Inequalities

TEXTS AND READINGS
IN MATHEMATICS 49
Inequalities
An Approach Through Problems
Texts and Readings in Mathematics
Advisory Editor
C. S. Seshadri, Chennai Mathematical Institute, Chennai.
Managing Editor
Rajendra Bhatia, Indian Statistical Institute, New De1hi.
Editors
R. B. Bapat, Indian Statistical Institute, New Delhi.
V. S. Borkar, Tata Inst. of Fundamental Research, Mumbai.
Prob al Chaudhuri, Indian Statistical Institute, Kolkata.
V. S. Sunder, Inst. of Mathematical Sciences, Chennai.
M. Vanninathan, TIFR Centre, Bangalore.
Inequalities
An Approach Through Problems
B JVenkatachala
Indian Institute of Science
Banglore.
rugl@ HINDUSTAN
U l.QJ UBOOK AGENCY
Published in lndia by
Hindustan Book Agency (lndia)

P 19 Green Park Extension
New Delhi 110 016
lndia
email: info@hindbook.com
http://www.hindbook.com
ISBN 978-81-85931-88-3 ISBN 978-93-86279-43-9 (eBook)

DOI 10.1007/978-93-86279-43-9
Copyright © 2009, Hindustan Book Agency (lndia)

Papercover edition 2015
No part of the material protected by this copyright notice may be

reproduced or utilized in any form or by any means, electronic or
mechanical, including photocopying, recording or by any information
storage and retrieval system, without written perm iss ion from the
copyright owner, who has also the sole right to grant licences for
translation into other languages and publication thereof.
All export rights for this edition vest exclusively with Hindustan Book
Agency (lndia) . Unauthorized export is a violation of Copyright Law
and is subject to legal action.
ISBN 978-93-80250-71-7
Preface
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 wh ich 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 apart 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 so me 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 geometrie
inequalities, is based on geometrie structures, like, triangles and quadrilaterals.
Some important geometrie 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, wh ich are discussed in the sixth chapter.
As I said earlier, the problems have been taken from various sourees. I have
tried to give reference to them where ever possible and whenever I had one.
vi
I deeply regret and apologise for any inadvertent omission in mentioning the
source.
This whole exercise of writing a book on inequalities is the out co me 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 für
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 acknowlerlge 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.
Table of Contents
Preface ...................................................................................................... v
1 Some basic inequalities ............................ .......... 1
1.1 Introduetion . 1
1.2 Arithmetie mean-Geometrie mean inequality . 2
1.3 Cauehy-Sehwarz 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 Coneave funetions, Jensen's inequality 34
1.8 Inequalities for symmetrie funetions 47
2 Techniques for proving inequalities ......... ............. 51
2.1 Introduetion . . 51
2.2 Use of induetion . 51
2.3 Applieation of known inequalities . 54
2.4 Use of ealculus in inequalities . 65
2.5 Trigonometrie substitutions . . 73
2.6 Properties of quadratie polynomials 75
2.7 A useful transformation . 78
2.8 Sehur's inequality . . . 80
2.9 Majorisation teehnique 82
2.10 Muirhead's theorem. 84
2.11 Homogenisation . 86
2.12 Normalisation .. 89
2.13 Stolarsky's theorem 91
3 Geometrie inequalities ...................................... 94
3.1 Introduetion. 94
3.2 Notations .. 94
3.3 Some identities involving elements of a triangle 96
3.4 Some geometrie inequalities . . . . . . . 98
3.5 T wo triangles one inseribed in the other . 124
3.6 Let P be a point. .. . ......... . ]30
4 Applieations involving inequalities ...................... 155
5 Problems on inequalities .................................. 187
6 Solutions to problems ...................................... 219

Chapter 1
Some basic inequalities
1.1 Introduction
As we all know, one of the important properties of real numbers is comparability.
'Ve can compare two distinct real numbers and say that one is srnaller
\Ve sm aller or larger
than the other. There is an inherent ordering < on the real number system
IR which helps us to compare two real numbers. The basic properties of this
ordering on IR
~ are:
Giveri any two real numbers a and b, one and only one of the following
(i) Given
true:
three relations is true:
= b or a > b;
a > 0 imply a + b >
> 00 and b > > 0;
(iii) a > 00 and b >
> 0 imply ab >
> O.
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
Hölder's
of convex and concave functions. Here are some easy consequences of these
fundamental properties of ordering on IR:
(1) a < b, then a + c 0 give ac < bc;
(2) a < band cc> bc; if c << 00, ,we have ac >
> bc;
bc;
(3) 0< a 0; a a < 00 and b > 0 imply ab < 0;
a < band b
(5) a< b<< c together imply a < c (transitivity);
> 0, we have a < b;
(6) if ac < bc and cc>
0 < a < 11implies a2 < a; if a
(7) 0< > 1, we
we have a 2 > a;
for any real a, aa2 2:
(8) für ~ 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 > band 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 > band c > d, none of
'h a b . . c dA' h .
a, b,C, d equa10 ,nelt er ~ > d 1S true nor 1S ;: > b' gam t e reason 1S
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
lxi ==
lxi {x
{~x
-x
if x 2 0
if x< O.
Note that lxi 2 0 and lxi == 0 if and only if x == O. The absolute value function
x -+ lxi has some ni ce properties:
(1) I - xl = lxi; Ixl 2 = x 2 ; Ixyl = Ixllyl; Ix/yl = Ixl/lyl for y -I O.

(2) Ix + yl ~ lxi + lyl and equality holds if and only if x and y have the same
sign. This is known as the triangle inequality.
(3) IIXI-lyll ~ Ix - yl·
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 nu mb er. For any complex number z = a + ib,
we associate its complex conjugate by z = a - ib and absolute value by Izl =
Ja 2 + b2 . Note that Izl 2 = Z· z. It is easy to check the following properties of
14
(1) Izl 20; and Izl = 0 if and only if z = 0;
(2) IZ1Z21 = Izd IZ21;

(3) IZ1 + z21 ~ IZ11 + IZ21 and equality holds if and only if Z1 = AZ2 for some
positive real nu mb er A, or one of Z1, Z2 is zero.
1.2 Arithmetic mean-Geometric mean inequality

The basic inequality in the real number system is x 2 2 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 Va and Vb are meaningful as real numbers. Sinee the
square of areal number is always non-negative, (Va - Vb) 2 2: O. This may be
rewritten in the form
a
- +-b >
a+b v'a:b.
"b
2 -> vao. (1.1)
The real num ber a; b is ealled the arithmetie mean of a and b; similarly ~
is known as the geometrie mean of a and b. Thus the property that x 2 2: 0
for areal number x implies that the arithmetic mean of two non-negative
real numbers eannot be smaller than their geometrie mean. Moreover, the
derivation also shows that equality in (1.1) holds if and only if a = b.
This may be eonsidered in a general setting. Starting with n non-negative
real numbers al, a2, ... ,an, we define their arithmetie mean A(al, a2, ... ,an)
and geometrie mean G(al, a2,··· ,an) by
aj + a2 + ... + an
A(al,a2' ... ,an)
n
( ala2··· l/n
G (al, a2, ... ,an) an ) .
As in the ease of two numbers, a eomparison between these two means leads
to the AM-GM inequality.
Theorem 1. Given any n non negative real numbers al, a2, ... ,an, they satisfy
the inequality
al + a2 + ... + an > (al a2 ... an) l/n . (1.2)
n
= a2 == ...
Equality holds in the inequality if and only if al = ... == an.
Proof: There are several proofs of this classical theorem. We give here a clever
induetion proof which was originatcd by Cauehy. It also illustrates how one
ean go ahead in an induetion proof leaving out the validity of induetion for
so me numbers and subsequently prove the validity for missing numbers using
an interpolation argument.
We have already obtained the result for n = 2;
al + az 2: (ala2) 1/2 . (1.3)

2
Here equality holds if and only if al = a2. Consider 4 non-negative real numbers
aj, a2, a3, a4. Divide them in to two groups; {al, a2} and {a3, a4}. Applying
the known inequality for eaeh group, we obtain
aj + a2 2: (ala2) 1/2 , a3 + a4 2: (a3 a4 ) 1/2 .

2 2
4 Inequalities
This leads to
(al; a2 ) + (a3 ; a4)

a1 + a2 + a3 + a4
4 2
( a1a2 ) 1/2 + (
a3 a4)1/2
>
2
1/2
{ 1/2 1/2 }
> (a1a2) (a3 a4)
( 1/4
a1 a2a3a4) .
Thus the inequality (1.2) is obtained for n = 4. It may again be observed that
al =
equality holds here if and only if a1 = a2, a3 = a4 and a1
al a2 = a3a4. Since all
numbers are non-negative, it follows that a1al = a2 = a3 = a4. Now this can be
= 8. Now induction shows that (1.2) holds for n == 2k
used to prove (1.2) for n =
for aB k E N. Here again thc condition for equality is that all the 2k numbers
be cqual.
Now consider any n non-negative real numbers al, a2, ... ,an. We choose a
natural number k such that 2 k - 1 S; n < 2 k . Put
A _ a1
al + a2
az + ... + an
A=
- , ,
n
and consider the set {al, a2, ... ,an, A, A, ... A} of 2k numbers; here A appears
2 k -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
1/2'
al
a1 + a2 + ... + an
an A +A ~
+,.:t ... + ~
+v ... A' ~
>
-
an ~
2k (al a2 ... an AA··· A )
'---v--"
2k-n
2k-n 22 k -n
-n
This reduces to
k
A>
A '? ((ala2 ... an
a1a2··· A2k-n) 1/2
anA2"-n)1/2k
A simple manipulation now yields A 2: a1

Ann ~ al 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.
•
al, a2, ... ,an be n positive real numbers whosc product is
Example 1.1. Let a1,
1. Prove that
(1 + a1)(1 + a2) ... (1 + an) ~ 2n .
Solution: Using the inequality (1 + aj) ;::

2: 2.;a;
2 JClj for 1 ::; j ::; n, we obtain
(1 + al)
a1) (1 + a2)'" 2: 2n (ala2"
a2) ... (1 + an) ;:: (al a2 ....aan) /2 = 2n .
n )11/2
•
Example 1.2. Show that for any natural number n > 1, the inequality
(2n)! < {n(n + l)r,
{n(n+1)r,
holds good.
Solution: We split (2n)! as two products:
5 (2n - 1)) (2 . 4 . 6 ... (2n))

(2n)! = (1 . 33 . .5 ... (2n) ) .
Using the AM-GM inequality, we get

2nn (1·2.3
(1 . 2 . 3 .... n)
C
2·4·6···(2n)
2·4·6··· (2n) = ··n)
< 2nC+2+3n+···+n)n
2n + 2 + 3n+'" + n) n =(n+1)n,
= (n + l)n,
and
1.3.5
1 ·3 . 5·... (2n -_ 1)
.. (2n 1) <
< C
( 1 + 3 + 5 + 'n'
~.. + (2n 1))
(2n -- 1)) n
n
= nn.
Combining these two inequalities, one can obtain
(2n)! < (n+ l)nnn = {n(n+ l)r,
as desired.
•
Example 1.3. If a, b, c are the sides of a triangle, prove that
3 abc
- < - - + - - + - - < 2. (1.4)
2 - b+c c+a a+b
Solution: The first part of the above inequality is equivalent to

9 abc
- < --+1+--+1+--+1
2 b+c c+a a+b
1 1 l'
( 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
9::; (x+y+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; fuH 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
abc a c c
- - +c+a
b+c
- - +a+b
-- < - - +c+a
a+c
- - +a+b
--
c
1+--
a+b
< 2,
since c < a + b by the triangle inequality. •
a1, a2, ... ,an are n positive real numbers, we define their harmonie mean
If al,
by
n
H(al'U2, ... ,an) =
-+-+ ... -1
1 1
al
a1 a2 an
Thus the harmonie mean of n positive real numbers is equal to the reciprocal
of the arithmetic mean of the reciprocals of the given numbers. Since
1 1 1 ( 1 ) l/n
l/n
-+-+ ... +- >n
al
a1 a2 an - a1 a2 ... an
al '
it follows that
G(al,a2, ... ,an) ~ H(al,a2, ... ,an).
Thus we have for n positive real numbers al, an, the inequality
a1, a2, ... ,,an,
A(al,a2' ... ,an) ~ (al,a2' ... ,an) ~H(al,a2' ... ,an),
or briefly AM ~;:::: GM ~HM.

;::::HM. This is often referred to as the AM-GM-HM in-
equality. Here again equality holds if and only if aH the numbers are equal.
Example 1.4. For any four positive real numbers al,a2,a3,a4,

a1,a2,a3,a4, prove the in-
equality
a1
- al
-
---+ - + - a2
- - + - a3 + - a4
- - +--- --
al + a2 a2 + a3 a3 + a4 a4 + al
a1
< al
a1 + a2 a2 + __a3
a-,-3_ + __a4
a_4_
<
- a2 + a3 + a3 + a4 + a4 + ala1
+--
a1 + a2
al
Solution: Adding
a2 a3 a4 al
---+ + +---
al + a2 a2 + a3 a3 + a4 a4 + al
to both sides, the inequality can be written in the following equivalent form:
a2 ( 1 + 1 ) + a3 ( 1 + _1_)
a3 + a4 a1 + a2 a4 + a1 0,2 + a3
+ a4 ( 1 + 1 ) + a1 ( 1 + 1 ) > 4.
~+~ ~+~ ~+~ ~+~
However, using the AM-HM inequality, we know that

1 1 4
-+ + a1 > ,
a3 + a4 al + a2 a1
al + a2 + a3 + a4
1 1 4
a4 + a1
al
+ a2 + a3 >
al
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
- - + - - + - - =8,
b+c c+a a+b
where 8 is the semi-perimeter of the triangle. Prove that the triangle is equi-
lateral.
Solution: Observe that

a+b b+c c+a
28 = -2
- +2
--+
2
--
222
> 1 1 + 1 1 + -1--1
-+-
a b
-+-
b c
-+-
c a
2ab 2bc 2ca
- - + b+c
a+b
- - +c+a
--
28.
Here the AM-HM inequality has been used. Thus equality holds in the AM-HM
inequality and hence a = b = c. •
c; c) (c; r/
Example 1.6. Suppose a, b, c are positive real numbers. Prove the inequality
( a ; b) a) ~ (a +: + c) (abc
3
. (1.5)
Solution: Introduce new variables x, y, z by
a b c
x----- z- - - - -
- a+b+c' y= 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) ~ ~(XYZ)2/3. (1.6)
This ean be rewritten using x + y + z == 1 in the following form:

(l-x)(l-Y)(l-z) ~ ~(XYZ)2/3. (1. 7)
Expanding the left hand side and using onee again x + y + z == 1, we may write
(1. 7) in the form
(XYZ)1/3(~+~+~_1)
(xy z)1/3 (1- + -1 + -1 - 1) ~> 8~
-.
- 33·
(1.8)
x y z
Now the AM-HM inequality gives
1/3
(XYZ)1/3~1
(xyz) ~ T l l'
-+ +-
i
3
x Y z
so timt
(XYZ)1/3(~+~+~_1)
(xy Z) 1/3(1- + -1 + 1
--) ~3-
1 ~ 3 - (xy z)1/3 .
(xYZ)1/3.
x y z
Henee it is suffieient to prove that
(xyz) <
(xYZ)1/3
1/3
!3'1
-~ 3.
Sinee x + y + z == 1, this follows from the AM-GM inequality.

•
Now let us see how one may proeeed to generalise the AM-GM inequality.
Consider n non-negative real numbers a1, al, a2, a3, ... , an and n positive integers
k 1l ,k2 ,k3 , ... ,kn . Form a set in whieh eaeh aj appears exaetly k j times, 1 :S ~
j :S ~ n. This set eontains k k1l + k 22 + k 3 + ... + k n non-negative real numbers.
The arithmetic me an and the geometrie mean of these numbers are
klal + k2a2 + ... + kna n

k1a1
kk 1l + k 22 + ... + k n
(a~'ak22 ... ann
k ) k,+k2~ ··+k n
respeetively. Now the AM-GM inequality gives
k 1a1 + k2a2 + ... + knan kn )

-----=----,--------"- > (k'
a a
k2 ... a
k 1 +k2+ ··+k n
k 1 + k 2 + ... + k n 1 2 n
Suppose we have positive rational numbers r1, ,rn. Then we ean

rl, r2, ... ,rn'
reduee all of them to a eommon denominator:
= -k ,,
r Jj. =
r
kj 1 :SJ:Sn,
.
~ J ~ n,
rn
where m and k j , 1 :S j :S n are natural numbers. It is not hard to see that
rIal + r2a2 + ... + rna n

n a 1 + k 2a2
k 1a1 a 2 + ... + kna n
r1 + r2 + ... + r nn k 1 + k 22 + ... + k n
Moreover, we have
(
ar 1 ar2
1 2'"
r
art
)"1 +r2~ .. +r
n ( a 1kl ak2
2 ...
k )
ann k, +k2~ ··+k n
It follows that
1
r1 a1 + r2a2 + ... + r. nan
> ( a~' a;2 ... a~n. ) ", +r2+"'+ n
r
r1 + r2 + .. , + r n
It is possible to furt her extend this inequality using a continuity argument. If
Al, A2,
A2 , ... ,A. n are positive reals and a1, a2, ... ,an are non-negative reals, then
... ,An
\ + A2
a2 + ... + Anan
> (Al
Al a1 A2 An) Al +A2+"'+"n
a 1 a 2 ... an (1.9)
Al + A2 + ... + An
This is known as the generalised or the weighted arithmetie mean- geometrie
mean inequality. With obvious modifications, this can be further carried to
the weighted arithmetie mean- geometrie mean-harmonie mean inequality.
Example 1. 7. Let a, b, e be positive real numbers. Prove that

aUbbc c :::: (abc)(u+b+ CC )/3.
Solution: Assigningthe 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~bbCC
vaUovc > a + b + c == a + b + e > vaoc
3~b
3/"lb
-!!+Q+f
ab
U b c c
3 - ,
which gives the required inequality.

•
Example 1.8. Let al :::: a2 :::: ... :::: an and b1, b22 ,, ...
..• ,bn be two sequences
of non-negative real numbers such that for each k, 1 :S k :S n, the inequality
b1b2 ... bk :::: a1 a2 ... ak holds. Prove that
bl + b2 + ... + bn :::: a1 + a2 + ... + an· (1.10)
Solution: We may assume that aj > 0 for all j; otherwise delete those par-
ticular aj 's which are equal to O. This does not affect the inequality. Consider
the set
{ ~, b2 , ... , bn },
al a2 an
10 Inequalities
where each bj / aj is attaehed with the weight aj. The weighted arithmetie
mean is
a1 (h) + (2.2) + ... +

al a2 a2 an (~)
an b1 + b2 + ... + bn
a1 + a2 + ... + an a1 + a2 + ... + an
and the weighted geometrie mean is
(G:f (!:f ... (!:r) .,+.,~+ ••
Since aj 2': aH 1 and b1 b2 ... bj 2': a1 a2 ... aj for 1 :::::: j :::::: n, we have
b1b2 ... b)aj-aj+l >1

( _ _ _..::...J
a1aZ· ··aj - ,
far 1 :::::: ::; n. Since

::; j :::::: Sinee
( ~)al (b2)a2 ... (bn)a n

a1 a2 an
b b ... bn ) an ( b1b2 ••. bn - 1 ) an -an-l (bI) a2- a,
( 1z
= a1a2·· ·a n a1 a 2·· ·a n -1 ... ~ ,
(!: ) (!:)
we conclude that
a,
aj (!~) a2a2 ...
... an 2: 1.
an 2'
It follows from the weighted AM-GM inequality that
b1 + b2 + ... + bn 2': a1 + a2 + ... + an·
Alternate Solution:
Put Aj == bj/aj, 1:::::: ::; n. We have to show that
1::; j ::::::
n
:Laj(Aj -1) 2'20.
: 0.
j=l
If L kk = L~=l (Aj - 1) then
Lk = Al + A2 + ... + Ak - k
b1 b2 bk
-+-+···+--k
a1 a2 ak
b b ... b l/k
)l/k
1 2 ··· bkk )
b1b2
> k (( - k 2'2: 0,
a1a2··· ak
for all k. On the other hand, we observe that
n n { k }
{; Lk(ak - ak+l) =
tLk(ak-ak+l) { ; (ak - ak+d ~ (Aj - 1)
t(ak-ak+d{t,(Aj-l)}
= (Al - 1)a1
= + (A2 - 1)a2 + ... + (An - l)a n .
(Here we have set an+1 = 0.) Since L k ~ 0 and ak ~ ak+1 for all k, it follows
that
(Al - 1)a1 + (A2 - 1)a2 + ... + (An - l)a n ~ 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 < a2 c2 + a 22 d2 + b2c2 + b2 d2
(a 22+ b22)(C
)(C22 + d22),),
where we have used 2abcd ::; a2d 22-I- b2c2 . This is a consequence of the inequality
(ad - bC)2 ~ O. Thus we obtain
lac + bdl ::; Va + b2 vc

Ja2--t:b Jc 2 + d
2 2 2 •.
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 dat product) of
two vectors. If we represent (a, b) by ~ and (c, d) by V;, then
ac + bd =~ . V;= I ~ I 11 V; I cosO,
where () is the angle between ~ and V;. This shows that
lac + bdl ::; I ~ 11 V; == Va
I J a2 + b2 J vcc + d 2 2.
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 {b 1,b 2,b3, ... ,bn } be any two sets
of real numbers. Then
II
n
ajb jf; ) (n
f;t,a,b,l"I (n
(t,a;f'
::; f; )
(t,b;f' a;
1/2
b;
1/2
(1.11)
Equality holds if and only if there exists a constant A such that aj = Ab

Abjj ,, for
1 ::; j ::; n.
12 Inequalities
Proof: Let us write

n n n
,",2
A = ~aj' B=
B ~b;,
= '"' 2
~ bJ' C= ~ajbj.
= l:ajb j .
j=1 j=1 j=1
The inequality to be proved is C 2 'S AB. If B = 0, then bj = 0 for all j so that

C == O. Hence the inequality trivially holds. We may therefore assurne that
B ::j:. O. Since B is the sum of squares of real numbers, B must be positive.
Using the fact that the square of areal number is non-negative, we have
n11 n n n
O'S l:
o 'S ~ (Baj - Cb j ) '2 ~a;
B 2 '~J
"' a2 + C 2'1 '~b; 2BC~ajbj
" ' b2 -- 2BC'"'
~J ~JJ a·b
j=1 j=1 j=1 j=1
B 2A + BC 22 -- 2BC 2
B(BA - C 2 ).
Since B > 0, it follows that C 2'1 'S AB. We also infer that equality holds if and
Cbjj = 0 for all j, 1 'S j 'S n. Taking A = C / B, the condition for
only if Baj - Cb
equality that aj == Ab j for 1 'S j 'S n is obtained. •
Remark 2.1: The Cauchy-Schwarz inequality extends to complex numbers as

weil with obvious modifications. Let us consider two sets {al, a2, a3, ... ,an}
and {bi, b2 , b3 , ... ,bn } of complex numbers. Then the inequality
I~ alJj
n I
'S
(n~ jaj 2) 1
1/2 ( n
~ jb j j2
) 1/2
holds good. Moreover equality holds here if and only if aj = Abj , 1 'S j 'S n,
aj = n,
for some constant A. The proof is similar to the one given above.
If we take n real numbers al, a2, ... ,an, then the Cauchy-Schwarz inequal-
ity gives
al + a2 + ... + an 'S Vr::/')
r;::/.)
V ai + a 22 + ... + an'
nnV
2
This may be written in the form
al + a2 + ... + an < lai

~----------~<
~----------~
jai + a§ + ... + a~
.
n - n
Here the left hand side is the arithmetic mean of al, a2, ... ,an and the right
side is the square root of the arithmetic mean of ai, a§, ... ,a;. 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
b2 + 2"
c
+ 2"
a
:::: -
c
+ -b +-.
a
Solution: We write
-a+c- + -b == --ab bc ca
.. -- + - . - + - . -.
c ba bc ca ab
Now applying the Cauchy-Schwarz inequality to the sets
{ ~b'c'a
~ ~} and {~c'a'b'
~ ~}
we get
a ce b a2 b2 C2 )1/2(b2 c2 a 2 )1/2
-+-+-
c b a
< ( -b')
~
+--2+-2
e: a -2+?+b
c 2 a~
a2 b2 c2
b')- + -2
c +?
a-
•
Example 1.10. Let al,a2, ... ,an and A be real numbers such that
n n 2
A +L a% < n ~ 1 ( k=l
La k)
k=l
Prove that A < 2alam for every pair {al, a m }, l 1:- m.

Solution: We may take l = 1 and m = 2; the proof for any other pair of
indices is the same. Suppose A 2: 2ala2. Then we see that
n n
A+ La% > 2ala2 + a2
1 + 2+
a2 """
~ ak2
k=l k=3
n
= ((al
al +a2)2
+a2 )2 + """
2:a~
~ak 2
k=3
k=3
> -
>
1
~ (n
-1 '"'ak
nn-l L (~akr)2 k=l
(Here we have applied the Cauchy-Schwarz inequality to the (n -1) tuples {aj +
a2, a3, ... ,an} and {1,
{I, 1, ... ,,I}.)
1 }.) But this contradicts the given condition. We
conclude that A < 2al a2. •
Example 1.11. (Short-List, IMO-1993) Let a, b, c, d be positive real numbers.

Prove that
a b c d 2
- -2e-+-3d+ c + 2d + 3a + d + 2a + 3b + a + 2b + 3e:
b + 2e:
>-.
3e - 3
14 Inequalities
Solution: We write
L .·
a =
= L .
v' Fa
y'bb + 2c + 3d
(Fav'b
(FaVb + 2c + 3d) ,
cyc I Je
cychc eye I·Je
cychc
where the surn is taken cyclically over a, b, c, d. Now we apply Cauchy-Schwarz

inequality to the sets
{ Fa ~ Vc Vc
~}
v'b + 2c + 3d' J c + 2d + 3a' V
Vb v'd + 2a + 3b v'a + 2b + 3c
3b'' va
and
Fav'b + 2c + 3d, JbJc + 2d + 3a, VcVd

{ FaVb Vcv'd + 2a + 3b, Jdva
Jdv'a + 2b + 3C}'
ar ~ (~
to get
(L
cychc cychc
b+2:+3d) (~a(b+2c+3d)).
cychc
However, we have
L a(b + 2c + 3d) = 4(ab + ac + ad + bc + bd + cd).

cyclic
eyclic
Thus it follows that

(a + b + c + d)2
2:. .,----."...--,--,:
a
.,.----,,--..,..-,: >
b + 2c + 3d - 4 (ab + ac + ad + bc + bd + cd)
.
cychc
eyehc
On the other hand,
3(a + b + c + d)2 - 8(ab + ac + ad + bc + bd + cd)

3(a 2 + b2 + c2 + d 2) - 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
> o.
Cornbining these two, it follows that
(a+b+c+d)2
--:--~---~----,..
~~~--~~~~-~ > -2
4( ab + ac + ad + bc + bd + cd) - 3
and hence
a 2
L
"""""
~ -=-b + 2ca + 3d 2>
b -+-2-c-+-3--=d
~3·
- 3·
cyclic
cyclie
•
ExaIllple 1.12. Let al ::::: a2 ::::: ... > an > 0 and bl1,b2, .. · ,bn be two se-
quences of real numbers such that
k k
2: a ~ 2: b
j j,
j=l j=l
for each k, 1 ~ k ~ n. Prove that

n n
2:
""
~a
j=l
~
aJ;2 -< 2: bJ'2
""
~
j=l
b;.
Solution: Take an+l = O. Since ak - ak+l ::::: 0 for 1 ~ k ~ n, we have
kk kk
( ak - ak+l) 2: aj ~ (ak - ak+l) 2: bjbj ,

j=l j=l
for 1 ~ k ~ n. Summing this over k, we obtain
a2l + a22 + '" + an2 ~ albl + a2 b2 + ... + anb n ·
Applying the Cauchy-Schwarz inequality to the right side, we get
al2 + a22 + ... + an2 ~

(2
a 2 2 ) 1/2 (2
l + a2 + ... + an bj + b22 + ... + b2)
n
1/2 .
This simplifies to
a2l + a22 + ... + an2 ~ b2l + b22 + ... + bn2 ·
•
ExaIllple 1.13. If a, b, c are positive real numbers, prove that
a2 3
2:
"" (a + b)(a + c) >
~(n..Lh\(n..Lr>\ -::::: -
4'
4'
where the summation is taken cyclically over a, b, c.
Solution: We have
( 2: a) 2 = ( 2: a J(a+b)(a+c)
)2
cyclic cyclic J(a + b)(a + c)
< (~(a+b~~a+C)) (~(a+b)(a+C))'

cychc cychc
16 Inequalities
using the Cauchy-Schwarz inequality. Thus we obtain
'" a2 (a+b+c)2
L...
cychc
(a+b)(a+c) ~ (a+b)(a+c)+(b+c)(b+a)+(c+a)(c+bf
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 num-
bers which add up to 1. Prove that
L A 1
nn n
aj >
L
j=1
j=l yIT-=aj ~
1 - aj -
Vn=1
1
L
L
n - 1 )=1
n
yfaj.
j=l yfaj.
Solution: We may write
L
L
n
aj =
L
n
n
1
-~
1
L viI -
n
n
'-1
J-
j=l
~
~
aj
V.L -- uJ
=
L
'-1
J-
j=l ~
-
J j=1 ~.
a· - '"
yIT-=aj j=l
V .L -
aj..
uJ.'
Using the A1-I-GM

AYI-GM inequality, we have
~
{;h, n{II7=1~
j~,~r
n { }1/n
> n
1
n l/n'
(TI
( IT7=1
n (
j=1 (11 - aj)
aj))
)1/n'
However,
n 1/n 1 n n- 1
(II(1-aj)) ~;;:L(l-aj)=-n-'
j=1 j=1
(i!!;
{i!!;
Thus we obtain,
n 1
'" yIT-=aj >n --.
~
~~- - nn-- l1
On the other hand, the Cauchy-Schwarz inequality gives
n n
L yIT-=aj ~ vIn, I L(1- aj) = y'n(n - 1).

j=1 j=1
We thus get
"Ln
~
1
- L. J1- aj 2 n ~
_n_ -
n
Jn(n -1) =
y'n
y'n
n ..
r
~ ~
j=l
J=l V.L -- aj
UJ
n-1 j=l
J=l n - 1 V 'n=l
vn=I
n -
.L
We also have
(~F; ~ ~"j ~ n n
Thus
n
v'n 2 Ly'{ij
j=l
and hence n
'" aj y'n 1 n
~ > > '"
j=l ~ - vn=I - vn=I ~ y'{ij.
J=l
•
1.4 Chebyshev's inequality
Let us consider the Cauchy-Schwarz inequality for two sets of real numbers
{ a1 , a2, ... , an} and {bI,
{b1, b2, . .. , bn } :
t
J=l
ajbj ~ (t a; ) ! ( t
J=l J=l
b; ) ! .
We may write this also in the form
~1(Ln ajb ) ~ (1~ Ln j a;

)!(1~ L n
b;) .
~
J=l J=l J=l
This shows that the Cauchy-Schwarz inequality relates the arithmetic means of
ai,a~, ... ,a;; and bi,b~, ... ,b;; with that ofthe numbers a1bl,a2b2,
a1b1,a2b2, ... ,anb n .
How do we relate the arithmetic means of the numbers al, a1, a2, ... , an and
b11,, b2, ... , bn with that of aa1bl,
l b1, a2b2, ... , anb n . Chebyshev's inequality answers
this question.
Theorem 3. Let {a1,a2, ... ,an} and {b 1,b2 , ... ,bn } be two sets of real num-
bers. Suppose either
a1 ~ a2 ~ ... ~ an and b1 ~ b2 ~ ... ~ bn ;
or
a1 2 a2 2 ... 2 an and b1 2 b2 2 ... 2 bn ;
18 Inequalities
i.e., both the sequences (ak) and (bk) are non-decreasing or both non-increasing.
Then the inequality
( al + a2 : ... + an) (bI + bz : ... + bn ) :s a1 b1 + az b2 + ... + an bn
holds. The inequality is strict unless at least one of the sequences is a constant
sequence.
Proof: We have
n n n n
LL (ajbj - ajbk ) L (najbj - aj L bk)

j=lk=l
j=lk=l j=l k=l
n t ajbj - ( t aj) ( t bk ).
j=l j=l k=l
Similarly, it is easy to obtain
t, ~ (akb k - akb j ) = n ~ akbk - (~ak) (~bj).

Using these two we obtain
n ~akbk - (~ak) (tbj)

1{ n n }
"2 ~ ~ (ajb j - ajbk + akbk - akbj)
1
LL
n n
"2 (aj - ak) (b j - bk).
j=l k=l
Since (aj - ak) (b j - bk ) ~ 0 whether both the sequences are non-decreasing or

both non-increasing, we get
n ~akbk - (tak) (t,b j ) > O.
Here equality holds if and only if für 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 (b j ) is a constant sequence .
•
Remark 3.1: Ifthe If the sequences (aj) and (bj ) are oppositely ordered, i.e., either
al ::::: az :::::
~ a2 ~ ...
.. . :::::
~ an and b1 1 2 b2 2 ... 2 bn , or al 2 a2 2 '" ' " 22an and
b1 ::::: ~ ...•.. :::::
~ b2z ::::: ~ bnn ,, then the inequality reverses;
( al + a2 : ... + an ) (bI + b2 : ... + bn ) 2 al b + a2 b2 + ... + anbn .

1
The proof is similar to the one given above, but with the crucial observation
that (aj -ak) (bj ~ 0 holds in this case for any pairs {aj, ad
(b j -bk) :::; ak} and {bj , bd.
bk }.
Remark 3.2: Chebyshev's inequality can be generalised to three or more sets

of real numbers. Suppose (al ,j),, (a2
(al,j) ,j), (a3,j) ... , (ar,j)
(aZ,j), (ar,j ),, 1 :::; ~ n be r sets
~ j :::::
of real numbers such that
o :::; ak,l :::; ak ,2 :::; ... :::; ak,n,
~ k :::;
for 1 :::; ~ r. Then the following inequality holds:
(~~a1,j) (~~a2,j) (~~a3,j) . .. (~~ar,j)

1
:<-:; -n- L
n
'"' al,j
~ al ,)'a2 'a3 . ...
a2,ja3,j
,),) .. . ar,j·
a '.
r ,)
n j=l
j=1
For r 2 3, it is essential to consider only non-negative sequences.
Remark 3.3: Suppose a1, az, ... ,an and b1 , b2, ... ,bn are two real sequences
al, a2,
such that either both are non-decreasing or both are non-increasing, and let
PI
PI,,P2,··· Pn be a sequence of non-negative real numbers such that L?=l
Pz, ... ,,Pn L7=1 Pj is
positive. Then the following inequality holds:
( L?=~ jaJbj)
Lj=l PJ
2 (L?~lPjaj) (L?~lPjbj).
Lj=l PJ Lj=l 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 cnn a n - 11 + bn - 11 + cn - 11
-- + -- + -- > - - -- -- -c --- ----- -
b+c c+a a+b - 2
Solution: Since
Sinee the expression is symmetrie in a, b, c, we may assume that
a2 b 2 c. Then we have
abc
->
-- >- --->
>----.
b+c - c+a - a+b'
b+c-c+a-a+b
20 Inequalities
In fact
a
_.> __b
-
b +-c -> -
c +-a ~
~
ac+a
ac
22
+ a 2 2 bc
bc+b 2
+ b2
b+c-c+a
~ (a-b)(c+a+b)20
~ a2b.
Thus we have
abc
--.
an-I> bn - 11 > cnn-- 11 and - - > - - > - -.
- - b+c-c+a-a+b
Using Chebyshev's inequality for these numbers, we get
( aan-1 + bn-1
n-l +b + cn-1)
n-l +c n-l) ( + _b_
a- +
(_a_
- + _c_)
-b- + - - <3
b+c c+a a+b -
(~
< 3 (an
--++~ n
+ ~).
-b - +
b+c c+a a+b
- c)
cn- ) .
However, we have
abc 3
--+--+-->-
- -+--+-->-
b+c c+a a+b - 2'
using the left part of the inequality (1.4). Hence
an bn ce an- 11 + bn- 11 + cn- 11

--+ + ---- +
+ ---->>
--------
b+c c+a a+b - 2
•
Example 1.16. Let al, az, a2, ... ,an be n real numbers in the interval [0,1) such
+ a2 ++ ...
that al + ... ++ an = 1. Prove that
2: n
n - aa·
---'}'-- >
. l+na
o
>-
n
__ _1
--
_ _...c)!...-
2·
0
J· -o
-
2 -
}=l
)=1
Solution: First we arrange aj's such that al ::; a2 ::; ... ::; an. Then we have
n - al 2 n - a2 2 ... 2 n - an·
+ nal ::; 1 ++ na2 ::; ... ::; 1 ++ nan, and

The arrangement of aj 's shows that, 1 +
hence
1 1 1
---> > ... > - - -
1++ nal 1 ++ na2 1+ + nan
Using Chebyshev's inequality, we obtain
~
n
n
'"'
2:
- aj
nn-a'
+ na)}
l+n;
1
.!.(~(n-a.))(t
1
> ;(t;(n-a
n ~ Jj )
n 1 )
1
l+naj
1 +na'
)
)(nt;
j=l
j=1 j=l j=l J
> n2
2 _ 1
> -n--
n::
1(n 1 )
2: 1 ++na/
naj .
(nt;
}=l J
Using the AM-HM inequality, we also get
> n2 n2 n
.t;
1
t(1 +
n
1 + naj - naj) 2n 2
j=l
Combining these, we finally get
to
)=1
n-aj
1 + naJo
~ (~)
n
(!!:)
2
n 2 -1
2
•
1.5 Rearrangement inequality
... ,bn }. We
Consider two sets of real numbers: {al, a2, ... ,an} and {bI, b22,, •..
pose the following question: among all permutations (aj) of (aj) and (bj) of
(b j ), wh ich permutation maximises the product aj bj and which minimizes L
it? The rearrangement inequality answers this question.
Theorem 4. Let a1 :::; a2 :::; ... :::; an and b1 :::; b2 :::; ... :::; bn (or a1 >
~
2: ... ~
a2 ~ 2: an and b1 ~
2: b22 ~
2: ... ~
2: bn ) be two sequences of real numbers. If
a~,a;, ... ,a~ is any permutation of a1,a2, ... ,an, the inequality
n n n
L aj bn+1_
bn+ 1- j :::; L ajbj :::; L ajbj ,
j=l j=l j=l j=l
j=l
n
holdso
holds. Thus the sum L ajbj is maximum when the two sequences (aj) and (bj )
j=l
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 bj 's are non-decreasing; the proof
is similar in the other case. Suppose (aj) f. (aj). Let r be the largest index
such that a~ f. aar;
r ; i.e., a~ f. ar and aj = aj for r < j :::; n. This implies that
a~ is from the set {al, a2, ... ,ar-I} and a~ < a ar.
r . Further this also shows that
a~, a~ ... ,a~ is apermutation of al, a2, ... ,ar. Thus we can find indices k < r
and l < r such that a~ = ar and a~ = a/. It follows that
akI - aIr = a r - a/ ~ 0, br - bk ~ o.
22 Inequalities
We now interchange a~ and a~ to get a permutation a~ a~,, a~, ... , a~ of a~, a~,
a~;; thus
... , a~
a"J = a'J' i- r, kk
if j -::f.
{ a"r = a,k = ar,
a% = a~ = al.
Let us consider the sums
5" = a~bl + a~b2 + ... + a~bn, 5' = a~bl + a~b2 + ... + a~bn.
S" - 5':
Consider the difference 5" S':
n
S" - 5'
5" S' (a'j - aj)bj
L (aj
j=1
j=l
(a% - a~)bk + (a~ - a~)br
a~)bk + (a~ -
(a~ - aDbk a~)br
(a~ - a~)(br - bk).
bk ).
Sinee a~ - a~ 2: 0 and br - bk 2: 0, we conclude that 5"

Since S" 2: 5'.
S'. Observe
that the permutation a~, a~, ... , a~ of a~, a~, ... ,a~ ,a:1 has the property that
a'j =
aj ai =
= aj :S n and a~ =
= aj for rr < j ::; = a~ = = a rr.. Hence the permutation (a'j)
(a'J) in
place (ai) may be considered and the procedure
plaee of (aj) proeedure can be continued as above.
After at most n - 1 such steps, we arrive at the original permutation (aj) from
(aj).
(ai). At each step the corresponding sum is non-decreasing. Hence it follows
that
a~bl + a;b 2 + ... + a~bn ::;
a;b2 albll + a2b2 + ... + anb
:S alb n.
anbn· (1.12)
To get the other inequality, let us put
Cj = a~+l_j' dj = -bn+l - j .
C2, ...
Then Cl, C2, ... ,C a2,, ... ,an
al,, a2
,Cn is apermutation of aj , an and d1 ::;:S d2 ::;
:S .... :S dn .
. . ::;
Using the inequality (1.12) for the sequences (Cj) and (dj (d j ),
), we get
c1d l + C2 d2 + .. + cndn ::; a1d 1 + a2 d2 + .. + and n .

Substituting back Cj and dj , we get
n n
- La~+l-j bn+l :S - Lajbn+l
bn+ l-- j ::; Lajbn+ l _j '.
j=1
j=l j=l
j=l
This reduces to
a~bl + a~b2 + ... + a~bn 2: alb n + a2bn-1 + ... + anbl , (1.13)
which gives the other part.

It is not difficult to find the condition und er which equality holds in the
inequality. If for each pair k, I with 1 ~ k a;
and bk = b!, then equality holds in (1.12). A similar condition is true for
equality in (1.13): for each k, I with 1 ~ k a~+1-1 and bn+1- k = bn+1-1. •
Many of the inequalities we have studied so far can be derived using the
rearrangement inequality.
Corollary 4.1: Let 01,02, ... ,On be n given real numbers and ß1, ß2, ... ,ßn
be apermutation of 01,02, ... ,On. Then
n n
LOjßj ~ L O;'
j=l j=l
with equality if and only if (Oj) = (ßj).

Proof: Let o~, o~, ... ,o~ be apermutation of 01,02, ... ,On such that o~ <
O~ ~ ... ~ O~. Then we can find a bijection a of {I, 2, 3, ... ,n} onto itself such
that oj = o,,-(j), 1 ~ j ~ n; i.e., a is a permutation on the set {I, 2, 3, ... ,n}.
Let ßj == ß,,-(j). Then ß~, ßb, ... ,ß~ is apermutation of o~, o~, ... ,o~. Ap-
plying the rearrangement inequality to o~, o~, ... ,o~ and ß~, ßb, ... ,ß~, we
get
n n n
L oj ßj ~ L (oj) 2 = L oJ .
j=l j=l j=l
On the other hand, we observe that
n n n
Lojßj = L0,,-(j)ß(T(j) = LOjßj,
j=l j=l j=l
since a is a bijection on {1,2,3, ... ,n}. It follows that

n TI
L Ojßj ~ L 0]-
j=l j=l
Suppose equality holds in the inequality and (Oj) =I- (ßj). Then (oj) =I- (ßj).
Let k be the largest index such that o~ =I- ß~; Le., o~ =I- ß~ and oj = ßj for
k < j ~ n. Let m be the least integer such that o~ = ß:n. If m > k, then
= o:n and hence o~ == o:n. This implies that o~ == 0~+1 == ...
ß:n = ... == o:n and
hence ß~+1 = ... = ß:n· Eut now we have a block of m + 1 - k equal elements
among o"s and m - k elements among ß"s. It follows that there is an m1 > m
such that o~ = = ß:n 1 • Using m1 as pivotal, we obtain o~ == 0~+1 == ... ... ==
= ...
o:n = ... == 0:n 1 and ß~H == ...
... = ß:n == ...
... == ß:n 1 • This process cannot be
continued indefinitely. We conclude that o~ == ß{ for some I < k, thus forcing
m< k.
24 Inequalities
Obviously ß~ i= ß~ by our choice of k. We know that equality holds if and

only if for any two indices r i= s, either a~ = a~ or ß~ = ß~. Since ß~ i= ß~,
we must have a~ == a~. But then we have a~ == a~+1 == ... ... == a~. Using the
minimality of m, we see that k - m + + 1 equal elements a~, a~+1 ' ... ,a~ must
be among ß~, ß~+l'··· ,ß~ and since ß~ is different from a~, we must have
a~ = ßf for some l > k. But then using ßf = = a; we get
am
I
= a m +1 = ... = ak
I I
= ... = az·
I
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 (aj) == (ßj). It
now follows that (aj) = (ßj). •
Corollary 4.2: Let a1, a2, ... ,an be positive real numbers; let ßl, ß2, ... ,ßn
be apermutation of a1, a2, ... ,an. Then
~ ßj 2': n,
L
La·aj
j=l J
and equality holds if and only if (aj) == (ßj).

Proof: Let a~, a~, ... ,a~ be apermutation of a1 , a2, ... , an such that a~ <
a~ ::; ... ::; a~. As in the proof of previous corollary, we can find apermutation
a of {I, 2, 3, ... ,n} such that aj == aa(j) for 1 ::; j ::; n. We define ßj == ßa(j).
Then (ßj) is apermutation of (aj). Using the rearrangement theorem, we get
t
j=l
ßj ( - ~/) Jj=l
t
~,) :s aj ( - ~J
~,) == -no
J
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

inequality.
Solution: Let a1, a2, ... , an be n positive real numbers. Put
(a1a2··· an) 11n

l/n al-a2
al-a2···... ak
G = (a1a2··· an ) ,and ak = .--,/.
Gk for 1 < k
,for < n.
Now we set
= a2,ß2
ß1 = = a3,··· ,ßn-1 = an,ßn == a1·
Now using corollary 4.2, we obtain
n ::; t ßj =
. a)
)=1
I:
aj+1 + a1 .
j=l a·J '-"n /V
However,
aj+1 _ (a1a2·· .aj+1)/GjH aj+1 a1 a1
aj - (a1a2···aj)/Gj C' an G
Thus we get
a1 + a2 + ... + an
n:S G '
which is same as the AM-GM inequality. Here equality holds if and only if
(aj) = (ßj). This is equivalent to a1 = a2 = ... = an which in turn is
equivalent to a1 = a2 = ... = an.
Using corollary 4.2, it may also be obtained that
~ aj G
n <~ - = - + -G + ... + -G + -.
G
- j=l ßj a2 a3 an a1
This implies that
_ n 1---
a1 + -.!...
1
a +_+ 1 ... + 1 -< G -- (a1 a 2··· a )l/n
2 a3 - n .
an
Equality holds if and only if a1 = a2 = ... = an. •

Example 1.18. Prove the Cauchy-Schwarz inequality using the rearrange-
ment inequality.
Solution: Let a1, a2, ... ,,an bb b

an and b1, b2 , ... ,,bn be real numbers. We have to
show that
n 2 n n
(L ajb j ) :S (L aJ) ( L bJ) .
j=l j=l j=l
may be assumed that

°
If either 1::7=1 a] = or 1::7=1 b] = 0, the inequality is immediate. Hence, it
n 1/2 n 1/2
A = (LaJ) (La;) and B = (LbJ)(Lb;)
j=l j=l
are both positive. Define
{ = ::t
{ aj -- ~
a·
A
.
a n +J -- B
an+j B -~
b·
for 1 :S j :S n,
for 1 :S j :S n.
n.
We thus obtain 2n numbers, a1,a2, ... ,an ,an+1, ... ,a2n. Now consider the
permutation
ßj = an+j, ßn+j = aj, 1:S j :S n.
26 Inequalities
Using c:orollary 4.1, we obtain

n n 2n
(Xj(Xn+j La;0:; == 2.
2: (lj(ln+j + L (ln+j(lj ::::; L (Xn+j(Xj ::::;
j=1 j=1 j=1

2:.;=1 ajbj) ::::; 2.
2( AB
Henc:e
2: ab
n n 1/2 n 1/2
j j ::::; AB = (La;) (L b;) .
j=1 j=1 j=1
Here equality holds if and only if O:j
aj == O:n+j,
(Xn+j, for 1 ::::; j ::::; n. This is equivalent
to
al _= a2 _= ...
... == an = A,
= A a constant.
b1 - bb22 -
f;; bbnn '
•
Example 1.19. Let x, y, z be positive real nurnbers. Prove that
L(z + X)2 (x + y)2 (y + z - x)(z - x) 2: 0,
where the surn is taken c:yclic:ally 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
L a2b2(3c - a - b)(c - b) 2: O. (1.14)
We rnay assurne 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-->
>---
cc-..
b - a -
In any c:ase the rearrangernent inequality gives
(s-a)2 b(s-b)2
(S-b)2 (S-C)2 (S-b)2 b(s-c)2
(s-c)2 (s-a)2
a a2 + b b2 + c c2 ::::; a b2 +b c2 +c a2
Sirnplific:ation gives the inequality (1.14). •
Example 1.20. Let al, a2, ... ,an be positive real nurnbers and put
A == al + a2 + ... + an·
Prove that
-al- + -a2- + ... + -an- > - n- .
A - al A - a2 A - an - n - 1
Solution: Note that the sum on the left side is symmetrie in aj 's and hence
we may assurne that al :S a2 :S ... :S an. This implies that
A - al 2: A
A A - a2 2: ... 2: A
A - an,
and hence
111
- - < -A-
A - al - A
A
<"'<
- a2 -
--.
A - an
- A
For any k, consider the permutation of al , a2, ... ,an defined by
ßj =
{a
ßJ. -- { a k +
+jj -- 1
ak+j-l-n
for 1 :S j
for n - k
:S n - k + 1,
+ 2 :S j :S n.
ak+j-l-n
Using the rearrangement inequality, we obtain
-al- + - a2 an
- + ... + - -
A - al
A A - a2
A A - an
A
>-
> ~
- + -~+1
~+ -+
ak+l + ...
... +
+ -~
-
an
A - al
- A A - a2
A A - a n+k-l
A
al ak-l
+ A- an+k
+"'+--.
A - an
This is true for every k. Now summing over k, 2 :S k :S n, we get
( n - l ) ( al
~ - - + -A -a2 + __
>. _ aj ~ n.
a2 an ))
- + ...
... + - -
~ a = ~ A _ aj
an
( A -al al + -- +
a2 A - an >. -
>. - a, >. - a, .;;..
an
(L
(L ) Ln A-a·
n
Ln
(n - 1) 1 )
> --1- l
al = ___J = n.
- L.t A
2: A-a·
- ajJ l#j
l#j
J
j=1
A-aJ
j=1
This implies that

L
n
a-
a'J _
__ n .
> __
j=1
A-a--n-l
A-a'-n-l
J
(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
(al,a2"" ,an) and
is known as Hölder's inequality. Given any two n-tuples (al,a2''''
(bI, b2 , ... ,b n ) of real numbers, the Cauchy-Schwarz inequality states that
n n 1/2 n 1/2
IL ajb I :S (L aJ) (L bJ) ,
j
j=l j=1 j=l
with equality if and only if aj = Abj , 1 :S j :S n, for some constant A. Here

squares and square-roots have special significance in the context of a more
general inequality.
28 Inequalities
Theorem 5. Let (a1,a2,

(a1,a2, ...
... ,an)
,an) and (b(b11,b,b22,, ...
... ,b
,bnn)) be two n-tuples of real
numbers and let p, q be two positive real numbers such that ~ + ~q = 1. (Such a
p
pair of indices is called a pair of conjugate indices.) Then the inequality
tajbjl ~ (t
I tajbjl ( t lajn
lajn l/p
l/p (t Ibjlqf/ q
( t Ibjlqf/ q,, (1.15)
j=l
j=l j=l
j=l j=l
j=l
holds. Moreover this is an equality if and only if lajlP Albjl q
lajlP == Albjl q,, 1 ~ j ~ n, for
some real constant A.
Proof: We first prove an auxiliary result which is useful for the proof of the
ahove 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
xx P yqyq
-+- ~xy.
P q
There are many ways to prove this, hut each one of them depends on some
continuityargument. We can use the generalised AM-GM inequality to get
((x P
~ + ~q ~ ((x P))l/p
l/p (yq) l/ q
(yq) l/ q)) = xy.
xy.
Or one may consider the function
tP r q
f(t) = - +-,
P q
on the interval (0,00). Then f has a unique minimum at t == 1 and hence
tP r q 1 1
-+->-+-=1
P q -p q ,
1/qy-1/ p,
for all t > o. Taking t == X 1/ qy-1/ p, we ohtain the desired inequality. It is also
easy to see that equality holds at the minimum point t = 1 which corresponds
yq.
to x P == yq.
Now taking
lakl Ibkl '
x= n
j=l
,y= n
(2:lajIPf/ P (2:lb j 1/q
j=l
n
we get
lakl P + nIbkl q >>

n
lakllbkl
lakllbkl
p(2: lajlP) q(2: Ibjl q) (t lajlPf/P
( t lajlPf/ P(t
( t Ibjlqf/q'
Ibjlqf/q'
j=l j=l J=l
j=l j=l
j=l
Now summing over k, we obtain

n
2: lakbkl
1 1
- + - > ---:::,-----..........:....------,,----
j=l
P q - (tlajnl/p(tlbjnl/q·
j=l j=l
Thus we have
n n p l/p n q l/q
2: lakbkl ~ (2: lajl) (2: Ibjl) .
j=l j=l j=l
This proves (1.15). Equality holds if and only if
lak
n
lak I
n
Ibkl q
P
2:lajlP
LlajlP 2:lbj l q
j=l
j=l j=l
:S k ~
for 1 ~ :S n. Taking
A= tt lajlP /tt Ibjl
lajlP / q
,
j=l
j=l j=l
j=l
it may seen that equality holds if and only if
lakl P= >.Ibkl q, for 1 ~ k ~ n.

•
Remark 5.1: If we take P = q == 2, Hölder's inequality reduces to the Cauchy-
p =
Schwarz inequality.
Remark 5.2: If either of P

p < 0 and q > 0 (obviously P
versed. For example, if P
negative simultaneously), we take
°
p and q is negative, the inequality (1.15) gets re-
p and q both cannot be
p
P 1
Pl
Pi = - -, ql = -.
q q
Pi
We observe that Pl > 0, P2 > 0 and °
~+~ = _2 + q = q(1- ~) = 1.
Pl ql P P
30 Inequalities
Introducing Uk = lak I~q, Vk = lak Iq Ib k q

k I and applying Hölder inequality to the
collections (Uk), (Vk) with the conjugate indices Pl,Ql,
Pl,ql, we obtain
'n n n
l: lakl~qlaklqlbklq::; (l: lakl~qPl) PI (l: (I ak ll b l)qq,) 01
1 1
k
k=1
k=l k=1
k=l k=1
k=l
= (tk=1 lakn ~ (tk=1 lakbklf·

k=] k=l
RearTanging this, we get
n n 1/ n 1/
l: lakbkl ~ (l: lakn P(l: Ibkn q,
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
(b 1 ,b2, ... ,bn);let (Wl,W2,'" ,wn)
n-tuplesofrealnumbers (al,a2,'" ,an), (b1,b2,
be an n-tuple of positive real numbers. Then the following inequality holds:
n n P l/p n q l/q
l:wklakbkl::; (l:wklakl) (l:wklbkl) ,
k=1 k=1 k=1
q of positive conjugate indices. If either of P and Q
für a pair P, Q q is negative, the
inequality gets reversed.
An inequality related to Hölder's inequality is Minkowski's inequality. This

is a generalisation of the weB known triangle inequality: if a and bare any two
arbitrary real numbers, then la+bl ::; lai + Ibl. It can easily be generalised to n-
(a],a2,". ,an) and (b 11 ,b2 , ... ,bn) aretwon-tuplesofrealnumbers,
tuples: if (al,a2,".
then
n n n
l: laj + bjl::; l: lajl + l: Ibjl·

j=1
j=l j=1
j=l j=1
j=l
Für an n-tuple a = IIal12

(al, a2, ... , an), we define its Euclidean norm IIal12 by
n
IIal12 = (~,aj'2r/2.
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 1/2 n 1/2

(l: laj + bjn ::; (l: lajn + (l: Ibjn '
j=1 j=1 j=1
(a1 ,a2,
for any two n-tuples (al , a2,'"
... ,,an) (b 1,b22 ,, ...
an) and (b],b , bn ). This can be easily
. •• ,b
seen by squaring on both sides. Thus we obtain
Ila + bl1 2 ::; IIal12 + Ilbll:J,

for any two vectors a and b in m n . As in the case of Hölder's inequality, we
jRn.
can replace 22 by any positive real nurnber p, with proper inequality sign.
Letp
Theorem 6. Let 2: 1 be areal number; let (al
p ~ (a1,a2,
,a2,'...
" ,an), (b 1,,bb:J2 ,, ... ,bn)
... ,b n)
be two n-tuples. Then
(2:
n
( t laj + bjl
j=l
j=1
bj n ~ (2: n
P)
11 n
l/pP::; ( t lajn
j=1
11 n
laj 1/pP + ( t Ibjl
j=l
j=1
11
(2:
P) p.
Ibj IPr/ p
· (1.16)
Here equality holds if and only if aj = Abjj for some constant .A,
= .Ab A, 1 ::;
~ j ::;
~ n.
Proof: We rnay assurne pp > 1, because the result for p = 1 is dear.

clear. Observe
that
n n
L laj +
2: bjl PP L laj + bj IP- 1 aj + bj I
2: 11
j=l
j=1 j=1
n n
< L laj + bjbjlP-
2: lP- 11lajl L laj + bbjIP-1Ibjl·
lajl + 2: I 1Ibjl ·
j P-
j=l
j=1 j=l
j=1
Let q be the conjugate index of p. We use Hölder's inequality to each surn on
the right hand side. Thus we have
2::
"'I aj + bj IP- 111
L laj
j=l
J=1
nn
aj I1::; L aj IP)1 P(2::
("'I
~ (:~.:::Iaj
P- 1 ("'1
L laj
nn
j=l
j=]
n
aj + bj I(p-l)qr
11/
l(p-l)q)l p
nn
j=l
j=1
1/
Q
q..
Since p, q are conjugate indices, we get (p -~ 1)q

l)q = p.
p . It follows that
tj=l
j=1
1~ (tlajn1
laj + bj IP- 1 aj I ::;
tlaj+bjIP-1Iajl
j=l
j=1
laj(t n (t
I Ip
j=l
j=1
laj + bj
/p (i=laj+bjn1 /Q
l /Q . n
Sirnilarly,
t
j=l
j=1
bj I ::;
laj + bj IP- 11Ibjl
tlaj+bjIP- 1
j=1
(t n (t
~ (tlbjIP)1
Ib j laj + bj
/P (tlaj+bjn1
IIp
j=l
j=1
/Q
l /Q.. n
It now follows that
n
~ laj + bjl P ::;
{( n
~ lajlP
)I /P + (n~ Ibjl P)l /P }( ~n laj + bjl )l/q P
•
32 Inequalities
If we use 1 - (l/q) = l/p, we finaBy get
n
( L: ) l/p (n ) l/p (n ) l/p
laj + bjl P S L: lajlP + L: Ibjl P .
j=l j=l j=l
Using the conditions for equality in Hölder's inequality, we may obtain the
conditions for equality in Minkowski's inequality: aj
aj = Abj,
Abj , for aB 1 S
:S j S
:S n,
where A is areal constant. •
Remark 6.1: For 0 0, we define
n
Ilall p = (L: lajIPf/ p
·
)=1
Thus Minkowski's inequality can be put in the form
Ila + bll p S Ilall p + Ilbll p ,

for p?: 1 and
Ila + bll p ?: Ilall p + Ilbll p ,
for 0 < P < 1, where a and bare two vectors in jRn .
Example 1.21. Let al, a2, ... , an be real numbers. Prove that
l~k~~313 s (~,ak'3/2r(~~).
Solution: We take p = 3/2 and q = 3. Then p, q are conjugate indices and
using Hölder's inequality we obtain,
It; k~~3 I S (nt; lakl 3/2)

n 2/3 (
t;n k1) 1/3
This gives the required inequality.

•
Example 1.22. Let p, q be two positive real numbers such that !p + !q = 1.
Suppose (al, a2, ... , an) is an n-tuple of real numbers. Show that
n
L:akbk SC, (1.17)
k=l
n
for every n-tuple (bI, b2, ... , bn ) of real numbers with L Ibk Iq = 1 if and only
k=l
if n
(L laklPr/p ~ c. (1.18)
k=l
Solution: Suppose (1.17) holds for all n-tuples (bI, b2, ... , bn ) such that
n
L Ib k Iq = 1. Consider the n-tuple defined by
k=l
lak IP - 1 Sgn(ak)
bk = l/q'
(
2:Z=ll ak IP)
where
Sgn(x) = {f' if ak i= 0,
if ak = 0.
For this n-tuple, we have
I I I
Ln ~n I(P-1)q ~n P
Ibklq = wk=l ak PP = wk=l ak PP = 1.
k=l L:~=1 lak
2:Z=l I
L:~=1 lak
2:Z=l I
Now the condition (1.17) gives
~ 11 ak IP L:~=1 (~I
n n
C ~ L... ak (la kIP - Sgn(ak) ) _ 2:Z=lllakl _ P)l/P
I
1/ - 1/ - L... ak .
k=l L:~=1 lak
( 2:Z=l q L:~=1 lak
( 2:Z=l q k=l
This shows that (1.18) holds.
Conversely, suppose (1.18) holds and (bI, b2 , ••• , bn ) an n-tuple satisfying
n
L:
L Ib I = 1. Then Hölder's inequality gives
k
q
k=l
n n 1/ n 1/
L:akbk ~ (L: lakl P) P (L Ibkn q~ c.
k=l k=l k=l
Thus (1.17) holds.
•
Example 1.23. Consider three sets {a1,a2, ... ,an}, {b 1,b2, ... ,bn }, and
{Cl, C2, ... , cn }, where aj's, bj's, cj's are positive real numbers. If p, q, rare
positive reals such that (llp) + (l/q) + (l/r) = = 1, then
n
n ) (n ) l/q (n ) l/r
La; L Lcj
I/p
l/p
'"'
L... abc<
JJJ-
( b]
j=l j=l j=l j=l
34 Inequalities
Solution: Using Hölder's inequality with exponents p and (p - 1)/p, we get
1/p ( ) (p-l)/p
t ajbjcj:S ( t a~) t(bj Cj)P/(P-1)
J=1 J=1 J=1
Now consider the exponents (p - l)q/p and (p - l)r/p. We observe that
(p
pp
- l)q
(p-l)q
+ (p -pp l)r =
+ (p-l)r
p (~ + ~) = 1
= -p-
_ 1 q
pp-l q +;:r = 1,'
(1 1)
because of (l/p) + (l/q) + (l/r) = 1. Thus Hölder's inequality with these
exponents can be used to get
n ) (p-1)/p
n ) 1/q
l/q (n ) 1/r
l/r
( ~ (bjcj )p/(p-l) < ( Lbj Lcj
Lej
j=1
j=l j=l
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 dass of functions, known as convex and concave func-
tions. Consider the function f (x) == x 2 defined on IR. If we take any two points
on the graph of f (x) = x 2 , then the chord joining these points always lies above
this graph. In fact taking a < b, and the point 'xaAa + (1 - 'x)b
A)b between a and b,
we see that
(,Xa
(Aa + (1- 'x)b)
A)b)22 -- 'xa
Aa2 - A)b2 == -'x(I-
(1- 'x)b A)(a - b)2:s
-A(I- 'x)(a b)2::::: o.
Thus we see that
f(,Xa
f(Aa + (1 - 'x)b) ::::: ,,\f(a) + (1 - 'x)f(b),
A)b) :S A)f(b),
for f (x) = x 2 . This is taken as defining property of a convex function. We
shall see that the dass of convex functions obey a general inequality known as
Jensen's inequality.
Let I be an interval in IR. A function f : I -+ IR is said to be convex if for
all x, y in Iland
and ,XA in the interval [0,1], the inequality
f('xx
f(AX + (1 - ,X)y) ::::: ,,\fex) + (1 - 'x)f(y),
A)Y) :S A)f(y), (1.19)
holds. If the inequality is strict for all x ::J y, we say that f is strictly convex
on I. If the inequality in (1.19) is reversed, Le.,
f('xx
f(AX + (1 - ,X)y)
A)Y) ~ ,Xf(x)
Af(x) + (1- 'x)f(y),
A)f(y),
for an x , y E land A EE [0,1], then we say J is eoneave. If the inequality is

all x,
striet for an all x i-
=I- yy,, we say that J is strietly
strictly eoneave.
eoneave.
There are other equivalent eharaeterisations of a eonvex funetion.function. Suppose
Jf : I ~ ~ is eonvex and Xl, X2, X3 are in I such
-+ IR sueh that Xl < X2 < X3. Take
A = X3 - X2
X3 - Xl '
so that
X2 - Xl
1 - A= , and X2 = AXI + (1 - A)X3.
X3 - Xl
We have
f(X2)
J(X2) = + (1 - A)X3)
f(AXI
J(AXI
::; V(XI) + (1 - A)J(X3)
< Af(xI) A)f(X3)
_ X3 - X2 JJ(xd
- + X2 - Xl
(Xl ) + Xl J (X3 ) .
f(X3).
X3 - Xl X3 -
X3 Xl
We may write this in the form
J(XI) - J(X2) J(X2) - J(X3)

::; ,
Xl - X2 X2 - X3
for an
all Xl < X2 < X3 in I.I . This mayaiso be written in a more symmetrie form:
form :
J(Xd + J(X2) + J(X3) > 0.

(Xl - X2) (Xl - X3) (X2 - Xl) (X2 - X3) (X3 - Xl) (X3 - X2) -
The eonvexity of a funetion
function has an intrinsic
intrinsie geometrie interpretation. If we
function,, this is a subset of the plane and the
look at the graph of a eonvex funetion
line joining any two points on the graph always lies above the graph. Suppose
Zl == (a, J(a))
f(a)) and Z2 == (b, J(b))
f(b)) are two points on the graph of J. f. Then the
equation of the line joining these two points may be written in the form
L(x ) =
L(x) = J(a) =
+ J(b~ - ~(a)
J(a) (x - a).
a) .
Now any point between a and b is of the form X == Aa + (1 - A)b for some
A
A E [0, 1]. Henee
[0,1].
L(x) = L(Aa + (1 - A)b)

J(a)
f(a) f(b) -=~(a)
+ f(b~ J(a) (Aa + (1-
(1 - A)b - a)
b-a
f(a) + (1 - A)
A) (f(b)
(J(b) - f(a))
Af(a) + (1 - >')f(b)
AJ(a) A)J(b)
f(Aa + (1 - A)b)
> J(Aa )"')b) = fex).
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 sub set E of the
plane IR2 is said to be convex if for every pair of points Zl and Z2 in E, the
line joining Zl and Z2 lies entirely in E. With every function f : I --t IR, we
associate a subset of IR2 by
EU) = {(x,y) : a S; x S; b,f(x) S; y}.
Theorem 7. The function : I --t IR is convex if and only if EU) is a convex

subset of IR2 .
= (xl,yd and Z2 == (X2,Y2) be two points

Proof: Suppose f is convex. Let Zl =
of EU). Consider any point on the line joining Zl and Z2. It is of the form
Z = AZI
>'Zl + (1 - >')Z2
A)Z2
(AXI + (1- >')X2,>'Yl
(>'Xl A)X2,AYl + (1- >')Y2)
A)Y2),,
for some >.A E [0,1]. Observe that a S; >'XI

AXI + (1 - >')X2
A)X2 S; b. Moreover,
f(AXI + (1 - >')X2)
f(>'XI A)X2) <
S;Af(Xl) + (1- >')f(X2)
A)f(X2)
<
< >'Yl
AYl + (1 - >')Y2.
A)Y2.
Hence it follows that Z E EU), proving that EU) is convex.

Conversely, suppose EU) is convex. Let Xl, X2 be two points in I and let
Zl = (Xl, f(xd), Z2 = (X2' f(X2»). Then Zl and Z2 are in EU). By convexity
AZI + (1 - A)Z2 also lies in EU) for each A E [0,1]. Thus
of EU), the point >'Zl
(AXI
(>'Xl + (1 - >')X2, Af(xd + (1- >')f(X2»)
A)X2, >'f(xd A)f(X2») E EU),
for all >.A E [0,1]. The definition of EU) shows that
f(>'Xl + (1 - >')X2) S; >'f(xd + (1 - >')f(X2),
for aB >.A E [0,1). This shows that fis convex on the interval I.
•
The convexity of a function implies something ab out the slope of its graph.
The foBowing theorem gives a complete description.
Theorem 8. Let f : I --t IR be a convex function and a EIbe a fixed point.

Define a function P : I \ {a} --t IR by
P(X) = fex) - f(a)

f(a).
x-a
Then P is a non-decreasi ng fu nction on I \ {a}.
Proof: Suppose f is convex on I and let x, y be two points in I, x f:. a, x f:. 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 :S --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) :S P(y). This shows that P(x) is a non-decreasing function for
x f:. a. •
Interestingly, the converse is also true; if P(x) is a non-decreasing function

on I \ {al for every a EI, then f(x) is convex. The proof is not difficult. Fix
x < y in I and let a = = .\x + (1 - .\)y where .\ E (0,1). (The cases .\ == 0 or 1
are obvious.) In this case
P(x) = f(x) - f(a) f(x) - f(a)

x-a (1- .\)(x - y)
P(y) = f(y) - f(a) f(y) - f(a)
y-a .\(y-x) .
The condition P(x) :S P(y) implies that f(a) :S .\f(x) + (1 - .\)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) 2: 0(> 0) for all x E I. (Here f"(x) denotes the second derivative of f
at x.) Similarly f is concave(strictly concave) on I if f"(x) :S 0« 0) for all
x E I.
Here are some examples of convex and concave functions.
1. The function f (x) = x a is concave for 0 < 0: :S 1 and convex for 1 :S 0: <
00 on the interval (0,00). We observe that
f"(x) = 0:(0: _1)x a - 2•
Hence f"(x) 2: 0 on (0, (0) if 0: 2: 1 and f"(x) :S 0 for 0 < 0: :S 1.

38 Inequalities
°
2. Consider f(x) == e X • This is convex on IR == (-00,00). Here f"(x)
e X > for all x in (- 00, 00) and hence e X is strictly convex.
3. The inverse of exponential function, namely the logarithmic function

f(x) = lnx is concave on (0,00). In this case f"(x) = -1/x 2 which
is negative on (0,00). Thus f(x) = lnx is strictly concave on (0,00).
= sinx is concave on (O,n). We have f"(x) == -sinx
4. The function f(x) =
°
and hence f"(x) < on (0, n).
5. If a function f is convex on I, then -fis 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 nu mb er of points.
Theorem 9. (Jensen's inequality) Let f : I -+ IR be a convex function. Suppose

Xl, X2, ... ,Xn are points in I and Al, A2, .. , ,An are real numbers in the interval
[0,1] such that Al + A2 + ... + An == 1. Then
n n
f(2':>jXj):S I>jf(xj). (1.20)
j=l j=l
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 Xl,X2,". ,Xk are k points in land Al,A2,'" ,Ak are real numbers
in [0,1] such that L:;=l Aj == 1, then
kk kk
f(LAjXj) :sLAjf(xj).
j=l j=l
Now consider any n points Xl, X2, ... ,Xn in the interval I and real numbers
Al, A2, ... ,An in the interval [0, 1] such that Al + A2 + ... + An = 1. Let us put
n-l \
",n-l
L: j =1 "jXj
L..j=l
Yz = Xn ,
= Xn, al =
a1 =
n-l
L Aj, a2 =
= An·
Yl = ",n l A'
" , n -1
6j=1
L..j=l J j=l
Observe that a2 = 1 - a1, Yl, Y2 are in I. Using the convexity of f, we get

al, and Y1,
f(alYl
f(a1Y1 + a2Y2) = + (1- adY2)
f(alYl
< ad(Yl) + (1 - adf(Y2)
<
ad(yt} + azf(Y2)'
However, we have
n
alYl + aZY2
a1Y1 azyz = L AjXj.
j=l
j=l
Now consider ff(Yl).

(Y1). If
AI
= --~
MI =
PI ",n-l -:; Il <
, .' 11'::;
" , n - l A.' .::; n - 11
L.. J =1
L..j=l /\JJ -
L~~/ MI = 1. Using the induction hypothesis,

then it can be easily verified that 2:~=~1
we get
n-1
n-l n-1
n- l
ff (2: M1Xl) (2:

.::; 2: Pt!
MiXt) -:; Mt! (Xl)' 2:
1=1
t=l 1=1
1=1
Since
n-l
n-1
Mixi =
2: PIXI Yl,
= Y1, 2:
1=1
we get
f(Y1) :S 2:~~l~~~f(xL) 2:;::11
= L;~ll Ajf(Xj)
Ajf(Xj)
2: j =l Aj Ctl
01
Thus we obtain
f(tAjf(Xj)) < 01 ( 2:;::11 Ajf(Xj)) + Anf(x n )

",n-1 A'
j=l
j=1 L..j=l J
n
L Ajf(xj).
2: Ajf(Xj).
jj=l
= l
This completes the proof by the induction. •

Remark 9.1: If f : I -+ ~ IR
lR is concave, then the inequality (1.20) gets reversed.
If Xl X2 , .. ,. ,X n are points in I and Al, A2, ... ,,An
Xl,, X2, An are real numbers in the
interval [0,1] such that Al + A2 + ... + An = 1, then we have the inequality
f (
n
~AjXj
)
2: ~Ajf(Xj)'
n
(1.21)
Remark 9.2: Using the concavity of f(x) fex) = lnx on (0,00), the AM-GM in-
Xl, X2, ... ,X n are points in (0,00) and Al, A2, ... ,,An
equality may be proved. If Xl, An
are real numbers in the interval [0, 1] such that Al + A2 + ... + An = 1, then we
[0,1]
have
n n
(2:
In (2: AjXj) 2:2:
~ 2:Ajln(xj).
Aj ln(xj).
j=l j=1
j=1
Taking Aj = l/n for all j,

n 1 n n
In (2: ~) 2: ;; 2: lnxj = 2: In (x~/n).
j=l j=1 j=l
40 Inequalities
Using the fact that g(x) = exp(x)(= e X ) is strict1y increasing on the interval
(-00,00), this leads to
1 n n
:;;:LXj L
(x~/n))
ln (x~/n))
> exp ( tIn
j=l j=l
j=l
fr
j=l
j=l
ex
exp
p ( In (x~/n))
1/n
= ((X1
X1X2··· x n ))l/n..
X 2 ... Xn
In fact, the generalised AM-GM inequality mayaiso be proved by this

method. We have
nn nn nn
(L
'\iXj) ~ L Aj
In ( L '\iXj) L
Aj In (Xj)
(Xj) = L In L x;j ,
j=l
j=l j=l
j=l j=l
j=l
for all points X1,X2,

X1,X2, ...
... ,X n in (0,00) and A1,A2,
,X n A1,A2, ... ,An in the interval [0,1]
... ,An
Al + A2
such that Al A2 + ... + An
An = 1. Using the property of exponential function,
we obtain
n n
L '\jXj ~ II x;j .
L '\jXj
j=1
j=l j=l
j=l
Now for any n positive real numbers Cl:

Q:1,1 , Cl:2,
Q:2, ••• Q:n,
.•. , Cl: n, consider
Cl:j
"n
A·J = L...k=l Cl:k
Aj are in [0,1] and

Observe that Aj 2::7=1
2:7=1 Aj
Aj = 1. These choices of Aj
Aj give
Cl:1X1 + Cl:2X2 + ... + Cl:nX n > ( XC>l X:'2 ... xC>n )1/(C>1+C>2+ .. +c>n)
Cl:1 + Cl:2 + ... + Cl: n 1 2 n ,
which is the generalised AM-GM inequality.
Remark 9.3: We use the function fex) = x P

°
P to derive Hölder's inequality.
We know that fex) = x P P is convex for p ~ 1 and concave for < P < 1 on
the interval (0,00). Hence for any n-tuple (Xl, X2, ...
(Xl, X2, ... ,, Xn )) ofreal numbers and
Xn
A1,A2,
A1,A2, ... ,An in [0,1], we have
... ,An
(L
n
( L AjXj
j=1
j=l
AjXj rL n
:S L AjX~ for p ~ 1,
j=1
j=l
and
(L
n
( L AjXj
j=l
j=l
AjXj rL n
~ L AjX~
j=l
j=l
for ° 1 be a glven real number. Let q be the conJugate of p; - + - = 1. We
°
may assurne bj =I- for all j; otherwise we may delete all those bj which are
zero without affecting the inequality. We now set
p q
n
t = L Ibjl
Ibjl q, q,
A =_ Ibkl
Ak
k -
Ibkl qq
tt ''
lakl
Xk = Ibkl q- l .
j=l
We observe that Ak are in [0,1] and Al + A2 + ... + An = 1. Using the convexity

of xP, we get
(
n
~ AjXj
)P :S ~ AjX;.
n
This implies that
~ Ibklq lakl)P ~ Ibkl q lakl P

( L...t 1~ P
-t-Ibklq-l :S L...t -t-lbkl(q-l)P = t L...t lakl ,
k=l k=l k=l
since (q - 1)p = q. Simplification gives
n ( n ) l/p ( n ) l/p ( n ) l/q

{; Ibkakl:S {; lakl P tl-(l/p) = { ; lakl P {; Ibkl q
If °< P < 1, then f(x) = x P is concave and the inequality is reversed:

P
n ( n ) l/p ( n ) l/q
{; Ibkakl ~ {; \akl P ( ; Ibkl q .
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 : II -+ lR is a convex function; al :S :::; a2 :S

:::; ... :S
:::; an and
bl :S
:::; b2 :S
:::; ...
•.• :S n are two sequences of real numbers in I such that al + bl E I
:::; bn
and an + bn E I. Let a~, a~ ... ,a~ be apermutation of al, a2, ... ,an. Then the
inequality
n n n
TIf(aj+bn+l - j ) :SLf(aj+bj
:::;Lf(aj+bj ) :sLf(aj+bj
:::;Lf(aj+bj ).
j=l j=l j=l
holds.
42 Inequalities
Proof: We follow the proof of theorem 4. Suppose (aj) :I (aj) and r be the
largest index such that a~ :I ar; i.e., a~ :I a r but aj = aj für r < j ::; n.
Hence a~ is in the set {al, a2, ... ,ar-I} , ar-I} and a~ < a r . Since aj = aj for
r < j ::; n, we see that (a~,a~ ... ,a~) is apermutation of (al,a2,'" (al,a2, ... ,ar).
Hence we can find k < r, l < r such that a~ = a r and a~ = az. We deduce that
a~ - a;. =
= a r - az 2: 0 and br - bk 2: O. We now interchange a~ and a~ to get a
permutation
permuta · (
t lOn a" "2 , •..
l ,a
(a~, a~, ... ,an") 0off ("
, a~) a l , aa~2 .•.
(a~, , ). th
... ,an'
, a~); thus
us
aj" = aj',for J.-/-r- r,',k a"r = ak, = ar , a~ = a~ = al.
Let us write
n n
5"
S" = L f(aj + b j ), 5'
S' = L f(aj + b j ).
j=1 j=J
We have
s" - S'
5"-5' f(a~ + br
br)) + f(a~ + bk)
bk) - f(a~ + brbr)) - f(a~ + bk bk))
f(a r + brr)) + f(at + bkk)) - f(at + brr)) - f(a r + bkk).
).
We observe that
at + bk < a,. + bk and a/ + br < ar + br .

These give
al bk < aarr + bk
at + bk bk ::; ar + bn
at + bkk ::; br < a r + br.
az + br
::; al br.
If Xl, X2, X3 are in I, then the convexity of f implies that
X2, X3
(X3 - xI)f(X2) ::; (X3 - x2)f(XI) + (X2 - xI)f(X3)'
Taking Xl = at + bk , X2 = a r + bk and X3 = ar + br , we get

(ar + br - at - bk)f(a r + bk) ::; (br - bk)f(al + bk) + (ar - at)f(ar + br ).
Similarly, taking Xl = al + bk , X2 = aZ + br and X3 = a r + br , we get

(ar + br - a/ - bk)f(at + br) ::; (ar - al)f(a/ + bk) + (br - bk)f(ar + br ).
Adding these two, we obtain
(ar + br - az - bk
bk ) {J (ar + bk)
bk) + f (az + br)
br) }
::; (ar + br bk)) {J(at + bk)
a, - bk
br - a/ bk) + f(a r + br)}.
br)}.
a,
Since a/ + bk < aar + br , we deduce that
f(a r + bk) + f(at + br ) ::; f(a/ + bk) + f(a r + br).
This proves that S" - S' 2: o.
.
Now we observe that the permutation (a~, a~, ... , a;;) has the property
a11rr --- a rr an d a11j -- aj, CClor r < J• < ,'d er tlIe ("
_ n. '~(~ may conSI a j ,a112 , . . . ,an"")) In
•
place (a~ , a~, ... , a~) 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 nondecreasing. We finally get
" "
L j(aj + bj ) ::; L j(aj + bj ).
j=l
j=l j=l
j=l
For proving the other inequality, we proceed along the same lines. We define
Cj =
= an+l-j so that Cl 2: C2 2: ... 2: C".
C". We have to show that
n n
Lj(an+1-j +b j )::; Lj(aj +b j ).

j=1
j=1 j=l
j=l
We also set cj == aj. Thus the inequality becomes

n
"
L f(cj + bj ) ::; L j(cj + bj ),
j=l
j=l j=l
j=l
, c~) is apermutation of (Cl,

where (c~, c~, ... ,c~) (Cl, C2,'"
C2, .. ' ,c,,).
, c,,). We assurne that
(cj) i- (Cj)
(Cj) and let r be the smallest index such that c;, i- Cr; Cr; i.e., c~ i- Cr but
cj = 1 ~ j < r. This forces that c~ E {C
= Cj far 1::; {Cr +l,C +2,'" ,cn } and c~ < Cr.
+l,Cr +2"" Cr.
Observe that (c~, C~+l"" ,c~) , c~) is apermutation of (cr, C +1, ... ,, cn). Hence
Cr +1,'"
we can find k > r, I > r such that c~ == Cr and c~ == Cl. Cl. This implies that
C~ - C~ == Cr - ClCl 2: 0 and bk - br 2: O. Now we interchange c~ and c~ to get a
· ("
permu t a t lOn Cl'
Cl' C211 , . .. ,C"" ) 0 f ('.I
, C" , ). thus
Cl , C2 ' ... , C",
C",
-r- r, k , = c~,
11 ,
11
Cj = Cj'
'C
lor J• --I- C~ = cr,
C~, = Cr, ck = c r = Cl·
Now we compute the difference between
L L
" n
S" = L j(c'j + bjj ),), S' = L j(cj + bjj ),),
j=l
j=l j=l
j=l
and obtain
S"-S' j(C~
j(c~ + br ) + j(c~ + bk ) - j(c~ + br) r ) -- j(c~ + bk )
j(Crr + br
j(C r )) + j(CI
j(CI + bk ) - j(CI
j(CI + br ) - j(cj(C r + bk ).
We observe that
~ Cl + bk < Ccr + bk ,
Cl + br ::;
Cl + br
Cl ::;
r ~ C cr + brr < Cr + bk ·
44 Inequalities
The convexity of f gives
(C r + bk - Cl - br)f(CI + bk ) ::; (cr - cl)f(CI + br ) + (bk - br)f(cr + bk ),

and
(Cr + bk - Cl - br)f(cr + br ) ::; (bk - br)f(CI + br ) + (cr - cl)f(cr + bk).

Addition of these two gives
(C rr
(C + bbkk - ClCl -- brr )) {J (Cl k ) + ff (Cr
(Cl + bbk) (C r + br
br )) }}
(C r + bk - ClCl -- br
~ (Cr
::; br )) {J (Cl br )) + f (Cr
(Cl + br (C r + bk)}'
Since Cr + bk - Cl
Cr Cl -- br
r f- 0, we get
f(CI + bk) + f(cr + br ) ::; f(CI + br ) + f(cr + bk ).

This shows that S" ::; ~ S'. We also observe that the new sequence (c'j) has
the property: c~ = Cr and c'j = Cj for 1 ::;~ j < r. We may consider now the
sequence (c'j) 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' = 2:7=1 f (cj + bj ).
We thus get
n n
L f(cj + bj ) ::;~ L f(cj + bj ),
j=l j=l
which is to be proved. •
Example 1.24. Show that for all n ;::: 1, the inequality
rrI1jj;:::
nn
J=l
j=l
) J> (n__
.'
-
+
2
2
1)n(n+l)/2
1)n(n+1)/2
holds.
Solution: Let us put aj == bj == j, aj == n + 1- j for 1 ::; ~ n. Then we see

~ j ::;
that a1 ::;
~ a2 ::; ~ an and b1 ::;
~ ... ::; ~ b2 ::; ~ bn . Moreover a~ ;::: a~ ;::: ... ;::: a~.
~ ... ::;
Consider the function f(x) = xlnx. This is a convex function on (0,00). In
fact we see that
1
j'(x) = lnx + 1, and j"(x) = - > 0 for x> O.
x
Now the rearrangement inequality applied to f(x) gives
n n
L(n + 1 - j + j) ln(n + 1 - j + j) ::;~ LU + j) lnU + j).
j=l j=l
This simplifies to
n
n(n + 1) In(n + 1) ~ 2 L j Inj + n(n + 1) In 2.
j=1
This furt her reduces to
n(n 2+ 1) In
n(n+1)1 n (n + 1) ~
(n+1)<~·1
-2- - tjlnj.= 1
~J nJ =
n
Inn (ftjj).
J. (rr -j)
2 2 j=1 J=1
j=1
Now taking exponentiation both sides we get the desired result. •

The concavity of the log function on (0,00) 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
(1 + xt ~ 1 + ax, if 0< a < 1,
and
(1 + xt 21 + ax if 1 ~ a < 00.
Solution: We use the concavity of f(x) = ln(x) on (0,00). Since x> -1, we
see that 1 + x> O. If 0< a < 1, we have
ln(l+ax) = ln(a(l+x)+l-a)
> aln(l+x)+(I-a)ln(l)
aln(l+x).
The exponentiation gives (1 + xt ~ 1 + ax. Suppose 1 ~ a < 00. Then again
a : 1 + ~ (1 + ax) )
a -1 1 )
In (1 + x)
In(l+x) = In ( -a-+;;:(I+ax)
a-1 1
> --ln(l) + -ln (1 + ax)
a a
1
-ln (1 + ax).
a
This gives (1 + xt 2 1 + ax. •

Example 1.26. Let a == (al,a2,a3, ... ,an ) be an n-tuple ofpositive 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
( L7~1 Wjaj) 1/r

Lj=l Wj
if r =I- 0, and Irl < 00,
M,.(a) = (n)=1 a .j) (l/'L-j~l Wj) ,

n w
j if r = 0,
min {a1,a2,a3, ... ,a n }, if r = -00,
max {aj ,a2,a3, ... ,a n }, if r == 00
Suppose r, s are two real numbers such that r < s. Prove that
M-oo(a) :::; Alr (a) :::; ,;Yfs (a) :::; MCXJ(a).
Solution: \Ve prove this by considering several cases.

(a) Suppose r == O. Then we have s > 0 and hence
(Mo (a))'
(Ir a;j) (S/'L-j~l Wj)
)=1
Cil, (aj)"") (,/,,";~, w,)

L7=1 Wja S
< "n
~j=l Wj
J (by the generalised AM-GM)
(Ms(a)r·
(b) Suppose 0 < r < s. Put Cl! = s / 7" and observe that Cl! > 1. Hence the
function f(x) = xO: is convex on (0,00). Hence Jensen's inequality gives
( 2:.7=1
(
"n
~j=l Wjawjaj")0: < 2:
j ) :::; ~
0:
n
~(aro:)
~ 2:.7=1
(aro:) == " n
~j=l wja
sj
wjaj.
2:.Wk
LWk - )=1
. LWk
~
J=l
2:.Wk J 2:. w k
LWk
::; (Ms(a)f.
This shows that (Mr(a))To: :::; Taking s-th root, we get
::; Ms(a).
Alr (a) :::;
Afr(a)
(c) If s = 0, then r < 0 and we take t = -r. Using the generalised AM-GM
inequality, we have
~ Wj
n _1 ) (l/'L- w k)
II atw
6 a
t.
J
> (
LWk j=1 j J
( Ir~j) (t/'L-wk)
J=l aj
Simplification gives
(
L ~) (l/t)
LWk >
(11 a;'; ) (,/,:,,)
( LWj a j )l/r<
(l/LWk)
rr a7
n )
(
LWk -
j
j=l
Thus we obtain Mr(a) ::; Mo(a).
(d) If r < 0 < 8, then we have

Mr(a) ::; Mo(a) ::; Ms(a)
from (c) and (d).
(e) Suppose r < 8 < o. Then we have 0< -8< -r and hence
M_ s (a- 1 ) ::; M_ r (a- 1 ),
where
a- 1 = (~,~, ... ,~)
al a2 an
Eut then
-1 1 1
M-s(a ) = Ms(a)' M_ r(a- 1 ) = Mr(a)·
Hence we get Mr(a) ::; Ms(a).
(f) Obviously
min {a1,a2,a3, ... ,a n } ::; Mr(a) ::; max {a1,a2,a3, ... ,a n }.
Combining all these, we get the desired inequality.

•
1.8 Inequalities for symmetrie functions
Given n complex numbers a = (al,a2,a3,

(a1,a2,a3, ... ,a n ), we consider the polynomial
whose roots are a1 ,a2 ,a3,· .. ,an;
P(x) == rrTI
n
j=l
(x - aj).
48 Inequalities
Expanding this we obtain
+
p(x) =x n +
P(x) ~(-1)k(~)UkXn-k,
where
L,'j=1 aj
UI = uI(a)
(7)
L,j<k ajak
U2 = u2(a)
(;)
IT'j=1 aj
Un = un(a) =
(~)
These functions UI,U2,U3, .. ',U n are ealled the elementary symmetrie funetions
in the variables al,a2,a3, ... ,a n . We may also write them in the form ur«ak))
to show the dependenee on the sequenee al,a2,a3, ... ,a n . We observe that the
permutation of al,a2,a3, ... ,a n does not alter any of the uj's, 1 :'S j :'S n. We
take uo(a) = 1.
We furt her introduee Vk = u~/k. We observe that VI VI is simply the arithmetie
mean of al,a2,a3, ... ,a n and Vn is their geometrie mean.
Theorem 11. (Newton's inequalities) Let al,a2,a3, ... ,an be a sequence of

non-negative real numbers. We have the inequalities
Vn :'S Vn-I :'S ... :'S VI·
Proof: Let a = min{ aj : 1:'S j :'S n} and b = max{ aj : 1:'S j :'S n}. Then
P(x) = 0 has n roots in [a, b]. Henee the derivative polynomial P'(x) has n-1
zeros in [a, b], eounted aeeording to multiplieity. (If P(x) = 0 has a root a of
multiplicity l, then a 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' 2 - 2UIX+U2).
p(n-2) (x) = --':'(x
2
This has real roots by previous observation. Henee we get
ur.
U2 :'S ur. (1.22)
We also observe that
( n) Un-2 '"' 1
2 ~ = ~ajak'
(( n)n) Un-I
11
Un-l t)=1j=1t~.:.
Un
U n
=
. a)o
J
Applying (1.22) for the numbers I/al, l/a2, ... ,I/an, we get
2
~ (7)' j~'2:-
1 1 1 1
-2:-<-
n
(
G) j<k ajak aj )
This reduces to
U n U n -2 :::; U;'_l· (1.23)
We show that Ur-l Ur+l :::;
Ur-l Ur+! ::; u; ,n -- 1. We use the induction on
u; for r = 2,3, ... ,n
the number of aj 'so For n = 3, we have r = 2 and hence we have to prove that
U3 :::;
Ul U3 ::; u~. This follows from (1.23) with n n == 3. Suppose the result holds for
a set of n - 1 numbers. We have
P'(x) = n {x n- 1 + ~(_I)k (n ~ I)Uk xn - 1- k } .
Let the roots of P' (x) =

= 0 be bll ,, b22 ,, ... n -- ll .• Introduce
..• ,bn
( n-1) Z - "
~
b-b
J1 -J2 ... bJ-.
r
r r - j1 <h< ... <jr
Then we see that
pl(x)
P'(x) bl)(x - b22 )···
n(x - b1)(x )··· (x - bn-d
n{x
n { x nn -- ll + ~(_I)k (n ~ (n 1) ZkXn-1-k } .
ZkXn-l-k } .
It follows that Zr
Zr = Ur ::; r :::;
Ur for 1 :::; n -- 1. Now, the induction hypothesis holds
::; n
for the set {bI, b22 ,, ...
... ,bn-d.
,bn-t}. This gives Zr-lZr+l Zr-lZr+l ::; ::; z;
z; for 2 ::; r ::; n n -- 2.
Using Zr Ur for 1 ::; r ::; n
Zr = Ur n -1, we infer that Ur-l Ur+l ::;
Ur-l Ur+! u; for 2 ::; r :::;
::; u; ::; n n -- 2.
For r = n - 1, this follows from (1.23). This completes the induction. We
observe that U2U2 ::;
::; ur, (UlU3)2 ::;
ur, (UIU3)2 ::; ui, (U2U4)3 ::;
ui, (U2U4)3 ::; ug
ug and so on. Multiplying all
these, we get
U2
U2UU4
4U
1 2 3
U6
6 ...
... u2r-2ur-lur
u2r-2ur-lur
r-l T
<U
r+l -
U2u4
2u
1 2
4 ... u 2r
... u 2r
r •
< U~+l and hence it follows that Vr+l

U~+l ::;
This reduces to U~+! Vr+l < VV rr 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 baekward
induction. One ean also use the known standard inequalities or use ideas from
induetion.
ealculus. In some eases trigonometrie substitutions simplify the result. We
take eaeh of these separately and illustrate them using examples.
2.2 Use of induction

In many problems on inequalities, the Principle 0/
01 Mathematical Induction is a
powerful tool that ean be deployed to prove the result. The teehnique is to use
induetion on the number of variables or some times on the degree (or power).
X1,X2, ... ,X n
°
Example 2.1. Let Xl, X2, ... ,Xn be n real numbers in [0,1] and suppose
Xj
"L.7=1 Xj = m + r, where m is an integer and 0 :::;~ r < 1. Prove that
2:.;=1
n
I: X] :::;m+r.
"~Xj2 ~ m + r 2. 2
j=l
j=l
Solution: induction on n that

We show by induetion
tx;:::; [tXj] + (t Xj - [t Xj ])2, (2.1)

J=l J=l J=1 J=1
= 1, then
where [x] denotes the integral part of x. If n = Xl
Xl lies in the interval
reduces to
[0,1] and (2.1) reduees
[x1]f,
xi ~ [Xl] + (Xl - [xI]f,
xi:::;
whieh is true sinee [Xl] = ° ~ n :::;
= 0 or 1. Suppose (2.1) holds for 1 :::; ~ N - 1. Put
N-l
~l
SN-1 I: Xj == k + t,
Xj 0 ~ t < 1,
0:::;
j=l
j=1
N
SN I: Xj == m +~,
Xj 0:::;
0 ~ rr < 1.
j=l
52 Inequalities
We show that
XJy :S [SN] + (SN - [SN])2 - [SN-I] - (SN-I - [SN_I])2. (2.2)
(m - k + r - t)2 :S m + r 2 - k - t 2.
We can further reduce it to another equivalent statement:
2(r - t)(m - k - t) :S (m - k) - (m - k)2. (2.3)
m=
If m = k, then t :S rand hence (2.3) reduces to -2(r - t)t :S 0, which is true. If
m = k+ 1, then r:S t since X n :S 1 and (2.3) is equivalent to 2(r-t)(1-t) :S O.
This again is true. Thus (2.2) holds for N. Hence
N N-l
N-I
LX; L
LX; +x~
x; +XJy
j=l j=l
< [SN-I] + (SN-I

(SN-l - [SN_d)2 + X~
< [SN] + (SN - [SN])2.
This completes induction.
Here is a generalisation of the previous result(CRUX-1992).
Let a > 0 and let jj : [0, a] -+ lR be a convex function. Let Xl,
Xl, X2, ... ,Xn
be n(~ 2) points in the interval [0, a] such that 2:7=1 Xj = ma + r, where m m
is an integer and 0 :S r < a. Then
n
L j(Xj) :S (n - m - l)j(O) + mj(a) + j(r).
j=l
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 Xl ~ X2 ~ ... ~ Xn > 0 be n real numbers. Prove that
rrfr (l+xj)
n
j=l
2
< 1+XI+XIX2+···+ XI X2··· Xn
(l+xj)
-
2 +
1+Xl+XlX2+···+XlX2···Xn.
nn+11
Solution: We use induction on n. If n = 1, the result is immediate. Suppose

it holds for all m :S n - 1. Take any n real numbers such that Xl ~ X2 ~ ... ~
Xn > O. We apply the induction hypothesis to the sequence X2 ~ X3 ~ ... '" ~
X n > 0 and obtain
frrr (l~Xj):s< 1+x2+x2x3+···+x2x3···xn

(l~Xj)1+X2+X2X3+~·'+X2X3"'Xn.
j=2
Techniques 53
Multiplying both the sides by (1 + Xl) /2 , this takes the form
fI (1 ~ Xj) ~ C~ Xl) C: R),

J=l
where
R = X2 + X2X3 + ... + X2X3 ... Xn ·
Thus the inequality to be proved is
(1 + xJ)(l + R) <
-'----'-'-----'- < 1 + Xl + xlR .
2n - n+ 1 .
R((n + 1) - (n - l)XI) ~ (n - 1)(1 + xd.

If n + 1 ~ (n -l)XI, 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 - l)XI > O. Thus it is sufficient to deduce that
R< (n-1)(1+xd
- (n + 1) - (n - l)XI
But, observe that
R = X2 + X2 X3 + ... + X2 X3 ... Xn <
_ Xl + Xl + ... + Xl
2 n-l
.
Therefore it is enough to show that
(Xl + xi + ... + x~-l )((n + 1) - (n - l)XI) ~ (n - 1)(1 + xd.

Equivalently, we need to prove that
(n - l)x~ - 2(XI + xi + ... + X~-l) + (n - 1) 2: O.

Consider the polynomial
P(X) = (n - l)x n - 2(x + x 2 + ... + x n - l ) + (n - 1).
Obviously P(l) = O. We also have
pl(x) = n(n - 1)x n - 2 - 2(1 + 2x + 3x 2 + ... + (n _1)x n - 2).
This shows that P I (l) = O. 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 x n P(l/x) = P(x). It
follows that P(x) > 0 for 0< X < 1. Thus P(x) 2: 0 for x> O. This completes
the proof. •
54 Inequalities
2.3 Application of known inequalities

We have already built up a library of so me 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, ce are positive real numbers such that a+b+c = 1, prove
that
8
8abe ~ (1 - a)(l - b)(l - c)
8abc e) ~ 27'
Solution: Since a, b, ce are positive and a + b + ce = 1, none of these can exceed
r
1. Thus each of 1 - a, 1 - b, 1 - ce is positive. Using the AM-GM inequality, we
obtain
(l-a)(l-b)(l-c) ( ~
3-a-b-C)3
(1 - a)(l - b)(l - c) ~ (3-a;b- 3 C 8
27'
which is the second part of the required inequality. On the other hand, we also
see that
(1 - a)(l - b)(l - c) 1 - (a + b + c)
e) + (ab + bc
be + ca) - abc
ab + bc + ca - abc.
Thus the first part of the inequality is equivalent to
9abe ~ ab + bc + ca.
9abc
We may write this last inequality in the form

1 1 1
-a+ -b +c-e >
a b --
9.
This is an immediate consequence oft he AM-RM inequality, since a+b+c = 1.

More generally, it is not hard to see that for any n positive real numbers
al, a2, ... ,an, the inequality
( al + a2 + ... + an ) ( -1 + -1 + ... + -1) 2: n 2 ,

al 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
abc<
b pa + qb
pa qb + rc
rc
ac< - .
88
Techniques 55
Solution: We make use of the generalised AM-GM inequality: if a, (x, ß, "( are
nonnegative real numbers such that a(X + ß + "( = 1, then for positive u, v, w,
we have
au + ßv + "(W 2:: uD: 1iw'.
Thus we obtain
.~+q.~+r.~ 2::~ (~)P(~)q(~)r
bc ca ab bc ca ab
and
(bc) P (ca) q ( ab r ~ p . bc + q . ca + r . ab.
These two lead to
abc 1
pa + qb + rc 1 1
p·-+q·-+r·-
1
bc ca ab
1
<
cr
(b1 c~r (:bf
(bct (ca) q (ab r
p' bc + q . ca + r . ab.
< p.
:::; c. Then
:::; b ~
Assurne a ~
p' bc + q. ca + r . ab
p. <
~ pbc + (q + r)ac
pbc + (1 - p)ac.
:::; p ~
Since 0 ~ :::; 1/2, we get
G- p) ac ~ G- p) bc.
This implies that
pbc+ (1- p)ac ~bc+ ~ac- G- p)bc+ (~- p)ac

1 1
< -bc+ -ac
2 2
1
2c(a + b).
But we know that

(
a + b) cc:::;
~ ( a+b+C)21
2 = 4·
4'
Combining all these, we end up with
abc 1 1 1
---,----
---,------ < - . - = -.
pa + qb + rc - 2 4 8
= b == 1/4, c == 1/2, p == q == 1/2, r == 0 gives equality.
We also observe that a =
•
56 Inequalities
Example 2.5. For nn 2': 2, and real numbers 0 < Xl :s: 1,

< X2 < ... < XXnn <
prove that
""n n
n
n L..Jk=l Xk > L ---.
2': '"'
L..t -
1
--.
""
2:k=l
L..Jk=l Xk +
Xk+nxlx2···X
nX1X2 ... Xn -
n k=l k=l +
1+Xk
1 Xk
:s: kk :S:s: n, then we know that

Solution: If ak > 0 for 1 :S
t a k ) ( t a ; l ) 2':n 22,
(tak)(ta;l)
(
k=l k=l
as an application of the AM-HM inequality. This implies that
( t a k ) ( t a ; l ( l - a k )) ( t ak) ( t a;l) - nn t aka;l) -

k=l k=l k=l k=l k=l
>
> n(n- t a k )
t
k=l
= n( (1 - ak) ) .
k=l
Taking
Xk
a =
ak - - - for 1 <
- -1--' <_ k <
<_ n,
n,
+ Xk'
k - 1 +Xk --
we obtain
nn (
x- 1l n -
' " x-
nn
'" ---
1) > nn'"n -+-1- . n 1
~ k
L
k=l
k ~ 1 + Xk
L
k=l Xk -- ~ + Xk
L
k=l
1 Xk
Xnn and
Multiplying by X1X2 ... X using the inequality
n n
L Xk 2': X1X1X2X2 ... XXnn LL X;l,
X;l,
k=l k=l
we get
nn nn 1 nn 1
((LXk)(n- L~)
LXk) (n - L -)
1 + Xk
2'2':nX1X2···XnL~·
: nX1X2·· ·Xn L - ·
1 + Xk
k=l k=l + k k=l + k
This implies that
nn nn nn 1
nn L
L Xk 2': (
(LL Xk + nX1X2 ... xxn) L 1 + Xk .
n) L
k=l k=l k=l
This is equivalent to the given inequality. •
al, a2, ... ,an be positive real numbers. Prove that
Example 2.6. Let a1,
aa22 aa22 aa22 1
_-,,-1_+
_--,,1,---+ 2 + ... + nn >-(al+a2+
>-(a1+ + aan).
a 2+ ... + n).
al + a2 a2 + a3
a1 an + a1al -- 2
Techniques 57
Solution: We put
aj _ aj
aj = and bj - - - ,
a1 + a2 + ... + an aj + aj+1
where an+l
an+! = a1. Then L: J=1 aj
2::7=1 O!j =
= 1. Now the weighted AM-HM inequality
gives
n n
Lajbj Laj
=-j=_1_ _ > _j=_1_
Tl. - n .
Laj L ~~
j=1 j=1 J
We observe that
( taj) ( t
J=1 J=1
~J) = t
J J=1
(aj + aj+I) = 2(t a
J=1
j),
n
so that L ~J~j
L
j=l
j=1 J
=
= 2. Since L: = 1, we get L:
J=1 aj =
2::7=10!j J=1 ajbj >2 1/2.
2::7=10!jbj Thus we
obtain
n a2
L
j=1
aj +a
J
J
= (taj) (tajb j )
J=1 J= 1
~ ~(taj),
J=1
which is the desired inequality.

•
a,b,c
Example 2.7. If a, b, c are positive real numbers, show that
abc( a + b + c + Ja 2 + b2 + c2 ) 3 + J3
~~-----:--c;----~-'- < --.
(a +b +c ){ab+bc+ca)
2 2 2 - 9
When does equality hold?
Solution: (a,, b, c) and (1 , 1,

The Cauchy-Schwarz inequality applied to (a 1,1)
1)
gives
a + b + c :::; v'3 .;a 2 + b2 + c2 .

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
abc( a + b + c + Ja 2 + b2 + c2 )
1
--,---'----,:-------::-,----,---------;--'-
abc(1+,;3) (a 2 + b2 + c2 ) 2"
~~--~~ 7--------~ <
(a 2 + b2 + c2 ) (ab + bc + ca) (a 2 + b2 + c2 ) (ab + bc + ca)
abc(1 +,;3)
1
(ab + bc + ca) (a 2 + 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 = l.
Prove that
1 113
-----+
- - - + 33
a 3 (b + c) b (c + a)
+ 33 >-.
c (a + b) - 2
Solution: Taking x == l/a, y = l/b, z = l/c, we have xyz = 1 and the

inequality to be proved is
x2 y2 Z2 3
- - +z+x
y+z
- - +x+y
- - >-.
- 2
Using the Cauchy-Schwarz inequality, we have
(x+y+z)2::;2 ( -X2- + -y2- + -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
~ 2 or x :s:
holds if x 2 ::::: -1. For each x, -1 < x < 2, we can find positive real
b, c such that abc = 1 and (2.4)is not true.
numbers a, b,
For any real x, define
1 1 1
fex; a, b, c) = ax(b + c) + ba;(c + a) + cX(a + ..
We show that fex; f(x; a, b, c) is a non-decreasing function far for x 2 ~ 1. Let x x>> y 2~ 1
be two reals. Suppose a ::; b :s: ::::: c. Then a X- Y :s: ::::: bX- Y :s:
::::: cX- Y is true. We also
observe that aY(b + c) :s: ::::: bY(c + a) :s: ::::: cY(a + b). In fact, aY(b + c) :s: ::::: bY(c + a) is
equivalent to ab(a Y- 1 - by- 1 ) :s: ::::: c(bY - aY) which is true because y -1 - 1 20
~ 0 and
a::::: b. Similarly, bY(c + a) :s:
a:S: ::::: cY(a + b) holds. Using Chebyshev's inequality, inequality, we
obtain
111
- --+
-,-----+
aX(b + c) bX(c + a)
+ ---
+---.,-----:-
cX(a + b)
1
1 1 1 1 1
------,---,-
- - - - - + bX ----:----,-
+- ---
aX - YaY(b + c)
aX- - YbY(c + a)
bX-YbY(c cX-YcY(a + b)
> ~ (_1_ + _1_ + _1_) ( 1 +_ 1 + 1 )
3 a XX-- Y bx-y
bx - y CcXX-- Y aY(b + c)
aY(b+c) bY(c + a)
bY(c+a) cY(a + b)
cY(a+b)
111 11 1
>
> + +-- b)"
aY(b + c) + bY(c + a) + cY(a + b)·
We have used the AM-GM inequality in the last leg. We get fex; f(x; a, b, c) 2
~
f(y; a, b, c) if x > y 2
~ 1. Thus fex;
f(x; a, b, c) is a non-decreasing function on the
interval [1, (0).
00).
Now consider f(2; a, b,b, c). We write
111
f(2 ;a, b, c)
1(2; -,,-,---+
--::-,------:-+ +...."...,--
+-:-:-----:-
b2 (c + a)
a 2 (b + c) c2 (a + b)
bc
- ca
-:--- + - -- -+- - ab
-- -
ab + ac bc + ab bc + ca
=
= (ab + bc + ca) ( -11-b-
(-b- bb + -b_1_ + 1 b) - 3
a + c c+ca ca+a
9 3
> -2 -- 3>
3 > -.
- 2
It follows that fex; a, b, c) ~ 3/2 tür
f(x; a, for x ~ 2.
We now show that the result is not true for 1/2 :s: ::::: x < 2. Take x = 2 - e
where e > O. Put a = d 2 ,b , b = l/d,c
l/d, c = l/d so that abc = 1. Then
1 2d 3 - e
f(2 - e; a, b, c) = 2d3-2e + d3 + 1
00, if e < 3/2. If e = 3/2 then 1(2
and f(2 - e; a, b, c) -+ 0 as d -+ 00, f(2 - e; a, b, c) -+
1/2 < 3/2. In any case (2.4) is not true. Thus (2.4) fails for für 1/2 :s:
::::: x < 2.
60 Inequalities
For x in the interval (-00,1/2), we replace a, b, c by l/p, l/q, l/r respec-

tively. Then j(x; a, b, c) = j(l-x;p, q, r). But I-x lies in the interval (1/2,00).
It follows that (2.4) holds for x :S -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,'"

Xl,X2,'" ,Xnn be positive
numbers such that
n 1
' -1 - 1
= 1.
L
n
"
~1+x·-·
l+xj
j=1
j=l J
Prove that
n 1
~
~
Fi'2 (n-l)L VX;"
j=l ]=1
Solution: We may take XlXl :S X2 < ... <

:S Xnn .. We claim that under this
condition we have the inequality:
-Fi
VXi>.,fik
->- .,fik
-
1 + Xj - 1 + Xk'
whenever 1 :S j :S kk :S n. In fact, we have

1 1 2 + Xj + Xk
1>--+--= ,
l+xj l+xk (l+xj)(I+xk)
from the given condition. This gives
1 + Xj + Xk + XjXk > 2 + Xj + Xk.

Hence XjXk > 1. But we also have
VXi
VXi _
_ .,fik _ (VXi - .,fik)(I- JXjXk)
(VXi-.,fik)(I-Jxj x k) > 0
-'-'----=---'-----'-'----'----"---'- 2 0,
+ Xj 1 ++ Xk
1+ + xj)(1 ++ Xk)
(1 + -,
since VXi:S .,fik and 1 :S JXjXk' Thus we obtain

y'Xl
y'Xl >- yIx2
yIx2 ... >> -..;x:;.
> ... yXn
- - > - > -.
+ Xl
1+ + X2 -
Xl -- 1 + - 1 + Xnn
Since we also know that
1 1 1
-->-->
- > - > ... >--
>--
y'Xl - yIx2 - --..;x:;'
yXn'
by the assumed ordering of Xj 's, we may apply Chebyshev's inequality. We get
-l(n
:l(n
; , j=l 1)(n
1 ) ( n VXi)
L rx;
j=l Y~J
L
j=l
j=l
VXi)
1+
nn 1
+ x :sL 1 ++ x .=1
j=l
JJ j=l JJ
(2.5)
Techniques 61
However using the AM-HM inequality, we obtain
(t 1 (t 1~') ~
j=l,;x;
+XXj)
j=l J
n2.
We may write this in the form
(t, l~J (t,ViJ) (t, l~J (t, ~) ~ n + 2

.
Now using (2.5), we obtain
n~ n+ (t, l~J (t,ViJ).

2
This reduces to
(t
j=l
J=l
1~') (t ViJ)
,;x;.)
1 + x JJ
2n
y'xj) ~ n
j=l
J=l
2 - n
n = n(n
n(n -1).
= (2.6)
However (2.5) mayaIso written in the form
~ x ; < __
n , ;y'xj nn__
-J~n 1
L-
'"'
~ 1 +Xj ' " ' _1
L..1+x·- n
j=l
j=l J ~,;x;
j=l y'xj
Using this in (2.6), we get
n(n -1) ~ (tViJ) (n/t ~).

This gives
tViJ ~ (n
n 1
j=l
-1) ~ ,;x;'
Alternate Solution:
Using
n
11 -1
L
n
+x' -= 1,,
1 +Xj
j=l J
we easily obtain
L --J-=n-l.
~~=n-l.
n
~ l+xj
1 +x·
x·
j=l J
62 Inequalities
Hence
n nn 1
Lftj - L
11 ~
~ rx-: - (n
~Y-J
rx:
- 1) ~. ~
(n-1)
j=1
j=l
j=1
,;x;
j=1 y-J
(~1 :xJ (~.;x;) - (~1 :JxJ (~~)

LL
nn. nn
Xj-Xk
LL
. k=1 (1
)=1
J=l k=1
(1~~~~
+ Xk). v
I"-"J
J
L (y'Xj
(,;x;yixk ,;x; -- yixk)
yIxk - 1) (y'Xj yIxk) 2( (,;x;
y'Xj + y'Xk)
JXk)
j>k
J>k ,;x;yIxk(l + xXj)(l
y'Xjy'Xk(1 J )(l + Xk) .
Here we have used

1 1 ((,;x; 1) (,;x;
yIxk - 1)(
y'Xj y'Xk y'Xj -- v'Xk)
JXk)
(1 + Xk)y'Xj + xjhlxk
Xk),;x; - (1 + Xj)>/Xk ,;x;JXk(l + xj)(l
y'Xjy'Xk(1 + Xk) .
+
Thus it is sufficient to prove that :rjXk 2 1 for any pair j, k such that j i:-
XjXk 2:: -::j:. k.
But this follows from the given condition:
1 > _1_ + _1_ 2 + Xj + Xk
= __2_+_x...:f.j_+_X.:.:.-k
- 1 + Xj + :rk
1+ Xk 1 + Xj + Xk + XjXk '
2 1.
which implies that XjXk 2:: •
ExaIllple 2.10. (Hardy's inequality) Let 0,1,
Example al, a2,
a2, ... ,aN be N positive real
... ,aN
numbers and let p > 1. Put
al
0,1 a2 + ... + ak
+ 0,2
O:k
ak = for 1 ::; ::; N.
:::; kk:::;
k
Prove that
N N
~ p
p
~ O:k:::;
a k ::;
(p
(P )P~
p _ 1
P
N
) ~ p
~a
P
k•
k=1
k=1 k=1
k=!
Solution: Let ih
Ak = al
0,1 + a2 + ... + ak, We have
P ~1 ~1 (A n
- - -PO n-l )
n -- A. n~l
p p-l p - --
an --
(ln --1
- - a(ln
n an an
O:n a:n
np-l
p-
p-1 p-1
a~n - P_P_o:
o:p ~a~~l p- l (no: - (n - 1)0:
_ 1 n (na nn - l)an~d
n-
1)
p-1
an
p
O:n
p((11 -- -np)
np )
- +
(n - l)p p-l
(n-1)p
an
~l
O:n an~l'
O:n-l·
p-1 p-1
p-1
Using the generalised AM-GM inequality, we have
(n-1)PaP~la
(n-1)p
---(ln
p- l ~1 < -
(ln-I:::; n-1{(
n -1{
- P_1)a~+a~~I}'
() pp}
- 1 (ln + (ln-l .
p-1
p-1 n n -p-1p-1
Techniques 63
Thus we get
P ~np -I)}} n 1 aPP
n --1a
1 + (n -1) + -
P p-1
an -- -
an - l ann an
--lO: an < aaP{l-
P { 1- -
p _- p _- 1
p- n n-1
_l_{(n -- l)a
_l_{(n
p _ 1 n-1
no:p}
l)o:PP - naP}
n'
Taking ao
0:0 = °and summing over n from 1 to N, we get
N
LN aP
o:P -- n -
p
P_
-1
N
~a
_-1L..." o:P-
n
L
P- 1 a <
n_
NT PP
N
-- -1-
aN <
- 0..
°
n=l P - n=l
p P-
p - 1
Thus we obtain,
N N
LO:~ < -P_~aP-1a

La~ _P- '"' aP-1a
1L.." n
p-1L..." n
n=l p- n=l n=l
N 1/ N
< P ~ 1 (L a~) l/pP(L a~-l)qr/q,
N N
a~-l)q)l/q,
p n=l n=1
n=l
where q is the conjugate index of p; we have used Hölder's inequality. Henc:e

(p - l)q = p and we obtain
N N l/p N l/q
L a~::; p ~ 1 (L a~ )
n=l n=l
( L a~ )
n=1
.
This simplifies to
N
L af ::; (p ~ 1 L af·
k=l
r N
k=l
•
Example 2.11. Let a1, a2, ... ,an be n positive real numbers. Prove that
rra~j
n
2 (rr n
a o
)("L-i=l aj)/n
J=l j=l J
Solution: Consider the function f(x) = xlnx. This is c:onvex on (0,00). In

fact f"(x) = I/x which is positive. Using Jensen's inequaIity,
inequality, we obtain
Ln 1
-;;aj In Uj 2
(nI)
L (nI'
In L n aj -;;a j ) .
J=l )=1 J=l
t.
However, the AM-GM inequality gives
(n) l/n
~ ~ajaj :>2 OJ,
]1 a,)""
1 n
-;; aj
64 Inequalities
Using the monotonicity of In x, we obtain
1
-;-n ""a·lna
~L aj lna
n
.
)=1
J
> ( LnI-;aj ) In (n aj ) l/n
>
)J --
)=1
rr
)=1
This implies that
L lna;; 2: (~taj) In (rr aj ).

n
j=1
j=l )=1 )=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 al,a1, a2, ... ,an be n positive real numbers. Show that for
any permutation a~, a~, ... ,a~ of the sequence al,
a1, a2, ... ,an, the inequality
rr a)a;; >~_ IIrr a),

n
II
j=1
j=l
a;
nn ,
a;; ,
j=1
j=l
aj
holds.
a1 2:
Solution: We may assume that al ~ a2 2:
~ ... 2:
~ an. Then we have
lna1 2: Ina2 2: ... 2: Inan.
We apply the rearrangement inequality to the sequences al,

a1, a2, ... ,an and
In al, ,ln an. We obtain
a1, In a2, ... ,In
n n
L aj In aj 2:~ L aj In aj.
j==1
j==l j=l
This implies
n n,
L lna;; 2:~ L lna;;.
j=l j=l
The exponentiation gives the result;
rr a;' rr a/.
n
j=l
a.
aJu; 2:~
n
j=l
a'.
u'.
•
Techniques 65
2.4 Use of calculus

calcu/us in inequalities
inequa/ities
direetly from the monotonicity of the function involved
Many inequalities follow directly
ealculating its global maxima and minima. For a differentiable function,
or by calculating funetion ,
ean be established using the nature of its derivatives. A point
the monotonicity can
loeal minima or maxima for a differentiable function is precisely that point
of local
vanishes. We state these resuIts
at which its derivative vanishes. results here
he re without giving
proofs; the proofs need the tecchniques
proofs; tecehniques of differential calculus.
f'(x)
fl(X) ~ °°
Theorem 12. Let f : (a, b) --7
2 0 for all x E (a
--+ IR
OC be a differentiable function and suppose
, b). Then fex) is a non-decreasing function on (a,
(a,b). b).
(a,b).
°°
f'(x) >> 0 on (a, b), then fex) is strictly increasing on (a, b)
If 1'(x) b)..
Similarly, 1'(x) :S 0 for all x E (a,
f'(x) ::::; (a , b) implies that fex) is non-increasing on
(a, b) and l' (x) < 0 for all x E (a, b) implies that fex)
f'(x) f(x) is strictly decreasing on
(a, b).
Theorem 13. Suppose f : (a , b) --7 OC be a differentiable function and A

--+ IR A. is
a point where a local extremum occurs for f; i.e. i.e.,, there exists a neighborhood
(A. - J,
(A A. + J)
8, A 8) of AA. such that either fex) ::::; :S f(A)
f(A.) for all x E (A (A. - J, A. + J)
8, A 8) or
fex) ~ 2 f(A)
f(A.) for all x E (A (A. - J,
8, A A. + J) f'(A) =
8).. Then 1'(A.) 0. Suppose further f is
= O.
twice differentiable and 1"(A.) f. O.
f"(A) -::j:. f"(A) < 0, f(A)
0. If 1"(A.) f(A.) is a local maximum; and
f"(A) >
if 1"(A.) > 0, then f(A)
f(A.) is a local minimum. If 1"(A.) f"(A) = 0, then f(A) f(A.) is neither a
maximum nor a minimum(a point of inftexion) infiexion}..
For proofs of these statements, refer to any book on Calculus. We consider

examples which make use of these theorems in their solutions.
x > 0,
Example 2.13. Show that for any x> 0, the inequality
{~o,
20, ifiX> 1 üriX < 0,
ifa>lora<O,
°
1
a
x a -- ax + iX - 1
iXX +a- { (2.7)
:s:S 0, if 0 < 0:
iX < 1.
Solution: We consider the function fex) fe x ) = x a - iXX + 0:

o:x iX -- 1 on the interval
(0,00). This is a differentiable function and
j'(x) =
f'(x) iXx a - 1l
= o:x -- a
0: = iX(X a - 1l - 1).
= a(x
This shows that f'

a(a
0:( 11)x
0: --
°
(x) = 0 only if x = 1 (Note that 0:
j'(x) a -::j:.
°
f. 1). Moreover 1" (x) =
1"(x)
)x a - 2 which is positive at x = 1 if either 0:a < 0 or 0:a >> 1, and negative
at x == 1 if 0 < 0:a < 1. Hence x = 1 is a unique point of minimum if either 0: a <0
a >
or 0: > 1, and a unique point of maximum if 0 < 0:a < 1. Since f(l) = 0, we
2 f(l) = 0 if either 0:a < 0 or 0:a >> 1, and fex) ::::;
°
conclude that fex) ~
if 0 < 0:
a < 1.1. This proves the result.
result.
:S f(l) = 0
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 Xl and X2 are two positive real numbers
66 Inequalities
and let x = xIi X2.

X2. Take any a
O! with 0 < aO! < 1. Then we have
((~:Xl)
X2
a (Xl)
(~:)
a _
) a _ O!
X2
+ a _11 :S~ O.
+ O. O! _
It may be rewritten as
xfx~-a :S aXl + (1- a)x2'

This proves the generalised AM-GM inequality for non-negative real num-
Xl,
bers Xl, X2 a, a.
X2 and weights O!, 1 - O!. Now we use the principle of induction to
prove the general case. Assume that for any n - 1 non-negative real numbers
A1,A2,A3,.' .,An-1 with 2:.;1~11 Aj = 1, we
Y1,Y2,Y3, .. ·,Yn-1 and positive weights Al,A2,A3,.'
Yl,Y2,Y3,
have
n-1 n-1
II y;i :S~ L
L AjYj·
j=l j=l
j=1
Take any n nonnegative numbers Xl,X2,X:3," .,X n , and positive real numbers
a1,a2,a3,
0!1,0!2,0!3, ... ,an such that 2:.7=1 aj
.. ·,O!n O!j = 1. Put
Aj=O!j) for1:Sj:Sn-2,
Yj=Xj, Aj=aj) for1~j~n-2,
Yn- 1 -- xan_I/An-l
n-l x na n JA n-l, An -1 = a n-1 + an·
We observe that Aj > 0 for 1 :S~ j :S
::; n - 1 and
n-1
71-1 n
L Aj = L aj
O!j = 1.
j=l j=l
Using the induction hypothesis, we obtain

n n-l
II X;i II
II Y/"A
j=l j=l
n-1
71-1
< L AjYj
j=l
n-2
L
L ajxj
O!jXj + (an-l
(O!n-l + an) (X~~l';"n-'x~n/"n-')
O!n) (X~~lI/An-' x~njAn-')
)=1
j=l
< ~ O!jXj + ( aO!n-1

L..t ajXj O!n ))(an-1
n -1 + an (O!n-1 a n)
O!n)
-A-- Xn - 1 + -A--Xn
j=l n-1
71-1 n-l
n-1
n
"
LajXj,
LO!jXj,
j=l
since An-1 == aO!n-1

A"-1 n-1 + an.
O!n. 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. We take a = 11p and x = alb. We get (%)I/P - ~ (%) + ~ -12: 0 if p <1
and :::; 0 if p > 1. This reduces to
a
a1/Pb 1/ q b
-+- > if p < 1,
P q
a
a1/pb 1/ q < -+- b
if p > 1.
P q
N ow 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 e X > 1 + x + (x2/2) for x > 0, and eX < 1+ x +
(x 2 /2) far x < O.
Solution: Let fex) = e X - 1 - x - (x 2 /2). We have

f'(x) = e X - 1 - x, fl/(x) = e X - 1.
Hence 1"(x) > 0 for x > 0 and 1"(x) < 0 for x < O. This implies that
1'(x) is increasing on (0,00) and decreasing on (-00,0). We thus infer that
f'(x) > 1'(0) = 0 for all real x i:- O. It follows that fex) is increasing on
(-00,0) and (0,00). This implies that fex) < f(O) far x < 0, and fex) > f(O)
far x> O. Since f(O) = 0, it follows that fex) > 0 for x> 0, and fex) < 0 for
x < 0, which is the desired inequality. •
Example 2.15. Prove Jordan's inequality that

2 sinx
-:::;--:::;1,
7f x
for all x E [0, 7f 12].
Solution: Consider fex) = = sinx. Since 1'(x) == cosx and 1"(x) == -sinx,
we see that 1" (x) :::; 0 on [0, 7f 12]. Hence fex) is a concave function on [0, 7f 12].
Take any x E [0,7f/2] and set A = 2xl7f. Using the concavity of fex) = sinx,
we have
sin(x) = sin((1-A)0+h/2)
> (l-A)sin(O) + Asin (7f/2)
A = 2x.
7f
68 Inequalities
Thus we obtain
sinx 2
-->-.
> -.
x - 7r
°
On the other hand, consider g(x) = sinx - x on [0,7r/2]. We again note that
g'(x) = cosx -1 :S for all xE [0,7r/2]. Hence 9 is a non-increasing function
on [0, 7r /2]. This implies that g(x) :S g(O) = 0. Thus we get, sin x :S x for all
xE [0,7r/2]. Using the known result that limx-tosinx/x = 1, we conclude that
sinx/x:S 1 for all x E [0,7r/2]. Thus we have Jordan's inequality,
2 sin x 1
- <
Ir
- --< -,X
valid for all x E [0, Ir /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)" > 1 + ax, if a > 1 or a < 0,
(1 + x)" < 1 + ax, if 0< a < 1.
Solution: Consider the function f(x) = (1 + x)" - 1 - ax for x > -1 and

a =I- 1. Then we see that
j'(x) =a{(1+x)a-l-1}.
Suppose a > 1. Since x > -1, we see that (1 + X),,-l > 1 if x > 0, and
°
(1 + X),,-l < 1 for -1 < x < 0. Thus f'(x) > if x > 0, and f'(x) < if
-1< x < 0. This shows that f(x) is decreasing on (-1,0), and increasing on
°
(0,00). Hence f(x) > f(O) for all x > -1, x =I- 0. We obtain
(1 + x)" > 1 + ax if x> -1, x =I- 0, a > 1.

°
If a < 0, then 1 - a > 1, and hence (1 + X),,-l - 1 < for 1 + x> 1; and
° ° °
(1 + X),,-l - 1 > for < 1 + x < 1. Thus f'(x) > for 1 + x > 1, and °
f'(x) < for 0< 1 +x < 1. Hence f(x) is decreasing in (-1,0), and increasing
° °
on (0,00). We see that f(x) > for x> -1, and x =I- in this case. This gives
(1 + x)" > 1 + ax, if x> -1, x =I- 0, a < 0.
° ° °
Consider the case in which < a < 1. Then a -1 < 0. Hence (1 + X),,-l -
1 < for x > 0, and (1 + X),,-l - 1 > for -1 < x < 1. Thus, it follows
° °
that f'(x) > on (-1,0), and f'(x) < on (0,00). We deduce that f(x) is
increasing on (-1,0), and decreasing on (0,00). This gives f(x) < f(O), and
hence
(1 + x)" < 1 + ax, if x> -1, x =I- 0, 0< a < 1.
•
Techniques 69
Example 2.17. Let a1,a2,a3, ... ,a nn be n > 1 distinct real numbers. Prove
that
min
min (. a· - ak )2
15cj<k5cn ( a)J - ak
15cj<k5cn
2
) <
:S
12
12
-,-~
- n(n 2 - 1) (
ta; -
)=1
. ~(taj)2)
)=1
Solution: We first prove that
mm
min.
15cj<k5cn a' - a
(a'
15cj<k5cn ()J k) <
ak) 22 12
-- --:--::----,--
n(n 2 _
n(n2 - 1)
1)
(n
(Ln )
L aj ,
j=l
a22
j ,
(2.8)
)=1
for any n real numbers a1,a2,ag, ... ,a nn .. We may assume 2:.7=1 a; =

2:.7=1 1. We may
also enumerate aj's such that
a = a1:Sa2:Sag:S" ·:San.
·:San .
Put
2 12
f-L =
f-L = - 2-
n(n - 1)
Suppose the inequality is not true. Then aj+1 - aj > f-L for 1 :S j :S n -1. Thus
aj > a + (j - 1)f-L,
for 2 :S j :S n. Hence
n n
1= La; > L(a+(j-1)f-L)2
j=l j=l
n-1
n-1
L (a + jf-L)2
j=O
n-1
n-1 n-1
n-1
na + 2af-L L j
2 + f-L2 L j2
j=l j=l
r
na2
na
2 + (1
+ af-Ln(n
af-Ln n -
) + f-L22 n(n
_ 1) + f-L
n (n-1)(2n-1)
- 1)(2n
6
- 1)
6
n( a +
n(a f-L(n _1))2
+ f-L(n;-
2 1) + 11
+
> 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 (b j ) to get
. . 2 2
5c n (aj - ak) :S n(n
12 ( n
~ (aj(~ t) 2)) ,,
+ t)2
mm
15cr;;!r
l<J<k<n
- -
n (2
n 2 _- 1) L
j=l )=1
70 Inequalities
valid for all t. Consider the function J(t) 'L.7=1 (aj

f(t) = 'L.7=1 (aj + t( Expanding this,
we obtain
f(t) = t,a; +2(t aj )t+nt2 ,
which is a quadratic polynomial in t. This quadratic polynomial attains its

extremum at ta, where J'
f' (ta) == O. Solving this, we see that
ta = -~(taj),
)=1
Since J" (ta) = 2n > 0, we infer that Jf has a unique minimum at ta. Hence
f"(ta)
J(t)
f(t) 2:: J(ta),
f(ta), for all t. But we see that
f(ta) =
n -;;:2(n)2
I>; L aj
l(n )2
+;;: L aj
)=1 )=1 )=1
f; a; - ~ (f; a
n n 2
j)
This implies that
12
min (.
l"'5.j<k"'5. n a) - ak
)2 < . ~-,
n(n2 _ 1/
f(ta)
(ta)
12 . ( La; -;;:
n(n2---
-1)
1
a; - ~ 2:n (n2:
Laja
) 2) . j .
)=1 )=1
)=1
•
Example 2.18. Let a, b, x, y be real numbers such that ay - bx == 1. Prove
that
a 2 + b2 + x 2 + y2 + ax + by 2:: V3.
Solution: We can find u, v, B, <jJ such that a = u cos B, b = u sin B, x = v cos <jJ
and y = v sin <jJ. Using ay - bx = 1, we infer that uvsin(B - <jJ) = 1. Hence
uv 2:: 1 (remember u and v are positive). We obtain
cos 2 (B - <jJ) = 1- sin 2 (B - <jJ) =1- U V

1
2 2 = (U 2 V 2 -1)/u 2 v2 .
We observe that u 22 = = a 2 + b22 ,, v 22 == x 2 + y2 and ax + by = uvcos(B - <jJ). Using
u 2 + v 2 2:: 2uv, we obtain
a2 + b2 + x 2 + y2 + ax + by 2:: 2uv - J u 2v2 - 1.

Techniques 71
Thus, it is sufficient to prove that
2uv - JU 2 V 2 -
";U =: 112> vi
Consider the function f(t) = 2t - Jt2=1 far t ~2 1. We have
'( )
t -2-
fj'(t)=2-
- ~,-
Jt2=1' t
which gives f'(to) = 0 on [1,00) if and only if to = 2/../3. We see that

1
f"(t) = (Vt2 _1)3'
Hence 1" (to) = = 3../3 > O. This shows that f has a unique minimum at
to == 2/../3 on [1,00). Thus we have
2 f(to) = ../3.
f(t) ~
It follows that
2uv - VU 2V 2 - 1~ J3,
which is what we have to prove. •
Another important technique far 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.
ExaIllple 2.19. Prove that in a tri angle with sides a, b, c, circum-radius R

and in-radius r, the inequality
9r (4R + r) :S 38 2 :S (4R + r) 2,
holds, where 8 =
= (a + b + c)/2 is the semi-perimeter.
Solution: Consider the monic polynomial whose roots are a. b, c. We have
82 + r(4R + r) 82 + 4Rr + r 2
abc 6,
8
2
+_._+- 6,2
2
6, 8 8
2 abc 6,2
8 +8- +8- 2
8 3 + abc + (8 - a)( 8 - b) (8 - c)
8
ab + bc + ca,
72 Inequalities
and
abc
4sRr = T .ß = abc.
Here ß denotes the area of the given triangle. Observe that we have used the
well known Heron's formula: ß2 = ses - a)(s - b)(s - c). Thus a, b, c are the
roots of the equation
p(x) = x 33 -- 2sx 2 + (S2

(S2 + r(4R + r))x - 4sRr = O.
Now p(x) = 0 has three real roots. Hence Rolle's theorem shows that p'(x) =0
has two real roots. But
p'(x) == 3x 22 -- 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
16s 2 -12(s2+ r (4R+r)) 20.
This leads to
9r (4R + r) :::; 3s 2 •
We also know that
ß ß
ra = - - , r =-.
s-a s
Hence
sera - r)
a= .
ra
Similarly, we can get
b=s(rb-r) s(r ec - r)
c = --'-------'--
rb rc
Here ra,rb,r ec are the ex-radii ofthe tri angle ABC.
= s(y - r)/y in p(x) == 0, we obtain an equation in y;
Putting x =
q(y) = y3 _ (4R + r)y2 + S2 y - S2 r = O.
Observe that ra,rb,r c are the roots of q(y) = O. Hence q'(y) = 0 also has real
roots. But
q' (y) = 3y2 - 2 (4R + r) + S2,
S2,
and its discriminant is 4 (4R + r)

r) 2 - 12s 22 .. We thus obtain the inequality
3s 2 :S (4R + r) 2 .
•
Techniques 73
2.5 Trigonometrie substitutions

Some of the inequalities may be proved using suitable trigonometrie substitu-
encountered one such use in Example 2.18. We shall
tions. We have already eneountered
study a few more such examples here.
Example 2.20. Let a,

that
a, b, c be real numbers such that °: :; a,
~ a, b, c ~
:::; 1. Prove
}a(1-
Ja(l b)(1 - c)
- b)(l + Jb(l
}b(l- c)(1 -
- c)(l a) + Jc(l-
}c(1 - a)(l
a)(1 - b) ~
:::; 1 +~.
Solution: Sinee
Since a, b, c are in [0,1],
[0, 1], we ean
can find x, y, z in the interval [0,7r/2]
[0, 7r /2]
such that
y,cc = sin 22 z.
a = sin22 x, b = sin22 y,
Thus the inequality to be proved gets transformed to
sin x eos cos z + sin y eos
cos y eos cos x + sin z eos
cos z eos cos x eos :::; 1 + sin x sin y sin z.
cos y ~
sin x ( eos cos z - sin y sin z)
cos y eos + eos cos z + eos
cos x ( sin y eos :::; 1.
cos y sin z) ~
On simplifieation, reduces to sin(x + y + z) ~
simplification, this reduees :::; 1, whieh
which follows from the
the
property of the sine funetion.
function. •
= 1.
Example 2.21. Let x, y, z be positive real numbers such that xy+yz+zx =
Prove that
2x(1 - x 2) 2y(1
2x(l-x - y2)
2y(1_y2) 2z(1
2z(l-z 2)
- Z2) xX Y zZ
-"'---,2~
-~-,f-+ + 2 + <
<--- -2+
+----+ - - 2.
+-
(1 + x 2)
(1+x2)2 (1 + y2)
(1+y2)2 (1 + Z2)2 - 1+x
(1+z2)2 1+ x 1 + y2
1+y2 1 + Z2
1+z
Solution: Consider the first term on the left side:

2x (1 - x 2) _ 2x 11-- x2
(1 + x 2) 2 - 1
1 + x 2
2 .
. 1 + x 22 .·
We know that
2tan(a/2) . 1 - tan 2 (a/2)
11 + Sin a,
= sina, 1 + tan2(a/2) = eosa.
+ tan (a/2)
tan2(a/2)
2 =
Hence, this suggests the substitutions

Henee,
x = tan(a/2), y = tan(ß/2), z = tan(r/2).
tanb/2).
We reeall
recall that the function x t-+
r-+ tan x is a one-one function of (0, 7r
7r /2) on to
(0, (0).. The inequality now transforms to
(0,00)
1
cos a + sin ß eos
sin a eos cos ß + sill'Y eos :::; 2
'Y ~
cos l' "2 ( sin a + sin ß + sin l''Y)) .
74 Inequalities
sin 20: + sin 2,

2ß8 + sin 2"( :S sin 0: + sin ß + sin "(.
2')' ~ ')'.
We observe that
1- xy 1 - tan(0:/2)
tan( 0:/2) tan(ß/2)
tan(ß /2)
z=--
+y
z = xx+y tan( 0:/2) + tan(ß /2)
tan %).
(~ - ~ - ~).
This shows that
tan(')'/2) = tan ((~
tanb/2) ~ - -~)..
-7r - -0: ß)
222
Since 0: and ß are in (0,7r/2), so is %.

~ - ~ --~. Since x t---t
r-t tanx is one-one on
(0, 7r /2), we conclude that

(0,7r/2),
')'
"( 7r 0: ß
2 222
This gives 0: + ß + "(')' =
= 7r. Now f(x) = -- sinx is convex on (0,7r). Suppose
fex) =
0: ~ ß
0::s ~ "(
ß:S ')' and let 0:', ß', "('
')" be apermutation of o:,ß,"(.
0:, ß, ')'. Using Theorem 10 on
page 41, we obtain
- sin (0:' + 0:) - sin (ß' + ß) - sin ("('

(!" + ')')
"() :S~ -- sin (20:) - sin (2ß) - sin (2"().
(2')').
Taking 0:' =
= ß, ß' = "(')' and "('
')" =
= 0:, we obtain
sin (20:) + sin (2ß) + sin (2')')

(2"() ~ sin (0: + ß)
:S + sin (ß + ')')
"() + sin (!'
("( + 0:).
This reduces to
sin (20:) + sin (2ß) + sin (2')')

(2"() ~ sino: + sinß
:S + sin,)"
sin"(,
proving the inequality needed.

•
Example 2.22. Show that for any three real numbers a, b, c, the inequality
(ab + bc + ca - 1) 2 ~ (a 2 + 1) W+ 1)(c2 + 1),

holds.
Solution: Put a = b = tanß, c =

= tano:, b = tan,)" o:,ß,"( are in the
tan,,(, where o:,ß,')'
interval (-7r/2,7r/2).
(-7r /2, 7r /2). Then
a2 +1= sec 2 0:, b2 +1= sec 2 ß, c2 +1= sec 2 "(,

Techniques 75
and the inequality to be proved is

2
( (ab + bc + ca - 1) cos 0: cos ß cos,) :::; 1.
a COS
However, we observe that
(ab + bc) cos 0: cos ß cos ,

a COS a sin ß cos ,
sin 0: + cos 0:a sin ß sin ,
(a + ,),
sin ß sin (0:
(ca - 1) cos 0: cos ß cos ,
a COS a cos ß sin , - COS
sin 0: cos 0: cos ß cos ,
a COS
- cos ß cos(o:
cos(a + ,).
Thus, we obtain
r
( (ab + bc + ca - 1) cos 0: cos ß cos , ) 2
a COS
( sin ,8 sin
sin((0:
a + ,) - cos ß cos
cos((0:
a + ,)
cos 2 (0:
(a + ß + ,) :::; 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\1'3
-----===2 +
--===+ <
< --.
-.
VI + x Ji+Y2 + VI + Z2 - 2
Solution: Since x + y + z = xyz, there is a tri angle with angles 0:,

a, ß" such
that x = tan 0:,
a, y = tan ß, z = tan ,. Then the inequality reduces to
.
sino:
sma . ß
+ sinß
sm + sin,
sm 3\1'3
' rv < --.
3\1'3
. 1-_ 2 .
This follows from 3.4.5 on page 101.

•
2.6 Properties of quadratic polynomials
The fact that a quadratic equation ax 2 +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, ... ,a n and
b1 ,b2 ,b3 , ... ,b n of real numbers. We associate the quadratic polynomial with
these sequences by
n
f(t) = L (aj + bj t)2 = A + 2Ct + Bt 2 ,
j=l
76 Inequalities
where
n n n
A = LaJ, B = LbJ, C = Lajbj .
j=l
j=l j=l
j=l j=l
j=l
We may assume that B =I- 0, for, otherwise bj = for all j forcing C = 0, and °
°
in this case the inequality C 2 :S AB is trivially true. Thus B > 0. Since f(t)
is the sum of several squares, we have f(t) 2: for all t. Hence its discriminant
is non-positive. This implies that C 2 :S AB and gives the Cauchy-Schwarz
inequality. We also observe that equality holds if and only if f(t) = has two °
coincident real roots. This is equivalent to aj + Ab j = 0, for 1 :S j :S n, where
A= _ 2::7=1 ajb j
",n b2 '
L-j=l j
is the double root of f(t) = 0.
Example 2.24. Let (al,a2), (b l ,b2) and (Cl,C2) be three pairs ofreal num-
bers. Prove that
(L al b2 + L a2 bl - 2 L al a2 ) 2 :S 4 ( L ai - L al bl ) ( L a~ - L a2b2 ) ,
where the sum is cyclically over a, b, c.
Solution: We know that for all real x, y, z, the following inequality holds:
x 2 + y2 + Z2 2: xy + yz + zx.
In fact this is equivalent to (X_y)2 + (Y_Z)2 + (z -X)2 2: 0. Moreover equality
holds if and only if x == y == z. Now consider the quadratic polynomial
P(t) = L (al + ta2)2 - L (al + ta2) (bI + tb2).
cyclic cyclic
Using x 2 + y2 + Z2 2: xy + yz + zx, we observe that P(t) 2: 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) ( L a~ - L a2b2 )t 2
cyclic cyclic
+(2 L ala2 - L al b2 - L a2bl )t)t

cyclic cyclic cyclic
L albl ).
+( L aiai -- Lalbl)'
cyclic cyclic
Techniques 77
Thus we obtain
(L
cyclic
a l b2 +
al L
cyclic
a2 bl - 2 L
cyclic
ala2 f
:::;4( L ai- L albl ) ( L a~- L a2b2).
cyclic cyclic cyclic cyclic
Here equality holds if and only if

al + ta2 == bl + tb2 == Cl + tC2,
where t is the coincident root
2 L:cyclic al a2 - L:cyclic al b'f - L:CYCliC a2 bl
t = - 2
L:cyclic a2 - L:CYClic a2 b2
This reduces after simplification to

blC2 - b2CI + Cla2 - C2al + al
a l b2 - a2bl = O.
•
Example 2.25. Let n > 2 and Xl, X2, ... ,X n be n real numbers. Put
n
p= LXj, q= L XjXk·
j=l l:Sj<k:Sn
Prove that
p _
_
-
n
-- _
-n_
n
-1 J 2nq p n - 1 / p2
p2---:::;x.:::;-+--
p2---:::;x·:::;-+--
n-l J n n
p2 -
_-2nq
_-_,,
n-l
J
for all j.
Solution: Consider p and q:

P XI+(X2+ X3+···+ Xn)
q XIX2 + XIX3 + ... + XIX n + L XjXk·
25,.j<k5,.n
2:Sj<k:Sn
We see that
(p-xI)2 = (X2+ X3+···+ Xn)2
< (n-l)(x~+x~+ ... +x;)
(n-l){(x2+ xx3+···+
3+···+ x n)2- 2 L XjXk}
25,.j<k5,.n
2:Sj<k:Sn
(n - l){ (p - xI) 2 - 2(q - Xl (p - Xl)) }
(n - 1)(p2 - 2q - xn.
78 Inequalities
Thus we get a quadratic inequality:
nxi - 2pXI - (n - 2)p2 + 2(n - l)q :::; O.
Solving this inequality, we get
p n- 1
- - --
n n
0
J 22
p - --
-2r~q
2nq < Xl < -p
n-1- -n
n - 1/
+- -
n
J 2
p - --.
2nq
n-1
Since we can choose any Xj in place of Xl, 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
literat ure.
Example 2.26. (CRMO, 1999) Let a, b, c be the sides of a triangle. Prove

that
abc
- --+
b+c-a c+a-b
+ a+b-c->3.
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 + 1t) + (1t + ~)
y+z z+x x+y
--+--+-- + (~ + ~)
X Y zz X Z Y z X
> 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..jXY ·2#· 2JZi ::; (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 + 2' b + 2' c + 2. Prove that
[A'B'C'] 2 ~[ABC].
Solution: It is easy to observe that there is a triangle with si des a + ~, b + ~,

2 2
c + ~. Using Heron's formula, we get
16[ABC)2 = (a + b + c)(a + b - c)(b + c - a)(c + a - b),

and
2 3
16[A'
16[A' B'C']
B'C' ] = 16 (a + b + c)(-a + b + 3c)( -b + c + 3a)( -c + a + 3b).
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
[ABCj2 _ 16xyz
[AIBICI]2 - 3(2x+y)(2y+z)(2z+x)"
Thus it is sufficient to prove that
(2x + y)(2y + z)(2z + x) 227xyz

for positive real numbers x, y, z. Using the AM-GM inequality, we get
2x + Y 2 3(X 2y)1/3, 2y + z 2 3(y2 Z )1/3, 2z + x 2 3(Z2 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 so me
results. It asserts that:
Let a, b, c be positive real numbers and let A be any real number. Then
aA(a - b)(a - c) + bA(b - c)(b - a) + cA(c - a)(c - b) 2': 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 assurne that no two are equal. Because of the symmetry, we may assurne
a > b > c. If A 2': 0, then
aA(a - b)(a - c) + bACb - c)(b - a) + cA(c - a)(c - b)

(a - b){ (a - c)a A - (b - c)b A} + (a - c)(b - c)c AA
2': o.
(a - b)(a - c){ a A - bA} 2':0.
> (a-b)(a-c){aA-b
If A < 0, then
aA(a - b)(a - c) + bA(b - c)(b - a) + cA(c - a)(c - b)

aA(a - b)(a - c) + (b - c){ (a - c)c A- (a - b)b A}
> (b - c)(a - c){ CA - bA} 2':2': o.

>
ExaIllple 2.29. (IMO, 2000) Let a, b, c be positive real numbers such that
abc = 1. Prove that
(a-l+~) (b-l+~) (C-l+~) ~1.

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.

x(y - x)(x - z) + y(z - y)(y - x) + z(x - ::;

z)(z - y) ~ o.
This follows from Schur's inequality. (For different solutions refer to problem
63 on page 254.) •
Techniques 81
Example 2.30. (APMO, 2004) Let x, y, z be positive real numbers. Prove

that
(x 2 + 2){y2 + 2){Z2 + 2) ~ 9(xy + yz + zx).
(X
Solution: Expanding, the inequality is
X2y2z 2 + 2 L X2y2 + 4 L x2 + 8 ~ 9 L xy.

cyc\ic cyclic cyc\ic
Note that
2 L X2y2 - 4 L xy +6 =2 L (xy - 1)2 ~ O.

cyc\ic cyc\ic cyc\ic
Moreover Lcyc\ic x2 ~ LCYc\iC xy. Thus it is sufficient to prove that
X2y2z 2 + L x2 + 2 ~ 2 L xy.
cyclic cyclic
If a, b, c are positive reals, Schur's inequality gives
L a3 + 3abc ~ L a2b + L ab 2
cyc\ic cyclic cyc\ic
ab(a + b) + bc(b + c) + ca(c + a).
Using u + v ~ 2,jUV for any positive reals u, v, we get
L a3 + 3abc ~ 2 L (ab)3/2.
cyclic cyclic
= X2/ 3, b == y2/3 and z == c2/ 3, this re duces to

Taking a =
x 2 + y2 + Z2 + 3{xYZ)2/3 ~ 2(xy + yz + zx).
X2y2z2 + 2 ~ 3{xYZ)2/3.
In fact, this is equivalent to t 3 + 2;::: 3t, where t = (xYZ)2 / 3; this follows from
(xyz)2/3;
t3-3t+2=(t-l)2(t+2)~0. Thus
x2y2Z2+X2+y2+Z2+2 > X2 +y2+ Z2+3(xyz)2/3

;::: 2+3(xyz)2 / 3
> 2(xy + yz + zx).
>
•
82 Inequalities
2.9 Majorisation technique

There is another useful technique, caBed the majorisation technique, often used
to derive new inequalities. Let x = (Xl,X2,." ,xn) and y = (Yl,Y2,oO. ,Yn)
be two vectors in ]Rn such that
x12 x 22 x 32·· '2xn and Y12Y22Y32·· ·2Yn.

We say X is majorised by y and write X -< YY if
(i) Xl +X2+X3+" '+Xk :S Yl +Y2+Y3+' '+Yk for 1 :S k :S n - 1;
(ii) Xl +X2+ X3+"
+X2+X3+" +xn = Yl +Y2+Y3+" '+Yn'
Let A be a subset of ]Rn. A function f :A ~ ]R is said to be Schur-eonvex if
X -< y
X y= } f(x) :S f(y).
=}
If the inequality is reversed, we say f is Sehur-concave. It is clear that f is

Sehur-eoneave if and only if - f is Sehur-convex.
There is a useful eriterion for eheeking whether a given funetion is Schur-
eonvex, at least when the funetion is reasonably smooth. Let
In = (a,b) X (a,b) x· 0
000 X (a,b),
be an 'interval' in ]Rn. Suppose f : In ~ ]R has partial derivatives of the first
order. Then f is Sehur-eonvex if and only if f is symmetrie in the variables
and
( Xj - Xk)
OXj OXk
(8
f (X) _ f (X)) ~ 0, 8
on In, for aB j :f. k. Here are some useful classes of Sehur-eonvex funetions.
(i) If 9 : (a, b) ~ ]R is a convex funetion, then
n
f (Xl ,X2,X3," .,Xn) = L g(Xj)
j=l
is Sehur-convex on In. This leads to a standard majorisation inequality:

if f : (a, b) ~ ]R is convex, then
n n
(Xl,X2,X3,oO .,xn) -< (Yl,Y2,Y3,oO.,Yn) =
=}} 'Lf(xj) :S 'Lf(Yj).
j=l j=l
(ii) A funetion f : In ~ ]R is said to be convex if
f(AX+(l-A)y) :SAf(x) +(l-A)f(y),
for aB veetors X, y in In, and A E [0, 1). Any eonvex function is Sehur-
eonvex.
Techniques 83
(iii) A function f : In JE.

--+ ~
~ is said to be quasi-convex if
ffC~x
( AX + (1
(1- ~ max
- A)Y) ::; max{{ff(x),
(x), ff(Y)},
(y) },
x, y in In, and A E [0,1].
for all vectors x,y [0, 1]. Any quasi-convex function is
Schur-convex.
Example 2.31. If a, b, c are the si des of a triangle, prove that

sides
Va + b - cc+ Vb
..;a:+b Jb + cc -- a + JVcc + a - b ::;
~ Va + Vb + ,jC.
VC.
Solution: Suppose a 2~ b 2~ c.c. Then
> a,
2(8 - c)
2(s
- c) + 2(s - b) > a+b,
2(8-C)+2(8-b)
2(s
2(8 - b)
2(s + 2(s
2(8 - a) + 2(s
2(8 - c) a + b+ c.
Thus
(a,b , c) -<
(a,b,c) (2(8 - c),2(s
~ (2(s C),2(8 - b),2(s
b),2(8 - a)).
Since f(t) =
= v't
0 is concave on (0,00),
(0, (0), using majorisation theorem,
}2(8 - c)
J2(s + }2(8
J2(s - b) + J2(s
}2(8 - ~
a) ::; Va + Vb +,jC.
+ VC.
This reduces to
Va + b - cc+ Jb
Ja Vb + cc -- a + Jc
Vc+ a - b::;
b~ Va + Vb +,jC.
+ VC.
•
Example 2.32. Let al,aZ,a3,
al,a2,a3, ... ,a nn be n natural numbers such that
al::;a2::;a3::;·· ·::;an , and al +a2+a3+· ·+an = nk + m,
where k, ~ m < n. Prove that

k , m are non-negative integers and 0 ::;
ala2a3·· ·an ::; (k + l)mk n - m .

Solution: Consider the vectors
(a n ,a
, an -l,a
-l,ann -2 '" ,aI)
-2 ,, ... ,al) and (k+1,k+1,
(k+1 , k+1 , ... ,k+1,k,k, ...
.. . ,k),
,k) ,
where k ++ 1 appears m times and kappears n - m times. If m > 0, then

we observe that an 2 ~ k + 1. Otherwise an ::; L aj ::;~ nk, a
~ k and hence L:
contradiction. Similarly, it is easy to see that if m > 1, then an + an-l 2 ~
2(k + 1). An easy induction proves that
an + an-l + ... + an- ~ m(k + 1).

m+l 2
an-rn+!
84 Inequalities
Similarly,
an m+1 + an-m + ... + an-m-j 2 m(k + 1) + Ci + 1)k,

+ an-i, + ... + an- m+1
for 0 :s: j :s: n - m - 1. It follows that
(k+ 1,k+ 1. .. ,k+ 1,k,k, ... ,k) -< (an,an-1,an-2,." ,al).

Since fex) = In x is concave on (0,00), we get
n
L)naj :s: mln(k + 1) + (n - m) lnk.
j=l
This simplifies to
a1a2a3" ·an :s: (k + 1)mk n- m .
•
Example 2.33. Show that in any triangle with sides a, b, c,
(a +b- c)(b + c - a)(c + a - b) :s: abc.

Solution: If we take a 2 b 2 c, then it is easy to see that
(a,b,c) -< (2(s - c),2(s - b),2(s - a)),

(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 fex) == lnx, we get
In2(s - c) + In 2(s - b) + In2(s - a) :s: lna + lnb + lnc.

This simplifies to
(a + b - c) (b + c - a)( c + a - b) :s: abc.

•
2.10 Muirhead's theorem
There is a classical theorem which was discovered by Muirhead. Consider a vec-
tor a = (al ,a2,a3, . . . ,a n ) in jRn. For each positive vector x = (Xl ,X2,X3, . . . ,x n ),
i.e., vector with positive components, consider the sum
S ( a; x ) =,. 1 ~ al a2 an
L...J xo-(1)xo-(2) ... xo-(n)'
n. o-ESn
where Sn denotes the set of aB permutations of {1, 2, 3, ... , n}. Then Muir-
head's theorem asserts that
S(a;x) :s: S(b;x)

Techniques 85
for an positive vectors x in IRn 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(x 2yz + y2 zx + Z2 xy ) :::; x 3y + y3 z + Z3 X + xy3 + yz3 + zx 3,
for an non-negative numbers x, y, z. Similarly (1,1,1) -< (3,0,0) and hence

3xyz :::; x 3 + y3 + z3 for an non-negative reals x, y, z. In fact, it is easy to derive
the AM-GM inequality. Observe that for any positive integer n
(l/n,l/n, ... ,1/n) -< (1,0,0, ... ,0).
Hence for any n non-negative real numbers al,a2,a3," .,a nn ,, we obtain
l/n
n (al a2 ... an ) :::; al + a2 + ... + an·
Example 2.34. 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 + ab2c + abc2 :::; a4 + b4 + c4 .
Henee Muirhead's theorem at onee gives the result.

°
However (2,1,1) -< (4,0,0) since 2 < 4, 2 + 1 < 4 + and 2 + 1 + 1 = 4 + °+ 0.
•
Example 2.35. Let a, b, c be non-negative real numbers. Prove that
1
a3 + b3 + c3 + abc 2: "7 (a + b + c)3 .
Solution: Expanding the right side, the inequality may be written in an

equivalent form:
6(a 3 + b3 + c3) + abc 2: 3(a 2b + b2c + c2a + ab2 + bc2 + ca 2).
Note that
2(a3 + b3 + c3) 2: a2b + b2c + c2a + ab2 + bc2 + ca 2
by Muirhead's theorem. Hence the result folIows.

•
86 Inequalities
2.11 Homogenisation
There is a large dass 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 2.36. Let x, y, Zz be positive real numbers such that x +y+z = 1.

Prove that
2(x 2 + y2 + z2) + 9xyz 2 1.
Solution: We put this in a homogeneous form using x +y+z = 1:
2(x+y+z)(x 2 +y2 +z2) +9xyz 2

~ (x+y+Z)3.
Expanding, we get
2 L x3 + 2 L xy2 +2 L x 2y + 9xyz
2 L x3 + 3 L xy2 +3 L x2y + 6xyz.

L
L x3 - L L xy2 - L L ~ o.
x 2y + 3xyz 2 O.
Equi valently,
L
L x(x-y)(x-z) 20,
~ 0,
cyclic
which follows from Schur's inequality.

•
Example 2.37. (Tournament of Towns, 1997) Let a, b, c be positive real num-
bers such that abc = 1. Prove that
111
--:-~+ + < 1.
a+b+1 b+c+1 c+a+l-
Solution: We introduce a = x 3, b = y3 and c = Z3 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
--;;-----;:----
--;:----;:---- + + < -- .
x3 + y3 + xyz y3 + Z3 + xyz Z3 + x + xyz
3 xyz'
xyz
Equivalently, we have
xyzL (X3+y3+xYZ)(y3+Z3+xyZ)::;
cyclic
rr
cyclic
(X 3 +y3+ xyZ ).
Techniques 87
Eut the left side reduces to

But
XYZ
xyz (3 I:
2: X3y3 + 4I:
cyclic
2: x 4yz + I:
X3y3 2: x
cyclic
x 4yz
cyclic
x6 + 3x
3X 2y2Z2 ).
Z2).
And the right side is
3z3 + 2X 2y2
3X 3yy3z3 y2zZ22 2: x 3 + xyz I:
I: 2: x 6
cyclic cyclic
+ 3xyz 2: x 3y3 + 2: x y 3 + 2: X3y 6 6.

cyclic cyclic cyc1ic
cyclic
The inequality is equivalent to
2 I:
2: X5y2Z2::; I:
5 2: X y3 + I:
2: Xx 3y 6 6.
cyc1ic
cyclic
This may be written as

S(5, 2, 2) ::; 3(6,3,0).
3(5,2,2) S(6, 3, 0).
Since (5,2,2) -< (6,3,0), the result follows from Muirhead's theorem. •
Example 2.38. (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 66 on page 257.
Using x + y + z == 1, we put the inequality in a homogeneous form:
o0::;::; (x + Y + z )( xy + y z + zx)
(x+y+z)(xy+yz+zx) --2xyz::; ;7 7
(x + Yy +
2xy z ::; 27(x+ +z)
z/ .
3
°
Observe
bserve that
+ Y + z) (xy + yz + zx) --2xyz=

(x+y+z)(xy+yz+zx)
(x 2xyz = I: x 2yy++ 2:
2: I: xy2 ++xyz.
xyz.
cyclic cyc1ic
cyclic
Hence the left-side inequality folIows.

follows. On the other hand
(x + y + z)3
Z)3 = I: x33 + 3 L
L x22yy + 3 L xy2 + 6xyz.
cyclic
cyc1ic cyc1ic
cyclic cyc1ic
cyclic
Hence the inequality reduces to
6 L x 2y + L xy2 ::; 7 L x 3 + 15xyz.

cyclic
cyc1ic cyclic
cyc1ic cyc1ic
cyc1ic
88 Inequalities
We may put it in the form
(2 L
cyclic
x3 _ L
cyclic
x 2y_ L
cyclic
x y2 ) +5( L cyclic
x(x-y)(x-z)) 2:0.
Since (2,1,0) -< (3,0,0), Muirhead's theorem gives

2 L x3 - L x 2y - L xy2 2: 0.
cyclic cyclic cydic
By Schur's inequality, we get
L x(x-y)(x-z) 2:0.
cyclic
Hence the result folIows.

•
Example 2.39. (USSR, 1962) Let a, b, c, d be non-negative real numbers such
that abcd = 1. Prove that
a2 + b2 + c2 + d 2 + ab + ac + ad + bc + bd + cd 2: 10.
Solution: The result at on ce follows from the AM-GM inequality. Eut we

give here a different solution using homogenisation. We may homogenise this
to get an equivalent inequality:
(a 2 + b2 + c 2 + d 2 + ab + ac + ad + bc + bd + cd) 2 ~ lOOabcd.
For a given 4-tuple (ml,m2,m3,m4) ofnon-negative integers, we use the no-

tation:
5(ml'
5(ml,m2,m3,m4)
1
m2, m3, m4) = -24 '"' am~(l)
24 ~
L
bm~(2)cm~(3)dm~(4)
am~(1)bm~(2)cm~(3)dm~(4),
,
I7E S 4
where 54 is the set of all permutations of {I, 2, 3}. Expanding the left-side, we
may write the inequality in the form
45( 4,0,0,0) + 245(3, 1,0,0) + 185(2,2,0,0) + 485(2, 1, 1,0) 2: 945(1,1,1,1).

Now Muirhead's theorem gives
5(4,0,0,0) 2: 5(3,1,0,0) 2: 5(2,2,0,0) 2: 5(2,1,1,0) 2: 5(1,1,1,1).
Hence
45( 4,0,0,0) + 245(3, 1,0,0) + 185(2,2,0,0) + 485(2, 1, 1,0)

2: (4 + 24 + 18 + 48)5(1, 1, 1, 1) = 945(1,1,1,1).
This gives the required inequality.
•
Technique8 89
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
aal1 -
- +a2 +···+an
- +···+a
- --- n >( al a2 ... )1/n
_ alaZ'" an ,
n
for nonnegative real numbers al,aZ,'"
al,a2, ... ,an' Since it is homogeneous, we may
normalise it by ala2a3 ... an == 1. Thus we have to prove that
al + a2 + ... + an :::: n,
for any n positive real numbers al,a2, ... ,an, under the additional condition
that ala2a3 ... an = 1. This may be proved by the principle induction. Für
n = 1, it is immediate. If n = 2, then
al + a2 - 2 = al + a2 - 2Jala2 = (yfaj - y'a2) 2 ::::~ o.O.

Suppose it holds for any k positive real numbers whose prodllct is 1, where
k < n. Take n positive real numbers al, a2, ... , an such that al a2a3 ... an =
= 1.
Among these n numbers, there mllst be some number :::: ~ 1 and there mllst be
some nllmber ::; 1. Thlls we may aSSllme that al :::: ~ 1 ::::
~ a2. We may apply
indllction to n - 1 numbers al a2, a3, ... , an:
+ a3 + ... an :::: n - 1.
al a2
Thus it is enollgh to prove that al + a2 ::::
~ 1 + al a2.
But this is equivalent to
(al - 1)( a2 - 1)::; 0 which is a conseqllence of al ::::
1)(a2 ~ 1 ::::
~ a2. Hence the proof is
complete.
Exalllpie 2.40. (INMO, 2007) Let x, y, z be positive real numbers. Prove

that
(x + y + Z)2(yz + zx + xy)2 ::; 3(y2 + yz + Z2) (Z2 + zx + x 2) (x 2 + xy + y2).
Solution: The ineqllality is homogeneolls of degree 6. Hence we may aSSllme
x + y + z = 1. Let 0; = xy + yz + zx. We see that
x 2 + xy + y2 (x + y)2 - xy
(x + y) (1 - z) - xy
x + y - 0; = 1 - z- 0;.
Thus
II(x 2 + xy + y2) (1 - 0; - z)(l - 0; - x)(l - 0; - y)
(1 - 0;)3 - (1 - 0;)2 + (1 - 0;)0; - xyz
0;2 - 0;3 - xyz.
90 Inequalities
Thus we need to prove that a 2 :'S 3(a 2 - a3 - xyz). This reduces to
3xyz :'S a 2 (2 - 3a).
However
3a = 3(xy + yz + zx) :'S (x + Y + Z)2 = 1,
so that 2 - 3a 2: 1. Thus it suffices to prove that 3xyz :'S a 2 . But
a2 - 3xyz (xy + yz + ZX)2 - 3xyz(x + Y + z)

L x 2y2 - xyz(x + Y + z)
cyclic
1
"2 L (xy - YZ)2 2: o.
cyclic
For different solutions, see solution to problem 236 on page 384.

•
Example 2.41. (IMO, 2001) Prove that
a b c
-r=:;;====c=
-r=::;;==:=;= + + >1
J a2 + 8bc Jb 2 + 8ca J c2 + 8ab -
for all positive real numbers a, band c.
Solution: Here we give a solution using normalisation a + b + c = 1. We use

the convexity of J(t) == 0.
Using the weighted Jensen inequality, we get
a b c
---r==:~~+ +--,=;~~
J a2 + 8bc Jb 2 + 8ca J c2 + 8ab
> 1
1
-2: -yll'a=;(=C
Ja(aa22;;==+==8=:=bc"7)=+===b(=:=b
+ 8bc) + b(b22;;==+==8=ca=;)=+=c:;=(
+ 8ca) + c(cC7<=22 =+=8==a=:=:=b)
+ 8ab)
Since J(t) == 0 is strictly decreasing and J(I) == 1, it suffices to prove that
a(a 22 + 8bc) + b(b 2 + 8ca) + c(c 22 + 8ab) :'S 1.
For this we go back to homogenisation:
a(a 22 + 8bc) + b(b 22 + 8ca) + c(c 2 + 8ab) :'S (a + b + c)3.
But this reduces to
3 L a2 b + 3 L ab 2 2: 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 177 on
page 348.) •
Techniques 91
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, x 3 + y3 + z3 -
3xyz or 2(x 3 + y3 + Z3) - (x 2 y + y2 z + z2 X + xy2 + yz2 + zx 2). 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) ~ ° and P(1, 0, 0) ~ 0;
(ii) P(A, JL, v) ~ ° for all non-negative reals A, JL, v.
Proof: It is sufficient to prove that (i) implies (ii). Let us put
P(x, y, z) == A 2:= x 3 + B ( 2:= x

3 2y + 2:= x y2 ) + Cxyz.
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:
P(x, y, z) = r 2:= x 3 + (q - r) ( 2:= x 2y + 2:= x y2 ) + (p - 6q + 3r)xyz.

cyclic cyclic cydic
Here we consider two cases.

Case 1: Suppose q ~ r. In this case we write
P(x, y, z) r( ~ x3- 3XYZ)

cychc
+ (q - r) ( ~ x 2y + 2:= xy2 - 6XY Z) + pxyz.

cychc cychc
Since r ~ 0, p ~ ° and q> r·, the result follows.

92 Inequalities
Case 2: Suppose q :S r. \\Te ,,;rite
P(x,y,z) =
= q( L x3 - 3XYZ)
eyclIc
+ (r - q) ( L x cyclIc
3 - ~
cyclIc
x2y - L
cyclIc
xy2 + 3XYZ) + pxyz.
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 asymmetrie homogeneous

form of degree 3:
P(x, y, z) == L (px:3 + q.T 2 y + rxyz) ,

sym
where p, q, r are real numbers. Suppose P(l, 1, 1) :::: 0, P(1, 1,0) :::: 0 and
P(2, 1, 1) :::: O. 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(l, 1,1) 6p + 6q + 6r :::: 0,
P(l, 1,0) 4p + 2q:::: 0,
P(2, 1, 1) 20]) + 14q + 12r :::: O.
Note that
P(.T, y, z) 2p(x 3 + y3 + Z3)
+q(x 2 y + y2 z + Z2 X + xy2 + yz2 + zx 2) + 6rxyz.
Hence some computation leads to
P(a, b, c) (4p + 2q) ( L x 3 ) + (6]) + 5q + 6r) ( ~ x2Y + ~ x y2 )

cyclIc cyclIc cyclIc
+12(q + r)xyz
Q(x,y,z).
Observe that
Q(l, 1, 1) 48(p + q + r) :::: 0
Q(l, 1,0) 20p+ 14q+ 12r:::: 0,
Q(l, 0, 0) 4p + 2q:::: O.
Techniques 93
Hence Q(x, y, z) ::::: 0 for all non-negative real numbers x, y, z. It follows that
P(a, b, e) ::::: O. •
Example 2.42. Let a, b, e be the sieles of a triangle. Prove that
3(a + b)(b + e)(e + a) ::; 8(a 3 + b3 + e:3 ).

Solution: Consieler the polynomial
P(a, b, e) =
= 8(a 3 + b3 + e3 ) - 3(a + b)(b + e)(e + a).
This is a homogeneous polynomial of elegree 3 in the variables a, b, e. Observe

that
P(l, 1, 1) 24 - 18 - 6::::: 0,
P(l, 1,0) 16 - 6 == 10> 0,
P(2, 1, 1) 80 - 42 - 12 =
= 16 > O.
By Stolarsky's theorem, P(a, b, e) ::::: 0, for all a, b, e which are the sieles of a
triangle. (In fact we can apply theorem 14 since P(l, 0, 0) == 8 > 0 anel hence
the result is true for positive a, b, e.) •
Example 2.43. Let a, b, e be the sieles of a triangle. Prove that
2(a + b + e)(a 2 + b2 + e2 ) ::::: 3(a 3 + b3 + e3 + 3abc)

Solution: If we take
P(x,y,z) = 2(x + Y + z)(x 2 + y2 + z2) _ 3(x 3 + y3 + z3 + 3xyz),

then P(x, y, z) is a homogeneous polynomial of elegree 3 in three variables.
Moreover, P(l, 1, 1) == 0, P(l, 1,0) == 2 anel P(2, 1, 1) == O. Hence Stolarsky's
theorem is applicable anel we concluele that P(a, b, e) ::::: 0 whenever a, b, e are
the sieles of a triangle. (Note that we cannot apply theorem 14 clirectly, since
P(l,O,O) == -1 < 0.) •
Chapter 3
Geometrie inequalities
3.1 Introduction
Many of the inequalities we have studied and the teehniques we have learnt have
their direet implieations in a dass of inequalities known as geometrie inequal-
ities. These inequalities explore relations among various geometrie elements.
For example, when we eonsider a triangle, we ean assoeiate many things with
it: angles, sides, area, medians, altitudes, eireum-radius, in-radius, ex-radii and
so on. We have already some inequalities, viz., tri angle inequalities associated
with the sides: a < b + c, b < e + a, C < a + b, where a, b, C are the sides
of a triangle; these eonditions are neeessary and sufficient for the existenee of
a triangle with sides a, b, c. We ean derive various relations among these ge-
ometrie elements. The dassic example is Euler's inequality: R 2': 2r, where
R is the eireum-radius and r is the in-radius. This ehapter provides several
inequalities of this kind, but the list is not exhaustive. For an exeellent and a
fairly exhaustive eollection of geometrie inequalities, please refer to [5) and [6).
3.2 Notations
For a triangle ABC, we use the following standard notations:
• a = IBCI, b = ICAI, e = IABI;

•• oe = LBAC, ß == LCBA, / == LACB;
Cl =
• m a , mb, m c are respeetively the lengths of the medians from A, B, C on

to BC, CA, AB;
• h a , h b , h c are respectively the lengths of the altitudes from A, B, C on

to BC, CA, AB;
• are respeetively the lengths of the angle bisectors of LBAC,

W a , Wb, W c
LCBA, LACB;
• R is the eireum-radius; r is the in-radius; r a , rb, r c are the ex-radii;

• Ll is the area of ABC;
• s is the semi-perimeter of ABC: s = (a + b + c)/2;

• 0 is the eireum-eentre; H is the ortho-eentre; I is the in-eentre; G is the
cent re of gravity; N is the nine-point eentre;
Geometrie Inequalities 95
• 0 1 denotes the first Brocard point of ABC; i.e., the unique point inside
ABC such that LCA0 1 = LAB0 1 = LBC0 1 = w, the Brocard angle of
ABC. It is known that w ~ 1f/6.
We also use a large number of results:

sin a sin ß sin ,
1. the sine rule: - -
a
= -b- = -c- = 2R;
2. the eosine rule: a 2 = b2 + c2 - 2bccosa, b2 = c2 + a 2 - 2cacosß, c2 =
a 2 + b2 - 2abcos,;
3. the half-angle rule:
. /2 V(S-b)(S-C)
j(S-b)(S-C) /2 Vs(s-a)
js(s-a)
sm a = bc ' cos a = bc'
etc.;
4 R = abc.
. 4ß'
5. T = 4Rsin(a/2) sin(ß/2) sin(r/2) = (s - a) tan(a/2)
= (s - b) tan(ß/2) = (s - c) tan(r/2);
6. ß = TS = (1/2)bcsina = 2R2 sinasinßsin,;

7. ß = )s(s - a)(s - b)(s - c) (Heron's formula);
8. 16ß 2 =
= 2a 2 b2 + 2b2 c2 + 2c2 a 2 - 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 = A : j.1,
j.1, then
Ab2 + j.1c 2 = (A + j.1)(AD 2 + BD . DC).

Note that Appolonius' theorem is a special case of this result: b2 + c2 ==
2AD 2 + 2BD 2 , where D is the mid-point of BC;
10. m a == (1/2)v2b2
(1/2)J2b 2 + 2c2 - a 2 , etc.;
11. Wa = (2bc/(b + c)) cos(a/2) = (2~/(b + c)) )s(s - a), etc.;
12. Ta = V(s(s - b)(s - c))/(s - a) = ß/(s - a), etc.;
13. 0/ 2 = R 2 - 2RT = R 2 (1 - 8 sin(a/2) sin(ß /2) sin(r /2));
14. OH 2 == R 2 (1- cosacosßcos,) == 9R 2 - (a 2 + b2 + c2 );

96 Inequalities
15. I H 2 = 2r 2 - 4R 2 eos a cos /) cos i;
16. L.: eos a = 1 + 4 TI sin(a/2). (Remark: the notation L.: always denote in
this ehapter the cyclieal sum and TI denotes the eyclieal produet unless
otherwise stated. Für example, L.:a = a+b+c, L.:a 2 b = a2 b+b2 c+c 2 a,
L.: sin a = sin a + sin /3 + sin i, TI eos a = cos a eos ß cos i ete.)
3.3 So me identities involving elements of a triangle
In this seetion we develop some identities involving various elements of a trian-
gle: sieles, angles, circum-radius, in-radius, ex-raelii, altitudes, sem i-perimeter.
ete. In fact we identify some polynomial equations satisfied by these elements
and use these equations für furt her explorations.
3.3.1 Equatiün für the sides

We start with the relation
a = 2Rsina = 4Rsin (a/2) cos (a/2), s- a = rcot (a/2).

Using these, we obtain
ar
. 2 ()
sm 2 (a/2) =
ar eos-"((a/2)
eos2 a/2 ) = a(s-a)
= a(s - a) .
sin a/2 4 R( s _- a)'
4R(s a)' 4Rr .
We thus get
1 sin 2 (0/2) + cos 2 (a/2)

a
4R
(r
(s - a)
(s-a))
+ - 7 - '- .
This simplifies to the relation
a:l - 2sa 2 + a(s2 + r 2 + 4Rr) - 4Rrs =

= O.
Thus, a is a root of the cubic equation
t3 - 2st 2 + (8
(S22 + r 2 + 4R1')t - 4R1's = O.
4R1'8 (3.1)
Similarly, we can prove that b, c also satisfy the cubic equation (3.1). Now
using the relations between the roots anel coefficients of a cubic equation, we
obtain
La = 28,2s, Lab = 8S22 + 1,2 + 4R1', abc = 4R1'8.
4R1's.
Using (3.1), we obtain the equation for the reciprocals of the sides of a triangle.
Thus, l/a, l/b, l/c satisfy the cqllation:
4Rrs(l -
4R1'8(l (S2
(8 2 + 1'2 + 4R1'
4Rr)t)t2 + 28t
2st - 1 = O. (3.2)
This gives
S2
' " ~1 = S2 + rr 22 +
+ + 4R1'
4R1' 1 1
L....t
L a = ~ 4R1's
.4 T'"'l
'
,
L ab = 2Rr·
3.3.2 Equatiün für Altitudes

Using (3.2), we have
1 - (2
4Rrs~
4Rrs- (s2
s + l'1'22 + 4R1' -~ + 2s-
4R1'))1 2s~1 - 1 = O.
a3 22 a a
Using a = 26./ h a , this re duces to
2Rh~ - (S2 + 1'2 + 4Rr)h~ + 4rs 2h aa -- 4r 2s 2 = o.
O.
We get similar expressions for h bb and h c . Thus, h aa ,, hb, h c are the roots of the
cubic equation
2Rt 3 - (S2 + r 2 + 4Rr)t 2 + 4rs 22 t - 4r 2S2
S2 = O. (3.3)
We hence obtain
s2 + r 2 + 4Rr Lhah b = 2s 2r
Lha
s2
Lha = ~~ ,' II h a = 2r 2 8 2
= 2R R' R .
Using the reciprocal equation
4r 22s 2 t 3 - 4rs 2 t 2 + (S2
(S2 + r 2 + 4R1')t - 2R = 0,
we also obtain
1 1 + r 2 + 4Rr
L h a = ;:,
L _1_ =
hah b
S2
4r 2s 2 ·
3.3.3 Equatiün für s - a, s - b, s - e

The equation (3.1) mayaIso be written in the form
(s - t)3 - s(s - t)2 + (1'2 + 4R1')(s - t) - s1' 2 = o.
O.
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 - se
se + (1'2 + 4Rr) t - s1' 22 = O. (3.4)
We thus obtain the relations
= s,
2)s - a) = = r(r
L(s - a)(s - b) = + 4R), II(s - a) == sr 2 .
Going to the reciprocal equation, we see that l/(s - a), 1/(8 l/(s - c) are
l/(s - b), 1/(8
the roots of the equation
s1' 2 t 3 - 1'(1' + 4R)t22 + st - 1 == O. (3.5)
We also obtain
L
""' 1 r' + 4R,
= r·+4R
L
1 1
6 s- a sr (,,_n)(,,_',\ r2'
(s - a)(s - b) -= rZ·
98 Inequalities
3.3.4 Equatiün für ex-radii

Now 1/(s - a) is a root of (3.5). Henee, we have
2 1 1 1
sr (
sr,
s-a
)3-r(r+4R)(
s-a
)2+s-(--)-1=0.
s-a
This ean be written in the form
r2 ( _ s ) 3 _ r(r + 4R) ( _ s ) 2 + S2
S2 ( _ s ) _ S2
S2 = O.
s-a s-a s-a
= ra(s - a) == 6.. Substituting s/(s - a) == ra/r, we

However we know that rs =
get the relation
r~ - (r + 4R)r~ + s2 ra - s2 r = O.
Similar relations for rb and r cc may be obtained. Thus, r a, rb, r e are the roots
of the equation
t 3 - (r + 4R)t2 + S2 t - s2 r = O. (3.6)
Consequently, we have the relations
2:
Lrara = r+4R, L....,;
'""' rarb = s 2 , rarbrc = S2 r.
Again the reeiproeal equation of (3.6) is
S2 rt 3 - s2 t2 + (r + 4R)t - 1 = 0, (3.7)
whose roots are I/ra, I/rb, I/re. We henee obtain the relations
2:~=~
ra r'
2:-1-
L
I = r+4R
r+4R
rarb =~.
S2 r '
3.4 Some geometrie inequalities
In this seetion, we prove several inequalities among various elements of a tri-

angle. Again the readers are reminded that the following seetion only gives a
pIe of geometrie inequalities, but not an exhaustive list of these inequalities.
sam pie
The exeellent eolleetion of examples in [5] and [6] gives an idea of the innu-
merable possibilities one ean have in the dass of geometrie inequalities. The
teehniques used are essentially what were developed earlier; we oeeasionally
resort to different ideas.
3.4.1. abc 2: 8(s - a)(s - b)(s - c).

Prüüf: We have a 2 - (b - C)2 ~ a 2 and equality holds if and only if b = c.

Similar inequalities hold: b2 - (c - a)2 ~ b2, c2 - (a - b)2 ~ c2. Hence,
abc > Ja 2 -=
# - (b
(b =-~)2y'b2 - (c - a)2Jc2 -
- c)2Jb 2 =-(Z=--a)2jc (a - b)2
(a + b - c)(b + c - a)(c + a - b)
8(s - a)(s - b)(s - c).
= b == c.
Equality holds if and only if a =
Alternatively, Stolarsky's theorem (see theorem 15 on page 92) 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
= 0, P(l, 1,0) == and P(2, 1, 1) == 2 > 0. Hence,
P(l, 1, 1) =
abc - (a + b - c)(b + c - a)(c + a - b) ?: 0,
and the result follows. •

3.4.2. abc< I: a2(s - a) ~ ~abc.
2
Prüüf: We have
2I:a 2 (s-a) a22 (b + c - a) + b2(c + a - b) + c2 (a + b - c)
I: a b + I: ab I: a
2 2 - 3.
On the other hand, we also see that
c)
(b + c - a)(c + a - b)(a + b - c) = (c 2 - a2 - b2 + 2ab)(a + b - c)
2
= a b + ab + b c + bc + c a + ca - a - b - c:3 - 2abc.
2 2 2 2 2 3 3
Thus, we obtain
2 I: a (s - a) == (b + c -
2 a)(c + a - b)(a + b - c) + 2abc.
Since a, b, c are the sides of a triangle, we know that b + c - a > 0, c + a - b >

and a + b - c > 0. Hence, abc< l:: a2 (s - a). Now using 3.4.1, we get
°
2 I: a2(s - a) ~ abc + 2abc == 3abc,
which proves the right hand side inequality.
100 Inequalities
Again,
Again , we may use
use Sto1arsky's theorem(theorem 15 on page 92). Consider-
ing
P(x,y,z)
P(x, L
y, z) = L xx22 (y + zz - x) - 2xyz,
cyclic
cyclic
that it is a homogeneous po1ynomia1 of degree 3 and P(l, 1, 1) =

we see that = 1,
P(l , 1,0) =
P(l, = 0, 1, 1) =
0, P(2, 1, = O. Henee, b, c) > 0,
Henee, P(a, b, 0, giving the 1eft-side inequa1-
the other hand,
ity. On thc hand , the po1ynomia1
po1ynomia1
Q(x, y, z) == 3xyz - L L x 2 (y + z - x),

cyclic
gives Q(l, 1, 1) = 0,
0, Q(l, 1,0) = 0 and Q(2,
Q(2 , 1, 1) Q(a, b, c) > 0, and
= 2. Thus Q(a,
inequa1ity.
we get the right-side inequality. •
3.4.3. ~ : :; L bb:: c < 2. Equa1ity ho1ds on the 1eft if and on1y if a = bb= c.
Proof: We have proved this in ehapter 1; refer to (1.4). The 1eft hand side
of the above inequa1ity is generally known as Nesbitt's
Nesbitt 's inequality. There are a
such proofs.
variety of ways of proving this. We give two sueh
(i) Using the Cauehy-Sehwarz inequality,

inequality, we have
(a+b+c)2 = (LVb:CJa(b+c)f
(LVb:CJa(b+C)r
< c) (
L a(b + c)).
b:: J (La(b+c)).
(L b
This gives
'L" aa (a+b+c)2 3(ab+bc+ca) 3

~ b+ ~
b + c 2: + bc + ca) 2:~ 2(ab
2(ab+bc+ca)
2(ab + bc + ca) = 2'
2(ab+bc+ca)
sinee (a + b + C)2 2:~ 3(ab + bc + ca).

< b :::; c, sinee
(ii) We may assume a :::; inequa1ity is symmetrie in a, b, c.
since the inequality
This imp1ies that
111
- -<
<- -- -<
<----.
b+c - c +a - a+b·
b+c - c+a - a+b
Using re arrangement inequality,
inequa1ity, we obtain
abc abc
- -+--+-- > -
b+c c+a a+b
- + - - +b+c'
c+a a+b b+
- -c'
abc abc
- - + - - + - - > --+--+--.
b+c c+a a+b
b+c c+a a+b a+b b+c c+a a+b b+c c+a
Adding these two, we obtain
ab
2 ( --+--+--
b+c c+a a+b
c) 2: ~ 3.
This gives the desired inequality.
•
3.4.4. y's < vs=a + ~ + vs:::c : ; J3S.
Proof: The first inequality follows from the fact that for
far any positive reals
y , Z, we have
x, y,
Jx ++ Y ++ Z < Vx ++ v'Y ++ ,,[Z. ,;y ,;z.
On the other hand f(x) = Vx
JX is a concave (0, (0).
concavc function on (0,00). Hence,
I 1
L 1
3~ ::; VL 3(8 - a),
which gives
~+~+ vs:::c : ; J3S.
•
3.4.5. · 0:
O < sma
sm + sm ß
.. ß+ ·sm"(::;
' 3V3
3.)3
-2-.
-2-'
Proof: We know that 0 H 2 = 9R 2 - (a 2 + b2 + c2 ), which gives

a2 + b2 + c2 ::; 9R 2 •.
Hence, we obtain
(a + b + C)2 ::; 3(a 2 + b2 + c 2 ) ::; 27R 2 •
This gives
o < a + b + c ::; 3V3R.
= 2Rsina,
Using a = 2Rsino:, etc., we get
· 0:
O < sm
sma .. ß
+ sm '. <
ß + sm"(
sm"(::; 3V3
3.)3
-2-'
- -2-.
•
3.4.6. 2a + sin 2ß + sin 2"( ::; sin 0:
sin 20: a + sin ß + sin "(.
102 Inequalities
Proof: Consider f(x) == -sinx on (O,K). This is a convex function. Suppose

a:S ß:S "(. Using Theorem 10, we see that for any permutation (a',ß',,') of
(a,ß,,), we have
L - sin (a' + a) :S L - sin (a + a) = - L sin 2a.

This reduces to
L sin 2a :S L sin (a' + a) .
Taking a' =
= ß, ß' =
= ,, and " =
= a, and using sin(a + ß) =
= siwy, etc., we get
sin2a + sin2ß + sin2,:s sina + sinß + sin"(.
•
3.4.7. II sin 3V3
3V3
a -< -
sina< -.
- 88 .
Proof: Using the AM-GM inequality and 3.4.5, we have
3V3
II sina:S (~~nar :S (V;r 8
•
3.4.8. 0<
.
II sin 1
sm (a/2) :S ~.
S·
Proof: We know that 0/2 = R 2 (1 - 8 TI sin (a/2) ). Hence, it follows that

1- 8TIsin (a/2) ~ 0. The first inequality is obvious. Equality holds if and
only if the triangle is equilateral. •
3
3.4.9. 1< "cosa <~.
~ -<-.
2
Proof: We use the known identity: L ~ cos a = 1+4TIsin(a/2). Since

TI sin (a/2) > 0, we get the left side inequality. Now using the inequality
3.4.8, we get the right side inequality.
•
3.4.10. 1< L sin(a/2) :S ~.
Proof: Whenever a,ß" are the angles of a triangle, (7r - a)/2, (7r - ß)/2,
(7r -,)/2 are also the angles of some triangle. Applying 3.4.9 to this triangle,
we get these inequalities. •
3.4.11. II eosa <-< ~.8-.1

Proof: Using 3.4.9 and the AM-GM inequality, we get
II eos
eosa<
-
(L:eosa)3
a :S (L: eos a ) <
3
~-.
1
-< 8·
3
8
•
3.4.12. If x, y, z are real numbers such that xyz > 0, then
y
xeosa
x eos a + yeosß z eos, ::; 2
y eos ß + zeos,:S -;; + 11
1 (YZ zx + ~
x ) .
Proof: We start with the obvious inequality
(x z eos a + y z eos ß - xy) 2 + (x z sin a - y z sin ß) 2 2: o.

This simplifies to
2x 2 yz eos a + 2xy 2 z eos ß - 2xyz 2 eos( a + ß) ::;

:S L y2 Z2.
Using eos(a + ß) = - eos" we obtain the desired inequality. Equality holds if
and only if xz eos a + yz eos ß - xy = 0 and xz sin a - yz sin ß = O. This takes
the form
1 1 1 . . ß .
- : - : - = sm a : sm : sm ,.
x y z
We easily infer that the right side of 3.4.9 is a eonsequenee of the present
inequality. •
3.4.13. In an obtuse-angled triangle, L:

L: eos 2 a > 1 and L:
L: sin 2 a < 2.
Proof: We have
IH2 =R2 ( 1 - II eos a) = 9R L a 2 - 2.
Sinee the triangle is obtuse, rr eos a << O. Henee, I H 2 > R 2 . This gives
R2 < IH = 9R L
2 2 - a2 .
104 Inequalities
Thus, L" a 22 < 8R 2 .. This may be put in the form

'~
sin 2 a + sin 2 /3 + sin 2 ~f < 2.

Equivalently
"'2 a> 1.
~cos
•
3.4.14. L tan a > 3V3 if the tri angle is acute, and L tan a < 0 if the
triangle is obtuse.
et, tan ß and tawy are

Proof: Suppose the triangle is acute. In this case tan a,
aU positive. We have
II tan a = L tan a :::: 3 ( II tan a )

1/3
.
This gives TI tan a :::: 3V3. Using TI tan ü == L tan a, we get the desired
inequality for acute-angled triangles. Alternatively, we can also use the fact
that f(x) = tanx is a convex function on (0,7r/2).
If the triangle is obtuse, say a > 90°, then tan a < 0 and tan ß, tan, are
positive. Hence, TI tan a < 0 and this gives L tan a < O. •
3.4.15. Lcota:::: V3.

Proof: We have
CoSü cosß
cot a + cot,ß --+--
sina sin ß
sin (a + ß)
sin a sinß
sina sin ß
2 sin,
cos(a - ß) + cos,
> 22 sin
sin,
-----'- , = = 2 tan (r
2tan (r/2).
/2).
1 + cos,
Similarly, wc may prove that
cot ß + cot, :::: 2 tan (a /2) , cot, + cot a :::: 2 tan (ß / 2) .

Adding, we obtain
cotet+cotß+cot,):::: 2(tan(a/2) +tan(ß/2) +tan(r/2)).

2( cota+cotß+cot,)::::
However using the convexity of f (x) = tan x on (0, 7f /2), we obtain
~(tan(OO/2)+tan(ß/2)+tan(l/2)) 2': tan(OO+~+I)

tan30° = ~.
It follows that
(0:/2) ++ tan (,8/2)

tan (00/2) (,8/2) ++ tan (1/2)
(1/2) 2': vi
This gives
cotoo + cotß ++ cot,2': tan (00/2)

coto: + (0:/2) ++ tan (,8/2)
(,8/2) ++ tan (1/2)
(1/2) 2': V3.
•
3.4.16. (0'/2) 2': 3y3.
Lcot (00/2)
Proof: Using the convexity of the function fex) = cotx on (0,7f/2), we get
the desired inequality. The convexity part can be proved using the second
derivative of fex). •
3.4.17. L cot 2 (00/2)

(0'/2) 2': ( (0'/2)) (( L cot (0).
L cot (00/2)) 0').
(0'/2) = x, cot
Proof: Let us put cot (00/2) (ß/2) = (1/2) = z. Then we know
y, cot (1/2)
+ y ++ z = xyz. We have
that x +
+ Y ++ z) 2 == (x ++ Y ++ z )xy z ==
(x + L +
(x 22 -- 1) y z + L xy.
This gives
x 2 + y2 + Z2 + L xy = L (x 2 - 1) y z.
It follows that
2 (x 2 + y2 + Z2) 2': x 2 + y2 + Z2 + L xy = L (x 2 - 1) y z.
We write this in the form
x 2 + y2 + Z2 2': xyz ( L X2
2:
1) .
However, observe that
x 2 -1
- - =cotO'.
~=cotoo.
2x
106 Inequalities
Thus, we obtain
L cot 2 (0/2) ~> (11 cot (0/2)) (L cot 0)

(L cot (0/2)) (L cotO).
•
3.4.18. TI cot 0 ~ 1/ (3V3) in an acute-angled triangle and TI cot 0 < °for
an obtuse-angled triangle.
Proof: This easily follows from the fact that
11 tano ~ (3V3),
in an acute-angled tri angle and this product is negative for an obtuse-angled
triangle (see the proof of 3.4.13). •
3.4.19. L: cot 2 0 ~ L: cot 0 cot ß = 1.
Proof: This follows from the rearrangement inequality.

•
3.4.20. l: sec 0
L: ~ 6.
Proof: Using l: coso :::; 3/2

L: (see 3.4.9) and AM-HM inequality, we get
L 9 2
l: coso ~ 9 x -3 = 6.
" sec 0 ~ '\"'
'~
•
3.4.21. l: cosec 0
L: ~ 2V3.
Proof: This follows from the convexity of f(x) = cosecx on (0,11"). In fact
f"(x) = cosecx(cosec 2 x + 2cot 2 x) ~ 0,

on (0,11"). Hence f(x) = cosecx is convex on (0,11"). This gives
3"1 '"'
'~
" cosec 0 ~ cosec (0 + ß + 1')
3 2
= cosec 60° = V3.
Hence, we get
L cosec 0 ~ 2V3.
•
3.4.22. L cosec 2 O' ~ 4.
Proof: Using the re arrangement inequality, we have
a
Lcosec 2 O' > L cosec a cosec ß
0'
L: sina
LsinO'
rrsina
sinO'
3
>
. )2/3·
((rrrr smO'
sina)
2/3·
We have used the AM-GM inequality at the end. However we know that
rr ::; 3V3/8 (see 3.4.7). It follows that
a ::;
sin 0'
4
'"'
~
a >
'""' cosec 2 O' - 3x ~
-3 = 4.
•
3.4.23. (L sin (0'/2) r: ; L cos 2 (0'/2).
a
Proof: We may arrange the angles such that either 0' > 60° > ß > ''YY or
0: ::;
0' 'Y. This implies that
::; 60° ::; ß ::; 'Y.
. -~
(( sm
sin
2 2
1) ( . 1)
ß - -~) (sin 'Y -- -~) > 0
sm -1:
2 2 -
4 sm
. 2 . "2-'Y >
-ß sm 2 ( sm ß
- + sm
. 2 -'Y) -- 1
. "2
2 --
=}
=} sm sm sm sm
2 2 2
=}
1 +4 II sm. 2"20' ~2: 2·sm
0:
sm"20' ( sm
2 . 2ß + sm
0: ( .
"2 'Y) + 1 - sm
. "2
'Y) 0'
. 2·
"2. 0:
However we know that

. ß . ''YY
2 sm-sm-
2 2 (% -"2
ß
cos ( 2 - 'Y)
~) -- sm
. ~
sin 2
0'
< 1 - sin~.
2
This gives
1
+ 4 II sm
. 2"20'0: >-- 2 sm 0: (( .. ß
. 0'
2
sm"2
. "2'Y) + 2·
"2 + sm
sm 2 sm 2
. "2'Y
"2ß sm
=} L cos 0:
0' >
-
2L sm·. -0'2 sm. -ß2
0: .
2,"",.0:. ß
=}
'"'
'""'
~ cos 2
20'
20:
2: ~ sm 2
"2 ~ "2
,",.20'
,"",.20:
+ 2'"'.0'.
~ sm 2 "2.
"2 sm 2·
108 Inequalities
This simplifies to
(Lsin (a/2)) 2 :S Lcos 2 (a/2).
•
3.4.24. a2 + b2 + c2 2: 4V3.6. (Weitzenböck's
(Weitzenböek's inequality).
Proof: Using Heron's formula for the area of a triangle, we have
16.6. 22 = (a + b + c)(a + b - c)(b + c - a)(c + a - b)

(a+b+c)(a+b-c)(b+c-a)(c+a-b)
< (a+b+C)(a+~+cr
(a+b+c)4
27
This gives
4v'3.6.< (a+b+c)2
4J3.6.<
- 3
However we know [rom the Cauchy-Sehwarz
Cauehy-Schwarz inequality that
(a+b+c)2 :S3(a2 +b2 +c2 ).
It follows that
a2 + b2 + c2 2: 4V3.6..
Solution 2: vVe prove astronger version of the above inequality which
whieh is
knmvn as thc Hadwiger-Finsler inequality. We show that
a2 + b2 + c2 2: 4J3.6. + L(a - b)2.
Using the eosine rule, we have
a2 = b2 + c2 - 2bceosa
2bccosa
(b - C)2 + 2bc(1 - cos
eos a)
(b - C)2 + 4.6. tan(a/2),
sinee
( 1 - eosa) = 4.6. tan(a/2).
eos a) = 4.6..
2bc(1 - cos
sma
Thus, we get
a2 + b2 + c2 = 4.6.( tan(a/2) + tan(ß/2) + tan(r/2)) + L(a - b)2.

However, we know that f(x) = tan;]: is a convex function on (O,1l-j2). Hence,

it follows that
l(
::3 tan(a/2) + tall(ß/2} + tan(r/2) 2':tall ) (a+ ß +r )
6 1
=taIl30o= )3.
This shows that

a2 + b2 + c2 2': 4v'3~ + :L)a - b)2.
We observc that equality holds if and only if a = b = c. •
3.4.25. ab + bc + ca 2': 4)3~.
Proof: We kllOW that
~ = "21 a b sm
. r = "2Ib·
c sm a = 1 sm ß.
"2 ca .
Hellce, we get
L.: ab == L.: -sina- >-

2~
1 6~
(1
II
) --
sina
1/:3
However we kllOW from 3.4.7 that TI Sil~ a 2': (8/3)3). This gives
L.: ab 2': 6~ ( 3)3

8 ) 1/3
= 4v'3~.
•
3.4.26. I: a4 2': 16~ 2 .
Proof: USillg 3.4.24 alld thc Cauchy-Schwart: inequality, we get

1 . 2
L.:a 4 2':::3(L.:a 2 ) 2':16~2.
•
3.4.27. 'i;""'
L.. a 2b2 2': 16~ 2 .
Proof: This follows from Heron's formula alld 3.4.26. •

3.4.28. (abc)2 2': ()3
(abc)22': )3 ( 4~)3
110 Inequalities
Proof: We use a + b + c = 2R( sinn + sinß + sin ,) :s (3V3)R (see 3.4.5) and
abc == 4Rß. Thus,
4ß abc 3abc ()2/3

-V3 =
V3 = -
RV3 2:
RV3 - a + b+ c - :s
- > a + b + c < (abc)2/3,
abc
,
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 :S Rand equality holds if and only if the

triangle is equilateral.
Proof: We use 01 2 = R(R - 2r-).

•
3.4.30. 9r(ro + 4R) :S 38 2 :S (r + 4R)2.
Proof: Wc know that a, b, c are the roots of the cubic equation (see section
3.3.1) t 33 -- 28t 2 + (8 2 + r 2 + 4Rr) t - 4Rr8 =
= O. Since all its roots are real, the
derivative polynomial 3t 2 - 48t + (8 2+ r 2 + 4Rr) = 0 has only real roots. This
imposes a condition on the discriminant, namely, 8 2 2: 2 3r(r + 4R). This gives
the left inequality.
Again we know that ra, rOb, r c are the roots of the equation (see section
3.3.4) t 33 -- (r + 4R)t 2 + 82 t - 82 r = O. We look at the derivative polynomial:
3t 2 - 2(ro + 4R)t + 8 2 = = O. This again has only real roots. Hence, using the
condition on the discriminant, we obtain 38 2 :S (r + 4R)2, which is the other
inequality. •
3.4.31. 82 2: 27r 2 .
Proof: We have
8
-
(8 - a) + (8 - b) + (8 - c) > {(8-a)(8-b)(8-C)}1/3
(~2r/3
3 3
(r
( r2
2 8)1/3.
8
) 1/3
.
On cubing the above relation and rearranging the terms, we get the result. •
3.4.32. 36r 2 :S L a 2 :S 9R 2 .
Proof: Using 3.4.31, we have
2
27r:S
2 (a+b+C)2
(a+b+c)2
2 -<~"
3,", 2
:S 4
4 ~ a2 ,
where we have used the Cauchy-Schwarz inequality. This gives the left side
inequality. On the other hand, using OH 2 == 9R 2 - (a 2 + b2 + c2 ), we get the
right side inequality. •
L::-2
a 2
3.4.33. 4.
rbrc
Proof: We know (see 3.3.4) that 2:rbrc

L:rbrc = S2. Using the Cauchy-Schwarz
inequality, we get
L:: ~ 2 (2: a)2 = 4s 2 _
rbrc 2: rbrc S2 - 4.
•
3.4.34. 5R - r 2 V3s.
~ V3s.
Proof: We use 2:
L: r a = 4R + r (3.3.4) and R 2
~ 2r (3.4.29). Thus,
2
= 4R + r + R - 2r ~ 4R + r ==
5R - r = L::
Lr a·
D./(s - a) = rs/(s - a) and similar expressions, we obtain

Using r a = tl/(s
5R-r 2 ~::>a tl(Ls~a)

D.(Ls~a)
La22}.}.
Lab -- L::a
4~ { 2 L::ab
4~ {2L::ab- L::a2 } 2 V3s.

2 L:: ab - L:: a 24V3tl.
2
This follows from the Hadwiger-Finsler inequality. (See 3.4.24.)

•
3.4.35. L:a(s - a):S 9rR.
2:a(s
112 Inequalities
Proof: We know that r' = (s - a) tan (0:/2), etc. Hence,
La(s - a) =L ~rr )\
( = 4Rr Lcos 2 (0:/2).
tan 0:/2
Thus, it is sufficient to prove that 2 L cos 2 (0:/2) <

:S ~~. This is equivalent to
L cOSü -::::S 3/2.
L This follows from 3.4.9. •
3.4.36. L ha -::: V3s.
Proof: We have
. ~~ ~
(a+b+c)2;:::3(ab+bc+ca)= 2~ (ha+hb+hc)=6RLha.
Thus, we obtain
""h < (a+b+c)2

Lh a -::: s(a+b+c)
~ a - 6R
6R 3R
But we know that a + b + c -:::
:S 3V3R. It follows that L h aa -::::S V3s.
L •
3.4.37. L~<V3
1V3
L-<-·
a - 2r·
2r
Proof: We have
L~~ == L
"" L,hh a .
~aa 2~
2~·
Lh a -::: V3~
r
This follows from ~ = rs
TS and 3.4.36.
•
3.4.38. L ~a >_ . 3V3_.
Proof: vVe know that
rT = 4R II sin (ü/2), 1+4 II sin (ü/2) = L cosü.

Using a == 2R sin a,
Ct, etc., we get an equivalent inequality:
1
"" _1_ > _3_V3_3_ 3V3
L
~ -si-n-a L cos
sin ;::: -L-co-s-a·
Ct - Ct .
This reduces, after cross multiplication, to
'L.....,
" cot
'"""
L......
'"""
cot a0: + 'L.....,
" (cos a0: + -.-
----;--ß
sin ß sma
L...... sm
cos
sin 0:
ß) ;:::;: : 3v3V3.3.
In
Since L:cota;:::
L:coto:;::: v'3 (3.4.15), it is sufficient to prove
L
'""" (c~sa c~s ß) ;:::;: : 2V3.
(c~s + c~sß)
0:
L...... smß sm
sma 0:
L sin,
'""" ---'---
. sin:
L...... sina
smo:smß
2V3.
> 2v3.
sinß -
In
By the AM-GM inequality, we have
L . sin,
siwy
---'---::- >3 (II)
sin
sina 0:
-1/3
..
sm
sin a0: sinß -
Using TI sin a0: :::;:::; 3v'3/8 (3.4.7), we get the desired inequality. •
9
3.4.39. 9r<
9r < L...... r a -<
-- '"
'"""
L....., - --R.
2 R.
Proof: We know from the section 3.3.4 that L: r a = 4R + r. Using 2r :::; R,

we obtain 9r :::; 4R + r :::; (9/2)R. •
3.4.40. 9r :::; L: h a :::; v'3 s.
Proof: We have
v'3
Lh a = 2~ L ~1 : :; 2~~ = V3 s,
and
Lh a ;:::
9
. = 9r.
(See 3.3.2.) •
3.4.41. L:h;:::; (3/4)L:a 2 .
Proof: We know that h a :::; m a , 4m; = 2b 2
4m~ = + 2c2 - a 2 and similar results for
h b, h c , mb, rn c · Thus, it follows that
2: h~ :::; 2: m~ = ~ ( a 2 + b2 + c2 ) .
•
114 Inequalities
2
3.4.42. ""
~ h 2
b
a
+ h c2 >
-
2.
Proof: We write
a2 a 2 b2 c 2
2:
L h2 h~ == 2:
h~ ++ h2
b L -~-
4LV(b + c
c
2 2 )"
Hence, it is sufficient to prove that

1 1
L
" " -cc22>
b22 +
~ b
-
-2: 2R2·
Using the AM-HM inequality, we get

1 9 1
L
" "
b22-
~ b +- >-->-
cc22 -2: 2 L: aa22 -2: 2R2'
since L:a2:s: 9R 22.. (This follows from GH 22 = 9R 22 -- L:a

L:a 22.).)
•
3.4.43. L: ha :s: L: W a :s: 3(R + r).
Proof: The first part is obvious. For the second inequality, we begin with the
expression
2bc
W a = -b- cos (a/2),
+c
:s: (b + C)2, we get W~ :s::s:
and similar express ions for Wb and Wc. Since 2bc :s:
bccos 2 (a/2), and similar estimates hold for and w~. Now the Cauchy- w;
Schwarz inequality gives,
(2: Wa f :s: (2: vbccos (a/2) f

< (2: bC) (2: cos2 (a/2))
9
< 42: bc ;
here we have used the estimateL: cos 2 :s:
(a/2) :s: 9/4. Hence, it is sufficient to
prove that
L bc :s::s: 4(R + r)2

2:
L sin asin ß :s::s: (1 + ~) = (2:
2: (L cos a)
2 2
{:::=}
{:::=} L sin a sin ß :s::s: 2:

2: L cos a + 2 2:
L cos a cos ß
2
{:::=} L cos a :s::s: 2:

2: L cos a + 2:L cos a cos ßß
2
{:::=} 2: (L cos a) + 2:
L 2 cos a :s::s: (2: L cos a. 2 2
Thus, it is sufficient to establish the quadratic inequality
(L: coso:) 2 - 2 L: coso: + L: cos 2 0: 2: o.

If the tri angle is obtuse, then I: cos 2 0: > 1 (3.4.13) and hence the inequality
follows from the fact that the discriminant is negative. Hence, we may assume
that I: cos 2 0: S 1. The above quadratic inequality is true if either cos 0: S L
1- )1 - L cos 0: or
2 L
cos 0: 2: + 1 )1 - L
cos 2 0:. However, we know that
I: cos 0: > 1 and hence the first alternative does not occur. It is sufficient to
prove that für an acute-angled triangle
L coso: 2: 1 + )1 -- L cos
1 +)1 2 0:.
We use
1- L cos 2 0: = 2 II coso:, L coso: = 1 + 4 II sin (0:/2).
Thus, we have to show that
8 II sin 2 (0:/2) 2: II cos 0:.

However we know that I H 2 = 2r 2 - 4R2 TI
TI COS 0:. This gives
4R2 II COS 0: S 2r 2 ,
and hence
COS
16R TI
II coso:0: S 2R2 = 16R sin
TIsin
r2
r2
2R2
22 2 (0:/2)
(0:/2) -
= II'sm22 (0:/2).
- 8 II sin
( / )
0: 2 .
This completes the proof of the desired inequality.

•
3.4.44. L hah b S 3V3.6..
Proof: We know (see 3.3.2) that
'""" h h _ 28 2~s
2s 2 r _ 2.6.8
~ab-R-R'
However wc also have
L.sm 0:) S -2-;

s == 2"1 (a. + b + c) == R (
8
3V3R
L
(see 3.4.7). Using this estimate, we obtain the required inequality. •
3.4.45. TI h a 2: 27r 3 •
116 Inequalities
Proof: We begin with the known identity
1 1
~ . --.
~ ha r
Hence, the AM-GM inequality gives
II hh1
1 :; (1
aa :;
1)3
(-3:31 2: _,1
h ) = a
~a
3
1
2711'3'
27r 3 '
Taking the reciprocals, we obtain
II h a 2: 27r 3 •
•
1 3
~
'" >-.
21' ->
-,
3.4.46.
~ ha - l'
Proof: Using 2: -,1 == -,-,r1 we get

~a l'
= 3 - 2 ==
1= 2: (ha
(ha -- 2~) = ha
= 2: ha-
- 2r.
21',
ha h a h a a
Using the AM-GM inequality, we obtain
(2: ha h- a
21') (2: h ':': 21' ));:::2: 9.9,
(2: a
This gives
'~
" hha > 9.
9,
~ ha - 21' -
2r
Eut the left hand side is equal to
3+
3+
L.h
L
21'
2r
a - 21"
1 3
~
'" >-.
~ ha - 21' ->-
r1"
•
3.4.47. J3~ :; 1'(4R
r(4R + 1'),
r).
Proof: We start with the identity, I[H 21'22 -- 4R 2

H 2 = 2r nrr cos et.
Cl:. However
4R 2 II COSet
COSCl: 2R 2(2: sin 2) 2 et -
~ 2: 4R
a2 - 2
~ (2: a) 2: ab - 4R
2 - 2
2 (2: a ) - (8 2 +1'(4R+1')) -4R 2

1 2
~ (2: a f - (2R+1')2.
Here we have used 'i:,

Lab ab = 82 + 1'(4R
r(4R + 1');
r)j see 3.3.1. Thus, it follows from
I H 2 2'>: 0 that
[H
8 2 ::; 4 (4R 2 + 4R1' + 31'2) .
This gives
.6.
ß2 41'22 (( 4R 22 + 4R1'
< 4r 4Rr + 3r
31'22 ) )
r 2 {{ (4R 22 + (8/3)R1'
1'2 (1/3)r 2 ) + ~ (4R1'
(8/3)Rr + (1/3)1'2) (tRr + 8r 2) }
81'2) } ..
41'22 ::;
Using 4r :S 2R1' :S R 22 ,, we obtain
2Rr ::;
.6.
ß22 < 1'2 (4R 2 + (8/3)R1'
r 2 {{(4R (8/3)Rr+ (1/3)r 2 ) + ~R2}
+ (1/3)1'2)
1
3"1'-
:3r (4R + l' )2 ..
1 ., (4R+r)
2 2
y'3ß ::; l'r (4R

This implies that y'3.6. :s
( 4R + 1')
r) . •
3
3.4.48. 28 < 2..= m a < 28.
Proof: Let D, E, F be the mid-points of BC BC,, CA

CA,, AB respectively. Let AI
A'
be the reflection of A in D. Similarly define BI CI. Then BA'CA
B' and C'. BAI CA is a
parallelogram. (See Fig. 3.1.)
118 Inequalities
c~~ ~-~D+-r=------'io)C
~" \ /c
A'
Fig. 3.1
Using 2m a = AA' < AB + BA' = b + c and similar inequalities for 2mb,

2m e , we get
m a + mb + m e < ~ (b + c + c + a + a + b) = 28.
Let G" be the point of intersection of BG with a line through A and parallel
to the median BE. Then the sides of AA' G" are 2m a, 2mb and 2m c . 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
2" (a + b + c) < 2 (m a + mb + m e ) .
This gives the left side inequality. •
3.4.49. I:m aa ~ 4R+r.
Proof: Let D, E, F be the mid-points of BG, GA, AB respectively. Join the

circum-centre 0 to D, E, F. In the triangle ADO, we see that AD ~ AO +
OD == R + OD, and equality holds if and only if A, 0, D are collinear. Thus,
we obtain m a ~ R + OD. Similarly we have mb ~ R + OE and m e ~ R + OF.
Adding, we get
Lma ~ 3R+OD +OE +OF.
= Rcosa, OE == Rcosß and OF == Rcos,. Using
Observe that OD =
L cos a = 1 + 4 II sin (a / 2) = 1 + i,
we get
Lma~3R+R(1+ ~) =4R+r.
•
3.4.50. s < I:w a ~ JS( I: vs=a) ~ V38.
Proof: Let AD (= w a ) denote the bisector of Q. Then BD + AD > AB = c

and CD + AD > AC = b. Adding we get
2w a + (BD + CD) > b + c.
This shows that Wa > (s - a) and hence s < L: W a . We also have
2 v'bC
Wa = 2v'bC
b + ccvs(s
Js(s -
I'---;-------"7""
a) :::; V s(s - a).
y's(s
Hence, we get
LW v's(L vs=a).
a :::;
Now using the Cauchy-Schwarz inequality, we have
L vs=a : :; V3v's.
Combining all these, we have the required inequalities.
•
3.4.51. L: w~ 2 3V3.6..
3V3.6..
Proof: Using Stewart's theorem or otherwise it is easy to compute that
2-b a2 bc
Wa - C - (b + c)2
Since bc/(b + C)2 :::; 1/4, we obtain
W~ a 2
w2 2
> bc _ a2
bc--.
a - 4
4'
This gives
L:w~::::: L:ab- 41 L:a 2.
But we know that

Lab 2 4V3.6. + L(a - b)2;
see 3.4.24. Hence,
Lab-~La2 = ~(6Lab-L(a-b)2)
= ~ (5 L ab + Lab - L (a - b)2)
> ~ (5 L ab + 4V3.6.
4V3.6.))
> ~ (20V3.6.
(20V3.6. + 4V3.6.
4V3.6.))
3V3.6..
3V3.6..
•
120
120 Inequalities
3.4.52. R + r :::; max { h a , h b , hc } •
Proof: This inequality is equivalent to
2,6,
2~ 2,6,
2~ 2,6,}
2~} 2,6,
2~
R + r < max { -;;:'b'--;:-
R+r:::;max - - - = ---:--:-----::---:-
- a ' b ' c =min{a,b,c}"
min {a, b, c} .
Hence,
Hence, it is sufficient to prove that min {a,, b,
min{a c} (R + r) :::;
b,c}(R+r) 2~. However R
~ 2,6,. +r=
R+r
R( L:
l: cosa)
coso:) = OD+OE+OF, where D, E, F F are the midpoints of BC, CA, AB
respectively and 0 is the cireum-centre.
circum-centre. Thus, the inequality reduces to the
form
min{a , b,c}(OD+OE+OF) ~ 2,6,.
min{a,b,c}(OD+OE+OF):::; 2~.
However we observe that
min{a,b,c}(OD+OE+OF) < a·OD+b·OE+c·OF

:::;
2[BOC] + 2[COA] + 2[AOB]
2,6,.
2~.
•
3.4.53. (0:/2) 2:
l:asin (a/2)
L:asin ~ s.
s.
Proof: We first prove that in a triangle ABC, the following inequality holds:
11 (0:/2) :::: 11
TI sin (a/2) TI (1 - sin (a/2)).
(0:/2)). (*)
We observe that a, 0:, ß, "( are the angles of a triangle if and only if (7r
(71" - a)/2,
0:)/2 ,
(7r
(71" - ß) /2, (7r
(71" - "() /2 are angles of another triangle. Henee,
Hence, it is suffieient
sufficient to
prove
11
TI eosa
coso: :::;
:::: (1 11
TI
(1 -- eosa)
coso:)
But we have
2
(1
(1-- coso:)(l-
cos a) (1 - eos ß)(l - cos
cosß)(l ~() =
cos"() = 8 (a/2) sin 2 (ß
sin 2 (0:/2)
8sin /2) sin 2 h
(ß/2) = 2~2'
/2) =
h/2)
where Rand rare eircum-radius

circum-radius and in-radius of the triangle whose angles are
0: , ß,
a, ß, "(. Since
cos 0:
4R 2 COS eos ß cos
a COS eos "( = 2r 2 - I H 2 :::;
~ 2r 2r· 2 ,
where land H are the in-centre and ortho-centre of the triangle, we get the
result.
Now, we may write the inequality (*) in the form
ITsin(a/2)
TIsin(0:/2) ITsin (9)
TIsin(9)
TI eos( 0:/2)
IT cos( ~ TI
a/2) :::; (7r~a)"
IT cos (1f 4") .
U sing the identities
L sin(oo/2)
sin(a/2) +
1+ rr sin ( 7f ~ 00a)) ,
L eos(oo/2)
eos(a/2) rr eos ( 7f ~ 00a ),
we obtain
Ilsin(a/2)
Ilsin(oo/2) + Lsin(oo/2)
-1 + L:sin(a/2)
eos(a/2):::;
Il eos(oo/2):::; L
L: eos(oo/2)
eos(a/2) .
However using r = = 4R Il sin(oo/2)

sin(a/2) and s == 4R Il eos(oo/2),
eos(a/2), the 1eft side is simp1y
r / s. Thus, the inequa1ity takes the form
::.-<
r < _---:1 + L:sin(a/2)
-1 =+,---:L=-s_in---.:.(00-.:./---.:.2. .:. ) .
s -
S - L
L: eos(oo/2)
eos(a/2)
But observe that (s - a) sin(oo/2)

sin(a/2) = reos(oo/2).
reos(a/2). Thus, we get
L(s - a) sin(oo/2) + L Sin(OO/2).

sin(a/2) :::; s ( - 1 + Sin(a/2).
This reduees to
L asin(oo/2)
asin(a/2) ;::: s.
•
3.4.54. 3L:eosa;::: 2Lsinoosinß.
3Leosoo;::: 2L:sinasinß.
Proof: We begin with the inequa1ity
L."
""'
""' (eosa - eosß) 22 ;::: O.
(eosoo
Expanding this we get
L eos 2 a ;:::
00 L eos 00a eos ß·
But then
L eos 2 a ;:::
00 L eos 00a eos ,6 ===> ( L eos 00a)) 3 L eos 00a eos ß
2 ;:::
3 L eos 00
a eos ß :::; 2" L eos 00
3
===> a
===> 2 L eosoo
eosa eosß :::; L eosoo
eosa
122 Inequalities
===} L cos(a - ß) L cos(a + ß) :::; L cosa

+L
===} Lcos(a-ß):::; 2Lcos a
===} L cos(a - ß) - L cos(a + ß) :::; 3 L cosa

===} 2 L sina sinß :::; 3 L cosa.
We have used the weH known result: Lcosa:::; 3/2. (3.4.9)

•
3
3.4.55. max{ra,rb,rc } 2 2R .
Proof: The required inequality is equivalent to
{ lI I} >-.
3R
max - - , - - , - -
s- a s- b s- c
3R
- 2ß
This is furt her equivalent to
3Rr 3R
max { tan(a/2),tan(ß/2),tan{J/2)
tan (a/2), tan (ß/2), tan (1/2) } 2
2: 2ß == 2;.
This mayaiso be written in the form
(Lsina)max{ tan(a/2),tan(ß/2),tan{J/2)}
tan(a/2),tan(ß/2),tan(l/2)} 2~.
2:~.
However, we have
(L sina) max { tan (a/2), tan (ß/2), tan ("(/2) }

22: L sinatan (a/2) == 2 L sin 2 (a/2).
On the other hand,
3 1
Lsin 2 (a/2) 2-2Lcosa
~ - ~ (1 + 4 rr sin (a/2))
~-~ (1 +~)
2 2 R
r
1--.
2R
Since 2r :::; R, the result foHows. In fact we have proved more:
max{ra,rb,rc } 22R-r.
•
3.4.56. R - 2r 2
2: Wa -
W ha · .
Proof: We use
r = 4R II sin(a/2), Wa
Wa
4R sin ß sin ry cos(a/2),
=4Rsinßsill'Y
.
= sin
smß
.
ß + siwy
smry
(/2)
cos a ,
and
2R sin ß sin ry.
h a == 2Rsinßsin,.
The inequality to be proved is
1- 8 II sin(a/2) 2:: cos2sinßsin,

2 sin ß sin ry
(( ß - , )/ 2)\ (1- cos ((ß - ,)/2)).
ß - ry ,,/)/2)).
= cos ((ß -,)/2)

Let us put t = -ry)/2) and x == sin(a/2). Then tE (0,1] and xE (0,1).
We have to prove
f (x; t) = 2 (1 + t) x 2 - 4e x + (t + 2e - t3 ) :::: O.
Consider this as a function of x, say g(x). We observe that
g'(x) == 4(1 + t)x - 4t 2 , gl/(x) == 4(1 + t) > o.

Hence, 9 has the minimum at x = t 2 /(1 + t). But observe that
f ( ~.
l+t'
t) = t(l - tt)(t++13) -> o.
= 3)
2:: o.
Hence, min xx fex; t) :::: 2:: 0 for all x E (0,1) and t E (0,1].
O. Thus, fex; t) ::::
Equality holds if and only if t = 1. Equivalently x = 1/2. This is equivalent to
a = 7r /3 and ß = "ry, which corresponds to the case of an equilateral triangle .
•
3.4.57. yIS(S~
Js(s - a) Js(s - b) + m c ::; J3 s.
+ }s(s 8.
Proof: Using the AM-GM inequality, observe that
2 J (s
2} (8 - (8 - b) ::; s8 - a + s8 - b = c.
a) (s
Equality holds if and only if a = b. Thus,
4m~ = 2a 2 + 2b2 - c 2
(a + b)2 + (a - b)2 - c2
(a+b)2 - (b+c-a)(a+c-b)
(a + b)2 - 4(s - a)(s
a)(8 - b)
(a + b + 2}(s a)(8 - b)) (a + b -
2J(8 - a)(s 2J(8 - a)(s
2}(s a)(8 - b))
(a + b + 2}(s
2J(8 - a)(s (28 -
a)(8 - b)) (2S - (vs=a + JS=b)2)
28(28- (vs=a+JS=b)2).
< 2s(2s-
124 Inequalities
This implies that
Js(s - a) + Js(s - b) :s; hJs2 - m~.
On the other hand

(hJ S 2 - m~ + mcr :s; 3s 2 ,
(s2 -- 33mn
as this is equivalent to the inequality (s2 m 2 ~ n 2
~ O. Hence, the result
folIows. •
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], [AFEl}.
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 a, ,8, 'Y, be the areas of the three corner triangles so arranged
that 0 < a :s; ß :s; 'Y and let 8 = [DEF]. We prove the stronger statement:
8 ~ jCi(J. We normalise the areas such that [ABC] = 1. Let
BD DC CE EA AF FB
--
- == -- = = = y, -b- =Y , ==Zz,, -- ==
I I I
xx,
, Xx,
, -b-
-b-=Y' -b-=Y , -- zz.
.
a a c e
Let AX and FY be the altitudes drawn respectively from A and F on to BC.
...
B Y X
~
D
.~c
Fig.3.2
Then we see that

[BDF] BD ·FY BD FB I
[ABC] =
= BC· AX =
= BC . AB =
= xz .
This gives [BDF] = = xz'. Similarly, we obtain [CED] =

= yx', [AFE] =
= zy'.
Using these we get
J 1 - (xz' + yx' + zy')

1 - x(l- z) + y(l- x) + z(l - y)
1 - (x + Y + z) + (xy + yz + zx)
(1 - x)(l - y)(l - z) + xyz
x'y'z' + xyz.
Now we consider two cases : rr < 1/4 and rr 2':

2': 1/4. Suppose 1r < 1/4. If
J< vaP, then J < ~ = I< r < 1/4 and hence
1
1= a +"2.
+ ß + 1r + J < a + ß + 2·
This gives a + ß > 1/2 and in turn we obtain
1 1
2-2 << aa + ß :S-< 21 < 2·
2"1 < -.
2
I
2': vaP, when 1r < 1/4.

This contradiction proves that J 2':
Suppose on the other hand that 12':r 2': 1/4. In this case
J x'y'z' + xyz
> , ,
2 ((XX'yy'zz')
xx yy zz 1/2
,) 1/2
(a ß1)
1/2
2 ßr)
N·2fi
> N·
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
2': min{p(BDF),p(CED),p(AFE)},
p(DEF) 2':
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.
126 Inequalities
Prüüf: For angles x, y, z, let U = tan (x/2), v = tan (v/2), W =

= tan (w/2).
Then, we have the following results.
R
\ s
Yl ~\,\2
Fig. 3.3
(a) If x + y + Z = 1f, then uv + VW + wu = 1. This follows immediately from

the property of tan function.
(b) If PQR and PQS are two triangles with the common side PQ and if
LRPQ = Xl, LRQP = YI, LSPQ = X2, LSQP = Y2, Y2, then p(PQR) <
p(PQS) or p(PQR) = p(PQS) or p(PQR) > p(PQS) according as
UI VI < U2V2 or UI VI = U2V2 or UI VI > U2V2 respectively.
prüüf: It is easy to prove that
p(PQR) 2 p(PQS) 2
PQ 1- UIVI' PQ 1- U2V2
Observe that UI VI < 1 and U2V2 < 1. The result follows from comparison.
Suppose p(AFE) 2: p(DEF) and p(FBD) 2: 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(FBD). This only increases the perimeter of the triangle CED.
Thus, it is sufficient to prove that p(DEF) 2: p(CED) under the assumption
that p(AFE) = p(DEF) = p(FBD).
Using (a) and (b), we get
UUI+ UIU2 + U2U 1,

VVI + VIV2 + V2V 1,
WWI + WIW2 + W2W 1,
U2UI = UW, VI W2 = VW, V2UI kvu.
B D
D c
Fig.3.4
Fig. 3.4
Further we also have uv + vw + wu = 1. We have to prove that k ~ 1. We can

get the following relations:
kuv VW UW
Ul = --,
V2
Vl = -,
W2
Wl = -;;.
Substituting these in the first three relations, we obtain
kuv kuv
u ''---
-++---''UU2 2+
+U U2
2U 1,
V2 V2
VW VW
V . - +- . +
V2 V2V 1,
W2 W2
UW UW
W . - +- . W2 W2W + = 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
ku 2 v 2 (1 + w 2 ) (w + W2) + (W2 - v 2 w) (U 2 W 2 + U2WW2 -1 + WW2) = o.
We show that
(W2 - v 2 w) (U 2 W 2 + U 2 WW2 - 1 + WW2) + U 2 V 2 (1 + w 2 ) (w + W2)

=w(1+u 2 )(w2- v t
128 Inequalities
We have
'Ne
W2
((W2 2W
- V2W)
) ((U +U
U22W22 + WW2
U2WW2
2 - 1+ + WW2 + U22vV22 (11( ++ W
WW2)) + 2) ((w
w2) W + W2
W2))
= U22VV2(2 (wW+ Wz +U
W2)) + U22 W22 W2 + 2 WW 2 - W2
U2WW~ 2
+WW~
+ww~ + v 22 w - V2W22W2
= U2V w+
v2 W + W2 (U 22V22 + Vv 2W2 + W
w 2U
u 2) - 2V 2W2W2
2V2W2W2
2
WW 22 -- W2
+U2WW~
+U + WW~
WW 22 + V
v 22wW
W{ U2V
v 2 - 2VW2 (UV + VW
vw + WU)
+ w~ - 2VW2) ++ 2VW2 }
+(U22 + 1) (v 22 +
+(u
= w(1
w(l + Uu 2 )(W2
) (W2 - V)2;
we have used uv + vw + wu = 1. It follows that
(k - 1)u 2 v 2 (1 + w 2) (w + W2) + w(l + u 2) (W2 - V)2 = o.

V2Ul :::; vu. This is
The expression shows that k :::; 1. This implies that V2UI
equivalent to p(DEF) 2 2: p(CED) in view of (b). Equality eorresponds
corresponds to
k = 1. This fore es W2 = v and henee
forces hence VI = w. Using the symmetry we eonclude
conclude
U2 = W
U2 W,, WI
Wl = 'UU and V2 = U, UI = v. Looking at the expressions involving
U, Ul
Ul,Vt,Wt, U2,V2,W2, we eonclude
U,V,W, UI,VI,WI, D,E,F
conclude that D ,E,F are the mid-points of
BC, CA, AB respeetively.
respectively. •
3.5.3. Let ABC be an aeute-angled

acute-angled triangle and DEF be an aeute-angled
acute-angled
triangle
tri inscribed in ABC. Show that
angle inseribed
,\ :::; R(DEF) :::; {t,

j)"
where
,\ = min {R(AFE),R(BDF),R(CED)},
{R(AFE),R(BDF) , R(CED)},
fL
(t = max{R(AFE) , R(BDF) , R(CED)}.
max{R(AFE),R(BDF),R(CED)}.
R(XY Z) denotes the eireum-radius

(Here R(XYZ) XY Z.)
circum-radius of the triangle XYZ.)
Proof: The segment EF is given by EF = 2R(AF E) sin 0:

a whieh
which is equal to
2R(DEF) sin L.F DE. This implies that
2R(DEF)sinLFDE.
R(DEF) _ sm
sina0:
R(AF E) - sinLFDE
R(AFE) sin L.F DE·
can obtain
Similarly, one ean
R(DEF) _ sinß
R(BDF) - sinLDEF'
sinL.DEF'
R(DEF) sill"Y
R(CED) sinLEFD·
However, 0: + ß + "( == 7r == LFDE + LDEF + LEFD and henee at least one
of the ratios 0: : LFDE, ß : LDEF, "( : LEFD is ~ 1, and at least one of
these ratios is also ~ 1. Sinee all the angles are aeute, at least one of the ratios
R(DEF) : R(AFE), 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
A ~ R(DEF) ~ fL.
Remark: If r(XY Z) denotes the in-radius of triangle XY Z, it is true that
r(DEF) ~ min {r(AFE), r(BDF), r(CED)}.
See [5J. •
3.5.4. Let ABC be an aeute-angled triangle and D, E, F be arbitrary points
on the segments BC, CA, AB respeetively. Show that
22::: a l eos 0: ~ 2::: a eos 0:,
a',b',c' are the sides of of DEF (i.e., EF =

where a',b',c' = a l , FD =
= b' , DE =
= Cl).
Proof: Let BD =
= Xl, DC =
= YI, CE =
= xz, EA =
= Yz, AF =
= X3, FB =
= Y3.
Then
al > a- Y3 eos ß - X2 eos ,,(,

b' > b -
b' YI eos"( - X3 eoso:,
Cl
> C- Y2 eoso: - Xl eosß·
Sinee eos 0:, eos ß, eos"( are positive, the above estimates give
2::: a l eos 0: ~ 2::: a eos 0: - 2::: a eos ß eos "( .
However,
2::: a eos 0: =
= 2::: 2R sin 0: eos 0: =
= R 2::: sin 20: =
= 4R sin 0: sin ß sin ,,(,
and
2::: a eos ß eos "( 2R 2::: sin 0: eos ß eos "(

2R eos "(sin(o: + ,8) + 2Reosaeosßsin"(
2R sin "( eos "( + 2R eos 0: eos ß sin "(
2R sin "( ( eos "( + eos 0: eos ß)
2Rsin 0: sin ß sin "(.
130 Inequalities
Thus, it follows that
L a' cos 0: > L acos L acos ßcos 'T

0: -
4R sin 0: sin ß sin 'T - 2R sin 0: sin ß sin 'T

2R sin 0: sin ß sin 'T
1
2" Lacoso:.
Hence, 2 L a' cos 0: ~ La 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 Rand 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: R l = = AP, R 2 == BP, R 3 == CP; similarly
Tl, T2, T3 denote respectively the distances of P from BC, CA, AB. Now we
can develop a large number of inequalities involving R l , R 2 , R 3 , Tl, T2, T3 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 8s be its semi perimeter.
Prove that
8 < R l + R 2 + R 3 < 28.
Proof: Since AB < PA + PB, BC < PB + PC and CA < PC + PA, we get

by adding 8s < R l + R 2 + R 3 .
F E
B D c
Fig. 3.5
Extend AP, BP, CP to meet opposite sides BC, CA, AB respectively in

D, E, F. Using the tri angle inequality, we have BE < BC + CE. Adding
AP both sides and using BE = BP + PE, AP < PE + AE, we obtain

BP+PE+AP < BC+CE+AP < BC+CE+PE+AEsothat BP+AP <
BC + CE + AE = BC + CA. Similarly, we obtain AP + C P < BC + AB and
BP + C P < AB + AC. Adding these three inequalities, we get R 1 + R 2 + R 3 <
AB+BC+CA=28.
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{PA, PB, PC} + PA + PB + PC < 28. This depends
on the following property: If P is any interior point of a convex quadrilateral
ABC D, then PA + PB< AD + DC + C B. The proof is easy. Extend
AP to meet the quadrilateral in Q and suppose, for example, Q lies on DC.
Then PA + PB < PA + PQ + QB ~ AQ + QB ~ AQ + QC + CB ~
AD + DQ + QC + C B = AD + DC + C B.
Let DEF be the medial triangle of ABC. This divides ABC into 4 regions.
Each region is covered by at least two ofthe three convex quadrilaterals ABDE,
BCEF, and CAFD. If, for example, P lies in quadrilaterals ABDE and
BCEF, then we have PA + PB< AE + FD + DB and PB + PC < BF +
FE + EC. Adding these two we get PB + (PA + PB + PC) < AE + ED +
DB+BF+FE+EC = AB +BC+CA.
As an interesting corollary to this, consider an acute tri angle and let P be
its circum-centre (which lies inside ABC). Then PA = = PB
PB = = PC == 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 a + sin ß + sin, > 2.
•
3.6.2. Let ABC be a triangle in which a > b > c. Let P be any interior
point and suppose AP, BP, CP meet the opposite si des in D, E, F respectively.
Prove that
PD + PE + P F < a.
Proof: Draw P X parallel to AB, PY parallel to AC, X K parallel to CF and

Y L parallel to PE. 0 bserve that a is larger than each of AD, BE, CF. Since
XPY is similar to BAC, we get XY > PD. Similarly, we obtain BX > XK =
PF and CY > YL = PE. Adding these we get
= BC == a.
PD +PE+PF < XY +CY +BX =
•
3.6.3. Let P be a point inside a tri angle ABC and let D, E, F be feet of
e
perpendiculars from P on to the sides Be, A, AB respectively. Prove that
R 1 + R 2 + R 3 ~ 2(rl + r2 + r3).
(This is known as the Erdös-Mordell inequality.)
132 Inequalities
Proof: Suppose Land !vI are arbitrary points chosen on the sides AB and
AC. Complete the parallelograms LAPS and MAPT. Then LMTS is also a
parallelogram and [LAPS) + [MAPT) == [LMTS).
F E
B c
s
Fig. 3.6
Observe that [LAPS) = LA· PF, [MAPT) == MA· PE ,[LMTS) ::; LA!·
AlT == LM . PA. Here, the points L, 114. can be chosen on the extended lines
AB and AC. Choosing L, 114. such that AM = AB and AL = AC, we get
AC . P F + AB . PE::; LM . PA.
But, the triangle ALM is similar to AC Band hence LM = C B. Thus, we get
bP F + cP E ::; aP A. It follows that
b c
PA> -PF+ -PE.
- a a
Similarly we get
c a a b
PB 2: bPD + -,;PF, PC>- -PE
c
+ -PD.
c
Adding these inequalities we get
PA + PB + PC >
>
(~+ n PD
2(PD + PE + PF).
+ (~ + ~) PE + (~ + ~) P F
Alternate Proof: Observe that AF P E is a cyclic quadrilateral. This gives

FE = AP sin A. Using eosine rule, we also get
FE 2 PE 2 + PF 2 - 2PEcotPF· COSL.FPE
PE 2 + PF 2 + 2PEcotPF· cosa
( PE siwy + P F sin ,8) 2 + (PE cos r - P F cos ß?
> (PEsin r + PFsinß)2.
It follows that
PA = FE
FE> > PE sin
siwy'/ + P F sin ß .
sino - sino
Similarly we get
, + P F sin 0
PD sin '/ PC> PE sin 0 + PDsinß
PEsino PD sin ß
PB > . , PC ~ . .
- smß sm,
sin '/
Adding these we get
PA+PB+PC > PD (s~n

(s~n,'/ + s~n
s~nß)
ß) + PE (s~n
(s~n,'/ + s~n
s~no)
0 )
sm
smßß sm,/
sm, sm
smo0 sm
sm,'/
+PF (s~no + s~nß)
smß smo
> 2(PD + PE + PF);
we have used the AM-GM inequality at the end.
Here are so me interesting eonsequenees of the Erdös-Mordell inequality.
(i) Take P = 0, the cireum-eentre of ABC. Then PA = PB = PC = R
and PD = ReosA, PE = Reosß and PF = Reos,/. Reos"(. Thus, we get (see 3.4.9)
L coso ::; 2'3

(ii) Taking P = H, the ortho-centre of ABC, we get PA = 2Rcoso,
PB = 2Reosß, PC = 2ReosT
2Reos,. Henee, we obtain PD = 2Reosßcos,/,
2Reosßcos" PE =
2Rcos,coso, and PF == 2Rcosocosß. Thus, we get
2Rcos,/coso,
2 (cos 0 cos ß + cos ß cos '/

, + eos
cos '/
, cos 0) ::; (cos 0 + eos ß + cos
cos,).
'/ ) .
Using the inequality in (i), we obtain

3
eos 0 cos ß + cos ß cos '/
, + eos
eos,'/ cos 0 ::; 4"'
(iii) Taking P = I, the in-centre, the inequality takes the form
AI + BI + CI ~ 6.
Using AI = rcosec (0/2), ete., we get
L cosee (0/2) ~ 6.
(iv) Let P = K, the symmedian point. If AK meets BC in D' D ' , then it is
BD ' / D'C = c2 /b 2 and similar ratios for other interseetion points.
known that BD'
Using this and Stewart's theorem, we ean eompute the symmedian length:
I 2bc
AD = -
2 b 2ma'
+c
134 Inequalities
This implies that

2bc
AK = 2 b2
m a,
')2ma,
a + +c
and similar expressions for BK, CK. Using these we can also get the lengths
ofperpendiculars KL,KM,KN on to BC,CA,AB:
2a
KL b2
a
2
+ +c 2 [ABC],
?
2b
2b b2 [ABC],
[ABC]
KM 2 -
a2 + c2
+ b22 +c '
2c
KN [ABC]
2c2 b2 [ABC].
2
+-
a2 + + c2
b2 +c .
In this case, the Erdös-Mordell inequality takes the form

mamb mc s
-+-+->-.
abc - R
•
1 1 1
3.6.4. L->2L->4L-·
rlr2 - R1rl - R R 1 2
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 Ri, R~, R~ and the distances
of P from the sides B' C', C' A', A' B' by ri, r~, r;, then the Erdös-Mordell
inequality applied to triangle A' B' C' gives
R~ +R~ +R; ~ 2(r~ +r~ +r;).

If D, E are the feet of perpendiculars drawn from P to BC, CA respectively,
then DE = R 3 sin ,. Using the fact that D and E are the mid-points of PA'
and PB', we get A' B' = 2R3 sin ,. Now let F' be the foot of perpendicular
from P on to A' B'. We then have
PF', = 2[A'~B']
2 [A'PB']
PF = A'B' .
On the other hand
[A'PB'] ~ PA' . PB' . sin LA' PB'

2
1 .
sm(Jr -,)
"2(2rd(2r2) sin(Jr
2rl r2 sin,.
r;
This gives r; = 2rlr2/ R 3 • Similarly, we may obtain
r~ = 2r2r3/Rl' r~ = 2r3rI/R2.
Moreover
R~ = 2rl, R~ = 2r2, R; = 2r3.
Using these we get
1 1
L
L :rlr2
rl-r 2>-2:
2 LR
:- 2L .
1 r l·
R1rl
process adopted to the point P and the triangle A'
The same proeess AI B'
BI C'
CI gives
1 1
L:->2L:-.
R1rl - R 1R 2
•
3.6.5. L:R1rl 2: 2L: rlr2.
Proof: We refer to Fig. 3.6. Taking L = Band M = C, there, we obtain
BA· PF + CA··PE:::;
BA ·PF+CA PE::; BC ··PA.
PA.
reduces to cr3
This reduees + br2 ::; 1 .. This may be written in the form
:::; aR 1
c b
R 1rl 2: -r3 rl
a
+ -r2
a
rl·
Similarly, we obtain
a c
R 2r2 > i/I
//1 r2 + 1/3 r 2
b a
R 3r3 > -r2 r3 + -rlr3
-rl r3··
c c
Adding these three inequalities, we get
LR1 r l
L:R1rl > L:
2: l r 2 (~+~)
L r lr2
> 2 L:rlr2;
Lrlr2;
we have used the AM-GM inequality. Equality holds if and only if ABC is
equilateral and P is its eentre.
centre.
Consequences:
(i) Taking P = in-cent re of ABC, we have rl =

= I, the in-eentre = r2 =
= r3 =
= rand
R 11 =
= r eosee (0:/2), R 2 =
cosec (a/2), cosec (ß
= r eosee /2), R 3 =
(ß/2), = r eosee (')'/2).
cosec ("(/2).
We obtain
L:
L eosee (0:/2) 2: 6.
cosec (a/2)
136 Inequalities
(ii) Taking P == 0, the circum-centre of ABC, we get
L cos Oea 2: 2 L cos Oea cos ß·

(iii) If P == H, the ortho-centre, then
R l1 = 2R
2Rcosa,
cosOe, R2 = 2Rcosß, R3 = 2Rcos,!,
and
Tl = 2Rcosßcos,!, T2 = 2Rcos'!cosOe,
2Rcos'!cosa, T3 = 2RcosOe
2Rcosacos,ß.
cos,ß.
The inequality we obtain is L cos Oea ::; ~. (See 3.4.9.)

(iv) If we take P == C, the centroid, then the inequality takes the form
L maha 2: L hah b•
Adopting the method developed in 3.6.4, we obtain
LRR I 2 2: 2 L RITl·
Repeating this process once more, we also get
L-2:
1
2 2:-.
R
1
Tl I
•
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. PD + PE + PF 2: 6;
AP BP CP
2. PD . PE . P F 2: 8;
AD BE CF 9
3.
AP + B P + CP 2: 2";
AD BE CF
4. PD . PE . PF 2: 27;
PD PE PF 3
5.
PA + PB + PC 2: 2"
Proof: Let us take

BD x z AF CEy
DC y' x' FB EAz
ax ay cy cz
Then we see that BD = = --,
- - , CD = = --,
- - , AF =
= -
-- - and FB =
= --.
--.
x+y x+y z+y y+z
Using Menelaus' theorem, we obtain
AP BC AF x+y
PD CD FB zz
BP z+x
z +x CP y+z
y + z. Adding
and similar expressions: PE -y--' PP
C
--y--' PFF =
= -x-'
x Adding these
these we
we get
get
PE
(~+;) +
+ (~+~) +
AP BP CP
PD + PE + PF + (~+~)
> 2 + 2 + 2 = 6.
Similarly,
AP BP CP (x +
+ +
+z)(z +
y)(y +x)
PD PE PF xyz
> 2vxy . 2 # . 2,jZX = 8.

xyz
AD AP x+y+z AD x+y+z
PD = 1 + PD = z ' AP x+y
Thus,
AD BE CF
AP + BP + CP (x+y+z) 1- + y+z
x+y
x+y
-1
- + -1)
(---+--+-- -
z+x
y+z z+x
9
> 2
And
AD BE.
AD. (x + y + z)3 > 27.
BE CF = (X+y+z)3
- .-._= >27.
PD PE PF xyz -
Finally
PD PE PF
PA + PB + PC
-
z
-
x+y
+--+--+--
-
x
-
y+z
+ -
x
y+z
-
(x+y+z) 1 y+z
1 z+x
(--+--+
(x+y+z)(_l_+_l_+
x+y
1)
1 )-3
-3
9 3
>
> -2 - 3 =-.
2 - 3 = 2'
2
•
138 Inequalities
3.6.7. Prove
(i) '"' h a > 9·,

~ rl -
(ii) hahbhc 2: 27 r lr2 r 3;
(iii) (ha - rl)(h b - r2)(hc - r3) 2: 8rlr2r3;
(iv) min{ha,hb,h c}:=::; rl +r2 +r3:=::; max{ha,hb,hc};
rl 3
(v) L
(v) ' " h 1'1
_ rl 2:
~ h a - 1'1 - 2
~.
> "2.
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: FB == y: x.
B T L D c
Fig. 3.7
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 PD Land ADT, we have
ha AT AD x + y + zZ
rl
1'1 PL PD x
Similarly, we may obtain
hb
an - -
+ YY ++ zz .
x+y+z andd hhcc =_ xx +
r2 Y r3
1'3 zZ
Now
2:: hTl
a
(x+y+z)(~+;+~)
(x+y+z)(~+t+~)
> 9.
This proves (i). Similarly,
ha
-0-0--
h c _(x+
h a • h b . hc = (X+y+z)3 > 27 ,
V+z)3>27
_
Tl
rl r2 T3
T2 r3 xyz -,
giving (ii). We also observe
(ha - Tl) (h b - T2) (he - T3)
Tl T2T3 (~~ - 1) (~: - 1) (~: - 1)
= (z:V) (x; z ) (V:x)

z+y x+
z+V x+zz y+x
= #.yXz.y'yX
yYZ·VXZ · y'yX
> 2·2·2 = 8,
proving (iii). Using the same expressions, we also get
""' Tl
6-';=l.
a
This implies (iv). Finally, we have
Tl PD xx
ha - Tl AP y+z''
V+z
r3 ..
. ·1ar expreSSlOns
. for h T2 ' h T3 . Th b tam
c r2
and
an d Slml
similar expressions tor Thus,
US, we 0obtam
bb -- T2
r2 c - r3
T3
'"'
" rl _
" ' -:--T_l '"' _x_ = (x+y+z) (_1_+_1_+_1_) -3 > ~-3 = ~.
= ""'
6~ h a - Tl
rl 6~ VY+z y + z zz+ x xx+ V
Y 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 + 2J3)
3(3+2V3)
--+-+->-----'-.,-:-------'-
+ - + - > --'----'-
TA
rA TB
rB rc -
TC R +Tr '
where TArA,, TB
rB,, TC
rc denote the in-radii of triangles PBC, PCA, PAB respec-
tively.
140 Inequalities
Proof: We refer to the figure Fig. 3.7. We have
a·PL PL PD
rA = a+PB+PC' ha AD'
AD·
Thus, we get
1 a+PB+PC AD
rA a· h a PD
Suppose BD : DC = z: y, CE : EA = x: z and AF: FB = y: x. Then
AD x+y+z BE x+y+z CF x+y+z

PD x 'PE y 'PF z
We hence obtain
PA 2 = (y + z)2 AD 2 .
(x+y+z)2
Using Stewart's theorem, we calculate AD;
AD 2 == (y + z)yb 2 + (y + z)zc2 - yza 2

(y + z)2
Thus, we obtain
PA 2 = (y + z)yb 2 + (y + z)zc 2 - yza 2

(x+y+z)2
= b2
If we use the standard formula a 2 = + c2 - 2bccosa, this expression reduces
to
PA 2 = Y 2b2 + z 2 c2 + 2bcyzcosa
(x+y+z)2
PA 2 == (. h 2a)2 { (Z cot r - y cot ß) 2+ (z + Y) 2} .

(z
x+y+z
Similar expressions may be obtained for P B 2 and PC 2 :
z cot r) 2+ (x + z) 2} ,
2
P B 2 =.
= ( h b)2 { (x cot a -
x+y+z
PC 2 = (. h 2C)2 { (y cot ß - x cot a) 2+ (y + x) 2} .

x+y+z
Observe that
PB 2:
PB> hh bb )
(x+z,
(x+z), PC >
2: hc (y + x) .
- x+y+z - x+y+z
Thus, we obtain
1 x + Yy + Z { a + PB + PC }
rA 2ß
2~ x
(x + Yy + z)a + hb(x + z) + hc(Y + x) .
> 2ßx 2ßx 2ßx
2~x 2~x 2~x
Similarly, we rnay
Sirnilarly, may obtain
> (x
-1 >
~ + y + z)b
(x+y+z)b + + x) +
hc(Y+x)
+ hc(Y + --'--------'--'-
ha(z+y)
_ha--,-(z_+_y,,--,-)
rB --
TB 2ßy
2~y 2ßy
2~y 2ßy'
2~y'
and
~1 > (x+y+z)c ha(z+y) hb(x+z)
->
rc -
TC
--,----+
2ßz
2~z
+ 2ßz
2~z
+ 2ßz
2~z
.
Adding, we have
111
111
-+-+- > x+ y + (~
x+y+z(a - z + -~b + ::) c) 1,,",
2: h ((z + Y))(1
- + -~ ~ h a (~ + -~1))
z y -
rA
TA rB
TB rc
TC ,. 2ß
2~ x y z 2ß
2~ z Y
y/
> x+y+z,,",~
> x+y+z",~ ~"""~"'h
2ß
2~ ~xx + ß
~ ~~~ ~a'
a·
However, we know that (see 3.4.38)
ha + hb
ß
+ hc = 2 L ~ > 3v'3 .
a-R+r
x + Yy + z ""'
'" ~ > ~
~(abc)1/3
(abc) 1/3
2ß
2~ ~ x - 2ß
2~
and (3.4.28)
2
2
(abc)2 ~
(abc)
(4ß)3
(~r
v'3
v'3~ :::::
Using 3.4.47, we can prove that v'3ß ~ (R+r( Cornbining all these, we get
(R+T)2. Combining
~1 +~
1 +~
1 > 3-,(-,3
3(3+2v'3)
_+_2v'3_3...:...)
-+-+->--'------'-
rA
TA rB
TB TC --
rc R
R + rT ''
•
3.6.9. Let P be a point in a triangle ABC. Then
2: R~:2 ~2 1.l.
L
Equality holds if and only if either P
P is one of the vertices or P
P = H
H,, the
ortho-centre of ABC.
142 Inequalities
Proof: Let us consider the vertices of ABC to be points in the complex plane;
let us denote A, B, C by complex numbers Zl, Z2, Z3 respectively. Let us write
z for P. N ow consider the function
() _~
g(z) " (z (z -- Zl)(Z
zI) (z -- Z2)
9 z - L..-
- L..- (Z3 - Zl) (Z3 - Z2) .
= 9 (Z2) == 9 (Z3) == 1. Hence, g( z) == 1
This is a quadratic polynomial and 9 (Zl) =
for a11 z. Thus, we have
1 = Ig(z)1 :::; I: Iz - zlll z- z21 =~ R1R2

IZ3 - zl11z3 - z21 L..- ab .
Equality holds if and only if z = Zl = Z2 or z == Z3 or

or z =
arg
(Z-ZI)(Z-Z2)
= arg
(z - Z2) (z - Z3)
(Z3 - Zl) (Z3 - Z2) (Zl - Z2)(Zl - Z3)
(z - Z3)(Z - Zl)
arg
(Z2 - Z3)(Z2 - Zl)'
The last case corresponds to P == H.
Consequences:
(i) Let us take P = I, the in-centre. Then
R 1 = rcosec (a/2),
R R = rcosec (ß/2),
R2 = = rcosec b/2).
R3 =
R
Then it is easy to calculate
~RIRz
"
L..-R
-
1 R2
-b- -=
a
ab
1~ 1" ..
= -- L..- asm (a/2).
s
We hence obtain
I: asin(a/2) ~ s.
(For a direct proof see 3.4.53.)
(ii) Taking P == G, the cent re of gravity, we have R 1 == 2m a /3, R 2 == 2mb/3
and R 3 = 2m c /3. Thus, we obtain the inequality
~mamb > ~~.

"mamb
L..- ab - 4'
4
= R z2 == R 3 = Rand
(iii) Taking P = 0, the circum-centre, we see that R 1 =
the inequality is
1 1
- > -
ab - RZ'
R2'
or a + b + c ~ 46./ 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 2 CE ~ AF ~
DC b~'
2 ' EA c&'
2 ' FB
F B a~.
2'
Using Stewart's theorem, we can calculate R 1 ; we get

2bc
R1 = L:a2ma
2ma ,,
and similar express ions for R 2 and R 3 . The inequality in this case is
Lc2mamb~ (a 2 +b2 +c2)2
(v) When P = f1 j , the first Brocard point, the sine rule gives
b. R c. R C.
R 1 = -.--
- . - slnw, 2 = --=---ß slnw, 3 = - . - smw,
smoo sm sm,
where w is the Brocard angle. We then get
L - -b >- TITIsinoo
~-->
asinoo
'"' a sin 00 sin
sin w
. 2
00
Using the known fact that w:S 7r/6, we have sinw:S 1/2 and hence
L --
asinoo
-b-
-> -
4II'smoo.
•
3.6.10. Let P be a point in a triangle ABC. Then
'"' rlr2 < !!.

~ ab - 4' 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: FB = Y : x. An easy computation shows that
xh a yhb zh c
rl =
= -- , r2 = - - - - r3 = - - - -
x+y+z
r2
x+y+z +Y+ z
r3 = xx+y+z
Thus, we get
r2
rl r2 xy sin 2 ,
ab (x + y + z)2'
144 Inequalities
The required inequality is

1
1
)2 {
(x+y+z)2 ...
)) ')
{ xysm-,+yzsin-0:+zxsin } 1
xysin-, +yzsin-a + zxsin2 ß <::;4'
-::;-.
2}
x+y+z 4
This simplifies to
LX2 + L2xycos2, ~ O.
Writing
cos
cos2a
20: cos(2ß+2,)
cos 2/3 cos 2, - sin 2,3 sin 2"
the inequality reduces to
(x + ycos2, + zCOS2ß)2 + (ysin2,- zsin2ß)2 ~ 0,
which is trivially true. Equality holds if and only if

x y z
2a
sin 20: sin 2ß sin 2,
This corresponds to P == 0, the circum-centre.

Consequences:
= G. Then rl == h a /3, r2 == h b /3 and r3 == h c /3. Then we have
(i) Take P =
L rlr2 =
ab
= .! L
9
hah b =
ab
= .! L
9
a.
sin 2 0:.
Thus, the inequality is
- ~
" sin 22 0:
'L..t a < ~.
..
4'
s:
or equivalently a 2 + b2 + c22 <::; 9R 2 •
= I. Then rl == r2 == r3 and the inequality takes the form
(ii) Let P =
1 1
-<-.
ab - 4r 2 '
-::; R.
This is equivalent to 2r <::;
(iii) Consider P = H, the ortho-centre. Then
= 2Rcosßcos"
rl = = 2Rcos,coso:,
r2 = 2Rcos,cosa, = coso:cosß.
r3 = cosacosß.
Using these values we may write the inequality in the form
L in220:
LSsin a <::;
-::; ~ rr tano:
tana = ~ LLtana.
tano:.
(iv) If P is the symmedian point, then
rl a2
ha a 2 + b2 + c2 '
and similar expressions far r2
T2 and r3.
r3. We thus get
L rlr2 = 12.6. 2
ab (2:: a 2 )2·
The inequality now is
La2 ;: : 4V3.6..
(v) Suppose P = fh, the first Brocard point. In this case
b . 2 c . 2 a . 2
rl
Tl . - - sm w,
-.-sm
= -
sm 0:
r2
T2 = -:--ß
sm
sm w, rl
Tl -.-sm
= -.--
sm,
sm w.
We obtain
sin4 w < rr sin 2
0:
- 4 2:: sin 2 0: sin 2 ß·
Equivalently
. 4 1 1
sm
sm w::::;
w::::; 16R2 (I:1/a
16R2 -(-- ).
2:: 1/a22 ) '
However using Cauchy-Schwarz inequality and 3.4.38, we get
2: -aa11 ;::::3
>
22
1( -1)2
-3 2: ~
-1( 1)2 ;>: : ~99 ,~.
a -4(R+T)2'
Hence, the inequality is

2 R+r
R+T
sin w:::; (jR.
(jR'
2T :::; R.
This is better than sin w :::; 1/2 in view of 2r •
r;:
3.6.11. Let P be a point in a triangle ABC. Prove that
(R 1 + R2 + R3 4V3 [ABC].
Proof: Draw AT parallel and equal to PB; BR parallel and equal to PC; and
C Sparallel and equal to PA. Then the area of the hexagon AT B RC S is equal
to 2 [ABC] , and perimeter of ATBRCS is 2(R 1l + R 2 + R:3 ). But among all
hexagons of fixed areas, a regular hexagon has the least perimeter. If a regular
146 Inequalities
hexagon is inscribed in a circle of radius R, then its perimeter is 6R and its

area is 3V3R 2 /2. Thus, we obtain
2( R 1 + R z + R3 ) ;::: 6R = 6y(2K
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 [ABC]
2 (R 1 + R 2 + R3 ) 6
;:::
3V3
This simplifies to
(R 1 + R 2 + R3 f ;: : 4V3 [ABC].
•
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
[DEF] :S ~ [ABC],
and equality holds if and only if P coincides with the circum-centre of ABC.
Proof:
A
B D c
Fig. 3.8
Let PD =
= Tl, PE =
= T2 and PF =
= T3. Then
[DPE] :t
1
1 T2
.
sm LD PE
1
"2T1T2 sin (7r -,,)
1 .
"2T1 T2 sm"
1
4RcTIT2,
and similar express ions for [EPF], [FPD] may be obtained. Adding these,
one obtains
1
[DEF] 4R (crl r2 + ar2r3 + br3rI)
L
abc ""' rlr2
rl r2
4R~ ab
4R
1
~[ABC],
< 4 [ABC] ,
using (3.6.10).
Consequences: Using the concyclicity of A, F, P, E, it is easy to get EF =

APsinA. Similarly, we get FD = BPsinB and DE = CPsinC. If R I , SI
and r l denote respectively the circum-radius, semi-perimeter and in-radius of
triangle DEF, then
EF·FD·DE R I TI R l1 sin 0:
RI
4 [DEF] L r lr2 c
SI
R 1 sin 0: + R 2 sin ß + R 3 sin 'f
2
rl
[DEF] L rlr2 C
SI 2R L R l1 sin 0: .
Now the inequality SI ;::: 27(r l )2 gives
(L R 27( L rlrzc)
2
l sino: f
> --'------'---
4
- 4R2 ( L R 1 sin 0: )
This simplifies to
3V3
R L
2
(
.LRlsino: ) ;::: r lr2 c .
Using a = 2Rsino:, etc., this may be written in the form
(LRlsino:f;::: 6V3L r lr2 sin "(.
However,
(R 1 sino:)2 = EF 2 r 22 + r 32 - 2rlr2 cos (

Ir - )
0:
r 2 + r 3 + 2rlr2 coso:,
2 2
148 Ineq'llalities
r
and similar expressions for the other two. Thus, we deduee
(2:R1
'2
6J3I>lr2 siwy <
::; (2: R 1 sinn)
sinn
2
< 32: (R 1 sin 00 ) 2
32:
32: (r~ + r~ + 2r2r3 eosn)
eoso:)
6( 2: rf + 2: r2 r 3 eosn),
eoso:) ,
01'
J32: rl r2 sin { ::; 2: ri + 2: r2 rr33 eos n.
0:.
Suppose P = I, the in-eentre of ABC. Then rl = r2 = r3 = rand henee
J3 r 2 ( sin 00 + sin ß + sin {) ::; 3r 2 + r 2 ( eos 00 + eos ß + eo8/') ,

01' whieh is the same as
J3 ( 2: sin A) ::; 3 + 2: eos A.

If we ehoose P = 0, the eireum-eentre, then rl = Reosn, r2 R eos ß,
Reosß,
= R eos { and henee,
r3 =
0: + 3
J32: eos 00 eos ß sin { ::; 2: eos 2 00 II eos n.
Using the standard identities 4eosneosßsin{
4eoso:eosßsin{ sin20: + sin2ß -
= sin2n sin2{, ete.,
00 = 22 + 22 TI eos 00,
and L sin 2 00 00, this reduees to
sin 200
200 + sin 2ß + sin 2{ ::; ~ (( sin 2 00
2{ ::; 00 + sin 2 .8 + sin 2 r).
•
3.6.13. (AMM, 1978) Let G denote the eentroid of a triangle ABC and let
() = LGAB, p = LGBC and v = LGCA. Prove that
. (). . 3
sm + sm ,..11 + sm v -< -.
2
Proof: Observe that

. , 6 ,6.
. ,6. ,6.
sin(}
sm(} = --, sinp = --, sinv =
= -b-.
b~c.
C~a a~b ~c
{ l+I- - + I}
Thus,
sin(} + sinp + sinv =,6. - - -b- .
C~a a~b ~c
( sin e + sin jt + sin v) 2 :S ß 2( 2: :2 ) ( 2: ~ ~ )
ß
2 (2: a2b2 ) (2:m;m~)
_ .
(abcmambmc)
166. 2 = 22:
Using the standard identities 16ß 22:: a 2 b22 -- 2:
2:: a\ 4m; = 2b22 + 2c22 -- a2 ,
etc., the above inequality reduces to
4 2 P
9 (sin eB + sin J.Ljt + sin v) :S:::; Q'
where
P = (22:a2b2 - 2:a4) (2:a2b2r,
Q =. a 2 b2c2 (2a 2 + 2b 2 - c2 ) (2b 2 + 2c2 - a 2) (2c 2 + 2a 2 _ b2 ).
Thus, it is sufficient to verify the inequality P / Q :S:::; 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 = = a22,, y == b2, Z == c22 ,, say, Q - P == f(x, y, z).
Thus,
f(x, y, z) == xyz(2x + 2y - z) (2y + 2z - x) (2z + 2x - y)

- (2xy + 2yz + 2zx - x 22 -- y2 _ Z2) (y2 + 2yz(
It is easy to verify that f(y,y,z) == 0 and fx(y,y,z) == O. 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 -

z)(z - X)]2 == R ~ 0 and hence P:S = y == z
P:::; Q. Equality holds if and only if x =
= b == c. Thus, equality holds if and only if ABC is
which is equivalent to a =
equilateral.
A Generalisation
The above inequality may be written in the form
~+~+~<3.
mb rn c ma -
150 Inequalities
However, a more general inequality holds. Let (h l1 , h 2 , h3 ) be any permutation

of (ha, h b , h c ). Then the inequality
~+~+~<3
~+~+~:::;3,
- ,
ma mb mc
m
is true. Suppose a :::; b :::; c. Then it is easy to see that h a 2: h b >

2: h c and
ma 2: mb 2: m c · Hence, the rearrangement inequality implies that
ha hb hc hl h2 h3 ha hb hc
- +- +- <-+- +- <-+- +- .
m amb m - m mb m - m
c mb ma c c a
~+~+~<3.
mc mb rn a -
Again the Cauchy-Schwarz inequality gives
(Lh~) (L ~~),
ha
(- hb hc ) 2
mc
+mb
- +m- <
-
a
and the rest is as in the early part.

•
3.6.14. Prove the inequality
L- r2 -
- - - < 1 < -1 L ---.
+ r3 R l1
r2 + 2R l1 + T3
T2 r3 - - 3 r2 + r3
Find conditions on P and triangle ABC for equality to hold in the above
inequality.
Proof: Let PD, PE, PF be the perpendiculars from P on to the sides BC,
CA, AB respectively. Thus, AP = R l1 , BP = R 2 , CP = R 3 , PD = rl,
PE = r2 and P F = r3. It is easy to see that a· R l1 2: b· r2 + c· r3, with equality
if and only if AP is perpendicular to BC. If we reflect P in the bisector of
L.BAC, we also get a· R l1 2: b· r3 + c· r2. Adding these we get
b+c
R l1 2: ~(r2 +r3).
Equality holds if and only if AP is is perpendicular to BC and bisects L.BAC.

This gives
1 s
R1+"2(r2+r3) 2: ~h+r3),
Rl+"2(r2+r3)
where s is the sem i-perimeter of the tri angle ABC. Dividing by r2 + r3, we
obtain
Rl 1 hl
--+->-
r2 + r3 2 - 2r'
where r is the in-radius and h 1 is the altitude from A on to BC. This simplifies
to
r2 + r3 < !....
r
----~----< --.
r2 + 2R 1 + r3 - h 1 .
Using
1 1 1 1
-+-+-=-
--+--+--=-
h h h r
1 2 3
we get the left hand side of the required inequality. We also observe that
R b+c
- -1 > - - .
r2 + r3 - 2a
Hence, we get
"
'"'R~
~-->~
1
~ r2 > "'"'+ r3 - ~ (~
(-+-
b a)
2a + ~)
2b ->3
> 3,
wh ich gives the right hand side. Equality holds if and only if ABC is equilateral
and P is its centre. •
3.6.15. Prove that

"
~ R 21 sin 2 a <
_ 3~
2
" r l'
and equality holds if and only if Pis the symmedian point of ABC.
Proof: Consider the cyclic quadrilateral AEPF. Ptolemy's theorem gives
Ap·EF r3AE + r2 AF
R 1 sin f}z . R 1 cos 01 + R 1 sin 01 . R 1 COS O2
Ri sin (0 1 + OO2 )
Ri = Ri sina,
where L.PAB = O2 and L.CAP = 0dSee Fig. 3.9.)
R3
E K F
B D c
Fig.3.9
Fig. 3.9 Fig. 3.10
Thus, FE = R 1 sin a. This gives
LRi sin 2 a = EF 2 + FD 2 + DE 2 •
152 Inequalities
Consider the triangle DEF and the point P inside it. We show that
DE 2 + EF 2 + FD 2 :::; 3(PD 2 + PE 2 + PF 2), (3.9)
valid for any tri angle DEF and any point P inside it. Let K be the mid-point
of EF. Join PK and DK. (See Fig. 3.10.) Appolonius' theorem applied to
the triangle P EF gives
- ~y'2PE2
PK -2~J2PE2 + 2PF2 - EF2 .
Similarly,
- ~y'2DE2
DK -2~J2DE2 + 2DF2 - EF2 .
DK ~ PK + KD, with equality if and only if P lies on DK. Thus,
But DK:::;
y'2DE2 + 2DF2 - EF2 :::; 2PD + y'2PE2 + 2PF2 - EF2.

Squaring and simplification gives
2DE 2 +2DF 2 < 4PD 2 +2PE2 +2PF 22

+PDJ2PE2
+PDy'2PE2 + 2PF2 - EF2.
Summing over the cyclic permutation of D, E, F, this leads to
DE 2 + EF 22 + FD 2 < 2(PD 2 + PE 2 + PF 2 )
+ ~ PPDDylr-2-P-E---C2-+-2-P-F---C2-_-E-F-"--2.
J~2-P-E-:2-+-2-P-F-2:--_-E-F--'--2.
If DE 2 + EF 2 + FD 22 :::;
~ 2(PD 2 + PE 2
2 + PF 2), then (3.9) is trivially valid.
Otherwise, we have
o < DE 2 + EF 2 + FD 2 - 2(PD 2 + PE 2 + PF 22 ))
~PDJ2PE2 + 2PF2 - EF2
< ~PDy'2PE2
< LPD2 )
((LPD2r/2
1/2(
(4 )1/2
LPD2 - LLDE2r/2
4LPD2 DE2
Setting 2::= P D 2 = x2 and 2::= D E 2 = y2, the above relation is

1/2
o :::; y2 _ 2x 2 :::; X ( 4x 2 _ y2 ) .
~ 3x 2 , which is
~ O. It follows that y2 :::;
This is equivalent to y2 (y2 - 3x 2 ) :::;
precisely (3.9). Equality holds if and only if P coincides with the centroid of
DEF. This is equivalent to
[PDE] = [PEF] = [PFD],

or
rj r2 sin 1= r2r3 sin Cl! = r3rl sin ß.
This shows that
rj
Tl :: r2 T3 = a : b : c.
T2 :: r3
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
·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 Xl ,X2,X3, ... ,Xnn be a sequence of n real nu m-
bers such that 2:::7=1 x; = 1. Prove that for every k ::::: 2, there are integers
a1,a2,a3, ... ,ann ,, not all zero, such that lajl :::; k - 1 and
la1 X1 + a2 X2 + ... + anXnl :::; (k.-1)vin.
Solution: For each n-tuple (a1,a2,a3,"

(a1,a2,a3, ....,a
,ann )) of non-negative integers with
0:::; aj :::; k - 1, associate the sum 2:::7=1 ajXj. Observe that there are k nn such
n-tuples. Using the Cauchy-Schwarz inequality, we obtain
<
n ) (nLX; )
( La; 1/2 1/2
! "tajXj!
lt,ajXjl <
J=l j=l j=l
< (k - l)y'n.
Now, we split the interval [0, (k - 1) vinJ v'7i"J into k n - 1 subintervals each of
length (k - 1) Vn vin / (k n - 1). By the pigeonhole principle, we can find two n-
(b1,bzz ,b3,... ,bn)
tuples (b1,b ,bn ) and (C1,CZ,C3," ,cn ) such that the sums 2:::7=1 bjxj and
(C1,C2,C3, ....,cn)
2:::7=1 CjXj lie in the same subinterval. Hence, it follows that
nn nn !I (k-1)v'7i"
! L bjXj L (k-1)vin
I bjxj - CjXj :::; (n __ ') .
j=l j=l k 1
Taking aj == bj
bj - Cj ; we see that laj I :::; k - 1 for 1 :::; j :::; n and
ItajXjl :::; (k -l)vin

]=1 kn - 1 .
•
Example 4.2. Find the maximum value of
x(l- y2) (1- z2) + y(l- z2) (1- x 2) + z(l- x 2) (1- y2),
156 Inequalities
subjeet to the eonditions that x, y, z are non-negative and xy + yz + zx == 1.
L ;r(l - Jl) (1 - Z2) L x - L (xy 2 + xz 2 ) + xyz L yz

cyclic cyclic cycJic cycJic
L x- L (X 2 Y+X 2 Z)+xyz
cycJic cyclic
L x- L x(xy+xz) +xyz
cycJic cyclic
L x - L x(l - yz) + xyz

cycJic cycJic
4xyz.
Now the AM-GM inequality gives
1
( xyz )
2/3
:::; 31 ( l:y + yz + zx )
3
L
~ x(l - y2) (1 - Z2)
4
3~.
= 4xyz :::; 3\1'3"
cyclic
cycllc
Here equality holds if and only if x == y z 1/V3. Thus, the maximum

value is 4/3V3.
•
Exarnple 4.3. (USSR, 1974) Consider a square grid S of 169 points which
are uniformly arranged in 13 rows and 13 eolumns. Show that no matter wh at
subset T eonsisting 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
ml + m2 + ... + m13 = 53.
Now to get a reet angle R frmn the points of T, with sides parallel to those of
S, we must get some two points from so me 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 eoineide with the eolumns determined by the two points on the k-th row.
Now the mlmber of pairs of columns determined by mj points on the j-th
row is (~j). Hence, the total pairs of columns eollectively determined by
Applications 157
13
points of T is ~ (~j). We observe that
t.l: (~j)
)=1
13 (
~j )
1 13
2 L (m; -mj)
j=1
1 (13 2) 1
1 13
13
Lmj
> 2 x1 13 (13 )
Lm;
~ m j -- 22" ~ mj
)=1 j=1
~(532)_~(53)
~ (53 2 ) - ~ (53)
26 2
> 2 xx 53 - 27
79 78 C2
(~3).
> 78 =
79> 3 ).
We have used the Cauchy-Schwarz inequality in the second step above. The
C} ). Hence, we may apply
nu mb er of pairs of columns in the grid S is C}).
total number
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 p air of columns
determined by some two points on some k-th row, j i:- 1:- k. We thus get a
rectangle with sides parallel to the sides of S. •
Example 4.4. Consider the t he set S = {al,a2

{al , a2,' ... ,an }. Let C = {P
, an}. {PI1 , PP22,,
....
.. ,Pn } be aa collection of n distinct 2-element subsets of SS such that whenever
whencver
Pj n P k 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, mj denote the number of Pr 's in which aj lies. We

j , let mj
have to show that mj
mj = :s :s
= 2 for each j, 1 ::; j ::; n. Since Pj is a 2-element set,
IP
we have j I = 2 for 1 :::; ::; n. Hence,
::; j :::; Hence, we obtain
ml
ml + 7n2
rn2 + ... + m n = 2n.
Pkk n Pt. Then the given condition implies that {ak'

Suppose aj E P {ak ' al} is in the
{p .. . , Pnn }.
l , P22,, •..
collection {PI' }. Thus,
Thus, for each j, (~j) number of Pr 's
j , there are (n;;)
being contributed to the collection {PI' P2 , .•. n }.

. . . ,Pn }. Moreover all such sets
are distinct. In fact if a set {ak' al} corresponds

corrcsponds to aaBs and at, then both aaBs
and at are Pkk n PI
a re in P Pk == {a
Pt forcing Pk {aBs ,' at} = Pz.
Pl. It follows that
t~ (m(~j)
~
)=1
J=1
2
j)
< n.
<no
This simplifies to
n
TI
l:
L(m;-mj) :s 2n.
(m; - mj) <2n.
j=1
)=1
158 Inequalities
Since 2::::7=1 mj = 2n, we obtain
"2
n
~mj:::; 4n.
j=l
However an application of the Cauchy-Schwarz inequality gives

n 2 n
(2.:
mj) :::;
4n 2 = (2:mj) ::; n 2:m;:::; 4n 2 . 2.: m; ::;
j=l j=l
This shows that equality holds in the Cauchy-Schwarz inequality. We conclude

mj == A, for 1 :::;::; j :::;::; n. Now using 2::::7=1 mj == 2n, we obtain A == 2. This gives
mj == 2, for 1 :::;
::; j :::;::; n. •
Example 4.5. Let ABC be a triangle in which C = 90°. Let m a , mb, m ec be

its medians from the vertices A, B, C on to BC, CA, AB respectively. Find
. m a +mb
the maXImum value of .
m ec
Solution: Let us start with the standard expression:
4m; = 2b 2 + 2c2 - a2 .
However ,, c2 = b2 + a2 and hence 4ma

2 =
a = 4b 2 +
" a2· = 4a 2
' 4m 2b =
Similarly + V,
4m~ = 2b2 + 2a 2 - c2 = b2 + a2 .
2
m~ +m b = 5.
m 2c
Now the Cauchy-Schwarz inequality gives
( ma + m b )2 : :; 2(m~ ~ m~) = 10.

mc mc
We obtain
ma + mb < Fa.
rn e
Equality holds if and only if a == b. Thus, the maximum value is v'IO.

Fa. •
Example 4.6. Let ABC be a triangle with circum-circle f. The medians AD,
BE, CF, when extended, meet f in Al, BI, Cl respectively. Show that the
surn of the areas of the triangles BAI C, CB 1 A, AC1 B is at least equal to the
area of ABC. When does equality hold?
Applications 159
Solution: Let h a , h b and h c be the altitudes; m a , mb, m c be the medians.

Let AlK, BIL and ClM be the perpendiculars from Al, BI, Clon to BC,
CA, AB respectively. Then
[BAlC] AlK DA l
[ABC] ha AD·
But DA I . AD = BD . DC = a2/4, since BD = DC = a/2. This gives

DA I = a 2/4AD, and hence
[BAlC] _ a2 a2
[ABC] - 4AD2 4m 2·
a
Similarly,
[CBIA] b2 [ACIB] c2
[ABC] 4m~' [ABC] 4m 2·
c
Thus,
a2 b2 c2 )
[BAIC] + [CBIA] + [ACIB] = ( -42 + -42 + -42 [ABC],
ma mb mc
and it is sufficient to prove that
a2 b2 c2
-4
m 2a +-4
m 2 +-4
m c2
2:l.
b
a2 b2 c2
----,----- + + . > l.
2b + 2c
2 2 2 - a 2c2 + 2a 2 - b2 2a 2 + 2b 2 - c2 -
Using the Cauchy-Schwarz inequality, observe that
(a 2 +b2 +c2 )2 = ( ~
cychc
v'2b2 + ~ C
2 _
a
2· aV2b 2 + 2c2 _ a2) 2
< ( ~
cychc
2b 2 + ~;2 _ a2 ) ( L
cyclic
a2 (2b 2 + 2c2 - a2)) .
Thus, we obtain
L n'')
a2
n') ') 2:
2
(a +b +c2 )
2 2
cyclic
(a 2 + b2 + C2 )2 2: 4(a 2 b2 + b2c2 + c2a2) _ (a 4 + b4 + c4 ).

160 Inequalities
This further reduces to

L a4 ~ L 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 aj , az,

a2, ... ,an be real numbers
2na2 < O. Prove that the roots of the equation
such that (n - l)ai - 2naz
x" + ajX n - j + ... + an-IX + an = 0
cannot all be real.
Solution: Let al, ... ,et"

Ct2, ...
Ctl, az, ,et n 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
z ') ,)
a l + a2 + ... + a;, ( Ct I + Ct2 + · · · + Ct nn))
(al+aZ+···+a
z2
-2L a a k
Ct j Ct k
j<k
ai - 2az.
Since al, Ct2, ...

Ctl, az, ... ,an
,Ct n are real, the Cauchy-Schwarz inequality is applicable. This
implies that
(al+a2+···+ a n)2 < n (a 2I + a 22 + ... + a 2n )

n(ai - 2a2)'
Thus,
a 2j :::; na 2j - 2naz,
and hence (n - l)ai - 2na2 ~ O. This violates the given condition. Hence not
all al, ... ,an
Ct2, ...
Ctl, a2, ,Ct n can be real. •
Example 4.8. (Indian Team Selection, 1998) Let M be a positive integer and
consider the set S = {n E 1"1 I M 2 :::;
::; n < (M + 1)2}. Show that the products
of the form ab where a, bE S are all distinct.
Solution: Suppose there are integers a, b, c, d in the given interval such that
ab = cd. We may assurne d is the lm'gest among them so that a < d and b < d. d,
This implies that c < a and c < b. Put c = a - k and d = a + II for some
positive numbers k and I.
I, Thell we have
ab = (a - k)(a + I) = a 2 + a(l - k) - kl.

klo
Applications 161
This shüws that a divides kl. Since c, d lie in the set 5, we übserve that
a - k 2 M 2 and a + Z :S M 2 + 2M. Thus, we have
l + k = (a + l) - (a - k) :S (M 2 + 2M) - M2 = 2M.
Using the AM-GM inequality, we now get
Jki < k+ Z < M.

- 2 -
This implies that kZ :S M 2 • Since a divides kZ, we get a :S M 2 • If a = M 2 ,
then c < a implies that c is not in the set 5. Thus, a < M 2 , a cüntradiction.
Thus, if ab = cd, then {a, b} and {c, d} must be the same sets. •
Example 4.9. Let f : N -+ [1, (0) such that

(a) f(2) == 2;
(b) f(mn) = f(m)f(n) für an m,n in N;
(c) f(m) < f(n) whenever m < n.
Prove tImt f(n) == n for an natural numbers n.
Solution: Observe that f(l) = = f(l . 1) == f(1)2 and hence f(l) = = 1. Nüw
inductiün shüws that f(2 k ) = 2k für an integers k 2 1. Let us take any mE N
and suppüse f(m) = Z. Then f(m n ) = zn für an n E No If k is such that
2k :S m n < 2k+l, then using (b) and (c) we obtain
2k :S zn < 2k+ 1 .
Thus, we get the inequality
-1 < (m)n
2< -
T < 2,
2 (4.1)
2 Z '
valid for all natural numbers n. If m > Z, choose n such that n > Z/(m - Z).
Then Bernüulli's inequality gives
m)n == (1m
(T - l ) n > 1 + n (m-l)
+ -Z- -Z- > 2,
by our choice of n. This cüntradicts the right hand side of inequality (4.1). If
m < Z, we choüse n > m/(Z - m). Again, Bernoulli's inequality is applicable
and we get
( mI)n = (1z+---:;;-
m ) n > 1 + n (I-m)
--:;;- > 2,
by our chüice üf n. We thus obtain (m/l) n < 1/2. But this contradicts the left
hand side of inequality (4.1).
We conclude that I = m, thus forcing f(m) = m für all natural numbers
m. •
162 Inequalities
Example 4.10. ([8]) Determine all real polynomials of degree n with each of
its coefficients in the set {+ 1,
I, -1}
-I} 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 mayaiso assume that
p( x) is 1 by changing the signs of all the coefficients,
the leading coefficient of p(x)
if necessary. Thus, p(x) is of the form
( ) =x n
px +Cn~IX
n~l + ",+cx+co,
where ICj I = 1, for 0 ::; j ::; n - 1. Let GI, G2, ... ,G n denote all the roots of
0:10:2'"
p(x) = O. We have 0:10:2'" O:n O:n = (-l)n
(-I)n co , so that O:j -:j:.
-:j:. 0 for every j. This
implies that
n
o < L 0:;
0:; == C;'~l -- 2Cn~2 == 1 ± 2.
j=l
Hence, Cn~2 = -1 and L'j=l a; = 3.

Consider the reciprocal polynomial of p(x). It is also of the same form, i.e.,
having coefficients ±1 and l/aj are the zeroes of this reciprocal polynomial.
The same analysis leads to the conclusion L'j=l (l/a;) = 3. These two lead to
t~(2 O:j
1)_6
(0:; + ~2)
L..J a2 -= 6.•
j=l J
However, the inequality

1I
>2
a+ -a-
holds good for any positive real a. This implies that
6= t
j=l
(0:; + ~2)
(0:;
JJ
:: 2n.
It follows that n ::; 3. We may verify that for n =

= 3, ±(x+ 1)2(x-l) and ±(x+
1)(x - 1)2 are the required polynomials. In the case n = 2, the polynomials
±(x2 - x-I) and ±(x2 + x-I) are the required ones. •
Example 4.11. ([8]) Suppose 0:, ß, '"(Y and 8 are real numbers such that
0:,
a+ß+'Y+ 8 =a 7 +ß 7 +'Y 7 +8 7 =0.
+ ß)(a ++ 'Y)(a
Prove that a(a + + 8)
"()(a + = O.
Applications 163
Solution: Here we use a tricky factorisation:

o =
= _a 77 -- ß7
ß7 -- "(7
"1 7 _- 0677
= (ß + "("I + 0)6) 7
= - ß77 -- "(7
,8 "1 7 _- 0677
= 7 (ß + "()
"I) ("(
("I + 0)
6) (ß + 0) "1 2 + 062
6) { (ß2 + "(2 2 + ß"(
ß'Y + "(Ii
"16 + ßO)
ß6) 2
6)}.
ß'Y6(ß + "("I + O)}.
+ ß"(O(ß
However we have
4{ (ß2 + "1
"(22 + 0622 + ß"(
ß'Y + "(0
"16 + ßO)2
ß6)2 + ß"(O(ß
ß'Y6(ß + "("I + 0)
6) }}
= (ß + "()
= [[(ß (ry + 0)
"I) 2 + ("t
2
6) 22 + (ß + 0)
6) 2] 2
2 f -- 4aß"(0
4aß'Y6
= [[(a+ + (a+"() +(a+o) -4aß"(0

2
ß )2+(a+'Y)2+(a+6)2f
(a+ß) -4aß'Y6 2 2] 2
=
= [3a 2 + 2a (ß + "("I + 0)
2a(ß 6) + ß2 + "(2 622 ] 2
"1 2 + 0 f- - 4aß"(0.
4aß'Y6.
The last expression reduces to
[a 22 + ß2 + "(2
"1 2 + 02]
62 ]22 -- 4aß"(0.
4aß'Y6.
However we see that
[a 2 + ß2 + "(2
"1 2 + 02f f-
62 - 4aß"(0
4aß'Y6 ~> [laßI + 21"(01f
21'Y61f - 4aß"(0
4aß'Y6
> [4Jlaß"(or
[4y'la 'Y61] - 4aß"(0
2 _ 4aß'Y6
> 12Iaß'YJI:::::~ O.o.

121aß"(01
The AM-GM inequality has been used. We conclude that
4{
4 { (ß2 + "1 "(22 + 062 + ß"(
ß'Y + "(0
')'6 + ßO)
ß6) 2 + ß"(o
ß'Y6 (ß + "( J) } ~::::: 0
"I + 0) 0
with equality if and only if lal=IßI=I"(I=lol=O.

lal=IßI=hl=161=O. Otherwise one of ß + ,,(,
"I, ß + 0,
6,
"("I + 06 must be zero.
It follows that
a(a + ß)(a + "()(a

'Y)(a + 0) o.
6) = O.
•
Example 4.12. Find
min { max { 11 + z I, 11 + Z21} }.

Izl:-;l
164 Inequalities
Solution: Consider the functions j, 9 defined by
j(z) =max{11+zl, 11+zZI},g(z) =max{11+zI 2, 11+z212}.
Both jj and 9 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
11 + zl2 :'S: c2 , and 11 + z212 :'S: c2 • (4.2)
11 ++ z 1I == 1111 ++ z21 == 11for z == 0, we must have c :'::;S: l.1.

Since 11
Let us put z = = reiB. Then
11 + z 12 = 1 + r Z + 2r cos (), 11 + z212 = (1 - r Z ) 2 + 4r 2 cos 2 e.
Using 11 + zl2 :'S: cZ, and 11 + z212 :'S: c2 , we get
cos 2 e> (r 2 +1_c2 )2 cos 2 e :'S: (:2 - (1 - r 2 ) 2

, ')
4r 2
11 ++ zl2 :'::;S: c22 and 1111 ++ z212 :'::;S: c22 if and only if
Thus, there is a z such that 11
(r Z + 1- C2)2:'S: c 2 - (1- r2 (
This simplifies to
2r 4 - 2r 2 c 2 + (c 4 - 3c 2
2 + 2) :'::;
S: O.
For getting a positive value for r 2 , we need the discriminant of the above
quadratic expression to be non-negative. This gives
c4 - 6c 2 + 4 :'S: O. (4.3)
Thus, if (4.2) holds for some z, then (4.3) holds. The least number c satisfying
(4.3) is
c=V3 -v'5.
Thus, for any z, we have
max{\l+z\, \1+z 2 \} 2: V3-v'5.

Taking
=
z = reiB,
reiB r =
,
= J3-
22 ',()e -= -211"
J5 _ 21r
3'
we see that 11 +
\1 + z\zl 11 + Z2\
= \1 + z21 = J3 - J5. Hence, the required minimum is
J3- V5. •
Applications 165
Example 4.13. (Romania) Define a sequence (x nn )) by

(Romania) Define by
XX n
Xl
Xl =
= 1,X n +1 = --nnnn + -,for
X
Xn
nn;::
>
--
l.
n
(xn )) is increasing and [x~J

Show that (x [x;] = nn for nn ~;:: 4.
Solution: We observe that
that XI
Xl = 1,
= X2 = 2,
2, X3 = 2. We show that for n ~;:: 3,
=
the inequality
n
C<'E
yn _. n - < Jn- l' (4.4)
holds good. Consider the function

X n
f(x) =
= --
n
+-,
+ -,
X
defined for aHall X > O. Then it is easy to see that f(x)

f( x) is a decreasing function
:S n. Assuming (4.4) for n,
for X:E ::; n , we have
n+1
Xn+l =
= f(x
f(x n ) ~
::; f( vn)
f( Vii) =
= vn'
Vii'
Xn+l = f(x n ) ~ f(~)

;::
n-1
n -l
= ~> v'n+1.
n--1
n-1
Vn+l.
Thus, (4.4) is true for aH
all n by induction. We now have
Xn+ l - X n
Xn+l Xn 1)
(~ -1)
Xn(~ -
(n - 1)
nX nn
(~
n - 1
X;)
_ x2)
n-
~> O.
Thus, :S Xn+l
X n ::; aB n. Now USillg
for aH using X n ::; :S n/vn=I,
n/~, we get
Xn+l =
= f(x n )} ~ f(
;:: ~)
f(~)
n-1 = ~
= Jnn_1·.
n-1
We thus obtain
n-1 n
---::::==
--=== <
vr;:=2 <xX n < - - ..
Jn-2 -- -- ~
vn=I
This gives
n-1)
-f( X ) < f (( n - l ) _- n----'-----===---
(n-2)+(n-1)2 n 2(n-2)+(n-1)2
x +
Xn
n+11 -= f(x nn ) ::; f ,~
- ,~ = l)~
n(n -_ 1)Jn
n(n - 2
We show that
n 2(n - 2) + (n - 1)2
(n-2)+(n-1)2 ~
n(n-1)~
-~--'---===--- < n + 2.
< vJn+2.
n(n-1)~
166 Inequalities
This is equivalent to the statement 2n 3 - 6n 2 + 4n - 1 > 0, which is true for

n ~ 4. Thus, X n < vInTI for n ~ 4. We obtain
yTi < X n < v'n+l,

for n ~ 4. Hence, [x;] = 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 + c 2 == 4.
1 [(I-a)+(a-b)+(b-c)+c[2
< 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
= 3/4, b == 1/2, c == 1/4.
Hence, a = •
Example 4.15. ([8]) Show that if all the roots of P(x) = ax 4 - bx 3 + cx 2 -
X + 1 = 0 are positive, then c ~ 80a + b.
Solution: Suppose a =I- O. Let a, ß, "( and 15 be the roots of P(x) = O. These
are positive reals. Taking
1 1 1 1
uu=-+-
a+ß13''
=~ v=-+-
"( 15 '
the following relations are easily obtained:
u + v == 1, =a
aßu = + ß, = "("( + 15.
"(15v =
We have to show that

c b
- - - > 80.
a a-
However, using the Viette's relations between the roots and coefficients,
c b
aß + a"( + a15 + ß"( + ß15 + "(15 - (a + ß + "( + 15)
a a
(a + ß)("( + 15) - (a + ß) - ("( + 15) + aß + "(15
aß"(15uv + aß(l - u) + "(15(1 - v)
aß"(15uv + aßv + "(15u.
Applications 167
But, observe that

...ß!j '2 _. 2
This gives aß '2

~ 4u- 2 . Similarly ,8 '2
~ 4v- 2 . Thus,
c b
> _
16 + __
4 ( u 3 +v 3 )
a a uv 2 2u V
(2
-16 + -24-2 u - uv + +v 2)
uv u V
3 1 1 }
4{
uv
+-+-
u2 v2
4{ u v +(~+ ~f}
1
4{4{ uvu11v + u21I}

v2 }
U 2V 2
(:v +lr-1.
But u + v == 1 implies that uv ::; 1/4. Thus,
b
-ac - -a-> 92 - 1 == 80.
Equality holds if and only if a = ß = "( = 8 = 4.
Suppose a = 0 and b -I- O. Then P(x) reduces to a cubic polynomial:
P(x) == -bx 3 + cx 2 - X - 1. Let p, q, r be its positive roots. Then
e I l 1
--b = pP + q + r, and - + - + - = 1.
b p q r
Hence,
~ = (p + q+ r) (~ + ~ + ~) '2~ 3,
by the AM-GM inequality. In particular c/b > 1 and this proves the result in
this case.
Finally, suppose a = b = O. Then P(x) = cx 2 - x-I 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

abc d
-~ + - + - +-,
c +dd +~a + b'
b
where a, b, c, d vary over all distinct real numbers such that
abc d
b - + ;:- == 4,
- + ~- + d = bd.
ac =
b c d a
168 Inequalities
Solution: Introducing the new variables u,v by u = alb and v = b/c, we see
that
-dc = -u'a
Id la
- = -v'c- = uv 'd-b = -u·
v
v 1 u
Thus, the maximum of uv + - + - + - has to be found under the constraint
u uv v
u + v + (l/u) + (I/v) = 4. Now for any nonzero real p, we have p + (l/p) 2 2
if p > 0 and p + (l/p) ::; 2 if p < O. The condition u + v + (l/u) + (I/v) == 4
shows that u and v cannot both be negative. If u and v are both positive then
u + (l/u) = 2 = v + (I/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
other negative. We may assume that U > 0 and v < O. In this case v + ~ ::; - 2
v
and hence
U +~ = 4- (v + ~) 2 6.
We finally get
uv + -~ ++u~-1+ -~ (u ++~)
uv v
u uv
-1)((v + -~)
v = (u
u =
1) <<u
v
v
-12.
-12.
This maximum is attained when
a = 3 + 2v2, b = 1, c = -1, d = -3 + 2v2.

•
Example 4.17. Solve the system of equations for real values of x, y, z:
3(x 22 + y2
y2+ Z2)
Z2) 1,
X2y 2 + y2 z 2 + Z2
Z2 X 2
2 xyz(x + Y + Z)3.
Solution: The second relation shows that xyz(x + y + z)3 2 0 and hence
xyz(x+y+z) 2 o. If xyz(x+y+z) = 0, then the second relation irnplies that
xy = yz = zx = O. Hence, two of x, y, z are zero. If x = y = 0, then z2 = 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) > O. Using the Cauchy-Schwarz inequality, we see that
(xy + yz + ZX)2 ::; 3(x 2y2 + y2 z 2 + Z2 X2).
This gives
xyz(x + Y + z) ::; X2y2 + y2 z 2 + Z2 X2 = xyz(x + y + Z)3.

Applications 169
Since xyz(x + y + z) > 0, it follows that (x + y + Z)2 :2: 1. Eut then
1 ::; (x + Y + Z)2 ::; 3(x 2 + y2 + Z2) = 1.
Hence, equality is forced every where. We obtain (X+Y+Z)2 = 3(x 2 +y2 +Z2)
or equivalently x 2 + y2 + Z2 = xy + yz + ZX. This leads to
(x - y)2 + (y - Z)2
Z)2 + (z - X)2 = 0,
X)2
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) == anx n + an_lX n - 1 + ... + alX + ao be areal

polynomial of degree n. Suppose n is even and
(i) ao > 0, an > 0;
(1..
'1') a 2
(11) 2 '2) 22 4 . {2{2 2}
1 + a2 + ... + an-I::; - - mln a o, an .
2}
n-l
n-1
Prove that P(x) :2: ° for all real values of x.
Solution: Suppose x is areal number. The Cauchy-Schwarz inequality gives
lan_1X n-1
lan_1X n - 1 + ... + a1xl ::; Ja;'-l ar vx
+ ... + ar 2 + x 44 + ... + x 2n - 2 ,
for any real x. Thus, for x E OC,
P(x) (anX n + ao) + (an_lX n- 1 + ... + alx)
> (anX n + ao) -lan_1Xn-1 + ... + a1xl
,ao}(l + x nn )) -- Ja;'-l + ... + ar vx

> min {a n ,ao}(1 ar
Vx 2 + x 44 + ... + X2n - 2 .
Using the condition (ii), it follows that
P(x) > min {a n ,ao}(1 + x n ) - 2m~an} Vx2 + x4 + ... + X2n-2

n-l
. [n 2Vx 22+x
2vx 4
+ x 4+ ... + xX2n - 2 ] 2n 2] .
mrn {an,
min ao } [11 + x n -
{ an, ao} Vn=1
n-1
n 2vx 2 + x 4 + '" + X2n - 2
2Jx2+x4+···+x2n-2
1 + x n -- r::::-1 :2: 0,
Vn=1
yn -1
for all real x. Equivalently, one has to prove that
n-l
n-l
(n - 1)( xn + 1) 2 :2: 4 L X2.i

X2.i ..
j=l
170 Inequalities
Since n is even, n = 2m for some positive integer m. Introducing x 2 = y, we

need to show that
2m-l
(2m - 1) (y2m + 2ym + 1) 2: 4 L yf.
j=l
We use the majorisation technique. The function f(t) = t k is a convex function

for all k 2: 1. Hence, we need to show that
(?m-l,.~. ,2m- ~,?m- 2,.~. ,2m- ~, ... ,~)

4 4 4
-< (2m, 2m, ... ,2m, m, m, ... ,m, 0, 0, ... , 0)

, ... ' ""
"--v
v- - ' ' - v - - '
2m-l 2(2m-l) 2m-l
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 6m 2 - 12m and the right-side is 6m 2 - 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
333
x+y+z + - - + - - + - - = 2(v'x+2+ vY+2+ v'Z+2).
x-I y-l z-1
Solution: For t > 1, the AM-GM inequality gives

t+2
t - 1 + - - > 2-v't+2.
t -1 -
Equality holds if and only if (t - 1) = (t + 2)/(t - 1), which is equivalent to

t = (3 + V13)/2. Thus,
333
x+y+z+--+--+--
x-I y-I z-I
x+2 y+2 z+2
x-I+y-l+z-l+ - -
x-I
+- - +z-I
y-I
--
> 2(v'x+2+vY+2+v'Z+2)
333
x+y+z+--+--+--
x-I y-I z-1
Applications 171
Hence, equality holds and this corresponds to
3+JI3
x=y=z=3+v'I3
x=y=z=
2
•
Example 4.20. (Bulgaria) Find the largest real number a such that
x y z
JJ y2 + Z2 + VZ2 + x 2 + JJ x 2 + y2
-r=:;===.
-r=:;=="" > a,
-
for all positive real numbers x, y, z.
Solution: Note that for any real a,

2a
1 + a2 ::; 1.
If a > 0, we also get
2a 2
-- <
1 + a2 - a.
x 2x 2/(y2 + Z2) 2x 2
-r=:;===.
-r===~ > --~~------~
--~~~----~
JJy2+Z21+(X2/(y2+z2))
y 2+ Z2 - 1+ (X 2/(y2+ Z2)) x2 + y2 + z2'
This implies that

x y Z ~ 2x 2
---r::::==~
---,=;=;===
y 2 + z2
. Il y2
+ VZ2 + x 2 + > ~ 2
. I x2 + y2 - cychc
. x + y2 + Z2
= 2
V
V V
V cychc
We show that a = 2 is the largest possible. In other words, for each 1010 > 0,
there are positive x, y, z such that
L x <2+E.
cyclic J
cyclic J y2 + Z2
For this choose x = y = 1 and z = EV2. Then
L
~
J
xx
~ ---r==.===;c =
22
VI ++ 2102 + 10 < 2 + E.
cyclic
.
cychc V
l y 22 ++ z2 210 2 + 10 < 2 + E.
•
Example 4.21. (Romania, 1993) Prove that if x, y, z are positive integers
such that x 2 + y2 + Z2 = 1993, then x + y + z cannot be the square of an
integer.
172 Inequalities
Solution: Suppose, if possible, x +y+z = k 2 for some integer k. Then
k2 = X + Y + zz ::;::; v'3vx2 + y2 + z2
z2 = v'3J1993 < 70.
Thus, k 2 E {I, 4, 9,16,25,36,49,64}. On the other hand x 2 == x (mod 2) for

any integer x. Thus, x 2 + y2 + Z2 == X + Y + z (mod 2) and hence
k 2 == X + Y + z == x 2 + y2 + z2 == 1993 == 1 (mod 2).
This shows that k 2 is odd and hence k 2 E {1,9,25,49}. Now observe that
(x + Y + Z)2 > x 2 + y2 + Z2 == 1993, and 25 2 == 625 < 1993.
Hence, x + y + z = = 49 is the only possibility. But then x + y + z == 49 and

x 2 + y2 + Z2 == 1993 give
(49 - X)2 = (y + z)2 > y2 + Z2 = 1993 _ x 2 .
Simplification leads to x 2 - 49x + 204 > o. It follows that
either < 49 - J1505 x> 49 + V1505

49
either xx < 49 - J1505
2 or x> 22
However
x > 49 + V1505 =* x > 49 + 39 =
= 44,
2 2
antI hence x :::: 45. But then x 2 :::: 45 2 == 2025 > 1993. Similarly,
49 -- J1505
49 J1505 =? x ::; 4.
x<
x< 2 ==? x <
-- 4.
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
lax 2 + bx + cl ::; 1, for all lxi::; 1,

8 '2
anel 3"a + 2b2 is maximum.
Applications 173
Solution: It is sufficient to maxi mise 4a 2

maximise + 3b2 , since
+ 3b22 =
4a 2 + ~ (~a2 ++ 2b22)-
) .
+ bx + c. Since IP(O)I
Consider the polynomial P(x) = ax 2 +bx+c. .;:::; 1 and IP(I)1
IP(O) I ~ IP(I) I ~
.;:::; 1,
we have
2 ::::: IP(I) - P(O)I = la + b + c - cl = la
22 la + bl·
IP(-I)1
Similarly using IP( .;: :; 1, we also obtain
-1)1 ~
2::::: IP( -1) - P(O)I ~

22 = la - b + c - cl = la -- bl·
= la
Thus,
4a 2 + 3b 22 = 2(a + b)2 + 2(a - b)2 - b22 ~
.;:::; 16 - b2 ~
.;:::; 16.
hnlds only if b == O. Thus, 4a 2

Here equality holds + 3b22 is maximum only if b = O.
In this case
la + bl
la = la - bl = lai = 2.
Thus,
IP(I) - P(O)I = la + c- cl = lai = 2.
Hence,
= IP(l)
2= IP(I) - P(O)I ~ IP(I)1
.;:::; IP(l)1 + IP(O)I
W(O)I ~.;: :; 2.
We conclude that IP(I)1 = IP(O)I = 1 and hence !cl lcl = 1, lala + cl = 1. It follows
that either c = 1, a = - 2, b = 0 or c = -1, a = 2, b = O.
Now
N ow the required maximum is
~a2 + 2b2 = ~ x 4 = 32.

3 3 3
•
Wh at is the maximum value of a 2
Example 4.23. ([7]) What + b2 , given that the
equation
x 4 + ax 3 + bx 2 + ax + 1 = 0
has real roots.
Solution: Consider the polynomial P(x) = x 4 + ax 3 + bx 2 + ax + 1. Observe

that P(x) == x 4 P(I/x). Hence,
Hence, P(x) is a reciprocal polynomial and hence the
substitution y = x + I/x is meaningful. This leads to a quadratic equation
y2 + ay + b - 2 = o.
O.
If x is real,
real , we observe that y is also real and hence
(2 - y2)2 = ((Ly + b)2 ~ (a 2 + b2)(1 + y2).

174 Inequalities
This gives
> ((22 -- yy2)2
+ bb22 >
aa22 +
2)2
- 1 + y2
We also note that y = x + I/x ~ -2 or y 2: 2. Hence, y2 2: 4. Consider the
function
f( t) = (2 - t)2
1+t '
> 4. This is a monotonically increasing function for t 2: 4, as may be
for t 2:
verified using its derivatives. Its minimum is attained at t = 4;
a2 + b2 2: f(4) = s·4
= 4, we have x + I/x == 2 and hence x == 1 is a real solution. This
If we take y2 =
corresponds to 2a + b = - 2 and we can take
4 2
=-S'
aa = -- b= -~
--
5' 5·
Then a 2 + b2 = 4/5. Thus, the minimum value of a 2 + b2 is 4/5. •

Example 4.24. Suppose a, b, c are real numbers such that
lai ~ Ib - cl, Ibl ~ le - al, and lei ~ la - bl

Prove that one of these nurnbers is the surn of the remaining two.
Solution: We get from the first relation, a2 ~ (b - e)2, or which is the same
as
(b - e - a)(b - e + a) 2: o.
Similarly
(e-a-b)(e-a+b) 2: 0, and (a-b+e)(a-b-e) 2: o.

Multiplying these (which is permissible since all are non-negative), we get
(a + b - e)2(b + e - a)2(e + a - b)2 ~ o.

It follows that at least 0I,2 of (a + b - e), (b + e - a), (e + a - b) is equal to zero .
•
Example 4.25. Let ABC be a tri angle 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 175
Solution: It is sufficient to locate D, E, F such that AD, BE, CF are

concurrent and [DEF] is maximal,
maximal, where [DEF] denotes the area of triangle
tri angle
DEF. Let us put BD == a - x, DC == x, CE == b - y, EA == y, AF = =c- z
and FB == z. Then by Ceva's theorem,
a-x
- _ b-y
.-
__ c-z
- -. _ .._-=1.
_-=l.
xx Yy zz
If we also introduce a0: = x / a, ß = y //bband

and 'ryY = z / c, then it takes the form
0:)(1 - ß)(l
(1 - a)(l- ß)(1 - 'Y)
ry) == aß'Y·
o:ßry.
Now observe that [BDF] = 'Y(1

ry(1 - a)[ABC]
o:)[ABC] and similar expressions
express ions for other
corner triangles. Thus,
[DEF] [ABC]- [BDF]- [CED]- [AFE]

[ABC]{I-
[ABCl{ ry(l-
1 - 'Y(1 0:) - a(l
- a) 0:(1- ß) - ß(l
- ß) ß(I- ry)}.
- 'Y)}.
This shows that we have to minimize 'Y(1 + a(l

0:) +
ry(1 - a) 0:(1 - ß) + ß(l
ß) + ß(1 - 'Y)
ry) in order
maxi mise [D
to maximise E F]. However
[DEF].
ry(1 - a)
'Y(1 + a(l
0:) + + ß(l
0:(1 - ß) + ry) =
ß(1 - 'Y) = a0: +
+ß + 'ryY - (aß
ß+ + ß'Y
(o:ß + + 'ryo:).
ßry + Y a ).
0:)(1 - ß)(l
Now using (1 - a)(l ß)(1 - 'Y)
ry) = aß'Y,
o:ßry, it is easy to get
+ ß ++ 'ryY - (aß
a0: + + ß'Y
(o:ß + + 'Ya)
ßry + ryo:) = 1 - 2aß'Y·
20:ßry·
Thus, we have to minimise 1- 20:ßry or which is same as maximising aß'Y.

1 - 2aß'Y o:ßT But
o:ßry
aß'Y V
)o:ßry(1 - a)(l
a ß'Y(1- 0:)(1 - ß)(1 - 'Y)
ß)(l ry)
)0:(1 - a)ß(l
va(l o:)ß(1 - ßh(l
ßh(1 - 'Y)
ry)
< (a+1-a+ß+~-ß+'Y+1-'Yr
(O:+I-O:+ß+~-ß+ry+l-ryr
1
8
Equality holds if and only if a0: =
= ß=
= "(.
ry. This is possible if and only if D, E, F
F
are the mid-points of BC, CA, AB respectively. We conclude that P = G, C, the
centroid of ABC. •
Example 4.26. (Short-list, IMO-1993) Let a, b, c, d be non-negative real num-

+ b ++ c ++ d = 1. Prove that
bers such that a +
1 176
227 + 2i
+ bcd ++ cda ++ dab -<~ -7
abc + + -2 abcd.
7 abcd .
176 Inequalities
Solution: Consider the expression

176
F(a, b, c, d) abc + bcd + cda + dab - TIabcd
bc(a + d) 176) .
+ ad ( b + c - TIbc
The funetion F (a, b, c, d) is symmetrie in a, b, c, d. There are two possibilities:
(i) If b + c - ~776 bc :S 0, then the AM-GM inequality implies that
F (a, b, c, d) :S bc(a+d):S
F(a,b,c,d):S C+c~a+dr
bc( a + d) :S (
b+c+a+d)3
3
1
27
Henee, we may assume that this never oeeurs in any subsequent steps.
176
("\8
(ii)
11) Suppose TI
uppose bb + c - TIbc
176 bc > O. Then
Tllen again the
t hAMGM
e AM-GM
- inequaIity implies
inequality
that
F(a, b, c, d) < bc(a+d) + ( -2- 176)

a+d)2( b+c- TI bc
F(a +2 d "c,
b a + d)
2 .
Now we iterate this under the assumption that we are in the seeond ease
each stage of the iteration and we also exploit the symmetry of F:
in eaeh
< F a+d
( -2-,b,c, a+d)
F(a,b,c,d) :S F(a;d,b,C, a;d)
-2-
F(b a + d a + d )
, 2 ' 2 ,c
< F(b+C a+d. a+d b+C)

2 ' 2 ' 2 ' 2
F(a+d b+c a+d b+C)
2 ' 2 ' 2 ' 2
< F(~ b+c a+d ~)
4' 2 ' 2 '4
F(b+C ~ ~ a+d)
2 '4' 4' 2
< F(~
( 1 ~1 ~1 ~)
< F 4'
4' 4'
4' 4'
1)
4' 4
~1
4 =- 27
-
This eompletes the solution. •
Applications 111
111
4.27. Determine the minimum value of 2:7=1

Example 4.21. 2:;'=1 aj where the mini-
mum is taken over all sequences (al ,a2,a3, ... ,a nn), (n 2: 2),
(al,a2,a3, 2) , with non-negative
n
terms such that Lajaj+l L
ajaj+1 =
= 1, where an+l = al·
= al.
j=1
j=l
Solution: Let
Mn = min { ~ aj Iaj 2: 0, for 1 ::; j ::; n, and ~ ajaj+1 = 1 }.

y2, M 33 = V3
We show that M 22 = v'2, J3 and Mn = 2 for n 2: 4. Put 0:Q: == 2:
2: aj and
ß = 2:
ß ajaj+!' If n = 2,
2: ajaj+l' Q: = al + a2, ßß = 2ala2 = 1, so that
2, then 0:
0: 2
Q:2 -2=0:
2 2
2: 0,
- 2ß = (al - a2)2 2:0,
_ 2 = Q:2 -2ß=(al-a2)
with equality if and only if al = a2 = 1/v'2. Thus , M 22 = v'2.

1/y2. Thus, y2. If n = 3, then
0: 2 - 3 Q:22 -_ 3ß
0: 3ß
(al + a2 + a3)2 - 3(ala2 + a2a3 + a3ad
a 2ll + aa222 + a 32Z- ala2
alaZ - a2a3 - a3al
2"1{ (a1- 1{
(al - a2) 2 + (a2 - a3) 2
2 +(a2- 2 +
+(a3 a d 2}
(a3 - ad z}
O.
> 0.
Here again, equality holds if and only if al = = 1/ V3.
a2 = a3 =
= az J3. Thus, 1M1.13 = V3.
v'3.
Consider the case n = 4.4. In this case,
0: 2 - 4 0:
Q:22 - 4ß
4jJ
2
r
{(al + a3) + (a2 + a4)} - 4(al + a3)(a2 + a4)
{(al + a3) - (a2 + a4) o.
2: 0.
aj + a3 = a2 + a4 = 1/2. Thus, M 4 = 2.
Equality holds if and only if al
Suppose n 2: .5.
5. We may assume that al is the least number by using the
cyclical nature of the condition 2:
eyclieal 2: ajaj+l = 1. Let
/ = (al + a2 + ... + a n )2 - (al - az + a3 - ... + (_1)n+1 an )2.

If n is even, say n = 2k, then
"Y
/ 4(al + a3 + ... + a2k-l)
4(al+a3+···+ (a2 + a4
a2k-I)(a2+ + ... + a2k)
a4+···+
> 4(a l aZ + a2a3 + ...
4(alaz ... + a2k-la2k + a2kal)
a2k a l)
4.
178 Inequalities
Suppose n is odd, say n == 2k - 1; then
r + a3 + ... + a2k-d (a2 + a4 + ... + a2k-2)

4(al
~ 4(ala2 + a2a3 + ... + a2k-2a2k-1 + a2k-Ia2)

>
> 4(ala2 + a2a3 + ... + a2k-Ial)
4,
using a2 ~ al. Thus, r ~ 4. This implies that
(al +a2 + ... +a n )2 ~ 4. (4.5)
We show that equality holds in (4.5) if and only if there is a kk such that
= 1, ak-l + ak+l == 1 and aj == 0 for all other j. (Here ao == an,
1 :S kk :S n, ak =
a n +1 = aIo) Clearly this is sufficient. Conversely, suppose equality holds in
(4.5). Then
(al + a2 + ... + a n )2 - (al - a2 + a3 + (_1)n+1 an )2

- ...
= 4(ala2 + a2 + a3 + ... + an-Ian + anal)'

=
Thus, for j, k with j + 2k + 1 :S n, it must be the case that ajaj+2k+1 = O. In

particular, ala4 = O. This gives al = 0, since al :S a4. Further we have
a1 + a2 + ... + an = 2, and al - a2 + a3 - ... ± an = O.

a1
al + a3 + ... 1,
a2 + a4 +... 1.
If a2 -::J 0, then the condition ajaj+2k+l = = 0 implies that a5 == a7 == ...

... == 0,
and a3 = = 1. Further it also follows that a6 == a8 == ...
... == O. Hence, a2 + a4 == 1.
Suppose a2 = O. Consider a3. Using a similar argument, it is easy to show
that there is some l such that al = 1 and al-l + al+1 = 1 and the remaining
are all equal to zero. •
Example 4.28. (Taiwan, 2002) Determine all positive integers n and integers
:S j :S n, and
al,a2,a3, ... ,a n such that aj ~ 0, for 1
Ln a; = 1 + 4n 4+ 1 (nL aj )2
~
j=1
a; 4n ~a
)=1
J
Solution: Let us put

n
A= Laj.
2::aj.
j=1
j=l
Applications 179
Suppose ajaj :S 1, for all j.

j. Then aj = 0 or 1, for each j.
aj = j. Let aa be the number
of those 1's for which aj =
those j's 1. Then nn -- aa is the number
= 1. nu mb er of those 1's
j's for which
aj =
aj = O. The given equation reduces to
4
a = 1+ _ _ a 2 • (4.6)
4n + 1
This shows that a > 1. Moreover,
4
4n+1=4(a+1)+--.
a-1
Hence, a - 1 divides 4, so that a - 1 = 1,2 or 4. The first two cases may be
disposed ofusing (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
mayassume 2 2 for some j. In this case
assurne aj 2:
A+ 2:= a;
< "~ aj2 = 11 +
+ 11 < 4n 4+ 11 A 2 •.
+ 4n+
4
Thus, A > n + (1/4). Since A is an integer, it follows that A 2:

2n + 1. Using
the Cauchy-Schwarz inequality, one obtains,
A2 <
- -
(n)
'" a2
L,.;)
= 1+ 4- A2.
4n-+ 1
n j=l
This implies that A 2 2 + n. Again the nature of A shows that A :S 2n.

2 :S 4n 2
Thus, n + 1 :S A :S 2n. We hence obtain

1 A
1< 1 + - < - < 2.
n-n-
But we also have
2:=
n (A)2
~ a· - -n --= 11 --
ajJ - -:;;:
A2
n(4n + 1) < 1,
I,
1, ,\
)=1
which gives
A
a -- --
-1< aj
) n
< 11,'
for 1 :S j :S n. Thus,
far
A A
0< -1 + -n < a· < 1 + -n < 3,
J
for n. This shows that aj =

far 1 :S j :S n. = 1 or 2 for all j. Let b denote
denüte number of
üf
für which aj = 2. Then there are nn - b values of j for which aj = 1.
thüse j's for
those
In this case (4.6) takes the form
4
+ n - b = 1+
4b+n-b=
4b 1+ - - (2b + n - b)2.
--(2b+n-b)2.
4n+
4n +1
180 Inequalities
1
This simplifies to n == b+ 4b _ 3' Since n and bare integers the only possibility
is b == 1 ancl n = 2.
\Ve conclude that n == 2 or n == 6. In these cases, aj 's can be computed
using the bounds we have derived:
n = 2: (al,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 5 in three-dimensional

space. Let 5 x , 5 y , 5 z denote the sets consisting of the orthogonal projec-
tions of the points of 5 on to yz-plane, zx-plane, xy-plane respectively. Prove
that
1512 ~ 15x1215y1215z12,
where lXI denotes the number of elements of a finite set X.
Solution: Let us put 15x l = a, 15y l = band 15z 1= c. The inequality 151 ~ abc
will be proved by induction on 151. If 151 = 1, then a = b = Cc = 1 and the
desired inequality follows trivially. So, assurne that the required inequality is
true for all finite sets 5 with 151 < n, where n > 1. Take any set in the three-
dimensional space with 151 = = n. Since 5 is finite, there is a plane a parallel
to the plane of one of the coordinates, say xy-plane, such that aseparates the
points of 5 into two disjoint non-empty subsets, say U and V. If IUI = nl and
IVI = nz, then 1 ~ Tll < Tl, 1 ~ nz < Tl and Tll + n2 = n. If we set IUx I = al,
IV"I = a2, IUyl = bl , Iv~1 = bz , IUzl = Cl and Iv~1 = C2, then
al + az = a, bl + b2 = b, Cl + C2 = C.
15 1
2
= (/VI + /VIf
2
< (j al bl Cl + j a2bZC2 ) (by induction hypothesis )
< (jalblc+ jazbzcf

z
c( Va;b; + j azb 2)
< c(al + az) (bI + bz ) (the Cauchy-Schwarz inequality)

abc.
This proves the result by induction.

•
Example 4.30. For k 2: 2, define f :N ~ N by
f(n) = [(n+n1/kf/k] +n,

Applications 181
whec [xl dcnotes the integer part of areal number x. Determine the range of
].
f·
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 E Pi! is not a k-th power. Then there is a natural number m such that
m k < a < (m + l)k. Using the binomial theorem, it is easy to derive that
(x + l)k 2: x k + X + 1,
for all x 2: o. Taking x = m - 1, it follows that (m - l)k ::::: m k - m. Taking
n = a - m, we also get m k - m < a - m = n < a. Thus,
mk < a-1 (a - m) + (m - 1)
< n+n 1 / k
< (a-m)+(m+1)
a+l
< (m+1)k.
](n) = m + n =
The definition of ]f shows that f(n) f(a - m) = a. This
= a, i.e., ](a
shows that if a E Pi! is not a k-th power, then a is in the range of f. Observe
of].
f(l) = 2, it follows that 1 is not in the range
] is strictly increasing. Since ](1)
that fis
of f. Now consider any k-th power, say, m k . We have proved that if a E Pi! is
not a k-th power, then
a=f(a- [a 1 / k]).
Thus, we obtain
f(mk-m) =f(m k -1-(m-1)) =m k -1,
and
f] ( m k + 1 - m) == m k + 1.
f. Thus, the range of f] consists of
of].
It follows that m k is not in the range of
precisely all those natural numbers which are not the k-th power.
•
Example 4.31. (IMO, 1992) Find all intcgers a, b, c with 1 < a < b < c such
that (a - l)(b - l)(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 + l)(y + l)(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
182 Inequalities
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.
- 232366
= 1 or p == 2.
Thus, p =
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 ::; 3" + 4" + "5 + 12 + 15 + 20 = 60 < 1
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,
= (3,7) and this gives (a,b,c) == (2,4,8).
the only solution is (y,z) =
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 tri angle is numerically equal to its perimeter.
Solution: Let a, b, c denote the three sides of such a triangle. If 8 = (a + b +

c)/2 is its semi perimeter, Heron's formula gives
6 2 == s(s - a)(8 - b)(s - c).
The given condition translates to
48 = (8 - a)(s - b)(s - c).
Taking 8 - a = x, s - b = y, s - c = z, we can write an equation for x, y, z:

1 1 1 1
-
xy
+ yz
- +-
zx
=-.
4
(4.7)
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·
4
Applications 183
This can be written as a quadratic equation in x: zx 2 - 8x - 4z == O. The

discriminant of this equation is D = 16(4 + Z2). The equation has an integer
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 :s
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
11111111
- = - + - + - < - + - + - = -.
4 xy yz zx - 12 20 15 5
This absurd implication shows that x :s 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 =I- O. Suppose
uv > 0 then u 3 - v 3 > 0 and hence u - v ::::: 1. We get
UV+1=U
UV+1=U 3 _V
_V 3 = (U-V)(U 2
(u-v)(u +UV+V 2
2 +UV+V 2 ))
= (u-v){(u-v)2+3uv}
=
> 3uv
This shows that 2uv < 1 which is impossible because uv ::::: 1.
UV + 1 = 0 so that u = v.
Suppose uv < O. If uv = -1, we have u 3 - v 3 == UV
But then u 2 == UV
UV = -1, wh ich is impossible. Hence, we may assume that
uv< -1. We have
luv + 11 lu - vllu 2 + UV
UV + v2 1
lu - vll(u + V)2 - uvl.

184 Inequalities
If u + v = 0, we see that u = -v and hence we get 2u 3 = -u 2 + 1 which has no

integer solution. Hence, we may assurne that u + v -I-
Using the condition uv < 0, we get
° and thus (u + v)2 2: 1.
1 - uv s luv + 11 = -(uv + 1)
because uv < -1. This leads to the absurd conclusion that 1 S -1.
"vVe conclude that uv = 0. If v = 0, we get v = -1. If v = 0, we get u
"Ve = 1.
Thus, the only solutions are (u, v) = (1,0), (0, -1).
•
Example 4.34. (Romania, 2003) Two unit squares with parallel sides over-
lap 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 ABC D 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 E [0,1]. Draw a line from X
parallel to AB and a li ne from Y parallel to BC. Let them intersect in Z. (See
Figure.)
~
I Dj
A
• r'-B
• •
x z
Then X Z == 1 - a and Y Z == 1 - band
Xy 2 = XZ 2 + YZ 2 (1 - a)2 + (1 - b)2
a2 + b2 - 2(a + b) + 2
a
2
+ 2ab + b2 - 2(a + b) - 4:1 + 2
7
(a+b)2- 2(a+b)+4:
(a+b-1)
2
+-3
4
3
> -
4
Thus, XY 2: ,;3/2 and hence the minimal distance is ,;3/2. This is achieved
Applications 185
when a + b = 1 and ab = 1/8,

I/S. This corresponds to
a = 22 + v'2
v'2 b = 22 -- v'2
v'2 or a = 22 -- v'2
v'2 b = 22 + v'2,
v'2.
4' 4 4' 4
Observe
o ::; (1(1 -- (1 -- b)
a) (1 = 11 -- a - b + ab = 8"9 - (a + b),
b).
This shows that a + b ::; 9/8,

9/S. On the other hand
1
- 2v;;,'b =
+ b ::::>
a+ = v'2'
vM'
2
Thus,
1 1
-v'2-
-< a + b -1<
v'2 -11 ::;a+b-l< ~-.
--S8'
However,
(8"1)2 ::; (1J2- )2 1
as may be easily verified,

verified. This implies that
3
Xy 2 (a+b-l)2+~
(a+b-l)2+-
4
1
< ( J2- 1
)2 +43
9
--V2
4
( 2 - 1- )2
v'2
v'2
(1/ J2),
Thus, it follows that XY ::; 22 -- (1/ J2). Equality holds only if a == b == 1/2v'2,
1/2v'2.
We hence have
,. 1
V3/2, and maxXY = 22 -- v'2'
mmXY = V3/2, v'2'
•
Chapter 5
Problems on inequalities
1. (Romania,1996) Let al, a2, ... ,an, anH be n + 1 positive real numbers such
that al + a2 + ... + an == anH' Prove that
n n
L Jaj(anH - aj) :::;
LVaj(anH LanH(an+l-aj).
j=l j=l
2. (Czech and Slovak, 1999) If a, b, c are positive real numbers, prove that
abc
- - + - - + -a -
b + 2c c + 2a +>2b -
l.
3. (Balkan Olympiads, 2001) Let a, b, c be positive real numbers such that

abc:::; a + b + c. Prove that
a2 + b2 + c2 2: J3(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)
c(b+c
+
)+ > - -.
a c+a) - (a+b+c)2 .
( >
a(c+a)
5. (Ireland, 2003) Let a, b, c be three sides of a tri angle 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
2': 1 + Vab + v'bc + y'Ca.

v'ab + c + v'bc + a + v'ca + b 2: vca.
7. (Romania, 2003) Let a, b, c, d be positive real numbers such that abcd = l.
Prove that
1 + ab 1 + bc 1 + cd 1 + da
- - + - - + - - + - - > 4.
l+a l+b l+c l+d-
8. If a, b, c, d are positive real numbers, prove that
/ {a2
y-
a2 b2 : c2+d2
+ b2+~2
4
+ d2 >
2':
-y
i3/ abc + abd
abc abd + acd + bcd.
: acd
4
bcd.
188 Problems
9. (Hungary, 1990) Let a, b, c be the sides of a triangle such that a + b + c = 2.

Prove that
a 2 + b2 + c2 + 2abc < 2.
10. If a, b, c are positive real numbers such that a 2 + b2 + c2 = 1, prove that
-1
(~
a+b
+ -1 + -1) + a + b + c :::: 4V3.
b +c~
11. (Austria, 2001) Find a11 triplets (a, b, c) of positive real numbers which
satisfy the system of equations:
a+b+c 6,
1 1 1 4
-+-+-
abc
2--.
abc
12. (Bosnia and Herzegovina, 2002) Let a, b, c be real numbers such that a 2 +
b2 + c2 = 1. Prove that
a2 b2 c2 3
---+
- - + 1 + 2ca + 1 + 2ab >-.
1 + 2bc - 5
13. (Ukraine, 2001) Let a, b, c and a, /3" be positive real numbers such that
a + ß + , = 1. Prove that
aa + ,Bb +
oa ,C + 2 J(a,B + ,B, + ,a) (ab + bc + ca) ::; a + b + c.
2
14. (Estonia, 1996-97) Prove that for all real numbers a, b, the inequality
a2 + b2 + 1 > a Jb2+1 + bva 2 + 1

holds.
15. (Poland, 1994-95) For a fixed positive integer n, compute the minimum
value of thc sum
X~ x~ X~
Xl + -+-+
22 + 33 + ... +-
+-;;:,
n'
where XI,X2,X3, ... ,X n are positive real numbers such that
1 1 1 1
- + - + - + ... + - = n.
Xl X2 X3 Xn
16. Let a, b, c, d be positive real numbers such that a + b + c + d ::; 1. Prove

that
abc d 1
-b + -c + -d + -a < --.
- 64abcd
Problems 189
17. Let a, b, c be positive real numbers, aliless than 1, such that a + b + c == 2.

Prove that
abc 2: 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

2a2+(b+c)2 +2b2 +(c+a)2 +2c2 +(a+b)2 ~8.
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
a 2b(a - b) + b2c(b - c) + c2a(c - a) 2: O.
21. Let a1,

a1, a2, ... ,an
,an be n(2: 2) real numbers whose sum is 1. Prove that
"'_J
n
n a.
a·
""' _J_
j=l
a· - 2n-l
~ 2 - aj
J
>_
> n
_
n
_.
22. Let a1, a2, ... ,an be n positive real numbers whose sum is 1. Prove that
n a2 1
'
""'
" J >_.
a· +
~ aj aj+1 -- 2
+a'+1
j=l J J
(Here an +1 == ad
23. Let a, b, c, d be foUf positive real numbers. Prove that
1 4 9 16 100
-+-+
abc
+->
d-a+b+c+d
.
24. Let al, a2,'" ,an

,an be n(> 2) positive real numbers such that a1 + az + ... +
an
an = 1 and ajaj < 1/2 for each j, 1 ~ j ~ n. Prove that
n a2 1
'
""
~
"' J >_.
1- 2a· - n - 2
j=l
j=l J
25. (Romania, 1999) Let X1,XZ,

X1,XZ, ...
... ,X
,Xnn ,, Y1,Y2,
Y1,Y2, ...
... ,Yn
,Yn be 2n positive real num-
bers such that Xl + X2
X2 + ... + x
xn X1Y1 + X2Y2
n 2: X1Y1 X2Y2 + ... + xnYn,
xnYn, where n 2: 2 is
an integer. Prove that
XI
Xl
Xl + X2
X2 + ... + X"
X"
XI
~ -- + -X2
X2 Xn
Xn
+ ... +-.
YI
YI Y2
Y2 Yn
Yn
190 Problems
26. (Proposed far IMO-200l) If Xl,X2, ...

XI,X2, ••• ,X n are n positive real nurnbers,
prove that
-
Xl
-2 + +
X2
+ ... +
Xn
<viii
~
< yn.
1 + xi xi 1 + xi
-- 2 2 2 2 2
Xl 1+ Xl + x~
X2 Xl + x~
X 2 + ... + x;;
X~ .
27. (CRUX, 2001) If a, b, Cc are positive real nurnbers, prove that
3(a 2b + b2c + c2a) (ab 2 + bc 2 + ca 2 ) 2: abc(a + b + c)3.
28. (Czech and Slovak, 2003) Let P(x) == ax 2 +bx+c be a quadratic polynornial
with nonnegative coefficients and let 0: be a positive real nurnber. Prove that
P(0:)P(1/0:) 2: p(1)2.
29. If a, b, c, d, e are positive reals, prove the inequality

a 5
5
"""a- >-,
"
Lb+c~2'
Lb+c-2
where the surn is taken cyclically over a, b, c, d, e.
30. If a, b, c are the sides of an acute-angled triangle, prove that
L JJaa2 +bb2-- c JJaa

2 2 - 2 2 - b22 + c2 :s: ab + bc + ca.
cyclic
31. (Iran, 2006) Let a, b, c be nonnegative real nurnbers such that
111
-2 - + - - - + -2- = 2
a + 1 b'2 + 1 c + 1 .
Prove that ab + bc + ca :s: 3/2.
32. (Thailand, 2006) Suppose a, b, c are positive real nurnbers. Prove that
3 ( a + b + c ) 2:
3(a+b+c) >
-
8 (abcf/3
( abc ) +
+
1/3
(a3+~+c3r/3
(a3+b3+c3)1/3
3
CI,C2,C3," .,C n be n real nurnbers such that either 0

33. Let Cl,C2,C3," :s: Cj :s: 1 for all j
~ 1 for all j, 1 :s: j :s: n. Prove that the inequality
or Cj 2:
n n
II (1 - P + PCj) :s: 1 - P + P II Cj
j=l j=l
holds, for any real P with 0 :s: P :s: 1.

Problems 191
34. (Taiwan, 2002) Let XI,X2,X3,X4 be real numbers in the interval (0,1/2].
Prove that
XIX2 X3X4
(1 - xI)(l - x2)(1 - x3)(1 - X4)
< X4I + x 42 + x 43 + x 44 .
- (1 - xd + (1 - X2)4 + (1 - X3)4 + (1 - X4)4
4
35. Let XI,X2,X3, ... ,X nn be n real numbers such that 0 < Xj :::; 1/2. Prove that
(gX)(f,Xjr) ~ (g (1-Xj)/(f, (1-Xj)r)

36. (CRUX) Consider a sequence (an) ofreal numbers satisfying aj+k :::; aj+ak.
Prove that
a2 a3 an
al + -2 + -3 + ... + -n >- an,
for all n.
37. For positive real numbers x, y, z, prove the inequality
L x
x+y'(x+y)(x+z) -
<1,
where the sum is taken cyclically over x, y, z.
38. (INMO, 2002) Let x, y be nonnegative real numbers such that x +y = 2.

Prove the inequality
X3y3(x 3 + y3) :::; 2.
39. (INMO, 1998) A convex quadrilateral ABCD is inscribed in a unit circle.

Suppose its sides satisfy the inequality AB· BC· CD· DA 2 4. Prove that the
quadrilateral is a square.
40. Let {al, a2, ... , an} and {bI, b2, ... , bn } be two sets of reals such that
0< h :::; aj :::; Hand 0 < m :::; bj :::; M for some reals h, H, m, lI/I. Prove that
1:::; (L:al)(L:b/) < ~ (JHM + J hm

(L:a j bj )2 -4 hm HM )2
41. Let ff :: [0, a] --t lR be a convex function. Consider n points Xl ,X2 ,X3,· . .,X nn
in [0, a] such that L:7=1
"L7=1 Xj is also in [0, a]. Prove that
f,f(Xj) :::;f(f,Xj) +(n-1)f(0).

192 Problems
42. Für any naturaillumber n, prove
C:)~ < 4n .
43. Let a, b, cC be positive real numbers and let x be a non-negative real number.
Prove that
a x + 2 + b,r+2 + C I + 2 2: a X bc + abI C + abc x .
44. Let (al, a2, ... ,an), (b 1 , b2 , ... ,bn ), and (Cl, C2, ... ,c
(bI, ,Cn ) be three sequenees
of positive real numbers. Prove that
L
n ajbjcj::; (nLa; )1/:,( Ln b; )1/3( Ln c; )1/3
)=1 )=1 )=1 )=1
45. ([9]) Prove for any three real numbers a, b, c, the inequality
3(a 2 - a - l)(b
1)(b2 - b - 1)(c
l)(c2 - C + 1) 2:::::: (abc)2 - abc + 1.
46. In a triangle ABC, show that
118
--
-++---->
>-- .
sin A sin B - 3 + 2 eos C
Find the eonditions for equality.
47. Consider areal polynomial of the form
P( x ) -- x., n + an-IX n-1

n-l + ... + a1X
alX + 1 ,
CLj 2:
where aj ::::: 0 for 1 ::; :::; (n - 1). Suppose P(x) = 0 has n real roots. Prove
:::; j ::;
that P(2) 2: ::::: 3n .
48. Let a1 .<CL n be n positive integers. Prove that

CLI <a2<a3<" .<a
(al +a2+a3+" .+a n )2 ::; ar + a~ + a~ + ... + a~.

49. (Poland, 1996) Consider a sequenee al ,CL3 ,... . ,a n of positive real numbers
aI ,a2 ,a3,'
which add up to 1, where n 2: ::::: 2 is an integer. Prove that for any positive real
XI,X2,X:~, ... ,X" with
numbers X1,X2,X;" 2:.7
Xj = 1, the inequality
L + L --)-,
n-2 "
ajx, 2
2 .1:j.Tk ::; - -
n-l l-a'
j<k j=l)
holds.
Problems 193
50. (Iran, 1998) Let Xl, XZ, X3, X4 be four

fOUf positive real numbers such that
XIX2X:JX4 =
XlXZX3X4 = 1. Prove that
3333
Xl +
x2 x3 x4 + + 2: mm
.{
Xl + X2 + X3 + X4, -
1 + -1 + -1 + -I} .
Xl X2 X3 X4
51. (Romania) Let {x} denote the fractional part of X;

X; i.e.
i.e.,, {x} = Xx -- [x].
Prove for any positive integer nn,,
t{n
2: {vJ}
j=l
j=]
JJ} :s~ n 2
;-
2
11.
.
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,
b, c are the si
sides
des of a triangle. Prove that
a2 (b +c- a) + b2 (c + a - b) + c2 (a + b - c) :s 3abc.
54. Let x, yy,, z be positive real numbers such that xyz 2: xy + yz + zx.
ZX. Prove
that
(x + Y + z).
xyzz 2: 33(x
xy
55. Let bll ,b2 ,b3 , ...

.•• ,b n be n nonnegative real numbers and let b denote the sum
of these numbers. Prove that
n-l
~ b2
~ bjbJ+l < -
j=l - 4 .
56. (AMM) Let a, b, c, d be complex numbers :such "I- o.

such that ac i- O. Prove that
max {Iacl, lad + bel, Ibdl} > _-1_+_v'

max{lacl,lad+bcl,lbdl} -1 + V5 5_5
-~~-~~--~> ------
max {lai, IblHlcl,
Ibl}{lcl, Idl} - 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(l-
(l-x)(l-y)(l- +y(1-x) + zz(1-y).
x(1- zz)) +y(l-x) (l-y).
58. Let Xl ,Xz ,X3 , ...

Xl,X2,X3, .. ',X
,X n be nonnegative real numbers such that
"n 1
1
~-<1.
" ' - - < 1.
~ 1+x-
~ l+xjJ
j=1
Prove that XIX2X3·· ·X" 2: (n - I)".

194 Problems
59. (Bulgaria, 1998) For positive real numhpfR a, b, c, prove the inequality
3(a+v'ab+~) :::;4(a+b+c).
60. ([1]) Show that for any two natural numbers m, n, the inequality
1 1 4
----- <-
m +n +1
(m + l)(n + 1) - 45
holds.
61. If a, bare 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
(aa+- l)P + (bb+-
+ ~a +
l)P 5P
+ bb ::::>--.
- 2P-1·
2p-l
63. (IMO, 2000) Let a, b, C be positive real numbers such that abc = 1. Prove
that
(a-1+~) (b-1+~) (C-1+~):::;1.
64. (Proposed for IMO-200l) Let x,y,z be real numbers in the interval [-1,2]
such that x + y + z = O. Prove that
(2-x)(2-y) + (2-y)(2-z)
-;---:-:-----;- + ..-(2_-_z-,-)-'..-(2_-_x-,--) > 3.
-'(2-z)(2-x)
(2+x)(2+y) + (2+y)(2+z) + (2+z)(2+x) -2:3.
65. (IMO, 1978) Let (an) be a sequence of distinct positive integers. Prove
that
"->"-
n ak n 1
~k2 -~k'
k=l k=l
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
o -< xy + yz + zx - 2xyz <- -.
27
67. Let Xl, X2, ... ,X n be n positive real numbers. Prove that
n x3 1 n
L 2
j=l X j +
J
XjXj+l +
2
X j +!
::::-LXj,
3 j=l
Problems 195
where Xn+l ::;:: Xl·

Xn +l =
68. Suppose X, y, z are nonnegative real numbers. Prove the inequality

+ y(y - Z)2 2~ (x-z)(y-z)(x+y-z).
x(x - Z)2 +Y(Y_Z)2
X(X-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,
abc 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

+ ab + b22 23(a+b-1).
a 22 +ab+b ~ 3(a + b - 1).
71. ([7]) Define a sequence (x n ) by
X~
X 4 +9
n __ .
Xl = 2, X n +l =
Xn+l 10x nn .
4 5
Prove that 5 <
- < XX n :S -
5 - 4"4 for all n > 1.
72. ([7]) Let a, b, c be positive real numbers such that a 22 -- ab + b22 == c22 .. Prove
that (a - c)(b - c) :S:::; o.
O.
73. ([7]) Let

Let a, b, c be positive real numbers.
numbers. Prove that
JJ a 2 - J
ab + b22 + Jbb22 -- bc + c2 2:
~ J a2 J + ac + c2 .
74. ([7]) For all real numbers a, show that
(a 33 -- a + 2)2 2
~ 4a 2 (a 2 + l)(a - 2)
holds.
75. (Thailand, 2004) Let a, b, c be distinct real numbers. Prove that
2a
(-
a-b
- b) 2
2a-b)2
- + (2b
(2b-c)2
b-c
b-c
c)
-_
--- 2 + (2C -a) 2 > 5.
(2c-a)2
c-a- c-a-
0:, ß, Xl,X2,X3,"
76. Let a, ,Xn be positive reals such that a0:
Xl,X2,X3,· ..',X +ß 1, and
+X2+X3+· +X
Xl +X2+X3+' +Xnn = 1. Prove that
n 2m + l
xX 2mH 1
> __
L
'"'
~ ax·
O:X·
)=1
j=l J
~ ßX'-I-l
J
+ ßX +1
J
~ n 2m - l '
- l '
196 Problems
for every positive integer m, where Xn+l =

= Xl.
77. Given positive reals a, b, c, d, prove that
(b + d)2 ::; Ja
J (a + c)2 + (b+d)2
J(a J a2 + b2 + Jc
J c22 + d22
21ad - bcll
-
vi (a + c)2 + (b + d)2 + ---r=c=2=la=-d=-=b=c
< J(a ==
---;=;=='=;=;;===:===.=;::
J(a+c)2+(b+d)2
78. (APMO, 1996) In a triangle ABC, prove that
~+~+~< fo+Vb+VC
- '2
79. Let ABC be an aeute-angled triangle with altitudes AD, BE, CF and
ortho-eentre H. Prove that
HD HE HF 3
HA+HB+HC~2'
80. (IMO, 1981) For any point P inside a triangle ABC, let rl, r2, r3 denote
the distances of P from the lines BC, CA, AB respectively. Find all points P
for which a/rl + b/r2 + c/r3 is minimal.
81. (IMO, 1996) Let ABCDEF be a eonvex hexagon in whieh AB, BC, CD
are respectively parallel to DE, EF, FA. Let RA, RB, Re be the eircum-radii
ofthe triangles FAB, BCD, DEF respeetively, and let p denote the perimeter
of the hexagon. Prove that
p
RA+RB+Re >-.
- 2
W c denote respeetively the altitude from A to BC, the

82. (AMM) Let h a , mb, W
median from B to CA and the internal angle bisector of angle C. Prove that
ha + mb + Wc ::;
V3
T(a + b + c).
83. (Bulgaria, 1997) Let ABC be a tri angle with centroid G. Provc that
2
sinLABG + sinLACG < M'
- v3
84. (CRUX) Let ABC be a triangle with in-radius r. Let r l , r 2 , r 3 be three

cirdes inseribed inside ABC such that each touehes other eil'des and also two
of the sides. (Such a configuration is ealled Malafatti circ:les.) Let 0 1 , O 2 , (h
Problems 197
be respectively the centres of the circles r 1, r 2, r 3. If r' denotes the in-radius

of 0 1 0 2 0 3 , prove that
~ > 1 + v3.
r' -
Find conditions for equality.
85. Let 0 be a Brocard point of a triangle ABC. Let AO, BO,Cn extended
meet the circum-circle of ABC in K, L, M respectively. Prove that
AO BO CO
OK + OL + OM 2: 3.
86. (Romania, 2006) Let P be a point inside a triangle ABC. Let r A, rB, rc
denote the in-radii of triangles P BC, PCA, P AB respectively. Prove that
abc r;;
- + - + - 2: 6(2 + v3).
rA rB rc
87. (Bulgaria, 1995) Suppose A)A 2 ·· ·A7 ,"B1 B 2 ·· ·B7 , C 1 C2 ·· ·C7 are three
regular heptagons which are such that A 1 A 2 = B)B 3 = C 1 C4 . If 6.),6. 2 ,6. 3
denote respectively their areas, prove that
+ 6. 3 < 2 - vIz.
-21 < 6. 2 6.1
A
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] 2: 2r 2 .
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 LABX 2: 120 0 and let C be a point between B
0
and X. Prove that 2AD ~ V3 (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, Fand the circum-circle
in L, A1, N respectively. Prove that
AD BE CF
DL + EM + F N 2: 9.
91. (IMO, 1995) Let ABCDEF be a convex hexagon with AB = BC = CD,
DE = EF = FA, and LBCD = LEFA = 7r/3. Suppose G and H are two
interior points of the hexagon such that LAG B = LD H E = 27r /3. Prove that
AG + G B + G H + DH + HE 2: CF.
92. (Balkan Olympiads, 1996) Let 0 and G be respectively the circum-centre
and centroid of a triangle ABC. If Rand rare its eircum-radius and in-radius,
prove that OG ~ y'R(R - 2r).
198 Problems
93. (St. Petersburg Olympiad, 1996) Let M be the point of interseetion of two
diagonals of a eyclie quadrilateral. Let N be the point of interseetion of two
lines joining the midpoints of opposite pair of sides. If 0 is the eentre of the
eireumseribing circle, prove that DM 2: ON.
94. (APMO, 1997) Let ABC be a triangle with internal angle biseetors AD,
BE, CF. Suppose AD, BE, CF when extended meet the cireum-circle again
in K, L, M respeetively. If la = AD/AK, lb = BE/BL and lc = CF/CM,
prove that
Za + _l_b
_la_ lb _ + _l,_c
lc _ > 3.
3
- -
sin A
2
+ - -
sin B
2
+ - , -
sin C -
2
> .
95. (Armenia, 1999) Let 0 be the cireumcircle of a tri angle ABC. Suppose
AO, BO, CO when extended meet the eireum-eircles oftriangles BOC, COA,
AOB in K, L, M respect.ively. Prove that
AK BL CM 9
OK + OL + DM 2: 2"'
96. (Romania, 2003) Show that in any triangle ABC
1
--+--+--<--.
1 1 J3
mamb mbmc mcm a - [ABC]
97. (Romania, 2003) In any triangle ABC, prove that
W a + Wb + W c :s V3s.
98. (Ireland, 1998) Let ABC be a triangle with points D, E, F respeetively on
the sides BC, CA, AB. Let the lines AD, BE, CF, when produeed meet the
eireum-eircle respeetively in K, L, M. Prove that
AD BE CF
DK+ EL + FM 2:9.
99. Show that in a triangle ABC,
max {am a , bmb, cm c } ::; sR.
100. Let AB CD be a eonvex quadrilateral of area 1 unit. Prove that
AB + BC + CD + DA + AC + BD 2: 4 + v8.
101. Let AB CD be a square inseribed in eircle. If M is a point on the are AB
(are not eontaining C and D), prove that
MC· MD 2: (3 + 2V2)MA· MB.

Problems 199
a, b, c be the sides of a triangle ABC with in-radius

102. (Estonia, 1996) Let a,
r.
r . Prove that
asinA + bsinB + csinC 2: 9r.
csinC;::
103. Suppose ABC is an acute-angled triangle with area ß

.6. and in-radius r.
r.
f ~ ~.
Prove that
( J cot A +J cot B + J cot C
104. (US Team Selection, 2000) Let ABC be a tri triangle

angle having the circum-
P be an interior point of ABC. Prove that
radius R. Let P
AP BP CP 1
BC2 R:
+ CA2 + AB2 2:;:: Ii,-
105. With every natural number n, associate areal number an by
111
= -- + - + ...
an = ... + -- , ,
PI P2
P2 Pk
where {PI, P2, ... ,,Pk}

Pk} is the set of all prime divisors of n. Show that for any
natural number N 2: ;:: 2,
N
L a2 a3 ...... an < l.
n=2
106. (IMO, 1999) Let n be a fixed integer, with n 2:

;:: 2.
(a) Determine the least constant C such that the inequality
L
l<;i<j<;n
XiXj(X; +x;) ~ C ( L
l<;i<;n
Xi)4
holds for all real numbers Xl

Xl,, ... ,,X ;:: o.
X nn 2: O.
(b) For this constant C , determine when equality holds.
107. Let a,
a, b, c, d be real numbers such that
C, d
(a 2 + b2 _ 1)(c2 + d2 - 1) > (ac + bd _ 1) 2.

Prove that aa22 + b22 -- 1 > 0 and c2 + d~2 -- 1 > o.
O.
108. ([9]) Let XI,X2,x3, .. .• ,XlOO be 100 positive integers such that
Xl ,X2 ,X3 , ..
111
-- ++- +
- · ·+·. +. .-
+- -2
-= 20..
=0
JXl Fz
.jXl.jX2 JXlOO
Prove that at least two of xj's are equal.

200 Problems
109. Let f(a;) be a polynomial with integer coefficients and of degree n > l.
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
lai;:::: 2n + 1.
110. Show that the equation

x y z w
- + - + - + - =m,
y z w x
has no sohItions in positive re als for m = 2,3.
111. Solve the system of equations
4z 2
4z 2 4x 2
4x 2 4x 2
z = ---,-
X = 1 + 4z
X = 4z 2
2 '' = 1 + 4:r2
Y= 4:r2 '' z = 1 +
1 4x2'
4x 2 '
for real numbers x, y, z.
112. Suppose a, bare nonzero real numbers and that all the roots of the real
polynomial
ax n - ax n - l + an_2Xn-2 + ... + a2x2 - n 2bx + 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
a4 + b4 + c 4
CLCOS(ß - ,) + bcos(r - a) + ccos(a - ß) == b.
CL c
'
where a, b, c are the sides and a, {3" are the angles opposite to the sides CL, b, c
respectively.
115. ([7)) Find all positive solutions of the system
1 1 1 1
Xl + - == 4, X2
X2 + - = I, ...
= ,Xl999 X2000 + - =
+ - - == 4, X2000 = l.
X2
X2 X3
X3 x2000
x2000 Xl
116. Find all positive solutions of the system
x+y+z
x+y+z I,
xx 33 + y3
y3 z3 + xy
+ z3 xy zz x4 + y4 + z4 + 1.
Problems 201
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) == x n + nxn~l + a2Xn~2 + ... + an
has real roots 01,02,03, ... ,On. If
oi 6 + 0~6 + ... + 0~6 = n.
Find 01,02,03, ... ,On.
119. (IMO, 1972) Find all the solutions of the following system of inequalities:
(i) (xi - X3X5)(X~ - X3X5) < 0,

::;
(ii) (x~ - X4X1) (x~ - X4X1) < 0,
::;
(iii) (x~ - X5X2) (x~ - X5X2) <
::; 0,
(iv) (x~ - X1X3) (x; - X1X3) < 0,
::;
(v) (x~ - X2X4) (xi - X2X4) < 0.
120. (Short-list, IMO-1993) Solve the following system of equations, when a is
areal number such that lai> 1:
Xl
2
aX2 + 1,
x 22 aX3 + 1,
X~99 axlOOO + 1,
2
X lOOO aXl + 1.
121. (Indian Team Selection,1999) Let a1,a2,a3, .. .,an be n positive integers

such that 2:.7=1
2:7=1 aj = TI7=1 aj. Let Vn denote this common value. Show that
2:
Vn 2: n + s, where s is the least positive integer such that 2 8 - s 2: n.2:
122. Let Zl,Z2,Z3, ... ,zn be n complex numbers such that 2:.7=1
2:7=1 IZjl = 1. Prove
that there exists a sub set S ofthe set {Zl,Z2,Z3, ... ,Zn} such that
l2: z l2: ~.4

zES
202 Problems
123. ([1]) Let (a1,a2,a3," .,a n ) and (b 1,b2,b3,... ,bn ) be two sequences of real
numbers which are not proportional. Let (X1,X2,X3," .,x n ) be another sequence
of real numbers such that
n n
Lajxj = 0, LbjXj = l.
j=l j=l
Prove that
n ",n 2
LX;
' " 2 2 _______L..Jj=l
> a j_ _ _ _ _~
L....-'J'-·=_l_a...:.j
~XJ - 2
j=l ( 2:7=1 a;) ( 2:7=1 b;) - ( 2:7=1 ajb j ) 2
124. ([1]) Let (a1,a2,a3," .,an ) and (b 1,b2,b3,... ,b n ) be two sequences of real
numbers such that
b2 - ... - bn2
2-2
b1 > 0 or a1
2-2 2 > 0'
a 2 - ... - a n
Prove that
(a12 - a22 - ... - an2)( b2l - b22 - ... - b2)

n :S (alb l - a2 b2 - ... - anb n )2 ,
and show that equality holds if and only if aj = >"bj , 1 :S

::; j :S
::; n, for some real
constant >...
125. ([1]) Let XI,X2,X3,"

X1,X2,X3," ',X n be n positive real numbers. Prove that
L
n
~ x.
XjJ <n
<n
2x'
j=l J
j=l
+
~2X'+Xj+l XJ
+ ... + X·+
'+1 +"'+Xj+n-2
J n
-2 -- ,,
where Xn+k = Xk.
126. ([1]) Let X1,X2,X3, ... ,x 2 2 positive real numbers and k be a fixed
,Xnn be n ~
integer such that 1 :::; k :::; n. Show that
'"' Xl + 2X2 + ... + 2Xk-1 + Xk

XI 2n(k - 1)
- ---------- >
~ ---------------------
L .
. Xk + XkH + ... + Xnn -- n - k + 1
cychc
127. ([1]) Let z and ~ be two complex numbers such that Izl :::; r, I~I :::; rand
z i-~. Show that for any natural number n, the inequality
IIznznz --- ~n~n I<

~
1
2
I :::; !n(n
-
- l)rn-2Iz - ~I
-n(n -1)rn-2Iz ~I
holds.
Problems 203
128. For any three vectors, x = = (Xl,X2,X3, ... ,xn ), Y = (YI,Y2,Y3," .,Yn), and
z == (Zl,Z2,Z3,.' .,zn) in ~n, prove that
Ilxll + Ilyll + Ilzll -llx + yll -IIY + zll -Ilz + xii + Ilx + YY + zll ~ o.
129. Let A 1 A 2A 3·· ·A n+1 be a polygon with centre 0, in which Al = An+! 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 - 2a n+l + a n+2 ~ 0 for aB

n ~ 1. Let al,a2,a3, ... ,a2n+! be a convex sequence. Show that
al + a3 + a5 + ... + a2n+l > a2 + a4 + aß + '" + a2n
n+1 - n '
and equality holds if and only if al,a2,a3," .,a2n+! is an arithmetic progression.
131. Suppose al,a2,a3," .,a n are n positive real numbers. For each k, define
Xk = ak+l + ak+2 + ... + ak+n-l - (n - 2)ak,
where aj = aj-n for j > n. Suppose Xk ~ 0 for 1 :s k :s n. Prove that

n n
11 ak >~ 11 Xk·
k=l k=l
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
fez) = alzI 2 + 2 Re(bz) + c ~ 0,

for all complex numbers z, where Re(z) denotes the real part of z. Prove that
IW :s ac,
and
fez) :s (a + c) (1 + IzI2).
Show that IW = ac only if fez) = 0 for some z E C.
133. (IMO, 2003) Suppose Xl:SX2:SX3:S ... :Sx n be n real numbers. Show that
n n ) 2 2(n2 _ 1) n n 2
(
~ ~ IXj - Xk I :s 3 ~ ~ (Xj - Xk) .
Prove also that equality holds if and only if the sequence (x j) is in arithmetic
progression.
204 P7'Oblerns
134. (Short-list, 1MO-2002) Suppose (an) is an infinite sequence of real nu m-

bers with the properties:
(i) there ~s süme real constant c such that 0 ::; an ::; c, for all n 2': 1;
(ii) laj - ak I 2': j +1 k for all j, k with j "I- k.
Prove that c > 1.
135. Let ABC be a right-angle triangle with medians m a , mb, m c . Let A' B'C'
denüte the triangle whose sides are m a , mb, m c . If Rand R' denote respectively
the circum-radii of ABC and A' B' C', prove that R' 2': ~ R.
136. Let ABC be an equilateral triangle and D, E, F be arbitrary points on

the segments BC, CA, AB respectively. Prove that
[DEF] { l+I
[BDF] [CED] +
I} 2':
[AFE] 3.
(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
J[APB] + J[CPD] ::; J[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

denüte the in-radii üf triangles ABC, ABD, ADC respectively. Prove that
1 1 (1 2)
-+->2
-
r1
+->2 - + - .
r2 - r BC
139. (Japan, 2005) Let a, b, c be positive reals such that a + + b ++ c = 1. Prove

that
0(1 ++ b - C)1/3 ++ b(l ++ c - a)1/3 ++ c(1 ++ a - b)1/3 ::; 1.
140. Show that in a triangle ABC,
(a2m~ + b2m~ + c2m~) (a 2 + b2 + c 2) 2': 16m~m~m~,
where m a , mb, m c denote the medians on to the sides BC, CA, AB from A, B, C
respecti vely.
Pmblems
Problems 205
141. Let Xl ,X n be n positive reals which add up to 1. Find the mini-

. "X
,X2 ,X3, ....
Xl,X2,X3,
mum value ofüf
n Xj
LI +
)=1
Lk#j Xk
142. (IMO, 1974) Find all possible values of
abc d
--++
a+ b +d a+b+c
a+b+d
+ b+c+d +--
+ a+c+d'
a+c+d
,
when a,
a , b, c, d vary over positive reals.
143. Let (Fn ) be the Fibonacci

Fibünacci sequence defined by
Fl1 = F2 = 1, + Fn ,, for
Fn +2 = Fn + l1 +F für n > 1.
Prove that
p.
"-...1.
p.
n
L
L 2)
j=1
j=l
2~ < 2,
2,
for all n > 1.
P(x) = x n + an_1X
144. ([7]) Let F(x) an_lX n - l1 + ...
.. + ao
ao be a polynomial with real
IP(O)I == F(l).
coefficients such that IF(O)I P(x) == 0 are
P(l). Suppose all the roots of F(x)
real and lie in the interval (0,1). Prove that the product of the roots does not
1 /2 n .
exceed 1/2
145. If xx,, y are real numbers such that
2x + Y + V8X2
j8x 2 + 4xy + 32y 22 == 3 + 3V2,
3v2,
:S 1.
prove that x 2 y ::;
0:, ß, I"( are the angles of a triangle whose circum-radius

146. (South Africa) If a,
is Rand in-radius r, prove that
cos 2 (a; ß) >-2r

(
0: -
- 2-
2r
R·
- 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 opp8site sides at A'
A',, B' and C'.
Prove that
1 AI· BI· CI 8
4<
4< AA'.BB'.CC':S 27
AA'.BB',CC'::; 27·'
206 Problems
148. (Proposed for IMO-1991) Determine the maximum value of
LXjXk(Xj + Xk),
j<k
over all n-tuples (Xl,X2,X3, ... ,x n )

(X1,X2,X3, of reals such that Xj 2': 0 for 1 :S j :S n.
149. If a, {J, I are the angles of a triangle, prove that
3 L cos a 2': 2 L sin a sin ß·

ß·
cyclic cyclic
150. (Indian Team Selection, 1994) Let X1,X2,X3,

Xl,X2,X3, ... ,XN be positive real num-
bers. Prove that
N 1/.
~ (X1 X 2·· .Xj) J < 3 ~Xj .
(N )
151. (Indian Team Selection, 2000) Let a1:Sa2:Sa3:S··

al:Sa2:Sa3:S·· ·:Sa nn be n real numbers
with the property 2:,7=1 aj = o. Prove that
n
na1an
nala n L aJ :S o.
j=1
152. (Indian Team Selection, 1997) Suppose a, b, c are positive real numbers.
Prove that
1 1 1 3
-+ + > -.
a(l+b) + b(l+c) + c(l+a) -2': l+abc
l+abc·
153. Let x, y, z be positive real numbers such that x 22 + y2 + z2 = 2. Prove that
x + Y + z :S 2 + xyz.
Find conditions under which equality holds.
154. (Wolschaum County Problem book) Let 0 :S Xl Xl :SX2:SX3:S·· ·:SX

·:SXn be such
that 2:,7=1 Xj == 1, where n 2': 2 is an integer. If X n :S 2/3, prove that there
exists a k such that 1 :S k :S n and
1 k
3:sLX.<~
. J - 3·
)=1
155. (Vietnam, 1996) Let x,
X, y, z be non negative real numbers such that xy +
yz + zx + xyz = 4. Prove that
X + Y + z 2': xy + yz + zx.
Problems 207
156. (Canada) Let x, y, z be nonnegative real numbers such that x + yy + z = 1.

Prove that
x 2y+y2 Z + Z2X < ~.
- 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 pp + qq + r == 1. Prove that
pqr(x + Y
y + Z)2
Z)2 ~2 xyr(1-
xyr(l - 2r) + yzp(1
yzp(l - 2p) + zxq(1-
zxq(l - 2q).
158. (Romania) Let Xl ,X2 ,X3 , ..

X1,X2,X3, .... ,X n be n real numbers in the interval [0,1).
[0,1].
(t (t
Prove that
Xj ) - XjXj +1) ~ [~],

where
wherex = Xl.
+11 =
Xn+
159. (IMO, 2005) Suppose x, y, z are positive real numbers such that xyz 2
~ 1.
Prove that the inequality
X5 _ x2 y5 _ y2 z5 _ Z2
-,.,.-----+ + >0
x5 + y2 + z2 y5 + Z2 + x2 Z5 + x2 + y2 -
holds.
160. (Indian Teanl

Team Selection, 1992) Consider two sequences of positive real
numbers, al~a2~a3~" ·::;an and bl:::;b2:::;b3~"
ac::;a2::;a3::;·· ':::;an ·::;b n , such that
b1 ::;b2 ::;b3::;·· ·~bn,
n n
Laj ~
2 Lbj.
Lbj .
j=l j=l
Suppose there exists a k, 1 ~ < k ~::; n,

n, such that bj
bj ::; aj
~ aj for 1 ~ ::; k and
::; j ~
bj 2 aj for j > k. Prove that
bj ~
n n
II aj ~2 II bjb .
aj j .
j=l j=l
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
l+a+b 1+b+c
l+b+c
+ 1+c+a <--+--+--.
l+c+a - 2+a 2+b 2+c
208 Problems
162. (PUTNAM) Let n 2: 4 and let aj

al ,a2,a3, ... ,a" be real nurnbers such that
al + a2 + ... + an 2: n, a 21 + a22 + ... + an2 2: n2.
Prove that
rnax {al,a2,a3," .,a n } 2: 2.
163. Let Xl:SX2:SX3:S"

Xl:SX2:SX3:S" ,:Sx
,:Sx n +1 be nn + 1 positive integers. Prove that
L L
n+l ~
VXj+1 - Xj < ~.
j=1 Xj+l j=1 J
164. Let a, b, c be three positive real nurnbers which satisfy abc == 1 and a3 >
36. Prove that
22,)')
-a~ < a + b + c - ab - bc - ca
'2 2 2
3 .
165. ([1]) Let ZI,Z2,Z3, .. ',Zn be n cornplex nurnbers and consider n positive
real nurnbers Al,A2,A3, ... ,A n which have the property that L I/Aj = 1. Prove
that
2
n
~Zj :S ~AjIZjI2.
n 1
I
166. ([1]) Let a, b, c be three distinct real nurnbers. Prove that
3 rnin { a, b, c} < :L::a - (I>2 - LL ab)

, 1/2
L
< L a + ( L a2 (L - L
L ab )
1 j"
~ < 3 rnax { a, b, c},
where the surn is cyclically over a, b, c.
167. Suppose a, b, c are real nurnbers such that a 3 + b3 + c3 == O. Prüve that
(Lazf:s (L(b-c)2)(La 4 ),
where the surn is cyclically over a, b, c.
168. Show that for all cornplex nurnbers Z with real part of Z > 1, the following
inequality holds:
IzIzn+1
n + 1 -- 11 > Iz"llz -11, for all n
n 2: 1.
169. Suppose a, b, c are positive real nurnbers and let
x == a + b - c, y == b + c - a, =
z = c + a - b.
Problems 209
Prove that
abc(xy + yz + zx) 2 xyz(ab + bc + ca).
170. Let a, b, c be positive real numbers. Prove that
"'""' a3 3 2:cyclic ab
~. b2 - bc + C2 2 '"
ucyclic a .
cychc
171. (CRUX, 2004) Let al,a2, ... ,an< 1 be non-negative real numbers satis-
fying
a = /
a 21 + a 22 + ... + a 2n >_.J3
n - 3
Prove that
al a2 an na
2
al + - 1
- a2
2+"'+-1 2
- an
2 -
1 - a 2'
172. (USAMO, 2001) Suppose a, b, c are nonnegative real numbers such that
a 2 + b2 + c2
2 + abc = 4. Prove that
o S ab + bc + ca - abc S 2.
173. (CRUX, 2003) Suppose a, b, c are complex numbers such that lai == Ibl ==
lei. Prove that
1 aa22 :: b2 11+ 1 b2 : cc2 11+ 1 c2 c: a212 V3.
a212
174. (CRUX, 2001) Suppose x, y, z are nonnegative real numbers such that
x 2 + y2
y2 + Z2 =
= 1. Prove that
(a) 1< l: -x 3V3

- <-'
- 1- z- 2 '
cyclic Y
(b) 1< "'""' _x_ <

- ~ l+yz-
V2.
cyclic
175. (CRUX, 2004) Let x, y, z be non-negative real numbers satisfying x + y +

z== 1. Prove that
2
222
xy + yz + zx 2 xy + yz + zx - 9'
176. (CRUX, 1990) Let a, b, C, d be four positive real numbers such that a +
b + c + d = 2. Prove that
a2
2 16
16
2:
"'""'
~ ( 22
cyclic (a
cyclic a
a
+ 1) 2<-
2 S 25'
1) - 25'
210 Problems
177. (IMO, 2001) Prove that

a b c
+ +
-;==;:;=======
-;:::;;;=======
va 2 + 8bc Vb 2 + 8ca vc
2
>1
+ 8ab -
for all positive real numbers a, band c.
178. Prove that in a triangle ABC, the inequality

ABC 9R 2
tan"2 + tan"2 + tan"2 :S 4[ABC] '
holds.
179. (CRUX, 2002) Prove that in a triangle with angles a, ß, 'Y, the inequality
2: .
2: sma:S. J15
V/15
2: cos(a -
4"4 + 2:cos(a ß)
cyclic
holds.
180. (KÖMAL) If x,y are real numbers such that x 3 + y4 :S x 2 + y3, prove
that x 3 + y3 :S 2.
181. (CRUX, 2003) Let a, b, c be three positive real numbers. Prove that
2: c(cab+ a) 2 2:
2: 2: c +a a'
where the sum is taken cyclically over a, b, c.
182. Prove that for any real x, and real numbers a, b,
a b
( sin x + cos x)( sin x + cos x) :S 1 + (a; b) 2
183. (PUTNAM, 1988) Let x, y be two real numbers, where y is non-negative
and y(y + 1):S (x + 1)2. Prove that y(y -1):S x 2.
184. Let x, y, z be positive real numbers. Prove that
(Xy+y;+zxr/2 :S C x + y )(y;Z)(z+X)r/ 3
185. (Romania, 1997) Let a, b, c be positive real numbers such that abc = l.
Show that
a 9 + b9
'"' a6 + a 3 b3 + b6 >
~ - 2,
where the sum is cyclical.

Problems 211
= 2(8 - a), y == 2(8 - b),

186. Let a, b, c be the sides of a triangle and set x =
z = 2(8 - c), where 8 denotes the semi-perimeter. Prove that
abc (ab + bc + ca) ;:::: xyz(xy + yz + zx).
187. (AMM) Let al,a2,a3, ... ,a nn (n > 2) be positive real numbers and let 8
denote their sumo Let 0 < ß ::; 1 be areal number. Prove that
t (~)ß ;: :
k=1 ak
(n _ 1)2ß t k=1
(~)ß
8 - ak
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 rl' Similarly define r2 and r3. Prove that
(1') 2rl
(i) 2
rl +
+ 22v!8(8-a)
/8(8a - a) r = h aa'·
h
a
(ii) 2(rl + r2 + r3) + 8 ;:::: ha + hb + hc ·
189. (Romania) For n ;:::: 4, let al,a2,a3, ... ,a n be n positive real numbers such
n
that La; = 1. Show that
j=1
al a2 an 4(
a~+l a~+l
2
+ + ... + ai+l;::::"5 alJal+a2v'a2+ ... +anFn) .
190. (Belarus) Does there exist an infinite sequence (x n ) of positive real num-
bers such that
Xn +2 = JX n +l - ,;x;;,
for all n ;:::: 2.
191. (Belarus) Let al,aZ,a3," .,a nn be n positive real numbers and consider a
permutation b1 ,b2 ,b3 ,... ,bnn of it. Prove that
n 2 n
L: Laj.
j=1
> '" aJ'.
" aJj ;::::
'Ltb -Lt
J j=1
al,aZ,a3, ... ,a nn and b11 ,b22 ,b3 ,... ,bnn be two sequences of positive real
192. Let al,a2,a3,
numbers such that 2::7=1 aj = 2::7=1 bj = = 1. Prove that
"'_J_
n
"'"
L... +b
Lt a +b·-2
a·
1
>_.
>
2
a2
_.
_J_
j -
1
j=1 J J
212 Pmblems
Problems
193. Let x, y, .z.z be positive

positive real numbers. Prove
Prove that
y2 _ x2 z2 _ y2 x 2 _ Z2
"----- + + > O.
z+x x+y y+z-
194. (Japan,
(Japan, 2003) Find the greatest real value of k such that far
for every triple
tripIe
(a, b,
b, c) of positive real numbers, the inequality
(a 2 - bC)2 > kW - ca) (c 2 - ab)
holds.
195. (Romania, 2003)Let a, b, c, d be positive

posit ive real numbers such that abcd = 1.
Prove that
+ ab+ -1 -
11 + ++
+ bc 1++ cd - - + da 4 .
11 +
---
- + -- +---+ - + -> - > 4.
l+a l+b l+c l+d-
196. (UK,2004)
(UK ,2004) Let a, b, + b ++ c == 1,1,
b, c be the sides of a triangle such that a +
2 2 be a natural
and let n > natura l number. Prove that
21/n
(( an+b n ))l/n + ((a"+b . )l/n( ))l/n
l/n . ) l/n ( l/n 1/n
an+bnn an+b n <1+-2-.
a, b, c, d be positive real numbers. Prove that

197. Let a,
'"'
'""' a > 1.
>1.
~ b+2c+d-
cychc
cyclic
198. (Balkan Olympiads,

Olympiads, 2005) Let a, a, b,
b, c be positive real numbers such that
+ b)(b +
(a + + a) == 1. Prove that
+ c)(c +
3
ab + bc + ca ::;
- 4.
< -.
4
199. (Iran, 2005) Let ABC be a right-angled tri angle with A = 90°.
triangle 90°. Let AD
bisector of angle A, and labe the ex-centre opposite to A. Prove that
be the bisec:tor
AD
AD <2v'2--11..
~
- <v
DIa -
200. Let x, y, z be non-negative real numbers such that x +

+ y+
+ z = 1. Prove
that
x 2 + y2 + z2 + 18xyz ::; 1.
circum-circ:le r,
201. Let ABC be a triangle with circum-circle f, and G be its centroid. Extend
AG, BG, CG to meet rf in D E , FFrespectively. Prove that
D,, E,
AG+BG+CG::;GD+GE+GF
AG +BG +CG::; GD +GE+GF.
P7'Oblems
Problems 213
202. Prove with usual notation that in a triangle ABC

ABC,, the inequality
(a+b+c)(ha+hb+h c ) ~ 18ß.
203. (Short-list, IMO-2004) Let a, b, c be

bethree positive real numbers such that
ab + bc + ca = 1. Prove that
1 ) 1/3
1/3 (1 / 3 (1
))11/3 ))11// 33 1
(~
-;;- + 6b + b + 6c + ~ + 6a ::; abc·
::::: abc '
204. (IMO
(IMO,, 1974) Let ABC be a triangle. Show that there exists a point D on
AB such that CD 22 = AD · BD if and only if VsinAsinB
AD· vsinAsinB :::::
::; sin(Cj2).
sin(C/2).
205. (Short-list, IMO-2004) Let a1,a2 a1 ,a2,a3," .,a n be n > 1 positive real
,a3, .. .,an real numbers.
numbers.
For each k, 1::::: k::; n, let A k = (al +a2+··
1 ::; k::::: ·+ak)/k. Let gn
+a2+" ·+ak)jk. (a 1a2'" ·an)l/n
gn = (a1a2·· an )l/ n
and G nn = (A 1A 2 ··· A
= (A1A:~··· A nn )l/n
)l/ .. Prove that
n
n
(
A:
G ) l1/n
/n
+ ~: :::;: : n + 1.
Find the cases of equality.
206. Let x, y,
y , z be real numbers in the interval [0,1].
[0, 1] . Prove that
3(X 2y2 + y2 z 2 + z2 X2) - 2xyz(x + Y + z) : : : 3.

207. (UK
(UK,, 1999) Let x , y ,zz be non-negative real numbers such that x+y+z
x+y+ z = 1.
tha t
Prove that
7(xy + yzy z + zzx S; 2 + 9xyz
x)) ::; 9xy z..
208. (South Africa, 2003-04) Let x, y , z be real numbers in the interval [0,1].
[0, 1].
Prove that
x y z
-
- --++ - -
- +- +- -
y z + 1 zx + 1 xy + 1 -
yz
-< -2 <.
2.
(Greece, 2003) Let a, b, c, d be positive reals such that a 3 + b3 + 3ab =

209. (Greece,
c + d = 1.
1. Prove that
(a+ ~r + (b+ ~r + (c+ ~r + (d+ ~r ~ 40.

210. Let x, y , zz be positive real numbers such that x + y + z == xyz.
xy z. Prove that
1 1
1 11 33
---=== + + <-.
Jl.+? -< -.
--===
VI
V1 Ji+Y2
+ x 2 }1 + y2 J1+Z2 2
214 Problems
211. Let P be an interior point of a triangle ABC whose sides are a, b, c. Let
R l1 = PA, R 2 = PB, and R 3 = PC. Prove that
(R 1 R 2 + R 2R 3 + R 3R l ) (R l + R 2 + R 3) ;::: a2 R l + b2 R 2 + c2 R 3.
t1hh, .. .,tn be positive real numbers such that

212. (IMO, 2004) Let tlhh,
n2 + 1 > (tl + t2 + ... + t n ) (~

tl
+ ~ + ... + ~) ,
t2 t n
2: 3 is an integer. Show that for each tripie (j, k, l) with 1 :S j

where n ;::: <k<
l :S n, there is a triangle with sides tj, tk, tl.
213. (Moldova, 2004) Let x, y, z be non-negative real numbers. Prove that
x 3 + y3 + z3 ;::: x 2yYi + y2y'ZX + z2"fiiY.
214. (Romania, 2004) Find all positive real numbers such that
4(ab+ bc+ca) -1;::: a2 + b2 +c2 ;::: 3(a 3 + 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)(c+1) - 4·
216. (CRUX, 2000) Suppose a, b, c are positive real numbers such that a 2 +
b2 + c2 = 1. Prove that
_
1 + 1
a2 b2 +
1
c2 ;::: 3+
2( a3 + b3 + c3 )
abc
217. (Taiwan, 1999) Let ABC be a triangle with circum-centre 0 and circum- °
radius R. Suppose the li ne AO, when extended, meets the circum-circle of
OBC in D; similarly define E and F. Prove that
OD· OE· OF ;::: 8R 3 .
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)
(l+y)(l+z) (1+z)(1+x)
(l+z)(l+x) (1+x)(1+y) - 4·
(l+x)(l+y)
Problems 215
219. (Nordic Contest, 2003-04) Let D, E, F be respectively the points of con-

tact of the in-circle of a triangle ABC with its sides BC, CA, AB. Prove
that
BC CA AB
FD + DE + EF 2: 6.
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
- - - + c+d+a + d+a+b + a+b+c >-.
b+c+d - 3
221. (Bulgaria, 1974) Find all real A for which the inequality
xi + x~ + x~ 2: A(XIX2 + X2 X 3),
holds for all real numbers Xl, X2, X3.
222. Let a, b, c be positive real numbers such that abc = 1. Prove that
abc 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 :S 1 + c, b + c :S 1 + a, c + a :S 1 + b.
Prove that
a2 + b2 + c2 :S 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
abc 9
- -+-
1 + bc
- + -1 +
1 + ca
-ab
>- -10.
225. (Indian Team Selection, 2006) Let ABC be a tri angle with sides a, b, c,
circum-radius Rand in-radius T. Prove that
2
R 64a 2 b2 c2 )
2T 2: ((4a 2 - (b - C)2) (4b - (c - a)2) (4c 2
2 - (a - b)2)
226. (INMO, 2004) Let R denote the circum-radius of a triangle ABC; a, b, c

its si des BC, CA, AB; and Ta, Tb, r c its ex-radii opposite A, B, C. If 2R :S Ta,
prove that
(i) a > band a > c;

216 Problems
(ii) 2R> 7'b and 2R > 7'C.

7'c.
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
abbccdd a > bacbdca d.
229. (Ukraine, 2001) In a tri angle ABC, let AAl, CC] 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 Ai, parallel to AAl and let it meet
CCI in N and BC in Q. Similarly, let the line through M parallel to CCI
meet AAl in Hand AB in P. Let d l , d l , d3 respectively denote the distances
of H, I, N from the line PQ. Prove that
dl d2 d 2ab 2bc 2ca
-+-+
d d
-d3 > -.--+-,--+-,--.
- a 2 + bc b2 + ca c2 + ab
2 3 l
230. (CRMO, 1996) Let ABC be a tri angle and h a be the altitude through A.
Prove that
(b + C)2 2: a 2 + 4h;.
(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 bc positive real numbers such that y3 + Y ::;
x - x 3 . Prove that
(a) y < x < 1; and
(b) x 2 + y2 < 1.
233. (CRMO, 2005) Let a, b, c be three positive real numbers such that a + b+
c = 1. Let
,X =min{a 3 +a 2 bc, b3 +ab2 c, c 3 +abc2 }.
Prove that the roots of the cquation x 2 + x + 4,X = 0 are real.
234. (CRMO, 2006) If a, b, c are three positive real numbers, prove that
a +1 b +1 c +1
2 2 2
- - + - - + - - > 3.
b+c c+a a+b-
Problems 217
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) :s; d 2 •
236. (INMO, 2007) If x, y, z are positive real numbers, prove that
(x + Y + z)2(yz + zx + xy)2 :s; 3(y2 + yz + Z2) (Z2 + zx + x 2) (x 2 + xy + y2).
237. (USAMO, 1974) Suppose a, b, c are positive real numbers. Prove that
aabbcc;::: (abc)(aH+c)/3.
238. (Bulgaria, 1988) Find aB real p and q for which the equation
2
x4 _ ~X3 + 4qx 3 - 3px + p2 == 0
q
has four positive roots.
239. (Russia, 2005) Let ai,a2,a3

al,a2,a3 be real numbers, each greater than 1. Let
S = ai
al + a2 + a3 and suppose S < a;/(aj -1) for j = 1,2,3. Prove that
111
---+
---+ +
+--->1. >1.
ai
al + a2
az az + a3
a2 a3 + ai
al
Chapter 6
Solutions to problems
1. We observe that
(t
n
L an+l (a(an+1
n+l - aj) na;+l -
na;'+1 an+1
aj ) an+l
j=l
j=1 J=l
J=1
l)a;,+!o
(n - l)a; +l '
n
L Jaj(an+1
jaj(an+! - aj) :S 1) an+l
::; J(n --1) an+l'o
j=l
j=1
However, this follows from the Cauchy-Schwarz inequality:
n n
L jaj (a n+l - aj)
Jaj(an+l L ra;j(an+1 - aj)
j=l
j=1 j=1
<~
n
L (an+!
(a +1 - n aj)
j=l
j=1
~ J(n --l)a n +l
l)an+l
J(n - 1) an+!o
an+l'
2. The inequality may be written in an equivalent form:

a2 b2 c2
- --+
ab + 2ca bc + 2ab
+ ca + 2bc >l.
>1.
--
(L
2
(L
2
(a+b+c)2 = a J ab + 2ca)
vab+2ca)
cyc 1
.
I lC
CYCIC
0 vab + 2ca
Jab+
<
< ((~ab
L ab :22ca )
cycltc
cychc
(~(ab + 2Ca)) .
cycltc
cychc
2Ca)) 0
It follows that
L. -
--
a2
---> >
(a + b + c) 2
(a+b+c)2
ab+2ca - 3(ab+bc+ca)
o 3(ab+bc+ ca)
. 0
cychc
cy cltc
220 Solutions
(a + b + c) 2 2: 3 (ab + bc + ca).
Equivalently, we need to prove that a 2+b 2 +c2 2: ab+bc+ca, wh ich is clear from
the Cauchy-Schwarz inequality. (Or one can use (a-b)2 + (b-C)2 + (c-a)2 2: 0.)
3. We observe that
(a + b + c)2 - 3(a 2 + b2 + c2 ) = 2(ab + bc + ca - a 2 - b2 - c2 ) ~ o.

Thus, we gct
3 (a 2 + b2 + c22)) 2: (a + b + c) 2 2: (abc) 2 .
On the other hand, we also have
a2 + b2 + c2 2: 3(abc)2/3,
as a consequence of the AM-GM inequality. This gives
(a 2 + b2 + C2 )3 2: 27(abc(
Multiplying these inequalities, we get
9(abc)4 ~ (a 2 + b2 + c2 (
Taking the fourth-root, we get the desired inequality.
4. Taking A == ~ 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
.1< a+b+c B~ 2(a+b+c)

B<
- .
- 3 ' 33
Thus
6 27
- >
>--- .
AB - (a + b + c)2 .
AB - (a + b + c)2
5. We have
(1 - a)(l - b)(l - c) 1 - (a+ b + c) + (ab + bc + ca) - abc

-1 + (ab + bc + ca) - abc.

1
o ~ (1 - a)(1- b)(l - c) ~
::; 27.
Solutions 221
But, this is a consequence of the AM-GM inequality: observe a 0, and similarly 1 - b > 0, 1 - c > 0;
(l-a)(I-b)(l-c) ~ (
(1 - a)(1 - b)(1 - c) S; C- a
1 - a + 1- b + 1- C
+ 1 ;b+l-
3 r
c)' '3
1
27
6. First we observe that .lab + c 2: v-;;:bv-;;.b + c. In fact, this is equivalent to

c 2: c2 + 2cv-;;:b,
2cv-;;.b, which is equivalent 2v-;;.b, since c > O. Using
to 1 - c 2: 2v-;;:b,
2v-;;.b. Similarly, we derive v'bC + a 2: v'bC + a,
1 - c = a + b, we get a + b 2: 2v-;;:b.
Vca + b 2: ,;ca + b. Thus it follows
.lab + c + v'bc + a + Vca + b 2:> Jbc + yea

v-;;.b +
v-;;:b yCa + a + b + c
1 + Vah + Jbc + yea.
yCa.
7. Using cd = I/ab and da = l/bc, the inequality to be proved is
(1 + {I:
ab) a + ab(/+ c) } + (1 + {I!
bc) b + bC(II+ d) } 2: 4.
But,
4
-1- + ----,-----:-
1
----,-----,- >
1+a ab(1 + c) 1+ a + ab + abc
1 1 4
- - + -=-----:-----,---:,-
-=-----::--:----:-:- >
1+b bc(1 + d) 1 + b + bc + bcd·
bcd'
Thus
+ ab 4(1 + ab) 4(1 + bc)

> ---'-------,=-----'-----:-+---'-----:---'---,:--:
"'~
" 1 -- > _----'-_---::---'-----:-+_-:c'---:-:---'----:--:
~
cyclic
l+a
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.
222 Solutions
8. Using the AM-GM inequality, we have
abc + abd + acd + bcd

4
~{ab(c+d) Cd(a+b)}
2 2 + 2
~ { (a + b)2 (c + d) (a + b) (c + d)2 }
< 2 8 + 8
(a+b). (c+d). (a+b+c+d)
224
< (a+b:C+dr
«
This shows that
y-
abc + abd + acd + bcd < a+b+c+d
3/abc+~bd+acd+-t;;;;j
4 -
a + b + c + d < /aa2 +b
4 -y
-
+ b2 +c
+ c2 +d
4
+ d2
'
j
where we have used the root-mean-square inequality in the last step.
9. 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 + c 2 + 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> o.
But this is precisely (1 - a)(l - b)(l - 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
L x2 + L xy + L x 2y + L xy2 + 2xyz < 1.

This reduces to 1 - xyz < 1 after using x + y + z = 1. The result follows from
xyz > o.
Solutions 223

1 1 1 9
-+-+->--
-+-+->
ab c-a+b+c
.
9
a+ + + aa + b + c ;:::>
a+ b +cc - 4J3.
Equivalently, we need to prove
(a + b + C)2 - 4J3 (a + b + c) + 9;:::

92: O.o.
Or
( a + b + c - J3) ( a + b + c - 3J3) ;:::
2: O.o.
However, we have
a + b + c :::; J3 V~2 +
J a 2 + b2 + c22 = J3.
Hence a + b + c - 3J3 :::; 0 and the inequality follows.
11. By the AM-GM inequality, we have
abc:::; (a +: + c) 3 = 8.
Thus
4 4 3
2- - <2- - =-.
abc - 8 2
On the other hand,
1 1 1 9 3
- +- +- > =-.
ab c-a+b+c 2
Thus,
3 1 1 1 4 3
< - + -b + -c == 2 - -abc-2
-2-a <-.
This shows that equality holds in the AM-GM inequality. Hence a = b =
= c,
giving (a,b,c) = (2,2,2).
12. Using the AM-GM inequality, we have 2bc :::; b2 + c2 = 1 - a 2 and similar
bounds hold for 2ca, 2ab. Thus,
a2 b2 c2 a2 b2 c2
-:----::-:,- +
-:----::-:- + 1 + 2ab > -- +- - +- -
I + 2bc 1 + 2ca - 2 - a2 2 - b2 2 - c2
=
= -3 + 2 ( 1a2 + 1b2 + 1)
2_ 2_ 2_ c2 .
224 Solutions
1 1 1 1 (3 ) 9
2-a2 + 2-b2 + 2-c2 2':"2 5 +3 = 5·
This follows from AM-HM inequality:
1 1 1 9 9
- - . +22-
22 - a2
- +22-- -c2>
- b2 6 - (a 2 + b2 + c2 )
- 6
=-
5·
13. Introduce new variables

a b c
x
x--------
- Y = ---,-- Z =
z =-- .
- a+b+c' a + b + c'
Y=a+b+c' a+b+c
The inequality may be written in the form
aX+ßy+,z+2v(a ß + ß ,+,a)(xy+yz+zx):s 1.
aX+ßY+rZ+2)(aß+ßr+ra)(XY+YZ+Zx):s
Observe that x + Y + zZ == 1. Using the AM-GM inequality
nx + /3y + ~(Z + 2v(a!3

2)(aI3 + /h + ,a)
ra) (xy + yz + zx)
0 2 + x2 ß2 + y2 "(2 + Z2
,2 z2
< 22 + 22 +
+' 22 + (aB+ß,+,a)
(aB+ßr+r a ) + (xy+yz+zx)
1 2 1 2
"2(a+ß+r) +"2(x+ y +z) =1.
"2(a+ß+,)
14. By the Cauchy-Schwarz inequality,
(ajb 2 + 1+ bj;2+1/ :S (a 2 + b2 ) (a 2 + b2 + 2).

But it is easy to see that x(x + 2) < (x + 1)2 for any positive x. Thus
(a 2 + b2 ) (a 2 + b2 + 2) < (a 2 + b2 + 1) 2,
which gives the desired inequality.
15. Let S denote the sum to be minimised. Define

1 1 1 1 .
H=l+-+-+···+- and 1O '=-.-,l<J·<n.
10jj jH' 1 -:S J -:S n.
= JH
2 3 n'
Then 10j > 0, for 1 :S j :S n, and 101 + 102 + ... + 10 n == 1. Using the weighted
AM-GM inequality, we obtain
S
J1 (xn
n n
H = L10jX
,
j
J
> Wj
J=1
(~xrH
Solutions 225
Using the GM-HM inequality, we also get
rr
n
n
Xj 2: 1 1 1 1 =l.
j=l -+-+-+ ... +-
Xl X2 X3 Xn
It follows that S 2: H, and equality holds if and only if Xl = X2 = ... = Xn.

1 1 1
Thus the least value of S is H = 1 + - + - + ... + -.
2 3 n
16. We have
abc d a2 cd + b2 da + c2 ab + d2 bc
-+-+-+-
b c d a abcd
(ab + cd)(ad + bc)
abcd
1
(ab + cd)(ad + bc) ::; 64'
The AM-GM inequality gives
(a+b)(c+d)::; (a+b+c+d)2 < ~

2 - 4'
and
< ( ac + bd ; ad + bC) 2
r
(ac + bd)(ad + bc)
(a + b)~c + d)
(a
1 1
<
82 64
17. Introducing new variables X, y, z by X = 1 - a, y = 1 - band z = 1 - c, we

see that x, y, z are positive and X + Y + z == l. The required inequality is
(1 - x)(l - y)(l - z) 2: 8xyz.
Equivalently, we have to prove that
xy + yz + zx 2: 9xyz.
1 1 1
-x + -y + -z-> 9.
Since x +y+z = 1, this follows from the AM-HM inequality.
226 Solutions
18. We introduce '\, f-t and v by 2'\ = + c, 2f-t =

= b+ + a and 2v =
= c+ + b. Then
= a+
a= + v - '\,
= f-t + b= + ,\,\ -- f-t,
= v+ c=
= ,\ + f-t - v.
,\ +
Thus, we obtain
+ b ++ c =
2a + + v),
= 2(f-t + + c ++ a =
2b + + ,\),
= 2(v + ,\), + a ++ b =
2c + + f-t - v).
= 2('\ +
The inequality to be proved is:
L
" 4(f-t + v)2
4(f-t
~--~--~--~ < .
v)2
+ v - '\)2 ++ 4A2
~. 2(f-t +
< S8
4,\2 - .
cychc

4(f-t + v)2 2(f-t + v)2 4(f-t + v)2
2(f-t + v - ,\)2 + 4,\2 + V - ,\)2 ++ 2A2
(f-t + + v)2 ++ 2,\2 .
2,\2 :::; (f-t +
Herewehaveused (f-t+v)2:::; (f-t+v-,\)2+2,\2. Using (f-t+V)2 :::;2(f-t2+ v2),

we get
4(f-t + V)2 = 4 < _--,--......,.-4_ _-,- 4(f-t2 + v 2)

(f-t+v)2+2,\2 1+ (2A2/(f-t+v)2) -1+(,\2/(f-t2+ v2)) f-t2 + v 2 +,\2'
<"
Thus,
" --,.-_4--,(,--f-t_+-:-V::--)2_-::- 4(f-t2 + V2) _ S
~ 2(f-t + v - ,\)2 + 4A2 - ~ f-t2 + V2 +,\2 - .
cychc cychc
+a
19. We know that 1 + 2Vb and 1 ++ c 2': 2>/C. Thus, we get
2': 2ya, 1 ++ b 2': 2Vb
S + a)(l
8 == (1 + + b)(l
a)(1 + + c) 2': Sv'O];;.
b)(1 + 8v'(;];;.
It follows that abc:::; 1.
20. 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
L (y + Z)2(Z + x)(y - x) 2': O.

cyclic
Expanding the left hand side, this reduces to

+ yz3 ++ zx 3
xy3 + 2': + xy2 Z ++ xyz2.
x 2yz +
x 2yz + xy2z + xyz2 = L (X 3/ 2Z1/ 2) (X 1/ 2yz l/2)

cycJic
:::; (x 3z + y3 x + Z3 y )1/2 ( xy 2z + xyz2 + x 2YZ)1/2.

Solutions 227
This irnplies that

x 2 yz + Xy2 Z + XYZ2 :::; Xy3 + YZ3 + ZX 3 .
This also follows frorn Muirhead's inequality, since (2,1, 1) ~ (3,0,1).
21. Using the Cauchy-Schwarz inequality, we see that

~
L
+ a2
n ~ > (al +a2 + an )2
+ ... +a ___1
2 - 2::
~ ~
j=l
",n
j=l 2 - aj
n
a"J - 2::}=1
"a"
L..j=l a")
ajJ (2 - aj)
}
-- ",n n
L..j=l
)"--1 aj
2.
a)"2'

1 n
--=::-;;;---".>--.
2 ~ --.
2::7=1 aj
",n
2 - L..j=l a; - 2n - 1
This is equivalent to n2::7=1 a; ~2: 1, which follows frorn the Cauchy-Schwarz
inequality:
1 = (al + a2 + '" + a n )2 :::; n(ai + a~ + ... + a;').
22. We observe that
n 2 2 n
'~" a·J - a+
J
1
= '~" (aj - aj+1) = 0.
j=l aj + aj+1 j=l
Thus, we get
n 2 n 2
'~" aajj _ '~ a+1
" aj+1
~
}=1 a
j=l J +
~a'+a'+l a J j=l J ~ aJ"+1 .
~
+1 -= ~a'+a'
a"
Hence, if S is the required surn, we have
4S
4S = fL 2(aJ ++ aJ+l)
n 2(aJ aJ+l)
a") + aj+1
a+1
j=l aj }
~(aj+aj+1)2
> t ( a j +a j +1)2
~
j=l
j=l + aj+1
aajJ +a+1
J
n
L (aj + aj+1) =
L(aj+aH1) 2.
=2.
j=l
2: 1/2.
It follows that S ~
23. This follows frorn the Cauchy-Schwarz inequality:
+ 2 + 3 + 4)2 Va
( Va+ Vb y'C + 4 v'd) 22
Vb + 3 yIc
Va + 22Vb+ v'd)
(1
Va Vb 3 y'C y1c+4v'dv'd
< ((~+~+~+
1 4 9 16) ((a+b+C+d).
~+b+~+d )
a+b+c+d. 1;)
228 Solutions
Thus, it füllüws that

1 4 9 16 100
-+-+-+->
-+-+-+->
abc d-a+b+c+d
..
24. Since aj < 1/2 für each j, we see that 1 - 2aj > 0, für 1 s: j s: n. Using
the Cauchy-Schwarz inequality, we have
(t,ajr (t VI -
j=l
aj
2aj
Vl- 2aj ) 2
< (t J=l
1 :L) (t(l- J J=l
2a j ))
(n - 2) (t, :;aJ· 1
Hence it füllüws that

n a2 1
""" J > _,
~ 1- 2aj - n - 2
j=1
because üf 2:.7=1 aj = 1.
25. The Cauchy-Schwarz inequality gives
(Xl + X2
(Xl + + ...
X2 + ... ++ Xx nn ))22 s: (Xl-Yl
(Xl
Yl
+ -X2
+ + ...
X2 +
Y2
Y2
... ++ -xxn)
n)( + X2Y2
XlYl +
(XIYI + ...
X2Y2 +
Yn
Yn
... ++ XnYn
xnYn).
) .
Xl YI
The inequality Xl + X2Y2
YI + + ...
X2Y2 + ... ++ XnYn
XnYn s: Xl ++ X2X2 ++ ...
Xl ... ++ XXnn gives the result.
26. Using the Cauchy-Schwarz inequality, it suffices tü prove that
2 2 2
Xl
Xl X2 Xn
---''--'''''2 +
--''--'''''2+ 2 + ... +
2+"'+ 2 < 1.
0<1.
((l+xi)
1 + Xl2) ( 2 2)
1 + Xl + X 2
(l+xi+x~) (
1 + Xl + X 2
2 2 + ... + X 2)
(l+xi+x~+"'+X~) n
But für j 2': 2, we have

X2
J
I+X 2lI
( 1+X +X 222 +"'+X 2)2
j
X 2j
<
+
( 1 + Xl
Xl2 + XX222 ++ ...
+ ... ++ xx 2j _ ll ) ( 1 ++ Xl + xx 222 ++ ...
Xl2 + ... ++ xx j2)
1 1
+
1+ 2
Xl
Xl + x 2 ++ ...
+ ... ++
2
x 2j _ l 1+ xi + x~ + ... + X]
We alsü have
xi 1
-(-------'-2....,.,-)2
1 + Xl
s: 1 - -(-
1 + .T l
2)
Solutions 229
Summing all these, we get
- xi + x~ x~
+ ... + -----'-"----....,.
(1+:y;i)2 (1+xi+x~)2 (1+xi+x~+ ... +x~)2
1
< 1-
1+ Xl
2
+ X z + ... + x.2n
2 < l.
27. Using the Cauchy-Schwarz inequality, we have
+ b'2 c ++ c2 a) (al? ++ bc2 ++ ca 2 )

(a 2 b +
> (a 2 vt;;; + b2,jCa + c2 yfab) 2
= abc(a 3 / 2 + b3 / 2 + c3 / 2 ) 2 .
2 3
+ b3 / 2 ++ c3 / 2 )
3(a 3 / 2 + ;::: + b ++ c) .
(a +
This follows from Hölder's inequality: taking p = 3/2 and q = 3, we obtain
+ b ++ c) :::; (a 3 / 2 ++ b3 / 2 ++ c3 / 2 ) 2/3 (1 ++ 11++ 1) 1/3,
(a +
which gives the desired inequality.
28. We have
P(a)P(l/a) + ba ++
( aa 2 + c) (:2 ++ ~ ++ c)
a2 + b2 + c2
+(ab ++ bc) ( a ++ ~) ++ ac ( a 2 ++ ~2 )
+
+ b2 ++ c2 ++ 2ab ++ 2bc ++ 2ac
> a2 +
(a + b + C)2 == P(1)2;
we have used the AM-GM inequality.
29. The inequality is equivalent to

~ ~ ? ~ ~ 5
--- + + - - - ++ +
+ +
+ > -.
+ ac bc ++ bd cd ++ ce de + da ea + eb - 2
ab +
Using the Cauchy-Schwarz inequality, we see that
(a+b+c+d+e)2
a2 b c2
+ dd'2 ++--"""C"
e
2 2) 2
......,,----+--+
< ( ......,,---+
+ ac
ab +
-- +
+ bd
bc +
---:-
+ ce de ++ da ea ++ eb
cd +
x + ac ++ bc + bd ++ cd ++ ce ++ de ++ da + ea + eb) .
( ab +
230 Solutions
Thus, we obtain
a2 b2 C2 d2 e2 (a+b+c+d+e)22
--:---+--+ +
ab + ac bc + bd cd + ce de + da
+ ea + eb >
- L:sym ab
.
.
where
Lab = ab + ac + ad + ae + bc + bd + be + cd + ce + deo
E~=~+~+~+~+k+M+~+~+a+~.
sym
rym
(a + b + c + d + e)2 2: ~ E
Lab.
ab.
sym
2(a 2 + b2 + c 2 + d2 + e2) 2: ab + ac + ad + ae + bc + bd + be + cd + ce + deo

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 2: o.
Hence the result follows.
30. Let us introduce a2 + b2 - c2 == z, b2 + c2 - a2 =

= x and c2 + a2 - b2 =
= y.
Then we have
E Va 2 +-b2~ c2Va 2 - b2 + c2
cyclic
fiY + yYZ + yZx

~{fiY + yYZ + yYZ + yZx + yZx + fiY}
~{fiY + vXz + JYZ + y'yX + yZx + v'zY}
< ~ {{ V
J (x + y) (x + z) + V
J (y + z) (y + x) + V
J (z + x) (z + y) }}
~ { v'2c2 . 2b2 ++ J2a 2 . 2c22 ++ J2b 2 . 2a22 }}
ab + bc + ca.
31. We have
a2
a2
+ 1 + b2 + 1 +
b2 c2
c2 + 1 = 3 -
(1+
a2 1
1
+ b2 + 1 + c2 + 1
1) = 1.
Solutions 231
Cauchy-Schwarz inequality, we observe that

Using the Cauehy-Sehwarz
(
a2 b2
a 2 + 1 + b2 + 1 + c2 C+ 1
2) (a + 1+ b + 1+ c2 + 1) > (a+b+Cr
2 2
Thus, we get
a2 + b2 + c2 + 3 2: (a + b + C)2.
This simplifies to
3
- 2"'
< -.
ab + bc + ca :S 2
concavity of the function f(x) = x 1 / 3 on the interval (0,00),

32. Using the eoneavity (0,00) , we
obtain
88 (abcr/3
(
abc )
1/3
+ (a3+~+c3Y/3
+ 3
+ a ±b ±c
< 9 ( 8abc +
(a3+b3+c3)1/3 < 9 ( 8abC
3
;'H;+,' ) 'I'
3 3 3 ) 1/3
(24abc + a3 + b3 + c3) 1/3

33(24abc 1 /3
sufficient to prove that

Thus, it is suffieient
24abc+a 3 +b3 +c3 :s (a+b+c)3.

This reduees
reduces to
a2 b + b2 c + c2 a + ab 2 + bc 2 + ca 2 2: 6abc,
whieh consequence of the AM-GM inequality.

which is a eonsequenee inequality. Equality holds if and only
= =
if a = b = c.
33. We use induetion

is true for all j :S
induetion
induction hypothesis
hypo thesis gives
°: :;
:::; 1 and Cj
Since 0 :S P :S
:::; n - 1. Sinee °
induction on n. The statement is immediate for n = 1. Suppose it
;::: 0 imply 1 - P + PCj 2:
Cj 2: ;::: 0, the
rrII (1 - rrII (1 -
n n-1
n- 1
P + PCj) P + PCj) . (1 - P + pcn )
j=1
j=l j=1
j=l
n-1
< (1- P + P II Cj) . (1 - P + PCn ) . n) .
j=1
j=l
The induetion
induction step is eomplete onee the inequality
n-1
n- 1 n
(1 - Pp + P II Cj) . (1 - P + PCn ) :S 1- P + P II Cj
Cj) . n ) :::; Cj
j=1
j=l j=1
j=l
232 Solution"
is proved. This is equivalent to
rr rr
n-l n-l
(p - p 2 ) { Cn + Cj - 1- Cn Cj} ::; O.
j=l j=l
Since p - p2 2': 0, this is equivalent to
('rrTI
n-l
Cj - 1) (c 1) 2': O.
n -
j=l
J=l
The given conditions on Cj'S now imply the result.
34. The inequality may be written in the form
L (1 -
x j ) 4 < L X]
TI(1-xj) - TIXj·
By symmetry, we may assume that Xl 2': X2 2': X3 2': X4. If we set 1 - Xj = Yj,
=
then Yl ::; Y2 ::; Y3 ::; Y4· An easy computation shows that
LX]
-- - 4 =
(xi - x~)2 + (x~ - x~)2 + 2(XIX2 - X3X4)2
..:.....:=----........::-'-----:........::..----"-'----'---------'---
TI Xj Xj X2X3X4
Xl2 -
((XI X~)2
X 2)2
2 X~)2
X23 -- X4
+ ((x§ 2)2 + 2 ( XIX2
2(XlX2 - X3X4 )2
X3 X 4)2
Xj X2X3X4
XIX2
> ( Yl2-2Y2
)2
+ (Y32-2Y4
)2
+ 2 ( Yl!J2 - Y3Y4
)2
YIY2Y3Y4
Since Xl + X2 ::; 1, we have Xl - X2 2': (XI - X~) which may ,)e written in the
form Xl Yl 2': X2Y2· Suppose p and q are real numbers such that p 2': q > l.
Then it is easy to see that
1 1
P+-2':q+-
P+-2':q+-
P q
holds. Taking p = x1/x2
xI/x2 and q = Y2/Yl, we obtain
Xl ;[;2 Y2 Yl
-+-
X2
> -+-.
Xl - Yl Y2

2 2 2 2
Xl +X2 > Yl +Y2
XIX2 YIY2
Solutions 233
or equivalently in the form

2 2
(Xl + X2) >
(Xl
-'-----------'- (YI + Y2)
(Yl
> ---'---------'-
XlX2
XIX2 - YlY2
YIY2
But observe that (Xl - X2)2 = (Yl

(YI - Y2)2 and X3X4 ::; (1- X3) (1- X4) = Y3Y4.
Thus we obtain
(2 2 (Xl
Xl - X22)2
Xl - (Xl
(Xl +X2) (Xl -
- X2)2
XIX2 X 3 X 4
XlX2 XlX2
XIX2 X3 X 4
(YI + Y2)2
(Yl (YI -
(Yl Y2)2
>
YlY2
YIY2 Y3Y4
(2
YI - Y22)2
Yl
YlY2Y3Y4
YIY2Y3Y4
Similarly,
(X~ _ X~)2 > (Y~ _ y~)2
XIX2 X 3 X 4
XlX2 - YIY2Y3Y4
YlY2Y3Y4
may be obtained. Taking P == Xl X2 / X3X4 and q == Y3Y4 / Yl

YI Y2, it can be easily
verified that p 2: q 2: 1 . We hence also obtain
2 2
2(XIX2 -
2(XlX2 X3X4)2
X3X4) > 2(YIY2 - Y3Y4)
2(YlY2 Y3Y4)2
XIX2 X 3 X 4
XlX2 - YlY2Y3Y4
YIY2Y3Y4
Xl =
Combining all these, we get the result. Equality holds if and only if Xl = X2 =
X3 = X4·
=
35. (By Kshipra Bawalkar) Since 0 < Xj ::; 1/2, we have 1/2::; 1 - Xj < 1 for,
L(1-
1::; j ::; n. Thus L LXj, rrXj
Xj), LXj,
( 1 - Xj), fIxj and rr(I
fI(1 - Xj)
Xj) are all positive. The
inequality may be written in the form
( L;'=l X J )
rr;=l Xj <
n
)n
r: (I-xj) -
rr J=l ("n ~J=l (I-x J )
Let A = (Xl + X2 + ... + x n ) In. Let Xl and X s denote respectively the largest
and the smallest among Xl,X2,X3,
XI,X2,X3, ... ,X n . We show that
rr;=l Xj
rr;=l (1 - Xj)
XlX2···
XIX2··· A··· + X s - A) ... X n
(Xl
<
xI)(1 -
(1 - xl)(1 X2) ... (1 - A) ... (1 - Xl - X s + A) ... (1 - x n )'
234 Solutions
where the right side is obtained by replacing Xl and Xs respectively by A and

Xl + Xs - A. We need to prove that
XIX s A(XI + Xs - A)
(1- xl)(l - x s ) ::; (1 - A)(l - Xl - Xs + Ar
We first observe that °< Xl + X s ::; 1. The above inequality simplifies to
(A 2 - A(XI + x s ) + XIX S ) (1 - Xl - x s ) ::; 0.

Since 1 - Xl - Xs 2: 0, this is equivalent to
A2 - A(XI + x s ) + XIX s ::; 0.
But
A2 - A(XI + x s ) + XIX s = (A - Xl) (A - x s ) ::; 0,
since Xl ::; A ::; x s . Now consider the set
{Xl,X2'" ,A, ... ,(xl+xs-A), ... ,X n }
°
obtained by replacing Xl, Xs respectively by A, (Xl +X s - A). This again satisfies
the hypothesis of the problem, as < A ::; 1/2 and < Xl + Xs - A ::; 1/2
(since Xs 2: 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
TI7=l X j <~
Ili=l(I-xj) - (I-Ar'
However
An (nA)n
(1- Ar (n - nAr
( L:7=1 Xj) n
(n - L7=1 Xj) n
( L7=1 Xj) n
(L7=1 (l_Xj))n"
r"
It follows that
( L7=1 Xj) n
TIi",Xj <
TI7=1(1-xj) - (L7=1(1-Xj)
Solutions 235
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 apower of 2 and then fill up the
remaining gaps by coming back. The case n == 2 is simple. The inequality
required is
XIX2 (1 - Xl) (1 - X2)
--=--.:=----".2 < n
(XI+X2) - (2-XI-X2)
Simplification gives the equivalent inequality:
(Xl - X2)2 (1 - Xl - X2) 2': o.

Since Xl + X2 ::::: 1, the result follows for n = 2. We also observe that equality
X2 :::::
holds if and only if Xl = X2.X2. We use the case n = 2 to prove the result for n = 4.
Consider positive numbers XI,X2,X3,X4 E (0,1/2]. Taking YI = (Xl (Xl + x2)/2
and Y2 = (X3 + X4) 12, the result for n = 2 gives
YIY2 (1 - YI)(I - Y2)

----=.--=----~2 < "
(YI + Y2) - (2 - YI - Y2)
Substituting for YI and Y2, this reduces to
Xt X2X3X4 (Xl + X2) 2 (X3 + X4) 2 xlx2 x 3x 4 (2 - XI- X2) 2 (2 - X3 - X4)2

----'------'-~4---'-- ::::: •
( 2:;=1 Xj) ( 2:;=1 (1 - Xj) )
However, using the result for n = 2, we also have
XIX2(2- XI-X2)2 (1 - Xl) <

::::: (Xl + X2)2
Xl) (1 - X2) (Xl
X3 X4(2- X3- X4)2 < (1- x3)(1 - X4) (X3 + X4)2.
Using these on the right side and effecting some cancellations (Xl
(Xl + X2) 2 (X3 ++
2
X4) , we get
rr;=IXj
--"-----;-4 :::::
rr;=l (I-Xj) 4.
( 2:;=1 Xj) ( 2:;=dI- Xj))

This proves the result for n = 4. Using induction on k, this proves the inequality
for aB n of the form n = 2kk .
Take any n and fix k such that 2k ::::: n < 2k+1. Let A be the average of
these n numbers; A = (Xl (Xl + X2 + ... + Xn ) In. We consider 2 k
k +1 numbers
Xl, X2, ...

Xl, X2, X n , A, A, ... , A,
... ,, X
236 Solutions
°
where A appears 2 k +1- n times in the above sequence. We apply the inequality
for these 2k+1 numbers. (Note that < A :::; 1/2.) Thus we obtain
X1 X 2 ... x n A 2k + 1 -n
2 k +1
( 2:.7=1 :rj + (2k+ 1 - n)A)
:::;~
:::; 2 k,. +
, ,
1
X2) ... (1-
Xl) (1 - X2)'"
(1 - x1)(1- (1 - A)-
(1 - x n ))(l- Ar
o)k+l_ n
o)k+l
-n
.•
2:.;1=1 (1 -
( 2:.;'=1 Xj)
Xj) + (2k+ 1 - n)(l- A))
This simplifies to
X1X2 ... X n < (1 - xI) (1 - X2) ... (1 - Xx n )

X2) ...
An - (l-A)n
which is the required inequality for n.
Alternate Solution 11:
Here is another solution using Jensen's inequality for the convex function
f(x) == log (~-1),

defined on the interval (0,1/2). Writing f(x) = log(l- x) -logx, it is easy to
check that
1 1
f'(X)
f'(x)
1- x x'
_ 1 1
f"(X)
f"(x) (1 - x)2 + x 2
- (1-x)2+x
1- 2x
x2 (1 _ x2) > 0,
for x E (0,1/2). Hence f is convex on (0, 1/2). This implies that
n 1 1) :::; -1 (1
- - 1) .
~ log
log ( - n
2:. j =l (xj/n) n Xj
The exponentiation gives
( nn
2:.-;=1 Xi
-1 ) n
:::;II (1- - 1).
j=l
n
Xj
Some simplification leads to the required inequality.

Solutions 237
36. \Ve have for any n,
an ::; aj + an-j, 1::; j ::; (n - 1).
Summing over j, we obtain

n-1
(n-I)a n ::;2Laj.
(n-1)a :S2Laj.
j=1
j=1
We prove the given inequality by induction on n. It is obviously true tür n == 1.

Suppose it is true for an k < n. Thus, we have
bk = a1
al + -a22 + -a33 + ... + -akk >
- ak ,
for an k < n. Summing these over k, we get
n-l
n-1 n-1
L bk 2': L ak·
k=1
k=l k=l
But, we have
~ (a2 ) ( a2 a3 )
~ bk = a1 + al +"2 + a1 + 2 +"3
k=l
+ ... + ( a1 an-I)
al + -a2 + ... + -
an-1)
- .
2 n-I
n-1
This simplifies to
n-1
~ n- 2 n - (n -1)
~b bkk =
= (n - 1)a1
l)al + -2-a2
-2- a2 + ... + n _ .1 an-1.
an-I·
k=l
k=1
We thus have
n-2 n-(n-1)
(n - 1)a1
l)al + --a2 + ... + an-I
an-1 2': a1
al + a2
az + ... + an-l·
an-I'
2 2 n-I
n-1
Adding a1
al + a2 + ... + an to both sides and simplifying, we get
n ( a1 a2 + ... + nan-1)
+ "2 _ 1 + an 2': 2 ( a1 + a2 + ... + an-l ) + an·
But we note that 2( al +a2 + ... +a

2 (al n - 1 )) +a n 2': (n-1)a
+an-I (n -l)ann +an = nano Dividing
by n, we get
a2 an-I
an-l an
al + - + ... + - - + - > an·
a1
2 n-I
n-1 n-
This completes the induction step and hence the inequality is valid for an n.
238 Solutions
37. We observe that
J(x + y)(x + z) ::::

v(x ,;xy + y'xZ.
In fact, squaring on both the sides, this reduces to
x 2 + yz :::: 2x"fiji.
which is immediate from the AM-GM inequality. Thus
L
L}. X + J~
(x : y) (x + z)
<
<
L
~ x+~+y'xZ X
cyclic
eye Je cyclic
eyehe
L
""' Vx
vx
~ vx+yfij+Vi
-=1.
1
eyclie
cychc Vx + vy + Vi - .
38. As a consequence of the AM-GM inequality, we have
2
xy
xy ~~ ( X+y
(x +2 y)2
-2- )
= 1.
=1.
Thus, we obtain 0 < xy ~ 1. We write
X3y3(X3 + y3) (xy)3 (x + y)(x 2 _ xy + y2)

2(xy)3 ((x + y)2 - 3xy
3xy))
2 ( xy ) 3 (4 - 3xy) .
Thus we need to prove that
(xy ) 3 (4 - 3xy) ~ 1.
Putting z = xy, this inequality re duces to
Z3 (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 33 + 1 :::: O.
Here the expression on the left hand side factors in to (z - 1)2 (3z 2 + 2z + 1)
and (3z 2 + 2z + 1) is positive since its discriminant D = -8 < O. Or applying
r
the AM-GM inequality to the positive reals 4 - 3z, z, z, z, we obtain
Z3 (4 - 3z) ~ ( 4 - 3: + 3z ~ 1.
Solutions 239
39. Let AB = a, BC = b, CD = c, DA = d. We are given that abcd 24. 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 üf the AM-GM
inequality gives
ac + bd 2 2Jabcd 2 2V4 == 4.
We cünclude that ac+bd = 4. This fürces AC ·BD = 4, giving AC = BD = 2.
Each of AC and BD is thus a diameter. This implies that AB CD is a rectangle.
Note that
(ac - bd)2 = (ac + bd)2 - 4abcd ::::; 16 - 16 = 0
and hence ac = bd = 2. Thus we get a = c = ,;ac ,;ac = v2, and similarly
b = d = v2. It now follows that ABCDis a square.
40. The first inequality is simply arestatement üf the Cauchy-Schwarz inequal-

ity. Define Cl = min { al , a2, . .. ,an} and having defined Cl, C2, ... , Cj -1, define
Cj = min { {al, a2, ... ,an} \ { Cl, C2, ... ,Cj-I} },
für 2::::; j ::::; n. Then Cl::::; C2::::; ... ::::; Cn and {Cl,C2' ... ,Cn } is simply the
rearrangement of {al, a2, ... ,an} in increasing order. Similarly, we get the
rearrangement { d l , d2, ... ,dn } of the elements in {bI, b2, ... ,bn } in decreasing
order: d l 2 d 2 2 ... 2 d n . Using the rearrangement inequality, we have
n n
LCjdj ::::; Lajbj .
j=l j=l
We may assurne that not all the aj 's are equal and not all the bj
bj 's are equal so
that Cl < Cn and dn < dl . Hence we übtain
cndl - cldn = (C n - cl)dl + Cl (dl - dn ) > O.
If k 2 2, define Uk and Vk by
Ck2 = UkCl2 + Vk Cn,
2 d 2k = Uk d21 + Vk d2n·
Then Uk 2 0, Vk 2 0 and
Ckdk = ( UkCl2 + Vk C2)1/2(

n ukd2l + Vkdn2)1/2 2 UkCldl + Vkcndn,
using the Cauchy-Schwarz inequality. We übserve that Uk = 0 für süme k
implies that Ck = Ck+l = ... = Cn , d k = d k+ l = ... = d n and Vk = 1. Similar
is the case when Vk = O. If Uk > 0 and Vk > 0, then Ckdk > UkCldl + Vkcndn.
Hence
1 < (L (2: dj 2) < (Pcr + Qc;) (Par + Qd;)
(2: Cj2) (L
1< 2 < 2
(2:Cjdj)2
(Lcjdj ) - (Pc l dl +QCn dn ))2
(Pcldl+Qcnd
240 Solutions
where P == 1 + U2
U2 + ... + U
Un =
n and Q = VI + V2
V2 + ... + Vn-l
Vn-l + 1. The expression
on the right hand side is equal to
2
C dl - cldn )
1 + PQ
(
PC~dl + Qcndn
Using Pcldl + Qc"dn ;::: 2,jPQcl dl c"d", we obtain
C dl - Cl d n )
2
< ~+
rz::;i;
(( Vc;d;;
Ic":i:)
Vc;:;I; 2
1 + PQ
(
PC~dl + Qcnd" 2
< ( , :~)'
(~:~y
Since '" a 2 = '"
L..J c2 ' " b2 = '"
L..J,L..J d2 and
L..J
n n
2:>jd
2:>jdj :::; Lajbj ,
j=l j=l
we get the desired inequality.
41. Using the convexity of f,

J, we have
( Xl(Xl+X2)+X2'0)< Xl f(Xl+X2)+ X2 f(O).

f Xl + X2 - Xl + X2 Xl + X2
f(xI):::; Xl f(xI + X2) + X2 f(O).

Xl + X2 XI + X2
Interchanging Xl and X2,
X2, we also obtain
f(X2):::; X2 f(Xl + X2) + Xl f(O).

Xl + X2 Xl + X2
Adding these two, we get
f(Xd + f(X2) :::; f(xI + X2) + f(O).

Now we use the induction on n. Suppose the result holds for (n - 1); say
f(Xl) + f(X2) + ... + f(xn-d :::; f(Xl + X2 + ... + xn-d + (n - 2)f(0),

for all points Xl, X2,
X2, ...
... ,Xn-l
, Xn-l in [0, aJ such that their surn is also in [0, aJ. Now
consider any n points Xl ,X2 ,X n in [0, aJ such that L:7=1

,X3,' .. .. ,Xn
,X2 ,X3,' L::7=1 Xj
Xj is also in [0, aJ.
Solutions 241
Then
f(xI) + f(X2) + ... + f(xn-J) + f(x n)

< f(Xl + X2 + ... + Xn-l) + (n - 2)f(0) + f(x n )
::;
< f(Xl + X2 + ... + Xn-l + Xn ) + f(o) + (n - 2)f(0)
f(Xl + X2 + ... Xn-l + Xn) + (n - l)f(O).
This completes induction and hence the proof as weIl.
42. We show by induction that
C:) v3n + 1 ::; 4n .
:s; n. Then we have

For n = 1, the result is obvious. Suppose it holds for all k ::;
( 2n +
n +1
2) J3(n + 1) + 1 = 2(2nn ++11) (2n)
n
J3n + 4.v3n + 1
3n + 1
< 2(2n+1)J3n+4
--·4
n
- n +1 3n + 1 '
by induction hypothesis. However, we can show that
2n + 1 J 3n + 4 < 2.
n + 1 3n + 1 -
In fact, this statement is equivalent to
(2n + 1)2(3n + 4) ::;

:s; 4(n + 1)2(3n + 1).
12n 3 + 28n 2 + 19n + 4 ::;:s; 12n 3 + 28n 2 + 20n + 4,
which is true for all natural numbers n. Hence we obtain
( 2n
n+1
+ 2) J3(n+1)+1::;4 n +1,
which completes induction. Since ffn

ffn < v3n + 1, we get the required inequal-
ity.
43. We observe that

a X b2 + a2 b
X ::; a x +2 + bx + 2 •
242 Solutions
This follows easily, by writing the inequality in the form (a x - bX ) (a 2 - b2 ) 2 O.

Thus we have,
2 (a X + 2 + bx +2 + C +2 )X
(a X + 2 + bX + 2 ) + (b X +2 + C X +2 ) + (C X +2 + aX + 2 )
> (a X b22 + bX a 22)) + (b X c22 + cX b2) + (c X a 22 + a X c22))
a (b 22 + c2 ) + b (c 2 + a 22 )) + C
X X
C (a 2 + b2 )X
> 2 (a bc + b ca + C
X c ab) .
X X
44. Using Hölder's inequality with exponents p = 3 and q = 3/2, we get

n
~ajbjCj::;
(n)
~aJ
1/3 ( n
~(bjCj)3/2
)2/3
Using the Cauchy-Schwarz inequality, we also get
n
~ (bjCj) 3/2::; ~ bJ
(n ) (n~ cJ )
1/2 1/2
Combining these two, we get the desired inequality.
45. Consider the function f(a) = 3(a 2 - a + 1)3 - (a 6 + a 33 + 1) for a E IR..

Some computation shows that f(l) = 0, f'(1)
f'(l) = 0, f"(l) = 0, fll/(l) = 0, but
f(iv)(l) =I O. 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(2a 22 -- a + 2).
But the discriminant of 2a 2 - a + 2 is -15 and hence 2a 22 -- a + 2 2 0 for all

real a. It follows that
3(a 2 - a + 1)3 2 a6 + a3 + 1,
r r
for all real a. Using the previous problem, we obtain
((XYZ)2 + (xyz) + 1 (x 2 . y2 . Z2 + X· y. z + 1 ·1·1
< ( x 6 + x 3 + 1) (y6 + y3 + 1) (z6 + Z3 + 1).

Hence we get
(( (xyz)2+(xyz)+1 )) 33 ::;27(x 33 33 33
:::::27(x2 -x+1) (y2_ y +1) (z2- z +1).
Solutions 243
But for arbitrary reals a, b, c, we have
(abc)2 - (abc) +1~ labcl 22 + labcl + 1.

((abc)2 - (abc) + 1)3 ~ 27(a 2 - a + 1)(b 2 - b + 1)(c2 - C + 1).

46. The function f(x) = l/sinx is convex on [O,K]. Jensen's theorem shows
that
1 1 2 2
--+--> =
=---;-:::-:.,..,-:- .
sinA sinB-sin(A!B) cos(C/2)
cos(C/2)"
Thus, it suffices to prove that
1 4
----,--- >
----,---->-- .
cos(C/2) - 3 + 2cosC
Using cosC = 2cos 2 (C/2) -1, this reduces to
(2cos(C/2) - l f ~ O.
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. Since the coefficients are all nonnegative, we see that P(t) ~ 1, for any
t ~ O. Thus P(x) = 0 has only negative roots; let these be -0:1, -0:2, ... ,
-O:n, where o:j's are all positive. We have
P(x) = (x + 0: I) (x + 0:2)'" (x + O:n).

0:10:2 ... O:n = 1.
2 + O:j = 1 + 1 + O:j ~ 3(O:j)1 / 3, for 1 ~ j ~ n.

Hence
P(2) = (2 + o:d(2 + 0:2)'" (2 + O:n)

1/3
> 3nn ((0:10:2"'o:n)1) /3
0:10:2'" O:n
3n .
In fact, this may be generalised as folIows: 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 + o:d(t + 0:2)'" (t + O:n).
244 Solutions
Consider a general term (t + ak). We have

t· 1 + 1 . ak
t + ak - - - ( t(t+1)
t+1 + 1)
> (t + l)(ak)I/(t+l),
1)(ak)1/(t+1),
by the weighted AM-GM inequality. Thus

P(t) 2: (t + 1)n(ala2··· )1/(t+1) == (t + 1)n.
1)n(a1a2··· a n )I/(t+1)
a1 > 0
48. We prove this by induction on n. The inequality is clear for n == 1: al
a1 2: 1 and hence
implies that al ai ::; ar.
Suppose it holds far all distinct positive
integers 0 < al
a1 < a2 < ... < an and an+! > an. Then we see that
an+l 2: 1 + an, an+l 2: an-l + 2, ... ,an+l 2: al + n.

Adding all these inequalities, we get
nan+l 2: (al + a2 + ... + an) + n(n + 1).

Thus
(al + a2 + ... + a n ) <
_ n a n +l - - - -+"1)
n(n -
Hence
(~aj+an+lr
(~aj+an+1r
n+l ) 2
(
Laj
(t r (t
j=1
aj + 2a n+l
+1 aj ) + a~+l
a~+1
~ aj3 + 2an+l ( nan+l
< L....- -
n(n+1))
2 + an+1
2
)=1
n
L aJ + (2n + l)a;+1
1)a;+1 - n(n + l)an+l.
j=1
o((2n
o
1) a n
2n + 1)a
22
+1 - n(n + l)a
33
n+l ::; an+!
1)an+1
{:::=} (2n + 1) an+1 - n (n + 1) ::; a;+ 11
{:::=} a;'+1 - (2n l)an+I + n(n + 1) 2: 0
+ l)an+l
{:::=} (a n+I - n - 1) 2: o.
(an+ 1I - n) (an+1
Since aj are integers and 0 < al a1 < a2 < ... < an < a n +l,
+1, it follows that
aj 2: j, far 1 ::; j ::; n + 1. Hence an+1 n + 1, and thus
an+I 2: TI
(an+I - n) (an+1
(an+1 (an+l - n - 1) 2: o.
Solutions 245
49. We have
)2 = 2: X] + 2 2: X Xk·
2: = t x ; + 2~XjXk.
n n
11 = (
(
t Xx j
j r j
j=l j=l j<k
Thus we have to prove that
1- 2:
n
L Xj ::; -
2
-+
L -ajxj
n - 2 2:
-.
n 2
n-l l-a-
l-a'
j=l j = l )J
j=l
1 n x2
_<"_J_.
- <"-)-.
n-1-L....-1-a
j=l
j=l J)
This follows from the Cauchy-Schwarz inequality:
1 = (tXj)2
)=1
< (t :7a ) (t(l- a
)=1
1
) )=1
j ))
(n-1)(~
(n-1)(~~).
1:7aJ.
L....-1-a·
j=l J
50. Let us put xy + x~ + x~ + xl == A and aj == A - x;. Then it is easy to check

that
A= ~ (al + a2 + a3 + a4) .
Moreover,
1 x3 + x3 + 1
+ x 33 > X2 3 4 == Xl
-a -- x~2 + x~
1 1 =
-al 33 x 44 2: X X _.
3
3 - Xl
Similarly, we get
1 1 1 1 1 1
-a2> - , -a3 >-, -a4 >-.
3 - X2 3 - X3 3 - X4
1(
A = - a1 + a2 + a3 + a4 ) 2: -1 + -1 + -1 + -.
1
3 Xl X2 X3 X4
Using Chebyshev's inequality, we also have
xy3
Xl + Xx~32 + Xx~33 + X4 xi2+, 2
xl3 > Xl ;r~ + X
:1:2 x~32 + X4
X~2 Xl + X2 +
+ X3 +.
+ X1
4 4 4
But
xi + xx~22 +4 xx~32 + x4
2
x~ 2: 1,
2: 1,
246 Solutions
by the AM-GM inequality. Thus we also obtain

A 2: Xl + X2 + X3 + X4·
It now follows that
A 2: max { Xl + X2 + X3 + X4, -l+I

- l
+ -I+}
- .
Xl X2 X3 X4
51. We use induction on n. For n == 1, the result is dear. Suppose it holds for
some n. We prove its validity for n + 1 also. We observe that n< Jn 2 + j <
n + 1 for 1 :S j :S 2n. Thus
{Jn 2 + j} = Jn 2 + j - n < Jn 2 + j + (j/2n)2 - n = 1n'

Hence
(n+1)2 n2 (n+I)2
L {JJ} L {{JJ} + L {JJ}

j=l j=l j=n2 +1
n2 _ 1 1 2n
< - -2+ - L
2n
j+O
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 weH.
52. After expanding, the inequality takes the form

4(a 2b + b2c + c2a) + 4(ab 2 + bc2 + ca 2) 2: 3(a + b) (b + c) (c + a).
This reduces to
(a 2b + b2c + c2a) + (ab 2 + bc22 + ca2) 2: 6abc,
which is a consequence of the AM-GM inequality.
53. Introducing X, y, z by
b + c - a == 2x, = 2y,
c+a - b= a + b - c == 2z,
= y + z, b == z
we get a = x, c == X + Y and the inequality takes the form
+ X,
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::::: x 2y + y2 z + Z2 X + xy2 + yz2 + zx 2,
which directly follows from the AM-GM inequality.
Solutions 247
54. We have
(xy + yz + ZX)2 - 3xyz(x + Y + z)

(xy)2 + (yz)2 + (zX)2 - (xy)(yz) - (yz)(zx) - (zx)(xy)
= ~{(Xy - YZ)2 + (yz - ZX)2 + (zx - Xy )2}.
Hence it follows that
:s: (xy + yz + zx)2 :s: (xyz)2.

3xyz(x + Y + z)
Since xyz > 0, we obtain 3(x + y + z) :s: xyz.
55. Let bk denote the largest among bj's. Then we have
n-1 k-l n-1

L bjbj+l L bjbj+l +L bj bj+1
j=1 j=1 j=k
k-1 n-1
< bk L bj + bk L bj+l
j=1 j=k
...... b k )
b k (b .....
bk(b
2
b _ (~ _ bk ) 2 < 2
b .
4 2 - 4
56. Let us take r =

= bla, s =
= die and k =
= (v'5 - 1)/2. Then k 2 =
= 1- k and
o < k < 1. We have to show that
max {I, Ir + si, Irsl} ;::: kmax {I, Irl} max {I, Isl}.
We consider several cases.
Case 1. Irl ;::: 1 and Isl ;::: 1.
In this case
max { 1, Ir + si, Irsl} ;::: Irsl > klrllsl = k max { 1, Irl} max { 1, Isl}.
Case 2. Irl < 1 and Isl < 1.

We obtain
max{ 1, Ir + si, Irsl} ~1>k= kmax{ 1, Irl} max{ 1, Isl}.
Case 3. Irl < 1 and Isl ;::: 1.

248 Solutions
We have to show that
max { 1, Ir + si, Irsl} 2: klsl·

If either Ir + si 2: klsl or 1 2: klsl, the result follows. Suppose Ir' + si < klsl
and 1 < klsl. Then
Irl + Ir + si = I - rl + Ir + si 2: 14
ami hence
Irl 2: Isl - Ir + si > Isl - klsl = k2 1si-
Thus, we obtain
Irllsl > k2 1s1 2 > klsl
since k Isi> 1. This shows that
max { 1, Ir + si, Irsl} 2: klsl·

By symmetry, the result follows in the case Irl 2: 1 and Isl < 1.
57. We show that
3
x(l - z) + y(l - x) + z(l - y) 2: 4'
= 0, we see that either y == 1 or z == 1 and the inequality follows. Similar
If x =
is the case when y = 0 or z = O. We may hence assurne that x > 0, Y > 0 and
zz > O. Now put
II-x
-x 1-y
1- 1I-z
- =
--x = =aa,, - y b
-y = = b,
.. - - cc.
z = .
x y zz
Then abc == 1 and we have to prove
'"" cc 33
L
~
cycJic
cychc
(1 + a)(l + c) 2: 4'
4(~c a+ ,~, ab) ,,3 (1 + ,~, a+ ,~, ab+abC).

This reduces to
L a+ Lab 2: 6.
cyclic cyclic
However, we have
a + b+ c 3(
> 3 ( abc ) f/3
1/3
3,
= 3,
2/3
ab + bc + ca > 3 ( abc) = 3.
Solutions 249
These two together give

L a+ L ab26.
ab?6.
cyclic cycJic
cyclic
= b == c, which is equivalent to x == Y == z.
Equality holds if and only if a =
58. Introduce Yj = 1/(1 + Xj), fürfor 1 :S ::; n. Then we get Xj = (1 - Yj)/Yj,

::; j :S
1 :S
::; j :S
::; n. 2:7=1 Yj <::; 1 and the conclusion we
n . The condition translates to 2:,7=1
fIrr (l- Y
have to derive is
j ) 2 (n _1)n.
j=l Y)
Yl
If we use the AM-HM

AM -HM inequality, we get
""'->
1
L-
n 2
n 1 _n_ > n.
2 -;.- 2n. n -
L
j=l Yj
j=l Yj - ""'y.
LYj
~ J
j=l
Using the AM-GM inequality, we get
rrrr(l-
n
Y >> ( ~t 1_Uj)n
(l-Y')
r
_ 1j )
j=l y)
Yj j=l YJ
(t,H:i -1)
(
1
;;:L~-l
n1 )n
j=l YJ
> (:: -1) n
(n - l)n.
Alternate solution:
We introduce the polynomial
P(z) = (z + xI) (z + X2)'" (z + x n ).

Then
n _1__ P'(l)
LI
)=1 +x
J=l J
- P(l)'
P(l)·
P'(1) :S
The given condition is P'(l) ::; P(l). We now write
P() n
Z = z +cr1Z
n-1
+cr2Z
n-2
+"'+crn ,
250 Solutions
O"'s are elementary symmetrie functions in X1,X2,X3,

where I7'S X1,X2,X3, ...
... ,x n
n .. Then the
derivative is given by
P'(z) = nzn-1 + (n - 1)171Zn-2 + (n - 2)172Zn-3 + ... + I7 n-1.
P'(l) :::; P(1)

The condition P'(1) P(l) is now equivalent to
I7O"nn 2: (n - 1) + (n - 2)171
2)0"1 + (n - 3}172
3)172 + ... + I7O"n-2·
n-2·
Now we use the inequalities
I7j
O"j (n) (
j {l7
2: (;) O"n
)j/n
n )j/n,, 1:::; j :::; n.
Taking r = (I7 n )l/n, we obtain
rr nn __ (( n ) rn-2
rn-2 _ rn
_ 2( n ) r n -- 3
3 _
_ ...
... _ r -- (n - 1) 2:
_ (n _ 2) (n) r o.
n-2 n-3 1
Consider the polynomial
(n) n)
~
Q()
X =x n - n ) x n-2 _- n-2 2 (( n ) x n-3 - ...
n-3 _ ...
Q(x) xn _ (n - 2
n-2
X
n-3
n-3 _ (n _2) (.)n x-(n-I).
-(n-2)(~)x-(n-1).
We observe that Q(x) has only one change of sign for x> o. 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) = O. Thus, it follows that Q(x) 2: 0 in (0,00) if and
only if x 2: (n - 1). Since Q(r) 2: 0, we conclude that r 2: (n - 1). This gives
I7O"nn 2: (n - 1)n.
l)n.
59. Here is a beautiful solution to the problem. We write
a + Vab + ~=a+
a+Vab+ ~ = a + ~~+
~~ + ~~a.4b.16C.
~ .va . 4b . 16c.
Using the AM-GM inequality, we have
a + ~~ + ~ ~
{Iaa . 4b . 16c
2 4
1 1
:::; a + 4 (a + 4b) + 12 (a + 4b + 16c)
4
3(a+b+c).
= 3{a+b+c).
Solutions 251
60. Suppose we introduce

1 1
f(m, n) == m+n+1
f(m,n) m+n+ 1 - ( m+1 )( n+1 )"
(m+1)(n+r'
Then it is easy to see that f(l, 1) = f(l, 2) = 1/12< 4/45. Now it is not hard
to prove that
1 4
~--~--~
~--~--~> > .
(m + l)(n + 1) - (m + n + 2)2
o. Thus if k = m
In fact, this is just equivalent to (m - n)2 :::: O. + n + 2, then
1 1
f(m,n)
m+n+1 (m + l)(n + 1)
1 4
<
m+n+1 (m + n + 2)2
1 4
----
k - 1 - k 2 '·
Consider the function g(k) defined for natural numbers k :::: 2 by
1 4
g(k) = k _ 1 - k2'
k2·
We show that it is a non-increasing function for k :::: 6. In fact for k :::: I,

1 4 1 4
g(k) - g(l) k - 1 - k2 - I- 1 + [2
I-k _4(I-k)(I+k)
(I - l)(k - 1) Z2k 2
(lk - 21 - 2k)2 - 41 - 4k
(1- k)
(I-k) " . , .. . '"~ . ~
(1-1)(k-1)t2k2 ..
Since 1 - k ::; 0, we have to show that Ik - 2(1 + k) :::: 2yT+k for 1 :::: 6 and
k :::: 6. For 1 :::: 6 and k :::: 6, we have (I - 4) :::: 2 and (k - 4) :::: 2 and hence
(I - 4)(k - 4) :::: (I - 4) + (k - 4). This gives Ik :::: 5(1 + k) - 24. Thus
Ik - 2(1 + k) :::: 3(1 + k) - 24.
It is sufficient to prove that 3(1 + k) - 24:::: 2yT+k, whenever 1 :::: 6 and k :::: 6.
Consider the quadratic function
3e -
q(t) = 3e - 2t - 24.
Then q(t) :::: 0 if and only if t :::: (1 + ,;73)/3. Since
vT+k :::: v'12:::: 11+~,

VT+k::::v'l2:::: +3,;73 ,
252 Solutions
for 1 2: 6 and k 2: 6, it follows that
3(l + k) - 2vT+k - 24 2: O.
This shows that f(m, n) ::::: g(6) = 4/45 for all m, n such that m + n + 22: 6.
61. If either a> 1 or b > 1, then the result holds. Thus it is sufficient to
consider the case 0 < a < 1 and 0 0, f'(x) == a X lna + ax a - 1 .
Suppose for some x E (0,1), thc inequality a X + x a ::::: 1 holds. Then f(x) ::::: 0,
f(O) = 0 and f(1) > O. Using Rolle's theorem, we can find b E (0,1) such that
f(b) ::::: 0 and f'(b) == O. This implies that
ab In a + ab a - 1 = 0, ab + ba - 1 ::::: O.
b
~ In a - a- b
1 - -ln
b
< O.
a -
We show that this is not true for any a, bE (0,1). Consider thc function
In a -x
9 (x ) =
= 1- - x -a ,
a
where a E (0,1) is fixed. It is easy to compute
, (x )
9g'(x) In a +a-Xlna=lna(a-
_l:a
--;:-+a- x I
na=lna ( a- xX - a1 ) ' _!),
g"(X) -a x (lna)2.
Thus g"(X) < 0 for all x E (0,1). This implies that g'(x) is a decreasing
function in (0,1). Hence g'(x) > g'(1) == 0 for all x E (0,1). But then g(x)
is an increasing function on (0,1). Hence g(b) > g(O) = o. Thus, there is no
bE (0,1) such that g(b) ::::: O.
1 1
62. Suppose p > 1. If q is the conjugate index of p, i.e., +- 1, then
P q
Hölder's inequality gives
( 1 1) {(
a+;-+b+Z;::::: a+;-
1)P + (b+Z;1)P}1/P 2 1/ q .
However, a +b= 1 and

1 1 aa+b
+-b _= ~1
- >
1
("tb)'
-1 + -1 = - 1 = 4.
=
~CL + bb ~ ---;;b
ab - ab -~ (at
> ----------,0-2
b)
4.
Solutions 253
Thus, we übtain
5::::; { (( a+
p )P } l/p 2(p-l)/p,
~ )) p ++ ( b+ ~ )P}l/P
(
since q = (p - l)/p. This proves that
( aa+-l)P + ( b+-l)P >--5P

+ ;:
a + b+ b b 2:
- 2p-l'
für p > 1. If p == 1, then the inequality tü be prüved is

1 1
a+-+b+->5
a b - ,
1 1
~ + b 2:
which is immediate using a + b = 1 and ;: 2': 4.
Suppüse 0 < P < 1. Cünsider the functiün
+
f(x) = (x + (1 -- x ++ 11 ~ x) P,
~) P ++ (1
für xE (0,1). Its derivative is
f'(x)=p ( 1-~1 ) ( x+-1 ) p-l +p ( -1+ 1.) ( 1 - x + -

1 - ) p-l
. x2 x (1 - X)2
X)2 1- x
For 0 < x < 1/2, we übserve that

1 1
x+->l-x+--.
x 1- x
In fact, this is equivalent tü (1 - 2x) ( - 1 + l/x(l - x)) > 0, which may be
seen to be true since 1- 2x > 0 and x(l - x) ::::; 1/4. Since 0< p < 1, it follows
that
(( x+;:l)P-l
l)P-l (( 1 )P-J
< 1-x+ 1 _ x
)P-1
This gives
f'(x) < (
P(1_X+_1_)P-l{1_~_1+
P 1-x+--
1- X
p-l
1---1+
2
2
1) {lI}
I.} x
.
(1 - x)2
( 1 -xx++
pP(l-
1
_l_)P-l
--
2x )P-l
2x --11
1- x x 2 (1 - :r)2
< o.
Similarly, we can prove that f' (x) > 0 für 1/2 < x < 1. It füllüws that f (x)
2': f(1/2) für all
is decreasing in (0,1/2) and increasing in (1/2,1). Thus f(x) 2:
xE (0,1). But
f(1/2) == (5/2)P + (5/2)P ==
5P
1. 2:: .
2P-J
254 Solutions
We conclude that
5P
f(x) ~ 2p-l'
for all x E (0,1).
63. A solution using Schur's inequality was given in chapter 2 (Example 2.29).
Here we give two more solutions.
Alternate Solution 1.
Since a, b, c are positive real numbers such that abc = 1, we can find positive
real numbers x, y, z such that a = x/y, b = y/ z and c = z/x. The inequality
reduces to
(z +x - y)(x + y - z)(y + z - x) ::::; xyz.
This is symmetrie in x, y, z. Hence it is sufficient to consider the case x::::; y ::::; z.
It follows that y+z-x > 0 and z+x-y > O. If x+y-z ::::; 0, then the left-side is
non-positive and the right-side is positive; hence the inequality holds. Suppose
x + y - z > O. Then x, y, z are the sides of a triangle. Let its area, eire um-
radius, in-radius, semi-perimeter be respectively ß, R, r, s. The inequality
may be written in the equivalent form:
(x + Y + z)(z + x - y)(x + y - z)(y + z - x) ::::; xyz(x + Y + z).

But the left-side is 16ß2 and the right-side is 8Rßs. Thus the equivalent
inequality is
16ß 2 ::::; 8Rßs.
Using ß == rs, this reduces to 2r <

::::; R, which is a standard result(Euler's
inequality) .
Alternate Solution 2.
If a - 1 + (l/b)
(1/b) ::::; 0, then we see that a < 1 and b> 1. Hence we obtain
1 1
b -1 + - > 0, c - 1+ - > O.
c a
Thus, the product on the left hand side of the inequality be comes negative
and forces the inequality. So is the case when either b - 1 + (1/c) ::::; 0 or
c-1 + (1/a) ::::; O. We may thus assume that a -1 + (1/b) > 0, b-1 + (1/c) > 0
and c - 1 + (1/a) > O. We observe that
( a - 1 + ~) (b - 1 + ~) (c - 1 + ~) = (ab - b+ 1) (bc - c+ 1) + (ca - a + 1).

Suppose
1 1 1
-a+b
- +c- <
-
a + b + c.
Solutions 255
This implies that ab + bc + ca :S a + b + c. Using the AM-GM inequality, we

obtain
- bb +
(ab _
(ab + 1) - cc +
(bc _
1) (bc + 1)
1) +
+ (ca - aa +
(ca _ + 1)
1) :S
< (( ab + bc
ab + + ca
bc + -3 3)
+ 3) 33
ca ~ aa -- bb -- cc +
< 1.
If
1 1 1
- + - + - > a + b + c,
abc
we introduce the variables 0: == 1/a, ß == 1/b and 'Y == 1/c. We see that 0:, ß
and 'Y are positive reals with o:ß'Y = 1. Moreover, these satisfy
111
- + - + - :S 0: + 'Y + ß·
0: 'Y ß
Applying the above reasoning to 0:, 'Y and ß, we obtain
( 0: - 1+ ~) ('Y - 1+ ~) (ß - 1 + ~) :S 1.
This reduces to
(a-1+~) (b-1+~) (C-1+~):S1.

64. The inequality we need to prove is
L~~~3.
J4-X2~
~ (2+x)(2+y) 2:3. (64.1)
cychc
cyclic
Using the condition x + y + z == 0, we obtain

3(4-x 2) = 12-3x 2 12-2x 2 -(Y+Z)2
(y - z)2 + 2(6 - ~ cychc

x 2)
(y - Z)2 + 4(3 + ~ cychc

x y) .
We observe that
3+ L xy= L(1+x)(1+y)~0.
L(1+x)(1+Y)2: 0.
cyclic cyclic
256 Solutions
Thus, we get
J4 - x 2 J4 - y2
~--------------
~-----------------
~ ( (3+ C~C
Z)' +
(y - z)' +44 x y) ) ( (x - z)2 + 4(3 + L cychc
Y) )
X Y)
> [(y - (3+ ,~C xy) 1

~ [(Y-Z)(X-Z)+4(3+c~cXy)l
Z)( x - Z) +4
1
3(12 + 5xy + 3xz + 3yz + Z2)
1
3(12 + 5xy + 2xz + 2yz)
(4+xy) + ~ (~C
(~, XYy )X ,
where we have used the Cauchy-Schwarz inequality and the given relation x +
y + zz = O. Using this estimate we obtain
L 1 1
(4 - x 2 ) 12" (4 -_ y2) 12"
~
cyclic
cychc
(2+x)(2+y)'
1 1
I:cycliC (4 - x 2 ) 2" (4 - y2) 2" (2 + z)
TI(2+x)
>
> TI (21+
TI (2 + x)
x) {{ c~c
2: (4 ++ xy)
cychc
xy) ((22 +
( + z) + ~ (( c~c
z) + 2: XY) ( ~ (22 +
XY) (c~c
cychc
+ z))
z)) }}
cychc
(
1
TI(2 + x) { {c~c
~ (8 + 4z + 2xy + xyz) + ~ (c~c
cychc
( 2: X
XY)6}
Y )6}
cychc
1
TI (2 + x) {{24+2(
24 + 2 ( c~c
~ XY) +3xyz+4(
cychc
+ 3xy z + 4 ( c~c
~ x Y))}}
cychc
TI (21+ x) {24 + 6( 6(c~c

~ xy) + 3X
cychc
3XYY Z},
where TI(2 + x) = (2 + x)(2 + y)(2 + z). However
24 + 6 (~c XY)
x y ) + 3xyz ~ 3(2 + x)(2 + y)(2 + z),
z).
We get the desired inequality (64.1).

Solutions 257
65. We use the rearrangement theorem (Theorem 4 on page 21). Fix n and
consider the first n elements al,a2, ... ,an' Let 0'(1),0'(2), ... ,O'(n) be a per-
mutation of 1,2, ... ,n such that
a.,-(l) < a"-(2) < ... < a.,-(n)·

Observe that
111
->->
12 - 22 - ... >-
- n2 '
Using theorem 4, it follows that

n n
""' ak > ""' ?,.,-(k)
~ k2 - ~ 2 k •
k=l k=l
However, a.,-(k) is a positive integer for each k. Since (a"-(k))

(a"-Ck)) is an increasing
sequence, it may be concluded that a.,-(k) ::::: k for 1 ::::: k ::::: n. This gives
n ak n k n 1
L
k=l
k2 ::::: L
k=l
k2 = L
k=l
k'
66. Let us introduce f(x, y, z) = xy + yz + zx - 2xyz. Assume

Assurne x ::::: y ::::: z,
which is permissible by the symmetry of the expression. Using x + y + z = 1,
it may be concluded that x ::::: 1/2. Write
f(x, y, z) = x(y + z) + yz(l - 2x).

Since 1 - 2x ::::: 0, it follows that f(x, y, z) ::::: O. On the other hand
(1 - X)2 = (y + Z)2 ::::: 4yz,

by the AM-GM inequality. Thus
7 7
f(x, y, z) - 27 x(y + z) + yz(l - 2x) - 27
x (1 _ x) + (1
< x(l (1 -- x)2(1
x) 2 (1 -- 2x) 7
2x) _ ~
< - x) + 4
4
- 27
27
_ (3x - 1)2(6x + 1) < O. 0
108
108 -- .
This proves the other part of the inequality.

Alternate Solution:
ass urne x:::::
As in the previous solution, we may assume x ::::: y ::::: z. And the same method
shows that f(x, y, z) ::::: O. Introducing new variables a, b, c by
1 1 1
= a+ 3'
xx=a+- 3'
Y =b+-
y=b+-
3'
zz=c+-
= c +-,
3' 3
258 Solutions
the expression f(x, y, z) takes the form
7
f(x, y, ,,) = 27 + 3"1 ( bc - a 2 - 6abc ) .
The eonditions on a, b, c are
1 2
a :S b :S c, a + b + c == 0, --3 < , , - -.
- abc< 3
Thus it follows that either a :S b :S 0 :S c or -(1/3) :S a :S 0 :S b :S c. In the

first ease bc :S 0, and abc 2: o. Henee bc - a 2 - 6abc :S O. In the seeond ease,
we rewrite the expression:
bc - a 2 - 6abc = -(b - C)2 - 3bc(1 + 2a).

Again bc 2: 0 and 1 + 2a 2: 0 sinee a 2: -(1/3) 2: -(1/2). Thus it follows that
f(x,y,z):S 7/27.
(For yet another proof, refer to example 2.38.)
67. Let us introduee
n x3
A = ~
'"'
x2 + XjXj+1 + X]+l
j=l J
J .
Consider
3
n Xj +1
L
.
B =
j=l
X2
J
+ XjXj+1 + X]+!
Observe that
n X~ - X1+1
A-B '"'
~
j=l
X2
J
+ XjXj+1 + X 2j +!
J
n
L (Xj - Xj+I)
(Xj - xj+I)
j=l
O.
This shows that A = Band henee A = (A+B)/2. However, A+B is symmetrie

and it is advantageous to deal with it. Observe that
3 3
Xj + Xj+1
-L--~~~2~ > Xj + Xj+1 .
X] + XjXj+1 + Xj+1 - 3
Solutions 259
Thus
1 1 n x 3j + x H1
3
A=2(A+B) =
L
-2 j=l x J2 + x'x'+l
J J
+ X2+J 1
1 1
L
n
> - X - (Xj + xHd
2 3 j=l
1 n
- LXj.
3 j=l
This proves the inequality.
68. Expanding and rearranging, the inequality may be written in the form
x 3 + y3
X + Z3
z3 + 3xyz ~
2: x 2y + y2 z + Z2 X
X + xy2 + yz2 + zx 2.
This reduces to
x(x - y)(x - z)
X(X + y(y - x)(y - z) + z(z - 2: 0,
x)(z - y) ~
which follows from Schur's inequality. Equality holds if and only if x =

= y=
= z;
x == 0, y == z; y == 0, z == x; z == 0, x == y.
69. Put
a bb c
b == x, ~ = y,
= - = z.
a
Then xyz =
= 1 and it is easy to calculate
c+a l+x_ X+y
x+y
c+b l+x- l+y'
1 + y'
a+b l+y_ y+z
Y +z
a+c l+y- l+z'
1 + z'
b+c z+X
z+x
= l+z--- .
l+z-l+x'
b+a l+x
Thus, the inequality reduces to
X+y
x+y y+z z+x y+z
3 z+X 3
--+--+-->
l+y l+z l+x-
.
X2 + y2 + Z2 + x 2z + y2 x + Z2 y - (x + Y + z) - 3 ~ 0.
Using xyz =
= 1 and the AM-GM inequality, we get
x 2z + y2 x + z2 2: 3xyz = 3.
Z2 y ~
260 Solutions
Using the Cauchy-Schwarz inequality, we also have

1/2
x + Y + z < J3 (x 2 + y2 + z2 )
(3(x z2)1/3) 1/2 (x
(3(X z2)1/3)
22y2
y2
1/2
2 z2)
+ y2 + z2)
1/2
1/2 ( 1/2
< ( x2 + y2 + Z2)
Z2 ) x + y2 + Z2 ))1/2
1/2( 2
x 2 + y'2 + Z2.
70. This may be written in the form
(a +b- 1)2 + (a - 2)2 + (b - 2)2 ~ 3.
By the Cauchy-Schwarz inequality, we have
I(a + b - 1) - (a - 2) - (b - 2)1 2 :::; 3((a + b - 1)2 + (a - 2)2 + (b - 2)2).
Hence the required inequality follows.
71. First observe that all the terms of the sequence are positive. Moreover,
x;' + 9
X4 X~
x~ 3 3 3
Xn +l =
-
n_
10x n -+--+--+--
10 10x 10x 10x
3 lOx3 10x3)
n n n
> 4 ( 10' lOx

> x~
(x~ 1/4
1/
4
lOX n . lOx n . 10x n
~(m) >~.
Thus X n > 4/5 for all n. To prove the othcr inequality, observe that X2 = 5/4.
If we show that XXn +1 ::; XXn for all n ::::: 2, it follows that XXn :::;
+l :::; ::; 5/4 for all n ::::: 2.
Thus we have to check that
4+9
~<Xn,
lOx n
for n ~ 2. Equivalently, we need to prove that x~ x;' - 10x; + 9 :::;
::; 0, for all n ~ 2.
If 1 :::;
::; XXn :::; ::; x; < 9 and hence x~
::; 5/4, we have 1 :::; x;' - 10x; + 9:::;
9::; O. If XXn < 1,
we see that
x;'4n + 9
X 10 1 5
+l =
xXn + l= 10x O-
X nl <
<-10
10x
X n =
=-<4'
X
X n < 4'
since Xn > 4/5, by the first part.
72. We have to show that ab - ac - bc + cc'22 :::;

::; O. This reduces to
a2 +b 2 :::;c(a+b).
Solutions 261
Squaring both sides, this may be written in the form
(a 2 + b2 )2 :s (a + b) (a 3 + b3 ).
This reduces to the standard inequality 2ab :s
::::: a2 + b2, which is true.
73. Consider the coordinate plane with 0 = (0,0) as the origin. Let A = (a, 0),
B = (b/2, bV3/2) and e = (-c/2,
(-e/2, cV3/2).
eV3/2). Then we see that
AB = J a22 + b2 - ab, Jb 2 + e2 - be,

Be = .,fb'2+c2-~b~, e A = J a2 + b22 + ac.
ae.
Moreover A, B, e are the vertices of a tri angle (may be degenerate) in the

plane. Hence triangle inequality gives
v'd2+b2-+-a-;; :s ~2+b2 ~ ab + Jb 2 + c 2 - bc.
74. Consider the quadratic function
f(x) == a 2(a - 2)x 2 - (a 3 - a + 2)x + (a 22 + 1).
We observe that f(O) = a2 + 1 > 0 and f(l) = _(a 2 - a + 1) < 0 for all a.
Hence f(x) = 0 has areal rüot.
root. But then the discriminant of f(x) must be
non-negative. Thus we get
(a 3 - a + 2)2 ::;:: 4a 2(a 2 + l)(a - 2).
75. Put
a b ce
x= --b' Y = -b-' - --.
Z -
a- y= b --cc' c-a
Then it is easy to see that
(x - l)(y - l)(z - 1) = xyz.
This reduces to
x +Y + z = xy + yz + zx + 1.
r (2:~ r (~ c~:r
Hence
(2: ~:
( 2a-b)2
a-b
+ (2b-c)2 + (~c_a)2
+ b-cce + c-a
(1 + X)2 + (1 + y)2 + (1 + Z)2
3 + x 2 + y2 + z2 + 2x + 2y + 2z
3 + x 2 + y2 + Z2 + 2(xy + yz + zx + 1)
5+(X+y+Z)2::;::5.
5+(X+y+Z)2~5.
262 Solutions
76. Using the Cauchy-Schwarz inequality, we get
( Ln x j+1 ) 2
:::::
(nL x2m+1
O:X.)~ ßX
O:X. ßx 1
) (
LXj(O:Xj
n + ßXH1)
).
j=l
J=l J +} +
j=l= l) } )+1 j=l
}=1
However,
n n n
LXj(O:Xj + ßXH1) L O:X; +L ßXjXH1
j=l j=l j=l
n
< (o:+ß)LX;,
j=l
Moreover, Hölder's inequality gives
( f; n
X;
)m+1
::::: n m - 1f;
( n
Xj+1
)2
It follows that
n X.2m+1
(t,X;) mH
>
L O:Xj
j=l
~ ßWJ n m - 1(o: + ß) ( L7=1 X])
(t
(t,x;)
x;)
)=1
m
= nm - 1
Again the Cauchy-Schwarz inequality shows that
n(tx;) 2: (tXj)2
J=l J=l
= 1.
Combining, we get
2m +l
x 2mH 1
L
n
> 2m
- -1
L.
J=l
) = 1 )
~
O:X. +
O:Xj
J
ßX'+l -2 n 2m-I'
ßXj-"-l
J
- '
77. We observe that

C)2 + (b+d)2::::: Ja 2 + b2 + Jc 2 +d2
J(a+ C)2
{::=} C)2 + (b + d)2 ::::: a22 + b2 + c2 + d2 + 2/(a
(a + C)2 2J(a2 + b2 ) (c 2 + d2)
{::=} )(C2 + d2)

2ac + 2bd::::: 2J (a 2 + b2 )(C
{::=} (ac + bd)2 ::::: (a 2 + b2) (c 2 + d22 ))
{::=} 2abcd::::: a 22d2 + b22c22,,
Solutions 263
which follows from the AM-GM inequality. The second inequality is equivalent
to
( Ja 2 + b2 + Je 2 + d2) ((J(a + e)2 + (b + d)2)

d2)
::::; (a + e)2 + (b + d)2 + 21ad - bel.
r: ;
In view of the first part, it suffices to prove that
( Ja 2 + b2 + Je 2 + d2 (a + e)2 + (b + d)2 + 21ad - bel.
Squaring, and simplifying, this reduces to
0::::; 2(ae + bd)lad - bel,

which is true by the positivity of a, b, e, d. Equality holds here if and only if
ad = be.
78. We use the standard transformations: s - a = x, s - b = y and s - e = z.

The inequality is equivalent to
,;x + y'y + vIz <- .fij+z + y'Z+X

J2
+ ..;x+Y
This follows from the root-mean-square inequality:
L
cyclic
rn:: =
v2x L
cyclic
ffx + V2Y <
2 -
L
cyclic
J 2x + 2y =
2
L
cyclic
VX+Y
x +y.
Note that equality holds if and only if x = y = z, which is equivalent to

a = b = e or that ABC is an equilateral triangle.
79.
A
B D c
Let us intro du ce x = [BHC], y = [CHA] and z = [AHB]. Then we see that
HD _ [BDH] [CDH] [BDH] + [CDH] [BHC] _ x
HA - [BHA] [CHA] [BHA] + [CHA] [BHA] + [CHA]
[BHA]+[CHA]- y+z
y+z
264 Solutions

HE y HF zz
HB z+x' HC x+y
Thus, we obtain
HD HE HF _x_ + _y_
x + _z_ y z
HA + HB + HC - -+
y+z
--+ --
z+x x+y
3
>
2'
see Nesbitt's inequality.
80. Let ~ denote the area of ABC. Then it is easy to see that
2~ = arl + br2 + cr3.

Let>. = air] + blr2 + clr3' Then we see that
2>'~ = (ar 1l + br2 + cr3)

cr 3 ) (~
rl
1'1
+ ~ + ~)
r2 1'3
1'2 r3
a2 +b2 +c22 + L .
cyclic
ab(r 1l
r2
1'2
+ rz )
rl
1'1
cychc
> a 22 +bZz +cZz +2ab+2bc+2ca=(a+b+c)2.

+2ab+2bc+2ca=(a+b+c)z.
This shows that

>.> (a+b+c)2
- 2~ ,
rl == rz == 1'3.
for any point P. Equality holds if and only if 1'1 r3. This is precisely
preeisely
when P is the in-eentre of ABC. Thus, the given expression is minimal only if
P is the in-centre
in-eentre of ABC.
Inseribe the hexagon in a reet angle MN PQ as shown in the figure. Note

81. Inscribe
that LA == LD, LB == LE and LC == LF.
o
L-
~
E
1
D
\
P
•
r
I
c
F
1
1
I
~"
~ ,
M A
A B N
Solutions 265
In triangle BCD,
BGD, we have
4RB = 2.BD > MQ.+NP.

smC
smG - smC
smG
But MQ == MF + FQ and NP = NC
NG + CP.
GP. Again sine rule gives
MF FAsinL.MAF = FAsinA,
FQ EFsinE = EFsinB,
NC
NG BCsinB,
BGsinB,
CP
GP GDsinD = CDsinA.
CDsinD GDsinA.
Thus we get,
4RB > FAs~nA +EFs~nB +BCs~nB

+B s~nB +CDs~nA.
+ Ds~nA.
- smC
smG smC
smG smC
smG smC
smG
Similarly, we can prove that
sinC
sinG ABsinAsinA EF sinC
sinG DE sinA
4Re
4Rc G --+
> BB C
> . BB + AB --+
. BB + EF --+
. BB + DE -.- B''
sin
sm sin
sm sin
sm sin
sm B
sinB + CDsinC sinB + FA sinC
sinG + AB --+
4RA > DDEE --+ GD --+
sin A sin A
--.
sinG.
sin A sin A
Adding these, we get
4(RA+RB+Re)
sin-A +sin B) +B C
G (sin
- -C G+sin B)
> AB ( -
sin B
--
sin A sin B
--
G
sin C
+GD (s~nC
+CD (s~n + s~nA) + DE(s~nA + s~nB)
smA smC
smG smB smA
+ E F (s~n C
G + s~n B) + FA (s~n C
G + s~n A )
smB smC
smG smA smC
smG
> 2(AB +BC +GD + DE +EF + FA)
+BG +CD
= 2p.
It follows that
P
RA + RB + Re >
- -.
2
82. We first prove an auxiliary result.

Lemma: Suppose a + b :::; 2c. Then
y'3
ma+mb+wc:::; 2(a+b+c),
= b == c.
and equality holds if and only if a =
266 Solutions
Proof of Lemma: Let a = u + x, b = u - x, and c = 2v so that -v ::; x ::; v

and v ::; u ::; 2v. Then
2_
Wc - (
ab 1 - (a
C
2) - (
+ b)2
_
U
2
- x
2
)(1 - u2) V
2 ::; U
2
-
2
x . (82.1)
Equality holds if and only if x = 0; i.e., a = b. We also have
2m aa = J2b 2 + 2c2 - a2 = J(3u - x)2 + 8v 2 - 8u 2 = I(x),

f(x),
say. Then 2mb = 1f ( - x). It is easy to verify that 1" (x) ::; O. Hence 1f (x) is
concave. This gives
1 1
m a + mb = 21 (x) + 21 ( -x) ::; 1(0) = Ju 2 + 8v 2 • (82.2)
Ju 2 + 8v 2 + Ju 2 - v 2 ::; V3(u + v), (82.3)
and equality holds if and only if u = 2v. Taking y = u/v, the inequality (82.3)
may be written in the form
2J(y2 + 8)(y2 - 1) ::; y2

2J(y2+8)(y2 + 6y - 4.
Since 1 < y ::; 2, this is equivalent to
4(y2 + 8)(y2 _ 1) ::; (y2 + 6y _ 4)2,

which reduces to
(2-y)3(2+y):::::0.
This proves the inequality (82.3) and hence the lemma.
Coming back to the solution of the problem, if a + b ::; 2c, then the lemma
gives
m aa + mb +W J3
W c ::; T(a + b + c).
Using ht ::; Wt ::; mt for t = a, b, c, we get
h a + Wb
ha Wb +m
m c ::; m aa + mb + W
J3
W c ::; T(a + b + c).
Similarly, if b + c ::; 2a, then the proof of lemma shows that
mb+mc+Wa::; T(a+b+c),
J3
and hence
ha + mb + W c ::; W a + mb + m c ::;
J3 + b + c).
T(a
Solutions 267
We may, therefore, assurne that a + b :::: 2c, b + c :::: 2a. Hence it follows that
a + 2b + c :::: 2(a + c),
so that a + c :::; 2b. This implies that
ma + m c + Wb :::;
V3 (a + b + c).
2
Moreover, we also have
4a+2c <
:::; 2(b+c)+(a+b)=a+3b+2c,
2a+4c < (b+c)+2(a+b)=2a+3b+c.
From these two we get a :::; band c :::; b. Thus
b < 2a + 2c - b :::; 2a + (a + b) - b == 3a,

and similarly b < 3c. Thus we obtain
a :::; b < 3a, c:::; b < 3c, a + b :::: 2c, b + c :::: 2a.
Suppose we show that

ha - hb :::; m a - mb·
Then
ha + mb + W c :::; m a + hb + Wc :::; m a + Wb + m c :::;

V3
2(a + b + c),
as observed earlier. Thus all we need to prove is h a - h bb :::; m

ma - mb.
We may assurne b == 1. Thus we have
1 1
- 1, 3
-3 < a:::;
3< a< "3 < c :::; 1, 2a:::; 1 + c, 2c:::; 1 + a, a +c > 1.
Heron's formula gives
16~2 = (1 + a + c)(1 + a - c)(1 + c - a)(c + a - 1)

(1+a+c)(1+a-c)(1+c-a)(c+a-1)
2(1 + a2 )c2 - c4 - (1 - a2 )2 = g(c),
say. Note that g is a strictly increasing function of c in the interval1/3

intervall/3 :::; c :::; 1.
Since 2c:::; 1 + a, it follows that g(c) :::; g((l
g((1 + a)/2). This gives
16~2:::; 136(3-a)(3a-1)(1+a)2.
268 Solutions
Thus
ha - hb 2Ll (~ - 1)
2Ll(1 - a)
a
<
< V;C~a2)V(3-a)(3a-1).
v'3(1-a 2 )V(3_a)(3a_1).
8 a
But we know that
2m a = y'2 + 2c2 - a2 , 2mb = y'2c2 + 2a2 - 1.
Moreover, for any positive 8s > t, we have

8-t
JS - Vi: ~ )2(8 + t)
Since a < b, we have m a > mb and hence

2(m a - mb) = V2 + 2c2 -
y'2 V2c2 + 2a 22 -- 11
y'2c a2 -
> 3(1- a 2 )(4c2 + a 2 + 1r 1 / 2 .

Note that the right side is a decreasing function of c. Hence, replacing c by
(1 + a)/2, we get
3
ma - mb ~ 4(1- a 2 )(1 + a + a 2 )-1/2.
M
y3(1- 3 - a 2 )(1
a 2) y'(3 _ a)(3a - 1) ::; 4(1 + a + a)-
2 1/2
.
This is simply
(1 - a)2(3a 2 - a + 3) ~ 0,
which is true.
83. Let AD, BE, CF be the medians oftriangle ABC and let C be its centroid.
First we solve a special case of this problem when the circum-circle of triangle
ACB is tangent to BC at B. In this case DB 2 == DC· DA. This gives
3a 2 = 4m~ = 2b 2 + 2c2 _ a2 .
= 2a 2 . This gives that 4m~ == 2a2

Hence b2 + c2 = + 2c2 - b2 == 3c2 and similarly,
4m~ = 3b •
2
Solutions 269
B D c
Thus
sin LABe + sin LAGe (2~b

( _b +
2mb 2:
+ _c
2m
)
cc )
sinA
sinA
~ _b
(( V3cc + vi b)
+ _c )
V3b A
sinA
sin
b2 + c2 .
V3
~smA.
bc sinA.
v3 bc
Using eosine rule, we get
2bceosA = b2 + c2 - a 2 = 2a 2 _ a2 = a2 •
Using this, we obtain

. A
2a 2 sinA 22 . 22
2a sm _ ._ sin 2A <
< _v'3
L.ABG + sin
sinLABG
sin sinLACG =
L.AGG = M a2 /2 eos A -= v'3
v'3 M sm2A - M
V 3 a 2 /2 eos A v3 - v3
B K D L c
In the general ease, there are two circles through A and touehing BG, say e
at K and L. Here K lies between Band D; L between D and G. Note that e
270 Solutions
is also the centroid of the triangle AK L. This follows from D L 2 = DG . DA =

2 = =
2
DK so that DK = DL showing AD is the median of AKL from A; and
2
AG : GD == 2 : 1. Moreover LABG :; LAKG, LAGG :; LALG. Suppose

LAKG :; 90°, and LALG :; 90°. Using the special case above, we get
sin LABG + sin LAGG :; sin LAKG + LALG :; ~.
In the other case, we may assume that LAKG > 90° and LALG :; 90°.
(Obviously, both cannot be larger than 90°.) Hence AG 2 2 + KG 2
2 > AK 2 2 .. Let
us introduce K L = a1,
al, LA = b1l and AK = Cl, the sides of triangle AK L.
Then
AG 2 = ~AD2
9
91 (2
2b 1 + 2C12 - a 2)
1 ,
KG 2 91 (2
2c1 + 2a 2
1 - b2)
1 .
Hence we obtain
91 (2
2b 1 + 2C12 - a 2) 2 1 (2 2 2)
1 > Cl + 9 2c 1 + 2a 1 - b1 .
By the property of tangency, we have DK 2 = DG ·DA. Using DK == at/2,

2 =
DG = = m a /3, DA = m a and m a also gives median length of triangle AKL, we

see that
3a~ == 4m~ == 2b~ + 2c~ - a~.
This reduces to 2ai = bi + ci. Thus we get
a2 c2
.-l.
3
> c21 + .2..
3
It bllows that ai > 4ci- Similarly, we get bi

bi > 7ci. Now we have
.
.
sm LALG =
= -
-~-
~
- sm LLAK .
.
= ~
~
M
J/ 1 -
V cos 2 LLAK.
2m Cl v3b 1l
However ai = bi
ai = bi + ci
ci -- 2blCl cos LLAK.
2b1C1 This gives cos LLAK (bi
(bi +
ci) / (4b 1l Cl) and hence
.
smLALG
Cl
Cl b2_
1- ( _1 +_
C21 )2
sin LALG = J3b
M
v3b 1l 4b 1 C1
Putting ci/bi = x, We get
sinLALG == \J14x -1- x2 •

4v3
Solutions 271
Using x < 1/7, this gives sin LALG < 1/7. Thus
sin LAKG + sin LALG < 1 + ~ < ~.

7 v3
This implies that
2
sin LABG + sin LACG < v'3'
84.
A
B E D F c
We have BD = = s - b. Hence BE/BD = = 02E/ID = = r2/r and this gives
BE == r2(s - b)/r. Thus ED = = BD - BE = = (r - r2)(s - b)/r. Similarly, we
obtain DF == (r - r3)(s - c)/r. Now P0 3 = 20 3 =
= r3 - r2, 0 20 = r2 + r3 and hence
02P == 2Jr2r3' Thus we have
2Jr 2r 3 =
= 02 P =
= EF =
= ED + DF =
= (r - r2)(s - b) + (r - r3)(s - c)
c).
r
We therefore get
2rJr2r3 = (r - r2)(s - b) + (r - r3)(s - c).
Similarly, we may obtain relations
2rJr3r l (r - r3)(s - c) + (r - rd(s - a)

2ry'r lr2 (r - rd(s - a) + (r - r2)(s - b).
Solving these, we obtain
s- a _r_ (- Jr2r3 + Jr3rl + Jr 1 r 2)

r - rl
s- b _r_ (Jr 2r 33 -- y'r3rl + Jrl r 2)
r - r2
s- c _r_ (Jr2r3 + Jr3rl - Jrl r2)'
r - r3
Using rs = [ABC] = Js(s - a)(s - b)(s - c), we may write
r 2{ (s - a) + (s - b) + (s - c)} = (s - a) (s - b) (s - c).
272 Solutions
Substituting for (8 - a), (8 - b), (8 - c), we get a quadratic equation in r:
(-,;r;1++ y'r2
( 1 -1+ 1I)?r--2
-
+ Vr3 ,;r; + y'r2 + vIr3)r
1).)
r- - 2 (,;r; yr:; r + 2J(
2v'r lr2
l r2 r 3 = )
= O.
This leads to
2v'rl r2 r3
2Jrl
r == -,;r;-;::r=1 --:+-y'r2----;::=r2=-:-+-.JT3'
r;7=r=1-:+-y' r27=r=2--:+-y'T3-'-;::r=-3 rJr=I=+:==r2=+=:==r=3
-;r=3=------'
---v't=r=1 =:=+=r=2=+=:==r3
+ r2, r2 + r3, r3 + rl. Hence, as in the
Now the triangle 0 110 220 3 has sides rl
case of ABC, two ways of expressing the area of a tri angle give
r'(rl + r2 + r3) =
= [0 110 220 3] = vrlr2r3(r1 + r2 + r3).
3] =
We therefore obtain
r 2v'r1 + r2 + r3
2Vr1
r' ,;r; + y'r2 + .JT3
y'T3 - Jv'rr l + r2 + r3 .
But ,;r;
+ y'r2 + .JT3
y'T3 :::; V3J r 1 + r2 + r3, where we have used the Cauchy-
v'3v' r1
Schwarz inequality. Hence
r
~> 22
->--=v3+1.
r.:;
r' -- V3
r' v'3 _- 11 = V3 + 1.
85.
M _____- - _
A
B c
K
K
= LKAC and LCML == LCBL == w, the Brocard
We observe that LKMC =
angle. Thus we get
LKML == LKMC + LCML == LKAC + LBAK == LBAC.
Similarly, we obtain LLKM = LCBA and LMLK = LACB. These relations
show that the triangles ABC and M KLare similar. Since both have the same
circum-eircle, they are indeed congruent. We also observe that
LAnC == 180 0 - LnCA - LnAC == 180 0 - w- (LBAC - w) == 180 0

- LA.
Solutions 273
Using the sine rule in triangle AnC, we get

An AC b
sinw - sin LAnC =
- sin A·
Thus, An == b sin w j sin A. Similarly, Bn == c sin w j sin B, cn == a sin w j sin C.
Using similar arguments in the tri angle MKL, we obtain Kn = asinwjsinB,
Ln == bsinwj sin C and Mn == csinwj sinA. Thus
An
nK
Bn cn
+ nL + 12M =
L
L: bsinB
a sin A =
L
L: b2
a2 2 3,
cyclic cyclic
by the AM-GM inequality. Observe that equality holds if and only if a = b = c,

wh ich corresponds to the case of an equilateral triangle.
86. We use the steps developed in example 3.6.8. Using the estimates there,
we get
a + y + z) a 2 ahb(x + z) ahc(Y + x)
(x
rA
> 2~
-+
x 2~x
+ 2~x
b (x + Y + z) b2 bhc(Y + x) bha(z + y)
rB
> 2~
-+
y 2~y
+ 2~y
c (x + Y + z) c2 cha(z + y) chb(x + z)
> 2~
-+
z 2~z
+ 2~z .
rc
Thus
L: rA
' " ->
~->
a (x+y+z)",a a22 ' " hb {
(x+ y +z)L: -+
~-+~-
az CX}
- a+c+-+- L:
- 2~ x 2~ x z'
where the sum is cyclically over a,b,c and x,y,z. However, the AM-GM in-
equality gives
az cx
a + c + - + - 2 a + c + 2vac,
x z
and the Cauchy-Schwarz inequality gives
a2 b2 c2 )
-+-+-
(x+y+z) ( ~+-+- 2 (a+b+c) 2 .
x
X Y z
Thus, we have
'"'~
" a > (a+b+c)2
(a + b + C)2 hb (
hb 2vac)
~
L. rA -> 2~ + 2~ (aa + c + 2vac)
ac
(a+b+c)2 ",a+c+2vac
2~ +~
L. b
> ~(abc)2/3
9 (abc) 2/3 +
2~
2~ + '"
L. b
~
{a b} '" vac
' " {~+~} + 2 ' " VaC
b + a~ + 2 ~
L. -b-
b
9
> ~(abc)2/3
2~ (abc)2/3 + 12.
2~
274 Solutions
However (abc)2 ~ (4ßjJ3).

(4ßjV3). (See 3.6.8.) Thus, we obtain
L~~
'~-~
" a 18 ;;;
1!+12=6(2+V3).
;q+12=6{2+v3).
rA v3
87.
As
Al
Let the septagons be inscribed in three concentric circles of radii RA, RB, Re
respectively. (See figure.) Then we have
C1 C4 Re C1 C2
A I1 A 4 RA A1A2
B I1 B 3 RB B 1 B2
A 1 A3 RA A1 A2 ·
Using A 1 A 2 = B I1 B 3 = C 1 C 4 , we get
C1 C2 = (A 1 A 2 )2 B 1 B 2 = (A 1 A 2 )2
A1 A4 ' A1 A3
Let A 1 A 2 = a, A 1 A 3 = band A 1 A 4 = c. Then we obtain C I1 C2 = a 2 jc and
B 1 B 2 = a 2 jb. Using Ptolemy's theorem, we also have
A 1 A 2 . A3 A 6 + A2 A 3 . A 1 A6 = A I A 3 . A 2 A6 .
Using the regularity of septagons, we see that A 3 A 6 = c = A 2 A 6 , A I A 6 = b

and A 2 A 3 = a. Thus ac + ab = bc or
1 1 1
- + - =-.
b c a
ß2 (B I B2 )2 a2
ß1 (A I A 2 FF b2 '
ß3 (Cl C2)2 a2
ßI (A 1 A 2 FF c2·
Solutions 275
Thus
~2 + ~3 _ a2 a2 a2 (~ ~) 2 _ ~
~1 - b2 + C2 > 2 b+ c - 2'
We also have
~2 + ~3 = a2 { (~ + ~) 2 _ ~} = 1 _ 2a 2 .
~1 bebe bc
However,
a = 4R~ sin 2 (7r /7), b = 4R~ sin 2 (27r /7), c = 4R~ sin 2 (37r /7).
Thus
a2 _ (7r/7)
sin 2 (7r /7)
be
bc - sin(27r /7) sin(37r /7)
sin( 7r /7)
2 cos( 7r /7) sin( 47r /7)
1
= 8 cos2 (7r /7) cos(27r /7)
1
=
4 cos(27r /7) (1
/7) (1 + cos(27r /7))
/7)) ..
But cos(27r/7) < cos(27r/8) == cos(7r/4) == 1/V2.
1/y2. Hence
a2
->
1
---
V2
y2 -- 1
be 2(1 + V2)
y2) -- 2 .
We thus obtain
~2 + ~3 = 1 _ 2 a2 < 1 _ 2 (V2 - 1) = 2 - V2.

~1 be 2
88.
A
A
B
B D
D D' c
276 Solutions
Choose D' on BC and E' on CA such thatD'IE' is perpendicular to CI. We

may assume that CD 2: CE. Since CI bisects LC, we have
ID CD
IE = CE 2: 1.
Hence
[DID') ~ID'. ID· sinLDID'

2
> ~IE'.IE.sinLEIE' = [EIE');
here we have used D' I = I E'. Thus it follows that [CD E] 2: [CD' E'). But
note that
CD' = ID' ID' cos (CI)
sin (CI 2) , r = ID' cos(C/2).
sin(C/2)' 2.
Thus
, , , , r 22 2r22
2r 2
[CDE]=2[CDI)=r·CD
[CD'E'] =2[CD'I) =r·CD'= sin(C/2~cos(C/2)
. (CI)
= sm 2 cos (CI)
2 = -:--C
sinC
sm 2:2r.
2:2r 2
We hence obtain [CDE] 2: [CD'E') 2: 2r 2.
89.
B
c
A( )x
A~------~----~~X----~D
=:""D
Construct an equilateral triangle AT X on the segment AX. Using the eosine

rule, we have
AX 2 = AB 2 + BX 2 - 2AB . BX cos LABX.
ButnotethatcosLABX ~ -1/2. ThusAX 2 2:AB 2 +BX 2 +AB·BX. This

leads to
4AX 2 > 4AB 2 +4BX 2 +4AB·BX

3(AB + BX)2 + (AB - BX)2
> 3(AB + BX)2.
Solutions 277
Thus 2AX ;::: V3 (AB + BX). Finally,
2AD = 2AX + 2XD

> V3(AB+BX)+2XD
V3 (AB + BC) + V3 CX + 2XD
>
;::: V3(AB+BC)+V3(CX+XD)
> V3(AB+BC+CD).
90.
Using the property of angle bisector, we have
ca ba 2 4bc
BD=~ ,DC== -b-' AD == (b + c)2., 8
AD (8 -- a).
8(8 a).
b + c'
BD=b+c' ,DC b+
+cc'
Thus,we get
AD AD 2 AD 2 _ 48(8 - a)
DK - AD·DK BD· DC - a2
BE 48(8 - b) CF 48(8 - c)
EL = b2 ' FM = c2
278 Solutions
Hence, we obtain
AD BE CF (8-a)
48 { - (8-b) (8-C)}
DK + EL + FM - + -b2- + -c2-
a2
r
48 2 { L (8 ~ a) :2 }
> 48 2 { L (8 ~ a) . ~
{ I 3}2
r
48 2 L ~ --;
{28 L ~ -6
> (9 - 6)2 = 9,
where we have used the convexity of the function fex)
f(x) = x 2 on the realline
realline,,
and the AM-HM inequality.
91.
F'
E
C
C'
Join BD and EA. Note that BDC and EF Aare equilateral triangles. Hence
BA == BD and EA == ED. It follows that BE bisects LABD and LAED.
Construct external equilateral triangles ABC' with AB as base and EF' D
with ED as base. Then C' is the reflection
refiection of C in BE and F' is that of F in
BE. Now AC' BG is a cyclic quadrilateral. Hence, using Ptolemy's theorem
and the equilaterality of AC' B, we get C'
C'GG == AG + G B. Similarly, we may
GB.
obtain HF' = DH + EH. Hence
AG + GB + G H +
+GB+GH + HE
DH +HE
+DH = C' G + G H + HF' 2:
C'G+GH 2 C' F' = CF,
C'F'
refiection. Note that equality holds if and only if G and H

by the property of reflection.
both lie on C' F'
F'..
Solutions 279
oa22= R 2 - (1/9)(a 2
92. We have OG + b2 + c2 ). Using rs = ß
~ = abc/4R, we get
2rR = abc/(a + b + c). Thus
R2 _ oa2 = a 2
2
+ bb22 + c22 > b
abc
a c = 2rR.
= 2r R.
R 2 - OG 2 = 9
9 + b+ c
-- aa+b+c
oa22:::;~ R 2
This gives OG - 2Rr, which is the needed inequality.
93.
c ________
___- ____
D' , /: I",....
D ~--~~~~--------~B
::AB
-+
-t
For any point T, let T denote the position vector of T with respect to some
coordinate frame in the plane. Then we see that
-+
-t -+
-t -+
-t -+
-t
-+
-t +D
A+D
A -+
-tB+C
S= --2-'
5=-2-' Q=-2- '
Q=-2-'
so that
-+
-t -+
-t -+
-t -+
-t -+
-t -+
-t
S+Q
5+Q A+B+C+D
2 4
On the other-hand,
-+
-t -+
-t -+
-t -+ -t
-+
-t D+
D +B
B -+
-t A+C
y= --2-' X= --2-'
---2-'
so that
-+
-t -+
-t -+
-t -+
-t -+
-t -+
-t
Y+X A+B+C+D
2 4
This shows that N is also the mid-point of X and Y. Note that OX and OY
are perpendicular to AC and BD respectively. Hence, the circle with GM
OM as a
diameter passes through X and Y.
Y . Now ON is the median of tri angle XOY.
triangle XOY .
It follows that ON:::;
0 N ~ diameter = 0 M.
280 Solutions
94.
A
Observe that AK ::; 2R and AD 2' h a . Thus

AD ha
la = AK 2' 2R·
Using similar estimates for the other two fractions, we get

h aa
'"""' la
~ sin 2 A
cyclic
>
L
'"""'
~
cychc
cyclic
h
2Rsin 2 A
ha
L
'"""'
~
cyclic
ha
asinA
1/3
3 hahbhc
> (
abc sin A sin B sin C
)
3.
(Here we have used h a = bsinC, hb = csinA and h c = asinB.)
95.
/~
I ~A
M
L
K
K
Solutions 281
Let us invert the configuration with respect to the circum-circle f of the triangle
ABC. For any point X -::j:. 0, let X' denote the inversion of X in f. The
inversion of the circum-circle of the triangle OBC is the li ne BC. Hence
K' = D, the point of intersection of AK and BC. The property uf inversion
shows that for any two points X, Y -::j:. 0,
X'Y' = R
R 2 ·XY
2
.XY.
X'y' = OX.
OX·OYOy·
Thus,
R 2 ·A'K'
AK = R ·A'K'
AK = OA'. OK"
This gives
AK R 2 ·A'K' A'K' ,.
. OK = R 2 )
OA' (since OK ·OK'
OK OA'·OK'·OK
AK'
OA (since A' = A)
AD
OA·
OA'
Let OD : DC = = x : y, CE : EA == z : x, AF : FB == y : z. Then it is
easy to calculate that AD/OA = (x + Y + z)/(x + y). This gives AK/OK =
(x + Y + z)/(x + y). Similarly, BL/OL = (x + y + z)/(y + z), CM/OM =
(x + Y + z)/(z + x). Thus we get
AK
OK
BL
+ OL + OM
CM
(x+y+z) - (1
-+--+--
(x+y+z) (_1_+_1_+_1_)
x+y y+z z+x
1 1)
9
>
2'
by the AM-GM inequality.
96. Let AD, BE, CF denote the medians of ABC so that m a = AD, mb = BE,
m c = CF. Let G be the centroid of ABC. Extend GD to K such that
GD == DK. We observe that
2 2 2
GB = -mb,
3
BK = -me,
3
KG = "3ma.
[GBK] = 2[GBD] = ~[ABC].

282 Solutions
B D c
Hence the area ß m of a tri angle whose si des are m mb, m c is

m a , mb,
9 3
ßm = 4[GBKj = 4[ABCj
Thus, we have to prove that
3V3
-1- + -1- + -1- < --.
mamb mbmc mcm a - 4ß m
Since this relates the area of a triangle with its sides, all we need to prove is
that in a tri angle with sides a, b, c and area ß the inequality
1 1 1 3v'3
-
ab
+bc- +ca--<4ß-
holds. But, this is the standard inequality
. A
sinA
sm . B + sinC
+ sinB
sm . C <
sm 3v'3
3v'3
::; -4-'
4 '
in a tri angle ABC.
97. We have
2 4bc
w~a = (b
w + c)2 s(s - a) :S
::; s(s - a),
and similar inequalities for w~ and w~. U sing Cauchy-Schwarz inequality, we

get
(W a + Wb + w c )2 < 3(w 2 +w
3(w~a 2 +w
w~
b cw~)
2)
< 3(s(s - a) + S(S

s(s - b) + s(s - c))
3s 2 .
Hence the inequality follows.
Solutions 283
98.
A
Let BD : DC = x : y. Using Stewart's theorem, we get
xb 2 + yc2 = (x + Y) (AD 2 + BD· DC).

This implies that
xb 2 + yc 2 = ---
- 1 -
AD 2 AD 2
---"-------
-,-_---,---C:-_ _ -
(x+y)BD·DC - BD·DC AD·DK·
Thus,
AD xb + yc
2
- 1
2
-,---,--"-------
-,-_---,---C:-_ _ -
DK (x + y)BD· DC .
But, we know that
BD=
BD-~ xa DC=~
DC=~.
- x+y' x+y·
x+y
Thus, we get
AD (xb 22 + yc22)(x + y)
y---,--) _ 1
-'-(
x_b_ +_YC_-'-,:-)(c-X_+_
DK xya 2
>
(vx
(,fi by'y + y'y cvx)
c,fi)22 _ 1
--'-----'---...,:----'-- - 1
xya 2
(b+c)2_ 1 ·
a2 ''
we have used the Cauchy-Schwarz inequality. Using similar lower-bounds for

the other two ratios, we obtain
AD BE CF ' " (b + C)2
DK + EL + FM > ~
cyclic
a2 -3
2
> ~ (b+C + c+a + a+b) -3
3 abc
6 2
> 33 -3 = 9..
--3=9
284 Solutions
99.
A
/
r\fl _.Q
p .-
F-f .
L / / ,
B
//-4-- D
~c
Let D, E, F be the mid-points of BC, CA, AB respectively. Let P and Q be
the reftections of D in AB and AD respectively. Join DP and DQ. Let these
intersect AB, AC in L, M respectively. We have
1
2LM == PQ::; PF + FE + EQ =
= DF + FE + ED =
= 2(a + b + c) == s.
But ALDM is a cyclic quadrilateral. Hence, Ptolemy's theorem gives
LM . AD =
= AL· MD + AM . LD.
Thus we obtain
LM AD {AL MD AM LD}
AD . AD + MD' AD
AD { cos a sin fJ + cos /3 sin a}
AD sin( a + /3) = AD sin A.
This shows that LM = m a sin A. Hence

. a am a
s 2': 2LM == 2m aa sin A == 2m
2maa .. 2R = R'
Ra.
Thus am ::; sR. Similarly, we may obtain bmb ::; sR and cm c

amaa ::; ::; sR. These
three estimates give
max {am a , bmb, cm c } ::; sR.
100. If () denotes one of the angles between the two diagonals, then we know
that the area of ABCD is AC· BD· sin()/2. Thus
1 1
1 = -AC· BD· sin() < -AC· BD.
2 - 2
This shows that AC· BD 2': 2. Now we know
1 1
[ABC] = -AB· BC· sinB < -AB· BC
2 - 2
Solutions
Solution8 285
1 1
[ADC] == -CD· DA· sinD< -CD· DA.
2 - 2
Adding, we
Adding, weget
1 == [AB CD] ::;

[ABCD] ::::: ~(AB. BC + CD· DA).
Similarly, we get
Similarly,
[AB CD] ::;

1 = [ABCD] ::::: ~(AB'
~(AB. DA + CD· BC).
two, we get
Adding these two,
(AB + CD) (BC + DA) :::::

~ 4.
This implies that
(AB + BC + CD + DA)2 :::::~ 4(AB + CD) ~ 16,

(BC + DA) :::::
CD)(BC
and hence AB + BC + CD + DA:::::

DA ~ 4. Again
(AC + BD)2 :::::

~ 4AC· BD:::::
BD ~ 8. 8.
It follows that AC + BD :::::

~ ,;8.V8. Hence
AB VB.
+ BC + CD + DA + AC + BD :::::~ 4 + V8.
101.
Join M ttoo A, B, C, D. Draw perpendiculars OL and MK to AB. We have

1
2MA. MB· sinLAMB
[MAB] = 2"MA. sinLAMB,,
1 .
[MCD] = 2MC. MD
= 2"MC. MD··smLDMC.
sin LDMC.
286 Solutions
But
=
LDMG = ~LDO == ~LAOB == 180 0
- LAMB,
and hence sinLDMG = sinLAMB. Thus
MG·MD _ [MGDj
[MGD] _ h+a
MA·MB - [MABj--h-'
[MAB] -h-'
where h = M K and adenotes the side length of ABGD. Note that h <
r - OL = r - (a/2). We hence obtain,
h+a r-(a/2)+a 2r+a

--> =--.
h - r-(a/2) 2r-a
However r = a/J2.
a/V2. Thus
h + a > 2 + J2 = 3 + 2V2.
-h- - 2 - J2
102. We may write this in the equivalent form
a2 + b2 + c2 2: 18Rr.
Using two different expressions for the area of a triangle, we have

abc r(a + b + c)
- =rs = .
4R 2
This gives Rr == abc/2(a + b + c) and hence the inequality is equivalent to
gabe:::: (a + b + c)(a 2 + b2 + c2 ).
However, observe that
(abc)2/3 < a2 + b2 + c2
3
a+b+c
(abc) 1/3 <
3
Multiplying these, we get the desired inequality.
103. We use the expression
cosA b +c - a 2 2 2 R(b2 + c2 - a2 )
cotA
cot A =- -=-- ----
---
sin A 2bc sin A abc
and similar expressions for cot Band cot G. Thus the inequality to be proved
is:
(Jb2 + c2 - a2 + Vc2 + a2 _ b2 + Ja2 + b2 _ c2) 2 :::: ~:!c = (a + b + C)2.

Solutions 287
Using the concavity of f(x) == JX on (0,00), we observe that

!2 V /b 2
!Jb + c2 - a2 !2 v/c2 + a2 -
+ !Jc b2 <
- c,
and we have similar expressions for other sums. Thus it follows that
Jb 2 + c2 - a2 +J c2 + a2 - b2 +J a2 + b2 - c2 ::::: a + b + c,
and that completes the proof.
104. Draw perpendiculars from P on to BC, CA, AB to meet them in X, Y,

Z respectively. Draw perpendiculars from Y and Z on to the extended line
X P to meet it in N and M respectively. Since P Z AY is a cyclic quadrilateral,
it is easy to see that ZY = PA sin LA.
B x c
But ZY > ZM + YN. We also note that BXPZ is a cyclic quadrilateral.

Hence
LXBZ = 180 0 - LXPZ = LZPM.
It follows that LZPM = Band hence ZM = ZPsinB. Similarly, it may be
proved that Y N == Y P sin C. Thus we get
PA sin A :::: P Z sin B + PY sin C.
This implies that a . PA:::: b . P Z + c . PY. Similarly, we may prove that
b· PB:::: c· PX + a· PZ, C· PC:::: a· PY + b· PX.
Thus we obtain
(b~ + ~ ) + PY ( ; + :2 ) + P Z (:2 + ~ )
AP BP CP
BC2 + CA2 + AB2 >
> PX
2PX 2PY 2PZ
> --+--+--
bc ca ab
4[ABC] 1
abc R'
where [ABC] denotes the area of triangle ABC.
288 Solutions
105. We observe that

N
Lak= L P
~ [N]
P
n=2 p<;,N
p a prinle
On the other hand
L P~~ [;]
~
p<;,N
[N]
P
<
< L
p<;,N
~
P
(N) P
p a priITle p a priITle
1
N L
~l'7
p2
p a prllTIe
1
< N ( =1 +
1
t; 00
(2k
1
+ 1)2
)
=1 + =1 t;
1 1 00 1 )
(
< N k(k + 1)
N
2
Thus we see that
N
N N
Lak< 2'
Lak<
k=2
k=2
N
2 .
(( a2 ' , , + aN
+ a3 + ... aN)
N-l
) N-1
a2a3" 'aN <
N-1
< 2N-1
1 ( + 1) 1 N - 1
N-1
N-l
3
< 2N - 1 '
Hence we get
N
L
L a2 a 3" ·a
'ann a2 + a2a3 + a2a3a4 + a2a3a4a5 + ...
n=2
< ~2 + ~6 + ~ + ~ + 3 (~5 + ~6 + ...

12 60 2 2
.. ,))
46 6
60 + 32
229
-<1
240 .'
Solutions 289
106. We show that C == 1/8 is the least constant and there is a configuration
for which this is attained. We have
(t
J=l
Xj )4
(t j=l
x; + 2 L
l'5.j<k'5.n
XjXk) 2
> 4(t,Xj) (2 "~5n XjX')

8 (t x;) ( L
j=l l'5.j<k'5.n
XjXk)
> 8 (. L XjXk(X; + x%)) .

l'5.j<k'5. n
The secünd inequality is an equality ünly if n - 2 üf xj's are zero. If we

take X3 = X4 = ... = X n = 0, then the first inequality becümes an equality
ünly if Xl = X2. Thus if we take Xl = X2 = a für süme real number a and
X3 = X4 = ... = X n = 0, then we get equality with cünstant C = 1/8.
Alternate Solution:
It is sufficient tü prüve that
'"'
~ XiXj(Xi2 + Xj)2 ::; 1
8'
l'5.i<j'5.n
under the assumptiün that L Xj = l.

We can make use üf Newtün's identities; if für k ::; n
n
5
Sk =
k = LX;,
"Lx;, and ak
o.k = L
"L xi2 ...
Xj, xi2 ... Xjk'
j=l 1'5.], ::;i2::;···::;jk::;n
l::;j, '5.i2'5. .. ·'5.jk'5.n
and ak
o.k = ° für k > n, then
52 - 0. 1 5 1 + 20.2 = 0,
54 - 0.153 + 0.252 - 0.351 + 40.4 = 0.
We thus have
(t (tXj)
n
L 2
XiXj(Xi +X j )
2
x ;) - "Lx;
LX;
l'5.i<j'5.n )=1 )=1 j=l
53
S3 -- 54
S4 = o.2S2 -
0.252 0.3 + 40.4
0.2(1 - 0.2) + 40.4 - 0:3·
290 Solutions
Here we have used 0:1 = SI = 1. However, we know that 0:2(1 - 20:2) ::; 1/8.
On the other hand
40:4 - 0:3 L
<12<h
), <12<is
x),x12 x is
h
Cj~hlXk)
Cj~") x,)
L x),x 12 xis
h<12<is
L
), <12<is
xj,x 12 x is
Cj~Jd -1) xk
L xhxhxjs(xh + x 12 + xis) o.
::; O.
h<12<13
Thus we get
'L.,;
" (2 +x j2)::;-,
XiXj Xi 1
1< " "
_'<J<:,n
8
for n 2: 4.
107. Suppose, if possible, one of a 2 + b2 , c22 + (f2 is less than 1. (Observe that
2
none can be equal to 1.) Since (ac+ bd - 1) 2: 0, the product (a 2 + b2 -1) (c 2 +
d 2 - 1) is positive. Hence it follows that a 22 + b22 -- 1 < 0 and c2 + d 22 -- 1 < O.
Let us put Xx = 1 - a2 - b2 and y = 1 - c2 - ~. Then 0 < x, y < 1 and hence
4xy (2 - 2a 22 -- 2b 22)) (2 - 2c2 - 2d22))
> (2ac + 2bd - 2)2
(a 22 + b22 + X
X + c2 + d2 + Y - 2ac - 2bd)2
((a -
((a - C)2
C)2 + (b - d)2 + X +y f
> (X+y)2.
This shows that (x _y)2 < 0, which is impossible. We conclude that a2 +b2 > 1
and c22 + ~ > 1.
Here is a generalisation used in the US Selection Tests-2004.
a1,a2,a3, ... ,an and b1I ,b22,b3 , ... ,b n real numbers such that
Let aI,a2,a3,
(ai +a~ + .. ·+a; -l)(bi +b§ + ... +b; -1) > (alb l +a2b2 + .. ·+anb n -1(
(a l b1
Show that ar + a§ + ... + a;, > 1, and bi + b§ + ... + b;, > 1.
The argument of the proof is essentially the same. But we prove an in-
equality due to Aczel which will help us to resolve the problem quickly. Let
X1,X2,X3, ... ,X m and YI,Y2,Y3,
XI,X2,X3, Y1,Y2,Y3, ... ,Ym be real numbers such that
2
Xl > x 22 + x 32 + ... + x 2m ·
Solutions 291
Then
22
(XIYI -- X2Y2 -- X3Y3 - .. ,. -- XmYm)
(XlYl
2 ((xi
~ Xl2 -- X
x~2 x~2 - ... -- X
2 - X3 2)(2
m (yr
x~) yi2 -- Y3
Yl -- Y2 y~2 -- ... 2) .
... -- Yy~).
m
(*)
This follows from the property of the following quadratic polynomial:
m
m
f(t) = (Xlt+yI)2_L:(Xjt+Yj)2
(Xlt+Yl)2_ (Xjt+Yj)2 L
t
j=2
xi -- f
((Xi
)=2
)=2
+ 2(XIYI - fXjYj)t+
xX;)t 22 +2(XlY1- fXjYj)t + (Yi
(Yi -- fy;).
)=2
)=2
Y;). f
)=2
)=2
Note that the leading coefficient is positive. Moreover,
L: -~
-)- Yj )2 ::;
Yl
Y1 X·Yl
XjYl
ff ((
-- -Xl )) =
=
m
m
~
((
- + Yj
+ o.
::; O.
Xl j=2 Xl
Hence the discriminant must be non-negative, which implies (*).

al , a2,
We apply (*) to the sequences 1, al, .. an
a2, .... an and 1, b1l , b22 , ..•
... bn . Since
ta; -1) (tb; -1) > (tajb _1)2 > 0,

( )=1 j
)=1 )=1
'LJ=1 aJ
it follows that both 'L-'j=l a; - 1 and 'L-'j=l
'LJ=1 bJb; - 1 have same sign. If both are
negative, then Aczel's inequality shows that
negative,
)2
n
~ajbj > (ta; -1) (t
(ta; (tb;bj -1)
-1)
t r
( 1-
)=1 )=1
> (1- )=1

( 1- aJbJ
taibi)2
This contradiction proves that 'LJ=1

'L-'j=l a;aJ - 1 and 'LJ=1 b; - 1 are both positive.
'L-'j=l bJ
108. Suppose all the xj's are distinct; say Xl

Xl < X2 < X3 < ... < XlOO.
XIOO. Then
Xj 2
~ j for 1 ::; j ::; 100. Thus
1 1 1 1 1 1
20= - + - + ... + - - < -+-+ ... +--.
-+-+"'+--.
y'xl..fX2 J X I00 --
JXlOO v1 v'2
J2 J100
J100
But for any natural number k, it is easy to see that
1 2
~ < Vk
v'k
vk v'kk++ ~
Jk=l
k-l
= 2 ( v'k -
2(v'k v'k"=l).
- Vk=-1).
292 Solutions

1 1 1
20 < -
VI++-
v'2++ ...
... + --
+ J100
J100
VI v'2
100
< 2 L (JA; - Jk=l)

k=l
2J100 == 20,
which is impossible. We conclude that not all xj's are distinct.
109. Observe that j(O) 1= 0 and j(l) 1= O. Hence 1 :S Ij(O)j(l)l. Let

a1,a2,a3," .,a n be the roots of j(x) == 0, which alliie in (0,1). Then
j(x) = a(x - ad(x - (2)'" (x - an).
Thus we have
1:S Ij(O)j(l)1 = lal 2 1a 1 . a2'" a n (l - ad . (1 - (2)'" (1 - an)l·
But we know that x(l - x) :S 1/4 for x E (0,1) with equality only if x = 1/2.
Since not all roots are equal, we conclude that
4n < lal 2.
Since a is an integer, it follows that lai 2: 2n + 1.
110. If x, y, Z, ware aB positive, then the AM-GM inequality shows that

x y Z w
m == --
y
+ -Z + -w + -x -> 4.
Hence the result folIows.
111. If any one of x, y, z is 0, then the remaining two are also O. Suppose none
of them is O. Then x, y, z are all positive. Multiplying all the relations,
64xyz = (1 + 4x 2 ) (1 + 4y 2) (1 + 4z 2 ) 2: 4x . 41/ ·4z,
where we have used the AM-GM inequality in the last step. This shows that
equality holds in the inequality, implying x = y = z = 1/2. Thus we have two
solutions:
(x, y, z) = (0,0,0), (1/2, 1/2, 1/2).
112. Let a1,a2,a3," .,a n be the roots of the given equation. Then
1 1 1 2
a1 +a2+a3+" '+a" = 1, -+-+"'+-=n
al a2 an
Solutions 293
This shows that
n2 = ( al+a2+a3+· ··+a n )(1 1

-
a]
1) >n,
+ - + ... + -
a2 an
2
by the AM-GM inequality. Hence equality holds in the inequality and this is
precisely the case when al =0:2=0:3=· .. ·=a
0:1 =a2=a3=· ·=O:n.
n.
1
1. Thus we may assume that 1 < a < b < c. Write bc =
= 1 ++ cz. From this we obtain
and ab =
1
113. Observe that no two can be equal. Nor any one of them can be equal to
1
= 1 ++ ax, ca == 1 ++ by
xyzabc = (bc - 1)(ca + b ++ c) ++ ab ++ bc ++ ca - 1.

l)(ab - 1) = (abc)2 - abc(a +
l)(ca - 1)(ab
This expression shows that abc divides ab + + bc ++ ca - 1. Since abc'and ab +

+
+ ca - 1 are positive integers, we can immediately get
bc +
abc< ab + bc + ca. (*)

Using a 1 now implies that a == 2. This can be put back in
+ c). Again using b < c, we get bc < 4c or
the inequality (*) to get bc < 2(b +
equivalently b < 4. Taking in to account 2 = a < b, we can only have b = 3.
Using a == 2 and b == 3, we see that the only possibilities for c are 4 and, 5 of
which 4 can be ruled out (why?). Thus the only solution (a, b, c) with a < b < c
is (2,3,5). Permutations of this solution give all the other solutions.
= ß=
114. If ABC is equilateral, then a0: = = '/ = b=
" and a = = c. Thus both the left
hand side and the right hand side reduce to 3a.
Suppose the above relation holds. Then the AM-GM inequality shows that
~+~+~ >
~ ~~+~2+2~
> a 2bc + bCa + cab
abc(a + b + c).
Thus,
a4 + b4 + c4 > a+b+c
abc
> acos(ß - ,/) + bcosh - a)
,,) + + ccos(a
0:) + ccos(o: - ß)
a4 + b4 + c4
abc
Hence equality holds every where. This implies that
+ b4 ++ c1 = abc(a ++ b ++ c),
a4 +
which in turn gives a = b = c

294 Solutions
Xl + -1 2:
X2
2 f!;l
-,"
X2
'X2000 + -1 2:
Xl
2 J X2000
- -.
Xl
41000 = (
Xl + X21 ) ( X2 + X31 ) • .•
( X2000 + Xl1 ) > 41000 .
We conclude that
1 1 1
Xl =-
-,, X2 = --,, •••
... ,X2000 = -.
=
X2 X3 Xl
It follows that
1
Xl = X;3 == X5 == ...
= ... == X1999 = 2, and
Xl999 = X2 = X4 == X6 == ...
= ... == X2000 == 2'
116. Using the first relation, we have
~ 3(xYZ)1/3,
1 =x+y+z 2: 3(xYZ)I/3,
so that xyz ::; 1/27. Using the second relation, we get

26
x 3(1 - x) + y3(1 - y) + z3(1 - z) = 1 - xyz 2: 27'
r
However, for any t E (0,1), we have
3t 3 (1 - t) = t· t· t· (3 - 3t) ::; C+ + t t4+ 3 - 3t

81
256
Thus,
81
x 3(1 - x) + y3(1 - y) + z3(1 - z) ::; 256'
which contradicts x 3(1- x) + y3(1- y) + z3(1- z) 2:~ 26/27. We conclude that

there are no solutions to the given system in positive real numbers.
117. Observe the equation (a + b - x)2 = a - b has two distinct roots if and
only if a > b. Similarly, the second equation has distinct roots if and only if
ab > 1. Since a > 0, we have a 2 > ab > 1 so that a > 1. Thus we obtain
1
a> b> -, a> 1.
a
Now, the bigger roots of the two equations are
a + b + ~, ab + 1 + J ab - 1.
Solutions 295
Equality of these two gives
= (a - 1) (b - 1)
~= + J ab - 1.
If a > b ~ 1, then
~~Jab-1.
It follows that (a + I)(b - 1) :::; 0, showing that b :::; 1. Hence b =
= 1. On the
other-hand if (I/a) < b:::; 1, then
~:::;Jab-I,
and hence (a + I)(b - 1) ~ o. It follows that b ~ 1, and thus b == 1. Thus the

bigger two roots are equal if and only if b = 1 and a > 1. In this case both the
equations reduce to (a + 1 - X)2 = (a - 1) and hence the smaller roots are also
equal.
118. We have
al +a2+a3+·· ·+a n = -no
If we use the Cauchy-Schwarz inequality, we get
n2 =
( ~n)2
aj
n
:::; n ~ a;.
Repeatedly applying this we obtain

n n
n4 < n 2 . n . "" 4
~ aJ = n3 "
~ "a4
J'
j=l j=]
n n
n8 < n ß . n . ""
~aJ
8 = n7 "
~
" a 8J'
j=l j=l
n
nIß
< nn 14
l4 - n .. L a}6 == n 16 .nIß.
j=1
j=l
Thus equality holds in the Cauchy-Schwarz inequality. This implies that all
the aj's are equal. Hence aj = -1 for all j.
119. It is easy to check that Xl = X2 = X3 = X4 = X5 = Ais a solution. Suppose

not all the X j's are equal. Thus among X3, X5, X2, X4, Xl, two consecutive num-
bers are distinct; say X3 f. X5. We observe that whenever (Xl,X2,X3,X4,X5)
(XI,X2,X3,X4,X5)
is a solution, (I/Xl, I/x2' I/x3' I/x4, I/x5) is also a solution. Hence we may
assurne that X3 < X5.
Suppose Xl :::; X2. Then xi - X3X5 :::; 0 and x~ - X3X5 ~ o. This shows that
Xl :::; JX3X5 < X5 and X2 ~ JX3X5 > X3- Thus XIX3 < x~ and x~ :::; XIX3 < <
296 Solutions
X3X5. But x~ < X2X3 < X2X5. Hence relation (iii) shows that x~ 2: X2X5 >
X3X5, because X2 > X3. This is impossible.
:s; J.T3X5 < X5·
Suppose Xl > X2· Then (i) gives Xl 2: JX3X5 > X3 and X2 :s;
Using (ii) and (iv), we infer that
XIX4:S; max{x~,xn:s; X3 X 5, X2 X 4 2: min{xi,xD 2: X3 X 5·
This shows that X2X4 2: Xl X4, contradicting Xl > X2. Thus all the X /s must
be equal.
120. \Ve may assume that a > 1. For, if a < -1, we can take a' == - a and
xj == -Xj, 1 :s;:s; j :s;:s; 1000, to get a similar system in which a' > 1. Since left
sides are squares, we get xj 2: -l/a > -1, for 1 :s; :s; j :s;
:s; 1000. We may assume
Xl 2: Xj, 1::; j::; 1000. Thus Xl 2: X2,X3· If Xl 2: 0, then xIooo 2: 1 and hence
XlOOO 2: 1. This implies that Xggg > 1. Using induction, we get Xj > 1 for all
j. Thus either Xj > 1 for all j or Xj < 0 for all j.
Suppose X j > 1 for all j. Then Xl 2: X2 and hence xi 2: x§. x§. This implies xi
that X2 2: X3· Using induction, we have XI2:X22:X32:·· ·2:XlOOO 2: Xl. Conse-
quently all Xj 's are equal. Thus we have to solve the equation xx 2 = ax ax + 1. We
obtain
Xj = ~ (a + J a 2 + 4), :s; j :s;
for 1 :s; :s; 1000.
Suppose Xj < 0 for all j. Then Xl 2: X3 implies that xi xi :s;:s; x~ and hence
X2 ::; X4· This implies that X3 2: X5. Again induction shows that Xl 2: X3 2:
X5 2: ... 2: Xggg 2: Xl and X2 :s; :s; X4 ::; ... :s;
:s; XlOOO :s;
:s; X2· Thus we get
X2 = X4 = ... = XlOOO and Xl = X3 = ... = X999. In this case we have two
xi = aX2 + 1, x§
equations: xi x§ = aXI + 1. Hence
xi - x~ = a(x2 - xI).
If Xl = X2, then all Xj 's are equal and each is (a - J a2 + 4) /2. Otherwise
Xl + X2 + a = O. In this case xi + aXI + xi 1) = o. The discriminant is
(a 2 -
4 - 3a 2 . If a > 2/,;3, then there is no solution to the system. If a :s;
:s; 2/,;3,
then
Xl = ~( - a - J a 2 + 4) and X2 = ~ ( - a + J a 2 + 4),
or the other way round.
121. (By Rishi Raj) We begin with the observation that for a, bEN,
2a - a 2: 2b - b {:::=} a 2: b.
This follows easily from the fact that the function f(x) 2X - X is a non-
decreasing function on IR. Thus we see that
Vn - n 2: s {:::=} 2V n -n - Vn +n 2: 28 - s.
Solutions 297
The definition of s shows that it is sufficient to prove
2Vn - n - (Vn - n) ~ n.
We also have 2m - 1 ~ m for m ~ 1, which din be easily established by in duc-
tion. Thus
2 Vn - n = 2L:j=l(aj-l) = 2(a,-1)2(a 2 -1) .. ·2(a n -l) ~ al· a2·· ·a n = Vn .
This gives
2Vn - n - (Vr, - n) ~ n.
The desired inequality follows.
Alternate Solution:(By Aravind Narayanan)

Let k be the number of aj which are strictly larger than 1. Let us take al > 1,
a2 > 1, ... , ak > 1, and aj = 1 for k + 1 :S j :S n. Then
Vn = al + a2 + ... + ak + (n - k) = al . a2 . a3 ... ak.
Observe that Vn > n. Suppose Vn == n + r for some r such that 1 :S r < s.

~
Then we have
(n - k) + al + a2 + ... + ak == n + r
===} al + a2 + ... + ak = r + k
===} r + k > 2 + 2 + ... + 2 == 2k
-~
k
===} r > k.
But we see that
1 + 1 + ... + 1 +al + a2 + ... + ak =r - k+r+k = 2r.

~
r-k
This gives
l/r
2r == 1 + 1 + ... + 1 +al + a2 + ... + ak
~
>
-
r ((al ak)) l/r..
al a2 ... ak
r-k
We have used the AM-GM inequality. Thus we obtain
ala2··· ak :S 2r .
Hence
n +r = Vn = ala2··· ak :S 2rr .
It follows that 2" - r ~ n. But this contradicts the minimality of s. We
conclude that r > s. This gives Vn = n + r ~ n + s.
298 Solutions
122. Let us express Zk

Zk = Xk + iYk
Xk iYk for 1 ::::: k ::::: n. Then we have
n n
=L
1= L IZjl :: :: L
L (IXjl + IYjl) == L
L IXjl + L
L IXjl + L
L IYjl + L
L IYjl·
j=l j=l Xj2:0 Xj<o Yj2:0 Yj<o
By the pigeonhole principle, at least one of the sums is not smaller than 1/4.
By symmetry, we may assume it to be the first sumo Thus we get
~ x~O IXjl = Ix~o Xk

:::::
It follows that
LZkl:::::ILXkl:::::~'
xk2:0 Xk2:0
We, in fact, show that the constant 1/4 may be replaced by 1/7f. For areal
number x, let x+ = max {x, O}. We write for 1 ::::: k ::::: n,
Zk = rk (COSOk + i sinOk),
where rk = IZkl, 0 ::::: fh < 27f. Define
n +
1(0) = Lrk( cOS(O-Ok)) .
k=l
We see that
~1 r21f
r 1(0) dO
21f 1 n
~ rk Jo
r 21f
27f J
27f Jo 1(0) dO 27f (cos (0 - Ok)) + dO
o
27f
1
Lrk
k=l
n
!
-1f
1f /2
/2
cosOdO
1 n
;Lrk.
k=l
Hence there is a number 00 such that
1 n
1(00 ) ::::: - L IZk I·
7f k=l
We set
T = {j 11::::: j::::: n,cos (0 0 - Ok) > o}.
Solutions 299
With this set T, we obtain
!LZj!
JET
!e-iOO LZj!
J
LZjJ
e-iOD
JET
>
> Re(Le-iOOZj)
Re(Le-iOOZj)
JET
L rj cos (0 Oj)
rj Oj) 0 -
JET
Lrj( cos(Oo -Oj))+

JET
1
L IZjl·
n
1(00 ) :::: ;:
j=1
Here 1/7f is the best constant, but the proof needs more work.
123. Put
n n n
A = '~
" aJ'
2 B = ' " b2 C = "'a.b.
~ J'
. _ Ab j - Caj
~ J J'YJ - .~ .
j=1 j=1 j=1
An easy computation shows that
nn A
LYjXj=
LYjXj = ~
j=1
AB _ C2' .-
j=l
The definition of Yj gives
n
AI:ajbj - CI:a;
LajYj AB _ C2 = 0,
j=1
n
AI:b2 - C'Lajbj
LbjYj J -1
AB - C2 -.
j=1
Using these, we may now obtain
~
n 22 '" Ab Caj _ ~
Abjj -- Caj A
LYj ~ Yj AB - C2 -= AB - C2 .'
~ Yj = LYj
j=1
j=l
300 Solutions
It follows that
11 11 n n
'"""'
~Xj2 - '"""'
~Yj2 LX] - LYjXj
j=1 j=1 j=1 j=1
11
LXj(Xj-Yj)
j=1
11 11 11
LXj(Xj -Yj) - LYjXj + LYJ

j=1 j=1 j=1
11 11
LXj(Xj -Yj) - LYj(Xj -Yj)

j=1 j=1
L
11
(Xj - Yj)2 2: O.
j=1
Hence we get
n n A
~Xj
L x]2
'"""' ~ '"""'
2: LYJ2 = ~
~Yj ~')''
AB _ C2
j=1 j=1
which is the required inequality.
Corollary:
If (al,a2,a3, ... ,a n ) and (b 1 ,b2 ,b3 ,... ,b n ), n 2:
~ 2, are two sequences of real num-
bers such that ajb k =f. akbj for j =f. k, then
r
aJ2· <
(~) ~ (~akbj ~ ajbJ
'D·
"
J
-2 2
( 2:j aJ) ( 2: j bJ) - (2: j ajbj
Define
-1 a
Xj = (~ )
L akbj
koj.J
~ ajb k '
for 1 ::; j ::; n. Then we get
f11
j=1
ajXj =
L ajxj = (~)
(n
2
) -1
-1 fn
LL akba~:bk.
ajak
L akb-a _
j=1 koj.j
k#-j JJ J
For any I =f. m, we may regroup the terms of the form

amal alam
amb 1 - al bm
+ a1b ,
m - amb 1
to get the sum equal to O. Thus we see that
n
Lajxj = O.
j=1
Solutions 301
n
Lbjxj = 1.
2:)jXj
j=l
j= 1
Now we can apply the inequality in the above problem.
124. We may suppose that bf - b~ .... -- b;

b§ - .. b;, bi - > O. Consider the function f(x)
defined on JE.
~ by
f ( x)
f(x) = ( b2l1 - n x 2 - 2 ((alb l - a2b2 -_ ... - anb n ) x
b22 - ... - b2)
+ ( a21l -- a z222 -- ...
••• -- an
2)
z 2 2
(blx - aI)
al)2 - (bzx
(b 2x - a2)
a2)2 - ... - (bnx - an)
an ( .
Now bi - b§ b;,
b~ - ... - b; > 0 implies that bl1 :f.
-::J O. Taking x aI/bl, we obtain
= aI/bi,
2 2
f(~~) = - (b ~~
f ( ~~) =-(b2~~ 2 - a 2 ) 2 __ ...
-a2) ... -_ (b(bn~~
n ~~ _ am ) 2
-am) ~ O.
~O.
However since the leading coefficient
coefficicnt of the quadratic expression f(x) is pos-
itive, f(x) -+ 00 as x -+ ±oo. Since f(aI/b l ) ~ 0, the equation f(x) = 0
f(aI/bI)
has one root each in the intervals (-00,
(-00,aI/b [aI/bI' +(0).
1 ] and [aI/bi'
aI/bi] +00). Hence the
discriminant of the quadratic function f(x) must be nonnegative. This gives
( ai') bl - b2z - ... - bn2 )

- a 22 - . .. - an2 ) (2 ~ (al bl - azb z - ... - anb n ) 2.
aI/bI is a double root of f(x) = O. This forces

Equality holds if and only if aI/bi
= Abj for 1 ~ j ~ n, where A == aI/bi
aj = aI/bI.'
125. The inequality is equivalent to

n
...::--. Xj+l + Xj+2 + ... + Xj+n-2
L ~---=------"----
Xj+n - 2 2: n - 2.
~ n - 2.
6j=1 2x JJ·. + Xj+l
XJ·+1 + ...
. . .++Xj+n-2
X+ -2
J=1 J n
Using the Cauchy-Schwarz inequality,

n
'""' XJ+l
"" Xj+n-2 >
Xj+l + Xj+2 + ... + XJ+n-2
~
j=1
2x·
2xjJ
+ X
Xj+!
J
..
·+1 + .... ++ X+ -2 --
XJ+n-2
J n
( L XJ+l Xj+2 + ...

Xj+l + XJ+2 ... + XJ+n
Xj+n-2
- 2 ))
2
cyclic
L (Xj+l + XJ+2
(Xj+! Xj+2 + ... + XJ+n-2)
Xj+n-2) (2Xj + XJ+l
XjH + ... + XJ+n-2)
Xj+n-2) .
cyclic
Let us put S == 'E.7=1

'L;'=l Xj. Then
XjH + Xj+2 + ... + Xj+n
Xj+l -2=
Xj+n-2 = S - Xj - l - Xj,
Xj-l
302 Solutions
for 1 :s; j :s; n. (Here the indices are taken modulo n.) Thus we have to prove
that
n 2 n
(~5 - Xj-l - Xj) ~ (n - 2) ~ (5 - Xj-l - Xj) (5 - Xj-l + Xj).
n
L
2
2)s22
(n - 2)5 ~> (5
(S - Xj-l) - x;
j=1
~ (5
~ S 2- 25Xj_l
2SXj_l + X2
j_l 2)
- Xjj ..
j=1
However usmg 2"n

. uj=1 Xj_l = 2 t heng
= uj=1 Xj' "n
. ht SI
SI·deis
.
eIs
L 52S2 - 2SXj-l
25xj-l = n5
nS 2 - 25
2S L Xj-l = (n - 2)5
2)S2.
2.
Thus the inequality follows.
126. Put ,\-1 = 2:7=1 Xj and Yj = '\Xj,

'\xj, for 1 :s; j :s; n. Then yj's are in (0,1)
and 2:7=1 Yj == 1. The inequality takes the form
L Yl + 22Y2
' " Yl+ + ... 22Yk-l
Y2+··· + Yk
Yk-l+Yk 2n(k - 1)
2n(k-1)
~ 1-(Yl+Y2+···+Yk-l)
CYclic1-(Yl+Y2+···+Yk-l)
cychc ~~n-k+1·
n-k+1·
Using the new symbols
0:1
0:1 1-(Yl+Y2+···+Yk-l),
0:2
0:2 = 1-(Y2+Y3+···+Yk),
O:n = 1-(Yn+Yl+···+Yk-2),
we may now write Yl + 2Y2 + ... 2Yk-l + Yk == 2 - 0:1 -- 0:2.

0:1 0:2. The inequality to
be proved is
L
'" 2 - 0:1
0:1 -- 0:2
0:2 > 2n(k - 1)
~
cyc1ic
cyc1ic 0:1
0:1 - n - k + 1'
-n-k+1'
where O:j E (0,1) for 1 :s; j :s; n. We observe that 2:7=IO:j = (n - k + 1). Let
us put
1
Zj =
Zj =~,- , Mj = 2 - O:j - O:j+1, 1 :s; j :s; n.
O:jJ
Solutions 303
Then
n n
LJLj L (2 - Cl'j - aj+t)

j=l j=l
2n - 2(n - k + 1) == 2(k - 1).
2:.7=1 Aj = 1 This entails us to use Jensen's
JLj/2(k -1), we have 2:.7=1
Taking Aj = {tj/2(k
inequality for the convex function j(z) = l/z in the interval (0,00). Thus
1 n 1
2:.nn . . ::; LAj~.
j=l AJZ)
AJZJ j=l z)
zJ
Using the definitions of Zj
Zj and Aj,
Aj, this reduces to
2(k-1) 1 n
2:. (2 - aj - aj+1)/aj ::; 2(k _ 1) L
)=1
(2 - aj - aj+1)aj.
This furt her simplifies to

(2 -- aj - aj+t)
~ (2 aj+I) 4(k-1)2
>
~
j=l
a·
J) 22:.7=1 aj - 2:.7=1
22:.7=1 2:.7=1 aj (aj + aj+1)
8(k- 1)2 .
+ 1) -
4(n - k + 2:.7=1
2:. n=l ((aj
j aj + aj+1 )2'
aj+1)2'
we have used
n n
L
2 L aj (aj + aj+t)
aj+I) L + aj+1)2.
= L (aj +
j=l j=l
Thus we need to prove that
8(k - 1)2 2n(k - 1)
----------~--~-------=2
2 > .
4(n-k+1)-2:.7=1 (aj+aj+1) - n-k+1
This in turn is equivalent to
~
~ (aj + aj+l )2 2: -(n
4 - k + 1) 2.
j=l n
However this is a consequence of the Cauchy-Schwarz inequality:
L
n
(aj + aj+t}2 1(
aj+I)2 > ~ f;
n
(aj + aj+1)
)2
j=l
= ~ (2(n - k + 1) f
-4 (n-k+1) 2 .
n
This completes the proof.
304 Solutions
127. We have
L ZkC-
_(n
zn.n_~n n-1
n-l
cn - 11 k-
------"-
Z . - nnC-
- ", zk~n-l-k
1- n~n-1
- n~n -l
zz-~
-~
k=O
n-1
L C-
n-l
1- k (zk _ ~k)
k=O
n-l k-l
{; (z - ~)C-l-k (~Zl~k-l -l).
This gives
n-l k-l
zn - ~n I < Iz - ~I~I (;
~ 1~ln-l-k
1~ln-1-k (~IZlll~lk-l-l)
(~IZlll~lk-1-1)
I z-~
k=O 1=0
n-1
n-l k-1
k -l
< Iz - ~I L rrnn-- 11-- k (L r1r
r1rkk-- 11-- 1t ))
k=O 1=0
n-1
n-l
Iz - ~I L kr"-2
kr n- 2
kk=O
=O
1
~n(n
2 ~lrn-2.
- l)l z -- ~lrn-2.
n (n -l)lz
2
128. It is easy to prove the following identity:
Ilzll-llx
{Ilxii + Ilyll + Ilzll- yll-lly
Ilx + yll - IIY + zll-Ilz
zll - Ilz + xii + Ilx + Y + zll}
{Ilxii + Ilyll + Ilzll + Ilx + Yy + zll}
= rr {llyll
II {IIYII + Ilzll-IIY
Ilzll -IIY + zll}{llxll-lly
zll}{llxll-IIY + zll + Ilx + Yy + zll},
where the product is taken cyclically. But triangle inequality for
für vectors shows
that the right side is non-negative. Hence the result folIows.
follows.
LAjOAHl =
129. Let LAjOAj+l = O'j.
aj. Then 0 < O'j
aj < r.
7f for 1 ::;
<::; < n.
j ::; Using the
concavi ty of sin x
concavity x on (0, r.),
7f) , we have
1 n ( n )
~ ~ sin (a j /2) <::; sin ~ (a j / 2n ) = sin (7f / n ) .
We observe that A j A]+l =

AjAj+l = 2Rsin (O'j/2),
(aj/2), where R is the radius of the circle.
cirde.
This gives
AjAj+l ::;
AjAj+l L
<::; 2nRsin (r./n).
(7f/n).
cyclic
Solutions 305
Equality holds if and only if al == a2 ... = an; i.e., when the polygon is
regular.
Sirnilarly, the total area is
~R2 L
'n
1 1
L;; sin (aj)
n
sin (aj) "2"2 nR2 L;;
2 j=1 j=1
< ~nR2 sin (t, aj/n)
1 2 sin (27r)
"2nR
"2nR -;; .
Equality holds if and only if al == a2 == ...

... == an.
130. We have
j(n - j + 1) (a2j-l - 2a2j + a2j+l) > 0, for 1 :s; j :s; n,

j(n - j)(a2j - 2a2j+l + a2j+2) > 0, for 1 :s; j :s; n - 1.
Adding these and surnrning over j, the resulting surn is non-negative. If aj

denotes the coefficient of aj in this surn, then
a2k+l k(n - k + 1) - 2k(n - k) + (k + l)(n - k)

k(n - k + 1) + (n - k)(l - k)
(n - k) + k = n,
a2k -2k(n - k + 1) + k(n - k) + (k - l)(n - k + 1)
k(n - k) - k(n - k + 1) - (n - k + 1)
-k - (n - k + 1) = -(n + 1).
n(al + a3 + ... + a2n+l) - (n + 1) (a2 + a4 + ... + a2n) ~ 0.
This sirnplifies to
al + a3 + a5 + ... + a2n+1
--------~--------~>
a2 + a4 + a6 + ... + a2n .
n+ 1 - n
Equality holds if and only if each aj is the average of aj_1 and aj+l. Equiva-
lently, the given sequence is an arithrnetic progression.
131. For each k == 1,2,3, ... ,n,

n-I
(n -l)ak =
= LXk+j,
j=1
306 Solutions
where the indices are read modulo n. This gives
rr
k=l
n
ak 11 (n ~ 1 ~ Xk+ j )
> rr rr
n (n-1
k=l j=l
Xk+j
) 1/(n-1)
rr
n
II
k=l
Xk·
Xk·
Equality holds if and only if Xl = X2 = ... = Xn . Equivalently aj's are an

equal.
Suppose n = 3. Then Xl = a2 +a3 - a1, al, X2 = a3 +a1
+al - a2, X3 = a1
al +a2 -a3.
Suppose only two of the xj's are negative, say, Xl < 0 and X2 < O. Then
Xl + X2 < 0, forcing a3 < 0 which is a contradiction. Thus either only one of
the xj's is negative or an three negative. But then X1X2X3 is negative and hence
al a2a3 2': Xl X2X3· This proves the result for n == 3 without any restrictions on
a1
Xj.
Xj'
al = a2 = 1 and a3 = a4 = ... = an = A > O. If the
For n 2': 4, we take a1
inequality holds for this sequence, then
An - 2 (2 - Ar- 2 ( - n + 3 + (n - 2)A)2.
As A -+ 0, we see that
2n - 2 (n _ 3)2 ~ O.
But this is impossible since n > 3.
132. Write z = re iO and b = pe it . Then
j(z) = ar 2 + 2prcos(B + t) + c 2': O.
In particular ar 2 - 2pr + c 2': O. Since this holds for

für an r, the discriminant
nünnegative. We thus get Ibl 2 = p2 ~ ac. Equality
üf the quadratic must be nonnegative.
of
ünly if (ar - p)2 = O. Equivalently r = p/a. Taking Zo = (p/a)e iOO ,
holds if and only
where cos(B o + t) = -1, we get j(zo) = O.
Since p2 ~ ac, we have
1
pr ~ rJ(iC= Va.cr 2 ~ 2"(a+cr 2).
Thus
j(z) = ar 2 + 2pr cos(B + t) + c

< (ar 2 + c) + (a + er 2)
(a + c) (1 + r2 ) •
Solutions 307
133. We start with the observation that it is sufficient to consider the case
~;= 1 Xj
where 2::7=1 = o.
Xj = O. Otherwise we may replace each Xj Xj Xj -- a,
by Xj a, where
a= (~;=1 Xj)
= (2::7=1 Xj) In, without affecting the inequality. Hence we assurne that
2::7=1
~;=1 XjXj = o.
O. Consider the right hand side. We have
n n
LL (Xj -Xk) 2(Xj - Xk)2
n
2n ~X; - 2
(n)2
~Xj
j=1k=1
j=lk= l
"'2
n
= LX;,
2n L X j '
j=1
j=l
as evident from expanding the square, rearranging it and the assumption

O. We also see that
~ Xj = O.
2::
)2
(
n
~EIXj
n
-Xkl 4( L
j<k
J<k
IXj -- xkl)2
4(L IXj Xklr
4(t,(2 j --1-
4(t(2j 1- n)Xj)2
n)Xj) 2
)=1
< 4(t(2
4(I:(2 j -1-n) 2) ( t x ;).
-1-n)2)(I:x;).
)=1
J=l )=1
J=l
Here we have used the Cauchy-Schwarz inequality. Hence it is sufficient to

show that
n
n n(n22 -1)
L(2j n)2 <
L(2j -- 11 _- n)2 ::; n(n 3- 1) .
j=1
j=l -
However, we see that
n
L(2j - 1 -_n)2
~(2j n)2
j=1
j=l
n n
4 ~j2 + 1) ~j
L j2 - 4(n + L j + n(n + 1)2
j=1
j=l j=1
j=l
n (n+1)(2n+1)
+ 1)(2n ( + 1)
+ 1) _-4n+ 1)n(n+1) ( + 1)2
n(n + 1) + n(n )2
44 n(n 6
6
4(n
22 +nn+1
n(n 2 - 1)
3
Moreover, equality holds if and only if equality
This proves the inequality. Moreover,
holds in the Cauchy-Schwarz inequality. This is equivalent to the fact that Xj
is proportional to 2j - 1- n for each j, i.e.; Xj = k(2j --1-
Le.; Xj n) , for 1 ::; j ::; n.
1- n),
Xj = k(l - n) + 2kj or (Xj)
This is same as Xj (Xj) is an arithmetic progression.
308 Solutions
134. Fix any integer n 2: 2 and consider the first n elements in the sequence:
u be that permutation of 1,2,3, ... ,n which orders these
al,a2, ... ,an. Let u
n elements as an increasing sequence:
o ~ ao-(l) ~ ao-(2) ~ ... ~ ao-(n) ~ c.
Then
c > ao-(n) - ao-(l)

n
2: (a(J(j) - a(J(j-l))
ao-(j) - aa-(j-l))
j=2
t
j=2
n 1
>
> ~
1
j=2
u(j) + u(j - 1)
(n - 1)2
>
2:,7=2 (u(j) + u(j - 1)) ,
where the Cauchy-Schwarz inequality has been used in the last step. On the
other hand
n
2: (u(j) + u(j - 1)) 2 (~u(j)) - u(1) - u(n)
j=2
j=2
n(n + 1) - u(l) - u(n)

< n 2 + n - 3,
since u(l) + u(n) 2: 3. Using n 2 +n - 3 ~ n 2 + 2n + 3 = (n - l)(n + 3), we

obtain
(n-l)2 n-l 4
c 2:
c>
- (n-l)(n+3)
=
= -
---
n+3
== 1
1---
---.
n+3
This holds for all values of n. It follows that c 2: 1. (If c < 1, then choose
a large n such that 4/ (n + 3) < 1 - c. This is possible since 1 - c > 0 and
4/(n + 3) can be made arbitrarily small by choosing a sufficiently large n.)
135. Let AD denote the median from A on to BC, and G the centroid. Extend
AD to M such that GD = DM. If m a denotes the length of AD, then GD ==
m a /3 and hence GM == 2m a /3. Note that BG == 2mb/3 and CG == 2m c /3. We
also observe that BMCG is a parallelogram and hence BM = CG = 2m c /3.
Thus GB}.! is a triangle whose sides are 2m a /3, 2mb/3 and 2mcl3. Moreover
[GBM] = 2[GBD] = [ABC]/3. Hence the area of a triangle whose sides are
m a , mb, m c is 6' = (9/4) x [ABC]/3 = 3[ABCl/4.
Solutions 309
A
A
We also have
mb = b/2, 4m~ = a 2 + 4c2 , 4m~ = 4a 2 + c2 .

Hence,
m a mbm c = ~-8 J a 2 +
mambmc
b J c2 ++ 4a .22 2:> ~b-(a . cc++ 2c . 2a) == --
+ 4c22 J 5abc
5~bc.
-8 8 .
We have used the Cauchy-Schwarz inequality here. This shows that
mambmc > 5abc

R' = mambmc 4 = ~R.
4~' - 8 12[ABC] 6
136. Let us take AB = BC = CA = = 1 unit and AF =

= x, BD =
= y, CE = z.
Then using [ABC] = (sinA)/2, we get
1
= "2AF.
[AEF] = 2AF. AE· sinA = x(l
x(l-- z)[ABC].
F
z
B y D c
310 Solutions
[BFD] = y(1 - x) [ABC], [CDE] = z(l- y)[ABC].
Thus, we have to prove

3
1- x(l- z) - y(l- x) - z(l- y) ~
l-x(l-z)-y(l-x)-z(l-y)~ 1 1 1
-~1---~1---~1-
-,.----:-
- --+ + +
+- -----
x(l-z) y(l-x) z(l-y)
Putting u = z(1 - y), v = x(1 - z), w = y(1 - x), we have to show that
( 1 - u - v - w) (~ + ~ + ~) ~ 3.
Expanding, this may be written in an equivalent form:
I-v-w l-w-u l-u-v
----+
u v
+ >6.
w-
However,
I-v-w 1 - x(1 - z) - y(1 - x)
u z(1 - y)
1 - x + xz - Y + yx
z(1 - y)
(l-y)(I-x)+zx
z(1 - y)
1- x x
--+--.
z 1- y
l-w-u l-y y l-u-v l-z z
- - - ==- -+ -- - - - == -----+----.
-+--.
x + -1----z'
--v--- --x-- -------
v l-z' w y I-x
Thus, we need to prove that
I-x x l-y y l-z z
- - + - - + - - + - - + - - + - - > 6.
z l-y x l-z y l-x-
This follows from the AM-GM inequality: alb + bla ~ 2 for positive a, b.
137. Let BP = x and PD = y. The inequality to be proved is:
[APB) + [CPD) + 2J[APB)J[CPD) ::::; [ABCD).
This may be reduced to
2J[APB)J[CPD)::::; [BPC) + [DPA).

Solutions 311
c
D
D
E
x
A B
But observe that

[APD] y
~ [CPB] x
[APB] x' [CPD] y
2J[APB]J[CPD] :::; ::[CPD] + ~[APB].

Yy x
This follows from the AM-GM inequality. Here equality holds if and only if
::[CPD] = ~[APB].
Yy x
If AE and CF are altitudes respectively from A, C on BD, we have
1 1
[CPD] = 2Y · CF, [APB] = 2x· AE.
Thus, equality holds if and only if
Y CF CP
- AE =
;; = - PA·
This is possible only if C P D and AP Bare similar. Equivalently AB is parallel

to CD.
138.
A
B D c
312 Solutions
Let p, PI, P2 be the perimeters of ABC, ABD, ADC respectively. Then

PI + P2 = P + 2AD. Hence
2[ABD] + 2[ADC] =
= 2[ABC] + 2AD.
rl r2 r
Since AD is a median, [ABD] = [ADC] = [ABC]/2. Thus we get
[ABC] + [ABC] == 2[ABC] + 2AD.

rl r2 r
Using AD 2: h a = 2[ABC]/ BC, we obtain
1 1 2 4
-+->-+-.
r1
rl r2 - r BC
139. We observe that 1 + b - cC = a + b+ c + b- c = a + 2b >> O. Hence the

AM-GM inequality gives
(1+b_c)1/3< 1 + 1 + 1 + b - c _ 1 b-c
(1 + b -c ) 1/3 -< l+l+l+b-c_ b- c
- 33 - 1+--.
- + -33 - .
Thus,
a(b - c)
1/3 a+ a(b-c)
a(l+b-c )
a(1+b_c)1/3:::;
:::;a+ 3 .
b(1+c_a)1/3 <
:::; b+ b(c-a)
b(c;a)
3
c(1+a_b)1/3 <
:::;
c+ c(a-b)
c+c(a-b)
3 .
3
Adding, we get the desired result.
140. Using
2b 2 + 2c2 - a2
m2 =-----
a 4
and similar express ions for m~, m~, we get
'
~ " a 2 m 2a ~ L a 22(2b 22 + 2c 22 -- a 2 )
cyclic cyclic
2
~( L a 2
) - ~L a4 .
cyclic cyclic
Solutions 313
Similarly,
16m~m~m~ ~ rr
cyc1ic
(2(a 2 +b2 +c2 )-3a 2 )
~{ _ 4(a 2 + b2 + c2 )3 _ 27a2 b2 c2
+18(a 2 + b2 + c2 ) (a 2 b2 + b2 c2 + c2 a2 ) }.
Thus, the inequality to be proved is
(2 (a 2 + b2 + c2 ) 2 _ 3(a 4 + b4 + c4 ) ) ( a2 + b2 + c2 )
2: -4(a 2 + b2 + C2 )3 - 27a 2 b2 c2 + 18(a 2 + b2 + c2 ) (a 2 b2 + b2 c2 + c2 a2 ).
Simplification reduces this to
L a 2 (a 2 - b2 ) (a 2 - c2 ) 2: O.
cyc1ic
This follows from Schur's inequality.
141. Let 5 denote the sumo Then
n __
= '"
n
X· X.Ln 2
- = -n + '" - - .
n 2
5
L _J
~2-x'
J_
2- x.
j=l
~2-x·
2-x·
J j=l J
Using the Cauchy-Schwarz inequality, we get
2n 2 = (t Vi) :s; (2 ~ (t (2 -
J=l
2 X .)
J J=l
X j )) .
Hence,
n 2 2n 2
'
" "''-
L.-
->
~ 2 - Xj
>2n-l
-
-
X· - 2n
.
- 1
j=l J
Thus,
2n 2 n
5> -n+-- = - - .
- 2n - 1 2n-l
Equality holds if and only if Xl = X2 = ... = X n = l/n. Thus the least value
of S is n/(2n - 1).
142. Let us write
5 _ a + b d
1 - a+b+d
c
-b-+-c-+-d' 52 =
a+b+c
+ - - ..
+ a+c+d
314 Solutions
= 1. Let us introduce a + c == x, b + d == y. Then

We rnay assurne a + b + c + d =
a c a+c-(a 22 +c2 )
SI =--+--
=--+--
1-c 1-a 1 - (a + c) + ac
2ac+ x - x 2
ac + 1- x
S1 2: x and SI == x whenever a == 0 or c == O. Sirnilarly, it is

We observe that SI
easy to see that
2x
SI< - - .
-2-x
- 2-x
(In fact this is equivalent to (a + C)2 2: 4ac.) Thus the set of values of SI is
(x, 2x j (2 - x)]. Sirnilarly, the set ofvalues of S2 is (y, 2y j (2 - y)]. Cornbining,
we get that the set of values of SI + S2 is
(x + y, 2xj(2 - x) + 2yj(2 - y)].

But x +Y = 1 and
~ +-
2x
~ = 2y 4
4 -- 4xy < 2.
4xy <2.
- -+
2- x
-=
2- Y 2 + xy -
Equality holds only if xy == O. Since x i= 0 and y i= 0, it follows that SI +S2 < 2.
Thus, the set of values of the given surn is the interval (1,2).
143. Let Sn denote the surn above. Then
F[ F2 F[ + F2 F2 + F3 Fn - 2 + Fn - l
Sn 2 + 22 + 23 + 24 + ... + 2n
1 1 1 n-2 Y 1 n-l Y
2+4+4 L
j=1
2~ +2 L
j=2
2~
~ + ~S + ~S _ ~ (Fn - l + F n ) _ ~ Fn _ ~
4 4 n 2 n 4 2n - l 2n 2 2n 4
1 3 Fn - l 3Fn
2 + 4 Sn - 2n+l - 2n+2'
It follows that Sn < 2.
144. Let 0:1,0:2,0:3, .. . ,O:n be the roots of P(X) = O. Then we have
0:1 . 0:2 ... O:n = (1 - o:d(l - 0:2) ... (1 - O:n).
Introduce
ß - 1- -O:j
ß j-- ---
j -
,,
O:j
1 :S j :S n.
O:j
Solutions 315
Then aj = 1/(1 + ßj) for 1 :::; j :::; n. Thus

1
a1 . a2'" an
(1 + ßd(l + ß2)'" (1 + ßn)
1
<
2n ..)ß1 . ß2'" ßn
Note that ß1
ß1 .. ß2 ... ßn = 1. Hence it follows that
1
a1 ·a2· ··a n <-.
- 2n
2x + y = x + x + Y 2': 3(X 22y)1/3.

Similarly,
8x 2 + 4xy + 32y 2 = 4(2x 2 ) + 4xy + 32y 2
4 4 ) 1/9
> 9 ( (2x 2 ) (xY) 32y2
18(x 2 y)2/3.
Thus,
3 + 3v'2 = 2x + Y + V8X2 + 4xy + 32y

32y2
2
> 3(x y)1/3 + 3v'2(x y) 1/3

2 2 1/3
(3 + 3..)2)
3V2) (X 2 y)1/3.
It follows that x 2 y :::; 1.
146. We have
' a . ß . ,
4R sm-sm-sm-
r Slll-Slll-Slll-
222
= . a(~
sin 2" ( cos -2-
2R sm (ß-,)
(ß ; ,) -_cos (ß+,))
(ß-2-
; ,) )
2Rsin ~ cos ((ß -,)/2) - 2Rsin2 ~.
This shows that sin(a/2) is a solution of the quadratic equation
((ß -,)/2)X + r = O.
2Rx 22 -- 2Rcos ((ß
Hence the discriminant of the quadratic expression is non-negative. We thus

get,
4R 2 cos 2 ß-,) 2': 8Rr.
(-2-
316 Solutions
This reduces to
cos 2 (a; ,8)ß) >-.
(
0:-
-2-'
-R
2r
147. Observe that

A'B A'I A'C
AB AI - AC'
Hence
A'I A'B + A'C a
AI AB + AC c + b'
A
A
Thus, we get
B
i!X:=~ A' c
AI AI c+ b
AA' AI + I A' a + b+ c
Similarly,
BI c+a CI a+b
= ----,:--
BB' a + b+ c' a+ b+c
CC' - a+b+c
The inequality to be proved is, therefore,
(a + b)(b + c)(c + a) < ~.

1 < (a+b)(b+c)(c+a) 8
-~ < <-
4 (a+b+c)3 - 27
(a+b)(b+c)(c+a) 1 (a+b 8
--------,--'-'--------::---'-
------,--'--'-----::c---'- < - + b+c + -c+a)3
--
(a+b+c)3 -- 27 a+b+c a+b+c a+b+c 27
For getting the other inequality, take x = s - a, y = s - b, z = s - c. Then

x, y, z are positive real numbers and x + y = c, y + z = a, z + x = b. The left
side in equality reduces to
(y + z + 2x)(z + x + 2y)(x + Y + 2z) > 2(x + Y + Z)3.
This can easily be verified after expanding the left side of the inequality.
Solutions 317
148. Observe that

1
LXjXk(Xj
l.::XjXk (Xj +
+Xk)
Xk) = "2 LXjxdxj
"2LXj + Xk)
Xk(Xj+Xk)
j<k j#k
jf.k
LX]Xk'
LXJXk.
j#
We may assume that Xl 2: X2

X2 2: ... 2: X n .. Introducing
Introdueing
Yl = Xl,Y2 = X2,··· ,Yn-2 = Xn-2,Yn-l = Xn-l + Xn ,

'2:,7::11 Yj = 1. Moreover
we see that Yj 2: 0 and I:,7::1
n-l
LY]Yk - LXJXk
LYJYk LX]Xk (Y;-l
(Y; X; _l - x;) LL Xj
- l - X;_l
j#
j# jj#k
# j=l
j=l
n-l
n-l
2X n -1
2Xn -lXXnn LL Xj 2: o.
O.
j=l

LXJXk ~ LY]Yk.
LX]Xk:::; LyIYk.
jjf.k
# jj#k
#
This shows that the number of variables can

ean be reduced
redueed by 1. Continuing by
induetion on the number of variables, we see that the required maximum is the
induction
aß(a + ß) under the condition
maximum of o:ß(o: a+ß
eondition 0: ß = 1. This maximum is 1/4.
Thus the required maximum is 1/4.
149. (Amit Diwadkar) We know that L (coso:

(eosa - COSß)2 > O. Expanding
eosß)2 2:
cyclic
this we get
L
L eos22 0:
cos a 2: L eosa
cos eosß
0: COS ß ..
cyclic cyclic

2
3 L L eosaeosß:::; ( L
coso:cosß ~ L eosa) 2
coso:) ~:::; ~ L
L eosa,
COSO:,
cyclic cyclic cycJ ic
cyclic
beeause
because L eos a ~
coso: :::; 3/2. Thus we obtain
cycJic
cyclic
2 L coso:cosß ~
eosaeosß:::; L coso:.
eosa.
cyclic
cycJic cyclic
cycJic
318 Solutions
U sing 2 cos a cos ß =

= cos( a + ß) + cos( a - ß), this reduces to
L cos(a - ß) :::: 2 L cos a.

cyclic cyclic
Adding L cos a both sides, this further reduces to

cyclic
L
L cos(a - ß) - L
L cos(a + ß) :::: 3 LL cosa.
After expanding, we get
2 L sin a sin ß :::: 3 L cos a.

cyclic cyclic
150. Let Cl,C2,C3, ... ,CN be N positive real numbers. Then for 1 :::: j :::: N, we
Cl,C2,C3, .. . ,CN
have
l/j ClXl + C2X2 + ... + CjXj
( XlX2"' X ) j:::: -,. .
j (Cl C2 ... Cj)
This follows from the AM-GM inequality. We choose Cj to meet our require-
ment. Taking
C' _ (j + 1)j
J -=
Cj jj-1
jj-1 ''
we see that
(Cl C2 ... Cj) ( ~1 . 322 . 43

32
... (j + 1)j) l/j
ji-l
ji-l
IJj
j + 1.
This leads to
l/i ClXl + C2X2 + ... + CjXj

(
Xl X2 ... X j::::
)
j (j + 1) .
Summing over j, we obtain
11
N N j
~
N j )
'"'
L.,..(XlX2''' X j) 1/J
j=l
< L (LCkXk)'e
j=1
({;CkXk
J J +
k=l
j(j+ 1).
N ( N 1 )
LCkXk L j(j + 1)
k=l J=k
Solutions 319
L
N
L CCkXk
kXk LL
N (1J - +1)
-:- - j- .-1
1
k=l )=k
j=k J J+
N
LCkXk
(1k - 1) N
k=l
N
< L Ck;k.
k=l
But we observe that
Ck
k
C:k)k (1 +~) < 3. k
We conclude that
N 1/'
~ (X1X2" 'Xj) ) < 3 ~Xj .
(N )
Here the best constant is e, not 3. This is known as Carleman's inequality.
151. Consider (al - aj )(aj - an), for 1 ~ j ~ n. This product is non-negative

al ~ aj ~ an. Hence we get
since a1
alaj - a1an - a 2j + ajan 2: O.

Summing this over j, w~ obtain
a1(taj) - na 1an- ta;+an(taj) 2:0.

)=1 )=1 )=1
This reduces to
n
o.
nalan + La; ~ O.
na1an
j=l
j=l
Alternate Solution:
Put aj = a1
al + rj for 1 ~ j ~ n. We get 0 = r1~r2~r3~" ·~rn and
rl~r2~r3~·· '~rn
n
nal
na1 + Lrj = O.
j=l
j=l
320 Solutions
Thus
n n
La; L (al + rj)2

j=l j=l
n
11
'"""
L.. (al2 + 2alrj + rj2)
j=l
n n
+ 2a1
nai +2al(Lrj) (2:>j) + L
L rr;;
j=l j=l
n
-na 21 + '"""
L..
~rj'
"""" r j2 •
j=l
Hence we obtain
n n
na1
,"",,2
+ an + L.. aj na1 (al + rn)r
n ) -- nai + L
L r; r;
j=l j=l
-rn (t,r t,r;

-rn (t,r jj )) + t,r;
11
Lrjh - rn)
j=l
< 0,
since rj - r n ::; 0, for 1 ::; j ::; n.
152. This is a tricky problem. We add 3/(1 + abc) to both sides to get an
equivalent inequality
{-a(-l-~-b-) + 1 +labc }+{ b(l ~ c) + 1 +labc }+{ c(l ~ a) + 1 +labc } 2 -1+_6abc '
Observe that
1
-----;----,--
-----;--,-:- + 1 = 1 {I + a b(l + c) }
+ -'------'-
a(l+b) l+abc l+abc a(l+b) l+b .
Thus we get
{-a(-l-~-b-) abc } + {b(l ~ c) + 1 +l

{-a(7"l-~-b:-:-) + 1 +l+labc} abc } + {C(l ~ a) + 1 +labc } ==
+labc}
I+a
-1- {----,-- + b(l + c) + 1 + b + c(l + a) + 1 + c +~_...:..
a(l + b)}
+ -'------"-
l+abc a(l+b) l+b b(l+c) l+c c(l+a) l+a
Solutions 321
1 { ( l+a a(l+b)) (b(I+C) l+b)

1 + abc a(1 + b) + 1 + a + 1 + b + b(1 + c)
+(C(I+a)+
1+C
I+C)}
I+C)}
c(1 + a)
6
> 1 + abc'
abc·
Alternate Solution:
We may write the inequality in the form
2: ---+
--:-1~
1 + 2: - bc
_bc_
-b)> > 33.
a(1 +
.
a(1
cyclic
cychc + b)
b) (1 +
. (1 + b) -- .
cyclic
cychc
This may be rearranged to
'"' (--;--1--:-:-
'" (
L..t.
L. + b) + ~)
1
a(l+b)
a(1
_ab_) > >3
+ a) - 3..
(l+a)
(1
cychc
Equivalently, the inequality to be proved is

Equivalently,
2:
~
cychc
cyclic
1
( ~. 1
-·--+b·
a 11 + bb + b · 1
1 ) >3
1 + (l/a)
(l/a) 23.
- .
If l/a 2
:::: b, then (1 + a)/a 2:::: 1 + band hence
1 1
-- > ----:-~
---;-c-...,.-:-
+ b - 1 + (I/ar
l+b-l+(I/a)"
1
Similarly, 1/ a :s;
:S b implies that
1 1
-- - < --,---,--c-
----,--...,-,-
1 + b - 1 + (I/ar
(l/a)"
Thus, the pairs
(~,b), C:b' I+~I/a))

are oppositely ordered. The rearrangement
re arrangement inequality gives
-1 ·. -
1
-
11 1
---++ bb 1
. ----,--...,-,-
·---
111
+ b. - -
> -. --,---,--c- +b·--
a 1l+b +b + (l/a)
11+(I/a) + (l/a)
a 11+(I/a) +b
1l+b
1 b
--+ --.
l+a + l+b
l+b·
322 Solutions
Thus,
L
~ ( ;:.
1
cyclic
cychc
1 1)
;:. 1 + b + b· 1 + (l/a) > "'-+'"
L1+a+
L....-
cyclic
l+a L l+b
1
L....- 1+b
cyclic
b
~ C~a + l:a)
cychc
3,
which is what we need to prove.
153. Let us put a = xy, b = yz, c = zx, 8 = X + Y + z and t = xyz. Since

x 2 + y2 + Z2 == 2, we have
X2 +(Y_Z)2 2 - 2b,
y2+(Z_X)2 2 - 2c,
Z2+(X_y)2 2 - 2a.
This shows that 1 - a, 1 - b, 1 - c are non-negative reals. Thus we obtain
(1 - a)(l - b)(l - c) ~ o.
2
82 = (x + Y + Z)2
=(X+y+Z)2 = xX2 +y2+
+ y2 + Z2 + 2(xy + yz + zx)
Z2+2(xy+yz+zx)
+ 2(a + b + c).
= 22+2(a+b+c).
Hence we obtain a + b + c == (8 2 - 2}/2. Moreover, we also have
ab+bc+ca=xyz(x+y+z)=ts, abc=t 2 .
We have to show that 8 - t :s 2. We observe that
(1 - a)(l - b)(l - c) 1 - (a + b + c) + ab + bc + ca - abc
- 2
1 -"2
1 (2)2 + t8t8 -- t 2
8 -
4- (8 - t)2 - t 2
2
11 ( 2 t 22
2- 2 8-t ) 2
2-"2(8-t) -2t
-2"
< 2-~(8-tr·
Since (1 - a)(l - b)(l - c) ~ 0, we conclude that (8 - t)2 4 or 18 - 2. :s tl :s
Equality holds if and only if a = 1 or b = 1 or c = 1 and t = O. But t = 0 holds
only if one of the x, y, z is zero. If, for example, x = 0, then yz = 1 giving
y = z = 1. Thus equality holds if and only if one of the x, y, z is zero and the
other two are equal to 1 each.
Solutions 323
154. We consider two cases.

Case 1. Suppose Xn > 1/3. Here we have
1 2
-3 <
- 1- x n <-.
3
Eut, we know that 1- X n = L.7~11 Xj. Hence it follows that
1 n-1 2
3:::; L X jJ <~.
-<""'X"<-.
3-L...- 3
j=l
j=l 3
Thus, the given inequality holds with k = n - 1.

Case 2. Suppose X n :::; 1/3. We take Sk = L.~=1 Xj. If we have
1 2
-3 <
-
sk < -
3'
for some k, then we are through. Otherwise, choose the smallest k such that
Sk 2: 2/3. Note that this is possible since So = = 0 and Sn == 1. The choice of
. . 1 2
th1S least k shows that Sk-1 < 1/3', for otherw1se we would have -3 -< Sk-1 <
- -3
and we would have oUf result. Eut then
2 1 1
Xk = Sk - Sk-1 > 3 - 3 = 3 2: X n ,
a contradiction.
Thus we conclude that the result holds for some k.
155. We solve für z to get

4-xy
z=----
Y + x +xy
This shows that xy :::; 4. We write the given inequality in the form
+ + (44 - xy) (x + y - 1)
xx + yy -- xy 2:
xy 2: zz (x
(x + Yy _- 1)
1) = ( - xy) (x + y - 1) .
=
y + x + xy
We may reduce this to
(x + Y - 2)2 2: xy(x - l)(y - 1).
If (x -l)(y -1) :::; 0, then the inequality is obvious. Suppose (x -l)(y -1) 2: O.
Then we have.
(X+y_2)2 = (x-1+y-1)2
> 4(x
2: - l)(y - 1)
4(x-1)(y-1)
> xy(x - l)(y - 1),
324 Solutions
since 0 ::::: xy ::::: 4.

A Generalisation:
Here is a generalisation due to M. S. Klamkin(CRUX-2000). If 0< a ::::: 1 and
if xy + yz + zx + xyz = 3a + a3, then x + y + z 2: xy + yz + zx. The symmetry
consideration shows that it is sufficient to assume x 2: y 2: z. This implies that
3x + x 3 2: 3a + a 3 2: 3z + z3.
It follows that x 2: a 2: z. Put ,\ = 3a + a3. Then the condition xy + yz + zx +

xyz == 3a + a3 gives
,\ - xz
y
y-- - - - -
- x + z + xz·
Substituting this in the required inequality, it is sufficient to prove that
x 2 (1 + z - z2) + X(Z2 + z - ,\) + z2 - '\z + ,\ 2: O.
Considering the left side as a quadratic expression in x, the inequality is equiv-

alent to the non-positivity of its discriminant. Observe that the discriminant
is given by
D = (Z2 + z - ,\) 2 - 4(Z2 - '\z + '\)(1 + z _ Z2).
Now D ::::: 0 is equivalent to
,\( 4 - ,\) + z(l - z) (2'\(1 - 2z) + z(5z + 3)) 2: O.
Since ,\ ::::: 4 and z ::::: 1, the inequality is true if z ::::: 1/2. If z > 1/2, then
2'\(1 - 2z) + z(5z + 3) 2: 8(1 - 2z) + z(5z + 3) == (1 - z)(8 - 5z) 2: O.
More generally it can be proved that if positive reals x, y, z satisfy xy +

yz + zx + xyz == Cl, then x + y + z 2: xy + yz + zx if and only if 0 < Cl ::::: 4.
(CRUX-2000.) Note that xy < Cl and
Cl -xy
zz=
= ----'-
x + Y + xy
Thus we have to determine positive Cl such that
Cl -xy (x + y )( Cl - xy)
x +y+ > xy + .
x + Y + xy - x + Y + xy
This simplifies to determine all positive Cl such that
f(x, y) == (x + y)2 - (xy)2 + (Cl - xy)(l - x - y) 2: 0,
whenever x, y > 0 and xy < Cl.

Solutions 325
0:/2 and making y approach zero, we see that

Taking x = a/2
a 2
-0:4 +a a)
1- --0:)
+0: ( 1- 2 -
> O.
O.
This implies that a0: ~ 4.

Conversely, suppose a0: ~ 4 and let x, y be positive reals such that xy < a.
0:.
If (x - l)(y - 1) 2'
2': 0, then
f(x,y) == (x - y)2 + (4 - a)xy
o:)xy + (a
(0: - xy)(x -l)(y -1) 2'2': O.
O.
If (x - l)(y - 1) < 0, we may write
f(x,y) = (x + y - 2)2 + (4 o:)xy + (a
(4 -- a)xy (0: - 44 -- xy)(x - l)(y - 1) 2'
2': o.
This proves the result.
156. We may assurne that z is the least number. Suppose 0 < ~ z ~ y < x.
Then
x 2y + y2 z + Z2 X ~ x 2y + xyz + xyz + Z2 y = (x + Z)2 y.
Here equality holds if and only if Z2 y = 0 and y2 z = xyz = z2 X • If y = 0,
we have z = 0 and hence the given inequality is trivially true. Hence we may
r
assurne that y -::f. O. Thus equality holds if and only if z = O. It may be observed
now that
(x + z)2 y = 4 (x; z) (x; z) y ~ 4 (x + ~ + Y 4

27'
We have used the AM-GM inequality in the last step. Here equality holds if
and only if x + z = 2y. Since z = 0, we must have x = 2y. Using x + y + z = 1,
we see that equality holds if and only if y = 1/3 and x = 2/3, and of course
z = O. The case 0 ~ z ~ x ~ y can be dealt in the same way. Thus equality
holds if and only if
= (2/3,1/3,0), (0,2/3,1/3), (1/3,0,2/3).

(x,y,z) =
157. Putting x = pu, y = qv, z = rw, we get an equivalent inequality:

pqr(pu + qv + rw)2 2'
2': pqr{ uv(l- 2r) + vw(l- 2p) + wu(l- 2q)}.
This furt her re duces to
(pu + qv + rw)2 2'

2': uV(l - 2r) + vw(l - 2p) + wu(l - 2q).
Taking 1 - 2p = a, 1 - 2q = band 1 - 2r = C, we see that a, b, c are positive
real numbers such that a + b + C = 1. The above inequality is equivalent to
2
2':4(avw+bwu+cuv).
(a(v+w)+b(w+u)+c(u+v)) 2'4(avw+bwu+cuv).
326 Solutions
Suppose u :::: v:::: w. Then

v(u+w-v) vu + vw - V2
UW + (W - V)(V - U) ~ UW.
Hence
f
r
( a( v + w) + b( w + u) + c( u + v)
r
{(a + b + c)v + aw + b(w + u - v) + cu
{v + aw + b(w + u - v) + cu
> 4v (aw + b( w + u - v) + cu)
4(avw + bv(w + u - v) + cuv)
> 4(avw+bwu+cuv).
We observe that equality holds if and only if v = = w or v == u and v == aw +
= w, then we get v == av + bu + cu == av + (1- a)u and
b(w + u - v) + cu. If v =
= u since a f- 1. Similarly, v == u implies that v == w. We conclude that
hence v =
= v == w. This is equivalent to x / p == y / q == z / r.
equality holds if and only if u =
158. Observe that

n n n
LXj - LXjXjH LXj(1-xj+1)
j=l j=l j=l
"L J=l x; + "L J=l (1- Xj+1)2
<
2
2 "L J J
=l x; - 2 "L =l Xj +n
2
n n
~2 + '"' 2
~ xJ - '~
" ' X J.
j=l j=l
n
<
2'
since x; :::: x j for aB j, in view of the given condition that x j's are from the
interval [0, 1], for 1:::: j :::: n. The quantity "L J=l (Xj-Xjxj+t) is maximum only
when 2:,J=l x; - 2:,J=l Xj = O. Writing this in the form 2:,J=l xj(l- Xj) = 0,
it foBows that Xj = 0 or 1 for each j. Moreover, the condition for equality in
the AM-GM inequality shows that equality holds in the above inequality only
if
Xl = = 1 - X2, X2 == 1 - X3,· .. ,Xn -1 == 1 - Xn , Xn == 1- Xl.
Thus, it follows that the quantity 2:, Xj - "L XjXj+1 is maximum if and only if
each Xj is either 0 or 1 and alternate xj's are equal.
Solutions 327
Case 1: Suppose nn = 2k. This irnplies that Xl = X3 = ... = X2k-1 and

X2 = X4 == ..... = X2k·
X2k. Since each Xj is either 0 or 1, it follows that
Xl = X3 =
= ...
... = X2k-1
X2k-l = 0 and X2 = X4 = '"... = X2k = 1,
or
Xl = X3 = ... = X2k-1
X2k-l = 1 and X2 = X4 = ...
' " ==X2k = O.o.
Thus, we obtain
n n
L X j - LXjXj+1
LXjxj+1 = k = [n/2].
j=l
j=l j=l
j=l
Case 2: Suppose n = 2k + 1. In this case, we have
Xl = X3 = ... = X2k+1
X2k+l = 1 - Xl and X2 = X4 = ... = X2k·
X2k.

1
Xl = X3 = ... = XZk+1 = 2'
But then X2 = X4 == ...
... = XZk
X2k = 1/2 and
n n
'"'
L...-Xj
LXj - '"'
L...-XjXj+l
L "2n
X jXj+1 = 2 -"4nn = "4nn < [n/2].
j=l
j=l j=l
159. (by a Moldovian student)

We first show that
X5 - x2 x5 x2
L L ~(" ,,"
-
x5 + y2 + Z2 2:
cyc\ic cyc\ic
where the surn is taken cyclically over X, y, z. In fact
'"' X 5 _X 2 _'"' X5_X2 x 2 (x 3 _1)2(y2+ z 2)

L...- x5 + y2 + Z2 L...- ~ I " ,,' ~ I " ,,' I e ,,' ,
cyclic cyclic
which is nonnegative. Thus it is sufficient to prove that
X5 _ x2
'"'
L...-. x3 (x 2 + y2 + z2 , >
-
o.
cychc
328 Solutions
However, we have
L x 5 _x 2 1
~, L (x 2 -~)
cyclic cyclic X
> (X 1
1
+ y2 + Z
(x 22 +y2+ 2)
Z2)
L
L
( yz
(xX 2 _yz)
.
2
-
)
cychc
cyclic
1
1 "'"' 2
2(x22 + y22
2(x + z2) L~ (x
(X - y)2
y)
o. Y cychc
cyclic
> O.
160. We define
b'J -_ bjak
1 :S j :S
I
aj = ak, ~-- n.
a·J '
Then, for 1 :S j :S n, we have
a Ij - bjI = ak ( bj ) .
1 - aj
This gives
aJJl -- b'.
a' b'oJJ - (a·JJ - b·)
JJ
= (a JJ - b)
J
(ak - a JJ.) jaJ>-0'
for 1 :S j :S n; this follows from aj - bbjj 2: 0, ak 2: aj, for 1 :S j :S k, and

aj - bj :S 0, ak :S aj, für k < j :S n. Summing over all j, we obtain
n
~aj -
n
~bj -
(n~aj - ~bj
n) 2: o.
Since we are given that 2:.7=1 aj 2: 2:.7=1 bj , we conclude that
n n
nak = Laj 2: Lbj.
j=l j=l
( bI1bI2 ... b'

n
)l/n < 2:.
__ b'
n j -< ak·
-
This implies that
( b1b b3 .. ·bnak
2 n) l/n :S ak·
a1a2a3·· ·a n
b1b2 b3 ·· ·bn :S a1 a 2 a3·· ·a n ·
Solutions 329
161. We introduee elementary symmetrie functions 0"1, O"z, 0"3 in a, b, c: 0"1 = =

= ab + bc + ca; 0"3 == abc. Note that 0"1 2: 3, o"z 2: 3 and 0"3 == l.
a + b + c; 0"2 =
The left side of the inequality is:
2:cyclic(1 + b + c)(l + c + a)
ITcycJic(l + a + b)
Now
L (1 + b + c)(l + c + a)
cycJic
= L (1 + a + b + c + c + ab + bc + ca + c2 )
cyclic
= 3 + 40"1 + 30"2 + L c2
cyclic
= 3 + 40"1 + O"Z + O"i,

and
II (1 + a + b) = 20"1 + O"Z + O"i
O"i + 0"20"1·
cycJic
Thus, the left side may be written in the form
3 + 40"1 + O"Z + O"i

20"1 + 0"2 + O"i + 0"20"1 .
Similarly, the right side of the inequality reduees to
12 + 40"1 + 0"2
9 + 40"1 + 20"2
The required inequality is, thus,
+
3 40"1 0"2 +
~------~2~--~<
+d 12 + 40"1 + O"Z
.
20"1+O"Z 0"1 +
0"20"1 - + 9 + 40"1 + 20"2
Equivalently, one needs to prove
30"i 0"2 + 0"1 O"~ + 60"10"2 - 50"r - O"~ - 240"1 - 30"2 - 27 2: 0.
Consider the quadratic expression:
Q(x) = (30"2 - 5)x 2 + (O"~ + 60"2 - 24)x - (O"~ + 30"2 + 27).

For x = 0, this is negative. For x = 3, the expression reduces to
20"~ + 420"2 - 144,

330 Solutions
which is non-negative because of IJ"2 2> 3. Note that Q(x) == 0 has a positive
and a negative root. Since the leading coefficient is 31J"2 - 5 2 ~ 9 - 5 = 4 > 0,
it follows that Q(x) 20,
~ 0, for all x 2 ~ o.
~ 3, it follows that Q(lJ"d 2
~ 3. Since IJ"1 23,
This proves the required inequality.
Alternate Solution:
We have to prove that
L:cyclic(1 + b ++ c)(l ++ c ++ a) L:Cyclic(2
L:cyclic(1 + + b)(2 ++ c)
L:Cyclic(2 +
IT < .
IT cyclic(l + a ++ b)
cyclic(l + - ITcyclic(2 + a)
ITcyclic(2 +
+ b ++ C, IJ"2 = ab ++ bc ++ ca,
As in the earlier solution, we introduce IJ"1 = a +
IJ"3 = abc = 1. Then we have
L + b ++ c)(l ++ c ++ a)
(1 + 3 + 41J"1 + IJ"2 + IJ" 12 ,
cyclic
cyclic
II(l+a+b) 21J"1 + IJ"2 + IJ"f + IJ"21J"1,

cyclic
cyclic
L(2+b)(2+c) 12+41J"1+1J"2,
cyclic
cyclic
II (2 + a) 9 + 41J"1 + 21J"2.
cyclic
cyclic
The equivalent inequality is

+ 241J"1 ++ 31J"2 ++ 51J"f
27 + + IJ"~ - 31J"flJ"2
51J"~ - 61J"11J"2 + 31J"~1J"2 - IJ"11J"~ ::; o.
We observe that
IJ"l=a+b+c > 3 (abc) 1/3 =3,
> 3(a 2b2c2)1/3=3,
IJ"2=ab+bc+ca
31J"2 =3(ab+bc+ca) < (a+b+c)2 =lJ"i.
=IJ"~.
+ 241J"1 ++ 51J"12 ++ IJ"2 (3 - 61J"1 ++ IJ"2 - 31J"12
27 + ) ::;
- IJ"11J"2 o.
Observe that
+ IJ"2 - 31J"i
3 - 61J"1 + 31J"~ - IJ"1 IJ"2 < 3 - 61J"1 ++ IJ"2 - 31J"f
31J"~ - 31J"1
+ IJ"2 - 31J"i
3 - 91J"1 + 31J"~
< 0,
since IJ"1 2~ 3 and IJ"2 ::; d /3. Thus, using IJ"2 2~ 3, we have
+ 241J"1 ++ 51J"i
27 + + IJ"2 (3 - 61J"1 ++ IJ"2 - 3d - IJ"1 IJ"2)
51J"~ +
+ 61J"1 - 41J"f
::; 36 + + 31J"2
41J"~ - 31J"1 IJ"2 +
+ 61J"1 - 4d - 31J"2 (1
= 36 + (1 -- IJ"1) .
Solutions 331
Since 1 - a1 < 0, we get

36 + 6a1 - 4ai - 3a2(1- a1)
:s: 36 + 6al - 4ai - 9(1 - al) (since a2 2: 3)
= 45 - 3a1 - 4ai
= -(al - 3)(4a1 + 15) :s: 0,
since a1 2: 3.
162. We assurne a1:S:a2:S:a3:S:·· ·:S:an. Put
a == (taj
( t a j )/n
j=l
and bj == aj - a, for 1 :s: j :s: n. Then we observe that
bl1 :S:b2 :S:b 3 :S:·· ·:S:bn ,

:S:b3:S:·· L nbj == O.
j=l
Consider (bI - bj ) (bj - bn ). This product is non-negative since bl1 - bj :s: 0 and
bj - bn :s: O. Summing these products over j, we observe that
n
nb1bn + L b; :s: O.
j=l
Substituting back bj = aj - a and simplifying, we obtain
n
na1an - na(al + an) + La; :s: 0;
j=l
we have used L:7=1 aj 2: n. Using L:;'=l a; 2: n 2, this furt her simplifies to
n :s: a(al + an) - alan·

Suppose an < 2. If laj I :s: 2 for all j :s: n - 1, then
n
n2 <" a 2 < 4n '
-~ J
j=l
and hence n < 4. Thus we may ass urne that laj I > 2 for so me j i- n (and of
course an < 2).
Suppose k is such that a1:S:a2:S:a3:S:·· ·:S:ak < -2 and -2:S: ak+1 :s: ak+2 :s:
... :s: an < 2. Since a1 + an :s: 0 and a 2: 1, we have
n :s: a1 + an - a1 an < al + 2 - 2a1 = -al + 2.
332 Solutions
Thus, we obtain < - n + 2. We further have

al
n < al + a2 + a3 + ... + ak + ak+l + ... + an

< al+a2+ a 3+···+ a A,+2(n-k),
and hence al + a2 + a3 + ... + ak > 2k - n. This estimate leads to
2k - n < al + a2 + a3 + ... + ak
<
< - n + 2 + a2 + a3 + ... + ak
< -n + 2 - 2k + 2.
It follows that k < 1. But this contradicts laj I > 2 for at least one j -I- n.
Thus, an < 2 forces n < 4. \Ve conclude that for n ::::: 4, the largest number
among aj 's must exceed 2.
163. Since Xj are positive integers, xj+! - Xj ::::: 1 for 1 :::: j :::: n. This forces
JXj+! - Xj :::: Xj+1
JXj+1 xj+! - Xj, 1:::: j :::: n. Thus
L
n+l
VXj+! - Xj
< L
n+!
n+1
Xj+l - Xj
)=1 Xj+l j=l Xj+!
X2 - Xl X3 - X2 Xn+l - X n
---+ + ... + - - - -
X2 X3 Xn
< 1+ (~ + ~ + ... + ~)
2 3 X2
+(_1_ + ... +~)

X2 +1 X3
+ ... + (_1_ + ... + _1_)

Xn +1 Xn+l
1 1 1
< 1 + - + + ... + - - .
2 3 Xn+l
\Ve prove the inequality by induction on n. Für n == 1, we have
-'----- VX2
J X 2 - Xl < - - 1
= -- < 1.
X2 X2 VX2
Suppose
L L
n (n-l)2
JXj+l -Xj < ~.
j=l Xj+l )=1
j=l J
If Xx n +! :::: n
n 2, then the abüve estimate proves the inequality:
L
n+l
JXj+l - Xj
< 1+-+-+···+--
1 1 1
j=l Xj+! 2 3 Xn+l
1 1 1
< 11 ++ "2- ++ "3- ++ ...
... +
+-.
n2'
2 3 n2
Solutions 333
Suppose Xn+l > n 2 •. Then we have

y'X n +1 - X n
y'Xn+l 1 1 1 1
-~---- < ----- < -.
Xn+l
X n +1 y'Xn+l
y'Xn +1 n
On the other hand,
1 1 1 2n - 1 1
-----,-- + + ... + -;-2 > --.22 - > -.
(n - 1)2 +1 (n - 1)2 +2 n n -- n
Thus, we get
y'x n +1 - X Xn 1 1 1
--'---'-'--'----'-
Xn+l
< + (n - 1)2
(n - 1)2 + 1 + (n-1)2+2
< (n-1)2+1 +2 +
+ ...
... +
+ --
n 22 ··
X n +1
Using the induction hypothesis, we now obtain
L
n+1
j=l
y'Xj+1 -
Xj+1
Xj
(t j=l
y'Xj+1 - Xj)
Xj+l
Xj+1
+V X n+1
y'x n +1 -
X n +1
Xn+l
Xn
(n_l)2
(n_1)2 1 n2 1
< L -:+ L -:
j=l J j=(n-l)2+1
j=(n-1)2+1 J
n2
L;·J
j=l
This completes the proof by induction.
164. Consider the quadratic expression

1
a(b + C)2 - a2 (b + c) + 3a3 - 3,
in b + c. Its discriminant is
D = a4- (1 3 ) = 3a ( 36 - a3) < 0,

3a 3- 3 )
4a "3a
and hence
1
a(b + C)2 - a 2(b + c) + "3a3
3a3 - 3> 0,
for all values for b + c. Using abc == 1, this may be written in the form
a{ (b + C)2 - a(b + c) + ~a2 - 3bc} > O.
Since a > 0, this reduces to

b2 + c2 - a(b + c) + ~a2 - bc > o.
It follows that
~a2 < a 2 + b2 + c2 - ab - bc - ca.
3
334 Solutions
165. Suppose ß1,ß2,'" ,ßn be any n real numbers. Then the Cauehy-Sehwarz
inequality gives
n
LAjß; ( t : ) (tAjß;)
j=l J=l J J=l
> (~ßjr
We may assume LJ=l
L}=l IZj 1 i O. Taking
ßßJ --~
- IZjl
L}=l 1
J - ",n
L..j=l Zj I'
IZjl'
t, (t,
we obtain
n (n)2
Ajlzjl2 --
~ AjlZjl2 IZj1r 2:2 O.
~ IZjl
However, triangle inequality gives 1t, I :s ~t,

I~ Zj 1 :S IZj I· Thus we obtain
2
n n
~Zj :S ~AjIZjI2.
1
I
Remark: For n == 2, this is known as Bohr's inequality.
166. Consider the monie polynomial whose roots are a, b, c:

p(x) (x - a)(x - b)(x - c)
x3 - ( La) x 2 + ( L ab) x - abc.
cyclic cyclic
We know that the roots Ofpl(X) == 0 lie between min{a,b,c} and max{a,b,c}.
But pI (x) == 3x 2 - 2 ( La) x + ( L ab) whose zeros are
L
cyclic
a+ ,I L
cyclic
a2 - L
cyclic
ab L
cycIic
a- JL cycIic
a2 -- L ab
cyclic
3 3
Sinee a, b, c are distinet, LCYclic a2 - LCYcJic ab > 0, and henee
1/2
3 min { a, b, c} < L a- (
L a2 - L ab )
1/2
<
<L a+ (
L a2 - L ab ) < 3 max { a, b, c}.
cyclic cyclic cycJic
Solutions 335
167. We have
(a - b)(b - e)(e - a) ab 2 - a 2b + be2 - b2e + ea 2 - e2 a

1 1 1
a b e
a 2 b2 e2
1 1 1 ,22
(a-b)2(b-e)2(e-a)2= a
(a-b)2(b-e)2(e-a)2=la b e
a2 b2 e2
But for any two matrices A and B, we have det(AB) = det(A) det(B) and
det(A) = det(At). This implies
1 1 1 ,22
(a - b)2(b - e)2(e - a)2 = laa b e
a 2 b2 e2 I
1 1 1 1 a a2
a b e 1 b b2
a 2 b2 e2 1 e e2
(:,G ,;),~) (lG a')

1 a
b b b2
b2 e e2
Introducing 8k
Sk = ak + bk + ek for k 2: 1, the above relation simplifies to
(a - b)2(b - e)2(c _ a)2
. 3 81
81 S2
82
= 81 82
=18182 S3
83
S2
82 83 84
= 84(382 - si)
8i) -
- 8d3s
8d3833 - 8182) + 82(S3S1
82(8381 -- 8~).
s~).
Since s3 = 0, we get
83 =
(a-b)2(b-e)2(e-a)2 =s4(382 -si) -s~.
This shows that

s~ ::; 84 (3s 2 - si) .
However,
3s2 8i = 3(a 2 + b2 + e
382 -- si c2) - (a + b + e)2
C)2 = L (b - e)2.
C)2.
cyclic
336 Solutions
Thus, we obtain
( L a2) 3 :::; ( L (b - L a4 .
C)2 ) ( )
168. We may easily verify this for n = = 1 and 2. We assurne n ~ 3. Put

Re(z) = = rcosB, where Re(z) denotes the real part of z. Then 1" > 1 and
0:::; B:::; 1f/2. We have
Iz n+l __
I znl2
Zn 12 r 2n +2 + r 2n _ 2r 2n+1 COS B,
I z n+l _ 11 2 2 + 1 - 2r n
2n +2
r 2n n +1 cos(n + l)B.
Thus, we have to show that
r 2n +2 + r 2n _ 2r 2n +1 cosB < r 2n +2 + 1 - 2r n+1 cos(n + 1)B.
This is equivalent to the relation
r 2n - 1 < 2r 2n +1 cosB - 2r n+1 cos(n + 1)8.
Here we consider two cases.

= 1 + d, where
Case 1. Suppose r =
d>
d> 22 , n 2: 3.
n(n _
-- n(n 2) , n 2: 3.
- 2)
Then
r 2 - 1 = 2d + d2 :::; (n - 1)2d2 = (1 + nd - r)2 < (rn - r)2.

Thus we get in this case r 2n - 2r n+1 + 1 > 0 and hence
2n +1 cosB - 2r n+1 cos(n + l)B
2r 2n > 2r 2n+1 cosB - 2r n+1
> 2r 2n _ 2r n
2n _ n +1
> r 2n - 1.
Case 2. Suppose
d< 22
d< n(n - 2)
Since r cos B > 1, we have
2 1 B2 B4
1- < 1- d < - < cos B < 1 - - + -.
n(n - 2) r 2 24
Solutions 337
This implies that

48
84 -_ 1282 + 48 > O.
n(n - 2)
Since 82 < 6, we get
82 :S 6 - 6VD,
where D = 1 - (4
3n n - 2,
2
)' If n
\' ~ 4, it is easy to check that
8n 4 + 10n 3 - 87n 2 - 50n - 16 > 0,
and hence
6
1-, .,~
1- (n+2)2 <
<VD.
VD.
This gives
2 Ir> 36 47f 2
8 :S6-6VD<
:S6-6vD< (n+2)2 «n+
< (n+2)2' '
Thus we obtain (n + 2)0 < 27f. Hence 37f/2 < (n + 1)8 :S 27f and this implies
2)8<
that cos(n + 1)8 < cos(27f - 8) = cos8. Obviously, this holds for n = 3 also,
since 0 :S 8 :S 7f /2. Thus we get
2r 2n +1 cos 8 - 2r n +1 cos( n + 1)8 ~ 2r 2n - 2r n > r 2n - 1.
169. Since the express ions are symmetric in a, b, c, we may assume that a ~
b ~ c > O. Note that x > 0 and z > O. If y > 0, then the inequality is
equivalent to
1 1 1 1 1 1
-x + -y + -z-a
> - + - + -.
b c
(169.1)
Eut using the AM-HM inequality, we see that

1 1 1 1 2
-Y + -z =
b+c-a
+ c+a-b > -.
- c
Similarly, we can prove that
1 1 2 1 1 2
-+->-
x Y - b'
-z + -x >-.
- a
Adding these we get (169.1).

If y = 0, then the inequality reduces to abczx ~ 0 which is true. If y < 0,
then the inequality is equivalent to
1 1 1 1 1 1
- + - + - < - + - + -. (169.2)
x Y z - abc
Since z ~ c and x ~ a, we have
1 1 1 1
-<-
- , -<-
- ,
z c x a
338 Solutions
and
1 1
11 < 0 < b'
Adding these we get (169.2).
170. The left hand side is equal to
L a 3 (b + c)
cyclic b3 + c3 '
and this may be written as
Lcyclic a 3 (a 3 + b3) (a 3 + c3) (b + c)

TIcycliC (b 3 + c3 )
However, we have
a3 (b + c) (a 3 + b3) (a 3 + c3)
= a3 (b+c) {a 3 ( ~ a3) +b3C3 }

cychc
cj { a'
(b+ c) (,~, a') + a'b'c'
{a\~, a'b'e' }
}
= (L a) ( L a 3)a 66 + ( L a)a 3b3c3 - a7 ( L a3) - a(a 3b3c3).

Thus, we obtain
I: (I: a)( I:
L a3(a3+b3)(a3+c3)(b+c) = (L L a3)( I:
L a6)
(I: a3)( I:
- (L L a7) +2( I:
L a)a 3b3c3.
cyc1ic cyclic cyclic
On the other hand, we also have
II W+ ( I: a3)(
W+ c3c3) = 2a 3b3c3 + (L ) ( I:
L a6) - I:
L a9.
Solutions 339
Thus, the inequality to be proved is
( (c~c a3)(c~c a
c~c a) {(c~c a) (c~c 3
) (
6
)
( L a3)( L aa +2(
- (L + 2( L
3 )(a)aa3b3c
La) 7) 3 3 C3 }
(,~, ab) {2a'b'h

3(~:b)
:>" 3 (,~:')(
{2a'b'c' + (~c ,~, a') - ,~, a'}
a3)(c~, a' }
{(c~c
{
(L
cyclic
ar - c~cL ab}ab} {{2a2a3bb3
33
cyclic
3 3 c33
c +( L a
+ (c~c 3 c~c
cyclic
a3 ))(( L aa6 ))
cyclic
6 -
-
L aa9}
c~c
cyc1ic
9
}
(c~c
+ (c~ca){
a) {(c~ca)(c~ca9)
a) ( a9 ) (c~c c~c - (c~c
- a c~c a ~~o.0.
(c~ca3)(c~ca7)} 3
) (
7
) }
We observe the following:
(L af -3-3 L ab=
( L af
cyclic
L (a - b)2 ~O,
ab = ~ L(a-b)2 ~ 0,
cyclic cyclic
2a 3b3cc3+ ( L a3)( L a
3 3 3 3) ( 6) - L aa9> 0,
9
cyclic cyclic
cyclic cyclic
cyclic
and
L aa3)( L aa7)
La)a)(( L aa9) - (L
((L 9) - ( 3) ( 7)
cyclic cyclic
cyclic
L ab7(b _aa2) + 2:
L ba (a _bb
7 2 - 2) 7 2 - 2)
cyclic cyclic
L ab(b a (b a2) 6 - 6) 2 - 2)
cyclic
L ab(b 2- af (b + ba2+ a
2 - 2) 2 4 2 2 4)
cyclic
> o.
This proves the given inequality. Equality holds if and only if a = b = c.
171. We first prove that it is sufficient to consider the case where all the aj's aj's
are positive. Suppose, for example, an = = O.o. Then
a=
a=
ai + a~ + ... + a;_l
ai a;_l
n
340 Solutions
If we set
a211 + aa2222 + ...
a + aa2n - 11
b=
n-1
then the inequality for (n - 1) numbers al, ... ,a n -l
a1, a2, ... -1 is
n-l
~~ > (n-1)b
L~
L...
. 11-- a 2 - 1 - b2 ·.
J=1
j=l J J-
Note that
a> - 1 = } b = a·
a>-~b=a' ~ >>a
- --
- a>>-.
1.
-
-y'3
-y13 n-
-11--y
y1 3
'3
Thus, we need to prove
(n - l)b > ~
(n-1)b na
-'---'-
1- b
> - -.
1 - a2 '
1- b2 - 1- a
JV
-
Using b = a·
a . n ,this reduces to
n-1
}n(n - 1)
Jn(n n
na 2 ~
> - - 22 '
1-
-l-a a ·
1
1---:--~
---
(n - 1)
Simplification leads to
1 _ a22 < n 22 + Jn(n

}n(n - 1)3
1 - a:S
- n 33 - (n
( - 1)3
n-1 ) 3 '.
Since a 2 2:
~ 1/3 implies 1 - a 2 :S
::; 2/3, it is sufficient to prove
n 2 +}n(n-1)3
-2 < --::---''-::-.'-.--,--::-~
3 - 3 n: - (n - 1)3
A further simplification re duces this to

reduces
n2 - 2n + ~ ::; }n(n - 1)3.
Finally we observe
2
n2 - 2n + 3" < (n - 1)2 < }n(n - 1)3,
proving our claim.

The above argument shows that we can discard all those aj 's which are
=I 0
equal to zero without affecting the inequality. We henceforth assume aj -=I-
for 1::::;
far 1 ::; j ::::;
::; n.
Solutions 341
""_J
n
"" _J
~ 2
a2
j=l na a'(l-
J=l J
1
> ~----,::-c-)'
aJ2 ) - a(l- a 2 )'
1
Setting Wj = aUna2 for 1 ::; j ::; n, we see that Wj > 0, for 1 ::; j ::; n and
'L,7=1 Wj = 1. Consider the function
'L,7=1
1
ff ((x) = x ((1 _ x 2)' for 0 < x < 1.
x 1- x 2 )'
It is easy to check that r(x) ;::: 0, for 0 < x < 1. Thus f(x) is a convex
function on the interval (0,1). Now Jensen's inequality gives
f,Wjf(a j );::: f(f,wja j ).
This takes the form 1

n I l
~ > )
"~waj-('L,j=l
n 1
Wj aj (1- aJ) ->
2::W· 2j
j=l Ja J (l-a ) 'L,7=1 Wjaj
0J=1 J
n w.a.
- ( 'L,7=1 Wja
J J
j )
3'
n
~ Wjaj -
( n
~ Wjaj
)3 ::; a(l- a2).
Taking ,\ = a;,
'L,7=1 a;, this reduces to
= 'L,7=1
,\3 _ n 22a 4 ,\ + n 33 a7 (1 - a 22)) ;:::

;::: O.
But using the factorisation
,\3 _ n 2 a 4 ,\ + n 33 a 7 (1 _ a 22 )) = (,\ _ na 33 )) (,\2 + na33 ,\,\ + n 2 a 4 (1 _ a 22 )),

)),
and the observation that for a 2 < 1 and ,\ ;::: 0, the expression ,\2 + na33 ,\ ,\ +
n 2a 4 (1
(1 -- a 22 )) remains positive, it suffices to prove that ,\ - na 33 ;:::
;::: O. Thus we
need to prove that
~n
La];:::
3 ( n a2)3~
n (~a2)3/2
L--1.
~aj ;::: n ~--1.
j=1
j=1 .
J=l
j=l nn
This reduces to the form
n( 2::n a1 )2 ;: : (n2:: a; )3
J=l J=l
342 Solutions
However, this follows from Hölder's inequality with exponents p = 3/2 and
q = 3:
n
~a; ~
(n)
~a;
2/3
n 1/ 3 .
= a2 == ...
Equality holds if and only if aj = ... == an.
172. The given condition on a, b, c shows that at least one of a, b, c cannot
:S 1. Then
exceed 1; say a ~
ab + bc + ca - abc = a(b + c) + bc(l - a) 2 o.

Here equality holds if and only if a(b + c) = = bc(l - a) == o.O. If a == 1, then
b+c = = 0 forcing b == c == 0 which is impossible in view of the condition
a22 + b22 + c2 + abc = 4. Thus 1 - a =I- =J 0, and only one of b, c can be equal to
zero. Suppose, for example, b = O. Then c cannot be zero and ca = O. This
implies that a = = 0 and hence c2 == 4. But then c == 2. Thus the equality holds
on the left side if and only if (a, b, c) = (0,0,2), (0,2,0), (2,0,0).
To prove the right side inequality, we show that the given condition forces
the existence of a tri angle with angles a,ß" such that a = 2sin(a/2), b =
2sin(ß/2), c = 2sinb/2). In fact, if a,ß" are the angles of a triangle, then
L sin 2 (a/2)
1
"2~ L 2sin 2 (a/2)
cyclic cyclic
1
"22" L (1 - cosa)
cyclic
3 1
"22" -"2
- 2" L cosa
cyclic
"22"3-"2
- 2"1( 1 + 4 II sm(a/2)
.)
cyclic
1- 2 II sin(a/2).
cyclic
Thus we obtain
L sin 2 (a/2) +2 II sin(a/2) = 1.
cyclic cyclic
Conversely, suppose there are non-negative real numbers x, y, z such that x 2 +

y2 + z2 + 2xyz = 1. We show that there is a triangle with angles a, ß" such
that x = sin(a/2), y = sin(ß/2), z = sinb/2). In fact solving for x, we obtain
x = -yz + J(1- y2) (1-

)(1- (1 - z2).
Solutions 343
Putting y = sinv, Z = sinw, 0 < V,w < 90°, we get

= -- sin v sin w + cos v cos w == cos( v + w).
x =
Let ß = 2v, "Y = 2w and a0: = 1f - ß - T We observe that 1 > y2 + Z2 =

sin 2 (ß/2) + sin 2 b/2) and hence cos
cos 2 (ß/2)
(ß/2) > sin 2 b/2). Since 0 < ß/2,
ß/2, "Y/2 <
90°, we must have cos(ß/2)
cos(ß/2) > sinb/2) = cos(1f/2 - "Y/2). Thus it follows that
ß ß
ß/2 < 1f /2 - "Y /2. Hence ß + "Y < 1f. The definition of a0: shows that a, ß,
0:, ß, "Yare
the angles of a triangle and x = = cos(v + w) == sin(a/2), = sinv == sin(ß/2),
sin(0:/2), y =
Z =
= sinw = sinb/2).
Coming back to the problem, since a 2 + b2 + c2 + abc = 4, there is a triangle
with angles a,0:, ß, "Y such that
= 2sin(a/2),
a= 2 sin(0:/2) , = 2 sin(ß/2) ,
b= c = 2sinb/2).
Then
ab 2 (( sin a
< sinatan(ß/2)
0:/2)
0: sin ßtan( a/2) tan(ß /2) )
sin0: tan(ß/2) + sinßtan(a/2)
sinßtan(0:/2)
r/1/22
·. 0: cot (a+"Y).
smacot
sm (0:
-2 +- . ßcot
"Y) + sm ß (ß+"Y)
(ß- 2+ -"Y) .
Similarly
bc < sm
sin ß (ß ;
· ßcot (ß+a). sin "Y cot ("Y+
0:) + sm
-2- ("Y
-2; a
0: ))
-
ca < sm . ("Y+ß).
· "Y cot ("Y+ß).
-2-
- +smo:cot (a+ß)
2 - +smacot -2- . (O:+ß)
Adding these we obtain
ab + bc + ca < (sino: + sinß) cot (0:; ,8)

~ (sina
cychc
(a;
2 "~
~ cos (a-ß)
cychc
(0:
-2 ; - cos ß) (a+
(0:;
-2- ß))
ß
2 L cosa
coso:
cyclic
6 - 4( L Sin2(a/2))
2
(0:/2))
cychc
= 6- (a 2 + b2 + c2 )
2 + abc.
344 Solutions
173. We may assume that a 2 , b2 , c2 are all distinct and abc 1=

=I- 0. Put a = reic>
reiO:,,
b=
= reiß, and c = re h . Then the inequality reduces to
= reh.
Icosec (0: - ß)I + Icosec (ß -1')1

- ,)1 + Icosec b 0:)1 2: 2V3.
b -- 0:)12: 2v'3.
Now the AM-GM inequality gives
L
cycJic
cyclic
Icosec ß)I2:
(0: - ß)
Icosec(o: I 2: 3311 rr cosec
II
cyclic
cycl ic
ß)
(0: - ß)1
cosec(o:
1/3
11 / 3 ..
Putting A =
= 0: - ß, B = -" C =
= ß -1', - ,) =
= 7r - (0: -1') = 7r - (A + B), This takes the
form
L
cyclic
cycJic
Icosec (0: - ß)
Icosec(o: ß) I12: 3 1I rr cosecA
II cosec A cosecB
cycJic
cyclic
cosec B cosecC
cosec C 1
1/3
,,
I II
1 rr
cychc
cosec A cosecB
cosecA 2:
cosec B cosecCI2:
cosec Cl 3~·
3~'
The function f(t) = log 11sin tl is a concave function and Jensen's inequality
gives
f(A) + f(B) + f(C) :S::; 3f ( A+B+C)

3 3 log ( sm(7r/3)
= 3log .
sin(7r/3) ).
This implies that
I Cl : ;
log 1 sin A sin B sin Cl :S 33log
log ( ~) .
(~).
Hence
I II
1 rr
cycJic
cychc
cosec A cosec
co sec B cosec Cl 2: Cl 3~'
is true.
174. (a):
Observe that x/(l - yz) 2: x, y/(l - zx) 2: y, and z/(l
z /(l - xy)
XV) 2: z. Thus, we
get
L 1 ~ z 2: x +Y+ z 2: x 2 + y2 + Z2 = l.
cycJic Y
Equality holds if and only if (x,y,z) == (1,0,0), (0,1,0) or (0,0,1).
(0,0,1) .
On the other hand,
z2) == 1 _ (1 ~
> 1 _ (y2 ;+ Z2)
1 _ yz 2: - x 2) == 1 ~
+ x2 ,
- 2 2 2'
Solutions 345
which gives
x 2x
1-
I
<--
- yz -:::; 1 + x 2 '
However, the inequality
-
2x
2x
-<
3V3
3V3
- - (l+x 22)
11 +
+ x2
x 2 ::;
- 8 (1 + x )
-8-
holds for all real x. In fact
+ X 22)2
(1 +x 16V3
)2 _- - -x x ~ (9x 4 + 18x 2 + 9 - 16V3 x)
9
1 2
g(V3x-1) (3x 2 +2V3x+9) 20.
Equality holds if and only if x == 1/ V3. It follows that
""
~ _x_ < ,,~3V3(
_x_
~ 1 - yzz -- ~
cyclic
cychc Y
" ~
~ 1 + xx 2 ::;
cyclic
-
3V3(3+
< -8- " x2 ) =
8 3+ ~ x
cyclic
L 2) 3V3
3V3.
2 .
2
= Yy == z == 1/V3.
Equality holds if and only if x =
(b ):
Using x 3 - 3x + 2 = (x - 1)2(x + 2) 20, the inequality
1
+ xyz
xx+xyz <
::; + 2"X(y2 + Z2)
xX+2"X(y2+
x+
X ~x(l
~X(l
2
- x 2)
1
2" (3x - x 3 ) :::;
::; 1
is obtained. This implies
X x2 •
"~ -
1 +-
cyclic
yz = "~ x
cyclic
+ xyz >
- "~ x = 1,
2
cyclic
which proves the first inequality in (b).

To prove the right side inequality in (b), we may assurne that x ::;
:::; z and
:::; z. Thus
y ::;
" __ x_ < _x_+------"-y_+_z
_x_ _x_+---=y_+_z
~ 1 + yz -
~l' 1 + xy ,
cychc
CYCIC
x + Y + z ::;:::; .J2 (1 + xy) ,

346 Solutions
under the condition that x :::; z, y :::; z and X2 + y2 + Z2 == 1. This is equivalent

to the inequality
J1 - X2 - y2 :::; J2(1 + xy) - x - y.
Since both sides are non-negative, this is furt her equivalent to
1- x 2 - y2 :::; (J2 (1 + xy) _ x _ y) 2 .
Setting x + y = 0, xy = ß, the inequality to be proved is
1- 0 2 + 2ß :::; (J2(1 + ß) - 0 f·
This simplifies to
( J20 - ß- 1) 2 + ß2 ~ 0.
Since the left side is the sum of squares of two real numbers, the result follows.
We also observe that the case of equality occurs only when ß == and V20 -
°
,8 - 1 == 0. This corresponds to xy == and x + y == 1/V2. We conclude that
°
(x, y, z) == (0, 1/J2, 1/J2) or (1/J2, 0, 1/J2).
By symmetry, we get one more case of equality:
(x, y, z) = (1/J2, 1/J2, 0).

175. Introduce new variables a, b, cC by
1 1 1
= a+ 3'
xx=a+- Y = b+- C+-3·
z -- c+-
3' 3'
Then a + b + c == 0, 3a
re duces to
+ 1 ~ 0, 3b + 1 ~ °and 3c + 1 ~ 0. The inequality
L ab:::; 3 L ab + L a
2 2,
see that b = c = °
We may assurne a to be the largest among a, b, c. Then a > 0. (If a = 0, we
and the inequality is, in fact, an equality.) Substituting
b == -a - c, and after some simplification, an equivalent inequality is obtained:
a 3 + (3c+1)a 2 +ca+ (C 2 _C 3 ) ~o.
If c ~ 0, this is true in view of the condition c < 1 since z :::; 1. Hence we may
assurne that c < 0. For a fixed c, consider the function
f(a) = a 3 + (3c + 1)a 2 + ca + (c 2 - c3 ).

Solutions 347
Its derivative is f'(a) = 3a 2 + 2(3e + 1) + e. Setting the derivative equal to

zero, two values of aare easily computed:
- (3e + 1) ± y'ge2 + 3e + 1
a = --
' -' - - - -
' -' - -
33-- - - - .
Since a ~ 0, the positive square root has to be chosen. Using 3e + 1 ~ 0, it

may be deduced that f(a) has an extremum at
- (3e + 1) + vge
y'ge2 + 3e + 1
ac = 3 .
The second derivative f"(a) = 2(3a + 3e + 1) > 0, since 3e + 1 ~ and a > 0.

Thus the function f(a) has minimum at a = a cc .. We show that f(a c ) ~ 0. An
°
involved computation gives
f (ac) = 217 { 27e3 + 54e2 + ge + 2 - 2 (ge 2 + 3e + 1) J ge2 + 3e + 1 }.

27e3 + 54e2 + ge + 2 - 2 (ge 2 + 3e + 1) J ge2 + 3e + 1 ~ 0,

:s
for -(1/3) e:S 0. Using 27e3 + 54e2 + ge+ 2 = 27e2 + (ge 2 + 3e+ 1) (3e+ 2)
and rationalisation process, the inequality reduces to
(ge 2 +3c+1)(-27c
+3e+1)(-27e2 ) 2
-::----::__ + 27e > 0.
~--::::--:::--;~==:;::=======
:::--;~==:;;===:=
3e + 2 + 2y'ge2 +3c+1
3e+2+2J9c + 3e + 1 -
This may be seen to be equivalent to
2 J 9c 2 + 3e + 1 ~ 9c 2 - 1.
But observe that

9c
ge2 - 1 == (3e + 1) (3c
(3e - 1) :s 0,
whereas the left side is non-negative.
176. From the symmetry, it may be assumed that a :s b :s e :s d. Suppose

a ~ 1/8. Then
(48a - 4) (a 2 + 1)2 - 125a 2 = (2a - 1)2 (12a 3 + lla2 + 32a - 4) ~ O.

a2 48a - 4
----,0-<---
--~<---
(a2 + 1)2 125
Summing over a, b, c,
e, d, we get
I:
I:. (a2 aa 22
:s
48
48
< -(a+b+e+d) -
16
-25
16
(a + 1) - 125 (a + b + c + d) -
+ 16 -=25'
cyclic 1) 125 22 -.
16
cychc
2 25 25
348 Solutions
Suppose a < 1/8. In this case, we observe that for x ::::: 0,
(540x + 108) (x 2 + 1)2 - 2197x 2

(3x - 2)2 (60x 3 + 92x 2 + 216x + 27) ::::: 0,
which gives
~x2 540x + 108
< 540x+
(x2 + 1)2 - 2197
b2 c2 d2 540(b + c + d) + 324
--+ +-- <
(b2 + 1)2 + (c2 + 1)2 + (d2 + 1)2 2197
108 540a
169 - 2197'
and
a 22 2 a 540a
--~ < a < - < --.
<a<-<-
(a2 + 1)2 8 2197·
2197
We thus obtain
L
'"
a2 108 16
~ (a2 + 1)2 < -<
< 169 -
< 25·
177. We show that 4

a a"3
::::: 4 4 4 • (177.1)
v'a 2 + 8bc a"3 + b"3 + c"3
This is equivalent to the inequality
(a"3
4+ b"34+ c"3)42 ::::: a"322
22
(a + 8bc). (177.2)
However we observe that
(a"34+b"34+c"34)2 - (4)2
(
4 4
a"3 (b"34+c"34) (4
a"3 +a"34+b"34+c"34)
2 2 2 1 1
> 2b 3c 3 . 4a"3b 3 c 3
2
8a 3 bc,
using the AM-GM inequality. This in turn gives the inequality (177.2) and
hence (177.1). Thus we obtain
a b c
---;=:;::==0==+
va 2 + 8bc Vb 2 + 8ca
+---;=;=====7
vc
2 + 8ab
aa11 b~ c~
::::: 4 4 4+ 4 4 4+ 4 4 4=l.
a"3 +b3 +C"3 a3 +b3 +c3 a"3 +b3 +c"3
Solutions 349
Alternate Solution.
Introducing
bc ca ab
a 2 = x, b2 = y, 2=Z,
cc2 = Z,
the inequality to be proved is
111
-===+
VI + 8x VI + 8y + VI + 8z ->1 ,
(177.3)
under the restrietion xyz = 1. Let us put .
1 1 1
VI + 8x == p, VI + 8y == q, VI + 8z == r,
so that (177.3) reduces to
p + q + r ;::: 1, (177.4)
and xyz = 1 transforms to
(1 - p2)(1 - q2)(1 - r 2 ) = 512p 2 q2 r 2. (177.5)
Suppose (177.4) is not true under the condition (177.5). Then we have q+r <
1 - p and q + r + 2p < 1 + p. Thus we obtain
2
> (q + r)(2p + q + r) ;::: 2(qr)21 ·4(p2 qr)"4
1 1 3
1- p qr) -.. = 8p2 (qr) "4-..,,
by an application of the AM-GM inequality. Similarly, we obtain

2 1 3 2 1 3
1-q >8q2(rp)4,I-r >8r 2 (pq)"4.
(pq)-".
It follows that,
(1 - p2)(1 _ q2)(1 _ r Z ) > 83 p2 q2 r 2,
contradicting (177.5). We conclude that (177.4) holds and hence the required
inequality is true.
A generalisation.
We prove a more general inequality(due to Oleg Mushkarov, Bulgarian leader
for IMO-200l). We show that
a b c 3
-r=-~~= +
)"bc + Vb 2 + Aca
Va 2 + Abc
+ 2 + Aab
)..ca + vc
> -~'
)..ab -;::: ~'
(177.6)
for all positive real numbers a, b, c and for all A

).. ;::: 8. As in the first solution,
introduce the new variables x, y, z, and the inequality to be proved is
1 1 1 3
VI - ++ VI + )..y +
+ )..X
Vl+Ax VI + )..z >
Vl+AY + Vl+Az
-
;::: ~'
-~'
(177.7)
350 Solutions
where xyz = 1 and >. ~ 8. Put
1+ >.X = u2 ,
= 1+ >.y = v2 ,
= 1+ >.Z = w2 ,
=
so that (177.7) reduces to

3uvw
uv+vw+wu>
+ vw + wu ~ \11
~, (177.8)
uv - v 1+ >. '
+>'
under the restrietion >. > 8 and
(u 2 _ 1)(v 2 _ 1)(w 2 _ 1) = >.3. (177.9)
We consider two possibilities:

+ >.)3/2. Then, an application of the AM-GM
Case 1. Suppose uvw ::; (1 +
inequality yields
2 3uvw
+ vw ++ wu ~> 3(
uv + 3(UVW)3
-
> ~.
uvw)3 ~ 'I + >. . - 1+>'
Case 2. Suppose uvw ~ (1 + >.)3/2. We write (177.9) in the form
>.3 U2 V2 W2 _ V 2 W2 + W2
(u 2 v 2 + V 2 U2
2 )) + (u 2
2 + v 2 + w 2) - 1
=
= (uvw+u+v+w)2_(uv+vw+wu+1)2.
Thus the inequality (177.8) reduces to
.
(UVW+U+V+W)2-
(uvw+U+V+W)2- (3UVW
yT+":\+1 )2 ~>.3. (~+1)2 ~>.3.
Since u + v + w ~ 3( uvw) 1/3, we see that by setting uvw =
= X 3 , it is sufficient
to prove that
(X3 +3X)2 _ ( 3X3 +
yT+":\
~
+ 1)2 -
-,
1)2
> >.3 (177.10)
~. We write (177.10) in the form

under the restriction X ~ yT+":\.
[X 3 (1 + k) + + 3X + 1] [X 3 (1 - k) + + 3X - 1] ~ >. 3.
Since >. >

-
8, we observe that 1 - k~ >
> O. Using X>
1+>'--
~, we obtain
X > yT+":\,
X3 (1 - k) + 3X - 1
> (1+>')~ (1- k) +3~-1 +3-J1"+1"-1
vT+1(>' +
-J1"+1"(>. + 4) - (3)' ++ 4).
Solutions 351
Similarly, we show that
(1 k)
X 3 (1 +
+ 1~ ~('x+4)
~) + 3X ++ 12: + 4) +
+3X+ (3'\ + 4)..
(3,X+4) v'l+:\(,\
Combining these, we get
(X 33 +
+ 3X)2 (~
3x3
3x)2 _- ( ~ + + 1) ~ (1 +
1 2 2: + ,\)(,\
,X)(,x +
)2 (3'x +
+ 4)2 _- (3'\ 4)2 == ,\3.
+ 4? ,X3.
This proves (177.10) and henee the required inequality (177.6).
178. We know that

ABc
r == (s - a) tan - == (s - b) tan - == (s - c) tan -.
22 2 2 2 2
Henee we get
ABC
+ tan - + tan -
tan - +
222
( 1 1 1)
r ( -1- + -1- + -1
s-a s-a
s-a s-a
-)
(s - a)(s - b) + (s - b)(s - c) + (s - c)(s - a)
[ABC]
9R 2
9R
(s - a)(s - b) + (s - b)(s - c) + (s - c)(s - a) :s
~ 4'
4.
Using
(s - a)(s - b) +
+ (s - b)(s - c) +
+ (s - c)(s - a) = r(4R +
+ r),
the inequality to be proved is
4r(4R + r) :s 9R2 .
(9R + 2r)(R - 2r) 2:
~ 0,
2: 2r.
whieh follows from R ~
179. Sinee both L sin an and

15
~+~
"4 ~ eos(n - ß) are positive, the in-
eos(a
cyc\ic cychc
equality is equivalent to
15
L L
2
(
sinn
sin a )) :S""4
~"4 + eos(n
eos( a -- ß)·
cyc\ic cyc\ic
cyclic
352 Solutions
Expansion and some cancellation leads to
3- L: cos 2 Cl'::;
a::; 145 + L: (cos Cl' a sin ß).
a cos ,8 - sin Cl'
cyclic cyclic
L: (cosa-l/2)2 20.
(COSCl'-1/2)2
cyclic
Equality holds if and only if cos Cl'

a = = cos ß == cos I == 1/2, which corresponds to
the case of an equilateral triangle.
180. It may be seen that
3x 2 - 2x3 - 1 = -(x - 1)2(2x + 1),
and
4 y3 - 3y4 _ 1 = - (y - 1) 2 (3 y2 + 2y + 1).
Thus, we obtain
+ y3 _ x3 _ y4) ++ x 3 ++ y3 - 2
3(x 2 +
= -(x -1)2(2x
= + 1) - (y _1)2(3y2 + 2y + 1).
Tt follows that
2 - x3 _ y3 (x -1)2(2x + 1) + (y _1)2(3y2 + 2y + 1)
+3(x2 + y3 _ x3 _ y4)
> 0
181. We introduce the variables x, y, z by

a b c
b == x, ~-c == y, -a == z.
Then xyz == 1 and the inequality takes the form
x-I y-l z-1
- - + - - + - - > O.
y+l z+1 x+l-
Expanding, this may be written in the form
+ y2 ++ z2 - X - Y - z
x2 + + xy2 ++ yz2 ++ zx 2 - 32 o.
However, we have
1
x 2 + y2 + Z2 > -(X+y+Z)2
3(X+ Y +Z)2
3
>
> + Y ++ Z)(xYZ)1/3 = X + y ++ z,
(x +
Solutions 353
and
xy2 + YZ2 + ZX 2 ~ 3(X 3y3Z3)1/3 == 3.
182. If cos x = 0, then the inequality is dear. Suppose cos x -::f. O. We may
write the inequality in the form
( tan x + a) (tan x + b) ::; sec x + (a; b)

2 2 sec 2 x.
r).
Taking t = tan x, this may be written as
t2 + t(a + b) + ab::; (1 + (1 + t2 ) (a; b
This reduces to
C(a; b) _ 1)
2
+ (a; b)
2
~ O.
Hence the inequality follows.
183. If 0 ::; y ::; I, then y(y - 1) ::; 0 ::; x 2. Suppose y > 1. If x::; y - (1/2),
then
y(y - 1) = y(y + 1) - 2y <

::; (x + 1)2 - 2y
+ 2x + 1- 2y
x2
< x2 •
If x ~ y - (1/2), we have
x2 ~y
2 1 ( )
-Y+"4>yy-1.
Thus, y(y - 1) ::; x 2 holds in all cases.
= c, Y + Z == a, Z + x == b. Then a, b, c are the sides of a triangle,

184. Put x + Y =
= s - a, y == s - b, Z == s - c, where s == (a + b + c)/2. The inequality
and x =
may be written in the form
1
( 3((s-a)(s-b)+(s-b)(s-c)+(s-c)(s-a))
) 1/2
::;
(b)
a8 c
1/3
But, we also know
(s - a)(s - b) + (s - b)(s - c) + (s - = r(4R + r),

c)(s - a) = = 4Rrs,
abc =
354 Solutions
where R, rare the circum-radius and in-radius of the triangle whose sides are
a, b, c. The inequality reduces to
4r(4R + r)3 ~ 27R 2 s 2 .
Using 3(ab + bc + ca) ~ (a + b + c)2, we obtain

(ab + bc + ca) - s2 ~ ~S2.
Hence, we get
r(4R+r) (s - a)(s - b) + (s - b)(s - c) + (s - c)(s - a)

(ab + bc + ca) - S2
1 2
< -s
3
Thus, it suffices to prove that
4(4R + r)2 ~ 81R 2 .
This equivalent to
(17R + 2r)(R - 2r) 20,
which follows from R 2 2r.
a +b9 9 1 (3 b3
+ a 3 b3 +b
aaß6 +a + b6ß -> 3- a + ).).
-;:---;:-::-::---::-;:-
This follows easily by cross-multiplication and rearrangement of terms. Hence
2: a9 + b9 2
> _(a 3 +b 3 +c3)
cyclic
3
> 2(abc) = 2.
186. Note that

y+z
y+z b=Z+X X+y
a==
a -- cc==-2
-
2 - ..
2 ' 2 '
Thus,
abc ( y; z) (Z; x;
X) ( x; y)
>
> ffzv'zXy'XY = xYZ.
Moreover,
1
ab = 4(xy + yz + zx + Z2),
Solutions 355
so that
Lab
Lab
~ C~c X
2
+3 c~c X
y)
cyclic
> ~ C~c xy + 3 c~c X

y)
L xy.
cyclic
abc L abab ~ XYZ L XY·

cyclic cyclic
187. Define
A t (8: ak)ß
k=l k
(n-1)ß Li31(1
n_1 Laj )13
n
k=l ak j#k
The function f(x) = xß is concave für x> O. Hence Jensen's inequality gives
( --
1
n-1L a· )13 > -1- L aß.
-n-1
J J'
j#k j#k
a1 a2
= a2 == ...
with equality if and only if a1 = ak-1 ak+1
... == ak-1 == ak+1 == ...
... = an.
an. Thus, we
have
A
A (n -
> (n - 1)13
1)13 t t a;ß n~ (L
k=l
(L an n 1
j#k
-1)13 ta~ n
(n(n -1)13 (La;ß).
n ~ 1 (La;ß).
j=l k#j
356 Solutions
The function g(x) = x- 8ß is convex for x

= X- > o. Hence we get
A > (n-1)8ta;(Ln~lak)-8
j=1 k=h
n
(n -1)2 8 La ; (
Lak )-8
j=1 k-j.j
n
(n - 1)28ß L a~(s
a~(8 - aj)-8
aj)-ß
t, C~j
j=1
(n - 1)28 aJß,
with equality if and only if al = a2 = ... = an-

188. Let 0] and O 2 be the in-cent res of ABD and ACD respectively; and let
81 and S2
SI 82 be their semi-perimeters. Then we have
[ABC] = [ABD] + [ACD] = rlS1 + rlS2 = rl (SI + S2).

But
1
81 + S2 2(AB + AD +BD +AD + DC +AC)
1
-2" (a
2
+ b + c) + AD =. s + AD.
Thus [ABC] == rI(8 + AD).

rJ{s
A
B p D Q c
Let P and Q denote the respective points of contact of the in-circles of ABD
and ACD with BC. Then 01PQ02 is a rectangle. Hence
0 10 2 = PQ = PD + DQ = SI
81 - AB + 052 - AC
= (81
(SI + 82)
82) -- (b + c) = S8 + AD - b - c.
Solutions 357
Let [ be the in-centre of ABC. Then, [0 1 0 2 is similar to [BC so that
0 10 2 r - r1
BC r
This gives (s + AD - b - c)r = a(r - r1). But rs = [ABC] = rds + AD). We

get a quadratic equation:
arf - 2rsr1 + r 2s = o.
Solving, we get
r1 = rC ± yI:(s - a)).
Since s + yls(s
J s( s - a) > a, we must take the negative sign. Thus, we obtain
r1 r(S-J:(s-a))
r(S-yl:(s-a))
rs rJs(s-a)
a a
V'-s('s---a""""'"")
~hh _ rrJs(s
~
~---,-
- a)
2 a a·
This reduces to
h aa == 2r1 + 2rJs(s -- a)
2r1 + 2ryls(s
a
a)
,
which proves the first part.
From the first part we have
ha + hb + h c = 2(r1 + r2 + r3) + 2r{ ylsCi=

a
a; + Psb-=b) + yls(sc - c) }
.
2{
{ J(S=a)
y'(S=a) y'(S=b) y'(S-=c)
+ V'(S=b)
b +
J(S-=c) } y's
:S-.
a c r
But, note that a = s - b + s - c. Hence
y'(S=a) y'(S=a) y'(S=a)

a s-b+s-c
< ~.
- 2V'(S=b)y(s-c)
358 Solutions
Thus, we obtain
{
2{
~ + yI(S=b)
b + y'(S-=c)}
a c
< ~ yI(S=b) y'(S-=c)

< +
--,=;===~o;====7' + --;=;,====;=;==:=;=
--,=;====;=::;===:=;=
J(s - b)(s - c) J(s - c)(s - a) J(s - a)(s - b)
s
J(s - a)(s - b)(s - c)
s
ryls
yIs
r

n
62:22)+2
""'
j=l aa·a-2·+1 + a2
2 a+ a·j
a
a3j3
t (aj y1ij) 2 2
j=l J J) 1 J j=l (/a;a;+l + a; )
(2:7=1 ajy1ijr
> "n (2 2 2) ,
~j=l a aj+1 + a j
j
where an+J
an+J = a1'
al' But
n n
L (a;a;+1 + aJ)
aD = 1 + La;a;+l'
j=l
j=l j=l
j=l

n 1
2.: a;a7+1 ~ 4'
j=l
under the restriction 2:7=1

2::7=1 = 1. a;
Xl,X2,X3, ... ,Xn be n positive real numbers such that 2:7=1
Let X1,X2,X3, 2::7=1 Xj = 1,
=
where n 2: 4. If n is even, then
X1X2
XlX2 + XZX3
X2X3 + ... + X n X1 ~
XnXl (Xl + X3 + ... ) (X2 + X4"')
:S (Xl = y(l - y),
where y = 2:2::j X2j-1' :S 1/4 since 0 ~

X2j-l' But, y(l - y) ~ :S y ~
:S 1. This takes care of
even values of n. If n is odd, then n 2: 5, and we may assume Xl Xl 2: X2 (since
Xj 2: Xj+l for some j). Thus
1
XIX2 + X2 X 3 + ... + XnXI :s: :S 4'
Xl (X2 + X3) + (X2 + X3)X4 + ... + XnXI ~
:S Xl
using the previous argument to n - 1 numbers Xl,
Xl, Xz
X2 + X3, X4, ... X n . This
completes the proof.
Solutions 359
190. The answer is NO. Suppose such a sequence (x n ) exists. Then yfxn+l >
,;x;;and hence (x n ) is strictly increasing. Moreover,
yfXn+1 - Fn == Xn+2 > Xn+1 =

= Fn - yfXn-1,
so that
yfXn+1 - Fn > Fn - yfXn-1 > ... > JX2 - VX1.
n
yfXn+l - VX1 L (yfXj+1 - y'xj)
j=l
> n(JX2 - VX1).

This gives
yfXn+1 > n(JX2 - VX1) + VX1,
für all n > 2. Choose k large such that
1
k(JX2 - VX1) + VX1 > "2.
Then yfX n+1 > 1/2 for aB n ~ k. Thus if n ~ k, then
Xn+3 > Xn+3 - Xn+2

( yfXn+3 - yfXn+2) ( yfXn+3 + yfXn+2 )
> ( yfxn+3 - yfXn+2 )
X n +4·
This contradicts the earlier observation that (x n ) is strictly increasing.
( n ) (n a 2 ) (n )
~ aj
2
~ ~ b; ~ bj .
But 2:.7=1 bj = 2:.7=1 aj. Hence we get

n
~ b·
a· >
2
'"'" -...L
b. -
L a)..
n
J
j = l )J
j=l j=l
192. The Cauchy-Schwarz inequality gives
(t)=1 aj ) 2 ~ (t
)=1
~ b.) (t a.
J J J=l
aj + bj ) .
Using 2:.7=1 aj = 2:.7=1 bj = 1, we get the desired inequality.

360 Solutions
193. Introducing a = y + z, b = z + x, c = x + y, we see that a, b, c are the sides

of a triangle, and x =
= s - a, y =
= s - b, z =
= s - c, where s is the semi-perimeter
of the triangle. We get
y2 _ x2 (a - b)c
y - x = a - b,
z+x b
We hence obtain
y2 _ x 2 Z2 _ y2 x 22 -- Z2 ac ba cb
-'---- + + = - + -- + - - (a + b + c).
(a+b+c).
z+x x+y y+z b c a
Thus, we have to prove
a 22b22 + b22c2 + c22a 22 2: abc(a + b + c).

This follows from Muirhead's inequality, since (2,1,1) --< (2,2,0).
194. We show that the largest value of k is 4. In fact, we show that
(a 22 __ bc) 2 > 4 (b 2 - ca) (c 2 - ab),

for every positive tripIe (a, b, c), and for any I > 4, it is possible to choose
positive a, b, c such that
(a 22 -- bC)2 < l(b 22 -- ca) (c 2 - ab).

Let us take ,X = a - v'bC. Then
(a 2 - bc)2 - 4(b 2 - ca)(c 22 -- ab)
a 4 _ 6bca 22 + 4(b 3 + c3 )a - 3b 22c22
('x + ~)4 _ 6bc(,X + ~)2 + 4W + c3 ) (,x +~) _ 3b 22c22
,X4 + 4~,X3 + 4(bVb - cye)2,X + 4~(bVb _ cye)2
> o.
On the other hand, suppose I > 4. Choose E > 0 such that 5E o.
= (1 + E)2 - 1 =
Hence
(a 22 _bC)2 = (2E+E 2)2 = E2 (E 22 +4E+4)
< E2 (5E + 4)
< lE 2 == I ( - E)( - E)
I (b 2 - ca)( c2 - ab) .
Thus, k = 4 is the largest constant for which the given inequality holds for all
choices of the positive reals a, b, c.
Solutions 361
195. Using cd = I/ab and da = l/bc,

I/bc, the inequality is equivalent to
1 + ab 1 + ab 1 + bc 1 + bc 4
-1 +
-a+ ab + abc +--+
1+b
> .
>4.
bc + bcd -
The AM-GM inequality gives
1 + ab 1 + ab 4(1 + ab)
-1 +
-a+ ab + abc > 1 + a + ab + abc'
l+bc l+bc 4(1 + bc) 4a(I + bc)
4a(1
--+ > =
l+b bc+bcd 1 + b + bc + bcd a + ab + abc + 1 .
It follows that
1 + ab 1 + ab 1 + bc 1 + bc
-- + -:----:
+ - - + -:---.,--
1+a ab + abc 1+b bc + bcd
> 4(1 + ab) + 4a(1 + bc)
4a(l+bc) = 4.
=4.
- 1 + a + ab + abc a + ab + abc + 1
196. Taking x == s - a, y == s - b, zZ == s - c, we see that x, y, zZ are positive
reals such that x + y + zZ == 1/2 and a == y + z, b == zZ + x, c =
= x + y. Using
Minkowski's inequality, we get
(an + bn r/ n ( (y + z)n + (x + z)n )

ll/n
/n
< ( yn + x n ) l/n + ( 2zn )l/n

) l/n
< Yy + Xx + 2 1 / n Z
+21 / n z.
c+2
C Z.
l/n l/n
( bn + cn ) < a + 2 1 / n x, ( cn + an ) < b + 21 / n y.
Adding, we get
( an + bn ) l/n + ( bn + cn ) l/n + ( n
c + an ) l/n
21 / n
< a + b + c + 211/ n (x + y + z) = 1 + -2-.
= -2-'
197. For any positive real numbers u, v, x, y, we have
u v uy + vx
uy+vx 4( uy + vx)
4(uy+vx)
-+-=
x Y xy
> + y)2 .
(X+y)2
- (x
.
362 Solutions
Thus, we gct
a c 2a 2 +
+ 2c 2 +
+ ab +
+ bc +
+ cd +
+ da
----:+ >
b+
+ 2c +
+ d d+
+ 2a +
+b (a ++ b+
+ c++ d)2
b d 2b 2 +
+ 2d2 +
+ ab +
+ bc +
+ cd +
+ da
----+ >
c+
+ 2d +
+ a a+
+ 2b +
+c (a+b+c+d)2
Adding, we get
2( a2 ) +
+ 2(ab +
+ bc +
+ cd +
+ da)
L
2:CYcliC
2:CycliC
a >
(a+b+c+d)2
cyclic
( 2:cycliC a) 2 + ( 2:cycliC a2 ) - 2ac - 2bd

(a+b+c+d)2
1+ (a-c)2+(b-d)2 1
> .
(a+b+c+d)2 -
198. We have
1 (a+b)(b+c)(c+a)
a2 b +
+ ab 2 +
+ b2 c +
+ bc2 +
+ c2 a +
+ ca 2 +
+ 2abc
(a +
+ b++ c)( ab ++ bc +
+ ca) - abc.
Thus, we obtain
1 + abc
a+ +
ab + bc
ab bc + ca = a+b
ca = b c
+c
a+b+c = ~((a+b)+(b+c)+(c+a))
3(
> 2 (a+b)(b+c)(c+a) )1/33
=
= 2·
Similarly,
1 = (a +
+ b)(b +
+ c)(c +
+ a) 2: 8Vab ~ ,;ca = 8abc.
8abe.
This shows abc::; 1/8. Finally,
ab + bc + ca = 1l+abe
ab+bc+ea= + abc < 11+1/8=~.
+ 1/8 _ 3
a + b+ e -
a+b+c 3/2 - 4·4
Solutions
Soltltions 363
199. Let h a be the altitude from A on BC and r a be the ex-radius correspond-

ver tex A. Then it is easy to see that
ing to the vertex
AD ha
DIa ra
But,
Eut, we observe that
h _ AB·AC
a - BC
Since ABC is right-angled at A, we know that r a = (a + bb+
+ c)/2.
c)j2. Thus, we
get
ha 2bc
a( a + b + c) .
r a - a(a+b+c)"
However
(b + C)2
a(b +
+ c) = (b +
+ c) Vb 2 +
c)Jb +~)2 2::::: 2V2bc,
+ c2 2::::: (b V2
and
a2 = b2 + c2 2: 2bc.
Hence, we get
AD < 2bc < 2bc _= V2

~2 _ 1
1.
- < < -VL..
DIa - a22 + a(b + c) - 2V2bc + 2bc
200. Note that
2 + y2
x2 Z2 =
+ Z2 = (x + Y + Z)2
z)2 - 2(xy + YZ
yz + zx) =
= 1- 2(xy + yz + zx).
Hence it is sufficient to prove that
9xyz ::::
~ xy + yz + zx.
Introducing
a = y + z, b = Zz + x, =x
c =x + y,
we see that a, b, c are the sides of a triangle, and
x =s- a, y =s - b, s = 1 - c,
where s = c)j2 =
= (a + b + c)/2 = 1. The inequality is
9(s - a)(s - b)(s - c) ::::

~ (s - a)(s - b) + (s - b)(s - c) + (s - c)(s - a).
We use
L (s - a)(s - b) = r(4R + r), (s - a)(s - b)(s - c) = r 2 s,

cyclic
364 Solutions
to get
91'2 S ::; l' ( 4R + 1').
But S == 1 and this reduces to 21' ::; R, which is Euler's inequality. The result
follows.
Alternate Solution: We use thc homogenisation technique. We may write
the inequality in the form
(x + y + z)( xy + yz + zx) + 18xyz ::; (x + Y + z(

2( L x 2y + L xy2 - 6XYZ) ~ O.
cyclic cyclic
This follows from the AM-GM inequality.
201. Let K, L, M be respectively the mid-points of BC, CA, AB. We have
a2
AK·KD =BK·KC= 4'
=BK·KC=-.
4
We thus obtain
a 22
KD=~
KD= -.
4m a
A
This implies that

ma a2
GD=GK+KD -
3
+4m-a
4m~ + 3a 2
12m a
a2 + b2 + c2
6m a
Solutions 365
Similarly,
~ a22 ~ a22
GE =- L... cyclic a
Lcyclic GF = L...cyclic a
LCYcliC
- 66mb
mb ' 6m c
Adding these, we get
2 c2 ) ( 1
GD+GE+GF= ((aa22 +
GD+GE+GF= +bb6 + - -1 +
+m
+c )(_1_+_1 +~)
-1 ) .
.
6 ma m
m c mc
m
But, we know that
4(m; + m~ + m~) = 3(a 2 + b2 + c2).
GD+GE+GF
GD +GE+GF = -92(m; +m~ +m~) (~+
~(m;
9
(_1_
ma
~c +
+m + ~).
mc
~ ( L m;)(
~(L
cy
m;) ( L~)
clic
cyclic
: : L ma ,
L ~a ) ~Lma, cycli
cyclicc
which is a consequence
consequence of Chebyshev's inequality.
202. If a :S b :S c, we see that h a ~:::: h b ~

:::: h ec . Hence Chebyshev's inequality
gives
((aa + b + c) (ha
(ha + h b + h ce ) >
~ 3 (aha + bh b + ch e )
3(66.) = 186..
3(6~) 18~.
203. We use the concavity of f(x) =

= x 1 / 3 to get
( ~1 + 6b ) 1/
1/33 (1
+ b + 6c
) 1/3
+
(1
~ + 6a
) 1/
1/33
3 (1 1 1 ) 1/3
l/:{
< -- -- + - + - -++6(a + b + c)
-.y3a
-?f;3a bb c
= 1 gives
But, ab + bc + ca =
3abc(a + b + c) :S (ab + bc + ca)2 = 1,
so that (a + b + c) :S 1/3abc. Moreover
1 1 1 ab + bc + ca 1
-+-+-=----
~ +
abc b + ~ = abc abc'
abc
366 Solutions
Hence, we obtain
1 1 1 3
- + - + - + 6(a + b + c) < - .
abc - abc
Thus we get
(
1
~+6b
)1/3 + (1b+ 6c)1/3 + (1c+ 6a)1/3 :::;(abc)l/3'
3
Again, note that

1 = ab + bc + ca::::: 3(abc)2/3,
so that
3 1 1 1
--:--::----:-::-=< . (abc)l/3 =-
(abc)l/3 - (abc)2/3 abc'
This completes the proof.
204. For a point D on AB, let LAC D = /1, LBC D = /2. We have
CD 2 sinA sinB
AD·BD sin /1 sin /2
But, we have
sin /1 sin /2 ~{ COS(r1 - / 2 ) - cosC}
< 1 - cos C . 2 (CI )

2 =sm 2.

CD 2 sinAsinB
-------
-:-:::--:=-= >
--~----
-n---
AD· BD - sin 2 (C/2) ,
far any point D on AB. Hence, the range of values of CD 2 lAD· BD is the
interval [sinAsinBIsin2 (C/2), (0). It follows that CD 2 =
= AD· BD if and
only if 1 is in the range of values of CD B D. This is equivalent to
lAD . BD.
C D 2 lAD·
v'sinAsinB :::; sin(C/2).
205. Put A o = 0 and define Ck = Ak-dA k for 1 :::; k :::; n. We observe that
ak = kAk - (k - 1)A k - 1 =k _ (k _ l)ck'

Ak Ak
We may write
1/n2
Gn) l/n ( 2 n-1 )
( C2C3 ... c
Cn ) ,
An
Gn
9n ( rr
n
k=l
(k - (k - l)ck)
) l/n
Solutions 367
n(1 n(n+l)/2c2C~ ... C~-l) I/n

2
< .!. (n(n + 1) +

n 2
"t(k _
k=2
l)Ck)
n + 1 1 nn
-n + 1 + .!.- 2)k - l)ck.
22 nn k=l
k=l
Similarly, the AM-GM inequality also gives
( g
n
(k - (k - l)ck)
) I/n
::; T+ -:;-;1 1 n
E(k - l)ck.
Adding these two, the required inequality is obtained.
3(X 2y2 + y2 z2 + Z2 X2) - 2xyz(x + Y + z)

= (xy + yz - ZX)2 + ( - xy + yz + ZX)2 + (xy - yz + ZX)2.
Note that
xy + yz - zx 2: -zx 2: -1,
and
xy + yz - zx y(z + x) - zx
< x + z - zx
(x -1)(1- z) + 1::; 1.
Thus, we obtain
(xy + yz - ZX)2 ::; 1.
Similarly, we may bound the remaining two terms also by 1 each. The inequal-
ity follows.
207. We use the technique of homogenisation to put it in the form
7(xy + yz + zx)(x + Y + z) ::; 2(x + Y + Z)3 + 9xyz.

This reduces to
7( I:
cyclic
x 2y + I: xy2 + 3XYZ)
cyclic
::; 2 ( I: x 3 + 3 I: x 2y + 3 I: xy2 + 6XYZ) + 9xyz.

368 Solutions
Equivalently, we have to prove
2( L x 3) 2: L x 2y + L xy2.
But, this is a direct consequence Muirhead's theorem.
208. First, observe that xyz cannot exceed each of xy, yz, zx. Thus, we see
that
x
_x_+_y_+_z_ y z < ---+ x y z
- - +zx
yz + 1
-+-1+xy-+-1 < --+ yzx + 1 +
xyz + 1
+----
zxy + 1
x+y+z
xyz + 1
Hence, it is sufficient to prove that (x + y + z) ::; 2(xyz + 1). Since x, y are
in [0,1], we have (1 - x)(l - y) 2: 0. This reduces to 1 + xy 2: x + y. Since
z 2: 0, we get z + xyz 2: zx + yz. Similarly, we get x + xyz 2: xy + zx and
y + xyz 2: xy + yz. Adding these, we get
(x + Y + z) + 3xyz 2: 2(xy + yz + zx).
Hence, it follows that
2xyz + 2 > 2xyz + 2 + 2(xy + yz + zx) - 3xyz - (x + Y + z)
2 (1 - x)( 1 - y) (1 - z) + xy z + (x + Y + z)
> x+y+z.
This proves the inequality.
209. We may write the relation in a, bin

b in the form
a 3 + b3 + (_1)3 - 3(a)(b)( -1) = 0.
(a + b- l)((a - b)2 + (a + 1)2 + (b + 1)2) = 0.
Thus, either a + b == 1 or a == b == -1. Since a, b are positive, we conclude that

a + b== 1 == c+ d.
If x, y are positive real numbers, then
x 3 + y3 2: xy(x + y).
(In fact, this is equivalent to (x - y)2 2: 0.) We therefore get
(a+~r+(b+~r >
2: (a+~)(b+~)(a+~+b+~)
2: 4 ( 1 + :b)
> 4(1 + 4) = 20;
Solutions 369
we have used a + l/a ~ 2 and ab >~ ((a + b)/2f

> 1/4. Similarly, using
c + d = 1, we obtain
(c+ ~r + (d+ ~r ~ 20.

Combining these two, we get the required inequality.
210. As in the example 2.23 on page 75, we use trigonometrie substitutions:

x = tan A, y = tan Band z = tan C. Then A, B, C are the angles of a triangle
and the inequality reduees to
3
eosA + eosB + eosC < -.
- 2
This follows from 3.4.9 on page 102.
211. The inequality may be written in the form
Ri2 +
'" R + RR~2 -- cc22 0
+ ~.
33+L.." 1 Rl~2
RR
1 2
~~.
o.
cychc
Let LBPC = Q, LCPA = ß and LAPB = ,. Then the eosine rule applied to
triangles BPC, CPA and APB reduee
reduce the inequality to the form
3
eos
cos Q + eos
cos ß + eos
cos,rv >
1_
"2.
~ -- -.
2
/.,
b
a
R3
B a c
We may assume Q ::; ß ::; ,. Sinee,
Since, < x, it follows that ß > x/2. Now the
function f(x) = eosx
funetion convex in [x/2,xJ. Henee
cosx is eonvex Hence Jensen's inequality gives
cos
eos ß+ eos
cos, cos (ß ; ,) = - 2 eos
l' ~ 2 eos cos (~ ).
370 Solutions
Taking'\ =
= cos(a/2), we see that it is sufficient to prove that
2 3
2'\ - 1 - 2'\ > - -.
- 2
This is equivalent to (2'\ - 1)2 2': 0, which is obviously true. Equality holds if
and only if ß == 'Y and ,\ == 1/2. This gives a == 120° and a being the least
= ß == 'Y == 120°. Thus P is the Fermat's point of ABC.
angle, we get a =
212. We may assurne that t l1 ::; t2 ::; ... ::; tn. Thus, it is enough to prove that
t nn < tl + t2· The given condition may be written as
n2 + 1 > n+ '" (ttk ttj

L...
l::;j<k::;n
-.2+_k )
n + tn (~
tl
+ ~)
t2
+ ~
tn
(tl + t2)
+ 2:
l<;<k<n
(ttk + ttj
j k ).
(j,k)#(I,n),(2,n)
t·
tj + -tk
Using -.2
tk > 2, it follows that
tk tj-
n 2 +1 >n+tnC~ + t~) + t~ (tl +t 2) +2{ (;) -2}.
This reduces to
tn (~ + ~) + ~ (tl + t2) - 5 < O.
tl t2 tn
1 1)
We observe that
-4t
- n< t n ( -+- .
tl + t2 - tlt2
Thus, we get
4
-+'\-5<0
,\ ,
where
,\ = h + t2
tn .
This shows that (,\ - 1)('\ --,-- 4) < O. We conclude that 1 < ,\ < 4. Thus, we
obtain
t n < tl + t2 < 4t n ,
and this completes the proof.
Solutions
Sol'utions 371
213. Put x = a 22 ,, y = b22 ,, ZZ = the form

c22 .. The inequality takes the
a66 + b66 + c66 ;::0:

2 a44 bc + ab 44 c + abc44 ..
This follows from Muirhead's theorem, since (4,1,1) -< (6,0,0).
214. Chebyshev's inequality gives
(a + b + c)(a 22 + b22 + c2 )) :S 3(a 33 + b33 + c3 ).

).
Hence, a + b + c :S 1. On the other hand, we also have
4(ab + bc + ca) - 112

;::0: a2 + b2 + c2 ;::0:2 ab + bc + ca.
This shows that 3 (ab + bc + ca) ;::0:
2 1. Thus,
Thus,
1 :S 3 (ab + bc + ca) :S (a + b + c) 2 :S 1.
We conclude that
a + b + c = 1, and 3(ab+bc+ca)
a+b+c= (a + b + c(
3(ab + bc + ca) = (a+b+ct
It follows that a = b = c = 1/3.
215. Clearing the denominators, this is equivalent to
4 (ab + bc + ca + a + b + c) ;::0:
2 3 (abc + ab + bc + ca + a + b + c + 1).
further
ab + bc + ca + a + b + c ;::0:
2 6.
Dividing by abc, this may be written in the form
1 1 1
-a+b
- +c- + a + b + c >
-
6.
6.
Using x + I/x ;::0:2 2, for positive x,

x , the result follows.
216. Clearing the denominators, we get an equivalent inequality:
a2 b2 + b2 + c2 + c2 a2 ;::0:2 3a 2 b2 c2 + 2abc(a3 + b3 + c3 ).
This may be homogenised:
(a 2 + b2 + c2 ) (a 2 b2 + b2 + c2 + c2 a2 ) 2;::0: 3a2 b2 c2 + 2abc(a 3 + b3 + c3 ).

I: a b + I: a b
4 2 2 4 ;::0: 2 I: a bc. 4
cyclic
cyc\ic cyclic
cycJic cyclic
cyc\ic
Since (4,1,1) -< (4,2,0), the result follows by Muirhead's theorem.

372 Solutions
217. Let D', E' and F' be the points of intersection of AD, BE, and CF
respectively with BC, CA and AB. Invert the whole configuration with the
circum-circle of ABC. Then BC is mapped on to the circum-circle of BOC;
CA on to that of OCA; and AB on to that of OAB. Moreover, D' moves to
D; E' to E; and F' to F. The property of inversion shows that
= OE . OE' == OF . OF' == R 2
OD . OD' =
= x, [OBC] == y and [OCA] == z. Observe that

Let [OAB] =
AO x+z BO x+y CO y+z
OD' -y-' OE' -z-' OF' - x
Thus
AO·BO·CO (x+y)(y+z)(z+x)
OD' . OE' ·OF' xyz
2VXfj· 2VfiZ'
2VXfj' 2VfiZ· 2,jZX
> yz
x
> 8.
A
F
E
B c
D
This shows that
OD' . OE' .OF' < AO· BO . CO = R 3

-- 88 =8 -
Hence, we get
R6
OD . OE· OF = OD'
OD·OE·OF= .__ , ?8R
~_. . OE' .OF' 2: 8R33 .•
Solutions 373
218. Suppose x :::; y :::; z. Then
(l+y)(l+z)
(1+y)(1+z) ~ (l+z)(l+x)
(1+z)(1+x) ~ (l+x)(l+y).
(1+x)(1+y).
Hence, Chebyshev's inequality is applicable and
C~C(l+y~;l+Z)
C~C(1+y~;1+Z) ~> ~C~CX3)C~c(1+y)1(1+Z))
1 (x 3 + y3 + z3)(3 + x + Y + z)
3 (l+x)(l+y)(l+z)
(1+x)(1+y)(1+z)
4(x 3 + y3 + z3)(3 + x + y + z) ~ 9(1 + x)(l + y)(l + z).

Since
3
( 3+x+y+z )
(l+x)(l+y)(l+z):::;
(1+x)(1+y)(1+z):::; (3+X;Y+Zr,
3 '
all we need to prove is
12(x 3 +y3 +Z3) ~ (3+x+y+z(
By Hölder's inequality, we have
9 (x 3 + y3 + z3) ~ (x + Yy + z) 3 .
Thus, it is enough to prove that
4(x + Y + Z)3 ~ 3(3 + x + Y + z(

Putting t = xX + Y + z, this reduces to
4t 3 - 3t 2 - 18t - 27 ~ o.
Equivalently (t - 3)(4t 2 + 9t + 9) ~ O. We need to check that t ~ 3. This
follows from
xX + Y + z ~ 3(xYZ)1/3 = 3.
219. Note that FD = 2(8

2(s - b)sin(B/2). Thus
BC a a
FD = 2(8 b)sin(B/2)
2(s-b)sin(B/2)
- = 2rcos(B/2)"
But cos(B/2) = hb/a. We thus get
BC a2 b
FD = 4r6.·
374 Solutions
The inequality reduees to
a 2 b + b2 c + c2 a ~ 24rß .
ab
a b cc 6r
-c + -a + -b >-.
- R
But
abc 6r
-+-+-
c a
>
- 3>-
-+-+->3>-
bb- --R'
R'
beeause 2r :S
::; R.
220. Observe that
a2 + b2 + c2 + d2 ~ ab + bc + cd + da = 1.
Note that the inequality is symmetrie in a, b, c, d. Suppose a >

2': b >
2': c ~2': d.
Then
1 1 1 1
".-----::
-----:: > > > -- .
b+c+d-c+d+a-d+a+b-a+b+c
b+c+d-c+d+a-d+a+b-a+b+c
Taking x = b+c+d, y = c+d+a, Z = d+a+b and w = a+b+c, Chebyshev's
inequality gives
_3
a3 + _
b3 + _
c3 + _
-+-+-+-2':-
3
d3 ~ _ + b3
1 ( a3 +b (3 3++cc3+d
+ d3)) ((1-+-+-+-..
3 1+_
1+_
31 + -1 )
_ . .
x Yy Z w 4 x y z
Z w
W
One more applieation of Chebyshev's inequality leads to
a3 + b3 + c3 + d 3 > ~(a2
~ (a 2 +b
+ b2 +c
+ c22 +d
+ d2 )(a+b+c+d)
)(a + b + c + d)
1
> 4(a+b+c+d).
~ ~ 2 ~
- --+ + +---
b+c+d c+d+a d+a+b a+b+c
> ~
16
(a + b+ c + d) (.!.x + .y!. + .!. + ~)
Z W
~(x+y+z+w)
48
(.!.x +.!.y +.!.Z +~)
w
1
>
3'
where we have used the AM-HM inequality in the last leg.
Solutions 375
221. We show that the range of values of >. is [-J2, J2). Taking Xl =
= X3 =
= 1
and X2 = p, we have
2 + pp22 2: 2pp2,
2,
which shows that p2 ::; 2. Thus pp E [-J2, J2).

p2 ::;
Conversely, suppose pp E [- J2, J2J, so that p2 p2 x~, x~,
::; 2. Consider xi, p2
p2 ::;
where Xl, X2, X3 are arbitrary real numbers. Now the AM-GM inequality gives
2
P
2
Xl +4
2
X2 > IP X 1X21 2: P X 1X2,
p2 2 2
4 X2 + x3 > IP X 2 X 31 2: P X 2 X 3·
Adding these, we obtain
2 p2 2 2
Xl +2 X2 + x 3 2: P(X1X2 + X2 X 3).
Since p2 ::; 2, it follows that
p2 ::;
2 2 2 2 p2 2 2
Xl + X 2 + X 3 2: Xl +2 X2 + x 3 2: P(X1X2 + X2 X 3).
Thus for all values of p E [- J2, J2], the inequality

2
Xl + x 22 + X 32 2: p(X1X2 + X2 X 3 ) ,
holds für all choices of real Xl, X2, X:~.
222. Suppose ab + bc + ca 2:
~ a + b + c. Using l/a, l/b, 1, c as weights, the
weighted AM-HM inequality gives
abc (1+1+1)2
-+-+- > b
1
c
1
a
1
ab + bC + ca
b c a
(ab + bc + ca) (t + ~ + ~)
a+b+c
1 1 1
> -+ - +-.
b c a
If ab + bc + ca ::; a + b + c, then we use a, b, c as weights to get
abc (a+b+c)2
-+-+- >
b c a ab + bc + ca
>
> ab + bc + ca
1 1 1
-b + -c +-.
a
376 Solutions
223. Adding a + b ::; 1 + c and b + c ::; 1 + a, we get 2b ::; 2 and hence b ::; 1.
Similarly, we get c ::; 1 and a ::; 1. Take a = 1 - a, ß = 1 - band, = 1 - c.
Then 0 ::; a, ß" ::; 1. Moreover a == b ::; 1 + c implies that a ~ ,. Similarly,
ß +, ~ a and , + a ~ ß. The inequality to be proved is:
(1 - a)2 + (1 - ß)2 + (1 _,)2 ::; 2(1 - a)(1 - ß)(1 -,) + 1.
This reduces to
a 2+ /3 2+,2 ::; 2(aß + ß, + ,a) - 2aß,.

We may assume that , is the largest among the three; i.e., a ::; , and ß ::; ,.
Then, ::; a + ß and hence ,2 ::; ,a + ß,. ;v!oreover, a 2 ::; a, and ß2 ::; ß,.
These two imply that
a 2+ ß2 +,2 ::; 2(ß, + ,a).

Thus it is sufficient to prove that 2aß, ::; 2aß. But this follows from , ::; 1,
and a, ß are non-negative.
Alternate Solution: (Arpit Amar Goel and Utkarsh Tripathi) As in the
earlier solution, we get a ::; 1, b ::; 1 and c::; 1. By symmetry, we may assume
c is the least and a the largest, so that 0 ::; c ::; b ::; a ::; 1. We may write the
inequality in the form
(a - b)2 + c2 ::; 2abc + 1 - 2bc,

or equivalently
(a - b)2 ::; (1 - c)(1 + c - 2ab).
Using a + c ::; 1 + b, we get 1 - c ~ a - b~ o. Also
1 + c - 2ab - a + b 1 + c - a(1 + b) - b(a - 1)
> 1+c - a(1 + b), (since a ::; 1)
1 + c - a - ab
>
~ l+c-a-b, (sincea::;l)
> 0,
because a + b ::; 1 + c. Thus, we get 1 + c - 2ab ~ a - b ~ O. It follows that
(a - b)2 ::; (1 - c)(l +c- 2ab).
Equality holds if and only if a = 1 and b = c, or b = 1 and c = a or c = 1 and

a = b.
224. Using the weighted AM-GM inequality, we have

abc
1 + bc
+1 - - +1 -
+ ca - >
+ ab a + b+ c
~~~~~~--~~~ > ~--~~~~--~--~--~
a + b+c - a(l + bc) + b(l + ca) + c(1 + ca)
Solutions 377
This reduces to
abc 1
- -+-
1 + bc
-+ +-
1 + ca
-- ->>-- .
1 + ab - 1 + 3abc
Using the AM-GM inequality, we also have
abc< (a
-
+ b+
3
c) = ~.
27
Thus, we get
10
1 + 3abc <
- -.
9
This implies that
abc 9
- -+-
1 + bc
- + -1 +
1 + ca
-ab
>- -10.
More generally, if Xl, Xz,
Xz, ... ,X n are non-negative real numbers such that
... ,X Xl +
X2 + ... + x n = 1, then
L rr -
n
Xj 1
""' --"x
L--,,--J--
=:-- >
> 1-+-1nl '
j=l 1 + l+n
j=ll+ Xk - n
l -n - '
Xk
l<k<n
kiej
225. (Riddhipratim Basu) We define X, = x, s - b == y, s - c == z,

X, y, z by s - a =
where s == (a + b + c)/2. Then x,y,z are positive, and a == y + z, b == z + x,
c = x + y. '\Te may express
R abcs (x + y)(y + z)(z + x)

2r 4~2 8xyz
Similarly
4a 2 - (b - C)2 = (3y + z)(3z + y),4b 2 - (c - a)2 = (3x + z)(3z + x),

4c2 - (a - b)2 = (3x + y)(3y + x).
Thus, we need to prove
(X + y)(y + z)(z + x)
8xyz
64 2 (x + y)4(y + Z)4(Z + X)4
>
-2: (3x + y)2(3y + x)2(3y + z)2(3z + y)2(3x + z)2(3z + x)2'
This may be writtcn in the form
(3x + y)2(3y + x)2(3y +z)2(3z + y)2(3x + z)2(3z + x)2

2: 32:3 xyz(x + y)3(y + z)(z + x)3.
~
378 Solutions
(3x + y)(3y + X)
x) 3x 2 + 3y2 + lQ:cy
(x + y)2 + (x + y)2 + (x + y)2 + 4xy
1/4
{ (x + y)6
> 44{(X+ ·4xyy }}1/4..
y )6'4x
It follows that (3x + y)2(3y + X)2 2: 32y'Xy(x + y)3 and

anel similar expressions for
the other products.
proelucts. Thus
(3x + y)2(3y + x)2(3y + z)2(3z + y)2(3x + z)2(3z + X)2

2: 32 3 xyz(x + y)3(y + z){z + X)3.
which gives the desired
elesireel inequality.
, abc 6
= 2/\ and
226. We know that 2R = anel r a
L.:>
=-
--,
-,
s-a
where a, b, c are the sides
sieles of
. a+b+c. .
the tnangle ABC, s = 2 and lS the area of ABC. Thus the glven
anel 6 lS given
condition
conelition 2R :::; r a translates to the condition
conelition
26 2
abc< - -
-s-a
Putting s - a = = p, s - b == q, s - c =
= r, we get a = = r + p, c == p + q
= q + r, b =
and
anel the condition
conelition now is
pep + q)(q + r)(r + p) :::; 26 2
But Heron's formula gives, 6 2 = s(s - a)(s - b)(s - c) = pqr(p + q + r). We
obtain (p + q)(q + r)(r + p) :::; 2qr(p + q + r). Expanding
Expaneling and
anel effecting some
cancellations, we get
p2(q + r) + p(q2 + r 2) :::; qr(q + r). (226.1)
Suppose a :::; b. This irnplies that q + r :::; r + panelp and hence q :::; p. This implies
that q2 r :::; p2 r anel qr 2 :::; pr 2 giving qr(q + r) :::; p2 r + pr2 < p2 r + pr2 + p'2 q +
pq2 = p2(q + r) + p(q2 + r 2) which contraelicts (226.1). Similarly, a :::; cis also
not possible. This proves (i).
Suppose 2R :::; rb. As above, this takes the form
l(r + p) + q(r 2 + p2) :::; pr(p + r). (226.2)

Since a > band
b anel a > c. we have q > p, r > p. Thus q2 r > p2 r and
anel qr 2 > pr2.
Heuce
q2(r + p) + q(r 2 + p'2) > q2 r + qr 2 > p2 r + pr z =
= pr(p + r)
which coutraelicts (226.2). Heuce 2R > rb. Similarly, we can prove that 2R >
r c . This proves (ii)
Solutions 379
227. We show that k = = 21 is the maximum possible value. The following

diagram shows that there is no reet angle for k = 21.
rectangle
We follow the argument in the example 4.3 on page 156. Let mj mj,, 1 ~ j ~ 7,
denote the number of elements of T in j-th row. In order to avoid a rectangle
reet angle
whose sides are parallel to the grids of T, we must have
1; (~j) ~ G).
Taking k = m1
ml + m2 + ... + m7, this may be written in the form
7
'Lm;
:Lm; ~ 42+
42+ k.
j=1
j=l
However, the Cauchy-Schwarz inequality gives

7
(mi +
(m1 m2 + ......+m7)2
+m2 L
+m7) 2 ~ 77:LmJ.
m;.
jj=l
=1
Thus, we obtain
k2 7
-7-L..t - 42 + k,
< '"' m J2 <
j=1
re duces to (k + 14)(k - 21) ~

which reduces ~ 21. If k = 21, then
~ O. This shows that k ~
we see that equality holds in the Cauchy-Schwarz inequality, forcing mj = 3
for 1 ~
~j ~ ~ 7. Since for k = 21, there is already an example, it follows that 21
is the maximum value of k which avoids a rectangle.
228. We use the following property of convex functions:

380 Solutions
1f f : [8, t] -+ IR is a convex function and if X, y, z are three points in [8, t]

such that x :S y :S z, then
(y - z)f(x) + (z - x)f(y) + (x - y)f(z) :S O.
Using the convexity of f (x) = In x on [0, 00), we have for 0 :S x :S y :S z,
(y - z) In x + (z - x) In y + (x - y) In z 2 o.
This simplifies to
xYyZ ZX 2 XZyx zY.
Thus, we have
abbcc a 2 bacba c, accdd a 2 cadca d.
Multiplying these two and effecting some cancellations, we get the required
inequality.
229. Join PQ. Let it intersect AAl in T and CCI in S respectively.

A
B AI Q c
Since M Pis parallel to CCl , we have
PH Cl 1 ACI
HM = 1C = AC·
But CCI is the bisector of angle C, so that ACI be/tb + a).

= bc/(b This gives
HM b+a PM a+b+e
a+b+c
and
PH ce PH ce
But PHT is similar to PMQ, so that
PT PH ce
PQ = PM a+b+c·
QS a
PQ b+e
a+ b+c
Solutions 381
Thus, we have
TS b
PQ a+ b+c
a+b+c
Now, using the similarity of triangles P
Now, HT, SSIT
PHT, IT and SN Q, we get
SNQ,
dd1 c d2 b d3 a
dd2 b' dd3 a'
a' dd1 c
Thus, the inequality to be proved is::
c b a 2ab
2~ 2bc
2~ 2ca
~a
- +- +- >-
-+-+ -.- - -++
--- -+ +---- ..
b a c -- a 2 + bc b2 + ca c2 + ab
2ab 2 1 (b a)
a 2 + bc = ~ +!:: :S 2" ~ + ~ ,
b a
and similar expressions may be obtained.
obtained. Therefore,
Therefore, we obtain
2ab 2bc 2ca c b a
- -- ++
a 2 + bc
-
-b2- -
++
ca
+
-c.2 -
+-
- < -+-+-.
ab - b a c
ne l parallel to BC through A and refleet

230. Draw a liline reflect AC in this line to get
intersect l in P. Join BD. Observe that CP = PD = AQ =
AD. Let CD interseet = ha ,
where AQ is the altitude through A. We have,
= AC + AB == AD + AB 22:: BD == JCD2
b+ c = + CB2 == J4h~ + a 2 ,
whieh occurs if and only if B, A, D are
which give the desired result. Equality oeeurs
eollinear, = AB (as AP is parallel to BC and biseets
collinear, i.e., if and only if AD = bisects
DC) and this is equivalent to AC = BC.
Alternate Solution:
The given inequality is equivalent to
16ß 2
+ C)2
(b + C)2 -- a2 2:: 4h
> 4h22 = 16ß 2
- a a2 '
tri angle ABC. Using the identity
where ß is the area of the triangle
16ß 2 = [(b + C)2 - a2 ][ a2 - (b - C)2 J
we see that the inequality to be proved is a 2 - (b - C)2 :S

:::; a 2 (here we have used
a < b + c) which is true. Observe that equality holds if and only if b = = c.
ar
231. By the AM-GM inequality, we have
a(l - a) :::; (a++ ~21 -- a)2

a) :s (a 1
-
4
382 Solutions
Similarly, we also have

1 1
b(1-b)~4 and c(1-c)~4·
Multiplying these, we obtain

1
abc(l - a)(l - b)(l - c) ~ 43 .
We may rewrite this in the form

1
a(l - b) . b(l - c) . c(l - a) ~ 43 .
Hence, at least one of the factors from among a(l - b), b(1- c), c(l- a) has to
1. 1
be less than or equal to 4; otherWlse lhs would exceed 4 3 .
Again, consider the sum a(l - b) + b(l - c) + c(l - a). This is equal to
a + b + c - ab - bc - ca. We observe that
3 (ab + bc + ca) ~ (a + b + c) 2 ,
which, in fact, is equivalent to (a - b)2 + (b - C)2 + (c - a)2 2: 0. This leads to
the inequality
1 1 2
a + b + c - ab - bc - ca 2: (a + b + c) - "3 (a + b + c) 2 = 1- "3 = "3.
Hence, at least one of the summands from among a(l- b), b(l- c), c(l- a) has
2. 2
to be greater than or equal to 9; (otherwlse lhs would be less than "3.)
232. (a) Since x and y are positive, we have y ~ x - x 3 - y3 < x. We also

have x - x 3 2: y + y3 > 0, so that x(l - x 22)) > O. This gives x < 1 and
thus y < x < 1, proving part (a).
(b) Again, we have x 3 + y3 ~ X - y, so that
x-y
x2
x2
_
-
xy
xy
+ Yy22<-~
+ __
+_y
xx+y
This implies that
2 2 x - Y x - Y + xy(x + y) .
x +Y < - - + xy =
- x+y x+y
But xy(x + y) < 1· y . (1 + 1) = 2y and hence
2 2 X - Y + 2y x +Y
x +y < =--=l.
x+y x+y
This proves (b).
Solutions 383
233. Suppose the equation x 2 + X + 4,\ = 0 has no real roots. Then 1-16'\
+x+4'\ < O.
This implies that
1 - 16(a 3 + a 2 bc) < 0, 1 - 16W + ab2 c) < 0, 1 - 16(c3 + abc2 ) < O.

Observe that
1 - 16(a3 + a 2 bc) < 0 ===}

==} 1-16a22 (a+bc) < <00
==}
===} 1 - 16a (1 - b - c + bc) < 0
22
==}
===} b)(1 - c) < 0
1 - 16a 22 (1 - b)(l
11 2
==} -16 <
<aa 2 (1 - b)(l - c) ..
(l-b)(I-c)
16
1 2 1 ?2
16 1~3.
1!3.
However, 00 < a < 1 implies that a(l ::; 1/4. Hence

a(1 - a) ~
a 2 b2c2 (1- a)2(1- b)2(1- C)2 = (a(l- a))2(b(1- b))2(c(1- C))2 ~ 1~3'
a contradiction. We conclude that the given equation has real roots.
234. We use the trivial inequalities a 2 + 1 2:2 2a, bb2 + 1 2:

2a, 2 2b and c2 + 1 2:2 2c.
Hence we obtain
a 22 + 1 b2 + 1 c2 + 1 2c
- - +- - +- - > -2a- + -2b- + --.
b+c c+a a+b - b+c c+a a+b
2a 2b 2c
--+
b+c c+a
+-- >3.
a+b-
Adding 6 both sides, this is equivalent to
(2a+
(2a + 2b + 2c) ( -
2b+2c) 1- +1
-b- -- 1 ) 2:
+ --b 29.9.
+c c+a a+
b+c
Taking x == b + c, y =
= c + a, Z = a + b, this is equivalent to
=
(x+Y+z) 1
x Y
y Z
(1 1 1)
1 1) 2:9,
(x+y+z) ( -+-+- 29,
384 Solutions
which is a consequence of the AM-GM inequality.

Alternate Solution:
The substitutions b + c = x, c + a = y, a + b = z lead to
"""'
L -2a- = """'
L y+ z - x = """'
L (x
- + -xy) - 3 :::: 6 - 3 = 3;
cyclic
b+c cyclic
x cyclic
y
we have used the AM-GM inequality in the last leg.
235. Consider the polynomial P(x) = x 2 - (a + b + c)x + (ab + bc + ca). Then

we see that
P(d) d2 - (a + b + c)d + (ab + bc + ca)

d2 - a(d - b) - b(d - c) - c(d - a).
Thus, we have to prove that P(d) :::: O. On the other hand,
xP(x) - abc = (x - a)(x - b)(x - c).
Thus
(d -- a)(d
(d) = (d
P(d) b)(d -- c)
a)(d -- b)(d c) + abc
abc> abc abc> O.
P = d
d - d::::
:::: d - O.
236. One solution using normalisation was given on page 89. We give here
three more solutions.
Solution 1: We begin with the observation that
x 2 +xy+y2 = ~(x+y)2 + ~(X_y)2 2: ~(X+y)2,

and similar bounds hold for y2 + yz + Z2, z2 + ZX + x 2. Hence,
3(x 2 + xy + y2) (y2 + yz + Z2) (Z2 + ZX + x 2 )

81 2 2 2
:::: 64(X+ Y) (y+z) (z+x) .

9
(x+y+z)(xy+yz+zx) s 8(x+ y )(y+z)(z+x).
8{x + Y + z)(xy + yz + zx) S 9(x + y)(y + z)(z + x).

However, we note that
(x+y)(y+z)(z+x) = (x+y+z)(yz+zx+xy) -xyz.

Solutions 385
Thus, the required inequality takes the form
(x + Y)(Y
y)(y + z)(z + x) ~
:::: 8xyz.
8xyz .
This follows from the AM-GM inequalities;
x + yY ~2: 2Vxfj,
2FY, y +z ~
Y :::: 2#, z +x 2..[ZX.
:::: 2,JiX.
~
Solution 2: Let us introduce x + y = c, y + z = a and z + x = b. Then a, b, c

are the sides of a triangle. If s8 = = (a + b + c)/2,
c)/2 , then it is easy to calculate
x = s8 -- a, y = s8 -- b, zZ = s8 -- c and x + y + zZ = s.
8. \Ve
xx 22 + xy + y2 = (x + y)2
y) 2 -- xy
~(c + a - b)(cc+b-a)
2
= c2 - -(c+a-b
=c
1 ~C2
+ b - a) == -c)( 3 2
~(a 3~C2.
1 - b)22 > -c.
+ -(a-b) 2
4 4 4 - 4
Moreover, xy + yz + zx
Moreover, z x = (s a)(8 - b) + (s
(8 -- a)(s b)(8 - c)
(8 -- b)(s + (s
(8 -- C)(8 - a). Thus
c)(s
it is sufficient to prove that
9
s8 2)s a)(8 - b) ~
L)8 - a)(s :::; ~abc.
Sabc.
But, L:(s a)(8 - b) = r(4R + r),

L:(8 - a)(s r) , where r,r,R
Rare
are respectively the in-radius,
in-radius,
4Rr8. Thus
the circum-radius of the triangle whose sides are a, b, c, and abc = 4Rrs.
the inequality reduces to
9
r(4R + r) ~:::; 2Rr.
:::; R, which is Euler

This simply reduces to 2r ~ Euler's
's inequality.
Solution 3: Dividing throughout by xX 22yy2 Z2, the inequality may be written in
2 Z2,
the form
((
x
3+-+-+-+-+-+-
y
yy yy
xx Zz y y x x Z z
zZ zZ
:::; 3 (( -+-+1
3 + - + - + - + - + - + - ) 2 ~3
xx
- + - + 1) ( -+-+1
y yx x
x
- + - + 1) ( -+-+1
Z zy y
- +- + )
x zx z
y y Zz zZ
1)x
x
If we set
x y y Z z xx
- +- + 2,,
= aa+2 ~- + ~- == b+
b + 2, - + - = c+ 2,
-+-=c+2
y x zZ Y y x Zz '
a, b, c are non-negative and the inequality to be proved is
then a,
(9
(9 + a + b + C)2 :::; 3(a + 3)(b + 3)(c + 3).
~ 3).
This reduces to
a2 + b2 + c2 ~:::; 3abc + 7(ab + bc + ca) + 9(a + b + c).

386 Solutions
However, it is easy to see that
or
( -+-
.r
Y x
x z
z)y (z-+-
y) (y-+-
z x
x)z
Y x z 2
x2
+ 2y + 2x +
y2 y2
2Z
Z2 Z2
+ 2y + 2x +
x2
2'
Z
This takes the form
a2 + b2 + c2 =
= abc + 2(ab + bc + ca).
Thus the inequality reduces to
2abc + 5(ab + bc + ca) + 9(a + b + c) ~ O.
This follows from the non-negativity of a, b, c.
237. Suppose x and y are positive real numbers. Then re arrangement inequal-
ity implies that
x In x + y In y ~ x In y + y In x.
Exponentiation gives xXyY ~ xYyx. Thus, we have
aabb ~ abba, bbcc ~ bCc b, cCa a ~ cbaa c.
Multiplying these out, we get
(aabb Cc)2 ~ ab+cbc+aca+b.
It follows that
(aabb Cc)3 ~ ab+c+abc+a+bca+b+c,
which reduces to
aabbCc ~ (abc)(a+b+ c)/3.
Alternately, we can use the weighted AM-HM inequality:
(aabbCc) l/(a+b+c) ~ a+b+c a+b+c

abc 3
-+-+-
abc
> (abc)1/3,
where we have used the AM-GM inequality in the last stage. This gives the
required inequality.
238. Let the positive roots of
8p 2
P(x) = x 4 - _ x3 + 4qx 2 - 3px + p2 = 0,
q
Solutions 387
be a, b, c, d. Then we have
8p 2
a+b+c+d
q
+ + + + +
ab ac ad bc bd cd
~+oc+~+k+W+~ 4q,
+ + +
abc abd acd bcd
~+~+~+~ 3p,
2
abcd p.
Since a, b, c, d are positive, p, q are also positive. Now using the AM-GM in-
equality, we have
8 2
L = a + b + c + d 2: 4(abcd)1/4 = 4Jp.
q
2p 2 2: q,jP. Similarly,
Thus, 2p
4q = ab + ac + ad + bc + bd + cd 2:> 6(abcd)1/2 = 6p,

3p = abc + abd + acd + bcd 2:
> 4(abcd)3/4 = 4pJp.
= b == c == d. Multiplying out
Equality holds in each of these, if and only if a =
all the three inequalities, we obtain
(2 p2)(4q)(3p) 2: (qJp) (6p) (4p,jP).

(2p2)(4q)(3p)
We see that this is an equality, so that a = b == c == d. Let us denote the

common value by a. Thus we obtain
8p 22
4a
q
6a 22 4q,
4a 3 3p,
a4 p2.
Using the first two relations, we get
3 4p
4p2
a == ---
-
3
Using the third relation, we have
16p22 = 2p,
16p
-3- = 2p,
3
= 9/16. Thus, a == 3/4 and hence q == 27/32.
which gives p =
388 Solutions
239. Observe that aI!(al - 1) > S is equivalent to ar > (al + a2 + a3)(al -1),
which in turn is equivalent to
1 al
> - -al
---> ---
a2 + a3 al + a2 + a3
Using similar inequalities from the other two, we get
L
""
~ --->
1 > al + a2 + a3 = l.
=l.
.
cychc
al + a2 al + a2 + a3
Bibliography
[1] D. S. Mitrinovic Analytie Inequalities, Springer-Verlag, 1970.
[2] D. S. Mitrinovic, J. E. Pecaric, A. M. Fink, Classieal and New Inequalities

in Analysis, Kluwer Academic Publisher, 1993.
[3] G. H. Hardy, J.E. LittIewood, .1. Polya. Inequalities, Cambridge University

Press, 1934.
[4] Titu Andreescu, Zuman Feng, 101 Problems in Algebra, AMT Publishing,
2001.
[5] O. Bottema, et al., Geometrie Inequalities, Wolters-Noordhoff Publishing,

Groningen, 1969.
[6] D. S. Mitrinovic, .1. E. Pecaric, V. Volenec, Reeent Advanees in Geometrie
Inequalities, Kluwer Academic Publisher, 1989.
[7] Arthur Engel, Problem-Solving Strategies, Springer-Verlag, 1998.

[8] E. J. Barbeau, Polynomials, Springer-Verlag, 1989
[9] S. Savachev, Titu Andreescu, Mathematieal Miniatures, Mathematical As-
sociation of America, 2002.
389
Abbreviations
We have used the following abbreviations in the text.
• AMM - American Mathematical Monthly;

• CRUX - Crux Mathematicorum;
•• IMO
IMO -~- International Mathematical Olympiad;
•• CRMO --
CRMO - Central Regional Mathematical Olympiad(this is the
first level examination for selecting students to rep-
resent India in IMO);
IN MO -~- Indian National Mathematical Olympiad(this is the
•• INMO
second level examination for selecting students to
represent India in IMO).
Index
Absolute value, 2 AM-GM-HM,6

weighted,9
Conjugate, 2 Bernoulli's,
Bernoulli's , 45, 68
Convex set, 36 Bohr's, 334
Carleman's, 319
Equation
Cauehy-Sehwarz, 11
a, SS - b, S - c, 97
for s - a,
Chebyshev's, 17
for altitudes, 97
with weights, 19
for ex-radii, 98
Erdös-Mordell, 131
for the sides of a triangle, 96
Euler's, 94, 110
llO
Euclidean
Euelidean distanee,
distanee , 30
geometrie, 94
Euclidean
Euelidean norm , 30
Hölder's, 27
Hölder's,
Formula weighted, 30
for angle biseetor,
bisector, 95 Hardy's,62
for median, 95 Jensen's , 38
for ex-radii, 95 Jordan's, 67
Heron 's, 72, 95
Heron's, Minkowski's,
Minkowski's,30 30
Stewart's, 95 Nesbitt's,
Nesbitt's, 100
Function
Ftmetion Newton's, 48
eoncave,
concave, 35 re arrangement , 21
rearrangement,
eonvex,
eonvex,34 34 for convex funetions, 41
equivalent eharaeterisations,
eharaeterisations RMS, 12
35 ' Schur's,80
Sehur's,80
geometrie interpretation, 35 Weitzenböek's, 108
twice
twiee differentiable, 37 Young's,67
strietly eoneave, 35
Law of trichotomy,
triehotomy, 1
eonvex, 34
strietly eonvex,
Malafatti circles,
cirdes, 196
Geometry
Mean
basie results, 95
arithmetie,
arithmetic, 3
notations, 94
geometrie, 3
Ineqllality
Inequality harmonie, 6
harmonie,6
Hadwiger-Finsler, 108
AM-GM,
AM-GM,33 Ordering on IR, 1
weightcd,9
weighted,9
391
392 Index
Problems Tournament of Towns, 86

AMM, 148, 193, 196, 211 UK, 212, 213
APMO, 196, 198 Ukraine, 187, 188, 216
Armenia, 198 USA, 189, 199, 209, 217
Austria, 188 USSR, 88, 156, 216
Balkan Olympiads, 187, 197,212 Vietnam, 206
Belarus, 211 Wolsehaum County, 206
Bosnia and Herzegovina, 188 Proposed for
Bulgaria, 171, 194, 196, 197, IMO-1991, 205
201, 207, 215, 217 IMO-1997,216
Canada, 207 IMO-1998,214
China, 16 IMO-2001, 190, 194
CRMO, 78, 216
Rule
CRUX, 167, 190, 191, 196,200,
eosine, 95
209, 210, 214
half-angle, 95
Czeeh and Slovak, 187, 190
sine, 95
Estonia, 188, 199
Germany, 197 Short-list
Greeee, 213 IMO-1989,201
Hungary, 188, 217 IMO-1990, 215
IMO, 58, 80, 87, 90, 155, 180, IMO-1993, 13, 175, 201
181, 189, 194, 196, 197, 199, IMO-2002, 204
201,203,205,207,210,213, IMO-2004, 213
214
INMO, 79, 89, 191, 215, 217 Teehniques
Iran, 190, 193, 212 homogenisation, 86
Ireland, 187, 197, 198 induetion, 51
Japan, 204, 212 known inequalities, 54
KÖMAL, 210 majorisation, 82
Moldova, 214 normalisation, 89
Nordie eontest, 170, 215 quadratie polynomials, 75
Poland, 188, 192 Ravi transformation, 78
PUTNAM, 208, 210 trigonometrie, 73
Romania, 165, 171, 184, 187, use of ealculus, 65
189, 193,197, 198,207,210~ Theorem
212, 214 Appolonius', 95
Russia, 217 Muirhead's, 84
Selection Tests, 160, 195, 201, Stewart's, 95
206, 207, 215 Stolarsky's, 91
South Afriea, 205, 213 Triangle inequality, 2
South Korea, 197
St. Petersburg Olympiads, 198
Taiwan, 178, 191, 214
Thailand, 190, 195
Texts and Readings in Mathematics
1. R. B. Bapat: Linear Algebra and Linear Models (Second Edition)

2. Rajendra Bhatia: Fourier Series (Second Edition)
3. C. Musili: Representations of Finite Groups
4. H. Helson: Linear Algebra (Second Edition)
5. D. Sarason: Complex Function Theory (Second Edition)
6. M. G. Nadkarni: Basic Ergodie Theory (Second Edition)
7. H. Helson: Harmonie Analysis (Second Edition)
8. K. Chandrasekharan: A Course on Integration Theory
9. K. Chandrasekharan: A Course on Topological Groups
10. R. Bhatia (ed.): Analysis, Geometry and Probability
11. K. R. Davidson: C* - Aigebras by Example
12. M. Bhattacharjee et al.: Notes on Infinite Permutation Groups
13. V. S. Sunder: Functional Analysis - Spectral Theory
14. V. S. Varadarajan: Algebra in Ancient and Modern Times
15. M. G. Nadkarni: Speetral Theory of Dynamical Systems
16. A. Borei: Semisimple Groups and Riemannian Symmetrie Spaces
17. M. Mareolli: Seiberg - Witten Gauge Theory
18. A. Botteher and S. M. Grudsky: Toeplitz Matrices, Asymptotie
Linear Algebra and Functional Analysis
19. A. R. Rao and P. Bhimasankaram: Linear Algebra (Second Edition)
20. C. Musili: Aigebraic Geometry for Beginners
21. A. R. Rajwade: Convex Polyhedra with Regularity Conditions
and Hilbert's Third Problem
22. S. Kumaresan: A Course in Differential Geometry and LieGroups
23. Stef Tijs: Introduetion to Game Theory
24. B. Sury: The Congruence Subgroup Problem
25. R. Bhatia (ed.): Connected at Infinity
26. K. Mukherjea: Differential Calculus in Normed Linear Spaces
(Second Edition)
27. Satya Deo: Aigebraic Topology: A Primer (Correeted Reprint)
28. S. Kesavan: Nonlinear Functional Analysis: A First Course
29. S. Szab6: Topies in Factorization of Abelian Groups
30. S. Kumaresan and G. Santhanam: An Expedition to Geometry
31. D. Mumford: Lectures on Curves on an Aigebraic Surface (Reprint)
32. J. W. Milnor and J. D. Stasheff: Charaeteristie Classes (Reprint)
33. K. R. Parthasarathy: Introduetion to Probability and Measure
(Correeted Reprint)
34. A. Mukherjee: Topics in Differential Topology
35. K. R. Parthasarathy: Mathematical Foundations of Quantum
Meehanics
36. K. B. Athreya and S. N. Lahiri: Measure Theory
37. Terence Tao: Analysis I
38. Terence Tao: Analysis 11
39. W. Decker and C. Lossen: Computing in Aigebraic Geometry
40. A. Goswami and B. V. Rao: A Course in Applied Stochastic
Processes
41. K. B. Athreya and S. N. Lahiri: Probability Theory
42. A. R. Rajwade and A. K. Bhandari: Surprises and Counterexamples
in Real Function Theory
43. G. H. Golub and C. F. Van Loan: Matrix Computations (Reprint of the
Third Edition)
44. Rajendra Bhatia: Positive Definite Matrices
45. K. R. Parthasarathy: Coding Theorems of Classical and Quantum
Information Theory
46. C. S. Seshadri: Introduction to the Theory of Standard Monomials
47. Alain Connes and Matilde Marcolli: Noncommutative Geometry,
Quantum Fields and Motives
48. Vivek S. Borkar: Stochastic Approximation: A Dynamical Systems
Viewpoint

Inequalities

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Inequalities

Uploaded by

Copyright:

Available Formats

TEXTS AND READINGS

Hindustan Book Agency (lndia)

ISBN 978-81-85931-88-3 ISBN 978-93-86279-43-9 (eBook)

Copyright © 2009, Hindustan Book Agency (lndia)

No part of the material protected by this copyright notice may be

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

4 Applieations involving inequalities ...................... 155

5 Problems on inequalities .................................. 187

6 Solutions to problems ...................................... 219

(9) if a and b are positive and a2 < b2 , we have a < b.

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

(1) I - xl = lxi; Ixl 2 = x 2 ; Ixyl = Ixllyl; Ix/yl = Ixl/lyl for y -I O.

(3) IIXI-lyll ~ Ix - yl·

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

(2) IZ1Z21 = Izd IZ21;

1.2 Arithmetic mean-Geometric mean inequality

al + az 2: (ala2) 1/2 . (1.3)

aj + a2 2: (ala2) 1/2 , a3 + a4 2: (a3 a4 ) 1/2 .

(al; a2 ) + (a3 ; a4)

A simple manipulation now yields A 2: a1

Solution: Using the inequality (1 + aj) ;::

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

5 (2n - 1)) (2 . 4 . 6 ... (2n))

Using the AM-GM inequality, we get

Combining these two inequalities, one can obtain

(2n)! < (n+ l)nnn = {n(n+ l)r,

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

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

since c < a + b by the triangle inequality. •

A(al,a2' ... ,an) ~ (al,a2' ... ,an) ~H(al,a2' ... ,an),

or briefly AM ~;:::: GM ~HM.

Example 1.4. For any four positive real numbers al,a2,a3,a4,

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

Solution: Observe that

Solution: Introduce new variables x, y, z by

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

(x;y)(y;z)(z;x) ~ ~(XYZ)2/3. (1.6)

This ean be rewritten using x + y + z == 1 in the following form:

Sinee x + y + z == 1, this follows from the AM-GM inequality.

klal + k2a2 + ... + kna n

respeetively. Now the AM-GM inequality gives

k 1a1 + k2a2 + ... + knan kn )

Suppose we have positive rational numbers r1, ,rn. Then we ean

where m and k j , 1 :S j :S n are natural numbers. It is not hard to see that

rIal + r2a2 + ... + rna n

Example 1. 7. Let a, b, e be positive real numbers. Prove that

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

which gives the required inequality.

bl + b2 + ... + bn :::: a1 + a2 + ... + an· (1.10)

a1 (h) + (2.2) + ... +

(G:f (!:f ... (!:r) .,+.,~+ ••

b1b2 ... b)aj-aj+l >1

far 1 :::::: ::; n. Since

( ~)al (b2)a2 ... (bn)a n

It follows from the weighted AM-GM inequality that

b1 + b2 + ... + bn 2': a1 + a2 + ... + an·

If L kk = L~=l (Aj - 1) then

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

lac + bdl ::; Va + b2 vc

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

Equality holds if and only if there exists a constant A such that aj = Ab

Proof: Let us write

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

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

(a1,j) (a2,j) (a3,j) . .. (ar,j)