < 1, then f (x) = xp is concave and the inequality is reversed:


n 
n 1/p  
n 1/q
|bk ak | ≥ |ak | p
|bk | q
.
k=1 k=1 k=1

We observe that for 0 < p < 1, its conjugate index q is negative.


There is an interesting generalisation of the rearrangement inequality to
convex functions.

Theorem 10. Suppose f : I → R is a convex function; a1 ≤ a2 ≤ · · · ≤ an and


b1 ≤ b2 ≤ · · · ≤ bn are two sequences of real numbers in I such that a1 + b1 ∈ I
and an + bn ∈ I. Let a1 , a2 . . . , an be a permutation of a1 , a2 , . . . , an . Then the
inequality


n
  n
  n
 
f aj + bn+1−j ≤ f aj + bj ≤ f a j + bj .
j=1 j=1 j=1

holds.
42 Inequalities

Proof: We follow the proof of theorem 4. Suppose aj = aj and r be the
largest index such that ar = ar ; i.e., ar = ar but aj = aj for r < j ≤ n.
 
Hence ar is in the set a1 , a2 , . . . , ar−1 and ar < ar . Since aj = aj for
r < j ≤ n, we see that (a1 , a2 . . . , ar ) is a permutation of (a1 , a2 , . . . , ar ).
Hence we can find k < r, l < r such that ak = ar and ar = al . We deduce that
ak − ar = ar − al ≥ 0 and br − bk ≥ 0. We now interchange ar and ak to get a
permutation (a1 , a2 , . . . , an ) of (a1 , a2 . . . , an ); thus

aj = aj , for j = r, k, ar = ak = ar , ak = ar = al .

Let us write

n
  
n
 
S  = f aj + bj , S = f aj + bj .
j=1 j=1

We have
       
S  − S  = f ar + br + f ak + bk − f ar + br − f ak + bk
       
= f a r + br + f a l + b k − f a l + b r − f a r + b k .

We observe that

al + bk < ar + bk and al + br < ar + br .

These give

a l + bk < a r + bk ≤ a r + br ,
a l + bk ≤ a l + b r < a r + b r .

If x1 , x2 , x3 are in I, then the convexity of f implies that


           
x3 − x1 f x2 ≤ x3 − x2 f x1 + x2 − x1 f x3 .

Taking x1 = al + bk , x2 = ar + bk and x3 = ar + br , we get


           
a r + b r − a l − bk f a r + b k ≤ b r − b k f a l + b k + a r − a l f a r + b r .

Similarly, taking x1 = al + bk , x2 = al + br and x3 = ar + br , we get


           
a r + br − a l − bk f a l + br ≤ a r − a l f a l + bk + br − bk f a r + br .

Adding these two, we obtain


     
a r + br − a l − b k f a r + bk + f a l + br
     
≤ a r + br − a l − b k f a l + b k + f a r + br .

Since al + bk < ar + br , we deduce that


       
f a r + bk + f a l + br ≤ f a l + b k + f a r + b r .
Basic Inequalities 43

This proves that S  − S  ≥ 0.


Now we observe that the permutation (a1 , a2 , . . . , an ) has the property
ar = ar and aj = aj , for r < j ≤ n. We may consider the (a1 , a2 , . . . , an ) in


place (a1 , a2 , . . . , an ) and proceed as above. After at most n−1 steps we arrive
at the original numbers aj from aj , and at each stage the corresponding
sum is non-decreasing. We finally get

n
  n
 
f aj + bj ≤ f a j + bj .
j=1 j=1

For proving the other inequality, we proceed along the same lines. We define
cj = an+1−j so that c1 ≥ c2 ≥ · · · ≥ cn . We have to show that


n
  n
 
f an+1−j + bj ≤ f aj + bj .
j=1 j=1

We also set cj = aj . Thus the inequality becomes


n
  n
 
f c j + bj ≤ f cj + bj ,
j=1 j=1

where (c1 , c2 , . . . , cn ) is a permutation of (c1 , c2 , . . . , cn ). We assume that


cj = cj and let r be the smallest index such that cr = cr ; i.e., cr = cr but
 
cj = cj for 1 ≤ j < r. This forces that cr ∈ cr+1 , cr+2 , . . . , cn and cr < cr .
Observe that (cr , cr+1 , . . . , cn ) is a permutation of (cr , cr+1 , . . . , cn ). Hence
we can find k > r, l > r such that ck = cr and cr = cl . This implies that
ck − cr = cr − cl ≥ 0 and bk − br ≥ 0. Now we interchange cr and ck to get a
permutation (c1 , c2 , . . . , cn ) of (c1 , c2 , . . . , cn ); thus

cj = cj , for j = r, k, cr = ck = cr , ck = cr = cl .

Now we compute the difference between



n
  
n
 

S = f cj + bj , 
S = f cj + bj ,
j=1 j=1

and obtain
       
S  − S  = f cr + br + f ck + bk − f cr + br − f ck + bk
       
= f c r + br + f c l + bk − f c l + br − f c r + bk .

We observe that

c l + br ≤ c l + b k < c r + bk ,
c l + br ≤ c r + b r < c r + b k .
44 Inequalities

The convexity of f gives


           
c r + bk − c l − b r f c l + b k ≤ c r − c l f c l + br + b k − br f c r + b k ,

and
           
c r + bk − c l − br f c r + br ≤ bk − b r f c l + br + c r − c l f c r + bk .

Addition of these two gives


     
c r + bk − c l − br f c l + bk + f c r + br
     
≤ c r + b k − c l − br f c l + b r + f c r + bk .

Since cr + bk − cl − br = 0, we get
       
f c l + b k + f c r + b r ≤ f c l + b r + f c r + bk .

This shows that S  ≤ S  . We also observe that the new sequence cj has
the property: cr = cr and cj = cj for 1 ≤ j < r. We may consider now the
sequence cj in place of cj and continue the above argument. At each stage
the sum never increases. After at most n − 1 steps we arrive  at the sequence

cj . Hence the corresponding sum cannot exceed that of S  = j=1 f cj +bj .
n

We thus get
n
   n
 
f c j + bj ≤ f cj + bj ,
j=1 j=1

which is to be proved.

Example 1.24. Show that for all n ≥ 1, the inequality


"
n  n + 1 n(n+1)/2
jj ≥
j=1
2

holds.

Solution: Let us put aj = bj = j, aj = n + 1 − j for 1 ≤ j ≤ n. Then we see


that a1 ≤ a2 ≤ · · · ≤ an and b1 ≤ b2 ≤ · · · ≤ bn . Moreover a1 ≥ a2 ≥ · · · ≥ an .
Consider the function f (x) = x ln x. This is a convex function on (0, ∞). In
fact we see that
1
f  (x) = ln x + 1, and f  (x) = > 0 for x > 0.
x
Now the rearrangement inequality applied to f (x) gives

n 
n
(n + 1 − j + j) ln(n + 1 − j + j) ≤ (j + j) ln(j + j).
j=1 j=1
Basic Inequalities 45

This simplifies to


n
n(n + 1) ln(n + 1) ≤ 2 j ln j + n(n + 1) ln 2.
j=1

This further reduces to

n(n + 1)  n + 1   " 
n n
ln ≤ j ln j = ln jj .
2 2 j=1 j=1

Now taking exponentiation both sides we get the desired result.


The concavity of the log function on (0, ∞) also helps us to derive a very
interesting inequality known as Bernoulli’s inequality.

Example 1.25. (Bernoulli’s inequality) Prove that for x > −1


 a
1+x ≤ 1 + ax, if 0 < a < 1,

and
 a
1+x ≥ 1 + ax if 1 ≤ a < ∞.

Solution: We use the concavity of f (x) = ln(x) on (0, ∞). Since x > −1, we
see that 1 + x > 0. If 0 < a < 1, we have
   
ln 1 + ax = ln a(1 + x) + 1 − a
   
≥ a ln 1 + x + 1 − a ln(1)
 
= a ln 1 + x .
 a
The exponentiation gives 1 + x ≤ 1 + ax. Suppose 1 ≤ a < ∞. Then again
 
  a − 1 1 
ln 1 + x = ln + 1 + ax
a a
a−1 1  
≥ ln(1) + ln 1 + ax
a a
1  
= ln 1 + ax .
a
 a
This gives 1 + x ≥ 1 + ax.
 
Example 1.26. Let a = a1 ,a2 ,a3 ,. . .,an be an n-tuple of positive real num-
bers and w1 ,w2 ,w3 ,. . .,wn be another n-tuple of positive real numbers (called
46 Inequalities

weights). Define the weighted mean of order r by


⎧  
⎪ n r 1/r

⎪ j=1 wj aj

⎪ n if r = 0, and |r| < ∞,


⎪ j=1 wj  * 

⎨  1 nj=1 wj
Mr (a) = n wj

⎪ j=1 aj , if r = 0,

⎪  



⎪ min a1 ,a2 ,a3 ,. . .,an , if r = −∞,

⎩  
max a1 ,a2 ,a3 ,. . .,an , if r = ∞
Suppose r, s are two real numbers such that r < s. Prove that
M−∞ (a) ≤ Mr (a) ≤ Ms (a) ≤ M∞ (a).

Solution: We prove this by considering several cases.


(a) Suppose r = 0. Then we have s > 0 and hence
* 
"  s nj=1 wj
 s n
w
M0 (a) = aj j
j=1
 * 
"  1 n
n
 s wj j=1 wj
= aj
j=1
n
wj asj
≤ j=1
n (by the generalised AM-GM)
j=1 wj
 s
= Ms (a) .

(b) Suppose 0 < r < s. Put α = s/r and observe that α > 1. Hence the
function f (x) = xα is convex on (0, ∞). Hence Jensen’s inequality gives
 n  n
wj  rα 
r α n s
j=1 wj aj j=1 wj aj
 ≤  aj =  .
wk j=1
wk wk
 rα  s
This shows that Mr (a) ≤ Ms (a) . Taking s-th root, we get
Mr (a) ≤ Ms (a).
(c) If s = 0, then r < 0 and we take t = −r. Using the generalised AM-GM
inequality, we have
* 
 wj ⎛ ⎞ 1  wk
at " n
1 ⎠
 j ≥ ⎝ twj
wk j=1 aj
* 
⎛ ⎞ t  wk
" 1
n
= ⎝ w

a j
j=1 j
Basic Inequalities 47

Simplification gives
  * 
  wj  1/t
⎛ ⎞ 1  wk
atj "
n
1 ⎠
 ≥⎝ wj .
wk a
j=1 j

This is equivalent to
 * 
⎛ ⎞
 1/r "n
1 wk
wj arj
≤⎝ aj j ⎠
w
 .
wk j=1

Thus we obtain Mr (a) ≤ M0 (a).


(d) If r < 0 < s, then we have

Mr (a) ≤ M0 (a) ≤ Ms (a)

from (c) and (d).


(e) Suppose r < s < 0. Then we have 0 < −s < −r and hence

M−s (a−1 ) ≤ M−r (a−1 ),

where  
−1 1 1 1
a = , ,... ,
a1 a2 an
But then
1 1
M−s (a−1 ) = , M−r (a−1 ) = .
Ms (a) Mr (a)
Hence we get Mr (a) ≤ Ms (a).
(f) Obviously
   
min a1 ,a2 ,a3 ,. . .,an ≤ Mr (a) ≤ max a1 ,a2 ,a3 ,. . .,an .

Combining all these, we get the desired inequality.

1.8 Inequalities for symmetric functions

Given n complex numbers a = a1 ,a2 ,a3 ,. . .,an , we consider the polynomial
whose roots are a1 ,a2 ,a3 ,. . .,an ;
"
n
P (x) = (x − aj ).
j=1
48 Inequalities

Expanding this we obtain



n  
n
P (x) = xn + (−1)k uk xn−k ,
k
k=1

where
n
j=1 aj
u1 = u1 (a) =  n ,
1

j<k aj ak
u2 = u2 (a) =  n ,
2
.. ..
. .
n
j=1 aj
un = un (a) =  n .
n

These functions u1 ,u2 ,u3 ,. . .,un are called the elementary symmetric functions
in the variables a1 ,a2 ,a3 ,. . .,an . We may also write them in the form ur (ak )
to show the dependence on the sequence a1 ,a2 ,a3 ,. . .,an . We observe that the
permutation of a1 ,a2 ,a3 ,. . .,an does not alter any of the uj ’s, 1 ≤ j ≤ n. We
take u0 (a) = 1.
1/k
We further introduce vk = uk . We observe that v1 is simply the arithmetic
mean of a1 ,a2 ,a3 ,. . .,an and vn is their geometric mean.

Theorem 11. (Newton’s inequalities) Let a1 ,a2 ,a3 ,. . .,an be a sequence of


non-negative real numbers. We have the inequalities
vn ≤ vn−1 ≤ · · · ≤ v1 .

Proof: Let a = min{aj : 1 ≤ j ≤ n} and b = max{aj : 1 ≤ j ≤ n}. Then


P (x) = 0 has n roots in [a, b]. Hence the derivative polynomial P  (x) has n − 1
zeros in [a, b], counted according to multiplicity. (If P (x) = 0 has a root α of
multiplicity l, then α is also a root of P  (x) = 0 of multiplicity l − 1.) It follows
that for any positive integer k, P (k) (x) = 0 has n − k roots in [a, b]. (Here
P (k) (x) denotes the k-th derivative of P (x).)
We observe that
n!
P (n−2) (x) = (x2 − 2u1 x + u2 ).
2
This has real roots by previous observation. Hence we get
u2 ≤ u21 . (1.22)
We also observe that
   1   n
n un−2 n un−1 1
= , = .
2 un aj ak 1 un a
j=1 j
j<k
Basic Inequalities 49

Applying (1.22) for the numbers 1/a1 , 1/a2 , . . . , 1/an , we get


⎛ ⎞2
1  1 ⎝ 1 ⎠
n
1
 n ≤  2 .
2
aj ak n a
j=1 j
j<k 1

This reduces to
un un−2 ≤ u2n−1 . (1.23)
We show that ur−1 ur+1 ≤ u2r for r = 2, 3, . . . , n − 1. We use the induction on
the number of aj ’s. For n = 3, we have r = 2 and hence we have to prove that
u1 u3 ≤ u22 . This follows from (1.23) with n = 3. Suppose the result holds for
a set of n − 1 numbers. We have
   

n−1
 n−1 k n−1 n−1−k
P (x) = n x + (−1) uk x .
k
k=1

Let the roots of P  (x) = 0 be b1 , b2 , . . . , bn−1 . Introduce


  
n−1
zr = bj 1 b j 2 · · · bj r .
r j <j <···<j
1 2 r

Then we see that

P  (x) = n(x − b1 )(x − b2 ) · · · (x − bn−1 )


   

n−1
n−1 k n−1 n−1−k
= n x + (−1) zk x .
k
k=1

It follows that zr = ur for 1 ≤ r ≤ n − 1. Now, the induction hypothesis holds


for the set {b1 , b2 , . . . , bn−1 }. This gives zr−1 zr+1 ≤ zr2 for 2 ≤ r ≤ n − 2.
Using zr = ur for 1 ≤ r ≤ n − 1, we infer that ur−1 ur+1 ≤ u2r for 2 ≤ r ≤ n − 2.
For r = n − 1, this follows from (1.23). This completes the induction. We
observe that u2 ≤ u21 , (u1 u3 )2 ≤ u42 , (u2 u4 )3 ≤ u63 and so on. Multiplying all
these, we get
u21 u42 u63 · · · u2r−2 ur+1 ≤ u21 u42 · · · u2r
r−1 r
r−1 ur r .

This reduces to urr+1 ≤ ur+1


r and hence it follows that vr+1 ≤ vr for r =
1, 2, . . . , n − 1.
Chapter 2
Techniques for proving inequalities

2.1 Introduction
There are several standard ways of proving a given inequality. We have al-
ready seen how to obtain the AM-GM inequality using forward and backward
induction. One can also use the known standard inequalities or use ideas from
calculus. In some cases trigonometric substitutions simplify the result. We
take each of these separately and illustrate them using examples.

2.2 Use of induction


In many problems on inequalities, the Principle of Mathematical Induction is a
powerful tool that can be deployed to prove the result. The technique is to use
induction on the number of variables or some times on the degree (or power).

Examplen2.1. Let x1 , x2 , . . . , xn be n real numbers in the interval [0, 1] and


suppose j=1 xj = m + r, where m is an integer and 0 ≤ r < 1. Prove that

n
x2j ≤ m + r2 .
j=1

Solution: We show by induction on n that


⎛ ⎞
n +
n , n +
n , 2
x2j ≤ xj + ⎝ xj − xj ⎠ , (2.1)
j=1 j=1 j=1 j=1

where [x] denotes the integral part of x. If n = 1, then x1 lies in the interval
[0, 1] and (2.1) reduces to
- .  - . 2
x21 ≤ x1 + x1 − x1 ,
- .
which is true since x1 = 0 or 1. Suppose (2.1) holds for 1 ≤ n ≤ N − 1. Put

N −1
SN −1 = xj = k + t, 0 ≤ t < 1,
j=1


N
SN = xj = m + r, 0 ≤ r < 1.
j=1
52 Inequalities

We show that
- .  - .2 - .  - .2
x2N ≤ SN + SN − SN − SN −1 − SN −1 − SN −1 . (2.2)

This is equivalent to

m − k + r − t)2 ≤ m + r2 − k − t2 .

We can further reduce it to another equivalent statement:

2(r − t)(m − k − t) ≤ (m − k) − (m − k)2 . (2.3)

If m = k, then t ≤ r and hence (2.3) reduces to −2(r − t)t ≤ 0, which is true. If


m = k + 1, then r ≤ t since xn ≤ 1 and (2.3) is equivalent to 2(r − t)(1 − t) ≤ 0.
This again is true. Thus (2.2) holds for N . Hence


N 
N −1
x2j = x2j + x2N
j=1 j=1
- .  - .2
≤ SN −1 + SN −1 − SN −1 + x2n
- .  - .2
≤ SN + SN − SN .

This completes induction.


Here is a generalisation of the previous result(CRUX-1992).
Let α > 0 and let f : [0, α] → R be a convex function. Let x1 , x2 , . . . , xn
n
be n(≥ 2) points in the interval [0, α] such that j=1 xj = mα + r, where m
is an integer and 0 ≤ r < α. Then

n
         
f xj ≤ n − m − 1 f 0 + mf α + f r .
j=1

A proof of this may be based on the properties of convex functions and the
principle of induction as used above.

Example 2.2. Let x1 ≥ x2 ≥ · · · ≥ xn > 0 be n real numbers. Prove that


" n  
1 + xj 1 + x1 + x1 x2 + · · · + x1 x2 · · · xn
≤ .
j=1
2 n+1

Solution: We use induction on n. If n = 1, the result is immediate. Suppose


it holds for all m ≤ n − 1. Take any n real numbers such that x1 ≥ x2 ≥ · · · ≥
xn > 0. We apply the induction hypothesis to the sequence x2 ≥ x3 ≥ · · · ≥
xn > 0 and obtain
n 
"

1 + xj 1 + x2 + x2 x3 + · · · + x2 x3 · · · xn
≤ .
j=2
2 n
Techniques 53

Multiplying both the sides by (1 + x1 )/2 , this takes the form


n   
" 1 + xj 1 + x1  1 + R 
≤ ,
j=1
2 2 n

where
R = x2 + x2 x3 + · · · + x2 x3 · · · xn .
Thus the inequality to be proved is
(1 + x1 )(1 + R) 1 + x1 + x1 R
≤ .
2n n+1
This is equivalent to
 
R (n + 1) − (n − 1)x1 ≤ (n − 1)(1 + x1 ).

If n + 1 ≤ (n − 1)x1 , then the left hand side is non-positive and the right hand
side is non-negative. Hence the result is true in this case. Thus we may assume
that (n + 1) − (n − 1)x1 > 0. Thus it is sufficient to deduce that
(n − 1)(1 + x1 )
R≤ .
(n + 1) − (n − 1)x1
But, observe that

R = x2 + x2 x3 + · · · + x2 x3 · · · xn ≤ x1 + x21 + · · · + xn−1
1 .

Therefore it is enough to show that


  
x1 + x21 + · · · + xn−1
1 (n + 1) − (n − 1)x1 ≤ (n − 1)(1 + x1 ).

Equivalently, we need to prove that


 
(n − 1)xn1 − 2 x1 + x21 + · · · + xn−1
1 + (n − 1) ≥ 0.

Consider the polynomial


 
P (x) = (n − 1)xn − 2 x + x2 + · · · + xn−1 + (n − 1).

Obviously P (1) = 0. We also have


 
P  (x) = n(n − 1)xn−2 − 2 1 + 2x + 3x2 + · · · + (n − 1)xn−2 .

This shows that P  (1) = 0. Thus P (x) = 0 has a double root at x = 1.


However, the number of sign changes in P (x) is 2 and hence the number of
positive roots of P (x) = 0 cannot exceed 2. We conclude that P (x) > 0 for
x > 1. But, P (x) is a reciprocal polynomial and hence xn P 1/x = P (x). It
follows that P (x) > 0 for 0 < x < 1. Thus P (x) ≥ 0 for x > 0. This completes
the proof.
54 Inequalities

2.3 Application of known inequalities


We have already built up a library of some basic inequalities. These may be
used in proving a new inequality. Only some insight into the problem leads to
the correct inequality to be chosen. Sometimes many inequalities may be used
in conjunction to prove the result.

Example 2.3. If a, b, c are positive real numbers such that a + b + c = 1, prove


that
8
8abc ≤ (1 − a)(1 − b)(1 − c) ≤ .
27
Solution: Since a, b, c are positive and a + b + c = 1, none of these can exceed
1. Thus each of 1 − a, 1 − b, 1 − c is positive. Using the AM-GM inequality, we
obtain  3
3−a−b−c 8
(1 − a)(1 − b)(1 − c) ≤ = ,
3 27
which is the second part of the required inequality. On the other hand, we also
see that

(1 − a)(1 − b)(1 − c) = 1 − (a + b + c) + (ab + bc + ca) − abc


= ab + bc + ca − abc.

Thus the first part of the inequality is equivalent to

9abc ≤ ab + bc + ca.

We may write this last inequality in the form


1 1 1
+ + ≥ 9.
a b c
This is an immediate consequence of the AM-HM inequality, since a+b+c = 1.
More generally, it is not hard to see that for any n positive real numbers
a1 , a2 , . . . , an , the inequality
  
1 1 1
a1 + a2 + · · · + an + + ··· + ≥ n2 ,
a1 a2 an

holds.

Example 2.4. Let a, b, c be positive real numbers, and let p, q, r be three numbers
in the interval [0, 1/2]. Suppose a + b + c = p + q + r = 1. Prove that

pa + qb + rc
abc ≤ .
8
Techniques 55

Solution: We make use of the generalised AM-GM inequality: if α, β, γ are


non-negative real numbers such that α + β + γ = 1, then for positive u, v, w,
we have
αu + βv + γw ≥ uα v β wγ .
Thus we obtain
1 1 1  1 p  1  q  1 r
p· +q· +r· ≥
bc ca ab bc ca ab
and  p  q  r
bc ca ab ≤ p · bc + q · ca + r · ab.
These two lead to
abc 1
=
pa + qb + rc 1 1 1
p· +q· +r·
bc ca ab
1
≤  1 p  1  q  1 r
bc ca ab
 p  q  r
= bc ca ab
≤ p · bc + q · ca + r · ab.
Assume a ≤ b ≤ c. Then
p · bc + q · ca + r · ab ≤ pbc + (q + r)ac
= pbc + (1 − p)ac.
Since 0 ≤ p ≤ 1/2, we get
1  1 
− p ac ≤ − p bc.
2 2
This implies that
1 1 1  1 
pbc + (1 − p)ac = bc + ac − − p bc + − p ac
2 2 2 2
1 1
≤ bc + ac
2 2
1
= c(a + b).
2
But we know that  a + b + c 2 1
(a + b)c ≤ = .
2 4
Combining all these, we end up with
abc 1 1 1
≤ · = .
pa + qb + rc 2 4 8
We also observe that a = b = 1/4, c = 1/2, p = q = 1/2, r = 0 gives equality.
56 Inequalities

Example 2.5. For n ≥ 2, and real numbers 0 < x1 < x2 < · · · < xn ≤ 1, prove
that n
n k=1 xk 
n
1
n ≥ .
k=1 xk + nx1 x2 · · · xn 1 + xk
k=1

Solution: If ak > 0 for 1 ≤ k ≤ n, then we know that


n  
n 
ak a−1
k ≥ n2 ,
k=1 k=1

as an application of the AM-HM inequality. This implies that


n  n
  
n  
n  
n
ak a−1
k 1 − ak ) = ak a−1
k −n ak
k=1 k=1 k=1 k=1 k=1
 
n 
≥ n n− ak
k=1

n
 
= n 1 − ak .
k=1

Taking
xk
ak = , for 1 ≤ k ≤ n,
1 + xk
we obtain  

n 
n
1 
n
1
x−1
k n− ≥n .
1 + xk 1 + xk
k=1 k=1 k=1
Multiplying by x1 x2 · · · xn and using the inequality

n 
n
xk ≥ x1 x2 · · · xn x−1
k ,>
k=1 k=1

we get

n  
n
1   1 n
xk n− ≥ nx1 x2 · · · xn .
1 + xk 1 + xk
k=1 k=1 k=1
This implies that

n 
n 
n
1
n xk ≥ xk + nx1 x2 · · · xn .
1 + xk
k=1 k=1 k=1

This is equivalent to the given inequality.

Example 2.6. Let a1 , a2 , · · · , an be positive real numbers. Prove that


a21 a22 a2n 1 
+ + ··· + ≥ a1 + a2 + · · · + an .
a1 + a2 a2 + a3 an + a1 2
Techniques 57

Solution: We put

aj aj
αj = and bj = ,
a1 + a2 + · · · + an aj + aj+1
n
where an+1 = a1 . Then j=1 αj = 1. Now the weighted AM-HM inequality
gives

n 
n
α j bj αj
j=1 j=1
≥ .
n n
αj
αj
j=1 j=1
bj

We observe that
      
n n
αj
n
  n
aj = aj + aj+1 =2 aj ,
j=1 j=1
bj j=1 j=1


n
αj n n
so that = 2. Since j=1 αj = 1, we get j=1 αj bj ≥ 1/2. Thus we
j=1
bj
obtain
      
n
a2j n n
1
n
= aj α j bj ≥ aj ,
j=1
aj + aj+1 j=1 j=1
2 j=1

which is the desired inequality.

Example 2.7. If a, b, c are positive real numbers, show that


 √ 
abc a + b + c + a2 + b2 + c2 √
3+ 3
   ≤ .
a2 + b2 + c2 ab + bc + ca 9

When does equality hold?

Solution: The Cauchy-Schwarz inequality applied to (a, b, c) and (1, 1, 1)


gives
√ 
a+b+c≤ 3 a 2 + b2 + c 2 .

On the other hand, the AM-GM inequality gives

a2 + b2 + c2 ≥ 3(abc)2/3 , ab + bc + ca ≥ 3(abc)2/3 .
58 Inequalities

Combining these two, we have


 √ 
 √  1
abc a + b + c + a2 + b2 + c2 abc 1 + 3 a2 + b2 + c2 2
   ≤   
a2 + b2 + c2 ab + bc + ca a2 + b2 + c2 ab + bc + ca
 √ 
abc 1 + 3
=   1
ab + bc + ca a2 + b2 + c2 2
 √ 
abc 1 + 3
≤ √  1  2
3 3 abc 3 abc 3

3+ 3
= .
9
Equality holds if and only if a = b = c.

Example 2.8. (IMO, 1995) Let a, b, c be positive reals such that abc = 1. Prove
that
1 1 1 3
+ + ≥ .
a3 (b + c) b3 (c + a) c3 (a + b) 2

Solution: Taking x = 1/a, y = 1/b, z = 1/c, we have xyz = 1 and the


inequality to be proved is

x2 y2 z2 3
+ + ≥ .
y+z z+x x+y 2
Using the Cauchy-Schwarz inequality, we have
 2 
2 x y2 z2
(x + y + z) ≤ 2 + + x + y + z).
y+z z+x x+y
Thus
x2 y2 z2 x+y+z
+ + ≥ .
y+z z+x x+y 2
But the AM-GM inequality gives

x + y + z ≥ 3(xyz)1/3 = 3.

It follows that
x2 y2 z2 3
+ + ≥ .
y+z z+x x+y 2

We prove, the following more general assertion:


Let a, b, c be positive reals such that abc = 1. Then
1 1 1 3
+ + ≥ , (2.4)
ax (b + c) bx (c + a) cx (a + b) 2
Techniques 59

holds if x ≥ 2 or x ≤ −1. For each x, −1 < x < 2, we can find positive real
numbers a, b, c such that abc = 1 and (2.4)is not true.
For any real x, define
1 1 1
f (x; a, b, c) = + + .
ax (b + c) bx (c + a) cx (a + b)
We show that f (x; a, b, c) is a non-decreasing function for x ≥ 1. Let x > y ≥ 1
be two reals. Suppose a ≤ b ≤ c. Then ax−y ≤ bx−y ≤ cx−y is true. We also
observe that ay (b + c) ≤ by (c + a) ≤ cy (a + b). In fact, ay (b + c) ≤ by (c + a) is
equivalent to ab(ay−1 − by−1 ) ≤ c(by − ay ) which is true because y − 1 ≥ 0 and
a ≤ b. Similarly, by (c + a) ≤ cy (a + b) holds. Using Chebyshev’s inequality, we
obtain
1 1 1
+ +
ax (b + c) bx (c + a) cx (a + b)
1 1 1
= + +
ax−y ay (b + c) bx−y by (c + a) cx−y cy (a + b)
  
1 1 1 1 1 1 1
≥ + + + +
3 ax−y bx−y cx−y ay (b + c) by (c + a) cy (a + b)
1 1 1
≥ y
+ y + y .
a (b + c) b (c + a) c (a + b)
We have used the AM-GM inequality in the last leg. We get f (x; a, b, c) ≥
f (y; a, b, c) if x > y ≥ 1. Thus f (x; a, b, c) is a non-decreasing function on the
interval [1, ∞).
Now consider f (2; a, b, c). We write
1 1 1
f (2; a, b, c) = + 2 + 2
a2 (b
+ c) b (c + a) c (a + b)
bc ca ab
= + +
ab + ac bc + ab bc + ca
 
1 1 1
= (ab + bc + ca) + + −3
ab + bc bc + ca ca + ab
9 3
≥ −3≥ .
2 2
It follows that f (x; a, b, c) ≥ 3/2 for x ≥ 2.
We now show that the result is not true for 1/2 ≤ x < 2. Take x = 2 − e
where e > 0. Put a = d2 , b = 1/d, c = 1/d so that abc = 1. Then

1 2d3−e
f (2 − e; a, b, c) = +
2d3−2e d3 + 1
and f (2 − e; a, b, c) → 0 as d → ∞, if e < 3/2. If e = 3/2 then f (2 − e; a, b, c) →
1/2 < 3/2. In any case (2.4) is not true. Thus (2.4) fails for 1/2 ≤ x < 2.
60 Inequalities

For x in the interval (−∞, 1/2), we replace a, b, c by 1/p, 1/q, 1/r respec-
tively. Then f (x; a, b, c) = f (1−x; p, q, r). But 1−x lies in the interval (1/2, ∞).
It follows that (2.4) holds for x ≤ −1 and for x in the interval (−1, 1/2) there
are a, b, c such that abc = 1 and (2.4) fails.

Example 2.9. Let n ≥ 3 be an integer and let x1 , x2 , . . . , xn be positive numbers


such that
n
1
= 1.
j=1
1 + xj

Prove that

n
√ n
1
xj ≥ (n − 1) √ .
j=1 j=1
xj

Solution: We may take x1 ≤ x2 ≤ · · · ≤ xn . We claim that under this


condition we have the inequality:
√ √
xj xk
≥ ,
1 + xj 1 + xk
whenever 1 ≤ j ≤ k ≤ n. In fact, we have
1 1 2 + xj + xk
1> + = ,
1 + xj 1 + xk (1 + xj )(1 + xk )
from the given condition. This gives

1 + xj + xk + xj xk > 2 + xj + xk .

Hence xj xk > 1. But we also have


√ √ √ √  √ 
xj xk xj − xk 1 − xj xk
− = ≥ 0,
1 + xj 1 + xk (1 + xj )(1 + xk )
√ √ √
since xj ≤ xk and 1 ≤ xj xk . Thus we obtain
√ √ √
x1 x2 xn
≥ ≥ ··· ≥ .
1 + x1 1 + x2 1 + xn
Since we also know that
1 1 1
√ ≥ √ ≥ ··· ≥ √ ,
x1 x2 xn

by the assumed ordering of xj ’s, we may apply Chebyshev’s inequality. We get



n  
n √  
n
1 1 xj 1
√ ≤ =1 (2.5)
n j=1
xj j=1
1 + xj j=1
1 + xj
Techniques 61

However using the AM-HM inequality, we obtain


 n  n √ 
1 + xj xj
√ ≥ n2 .
j=1
x j j=1
1 + x j

We may write this in the form


 n √  n  n √  n 
xj √ xj 1
xj + √ ≥ n2 .
j=1
1 + x j j=1 j=1
1 + x j j=1
x j

Now using (2.5), we obtain



n √  
n 
2
xj √
n ≤n+ xj .
j=1
1 + xj j=1

This reduces to

n √  
n 
xj √
xj ≥ n2 − n = n(n − 1). (2.6)
j=1
1 + xj j=1

However (2.5) may also written in the form

n √
xj n
≤ n .
1 + x j  1
j=1

j=1
xj

Using this in (2.6), we get



n  / n

√ 1
n(n − 1) ≤ xj n √ .
j=1 j=1
xj

This gives

n
√ n
1
xj ≥ (n − 1) √ .
j=1 j=1
xj

Alternate Solution:
Using

n
1
= 1,
j=1
1 + xj

we easily obtain

n
xj
= n − 1.
j=1
1 + xj
62 Inequalities

Hence

n
√ n
1
xj − (n − 1) √
j=1 j=1
xj
n   n  n  n 
1 √ xj 1
= xj − √
j=1
1 + xj j=1 j=1
1 + xj j=1
xj

n 
n
xj − xk
= √
j=1 k=1
(1 + xk ) xj
    
 √x j √x k − 1 √ x j − √x k 2 √x j + √x k
= √ √ .
xj xk (1 + xj )(1 + xk )
j>k

Here we have used


√ √ √ √ 
1 1 xj xk − 1 xj − xk
√ − √ = √ √ .
(1 + xk ) xj (1 + xj ) xk xj xk (1 + xj )(1 + xk )
Thus it is sufficient to prove that xj xk ≥ 1 for any pair j, k such that j = k.
But this follows from the given condition:
1 1 2 + xj + xk
1≥ + = ,
1 + xj 1 + xk 1 + xj + xk + xj xk
which implies that xj xk ≥ 1.

Example 2.10. (Hardy’s inequality) Let a1 , a2 , . . . , aN be N positive real


numbers and let p > 1. Put
a1 + a2 + · · · + ak
αk = , for 1 ≤ k ≤ N.
k
Prove that

N  p p 
N
αkp ≤ apk .
p−1
k=1 k=1

Solution: Let Ak = a1 + a2 + · · · + ak , We have


p p  
αnp − αnp−1 an = αnp − αnp−1 An − An−1
p−1 p−1
p  
= αn −p
αnp−1 nαn − (n − 1)αn−1
p−1
 np  (n − 1)p p−1
= αnp 1 − + α αn−1 .
p−1 p−1 n
Using the generalised AM-GM inequality, we have
(n − 1)p p−1 n − 10 p
1
αn αn−1 ≤ (p − 1)αnp + αn−1 .
p−1 p−1
Techniques 63

Thus we get
p 0 np 1 n−1
αnp − αp−1 an ≤ αnp 1 − + (n − 1) + αp
p−1 n p−1 p − 1 n−1
1 0 p
1
= (n − 1)αn−1 − nαnp .
p−1
Taking α0 = 0 and summing over n from 1 to N , we get

N
p  p−1
N p
N αN
αnp − αn an ≤ − ≤ 0.
n=1
p − 1 n=1 p−1

Thus we obtain,

N
p  p−1
N
αnp ≤ α an
n=1
p − 1 n=1 n

p   p 1/p   (p−1)q 1/q


N N
≤ a αn ,
p − 1 n=1 n n=1

where q is the conjugate index of p; we have used Hölder’s inequality. Hence


(p − 1)q = p and we obtain
 p   p 1/p   p 1/q
N N N
αnp ≤ a αn .
n=1
p − 1 n=1 n n=1

This simplifies to

N  p p 
N
αkp ≤ apk .
p−1
k=1 k=1

Example 2.11. Let a1 , a2 , . . . , an be n positive real numbers. Prove that


 *
" n " n  nj=1 aj n
a
aj j ≥ aj .
j=1 j=1

Solution: Consider the function f (x) = x ln x. This is convex on (0, ∞). In


fact f  (x) = 1/x which is positive. Using Jensen’s inequality, we obtain
n  n   n 
1 1 1
aj ln aj ≥ aj ln aj .
j=1
n j=1
n j=1
n

However, the AM-GM inequality gives


" 1/n
1
n n
aj ≥ aj .
n j=1 j=1
64 Inequalities

Using the monotonicity of ln x, we obtain


  " 1/n
1
n n n
1
aj ln aj ≥ aj ln aj .
n j=1 j=1
n j=1

This implies that

   " 
1
n n n
a
ln aj j ≥ aj ln aj .
j=1
n j=1 j=1

Using the fact that the logarithm takes multiplication to addition, and using
the monotonicity of exponential function, we obtain the desired inequality.

Example 2.12. Let a1 , a2 , . . . , an be n positive real numbers. Show that for any
permutation a1 , a2 , . . . , an of the sequence a1 , a2 , . . . , an , the inequality

"
n
a
"
n
a
aj j ≥ aj j ,
j=1 j=1

holds.

Solution: We may assume that a1 ≥ a2 ≥ · · · ≥ an . Then we have

ln a1 ≥ ln a2 ≥ · · · ≥ ln an .

We apply the rearrangement inequality to the sequences a1 , a2 , . . . , an and


ln a1 , ln a2 , . . . , ln an . We obtain


n 
n
aj ln aj ≥ aj ln aj .
j=1 j=1

This implies

n
a

n
a
ln aj j ≥ ln aj j .
j=1 j=1

The exponentiation gives the result;

"
n
a
"
n
a
aj j ≥ aj j .
j=1 j=1
Techniques 65

2.4 Use of calculus in inequalities


Many inequalities follow directly from the monotonicity of the function involved
or by calculating its global maxima and minima. For a differentiable function,
the monotonicity can be established using the nature of its derivatives. A point
of local minima or maxima for a differentiable function is precisely that point
at which its derivative vanishes. We state these results here without giving
proofs; the proofs need the techniques of differential calculus.

Theorem 12. Let f : (a, b) → R be a differentiable function and suppose


f  (x) ≥ 0 for all x ∈ (a, b). Then f (x) is a non-decreasing function on (a, b). If
f  (x) > 0 on (a, b), then f (x) is strictly increasing on (a, b).
Similarly, f  (x) ≤ 0 for all x ∈ (a, b) implies that f (x) is non-increasing on
(a, b) and f  (x) < 0 for all x ∈ (a, b) implies that f (x) is strictly decreasing on
(a, b).

Theorem 13. Suppose f : (a, b) → R be a differentiable function and λ is


a point where a local extremum occurs for f ; i.e., there exists a neighborhood
(λ − δ, λ + δ) of λ such that either f (x) ≤ f (λ) for all x ∈ (λ − δ, λ + δ) or
f (x) ≥ f (λ) for all x ∈ (λ − δ, λ + δ). Then f  (λ) = 0. Suppose further f is
twice differentiable and f  (λ) = 0. If f  (λ) < 0, f (λ) is a local maximum; and
if f  (λ) > 0, then f (λ) is a local minimum. If f  (λ) = 0, then f (λ) is neither a
maximum nor a minimum(a point of inflexion).

For proofs of these statements, refer to any book on Calculus. We consider


examples which make use of these theorems in their solutions.

Example 2.13. Show that for any x > 0, the inequality



≥ 0, if α > 1 or α < 0,
xα − αx + α − 1 (2.7)
≤ 0, if 0 < α < 1.

Solution: We consider the function f (x) = xα − αx + α − 1 on the interval


(0, ∞). This is a differentiable function and
 
f  (x) = αxα−1 − α = α xα−1 − 1 .
This
 shows
 α−2that f  (x) = 0 only if x = 1 (Note that α = 1). Moreover f  (x) =
α α−1 x which is positive at x = 1 if either α < 0 or α > 1, and negative
at x = 1 if 0 < α < 1. Hence x = 1 is a unique point of minimum if either α < 0
or α > 1, and a unique point of maximum if 0 < α < 1. Since f (1) = 0, we
conclude that f (x) ≥ f (1) = 0 if either α < 0 or α > 1, and f (x) ≤ f (1) = 0
if 0 < α < 1. This proves the result.
Interestingly, the inequalities (2.7) may be used to derive many of the known
inequalities. We show here how we can derive the generalised AM-GM inequal-
ity and Young’s inequality. Suppose x1 and x2 are two positive real numbers
66 Inequalities

and let x = x1 /x2 . Take any α with 0 < α < 1. Then we have
 α  
x1 x1
−α + α − 1 ≤ 0.
x2 x2
It may be rewritten as
1−α
 

1 x2 ≤ αx1 + 1 − α x2 .

This proves the generalised AM-GM inequality for non-negative real num-
bers x1 , x2 and weights α, 1 − α. Now we use the principle of induction to
prove the general case. Assume that for any n − 1 non-negative real numbers
n−1
y1 ,y2 ,y3 ,. . .,yn−1 and positive weights λ1 ,λ2 ,λ3 ,. . .,λn−1 with j=1 λj = 1, we
have
" λ
n−1 
n−1
yj j ≤ λj yj .
j=1 j=1

Take any n non-negative numbers


n x1 ,x2 ,x3 ,. . .,xn , and positive real numbers
α1 ,α2 ,α3 ,. . .,αn such that j=1 αj = 1. Put

yj = x j , λj = αj , for 1 ≤ j ≤ n − 2,
αn−1 /λn−1 αn /λn−1
yn−1 = xn−1 xn , λn−1 = αn−1 + αn .

We observe that λj > 0 for 1 ≤ j ≤ n − 1 and


n−1 
n
λj = αj = 1.
j=1 j=1

Using the induction hypothesis, we obtain

"
n
α
"
n−1
λ
xj j = yj j
j=1 j=1


n−1
≤ λj yj
j=1


n−2
  αn−1 /λn−1 αn /λn−1 
= αj xj + αn−1 + αn xn−1 xn
j=1

  
n−2
  αn−1 αn
≤ αj xj + αn−1 + αn xn−1 + xn
j=1
λn−1 λn−1

n
= α j xj ,
j=1

since λn−1 = αn−1 + αn . Equality holds if and only if all xj ’s are equal.
Techniques 67

We also establish Young’s inequality which is key to Hölder’s inequality.


1 1
Suppose a and b are positive reals and p, q be a nonzero reals such that + =
p q
 1/p 1  a  1
1. We take α = 1/p and x = a/b. We get ab − p b + p − 1 ≥ 0 if p < 1
and ≤ 0 if p > 1. This reduces to
a b
a1/p b1/q ≥ + if p < 1,
p q
a b
a1/p b1/q ≤ + if p > 1.
p q
Now replacing a by ap and b by bq , we get
ap bq
ab ≥ + if p < 1,
p q
ap bq
ab ≤ + if p > 1,
p q
which is Young’s inequality.

Example 2.14. Show that ex > 1+x+(x2 /2) for x > 0, and ex < 1+x+(x2 /2)
for x < 0.

Solution: Let f (x) = ex − 1 − x − (x2 /2). We have


f  (x) = ex − 1 − x, f  (x) = ex − 1.
Hence f  (x) > 0 for x > 0 and f  (x) < 0 for x < 0. This implies that
f  (x) is increasing on (0, ∞) and decreasing on (−∞, 0). We thus infer that
f  (x) > f  (0) = 0 for all real x = 0. It follows that f (x) is increasing on
(−∞, 0) and (0, ∞). This implies that f (x) < f (0) for x < 0, and f (x) > f (0)
for x > 0. Since f (0) = 0, it follows that f (x) > 0 for x > 0, and f (x) < 0 for
x < 0, which is the desired inequality.

Example 2.15. Prove Jordan’s inequality that


2 sin x
≤ ≤ 1,
π x
for all x ∈ [0, π/2].

Solution: Consider f (x) = sin x. Since f  (x) = cos x and f  (x) = − sin x,
we see that f  (x) ≤ 0 on [0, π/2]. Hence f (x) is a concave function on [0, π/2].
Take any x ∈ [0, π/2] and set λ = 2x/π. Using the concavity of f (x) = sin x,
we have
   
sin x = sin (1 − λ)0 + λπ/2
   
≥ (1 − λ) sin 0 + λ sin π/2
2x
= λ= .
π
68 Inequalities

Thus we obtain
sin x 2
≥ .
x π
On the other hand, consider g(x) = sin x − x on [0, π/2]. We again note that
g  (x) = cos x − 1 ≤ 0 for all x ∈ [0, π/2]. Hence g is a non-increasing function
on [0, π/2]. This implies that g(x) ≤ g(0) = 0. Thus we get, sin x ≤ x for all
x ∈ [0, π/2]. Using the known result that limx→0 sin x/x = 1, we conclude that
sin x/x ≤ 1 for all x ∈ [0, π/2]. Thus we have Jordan’s inequality,

2 sin x
≤ ≤ 1,
π x
valid for all x ∈ [0, π/2]. Here we have to take the limiting value for x = 0.

Example 2.16. Prove Bernoulli’s inequality that for x > −1,

(1 + x)α > 1 + αx, if α > 1 or α < 0,


(1 + x)α < 1 + αx, if 0 < α < 1.

Solution: Consider the function f (x) = (1 + x)α − 1 − αx for x > −1 and


α = 1. Then we see that
 
f  (x) = α (1 + x)α−1 − 1 .

Suppose α > 1. Since x > −1, we see that (1 + x)α−1 > 1 if x > 0, and
(1 + x)α−1 < 1 for −1 < x < 0. Thus f  (x) > 0 if x > 0, and f  (x) < 0 if
−1 < x < 0. This shows that f (x) is decreasing on (−1, 0), and increasing on
(0, ∞). Hence f (x) > f (0) for all x > −1, x = 0. We obtain

(1 + x)α > 1 + αx if x > −1, x = 0, α > 1.

If α < 0, then 1 − α > 1, and hence (1 + x)α−1 − 1 < 0 for 1 + x > 1; and
(1 + x)α−1 − 1 > 0 for 0 < 1 + x < 1. Thus f  (x) > 0 for 1 + x > 1, and
f  (x) < 0 for 0 < 1 + x < 1. Hence f (x) is decreasing in (−1, 0), and increasing
on (0, ∞). We see that f (x) > 0 for x > −1, and x = 0 in this case. This gives

(1 + x)α > 1 + αx, if x > −1, x = 0, α < 0.

Consider the case in which 0 < α < 1. Then α − 1 < 0. Hence (1 + x)α−1 −
1 < 0 for x > 0, and (1 + x)α−1 − 1 > 0 for −1 < x < 1. Thus, it follows
that f  (x) > 0 on (−1, 0), and f  (x) < 0 on (0, ∞). We deduce that f (x) is
increasing on (−1, 0), and decreasing on (0, ∞). This gives f (x) < f (0), and
hence
(1 + x)α < 1 + αx, if x > −1, x = 0, 0 < α < 1.
Techniques 69

Example 2.17. Let a1 ,a2 ,a3 ,. . .,an be n > 1 distinct real numbers. Prove that
⎛ ⎞
  2
 2 12
n
1
n
min aj − ak ≤ ⎝ a2 − aj ⎠ .
1≤j<k≤n n(n2 − 1) j=1 j n j=1

Solution: We first prove that


⎛ ⎞
 2 12 
n
min aj − ak ≤ ⎝ a2 ⎠ , (2.8)
1≤j<k≤n n(n2 − 1) j=1 j
n
for any n real numbers a1 ,a2 ,a3 ,. . .,an . We may assume j=1 a2j = 1. We may
also enumerate aj ’s such that
a = a1 ≤a2 ≤a3 ≤· · ·≤an .
Put
12
μ2 = .
n(n2 − 1)
Suppose the inequality is not true. Then aj+1 − aj > μ for 1 ≤ j ≤ n − 1. Thus
aj > a + (j − 1)μ,
for 2 ≤ j ≤ n. Hence

n 
n
 2
1= a2j > a + (j − 1)μ
j=1 j=1


n−1
 2
= a + jμ
j=0


n−1 
n−1
= na2 + 2aμ j + μ2 j2
j=1 j=1
n(n − 1)(2n − 1)
= na2 + aμn(n − 1) + μ2
6
 2
μ(n − 1)
= n a+ +1
2
> 1.
This contradiction proves the inequality (2.8).
We take bj = aj + t where t is a real variable and apply the inequality (2.8)
to the sequence bj to get
⎛ ⎞
 2 12 n
  2
min aj − ak ≤ ⎝ aj + t ⎠ ,
1≤j<k≤n n(n2 − 1) j=1
70 Inequalities
n  2
valid for all t. Consider the function f (t) = j=1 aj + t . Expanding this,
we obtain  
n n
f (t) = a2j + 2 aj t + nt2 ,
j=1 j=1

which is a quadratic polynomial


  in t. This quadratic polynomial attains its
extremum at t0 , where f  t0 = 0. Solving this, we see that

n 
1
t0 = − aj .
n j=1


Since
  f (t 0 ) = 2n > 0, we infer that f has a unique minimum at t0 . Hence
f t ≥ f t0 , for all t. But we see that

  2  2
  n
2
n
1
n
f t0 = a2j − aj + aj
j=1
n j=1
n j=1
n n 2
1
= a2j − aj .
j=1
n j=1

This implies that


 2 12  
min aj − ak ≤ 2
f t0
1≤j<k≤n n(n − 1)
⎛ ⎞
n 
n 2
12 ⎝ 1
= a2 − aj ⎠ .
n(n2 − 1) j=1 j n j=1

Example 2.18. Let a, b, x, y be real numbers such that ay − bx = 1. Prove that



a2 + b2 + x2 + y 2 + ax + by ≥ 3.

Solution: We can find u, v, θ, φ such that a = u cos θ, b = u sin θ, x = v cos φ


and y = v sin φ. Using ay − bx = 1, we infer that uv sin(θ − φ) = 1. Hence
uv ≥ 1 (remember u and v are positive). We obtain

1  *
cos2 (θ − φ) = 1 − sin2 (θ − φ) = 1 − = u2 v 2 − 1 u2 v 2 .
u2 v 2
We observe that u2 = a2 + b2 , v 2 = x2 + y 2 and ax + by = uv cos(θ − φ). Using
u2 + v 2 ≥ 2uv, we obtain

a2 + b2 + x2 + y 2 + ax + by ≥ 2uv − u2 v 2 − 1.
Techniques 71

Thus, it is sufficient to prove that


 √
2uv − u2 v 2 − 1 ≥ 3.

Consider the function f (t) = 2t − t2 − 1 for t ≥ 1. We have
t
f  (t) = 2 − √ ,
t2 −1
  √
which gives f  t0 = 0 on [1, ∞) if and only if t0 = 2/ 3. We see that
1
f  (t) = √ 3 .
t2 −1
  √
Hence f  t0 = 3 3 > 0. This shows that f has a unique minimum at

t0 = 2/ 3 on [1, ∞). Thus we have
  √
f (t) ≥ f t0 = 3.

It follows that  √
2uv − u2 v 2 − 1 ≥ 3,
which is what we have to prove.
Another important technique for proving inequalities is to use the fact that
between any two real zeros of a differentiable function, there is always a zero
of its derivative. This follows from Rolle’s theorem in calculus. We have
already used this idea while proving Newton’s inequality in an earlier chapter.
Existence of real zeros for the derivative puts certain restrictions on the function
and this would lead to inequalities.

Example 2.19. Prove that in a triangle with sides a, b, c, circum-radius R and


in-radius r, the inequality
   2
9r 4R + r ≤ 3s2 ≤ 4R + r ,

holds, where s = (a + b + c)/2 is the semi-perimeter.

Solution: Consider the monic polynomial whose roots are a, b, c. We have


 
s2 + r 4R + r = s2 + 4Rr + r2
abc Δ Δ2
= s2 + · + 2
Δ s s
2
abc Δ
= s2 + + 2
s s
s3 + abc + (s − a)(s − b)(s − c)
=
s
= ab + bc + ca,
72 Inequalities

and
abc
4sRr = · Δ = abc.
Δ
Here Δ denotes the area of the given triangle. Observe that we have used the
well known Heron’s formula: Δ2 = s(s − a)(s − b)(s − c). Thus a, b, c are the
roots of the equation
  
p(x) = x3 − 2sx2 + s2 + r 4R + r x − 4sRr = 0.

Now p(x) = 0 has three real roots. Hence Rolle’s theorem shows that p (x) = 0
has two real roots. But
 
p (x) = 3x2 − 4sx + s2 + r 4R + r .

This is a quadratic equation and it has two real roots if and only if its discrim-
inant is non-negative. Computing the discriminant, we obtain
  
16s2 − 12 s2 + r 4R + r ≥ 0.

This leads to  
9r 4R + r ≤ 3s2 .
We also know that
Δ Δ
ra = , r= .
s−a s
Hence
s(ra − r)
a= .
ra
Similarly, we can get

s(rb − r) s(rc − r)
b= , c= .
rb rc
Here ra , rb , rc are the ex-radii of the triangle ABC.
Putting x = s(y − r)/y in p(x) = 0, we obtain an equation in y;
 
q(y) = y 3 − 4R + r y 2 + s2 y − s2 r = 0.

Observe that ra , rb , rc are the roots of q(y) = 0. Hence q  (y) = 0 also has real
roots. But  
q  (y) = 3y 2 − 2 4R + r + s2 ,
 2
and its discriminant is 4 4R + r − 12s2 . We thus obtain the inequality
 2
3s2 ≤ 4R + r .
Techniques 73

2.5 Trigonometric substitutions


Some of the inequalities may be proved using suitable trigonometric substitu-
tions. We have already encountered one such use in Example 2.18. We shall
study a few more such examples here.

Example 2.20. Let a, b, c be real numbers such that 0 ≤ a, b, c ≤ 1. Prove that


   √
a(1 − b)(1 − c) + b(1 − c)(1 − a) + c(1 − a)(1 − b) ≤ 1 + abc.

Solution: Since a, b, c are in [0, 1], we can find x, y, z in the interval [0, π/2]
such that
a = sin2 x, b = sin2 y, c = sin2 z.
Thus the inequality to be proved gets transformed to

sin x cos y cos z + sin y cos z cos x + sin z cos x cos y ≤ 1 + sin x sin y sin z.

This is equivalent to
   
sin x cos y cos z − sin y sin z + cos x sin y cos z + cos y sin z ≤ 1.

On simplification, this reduces to sin(x + y + z) ≤ 1, which follows from the


property of the sine function.

Example 2.21. Let x, y, z be positive real numbers such that xy + yz + zx = 1.


Prove that
     
2x 1 − x2 2y 1 − y 2 2z 1 − z 2 x y z
 2 +  2 +  2 ≤ 2
+ 2
+ .
1+x 2 1+y 2 1+z 2 1+x 1+y 1 + z2

Solution: Consider the first term on the left side:


 
2x 1 − x2 2x 1 − x2
 2 = · .
1 + x2 1 + x 1 + x2
2

We know that
2 tan(α/2) 1 − tan2 (α/2)
= sin α, = cos α.
1 + tan2 (α/2) 1 + tan2 (α/2)
Hence, this suggests the substitutions

x = tan(α/2), y = tan(β/2), z = tan(γ/2).

We recall that the function x → tan x is a one-one function of (0, π/2) on to


(0, ∞). The inequality now transforms to
1 
sin α cos α + sin β cos β + sin γ cos γ ≤ sin α + sin β + sin γ .
2
74 Inequalities

This is equivalent to

sin 2α + sin 2β + sin 2γ ≤ sin α + sin β + sin γ.

We observe that
1 − xy 1 − tan(α/2) tan(β/2)
z= =
x+y tan(α/2) + tan(β/2)
π α β 
= tan − − .
2 2 2
This shows that π
α β
tan(γ/2) = tan − . −
2 2 2
π α β
Since α and β are in (0, π/2), so is − − . Since x → tan x is one-one on
2 2 2
(0, π/2), we conclude that

γ π α β
= − − .
2 2 2 2
This gives α + β + γ = π. Now f (x) = − sin x is convex on (0, π). Suppose
α ≤ β ≤ γ and let α , β  , γ  be a permutation of α, β, γ. Using Theorem 10 on
page 41, we obtain
           
− sin α + α − sin β  + β − sin γ  + γ ≤ − sin 2α − sin 2β − sin 2γ .

Taking α = β, β  = γ and γ  = α, we obtain


           
sin 2α + sin 2β + sin 2γ ≤ sin α + β + sin β + γ + sin γ + α .

This reduces to
     
sin 2α + sin 2β + sin 2γ ≤ sin α + sin β + sin γ,

proving the inequality needed.

Example 2.22. Show that for any three real numbers a, b, c, the inequality
 2    
ab + bc + ca − 1 ≤ a2 + 1 b2 + 1 c2 + 1 ,

holds.

Solution: Put a = tan α, b = tan β, c = tan γ, where α, β, γ are in the


interval (−π/2, π/2). Then

a2 + 1 = sec2 α, b2 + 1 = sec2 β, c2 + 1 = sec2 γ,


Techniques 75

and the inequality to be proved is


 2
(ab + bc + ca − 1) cos α cos β cos γ ≤ 1.

However, we observe that

(ab + bc) cos α cos β cos γ = sin α sin β cos γ + cos α sin β sin γ
= sin β sin(α + γ),
(ca − 1) cos α cos β cos γ = sin α cos β sin γ − cos α cos β cos γ
= − cos β cos(α + γ).

Thus, we obtain
 2
(ab + bc + ca − 1) cos α cos β cos γ
 2
= sin β sin(α + γ) − cos β cos(α + γ)
 
= cos2 α + β + γ ≤ 1.

Example 2.23. For any three positive real numbers x, y, z such that x + y + z =
xyz, prove the inequality

x y z 3 3
√ + +√ ≤ .
1 + x2 1 + y2 1 + z2 2

Solution: Since x + y + z = xyz, there is a triangle with angles α, β, γ such


that x = tan α, y = tan β, z = tan γ. Then the inequality reduces to

3 3
sin α + sin β + sin γ ≤ .
2
This follows from 3.4.5 on page 118.

2.6 Properties of quadratic polynomials


The fact that a quadratic equation ax2 +bx+c = 0 has real roots if and only if its
discriminant is non-negative helps in establishing many inequalities. The classic
case is the Cauchy-Schwarz inequality. Take two sequences a1 ,a2 ,a3 ,. . .,an and
b1 ,b2 ,b3 ,. . .,bn of real numbers. We associate the quadratic polynomial with
these sequences by

n
 2
f (t) = a j + bj t = A + 2Ct + Bt2 ,
j=1
76 Inequalities

where

n 
n 
n
A= a2j , B= b2j , C= a j bj .
j=1 j=1 j=1

We may assume that B = 0, for, otherwise bj = 0 for all j forcing C = 0, and


in this case the inequality C 2 ≤ AB is trivially true. Thus B > 0. Since f (t)
is the sum of several squares, we have f (t) ≥ 0 for all t. Hence its discriminant
is non-positive. This implies that C 2 ≤ AB and gives the Cauchy-Schwarz
inequality. We also observe that equality holds if and only if f (t) = 0 has two
coincident real roots. This is equivalent to aj + λbj = 0, for 1 ≤ j ≤ n, where
n
a j bj
λ = − n
j=1
2 ,
j=1 bj

is the double root of f (t) = 0.

Example 2.24. Let (a1 , a2 ), (b1 , b2 ) and (c1 , c2 ) be three pairs of real numbers.
Prove that
   2      
a 1 b2 + a 2 b1 − 2 a1 a2 ≤4 a21 − a 1 b1 a22 − a 2 b2 ,

where the sum is cyclically over a, b, c.

Solution: We know that for all real x, y, z, the following inequality holds:

x2 + y 2 + z 2 ≥ xy + yz + zx.

In fact this is equivalent to (x − y)2 + (y − z)2 + (z − x)2 ≥ 0. Moreover equality


holds if and only if x = y = z. Now consider the quadratic polynomial
  2    
P (t) = a1 + ta2 − a1 + ta2 b1 + tb2 .
cyclic cyclic

Using x2 + y 2 + z 2 ≥ xy + yz + zx, we observe that P (t) ≥ 0 for all real t.


Hence the discriminant of P (t) must be non-positive. We can also write P (t)
in the form
   
P (t) = a22 − a 2 b2 t2
cyclic cyclic
    
+ 2 a1 a2 − a 1 b2 − a 2 b1 t
cyclic cyclic cyclic
   
+ a21 − a 1 b1 .
cyclic cyclic
Techniques 77

Thus we obtain
    2
a 1 b2 + a 2 b1 − 2 a1 a2
cyclic cyclic cyclic
      
≤4 a21 − a 1 b1 a22 − a 2 b2 .
cyclic cyclic cyclic cyclic

Here equality holds if and only if


a1 + ta2 = b1 + tb2 = c1 + tc2 ,
where t is the coincident root
  
2 cyclic a1 a2 − cyclic a1 b2 − cyclic a2 b1
t=−  2
 .
cyclic a2 − cyclic a2 b2

This reduces after simplification to


b1 c2 − b2 c1 + c1 a2 − c2 a1 + a1 b2 − a2 b1 = 0.

Example 2.25. Let n > 2 and x1 , x2 , . . . , xn be n real numbers. Put



n 
p= xj , q= xj xk .
j=1 1≤j<k≤n

Prove that
 
p n−1 2nq p n−1 2nq
− p2 − ≤ xj ≤ + p2 − ,
n n n−1 n n n−1
for all j.

Solution: Consider p and q:


 
p = x1 + x 2 + x 3 + · · · + x n

q = x1 x2 + x 1 x3 + · · · + x 1 xn + xj xk .
2≤j<k≤n

We see that
 2  2
p − x1 = x2 + x3 + · · · + xn
  
≤ n − 1 x22 + x23 + · · · + x2n
 0  2  1
= n−1 x 2 + x3 + · · · + xn − 2 xj xk
2≤j<k≤n
 0   2 1
= n−1 p − x1 − 2 q − x1 (p − x1 )
  
= n − 1 p2 − 2q − x21 .
78 Inequalities

Thus we get a quadratic inequality:

nx21 − 2px1 − (n − 2)p2 + 2(n − 1)q ≤ 0.

Solving this inequality, we get


 
p n−1 2
2nq p n−1 2nq
− p − ≤ x1 ≤ + p2 − .
n n n−1 n n n−1

Since we can choose any xj in place of x1 , we get the desired inequality.

2.7 A useful transformation


Sometimes, an inequality involving the sides of a triangle can be converted
to an inequality for positive real numbers and vice versa. This achieves a lot
of simplification in solutions. In a triangle with sides a, b, c, we know that
a < b + c, b < c + a and c < a + b. Hence, if we introduce

2x = b + c − a, 2y = c + a − b, 2z = a + b − c,

then we see that x, y, z are positive reals and a = y+z, b = z+x, c = x+y. This
transformation is often referred as Ravi transformation in the mathematical
literature.

Example 2.26. (CRMO, 1999) Let a, b, c be the sides of a triangle. Prove that

a b c
+ + ≥ 3.
b+c−a c+a−b a+b−c

Solution: Using the transformations described above, this reduces to


y+z z+x x+y
+ + ≥ 6.
x y z

This follows by the AM-GM inequality:


    
y+z z+x x+y x y y z z x
+ + = + + + + +
x y z y x z y x z
≥ 2 + 2 + 2 = 6.

Example 2.27. If a, b, c are the sides of a triangle, prove

(b + c − a)(c + a − b)(a + b − c) ≤ abc.


Techniques 79

Solution: Using x, y, z as described above, this reduces to the inequality

8xyz ≤ (x + y)(y + z)(z + x).

Using the AM-GM inequality, we see that


√ √ √
8xyz = 2 xy · 2 yz · 2 zx ≤ (x + y)(y + z)(z + x).

Example 2.28. (INMO-2003) Let a, b, c be the sides of a triangle ABC. Let


b c a
A B  C  be the triangle whose sides are a + , b + , c + . Prove that
2 2 2
- . 9- .
A B  C  ≥ ABC .
4

b c
Solution: It is easy to observe that there is a triangle with sides a + , b + ,
2 2
a
c + . Using Heron’s formula, we get
2

16[ABC]2 = (a + b + c)(a + b − c)(b + c − a)(c + a − b),

and
- .2 3
16 A B  C  = (a + b + c)(−a + b + 3c)(−b + c + 3a)(−c + a + 3b).
16

Since a, b, c are the sides of a triangle, there are positive real numbers x, y, z
such that a = y + z, b = z + x, c = x + y. Using these, we obtain

[ABC]2 16xyz
- .2 = .

ABC  3(2x + y)(2y + z)(2z + x)

Thus it is sufficient to prove that

(2x + y)(2y + z)(2z + x) ≥ 27xyz

for positive real numbers x, y, z. Using the AM-GM inequality, we get

2x + y ≥ 3(x2 y)1/3 , 2y + z ≥ 3(y 2 z)1/3 , 2z + x ≥ 3(z 2 x)1/3 .

Multiplying these, we obtain the desired result. We also observe that equality
holds if and only if x = y = z. This is equivalent to the statement that ABC
is equilateral.
80 Inequalities

2.8 Schur’s inequality


There is a nice inequality, due to Schur, which is often helpful to prove some
results. It asserts that:
Let a, b, c be positive real numbers and let λ be any real number. Then

aλ (a − b)(a − c) + bλ (b − c)(b − a) + cλ (c − a)(c − b) ≥ 0,

with equality if and only if a = b = c.


Proof: If any two of a, b, c are equal, the result is immediate. Hence we
may assume that no two are equal. Because of the symmetry, we may assume
a > b > c. If λ ≥ 0, then

aλ (a − b)(a − c) + bλ (b − c)(b − a) + cλ (c − a)(c − b)


0 1
= (a − b) (a − c)aλ − (b − c)bλ + (a − c)(b − c)cλ
0 1
≥ (a − b)(a − c) aλ − bλ ≥ 0.

If λ < 0, then

aλ (a − b)(a − c) + bλ (b − c)(b − a) + cλ (c − a)(c − b)


0 1
= aλ (a − b)(a − c) + (b − c) (a − c)cλ − (a − b)bλ
0 1
≥ (b − c)(a − c) cλ − bλ ≥ 0.

Example 2.29. (IMO, 2000) Let a, b, c be positive real numbers such that abc =
1. Prove that
   
1 1 1
a−1+ b−1+ c−1+ ≤ 1.
b c a

Solution: Introduce a = x/y, b = y/z, c = z/x; this is possible since a, b, c


are positive real numbers such that abc = 1. The inequality reduces to

(z + x − y)(x + y − z)(y + z − x) ≤ xyz.

Expanding the left hand side of the above inequality, we obtain

x(y − x)(x − z) + y(z − y)(y − x) + z(x − z)(z − y) + xyz.

Thus it is sufficient to prove that

x(y − x)(x − z) + y(z − y)(y − x) + z(x − z)(z − y) ≤ 0.

This follows from Schur’s inequality. (For different solutions refer to problem
3.6 on page 288.)
Techniques 81

Example 2.30. (APMO, 2004) Let x, y, z be positive real numbers. Prove that
 2     
x + 2 y 2 + 2 z 2 + 2 ≥ 9 xy + yz + zx .

Solution: Expanding, the inequality is


  
x2 y 2 z 2 + 2 x2 y 2 + 4 x2 + 8 ≥ 9 xy.
cyclic cyclic cyclic

Note that
  
2 x2 y 2 − 4 xy + 6 = 2 (xy − 1)2 ≥ 0.
cyclic cyclic cyclic
 
Moreover cyclic x2 ≥ cyclic xy. Thus it is sufficient to prove that
 
x2 y 2 z 2 + x2 + 2 ≥ 2 xy.
cyclic cyclic

If a, b, c are positive reals, Schur’s inequality gives


  
a3 + 3abc ≥ a2 b + ab2
cyclic cyclic cyclic
= ab(a + b) + bc(b + c) + ca(c + a).

Using u + v ≥ 2 uv for any positive reals u, v, we get
 
a3 + 3abc ≥ 2 (ab)3/2 .
cyclic cyclic

2/3 2/3 2/3


Taking a = x ,b=y and z = c , this reduces to
 2/3
x2 + y 2 + z 2 + 3 xyz ≥ 2(xy + yz + zx).
However, we observe that
 2/3
x2 y 2 z 2 + 2 ≥ 3 xyz .
 2/3
In fact, this is equivalent to t3 + 2 ≥ 3t, where t = xyz ; this follows from
t3 − 3t + 2 = (t − 1)2 (t + 2) ≥ 0. Thus
 2/3
x2 y 2 z 2 + x2 + y 2 + z 2 + 2 ≥ x2 + y 2 + z 2 + 3 xyz
≥ 2(xy + yz + zx).

Example 2.31. Let x, y, z be positive real numbers and define p = x + y + z,


q = xy + yz + zx and r = xyz. Then
(i)p3 − 4pq + 9r ≥ 0; (ii)p4 − 5p2 q + 4q 2 + 6pr ≥ 0.
82 Inequalities

Solution: We use Schur’s inequality that


xt (x − y)(x − z) + y t (y − x)(y − z) + z t (z − x)(z − y) ≥ 0
for any positive reals x, y, z and real number t. Taking t = 1, we get

(x3 + y 3 + z 3 ) − x2 y + 3xyz ≥ 0.
sym

Using the known identities


(x + y + z)3 = x3 + y 3 + z 3 + 3(x + y + z)(xy + yz + zx) − 3xyz,
and 
x2 y = (x + y + z)(xy + yz + zx) − 3xyz,
sym

we obtain
(x + y + z)3 − 4(x + y + z)(xy + yz + zx) + 9xyz ≥ 0.
This proves (i). A similar proof works for (ii) with t = 2 in Schur’s inequality.

Example 2.32. Let a, b, c be non-negative real numbers such that a + b + c = 2.


Prove that
a4 + b4 + c4 + abc ≥ a3 + b3 + c3 .

Solution: Using Schur’s inequality, we have


a2 (a − b)(a − c) + b2 (b − c)(b − a) + c2 (c − a)(c − b) ≥ 0.
This gives
a4 + b4 + c4 + abc(a + b + c) ≥ a3 (b + c) + b3 (c + a)c3 (a + b)
= (a3 + b3 + c3 )(a + b + c) − a4 + b4 + c4 .
This simplifies to
2(a4 + b4 + c4 ) + abc(a + b + c) ≥ (a3 + b3 + c3 )(a + b + c).
Since a + b + c = 2, we obtain
a4 + b4 + c4 + abc ≥ a3 + b3 + c3 .

Example 2.33. Suppose a, b, c are non-negative real numbers. Prove that


a3 + b3 + c3 + 4(a + b + c) + abc ≥ 8(ab + bc + ca).
Techniques 83

Solution: Suppose we prove that


4(ab + bc + ca)2
a3 + b3 + c3 + 9abc ≥ .
a+b+c
Then we see that
4(ab + bc + ca)2
a3 + b3 + c3 + 4(a + b + c) + abc ≥ 4(a + b + c) +
a+b+c
≥ 8(ab + bc + ca),

where we have used AM-GM inequality. Therefore it remains to prove


4(ab + bc + ca)2
a3 + b3 + c3 + 9abc ≥ .
a+b+c
This may be put in an equivalent form by clearing the denominators:

a4 + b4 + c4 + 9abc(a + b + c) + a3 (b + c)
cyclic
⎛ ⎞

≥ 4⎝ a2 b2 ⎠ + 8abc(a + b + c).
cyclic

Further reduction leads to


⎛ ⎞
 
a4 + b4 + c4 + abc(a + b + c) + ab(a2 + b2 ) ≥ 4 ⎝ a 2 b2 ⎠ .
cyclic cyclic

We now use Schur’s inequality:

a2 (a − b)(a − c) + b2 (b − c)(b − a) + c2 (c − a)(c − b) ≥ 0.

Expansion gives

a4 + b4 + c4 + abc(a + b + c) ≥ a3 (b + c).
cyclic

Therefore

a4 + b4 + c4 + 9abc(a + b + c) + ab(a2 + b2 )
cyclic
 
3
≥ a (b + c) + ab(a2 + b2 )
cyclic cyclic
  
2 2
=2 ab(a + b ) ≥ 2 ab(2ab) = 4 (a2 + b2 ).
cyclic cyclic cyclic
84 Inequalities

2.9 Majorisation technique


There is another useful technique, called
 the majorisation technique,
 often used
to derive new inequalities. Let x = x1 , x2 , . . . , xn and y = y1 , y2 , . . . , yn
be two vectors in Rn such that
x1 ≥x2 ≥x3 ≥· · ·≥xn and y1 ≥y2 ≥y3 ≥· · ·≥yn .
We say x is majorised by y and write x ≺ y if
(i) x1 +x2 +x3 +· · ·+xk ≤ y1 +y2 +y3 +· · ·+yk for 1 ≤ k ≤ n − 1;
(ii) x1 +x2 +x3 +· · ·+xn = y1 +y2 +y3 +· · ·+yn .
Let A be a subset of Rn . A function f : A → R is said to be Schur-convex if
   
x ≺ y =⇒ f x ≤ f y .
If the inequality is reversed, we say f is Schur-concave. It is clear that f is
Schur-concave if and only if −f is Schur-convex.
There is a useful criterion for checking whether a given function is Schur-
convex, at least when the function is reasonably smooth. Let
In = (a, b) × (a, b) × · · · × (a, b),
be an ‘interval’ in Rn . Suppose f : In → R has partial derivatives of the first
order. Then f is Schur-convex if and only if f is symmetric in the variables
and   
∂f (x) ∂f (x)
xj − xk − ≥ 0,
∂xj ∂xk
on In , for all j = k. Here are some useful classes of Schur-convex functions.
(i) If g : (a, b) → R is a convex function, then
  
n
 
f x1 ,x2 ,x3 ,. . .,xn = g xj
j=1

n
is Schur-convex on I . This leads to a standard majorisation inequality:
if f : (a, b) → R is convex, then
    
n
   n
 
x1 ,x2 ,x3 ,. . .,xn ≺ y1 ,y2 ,y3 ,. . .,yn =⇒ f xj ≤ f yj .
j=1 j=1

(ii) A function f : In → R is said to be convex if


     
f λx + (1 − λ)y ≤ λf x + (1 − λ)f y ,

for all vectors x, y in In , and λ ∈ [0, 1]. Any convex function is Schur-
convex.
Techniques 85

(iii) A function f : In → R is said to be quasi-convex if


  0    1
f λx + (1 − λ)y ≤ max f x , f y ,

for all vectors x, y in In , and λ ∈ [0, 1]. Any quasi-convex function is


Schur-convex.

Example 2.34. If a, b, c are the sides of a triangle, prove that


√ √ √ √ √ √
a + b − c + b + c − a + c + a − b ≤ a + b + c.

Solution: Suppose a ≥ b ≥ c. Then

2(s − c) ≥ a,
2(s − c) + 2(s − b) ≥ a + b,
2(s − b) + 2(s − a) + 2(s − c) = a + b + c.

Thus    
a, b, c ≺ 2(s − c), 2(s − b), 2(s − a) .

Since f (t) = t is concave on (0, ∞), using majorisation theorem,
   √ √ √
2(s − c) + 2(s − b) + 2(s − a) ≤ a + b + c.

This reduces to
√ √ √ √ √ √
a + b − c + b + c − a + c + a − b ≤ a + b + c.

Example 2.35. Let a1 ,a2 ,a3 ,. . .,an be n natural numbers such that

a1 ≤a2 ≤a3 ≤· · ·≤an , and a1 +a2 +a3 +· · ·+an = nk + m,

where k, m are non-negative integers and 0 ≤ m < n. Prove that

a1 a2 a3 · · ·an ≤ (k + 1)m k n−m .

Solution: Consider the vectors


   
an , an−1 , an−2 , . . . , a1 and k + 1, k + 1, . . . , k + 1, k, k, . . . , k ,

where k + 1 appears m times and k appears n − m times. Ifm > 0, then


we observe that an ≥ k + 1. Otherwise an ≤ k and hence aj ≤ nk, a
contradiction. Similarly, it is easy to see that if m > 1, then an + an−1 ≥
2(k + 1). An easy induction proves that

an + an−1 + · · · + an−m+1 ≥ m(k + 1).


86 Inequalities

Similarly,

an + an−1 + · · · + an−m+1 + an−m + · · · + an−m−j ≥ m(k + 1) + (j + 1)k,

for 0 ≤ j ≤ n − m − 1. It follows that


   
k + 1, k + 1 . . . , k + 1, k, k, . . . , k ≺ an , an−1 , an−2 , . . . , a1 .

Since f (x) = ln x is concave on (0, ∞), we get



n
ln aj ≤ m ln(k + 1) + (n − m) ln k.
j=1

This simplifies to
a1 a2 a3 · · ·an ≤ (k + 1)m k n−m .

Example 2.36. Show that in any triangle with sides a, b, c,

(a + b − c)(b + c − a)(c + a − b) ≤ abc.

Solution: If we take a ≥ b ≥ c, then it is easy to see that


 
(a, b, c) ≺ 2(s − c), 2(s − b), 2(s − a) ,

where s = (a + b + c)/2 is the semi-perimeter of the triangle. Using the concave


function f (x) = ln x, we get

ln 2(s − c) + ln 2(s − b) + ln 2(s − a) ≤ ln a + ln b + ln c.

This simplifies to

(a + b − c)(b + c − a)(c + a − b) ≤ abc.

2.10 Muirhead’s theorem


There isa classical theorem  which was discovered by Muirhead.  Consider a vec-

tor a = a1 ,a2 ,a3 ,. . .,an in Rn . For each positive vector x = x1 ,x2 ,x3 ,. . .,xn ,
i.e., vector with positive components, consider the sum
  1  a1 a2
S a; x = xσ(1) xσ(2) · · · xaσ(n)
n
,
n!
σ∈Sn

where Sn denotes the set of all permutations of {1, 2, 3, . . . , n}. Then Muir-
head’s theorem asserts that
   
S a; x ≤ S b; x
Techniques 87

for all positive vectors x in Rn if and only if a ≺ b. For a proof of this


interesting theorem, please refer to [3].
For example (2, 1, 1) ≺ (3, 1, 0). Hence Muirhead’s theorem gives
 
2 x2 yz + y 2 zx + z 2 xy ≤ x3 y + y 3 z + z 3 x + xy 3 + yz 3 + zx3 ,

for all non-negative numbers x, y, z. Similarly (1, 1, 1) ≺ (3, 0, 0) and hence


3xyz ≤ x3 + y 3 + z 3 for all non-negative reals x, y, z. In fact, it is easy to derive
the AM-GM inequality. Observe that for any positive integer n
   
1/n, 1/n, . . . , 1/n ≺ 1, 0, 0, . . . , 0 .

Hence for any n non-negative real numbers a1 ,a2 ,a3 ,. . .,an , we obtain
 1/n
n a1 a2 · · · an ≤ a1 + a2 + · · · + an .

Example 2.37. Show that for any positive reals a, b, c, the inequality

a3 b3 c3
+ + ≥ a + b + c,
bc ca ab
holds.

Solution: We observe that the inequality is equivalent to

a2 bc + ab2 c + abc2 ≤ a4 + b4 + c4 .

However (2, 1, 1) ≺ (4, 0, 0) since 2 < 4, 2 + 1 < 4 + 0 and 2 + 1 + 1 = 4 + 0 + 0.


Hence Muirhead’s theorem at once gives the result.

Example 2.38. Let a, b, c be non-negative real numbers. Prove that

1
a3 + b3 + c3 + abc ≥ (a + b + c)3 .
7

Solution: Expanding the right side, the inequality may be written in an


equivalent form:
   
6 a3 + b3 + c3 + abc ≥ 3 a2 b + b2 c + c2 a + ab2 + bc2 + ca2 .

Note that
 
2 a3 + b3 + c3 ≥ a2 b + b2 c + c2 a + ab2 + bc2 + ca2

by Muirhead’s theorem. Hence the result follows.


88 Inequalities

2.11 Homogenisation
There is a large class of inequalities which are non-homogeneous and come
with constraints. In many cases, the given constraint may be used to put the
inequality in a homogeneous form and then one may use results like Muirhead’s
theorem or Schur’s inequality to prove them.

Example 2.39. Let x, y, z be positive real numbers such that x + y + z = 1.


Prove that  
2 x2 + y 2 + z 2 + 9xyz ≥ 1.

Solution: We put this in a homogeneous form using x + y + z = 1:


    3
2 x + y + z x2 + y 2 + z 2 + 9xyz ≥ x + y + z .
Expanding, we get
  
2 x3 + 2 xy 2 + 2 x2 y + 9xyz
cyclic cyclic cyclic
  
≥ x3 + 3 xy 2 + 3 x2 y + 6xyz.
cyclic cyclic cyclic

This may be written in the form


  
x3 − xy 2 − x2 y + 3xyz ≥ 0.
cyclic cyclic cyclic

Equivalently, 
x(x − y)(x − z) ≥ 0,
cyclic

which follows from Schur’s inequality.

Example 2.40. (Tournament of Towns, 1997) Let a, b, c be positive real numbers


such that abc = 1. Prove that
1 1 1
+ + ≤ 1.
a+b+1 b+c+1 c+a+1

Solution: We introduce a = x3 , b = y 3 and c = z 3 so that x, y, z are


positive real numbers such that xyz = 1. We may now write the inequality in
a homogeneous form:
1 1 1 1
+ 3 + 3 ≤ .
x3 + y 3 + xyz y + z 3 + xyz z + x3 + xyz xyz
Equivalently, we have
    "  
xyz x3 + y 3 + xyz y 3 + z 3 + xyz ≤ x3 + y 3 + xyz .
cyclic cyclic
Techniques 89

But the left side reduces to


    
3 3 4 6 2 2 2
xyz 3 x y +4 x yz + x + 3x y z .
cyclic cyclic cyclic

And the right side is


 
3x3 y 3 z 3 + 2x2 y 2 z 2 x3 + xyz x6
cyclic cyclic
  
+ 3xyz x3 y 3 + x6 y 3 + x3 y 6 .
cyclic cyclic cyclic

The inequality is equivalent to


  
2 x5 y 2 z 2 ≤ x6 y 3 + x3 y 6 .
cyclic cyclic cyclic

This may be written as


S(5, 2, 2) ≤ S(6, 3, 0).
Since (5, 2, 2) ≺ (6, 3, 0), the result follows from Muirhead’s theorem.

Example 2.41. (IMO, 1984) Let x, y, z be three non-negative real numbers such
that x + y + z = 1. Prove that
2
0 ≤ xy + yz + zx − 2xyz ≤ .
27
Solution: Here we give a proof using homogenisation. For a different proof,
see solution to problem 3.6 on page 291.
Using x + y + z = 1, we put the inequality in a homogeneous form:
   7 3
0 ≤ x + y + z xy + yz + zx − 2xyz ≤ x+y+z .
27
Observe that
    
x + y + z xy + yz + zx − 2xyz = x2 y + xy 2 + xyz.
cyclic cyclic

Hence the left-side inequality follows. On the other hand


 3   
x+y+z = x3 + 3 x2 y + 3 xy 2 + 6xyz.
cyclic cyclic cyclic

Hence the inequality reduces to


  
6 x2 y + xy 2 ≤ 7 x3 + 15xyz.
cyclic cyclic cyclic
90 Inequalities

We may put it in the form


       
2 x3 − x2 y − xy 2 + 5 x(x − y)(x − z) ≥ 0.
cyclic cyclic cyclic cyclic

Since (2, 1, 0) ≺ (3, 0, 0), Muirhead’s theorem gives


  
2 x3 − x2 y − xy 2 ≥ 0.
cyclic cyclic cyclic

By Schur’s inequality, we get



x(x − y)(x − z) ≥ 0.
cyclic

Hence the result follows.

Example 2.42. (USSR, 1962) Let a, b, c, d be non-negative real numbers such


that abcd = 1. Prove that

a2 + b2 + c2 + d2 + ab + ac + ad + bc + bd + cd ≥ 10.

Solution: The result at once follows from the AM-GM inequality. But we
give here a different solution using homogenisation. We may homogenise this
to get an equivalent inequality:
 2 2
a + b2 + c2 + d2 + ab + ac + ad + bc + bd + cd ≥ 100abcd.
 
For a given 4-tuple m1 , m2 , m3 , m4 of non-negative integers, we use the no-
tation:
  1  mσ(1) mσ(2) mσ(3) mσ(4)
S m1 , m2 , m3 , m4 = a b c d ,
24
σ∈S4

where S4 is the set of all permutations of {1, 2, 3}. Expanding the left-side, we
may write the inequality in the form

4S(4, 0, 0, 0) + 24S(3, 1, 0, 0) + 18S(2, 2, 0, 0) + 48S(2, 1, 1, 0) ≥ 94S(1, 1, 1, 1).

Now Muirhead’s theorem gives

S(4, 0, 0, 0) ≥ S(3, 1, 0, 0) ≥ S(2, 2, 0, 0) ≥ S(2, 1, 1, 0) ≥ S(1, 1, 1, 1).

Hence

4S(4, 0, 0, 0) + 24S(3, 1, 0, 0) + 18S(2, 2, 0, 0) + 48S(2, 1, 1, 0)


≥ (4 + 24 + 18 + 48)S(1, 1, 1, 1) = 94S(1, 1, 1, 1).

This gives the required inequality.


Techniques 91

2.12 Normalisation
Yet another technique used while proving inequalities is normalisation. This is
the reverse process to homogenisation. Many times homogeneous inequalities
may be normalised to simplify the proofs. The standard application is the
AM-GM inequality. We have to prove that
a1 + a2 + · · · + an
≥ (a1 a2 · · · an )1/n ,
n
for non-negative real numbers a1 , a2 , . . . , an . Since it is homogeneous, we may
normalise it by a1 a2 a3 · · · an = 1. Thus we have to prove that
a1 + a2 + · · · + an ≥ n,
for any n positive real numbers a1 , a2 , . . . , an , under the additional condition
that a1 a2 a3 · · · an = 1. This may be proved by the principle induction. For
n = 1, it is immediate. If n = 2, then
√ √ √ 2
a1 + a2 − 2 = a1 + a2 − 2 a1 a2 = a1 − a2 ≥ 0.
Suppose it holds for any k positive real numbers whose product is 1, where
k < n. Take n positive real numbers a1 , a2 , . . . , an such that a1 a2 a3 · · · an = 1.
Among these n numbers, there must be some number ≥ 1 and there must be
some number ≤ 1. Thus we may assume that a1 ≥ 1 ≥ a2 . We may apply
induction to n − 1 numbers a1 a2 , a3 , . . . , an :
a1 a2 + a3 + · · · an ≥ n − 1.
Thus it is enough to prove that a1 + a2 ≥ 1 + a1 a2 . But this is equivalent to
(a1 − 1)(a2 − 1) ≤ 0 which is a consequence of a1 ≥ 1 ≥ a2 . Hence the proof is
complete.

Example 2.43. (INMO, 2007) Let x, y, z be positive real numbers. Prove that
  2    
x + y + z)2 yz + zx + xy ≤ 3 y 2 + yz + z 2 z 2 + zx + x2 x2 + xy + y 2 .

Solution: The inequality is homogeneous of degree 6. Hence we may assume


x + y + z = 1. Let α = xy + yz + zx. We see that
x2 + xy + y 2 = (x + y)2 − xy
= (x + y)(1 − z) − xy
= x + y − α = 1 − z − α.
Thus
"
(x2 + xy + y 2 ) = (1 − α − z)(1 − α − x)(1 − α − y)
= (1 − α)3 − (1 − α)2 + (1 − α)α − xyz
= α2 − α3 − xyz.
92 Inequalities

Thus we need to prove that α2 ≤ 3(α2 − α3 − xyz). This reduces to


3xyz ≤ α2 (2 − 3α).
However
3α = 3(xy + yz + zx) ≤ (x + y + z)2 = 1,
so that 2 − 3α ≥ 1. Thus it suffices to prove that 3xyz ≤ α2 . But
α2 − 3xyz = (xy + yz + zx)2 − 3xyz(x + y + z)

= x2 y 2 − xyz(x + y + z)
cyclic
1 
= (xy − yz)2 ≥ 0.
2
cyclic

For different solutions, see solution to problem 3.6 on page 445.

Example 2.44. (IMO, 2001) Prove that


a b c
√ +√ +√ ≥1
a2 + 8bc 2
b + 8ca 2
c + 8ab
for all positive real numbers a, b and c.

Solution: Here we give a solution using normalisation a + b + c = 1. We use


the convexity of f (t) = √1t . Using the weighted Jensen inequality, we get

a b c
√ +√ +√
a2 + 8bc 2
b + 8ca 2
c + 8ab
1
≥ .
a(a2 + 8bc) + b(b2 + 8ca) + c(c2 + 8ab)
1

Since f (t) = t
is strictly decreasing and f (1) = 1, it suffices to prove that

a(a2 + 8bc) + b(b2 + 8ca) + c(c2 + 8ab) ≤ 1.


For this we go back to homogenisation:
a(a2 + 8bc) + b(b2 + 8ca) + c(c2 + 8ab) ≤ (a + b + c)3 .
But this reduces to
 
3 a2 b + 3 ab2 ≥ 18abc.
cyclic cyclic

This follows from the AM-GM inequality (or one can use Muirhead’s theorem).
(For a different normalisation and a generalisation, refer to problem 3.6 on
page 401.)
Techniques 93

2.13 Stolarsky’s theorem


Suppose we have a homogeneous polynomial of degree 3 in three variables. We
have already encountered many such polynomials: for example, x3 + y 3 + z 3 −
3xyz or 2(x3 + y 3 + z 3 ) − (x2 y + y 2 z + z 2 x + xy 2 + yz 2 + zx2 ). There is a
beautiful result which gives us conditions under which such polynomials take
only non-negative values for all non-negative real numbers x, y, z.

Theorem 14. Let P (x, y, z) be a homogeneous polynomial of degree 3 in three


variables. Then the following statements are equivalent:

(i) P (1, 1, 1) ≥ 0, P (1, 1, 0) ≥ 0 and P (1, 0, 0) ≥ 0;

(ii) P (λ, μ, ν) ≥ 0 for all non-negative reals λ, μ, ν.

Proof: It is sufficient to prove that (i) implies (ii). Let us put


    
3 2 2
P (x, y, z) = A x +B x y+ xy + Cxyz.
cyclic cyclic cyclic

We also introduce

p = P (1, 1, 1) = 3A + 6B + C,
2q = P (1, 1, 0) = 2A + 2B,
r = P (1, 0, 0) = A.

It is easy to get

A = r, B = q − r, C = p − 6q + 3r.

We may write P (x, y, z) in terms of p, q, r:


    
3 2 2
P (x, y, z) = r x + (q − r) x y+ xy + (p − 6q + 3r)xyz.
cyclic cyclic cyclic

Here we consider two cases.


Case 1: Suppose q ≥ r. In this case we write
  
3
P (x, y, z) = r x − 3xyz
cyclic
    
2 2
+ q−r x y+ xy − 6xyz + pxyz.
cyclic cyclic

Since r ≥ 0, p ≥ 0 and q > r, the result follows.


94 Inequalities

Case 2: Suppose q ≤ r. We write


  
P (x, y, z) = q x3 − 3xyz
cyclic
     
3 2 2
+ r−q x − x y− xy + 3xyz + pxyz.
cyclic cyclic cyclic

Again the result follows from q ≥ 0, p ≥ 0, r ≥ q, and Schur’s inequality.


As a consequence of this, here is a striking result which is due to Stolarsky.

Theorem 15. (Stolarsky’s theorem) Let P (x, y, z) be a symmetric homogeneous


form of degree 3:
 
P (x, y, z) = px3 + qx2 y + rxyz ,
sym

where p, q, r are real numbers. Suppose P (1, 1, 1) ≥ 0, P (1, 1, 0) ≥ 0 and


P (2, 1, 1) ≥ 0. Then P (a, b, c) ≥ 0, whenever a, b, c are the sides of a trian-
gle.

Proof: Let a, b, c be the sides of a triangle. Then we can find positive reals
x, y, z such that
a = y + z, b = z + x, c = x + y.
It is easy to compute
P (1, 1, 1) = 6p + 6q + 6r ≥ 0,
P (1, 1, 0) = 4p + 2q ≥ 0,
P (2, 1, 1) = 20p + 14q + 12r ≥ 0.
Note that
 
P (x, y, z) = 2p x3 + y 3 + z 3
 
+q x2 y + y 2 z + z 2 x + xy 2 + yz 2 + zx2 + 6rxyz.
Hence some computation leads to
      
P (a, b, c) = (4p + 2q) x3 + (6p + 5q + 6r) x2 y + xy 2
cyclic cyclic cyclic
+12(q + r)xyz
= Q(x, y, z).
Observe that
Q(1, 1, 1) = 48(p + q + r) ≥ 0
Q(1, 1, 0) = 20p + 14q + 12r ≥ 0,
Q(1, 0, 0) = 4p + 2q ≥ 0.
Techniques 95

Hence Q(x, y, z) ≥ 0 for all non-negative real numbers x, y, z. It follows that


P (a, b, c) ≥ 0.

Example 2.45. Let a, b, c be the sides of a triangle. Prove that


 
3(a + b)(b + c)(c + a) ≤ 8 a3 + b3 + c3 .

Solution: Consider the polynomial


 
P (a, b, c) = 8 a3 + b3 + c3 − 3(a + b)(b + c)(c + a).
This is a homogeneous polynomial of degree 3 in the variables a, b, c. Observe
that
P (1, 1, 1) = 24 − 18 − 6 ≥ 0,
P (1, 1, 0) = 16 − 6 = 10 > 0,
P (2, 1, 1) = 80 − 42 − 12 = 16 > 0.
By Stolarsky’s theorem, P (a, b, c) ≥ 0, for all a, b, c which are the sides of a
triangle. (In fact we can apply theorem 14 since P (1, 0, 0) = 8 > 0 and hence
the result is true for positive a, b, c.)

Example 2.46. Let a, b, c be the sides of a triangle. Prove that


    
2 a + b + c a2 + b2 + c2 ≥ 3 a3 + b3 + c3 + 3abc

Solution: If we take
    
P (x, y, z) = 2 x + y + z x2 + y 2 + z 2 − 3 x3 + y 3 + z 3 + 3xyz ,
then P (x, y, z) is a homogeneous polynomial of degree 3 in three variables.
Moreover, P (1, 1, 1) = 0, P (1, 1, 0) = 2 and P (2, 1, 1) = 0. Hence Stolarsky’s
theorem is applicable and we conclude that P (a, b, c) ≥ 0 whenever a, b, c are
the sides of a triangle. (Note that we cannot apply theorem 14 directly, since
P (1, 0, 0) = −1 < 0.)

2.14 Methods for symmetric inequalities

In this section we deal with more inequalities connecting elementary symmetric


functions and their application.

2.14.1 Use of elementary symmetric functions


For any non-negative real numbers x, y, z, we introduce their elementary sym-
metric polynomials:
p = x + y + z, q = xy + yz + zx, r = xyz.
96 Inequalities

We prove several identities involving symmetric functions.

1. x2 + y 2 + z 2 = (x + y + z)2 − 2(xy + yz + zx) = p2 − 2q.


2.

x3 + y 3 + z 3 = (x + y + z)3 − 3(x + y + z)(xy + yz + zx) + 3xyz


= p3 − 3pq + 3r.

3. x2 y 2 + y 2 z 2 + z 2 x2 = (xy + yz + zx)2 − 2xyz(x + y + z) = q 2 − 2pr.


4.

x4 + y 4 + z 4 = (x2 + y 2 + z 2 )2 − 2(x2 y 2 + y 2 z 2 + z 2 x2 )
 2 2  
= p − 2q − 2 q 2 − 2pr
= p4 − 2p2 q + 2q 2 + 4pr.

5.

(x + y)(y + z)(z + x)
= (p − x)(p − y)(p − z)
= p3 − p2 (x + y + z) + p(xy + yz + zx) − xyz
= pq + r.

6.

(x + y)(y + z) + (y + z)(z + x) + (z + x)(z + y)


= (p − z)(p − x) + (p − x)(p − y) + (p − y)(p − z)
= 3p2 − 2p(x + y + x) + (xy + yz + zx)
= p2 + q.

7.

(x + y)2 (y + z)2 + (y + z)2 (z + x)2 + (z + x)2 (x + y)2


 2
= (x + y)(y + z) + (y + z)(z + x) + (z + x)(z + y)
 
− 2(x + y)(y + z)(z + x) 2(x + y + z)
= (p2 − q)2 − 4p(pq − r).

8.

xy(x + y) + yz(y + z) + zx(z + x)


= xy(p − z) + yz(p − x) + zx(p − y)
= p(xy + yz + zx) − 3xyz = pq − 3r.
Techniques 97

9.

x2 (y + z) + y 2 (z + x) + z 2 (x + y)
= x2 (p − x) + y 2 (p − y) + z 2 (p − z)
= p(x2 + y 2 + z 2 ) − (x3 + y 3 + z 3 )
= p(p2 − 2q) − (p3 − 3pq + 3r) = pq − 3r.

10.

x 3 y 3 + y 3 z 3 + z 3 x3
= (xy + yz + zx)3 − 3(xy + yz + zx)xyz(x + y + z) + 3x2 y 2 z 2
= q 3 − 3pqr + 3r2 .

11.

xy(x2 + y 2 ) + yz(y 2 + z 2 ) + zx(z 2 + x2 )


= xy(p2 − 2q − z 2 ) + yz(p2 − 2q − x2 ) + zx(p2 − 2q − y 2 )
= (p2 − 2q)(xy + yz + zx) − xyz(x + y + z)
= p2 q − 2q 2 − pr.

Here are some inequalities connecting p, q, r.

Example 2.47. p2 ≥ 3q.

Solution: Follows from (x + y + z)2 ≥ 3(xy + yz + zx).

Example 2.48. p3 ≥ 27r.

Solution: Follows from (x + y + z)3 ≥ 27xyz.

Example 2.49. q 3 ≥ 27r2 .

Solution: Follows from (xy + yz + zx)3 ≥ 27(xyz)2 .

Example 2.50. q 2 ≥ 3pr.

Solution: We have

(xy + yz + zx)2 = (x2 y 2 + y 2 z 2 + z 2 x2 ) + 2xyz(xy + yz + zx).

But
x2 y 2 + y 2 z 2 + z 2 x2 ≥ (xy)(yz) + (yz)(zx) + (zx)(xy),
98 Inequalities

by C-S inequality. It follows that


q 3 ≥ pr + 2pr = 3pr.

Example 2.51. 2p3 + 9r ≥ 7pq.

Solution: This reduces to



2(x3 + y 3 + z 3 ) ≥ x2 y.
sym

Since (2, 1, 0) ≺ (3, 0, 0), the result follows from Muirhead’s theorem.

Example 2.52. p2 q + 3pr ≥ 4q 2 .

Solution: After expanding this reduces to



x3 y ≥ 2(x2 y 2 + y 2 z 2 + z 2 x2 ).
sym

Since (2, 2, 0) ≺ (3, 1, 0), this inequality is a consequence of Muirhead’s theo-


rem.

Example 2.53. q 3 + 9r2 ≥ 4pqr.

Solution: Again, expansion reduces this to



x3 y 3 + y 3 z 3 + z 3 x3 + 3x2 y 2 z 2 ≥ x3 y 2 z.
sym

By Schur’s inequality, we have for non-negative reals a, b, c


a3 + b3 + c3 + 3abc ≥ a2 b + bc + c2 a + ab2 + bc2 + ca2 .
Take a = xy, b = yz and c = zx. We obtain

x3 y 3 + y 3 z 3 + z 3 x3 + 3x2 y 2 z 2 ≥ x3 y 2 z.
sym

Example 2.54. p4 + 3q 2 ≥ 4p2 q.

Solution: This factorises as (3q − p2 )(q − p2 ) ≥ 0. We know that 3q ≤ p2 .


Hence q ≤ p2 . Thus (3q − p2 ) and (q − p2 ) are both non-positive. This implies
that their product is non-negative.

Example 2.55. pq 2 ≥ 2p2 r + 3qr.


Techniques 99

Solution: We know that 3q ≤ p2 and this gives 3qr ≤ p2 r. Therefore


3pr + 2p2 r ≤ 3p2 r ≤ q 2 p,
since q 2 ≥ 3pr.

1 1
Example 2.56. [10] Let a, b, c be positive real number such that + +
a+1 b+1
1
= 1. Prove that
c+1
1 1 1
+ ++ ≥ 1.
8ab + 1 8bc + 1 8ca + 1

Solution: Introducing p = a + b + c, q = ab + bc + ca and r = abc, the given


condition reduces to q + 2r = 1. The inequality to be proved reduces to
64pr + 16q + 3 ≥ 512r2 + 64pr + 8q + 1.
Hence it is sufficient to prove that
8q + 2 ≥ 512r2 .
We know that q 3 ≥ 27r2 . Using q = 1 − 2r, this is equivalent to
(8r − 1)(r + 1)2 ≤ 0.
Hence 8r − 1 ≤ 0. Now the inequality to be proved is
8(1 − 2r) + 2 ≥ 512r2 .
This can be written as
(8r − 1)(64r + 10) ≥ 0.
Since 8r − 1 ≤ 0, the result follows.

Example 2.57. Let a, b, c be positive real numbers such that a+b+c = 1. Prove
that    
6 a 3 + b3 + c 3 + 1 ≥ 5 a 2 + b2 + c 2 .

Solution: We use p = a + b + c, q = ab + bc + ca and r = abc.Observe that


p = 1. We also know
a 3 + b3 + c 3 = p(p2 − 3q) + 3r = 1 − 3q + 3r,
a 2 + b2 + c 2 = p2 − 2q = 1 − 2q.
Hence we have to prove
6 − 18q + 18r + 1 ≥ 5 − 10q.
This reduces to 9r + 1 ≥ 4q. But we know that p3 − 4pq + 9r ≥ 0. This follows
from Schur’s lemma. Since p = 1, this gives 9r + 1 ≥ 4q.
100 Inequalities

Example 2.58. Suppose x, y, z are positive real numbers such that x2 +y 2 +z 2 =


1. Prove that
8
(1 − xy)(1 − yz)(1 − zx) ≥ .
27
Solution: Introduce p = x + y + z, q = xy + yz + zx and r = xyz. The
inequality reduces to
8
1 − q + pr − r2 ≥ .
27
Using p3 − 4pq + 9r ≥ 0, we also get
9r ≥ p(4q − p2 ) = p(2q − 1),
since p2 −2q = x2 +y 2 +z 2 = 1. We also have p2 ≥ 3q. Hence 2q +1 = p2 ≥ 3q,
which shows that q ≤ 1. Using pq − 9r ≥ 0, we get
p ≥ pq ≥ 9r.
Therefore 9p − 9r ≥ 8p. This gives us (p − r) ≥ 8p/9. Thus we obtain
8 8 p(2q − 1) 8 2 8
r(p − r) ≥ pr ≥ p · = p (2q − 1) = (2q − 1)(2q + 1).
9 9 9 81 81
Using this estimate, we obtain
8
1 − q + pr − r2 ≥ 1 − q + (4q 2 − 1).
81
Therefore, we need to prove that
8 8
1−q+ (4q 2 − 1) ≥ .
81 27
Simplification leads to the inequality
(1 − q)(49 − 32q) ≥ 0.
Since q ≤ 1, the result holds.

2.14.2 An alternate approach for symmetric inequalities


Some times, the introduction of elementary symmetric functions may not give
sharp bounds the inequality may require sharp bounds. We use a slightly
different approach in such cases. For real numbers a, b, c, we introduce

p = a + b + c, q = a2 + b2 + c2 − ab − bc − ca, r = abc.
Note that q ≥ 0 and ab + bc + ca is related to p, q by
p2 − q 2
= ab + bc + ca.
3
Then the following inequality holds:
Techniques 101

Example 2.59.
(p + q)2 (p − 2q) (p − q)2 (p + 2q)
≤r≤ . (2.9)
27 27
Equality holds if and only if some two of a, b, c are equal.

Solution: Consider the polynomial whose root are a, b, c:


p2 − q 2
f (x) = (x − a)(x − b)(x − c) = x3 − px2 + x − r.
3
Its derivative is
p2 − q 2
f  (x) = 3x2 − 2ux + .
3
Hence f  (x) = 0 if and only if
p±q
x= .
3
Let
p−q p+q
x1 = , x2 = .
3 3
Then f  (x) < 0 for x1 < x < x2 . Hence f  (x) > 0 for x < x1 and x > x2 . Next
we compute the second derivative of f (x):

f  (x) = 6x − 2p.

Hence  
p+q
f  (x2 ) = 6 − 2p = 2q > 0
3
Hence f has a local minimum at x = x2 . Similarly, we see that f  (x1 ) =
−2q < 0 and hence f has a local maximum at x2 . Thus we see that f (x2 ) ≤ 0
and f (x1 ) ≥ 0. But it is easy to compute that
(p + q)2 (p − 2q)
f (x2 ) = − r,
27
and
(p − q)2 (p + 2q)
f (x1 ) = − r.
27
Combining, we get
(p + q)2 (p − 2q) (p − q)2 (p + 2q)
≤r≤ .
27 27
Equality holds if and only if f (x1 ) = 0 = f (x2 ). Since f  (x1 ) = 0 = f  (x2 ), it
follows that equality holds if and only if f has a double root. This means some
two of a, b, c are equal.
We can use this in proving several inequalities.
102 Inequalities

Example 2.60. [10] Let a, b, c be real numbers. Prove that

a4 + b4 + c4 ≥ abc(a + b + c).

Solution: We may assume that a + b + c = 1, since the inequality is homo-


geneous. Observe that

p2 + 2q 2
a 2 + b2 + c 2 = ,
3
and
−p4 + 8p2 q 2 + 2q 4
a 4 + b4 + c 4 = + 4pr.
9
Using p = 1, the inequality reduces to

−1 + 8q 2 + 2q 4
+ 4r ≥ r.
9
This can be written in the form

−1 + 8q 2 + 2q 4 + 27r ≥ 0.

Using the inequality (2.9), it is enough to prove that

1 − 8q 2 + 2q 4 + (1 + q)2 (1 − 2q) ≥ 0.

This simplifies to
q 2 (2q 2 − 2q + 5) ≥ 0.
However, we can write
 
2 2 q 2 (2q − 1)2 + 9
q (2q − 2q + 5) = ,
2
which is non-negative.

Example 2.61. [10] Let a, b, c be real numbers such that a2 + b2 + c2 = 9. Prove


that
2(a + b + c) − abc ≤ 10.

Solution: Using the neqw variables p, q, r as earlier, we get

p2 + 2q 2
9 = a 2 + b2 + c 2 = .
3
Hence p2 + 2q 2 = 27. Using (2.9), we get
Techniques 103

(p + q)2 (p − 2q)
2(a + b + c) − abc = 2p − r ≤ 2p −
27
54p − p3 + 3pq 2 + 2q 3
= .
27
However, we see that
54p − p3 + 3pq 2 + 2q 3 = 54p − p(p2 + 2q 2 ) + 5pq 2 + 2q 3
= p(27 + 5q 2 ) + 2q 3 .
Hence we have to prove that
p(27 + 5q 2 ) ≤ 270 − 2q 3 .
However,
 2  2
270 − 2q 3 − p(27 + 5q 2 )
 
= 27(q − 3)2 2q 4 + 12q 3 + 49q 2 + 146q + 219 ≥ 0.
This implies the required result.

Example 2.62. [10] Let a, b, c be positive real numbers such that a + b + c = 1.


Prove that
1 1 1
+ + + +48(ab + bc + ca) ≥ 25.
a b c

Solution: We begin with ab + bc + ca = (1 − q 2 )/3 and r = abc. Observe


that q ∈ [0, 1]and
1 1 1 1 − q2
+ + + +48(ab + bc + ca) = + 16(1 − q 2 ).
a b c 3r
Thus we need to prove that
1 − q2
+ 16(1 − q 2 ) ≥ 25.
3r
Using (2.9), we have
1 − q2 1 − q2
+ 16(1 − q 2 ) ≥ 27 + 16(1 − q 2 )
3r 3(1 − q)2 (1 + 2q)
1+q
= 9 + 16(1 − q 2 )
(1 − q)(1 + 2q)
2q 2 (4q − 1)2
= + 25 ≥ 25.
(1 − q)(1 + 2q)
Equality holds if and only if a = b = c = 1/3 or (a, b, c) = (1/2, 1/4, 1/4) and
permutations thereof.
104 Inequalities

Example 2.63. Prove Schur’s inequality: for non-negative real numbers a, b, c,


we have
a(a − b)(a − c) + b(b − c)(b − a) + c(c − a)c − b) ≥ 0.

Solution: We write the inequality in the form

a3 + b3 + c3 + 3abc ≥ ab(a + b) + bc(b + c) + ca(c + a).

Introduce p, q, r by

p = a + b + c, q= a2 + b2 + c2 − ab − bc − ca, r = abc.

Since the inequality is homogeneous, we may assume p = a + b + c = 1. The


inequality now changes to

27r + 4q 2 − 1 ≥ 0.

Observe that
0 ≤ q = (a + b + c)2 − 3(ab + bc + ca) ≤ 1.
Hence q ∈ [0, 1]. If q ≥ 1/2, the inequality is obvious. If q ≤ 1/2, then (2.9)
shows that

27r + 4q 2 − ≥ (1 + q)2 (1 − 2q) + 4q 2 − 1 = q 2 (1 − 2q) ≥ 0.

2.14.3 Squares in handling symmetric inequalities


Another useful method for proving symmetrical inequality is to transform it in
to a sum of squares with positive coefficients. Just to give an idea, consider the
standard inequality a3 + b3 + c3 ≥ 3abc for non-negative real numbers a, b, c.
We write this as

a3 + b3 + c3 − 3abc = s(a − b)2 + s(b − c)2 + s(c − a)2 ,

where s = (a + b + c)/2. Since s ≥ 0, it follows that a3 + b3 + c3 ≥ 3abc. In


general, we try to write the given inequality in the form

fa (b − c)2 + fb (c − a)2 + fc (a − b)2 ≥ 0

where fa , fb , fc are functions of a, b, c. The proof of the required inequality now


depends on the analysis of these functions fa , fb , fc .

Example 2.64. [10] Let a, b, c be positive real numbers. Prove that

a 2 + b2 + c 2 8abc
+ ≥ 2.
ab + bc + ca (a + b)(b + c)(c + a)
Techniques 105

Solution: We write the inequality in the form

a 2 + b2 + c 2 8abc
−1+ − 1 ≥ 0.
ab + bc + ca (a + b)(b + c)(c + a)
This reduces to

fa (b − c)2 + fb (c − a)2 + fc (a − b)2 ≥ 0,

where
abc
fa = b + c − a −
ab + bc + ca
and similar expressions for fb , fc . We may assume a ≥ b ≥ c Hence

(a−c)2 = (a−b+b−c)2 = (a−b)2 +(b−c)2 +2(a−b)(b−c) ≥ (a−b)2 +(b−c)2 .

Thus we have

fa (b − c)2 + fb (c − a)2 + fc (a − b)2


 
≥ fa (b − c)2 + fb (a − b)2 + (b − c)2 + fc (a − b)2
   
= fa + fb (b − c)2 + fc + fb (a − b)2 .

But
2abc 4c(bc + ca)
fa + fb = 2c − = ≥ 0.
ab + bc + ca ab + bc + ca
Similarly, fc + fb ≥ 0. Hence we obtain

fa (b − c)2 + fb (c − a)2 + fc (a − b)2 ≥ 0.

Example 2.65. [10] Let a, b, c be real numbers. Prove that

3(a2 − ab + b2 )(b2 − bc + c2 )(c2 − ca + a2 ) ≥ a3 b3 + b3 c3 + c3 a3 .

Solution: We may assume a, b, c ≥ 0. Expanding the left side, we get


  
3(a2 −ab+b2 )(b2 −bc+c2 )(c2 −ca+a2 ) = a 4 b2 − a 3 b3 − a4 bc+a2 b2 c2 .
sym cyclic cyclic

Therefore the inequality can be written in an equivalent form


  
3 a 4 b2 − 4 a 3 b3 − 3 a4 bc + 3a2 b2 c2 ≥ 0.
sym cyclic cyclic

This further reduces to the form

fa (b − c)2 + fb (c − a)2 + fc (a − b)2 ≥ 0, (2.10)


106 Inequalities

where

fa = 2a4 + 3b2 c2 − abc(a + b + c),


fb = 2b4 + 3c2 a2 − abc(a + b + c),
fc = 2c4 + 3a2 b2 − abc(a + b + c).

We may assume a ≥ b ≥ c. We see that

fa = 2a4 + 3b2 c2 − abc(a + b + c) ≥ a4 + 2a2 bc − abc(a + b + c),

since
a4 + 3b2 c2 − 2a2 bc = (a2 − bc)2 + 2b2 c2 ≥ 0.
Hence fa ≥ 0 using a ≥ b ≥ c. Similarly

fc = 2c4 + 3a2 b2 − abc(a + b + c) ≥ 3a2 b2 − abc(a + b + c) ≥ 0.

If fb ≥ 0, then (2.10) holds. Suppose fb < 0. It is easy to check that

(a − c)2 ≤ 2(a − b)2 + 2(b − c)2 .

Therefore

fa (b − c)2 + fb (c − a)2 + fc (a − b)2


 
≥ fa (b − c)2 + fb 2(a − b)2 + 2(b − c)2 + fc (a − b)2
   
= fa + 2fb (b − c)2 + fc + 2fb (a − b)2 .

However

fa + 2fb = 2a4 + 3b2 c2 + 4b4 + 6a2 c2 − 3abc(a + b + c)


≥ a4 + 2a2 bc + 8b2 ca − 3abc(a + b + c) ≥ 0,

and

fc + 2fb = 2c4 + 3a2 b2 + 4b4 + 6a2 c2 − 3abc(a + b + c)


≥ (3a2 b2 + 3a2 c2 ) + 3a2 c2 − 3abc(a + b + c) ≥ 0.

If a ≤ b ≤ c, we do the same type of analysis. It follows that (2.10) holds. This


completes the solution.

2.14.4 Strong mixing of variables


Another useful method in proving inequalities is a method known as strong
mixing of variables or SMV method for short. The crucial result of the method
is the following theorem:
Techniques 107

Theorem 16. If f : Rn → R is a continuous function which is symmetric


and satisfies f (a1 , a2 , a3 , . . . , an ) ≥ f (b1 , b2 , b3 , . . . , bn ) where (b1 , b2 , . . . , bn ) is
obtained from (a1 , a2 , a3 , . . . , an ) by replacing aj , ak by their average (aj +ak )/2,
where
aj = min{a1 , a2 , . . . , an }, ak = max{a1 , a2 , . . . , an },
then
f (a1 , a2 , . . . , an ) ≥ f (a, a, . . . , a)
where a = (a1 + a2 + · · · + an )/n.

For a proof of this, refer [11]. It is not necessary, we should take the average

(aj + ak )/2. We can also take geometric mean aj ak or root-mean-square
2
a2j + a2k
, whichever is convenient. The method is extremely useful in four
2
variables inequalities.
The method can be described as follows. Suppose we take some numbers
ap and aq among a1 , a2 , . . . , an and replace them by (ap + aq )/2. Let the new
sequence be (b1 , b2 , . . . , bn ). Check that f (a1 , a2 , . . . , an ) ≥ f (b1 , b2 , . . . , bn ).
Suppose we check that f (a, a, . . . , x) ≥ 0 for all a and x. Then it follows that
f (a1 , a2 , . . . , an ) ≥ 0. This is because replacing ap , aq by (ap + aq )/2 infinitely
many times leads to taking all equal to the average (a1 + a2 + · · · + an )/n. We
explain this by examples.

Example 2.66. (Short-list, IMO-1997) Suppose that a, b, c, d are non-negative


real numbers such that a + b + c + d = 1. Prove that
1 176
abc + bcd + cda + dab ≤ + abcd.
27 27

Solution: Let f (a, b, c, d) = abc + bcd + cda + dab − (176/27)abcd. We may


assume that a ≤ b ≤ c ≤ d. In all other cases, the same method works. We
write f (a, b, c, d) as
 
176
f (a, b, c, d) = ac(b + d) + bd a + c − .
27
Using a ≤ b ≤ c ≤ d, we obtain
1 1
a+c≤
(a + b + c + d) = .
2 2
Using AM-HM inequality, we see that
1 1 4 176
+ ≥ ≥8≥ .
a c a+c 27
Therefore,  
b+d b+d
f (a, b, c, d) ≥ f a, , c, .
2 2
108 Inequalities

In order to apply the theorem 16, it is enough to check that f (a, x, x, x) ≤ 1/27
whenever a + 3x = 1. But we have
176 3
f (a, x, x, x) = 3ax2 + x3 − ax .
27
Hence we have to show that
176 3 1
3ax2 + x3 − ax ≤
27 27
under the condition 3x + a = 1. Replacing a by (1 − 3x), this can be written
as
(1 − 3x)(4x − 1)2 (11x + 1) ≥ 0.
Since x ≤ 1/3, the result follows. Equality holds whenever a = b = c = d = 1/4
and a = b = c = 1/3, d = 0 and permutations thereof.

Example 2.67. [10] Let a, b, c, d be non-negative real numbers such that a + b +


c + d = 4. Prove that

(1 + 3a)(1 + 3b)(1 + 3c)(1 + 3d) ≤ 125 + 131abcd.

Solution: We consider the expression

f (a, b, c, d) = (1 + 3a)(1 + 3b)(1 + 3c)(1 + 3d) − 131abcd.

Again it is enough to consider the case a ≥ b ≥ c ≥ d. Consider the difference


 
a+c a+c
f (a, b, c, d) − f , b, ,d .
2 2

Some computation shows that this expression is equal to


   
(a + c)2 (a + c)2
9(1 + 3b)(1 = 3d) ac − − 131bd ac − .
4 4

Observe that

9(1 + 3b)(1 + 3d) ≥ 131bd if and only if 9 + +27(b + d) ≥ 50bd.

We also observe that 2(b + d) ≤ a + b + c + d = 4 so that b + d ≤ 2 and bd ≤ 1.


Therefore √
9 + 27(b + d) ≥ 54 bd ≥ 54bd ≥ 50bd.
This shows that  
a+c a+c
f (a, b, c, d) ≤ f , b, ,d .
2 2
Techniques 109

Therefore it is enough to prove f (a, a, a, x) ≤ 125, whenever 3a + x = 4. But

f (a, a, a, x) = (1 + 3a)3 (1 + 3x) − 131a3 x ≤ 125

if and only if
(a − 1)2 (3a − 4)(5a + 28) ≤ 0.
Since 3a + x = 4, we have 3a ≤ 4 and the result follows.

Example 2.68. (Pham King Hung) Let a1 , a2 , . . . , an be positive real numbers


such that a1 a2 a3 · · · an = 1. Prove that
1 1 1 1 3n
+ + + ··· + + ≥n+3
a1 a2 a3 an a1 + a2 + a3 + · · · + an
for all n ≥ 4.

Solution: As earlier, we assume a1 ≥ a2 ≥ · · · ≥ an . We show that


√ √
f (a1 , a2 , a3 , . . . , an ) ≥ f (a1 , a2 an , a3 , . . . , an−1 , a2 an ).

The difference can be written as


 2 √ √
1 1 3n( a2 − an )2
√ −√ − √ .
a2 an (a1 + a2 + · · · + an )(a1 + 2 a2 an + a3 + · · · + an−1 )

Hence we have to prove



(a1 + a2 + · · · + an )(a1 + 2 a2 an + a3 + · · · + an−1 ) ≥ 3na2 an .

Using a1 ≥ a2 ≥ · · · ≥ an , we get

(a1 + a2 + · · · + an )(a1 + 2 a2 an + a3 + · · · + an−1 )

≥ (2a2 + (n − 2)an )(a2 + 2 a2 an + (n − 3)an )
  √ 
≥ 2 2(n − 2) 2 + 2 n − 3 a2 an ≥ 3na2 an ,

since n ≥ 4. Hence it is enough to prove that

f (a1 , a, a, a, . . . , a) ≥ n + 3

where a = n−1
a2 a3 · · · an and a1 = 1/an−1 . Therefore we get

n−1 3nan−1
f (a1 , a, a, . . . , a) = an−1 + ++ .
a 1 + (n − 1)an

Consider the new function g(a) obtained after replacing n by n + 1:

n 3(n + 1)an
g(a) = an + + .
a 1 + nan+1
110 Inequalities

We show that g(a) ≥ n + 4. We first prove that g is a decreasing function for


a ≤ 1. Its derivative is
n 3n(n + 1)  n−1 
g  (a) = nan−1 − 2
+ n+1 2
a − a2n .
a (na + 1)

Consider the numerator obtained by clearing the denominators:


 n+1  
a − 1 (nan+1 + 1)2 − 3(n + 1)an+1 .

Using AM-GM inequality, we have


 2
nan+1 + 1 ≥ 4nan+1 ≥ 3(n + 1)an+1 .

Hence g  (a) ≤ 0 for all 0 ≤ a ≤ 1. Therefore g is decreasing on [0,1]. Since


g(1) = n + 4, we get g(a) ≥ g(1) = n + 4.
Chapter 3
Geometric inequalities

3.1 Introduction
Many of the inequalities we have studied and the techniques we have learnt have
their direct implications in a class of inequalities known as geometric inequal-
ities. These inequalities explore relations among various geometric elements.
For example, when we consider a triangle, we can associate many things with
it: angles, sides, area, medians, altitudes, circum-radius, in-radius, ex-radii and
so on. We have already some inequalities, viz., triangle inequalities associated
with the sides: a < b + c, b < c + a, c < a + b, where a, b, c are the sides
of a triangle; these conditions are necessary and sufficient for the existence of
a triangle with sides a, b, c. We can derive various relations among these ge-
ometric elements. The classic example is Euler’s inequality: R ≥ 2r, where
R is the circum-radius and r is the in-radius. This chapter provides several
inequalities of this kind, but the list is not exhaustive. For an excellent and a
fairly exhaustive collection of geometric inequalities, please refer to [5] and [6].

3.2 Notations
For a triangle ABC, we use the following standard notations:
• a = |BC|, b = |CA|, c = |AB|;
• α = ∠BAC, β = ∠CBA, γ = ∠ACB;
• ma , mb , mc are respectively the lengths of the medians from A, B, C on
to BC, CA, AB;
• ha , hb , hc are respectively the lengths of the altitudes from A, B, C on
to BC, CA, AB;
• wa , wb , wc are respectively the lengths of the angle bisectors of ∠BAC,
∠CBA, ∠ACB;
• R is the circum-radius; r is the in-radius; ra , rb , rc are the ex-radii;
• Δ is the area of ABC;
• s is the semi-perimeter of ABC: s = (a + b + c)/2;
• O is the circum-centre; H is the ortho-centre; I is the in-centre; G is the
centre of gravity; N is the nine-point centre;
112 Inequalities

• Ω1 denotes the first Brocard point of ABC; i.e., the unique point inside
ABC such that ∠CAΩ1 = ∠ABΩ1 = ∠BCΩ1 = ω, the Brocard angle of
ABC. It is known that ω ≤ π/6.

We also use a large number of results:


sin α sin β sin γ
1. the sine rule: = = = 2R;
a b c
2. the cosine rule: a2 = b2 + c2 − 2bc cos α, b2 = c2 + a2 − 2ca cos β, c2 =
a2 + b2 − 2ab cos γ;

3. the half-angle rule:


 
(s − b)(s − c) s(s − a)
sin α/2 = , cos α/2 = ,
bc bc
etc.;
abc
4. R = ;

5. r = 4R sin(α/2) sin(β/2) sin(γ/2) = (s − a) tan(α/2)
= (s − b) tan(β/2) = (s − c) tan(γ/2);

6. Δ = rs = (1/2)bc sin α = 2R2 sin α sin β sin γ;



7. Δ = s(s − a)(s − b)(s − c) (Heron’s formula);

8. 16Δ2 = 2a2 b2 + 2b2 c2 + 2c2 a2 − a4 − b4 − c4 (another form of Heron’s


formula);

9. (Stewart’s theorem) If D is a point on the side BC of a triangle ABC


such that BD : DC = λ : μ, then

λb2 + μc2 = (λ + μ)(AD2 + BD · DC).

Note that Appolonius’ theorem is a special case of this result: b2 + c2 =


2AD2 + 2BD2 , where D is the mid-point of BC;

10. ma = (1/2) 2b2 + 2c2 − a2 , etc.;
   √ 
11. wa = 2bc/(b + c) cos(α/2) = 2 bc/(b + c) s(s − a), etc.;
 
12. ra = s(s − b)(s − c) /(s − a) = Δ/(s − a), etc.;
 
13. OI 2 = R2 − 2Rr = R2 1 − 8 sin(α/2) sin(β/2) sin(γ/2) ;
   
14. OH 2 = R2 1 − cos α cos β cos γ = 9R2 − a2 + b2 + c2 ;
Geometric Inequalities 113

15. IH 2 = 2r2 − 4R2 cos α cos β cos γ;


  
16. cos α = 1 + 4 sin(α/2). (Remark:  the notation always denote in
this chapter the cyclical sum and denotes the cyclical
 2 product unless
2
otherwise
 stated. For example,  a = a + b + c, a b = a b + b2 c + c2 a,
sin α = sin α + sin β + sin γ, cos α = cos α cos β cos γ etc.)

3.3 Some identities involving elements of a triangle


In this section we develop some identities involving various elements of a trian-
gle: sides, angles, circum-radius, in-radius, ex-radii, altitudes, semi-perimeter.
etc. In fact we identify some polynomial equations satisfied by these elements
and use these equations for further explorations.

3.3.1 Equation for the sides


We start with the relation
     
a = 2R sin α = 4R sin α/2 cos α/2 , s − a = r cot α/2 .

Using these, we obtain


  ar   a(s − a)
sin2 α/2 = , cos2 α/2 = .
4R(s − a) 4Rr
We thus get
   
1 = sin2 α/2 + cos2 α/2
 
a r (s − a)
= + .
4R (s − a) r
This simplifies to the relation
 
a3 − 2sa2 + a s2 + r2 + 4Rr − 4Rrs = 0.

Thus, a is a root of the cubic equation

 
t3 − 2st2 + s2 + r2 + 4Rr t − 4Rrs = 0. (3.1)

Similarly, we can prove that b, c also satisfy the cubic equation (3.1). Now
using the relations between the roots and coefficients of a cubic equation, we
obtain  
a = 2s, ab = s2 + r2 + 4Rr, abc = 4Rrs.
Using (3.1), we obtain the equation for the reciprocals of the sides of a triangle.
Thus, 1/a, 1/b, 1/c satisfy the equation:
 
4Rrst3 − s2 + r2 + 4Rr t2 + 2st − 1 = 0. (3.2)
114 Inequalities

This gives
1 s2 + r2 + 4Rr  1 1
= , = .
a 4Rrs ab 2Rr

3.3.2 Equation for Altitudes


Using (3.2), we have
1  1 1
4Rrs − s2 + r2 + 4Rr 2 + 2s − 1 = 0.
a3 a a
Using a = 2Δ/ha , this reduces to
 
2Rh3a − s2 + r2 + 4Rr h2a + 4rs2 ha − 4r2 s2 = 0.
We get similar expressions for hb and hc . Thus, ha , hb , hc are the roots of the
cubic equation
 
2Rt3 − s2 + r2 + 4Rr t2 + 4rs2 t − 4r2 s2 = 0. (3.3)
We hence obtain
 s2 + r2 + 4Rr  2s2 r " 2r2 s2
ha = , ha hb = , ha = .
2R R R
Using the reciprocal equation
 
4r2 s2 t3 − 4rs2 t2 + s2 + r2 + 4Rr t − 2R = 0,
we also obtain
 1 1  1 s2 + r2 + 4Rr
= , = .
ha r ha hb 4r2 s2

3.3.3 Equation for s − a, s − b, s − c


The equation (3.1) may also be written in the form
 
(s − t)3 − s(s − t)2 + r2 + 4Rr (s − t) − sr2 = 0.
However we know that a, b, c are the roots of the equation (3.1). Thus, we see
that s − a, s − b, s − c are the roots of the equation
 
t3 − st2 + r2 + 4Rr t − sr2 = 0. (3.4)
We thus obtain the relations
  "
(s − a) = s, (s − a)(s − b) = r(r + 4R), (s − a) = sr2 .
Going to the reciprocal equation, we see that 1/(s − a), 1/(s − b), 1/(s − c) are
the roots of the equation
sr2 t3 − r(r + 4R)t2 + st − 1 = 0. (3.5)
We also obtain
 1 r + 4R  1 1
= , = 2.
s−a sr (s − a)(s − b) r
Geometric Inequalities 115

3.3.4 Equation for ex-radii


Now 1/(s − a) is a root of (3.5). Hence, we have

1 1 1
sr2 − r(r + 4R) +s − 1 = 0.
(s − a)3 (s − a)2 (s − a)

This can be written in the form


 3  2  
2 s s 2 s
r − r(r + 4R) +s − s2 = 0.
s−a s−a s−a

However we know that rs = ra (s − a) = Δ. Substituting s/(s − a) = ra /r, we


get the relation
ra3 − (r + 4R)ra2 + s2 ra − s2 r = 0.
Similar relations for rb and rc may be obtained. Thus, ra , rb , rc are the roots
of the equation

t3 − (r + 4R)t2 + s2 t − s2 r = 0. (3.6)

Consequently, we have the relations


 
ra = r + 4R, r a rb = s 2 , ra rb rc = s2 r.

Again the reciprocal equation of (3.6) is

s2 rt3 − s2 t2 + (r + 4R)t − 1 = 0, (3.7)

whose roots are 1/ra , 1/rb , 1/rc . We hence obtain the relations
 1 1  1 r + 4R
= , = .
ra r ra rb s2 r

3.4 Some geometric inequalities


In this section, we prove several inequalities among various elements of a tri-
angle. Again the readers are reminded that the following section only gives a
sample of geometric inequalities, but not an exhaustive list of these inequalities.
The excellent collection of examples in [5] and [6] gives an idea of the innu-
merable possibilities one can have in the class of geometric inequalities. The
techniques used are essentially what were developed earlier; we occasionally
resort to different ideas.

3.4.1. abc ≥ 8(s − a)(s − b)(s − c).


116 Inequalities

Proof: We have a2 − (b − c)2 ≤ a2 and equality holds if and only if b = c.


Similar inequalities hold: b2 − (c − a)2 ≤ b2 , c2 − (a − b)2 ≤ c2 . Hence,
  
abc ≥ a2 − (b − c)2 b2 − (c − a)2 c2 − (a − b)2
= (a + b − c)(b + c − a)(c + a − b)
= 8(s − a)(s − b)(s − c).

Equality holds if and only if a = b = c.


Alternatively, Stolarsky’s theorem (see theorem 15 on page 94) may be used.
Considering the polynomial

P (x, y, z) = xyz − (x + y − z)(y + z − x)(z + x − y),

we have a homogeneous polynomial of degree 3 in the variables x, y, z. Moreover


P (1, 1, 1) = 0, P (1, 1, 0) = 0 and P (2, 1, 1) = 2 > 0. Hence,

abc − (a + b − c)(b + c − a)(c + a − b) ≥ 0,

and the result follows.

 3
3.4.2. abc < a2 (s − a) ≤ abc.
2

Proof: We have

2 a2 (s − a) = a2 (b + c − a) + b2 (c + a − b) + c2 (a + b − c)
  
= a2 b + ab2 − a3 .

On the other hand, we also see that


  
(b + c − a)(c + a − b)(a + b − c) = c2 − a2 − b2 + 2ab a + b − c
= a2 b + ab2 + b2 c + bc2 + c2 a + ca2 − a3 − b3 − c3 − 2abc.

Thus, we obtain

2 a2 (s − a) = (b + c − a)(c + a − b)(a + b − c) + 2abc.

 2 we know that b + c − a > 0, c + a − b > 0


Since a, b, c are the sides of a triangle,
and a + b − c > 0. Hence, abc < a (s − a). Now using 3.4.1, we get

2 a2 (s − a) ≤ abc + 2abc = 3abc,

which proves the right hand side inequality.


Geometric Inequalities 117

Again, we may use Stolarsky’s theorem(theorem 15 on page 94). Consider-


ing 
P (x, y, z) = x2 (y + z − x) − 2xyz,
cyclic

we see that it is a homogeneous polynomial of degree 3 and P (1, 1, 1) = 1,


P (1, 1, 0) = 0, P (2, 1, 1) = 0. Hence, P (a, b, c) > 0, giving the left-side inequal-
ity. On the other hand, the polynomial

Q(x, y, z) = 3xyz − x2 (y + z − x),
cyclic

gives Q(1, 1, 1) = 0, Q(1, 1, 0) = 0 and Q(2, 1, 1) = 2. Thus Q(a, b, c) > 0, and


we get the right-side inequality.

3  a
3.4.3. ≤ < 2. Equality holds on the left if and only if a = b = c.
2 b+c

Proof: We have proved this in chapter 1; refer to (1.4). The left hand side
of the above inequality is generally known as Nesbitt’s inequality. There are a
variety of ways of proving this. We give two such proofs.

(i) Using the Cauchy-Schwarz inequality, we have


 2
2 a 
(a + b + c) = a(b + c)
b+c
   
a
≤ a(b + c) .
b+c

This gives
 a (a + b + c)2 3(ab + bc + ca) 3
≥ ≥ = ,
b+c 2(ab + bc + ca) 2(ab + bc + ca) 2

since (a + b + c)2 ≥ 3(ab + bc + ca).

(ii) We may assume a ≤ b ≤ c, since the inequality is symmetric in a, b, c.


This implies that
1 1 1
≤ ≤ .
b+c c+a a+b
Using rearrangement inequality, we obtain
a b c a b c
+ + ≥ + + ,
b+c c+a a+b c+a a+b b+c
a b c a b c
+ + ≥ + + .
b+c c+a a+b a+b b+c c+a
118 Inequalities

Adding these two, we obtain


 
a b c
2 + + ≥ 3.
b+c c+a a+b
This gives the desired inequality.

√ √ √ √ √
3.4.4. s< s−a+ s−b+ s−c≤ 3s.

Proof: The first inequality follows from the fact that for any positive reals
x, y, z, we have √ √ √

x + y + z < x + y + z.

On the other hand f (x) = x is a concave function on (0, ∞). Hence,

 1√ 1
s−a≤ (s − a),
3 3
which gives √ √ √ √
s−a+ s−b+ s−c≤ 3s.


3 3
3.4.5. 0 < sin α + sin β + sin γ ≤
.
2
 
Proof: We know that OH 2 = 9R2 − a2 + b2 + c2 , which gives

a2 + b2 + c2 ≤ 9R2 .

Hence, we obtain
 
(a + b + c)2 ≤ 3 a2 + b2 + c2 ≤ 27R2 .

This gives √
0 < a + b + c ≤ 3 3R.
Using a = 2R sin α, etc., we get

3 3
0 < sin α + sin β + sin γ ≤ .
2

3.4.6. sin 2α + sin 2β + sin 2γ ≤ sin α + sin β + sin γ.


Geometric Inequalities 119

Proof: Consider f (x) = − sin x on (0, π). This is a convex function. Suppose
α ≤ β ≤ γ. Using Theorem 10, we see that for any permutation (α , β  , γ  ) of
(α, β, γ), we have
      
− sin α + α ≤ − sin α + α = − sin 2α.

This reduces to    
sin 2α ≤ sin α + α .

Taking α = β, β  = γ and γ  = α, and using sin(α + β) = sin γ, etc., we get

sin 2α + sin 2β + sin 2γ ≤ sin α + sin β + sin γ.


" 3 3
3.4.7. sin α ≤ .
8

Proof: Using the AM-GM inequality and 3.4.5, we have

"  3  √ 3 √
sin α 3 3 3
sin α ≤ ≤ = .
3 2 8

"   1
3.4.8. 0< sin α/2 ≤ .
8
 
2 2
  
Proof: We know that OI = R 1 − 8 sin α/2 . Hence, it follows that
  
1 − 8 sin α/2 ≥ 0. The first inequality is obvious. Equality holds if and
only if the triangle is equilateral.

 3
3.4.9. 1< cos α ≤ .
2
   
Proof:
  We use the known identity: cos α = 1 + 4 sin α/2 . Since
sin α/2 > 0, we get the left side inequality. Now using the inequality
3.4.8, we get the right side inequality.

 3
3.4.10. 1< sin(α/2) ≤ .
2
120 Inequalities

Proof: Whenever α, β, γ are the angles of a triangle, (π − α)/2, (π − β)/2,


(π − γ)/2 are also the angles of some triangle. Applying 3.4.9 to this triangle,
we get these inequalities.

" 1
3.4.11. cos α ≤ .
8
Proof: Using 3.4.9 and the AM-GM inequality, we get

"  3
cos α 1
cos α ≤ ≤ .
3 8

3.4.12. If x, y, z are real numbers such that xyz > 0, then


 
1 yz zx xy
x cos α + y cos β + z cos γ ≤ + + .
2 x y z

Proof: We start with the obvious inequality


 2  2
xz cos α + yz cos β − xy + xz sin α − yz sin β ≥ 0.

This simplifies to

2x2 yz cos α + 2xy 2 z cos β − 2xyz 2 cos(α + β) ≤ y2 z2 .

Using cos(α + β) = − cos γ, we obtain the desired inequality. Equality holds if


and only if xz cos α + yz cos β − xy = 0 and xz sin α − yz sin β = 0. This takes
the form
1 1 1
: : = sin α : sin β : sin γ.
x y z
We easily infer that the right side of 3.4.9 is a consequence of the present
inequality.

 
3.4.13. In an obtuse-angled triangle, cos2 α > 1 and sin2 α < 2.

Proof: We have
 "  
2 2
IH = R 1− cos α = 9R2 − a2 .


Since the triangle is obtuse, cos α < 0. Hence, IH 2 > R2 . This gives

R2 < IH 2 = 9R2 − a2 .
Geometric Inequalities 121


Thus, a2 < 8R2 . This may be put in the form

sin2 α + sin2 β + sin2 γ < 2.

Equivalently

cos2 α > 1.

 √ 
3.4.14. tan α ≥ 3 3 if the triangle is acute, and tan α < 0 if the
triangle is obtuse.

Proof: Suppose the triangle is acute. In this case tan α, tan β and tan γ are
all positive. We have
"  " 1/3
tan α = tan α ≥ 3 tan α .
 √  
This gives tan α ≥ 3 3. Using tan α = tan α, we get the desired
inequality for acute-angled triangles. Alternatively,
  we can also use the fact
that f (x) = tan x is a convex function on 0, π/2 .
If the triangleis obtuse, say α > 90◦ , then
 tan α < 0 and tan β, tan γ are
positive. Hence, tan α < 0 and this gives tan α < 0.

 √
3.4.15. cot α ≥ 3.

Proof: We have
cos α cos β
cot α + cot β = +
sin α sin β
 
sin α + β
=
sin α sin β
2 sin γ
=
cos(α − β) + cos γ
2 sin γ  
≥ = 2 tan γ/2 .
1 + cos γ

Similarly, we may prove that


   
cot β + cot γ ≥ 2 tan α/2 , cot γ + cot α ≥ 2 tan β/2 .

Adding, we obtain
        
2 cot α + cot β + cot γ ≥ 2 tan α/2 + tan β/2 + tan γ/2 .
122 Inequalities
 
However using the convexity of f (x) = tan x on 0, π/2 , we obtain
   
1       α+β+γ
tan α/2 + tan β/2 + tan γ/2 ≥ tan
3 6
1
= tan 30◦ = √ .
3
It follows that
      √
tan α/2 + tan β/2 + tan γ/2 ≥ 3.

This gives
      √
cot α + cot β + cot γ ≥ tan α/2 + tan β/2 + tan γ/2 ≥ 3.

   √
3.4.16. cot α/2 ≥ 3 3.

Proof: Using the convexity of the function f (x) = cot x on (0, π/2), we get
the desired inequality. The convexity part can be proved using the second
derivative of f (x).

  
 2
     
3.4.17. cot α/2 ≥ cot α/2 cot α .

     
Proof: Let us put cot α/2 = x, cot β/2 = y, cot γ/2 = z. Then we know
that x + y + z = xyz. We have
  
(x + y + z)2 = (x + y + z)xyz = x2 − 1 yz + xy.

This gives   
x2 + y 2 + z 2 + xy = x2 − 1 yz.
It follows that
    
2 x2 + y 2 + z 2 ≥ x2 + y 2 + z 2 + xy = x2 − 1 yz.

We write this in the form


 
x2 − 1
x2 + y 2 + z 2 ≥ xyz .
2x

However, observe that


x2 − 1
= cot α.
2x
Geometric Inequalities 123

Thus, we obtain
 "   
2
  
cot α/2 ≥ cot α/2 cot α
   
 
= cot α/2 cot α .

  √  
3.4.18. cot α ≤ 1/ 3 3 in an acute-angled triangle and cot α < 0 for
an obtuse-angled triangle.

Proof: This easily follows from the fact that


"  √ 
tan α ≥ 3 3 ,

in an acute-angled triangle and this product is negative for an obtuse-angled


triangle (see the proof of 3.4.13).

 
3.4.19. cot2 α ≥ cot α cot β = 1.

Proof: This follows from the rearrangement inequality.


3.4.20. sec α ≥ 6.

Proof: Using cos α ≤ 3/2 (see 3.4.9) and AM-HM inequality, we get
 9 2
sec α ≥  ≥ 9 × = 6.
cos α 3

 √
3.4.21. cosec α ≥ 2 3.

Proof: This follows from the convexity of f (x) = cosec x on (0, π). In fact

f  (x) = cosec x cosec 2 x + 2 cot2 x) ≥ 0,
on (0, π). Hence f (x) = cosec x is convex on (0, π). This gives
 
1 α+β+γ 2
cosec α ≥ cosec = cosec 60◦ = √ .
3 3 3
Hence, we get  √
cosec α ≥ 2 3.
124 Inequalities

3.4.22. cosec 2 α ≥ 4.

Proof: Using the rearrangement inequality, we have


 
cosec 2 α ≥ cosec α cosec β

sin α
= 
sin α
3
≥  2/3 .

sin α

We
 have used
√ the AM-GM inequality at the end. However we know that
sin α ≤ 3 3/8 (see 3.4.7). It follows that
 4
cosec 2 α ≥ 3 × = 4.
3

 2
     
3.4.23. sin α/2 ≤ cos2 α/2 .

Proof: We may arrange the angles such that either α ≥ 60◦ ≥ β ≥ γ or


α ≤ 60◦ ≤ β ≤ γ. This implies that
  
β 1 γ 1
sin − sin − ≥0
2 2 2 2
 
β γ β γ
⇒ 4 sin sin ≥ 2 sin + sin −1
2 2 2 2
"  
α α β γ α
⇒ 1+4 sin ≥ 2 sin sin + sin + 1 − sin .
2 2 2 2 2
However we know that
 
β γ β γ α
2 sin sin = cos − − sin
2 2 2 2 2
α
≤ 1 − sin .
2
This gives
"  
α α β γ β γ
1+4 sin ≥ 2 sin sin + sin + 2 sin sin
2 2 2 2 2 2
  α β
⇒ cos α ≥ 2 sin sin
2 2
  
2 α 2 α α β
⇒ cos ≥ sin +2 sin sin .
2 2 2 2
Geometric Inequalities 125

This simplifies to
 2 
   
sin α/2 ≤ cos2 α/2 .


3.4.24. a2 + b2 + c2 ≥ 4 3Δ (Weitzenböck’s inequality).

Proof: Using Heron’s formula for the area of a triangle, we have

16Δ2 = (a + b + c)(a + b − c)(b + c − a)(c + a − b)


  3
a+b+c
≤ a+b+c
3
4
(a + b + c)
= .
27
This gives
√ (a + b + c)2
4 3Δ ≤ .
3
However we know from the Cauchy-Schwarz inequality that
 
(a + b + c)2 ≤ 3 a2 + b2 + c2 .

It follows that √
a2 + b2 + c2 ≥ 4 3Δ.

Solution 2: We prove a stronger version of the above inequality which is


known as the Hadwiger-Finsler inequality. We show that
√ 
a2 + b2 + c2 ≥ 4 3Δ + (a − b)2 .

Using the cosine rule, we have

a2 = b2 + c2 − 2bc cos α
= (b − c)2 + 2bc(1 − cos α)
= (b − c)2 + 4Δ tan(α/2),

since  
1 − cos α
2bc(1 − cos α) = 4Δ = 4Δ tan(α/2).
sin α
Thus, we get
  
a2 + b2 + c2 = 4Δ tan(α/2) + tan(β/2) + tan(γ/2) + (a − b)2 .
126 Inequalities

However, we know that f (x) = tan x is a convex function on (0, π/2). Hence,
it follows that
 
1  α+β+γ 1
tan(α/2) + tan(β/2) + tan(γ/2) ≥ tan = tan 30◦ = √ .
3 6 3
This shows that √ 
a2 + b2 + c2 ≥ 4 3Δ + (a − b)2 .
We observe that equality holds if and only if a = b = c.


3.4.25. ab + bc + ca ≥ 4 3Δ.

Proof: We know that


1 1 1
Δ= ab sin γ = bc sin α = ca sin β.
2 2 2
Hence, we get
  " 1/3
1 1
ab = 2Δ ≥ 6Δ .
sin α sin α
 1  √ 
However we know from 3.4.7 that sin α ≥ 8/3 3 . This gives

  1/3
8 √
ab ≥ 6Δ √ = 4 3Δ.
3 3


3.4.26. a4 ≥ 16Δ2 .

Proof: Using 3.4.24 and the Cauchy-Schwarz inequality, we get


 1  2 2
a4 ≥ a ≥ 16Δ2 .
3


3.4.27. a2 b2 ≥ 16Δ2 .

Proof: This follows from Heron’s formula and 3.4.26.

 3
2 4Δ
3.4.28. (abc) ≥ √ .
3
Geometric Inequalities 127
  √
Proof: We use a + b + c = 2R sin α + sin β + sin γ ≤ (3 3)R (see 3.4.5) and
abc = 4RΔ. Thus,

4Δ abc 3abc
√ = √ ≥ ≤ (abc)2/3 ,
3 R 3 a+b+c

where we have used the AM-GM inequality in the last step. The result follows
by taking cubes on both the sides.

3.4.29. (Euler’s inequality) 2r ≤ R and equality holds if and only if the


triangle is equilateral.

Proof: We use OI 2 = R(R − 2r).

3.4.30. 9r(r + 4R) ≤ 3s2 ≤ (r + 4R)2 .

Proof: We know that a, b, c are the roots of the cubic equation (see section
3.3.1) t3 − 2st2 + s2 + r2 + 4Rr t − 4Rrs = 0. Since all its roots are real, the
derivative polynomial 3t2 − 4st + s2 + r2 + 4Rr = 0 has only real roots. This
imposes a condition on the discriminant, namely, s2 ≥ 3r(r + 4R). This gives
the left inequality.
Again we know that ra , rb , rc are the roots of the equation (see section
3.3.4) t3 − (r + 4R)t2 + s2 t − s2 r = 0. We look at the derivative polynomial:
3t2 − 2(r + 4R)t + s2 = 0. This again has only real roots. Hence, using the
condition on the discriminant, we obtain 3s2 ≤ (r + 4R)2 , which is the other
inequality.

3.4.31. s2 ≥ 27r2 .

Proof: We have

s (s − a) + (s − b) + (s − c)  1/3
= ≥ (s − a)(s − b)(s − c)
3 3
 2 1/3
Δ
=
s
 2 1/3
= r s .

On cubing the above relation and rearranging the terms, we get the result.


3.4.32. 36r2 ≤ a2 ≤ 9R2 .
128 Inequalities

Proof: Using 3.4.31, we have


 2
a+b+c 3 2
27r2 ≤ ≤ a ,
2 4

where we have used the Cauchy-Schwarz inequality. This gives the left side
inequality. On the other hand, using OH 2 = 9R2 − a2 + b2 + c2 , we get the
right side inequality.

 a2
3.4.33. ≥ 4.
r b rc

Proof: We know (see 3.3.4) that rb rc = s2 . Using the Cauchy-Schwarz
inequality, we get
 a2  2
( a) 4s2
≥  = 2 = 4.
r b rc r b rc s


3.4.34. 5R − r ≥ 3s.

Proof: We use ra = 4R + r (3.3.4) and R ≥ 2r (3.4.29). Thus,

5R − r = 4R + r + R − 2r ≥ 4R + r = ra .

Using ra = Δ/(s − a) = rs/(s − a) and similar expressions, we obtain


  
1
5R − r ≥ ra = Δ
s−a
s 0   1
= 2 ab − a2 .

Thus, it is sufficient to prove that
s 0   1 √
2 ab − a2 ≥ 3s.

This is equivalent to   √
2 ab − a2 ≥ 4 3Δ.

This follows from the Hadwiger-Finsler inequality. (See 3.4.24.)


3.4.35. a(s − a) ≤ 9rR.
Geometric Inequalities 129
 
Proof: We know that r = (s − a) tan α/2 , etc. Hence,
  ar   
a(s − a) =   = 4Rr cos2 α/2 .
tan α/2
   9
Thus, it is sufficient to prove that 2 cos2 α/2 ≤ . This is equivalent to
 2
cos α ≤ 3/2. This follows from 3.4.9.

 √
3.4.36. ha ≤ 3s.

Proof: We have
3abc  
(a + b + c)2 ≥ 3(ab + bc + ca) = ha + hb + hc ) = 6R ha .

Thus, we obtain
 (a + b + c)2 s(a + b + c)
ha ≤ = .
6R 3R
√  √
But we know that a + b + c ≤ 3 3R. It follows that ha ≤ 3s.


1 3
3.4.37. ≤ .
a 2r
Proof: We have
1 
ha
= .
a 2Δ
Thus, it is sufficient to prove that

 3Δ
ha ≤ .
r
This follows from Δ = rs and 3.4.36.


1 3 3
3.4.38. ≥ .
a 2(r + R)

Proof: We know that


"   "   
r = 4R sin α/2 , 1+4 sin α/2 = cos α.

Using a = 2R sin α, etc., we get an equivalent inequality:



 1 3 3
≥ .
sin α cos α
130 Inequalities

This reduces, after cross multiplication, to


   cos α cos β  √
cot α + + ≥ 3 3.
sin β sin α
 √
Since cot α ≥ 3 (3.4.15), it is sufficient to prove
  cos α cos β  √
+ ≥ 2 3.
sin β sin α
This is equivalent to

sin γ √
≥ 2 3.
sin α sin β
By the AM-GM inequality, we have
 sin γ " −1/3
≥3 sin α .
sin α sin β
 √
Using sin α ≤ 3 3/8 (3.4.7), we get the desired inequality.

 9
3.4.39. 9r ≤ ra ≤ R.
2

Proof: We know from the section 3.3.4 that ra = 4R + r. Using 2r ≤ R,
we obtain 9r ≤ 4R + r ≤ (9/2)R.

 √
3.4.40. 9r ≤ ha ≤ 3 s.

Proof: We have

 1 3 √
ha = 2Δ ≤ 2Δ = 3 s,
a 2r
and  9
ha ≥  = 9r.
(1/ha )
(See 3.3.2.)

 
3.4.41. h2a ≤ (3/4) a2 .

Proof: We know that ha ≤ ma , 4m2a = 2b2 + 2c2 − a2 and similar results for
hb , hc , mb , mc . Thus, it follows that
  3 2 
h2a ≤ m2a = a + b2 + c 2 .
4
Geometric Inequalities 131

 a2
3.4.42. ≥ 2.
h2b + h2c

Proof: We write
 a2  a 2 b2 c 2
=  .
h2b 2
+ hc 4Δ b2 + c2
2

Hence, it is sufficient to prove that


 1 1
≥ .
b2 + c 2 2R2
Using the AM-HM inequality, we get
1 9 1
≥  2 ≥ ,
+c b2
2 2 a 2R2
 2 
since a ≤ 9R2 . (This follows from OH 2 = 9R2 − a2 .)

 
3.4.43. ha ≤ wa ≤ 3(R + r).

Proof: The first part is obvious. For the second inequality, we begin with the
expression
2bc  
wa = cos α/2 ,
b+c
and similar
 expressions for wb and wc . Since 2bc ≤ (b + c)2 , we get wa2 ≤
bc cos α/2 , and similar estimates hold for wb and wc2 . Now the Cauchy-
2 2

Schwarz inequality gives,


 2  √   2
wa ≤ bc cos α/2
    
≤ bc cos2 α/2
9
≤ bc ;
4
  
here we have used the estimate cos2 α/2 ≤ 9/4. Hence, it is sufficient to
prove that

bc ≤ 4(R + r)2
  r 2  2
⇐⇒ sin α sin β ≤ 1 + = cos α
  R 
⇐⇒ sin α sin β ≤ cos2 α + 2 cos α cos β
  
⇐⇒ cos α ≤ cos2 α + cos α cos β
  2 
⇐⇒ 2 cos α ≤ cos α + cos2 α.
132 Inequalities

Thus, it is sufficient to establish the quadratic inequality


 2  
cos α − 2 cos α + cos2 α ≥ 0.

If the triangle is obtuse, then cos2 α > 1 (3.4.13) and hence the inequality
follows from the fact that the discriminant is negative. Hence, we may assume

that cos2 α ≤ 1. The above quadratic inequality is true if either cos α ≤
    
1− 1 − cos2 α or cos α ≥ 1+ 1 − cos2 α. However, we know that

cos α > 1 and hence the first alternative does not occur. It is sufficient to
prove that for an acute-angled triangle
  
cos α ≥ 1 + 1 − cos2 α.

We use
 "  "  
1− cos2 α = 2 cos α, cos α = 1 + 4 sin α/2 .

Thus, we have to show that


"   "
8 sin2 α/2 ≥ cos α.

However we know that IH 2 = 2r2 − 4R2 cos α. This gives
"
4R2 cos α ≤ 2r2 ,

and hence
  
" r2 16R2 sin2 α/2 "  
cos α ≤ = = 8 sin2 α/2 .
2R2 2R2
This completes the proof of the desired inequality.

 √
3.4.44. ha hb ≤ 3 3Δ.

Proof: We know (see 3.3.2) that


 2s2 r 2Δs
ha h b = = .
R R
However we also have
1    3√3R
s= a+b+c =R sin α ≤ ;
2 2
(see 3.4.7). Using this estimate, we obtain the required inequality.


3.4.45. ha ≥ 27r3 .
Geometric Inequalities 133

Proof: We begin with the known identity


 1 1
= .
ha r

Hence, the AM-GM inequality gives

" 1   3
1 1 1
≤ = .
ha 3 ha 27r3

Taking the reciprocals, we obtain


"
ha ≥ 27r3 .

 1 3
3.4.46. ≥ .
ha − 2r r
 1 1
Proof: Using = , we get
ha r
  ha r
 
ha − 2r
1=3−2= −2 = .
ha ha ha

Using the AM-GM inequality, we obtain


   
ha − 2r ha
≥ 9.
ha ha − 2r

This gives
 ha
≥ 9.
ha − 2r
But the left hand side is equal to
 2r
3+ .
ha − 2r

Simplification gives
 1 3
≥ .
ha − 2r r


3.4.47. 3Δ ≤ r(4R + r).
134 Inequalities

Proof: We start with the identity, IH 2 = 2r2 − 4R2 cos α. However

"  
4R2 cos α = 2R2 sin2 α − 2
1 2
= a − 4R2
2
1   2 
= a − ab − 4R2
2
1   2  2 
= a − s + r(4R + r) − 4R2
2
1   2  2
= a − 2R + r .
4

Here we have used ab = s2 + r(4R + r); see 3.3.1. Thus, it follows from
IH 2 ≥ 0 that
 
s2 ≤ 4 4R2 + 4Rr + 3r2 .

This gives

 
Δ2 ≤ 4r2 4R2 + 4Rr + 3r2
 
     1
= r2 4R2 + 8/3 Rr + 1/3 r2 + 4Rr + 8r2 .
3

Using 4r2 ≤ 2Rr ≤ R2 , we obtain


2 2
     4
Δ ≤ r 4R + 8/3 Rr + 1/3 r2 + R2
2
3
1 2 2
= r 4R + r .
3
√  
This implies that 3Δ ≤ r 4R + r .

3 
3.4.48. s< ma < 2s.
2

Proof: Let D, E, F be the mid-points of BC, CA, AB respectively. Let A


be the reflection of A in D. Similarly define B  and C  . Then BA CA is a
parallelogram. (See Fig. 3.1.)
Geometric Inequalities 135

Using 2ma = AA < AB + BA = b + c and similar inequalities for 2mb ,
2mc , we get
1 
m a + mb + mc < b + c + c + a + a + b = 2s.
2
Let C  be the point of intersection of BC with a line through A and parallel
to the median BE. Then the sides of AA C  are 2ma , 2mb and 2mc . The
lengths of its medians are 3a/2, 3b/2 and 3c/2. Applying the inequality we
have just proved to this triangle, we get
3   
a + b + c < 2 m a + mb + m c .
2
This gives the left side inequality.


3.4.49. ma ≤ 4R + r.

Proof: Let D, E, F be the mid-points of BC, CA, AB respectively. Join the


circum-centre O to D, E, F . In the triangle ADO, we see that AD ≤ AO +
OD = R + OD, and equality holds if and only if A, O, D are collinear. Thus,
we obtain ma ≤ R + OD. Similarly we have mb ≤ R + OE and mc ≤ R + OF .
Adding, we get 
ma ≤ 3R + OD + OE + OF.
Observe that OD = R cos α, OE = R cos β and OF = R cos γ. Using
 "   r
cos α = 1 + 4 sin α/2 = 1 + ,
R
we get 
 r
ma ≤ 3R + R 1 + = 4R + r.
R

 √ √  √
3.4.50. s< wa ≤ s s − a ≤ 3s.
136 Inequalities

Proof: Let AD (= wa ) denote the bisector of α. Then BD + AD > AB = c


and CD + AD > AC = b. Adding we get
2wa + (BD + CD) > b + c.

This shows that wa > (s − a) and hence s < wa . We also have

2 bc  
wa = s(s − a) ≤ s(s − a).
b+c
Hence, we get  √ √ 
wa ≤ s s−a .
Now using the Cauchy-Schwarz inequality, we have
√ √ √
s − a ≤ 3 s.
Combining all these, we have the required inequalities.

 √
3.4.51. wa2 ≥ 3 3Δ.

Proof: Using Stewart’s theorem or otherwise it is easy to compute that


a2 bc
wa2 = bc − .
(b + c)2
*
Since bc (b + c)2 ≤ 1/4, we obtain
a2
wa2 ≥ bc − .
4
This gives
  1 2
wa2 ≥ ab −a .
4
But we know that  √ 
ab ≥ 4 3Δ + (a − b)2 ;
see 3.4.24. Hence,
 1 2 1   
ab − a = 6 ab − (a − b)2
4 8
1    
= 5 ab + ab − (a − b)2
8
1  √ 
≥ 5 ab + 4 3Δ
8
1 √ √ 
≥ 20 3Δ + 4 3Δ
8√
= 3 3Δ.
Geometric Inequalities 137
 
3.4.52. R + r ≤ max ha , hb , hc .

Proof: This inequality is equivalent to



2Δ 2Δ 2Δ 2Δ
R + r ≤ max , , = .
a b c min{a, b, c}

  it issufficient to prove that min{a, b, c}(R + r) ≤ 2Δ. However R + r =


Hence,
R cos α = OD+OE+OF , where D, E, F are the midpoints of BC, CA, AB
respectively and O is the circum-centre. Thus, the inequality reduces to the
form
min{a, b, c}(OD + OE + OF ) ≤ 2Δ.
However we observe that

min{a, b, c}(OD + OE + OF ) ≤ a · OD + b · OE + c · OF
= 2[BOC] + 2[COA] + 2[AOB]
= 2Δ.

  
3.4.53. a sin α/2 ≥ s.

Proof: We first prove that in a triangle ABC, the following inequality holds:
"   "  
sin α/2 ≤ 1 − sin α/2 . ()

We observe that α, β, γ are the angles of a triangle if and only if (π − α)/2,


(π − β)/2, (π − γ)/2 are angles of another triangle. Hence, it is sufficient to
prove " " 
cos α ≤ 1 − cos α
But we have
r2
(1 − cos α)(1 − cos β)(1 − cos γ) = 8 sin2 (α/2) sin2 (β/2) sin2 (γ/2) = ,
2R2
where R and r are circum-radius and in-radius of the triangle whose angles are
α, β, γ. Since
4R2 cos α cos β cos γ = 2r2 − IH 2 ≤ 2r2 ,
where I and H are the in-centre and ortho-centre of the triangle, we get the
result.
Now, we may write the inequality () in the form
   
sin(α/2) sin π−α
4 
 ≤  π−α .
cos(α/2) cos 4
138 Inequalities

Using the identities


 "  
π−α
sin(α/2) = 1+ sin ,
4
 "  
π−α
cos(α/2) = cos ,
4

we obtain  
sin(α/2) −1 + sin(α/2)
 ≤  .
cos(α/2) cos(α/2)
 
However using r = 4R sin(α/2) and s = 4R cos(α/2), the left side is simply
r/s. Thus, the inequality takes the form

r −1 + sin(α/2)
≤  .
s cos(α/2)

But observe that (s − a) sin(α/2) = r cos(α/2). Thus, we get

   
(s − a) sin(α/2) ≤ s − 1 + sin(α/2) .

This reduces to

a sin(α/2) ≥ s.

 
3.4.54. 3 cos α ≥ 2 sin α sin β.

Proof: We begin with the inequality


 2
cos α − cos β ≥ 0.

Expanding this we get


 
cos2 α ≥ cos α cos β.

But then
   2 
cos2 α ≥ cos α cos β =⇒ cos α ≥3
cos α cos β
 3 
=⇒ 3 cos α cos β ≤ cos α
2
 
=⇒ 2 cos α cos β ≤ cos α
Geometric Inequalities 139
  
=⇒ cos(α − β) + cos(α + β) ≤ cos α
 
=⇒ cos(α − β) ≤ 2 cos α
  
=⇒ cos(α − β) − cos(α + β) ≤ 3 cos α
 
=⇒ 2 sin α sin β ≤ 3 cos α.

We have used the well known result: cos α ≤ 3/2. (3.4.9)

  3
3.4.55. max ra , rb , rc ≥ R.
2
Proof: The required inequality is equivalent to

1 1 1 3R
max , , ≥ .
s−a s−b s−c 2Δ
This is further equivalent to
0      1 3Rr 3R
max tan α/2 , tan β/2 , tan γ/2 ≥ = .
2Δ 2s
This may also be written in the form
  0      1 3
sin α max tan α/2 , tan β/2 , tan γ/2 ≥ .
2
However, we have
  0      1
sin α max tan α/2 , tan β/2 , tan γ/2
     
≥ sin α tan α/2 = 2 sin2 α/2 .
On the other hand,
   3 1
sin2 α/2 = − cos α
2 2
3 1  "  
= − 1+4 sin α/2
2 2
3 1 r
= − 1+
2 2 R
r
= 1− .
2R
Since 2r ≤ R, the result follows. In fact we have proved more:
 
max ra , rb , rc ≥ 2R − r.

3.4.56. R − 2r ≥ wa − ha .
140 Inequalities

Proof: We use
" 4R sin β sin γ
r = 4R sin(α/2), wa = cos(α/2),
sin β + sin γ
and
ha = 2R sin β sin γ.
The inequality to be proved is
" 2 sin β sin γ   
1−8 sin(α/2) ≥   1 − cos (β − γ)/2 .
cos (β − γ)/2
 
Let us put t = cos (β − γ)/2 and x = sin(α/2). Then t ∈ (0, 1] and x ∈ (0, 1).
We have to prove
   
f (x; t) = 2 1 + t x2 − 4t2 x + t + 2t2 − t3 ≥ 0.
Consider this as a function of x, say g(x). We observe that
g  (x) = 4(1 + t)x − 4t2 , g  (x) = 4(1 + t) > 0.
Hence, g has the minimum at x = t2 /(1 + t). But observe that
 2 
t t(1 − t)(t + 3)
f ;t = ≥ 0.
1+t t+1
Hence, minx f (x; t) ≥ 0. Thus, f (x; t) ≥ 0 for all x ∈ (0, 1) and t ∈ (0, 1].
Equality holds if and only if t = 1. Equivalently x = 1/2. This is equivalent to
α = π/3 and β = γ, which corresponds to the case of an equilateral triangle.

  √
3.4.57. s(s − a) + s(s − b) + mc ≤ 3 s.

Proof: Using the AM-GM inequality, observe that



2 (s − a)(s − b) ≤ s − a + s − b = c.
Equality holds if and only if a = b. Thus,
4m2c = 2a2 + 2b2 − c2
= (a + b)2 + (a − b)2 − c2
= (a + b)2 − (b + c − a)(a + c − b)
= (a + b)2 − 4(s − a)(s − b)
    
= a + b + 2 (s − a)(s − b) a + b − 2 (s − a)(s − b)
   √ √ 2 
= a + b + 2 (s − a)(s − b) 2s − s−a+ s−b
 √ √ 2 
≤ 2s 2s − s−a+ s−b .
Geometric Inequalities 141

This implies that


  √ 
s(s − a) + s(s − b) ≤ 2 s2 − m2c .

On the other hand √  2


2 s2 − m2c + mc ≤ 3s2 ,
 2
as this is equivalent to the inequality s2 − 3m2c ≥ 0. Hence, the result
follows.

3.5 Two triangles one inscribed in the other


3.5.1. Let D, E, F be points in the interior of the segments BC, CA, AB
of a triangle ABC. Prove that
 
[DEF ] ≥ min [BDF ], [CED], [AF E] .

Here equality holds if and only if D, E, F are the mid-points of the respective
line segments on which they lie.

Proof: Let α, β, γ, be the areas of the three corner triangles so arranged


that√0 < α ≤ β ≤ γ and let δ = [DEF ]. We prove the stronger statement:
δ ≥ αβ. We normalise the areas such that [ABC] = 1. Let

BD DC CE EA AF FB
= x, = x , = y, = y , = z, = z.
a a b b c c
Let AX and F Y be the altitudes drawn respectively from A and F on to BC.
142 Inequalities

Then we see that


[BDF ] BD · F Y BD F B
= = · = xz  .
[ABC] BC · AX BC AB
This gives [BDF ] = xz  . Similarly, we obtain [CED] = yx , [AF E] = zy  .
Using these we get
 
δ = 1 − xz  + yx + zy 
     
= 1−x 1−z +y 1−x +z 1−y
   
= 1 − x + y + z + xy + yz + zx
   
= 1 − x 1 − y 1 − z + xyz
= x y  z  + xyz.
Now√we consider two cases : γ < 1/4 and γ ≥ 1/4. Suppose γ < 1/4. If

δ < αβ, then δ < γ · γ = γ < 1/4 and hence
1
1=α+β+γ+δ <α+β+ .
2
This gives α + β > 1/2 and in turn we obtain
1 1
< α + β ≤ 2γ < .
2 2

This contradiction proves that δ ≥ αβ, when γ < 1/4.
Suppose on the other hand that γ ≥ 1/4. In this case
δ = x y  z  + xyz
 1/2
≥ 2 xx yy  zz 
 1/2
= 2 αβγ
 √
= αβ · 2 γ

≥ αβ.
This completes the solution.

3.5.2. Let D, E, F be points in the interior of the segments BC, CA, AB


of a triangle ABC. Prove that
 
p(DEF ) ≥ min p(BDF ), p(CED), p(AF E) ,
where p(XY Z) denotes the perimeter of the triangle XY Z. Here again equality
holds if and only if D, E, F are the midpoints of the respective line segments
on which they lie.
Geometric Inequalities 143
     
Proof: For angles x, y, z, let u = tan x/2 , v = tan v/2 , w = tan w/2 .
Then, we have the following results.

(a) If x + y + z = π, then uv + vw + wu = 1. This follows immediately from


the property of tan function.

(b) If P QR and P QS are two triangles with the common side P Q and if
∠RP Q = x1 , ∠RQP = y1 , ∠SP Q = x2 , ∠SQP = y2 , then p(P QR) <
p(P QS) or p(P QR) = p(P QS) or p(P QR) > p(P QS) according as
u1 v1 < u2 v2 or u1 v1 = u2 v2 or u1 v1 > u2 v2 respectively. .
proof: It is easy to prove that

p(P QR) 2 p(P QS) 2


= , = .
PQ 1 − u1 v 1 PQ 1 − u2 v 2

Observe that u1 v1 < 1 and u2 v2 < 1. The result follows from comparison.

Suppose p(AF E) ≥ p(DEF ) and p(F BD) ≥ p(DEF ). We move A and


B towards F , if necessary, while keeping D, E, F fixed until we get p(AF E) =
p(DEF ) = p(F BD). This only increases the perimeter of the triangle CED.
Thus, it is sufficient to prove that p(DEF ) ≥ p(CED) under the assumption
that p(AF E) = p(DEF ) = p(F BD).
Using (a) and (b), we get

uu1 + u1 u2 + u2 u = 1,
vv1 + v1 v2 + v2 v = 1,
ww1 + w1 w2 + w2 w = 1,
u2 u1 = uw, v1 w2 = vw, v 2 u1 = kvu.
144 Inequalities

Further we also have uv + vw + wu = 1. We have to prove that k ≤ 1. We


can get the following relations:

kuv vw uw
u1 = , v1 = , w1 = .
v2 w2 u2

Substituting these in the first three relations, we obtain

kuv kuv
u· + · u2 + u2 u = 1,
v2 v2
vw vw
v· + · v2 + v2 v = 1,
w2 w2
uw uw
w· + · w2 + w2 w = 1.
u2 u2

Solving for u2 and v2 from the last two relations, we get

w2 − v 2 w uw(w + w2 )
v2 = , u2 = .
v(w2 + w) 1 − ww2

Substituting this in the first relation above, we obtain


     
ku2 v 2 1 + w2 w + w2 + w2 − v 2 w u2 w2 + u2 ww2 − 1 + ww2 = 0.

We show that
     
w2 − v 2 w u2 w2 + u2 ww2 − 1 + ww2 + u2 v 2 1 + w2 w + w2
  2
= w 1 + u2 w 2 − v .
Geometric Inequalities 145

We have
     
w2 − v 2 w u2 w2 + u2 ww2 − 1 + ww2 + u2 v 2 1 + w2 w + w2
 
= u2 v 2 w + w2 + u2 w2 w2 + u2 ww22 − w2
+ww22 + v 2 w − v 2 w2 w2
 
= u2 v 2 w + w2 u2 v 2 + v 2 w2 + w2 u2 − 2v 2 w2 w2
+u2 ww22 − w2 + ww22 + v 2 w
0  
= w u2 v 2 − 2vw2 uv + vw + wu
   1
+ u2 + 1 v 2 + w22 − 2vw2 + 2vw2
  2
= w 1 + u2 w 2 − v ;

we have used uv + vw + wu = 1. It follows that


       2
k − 1 u2 v 2 1 + w2 w + w2 + w 1 + u2 w2 − v = 0.

The expression shows that k ≤ 1. This implies that v2 u1 ≤ vu. This is


equivalent to p(DEF ) ≥ p(CED) in view of (b). Equality corresponds to
k = 1. This forces w2 = v and hence v1 = w. Using the symmetry we conclude
u2 = w, w1 = u and v2 = u, u1 = v. Looking at the expressions involving
u, v, w, u1 , v1 , w1 , u2 , v2 , w2 , we conclude that D, E, F are the mid-points of
BC, CA, AB respectively.

3.5.3. Let ABC be an acute-angled triangle and DEF be an acute-angled


triangle inscribed in ABC. Show that

λ ≤ R(DEF ) ≤ μ,

where
 
λ = min R(AF E), R(BDF ), R(CED) ,
 
μ = max R(AF E), R(BDF ), R(CED) .

(Here R(XY Z) denotes the circum-radius of the triangle XY Z.)

Proof: The segment EF is given by EF = 2R(AF E) sin α which is equal to


2R(DEF ) sin ∠F DE. This implies that
R(DEF ) sin α
= .
R(AF E) sin ∠F DE
Similarly, one can obtain
R(DEF ) sin β
= ,
R(BDF ) sin ∠DEF
146 Inequalities

R(DEF ) sin γ
= .
R(CED) sin ∠EF D
However, α + β + γ = π = ∠F DE + ∠DEF + ∠EF D and hence at least one
of the ratios α : ∠F DE, β : ∠DEF , γ : ∠EF D is ≥ 1, and at least one of
these ratios is also ≤ 1. Since all the angles are acute, at least one of the ratios
R(DEF ) : R(AF E), R(DEF ) : R(BDF ), R(DEF ) : R(CED) is greater than
or equal to 1, and at least one of these ratios is less than or equal to 1. This
proves that
λ ≤ R(DEF ) ≤ μ.
Remark: If r(XY Z) denotes the in-radius of triangle XY Z, it is true that
 
r(DEF ) ≥ min r(AF E), r(BDF ), r(CED) .

See [5].

3.5.4. Let ABC be an acute-angled triangle and D, E, F be arbitrary points


on the segments BC, CA, AB respectively. Show that
 
2 a cos α ≥ a cos α,

where a , b , c are the sides of of DEF (i.e., EF = a , F D = b , DE = c ).

Proof: Let BD = x1 , DC = y1 , CE = x2 , EA = y2 , AF = x3 , F B = y3 .
Then

a ≥ a − y3 cos β − x2 cos γ,
b ≥ b − y1 cos γ − x3 cos α,

c ≥ c − y2 cos α − x1 cos β.

Since cos α, cos β, cos γ are positive, the above estimates give
  
a cos α ≥ a cos α − a cos β cos γ.

However,
  
a cos α = 2R sin α cos α = R sin 2α = 4R sin α sin β sin γ,

and
 
a cos β cos γ = 2R sin α cos β cos γ
= 2R cos γ sin(α + β) + 2R cos α cos β sin γ
= 2R sin γ cos γ + 2R cos α cos β sin γ
 
= 2R sin γ cos γ + cos α cos β
= 2R sin α sin β sin γ.
Geometric Inequalities 147

Thus, it follows that

  
a cos α ≥ a cos α − a cos β cos γ
= 4R sin α sin β sin γ − 2R sin α sin β sin γ
= 2R sin α sin β sin γ
1
= a cos α.
2

 
Hence, 2 a cos α ≥ a cos A, as required. It may be seen that equality
holds if and only if D, E, F are mid-points of BC, CA, AB respectively.

3.6 Let P be a point . . .

In this section, we consider inequalities involving a point inside a triangle and


the quantities associated with such a point. We have already seen some such
inequalities. For example, we can take the point P to be the circum-centre
of ABC and then the distance of P from each vertex is equal to R. We have
studied inequalities involving R and other elements of a triangle. Similarly, we
have gone through several inequalities involving I, H or G.
Let us start with a point P inside a triangle ABC. We use the following
notations for elements involving P : R1 = AP , R2 = BP , R3 = CP ; similarly
r1 , r2 , r3 denote respectively the distances of P from BC, CA, AB. Now we
can develop a large number of inequalities involving R1 , R2 , R3 , r1 , r2 , r3 and
other elements of the triangle. We explore some of them below.

3.6.1. Let P be any point in a triangle ABC and s be its semi perimeter.
Prove that

s < R1 + R2 + R3 < 2s.

Proof: Since AB < P A + P B, BC < P B + P C and CA < P C + P A, we get


by adding s < R1 + R2 + R3 .
148 Inequalities

Extend AP, BP, CP to meet opposite sides BC, CA, AB respectively in


D, E, F . Using the triangle inequality, we have BE < BC + CE. Adding
AP both sides and using BE = BP + P E, AP < P E + AE, we obtain
BP + P E + AP < BC + CE + AP < BC + CE + P E + AE so that BP + AP <
BC + CE + AE = BC + CA. Similarly, we obtain AP + CP < BC + AB and
BP + CP < AB + AC. Adding these three inequalities, we get R1 + R2 + R3 <
AB + BC + CA = 2s.
Here is a beautiful extension of this result (It was short-listed in the 40-th
IMO held in Romania during July 2000). The right side inequality can be
strengthened to min{P A, P B, P C} + P A + P B + P C < 2s. This depends
on the following property: If P is any interior point of a convex quadrilateral
ABCD, then P A + P B < AD + DC + CB. The proof is easy. Extend
AP to meet the quadrilateral in Q and suppose, for example, Q lies on DC.
Then P A + P B < P A + P Q + QB ≤ AQ + QB ≤ AQ + QC + CB ≤
AD + DQ + QC + CB = AD + DC + CB.
Let DEF be the medial triangle of ABC. This divides ABC into 4 regions.
Each region is covered by at least two of the three convex quadrilaterals ABDE,
BCEF , and CAF D. If, for example, P lies in quadrilaterals ABDE and
BCEF , then we have P A + P B < AE + ED + DB and P B + P C < BF +
F E + EC. Adding these two we get P B + (P A + P B + P C) < AE + ED +
DB + BF + F E + EC = AB + BC + CA.
As an interesting corollary to this, consider an acute triangle and let P be
its circum-centre (which lies inside ABC). Then P A = P B = P C = R, the
circum-radius of ABC. The inequality takes the form 4R < a + b + c. Using
a = 2R sin A and similar expressions, we get sin α + sin β + sin γ > 2.

3.6.2. Let ABC be a triangle in which a > b > c. Let P be any interior
Geometric Inequalities 149

point and suppose AP, BP, CP meet the opposite sides in D, E, F respectively.
Prove that

P D + P E + P F < a.

Proof: Draw P X parallel to AB, P Y parallel to AC, XK parallel to CF and


Y L parallel to P E. Observe that a is larger than each of AD, BE, CF . Since
XP Y is similar to BAC, we get XY > P D. Similarly, we obtain BX > XK =
P F and CY > Y L = P E. Adding these we get

P D + P E + P F < XY + CY + BX = BC = a.

3.6.3. Let P be a point inside a triangle ABC and let D, E, F be feet of


perpendiculars from P on to the sides BC, CA, AB respectively. Prove that

R1 + R2 + R3 ≥ 2(r1 + r2 + r3 ).

(This is known as the Erdös-Mordell inequality.)

Proof: Suppose L and M are arbitrary points chosen on the sides AB and
AC. Complete the parallelograms LAP S and M AP T . Then LM T S is also a
parallelogram and [LAP S] + [M AP T ] = [LM T S].
Observe that [LAP S] = LA · P F , [M AP T ] = M A · P E ,[LM T S] ≤ LM ·
M T = LM · P A. Here, the points L, M can be chosen on the extended lines
AB and AC. Choosing L, M such that AM = AB and AL = AC, we get

AC · P F + AB · P E ≤ LM · P A.

But, the triangle ALM is similar to ACB and hence LM = CB. Thus, we get
bP F + cP E ≤ aP A. It follows that

b c
PA ≥ P F + P E.
a a

Similarly we get

c a a b
P B ≥ P D + P F, PC ≥ P E + P D.
b b c c
150 Inequalities

Adding these inequalities we get


  a  
c b c a b
PA + PB + PC ≥ + PD + + PE + + PF
b c c a b a
≥ 2(P D + P E + P F ).

Alternate Proof: Observe that AF P E is a cyclic quadrilateral. This gives


F E = AP sin A. Using cosine rule, we also get

F E2 = P E 2 + P F 2 − 2P E cot P F · cos ∠F P E
= P E 2 + P F 2 + 2P E cot P F · cos α
= (P E sin γ + P F sin β)2 + (P E cos γ − P F cos β)2
≥ (P E sin γ + P F sin β)2 .

It follows that
FE P E sin γ + P F sin β
PA = ≥ .
sin α sin α
Similarly we get
P D sin γ + P F sin α P E sin α + P D sin β
PB ≥ , PC ≥ .
sin β sin γ
Adding these we get
   
sin γ sin β sin γ sin α
PA + PB + PC ≥ PD + + PE +
sin β sin γ sin α sin γ
 
sin α sin β
+P F +
sin β sin α
≥ 2(P D + P E + P F );
Geometric Inequalities 151

we have used the AM-GM inequality at the end.


Here are some interesting consequences of the Erdös-Mordell inequality.
(i) Take P = O, the circum-centre of ABC. Then P A = P B = P C = R
and P D = R cos A, P E = R cos β and P F = R cos γ. Thus, we get (see 3.4.9)
 3
cos α ≤ .
2
(ii) Taking P = H, the ortho-centre of ABC, we get P A = 2R cos α,
P B = 2R cos β, P C = 2R cos γ. Hence, we obtain P D = 2R cos β cos γ, P E =
2R cos γ cos α, and P F = 2R cos α cos β. Thus, we get
2(cos α cos β + cos β cos γ + cos γ cos α) ≤ (cos α + cos β + cos γ).
Using the inequality in (i), we obtain
3
cos α cos β + cos β cos γ + cos γ cos α ≤.
4
(iii) Taking P = I, the in-centre, the inequality takes the form
AI + BI + CI ≥ 6.
Using AI = rcosec (α/2), etc., we get

cosec (α/2) ≥ 6.

(iv) Let P = K, the symmedian point. If AK meets BC in D , then it is


known that BD /D C = c2 /b2 and similar ratios for other intersection points.
Using this and Stewart’s theorem, we can compute the symmedian length:
2bc
AD = ma .
b2 + c 2
This implies that
2bc
AK = ma ,
a2
+ b2 + c 2
and similar expressions for BK, CK. Using these we can also get the lengths
of perpendiculars KL, KM, KN on to BC, CA, AB:
2a
KL = [ABC],
a 2 + b2 + c 2
2b
KM = [ABC],
a + b2 + c 2
2

2c
KN = [ABC].
a + b2 + c 2
2

In this case, the Erdös-Mordell inequality takes the form


ma mb mc s
+ + ≥ .
a b c R
152 Inequalities
 1  1  1
3.6.4. ≥2 ≥4 .
r1 r2 R 1 r1 R1 R2

Proof: Reflect P in the sides BC, CA, AB to get points A , B  , C  . Consider


the triangle A B  C  and its interior point P . If we denote the distances of
P from the vertices A , B  , C  respectively by R1 , R2 , R3 and the distances
of P from the sides B  C  , C  A , A B  by r1 , r2 , r3 , then the Erdös-Mordell
inequality applied to triangle A B  C  gives
 
R1 + R2 + R3 ≥ 2 r1 + r2 + r3 .

If D, E are the feet of perpendiculars drawn from P to BC, CA respectively,


then DE = R3 sin γ. Using the fact that D and E are the mid-points of P A
and P B  , we get A B  = 2R3 sin γ. Now let F  be the foot of perpendicular
from P on to A B  . We then have
- .
 2 A P B 
PF = .
A B 
On the other hand
- . 1
A P B  = P A · P B  · sin ∠A P B 
2
1
= (2r1 )(2r2 ) sin(π − γ)
2
= 2r1 r2 sin γ.
*
This gives r3 = 2r1 r2 R3 . Similarly, we may obtain
* *
r1 = 2r2 r3 R1 , r2 = 2r3 r1 R2 .

Moreover
R1 = 2r1 , R2 = 2r2 , R3 = 2r3 .
Using these we get
 1  1
≥2 .
r1 r2 R 1 r1
The same process adopted to the point P and the triangle A B  C  gives
 1  1
≥2 .
R 1 r1 R1 R2

 
3.6.5. R 1 r1 ≥ 2 r1 r 2 .
Geometric Inequalities 153

Proof: We refer to Fig. 3.6. Taking L = B and M = C, there, we obtain

BA · P F + CA · P E ≤ BC · P A.

This reduces to cr3 + br2 ≤ aR1 . This may be written in the form

c b
R 1 r1 ≥ r3 r1 + r2 r1 .
a a
Similarly, we obtain
a c
R 2 r2 ≥ r 1 r2 + r3 r2
b b
b a
R 3 r3 ≥ r2 r3 + r 1 r3 .
c c
Adding these three inequalities, we get
   
b a
R 1 r1 ≥ r1 r2 +
a b

≥ 2 r1 r2 ;

we have used the AM-GM inequality. Equality holds if and only if ABC is
equilateral and P is its centre.
Consequences:

(i) Taking P = I, the in-centre of ABC, we have r1 = r2 = r3 = r and


     
R1 = r cosec α/2 , R2 = r cosec β/2 , R3 = r cosec γ/2 .

We obtain   
cosec α/2 ≥ 6.

(ii) Taking P = O, the circum-centre of ABC, we get


 
cos α ≥ 2 cos α cos β.

(iii) If P = H, the ortho-centre, then

R1 = 2R cos α, R2 = 2R cos β, R3 = 2R cos γ,

and

r1 = 2R cos β cos γ, r2 = 2R cos γ cos α, r3 = 2R cos α cos β.


 3
The inequality we obtain is cos α ≤ . (See 3.4.9.)
2
154 Inequalities

(iv) If we take P = G, the centroid, then the inequality takes the form
 
ma h a ≥ ha hb .

Adopting the method developed in 3.6.4, we obtain


 
R1 R2 ≥ 2 R 1 r1 .

Repeating this process once more, we also get


 1  1
≥2 .
r1 R1

3.6.6. Let P be a point inside a triangle ABC and suppose AP , BP , CP


extended meet BC, CA, AB respectively in D, E, F . Prove that
AP BP CP
1. + + ≥ 6;
PD PE PF
AP BP CP
2. · · ≥ 8;
PD PE PF
AD BE CF 9
3. + + ≥ ;
AP BP CP 2
AD BE CF
4. · · ≥ 27;
PD PE PF
PD PE PF 3
5. + + ≥
PA PB PC 2
Proof: Let us take
BD x CE z AF y
= , = , = .
DC y EA x FB z
ax ay cy cz
Then we see that BD = , CD = , AF = and F B = .
x+y x+y z+y y+z
Using Menelaus’ theorem, we obtain
AP BC AF x+y
= · = ,
PD CD F B z
BP z + x CP y+z
and similar expressions:
= , = . Adding these we get
PE y PF x

AP BP CP x z  y x
 
z y

+ + = + + + + +
PD PE PF z x x y y z
≥ 2 + 2 + 2 = 6.
Geometric Inequalities 155

Similarly,
AP BP CP (x + y)(y + z)(z + x)
· · =
PD PE PF xyz
√ √ √
2 xy · 2 yz · 2 zx
≥ = 8.
xyz
We also observe that
AD AP x+y+z AD x+y+z
=1+ = , = .
PD PD z AP x+y
Thus,
 
AD BE CF 1 1 1
+ + = (x + y + z) + +
AP BP CP x+y y+z z+x
9
≥ .
2
And
AD BE CF (x + y + z)3
· · = ≥ 27.
PD PE PF xyz
Finally
PD PE PF
+ +
PA PB PC
z x x
= + +
x+y y+z y+z
 
1 1 1
= (x + y + z) + + −3
x+y y+z z+x
9 3
≥ −3= .
2 2

3.6.7. Prove
 ha
(i) ≥ 9;
r1
(ii) ha hb hc ≥ 27r1 r2 r3 ;
   
(iii) ha − r1 hb − r2 hc − r3 ≥ 8r1 r2 r3 ;
  
(iv) min ha , hb , hc ≤ r1 + r2 + r3 ≤ max{ha , hb , hc ;
 r1 3
(v) ≥ .
ha − r 1 2
156 Inequalities

Proof: Let D, E, F respectively be the points where AP , BP , CP meet BC,


CA, AB. Put BD : DC = z : y, CE : EA = x : z and AF : F B = y : x.

We easily compute
AP y+z BP z+x CP y+x
= , = , = .
PD x PE y PF z
Thus, we obtain
AD x+y+z BE x+y+z CF x+y+z
= , = , = .
PD x PE y PF z
Using similar triangles P DL and ADT , we have
ha AT AD x+y+z
= = = .
r1 PL PD x
Similarly, we may obtain
hb x+y+z hc x+y+z
= , and = .
r2 y r3 z
Now
 ha   
1 1 1
= x+y+z + +
r1 x y z
≥ 9.

This proves (i). Similarly,

ha hb h c (x + y + z)3
· · = ≥ 27,
r1 r 2 r3 xyz
Geometric Inequalities 157

giving (ii). We also observe


       
h a − r 1 h b − r2 h c − r 3 ha hb hc
= −1 −1 −1
r1 r2 r3 r1 r2 r3
   
z+y x+z y+x
=
x y z
z+y x+z y+x
= √ · √ · √
yz xz yx
≥ 2 · 2 · 2 = 8,

proving (iii). Using the same expressions, we also get


 r1
= 1.
ha
This implies (iv). Finally, we have
r1 PD x
= = ,
ha − r 1 AP y+z
r2 r3
and similar expressions for, . Thus, we obtain
hb − r 2 hc − r3
 r1  x   
1 1 1 9 3
= = x+y+z + + −3 ≥ −3 = .
ha − r 1 y+z y+z z+x x+y 2 2

We thus obtain (v).

3.6.8. Let P be an interior point of a triangle ABC. Prove


 √ 
1 1 1 3 3+2 3
+ + ≥ ,
rA rB rC R+r
where rA , rB , rC denote the in-radii of triangles P BC, P CA, P AB respec-
tively.

Proof: We refer to the figure Fig. 3.7. We have


a · PL PL PD
rA = , = .
a + PB + PC ha AD
Thus, we get
1 a + P B + P C AD
= · .
rA a · ha PD
Suppose BD : DC = z : y, CE : EA = x : z and AF : F B = y : x. Then
AD x + y + z BE x + y + z CF x+y+z
= , = , = .
PD x PE y PF z
158 Inequalities

We hence obtain
(y + z)2
P A2 = AD2 .
(x + y + z)2
Using Stewart’s theorem, we calculate AD;
(y + z)yb2 + (y + z)zc2 − yza2
AD2 = .
(y + z)2
Thus, we obtain
(y + z)yb2 + (y + z)zc2 − yza2
P A2 = .
(x + y + z)2

If we use the standard formula a2 = b2 + c2 − 2bc cos α, this expression reduces


to
y 2 b2 + z 2 c2 + 2bcyz cos α
P A2 = .
(x + y + z)2
This may be written in the form

2 h2a  2  2
PA = z cot γ − y cot β + z + y .
(x + y + z)2

Similar expressions may be obtained for P B 2 and P C 2 :



h2b  2  2
P B2 = x cot α − z cot γ + x + z ,
(x + y + z)2

2 h2c  2  2
PC = y cot β − x cot α + y + x .
(x + y + z)2
Observe that
hb   hc  
PB ≥ x+z , PC ≥ y+x .
x+y+z x+y+z
Thus, we obtain

1 x + y + z a + PB + PC
=
rA 2Δ x
(x + y + z)a hb (x + z) hc (y + x)
≥ + + .
2Δx 2Δx 2Δx
Similarly, we may obtain
1 (x + y + z)b hc (y + x) ha (z + y)
≥ + + ,
rB 2Δy 2Δy 2Δy
and
1 (x + y + z)c ha (z + y) hb (x + z)
≥ + + .
rC 2Δz 2Δz 2Δz
Geometric Inequalities 159

Adding, we have
    
1 1 1 x+y+z a b c 1  1 1
+ + ≥ + + + ha z + y +
rA rB rC 2Δ x y z 2Δ z y
x+y+z  a 2 
≥ + ha .
2Δ x Δ
However, we know that (see 3.4.38)

ha + h b + h c 1 3 3
=2 ≥ .
Δ a R+r
We also observe that
x+y+z  a 9 
≥ abc)1/3
2Δ x 2Δ
and (3.4.28)
 3
 2 4Δ
abc ≥ √ .
3
√  2
Using 3.4.47, we can prove that 3Δ ≤ R + r . Combining all these, we get
 √ 
1 1 1 3 3+2 3
+ + ≥ ,
rA rB rC R+r

3.6.9. Let P be a point in a triangle ABC. Then


 R1 R2
≥ 1.
ab
Equality holds if and only if either P is one of the vertices or P = H, the
ortho-centre of ABC.

Proof: Let us consider the vertices of ABC to be points in the complex plane;
let us denote A, B, C by complex numbers z1 , z2 , z3 respectively. Let us write
z for P . Now consider the function
  
 z − z1 z − z2
g(z) =   .
z3 − z 1 z 3 − z 2
     
This is a quadratic polynomial and g z1 = g z2 = g z3 = 1. Hence, g(z) = 1
for all z. Thus, we have
  
    z − z1   z − z 2   R1 R2
1 = g(z) ≤   =
 z3 − z1   z 3 − z 2  .
ab
160 Inequalities

Equality holds if and only if z = z1 or z = z2 or z = z3 or


     
z − z1 z − z2 z − z2 z − z 3
arg    = arg   
z3 − z 1 z3 − z 2 z1 − z2 z1 − z3
  
z − z3 z − z 1
= arg   .
z2 − z3 z2 − z1
The last case corresponds to P = H.
Consequences:
(i) Let us take P = I, the in-centre. Then

R1 = rcosec (α/2), R2 = rcosec (β/2), R3 = rcosec (γ/2).

Then it is easy to calculate


 R1 R2 1
= a sin(α/2).
ab s
We hence obtain 
a sin(α/2) ≥ s.
(For a direct proof see 3.4.53.)
(ii) Taking P = G, the centre of gravity, we have R1 = 2ma /3, R2 = 2mb /3
and R3 = 2mc /3. Thus, we obtain the inequality
 ma mb 9
≥ .
ab 4

(iii) Taking P = O, the circum-centre, we see that R1 = R2 = R3 = R and


the inequality is
1 1
≥ 2,
ab R
or a + b + c ≥ 4Δ/R.
(iv) Let us take P = K, the symmedian point. If D, E, F are respectively
the points where AP , BP , CP meet the sides BC, CA, AB, then it is easy to
check that
BD c2 CE a2 AF b2
= 2, = 2, = 2.
DC b EA c FB a
Using Stewart’s theorem, we can calculate R1 ; we get
2bc
R1 =  2 ma ,
a
and similar expressions for R2 and R3 . The inequality in this case is
 2 2
 a + b2 + c 2
c2 ma mb ≥ .
4
Geometric Inequalities 161

(v) When P = Ω1 , the first Brocard point, the sine rule gives
b c c
R1 = sin ω, R2 = sin ω, R3 = sin ω,
sin α sin β sin γ
where ω is the Brocard angle. We then get
 a sin α 
sin α
≥ .
b sin2 ω
Using the known fact that ω ≤ π/6, we have sin ω ≤ 1/2 and hence
 a sin α "
≥4 sin α.
b

3.6.10. Let P be a point in a triangle ABC. Then


 r1 r2 1
≤ .
ab 4
Equality holds if and only if P is the circum-centre of ABC.

Proof: As in 3.6.8, we produce AP , BP , CP to meet BC, CA, AB respec-


tively in points D, E, F . Let BD : DC = z : y, CE : EA = x : z and
AF : F B = y : x. An easy computation shows that
xha yhb zhc
r1 = , r2 = , r3 = .
x+y+z x+y+z x+y+z
Thus, we get
r1 r2 xy sin2 γ
= .
ab (x + y + z)2
The required inequality is
1 0 1 1
xy sin2 γ + yzsin2 α + zx sin2 β ≤ .
(x + y + z)2 4
This simplifies to  
x2 + 2xy cos 2γ ≥ 0.
Writing
 
cos 2α = cos 2β + 2γ
= cos 2β cos 2γ − sin 2β sin 2γ,

the inequality reduces to


 2  2
x + y cos 2γ + z cos 2β + y sin 2γ − z sin 2β ≥ 0,
162 Inequalities

which is trivially true. Equality holds if and only if


x y z
= = .
sin 2α sin 2β sin 2γ
This corresponds to P = O, the circum-centre.
Consequences:
(i) Take P = G. Then r1 = ha /3, r2 = hb /3 and r3 = hc /3. Then we have
 r1 r2 1  ha h b 1 2
= = sin α.
ab 9 ab 9
Thus, the inequality is
 9
sin2 α ≤ ;
4
or equivalently a2 + b2 + c2 ≤ 9R2 .
(ii) Let P = I. Then r1 = r2 = r3 and the inequality takes the form
1 1
≤ 2.
ab 4r
This is equivalent to 2r ≤ R.
(iii) Consider P = H, the ortho-centre. Then

r1 = 2R cos β cos γ, r2 = 2R cos γ cos α, r3 = cos α cos β.

Using these values we may write the inequality in the form


 1" 1
sin 2α ≤ tan α = tan α.
2 2

(iv) If P is the symmedian point, then

r1 a2
= 2 ,
ha a + b2 + c 2
and similar expressions for r2 and r3 . We thus get
 r1 r2 12Δ2
=  2 2.
ab ( a )
The inequality now is  √
a2 ≥ 4 3Δ.

(v) Suppose P = Ω1 , the first Brocard point. In this case


b c a
r1 = sin2 ω, r2 = sin2 ω, r1 = sin2 ω.
sin α sin β sin γ
Geometric Inequalities 163

We obtain  2
sin α
sin4 ω ≤  2 .
4 sin α sin2 β
Equivalently
1 1
sin4 ω ≤ 2
 .
16R 1/a2

However using Cauchy-Schwarz inequality and 3.4.38, we get

 1  2
1 1 9
≥ ≥ .
a2 3 a 4(R + r)2

Hence, the inequality is


R+r
sin2 ω ≤ .
6R
This is better than sin ω ≤ 1/2 in view of 2r ≤ R.

3.6.11. Let P be a point in a triangle ABC. Prove that


 2 √ - .
R1 + R2 + R 3 ≥ 4 3 ABC .

Proof: Draw AT parallel and equal to P B; BR parallel and equal to P C; and


CS parallel
- . and equal to P A. Then the area of the hexagon AT
 BRCS is equal
to 2 ABC , and perimeter of AT BRCS is 2 R1 + R2 + R3 . But among all
hexagons of fixed areas, a regular hexagon has the least perimeter. If a regular
hexagon √is inscribed in a circle of radius R, then its perimeter is 6R and its
area is 3 3R2 /2. Thus, we obtain
2
  2K
2 R1 + R2 + R3 ≥ 6R = 6 √ ,
3 3

where K is the area- of the. hexagon. Since we are minimising the perimeter of
a hexagon of area 2 ABC , we obtain
2 - .
  2 · 2 ABC
2 R1 + R2 + R 3 ≥ 6 √ .
3 3

This simplifies to
 2 √ - .
R 1 + R2 + R 3 ≥ 4 3 ABC .
164 Inequalities

3.6.12. Let P be an interior point of a triangle ABC and D, E, F be the


feet of perpendiculars from P on to BC, CA, AB respectively. Show that
- . 1- .
DEF ≤ ABC ,
4
and equality holds if and only if P coincides with the circum-centre of ABC.

Proof:

Let P D = r1 , P E = r2 and P F = r3 . Then


- . 1
DP E = r1 r2 sin ∠DP E
2
1  
= r1 r2 sin π − γ
2
1
= r1 r2 sin γ
2
1
= cr1 r2 ,
4R
- . - .
and similar expressions for EP F , F P D may be obtained. Adding these,
one obtains
- . 1  
DEF = cr1 r2 + ar2 r3 + br3 r1
4R
abc  r1 r2
=
4R ab
1- .
≤ ABC ,
4
using (3.6.10).
Geometric Inequalities 165

Consequences: Using the concyclicity of A, F, P, E, it is easy to get EF =


AP sin A. Similarly, we get F D = BP sin B and DE = CP sin C. If R , s
and r denote respectively the circum-radius, semi-perimeter and in-radius of
triangle DEF , then

EF · F D · DE R R1 sin α
R = - . = 
4 DEF r1 r 2 c
R1 sin α + R2 sin β + R3 sin γ
s =
- . 2

DEF r r c

r = 
=  1 2 .
s 2R R1 sin α
 2
Now the inequality s ≥ 27 r gives
 2  2
R1 sin α 27 r1 r2 c
≥  2 .
4
4R2 R1 sin α

This simplifies to √
 2
3 3
R1 sin α ≥ r1 r2 c.
R
Using a = 2R sin α, etc., this may be written in the form
 2 √ 
R1 sin α ≥ 6 3 r1 r2 sin γ.

However,
  
R1 sin α)2 = EF 2 = r22 + r32 − 2r1 r2 cos π − α
= r22 + r32 + 2r1 r2 cos α,

and similar expressions for the other two. Thus, we deduce


√   2
6 3 r1 r2 sin γ ≤ R1 sin α
 2
≤ 3 R1 sin α
 
= 3 r22 + r32 + 2r2 r3 cos α
  
= 6 r12 + r2 r3 cos α ,

or √   
3 r1 r2 sin γ ≤ r12 + r2 r3 cos α.
Suppose P = I, the in-centre of ABC. Then r1 = r2 = r3 = r and hence
√ 2   
3 r sin α + sin β + sin γ ≤ 3r2 + r2 cos α + cos β + cos γ ,
166 Inequalities

or which is the same as


√   
3 sin A ≤ 3 + cos A.

If we choose P = O, the circum-centre, then r1 = R cos α, r2 = R cos β,


r3 = R cos γ and hence,
√   "
3 cos α cos β sin γ ≤ cos2 α + 3 cos α.

Using
the 2standard identities
 4 cos α cos β sin γ = sin 2α + sin 2β − sin 2γ, etc.,
and sin α = 2 + 2 cos α, this reduces to
2  
sin 2α + sin 2β + sin 2γ ≤ √ sin2 α + sin2 β + sin2 γ .
3

3.6.13. (AMM, 1978) Let G denote the centroid of a triangle ABC and let
θ = ∠GAB, μ = ∠GBC and ν = ∠GCA. Prove that
3
sin θ + sin μ + sin ν ≤ .
2
Proof: Observe that
Δ Δ Δ
sin θ = , sin μ = , sin ν = .
cma amb bmc
Thus, 
1 1 1
sin θ + sin μ + sin ν = Δ + + .
cma amb bmc
Using the Cauchy-Schwarz inequality, we have
    
 2 1 1
sin θ + sin μ + sin ν ≤ Δ2
a2 m2a
  
 2 2  2 2
a b ma mb
= Δ2  2 .
abcma mb mc
 
Using the standard identities 16Δ2 = 2 a2 b2 − a4 , 4m2a = 2b2 + 2c2 − a2 ,
etc., the above inequality reduces to
4 2 P
sin θ + sin μ + sin ν ≤ ,
9 Q
where
     2
P = 2 a 2 b2 − a4 a 2 b2 ,
   
Q = a2 b2 c2 2a2 + 2b2 − c2 2b2 + 2c2 − a2 2c2 + 2a2 − b2 .
Geometric Inequalities 167

Thus, it is sufficient to verify the inequality P/Q ≤ 1. For this, it is sufficient


to prove that Q = P + R for some non-negative quantity R. Consider Q − P
as a function of the variables x = a2 , y = b2 , z = c2 , say, Q − P = f (x, y, z).
Thus,
   
f (x, y, z) = xyz 2x + 2y − z 2y + 2z − x 2z + 2x − y
  2
− 2xy + 2yz + 2zx − x2 − y 2 − z 2 y 2 + 2yz .

It is easy to verify that f (y, y, z) = 0 and fx (y, y, z) = 0. It follows that


(x − y)2 is a factor of f (x, y, z). The symmetry of f (x, y, z) in x, y, z shows
that (x − y)2 (y − z)2 (z − x)2 is a factor of f (x, y, z). Looking at the degrees,
it may be concluded that

f (x, y, z) = K(x − y)2 (y − z)2 (z − x)2 .


-
Now K may be seen equal to 1. Thus, it follows that Q − P = (x − y)(y −
.2
z)(z − x) = R ≥ 0 and hence P ≤ Q. Equality holds if and only if x = y = z
which is equivalent to a = b = c. Thus, equality holds if and only if ABC is
equilateral.
A Generalisation
The above inequality may be written in the form
ha hb hc
+ + ≤ 3.
mb mc ma
 
However,
 a more
 general inequality holds. Let h1 , h2 , h3 be any permutation
of ha , hb , hc . Then the inequality
h1 h2 h3
+ + ≤ 3,
ma mb mc
is true. Suppose a ≤ b ≤ c. Then it is easy to see that ha ≥ hb ≥ hc and
ma ≥ mb ≥ mc . Hence, the rearrangement inequality implies that
ha hb hc h1 h2 h3 ha hb hc
+ + ≤ + + ≤ + + .
ma mb mc ma mb mc mc mb ma
Thus, it is sufficient to prove that
ha hb hc
+ + ≤ 3.
mc mb ma
Again the Cauchy-Schwarz inequality gives
 2     
ha hb hc 2 1
+ + ≤ ha ,
mc mb ma m2a
and the rest is as in the early part.
168 Inequalities

3.6.14. Prove the inequality


 r2 + r 3 1  R1
≤1≤ .
r2 + 2R1 + r3 3 r2 + r3
Find conditions on P and triangle ABC for equality to hold in the above
inequality.

Proof: Let P D, P E, P F be the perpendiculars from P on to the sides BC,


CA, AB respectively. Thus, AP = R1 , BP = R2 , CP = R3 , P D = r1 ,
P E = r2 and P F = r3 . It is easy to see that a · R1 ≥ b · r2 + c · r3 , with equality
if and only if AP is perpendicular to BC. If we reflect P in the bisector of
∠BAC, we also get a · R1 ≥ b · r3 + c · r2 . Adding these we get
b + c 
R1 ≥ r 2 + r3 .
2a
Equality holds if and only if AP is is perpendicular to BC and bisects ∠BAC.
This gives
1  s 
R1 + r 2 + r 3 ≥ r2 + r3 ,
2 a
where s is the semi-perimeter of the triangle ABC. Dividing by r2 + r3 , we
obtain
R1 1 h1
+ ≥ ,
r2 + r 3 2 2r
where r is the in-radius and h1 is the altitude from A on to BC. This simplifies
to
r 2 + r3 r
≤ .
r2 + 2R1 + r3 h1
Using
1 1 1 1
+ + =
h1 h2 h3 r
we get the left hand side of the required inequality. We also observe that
R1 b+c
≥ .
r 2 + r3 2a
Hence, we get
 R1  b a

≥ + ≥ 3,
r2 + r3 2a 2b
which gives the right hand side. Equality holds if and only if ABC is equilateral
and P is its centre.

3.6.15. Prove that  


R12 sin2 α ≤ 3 r12 ,
and equality holds if and only if P is the symmedian point of ABC.
Geometric Inequalities 169

Proof: Consider the cyclic quadrilateral AEP F . Ptolemy’s theorem gives

AP · EF = r3 AE + r2 AF
= R1 sin θ2 · R1 cos θ1 + R1 sin θ1 · R1 cos θ2
 
= R12 sin θ1 + θ2 = R12 sin α,

where ∠P AB = θ2 and ∠CAP = θ1 .(See Fig. 3.9.)

Thus, F E = R1 sin α. This gives



R12 sin2 α = EF 2 + F D2 + DE 2 .

Consider the triangle DEF and the point P inside it. We show that
 
DE 2 + EF 2 + F D2 ≤ 3 P D2 + P E 2 + P F 2 , (3.11)

valid for any triangle DEF and any point P inside it. Let K be the mid-point
of EF . Join P K and DK. (See Fig. 3.10.) Appolonius’ theorem applied to
the triangle P EF gives
1
PK = 2P E 2 + 2P F 2 − EF 2 .
2
Similarly,
1
DK = 2DE 2 + 2DF 2 − EF 2 .
2
But DK ≤ P K + KD, with equality if and only if P lies on DK. Thus,
 
2DE 2 + 2DF 2 − EF 2 ≤ 2P D + 2P E 2 + 2P F 2 − EF 2 .
170 Inequalities

Squaring and simplification gives

2DE 2 + 2DF 2 ≤ 4P D2 + 2P E 2 + 2P F 2

+P D 2P E 2 + 2P F 2 − EF 2 .

Summing over the cyclic permutation of D, E, F , this leads to


 
DE 2 + EF 2 + F D2 ≤ 2 P D2 + P E 2 + P F 2
 
+ P D 2P E 2 + 2P F 2 − EF 2 .
 
If DE 2 + EF 2 + F D2 ≤ 2 P D2 + P E 2 + P F 2 , then (3.11) is trivially valid.
Otherwise, we have
 
0 ≤ DE 2 + EF 2 + F D2 − 2 P D2 + P E 2 + P F 2
 
≤ P D 2P E 2 + 2P F 2 − EF 2
 1/2    1/2
2 2 2
≤ PD 4 PD − DE .

 
Setting P D2 = x2 and DE 2 = y 2 , the above relation is
 1/2
0 ≤ y 2 − 2x2 ≤ x 4x2 − y 2 .
 
This is equivalent to y 2 y 2 − 3x2 ≤ 0. It follows that y 2 ≤ 3x2 , which is
precisely (3.11). Equality holds if and only if P coincides with the centroid of
DEF . This is equivalent to

[P DE] = [P EF ] = [P F D],
Geometric Inequalities 171

or
r1 r2 sin γ = r2 r3 sin α = r3 r1 sin β.
This shows that
r1 : r2 : r3 = a : b : c.
Thus, equality holds if and only if P is the symmedian point.
Chapter 4
Applications involving inequalities
Some problems, though not direct inequality problems, use inequalities
in their solutions. This is the case when the problems on maximisation and
minimisation are considered. Inequalities are also useful in solving some Dio-
phantine equations by way of estimating the bounds for solutions. They also
help us to determine whether some polynomial equations have real roots. Here
we consider several problems whose solutions use inequalities.

Example 4.1.  (IMO, 1987) Let x1 ,x2 ,x3 ,. . .,xn be a sequence of n real num-
n 2
bers such that j=1 xj = 1. Prove that for every k ≥ 2, there are integers
a1 ,a2 ,a3 ,. . .,an , not all zero, such that |aj | ≤ k − 1 and

 
a1 x1 + a2 x2 + · · · + an xn  ≤ (k − 1) n .
kn − 1
 
Solution: For each n-tuple a1 ,a 2n,a3 ,. . .,an of non-negative integersn with
0 ≤ aj ≤ k − 1, associate the sum j=1 aj xj . Observe that there are k such
n-tuples. Using the Cauchy-Schwarz inequality, we obtain
   1/2   1/2
 n  n n
 a x  ≤ a 2
x 2
 j j  j j
j=1 j=1 j=1
 √
≤ k − 1 n.
- √ .
Now, we  split√the*interval  0, (k − 1 n into k n − 1 sub-intervals each of
length  k − 1 n kn − 1 . By the pigeonhole  principle, we can find two n-
n
tuples b1 ,b2 ,b3 ,. . .,bn and c1 ,c2 ,c3 ,. . .,cn such that the sums j=1 bj xj and
n
j=1 cj xj lie in the same sub-interval. Hence, it follows that
    √
 n n
 k−1 n
 bj x j − 
c j xj  ≤  n  .
 k −1
j=1 j=1
 
Taking aj = bj − cj , we see that aj  ≤ k − 1 for 1 ≤ j ≤ n and
 n  √
  (k − 1) n
 
aj xj  ≤ .
 kn − 1
j=1

Example 4.2. Find the maximum value of


        
x 1 − y 2 1 − z 2 + y 1 − z 2 1 − x2 + z 1 − x2 1 − y 2 ,
174 Inequalities

subject to the conditions that x, y, z are non-negative and xy + yz + zx = 1.

Solution: Observe that


        
x 1 − y2 1 − z2 = x− xy 2 + xz 2 + xyz yz
cyclic cyclic cyclic cyclic
   
2 2
= x− x y + x z + xyz
cyclic cyclic
   
= x− x xy + xz + xyz
cyclic cyclic
   
= x− x 1 − yz + xyz
cyclic cyclic
= 4xyz.

Now the AM-GM inequality gives


 2/3 1  1
xyz ≤ xy + yz + zx = .
3 3
Thus, it follows that
    4
x 1 − y 2 1 − z 2 = 4xyz ≤ √ .
cyclic
3 3

Here equality
√ holds if and only if x = y = z = 1/ 3. Thus, the maximum
value is 4/3 3.

Example 4.3. (USSR, 1974) Consider a square grid S of 169 points which are
uniformly arranged in 13 rows and 13 columns. Show that no matter what subset
T consisting of 53 points is selected from these 169 points, some four points of T
will always form the vertices of a rectangle R whose sides are parallel to the sides
of S.

Solution: Let mj , 1 ≤ j ≤ 13, denote the number of elements of T in the


j-th row. Then
m1 + m2 + · · · + m13 = 53.
Now to get a rectangle R from the points of T , with sides parallel to those of
S, we must get some two points from some j-th row and some two points from
some k-th row such that the columns determined by the two points on the j-th
row coincide with the columns determined by the two points on the k-th row.
Nowthenumber of pairs of columns determined by mj points on the j-th
mj
row is . Hence, the total pairs of columns collectively determined by
2
Applications 175

13 
 
mj
points of T is . We observe that
j=1
2
13 
  13
mj 1  2 
= m j − mj
j=1
2 2 j=1

13  13
1 1
≥ m2j − mj
2 × 13 j=1
2 j=1
1  2 1  
= 53 − 53
26 2
> 2 × 53 − 27
 
13
= 79 > 78 = .
2
We have used the Cauchy-Schwarz inequality in the  second step above. The
total number of pairs of columns in the grid S is 13 2 . Hence, we may apply
the pigeonhole principle to conclude that there is a pair of columns determined
by some two points on some j-th row which coincides with a pair of columns
determined by some two points on some k-th row, j = k. We thus get a
rectangle with sides parallel to the sides of S.
 
Example 4.4. Consider the set S = a1 , a2 , . . . , an . Let C = {P1 , P2 , . . . , Pn }
be a collection of n  2-element subsets of S such that whenever Pj ∩ Pk
 distinct
is not empty then aj , ak is also one of the sets in the collection C . Prove that
each aj appears exactly in two elements of C.

Solution: For each j, let mj denote the number of Pr ’s in which aj lies. We


  that mj = 2 for each j, 1 ≤ j ≤ n. Since Pj is a 2-element set,
have to show
we have Pj  = 2 for 1 ≤ j ≤ n. Hence, we obtain
m1 + m2 + · · · + mn = 2n.
 
Suppose aj0∈ Pk ∩ Pl . Then 1 the given condition implies that ak , al is in the
 
collection P1 , P2 , . . . , Pn . Thus, for each j, there are m2j number of Pr ’s
0 1
being contributed to the collection P1 , P2 , . . . , Pn . Moreover all such sets
 
are distinct. In fact if a set ak , a l corresponds
 to as and at , then both as
and at are in Pk ∩ Pl forcing Pk = as , at = Pl . It follows that
n  
mj
≤ n.
j=1
2

This simplifies to

n
 
m2j − mj ≤ 2n.
j=1
176 Inequalities
n
Since j=1 mj = 2n, we obtain


n
m2j ≤ 4n.
j=1

However an application of the Cauchy-Schwarz inequality gives



n 2 
n
4n2 = mj ≤n m2j ≤ 4n2 .
j=1 j=1

This shows that equality holds in the


nCauchy-Schwarz inequality. We conclude
mj = λ, for 1 ≤ j ≤ n. Now using j=1 mj = 2n, we obtain λ = 2. This gives
mj = 2, for 1 ≤ j ≤ n.

Example 4.5. Let ABC be a triangle in which C = 90◦ . Let ma , mb , mc be


its medians from the vertices A, B, C on to BC, CA, AB respectively. Find the
ma + mb
maximum value of .
mc

Solution: Let us start with the standard expression:

4m2a = 2b2 + 2c2 − a2 .

However, c2 = b2 + a2 and hence 4m2a = 4b2 + a2 . Similarly, 4m2b = 4a2 + b2 ;


4m2c = 2b2 + 2a2 − c2 = b2 + a2 . Thus, it follows that

m2a + m2b
= 5.
m2c

Now the Cauchy-Schwarz inequality gives


 2
ma + mb 2(m2a + m2b )
≤ = 10.
mc m2c

We obtain
ma + mb √
≤ 10.
mc

Equality holds if and only if a = b. Thus, the maximum value is 10.

Example 4.6. Let ABC be a triangle with circum-circle Γ. The medians AD,
BE, CF , when extended, meet Γ in A1 , B1 , C1 respectively. Show that the sum
of the areas of the triangles BA1 C, CB1 A, AC1 B is at least equal to the area of
ABC. When does equality hold?
Applications 177

Solution: Let ha , hb and hc be the altitudes; ma , mb , mc be the medians.


Let A1 K, B1 L and C1 M be the perpendiculars from A1 , B1 , C1 on to BC,
CA, AB respectively. Then
[BA1 C] A1 K DA1
= = .
[ABC] ha AD

But DA1 · AD = BD · DC = a2 /4, since BD = DC = a/2. This gives


DA1 = a2 /4AD, and hence

[BA1 C] a2 a2
= 2
= .
[ABC] 4AD 4m2a
Similarly,
[CB1 A] b2 [AC1 B] c2
= , = .
[ABC] 4m2b [ABC] 4m2c
Thus,
 
a2 b2 c2
[BA1 C] + [CB1 A] + [AC1 B] = + + [ABC],
4m2a 4m2b 4m2c

and it is sufficient to prove that

a2 b2 c2
+ 2 + ≥ 1.
4m2a 4mb 4m2c

Equivalently, we need to prove that

a2 b2 c2
+ + ≥ 1.
2b2 + 2c2 − a2 2c2 + 2a2 − b2 2a2 + 2b2 − c2
Using the Cauchy-Schwarz inequality, observe that
   2
 2 2

2 2 a
a +b +c = √ 2 2
· a 2b + 2c − a 2

cyclic
2b2 + 2c2 − a2
     
a2 2 2 2 2
≤ a 2b + 2c − a .
2b2 + 2c2 − a2
cyclic cyclic

Thus, we obtain
 2 2
 a2 a + b2 + c 2
≥  2 2   .
2b2 + 2c2 − a2 4 a b + b2 c 2 + c 2 a 2 − a 4 + b4 + c 4
cyclic

Thus, it is sufficient to prove that


 2 2    
a + b 2 + c 2 ≥ 4 a 2 b2 + b 2 c 2 + c 2 a 2 − a 4 + b4 + c 4 .
178 Inequalities

This further reduces to  


a4 ≥ a 2 b2 ,
cyclic cyclic

which is a direct consequence of another application of the Cauchy-Schwarz


inequality. Equality holds if and only if a = b = c, which corresponds to the
case of an equilateral triangle.

Example 4.7. Let n > 1 be an integer, and let a1 , a2 , . . . , an be real numbers


such that (n − 1)a21 − 2na2 < 0. Prove that the roots of the equation

xn + a1 xn−1 + · · · + an−1 x + an = 0

cannot all be real.

Solution: Let α1 , α2 , . . . , αn be the roots of the equation, and suppose all of


them are real. Using Viette’s formulae relating the roots and the coefficients
of a polynomial equation, we get
 2 
α12 + α22 + · · · + αn2 = α1 + α2 + · · · + αn −2 αj αk
j<k

= a21 − 2a2 .

Since α1 , α2 , . . . , αn are real, the Cauchy-Schwarz inequality is applicable. This


implies that
 2  
α1 + α2 + · · · + αn ≤ n α12 + α22 + · · · + αn2
 
= n a21 − 2a2 .

Thus,
a21 ≤ na21 − 2na2 ,
and hence (n − 1)a21 − 2na2 ≥ 0. This violates the given condition. Hence not
all α1 , α2 , . . . , αn can be real.

Example 4.8. (Indian Team Selection, 1998) Let M be a positive integer and
consider the set S = {n ∈ N | M 2 ≤ n < (M + 1)2 }. Show that the products of
the form ab where a, b ∈ S are all distinct.

Solution: Suppose there are integers a, b, c, d in the given interval such that
ab = cd. We may assume d is the largest among them so that a < d and b < d.
This implies that c < a and c < b. Put c = a − k and d = a + l for some
positive numbers k and l. Then we have

ab = (a − k)(a + l) = a2 + a(l − k) − kl.


Applications 179

This shows that a divides kl. Since c, d lie in the set S, we observe that
a − k ≥ M 2 and a + l ≤ M 2 + 2M . Thus, we have
 
l + k = (a + l) − (a − k) ≤ M 2 + 2M − M 2 = 2M.
Using the AM-GM inequality, we now get
√ k+l
kl ≤ ≤ M.
2
This implies that kl ≤ M 2 . Since a divides kl, we get a ≤ M 2 . If a = M 2 ,
then c < a implies that c is not in the set S. Thus, a < M 2 , a contradiction.
Thus, if ab = cd, then {a, b} and {c, d} must be the same sets.

Example 4.9. Let f : N → [1, ∞) such that


(a) f (2) = 2;
(b) f (mn) = f (m)f (n) for all m, n in N;
(c) f (m) < f (n) whenever m < n.
Prove that f (n) = n for all natural numbers n.

Solution: Observe that f (1) = f (1 · 1) = f (1)2 and hence f (1) = 1. Now


induction shows that f (2k ) = 2k for all integers k ≥ 1. Let us take any m ∈ N
and suppose f (m) = l. Then f (mn ) = ln for all n ∈ N. If k is such that
2k ≤ mn < 2k+1 , then using (b) and (c) we obtain
2k ≤ ln < 2k+1 .
Thus, we get the inequality
1  m n
< < 2, (4.1)
2 l
valid for all natural numbers n. If m > l, choose n such that n > l/(m − l).
Then Bernoulli’s inequality gives
 m n  m−l
n 
m−l

= 1+ >1+n > 2,
l l l
by our choice of n. This contradicts the right hand side of inequality (4.1). If
m < l, we choose n > m/(l − m). Again, Bernoulli’s inequality is applicable
and we get
 n  n  
l l−m l−m
= 1+ >1+n > 2,
m m m
 n
by our choice of n. We thus obtain m/l < 1/2. But this contradicts the left
hand side of inequality (4.1).
We conclude that l = m, thus forcing f (m) = m for all natural numbers
m.
180 Inequalities

Example 4.10. ([8]) Determine all real polynomials of degree n with each of its
coefficients in the set {+1, −1} and having only real zeros.

Solution: If n = 1 then p(x) = (x − 1), (x + 1), −(x − 1) and −(x + 1) give all
the polynomials of degree 1. We may assume n > 1. We may also assume that
the leading coefficient of p(x) is 1 by changing the signs of all the coefficients,
if necessary. Thus, p(x) is of the form

p(x) = xn + cn−1 xn−1 + · · · + cx + c0 ,


 
where cj  = 1, for 0 ≤ j ≤ n − 1. Let α1 , α2 , . . . , αn denote all the roots of
p(x) = 0. We have α1 α2 · · · αn = (−1)n c0 , so that αj = 0 for every j. This
implies that
n
0< αj2 = c2n−1 − 2cn−2 = 1 ± 2.
j=1
n
Hence, cn−2 = −1 and j=1 αj2 = 3.
Consider the reciprocal polynomial of p(x). It is also of the same form, i.e.,
having coefficients ±1 and 1/αj are the zeroes
n of this reciprocal polynomial.
The same analysis leads to the conclusion j=1 (1/αj2 ) = 3. These two lead to

n 
 
1
αj2 + = 6.
j=1
αj2

However, the inequality


1
a+ ≥2
a
holds good for any positive real a. This implies that
n 
 
1
6= αj2 + ≥ 2n.
j=1
αj2

It follows that n ≤ 3. We may verify that for n = 3, ±(x+1)2 (x−1) and ±(x+
1)(x − 1)2 are the required polynomials. In the case n = 2, the polynomials
±(x2 − x − 1) and ±(x2 + x − 1) are the required ones.

Example 4.11. ([8]) Suppose α, β, γ and δ are real numbers such that

α + β + γ + δ = α7 + β 7 + γ 7 + δ 7 = 0.

Prove that α(α + β)(α + γ)(α + δ) = 0.


Applications 181

Solution: Here we use a tricky factorisation:


0 = −α7 − β 7 − γ 7 − δ 7
 7
= β + γ + δ − β 7 − γ 7 − δ7

     2 2
= 7 β+γ γ+δ β+δ β + γ 2 + δ 2 + βγ + γδ + βδ

 
+ βγδ β + γ + δ .

However we have
0 2  1
4 β 2 + γ 2 + δ 2 + βγ + γδ + βδ + βγδ β + γ + δ
3 2  2   2 42
= β+γ + γ+δ + β+δ − 4αβγδ
3 2  2  2 24
= α+β + α+γ + α+δ − 4αβγδ
3   42
= 3α2 + 2α β + γ + δ + β 2 + γ 2 + δ 2 − 4αβγδ.

The last expression reduces to


3 42
α2 + β 2 + γ 2 + δ 2 − 4αβγδ.

However we see that


3 42 3   42
α2 + β 2 + γ 2 + δ 2 − 4αβγδ ≥ αβ  + 2γδ  − 4αβγδ
+  ,
  2
≥ 
4 αβγδ  − 4αβγδ
 
≥ 12αβγδ  ≥ 0.
The AM-GM inequality has been used. We conclude that
0 2  1
4 β 2 + γ 2 + δ 2 + βγ + γδ + βδ + βγδ β + γ + δ ≥ 0

with equality if and only if |α|=|β|=|γ|=|δ|=0. Otherwise one of β + γ, β + δ,


γ + δ must be zero. It follows that
α(α + β)(α + γ)(α + δ) = 0.

Example 4.12. Find


0     1
min max 1 + z , 1 + z 2  .
|z|≤1
182 Inequalities

Solution: Consider the functions f, g defined by


     2  2 
f (z) = max 1 + z , 1 + z 2  , g(z) = max 1 + z  , 1 + z 2  .

Both f and g attain their minimum at the same point(Why?). Hence, it is


sufficient to consider the latter and find the point where it attains its minimum.
We have to find the least c for which there exists a z such that
   
1 + z 2 ≤ c2 , and 1 + z 2 2 ≤ c2 . (4.2)
   
Since 1 + z  = 1 + z 2  = 1 for z = 0, we must have c ≤ 1.
Let us put z = reiθ . Then
     
1 + z 2 = 1 + r2 + 2r cos θ, 1 + z 2 2 = 1 − r2 2 + 4r2 cos2 θ.
 2  2
Using 1 + z  ≤ c2 , and 1 + z 2  ≤ c2 , we get
 2  2
2 r 2 + 1 − c2 2 c2 − 1 − r 2
cos θ ≥ , cos θ ≤ .
4r2 4r2
 2  2
Thus, there is a z such that 1 + z  ≤ c2 and 1 + z 2  ≤ c2 if and only if
 2  2
r 2 + 1 − c2 ≤ c2 − 1 − r 2 .

This simplifies to 
2r4 − 2r2 c2 + c4 − 3c2 + 2) ≤ 0.
For getting a positive value for r2 , we need the discriminant of the above
quadratic expression to be non-negative. This gives

c4 − 6c2 + 4 ≤ 0. (4.3)

Thus, if (4.2) holds for some z, then (4.3) holds. The least number c satisfying
(4.3) is 

c = 3 − 5.
Thus, for any z, we have

    √
max 1 + z , 1 + z 2  ≥ 3− 5.

Taking 2

3− 5 2π
z = reiθ , r = ,θ = ,
2 3
     √
we see that 1 + z  = 1 + z 2  = 3 − 5. Hence, the required minimum is
 √
3 − 5.
Applications 183

Example 4.13. (Romania) Define a sequence xn by


xn n
x1 = 1, xn+1 = + , for n ≥ 1.
n xn
- .
Show that xn is increasing and x2n = n for n ≥ 4.

Solution: We observe that x1 = 1, x2 = 2, x3 = 2. We show that for n ≥ 3,


the inequality
√ n
n ≤ xn ≤ √ , (4.4)
n−1
holds good. Consider the function
x n
f (x) = + ,
n x
defined for all x > 0. Then it is easy to see that f (x) is a decreasing function
for x ≤ n. Assuming (4.4) for n, we have
  √  n + 1
xn+1 = f xn ≤ f n = √ ,
n
 
  n n √
xn+1 = f xn ≥ f √ =√ > n + 1.
n−1 n−1
Thus, (4.4) is true for all n by induction. We now have
 
1
xn+1 − xn = xn −1
n
 2 
(n − 1) n
= − x2n ≥ 0.
nxn n−1

Thus, xn ≤ xn+1 for all n. Now using xn ≤ n/ n − 1, we get
 
  n n
xn+1 = f xn ≥ f √ =√ .
n−1 n−1
We thus obtain
n−1 n
√ ≤ xn ≤ √ .
n−2 n−1
This gives
 
  n−1 n2 (n − 2) + (n − 1)2
xn+1 = f xn ≤f √ = √ .
n−2 n(n − 1) n − 2

We show that
n2 (n − 2) + (n − 1)2 √
√ < n + 2.
n(n − 1) n − 2
184 Inequalities

3 2
√ the statement 2n − 6n + 4n − 1 > 0, which is true for
This is equivalent to
n ≥ 4. Thus, xn < n + 1 for n ≥ 4. We obtain
√ √
n < xn < n + 1,
- .
for n ≥ 4. Hence, x2n = n for n ≥ 4.

Example 4.14. ([8]) Find all real numbers a, b, c such that


1
(1 − a)2 + (a − b)2 + (b − c)2 + c2 = .
4
Solution: Observe that
 2
1 = (1 − a) + (a − b) + (b − c) + c
 
≤ 4 (1 − a)2 + (a − b)2 + (b − c)2 + c2
= 1.

Thus, equality holds in the Cauchy-Schwarz inequality. It follows that


1
1−a=a−b=b−c=c= .
4
Hence, a = 3/4, b = 1/2, c = 1/4.

Example 4.15. ([8]) Show that if all the roots of P (x) = ax4 −bx3 +cx2 −x+1 =
0 are positive, then c ≥ 80a + b.

Solution: Suppose a = 0. Let α, β, γ and δ be the roots of P (x) = 0. These


are positive reals. Taking
1 1 1 1
u= + , v= + ,
α β γ δ
the following relations are easily obtained:

u + v = 1, αβu = α + β, γδv = γ + δ.

We have to show that


c b
− ≥ 80.
a a
However, using the Viette’s relations between the roots and coefficients,
c b
− = αβ + αγ + αδ + βγ + βδ + γδ − (α + β + γ + δ)
a a
= (α + β)(γ + δ) − (α + β) − (γ + δ) + αβ + γδ
= αβγδuv + αβ(1 − u) + γδ(1 − v)
= αβγδuv + αβv + γδu.
Applications 185

But, observe that


 2
αβ ≥ .
α−1 + β −1
This gives αβ ≥ 4u−2 . Similarly γδ ≥ 4v −2 . Thus,
c b 16 4  
− ≥ + 2 2 u3 + v 3
a a uv u v
16 4  
= + 2 2 u2 − uv + +v 2
uv
 u v
3 1 1
= 4 + + 2
uv u2 v
 1
1 1 2
= 4 + +
uv u v

1 1
= 4 +
uv u2 v 2
 2
2
= + 1 − 1.
uv
But u + v = 1 implies that uv ≤ 1/4. Thus,
c b
− ≥ 92 − 1 = 80.
a a
Equality holds if and only if α = β = γ = δ = 4.
Suppose a = 0 and b = 0. Then P (x) reduces to a cubic polynomial:
P (x) = −bx3 + cx2 − x − 1. Let p, q, r be its positive roots. Then
c 1 1 1
= p + q + r, and + + = 1.
b p q r
Hence,   
c 1 1 1
= p+q+r + + ≥ 3,
b p q r
by the AM-GM inequality. In particular c/b > 1 and this proves the result in
this case.
Finally, suppose a = b = 0. Then P (x) = cx2 − x − 1 and hence c > 0;
otherwise P (x) = 0 has no positive root. This completes the proof.

Example 4.16. (CRUX, 1995) Find the maximum value of


a b c d
+ + + ,
c d a b
where a, b, c, d vary over all distinct real numbers such that
a b c d
+ + + = 4, ac = bd.
b c d a
186 Inequalities

Solution: Introducing the new variables u, v by u = a/b and v = b/c, we see


that
c 1 d 1 a b v
= , = , = uv, = .
d u a v c d u
v 1 u
Thus, the maximum of uv + + + has to be found under the constraint
u uv v
u + v + (1/u) + (1/v) = 4. Now for any nonzero real p, we have p + (1/p) ≥ 2
if p > 0 and p + (1/p) ≤ 2 if p < 0. The condition u + v + (1/u) + (1/v) = 4
shows that u and v cannot both be negative. If u and v are both positive then
u + (1/u) = 2 = v + (1/v) which forces that u = v = 1. But then a = b = c
contradicting the distinctness of a, b, c. Thus, one of u, v is positive and the
1
other negative. We may assume that u > 0 and v < 0. In this case v + ≤ −2
v
and hence  
1 1
u+ =4− v+ ≥ 6.
u v
We finally get
  
v 1 v 1 1
uv + + + = u+ v+ ≤ −12.
u uv u u v

This maximum is attained when


√ √
a = 3 + 2 2, b = 1, c = −1, d = −3 + 2 2.

Example 4.17. Solve the system of equations for real values of x, y, z:


 
3 x2 + y 2 + z 2 = 1,
x2 y 2 + y 2 z 2 + z 2 x2 = xyz(x + y + z)3 .

Solution: The second relation shows that xyz(x + y + z)3 ≥ 0 and hence
xyz(x + y + z) ≥ 0. If xyz(x + y + z) = 0, then the second relation implies that
xy = yz = zx = 0. Hence, two of x, y, z are zero. If x = y = 0, then z 2 = 1/3
giving z = ±1/3. We get six solutions:
   √   √   √ 
x, y, z = 0, 0, ±1/ 3 , 0, ±1/ 3, 0 , ± 1/ 3, 0, 0 .

Suppose xyz(x + y + z) > 0. Using the Cauchy-Schwarz inequality, we see that


 2  
xy + yz + zx ≤ 3 x2 y 2 + y 2 z 2 + z 2 x2 .

This gives

xyz(x + y + z) ≤ x2 y 2 + y 2 z 2 + z 2 x2 = xyz(x + y + z)3 .


Applications 187

Since xyz(x + y + z) > 0, it follows that (x + y + z)2 ≥ 1. But then


 
1 ≤ (x + y + z)2 ≤ 3 x2 + y 2 + z 2 = 1.
 
Hence, equality is forced every where. We obtain (x + y + z)2 = 3 x2 + y 2 + z 2
or equivalently x2 + y 2 + z 2 = xy + yz + zx. This leads to
(x − y)2 + (y − z)2 + (z − x)2 = 0,
which gives x = y = z. We obtain from the first equation x = y = z = 1/3 or
x = y = z = −1/3, giving two more solutions.

Example 4.18. Let P (x) = an xn +an−1 xn−1 +· · ·+a1 x+a0 be a real polynomial
of degree n. Suppose n is even and
(i) a0 > 0, an > 0;
4  
(ii) a21 + a22 + · · · + a2n−1 ≤ min a20 , a2n .
n−1
Prove that P (x) ≥ 0 for all real values of x.

Solution: Suppose x is a real number. The Cauchy-Schwarz inequality gives


   
 
an−1 xn−1 + · · · + a1 x ≤ a2n−1 + · · · + a21 x2 + x4 + · · · + x2n−2 ,

for any real x. Thus, for x ∈ R,


   
P (x) = an xn + a0 + an−1 xn−1 + · · · + a1 x
   

≥ an xn + a0 − an−1 xn−1 + · · · + a1 x
    
≥ min an , a0 1 + xn − a2n−1 + · · · + a21 x2 + x4 + · · · + x2n−2 .

Using the condition (ii), it follows that


   2 min{a0 , an }  2
P (x) ≥ min an , a0 1 + xn − √ x + x4 + · · · + x2n−2
n−1
5 √ 6
  2 x2 + x4 + · · · + x2n−2
= min an , a0 1 + x −
n
√ .
n−1

Thus, it is sufficient to prove that



2 x2 + x4 + · · · + x2n−2
1 + xn − √ ≥ 0,
n−1
for all real x. Equivalently, one has to prove that

  2 
n−1
n−1 x +1 n
≥4 x2j .
j=1
188 Inequalities

Since n is even, n = 2m for some positive integer m. Introducing x2 = y, we


need to show that
2m−1

  
2m − 1 y 2m + 2y m + 1 ≥ 4 yj .
j=1

We use the majorisation technique. The function f (t) = tk is a convex function


for all k ≥ 1. Hence, we need to show that
 
2m − 1, . . . , 2m − 1, 2m − 2, . . . , 2m − 2, . . . , 1, 1, 1, 1




4 4 4
 
≺ 2m, 2m, . . . , 2m, m, m, . . . , m, 0, 0, . . . , 0




2m−1 2(2m−1) 2m−1

Note that after 4(m − 1) terms in the first vector, the remaining terms are not
more than m. Hence, it is sufficient to prove that
 
4 2m − 1 + 2m − 2 + · · · + 2m − (m − 1) < 2m(2m − 1) + m(2m − 3).

But the left-side is 6m2 − 12m and the right-side is 6m2 − 5m and hence the
inequality holds. This completes the proof.

Example 4.19. (Nordic Contest-1992) Find all real numbers x > 1, y > 1, z > 1
such that
3 3 3 √  √ 
x+y+z+ + + =2 x+2+ y+2+ z+2 .
x−1 y−1 z−1

Solution: For t > 1, the AM-GM inequality gives

t+2 √
t−1+ ≥ 2 t + 2.
t−1

holds if and only if (t − 1) = (t + 2)/(t − 1), which is equivalent to


Equality √
t = (3 + 13)/2. Thus,

3 3 3
x+y+z+ + +
x−1 y−1 z−1
x+2 y+2 z+2
= x−1+y−1+z−1+ + +
x−1 y−1 z−1
√  √ 
≥ 2 x+2+ y+2+ z+2
3 3 3
= x+y+z+ + +
x−1 y−1 z−1
Applications 189

Hence, equality holds and this corresponds to



3 + 13
x=y=z= .
2

Example 4.20. (Bulgaria) Find the largest real number a such that
x y z
 +√ + ≥ a,
y2 + z2 z2 +x 2 x + y2
2

for all positive real numbers x, y, z.

Solution: Note that for any real α,



≤ 1.
1 + α2
If α > 0, we also get
2α2
≤ α.
1 + α2
Thus, it follows that
*
x 2x2 (y 2 + z 2 ) 2x2
 ≥  * = 2 .
y2 + z2 1 + x2 (y 2 + z 2 ) x + y2 + z2

This implies that


x y z  2x2
 +√ + ≥ =2
y2 + z2 z 2 + x2 x2 + y 2 x + y2 + z2
2
cyclic

We show that a = 2 is the largest possible. In other words, for each  > 0,
there are positive x, y, z such that
 x
 < 2 + .
cyclic
y + z2
2


For this choose x = y = 1 and z =  2. Then
 x 2
 =√ +  < 2 + .
cyclic
y2 + z2 1 + 22

Example 4.21. (Romania, 1993) Prove that if x, y, z are positive integers such
that x2 + y 2 + z 2 = 1993, then x + y + z cannot be the square of an integer.
190 Inequalities

Solution: Suppose, if possible, x + y + z = k 2 for some integer k. Then


√  √ √
k2 = x + y + z ≤ 3 x2 + y 2 + z 2 = 3 1993 < 70.
 
Thus, k 2 ∈ 1, 4, 9, 16, 25, 36, 49, 64 . On the other hand x2 ≡ x (mod 2) for
any integer x. Thus, x2 + y 2 + z 2 ≡ x + y + z (mod 2) and hence

k 2 = x + y + z ≡ x2 + y 2 + z 2 ≡ 1993 ≡ 1 (mod 2).


 
This shows that k 2 is odd and hence k 2 ∈ 1, 9, 25, 49 . Now observe that

(x + y + z)2 > x2 + y 2 + z 2 = 1993, and 252 = 625 < 1993.

Hence, x + y + z = 49 is the only possibility. But then x + y + z = 49 and


x2 + y 2 + z 2 = 1993 give

(49 − x)2 = (y + z)2 > y 2 + z 2 = 1993 − x2 .

Simplification leads to x2 − 49x + 204 > 0. It follows that


√ √
49 − 1505 49 + 1505
either x < or x> .
2 2
However

49 + 1505 49 + 39
x> =⇒ x > = 44,
2 2
and hence x ≥ 45. But then x2 ≥ 452 = 2025 > 1993. Similarly,

49 − 1505
x< =⇒ x ≤ 4.
2

Symmetry shows that y ≤ 4 and z ≤ 4. It follows that

x + y + z ≤ 12 < 49.

Thus, x + y + z cannot be a square.

Example 4.22. ([7]) Find all real numbers a, b, c such that


 2 
ax + bx + c ≤ 1, for all |x| ≤ 1,

8 2
and a + 2b2 is maximum.
3
Applications 191

Solution: It is sufficient to maximise 4a2 + 3b2 , since


 
2 2 3 8 2 2
4a + 3b = a + 2b .
2 3

Consider the polynomial P (x) = ax2 + bx + c. Since |P (0)| ≤ 1 and |P (1)| ≤ 1,


we have
2 ≥ |P (1) − P (0)| = |a + b + c − c| = |a + b|.
Similarly using |P (−1)| ≤ 1, we also obtain

2 ≥ |P (−1) − P (0)| = |a − b + c − c| = |a − b|.

Thus,
4a2 + 3b2 = 2(a + b)2 + 2(a − b)2 − b2 ≤ 16 − b2 ≤ 16.
Here equality holds only if b = 0. Thus, 4a2 + 3b2 is maximum only if b = 0.
In this case
|a + b| = |a − b| = |a| = 2.
Thus,
|P (1) − P (0)| = |a + c − c| = |a| = 2.
Hence,
2 = |P (1) − P (0)| ≤ |P (1)| + |P (0)| ≤ 2.
We conclude that |P (1)| = |P (0)| = 1 and hence |c| = 1, |a + c| = 1. It follows
that either c = 1, a = −2, b = 0 or c = −1, a = 2, b = 0.
Now the required maximum is
8 2 8 32
a + 2b2 = × 4 = .
3 3 3

Example 4.23. ([7]) What is the maximum value of a2 + b2 , given that the
equation
x4 + ax3 + bx2 + ax + 1 = 0
has real roots.

Solution: Consider the polynomial P (x) = x4 + ax3 + bx2 + ax + 1. Observe


that P (x) = x4 P (1/x). Hence, P (x) is a reciprocal polynomial and hence the
substitution y = x + 1/x is meaningful. This leads to a quadratic equation

y 2 + ay + b − 2 = 0.

If x is real, we observe that y is also real and hence

(2 − y 2 )2 = (ay + b)2 ≤ (a2 + b2 )(1 + y 2 ).


192 Inequalities

This gives
(2 − y 2 )2
a 2 + b2 ≥ .
1 + y2
We also note that y = x + 1/x ≤ −2 or y ≥ 2. Hence, y 2 ≥ 4. Consider the
function
(2 − t)2
f (t) = ,
1+t
for t ≥ 4. This is a monotonically increasing function for t ≥ 4, as may be
verified using its derivatives. Its minimum is attained at t = 4;
4
a2 + b2 ≥ f (4) = .
5
If we take y 2 = 4, we have x + 1/x = 2 and hence x = 1 is a real solution. This
corresponds to 2a + b = −2 and we can take
4 2
a=− , b=− .
5 5
Then a2 + b2 = 4/5. Thus, the minimum value of a2 + b2 is 4/5.

Example 4.24. Suppose a, b, c are real numbers such that

|a| ≤ |b − c|, |b| ≤ |c − a|, and |c| ≤ |a − b|

Prove that one of these numbers is the sum of the remaining two.

Solution: We get from the first relation, a2 ≤ (b − c)2 , or which is the same
as
(b − c − a)(b − c + a) ≥ 0.
Similarly

(c − a − b)(c − a + b) ≥ 0, and (a − b + c)(a − b − c) ≥ 0.

Multiplying these (which is permissible since all are non-negative), we get

(a + b − c)2 (b + c − a)2 (c + a − b)2 ≤ 0.

It follows that at least one of (a + b − c), (b + c − a), (c + a − b) is equal to zero.

Example 4.25. Let ABC be a triangle and P be an interior point of ABC. Let
the lines AP , BP , CP meet BC, CA, AB respectively in D, E, F . Find all
positions of the point P for which the area of triangle DEF is maximal.
Applications 193

Solution: It is sufficient to locate D, E, F such that AD, BE, CF are


concurrent and [DEF ] is maximal, where [DEF ] denotes the area of triangle
DEF . Let us put BD = a − x, DC = x, CE = b − y, EA = y, AF = c − z
and F B = z. Then by Ceva’s theorem,

a−x b−y c−z


· · = 1.
x y z

If we also introduce α = x/a, β = y/b and γ = z/c, then it takes the form

(1 − α)(1 − β)(1 − γ) = αβγ.

Now observe that [BDF ] = γ(1 − α)[ABC] and similar expressions for other
corner triangles. Thus,

[DEF ] = [ABC] − [BDF ] − [CED] − [AF E]


 
= [ABC] 1 − γ(1 − α) − α(1 − β) − β(1 − γ) .

This shows that we have to minimize γ(1 − α) + α(1 − β) + β(1 − γ) in order


to maximise [DEF ]. However

γ(1 − α) + α(1 − β) + β(1 − γ) = α + β + γ − (αβ + βγ + γα).

Now using (1 − α)(1 − β)(1 − γ) = αβγ, it is easy to get

α + β + γ − (αβ + βγ + γα) = 1 − 2αβγ.

Thus, we have to minimise 1 − 2αβγ or which is same as maximising αβγ. But



αβγ = αβγ(1 − α)(1 − β)(1 − γ)

= α(1 − α)β(1 − β)γ(1 − γ)
 3
α+1−α+β+1−β+γ+1−γ

3
1
= .
8
Equality holds if and only if α = β = γ. This is possible if and only if D, E, F
are the mid-points of BC, CA, AB respectively. We conclude that P = G, the
centroid of ABC.

Example 4.26. (Short-list, IMO-1993) Let a, b, c, d be non-negative real numbers


such that a + b + c + d = 1. Prove that
1 176
abc + bcd + cda + dab ≤ + abcd.
27 27
194 Inequalities

Solution: Consider the expression


  176
F a, b, c, d = abc + bcd + cda + dab − abcd
 27 
176
= bc(a + d) + ad b + c − bc .
27
 
The function F a, b, c, d is symmetric in a, b, c, d. There are two possibilities:
176
(i) If b + c − bc ≤ 0, then the AM-GM inequality implies that
27
 3
  b+c+a+d 1
F a, b, c, d ≤ bc(a + d) ≤ = .
3 27

Hence, we may assume that this never occurs in any subsequent steps.
176
(ii) Suppose b + c − bc > 0. Then again the AM-GM inequality implies
27
that
 2  
  a+d 176
F a, b, c, d ≤ bc(a + d) + b+c− bc
2 27
 
a+d a+d
= F , b, c, .
2 2

Now we iterate this under the assumption that we are in the second case
in each stage of the iteration and we also exploit the symmetry of F :
 
  a+d a+d
F a, b, c, d ≤ F , b, c,
2 2
 
a+d a+d
= F b, , ,c
2 2
 
b+c a+d a+d b+c
≤ F , , ,
2 2 2 2
 
a+d b+c a+d b+c
= F , , ,
2 2 2 2
 
1 b+c a+d 1
≤ F , , ,
4 2 2 4
 
b+c 1 1 a+d
= F , , ,
2 4 4 2
 
1 1 1 1 1
≤ F , , , =
4 4 4 4 27

This completes the solution.


Applications 195

n
Example 4.27. Determine the minimum value of j=1 aj where the minimum
is taken over all sequences a1 ,a2 ,a3 ,. . .,an , (n ≥ 2), with non-negative terms
n
such that aj aj+1 = 1, where an+1 = a1 .
j=1

Solution: Let

n  
n

Mn = min aj  aj ≥ 0, for 1 ≤ j ≤ n, and aj aj+1 = 1 .
j=1 j=1

√ √ 
 that M2 = 2, M3 = 3 and Mn = 2 for n ≥ 4. Put α =
We show aj and
β= aj aj+1 . If n = 2, then α = a1 + a2 , β = 2a1 a2 = 1, so that

α2 − 2 = α2 − 2β = (a1 − a2 )2 ≥ 0,
√ √
with equality if and only if a1 = a2 = 1/ 2. Thus, M2 = 2. If n = 3, then

α2 − 3 = α2 − 3β
= (a1 + a2 + a3 )2 − 3(a1 a2 + a2 a3 + a3 a1 )
= a21 + a22 + a23 − a1 a2 − a2 a3 − a3 a1
10 1
= (a1 − a2 )2 + (a2 − a3 )2 + (a3 − a1 )2
2
≥ 0.
√ √
Here again, equality holds if and only if a1 = a2 = a3 = 1/ 3. Thus, M3 = 3.
Consider the case n = 4. In this case,

α2 − 4 = α2 − 4β
0 12
= (a1 + a3 ) + (a2 + a4 ) − 4(a1 + a3 )(a2 + a4 )
0 12
= (a1 + a3 ) − (a2 + a4 ) ≥ 0.

Equality holds if and only if a1 + a3 = a2 + a4 = 1/2. Thus, M4 = 2.


Suppose n ≥ 5. We may assume  that a1 is the least number by using the
cyclical nature of the condition aj aj+1 = 1. Let
 2  2
γ = a1 + a2 + · · · + an − a1 − a2 + a3 − · · · + (−1)n+1 an .

If n is even, say n = 2k, then


  
γ = 4 a1 + a3 + · · · + a2k−1 a2 + a4 + · · · + a2k
 
≥ 4 a1 a2 + a2 a3 + · · · + a2k−1 a2k + a2k a1
= 4.
196 Inequalities

Suppose n is odd, say n = 2k − 1; then


  
γ = 4 a1 + a3 + · · · + a2k−1 a2 + a4 + · · · + a2k−2
 
≥ 4 a1 a2 + a2 a3 + · · · + a2k−2 a2k−1 + a2k−1 a2
 
≥ 4 a1 a2 + a2 a3 + · · · + a2k−1 a1
= 4,

using a2 ≥ a1 . Thus, γ ≥ 4. This implies that


 2
a1 + a2 + · · · + an ≥ 4. (4.5)

We show that equality holds in (4.5) if and only if there is a k such that
1 ≤ k ≤ n, ak = 1, ak−1 + ak+1 = 1 and aj = 0 for all other j. (Here a0 = an ,
an+1 = a1 .) Clearly this is sufficient. Conversely, suppose equality holds in
(4.5). Then
 2  2
a1 + a2 + · · · + an − a1 − a2 + a3 − · · · + (−1)n+1 an
 
= 4 a1 a2 + a2 + a3 + · · · + an−1 an + an a1 .

Thus, for j, k with j + 2k + 1 ≤ n, it must be the case that aj aj+2k+1 = 0. In


particular, a1 a4 = 0. This gives a1 = 0, since a1 ≤ a4 . Further we have

a1 + a2 + · · · + an = 2, and a1 − a2 + a3 − · · · ± an = 0.

Thus, it follows that

a1 + a3 + · · · = 1,
a2 + a4 + · · · = 1.

If a2 = 0, then the condition aj aj+2k+1 = 0 implies that a5 = a7 = · · · = 0,


and a3 = 1. Further it also follows that a6 = a8 = · · · = 0. Hence, a2 + a4 = 1.
Suppose a2 = 0. Consider a3 . Using a similar argument, it is easy to show
that there is some l such that al = 1 and al−1 + al+1 = 1 and the remaining
are all equal to zero.

Example 4.28. (Taiwan, 2002) Determine all positive integers n and integers
a1 ,a2 ,a3 ,. . .,an such that aj ≥ 0, for 1 ≤ j ≤ n, and


n 
n 2
4
a2j = 1 + aj .
j=1
4n + 1 j=1

Solution: Let us put



n
A= aj .
j=1
Applications 197

Suppose aj ≤ 1, for all j. Then aj = 0 or 1, for each j. Let α be the number


of those j’s for which aj = 1. Then n − α is the number of those j’s for which
aj = 0. The given equation reduces to
4
α=1+ α2 . (4.6)
4n + 1
This shows that α > 1. Moreover,
4
4n + 1 = 4(α + 1) + .
α−1
Hence, α − 1 divides 4, so that α − 1 = 1, 2 or 4. The first two cases may be
disposed of using (4.6). The third choice gives n = 6. In this case (0, 1, 1, 1, 1, 1)
and its permutations give all the solutions.
We may assume aj ≥ 2 for some j. In this case
 4
A+1< a2j = 1 + A2 .
4n + 1
Thus, A > n + (1/4). Since A is an integer, it follows that A ≥ n + 1. Using
the Cauchy-Schwarz inequality, one obtains,
 n 
A2 2 4
≤ aj = 1 + A2 .
n j=1
4n + 1

This implies that A2 ≤ 4n2 + n. Again the nature of A shows that A ≤ 2n.
Thus, n + 1 ≤ A ≤ 2n. We hence obtain
1 A
1<1+ ≤ ≤ 2.
n n
But we also have
n 
 2
A A2
aj − =1− < 1,
j=1
n n(4n + 1)

which gives
A
−1 < aj − < 1,
n
for 1 ≤ j ≤ n. Thus,
A A
0 < −1 + < aj < 1 + < 3,
n n
for 1 ≤ j ≤ n. This shows that aj = 1 or 2 for all j. Let b denote number of
those j’s for which aj = 2. Then there are n − b values of j for which aj = 1.
In this case (4.6) takes the form
4
4b + n − b = 1 + (2b + n − b)2 .
4n + 1
198 Inequalities

1
This simplifies to n = b + . Since n and b are integers the only possibility
4b − 3
is b = 1 and n = 2.
We conclude that n = 2 or n = 6. In these cases, aj ’s can be computed
using the bounds we have derived:
n = 2: (a1 , a2 ) = (1, 2), (2, 1);
n = 6: the 6-tuples are (0, 1, 1, 1, 1, 1) and permutations of this.

Example 4.29. (IMO, 1992) Consider a finite set V in three-dimensional space


consisting of points with integer coordinates. Let Sx , Sy , Sz denote the sets
consisting of the orthogonal projections of the points of S on to yz-plane, zx-
plane, xy-planes respectively. Prove that
  2  2  2  2
V  ≤ S x  Sy   Sz  ,

where |X| denotes the number of elements of a finite set X.

Solution: Let us introduce


S(x) = {(y, z) : (x, y, z) ∈ V };
Sy (x) = {z : (x, z) ∈ Sy };
Sz (x) = {y : (x, y) ∈ Sz }.
Observe that S(x) ⊂ Sx and S(x) ∈ Sy (x) × Sz (x). Thus
 
|V | = |S(x)| ≤ |Sx ||Sy (x)|Sz (x)|
x x
 
= |Sx | |Sy (x)|Sz (x)|.
x

But Cauchy-Schwarz inequality gives


2 2
   
|Sy (x)|Sz (x)| ≤ |Sy (x)| |Sz (x)| = |Sy ||Sz |.
x x x

Combining the two inequalities, it follows that



|V | ≤ |Sx ||Sy ||Sz |.

Example 4.30. For k ≥ 2, define f : N → N by


+ 1/k ,
f (n) = n + n1/k + n,

whee [x] denotes the integer part of a real number x. Determine the range of f .
Applications 199

Solution: We show that the range of f is precisely the set of all those
natural numbers which are not k-th power of some natural number. Suppose
a ∈ N is not a k-th power. Then there is a natural number m such that
mk < a < (m + 1)k . Using the binomial theorem, it is easy to derive that
(x + 1)k ≥ xk + x + 1,
for all x ≥ 0. Taking x = m − 1, it follows that (m − 1)k ≤ mk − m. Taking
n = a − m, we also get mk − m < a − m = n < a. Thus,
mk ≤ a − 1 = (a − m) + (m − 1)
< n + n1/k
< (a − m) + (m + 1)
= a+1
≤ (m + 1)k .
The definition of f shows that f (n) = m + n = a, i.e., f (a − m) = a. This
shows that if a ∈ N is not a k-th power, then a is in the range of f . Observe
that f is strictly increasing. Since f (1) = 2, it follows that 1 is not in the range
of f . Now consider any k-th power, say, mk . We have proved that if a ∈ N is
not a k-th power, then  - .
a = f a − a1/k .
Thus, we obtain
   
f mk − m = f mk − 1 − (m − 1) = mk − 1,

and  
f mk + 1 − m = mk + 1.

It follows that mk is not in the range of f . Thus, the range of f consists of


precisely all those natural numbers which are not the k-th power.

Example 4.31. (IMO, 1992) Find all integers a, b, c with 1 < a < b < c such
that (a − 1)(b − 1)(c − 1) is a divisor of abc − 1.

Solution: Putting a − 1 = x, b − 1 = y and c − 1 = z, the problem is to find


integers 1 ≤ x < y < z such that xyz divides (x + 1)(y + 1)(z + 1) − 1. This is
equivalent to find integers 1 ≤ x < y < z for which
1 1 1 1 1 1
+ + + + + =p
x y z xy yz zx
is also an integer. We observe that x ≥ 1, y ≥ 2 and z ≥ 3, and hence
1 1 1 1 1 17
p≤1+ + + + + = < 3.
2 3 2 3 6 6
200 Inequalities

Thus, p = 1 or p = 2.
Case 1. p = 1.
Here we have
1 1 1 1 1 1
+ + + + + = 1.
x y z xy yz zx
Obviously x ≥ 2. If x ≥ 3 then y ≥ 4 and z ≥ 5 and we get an estimate
1 1 1 1 1 1 59
p≤ + + + + + = <1
3 4 5 12 15 20 60
which is not tenable. We conclude that x = 2 and hence
1 1 1 1 1 1
+ + + + + = 1.
2 y z 2y yz 2z

This simplifies to (y − 3)(z − 3) = 11. Keeping in mind that y < z, we see that
the only solution is (y, z) = (4, 14). This leads to the triple (a, b, c) = (3, 5, 15).
Case 2. p = 2.
Here again, we see that x ≥ 2 forces p < 2, and hence x = 1. Using this
information in the expression for p, we obtain (y − 2)(z − 2) = 5. Since y < z,
the only solution is (y, z) = (3, 7) and this gives (a, b, c) = (2, 4, 8).
We obtain two solutions to the problem: (a, b, c) = (3, 5, 15), (2, 4, 8).

Example 4.32. Find all integer sided triangles with the property that the area of
each triangle is numerically equal to its perimeter.

Solution: Let a, b, c denote the three sides of such a triangle. If s = (a + b +


c)/2 is its semi perimeter, Heron’s formula gives

Δ2 = s(s − a)(s − b)(s − c).

The given condition translates to

4s = (s − a)(s − b)(s − c).

Taking s − a = x, s − b = y, s − c = z, we can write an equation for x, y, z:


1 1 1 1
+ + = . (4.7)
xy yz zx 4

We can rule out isosceles triangles. If, for example, a = b, we get x = y and
the equation reduces to
1 2 1
+ = .
x2 xz 4
This can be written as a quadratic equation in x: zx2 − 8x − 4z = 0. The
discriminant of this equation is D = 16(4 + z 2 ). The equation has an integer
Applications 201

root only if D is the square of an integer. This can happen only if z = 0 which
is impossible.
Thus, we may assume that 1 ≤ x < y < z. If x ≥ 3, we have y ≥ 4 and
z ≥ 5. Going back to the equation (4.7) with this information, we see that
1 1 1 1 1 1 1 1
= + + ≤ + + = .
4 xy yz zx 12 20 15 5
This absurd implication shows that x ≤ 2. Thus, x = 1 or x = 2.
Case 1. x = 1.
Here the equation (4.7) takes the form
1 1 1 1
+ + = .
y yz z 4
This can be written as (y − 4)(z − 4) = 20. In this case, we get (y, z) = (5, 24),
(6, 14), (8, 9). We obtain the triangles (with c < b < a) (6, 25, 29), (7, 15, 20),
and (9, 10, 17).
Case 2. x = 2
By a similar analysis, the only triangles when x = 2 are (5, 12, 13) and (6, 8, 10).

Example 4.33. Show that the only solutions of the equation


u3 − v 3 = uv + 1
in integers are (u, v) = (1, 0) and (0, −1).

Solution: We show that there is no solution (u, v) such that uv = 0. Suppose


uv > 0 then u3 − v 3 > 0 and hence u − v ≥ 1. We get
uv + 1 = u3 − v 3 = (u − v)(u2 + uv + v 2 )
= (u − v){(u − v)2 + 3uv}
> 3uv
This shows that 2uv < 1 which is impossible because uv ≥ 1.
Suppose uv < 0. If uv = −1, we have u3 − v 3 = uv + 1 = 0 so that u = v.
But then u2 = uv = −1, which is impossible. Hence, we may assume that
uv < −1. We have
|uv + 1| = |u − v||u2 + uv + v 2 |
= |u − v||(u + v)2 − uv|.
If u + v = 0, we see that u = −v and hence we get 2u3 = −u2 + 1 which has no
integer solution. Hence, we may assume that u + v = 0 and thus (u + v)2 ≥ 1.
Using the condition uv < 0, we get
1 − uv ≤ |uv + 1| = −(uv + 1)
202 Inequalities

because uv < −1. This leads to the absurd conclusion that 1 ≤ −1.
We conclude that uv = 0. If u = 0, we get v = −1. If v = 0, we get u = 1.
Thus, the only solutions are (u, v) = (1, 0), (0, −1).

Example 4.34. ( Romania, 2003) Two unit squares with parallel sides overlap in
a rectangular region of area 1/8 square units. Find the minimum and maximum
possible distances between the centres of these two squares.

Solution: Let ABCD be the rectangle obtained as the intersection of two


unit squares. Let X and Y denote the centres of the two squares. Let AB = a,
BC = b so that ab = 1/8 sq. units, where a, b ∈ [0, 1]. Draw a line from X
parallel to AB and a line from Y parallel to BC. Let them intersect in Z. (See
Fig. 4.1)
Then XZ = 1 − a and Y Z = 1 − b and

XY 2 = XZ 2 + Y Z 2 = (1 − a)2 + (1 − b)2
= a2 + b2 − 2(a + b) + 2
1
= a2 + 2ab + b2 − 2(a + b) − +2
4
7
= (a + b)2 − 2(a + b) +
4
3
= (a + b − 1)2 +
4
3
≥ .
4

√ √
Thus, XY ≥ 3/2 and hence the minimal distance is 3/2. This is achieved
Applications 203

when a + b = 1 and ab = 1/8. This corresponds to


√ √ √ √
2+ 2 2− 2 2− 2 2+ 2
a= ,b = or a = ,b = .
4 4 4 4
Observe
9
0 ≤ (1 − a)(1 − b) = 1 − a − b + ab = − (a + b).
8
This shows that a + b ≤ 9/8. On the other hand
√ 1
a + b ≥ 2 ab = √ .
2
Thus,
1 1
√ −1≤a+b−1≤ .
2 8
However,
 2  2
1 1
≤ √ −1 ,
8 2
as may be easily verified. This implies that
3
XY 2 =(a + b − 1)2 +
4
 2
1 3
≤ √ −1 +
2 4
9 √
= − 2
4
 2
1
= 2− √ .
2
 √  √
Thus, it follows that XY ≤ 2 − 1/ 2 . Equality holds only if a = b = 1/2 2.
We hence have
√ 1
min XY = 3/2, and max XY = 2 − √ .
2
Chapter 5
Problems on inequalities

1. (Romania,1996) Let a1 , a2 , . . . , an , an+1 be n + 1 positive real numbers such


that a1 + a2 + · · · + an = an+1 . Prove that

 n  
  n
 
aj an+1 − aj ≤ ! an+1 an+1 − aj .
j=1 j=1

2. (Czech and Slovak, 1999) If a, b, c are positive real numbers, prove that
a b c
+ + ≥ 1.
b + 2c c + 2a a + 2b

3. (Balkan Olympiads, 2001) Let a, b, c be positive real numbers such that


abc ≤ a + b + c. Prove that

a2 + b2 + c2 ≥ 3(abc).

4. (Balkan Olympiads, 2002) For any positive real numbers a, b, c, prove that
2 2 2 27
+ + ≥ .
b(a + b) c(b + c) a(c + a) (a + b + c)2

5. (Ireland, 2003) Let a, b, c be three sides of a triangle such that a + b + c = 2.


Prove that
1
1 ≤ ab + bc + ca − abc ≤ 1 + .
27

6. (Ukraine, 2004) If a, b, c be positive real numbers such that a + b + c = 1,


prove that
√ √ √ √ √ √
ab + c + bc + a + ca + b ≥ 1 + ab + bc + ca.

7. (Romania, 2003) Let a, b, c, d be positive real numbers such that abcd = 1.


Prove that
1 + ab 1 + bc 1 + cd 1 + da
+ + + ≥ 4.
1+a 1+b 1+c 1+d

8. If a, b, c, d are positive real numbers, prove that


 
a 2 + b2 + c 2 + d 2 3 abc + abd + acd + bcd
≥ .
4 4
206 Problems

9. (Hungary, 1990) Let a, b, c be the sides of a triangle such that a + b + c = 2.


Prove that
a2 + b2 + c2 + 2abc < 2.

10. If a, b, c are positive real numbers such that a2 + b2 + c2 = 1, prove that


 
1 1 1 √
+ + + a + b + c ≥ 4 3.
a b c

11. (Austria, 2001) Find all triplets (a, b, c) of positive real numbers which
satisfy the system of equations:

a+b+c = 6,
1 1 1 4
+ + = 2− .
a b c abc

12. (Bosnia and Herzegovina, 2002) Let a, b, c be real numbers such that
a2 + b2 + c2 = 1. Prove that

a2 b2 c2 3
+ + ≥ .
1 + 2bc 1 + 2ca 1 + 2ab 5

13. (Ukraine, 2001) Let a, b, c and α, β, γ be positive real numbers such that
α + β + γ = 1. Prove that
  
αa + βb + γc + 2 αβ + βγ + γα ab + bc + ca ≤ a + b + c.

14. (Estonia, 1996-97) Prove that for all real numbers a, b, the inequality
 
a 2 + b2 + 1 > a b2 + 1 + b a 2 + 1

holds.

15. (Poland, 1994-95) For a fixed positive integer n, compute the minimum
value of the sum
x2 x3 xn
x1 + 2 + 3 + · · · + n ,
2 3 n
where x1 , x2 , x3 , . . . , xn are positive real numbers such that
1 1 1 1
+ + + ··· + = n.
x1 x2 x3 xn

16. Let a, b, c, d be positive real numbers such that a + b + c + d ≤ 1. Prove


that
a b c d 1
+ + + ≤ .
b c d a 64abcd
Problems 207

17. Let a, b, c be positive real numbers, all less than 1, such that a + b + c = 2.
Prove that
abc ≥ 8(1 − a)(1 − b)(1 − c).

18. (USAMO, 2003) Let a, b, c be three positive real numbers. Prove that

(2a + b + c)2 (2b + c + a)2 (2c + a + b)2


+ + ≤ 8.
2a2 + (b + c)2 2b2 + (c + a)2 2c2 + (a + b)2

19. Three positive real numbers a, b, c are such that (1 + a)(1 + b)(1 + c) = 8.
Prove that abc ≤ 1.

20. (IMO, 1983) If a, b, c are the sides of a triangle, prove that

a2 b(a − b) + b2 c(b − c) + c2 a(c − a) ≥ 0.

21. Let a1 , a2 , . . . , an be n(≥ 2) real numbers whose sum is 1. Prove that


n
aj n
≥ .
j=1
2 − aj 2n − 1

22. Let a1 , a2 , . . . , an be n positive real numbers whose sum is 1. Prove that



n
a2j 1
≥ .
j=1
aj + aj+1 2

(Here an+1 = a1 .)

23. Let a, b, c, d be four positive real numbers. Prove that


1 4 9 16 100
+ + + ≥ .
a b c d a+b+c+d

24. Let a1 , a2 , . . . , an be n(> 2) positive real numbers such that a1 + a2 + · · · +


an = 1 and aj < 1/2 for each j, 1 ≤ j ≤ n. Prove that


n
a2j 1
≥ .
j=1
1 − 2aj n−2

25. (Romania, 1999) Let x1 , x2 , . . . , xn , y1 , y2 , . . . , yn be 2n positive real num-


bers such that x1 + x2 + · · · + xn ≥ x1 y1 + x2 y2 + · · · + xn yn , where n ≥ 2 is
an integer. Prove that
x1 x2 xn
x1 + x 2 + · · · + x n ≤ + + ··· + .
y1 y2 yn
208 Problems

26. (Proposed for IMO-2001) If x1 , x2 , . . . , xn are n positive real numbers,


prove that
x1 x2 xn √
+ + ··· + < n.
1 + x21 1 + x21 + x22 1 + x21 + x22 + · · · + x2n

27. (CRUX, 2001) If a, b, c are positive real numbers, prove that


  
3 a2 b + b2 c + c2 a ab2 + bc2 + ca2 ≥ abc(a + b + c)3 .

28. (Czech and Slovak, 2003) Let P (x) = ax2 +bx+c be a quadratic polynomial
with non-negative coefficients and let α be a positive real number. Prove that

P (α)P (1/α) ≥ P (1)2 .

29. If a, b, c, d, e are positive reals, prove the inequality


 a 5
≥ ,
b+c 2
where the sum is taken cyclically over a, b, c, d, e.

30. If a, b, c are the sides of an acute-angled triangle, prove that


  
a2 + b2 − c2 a2 − b2 + c2 ≤ ab + bc + ca.
cyclic

31. (Iran, 2006) Let a, b, c be non-negative real numbers such that


1 1 1
+ + = 2.
a 2 + 1 b2 + 1 c 2 + 1
Prove that ab + bc + ca ≤ 3/2.

32. (Thailand, 2006) Suppose a, b, c are positive real numbers. Prove that
   1/3  1/3
a 3 + b3 + c 3
3 a + b + c ≥ 8 abc + .
3
When does equality hold?

33. Let c1 ,c2 ,c3 ,. . .,cn be n real numbers such that either 0 ≤ cj ≤ 1 for all j
or cj ≥ 1 for all j, 1 ≤ j ≤ n. Prove that the inequality
"
n
  "
n
1 − p + pcj ≤ 1 − p + p cj
j=1 j=1

holds, for any real p with 0 ≤ p ≤ 1.


Problems 209

34. (Taiwan, 2002) Let x1 , x2 , x3 , x4 be real numbers in the interval (0, 1/2].
Prove that
x1 x2 x3 x4
(1 − x1 )(1 − x2 )(1 − x3 )(1 − x4 )
x41 + x42 + x43 + x44
≤ .
(1 − x1 )4 + (1 − x2 )4 + (1 − x3 )4 + (1 − x4 )4

35. Let x1 ,x2 ,x3 ,. . .,xn be n real numbers such that 0 < xj ≤ 1/2. Prove that
⎛ ⎞ ⎛ ⎞
" /  n " /  
n n n
  n
  n
⎝ xj xj ⎠ ≤ ⎝ 1 − xj 1 − xj ⎠.
j=1 j=1 j=1 j=1

36. (CRUX) Consider a sequence an of real numbers satisfying aj+k ≤ aj +ak .
Prove that
a2 a3 an
a1 + + + ··· + ≥ an ,
2 3 n
for all n.

37. For positive real numbers x, y, z, prove the inequality


 x
 ≤ 1,
x + (x + y)(x + z)
where the sum is taken cyclically over x, y, z.

38. (INMO, 2002) Let x, y be non-negative real numbers such that x + y = 2.


Prove the inequality  
x3 y 3 x3 + y 3 ≤ 2.

39. (INMO, 1998) A convex quadrilateral ABCD is inscribed in a unit circle.


Suppose its sides satisfy the inequality AB · BC · CD · DA ≥ 4. Prove that the
quadrilateral is a square.
   
40. Let a1 , a2 , . . . , an and b1 , b2 , . . . , bn be two sets of reals such that
0 < h ≤ aj ≤ H and 0 < m ≤ bj ≤ M for some reals h, H, m, M . Prove that
    2
( aj 2 )( bj 2 ) 1 HM hm
1≤  ≤ + .
( a j bj ) 2 4 hm HM

41. Let f : [0, a] →


Rn be a convex function. Consider n points x1 ,x2 ,x3 ,. . .,xn
in [0, a] such that j=1 xj is also in [0, a]. Prove that
  
n
  n
   
f xj ≤ f xj + n − 1 f 0 .
j=1 j=1
210 Problems

42. For any natural number n, prove


 
2n √
3n < 4n .
n

43. Let a, b, c be positive real numbers and let x be a non-negative real number.
Prove that
ax+2 + bx+2 + cx+2 ≥ ax bc + abx c + abcx .

44. Let (a1 , a2 , . . . , an ), (b1 , b2 , . . . , bn ), and (c1 , c2 , . . . , cn ) be three sequences


of positive real numbers. Prove that


n 
n 1/3  
n 1/3  
n 1/3
a j bj c j ≤ a3j b3j c3j .
j=1 j=1 j=1 j=1

45. ([9]) Prove for any three real numbers a, b, c, the inequality
     2
3 a2 − a − 1 b2 − b − 1 c2 − c + 1 ≥ abc − abc + 1.

46. In a triangle ABC, show that


1 1 8
+ ≥ .
sin A sin B 3 + 2 cos C
Find the conditions for equality.

47. Consider a real polynomial of the form

P (x) = xn + an−1 xn−1 + · · · + a1 x + 1,

where aj ≥ 0 for 1 ≤ j ≤ (n − 1). Suppose P (x) = 0 has n real roots. Prove


that P (2) ≥ 3n .

48. Let a1 <a2 <a3 <. . .<an be n positive integers. Prove that
 2
a1 +a2 +a3 +· · ·+an ≤ a31 + a32 + a33 + · · · + a3n .

49. (Poland, 1996) Consider a sequence a1 ,a2 ,a3 ,. . .,an of positive real numbers
which add up to 1, where n ≥ 2 is an integer. Prove that for any positive real
n
numbers x1 ,x2 ,x3 ,. . .,xn with j xj = 1, the inequality

 n − 2  aj x2j
n
2 xj xk ≤ + ,
n − 1 j=1 1 − aj
j<k

holds.
Problems 211

50. (Iran, 1998) Let x1 , x2 , x3 , x4 be four positive real numbers such that
x1 x2 x3 x4 = 1. Prove that

1 1 1 1
x31 + x32 + x33 + x34 ≥ min x1 + x2 + x3 + x4 , + + + .
x1 x2 x3 x4

51. (Romania) Let {x} denote the fractional part of x; i.e., {x} = x − [x].
Prove for any positive integer n,
n
   n2 − 1
j ≤ .
j=1
2

52. If a, b, c are positive real numbers, prove that

a2 b2 c2 3
+ + ≥ .
(a + b)(a + c) (b + c)(b + a) (c + a)(c + b) 4

53. Suppose a, b, c are the sides of a triangle. Prove that

a2 (b + c − a) + b2 (c + a − b) + c2 (a + b − c) ≤ 3abc.

54. Let x, y, z be positive real numbers such that xyz ≥ xy + yz + zx. Prove
that
xyz ≥ 3(x + y + z).

55. Let b1 ,b2 ,b3 ,. . .,bn be n non-negative real numbers and let b denote the
sum of these numbers. Prove that

n−1
b2
bj bj+1 ≤ .
j=1
4

56. (AMM) Let a, b, c, d be complex numbers such that ac = 0. Prove that


  √
max |ac|, |ad + bc|, |bd| −1 + 5
   ≥ .
max |a|, |b| |c|, |d| 2

57. Let x, y, z be three real numbers in the interval [0, 1] such that xyz =
(1−x)(1−y)(1−z). Find the least possible value of x(1−z)+y(1−x)+z(1−y).

58. Let x1 ,x2 ,x3 ,. . .,xn be non-negative real numbers such that

n
1
≤ 1.
j=1
1 + xj

Prove that x1 x2 x3 · · ·xn ≥ (n − 1)n .


212 Problems

59. (Bulgaria, 1998) For positive real numbers a, b, c, prove the inequality
 √ √   
3
3 a + ab + abc ≤ 4 a + b + c .

60. ([1]) Show that for any two natural numbers m, n, the inequality
1 1 4
− ≤
m + n + 1 (m + 1)(n + 1) 45
holds.

61. If a, b are two positive real numbers, prove that

ab + ba > 1.

62. Let a, b be positive real numbers such that a + b = 1 and let p be a positive
real. Prove that  p  p
1 1 5p
a+ + b+ ≥ p−1 .
a b 2

63. (IMO, 2000) Let a, b, c be positive real numbers such that abc = 1. Prove
that    
1 1 1
a−1+ b−1+ c−1+ ≤ 1.
b c a

64. (Proposed for IMO-2001) Let x, y, z be real numbers in the interval [−1, 2]
such that x + y + z = 0. Prove that
2 2 2
(2 − x)(2 − y) (2 − y)(2 − z) (2 − z)(2 − x)
+ + ≥ 3.
(2 + x)(2 + y) (2 + y)(2 + z) (2 + z)(2 + x)

65. (IMO, 1978) Let an be a sequence of distinct positive integers. Prove
that
n
ak 
n
1
2
≥ ,
k k
k=1 k=1
for every positive integer n.

66. (IMO, 1984) Let x, y, z be non-negative real numbers such that x+y+z = 1.
Prove that
7
0 ≤ xy + yz + zx − 2xyz ≤ .
27
67. Let x1 , x2 , . . . , xn be n positive real numbers. Prove that

n
x3j 1
n
≥ xj ,
j=1
x2j + xj xj+1 + x2j+1 3 j=1
Problems 213

where xn+1 = x1 .

68. Suppose x, y, z are non-negative real numbers. Prove the inequality


x(x − z)2 + y(y − z)2 ≥ (x − z)(y − z)(x + y − z).
Find conditions for equality.

69. (Indian Team Selection, 2002) Prove that for any positive reals a, b, c, the
inequality,
a b c c+a a+b b+c
+ + ≥ + +
b c a c+b a+c b+a
holds.

70. If a, b are real numbers, prove that


a2 + ab + b2 ≥ 3(a + b − 1).

71. ([7]) Define a sequence xn by


x4n + 9
x1 = 2, xn+1 = .
10xn
4 5
Prove that < xn ≤ for all n > 1.
5 4

72. ([7]) Let a, b, c be positive real numbers such that a2 − ab + b2 = c2 . Prove


that (a − c)(b − c) ≤ 0.

73. ([7]) Let a, b, c be positive real numbers. Prove that


  
a2 − ab + b2 + b2 − bc + c2 ≥ a2 + ac + c2 .

74. ([7]) For all real numbers a, show that


(a3 − a + 2)2 ≥ 4a2 (a2 + 1)(a − 2)
holds.

75. (Thailand, 2004) Let a, b, c be distinct real numbers. Prove that


 2  2  2
2a − b 2b − c 2c − a
+ + ≥ 5.
a−b b−c c−a

76. Let α, β, x1 ,x2 ,x3 ,. . .,xn be positive reals such that α + β = 1, and
x1 +x2 +x3 +· · ·+xn = 1. Prove that

n
x2m+1
j 1
≥ 2m−1 ,
j=1
αxj + βxj+1 n
214 Problems

for every positive integer m, where xn+1 = x1 .

77. Given positive reals a, b, c, d, prove that


  
(a + c)2 + (b + d)2 ≤ a 2 + b2 + c 2 + d 2
 2|ad − bc|
≤ (a + c)2 + (b + d)2 +  .
(a + c)2 + (b + d)2

78. (APMO, 1996) In a triangle ABC, prove that


√ √ √
√ √ √ a+ b+ c
s−a+ s−b+ s−c≤ √ .
2

79. Let ABC be an acute-angled triangle with altitudes AD, BE, CF and
ortho-centre H. Prove that
HD HE HF 3
+ + ≥ .
HA HB HC 2

80. (IMO, 1981) For any point P inside a triangle ABC, let r1 , r2 , r3 denote
the distances of P from the lines BC, CA, AB respectively. Find all points P
for which a/r1 + b/r2 + c/r3 is minimal.

81. (IMO, 1996) Let ABCDEF be a convex hexagon in which AB, BC, CD
are respectively parallel to DE, EF , F A. Let RA , RB , RC be the circum-radii
of the triangles F AB, BCD, DEF respectively, and let p denote the perimeter
of the hexagon. Prove that
p
RA + R B + RC ≥ .
2

82. (AMM) Let ha , mb , wc denote respectively the altitude from A to BC, the
median from B to CA and the internal angle bisector of angle C. Prove that

3
h a + mb + w c ≤ (a + b + c).
2

83. (Bulgaria, 1997) Let ABC be a triangle with centroid G. Prove that

2
sin ∠ABG + sin ∠ACG ≤ √ .
3

84. (CRUX) Let ABC be a triangle with in-radius r. Let Γ1 , Γ2 , Γ3 be three


circles inscribed inside ABC such that each touches other circles and also two
of the sides. (Such a configuration is called Malafatti circles.) Let O1 , O2 , O3
Problems 215

be respectively the centres of the circles Γ1 , Γ2 , Γ3 . If r denotes the in-radius


of O1 O2 O3 , prove that
r √

≥ 1 + 3.
r
Find conditions for equality.

85. Let Ω be a Brocard point of a triangle ABC. Let AΩ, BΩ,CΩ extended
meet the circum-circle of ABC in K, L, M respectively. Prove that
AΩ BΩ CΩ
+ + ≥ 3.
ΩK ΩL ΩM

86. (Romania, 2006) Let P be a point inside a triangle ABC. Let rA , rB , rC


denote the in-radii of triangles P BC, P CA, P AB respectively. Prove that
a b c  √ 
+ + ≥6 2+ 3 .
rA rB rC

87. (Bulgaria, 1995) Suppose A1 A2 · · · A7 , B1 B2 · · · B7 , C1 C2 · · · C7 are three


regular heptagons which are such that A1 A2 = B1 B3 = C1 C4 . If Δ1 , Δ2 , Δ3
denote respectively their areas, prove that
1 Δ2 + Δ 3 √
< < 2 − 2.
2 Δ1

88. (Germany, 1995) Let ABC be a triangle, and let D, E be points on BC,
CA such that the in-centre of ABC lies on DE. Prove that [ABC] ≥ 2r2 .

89. (Ireland, 1995) Let A, X, D be points on a line with X between A and D.


Let B be a point such that√∠ABX ≥ 120◦ andlet C be a point between B
and X. Prove that 2AD ≥ 3 AB + BC + CD .

90. (South Korea, 1995) Let the internal bisectors of the angles A, B, C of a
triangle ABC meet the sides BC, CA, AB in D, E, F and the circum-circle
in L, M , N respectively. Prove that
AD BE CF
+ + ≥ 9.
DL EM FN

91. (IMO, 1995) Let ABCDEF be a convex hexagon with AB = BC = CD,


DE = EF = F A, and ∠BCD = ∠EF A = π/3. Suppose G and H are two
interior points of the hexagon such that ∠AGB = ∠DHE = 2π/3. Prove that
AG + GB + GH + DH + HE ≥ CF.

92. (Balkan Olympiads, 1996) Let O and G be respectively the circum-centre


and centroid of a triangle
 ABC. If R and r are its circum-radius and in-radius,
prove that OG ≤ R(R − 2r).
216 Problems

93. (St. Petersburg Olympiad, 1996) Let M be the point of intersection of two
diagonals of a cyclic quadrilateral. Let N be the point of intersection of two
lines joining the midpoints of opposite pair of sides. If O is the centre of the
circumscribing circle, prove that OM ≥ ON .

94. (APMO, 1997) Let ABC be a triangle with internal angle bisectors AD,
BE, CF . Suppose AD, BE, CF when extended meet the circum-circle again
in K, L, M respectively. If la = AD/AK, lb = BE/BL and lc = CF/CM ,
prove that
la lb lc
2 + 2 + ≥ 3.
sin A sin B sin2 C

95. (Armenia, 1999) Let O be the circum-circle of a triangle ABC. Suppose


AO, BO, CO when extended meet the circum-circles of triangles BOC, COA,
AOB in K, L, M respectively. Prove that
AK BL CM 9
+ + ≥ .
OK OL OM 2

96. (Romania, 2003) Show that in any triangle ABC



1 1 1 3
+ + ≤ .
ma mb mb mc mc ma [ABC]

97. (Romania, 2003) In any triangle ABC, prove that



wa + wb + wc ≤ 3s.

98. (Ireland, 1998) Let ABC be a triangle with points D, E, F respectively


on the sides BC, CA, AB. Let the lines AD, BE, CF , when produced meet
the circum-circle respectively in K, L, M . Prove that
AD BE CF
+ + ≥ 9.
DK EL FM
99. Show that in a triangle ABC,
 
max ama , bmb , cmc ≤ sR.

100. Let ABCD be a convex quadrilateral of area 1 unit. Prove that



AB + BC + CD + DA + AC + BD ≥ 4 + 8.

101. Let ABCD be a square inscribed in circle. If M is a point on the arc AB


(arc not containing C and D), prove that

M C · M D ≥ (3 + 2 2)M A · M B.
Problems 217

102. (Estonia, 1996) Let a, b, c be the sides of a triangle ABC with in-radius
r. Prove that
a sin A + b sin B + c sin C ≥ 9r.

103. Suppose ABC is an acute-angled triangle with area Δ and in-radius r.


Prove that √ 2
√ √ Δ
cot A + cot B + cot C ≤ 2 .
r

104. (US Team Selection, 2000) Let ABC be a triangle having the circum-
radius R. Let P be an interior point of ABC. Prove that
AP BP CP 1
+ + ≥ .
BC 2 CA2 AB 2 R
105. With every natural number n, associate a real number an by
1 1 1
an = + + ··· + ,
p1 p2 pk
 
where p1 , p2 , . . . , pk is the set of all prime divisors of n. Show that for any
natural number N ≥ 2,
N
a2 a3 · · · an < 1.
n=2

106. (IMO, 1999) Let n be a fixed integer, with n ≥ 2.


(a) Determine the least constant C such that the inequality
⎛ ⎞4
 
xi xj (x2i + x2j ) ≤ C ⎝ xi ⎠
1≤i<j≤n 1≤i≤n

holds for all real numbers x1 , . . . , xn ≥ 0.


(b) For this constant C, determine when equality holds.

107. Let a, b, c, d be real numbers such that


 2    2
a + b2 − 1 c2 + d2 − 1 > ac + bd − 1 .

Prove that a2 + b2 − 1 > 0 and c2 + d2 − 1 > 0.

108. ([9]) Let x1 ,x2 ,x3 ,. . .,x100 be 100 positive integers such that
1 1 1
√ + √ + ··· + √ = 20.
x1 x2 x100

Prove that at least two of xj ’s are equal.


218 Problems

109. Let f (x) be a polynomial with integer coefficients and of degree n > 1.
Suppose f (x) = 0 has n real roots in the interval (0, 1), not all equal. If a is
the leading coefficient of f (x), prove that

|a| ≥ 2n + 1.

110. Show that the equation


x y z w
+ + + = m,
y z w x

has no solutions in positive reals for m = 2, 3.

111. Solve the system of equations

4z 2 4x2 4x2
x= , y= , z= ,
1 + 4z 2 1 + 4x2 1 + 4x2
for real numbers x, y, z.

112. Suppose a, b are nonzero real numbers and that all the roots of the real
polynomial

axn − axn−1 + an−2 xn−2 + · · · + a2 x2 − n2 bx + b = 0

are real and positive. Prove that all the roots are in fact equal.

113. Find all triples (a, b, c) of positive integers such that the product of any
two leaves the remainder 1 when divided by the third number.

114. (CRUX, 2000) Show that a triangle is equilateral if and only if

a 4 + b4 + c 4
a cos(β − γ) + b cos(γ − α) + c cos(α − β) = ,
abc
where a, b, c are the sides and α, β, γ are the angles opposite to the sides a, b, c
respectively.

115. ([7]) Find all positive solutions of the system

1 1 1 1
x1 + = 4, x2 + = 1, . . . , x1999 + = 4, x2000 + = 1.
x2 x3 x2000 x1

116. Find all positive solutions of the system

x+y+z = 1,
x3 + y 3 + z 3 + xyz = x4 + y 4 + z 4 + 1.
Problems 219

117. (Bulgaria, 2001) Let a, b be positive integers such that each equation
(a + b − x)2 = a − b, (ab + 1 − x)2 = ab − 1
has two distinct real roots. Suppose the bigger of these roots are the same.
Show that the smaller roots are also the same.

118. (Short-list, IMO-1989) Suppose the polynomial


P (x) = xn + nxn−1 + a2 xn−2 + · · · + an
has real roots α1 ,α2 ,α3 ,. . .,αn . If
α116 + α216 + · · · + αn16 = n.
Find α1 ,α2 ,α3 ,. . .,αn .

119. (IMO, 1972) Find all the solutions of the following system of inequalities:
 2  
(i) x1 − x3 x5 x22 − x3 x5 ≤ 0,
 2  2 
(ii) x2 − x4 x1 x3 − x4 x1 ≤ 0,
 2  2 
(iii) x3 − x5 x2 x4 − x5 x2 ≤ 0,
 2  2 
(iv) x4 − x1 x3 x5 − x1 x3 ≤ 0,
 2  2 
(v) x5 − x2 x4 x1 − x2 x4 ≤ 0.

120. (Short-list, IMO-1993) Solve the following system of equations, when a


is a real number such that |a| > 1:
x21 = ax2 + 1,
x22 = ax3 + 1,
.. .. ..
. . .
x2999 = ax1000 + 1,
x21000 = ax1 + 1.

121. (Indian
n Team Selection,1999)
n Let a1 ,a2 ,a3 ,. . .,an be n positive integers
such that j=1 aj = j=1 aj . Let Vn denote this common value. Show that
Vn ≥ n + s, where s is the least positive integer such that 2s − s ≥ n.
n
122. Let z1 ,z2 ,z3 ,. . .,zn be n complex numbers such that j=1 |zj | = 1. Prove
 
that there exists a subset S of the set z1 ,z2 ,z3 ,. . .,zn such that
 
  1
 z  ≥ .
 4
z∈S
220 Problems
   
123. ([1]) Let a1 ,a2 ,a3 ,. . .,an and b1 ,b2 ,b3 ,. . .,bn be two sequences of real
numbers which are not proportional. Let x1 ,x2 ,x3 ,. . .,xn be another sequence
of real numbers such that

n 
n
aj xj = 0, bj xj = 1.
j=1 j=1

Prove that
n

n
a2j
x2j
j=1
≥      2 .
a2j 2 −
n n n
j=1
j=1 b
j=1 j a b
j=1 j j

When does equality hold?


   
124. ([1]) Let a1 ,a2 ,a3 ,. . .,an and b1 ,b2 ,b3 ,. . .,bn be two sequences of real
numbers such that

b21 − b22 − · · · − b2n > 0 or a21 − a22 − · · · − a2n > 0.

Prove that
 2    2
a1 − a22 − · · · − a2n b21 − b22 − · · · − b2n ≤ a1 b1 − a2 b2 − · · · − an bn ,

and show that equality holds if and only if aj = λbj , 1 ≤ j ≤ n, for some real
constant λ.

125. ([1]) Let x1 ,x2 ,x3 ,. . .,xn be n positive real numbers. Prove that

n
xj
≤ n,
j=1
2xj + xj+1 + · · · + xj+n−2

where xn+k = xk .

126. ([1]) Let x1 ,x2 ,x3 ,. . .,xn be n ≥ 2 positive real numbers and k be a fixed
integer such that 1 ≤ k ≤ n. Show that
 x1 + 2x2 + · · · + 2xk−1 + xk 2n(k − 1)
≥ .
xk + xk+1 + · · · + xn n−k+1
cyclic

127. ([1]) Let z and ξ be two complex numbers such that |z| ≤ r, |ξ| ≤ r and
z = ξ. Show that for any natural number n, the inequality
 n 
 z − ξn  1  
  n−2 
z − ξ
 z − ξ  ≤ 2 n(n − 1)r

holds.
Problems 221
   
128. For any three vectors, x = x1 ,x2 ,x3 ,. . .,xn , y = y1 ,y2 ,y3 ,. . .,yn , and
z = z1 ,z2 ,z3 ,. . .,zn in Rn , prove that

||x|| + ||y|| + ||z|| − ||x + y|| − ||y + z|| − ||z + x|| + ||x + y + z|| ≥ 0.

129. Let A1 A2 A3 · · ·An+1 be a polygon with centre O, in which A1 = An+1 is


fixed and the remaining Aj ’s vary on the circle. Show that the area and the
perimeter of the polygon are the largest when the polygon is regular.

130. ([1]) A sequence an is said to be convex if an − 2an+1 + an+2 ≥ 0 for


all n ≥ 1. Let a1 ,a2 ,a3 ,. . .,a2n+1 be a convex sequence. Show that
a1 + a3 + a5 + · · · + a2n+1 a2 + a4 + a6 + · · · + a2n
≥ ,
n+1 n
and equality holds if and only if a1 ,a2 ,a3 ,. . .,a2n+1 is an arithmetic progression.

131. Suppose a1 ,a2 ,a3 ,. . .,an are n positive real numbers. For each k, define

xk = ak+1 + ak+2 + · · · + ak+n−1 − (n − 2)ak ,

where aj = aj−n for j > n. Suppose xk ≥ 0 for 1 ≤ k ≤ n. Prove that


"
n "
n
ak ≥ xk .
k=1 k=1

Show that for n = 3 the inequality is still true without the non-negativity of
xk ’s, but for n > 3 these conditions are essential.

132. Let a, c be positive reals and b be a complex number such that

f (z) = a|z|2 + 2 Re(bz) + c ≥ 0,

for all complex numbers z, where Re(z) denotes the real part of z. Prove that

|b|2 ≤ ac,

and   
f (z) ≤ a + c 1 + |z|2 .
Show that |b|2 = ac only if f (z) = 0 for some z ∈ C.

133. (IMO, 2003) Suppose x1 ≤x2 ≤x3 ≤. . .≤xn be n real numbers. Show that
 n    2  n n
n
  2 2 n − 1  2
x j − x k  ≤ xj − xk .
j=1
3 j=1
k=1 k=1

Prove also that equality holds if and only if the sequence xj is in arithmetic
progression.
222 Problems

134. (Short-list, IMO-2002) Suppose an is an infinite sequence of real num-


bers with the properties:

(i) there is some real constant c such that 0 ≤ an ≤ c, for all n ≥ 1;


  1
(ii) aj − ak  ≥ for all j, k with j = k.
j+k

Prove that c ≥ 1.

135. Let ABC be a right-angle triangle with medians ma , mb , mc . Let A B  C 


denote the triangle whose sides are ma , mb , mc . If R and R denote respectively
5
the circum-radii of ABC and A B  C  , prove that R ≥ R.
6

136. Let ABC be an equilateral triangle and D, E, F be arbitrary points on


the segments BC, CA, AB respectively. Prove that

1 1 1
[DEF ] + + ≥ 3.
[BDF ] [CED] [AF E]

(Here [XY Z] denotes the area of the triangle XY Z.)

137. Let the diagonals of a convex quadrilateral ABCD meet in P . Prove that
  
[AP B] + [CP D] ≤ [ABCD],

where as usual square-bracket denotes the area.

138. Let AD be the median from A on to BC of a triangle ABC; let r, r1 , r2


denote the in-radii of triangles ABC, ABD, ADC respectively. Prove that
 
1 1 1 2
+ ≥2 + .
r1 r2 r BC

139. (Japan, 2005) Let a, b, c be positive reals such that a + b + c = 1. Prove


that
 1/3  1/3  1/3
a 1+b−c +b 1+c−a +c 1+a−b ≤ 1.

140. Show that in a triangle ABC,


  
a2 m2a + b2 m2b + c2 m2c a2 + b2 + c2 ≥ 16m2a m2b m2c ,

where ma , mb , mc denote the medians on to the sides BC, CA, AB from A, B, C


respectively.
Problems 223

141. Let x1 ,x2 ,x3 ,. . .,xn be n positive reals which add up to 1. Find the
minimum value of
n
x
j .
j=1
1 + k=j xk

142. (IMO, 1974) Find all possible values of

a b c d
+ + + ,
a+b+d a+b+c b+c+d a+c+d
when a, b, c, d vary over positive reals.

143. Let Fn be the Fibonacci sequence defined by

F1 = F2 = 1, Fn+2 = Fn+1 + Fn , for n ≥ 1.

Prove that

n
Fj
< 2,
j=1
2j

for all n ≥ 1.

144. ([7]) Let P (x) = xn + an−1 xn−1 + · · · + a0 be a polynomial with real


coefficients such that |P (0)| = P (1). Suppose all the roots of P (x) = 0 are
real and lie in the interval (0, 1). Prove that the product of the roots does not
exceed 1/2n .

145. If x, y are real numbers such that


 √
2x + y + 8x2 + 4xy + 32y 2 = 3 + 3 2,

prove that x2 y ≤ 1.

146. (South Africa) If α, β, γ are the angles of a triangle whose circum-radius


is R and in-radius r, prove that
 
2 α−β 2r
cos ≥ .
2 R

147. (IMO, 1991) Let I be the in-centre of a triangle ABC. Suppose the
internal bisectors of angles A, B, C meet the opposite sides at A , B  and C  .
Prove that
1 AI · BI · CI 8
<   
≤ .
4 AA · BB · CC 27
224 Problems

148. (Proposed for IMO-1991) Determine the maximum value of


  
xj xk xj + xk ,
j<k

over all n-tuples (x1 ,x2 ,x3 ,. . .,xn ) of reals such that xj ≥ 0 for 1 ≤ j ≤ n.

149. If α, β, γ are the angles of a triangle, prove that


 
3 cos α ≥ 2 sin α sin β.
cyclic cyclic

150. (Indian Team Selection, 1994) Let x1 ,x2 ,x3 ,. . .,xN be positive real num-
bers. Prove that
  
N
 1/j N
x1 x2 · · · xj <3 xj .
j=1 j=1

151. (Indian TeamSelection, 2000) Let a1 ≤a2 ≤a3 ≤· · ·≤an be n real numbers
n
with the property j=1 aj = 0. Prove that


n
na1 an a2j ≤ 0.
j=1

152. (Indian Team Selection, 1997) Suppose a, b, c are positive real numbers.
Prove that
1 1 1 3
+ + ≥ .
a(1 + b) b(1 + c) c(1 + a) 1 + abc

153. Let x, y, z be positive real numbers such that x2 + y 2 + z 2 = 2. Prove


that
x + y + z ≤ 2 + xyz.
Find conditions under which equality holds.

154. (Wolschaum County Problem book) Let 0 ≤ x1 ≤x2 ≤x3 ≤· · ·≤xn be such
n
that j=1 xj = 1, where n ≥ 2 is an integer. If xn ≤ 2/3, prove that there
exists a k such that 1 ≤ k ≤ n and

1 
k
2
≤ xj ≤ .
3 j=1 3

155. (Vietnam, 1996) Let x, y, z be non-negative real numbers such that xy +


yz + zx + xyz = 4. Prove that

x + y + z ≥ xy + yz + zx.
Problems 225

156. (Canada) Let x, y, z be non-negative real numbers such that x+y +z = 1.


Prove that
4
x2 y + y 2 z + z 2 x ≤ .
27

157. Let x, y, z be real numbers and let p, q, r be real numbers in the interval
(0, 1/2) such that p + q + r = 1. Prove that
 2      
pqr x + y + z ≥ xyr 1 − 2r + yzp 1 − 2p + zxq 1 − 2q .

When does equality hold?

158. (Romania) Let x1 ,x2 ,x3 ,. . .,xn be n real numbers in the interval [0, 1].
Prove that
 n   n  + ,
n
xj − xj xj+1 ≤ ,
j=1 j=1
2

where xn+1 = x1 .

159. (IMO, 2005) Suppose x, y, z are positive real numbers such that xyz ≥ 1.
Prove that the inequality

x5 − x2 y5 − y2 z5 − z2
+ 5 + 5 ≥0
x5 2
+y +z 2 y +z +x2 2 z + x2 + y 2

holds.

160. (Indian Team Selection, 1992) Consider two sequences of positive real
numbers, a1 ≤a2 ≤a3 ≤· · ·≤an and b1 ≤b2 ≤b3 ≤· · ·≤bn , such that


n 
n
aj ≥ bj .
j=1 j=1

Suppose there exists a k, 1 ≤ k ≤ n, such that bj ≤ aj for 1 ≤ j ≤ k and


bj ≥ aj for j > k. Prove that

"
n "
n
aj ≥ bj .
j=1 j=1

161. (Bulgaria, 1997) Let a, b, c be positive real numbers such that abc = 1.
Prove that
1 1 1 1 1 1
+ + ≤ + + .
1+a+b 1+b+c 1+c+a 2+a 2+b 2+c
226 Problems

162. (PUTNAM) Let n ≥ 4 and let a1 ,a2 ,a3 ,. . .,an be real numbers such that

a1 + a2 + · · · + an ≥ n, a21 + a22 + · · · + a2n ≥ n2 .

Prove that
 
max a1 ,a2 ,a3 ,. . .,an ≥ 2.

163. Let x1 ≤x2 ≤x3 ≤· · ·≤xn+1 be n + 1 positive integers. Prove that


n+1 √ 1n2
xj+1 − xj
< .
j=1
xj+1 j=1
j

164. Let a, b, c be three positive real numbers which satisfy abc = 1 and
a3 > 36. Prove that
2 2
a < a2 + b2 + c2 − ab − bc − ca.
3

165. ([1]) Let z1 ,z2 ,z3 ,. . .,zn be n complex numbers and consider
 n positive
real numbers λ1 ,λ2 ,λ3 ,. . .,λn which have the property that 1/λj = 1. Prove
that
 2 
 n  n
 2
 z  ≤ λ j  zj  .
 j 
j=1 j=1

166. ([1]) Let a, b, c be three distinct real numbers. Prove that

     1/2
3 min a, b, c < a− a2 − ab
    1/2  
< a+ a2 − ab < 3 max a, b, c ,

where the sum is cyclically over a, b, c.

167. Suppose a, b, c are real numbers such that a3 + b3 + c3 = 0. Prove that


 3    
a2 ≤ (b − c)2 a4 ,

where the sum is cyclically over a, b, c.

168. Show that for all complex numbers z with real part of z > 1, the following
inequality holds:

|z n+1 − 1| > |z n ||z − 1|, for all n ≥ 1.


Problems 227

169. Suppose a, b, c are positive real numbers and let

x = a + b − c, y = b + c − a, z = c + a − b.

Prove that
abc(xy + yz + zx) ≥ xyz(ab + bc + ca).

170. Let a, b, c be positive real numbers. Prove that



 a3 3 cyclic ab
≥  .
b2 − bc + c2 cyclic a
cyclic

171. (CRUX, 2004) Let a1 , a2 , . . . , an < 1 be non-negative real numbers sat-


isfying
 √
a21 + a22 + · · · + a2n 3
a= ≥ .
n 3
Prove that
a1 a2 an na
2 + 2 + ··· + 2
≥ .
1 − a1 1 − a2 1 − an 1 − a2

172. (USAMO, 2001) Suppose a, b, c are non-negative real numbers such that
a2 + b2 + c2 + abc = 4. Prove that

0 ≤ ab + bc + ca − abc ≤ 2.

173. (CRUX, 2003) Suppose a, b, c are complex numbers such that |a| = |b| =
|c|. Prove that      
 ab   bc   ca  √
 + + 
 a2 − b2   b2 − c2   c2 − a2  ≥ 3.

174. (CRUX, 2001) Suppose x, y, z are non-negative real numbers such that
x2 + y 2 + z 2 = 1. Prove that

 x 3 3
(a) 1 ≤ ≤ ;
1 − yz 2
cyclic

 x √
(b) 1 ≤ ≤ 2.
1 + yz
cyclic

175. (CRUX, 2004) Let x, y, z be non-negative real numbers satisfying x + y +


z = 1. Prove that
2
xy 2 + yz 2 + zx2 ≥ xy + yz + zx − .
9
228 Problems

176. (CRUX, 1990) Let a, b, c, d be four positive real numbers such that a +
b + c + d = 2. Prove that
 a2 16
 2 ≤ .
2 25
cyclic a + 1

177. (IMO, 2001) Prove that


a b c
√ +√ +√ ≥1
a2 + 8bc 2
b + 8ca 2
c + 8ab
for all positive real numbers a, b and c.

178. Prove that in a triangle ABC, the inequality


A B C 9R2
tan + tan + tan ≤ ,
2 2 2 4[ABC]
holds.

179. (CRUX, 2002) Prove that in a triangle with angles α, β, γ, the inequality

 15 
sin α ≤ + cos(α − β)
4
cyclic

holds.

180. (KÖMAL) If x, y are real numbers such that x3 + y 4 ≤ x2 + y 3 , prove


that x3 + y 3 ≤ 2.

181. (CRUX, 2003) Let a, b, c be three positive real numbers. Prove that
 ab  a
≥ ,
c(c + a) c+a
where the sum is taken cyclically over a, b, c.

182. Prove that for any real x, and real numbers a, b,


 2
   a+b
sin x + a cos x sin x + b cos x ≤ 1 + .
2

183. (PUTNAM, 1988) Let x, y be two real numbers, where y is non-negative


and y(y + 1) ≤ (x + 1)2 . Prove that y(y − 1) ≤ x2 .

184. Let x, y, z be positive real numbers. Prove that


 1/2  1/3
xy + yz + zx (x + y)(y + z)(z + x)
≤ .
3 8
Problems 229

185. (Romania, 1997) Let a, b, c be positive real numbers such that abc = 1.
Show that
 a 9 + b9
≥ 2,
a 6 + a 3 b3 + b 6
where the sum is cyclical.

186. Let a, b, c be the sides of a triangle and set x = 2(s − a), y = 2(s − b),
z = 2(s − c), where s denotes the semi-perimeter. Prove that

abc(ab + bc + ca) ≥ xyz(xy + yz + zx).

187. (AMM) Let a1 ,a2 ,a3 ,. . .,an (n > 2) be positive real numbers and let s
denote their sum. Let 0 < β ≤ 1 be a real number. Prove that
n 
 β n 
 β
s − ak 2β ak
≥ (n − 1) .
ak s − ak
k=1 k=1

When does equality hold?

188. A point D on the segment BC of a triangle ABC is such that the in-radii
of ABD and ACD are equal, say r1 . Similarly define r2 and r3 . Prove that

s(s − a)
(i) 2r1 + 2 r = ha .
a
 
(ii) 2 r1 + r2 + r3 + s ≥ ha + hb + hc .

189. (Romania) For n ≥ 4, let a1 ,a2 ,a3 ,. . .,an be n positive real numbers such

n
that a2j = 1. Show that
j=1

a1 a2 an 4 √ √ √ 2
+ + · · · + ≥ a 1 a 1 + a 2 a 2 + · · · + a n an .
a22 + 1 a23 + 1 a21 + 1 5

190. (Belarus) Does there exist an infinite sequence xn of positive real num-
bers such that
√ √
xn+2 = xn+1 − xn ,
for all n ≥ 2.

191. (Belarus) Let a1 ,a2 ,a3 ,. . .,an be n positive real numbers and consider a
permutation b1 ,b2 ,b3 ,. . .,bn of it. Prove that

n
a2j 
n
≥ aj .
j=1
bj j=1
230 Problems

192. Let a1 ,a2 ,a3 ,.


. .,an and b
1 ,b2 ,b3 ,. . .,bn be two sequences of positive real
n n
numbers such that j=1 aj = j=1 bj = 1. Prove that


n
a2j 1
≥ .
j=1
a j + bj 2

193. Let x, y, z be positive real numbers. Prove that

y 2 − x2 z2 − y2 x2 − z 2
+ + ≥ 0.
z+x x+y y+z

194. (Japan, 2003) Find the greatest real value of k such that for every triple
(a, b, c) of positive real numbers, the inequality
 2   
a2 − bc > k b2 − ca c2 − ab

holds.

195. (Romania, 2003) Let a, b, c, d be positive real numbers such that abcd = 1.
Prove that
1 + ab 1 + bc 1 + cd 1 + da
+ + + ≥ 4.
1+a 1+b 1+c 1+d

196. (UK, 2004) Let a, b, c be the sides of a triangle such that a + b + c = 1,


and let n ≥ 2 be a natural number. Prove that
 1/n  1/n  1/n 21/n
a n + bn + bn + c n + cn + an <1+ .
2

197. Let a, b, c, d be positive real numbers. Prove that


 a
≥ 1.
b + 2c + d
cyclic

198. (Balkan Olympiads, 2005) Let a, b, c be positive real numbers such that
(a + b)(b + c)(c + a) = 1. Prove that

3
ab + bc + ca ≤ .
4

199. (Iran, 2005) Let ABC be a right-angled triangle with A = 90◦ . Let AD
be the bisector of angle A, and Ia be the ex-centre opposite to A. Prove that

AD √
≤ 2 − 1.
DIa
Problems 231

200. Let x, y, z be non-negative real numbers such that x + y + z = 1. Prove


that
x2 + y 2 + z 2 + 18xyz ≤ 1.

201. Let ABC be a triangle with circum-circle Γ, and G be its centroid. Extend
AG, BG, CG to meet Γ in D, E, F respectively. Prove that

AG + BG + CG ≤ GD + GE + GF.

202. Prove with usual notation that in a triangle ABC, the inequality
  
a + b + c ha + hb + hc ≥ 18Δ.

203. (Short-list, IMO-2004) Let a, b, c be three positive real numbers such that
ab + bc + ca = 1. Prove that
 1/3  1/3  1/3
1 1 1 1
+ 6b + + 6c + + 6a ≤ .
a b c abc

204. (IMO, 1974) Let ABC be a triangle. Show√that there exists a point D
on AB such that CD2 = AD · BD if and only if sin A sin B ≤ sin(C/2).

205. (Short-list, IMO-2004) Let a1 ,a2 ,a3 ,. . .,an be n > 1 positive real numbers.
For each k, 1 ≤ k ≤ n, let Ak = (a1 +a2 +· · ·+ak )/k. Let gn = (a1 a2 · · · an )1/n
and Gn = (A1 A2 · · · An )1/n . Prove that
 1/n
Gn gn
n + ≤ n + 1.
An Gn

Find the cases of equality.

206. Let x, y, z be real numbers in the interval [0, 1]. Prove that
   
3 x2 y 2 + y 2 z 2 + z 2 x2 − 2xyz x + y + z ≤ 3.

207. (UK, 1999) Let x, y, z be non-negative real numbers such that x+y+z = 1.
Prove that
7(xy + yz + zx) ≤ 2 + 9xyz.

208. (South Africa, 2003-04) Let x, y, z be real numbers in the interval [0, 1].
Prove that
x y z
+ + ≤ 2.
yz + 1 zx + 1 xy + 1
232 Problems

209. (Greece, 2003) Let a, b, c, d be positive reals such that a3 + b3 + 3ab =


c + d = 1. Prove that
 3  3  3  3
1 1 1 1
a+ + b+ + c+ + d+ ≥ 40.
a b c d

210. Let x, y, z be positive real numbers such that x + y + z = xyz. Prove that
1 1 1 3
√ + +√ ≤ .
1+x 2 1+y 2 1+z 2 2

211. Let P be an interior point of a triangle ABC whose sides are a, b, c. Let
R1 = P A, R2 = P B, and R3 = P C. Prove that
  
R1 R2 + R 2 R 3 + R3 R 1 R1 + R 2 + R3 ≥ a 2 R1 + b2 R2 + c 2 R 3 .
When does equality hold?

212. (IMO, 2004) Let t1 ,t2 ,t3 ,. . .,tn be positive real numbers such that
  
1 1 1
n 2 + 1 > t 1 + t2 + · · · + t n + + ··· + ,
t1 t2 tn
where n ≥ 3 is an integer. Show that for each triple (j, k, l) with 1 ≤ j < k <
l ≤ n, there is a triangle with sides tj , tk , tl .

213. (Moldova, 2004) Let x, y, z be non-negative real numbers. Prove that


√ √ √
x3 + y 3 + z 3 ≥ x2 yz + y 2 zx + z 2 xy.

214. (Romania, 2004) Find all positive real numbers such that
   
4 ab + bc + ca − 1 ≥ a2 + b2 + c2 ≥ 3 a3 + b3 + c3 .

215. Let a, b, c be positive real numbers such that abc = 1. Prove that
a b c 3
+ + ≥ .
(a + 1)(b + 1) (b + 1)(c + 1) (c + 1)(a + 1) 4

216. (CRUX, 2000) Suppose a, b, c are positive real numbers such that a2 +
b2 + c2 = 1. Prove that
 
1 1 1 2 a 3 + b3 + c 3
+ 2 + 2 ≥3+ .
a2 b c abc

217. (Taiwan, 1999) Let ABC be a triangle with circum-centre O and circum-
radius R. Suppose the line AO, when extended, meets the circum-circle of
OBC in D; similarly define E and F . Prove that
OD · OE · OF ≥ 8R3 .
Problems 233

218. (Proposed for IMO-1998) Let x, y, z be positive real numbers such that
xyz = 1. Prove that

x3 y3 z3 3
+ + ≥ .
(1 + y)(1 + z) (1 + z)(1 + x) (1 + x)(1 + y) 4

219. (Nordic Contest, 2003-04) Let D, E, F be respectively the points of


contact of the in-circle of a triangle ABC with its sides BC, CA, AB. Prove
that
BC CA AB
+ + ≥ 6.
F D DE EF

220. (Short-list, IMO-1990) Let a, b, c, d be non-negative real numbers such


that ab + bc + cd + da = 1. Show that

a3 b3 c3 d3 1
+ + + ≥ .
b+c+d c+d+a d+a+b a+b+c 3

221. (Bulgaria, 1974) Find all real λ for which the inequality

x21 + x22 + x23 ≥ λ(x1 x2 + x2 x3 ),

holds for all real numbers x1 , x2 , x3 .

222. Let a, b, c be positive real numbers such that abc = 1. Prove that
a b c 1 1 1
+ + ≥ + + .
b c a a b c

223. (Indian Team Selection, 2007) Let a, b, c be non-negative reals such that

a + b ≤ 1 + c, b + c ≤ 1 + a, c + a ≤ 1 + b.

Prove that
a2 + b2 + c2 ≤ 2abc + 1.

224. (Indian Team Selection, 1997) If a, b, c are non-negative real numbers


such that a + b + c = 1 then show that
a b c 9
+ + ≥ .
1 + bc 1 + ca 1 + ab 10

225. (Indian Team Selection, 2006) Let ABC be a triangle with sides a, b, c,
circum-radius R and in-radius r. Prove that
 2
R 64a2 b2 c2
≥  2    .
2r 4a − (b − c)2 4b2 − (c − a)2 4c2 − (a − b)2
234 Problems

226. (INMO, 2004) Let R denote the circum-radius of a triangle ABC; a, b, c


its sides BC, CA, AB; and ra , rb , rc its ex-radii opposite A, B, C. If 2R ≤ ra ,
prove that
(i) a > b and a > c;
(ii) 2R > rb and 2R > rc .

227. (USSR, 1974) Given a square grid S containing 49 points in 7 rows and 7
columns, a subset T consisting of k points is selected. What is the maximum
value of k such that no four points of T determine a rectangle R having sides
parallel to the sides of S?

228. (Proposed for IMO-1977) Let a, b, c, d be real numbers such that 0 ≤ a ≤


b ≤ c ≤ d. Prove that
a b bc c d d a ≥ b a c b d c a d .

229. (Ukraine, 2001) In a triangle ABC, let AA1 , CC1 be the bisectors of the
angles A, C respectively. Let M be point on the segment AC, and I be the
in-centre of ABC. Draw a line through M , parallel to AA1 and let it meet
CC1 in N and BC in Q. Similarly, let the line through M parallel to CC1
meet AA1 in H and AB in P . Let d1 , d1 , d3 respectively denote the distances
of H, I, N from the line P Q. Prove that
d1 d2 d3 2ab 2bc 2ca
+ + ≥ 2 + 2 + 2 .
d2 d3 d1 a + bc b + ca c + ab

230. (CRMO, 1996) Let ABC be a triangle and ha be the altitude through
A. Prove that
(b + c)2 ≥ a2 + 4h2a .
(As usual a, b, c denote the sides BC, CA, AB respectively.)

231. (CRMO, 2003) Let a, b, c be three positive real numbers such that
a + b + c = 1. Prove that among the three numbers a − ab, b − bc, c − ca there
is one which is at most 1/4 and there is one which is at least 2/9.

232. (CRMO, 2004) Let x and y be positive real numbers such that y 3 + y ≤
x − x3 . Prove that
(a) y < x < 1; and
(b) x2 + y 2 < 1.

233. (CRMO, 2005) Let a, b, c be three positive real numbers such that a +
b + c = 1. Let
 
λ = min a3 + a2 bc, b3 + ab2 c, c3 + abc2 .
Problems 235

Prove that the roots of the equation x2 + x + 4λ = 0 are real.

234. (CRMO, 2006) If a, b, c are three positive real numbers, prove that
a 2 + 1 b2 + 1 c 2 + 1
+ + ≥ 3.
b+c c+a a+b

235. (Hungary, 1990) If d is the largest among the positive numbers a, b, c, d,


prove that
a(d − b) + b(d − c) + c(d − a) ≤ d2 .

236. (INMO, 2007) If x, y, z are positive real numbers, prove that


  2    
x + y + z)2 yz + zx + xy ≤ 3 y 2 + yz + z 2 z 2 + zx + x2 x2 + xy + y 2 .

237. (USAMO, 1974) Suppose a, b, c are positive real numbers. Prove that
aa bb cc ≥ (abc)(a+b+c)/3 .

238. (Bulgaria, 1988) Find all real p and q for which the equation
8p2 3
x4 − x + 4qx3 − 3px + p2 = 0
q
has four positive roots.

239. (Russia, 2005) Let a1 , a2 , a3 be real numbers, each greater than 1. Let
S = a1 + a2 + a3 and suppose S < a2j /(aj − 1) for j = 1, 2, 3. Prove that
1 1 1
+ + > 1.
a1 + a2 a2 + a3 a3 + a1

240. (Macedonia, 2010) Let a, b, c be positive real numbers such that ab + bc +


ca = 1/3. Prove that
a b c 1
+ 2 + 2 ≥ .
a2 − bc + 1 b − ca + 1 c − ab + 1 a+b+c

241. (Balkan Olympiads, 2010) Suppose a, b, c are positive real numbers. Prove
that
a2 b(b − c) b2 c(c − a) c2 a(a − b)
+ + ≥ 0.
a+b b+c c+a

242. (Macedonia, 2010) Let a1 , a2 , a3 , . . . , an be n > 2 positive real numbers


such that a1 + a2 + a3 + · · · + an = 1. Prove that

n
a1 a2 · · · aj−1 aj+1 · · · an 1
≤ .
j=1
aj + n − 2 (n − 1)2
236 Problems

243. (Thailand, 2014) Determine the largest value of k such that the inequality
     
a b  c a b c b c a
k+ k+ k+ ≥ + + + +
b c ba b c a a b c

holds for all positive real numbers a, b, c.

244. (Macedonia, 2010) Let x1 , x2 , x3 , . . . , xn be n ≥ 3 positive real numbers.


Prove that
x1 x3 x2 x4 xn−1 x1 xn x2
+ + ··· + + ≤ n − 1.
x1 x3 + x2 x4 x2 x4 + x 3 x5 xn−1 x1 + xn x2 xn x2 + x1 x3

245. Let a1 , a2 , a3 , . . . , a2017 be positive real numbers. Prove that


2017
 aj 2017
≥ ,
j=1
aj+1 + aj+2 + · · · + aj+1008 1008

where the indices are taken modulo 2017.

246. (Russia, 2014) Let a, b, c be three positive real numbers such that ab +
bc + ca = 1. Prove that
  
1 1 1 √ √ √
a + + b + + c + ≥ 2( a + b + c).
a b c

247. (Macedonia, 2010) Let a, b, c be positive real numbers such that a+b+c =
3. Prove that
a 3 + 2 b3 + 2 c 3 + 2
+ + ≥ 3.
b+2 c+2 a+2

248. (Russia, 2014) Let a, b, c, d be real numbers such that a2 +b2 +c2 +d2 = 4.
Prove that (2 + a)(2 + b) ≥ cd.

249. (Belarus, 2014) Find all real λ such that



a+b √ a 2 + b2
≥ λ ab + (1 − λ)
2 2
holds for all positive real numbers a, b.

250. (Russia, 2014) Let a, b, c, d be real numbers having absolute value greater
than 1 and such that abc + abd + acd + bcd + a + b + c + d = 0. Prove that
1 1 1 1
+ + + > 0.
a−1 b−1 c−1 d−1
Problems 237

251. (Belarus, 2014) Show that

1 1 1
− < ,
x + y + 1 (x + 1)(y + 1) 11

for all positive real numbers x, y.

252. (Kazaksthan, 2008) Let a, b, c be three positive real numbers such that
abc = 1. Prove that
1 1 1 3
+ + ≥ .
b(a + b) c(b + c) a(c + a) 2

253. (Belarus, 2014) Let a, b, c be positive real numbers such that a + b + c = 1.


Prove that
a2 b2 c2 9
3
+ 3
+ 3

(b + c) (c + a) (a + b) 8

254. (Belarus, 2014) Suppose a, b, c are positive reals such that ab + bc + ca ≥


a + b + c. Prove that

(a + b + c)(ab + bc + ca) + 3abc ≥ 4(ab + bc + ca).

255. (Kazaksthan, 2008) Let a, b, c, d be four real numbers such that a + b +


c + d = 0. Prove that

(ab + ac + ad + bc + bd + cd)2 + 12 ≥ 6(abc + abd + acd + bcd).

256. (Vietnam, 2014) Consider the expression

x3 y 4 z 3 y 3 z 4 x3 z 3 x4 y 3
P = + 4 + 4 .
(x4 4 2
+ y )(xy + z ) 3 4 2
(y + z )(yz + x ) 3 (z + x4 )(zx + y 2 )3

Find the maximum value of P when x, y, z vary over the set of all positive real
numbers.

257. (Ukraine, 2014) Let a, b, c be the sides of an acute-angled triangle. Prove


that
   
a2 + b2 − c2 + b2 + c2 − a2 + c2 + a2 − b2 ≤ 3(ab + bc + ca).

258. (Ukraine, 2014) Let x1 , x2 , x3 , . . . , xn be Positive real numbers such that


x1 x2 x3 · · · xn = 1. Let S = x31 + x32 + x33 + · · · + x3n . Prove that
x1 x2 x3 xn
+ + + ··· + ≤ 1.
S − x31 + x21 S − x32 + x22 S − x33 + x23 S − x3n + x2n
238 Problems

259. (Ireland, 2014) Let a1 , a2 , a3 , . . . , an be n(> 1) positive real numbers


a2
whose sum is 1. Define bk = n k a2 , 1 ≤ k ≤ n. Prove that
j=1 j


n
ak  bk
n
≤ .
1 − ak 1 − bk
k=1 k=1

260. (Bulgaria, 2014) Suppose a, b, c, d are positive real numbers. Prove that
 a4 a+b+c+d
≥ .
a + a b + ab2 + b3
3 2 4
cyclic

261. (Turkey, 2014) Let a, b, c be non-negative real numbers satisfying a2 +


b2 + c2 = 1. Prove that
√ √ √
a + b + b + c + c + a ≥ 5abc + 2.

262. (Turkey, 2014) Let x, y, z be positive real numbers such that x2 +y 2 +z 2 ≤


x + y + z. Prove that
x2 + 3 y 2 + 3 z 2 + 3
+ + ≥ 6.
x3 + 1 y 3 + 1 z 3 + 1

263. (Belarus, 2010) For any three positive real numbers a, b, c, prove that
a2 b2 3a + 2b − c
+ ≥ .
a+b b+c 4

264. (APMO, 2003) Let a, b, c be the sides of a triangle with perimeter equal
to 1. Prove that

   2
2 2 2 2
a +b + b +c + c +a <1+ 2 2 .
2

265. (Tajikisthan, 2014) Let a, b, c be the sides of a triangle. Prove that


  
a2 + ab + b2 + b2 + bc + c2 + c2 + ca + a2

≤ 5(a2 + b2 + c2 ) + 4(ab + bc + ca).

266. (Tajikisthan, 2014) Suppose a, b, c are non-negative real numbers such


that a3 + b3 + c3 + abc = 4. Prove that
a3 b + b3 c + c3 a ≤ 3.

267. (JBMO, 2014) Let a, b, c be positive real numbers such that abc = 1.
Prove that
 2  2  2
1 1 1
a+ + b+ + c+ ≥ 3(a + b + c + 1).
b c a
Problems 239

268. (Tajikisthan, 2014) Let a, b, c be positive real numbers with abc = 1.


Prove that
a b c 3
+ + ≥ .
c(a + 1) a(b + 1) b(c + 1) 2

269. (Estonia, 2014) Let a, b, c be positive real number such that abc = 1.
Prove that
1 1 1
+ + > 1.
1 + a2014 1 + b2014 1 + c2014

270. (CRUX, 2000) For positive real numbers a, b, c, prove the inequality
  
1 1 1 1 1 1 9
+ + + + ≥ .
a b c 1+a 1+b 1+c 1 + abc

271. ([10]) Let x, y, z be positive real numbers such that x + y + z = 3. Prove


that √ √

x + y + z ≥ xy + yz + zx.

272. ([10]) Let a, b, c be positive real numbers. Prove that


9abc ab2 bc2 ca2 a 2 + b2 + c 2
≤ + + ≤ .
2(a + b + c) a+b b+c c+a 2

273. ([10]) For positive real numbers a, b, c, prove that


abc 1
≤ .
(1 + a)(a + b)(b + c)(c + 16) 81

274. ([10]) Let a, b, c, d be positive real numbers such that a + b + c + d =


4. Prove that
1 1 1 1
2
+ 2 + 2 + 2 ≥ 2.
a +1 b +1 c +1 d +1

275. Let a, b, c be the sides of a triangle. Prove that inequality


64(s − a)(s − b)(s − c) ≤ (a + b)(b + c)(c + a),
where s = (a + b + c)/2 is the semi-perimeter of the triangle.

276. (Estonia, 2014) Let a, b, c be positive real numbers. Prove that


1 + ab 1 + bc 1 + ca  2  
+ + ≥ a + 2 + b2 + 2 + c2 + 2.
c a b

277. (Estonia, 2014) Let I Be the incentre of a triangle ABC. Let RA , RB , RC


be respectively the circum-radii of triangles BCI, CAI, ABI. If R is the
circum-radius of ABC, prove that
RA + RB + RC ≤ 3R.
240 Problems

278. (Estonia, 2014) Let a, b, c be positive real numbers such that a+b+c = 1.
Prove that
a2 b2 c2 1
+ + > .
b 3 + c 4 + 1 c 3 + a 4 + 1 a 3 + b4 + 1 5

279. (Balkan Olympiads, 2014) Let x, y, z be three positive real numbers such
that xy + yz + zx = 3xyz. Prove that

x2 y + y 2 z + z 2 x ≥ 2(x + y + z) − 3.

280. (Janous inequality) Let a, b, c and x, y, z be two sets of positive real


numbers. Prove that
x y z 
(b + c) + (c + a) + (a + b) ≥ 3(ab + bc + ca).
y+z z+x x+y

281. Let x, y, z be positive real numbers such that xy + yz + zx = 1. Prove


that √
x y z 3 3
+ + ≤ .
x2 + 1 y 2 + 1 z 2 + 1 4

282. Suppose x, y, z are positive real numbers such that x + y + z = 1. Prove


that    
1 1 1
1+ 1+ 1+ ≥ 64.
x y z

283. Let x, y, z be positive real numbers such that x + y + z = 1. Prove that


1 1 1 27
+ + ≤ .
1 − xy 1 − yz 1 − zx 8

284. Let x, y, z be positive real numbers such that x + y + z = 1. Show that


z − xy x − yz y − zx
+ 2 + 2 ≥ 2.
x2 + xy + y 2 y + yz + z 2 z + zx + x2

285. ([10]) Let a, b, c be positive real numbers define

u2 − v 2
u = a + b + c, = ab + bc + ca, w = abc,
3
where v ≥ 0. Then
(u + v)2 (u − 2v) (u − v)2 (u + 2v)
≤r≤ .
27 27

286. ([10]) Let a, b, c be positive real numbers. Prove that

a4 + b4 + c4 ≥ abc(a + b + c).
Problems 241

287. Let a, b, c be real numbers such that a2 + b2 + c2 = 9. Prove that

2(a + b + c) − abc ≤ 10.

288. (Serbia, 2008) Let a, b, c be positive real numbers such that a + b + c = 1.


Prove that
9
a2 + b2 + c2 + 3abc ≥ .
4

289. (China, 2004) Determine the maximum value of λ such that

a+b+c≥λ
√ √ √
for all positive reals a, b, c with a bc + b c + c a ≥ 1.

290. (China, 2004) If a, b, c are real numbers such that a + b + c = 1, prove


that
10(a3 + b3 + c3 ) − 9(a5 + b5 + c5 ) ≥ 1.

291. (IMO, 2004) Suppose a1 , a2 , a3 , . . . , an are n positive real numbers such


that
  1 1 1 1

a1 + a2 + a3 + · · · + an + + + ··· + < n2 + 1.
a1 a2 a3 an

Show that for any three distinct numbers j, k, l the numbers aj , ak , al form the
sides of a triangle.

292. (Colombia, 2013) Let a, b, c be positive real numbers. Prove that


  8
24abc ≤ a3 + b3 + c3 − (a + b + c)3  ≤ (a + b + c)3 .
9
Show further that equality holds in both the inequalities if and only if a = b = c.

293. (Czech and Slovak, 2013) Two circles Γ1 , Γ2 with respective centre S1 , S2
and radii r1 , r2 are externally tangent to each other and lie in a square ABCD
of side a units so that Γ1 touches DC, DA while Γ2 touches CD, CB. Prove
that the area of at least one of the triangles AS1 S2 and BS1 S2 is no more than
3 2
16 a units.

294. (Czech and Slovak, 2013) Find all λ > 0 such that the inequality
  √
a2 + λb2 + b2 + λa2 ≥ a + b + (λ − 1) ab

holds for all positive real numbers a and b.


242 Problems

295. (Netherlands, 2013) Let a, b, c be positive real numbers such that abc = 1.
Prove that 
1
a+b+c≥ (a + 2)(b + 2)(c + 2).
3

296. (Belarus, 2013) Let x1 , x2 , x3 , . . . , xn be n ≥ 3 positive real numbers such


that x1 x2 x3 · · · xn = 1. Prove that

n
x8j n
≥ ,
x (x4
j=1 j+1 j
+ x4j+1 ) 2

where xn+1 = x1 .

1 1 1
297. (Belarus, 2013) Let a, b, c be positive real number such that ab + bc + ca =
1. Prove that
a2 + b2 + c2 + ab + bc + ca − 3 a b c
≥ + + .
5 b c a

298. (Turkey, 2012) Show that for all positive real numbers x, y, z, the inequal-
ity
x(2x − y) y(2y − z) z(2z − x)
+ + ≥ 1.
y(2z + x) z(2x + y) x(2y + z)

299. (Turkey, 2012) Suppose


z(xz + yz + y)
≤ K,
xyy2 + z 2 + 1
for all real numbers x, y, z ∈ (−2, 2) with x2 + y 2 + z 2 + xyz = 4. Find the
smallest value of K.

300. (Turkey, 2012) Suppose a, b, c are positive real numbers such that a3 +
b3 + c3 = a4 + b4 + c4 . Prove that
a b c
+ 2 + 2 ≥ 1.
a2 3
+b +c 3 3
b +c +a 3 c + c3 + a3

301. (Turkey, 2013) Let a, b, c be positive real numbers such that a + b + c = 1.


Prove that
a4 + 5b4 b4 + 5c4 c4 + 5a4
+ + ≥ 1 − (ab + bc + ca).
a(a + 2b) b(b + 2c) c(c + 2a)

302.(USAMO, 1998) Let a0 , a1 , a2 , . . . , an be real numbers in the interval


0, π2 such that
 π  π  π  π
tan a0 − tan a1 − tan a2 − · · · tan an − ≥ n − 1.
4 4 4 4
Problems 243

Prove that
tan(a0 ) tan(a1 ) tan(a2 ) · · · tan(an ) ≥ nn+1 .

303. (Iran, 1996) Let x, y, z be positive real numbers. Prove that


 
1 1 1 9
(xy + yz + zx) 2
+ 2
+ ≥ .
(x + y) (y + z) (z + x)2 4

304. Suppose a, b, c are positive real numbers such that abc = 1. Prove that
 a2 + bc
≥ ab + bc + ca.
a2 (b + c)
cyclic

305. Let a, b, c be non-negative real numbers. Prove that


 
4 a3 + b3 + c3 + 15abc ≥ (a + b + c)3 .

306. Let a, b, c be positive real numbers such that a + b + c = 1. Prove that

1 1 1 3
+ + ≤ .
a 4 + b + c b4 + c + a c 4 + a + b a+b+c

307. Let a, b, c be positive reals. Prove that

1
a4 (b + c) + b4 (c + a) + c4 (a + b) ≤ (a + b + c)5 .
12

308. Suppose a, b, c are positive reals such that ab + bc + ca = 1. Prove that

1 1 1 1
+ + − ≥ 2.
a+b b+c c+a a+b+c

309. Let a, b, c be positive real numbers such that ab + bc + ca = 1. Prove that

1 + a 2 b2 1 + b2 c 2 1 + c 2 a2 5
2
+ 2
+ ≥ .
(a + b) (b + c) (c + a)2 2

310. [10] Let a, b, c, d be non-negative real numbers. Prove that

a4 + b4 + c4 + d4 + 2abcd ≥ a2 b2 + a2 c2 + a2 d2 + b2 c2 + b2 d2 + c2 d2 .

311. (Kvant, 1988) Let a, b, c be positive real numbers. Prove that


 
1 1 1 a b c (a + 1)(b + 1)(c + 1)
3+a+b+c+ + + + + + ≥3 .
a b c b c a 1 + abc
244 Problems

312. (Indian Team Selection, 2017) Let a, b, c be distinct positive real numbers
such that abc = 1. Prove that
 a6
> 15.
(a − b)(a − c)
cyclic

313. (Indian Team Selection, 2010) Let a, b, c be real numbers such that a2 +
b2 + c2 = 1. Prove that

a + b + c ≤ 2abc + 2.

314. (Japan, 1997) Let a, b, c be positive real numbers. Prove that

(b + c − a)2 (c + a − b)2 (a + b − c)2 3


+ + ≥ .
a2 + (b + c)2 b2 + (c + a)2 c2 + (a + b)2 5

315. (Ukraine, 2005) Let a, b, c be positive real numbers such that a+b+c = 1.
Prove that
     
1 1 1 1 1 1
−1 −1+ −1 −1+ −1 − 1 ≥ 6.
a b b c c a
Chapter 6
Solutions to problems

1. Let a1 , a2 , . . . , an , an+1 be n + 1 positive real numbers such that a1 + a2 +


· · · + an = an+1 . Prove that

n 
 
  n
 
aj an+1 − aj ≤ ! an+1 an+1 − aj .
j=1 j=1

Solution: We observe that


  
n
  n
an+1 an+1 − aj = na2n+1 − aj an+1
j=1 j=1

= (n − 1)a2n+1 .

Thus, it is sufficient to prove that


n 
 
aj (an+1 − aj ) ≤ (n − 1) an+1 .
j=1

However, this follows from the Cauchy-Schwarz inequality:


n 
 n 
 
 
aj an+1 − aj = aj (an+1 − aj )
j=1 j=1
 

n 
n
 
≤ ! aj ! an+1 − aj
j=1 j=1
 
= an+1 (n − 1)an+1

= (n − 1) an+1 .

2. If a, b, c are positive real numbers, prove that

a b c
+ + ≥ 1.
b + 2c c + 2a a + 2b
246 Solutions

Solution: The inequality may be written in an equivalent form:


a2 b2 c2
+ + ≥ 1.
ab + 2ca bc + 2ab ca + 2bc
Using the Cauchy-Schwarz inequality, we have
⎛ ⎞2
 a √
(a + b + c)2 = ⎝ √ ab + 2ca⎠
cyclic
ab + 2ca
⎛ ⎞⎛ ⎞
 a 2 
≤ ⎝ ⎠⎝ (ab + 2ca)⎠ .
ab + 2ca
cyclic cyclic

It follows that
 a2 (a + b + c)2
≥ .
ab + 2ca 3(ab + bc + ca)
cyclic

Thus, it is sufficient to prove that

(a + b + c)2 ≥ 3(ab + bc + ca).

Equivalently, we need to prove that a2 +b2 +c2 ≥ ab+bc+ca, which is clear from
the Cauchy-Schwarz inequality. (Or one can use (a−b)2 +(b−c)2 +(c−a)2 ≥ 0.)

3. Let a, b, c be positive real numbers such that abc ≤ a + b + c. Prove that



a2 + b2 + c2 ≥ 3(abc).

Solution: We observe that


 2    
a + b + c − 3 a2 + b2 + c2 = 2 ab + bc + ca − a2 − b2 − c2 ≤ 0.

Thus, we get
    2  2
3 a2 + b2 + c2 ≥ a + b + c ≥ abc .
On the other hand, we also have
 2/3
a2 + b2 + c2 ≥ 3 abc ,

as a consequence of the AM-GM inequality. This gives


 2 3  2
a + b2 + c2 ≥ 27 abc .

Multiplying these inequalities, we get


 4  4
9 abc ≤ a2 + b2 + c2 .

Taking the fourth-root, we get the desired inequality.


Solutions 247

4. For any positive real numbers a, b, c, prove that

2 2 2 27
+ + ≥ .
b(a + b) c(b + c) a(c + a) (a + b + c)2

3

3
Solution: Taking A = abc and B = (a + b)(b + c)(c + a), we have

2 2 2 6
+ + ≥ ,
b(a + b) c(b + c) a(c + a) AB

by the AM-GM inequality. However, the AM-GM inequality also gives

a+b+c 2(a + b + c)
A≤ , B≤ .
3 3
Thus
6 27
≥ .
AB (a + b + c)2

5. Let a, b, c be three sides of a triangle such that a + b + c = 2. Prove that

1
1 ≤ ab + bc + ca − abc ≤ 1 + .
27

Solution: We have

(1 − a)(1 − b)(1 − c) = 1 − (a + b + c) + (ab + bc + ca) − abc


= −1 + (ab + bc + ca) − abc.

Thus it is sufficient to prove that

1
0 ≤ (1 − a)(1 − b)(1 − c) ≤ .
27
But, this is a consequence of the AM-GM inequality: observe a < b + c = 2 − a
so that 1 − a > 0, and similarly 1 − b > 0, 1 − c > 0;
 3
1−a+1−b+1−c 1
(1 − a)(1 − b)(1 − c) ≤ = .
3 27

6. If a, b, c be positive real numbers such that a + b + c = 1, prove that


√ √ √ √ √ √
ab + c + bc + a + ca + b ≥ 1 + ab + bc + ca.
248 Solutions
√ √
Solution: First √ we observe that ab + c ≥ ab + c. √ In fact, this is equivalent
to c ≥ c2 + 2c ab, which is equivalent
√ to 1 − c ≥ 2 ab, √since c > 0.√ Using
1
√ − c = a +
√ b, we get a + b ≥ 2 ab. Similarly, we derive bc + a ≥ bc + a,
ca + b ≥ ca + b. Thus it follows
√ √ √ √ √ √
ab + c + bc + a + ca + b ≥ ab + bc + ca + a + b + c
√ √ √
= 1 + ab + bc + ca.

7. Let a, b, c, d be positive real numbers such that abcd = 1. Prove that

1 + ab 1 + bc 1 + cd 1 + da
+ + + ≥ 4.
1+a 1+b 1+c 1+d

Solution: Using cd = 1/ab and da = 1/bc, the inequality to be proved is


 
1 1 1 1
(1 + ab) + + (1 + bc) + ≥ 4.
1 + a ab(1 + c) 1 + b bc(1 + d)

But,

1 1 4
+ ≥
1 + a ab(1 + c) 1 + a + ab + abc
1 1 4
+ ≥ .
1 + b bc(1 + d) 1 + b + bc + bcd

Thus
 1 + ab 4(1 + ab) 4(1 + bc)
≥ +
1+a 1 + a + ab + abc 1 + b + bc + bcd
cyclic
4(1 + ab) 4a(1 + bc)
= +
1 + a + ab + abc a + ab + abc + abcd
4(1 + ab) 4a(1 + bc)
= +
1 + a + ab + abc 1 + a + ab + abc
= 4.

8. If a, b, c, d are positive real numbers, prove that


 
a 2 + b2 + c 2 + d 2 3 abc + abd + acd + bcd
≥ .
4 4
Solutions 249

Solution: Using the AM-GM inequality, we have


abc + abd + acd + bcd
4 
1 ab(c + d) cd(a + b)
= +
2 2 2

1 (a + b)2 (c + d) (a + b)(c + d)2
≤ +
2 8 8
(a + b) (c + d) (a + b + c + d)
= · ·
2 2 4
 3
a+b+c+d
≤ .
4
This shows that
 
3 abc + abd + acd + bcd a+b+c+d a 2 + b2 + c 2 + d 2
≤ ≤ ,
4 4 4
where we have used the root-mean-square inequality in the last step.

9. Let a, b, c be the sides of a triangle such that a + b + c = 2. Prove that


a2 + b2 + c2 + 2abc < 2.

Solution: Note that s = (a + b + c)/2 = 1. Using s − a > 0, it follows that


a < 1. Similarly, b < 1 and c < 1. We have
a2 + b2 + c2 + 2abc = (a + b + c)2 − 2(ab + bc + ca) + 2abc
 
= 2 2 − (ab + bc + ca) + abc .
Thus, we have to prove that
2 − (ab + bc + ca) + abc < 1.
Equivalently,
1 − (a + b + c) + (ab + bc + ca) − abc > 0.
But this is precisely (1 − a)(1 − b)(1 − c) > 0, which follows from a < 1, b < 1,
c < 1.
Alternate Solution:
We use the transformation a = y + z, b = z + x, c = x + y. This is possible
because a, b, c are the sides of a triangle. Then, x, y, z are positive real numbers
such that x + y + z = 1. The inequality gets transformed to
   
x2 + xy + x2 y + xy 2 + 2xyz < 1.
cyclic cyclic cyclic cyclic

This reduces to 1 − xyz < 1 after using x + y + z = 1. The result follows from
xyz > 0.
250 Solutions

10. If a, b, c are positive real numbers such that a2 + b2 + c2 = 1, prove that


 
1 1 1 √
+ + + a + b + c ≥ 4 3.
a b c

Solution: Using the AM-GM inequality, we have


1 1 1 9
+ + ≥ .
a b c a+b+c
Thus, it is sufficient to prove that
9 √
+ a + b + c ≥ 4 3.
a+b+c
Equivalently, we need to prove

(a + b + c)2 − 4 3 (a + b + c) + 9 ≥ 0.
Or  √  √ 
a+b+c− 3 a + b + c − 3 3 ≥ 0.
However, we have
√  √
a + b + c ≤ 3 a2 + b2 + c2 = 3.

Hence a + b + c − 3 3 ≤ 0 and the inequality follows.

11. Find all triplets (a, b, c) of positive real numbers which satisfy the system
of equations:
a+b+c = 6,
1 1 1 4
+ + = 2− .
a b c abc
Solution: By the AM-GM inequality, we have
 3
a+b+c
abc ≤ = 8.
3
Thus
4 4 3
2− ≤2− = .
abc 8 2
On the other hand,
1 1 1 9 3
+ + ≥ = .
a b c a+b+c 2
Thus,
3 1 1 1 4 3
≤ + + =2− ≤ .
2 a b c abc 2
This shows that equality holds in the AM-GM inequality. Hence a = b = c,
giving (a, b, c) = (2, 2, 2).
Solutions 251

12. Let a, b, c be real numbers such that a2 + b2 + c2 = 1. Prove that


a2 b2 c2 3
+ + ≥ .
1 + 2bc 1 + 2ca 1 + 2ab 5

Solution: Using the AM-GM inequality, we have 2bc ≤ b2 + c2 = 1 − a2 and


similar bounds hold for 2ca, 2ab. Thus,

a2 b2 c2 a2 b2 c2
+ + ≥ + +
1 + 2bc 1 + 2ca 1 + 2ab 2 − a2 2 − b2 2 − c2
 
1 1 1
= −3 + 2 + + .
2 − a2 2 − b2 2 − c2
Thus, it is sufficient to prove that
 
1 1 1 1 3 9
+ + ≥ +3 = .
2 − a2 2 − b2 2 − c2 2 5 5
This follows from AM-HM inequality:
1 1 1 9 9
+ + ≥ = .
2 − a2 2 − b2 2 − c2 6 − (a2 + b2 + c2 ) 5

13. Let a, b, c and α, β, γ be positive real numbers such that α + β + γ = 1.


Prove that
  
αa + βb + γc + 2 αβ + βγ + γα ab + bc + ca ≤ a + b + c.

Solution: Introduce new variables


a b c
x= , y= , z= .
a+b+c a+b+c a+b+c
The inequality may be written in the form
  
αx + βy + γz + 2 αβ + βγ + γα xy + yz + zx ≤ 1.

Observe that x + y + z = 1. Using the AM-GM inequality


  
αx + βy + γz + 2 αβ + βγ + γα xy + yz + zx
α 2 + x2 β 2 + y2 γ2 + z2    
≤ + + + αβ + βγ + γα + xy + yz + zx
2 2 2
1 1
= (α + β + γ) + (x + y + z)2 = 1.
2
2 2
252 Solutions

14. Prove that for all real numbers a, b, the inequality


 
a 2 + b2 + 1 > a b2 + 1 + b a 2 + 1

holds.

Solution: By the Cauchy-Schwarz inequality,


   2   
a b2 + 1 + b a 2 + 1 ≤ a 2 + b2 a 2 + b 2 + 2 .

But it is easy to see that x(x + 2) < (x + 1)2 for any positive x. Thus
    2
a 2 + b 2 a 2 + b2 + 2 < a 2 + b 2 + 1 ,

which gives the desired inequality.

15. For a fixed positive integer n, compute the minimum value of the sum

x22 x3 xn
x1 + + 3 + ··· + n,
2 3 n
where x1 , x2 , x3 , . . . , xn are positive real numbers such that
1 1 1 1
+ + + ··· + = n.
x1 x2 x3 xn

Solution: Let S denote the sum to be minimised. Define


1 1 1 1
H =1+ + + ··· + , and wj = , 1 ≤ j ≤ n.
2 3 n jH
Then wj > 0, for 1 ≤ j ≤ n, and w1 + w2 + · · · + wn = 1. Using the weighted
AM-GM inequality, we obtain
 n n  w
"
S j
= wj xjj ≥ xjj
H j=1 j=1
"
n 1/H
= xj .
j=1

Using the GM-HM inequality, we also get


"
n
n
xj ≥ = 1.
1 1 1 1
j=1 + + + ··· +
x1 x2 x3 xn
It follows that S ≥ H, and equality holds if and only if x1 = x2 = · · · = xn .
1 1 1
Thus the least value of S is H = 1 + + + · · · + .
2 3 n
Solutions 253

16. Let a, b, c, d be positive real numbers such that a + b + c + d ≤ 1. Prove


that
a b c d 1
+ + + ≤ .
b c d a 64abcd
Solution: We have
a b c d a2 cd + b2 da + c2 ab + d2 bc
+ + + =
b c d a abcd
(ab + cd)(ad + bc)
= .
abcd
Thus, it is sufficient to prove that
1
(ab + cd)(ad + bc) ≤ .
64
The AM-GM inequality gives
 2
a+b+c+d 1
(a + b)(c + d) ≤ ≤ ,
2 4
and
 2
ac + bd + ad + bc
(ac + bd)(ad + bc) ≤
2
 2
a + b)(c + d)
=
2
1 1
≤ = .
82 64

17. Let a, b, c be positive real numbers, all less than 1, such that a + b + c = 2.
Prove that
abc ≥ 8(1 − a)(1 − b)(1 − c).

Solution: Introducing new variables x, y, z by x = 1 − a, y = 1 − b and


z = 1 − c, we see that x, y, z are positive and x + y + z = 1. The required
inequality is
(1 − x)(1 − y)(1 − z) ≥ 8xyz.
Equivalently, we have to prove that
xy + yz + zx ≥ 9xyz.
This may be written in the form
1 1 1
+ + ≥ 9.
x y z
Since x + y + z = 1, this follows from the AM-HM inequality.
254 Solutions

18. Let a, b, c be three positive real numbers. Prove that

(2a + b + c)2 (2b + c + a)2 (2c + a + b)2


2 2
+ 2 2
+ 2 ≤ 8.
2a + (b + c) 2b + (c + a) 2c + (a + b)2

Solution: We introduce λ, μ and ν by 2λ = b + c, 2μ = c + a and 2ν = a + b.


Then
a = μ + ν − λ, b = ν + λ − μ, c = λ + μ − ν.
Thus, we obtain

2a + b + c = 2(μ + ν), 2b + c + a = 2(ν + λ), 2c + a + b = 2(λ + μ − ν).

The inequality to be proved is:


 4(μ + ν)2
≤ 8.
2(μ + ν − λ)2 + 4λ2
cyclic

However, we observe that

4(μ + ν)2 2(μ + ν)2 4(μ + ν)2


= ≤ .
2(μ + ν − λ)2 + 4λ2 (μ + ν − λ)2 + 2λ2 (μ + ν)2 + 2λ2
 2
Here we have used (μ + ν)2 ≤ μ + ν − λ + 2λ2 . Using (μ + ν)2 ≤ 2(μ2 + ν 2 ),
we get

4(μ + ν)2 4 4 4(μ2 + ν 2 )


=  *  ≤  *  = .
(μ + ν)2 + 2λ2 1 + 2λ2 (μ + ν)2 1 + λ2 (μ2 + ν 2 ) μ 2 + ν 2 + λ2

Thus,
 4(μ + ν)2  4(μ2 + ν 2 )
2 2
≤ = 8.
2(μ + ν − λ) + 4λ μ 2 + ν 2 + λ2
cyclic cyclic

19. Three positive real numbers a, b, c are such that (1 + a)(1 + b)(1 + c) = 8.
Prove that abc ≤ 1.
√ √ √
Solution: We know that 1 + a ≥ 2 a, 1 + b ≥ 2 b and 1 + c ≥ 2 c. Thus,
we get √
8 = (1 + a)(1 + b)(1 + c) ≥ 8 abc.
It follows that abc ≤ 1.

20. If a, b, c are the sides of a triangle, prove that

a2 b(a − b) + b2 c(b − c) + c2 a(c − a) ≥ 0.


Solutions 255

Solution: Taking s − a = x, s − b = y and s − c = z, we see that a = y + z,


b = z + x and c = x + y. The inequality is

(y + z)2 (z + x)(y − x) ≥ 0.
cyclic

Expanding the left hand side, this reduces to


xy 3 + yz 3 + zx3 ≥ x2 yz + xy 2 z + xyz 2 .
Using the Cauchy-Schwarz inequality, we have
   
x2 yz + xy 2 z + xyz 2 = x3/2 z 1/2 x1/2 yz 1/2
cyclic
 1/2  2 1/2
≤ x3 z + y 3 x + z 3 y xy z + xyz 2 + x2 yz .
This implies that
x2 yz + xy 2 z + xyz 2 ≤ xy 3 + yz 3 + zx3 .

This also follows from Muirhead’s inequality, since (2, 1, 1) ≺ (3, 0, 1).

21. Let a1 , a2 , . . . , an be n(≥ 2) real numbers whose sum is 1. Prove that



n
aj n
≥ .
j=1
2 − aj 2n − 1

Solution: Using the Cauchy-Schwarz inequality, we see that



n
aj (a1 + a2 + · · · + an )2 1
≥ n = n 2.
j=1
2 − aj a
j=1 j (2 − a j ) 2 − j=1 aj

Thus, it is sufficient to prove that


1 n
n ≥ .
2 − j=1 a2j 2n − 1
n
This is equivalent to n j=1 a2j ≥ 1, which follows from the Cauchy-Schwarz
inequality:
1 = (a1 + a2 + · · · + an )2 ≤ n(a21 + a22 + · · · + a2n ).

22. Let a1 , a2 , . . . , an be n positive real numbers whose sum is 1. Prove that



n
a2j 1
≥ .
j=1
aj + aj+1 2

(Here an+1 = a1 .)
256 Solutions

Solution: We observe that



n
a2j − a2j+1 
n
 
= aj − aj+1 = 0.
j=1
aj + aj+1 j=1

Thus, we get

n
a2j n
a2j+1
= .
j=1
aj + aj+1 a + aj+1
j=1 j

Hence, if S is the required sum, we have


 
n
2 a2j + a2j+1
4S =
j=1
aj + aj+1
 2
n
aj + aj+1

j=1
aj + aj+1

n
 
= aj + aj+1 = 2.
j=1

It follows that S ≥ 1/2.

23. Let a, b, c, d be four positive real numbers. Prove that


1 4 9 16 100
+ + + ≥ .
a b c d a+b+c+d

Solution: This follows from the Cauchy-Schwarz inequality:


√ √ √ √ 2
2 a b c d
(1 + 2 + 3 + 4) = √ + 2√ + 3√ + 4√
a b c d
  
1 4 9 16
≤ + + + a+b+c+d .
a b c d
Thus, it follows that
1 4 9 16 100
+ + + ≥ .
a b c d a+b+c+d

24. Let a1 , a2 , . . . , an be n(> 2) positive real numbers such that a1 + a2 + · · · +


an = 1 and aj < 1/2 for each j, 1 ≤ j ≤ n. Prove that

n
a2j 1
≥ .
j=1
1 − 2aj n−2
Solutions 257

Solution: Since aj < 1/2 for each j, we see that 1 − 2aj > 0, for 1 ≤ j ≤ n.
Using the Cauchy-Schwarz inequality, we have

 2  2
n n
a 
aj =  j 1 − 2aj
j=1 j=1
1 − 2aj
   
n
a2j n
≤ (1 − 2aj )
j=1
1 − 2aj j=1
 n 2 
aj
= (n − 2) .
j=1
1 − 2aj

Hence it follows that



n
a2j 1
≥ ,
j=1
1 − 2aj n−2

n
because of j=1 aj = 1.

25. Let x1 , x2 , . . . , xn , y1 , y2 , . . . , yn be 2n positive real numbers such that


x1 + x2 + · · · + xn ≥ x1 y1 + x2 y2 + · · · + xn yn , where n ≥ 2 is an integer. Prove
that
x1 x2 xn
x1 + x2 + · · · + xn ≤ + + ··· + .
y1 y2 yn

Solution: The Cauchy-Schwarz inequality gives


  
x1 x2 xn
(x1 + x2 + · · · + xn )2 ≤ + + ··· + x 1 y1 + x 2 y 2 + · · · + x n y n .
y1 y2 yn

The inequality x1 y1 + x2 y2 + · · · + xn yn ≤ x1 + x2 + · · · + xn gives the result.

26. If x1 , x2 , . . . , xn are n positive real numbers, prove that

x1 x2 xn √
+ + ··· + < n.
1 + x21 1 + x21 + x22 1 + x21 + x22 + · · · + x2n

Solution: Using the Cauchy-Schwarz inequality, it suffices to prove that

x21 x22 x2n


  +
2 2
 + ··· + 
2 2
2 < 1.
1 + x1 1 + x21 + x2 1 + x21 + x22 + · · · + x2n
258 Solutions

But for j ≥ 2, we have


x2j
 2
1 + x21 + x22 + · · · + x2j
x2j
≤   
1 + x21 + x22 + · · · + x2j−1 1 + x21 + x22 + · · · + x2j
1 1
= − .
1 + x21 + x22 + · · · + x2j−1 1 + x21 + x22 + · · · + x2j
We also have
x21 1
 2 ≤1−  2 .
1+ x21 1 + x21
Summing all these, we get
x21 x22 x2n
  +
2 2
 + ··· + 
2 2
2
1 + x1 1 + x21 + x2 1 + x21 + x22 + · · · + x2n
1
≤ 1− < 1.
1 + x21 + x22 + · · · + x2n

27. If a, b, c are positive real numbers, prove that


  
3 a2 b + b2 c + c2 a ab2 + bc2 + ca2 ≥ abc(a + b + c)3 .

Solution: Using the Cauchy-Schwarz inequality, we have


 2  
a b + b2 c + c2 a ab2 + bc2 + ca2
 √ √ √ 2
≥ a2 bc + b2 ca + c2 ab
 2
= abc a3/2 + b3/2 + c3/2 .
Thus, it is sufficient to prove that
 2  3
3 a3/2 + b3/2 + c3/2 ≥ a + b + c .
This follows from Hölder’s inequality: taking p = 3/2 and q = 3, we obtain
 2/3  1/3
(a + b + c) ≤ a3/2 + b3/2 + c3/2 1+1+1 ,
which gives the desired inequality.

28. Let P (x) = ax2 + bx + c be a quadratic polynomial with non-negative


coefficients and let α be a positive real number. Prove that
P (α)P (1/α) ≥ P (1)2 .
Solutions 259

Solution: We have
  
2 a b
P (α)P (1/α) = aα + bα + c + +c
α2 α
= a 2 + b2 + c 2
   
1 21
+(ab + bc) α + + ac α + 2
α α
≥ a2 + b2 + c2 + 2ab + 2bc + 2ac
= (a + b + c)2 = P (1)2 ;
we have used the AM-GM inequality.

29. If a, b, c, d, e are positive reals, prove the inequality


 a 5
≥ ,
b+c 2
where the sum is taken cyclically over a, b, c, d, e.

Solution: The inequality is equivalent to


a2 b2 c2 d2 e2 5
+ + + + ≥ .
ab + ac bc + bd cd + ce de + da ea + eb 2
Using the Cauchy-Schwarz inequality, we see that
(a + b + c + d + e)2
 
a2 b2 c2 d2 e2
≤ + + + +
ab + ac bc + bd cd + ce de + da ea + eb
 
× ab + ac + bc + bd + cd + ce + de + da + ea + eb .

Thus, we obtain
a2 b2 c2 d2 e2 (a + b + c + d + e)2 2
+ + + + ≥  ,
ab + ac bc + bd cd + ce de + da ea + eb sym ab

where 
ab = ab + ac + ad + ae + bc + bd + be + cd + ce + de.
sym

Hence, it is sufficient to prove that


5
(a + b + c + d + e)2 ≥ ab.
2 sym

This is equivalent to
2(a2 + b2 + c2 + d2 + e2 ) ≥ ab + ac + ad + ae + bc + bd + be + cd + ce + de.
260 Solutions

After multiplying by 2 on both sides, we may write this in the form

(a − b)2 + (a − c)2 + (a − d)2 + (a − e)2 + (b − c)2 + (b − d)2


+ (b − e)2 + (c − d)2 + (c − e)2 + (d − e)2 ≥ 0.

Hence the result follows.

30. If a, b, c are the sides of an acute-angled triangle, prove that


  
a2 + b2 − c2 a2 − b2 + c2 ≤ ab + bc + ca.
cyclic

Solution: Let us introduce a2 +b2 −c2 = z, b2 +c2 −a2 = x and c2 +a2 −b2 = y.
Then we have
  
a 2 + b2 − c 2 a 2 − b2 + c 2
cyclic
√ √ √
= xy + yz + zx
1 0√ √ √ √ √ √ 1
= xy + yz + yz + zx + zx + xy
2
1 0√ √ √ √ √ √ 1
= xy + xz + yz + yx + zx + zy
2
1 0   1
≤ (x + y)(x + z) + (y + z)(y + x) + (z + x)(z + y)
2
1 0√ 2 √ √ 1
= 2c · 2b2 + 2a2 · 2c2 + 2b2 · 2a2
2
= ab + bc + ca.

31. Let a, b, c be non-negative real numbers such that


1 1 1
+ + = 2.
a 2 + 1 b2 + 1 c 2 + 1
Prove that ab + bc + ca ≤ 3/2.

Solution: We have
 
a2 b2 c2 1 1 1
+ + =3− + + = 1.
a 2 + 1 b2 + 1 c 2 + 1 a 2 + 1 b2 + 1 c 2 + 1

Using the Cauchy-Schwarz inequality, we observe that


    2
a2 b2 c2 2 2 2
+ + a +1+b +1+c +1 ≥ a+b+c .
a 2 + 1 b2 + 1 c 2 + 1
Solutions 261

Thus, we get
a2 + b2 + c2 + 3 ≥ (a + b + c)2 .

This simplifies to
3
ab + bc + ca ≤ .
2

32. Suppose a, b, c are positive real numbers. Prove that

   1/3  a3 + b3 + c3 1/3
3 a + b + c ≥ 8 abc + .
3

When does equality hold?

Solution: Using the concavity of the function f (x) = x1/3 on the interval
(0, ∞), we obtain

 1/3  1/3
 1/3 a 3 + b3 + c 3 8abc + a3 +b3 +c3
3
8 abc + ≤ 9
3 9
 1/3
= 3 24abc + a3 + b3 + c3

Thus, it is sufficient to prove that

24abc + a3 + b3 + c3 ≤ (a + b + c)3 .

This reduces to

a2 b + b2 c + c2 a + ab2 + bc2 + ca2 ≥ 6abc,

which is a consequence of the AM-GM inequality. Equality holds if and only


if a = b = c.

33. Let c1 ,c2 ,c3 ,. . .,cn be n real numbers such that either 0 ≤ cj ≤ 1 for all j
or cj ≥ 1 for all j, 1 ≤ j ≤ n. Prove that the inequality

"
n
  "
n
1 − p + pcj ≤ 1 − p + p cj
j=1 j=1

holds, for any real p with 0 ≤ p ≤ 1.


262 Solutions

Solution: We use induction on n. The statement is immediate for n = 1.


Suppose it is true for all j ≤ n − 1. Since 0 ≤ p ≤ 1 and cj ≥ 0 imply
1 − p + pcj ≥ 0, the induction hypothesis gives

"
n
  "
n−1
   
1 − p + pcj = 1 − p + pcj · 1 − p + pcn
j=1 j=1
 "
n−1   
≤ 1−p+p cj · 1 − p + pcn .
j=1

The induction step is complete once the inequality


 "
n−1    "
n
1−p+p cj · 1 − p + pcn ≤ 1 − p + p cj
j=1 j=1

is proved. This is equivalent to


 0 "
n−1 "
n−1 1
p − p2 cn + cj − 1 − cn cj ≤ 0.
j=1 j=1

Since p − p2 ≥ 0, this is equivalent to


 n−1
"  
cj − 1 cn − 1 ≥ 0.
j=1

The given conditions on cj ’s now imply the result.

34. Let x1 , x2 , x3 , x4 be real numbers in the interval (0, 1/2]. Prove that
x1 x2 x3 x4
(1 − x1 )(1 − x2 )(1 − x3 )(1 − x4 )
x41 + x42 + x43 + x44
≤ .
(1 − x1 )4 + (1 − x2 )4 + (1 − x3 )4 + (1 − x4 )4

Solution: The inequality may be written in the form


 4  4
1 − xj x
  ≤  j.
1 − xj xj

By symmetry, we may assume that x1 ≥ x2 ≥ x3 ≥ x4 . If we set 1 − xj = yj ,


then y1 ≤ y2 ≤ y3 ≤ y4 . An easy computation shows that
 4  2 2  2  2
xj x1 − x22 + x23 − x24 + 2 x1 x2 − x3 x4
 −4= .
xj x1 x2 x3 x4
Solutions 263

Thus, it is sufficient to prove that


  2 2  2
x21 − x22
+ x23 − x24 + 2 x1 x2 − x3 x4
x1 x2 x3 x4
 2 2  2  2
y1 − y22 + y32 − y42 + 2 y1 y2 − y3 y4
≥ .
y1 y 2 y 3 y 4
 
Since x1 + x2 ≤ 1, we have x1 − x2 ≥ x21 − x22 which may be written in the
form x1 y1 ≥ x2 y2 . Suppose p and q are real numbers such that p ≥ q ≥ 1.
Then it is easy to see that
1 1
p+ ≥q+
p q
holds. Taking p = x1 /x2 and q = y2 /y1 , we obtain
x1 x2 y2 y1
+ ≥ + .
x2 x1 y1 y2
This may be written as
x21 + x22 y 2 + y22
≥ 1 ,
x1 x2 y1 y2
or equivalently in the form
 2  2
x1 + x2 y1 + y 2
≥ .
x1 x2 y1 y2
 2  2   
But observe that x1 − x2 = y1 − y2 and x3 x4 ≤ 1 − x3 1 − x4 = y3 y4 .
Thus we obtain
 2 2  2  2
x1 − x22 x1 + x2 x1 − x2
= ·
x1 x2 x3 x4 x1 x2 x3 x4
 2  2
y1 + y2 y 1 − y2
≥ ·
y1 y2 y3 y4
 2 
2 2
y1 − y2
= .
y1 y 2 y 3 y 4
Similarly,
 2  2 2
x23 − x24 y3 − y42

x1 x2 x3 x4 y1 y2 y 3 y 4
* *
may be obtained. Taking p = x1 x2 x3 x4 and q = y3 y4 y1 y2 , it can be easily
verified that p ≥ q ≥ 1 . We hence also obtain
 2  2
2 x1 x2 − x3 x4 2 y1 y2 − y3 y4
≥ .
x1 x2 x3 x4 y1 y2 y3 y4
Combining all these, we get the result. Equality holds if and only if x1 = x2 =
x3 = x4 .
264 Solutions

35. Let x1 ,x2 ,x3 ,. . .,xn be n real numbers such that 0 < xj ≤ 1/2. Prove that
⎛ ⎞ ⎛ ⎞
" /  n " /  
n n n
  n
  n
⎝ xj xj ⎠ ≤ ⎝ 1 − xj 1 − xj ⎠.
j=1 j=1 j=1 j=1

Solution: (By KshipraBawalkar)  Since 0< xj ≤ 1/2,


 we have 1/2 ≤ 1−xj <
1 for, 1 ≤ j ≤ n. Thus (1 − xj ), xj , xj and (1 − xj ) are all positive.
The inequality may be written in the form
 n
n
n
xj j=1 xj
n j=1 ≤  .
j=1 (1 − xj )
n   n
j=1 1 − xj

 
Let A = x1 + x2 + · · · + xn /n. Let xl and xs denote respectively the largest
and the smallest among x1 ,x2 ,x3 ,. . .,xn . We show that
n
xj
n j=1
j=1 (1 − xj )
 
x1 x2 · · · A · · · xl + xs − A · · · xn
≤         ,
1 − x1 1 − x2 · · · 1 − A · · · 1 − xl − xs + A · · · 1 − xn

where the right side is obtained by replacing xl and xs respectively by A and


xl + xs − A. We need to prove that
 
xl xs A xl + xs − A
  ≤  .
1 − xl 1 − xs 1 − A 1 − xl − xs + A

We first observe that 0 < xl + xs ≤ 1. The above inequality simplifies to


    
A2 − A xl + xs + xl xs 1 − xl − xs ≤ 0.

Since 1 − xl − xs ≥ 0, this is equivalent to


 
A2 − A xl + xs + xl xs ≤ 0.

But     
A2 − A xl + xs + xl xs = A − xl A − xs ≤ 0,
since xl ≤ A ≤ xs . Now consider the set
0   1
x1 , x2 . . . , A, . . . , xl + xs − A , . . . , xn
 
obtained by replacing xl , xs respectively by A, xl +xs −A . This again satisfies
the hypothesis of the problem, as 0 < A ≤ 1/2 and 0 < xl + xs − A ≤ 1/2
Solutions 265

(since xs ≥ A and xl ≤ A). Moreover the average of the set does not change
by this process and it remains A. We may continue this process and after at
most n − 1 steps all the numbers in the set are equal to A, the average. Thus
we obtain the inequality
n
xj An
n j=1 ≤ n .
j=1 (1 − xj ) 1−A
However
An (nA)n
 n =  n
1−A n − nA
 n
n
j=1 x j
=  n
n
n − j=1 xj
 n
n
j=1 xj
=   .
n   n
j=1 1 − xj

It follows that  n
n
n xj
xj j=1
n j=1
≤ n .
j=1 (1 − xj ) n  
j=1 1 − xj

Alternate Solution I:
We give here another solution based on induction on n. We use the classical
technique of proving the result whenever n is a power of 2 and then fill up the
remaining gaps by coming back. The case n = 2 is simple. The inequality
required is   
x1 x2 1 − x1 1 − x2
 2 ≤  2 .
x1 + x2 2 − x1 − x2
Simplification gives the equivalent inequality:
 2  
x1 − x2 1 − x1 − x2 ≥ 0.
Since x1 + x2 ≤ 1, the result follows for n = 2. We also observe that equality
holds if and only if x1 = x2 . We use the case n = 2 to prove the result
 for n = 4.
Consider positive numbers x1 , x2 , x3 , x4 ∈ (0, 1/2]. Taking y1 = x1 + x2 /2
and y2 = x3 + x4 /2, the result for n = 2 gives
  
y1 y2 1 − y1 1 − y2
 2 ≤  2 .
y1 + y 2 2 − y 1 − y2
266 Solutions

Substituting for y1 and y2 , this reduces to


 2  2  2  2
x1 x2 x3 x4 x1 + x2 x3 + x4 x1 x2 x3 x4 2 − x1 − x2 2 − x3 − x4
 4 ≤   .
4 4   4
j=1 xj j=1 1 − xj

However, using the result for n = 2, we also have


 2    2
x1 x2 2 − x1 − x2 ≤ 1 − x1 1 − x2 x1 + x2
 2    2
x3 x4 2 − x3 − x4 ≤ 1 − x3 1 − x4 x3 + x4 .
 2 
Using these on the right side and effecting some cancellations x1 + x2 x3 +
2
x4 , we get
4 4  
j=1 xj j=1 1 − xj
 4 ≤   .
4 4   4
j=1 xj j=1 1 − xj

This proves the result for n = 4. Using induction on k, this proves the inequality
for all n of the form n = 2k .
Take any n and fix k such that 2k ≤ n < 2k+1 . Let A be the average of
these n numbers; A = x1 + x2 + · · · + xn /n. We consider 2k+1 numbers

x1 , x2 , . . . , xn , A, A, . . . , A,

where A appears 2k+1 −n times in the above sequence. We apply the inequality
for these 2k+1 numbers. (Note that 0 < A ≤ 1/2.) Thus we obtain

k+1
x 1 x 2 · · · x n A2 −n
 2k+1
n  
j=1 xj + 2k+1 −n A
     2k+1 −n
1 − x1 1 − x2 · · · 1 − xn 1 − A
≤  k+1 .
n      2
j=1 1 − xj + 2 −n 1−A
k+1

This simplifies to
    
x1 x2 · · · xn 1 − x1 1 − x2 · · · 1 − xn
≤   n ,
An 1−A

which is the required inequality for n.


Alternate Solution II:
Solutions 267

Here is another solution using Jensen’s inequality for the convex function
 
1
f (x) = log −1 ,
x

defined on the interval (0, 1/2). Writing f (x) = log(1 − x) − log x, it is easy to
check that
1 1
f  (x) = − − ,
1−x x
1 1
f  (x) = − + 2
(1 − x)2 x
1 − 2x
=   > 0,
x 1 − x2
2

for x ∈ (0, 1/2). Hence f is convex on (0, 1/2). This implies that
   
1
n
1 1
log n −1 ≤ log −1 .
j=1 (xj /n) n j=1 xj

The exponentiation gives


 n " n  
n 1
n −1 ≤ −1 .
j=1 xj j=1
xj

Some simplification leads to the required inequality.

36. Consider a sequence an of real numbers satisfying aj+k ≤ aj + ak . Prove


that
a2 a3 an
a1 + + + ··· + ≥ an ,
2 3 n
for all n.

Solution: We have for any n,

an ≤ aj + an−j , 1 ≤ j ≤ (n − 1).

Summing over j, we obtain


n−1
(n − 1)an ≤ 2 aj .
j=1

We prove the given inequality by induction on n. It is obviously true for n = 1.


Suppose it is true for all k < n. Thus, we have
a2 a3 ak
bk = a 1 + + + ··· + ≥ ak ,
2 3 k
268 Solutions

for all k < n. Summing these over k, we get



n−1 
n−1
bk ≥ ak .
k=1 k=1

But, we have


n−1  a2   a2 a3 
bk = a 1 + a 1 + + a1 + +
2 2 3
k=1
 
a2 an−1
+ · · · + a1 + + ··· + .
2 n−1
This simplifies to

n−1
n−2 n − (n − 1)
bk = (n − 1)a1 + a2 + · · · + an−1 .
2 n−1
k=1

We thus have
n−2 n − (n − 1)
(n − 1)a1 + a2 + · · · + an−1 ≥ a1 + a2 + · · · + an−1 .
2 n−1
Adding a1 + a2 + · · · + an to both sides and simplifying, we get
   
a2 an−1
n a1 + + ··· + + an ≥ 2 a1 + a2 + · · · + an−1 + an .
2 n−1
 
But we note that 2 a1 +a2 +· · ·+an−1 +an ≥ (n−1)an +an = nan . Dividing
by n, we get
a2 an−1 an
a1 + + ··· + + ≥ an .
2 n−1 n
This completes the induction step and hence the inequality is valid for all n.

37. For positive real numbers x, y, z, prove the inequality


 x
 ≤ 1,
x + (x + y)(x + z)
where the sum is taken cyclically over x, y, z.

Solution: We observe that


 √ √
(x + y)(x + z) ≥ xy + xz.

In fact, squaring on both the sides, this reduces to



x2 + yz ≥ 2x yz.
Solutions 269

which is immediate from the AM-GM inequality. Thus


 x  x
 ≤ √ √
x + (x + y)(x + z) x + xy + xz
cyclic cyclic
 √
x
= √ √ √ = 1.
x+ y+ z
cyclic

38. Let x, y be non-negative real numbers such that x + y = 2. Prove the


inequality  
x3 y 3 x3 + y 3 ≤ 2.

Solution: As a consequence of the AM-GM inequality, we have


 2
x+y
xy ≤ = 1.
2
Thus, we obtain 0 < xy ≤ 1. We write
   3   
x3 y 3 x 3 + y 3 = xy x + y x2 − xy + y 2
  3  2 
= 2 xy x + y − 3xy
 3  
= 2 xy 4 − 3xy .
Thus we need to prove that
 3  
xy 4 − 3xy ≤ 1.
Putting z = xy, this inequality reduces to
 
z 3 4 − 3z ≤ 1,
for 0 < z ≤ 1. We can prove this in different ways. We can put the inequality
in the form
3z 4 − 4z 3 + 1 ≥ 0.
2
 2 
Herethe expression
 on the left hand side factors in to (z − 1) 3z + 2z + 1
and 3z 2 + 2z + 1 is positive since its discriminant D = −8 < 0. Or applying
the AM-GM inequality to the positive reals 4 − 3z, z, z, z, we obtain
 4
3
  4 − 3z + 3z
z 4 − 3z ≤ ≤ 1.
4

39. A convex quadrilateral ABCD is inscribed in a unit circle. Suppose its


sides satisfy the inequality AB ·BC ·CD ·DA ≥ 4. Prove that the quadrilateral
is a square.
270 Solutions

Solution: Let AB = a, BC = b, CD = c, DA = d. We are given that


abcd ≥ 4. Using Ptolemy’s theorem and the fact that each diagonal cannot
exceed the diameter of the circle, we get ac + bd = AC · BD ≤ 4. But an
application of the AM-GM inequality gives
√ √
ac + bd ≥ 2 abcd ≥ 2 4 = 4.

We conclude that ac + bd = 4. This forces AC · BD = 4, giving AC = BD = 2.


Each of AC and BD is thus a diameter. This implies that ABCD is a rectangle.
Note that
(ac − bd)2 = (ac + bd)2 − 4abcd ≤ 16 − 16 = 0
√ √
√ ac = bd = 2. Thus we get a = c = ac = 2, and similarly
and hence
b = d = 2. It now follows that ABCD is a square.
   
40. Let a1 , a2 , . . . , an and b1 , b2 , . . . , bn be two sets of reals such that
0 < h ≤ aj ≤ H and 0 < m ≤ bj ≤ M for some reals h, H, m, M . Prove that
    2
( aj 2 )( bj 2 ) 1 HM hm
1≤  ≤ + .
( a j bj ) 2 4 hm HM

Solution: The first inequality


 is simply a restatement
 of the Cauchy-Schwarz
inequality. Define c1 = min a1 , a2 , . . . , an and having defined c1 , c2 , . . . ,
cj−1 , define
0   1
cj = min a1 , a2 , . . . , an \ c1 , c2 , . . . , cj−1 ,
 
for 2 ≤ j ≤ n. Then c1 ≤ c2 ≤· · · ≤ cn and c1 , c2 , . . . , cn is simply the
rearrangement of a1 , a2 , . . . ,an in increasing order.
 Similarly, we get the
rearrangement d1 , d2 , . . . , dn of the elements in b1 , b2 , . . . , bn in decreasing
order: d1 ≥ d2 ≥ · · · ≥ dn . Using the rearrangement inequality, we have

n 
n
c j dj ≤ a j bj .
j=1 j=1

We may assume that not all the aj ’s are equal and not all the bj ’s are equal so
that c1 < cn and dn < d1 . Hence we obtain
  
cn d1 − c1 dn = cn − c1 d1 + c1 d1 − dn ) > 0.

If k ≥ 2, define uk and vk by

c2k = uk c21 + vk c2n , d2k = uk d21 + vk d2n .

Then uk ≥ 0, vk ≥ 0 and
 1/2  1/2
ck dk = uk c21 + vk c2n uk d21 + vk d2n ≥ uk c 1 d1 + v k c n dn ,
Solutions 271

using the Cauchy-Schwarz inequality. We observe that uk = 0 for some k


implies that ck = ck+1 = · · · = cn , dk = dk+1 = · · · = dn and vk = 1. Similar
is the case when vk = 0. If uk > 0 and vk > 0, then ck dk > uk c1 d1 + vk cn dn .
Hence   2   2   2  
cj dj P c1 + Qc2n P d21 + Qd2n
1≤  2 ≤  2
c j dj P c1 d1 + Qcn dn
where P = 1 + u2 + · · · + un and Q = v1 + v2 + · · · + vn−1 + 1. The expression
on the right hand side is equal to
 2
c n d1 − c 1 dn
1 + PQ .
P c1 d1 + Qcn dn

Using P c1 d1 + Qcn dn ≥ 2 P Qc1 d1 cn dn , we obtain
⎛  ⎞2
 2 cn d1
+ c1 d n
c n d1 − c 1 dn ⎝ c1 dn cn d 1

1 + PQ ≤
P c1 d1 + Qcn dn 2
⎛  ⎞2
HM hm
+
≤ ⎝ hm HM
⎠ .
2

   
Since a2j = c2j , b2j = d2j and


n 
n
c j dj ≤ a j bj ,
j=1 j=1

we get the desired inequality.

41. Let f : [0, a] →


Rn be a convex function. Consider n points x1 ,x2 ,x3 ,. . .,xn
in [0, a] such that j=1 xj is also in [0, a]. Prove that

  
n
  n
   
f xj ≤ f xj + n − 1 f 0 .
j=1 j=1

Solution: Using the convexity of f , we have


   
x 1 x1 + x2 + x 2 · 0 x1   x2  
f ≤ f x1 + x2 + f 0 .
x1 + x2 x1 + x2 x1 + x2

Thus it follows that


  x1   x2  
f x1 ≤ f x1 + x2 + f 0 .
x1 + x2 x1 + x2
272 Solutions

Interchanging x1 and x2 , we also obtain


  x2   x1  
f x2 ≤ f x1 + x2 + f 0 .
x1 + x2 x1 + x2
Adding these two, we get
       
f x1 + f x2 ≤ f x1 + x2 + f 0 .

Now we use the induction on n. Suppose the result holds for (n − 1); say
           
f x1 + f x2 + · · · + f xn−1 ≤ f x1 + x2 + · · · + xn−1 + n − 2 f 0 ,

for all points x1 , x2 , . . . , xn−1 in [0, a] such that their sum


nis also in [0, a]. Now
consider any n points x1 ,x2 ,x3 ,. . .,xn in [0, a] such that j=1 xj is also in [0, a].
Then
       
f x1 + f x2 + · · · + f xn−1 + f xn
       
≤ f x1 + x2 + · · · + xn−1 + n − 2 f 0 + f xn
       
≤ f x1 + x2 + · · · + xn−1 + xn + f 0 + n − 2 f 0
     
= f x1 + x2 + · · · xn−1 + xn + n − 1 f 0 .

This completes induction and hence the proof as well.

42. For any natural number n, prove


 
2n √
3n < 4n .
n

Solution: We show by induction that


 
2n √
3n + 1 ≤ 4n .
n
For n = 1, the result is obvious. Suppose it holds for all k ≤ n. Then we have
   
2n + 2  2(2n + 1) 2n 3n + 4 √
3(n + 1) + 1 = · 3n + 1
n+1 n+1 n 3n + 1

2(2n + 1) 3n + 4 n
≤ ·4 ,
n+1 3n + 1
by induction hypothesis. However, we can show that

2n + 1 3n + 4
≤ 2.
n + 1 3n + 1
In fact, this statement is equivalent to

(2n + 1)2 (3n + 4) ≤ 4(n + 1)2 (3n + 1).


Solutions 273

This further reduces to


12n3 + 28n2 + 19n + 4 ≤ 12n3 + 28n2 + 20n + 4,
which is true for all natural numbers n. Hence we obtain
 
2n + 2 
3(n + 1) + 1 ≤ 4n+1 ,
n+1
√ √
which completes induction. Since 3n < 3n + 1, we get the required inequal-
ity.

43. Let a, b, c be positive real numbers and let x be a non-negative real number.
Prove that
ax+2 + bx+2 + cx+2 ≥ ax bc + abx c + abcx .

Solution: We observe that


ax b2 + a2 bx ≤ ax+2 + bx+2 .
  
This follows easily, by writing the inequality in the form ax − bx a2 − b2 ≥ 0.
Thus we have,
 
2 ax+2 + bx+2 + cx+2
     
= ax+2 + bx+2 + bx+2 + cx+2 + cx+2 + ax+2
     
≥ a x b 2 + b x a 2 + b x c 2 + c x b2 + c x a 2 + a x c 2
     
= a x b 2 + c 2 + b x c 2 + a 2 + c x a 2 + b2
 
≥ 2 ax bc + bx ca + cx ab .
Hence the result follows.

44. Let (a1 , a2 , . . . , an ), (b1 , b2 , . . . , bn ), and (c1 , c2 , . . . , cn ) be three sequences


of positive real numbers. Prove that
 n  n 1/3   n 1/3   n 1/3
a j bj c j ≤ a3j b3j c3j .
j=1 j=1 j=1 j=1

Solution: Using Hölder’s inequality with exponents p = 3 and q = 3/2, we


get
  1/3   2/3
n n n
 3/2
a j bj c j ≤ a3j bj c j .
j=1 j=1 j=1
Using the Cauchy-Schwarz inequality, we also get
  1/2   1/2
n
 3/2 n
3
n
3
bj c j ≤ bj cj .
j=1 j=1 j=1

Combining these two, we get the desired inequality.


274 Solutions

45. Prove for any three real numbers a, b, c, the inequality


     2
3 a2 − a − 1 b2 − b − 1 c2 − c + 1 ≥ abc − abc + 1.

 3  
Solution: Consider the function f (a) = 3 a2 −a+1 − a6 +a3 +1 for a ∈ R.
Some computation shows that f (1) = 0, f  (1) = 0, f  (1) = 0, f  (1) = 0, but
f (iv) (1) = 0. Thus (a − 1)4 divides f (a) but not any higher power of (a − 1).
Expanding f (a), we obtain
 
f (a) = (a − 1)4 2a2 − a + 2 .

But the discriminant of 2a2 − a + 2 is −15 and hence 2a2 − a + 2 ≥ 0 for all
real a. It follows that
 3
3 a2 − a + 1 ≥ a6 + a3 + 1,

for all real a. Using the previous problem, we obtain


 3  3
(xyz)2 + (xyz) + 1 = x2 · y 2 · z 2 + x · y · z + 1 · 1 · 1
   
≤ x6 + x3 + 1 y 6 + y 3 + 1 z 6 + z 3 + 1 .

Hence we get
 3  3  3  3
(xyz)2 + (xyz) + 1 ≤ 27 x2 − x + 1 y 2 − y + 1 z 2 − z + 1 .

But for arbitrary reals a, b, c, we have

(abc)2 − (abc) + 1 ≤ |abc|2 + |abc| + 1.

Thus it follows that


 3    
(abc)2 − (abc) + 1 ≤ 27 a2 − a + 1 b2 − b + 1 c2 − c + 1 .

46. In a triangle ABC, show that

1 1 8
+ ≥ .
sin A sin B 3 + 2 cos C

Find the conditions for equality.


Solutions 275

Solution: The function f (x) = 1/ sin x is convex on [0, π]. Jensen’s theorem
shows that
1 1 2 2
+ ≥  = .
sin A sin B sin A+B 2
cos(C/2)
Thus, it suffices to prove that
1 4
≥ .
cos(C/2) 3 + 2 cos C
Using cos C = 2 cos2 (C/2) − 1, this reduces to
 2
2 cos(C/2) − 1 ≥ 0.

Equality holds here only if cos(C/2) = 1/2 which is equivalent to C = 120◦ .


Equality holds in Jensen’s inequality if and only if A = B. Thus equality holds
in the inequality only for that triangle with C = 120◦ and A = B = 30◦ .

47. Consider a real polynomial of the form

P (x) = xn + an−1 xn−1 + · · · + a1 x + 1,

where aj ≥ 0 for 1 ≤ j ≤ (n − 1). Suppose P (x) = 0 has n real roots. Prove


that P (2) ≥ 3n .

Solution: Since the coefficients are all non-negative, we see that P (t) ≥ 1,
for any t ≥ 0. Thus P (x) = 0 has only negative roots; let these be −α1 , −α2 ,
. . . , −αn , where αj ’s are all positive. We have

P (x) = (x + α1 )(x + α2 ) · · · (x + αn ).

We also observe that


α1 α2 · · · αn = 1.
Using the AM-GM inequality, we obtain
 1/3
2 + αj = 1 + 1 + αj ≥ 3 αj , for 1 ≤ j ≤ n.

Hence

P (2) = (2 + α1 )(2 + α2 ) · · · (2 + αn )
 1/3
≥ 3n α 1 α 2 · · · α n
= 3n .

In fact, this may be generalised as follows: for any t ≥ 0, the inequality p(t) ≥
(1 + t)n . We use the weighted AM-GM inequality. As in the above solution,
we have
P (t) = (t + α1 )(t + α2 ) · · · (t + αn ).
276 Solutions

Consider a general term (t + αk ). We have


t · 1 + 1 · αk
t + αk = (t + 1)
t+1
≥ (t + 1)(αk )1/(t+1) ,
by the weighted AM-GM inequality. Thus
P (t) ≥ (t + 1)n (α1 α2 · · · αn )1/(t+1) = (t + 1)n .

48. Let a1 <a2 <a3 <. . .<an be n positive integers. Prove that
 2
a1 +a2 +a3 +· · ·+an ≤ a31 + a32 + a33 + · · · + a3n .

Solution: We prove this by induction on n. The inequality is clear for n = 1:


a1 > 0 implies that a1 ≥ 1 and hence a21 ≤ a31 . Suppose it holds for all distinct
positive integers 0 < a1 < a2 < · · · < an and an+1 > an . Then we see that
an+1 ≥ 1 + an , an+1 ≥ an−1 + 2, . . . , an+1 ≥ a1 + n.
Adding all these inequalities, we get
  n(n + 1)
nan+1 ≥ a1 + a2 + · · · + an + .
2
Thus
  n(n + 1)
a1 + a2 + · · · + an ≤ nan+1 − .
2
Hence
 n+1
 2 
n 2
aj = aj + an+1
j=1 j=1

n 2 
n 
= aj + 2an+1 aj + a2n+1
j=1 j=1

n  
n(n + 1)
≤ a3j + 2an+1 nan+1 − + a2n+1
j=1
2

n
 
= a3j + 2n + 1 a2n+1 − n(n + 1)an+1 .
j=1

Thus it is sufficient to prove that


 
2n + 1 a2n+1 − n(n + 1)an+1 ≤ a3n+1
 
⇐⇒ 2n + 1 an+1 − n(n + 1) ≤ a2n+1
 
⇐⇒ a2n+1 − 2n + 1 an+1 + n(n + 1) ≥ 0
  
⇐⇒ an+1 − n an+1 − n − 1 ≥ 0.
Solutions 277

Since aj are integers and 0 < a1 < a2 < · · · < an < an+1 , it follows that
aj ≥ j, for 1 ≤ j ≤ n + 1. Hence an+1 ≥ n + 1, and thus
  
an+1 − n an+1 − n − 1 ≥ 0.

49. Consider a sequence a1 ,a2 ,a3 ,. . .,an of positive real numbers which add
up to 1, where n ≥ 2is an integer. Prove that for any positive real numbers
n
x1 ,x2 ,x3 ,. . .,xn with j xj = 1, the inequality

 n − 2  aj x2j
n
2 xj xk ≤ + ,
n − 1 j=1 1 − aj
j<k

holds.

Solution: We have

n 2 
n 
1= xj = x2j + 2 xj xk .
j=1 j=1 j<k

Thus we have to prove that



n
n − 2  aj x2j
n
1− x2j ≤ + .
j=1
n − 1 j=1 1 − aj

This is equivalent to
1 
n
x2j
≤ .
n − 1 j=1 1 − aj

This follows from the Cauchy-Schwarz inequality:


 2    
n n
x2j n
1= xj ≤ (1 − aj )
j=1 j=1
1 − aj j=1
 
n
x2j
= (n − 1) .
j=1
1 − aj

50. Let x1 , x2 , x3 , x4 be four positive real numbers such that x1 x2 x3 x4 = 1.


Prove that

3 3 3 3 1 1 1 1
x1 + x2 + x3 + x4 ≥ min x1 + x2 + x3 + x4 , + + + .
x1 x2 x3 x4
278 Solutions

Solution: Let us put x31 + x32 + x33 + x34 = A and aj = A − x3j . Then it is easy
to check that
1 
A= a1 + a2 + a3 + a4 .
3
Moreover,
1 x3 + x33 + x34 1
a1 = 2 ≥ x2 x3 x4 = .
3 3 x1
Similarly, we get
1 1 1 1 1 1
a2 ≥ , a3 ≥ , a4 ≥ .
3 x2 3 x3 3 x4
Thus it follows that
1  1 1 1 1
A= a1 + a2 + a3 + a4 ≥ + + + .
3 x1 x2 x3 x4
Using Chebyshev’s inequality, we also have

x31 + x32 + x33 + x34 x2 + x22 + x23 + x24 x1 + x2 + x3 + x4


≥ 1 · .
4 4 4
But
x21 + x22 + x23 + x24
≥ 1,
4
by the AM-GM inequality. Thus we also obtain

A ≥ x 1 + x2 + x3 + x4 .

It now follows that



1 1 1 1
A ≥ max x1 + x2 + x3 + x4 , + + + .
x1 x2 x3 x4

51. Let {x} denote the fractional part of x; i.e., {x} = x − [x]. Prove for any
positive integer n,
n
   n2 − 1
j ≤ .
j=1
2

Solution: We use induction on n. For n = 1, the result is clear. Suppose


it holds
 for some n. We prove its validity for n + 1 also. We observe that
n < n2 + j < n + 1 for 1 ≤ j ≤ 2n. Thus
    j
n2 + j = n2 + j − n < n2 + j + (j/2n)2 − n = .
2n
Solutions 279

Hence
(n+1)2 2 (n+1) 2
   
n
    
j = j + j
j=1 j=1 j=n2 +1
2n
n2 − 1 1 
≤ + j+0
2 2n j=1
n2 − 1 1 2n(2n + 1)
= + ·
2 2n 2
(n + 1)2 − 1
= .
2
This completes the inductive step and hence the proof as well.

52. If a, b, c are positive real numbers, prove that


a2 b2 c2 3
+ + ≥ .
(a + b)(a + c) (b + c)(b + a) (c + a)(c + b) 4

Solution: After expanding, the inequality takes the form


       
4 a2 b + b2 c + c2 a + 4 ab2 + bc2 + ca2 ≥ 3 a + b b + c c + a .
This reduces to
   
a2 b + b2 c + c2 a + ab2 + bc2 + ca2 ≥ 6abc,
which is a consequence of the AM-GM inequality.

53. Suppose a, b, c are the sides of a triangle. Prove that


a2 (b + c − a) + b2 (c + a − b) + c2 (a + b − c) ≤ 3abc.

Solution: Introducing x, y, z by
b + c − a = 2x, c + a − b = 2y, a + b − c = 2z,
we get a = y + z, b = z + x, c = x + y and the inequality takes the form
2x(y + z)2 + 2y(z + x)2 + 2z(x + y)2 ≤ 3(x + y)(y + z)(z + x).
Note that x, y, z are positive. The inequality may be reduced to
6xyz ≤ x2 y + y 2 z + z 2 x + xy 2 + yz 2 + zx2 ,
which directly follows from the AM-GM inequality.

54. Let x, y, z be positive real numbers such that xyz ≥ xy + yz + zx. Prove
that
xyz ≥ 3(x + y + z).
280 Solutions

Solution: We have
 2
xy + yz + zx − 3xyz(x + y + z)
= (xy)2 + (yz)2 + (zx)2 − (xy)(yz) − (yz)(zx) − (zx)(xy)
10 1
= (xy − yz)2 + (yz − zx)2 + (zx − xy)2 .
2
Hence it follows that
 2
3xyz(x + y + z) ≤ xy + yz + zx ≤ (xyz)2 .

Since xyz > 0, we obtain 3(x + y + z) ≤ xyz.

55. Let b1 ,b2 ,b3 ,. . .,bn be n non-negative real numbers and let b denote the
sum of these numbers. Prove that

n−1
b2
bj bj+1 ≤ .
j=1
4

Solution: Let bk denote the largest among bj ’s. Then we have


n−1 
k−1 
n−1
bj bj+1 = bj bj+1 + bj bj+1
j=1 j=1 j=k


k−1 
n−1
≤ bk bj + bk bj+1
j=1 j=k
 
= bk b − b k
 2
b2 b b2
= − − bk ≤ .
4 2 4

56. Let a, b, c, d be complex numbers such that ac = 0. Prove that


  √
max |ac|, |ad + bc|, |bd| −1 + 5
   ≥ .
max |a|, |b| |c|, |d| 2
√ 
Solution: Let us take r = b/a, s = d/c and k = 5 − 1 /2. Then k 2 = 1 − k
and 0 < k < 1. We have to show that
0 1 0 1 0 1
max 1, |r + s|, |rs| ≥ k max 1, |r| max 1, |s| .

We consider several cases.


Case 1. |r| ≥ 1 and |s| ≥ 1.
Solutions 281

In this case
0 1 0 1 0 1
max 1, |r + s|, |rs| ≥ |rs| > k|r||s| = k max 1, |r| max 1, |s| .

Case 2. |r| < 1 and |s| < 1.


We obtain
0 1 0 1 0 1
max 1, |r + s|, |rs| ≥ 1 > k = k max 1, |r| max 1, |s| .

Case 3. |r| < 1 and |s| ≥ 1.


We have to show that
0 1
max 1, |r + s|, |rs| ≥ k|s|.

If either |r + s| ≥ k|s| or 1 ≥ k|s|, the result follows. Suppose |r + s| < k|s|


and 1 < k|s|. Then
|r| + |r + s| = | − r| + |r + s| ≥ |s|,
and hence
|r| ≥ |s| − |r + s| > |s| − k|s| = k 2 |s|.
Thus, we obtain
|r||s| > k 2 |s|2 > k|s|
since k|s| > 1. This shows that
0 1
max 1, |r + s|, |rs| ≥ k|s|.

By symmetry, the result follows in the case |r| ≥ 1 and |s| < 1.

57. Let x, y, z be three real numbers in the interval [0, 1] such that xyz =
(1−x)(1−y)(1−z). Find the least possible value of x(1−z)+y(1−x)+z(1−y).

Solution: We show that


3
x(1 − z) + y(1 − x) + z(1 − y) ≥ .
4
If x = 0, we see that either y = 1 or z = 1 and the inequality follows. Similar
is the case when y = 0 or z = 0. We may hence assume that x > 0, y > 0 and
z > 0. Now put
1−x 1−y 1−z
= a, = b, = c.
x y z
Then abc = 1 and we have to prove
 c 3
≥ .
(1 + a)(1 + c) 4
cyclic
282 Solutions

This is equivalent to
⎛ ⎞ ⎛ ⎞
   
4⎝ a+ ab⎠ ≥ 3 ⎝1 + a+ ab + abc⎠ .
cyclic cyclic cyclic cyclic

This reduces to
 
a+ ab ≥ 6.
cyclic cyclic

However, we have
 1/3
a+b+c ≥ 3 abc = 3,
 2/3
ab + bc + ca ≥ 3 abc = 3.

These two together give


 
a+ ab ≥ 6.
cyclic cyclic

Equality holds if and only if a = b = c, which is equivalent to x = y = z.

58. Let x1 ,x2 ,x3 ,. . .,xn be non-negative real numbers such that


n
1
≤ 1.
j=1
1 + xj

Prove that x1 x2 x3 · · ·xn ≥ (n − 1)n .

Solution:
  Introduce yj = 1/(1 + xj ), for 1 ≤ j ≤ n.
nThen we get xj =
1 − yj /yj , 1 ≤ j ≤ n. The condition translates to j=1 yj ≤ 1 and the
conclusion we have to derive is
n 
" 
1 − yj
≥ (n − 1)n .
j=1
yj

If we use the AM-HM inequality, we get

n
1 n2
≥ n ≥ n.
y 
j=1 j
yj
j=1
Solutions 283

Using the AM-GM inequality, we get


⎛ ⎞n
n 
"  n
1 − yj ⎝ 1 1 − y j⎠

j=1
yj n j=1 yj
 n  n
11
= −1
j=1
n yj
 n n
1 1
= −1
n j=1 yj
 n
n2
≥ −1
n
= (n − 1)n .

Alternate solution:
We introduce the polynomial
    
P (z) = z + x1 z + x2 · · · z + xn .

Then

n
1 P  (1)
= .
j=1
1 + xj P (1)

The given condition is P (1) ≤ P (1). We now write

P (z) = z n + σ1 z n−1 + σ2 z n−2 + · · · + σn ,

where σ’s are elementary symmetric functions in x1 ,x2 ,x3 ,. . .,xn . Then the
derivative is given by

P  (z) = nz n−1 + (n − 1)σ1 z n−2 + (n − 2)σ2 z n−3 + · · · + σn−1 .

The condition P  (1) ≤ P (1) is now equivalent to

σn ≥ (n − 1) + (n − 2)σ1 + (n − 3)σ2 + · · · + σn−2 .

Now we use the inequalities


 
n  j/n
σj ≥ σn , 1 ≤ j ≤ n.
j
 1/n
Taking r = σn , we obtain
     
n n n
r −
n
r n−2
−2 r n−3
− · · · − (n − 2) r − (n − 1) ≥ 0.
n−2 n−3 1
284 Solutions

Consider the polynomial


   
n n
Q(x) = x −
n
x n−2
−2 xn−3 − · · ·
n−2 n−3
 
n
− (n − 2) x − (n − 1).
1
We observe that Q(x) has only one change of sign for x > 0. Hence the number
of positive roots of Q(x) = 0 cannot exceed 1. However, it is an easy exercise
to check that Q(n − 1) = 0. Thus, it follows that Q(x) ≥ 0 in (0, ∞) if and
only if x ≥ (n − 1). Since Q(r) ≥ 0, we conclude that r ≥ (n − 1). This gives
σn ≥ (n − 1)n .

59. For positive real numbers a, b, c, prove the inequality


 √ √   
3
3 a + ab + abc ≤ 4 a + b + c .

Solution: Here is a beautiful solution to the problem. We write


√ √
3 1√ 1√3
a+ ab + abc = a + a · 4b + a · 4b · 16c.
2 4
Using the AM-GM inequality, we have
1√ 1√3
a+ a · 4b + a · 4b · 16c
2 4
1  1 
≤ a + a + 4b + a + 4b + 16c
4 12
4 
= a+b+c .
3

60. Show that for any two natural numbers m, n, the inequality
1 1 4
− ≤
m + n + 1 (m + 1)(n + 1) 45
holds.

Solution: Suppose we introduce


1 1
f (m, n) = − .
m + n + 1 (m + 1)(n + 1)
Then it is easy to see that f (1, 1) = f (1, 2) = 1/12 < 4/45. Now it is not hard
to prove that
1 4
≥ .
(m + 1)(n + 1) (m + n + 2)2
Solutions 285

In fact, this is just equivalent to (m − n)2 ≥ 0. Thus if k = m + n + 2, then


1 1
f (m, n) = −
m + n + 1 (m + 1)(n + 1)
1 4
≤ −
m + n + 1 (m + n + 2)2
1 4
= − 2.
k−1 k
Consider the function g(k) defined for natural numbers k ≥ 2 by
1 4
g(k) = − 2.
k−1 k
We show that it is a non-increasing function for k ≥ 6. In fact for k ≥ l,
1 4 1 4
g(k) − g(l) = − 2− + 2
k−1 k l−1 l
l−k (l − k)(l + k)
= −4
(l − 1)(k − 1) l2 k 2
(lk − 2l − 2k)2 − 4l − 4k
= (l − k) .
(l − 1)(k − 1)l2 k 2

Since l − k ≤ 0, we have to show that lk − 2(l + k) ≥ 2