Venkatachala
Inequalities
An Approach Through Problems
Second Edition
123
B. J. Venkatachala
(emeritus) Homi Bhabha Centre
for Science Education
Mumbai, Maharashtra
India
Eight years have elapsed after the ﬁrst edition of this book has come out.
More inequalities have been generated as a need for various national math-
ematical olympiads and International Mathematical Olympiads. New books
have emerged and newer results have been added during these years. This has
prompted me to revise the book adding more material to the existing one. New
problems based on the old methods, new solutions to the old problems and new
methods have been added to the book to make it richer.
A section on proving symmetric inequalities has been added to chapter
2. More problems have been solved based on these methods. More than 70
problems have been added to chapter 5 and their solutions have been given
in chapter 6. It is my sincere wish that the new edition will help all those
who aspire to study inequalities for mathematics competitions and for general
interest.
I thank all my colleagues and students who have enriched me with their
lively conversations and made this revision possible.
B J Venkatachala,
Professor (Retired),
HBCSE, TIFR,
MUMBAI-400088, INDIA.
Contents
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2.11 Homogenisation . . . . . . . . . . . . . . . . . . . . . . . . . . 88
2.12 Normalisation . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
2.13 Stolarsky’s theorem . . . . . . . . . . . . . . . . . . . . . . . . 93
8 Abbreviation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 516
9 Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 517
Chapter 1
Some basic inequalities
1.1 Introduction
As we all know, one of the important properties of real numbers is comparability.
We can compare two distinct real numbers and say that one is smaller or larger
than the other. There is an inherent ordering < on the real number system
R which helps us to compare two real numbers. The basic properties of this
ordering on R are:
(i) Given any two real numbers a and b, one and only one of the following
three relations is true:
a < b or a = b or a > b;
(law of trichotomy.)
(ii) a > 0 and b > 0 imply a + b > 0;
(iii) a > 0 and b > 0 imply ab > 0.
Any new inequality we derive is totally dependent on these basic properties.
These properties are used to derive the arithmetic mean-geometric inequality,
the Cauchy-Schwarz inequality, Chebyshev’s inequality, the rearrangement in-
equality, Hölder’s and Minkowski’s inequalities, and are also used in the study
of convex and concave functions. Here are some easy consequences of these
fundamental properties of ordering on R:
(1) a < b, then a + c < b + c, for any real c;
(2) a < b and c > 0 give ac < bc; if c < 0, we have ac > bc;
(3) 0 < a < b implies 0 < 1/b < 1/a;
(4) a < 0 and b < 0, then ab > 0; a < 0 and b > 0 imply ab < 0;
(5) a < b and b < c together imply a < c (transitivity);
(6) if ac < bc and c > 0, we have a < b;
(7) 0 < a < 1 implies a2 < a; if a > 1, we have a2 > a;
(8) for any real a, a2 ≥ 0;
(9) if a and b are positive and a2 < b2 , we have a < b.
2 Inequalities
Note that |x| ≥ 0 and |x| = 0 if and only if x = 0. The absolute value function
x → |x| has some nice properties:
(2) |x + y| ≤ |x| + |y| and equality holds if and only if x and y have the same
sign. This is known as the triangle inequality.
(3) |x| − |y| ≤ |x − y|.
(3) |z1 + z2 | ≤ |z1 | + |z2 | and equality holds if and only if z1 = λz2 for some
positive real number λ, or one of z1 , z2 is zero.
√ √
Then we know that a and b are meaningful as real numbers. Since the
√ √ 2
square of a real number is always non-negative, a − b ≥ 0. This may be
rewritten in the form
a+b √
≥ ab. (1.1)
2
a+b √
The real number is called the arithmetic mean of a and b; similarly ab
2
is known as the geometric mean of a and b. Thus the property that x2 ≥ 0
for a real number x implies that the arithmetic mean of two non-negative
real numbers cannot be smaller than their geometric mean. Moreover, the
derivation also shows that equality in (1.1) holds if and only if a = b.
This may be considered in a general setting. Starting with n non-negative
real numbers a1 , a2 , . . . , an , we deﬁne their arithmetic mean A(a1 , a2 , . . . , an )
and geometric mean G(a1 , a2 , . . . , an ) by
a1 + a2 + · · · + an
A a 1 , a2 , . . . , an = ,
n
1/n
G a 1 , a2 , . . . , an = a1 a2 · · · an .
As in the case of two numbers, a comparison between these two means leads
to the AM-GM inequality.
Proof: There are several proofs of this classical theorem. We give here a clever
induction proof which was originated by Cauchy. It also illustrates how one
can go ahead in an induction proof leaving out the validity of induction for
some numbers and subsequently prove the validity for missing numbers using
an interpolation argument.
We have already obtained the result for n = 2;
a1 + a2 1/2
≥ a1 a2 . (1.3)
2
a1 + a2 1/2 a3 + a4 1/2
≥ a1 a2 , ≥ a3 a4 .
2 2
4 Inequalities
This leads to
a1 + a2 a3 + a4
+
a1 + a2 + a3 + a4 2 2
=
4 2
1/2 1/2
a1 a2 + a3 a4
≥
2
1/2 1/2 1/2
≥ a1 a2 a3 a4
1/4
= a1 a2 a3 a4 .
Thus the inequality (1.2) is obtained for n = 4. It may again be observed that
equality holds here if and only if a1 = a2 , a3 = a4 and a1 a2 = a3 a4 . Since all
numbers are non-negative, it follows that a1 = a2 = a3 = a4 . Now this can be
used to prove (1.2) for n = 8. Now induction shows that (1.2) holds for n = 2k
for all k ∈ N. Here again the condition for equality is that all the 2k numbers
be equal.
Now consider any n non-negative real numbers a1 , a2 , . . . , an . We choose a
natural number k such that 2k−1 ≤ n < 2k . Put
a1 + a2 + · · · + an
A= ,
n
This reduces to
1/2k
2k −n
A≥ a1 a2 · · · an A .
Example 1.2. Show that for any natural number n > 1, the inequality
n
(2n)! < n(n + 1) ,
holds good.
it follows that
G a 1 , a2 , . . . , a n ≥ H a 1 , a2 , . . . , an .
Thus we have for n positive real numbers a1 , a2 , . . . , an , the inequality
A a 1 , a 2 , . . . , a n ≥ G a 1 , a 2 , . . . , a n ≥ H a 1 , a2 , . . . , a n ,
Example 1.4. For any four positive real numbers a1 , a2 , a3 , a4 , prove the inequal-
ity
a1 a2 a3 a4
+ + +
a1 + a2 a2 + a3 a3 + a4 a4 + a1
a1 a2 a3 a4
≤ + + + .
a2 + a3 a3 + a4 a4 + a1 a1 + a2
Solution: Adding
a2 a3 a4 a1
+ + +
a1 + a2 a2 + a3 a3 + a4 a4 + a1
Basic Inequalities 7
to both sides, the inequality can be written in the following equivalent form:
1 1 1 1
a2 + + a3 +
a3 + a4 a1 + a2 a4 + a1 a2 + a3
1 1 1 1
+ a4 + + a1 + ≥ 4.
a1 + a2 a3 + a4 a4 + a1 a2 + a3
1 1 4
+ ≥ ,
a3 + a4 a1 + a2 a1 + a2 + a3 + a4
1 1 4
+ ≥ .
a4 + a1 a2 + a3 a1 + a2 + a3 + a4
bc ca ab
+ + = s,
b+c c+a a+b
where s is the semi-perimeter of the triangle. Prove that the triangle is equilateral.
Here the AM-HM inequality has been used. Thus equality holds in the AM-HM
inequality and hence a = b = c.
Example 1.6. Suppose a, b, c are positive real numbers. Prove the inequality
2/3
a+b b+c c+a a+b+c
≥ abc . (1.5)
2 2 2 3
a b c
x= , y= , z= ,
a+b+c a+b+c a+b+c
8 Inequalities
It follows that
r 1
r1 a 1 + r2 a 2 + · · · + rn a n 1 +r2 +···+rn
≥ ar11 ar22 · · · arnn .
r1 + r2 + · · · + r n
It is possible to further extend this inequality using a continuity argument. If
λ1 , λ2 , . . . , λn are positive reals and a1 , a2 , . . . , an are non-negative reals, then
λ 1
λ1 a 1 + λ2 a 2 + · · · + λn a n 1 +λ2 +···+λn
≥ aλ1 1 aλ2 2 · · · aλnn . (1.9)
λ1 + λ2 + · · · + λn
This is known as the generalised or the weighted arithmetic mean- geometric
mean inequality. With obvious modiﬁcations, this can be further carried to
the weighted arithmetic mean- geometric mean-harmonic mean inequality.
aa bb cc ≥ (abc)(a+b+c)/3 .
b1 + b 2 + · · · + b n ≥ a 1 + a 2 + · · · + a n . (1.10)
Solution: We may assume that aj > 0 for all j; otherwise delete those par-
ticular aj ’s which are equal to 0. This does not aﬀect the inequality. Consider
the set
b1 b2 bn
, ,... , ,
a1 a2 an
10 Inequalities
where each bj /aj is attached with the weight aj . The weighted arithmetic
mean is
a1 a1 + a2 a2 + · · · + an abnn
b1 b2
b1 + b 2 + · · · + b n
=
a1 + a2 + · · · + an a1 + a2 + · · · + an
and the weighted geometric mean is
a 1 a 2 an a1 +a2 +···+a
1
n
b1 b2 bn
··· .
a1 a2 an
for 1 ≤ j ≤ n. Since
a 1 a 2 a n
b1 b2 bn
···
a1 a2 an
a n an −an−1 a2 −a1
b1 b2 · · · bn b1 b2 · · · bn−1 b1
= ··· ,
a1 a2 · · · an a1 a2 · · · an−1 a1
we conclude that a 1 a 2 a n
b1 b2 bn
··· ≥ 1.
a1 a2 an
It follows from the weighted AM-GM inequality that
b1 + b 2 + · · · + b n ≥ a 1 + a 2 + · · · + a n .
Alternate Solution:
Put λj = bj /aj , 1 ≤ j ≤ n. We have to show that
n
aj λj − 1 ≥ 0.
j=1
k
If Lk = j=1 λj − 1 then
Lk = λ 1 + λ2 + · · · + λk − k
b1 b2 bk
= + + ··· + −k
a1 a2 ak
1/k
b 1 b2 · · · bk
≥ k − k ≥ 0,
a1 a2 · · · ak
Basic Inequalities 11
Thus the Cauchy-Schwarz inequality here represents the fact that the absolute
value of the dot product of two vectors does not exceed the product of their
lengths. This property extends to higher dimensions giving us the general form
of the Cauchy-Schwarz inequality.
Theorem 2. Let a1 , a2 , a3 , . . . , an and b1 , b2 , b3 , . . . , bn be any two sets
of real numbers. Then
1/2 1/2
n n n
a b ≤ a 2
b 2
. (1.11)
j j j j
j=1 j=1 j=1
Equality holds if and only if there exists a constant λ such that aj = λbj , for
1 ≤ j ≤ n.
12 Inequalities
= B 2 A + BC 2 − 2BC 2
= B BA − C 2 .
Since B > 0, it follows that C 2 ≤ AB. We also infer that equality holds if and
only if Baj − Cbj = 0 for all j, 1 ≤ j ≤ n. Taking λ = C/B, the condition for
equality that aj = λbj for 1 ≤ j ≤ n is obtained.
Solution: We write
a c b a b b c c a
+ + = · + · + · .
c b a b c c a a b
Now applying the Cauchy-Schwarz inequality to the sets
a b c b c a
, , and , , ,
b c a c a b
we get
1/2 1/2
a c b a2 b2 c2 b2 c2 a2
+ + ≤ 2
+ 2+ 2 2
+ 2+ 2
c b a b c a c a b
2 2 2
a b c
= + 2 + 2.
b2 c a
Solution: We may take l = 1 and m = 2; the proof for any other pair of
indices is the same. Suppose A ≥ 2a1 a2 . Then we see that
n
n
A+ a2k ≥ 2a1 a2 + a21 + a22 + a2k
k=1 k=3
2
n
= a1 + a2 + a2k
k=3
n 2
1
≥ ak .
n−1
k=1
(Here we haveapplied the Cauchy-Schwarz
inequality to the (n−1) tuples a1 +
a2 , a3 , . . . , an and 1, 1, . . . , 1 .) But this contradicts the given condition. We
conclude that A < 2a1 a2 .
Solution: We write
√
a √ √
a= √ a b + 2c + 3d ,
cyclic cyclic
b + 2c + 3d
to get
2 a
a ≤ a(b + 2c + 3d) .
b + 2c + 3d
cyclic cyclic cyclic
However, we have
a(b + 2c + 3d) = 4(ab + ac + ad + bc + bd + cd).
cyclic
(a + b + c + d)2 2
≥
4(ab + ac + ad + bc + bd + cd) 3
and hence a 2
≥ .
b + 2c + 3d 3
cyclic
Basic Inequalities 15
n
n
a2j ≤ b2j .
j=1 j=1
k
k
ak − ak+1 aj ≤ ak − ak+1 bj ,
j=1 j=1
This simpliﬁes to
Solution: We have
⎛ ⎞2 ⎛ ⎞2
a
⎝ a⎠ = ⎝ (a + b)(a + c)⎠
cyclic cyclic
(a + b)(a + c)
⎛ ⎞⎛ ⎞
a 2
≤ ⎝ ⎠⎝ (a + b)(a + c)⎠ ,
(a + b)(a + c)
cyclic cyclic
16 Inequalities
This simpliﬁes to
a2 + b2 + c2 ≥ ab + bc + ca,
which again follows from the Cauchy-Schwarz inequality.
However,
"
n 1/n 1
n
n−1
(1 − aj ) ≤ (1 − aj ) = .
j=1
n j=1 n
Thus we obtain,
n
1 n
≥n .
j=1
1 − aj n−1
On the other hand, the Cauchy-Schwarz inequality gives
n
√
n
1 − aj ≤ n! (1 − aj ) = n(n − 1).
j=1 j=1
Basic Inequalities 17
We thus get
√
n
1
n
n n
− 1 − aj ≥ n − n(n − 1) = √ .
j=1
1 − aj j=1 n−1 n−1
We also have ⎛ ⎞2
n
√
n
⎝ aj ⎠ ≤ n aj = n.
j=1 j=1
Thus
√
n
√
n≥ aj
j=1
and hence √
n
aj n 1 n
√
≥√ ≥√ aj .
j=1
1 − aj n−1 n − 1 j=1
n
n 21
n 12
a j bj ≤ a2j b2j .
j=1 j=1 j=1
This shows that the Cauchy-Schwarz inequality relates the arithmetic means of
a21 , a22 , . . . , a2n and b21 , b22 , . . . , b2n with that of the numbers a1 b1 , a2 b2 , . . . , an bn .
How do we relate the arithmetic means of the numbers a1 , a2 , . . . , an and
b1 , b2 , . . . , bn with that of a1 b1 , a2 b2 , . . . , an bn . Chebyshev’s inequality answers
this question.
Theorem 3. Let a1 , a2 , . . . , an and b1 , b2 , . . . , bn be two sets of real num-
bers. Suppose either
a1 ≤ a2 ≤ · · · ≤ an and b1 ≤ b2 ≤ · · · ≤ bn ;
or
a1 ≥ a2 ≥ · · · ≥ an and b1 ≥ b2 ≥ · · · ≥ bn ;
18 Inequalities
i.e., both the sequences ak and bk are non-decreasing or both non-increasing.
Then the inequality
a1 + a2 + · · · + a n b1 + b2 + · · · + bn a 1 b1 + a 2 b2 + · · · + a n bn
≤
n n n
holds. The inequality is strict unless at least one of the sequences is a constant
sequence.
Proof: We have
n n
n
n
a j bj − a j bk = naj bj − aj bk
j=1 k=1 j=1 k=1
n
n
n
= n a j bj − aj bk .
j=1 j=1 k=1
n
n
n
n
n
a k b k − a k bj =n a k bk − ak bj .
j=1 k=1 k=1 k=1 j=1
n
n
n
n a k bk − ak bj
k=1 k=1 j=1
n
1
n
= a j b j − a j b k + a k bk − a k bj
2 j=1 k=1
1
n n
= a j − a k b j − bk .
2 j=1 k=1
Since aj − ak bj − bk ≥ 0 whether both the sequences are non-decreasing or
both non-increasing, we get
n
n
n
n a k bk − ak bj ≥ 0.
k=1 k=1 j=1
Here equality holds if and only if for each pair of indices j, k either aj = ak or
bj = bk . In particular taking j = 1, k = n we see that a1 = an or b1 = bn . It
follows that either aj is a constant sequence or bj is a constant sequence.
Basic Inequalities 19
Remark 3.1: If the sequences aj and bj are oppositely ordered, i.e., either
a1 ≤ a2 ≤ · · · ≤ an and b1 ≥ b2 ≥ · · · ≥ bn , or a1 ≥ a2 ≥ · · · ≥ an and
b1 ≤ b2 ≤ · · · ≤ bn , then the inequality reverses;
a1 + a2 + · · · + an b1 + b2 + · · · + bn a 1 b1 + a 2 b2 + · · · + a n bn
≥ .
n n n
1
n
≤ a1,j a2,j a3,j · · · ar,j .
n j=1
The proof runs on similar lines. This gives Chebyshev’s inequality with weights.
In fact
a b
≥ ⇐⇒ ac + a2 ≥ bc + b2
b+c c+a
⇐⇒ (a − b)(c + a + b) ≥ 0
⇐⇒ a ≥ b.
Thus we have
a b c
an−1 ≥ bn−1 ≥ cn−1 and ≥ ≥ .
b+c c+a a+b
Using Chebyshev’s inequality for these numbers, we get
a b c
n
a bn cn
a n−1
+b n−1
+c n−1
+ + ≤3 + + .
b+c c+a a+b b+c c+a a+b
However, we have
a b c 3
+ + ≥ ,
b+c c+a a+b 2
using the left part of the inequality (1.4). Hence
n − a1 ≥ n − a2 ≥ · · · ≥ n − an .
n
1 n2 n2 n
≥ n = = .
1 + naj 2n 2
j=1
(1 + naj )
j=1
n 2
n − aj n −1 n n2 − 1
≥ = .
j=1
1 + naj n 2 2
n
n
n
aj bn+1−j ≤ aj bj ≤ a j bj ,
j=1 j=1 j=1
n
holds. Thus the sum aj bj is maximum when the two sequences aj and bj
j=1
are ordered similarly (i.e., either both non-decreasing or both non-increasing). And
the sum is minimum when aj and bj are ordered in opposite manner (i.e., one
of them is increasing and the other decreasing).
Proof: We may assume that both aj ’s and bj ’s are non-decreasing; the proof
is similar in the other case. Suppose aj = aj . Let r be the largest index
such that ar = ar ; i.e., ar = ar and aj = aj for r < j ≤ n. This implies that
ar is from the set a1 , a2 , . . . , ar−1 and ar < ar . Further this also shows that
a1 , a2 . . . , ar is a permutation of a1 , a2 , . . . , ar . Thus we can ﬁnd indices k < r
and l < r such that ak = ar and ar = al . It follows that
ak − ar = ar − al ≥ 0, br − bk ≥ 0.
22 Inequalities
We now interchange ar and ak to get a permutation a1 , a2 , . . . , an of a1 , a2 ,
. . . , an ; thus ⎧
⎪
⎨ aj = aj , if j = r, k
ar = ak = ar ,
⎪
⎩
ak = ar = al .
Let us consider the sums
cj = an+1−j , dj = −bn+1−j .
c 1 d1 + c 2 d2 + · · · + c n dn ≤ a 1 d1 + a 2 d2 + · · · + a n dn .
n
n
− an+1−j bn+1−j ≤ − aj bn+1−j .
j=1 j=1
This reduces to
It is not diﬃcult to ﬁnd the condition under which equality holds in the
inequality. If for each pair k, l with 1 ≤ k < l ≤ n, either ak = al or ak > al
and bk = bl , then equality holds in (1.12). A similar condition is true for
equality in (1.13): for each k, l with 1 ≤ k < l ≤ n, either an+1−k = an+1−l or
an+1−k > an+1−l and bn+1−k = bn+1−l .
Many of the inequalities we have studied so far can be derived using the
rearrangement inequality.
Corollary 4.1: Let α1 , α2 , . . . , αn be n given real numbers and β1 , β2 , . . . , βn
be a permutation of α1 , α2 , . . . , αn . Then
n
n
αj βj ≤ αj2 ,
j=1 j=1
n
n
αj βj ≤ αj2 .
j=1 j=1
Suppose equality holds in the inequality and αj = βj . Then αj =
βj .
Let k be the largest index such that αk = βk ; i.e., αk = βk and αj = βj for
Obviously βm = βk by our choice of k. We know that equality holds if and
only if for any two indices r = s, either αr = αs or βr = βs . Since βm
= βk ,
we must have αm = αk . But then we have αm = αm+1 = · · · = αk . Using the
minimality of m, we see that k − m + 1 equal elements αm , αm+1 , · · · , αk must
be among βm , βm+1 , · · · , βn and since βk is diﬀerent from αk , we must have
αk = βl for some l > k. But then using βl = αl we get
αm = αm+1 = · · · = αk = · · · = αl .
Thus the number of equal elements gets enlarged to l −m+1 > k −m+1. Since
this process cannot be continued indeﬁnitely, we conclude that αj = βj . It
now follows that αj = βj .
n
n
1 1
βj − ≤ αj − = −n.
j=1
αj j=1
αj
This gives the desired inequality. The condition for equality can be derived as
earlier.
Example 1.17. Prove the AM-GM-HM inequality using the rearrangement in-
equality.
n
βj αj+1
n−1
α1
n≤ = + .
α
j=1 j j=1
α j α n
Basic Inequalities 25
However,
αj+1 a1 a2 · · · aj+1 /Gj+1 aj+1 α1 a1
= = , = .
αj a1 a2 · · · aj /Gj G αn G
Thus we get
a1 + a2 + · · · + an
n≤ ,
G
which is same as the AM-GM inequality. Here equality holds if and only if
αj = βj . This is equivalent to α1 = α2 = · · · = αn which in turn is
equivalent to a1 = a2 = · · · = an .
Using corollary 4.2, it may also be obtained that
n
αj G G G G
n≤ = + + ··· + + .
j=1
βj a2 a3 an a1
Solution: Note that the sum on the left side is symmetric in aj ’s and hence
we may assume that a1 ≤ a2 ≤ · · · ≤ an . This implies that
λ − a1 ≥ λ − a2 ≥ · · · ≥ λ − an ,
and hence
1 1 1
≤ ≤ ··· ≤ .
λ − a1 λ − a2 λ − an
For any k, consider the permutation of a1 , a2 , . . . , an deﬁned by
ak+j−1 for 1 ≤ j ≤ n − k + 1,
βj =
ak+j−1−n for n − k + 2 ≤ j ≤ n.
Proof: We ﬁrst prove an auxiliary result which is useful for the proof of the
above theorem.
If x, y are two positive real numbers, and p, q are positive
reals such that {p, q} is a pair of conjugate indices, then
xp yq
+ ≥ xy.
p q
There are many ways to prove this, but each one of them depends on some
continuity argument. We can use the generalised AM-GM inequality to get
xp yq
1/p q 1/q
+ ≥ xp y = xy.
p q
Or one may consider the function
tp t−q
f (t) = + ,
p q
on the interval (0, ∞). Then f has a unique minimum at t = 1 and hence
tp t−q 1 1
+ ≥ + = 1,
p q p q
for all t > 0. Taking t = x1/q y −1/p , we obtain the desired inequality. It is also
easy to see that equality holds at the minimum point t = 1 which corresponds
to xp = y q .
Now taking
a k bk
x= n ,y= n ,
p 1/p q 1/q
a j bj
j=1 j=1
we get
p q
a k bk a k bk
+ ≥ 1/p .
n n n n
q 1/q
p aj p
q bj q
aj p
bj
j=1 j=1 j=1 j=1
Basic Inequalities 29
n
a k bk
1 1 j=1
+ ≥ n .
p q p 1/p
n
q 1/q
a j bj
j=1 j=1
Thus we have
n
n
p 1/p
n
q 1/q
a k bk ≤ a j bj .
j=1 j=1 j=1
for 1 ≤ k ≤ n. Taking
n
p '
n
q
λ= a j bj ,
j=1 j=1
Remark 5.2: If either of p and q is negative, the inequality (1.15) gets re-
versed. For example, if p < 0 and q > 0 (obviously p and q both cannot be
negative simultaneously), we take
p 1
p 1 = − , q1 = .
q q
1 1 q 1
+ =− +q =q 1− = 1.
p1 q1 p p
30 Inequalities
−q q q
Introducing uk = ak , vk = ak bk and applying Hölder inequality to the
collections uk , vk with the conjugate indices p1 , q1 , we obtain
n
−q q q n
−qp1 p11
n
qq1 q11
a k a k bk ≤ a k a k bk
k=1 k=1 k=1
n
p − pq
n
q
= a k a k bk .
k=1 k=1
Proof: We may assume p > 1, because the result for p = 1 is clear. Observe
that
n
n
a j + bj p = aj + bj p−1 aj + bj
j=1 j=1
n
n
≤ aj + bj p−1 aj + aj + bj p−1 bj .
j=1 j=1
n
n
p 1/p
n
1/q
aj + bj p−1 aj ≤ a j a j + bj p .
j=1 j=1 j=1
Similarly,
n
n
p 1/p
n
1/q
aj + bj p−1 bj ≤ bj a j + bj p .
j=1 j=1 j=1
Using the conditions for equality in Hölder’s inequality, we may obtain the
conditions for equality in Minkowski’s inequality: aj = λbj , for all 1 ≤ j ≤ n,
where λ is a real constant.
Remark 6.1: For 0 < p < 1, the inequality (1.16) gets reversed.
Remark 6.2: For any vector a = a1 , a2 , . . . , an and p > 0, we deﬁne
n 1/p
||a||p = |aj |p .
j=1
for p ≥ 1 and
||a + b||p ≥ ||a||p + ||b||p ,
for 0 < p < 1, where a and b are two vectors in Rn .
1 1
Example 1.22. Let p, q be two positive real numbers such that + = 1.
p q
Suppose a1 , a2 , . . . , an is an n-tuple of real numbers. Show that
n
ak bk ≤ C, (1.17)
k=1
Basic Inequalities 33
n
q
for every n-tuple b1 , b2 , . . . , bn of real numbers with bk = 1 if and only if
k=1
n
p 1/p
a k ≤ C. (1.18)
k=1
Solution: Suppose (1.17) holds for all n-tuples b1 , b2 , . . . , bn such that
n
q
bk = 1. Consider the n-tuple deﬁned by
k=1
p−1
a k Sgn(ak )
bk =
n p 1/q
,
a k
k=1
where
ak
|ak | if ak = 0,
Sgn(x) =
0 if ak = 0.
For this n-tuple, we have
n (p−1)q n p
n
q a k
bk = k=1 ak
n p = k=1
n p = 1.
a k
k=1 k=1 ak k=1
n
n
p 1/p
n
q 1/q
a k bk ≤ a k bk ≤ C.
k=1 k=1 k=1
Remark: This can be generalised to any number of sets with proper ex-
ponents.
holds. If the inequality is strict for all x = y, we say that f is strictly convex
on I. If the inequality in (1.19) is reversed, i.e.,
f λx + (1 − λ)y ≥ λf (x) + (1 − λ)f (y),
Basic Inequalities 35
for all x, y ∈ I and λ ∈ [0, 1], then we say f is concave. If the inequality is
strict for all x = y, we say that f is strictly concave.
There are other equivalent characterisations of a convex function. Suppose
f : I → R is convex and x1 , x2 , x3 are in I such that x1 < x2 < x3 . Take
x3 − x2
λ= ,
x3 − x1
so that
x2 − x1
1−λ= , and x2 = λx1 + (1 − λ)x3 .
x3 − x1
We have
f x2 = f λx1 + (1 − λ)x3
≤ λf x1 + (1 − λ)f x3
x3 − x2 x2 − x1
= f x1 + f x3 .
x3 − x1 x3 − x1
We may write this in the form
f x1 − f x2 f x2 − f x3
≤ ,
x1 − x2 x2 − x3
for all x1 < x2 < x3 in I. This may also be written in a more symmetric form:
f x1 f x2 f x3
+ + ≥ 0.
x1 − x2 x1 − x3 x2 − x1 x2 − x3 x3 − x1 x3 − x2
f (b) − f (a)
L(x) = f (a) + (x − a).
b−a
Now any point between a and b is of the form x = λa + (1 − λ)b for some
λ ∈ [0, 1]. Hence
L(x) = L λa + (1 − λ)b
f (b) − f (a)
= f (a) + λa + (1 − λ)b − a
b−a
= f (a) + 1 − λ f (b) − f (a)
= λf (a) + (1 − λ)f (b)
≥ f λa + (1 − λ)b = f (x).
36 Inequalities
Thus x, L(x) lies above (x, f (x)), a point on the graph of f .
There is still another way of looking at these things. A subset E of the
plane R2 is said to be convex if for every pair of points z 1 and z 2 in E, the
line joining z 1 and z 2 lies entirely in E. With every function f : I → R, we
associate a subset of R2 by
E(f ) = (x, y) : a ≤ x ≤ b, f (x) ≤ y .
z = λz 1 + (1 − λ)z 2
= λx1 + (1 − λ)x2 , λy1 + (1 − λ)y2 ,
for all λ ∈ [0, 1]. This shows that f is convex on the interval I.
The convexity of a function implies something about the slope of its graph.
The following theorem gives a complete description.
f (x) − f (a)
P (x) = .
x−a
Then P is a non-decreasing function on I \ {a}.
Basic Inequalities 37
Consider the case a < x < y; other cases may be taken care of similarly. We
write
x−a y−x
x= y+ a.
y−a y−a
The convexity of f shows that
x−a y−x x−a y−x
f y+ a ≤ f (y) + f (a).
y−a y−a y−a y−a
This is equivalent to
f (x) − f (a) f (y) − f (a)
≤ .
x−a y−a
Thus P (x) ≤ P (y). This shows that P (x) is a non-decreasing function for
x = a.
Interestingly, the converse is also true; if P (x) is a non-decreasing function
on I \ {a} for every a ∈ I, then f (x) is convex. The proof is not diﬃcult. Fix
x < y in I and let a = λx + (1 − λ)y where λ ∈ (0, 1). (The cases λ = 0 or 1
are obvious.) In this case
The condition P (x) ≤ P (y) implies that f (a) ≤ λf (x) + (1 − λ)f (y). Hence
the convexity of f follows.
There is a useful, easy way of deciding whether a function is convex or
concave for twice diﬀerentiable functions. If f is convex on an interval I and
if its second derivative exists on I, then f is convex(strictly convex) on I if
f (x) ≥ 0(> 0) for all x ∈ I. (Here f (x) denotes the second derivative of f
at x.) Similarly f is concave(strictly concave) on I if f (x) ≤ 0(< 0) for all
x ∈ I.
Here are some examples of convex and concave functions.
1. The function f (x) = xα is concave for 0 < α ≤ 1 and convex for 1 ≤ α <
∞ on the interval (0, ∞). We observe that
k k
f λj x j ≤ λj f xj .
j=1 j=1
However, we have
n
α 1 y1 + α 2 y2 = λj x j .
j=1
Basic Inequalities 39
Now consider f y1 . If
λl
μl = n−1 , 1≤l ≤n−1
j=1 λj
n−1
then it can be easily veriﬁed that l=1 μl = 1. Using the induction hypothesis,
we get
n−1
n−1
f μl xl ≤ μl f xl .
l=1 l=1
Since
n−1
μ l x l = y1 ,
l=1
we get n−1
n−1
l=1 λl f xl j=1 λj f x j
f y1 ≤ n−1 = .
j=1 λj
α1
Thus we obtain
n−1
n
j=1 λj f xj
f λj f x j ≤ α1 n−1 + λn f x n
j=1 j=1 λj
n
= λj f x j .
j=1
Remark 9.2: Using the concavity of f (x) = ln x on (0, ∞), the AM-GM in-
equality may be proved. If x1 , x2 , . . . , xn are points in (0, ∞) and λ1 , λ2 , . . . , λn
are real numbers in the interval [0, 1] such that λ1 + λ2 + · · · + λn = 1, then we
have
n n
ln λj x j ≥ λj ln xj .
j=1 j=1
Using the fact that g(x) = exp(x)(= ex ) is strictly increasing on the interval
(−∞, ∞), this leads to
1
n
n
1/n
xj ≥ exp ln xj
n j=1 j=1
"
n
1/n
= exp ln xj
j=1
1/n
= x1 x2 · · · xn .
and
n p
n
λj x j ≥ λj xpj for 0 < p < 1.
j=1 j=1
Basic Inequalities 41
Let a1 , a2 , . . . , an and b1 , b2 , . . . , bn be two n-tuples of real numbers and
1 1
p > 1 be a given real number. Let q be the conjugate of p; + = 1. We
p q
may assume bj = 0 for all j; otherwise we may delete all those bj which are
zero without aﬀecting the inequality. We now set
n
|bk |q |ak |
t= |bj |q , λk = , xk = .
j=1
t |bk |q−1
n
n 1/p
n 1/p
n 1/q
1−(1/p)
|bk ak | ≤ |ak | p
t = |ak | p
|bk | q
.
k=1 k=1 k=1 k=1
n
n 1/p
n 1/q
|bk ak | ≥ |ak | p
|bk | q
.
k=1 k=1 k=1
n
n
n
f aj + bn+1−j ≤ f aj + bj ≤ f a j + bj .
j=1 j=1 j=1
holds.
42 Inequalities
Proof: We follow the proof of theorem 4. Suppose aj = aj and r be the
largest index such that ar = ar ; i.e., ar = ar but aj = aj for r < j ≤ n.
Hence ar is in the set a1 , a2 , . . . , ar−1 and ar < ar . Since aj = aj for
r < j ≤ n, we see that (a1 , a2 . . . , ar ) is a permutation of (a1 , a2 , . . . , ar ).
Hence we can ﬁnd k < r, l < r such that ak = ar and ar = al . We deduce that
ak − ar = ar − al ≥ 0 and br − bk ≥ 0. We now interchange ar and ak to get a
permutation (a1 , a2 , . . . , an ) of (a1 , a2 . . . , an ); thus
Let us write
n
n
S = f aj + bj , S = f aj + bj .
j=1 j=1
We have
S − S = f ar + br + f ak + bk − f ar + br − f ak + bk
= f a r + br + f a l + b k − f a l + b r − f a r + b k .
We observe that
These give
a l + bk < a r + bk ≤ a r + br ,
a l + bk ≤ a l + b r < a r + b r .
place (a1 , a2 , . . . , an ) and proceed as above. After at most n−1 steps we arrive
at the original numbers aj from aj , and at each stage the corresponding
sum is non-decreasing. We ﬁnally get
n
n
f aj + bj ≤ f a j + bj .
j=1 j=1
For proving the other inequality, we proceed along the same lines. We deﬁne
cj = an+1−j so that c1 ≥ c2 ≥ · · · ≥ cn . We have to show that
n
n
f an+1−j + bj ≤ f aj + bj .
j=1 j=1
n
n
f c j + bj ≤ f cj + bj ,
j=1 j=1
and obtain
S − S = f cr + br + f ck + bk − f cr + br − f ck + bk
= f c r + br + f c l + bk − f c l + br − f c r + bk .
We observe that
c l + br ≤ c l + b k < c r + bk ,
c l + br ≤ c r + b r < c r + b k .
44 Inequalities
and
c r + bk − c l − br f c r + br ≤ bk − b r f c l + br + c r − c l f c r + bk .
Since cr + bk − cl − br = 0, we get
f c l + b k + f c r + b r ≤ f c l + b r + f c r + bk .
This shows that S ≤ S . We also observe that the new sequence cj has
the property: cr = cr and cj = cj for 1 ≤ j < r. We may consider now the
sequence cj in place of cj and continue the above argument. At each stage
the sum never increases. After at most n − 1 steps we arrive at the sequence
cj . Hence the corresponding sum cannot exceed that of S = j=1 f cj +bj .
n
We thus get
n
n
f c j + bj ≤ f cj + bj ,
j=1 j=1
which is to be proved.
holds.
This simpliﬁes to
n
n(n + 1) ln(n + 1) ≤ 2 j ln j + n(n + 1) ln 2.
j=1
n(n + 1) n + 1 "
n n
ln ≤ j ln j = ln jj .
2 2 j=1 j=1
and
a
1+x ≥ 1 + ax if 1 ≤ a < ∞.
Solution: We use the concavity of f (x) = ln(x) on (0, ∞). Since x > −1, we
see that 1 + x > 0. If 0 < a < 1, we have
ln 1 + ax = ln a(1 + x) + 1 − a
≥ a ln 1 + x + 1 − a ln(1)
= a ln 1 + x .
a
The exponentiation gives 1 + x ≤ 1 + ax. Suppose 1 ≤ a < ∞. Then again
a − 1 1
ln 1 + x = ln + 1 + ax
a a
a−1 1
≥ ln(1) + ln 1 + ax
a a
1
= ln 1 + ax .
a
a
This gives 1 + x ≥ 1 + ax.
Example 1.26. Let a = a1 ,a2 ,a3 ,. . .,an be an n-tuple of positive real num-
bers and w1 ,w2 ,w3 ,. . .,wn be another n-tuple of positive real numbers (called
46 Inequalities
(b) Suppose 0 < r < s. Put α = s/r and observe that α > 1. Hence the
function f (x) = xα is convex on (0, ∞). Hence Jensen’s inequality gives
n n
wj rα
r α n s
j=1 wj aj j=1 wj aj
≤ aj = .
wk j=1
wk wk
rα s
This shows that Mr (a) ≤ Ms (a) . Taking s-th root, we get
Mr (a) ≤ Ms (a).
(c) If s = 0, then r < 0 and we take t = −r. Using the generalised AM-GM
inequality, we have
*
wj ⎛ ⎞ 1 wk
at " n
1 ⎠
j ≥ ⎝ twj
wk j=1 aj
*
⎛ ⎞ t wk
" 1
n
= ⎝ w
⎠
a j
j=1 j
Basic Inequalities 47
Simpliﬁcation gives
*
wj 1/t
⎛ ⎞ 1 wk
atj "
n
1 ⎠
≥⎝ wj .
wk a
j=1 j
This is equivalent to
*
⎛ ⎞
1/r "n
1 wk
wj arj
≤⎝ aj j ⎠
w
.
wk j=1
where
−1 1 1 1
a = , ,... ,
a1 a2 an
But then
1 1
M−s (a−1 ) = , M−r (a−1 ) = .
Ms (a) Mr (a)
Hence we get Mr (a) ≤ Ms (a).
(f) Obviously
min a1 ,a2 ,a3 ,. . .,an ≤ Mr (a) ≤ max a1 ,a2 ,a3 ,. . .,an .
Given n complex numbers a = a1 ,a2 ,a3 ,. . .,an , we consider the polynomial
whose roots are a1 ,a2 ,a3 ,. . .,an ;
"
n
P (x) = (x − aj ).
j=1
48 Inequalities
where
n
j=1 aj
u1 = u1 (a) = n ,
1
j<k aj ak
u2 = u2 (a) = n ,
2
.. ..
. .
n
j=1 aj
un = un (a) = n .
n
These functions u1 ,u2 ,u3 ,. . .,un are called the elementary symmetric functions
in the variables a1 ,a2 ,a3 ,. . .,an . We may also write them in the form ur (ak )
to show the dependence on the sequence a1 ,a2 ,a3 ,. . .,an . We observe that the
permutation of a1 ,a2 ,a3 ,. . .,an does not alter any of the uj ’s, 1 ≤ j ≤ n. We
take u0 (a) = 1.
1/k
We further introduce vk = uk . We observe that v1 is simply the arithmetic
mean of a1 ,a2 ,a3 ,. . .,an and vn is their geometric mean.
This reduces to
un un−2 ≤ u2n−1 . (1.23)
We show that ur−1 ur+1 ≤ u2r for r = 2, 3, . . . , n − 1. We use the induction on
the number of aj ’s. For n = 3, we have r = 2 and hence we have to prove that
u1 u3 ≤ u22 . This follows from (1.23) with n = 3. Suppose the result holds for
a set of n − 1 numbers. We have
n−1
n−1 k n−1 n−1−k
P (x) = n x + (−1) uk x .
k
k=1
2.1 Introduction
There are several standard ways of proving a given inequality. We have al-
ready seen how to obtain the AM-GM inequality using forward and backward
induction. One can also use the known standard inequalities or use ideas from
calculus. In some cases trigonometric substitutions simplify the result. We
take each of these separately and illustrate them using examples.
where [x] denotes the integral part of x. If n = 1, then x1 lies in the interval
[0, 1] and (2.1) reduces to
- . - . 2
x21 ≤ x1 + x1 − x1 ,
- .
which is true since x1 = 0 or 1. Suppose (2.1) holds for 1 ≤ n ≤ N − 1. Put
N −1
SN −1 = xj = k + t, 0 ≤ t < 1,
j=1
N
SN = xj = m + r, 0 ≤ r < 1.
j=1
52 Inequalities
We show that
- . - .2 - . - .2
x2N ≤ SN + SN − SN − SN −1 − SN −1 − SN −1 . (2.2)
This is equivalent to
m − k + r − t)2 ≤ m + r2 − k − t2 .
N
N −1
x2j = x2j + x2N
j=1 j=1
- . - .2
≤ SN −1 + SN −1 − SN −1 + x2n
- . - .2
≤ SN + SN − SN .
A proof of this may be based on the properties of convex functions and the
principle of induction as used above.
where
R = x2 + x2 x3 + · · · + x2 x3 · · · xn .
Thus the inequality to be proved is
(1 + x1 )(1 + R) 1 + x1 + x1 R
≤ .
2n n+1
This is equivalent to
R (n + 1) − (n − 1)x1 ≤ (n − 1)(1 + x1 ).
If n + 1 ≤ (n − 1)x1 , then the left hand side is non-positive and the right hand
side is non-negative. Hence the result is true in this case. Thus we may assume
that (n + 1) − (n − 1)x1 > 0. Thus it is suﬃcient to deduce that
(n − 1)(1 + x1 )
R≤ .
(n + 1) − (n − 1)x1
But, observe that
R = x2 + x2 x3 + · · · + x2 x3 · · · xn ≤ x1 + x21 + · · · + xn−1
1 .
9abc ≤ ab + bc + ca.
holds.
Example 2.4. Let a, b, c be positive real numbers, and let p, q, r be three numbers
in the interval [0, 1/2]. Suppose a + b + c = p + q + r = 1. Prove that
pa + qb + rc
abc ≤ .
8
Techniques 55
Example 2.5. For n ≥ 2, and real numbers 0 < x1 < x2 < · · · < xn ≤ 1, prove
that n
n k=1 xk
n
1
n ≥ .
k=1 xk + nx1 x2 · · · xn 1 + xk
k=1
Taking
xk
ak = , for 1 ≤ k ≤ n,
1 + xk
we obtain
n
n
1
n
1
x−1
k n− ≥n .
1 + xk 1 + xk
k=1 k=1 k=1
Multiplying by x1 x2 · · · xn and using the inequality
n
n
xk ≥ x1 x2 · · · xn x−1
k ,>
k=1 k=1
we get
n
n
1 1 n
xk n− ≥ nx1 x2 · · · xn .
1 + xk 1 + xk
k=1 k=1 k=1
This implies that
n
n
n
1
n xk ≥ xk + nx1 x2 · · · xn .
1 + xk
k=1 k=1 k=1
Solution: We put
aj aj
αj = and bj = ,
a1 + a2 + · · · + an aj + aj+1
n
where an+1 = a1 . Then j=1 αj = 1. Now the weighted AM-HM inequality
gives
n
n
α j bj αj
j=1 j=1
≥ .
n n
αj
αj
j=1 j=1
bj
We observe that
n n
αj
n
n
aj = aj + aj+1 =2 aj ,
j=1 j=1
bj j=1 j=1
n
αj n n
so that = 2. Since j=1 αj = 1, we get j=1 αj bj ≥ 1/2. Thus we
j=1
bj
obtain
n
a2j n n
1
n
= aj α j bj ≥ aj ,
j=1
aj + aj+1 j=1 j=1
2 j=1
a2 + b2 + c2 ≥ 3(abc)2/3 , ab + bc + ca ≥ 3(abc)2/3 .
58 Inequalities
Example 2.8. (IMO, 1995) Let a, b, c be positive reals such that abc = 1. Prove
that
1 1 1 3
+ + ≥ .
a3 (b + c) b3 (c + a) c3 (a + b) 2
x2 y2 z2 3
+ + ≥ .
y+z z+x x+y 2
Using the Cauchy-Schwarz inequality, we have
2
2 x y2 z2
(x + y + z) ≤ 2 + + x + y + z).
y+z z+x x+y
Thus
x2 y2 z2 x+y+z
+ + ≥ .
y+z z+x x+y 2
But the AM-GM inequality gives
x + y + z ≥ 3(xyz)1/3 = 3.
It follows that
x2 y2 z2 3
+ + ≥ .
y+z z+x x+y 2
holds if x ≥ 2 or x ≤ −1. For each x, −1 < x < 2, we can ﬁnd positive real
numbers a, b, c such that abc = 1 and (2.4)is not true.
For any real x, deﬁne
1 1 1
f (x; a, b, c) = + + .
ax (b + c) bx (c + a) cx (a + b)
We show that f (x; a, b, c) is a non-decreasing function for x ≥ 1. Let x > y ≥ 1
be two reals. Suppose a ≤ b ≤ c. Then ax−y ≤ bx−y ≤ cx−y is true. We also
observe that ay (b + c) ≤ by (c + a) ≤ cy (a + b). In fact, ay (b + c) ≤ by (c + a) is
equivalent to ab(ay−1 − by−1 ) ≤ c(by − ay ) which is true because y − 1 ≥ 0 and
a ≤ b. Similarly, by (c + a) ≤ cy (a + b) holds. Using Chebyshev’s inequality, we
obtain
1 1 1
+ +
ax (b + c) bx (c + a) cx (a + b)
1 1 1
= + +
ax−y ay (b + c) bx−y by (c + a) cx−y cy (a + b)
1 1 1 1 1 1 1
≥ + + + +
3 ax−y bx−y cx−y ay (b + c) by (c + a) cy (a + b)
1 1 1
≥ y
+ y + y .
a (b + c) b (c + a) c (a + b)
We have used the AM-GM inequality in the last leg. We get f (x; a, b, c) ≥
f (y; a, b, c) if x > y ≥ 1. Thus f (x; a, b, c) is a non-decreasing function on the
interval [1, ∞).
Now consider f (2; a, b, c). We write
1 1 1
f (2; a, b, c) = + 2 + 2
a2 (b
+ c) b (c + a) c (a + b)
bc ca ab
= + +
ab + ac bc + ab bc + ca
1 1 1
= (ab + bc + ca) + + −3
ab + bc bc + ca ca + ab
9 3
≥ −3≥ .
2 2
It follows that f (x; a, b, c) ≥ 3/2 for x ≥ 2.
We now show that the result is not true for 1/2 ≤ x < 2. Take x = 2 − e
where e > 0. Put a = d2 , b = 1/d, c = 1/d so that abc = 1. Then
1 2d3−e
f (2 − e; a, b, c) = +
2d3−2e d3 + 1
and f (2 − e; a, b, c) → 0 as d → ∞, if e < 3/2. If e = 3/2 then f (2 − e; a, b, c) →
1/2 < 3/2. In any case (2.4) is not true. Thus (2.4) fails for 1/2 ≤ x < 2.
60 Inequalities
For x in the interval (−∞, 1/2), we replace a, b, c by 1/p, 1/q, 1/r respec-
tively. Then f (x; a, b, c) = f (1−x; p, q, r). But 1−x lies in the interval (1/2, ∞).
It follows that (2.4) holds for x ≤ −1 and for x in the interval (−1, 1/2) there
are a, b, c such that abc = 1 and (2.4) fails.
Prove that
n
√ n
1
xj ≥ (n − 1) √ .
j=1 j=1
xj
1 + xj + xk + xj xk > 2 + xj + xk .
This reduces to
n √
n
xj √
xj ≥ n2 − n = n(n − 1). (2.6)
j=1
1 + xj j=1
n √
xj n
≤ n .
1 + x j 1
j=1
√
j=1
xj
This gives
n
√ n
1
xj ≥ (n − 1) √ .
j=1 j=1
xj
Alternate Solution:
Using
n
1
= 1,
j=1
1 + xj
we easily obtain
n
xj
= n − 1.
j=1
1 + xj
62 Inequalities
Hence
n
√ n
1
xj − (n − 1) √
j=1 j=1
xj
n n n n
1 √ xj 1
= xj − √
j=1
1 + xj j=1 j=1
1 + xj j=1
xj
n
n
xj − xk
= √
j=1 k=1
(1 + xk ) xj
√x j √x k − 1 √ x j − √x k 2 √x j + √x k
= √ √ .
xj xk (1 + xj )(1 + xk )
j>k
Thus we get
p 0 np 1 n−1
αnp − αp−1 an ≤ αnp 1 − + (n − 1) + αp
p−1 n p−1 p − 1 n−1
1 0 p
1
= (n − 1)αn−1 − nαnp .
p−1
Taking α0 = 0 and summing over n from 1 to N , we get
N
p p−1
N p
N αN
αnp − αn an ≤ − ≤ 0.
n=1
p − 1 n=1 p−1
Thus we obtain,
N
p p−1
N
αnp ≤ α an
n=1
p − 1 n=1 n
This simpliﬁes to
N p p
N
αkp ≤ apk .
p−1
k=1 k=1
"
1
n n n
a
ln aj j ≥ aj ln aj .
j=1
n j=1 j=1
Using the fact that the logarithm takes multiplication to addition, and using
the monotonicity of exponential function, we obtain the desired inequality.
Example 2.12. Let a1 , a2 , . . . , an be n positive real numbers. Show that for any
permutation a1 , a2 , . . . , an of the sequence a1 , a2 , . . . , an , the inequality
"
n
a
"
n
a
aj j ≥ aj j ,
j=1 j=1
holds.
ln a1 ≥ ln a2 ≥ · · · ≥ ln an .
n
n
aj ln aj ≥ aj ln aj .
j=1 j=1
This implies
n
a
n
a
ln aj j ≥ ln aj j .
j=1 j=1
"
n
a
"
n
a
aj j ≥ aj j .
j=1 j=1
Techniques 65
and let x = x1 /x2 . Take any α with 0 < α < 1. Then we have
α
x1 x1
−α + α − 1 ≤ 0.
x2 x2
It may be rewritten as
1−α
xα
1 x2 ≤ αx1 + 1 − α x2 .
This proves the generalised AM-GM inequality for non-negative real num-
bers x1 , x2 and weights α, 1 − α. Now we use the principle of induction to
prove the general case. Assume that for any n − 1 non-negative real numbers
n−1
y1 ,y2 ,y3 ,. . .,yn−1 and positive weights λ1 ,λ2 ,λ3 ,. . .,λn−1 with j=1 λj = 1, we
have
" λ
n−1
n−1
yj j ≤ λj yj .
j=1 j=1
yj = x j , λj = αj , for 1 ≤ j ≤ n − 2,
αn−1 /λn−1 αn /λn−1
yn−1 = xn−1 xn , λn−1 = αn−1 + αn .
n−1
n
λj = αj = 1.
j=1 j=1
"
n
α
"
n−1
λ
xj j = yj j
j=1 j=1
n−1
≤ λj yj
j=1
n−2
αn−1 /λn−1 αn /λn−1
= αj xj + αn−1 + αn xn−1 xn
j=1
n−2
αn−1 αn
≤ αj xj + αn−1 + αn xn−1 + xn
j=1
λn−1 λn−1
n
= α j xj ,
j=1
since λn−1 = αn−1 + αn . Equality holds if and only if all xj ’s are equal.
Techniques 67
Example 2.14. Show that ex > 1+x+(x2 /2) for x > 0, and ex < 1+x+(x2 /2)
for x < 0.
Solution: Consider f (x) = sin x. Since f (x) = cos x and f (x) = − sin x,
we see that f (x) ≤ 0 on [0, π/2]. Hence f (x) is a concave function on [0, π/2].
Take any x ∈ [0, π/2] and set λ = 2x/π. Using the concavity of f (x) = sin x,
we have
sin x = sin (1 − λ)0 + λπ/2
≥ (1 − λ) sin 0 + λ sin π/2
2x
= λ= .
π
68 Inequalities
Thus we obtain
sin x 2
≥ .
x π
On the other hand, consider g(x) = sin x − x on [0, π/2]. We again note that
g (x) = cos x − 1 ≤ 0 for all x ∈ [0, π/2]. Hence g is a non-increasing function
on [0, π/2]. This implies that g(x) ≤ g(0) = 0. Thus we get, sin x ≤ x for all
x ∈ [0, π/2]. Using the known result that limx→0 sin x/x = 1, we conclude that
sin x/x ≤ 1 for all x ∈ [0, π/2]. Thus we have Jordan’s inequality,
2 sin x
≤ ≤ 1,
π x
valid for all x ∈ [0, π/2]. Here we have to take the limiting value for x = 0.
Suppose α > 1. Since x > −1, we see that (1 + x)α−1 > 1 if x > 0, and
(1 + x)α−1 < 1 for −1 < x < 0. Thus f (x) > 0 if x > 0, and f (x) < 0 if
−1 < x < 0. This shows that f (x) is decreasing on (−1, 0), and increasing on
(0, ∞). Hence f (x) > f (0) for all x > −1, x = 0. We obtain
If α < 0, then 1 − α > 1, and hence (1 + x)α−1 − 1 < 0 for 1 + x > 1; and
(1 + x)α−1 − 1 > 0 for 0 < 1 + x < 1. Thus f (x) > 0 for 1 + x > 1, and
f (x) < 0 for 0 < 1 + x < 1. Hence f (x) is decreasing in (−1, 0), and increasing
on (0, ∞). We see that f (x) > 0 for x > −1, and x = 0 in this case. This gives
Consider the case in which 0 < α < 1. Then α − 1 < 0. Hence (1 + x)α−1 −
1 < 0 for x > 0, and (1 + x)α−1 − 1 > 0 for −1 < x < 1. Thus, it follows
that f (x) > 0 on (−1, 0), and f (x) < 0 on (0, ∞). We deduce that f (x) is
increasing on (−1, 0), and decreasing on (0, ∞). This gives f (x) < f (0), and
hence
(1 + x)α < 1 + αx, if x > −1, x = 0, 0 < α < 1.
Techniques 69
Example 2.17. Let a1 ,a2 ,a3 ,. . .,an be n > 1 distinct real numbers. Prove that
⎛ ⎞
2
2 12
n
1
n
min aj − ak ≤ ⎝ a2 − aj ⎠ .
1≤j<k≤n n(n2 − 1) j=1 j n j=1
n−1
2
= a + jμ
j=0
n−1
n−1
= na2 + 2aμ j + μ2 j2
j=1 j=1
n(n − 1)(2n − 1)
= na2 + aμn(n − 1) + μ2
6
2
μ(n − 1)
= n a+ +1
2
> 1.
This contradiction proves the inequality (2.8).
We take bj = aj + t where t is a real variable and apply the inequality (2.8)
to the sequence bj to get
⎛ ⎞
2 12 n
2
min aj − ak ≤ ⎝ aj + t ⎠ ,
1≤j<k≤n n(n2 − 1) j=1
70 Inequalities
n 2
valid for all t. Consider the function f (t) = j=1 aj + t . Expanding this,
we obtain
n n
f (t) = a2j + 2 aj t + nt2 ,
j=1 j=1
Since
f (t 0 ) = 2n > 0, we infer that f has a unique minimum at t0 . Hence
f t ≥ f t0 , for all t. But we see that
2 2
n
2
n
1
n
f t0 = a2j − aj + aj
j=1
n j=1
n j=1
n n 2
1
= a2j − aj .
j=1
n j=1
1 *
cos2 (θ − φ) = 1 − sin2 (θ − φ) = 1 − = u2 v 2 − 1 u2 v 2 .
u2 v 2
We observe that u2 = a2 + b2 , v 2 = x2 + y 2 and ax + by = uv cos(θ − φ). Using
u2 + v 2 ≥ 2uv, we obtain
a2 + b2 + x2 + y 2 + ax + by ≥ 2uv − u2 v 2 − 1.
Techniques 71
It follows that √
2uv − u2 v 2 − 1 ≥ 3,
which is what we have to prove.
Another important technique for proving inequalities is to use the fact that
between any two real zeros of a diﬀerentiable function, there is always a zero
of its derivative. This follows from Rolle’s theorem in calculus. We have
already used this idea while proving Newton’s inequality in an earlier chapter.
Existence of real zeros for the derivative puts certain restrictions on the function
and this would lead to inequalities.
and
abc
4sRr = · Δ = abc.
Δ
Here Δ denotes the area of the given triangle. Observe that we have used the
well known Heron’s formula: Δ2 = s(s − a)(s − b)(s − c). Thus a, b, c are the
roots of the equation
p(x) = x3 − 2sx2 + s2 + r 4R + r x − 4sRr = 0.
Now p(x) = 0 has three real roots. Hence Rolle’s theorem shows that p (x) = 0
has two real roots. But
p (x) = 3x2 − 4sx + s2 + r 4R + r .
This is a quadratic equation and it has two real roots if and only if its discrim-
inant is non-negative. Computing the discriminant, we obtain
16s2 − 12 s2 + r 4R + r ≥ 0.
This leads to
9r 4R + r ≤ 3s2 .
We also know that
Δ Δ
ra = , r= .
s−a s
Hence
s(ra − r)
a= .
ra
Similarly, we can get
s(rb − r) s(rc − r)
b= , c= .
rb rc
Here ra , rb , rc are the ex-radii of the triangle ABC.
Putting x = s(y − r)/y in p(x) = 0, we obtain an equation in y;
q(y) = y 3 − 4R + r y 2 + s2 y − s2 r = 0.
Observe that ra , rb , rc are the roots of q(y) = 0. Hence q (y) = 0 also has real
roots. But
q (y) = 3y 2 − 2 4R + r + s2 ,
2
and its discriminant is 4 4R + r − 12s2 . We thus obtain the inequality
2
3s2 ≤ 4R + r .
Techniques 73
Solution: Since a, b, c are in [0, 1], we can ﬁnd x, y, z in the interval [0, π/2]
such that
a = sin2 x, b = sin2 y, c = sin2 z.
Thus the inequality to be proved gets transformed to
sin x cos y cos z + sin y cos z cos x + sin z cos x cos y ≤ 1 + sin x sin y sin z.
This is equivalent to
sin x cos y cos z − sin y sin z + cos x sin y cos z + cos y sin z ≤ 1.
We know that
2 tan(α/2) 1 − tan2 (α/2)
= sin α, = cos α.
1 + tan2 (α/2) 1 + tan2 (α/2)
Hence, this suggests the substitutions
This is equivalent to
We observe that
1 − xy 1 − tan(α/2) tan(β/2)
z= =
x+y tan(α/2) + tan(β/2)
π α β
= tan − − .
2 2 2
This shows that π
α β
tan(γ/2) = tan − . −
2 2 2
π α β
Since α and β are in (0, π/2), so is − − . Since x
→ tan x is one-one on
2 2 2
(0, π/2), we conclude that
γ π α β
= − − .
2 2 2 2
This gives α + β + γ = π. Now f (x) = − sin x is convex on (0, π). Suppose
α ≤ β ≤ γ and let α , β , γ be a permutation of α, β, γ. Using Theorem 10 on
page 41, we obtain
− sin α + α − sin β + β − sin γ + γ ≤ − sin 2α − sin 2β − sin 2γ .
This reduces to
sin 2α + sin 2β + sin 2γ ≤ sin α + sin β + sin γ,
Example 2.22. Show that for any three real numbers a, b, c, the inequality
2
ab + bc + ca − 1 ≤ a2 + 1 b2 + 1 c2 + 1 ,
holds.
(ab + bc) cos α cos β cos γ = sin α sin β cos γ + cos α sin β sin γ
= sin β sin(α + γ),
(ca − 1) cos α cos β cos γ = sin α cos β sin γ − cos α cos β cos γ
= − cos β cos(α + γ).
Thus, we obtain
2
(ab + bc + ca − 1) cos α cos β cos γ
2
= sin β sin(α + γ) − cos β cos(α + γ)
= cos2 α + β + γ ≤ 1.
Example 2.23. For any three positive real numbers x, y, z such that x + y + z =
xyz, prove the inequality
√
x y z 3 3
√ + +√ ≤ .
1 + x2 1 + y2 1 + z2 2
where
n
n
n
A= a2j , B= b2j , C= a j bj .
j=1 j=1 j=1
Example 2.24. Let (a1 , a2 ), (b1 , b2 ) and (c1 , c2 ) be three pairs of real numbers.
Prove that
2
a 1 b2 + a 2 b1 − 2 a1 a2 ≤4 a21 − a 1 b1 a22 − a 2 b2 ,
Solution: We know that for all real x, y, z, the following inequality holds:
x2 + y 2 + z 2 ≥ xy + yz + zx.
Thus we obtain
2
a 1 b2 + a 2 b1 − 2 a1 a2
cyclic cyclic cyclic
≤4 a21 − a 1 b1 a22 − a 2 b2 .
cyclic cyclic cyclic cyclic
Prove that
p n−1 2nq p n−1 2nq
− p2 − ≤ xj ≤ + p2 − ,
n n n−1 n n n−1
for all j.
We see that
2 2
p − x1 = x2 + x3 + · · · + xn
≤ n − 1 x22 + x23 + · · · + x2n
0 2 1
= n−1 x 2 + x3 + · · · + xn − 2 xj xk
2≤j<k≤n
0 2 1
= n−1 p − x1 − 2 q − x1 (p − x1 )
= n − 1 p2 − 2q − x21 .
78 Inequalities
2x = b + c − a, 2y = c + a − b, 2z = a + b − c,
then we see that x, y, z are positive reals and a = y+z, b = z+x, c = x+y. This
transformation is often referred as Ravi transformation in the mathematical
literature.
Example 2.26. (CRMO, 1999) Let a, b, c be the sides of a triangle. Prove that
a b c
+ + ≥ 3.
b+c−a c+a−b a+b−c
b c
Solution: It is easy to observe that there is a triangle with sides a + , b + ,
2 2
a
c + . Using Heron’s formula, we get
2
and
- .2 3
16 A B C = (a + b + c)(−a + b + 3c)(−b + c + 3a)(−c + a + 3b).
16
Since a, b, c are the sides of a triangle, there are positive real numbers x, y, z
such that a = y + z, b = z + x, c = x + y. Using these, we obtain
[ABC]2 16xyz
- .2 = .
ABC 3(2x + y)(2y + z)(2z + x)
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
If λ < 0, then
Example 2.29. (IMO, 2000) Let a, b, c be positive real numbers such that abc =
1. Prove that
1 1 1
a−1+ b−1+ c−1+ ≤ 1.
b c a
This follows from Schur’s inequality. (For diﬀerent solutions refer to problem
3.6 on page 288.)
Techniques 81
Example 2.30. (APMO, 2004) Let x, y, z be positive real numbers. Prove that
2
x + 2 y 2 + 2 z 2 + 2 ≥ 9 xy + yz + zx .
Note that
2 x2 y 2 − 4 xy + 6 = 2 (xy − 1)2 ≥ 0.
cyclic cyclic cyclic
Moreover cyclic x2 ≥ cyclic xy. Thus it is suﬃcient to prove that
x2 y 2 z 2 + x2 + 2 ≥ 2 xy.
cyclic cyclic
we obtain
(x + y + z)3 − 4(x + y + z)(xy + yz + zx) + 9xyz ≥ 0.
This proves (i). A similar proof works for (ii) with t = 2 in Schur’s inequality.
Expansion gives
a4 + b4 + c4 + abc(a + b + c) ≥ a3 (b + c).
cyclic
Therefore
a4 + b4 + c4 + 9abc(a + b + c) + ab(a2 + b2 )
cyclic
3
≥ a (b + c) + ab(a2 + b2 )
cyclic cyclic
2 2
=2 ab(a + b ) ≥ 2 ab(2ab) = 4 (a2 + b2 ).
cyclic cyclic cyclic
84 Inequalities
n
is Schur-convex on I . This leads to a standard majorisation inequality:
if f : (a, b) → R is convex, then
n
n
x1 ,x2 ,x3 ,. . .,xn ≺ y1 ,y2 ,y3 ,. . .,yn =⇒ f xj ≤ f yj .
j=1 j=1
for all vectors x, y in In , and λ ∈ [0, 1]. Any convex function is Schur-
convex.
Techniques 85
2(s − c) ≥ a,
2(s − c) + 2(s − b) ≥ a + b,
2(s − b) + 2(s − a) + 2(s − c) = a + b + c.
Thus
a, b, c ≺ 2(s − c), 2(s − b), 2(s − a) .
√
Since f (t) = t is concave on (0, ∞), using majorisation theorem,
√ √ √
2(s − c) + 2(s − b) + 2(s − a) ≤ a + b + c.
This reduces to
√ √ √ √ √ √
a + b − c + b + c − a + c + a − b ≤ a + b + c.
Example 2.35. Let a1 ,a2 ,a3 ,. . .,an be n natural numbers such that
Similarly,
This simpliﬁes to
a1 a2 a3 · · ·an ≤ (k + 1)m k n−m .
This simpliﬁes to
where Sn denotes the set of all permutations of {1, 2, 3, . . . , n}. Then Muir-
head’s theorem asserts that
S a; x ≤ S b; x
Techniques 87
Hence for any n non-negative real numbers a1 ,a2 ,a3 ,. . .,an , we obtain
1/n
n a1 a2 · · · an ≤ a1 + a2 + · · · + an .
Example 2.37. Show that for any positive reals a, b, c, the inequality
a3 b3 c3
+ + ≥ a + b + c,
bc ca ab
holds.
a2 bc + ab2 c + abc2 ≤ a4 + b4 + c4 .
1
a3 + b3 + c3 + abc ≥ (a + b + c)3 .
7
Note that
2 a3 + b3 + c3 ≥ a2 b + b2 c + c2 a + ab2 + bc2 + ca2
2.11 Homogenisation
There is a large class of inequalities which are non-homogeneous and come
with constraints. In many cases, the given constraint may be used to put the
inequality in a homogeneous form and then one may use results like Muirhead’s
theorem or Schur’s inequality to prove them.
Equivalently,
x(x − y)(x − z) ≥ 0,
cyclic
Example 2.41. (IMO, 1984) Let x, y, z be three non-negative real numbers such
that x + y + z = 1. Prove that
2
0 ≤ xy + yz + zx − 2xyz ≤ .
27
Solution: Here we give a proof using homogenisation. For a diﬀerent proof,
see solution to problem 3.6 on page 291.
Using x + y + z = 1, we put the inequality in a homogeneous form:
7 3
0 ≤ x + y + z xy + yz + zx − 2xyz ≤ x+y+z .
27
Observe that
x + y + z xy + yz + zx − 2xyz = x2 y + xy 2 + xyz.
cyclic cyclic
a2 + b2 + c2 + d2 + ab + ac + ad + bc + bd + cd ≥ 10.
Solution: The result at once follows from the AM-GM inequality. But we
give here a diﬀerent solution using homogenisation. We may homogenise this
to get an equivalent inequality:
2 2
a + b2 + c2 + d2 + ab + ac + ad + bc + bd + cd ≥ 100abcd.
For a given 4-tuple m1 , m2 , m3 , m4 of non-negative integers, we use the no-
tation:
1 mσ(1) mσ(2) mσ(3) mσ(4)
S m1 , m2 , m3 , m4 = a b c d ,
24
σ∈S4
where S4 is the set of all permutations of {1, 2, 3}. Expanding the left-side, we
may write the inequality in the form
Hence
2.12 Normalisation
Yet another technique used while proving inequalities is normalisation. This is
the reverse process to homogenisation. Many times homogeneous inequalities
may be normalised to simplify the proofs. The standard application is the
AM-GM inequality. We have to prove that
a1 + a2 + · · · + an
≥ (a1 a2 · · · an )1/n ,
n
for non-negative real numbers a1 , a2 , . . . , an . Since it is homogeneous, we may
normalise it by a1 a2 a3 · · · an = 1. Thus we have to prove that
a1 + a2 + · · · + an ≥ n,
for any n positive real numbers a1 , a2 , . . . , an , under the additional condition
that a1 a2 a3 · · · an = 1. This may be proved by the principle induction. For
n = 1, it is immediate. If n = 2, then
√ √ √ 2
a1 + a2 − 2 = a1 + a2 − 2 a1 a2 = a1 − a2 ≥ 0.
Suppose it holds for any k positive real numbers whose product is 1, where
k < n. Take n positive real numbers a1 , a2 , . . . , an such that a1 a2 a3 · · · an = 1.
Among these n numbers, there must be some number ≥ 1 and there must be
some number ≤ 1. Thus we may assume that a1 ≥ 1 ≥ a2 . We may apply
induction to n − 1 numbers a1 a2 , a3 , . . . , an :
a1 a2 + a3 + · · · an ≥ n − 1.
Thus it is enough to prove that a1 + a2 ≥ 1 + a1 a2 . But this is equivalent to
(a1 − 1)(a2 − 1) ≤ 0 which is a consequence of a1 ≥ 1 ≥ a2 . Hence the proof is
complete.
Example 2.43. (INMO, 2007) Let x, y, z be positive real numbers. Prove that
2
x + y + z)2 yz + zx + xy ≤ 3 y 2 + yz + z 2 z 2 + zx + x2 x2 + xy + y 2 .
a b c
√ +√ +√
a2 + 8bc 2
b + 8ca 2
c + 8ab
1
≥ .
a(a2 + 8bc) + b(b2 + 8ca) + c(c2 + 8ab)
1
√
Since f (t) = t
is strictly decreasing and f (1) = 1, it suﬃces to prove that
This follows from the AM-GM inequality (or one can use Muirhead’s theorem).
(For a diﬀerent normalisation and a generalisation, refer to problem 3.6 on
page 401.)
Techniques 93
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.
Proof: Let a, b, c be the sides of a triangle. Then we can ﬁnd positive reals
x, y, z such that
a = y + z, b = z + x, c = x + y.
It is easy to compute
P (1, 1, 1) = 6p + 6q + 6r ≥ 0,
P (1, 1, 0) = 4p + 2q ≥ 0,
P (2, 1, 1) = 20p + 14q + 12r ≥ 0.
Note that
P (x, y, z) = 2p x3 + y 3 + z 3
+q x2 y + y 2 z + z 2 x + xy 2 + yz 2 + zx2 + 6rxyz.
Hence some computation leads to
P (a, b, c) = (4p + 2q) x3 + (6p + 5q + 6r) x2 y + xy 2
cyclic cyclic cyclic
+12(q + r)xyz
= Q(x, y, z).
Observe that
Q(1, 1, 1) = 48(p + q + r) ≥ 0
Q(1, 1, 0) = 20p + 14q + 12r ≥ 0,
Q(1, 0, 0) = 4p + 2q ≥ 0.
Techniques 95
Solution: If we take
P (x, y, z) = 2 x + y + z x2 + y 2 + z 2 − 3 x3 + y 3 + z 3 + 3xyz ,
then P (x, y, z) is a homogeneous polynomial of degree 3 in three variables.
Moreover, P (1, 1, 1) = 0, P (1, 1, 0) = 2 and P (2, 1, 1) = 0. Hence Stolarsky’s
theorem is applicable and we conclude that P (a, b, c) ≥ 0 whenever a, b, c are
the sides of a triangle. (Note that we cannot apply theorem 14 directly, since
P (1, 0, 0) = −1 < 0.)
x4 + y 4 + z 4 = (x2 + y 2 + z 2 )2 − 2(x2 y 2 + y 2 z 2 + z 2 x2 )
2 2
= p − 2q − 2 q 2 − 2pr
= p4 − 2p2 q + 2q 2 + 4pr.
5.
(x + y)(y + z)(z + x)
= (p − x)(p − y)(p − z)
= p3 − p2 (x + y + z) + p(xy + yz + zx) − xyz
= pq + r.
6.
7.
8.
9.
x2 (y + z) + y 2 (z + x) + z 2 (x + y)
= x2 (p − x) + y 2 (p − y) + z 2 (p − z)
= p(x2 + y 2 + z 2 ) − (x3 + y 3 + z 3 )
= p(p2 − 2q) − (p3 − 3pq + 3r) = pq − 3r.
10.
x 3 y 3 + y 3 z 3 + z 3 x3
= (xy + yz + zx)3 − 3(xy + yz + zx)xyz(x + y + z) + 3x2 y 2 z 2
= q 3 − 3pqr + 3r2 .
11.
Solution: We have
But
x2 y 2 + y 2 z 2 + z 2 x2 ≥ (xy)(yz) + (yz)(zx) + (zx)(xy),
98 Inequalities
Since (2, 1, 0) ≺ (3, 0, 0), the result follows from Muirhead’s theorem.
1 1
Example 2.56. [10] Let a, b, c be positive real number such that + +
a+1 b+1
1
= 1. Prove that
c+1
1 1 1
+ ++ ≥ 1.
8ab + 1 8bc + 1 8ca + 1
Example 2.57. Let a, b, c be positive real numbers such that a+b+c = 1. Prove
that
6 a 3 + b3 + c 3 + 1 ≥ 5 a 2 + b2 + c 2 .
Example 2.59.
(p + q)2 (p − 2q) (p − q)2 (p + 2q)
≤r≤ . (2.9)
27 27
Equality holds if and only if some two of a, b, c are equal.
f (x) = 6x − 2p.
Hence
p+q
f (x2 ) = 6 − 2p = 2q > 0
3
Hence f has a local minimum at x = x2 . Similarly, we see that f (x1 ) =
−2q < 0 and hence f has a local maximum at x2 . Thus we see that f (x2 ) ≤ 0
and f (x1 ) ≥ 0. But it is easy to compute that
(p + q)2 (p − 2q)
f (x2 ) = − r,
27
and
(p − q)2 (p + 2q)
f (x1 ) = − r.
27
Combining, we get
(p + q)2 (p − 2q) (p − q)2 (p + 2q)
≤r≤ .
27 27
Equality holds if and only if f (x1 ) = 0 = f (x2 ). Since f (x1 ) = 0 = f (x2 ), it
follows that equality holds if and only if f has a double root. This means some
two of a, b, c are equal.
We can use this in proving several inequalities.
102 Inequalities
a4 + b4 + c4 ≥ abc(a + b + c).
p2 + 2q 2
a 2 + b2 + c 2 = ,
3
and
−p4 + 8p2 q 2 + 2q 4
a 4 + b4 + c 4 = + 4pr.
9
Using p = 1, the inequality reduces to
−1 + 8q 2 + 2q 4
+ 4r ≥ r.
9
This can be written in the form
−1 + 8q 2 + 2q 4 + 27r ≥ 0.
1 − 8q 2 + 2q 4 + (1 + q)2 (1 − 2q) ≥ 0.
This simpliﬁes to
q 2 (2q 2 − 2q + 5) ≥ 0.
However, we can write
2 2 q 2 (2q − 1)2 + 9
q (2q − 2q + 5) = ,
2
which is non-negative.
p2 + 2q 2
9 = a 2 + b2 + c 2 = .
3
Hence p2 + 2q 2 = 27. Using (2.9), we get
Techniques 103
(p + q)2 (p − 2q)
2(a + b + c) − abc = 2p − r ≤ 2p −
27
54p − p3 + 3pq 2 + 2q 3
= .
27
However, we see that
54p − p3 + 3pq 2 + 2q 3 = 54p − p(p2 + 2q 2 ) + 5pq 2 + 2q 3
= p(27 + 5q 2 ) + 2q 3 .
Hence we have to prove that
p(27 + 5q 2 ) ≤ 270 − 2q 3 .
However,
2 2
270 − 2q 3 − p(27 + 5q 2 )
= 27(q − 3)2 2q 4 + 12q 3 + 49q 2 + 146q + 219 ≥ 0.
This implies the required result.
Introduce p, q, r by
p = a + b + c, q= a2 + b2 + c2 − ab − bc − ca, r = abc.
27r + 4q 2 − 1 ≥ 0.
Observe that
0 ≤ q = (a + b + c)2 − 3(ab + bc + ca) ≤ 1.
Hence q ∈ [0, 1]. If q ≥ 1/2, the inequality is obvious. If q ≤ 1/2, then (2.9)
shows that
a 2 + b2 + c 2 8abc
+ ≥ 2.
ab + bc + ca (a + b)(b + c)(c + a)
Techniques 105
a 2 + b2 + c 2 8abc
−1+ − 1 ≥ 0.
ab + bc + ca (a + b)(b + c)(c + a)
This reduces to
where
abc
fa = b + c − a −
ab + bc + ca
and similar expressions for fb , fc . We may assume a ≥ b ≥ c Hence
Thus we have
But
2abc 4c(bc + ca)
fa + fb = 2c − = ≥ 0.
ab + bc + ca ab + bc + ca
Similarly, fc + fb ≥ 0. Hence we obtain
where
since
a4 + 3b2 c2 − 2a2 bc = (a2 − bc)2 + 2b2 c2 ≥ 0.
Hence fa ≥ 0 using a ≥ b ≥ c. Similarly
Therefore
However
and
For a proof of this, refer [11]. It is not necessary, we should take the average
√
(aj + ak )/2. We can also take geometric mean aj ak or root-mean-square
2
a2j + a2k
, whichever is convenient. The method is extremely useful in four
2
variables inequalities.
The method can be described as follows. Suppose we take some numbers
ap and aq among a1 , a2 , . . . , an and replace them by (ap + aq )/2. Let the new
sequence be (b1 , b2 , . . . , bn ). Check that f (a1 , a2 , . . . , an ) ≥ f (b1 , b2 , . . . , bn ).
Suppose we check that f (a, a, . . . , x) ≥ 0 for all a and x. Then it follows that
f (a1 , a2 , . . . , an ) ≥ 0. This is because replacing ap , aq by (ap + aq )/2 inﬁnitely
many times leads to taking all equal to the average (a1 + a2 + · · · + an )/n. We
explain this by examples.
In order to apply the theorem 16, it is enough to check that f (a, x, x, x) ≤ 1/27
whenever a + 3x = 1. But we have
176 3
f (a, x, x, x) = 3ax2 + x3 − ax .
27
Hence we have to show that
176 3 1
3ax2 + x3 − ax ≤
27 27
under the condition 3x + a = 1. Replacing a by (1 − 3x), this can be written
as
(1 − 3x)(4x − 1)2 (11x + 1) ≥ 0.
Since x ≤ 1/3, the result follows. Equality holds whenever a = b = c = d = 1/4
and a = b = c = 1/3, d = 0 and permutations thereof.
Observe that
if and only if
(a − 1)2 (3a − 4)(5a + 28) ≤ 0.
Since 3a + x = 4, we have 3a ≤ 4 and the result follows.
Using a1 ≥ a2 ≥ · · · ≥ an , we get
√
(a1 + a2 + · · · + an )(a1 + 2 a2 an + a3 + · · · + an−1 )
√
≥ (2a2 + (n − 2)an )(a2 + 2 a2 an + (n − 3)an )
√
≥ 2 2(n − 2) 2 + 2 n − 3 a2 an ≥ 3na2 an ,
f (a1 , a, a, a, . . . , a) ≥ n + 3
√
where a = n−1
a2 a3 · · · an and a1 = 1/an−1 . Therefore we get
n−1 3nan−1
f (a1 , a, a, . . . , a) = an−1 + ++ .
a 1 + (n − 1)an
n 3(n + 1)an
g(a) = an + + .
a 1 + nan+1
110 Inequalities
3.1 Introduction
Many of the inequalities we have studied and the techniques we have learnt have
their direct implications in a class of inequalities known as geometric inequal-
ities. These inequalities explore relations among various geometric elements.
For example, when we consider a triangle, we can associate many things with
it: angles, sides, area, medians, altitudes, circum-radius, in-radius, ex-radii and
so on. We have already some inequalities, viz., triangle inequalities associated
with the sides: a < b + c, b < c + a, c < a + b, where a, b, c are the sides
of a triangle; these conditions are necessary and suﬃcient for the existence of
a triangle with sides a, b, c. We can derive various relations among these ge-
ometric elements. The classic example is Euler’s inequality: R ≥ 2r, where
R is the circum-radius and r is the in-radius. This chapter provides several
inequalities of this kind, but the list is not exhaustive. For an excellent and a
fairly exhaustive collection of geometric inequalities, please refer to [5] and [6].
3.2 Notations
For a triangle ABC, we use the following standard notations:
• a = |BC|, b = |CA|, c = |AB|;
• α = ∠BAC, β = ∠CBA, γ = ∠ACB;
• ma , mb , mc are respectively the lengths of the medians from A, B, C on
to BC, CA, AB;
• ha , hb , hc are respectively the lengths of the altitudes from A, B, C on
to BC, CA, AB;
• wa , wb , wc are respectively the lengths of the angle bisectors of ∠BAC,
∠CBA, ∠ACB;
• R is the circum-radius; r is the in-radius; ra , rb , rc are the ex-radii;
• Δ is the area of ABC;
• s is the semi-perimeter of ABC: s = (a + b + c)/2;
• O is the circum-centre; H is the ortho-centre; I is the in-centre; G is the
centre of gravity; N is the nine-point centre;
112 Inequalities
• Ω1 denotes the ﬁrst Brocard point of ABC; i.e., the unique point inside
ABC such that ∠CAΩ1 = ∠ABΩ1 = ∠BCΩ1 = ω, the Brocard angle of
ABC. It is known that ω ≤ π/6.
t3 − 2st2 + s2 + r2 + 4Rr t − 4Rrs = 0. (3.1)
Similarly, we can prove that b, c also satisfy the cubic equation (3.1). Now
using the relations between the roots and coeﬃcients of a cubic equation, we
obtain
a = 2s, ab = s2 + r2 + 4Rr, abc = 4Rrs.
Using (3.1), we obtain the equation for the reciprocals of the sides of a triangle.
Thus, 1/a, 1/b, 1/c satisfy the equation:
4Rrst3 − s2 + r2 + 4Rr t2 + 2st − 1 = 0. (3.2)
114 Inequalities
This gives
1 s2 + r2 + 4Rr 1 1
= , = .
a 4Rrs ab 2Rr
1 1 1
sr2 − r(r + 4R) +s − 1 = 0.
(s − a)3 (s − a)2 (s − a)
t3 − (r + 4R)t2 + s2 t − s2 r = 0. (3.6)
whose roots are 1/ra , 1/rb , 1/rc . We hence obtain the relations
1 1 1 r + 4R
= , = .
ra r ra rb s2 r
3
3.4.2. abc < a2 (s − a) ≤ abc.
2
Proof: We have
2 a2 (s − a) = a2 (b + c − a) + b2 (c + a − b) + c2 (a + b − c)
= a2 b + ab2 − a3 .
Thus, we obtain
2 a2 (s − a) = (b + c − a)(c + a − b)(a + b − c) + 2abc.
3 a
3.4.3. ≤ < 2. Equality holds on the left if and only if a = b = c.
2 b+c
Proof: We have proved this in chapter 1; refer to (1.4). The left hand side
of the above inequality is generally known as Nesbitt’s inequality. There are a
variety of ways of proving this. We give two such proofs.
This gives
a (a + b + c)2 3(ab + bc + ca) 3
≥ ≥ = ,
b+c 2(ab + bc + ca) 2(ab + bc + ca) 2
√ √ √ √ √
3.4.4. s< s−a+ s−b+ s−c≤ 3s.
Proof: The ﬁrst inequality follows from the fact that for any positive reals
x, y, z, we have √ √ √
√
x + y + z < x + y + z.
√
On the other hand f (x) = x is a concave function on (0, ∞). Hence,
1√ 1
s−a≤ (s − a),
3 3
which gives √ √ √ √
s−a+ s−b+ s−c≤ 3s.
√
3 3
3.4.5. 0 < sin α + sin β + sin γ ≤
.
2
Proof: We know that OH 2 = 9R2 − a2 + b2 + c2 , which gives
a2 + b2 + c2 ≤ 9R2 .
Hence, we obtain
(a + b + c)2 ≤ 3 a2 + b2 + c2 ≤ 27R2 .
This gives √
0 < a + b + c ≤ 3 3R.
Using a = 2R sin α, etc., we get
√
3 3
0 < sin α + sin β + sin γ ≤ .
2
Proof: Consider f (x) = − sin x on (0, π). This is a convex function. Suppose
α ≤ β ≤ γ. Using Theorem 10, we see that for any permutation (α , β , γ ) of
(α, β, γ), we have
− sin α + α ≤ − sin α + α = − sin 2α.
This reduces to
sin 2α ≤ sin α + α .
√
" 3 3
3.4.7. sin α ≤ .
8
" 3 √ 3 √
sin α 3 3 3
sin α ≤ ≤ = .
3 2 8
" 1
3.4.8. 0< sin α/2 ≤ .
8
2 2
Proof: We know that OI = R 1 − 8 sin α/2 . Hence, it follows that
1 − 8 sin α/2 ≥ 0. The ﬁrst inequality is obvious. Equality holds if and
only if the triangle is equilateral.
3
3.4.9. 1< cos α ≤ .
2
Proof:
We use the known identity: cos α = 1 + 4 sin α/2 . Since
sin α/2 > 0, we get the left side inequality. Now using the inequality
3.4.8, we get the right side inequality.
3
3.4.10. 1< sin(α/2) ≤ .
2
120 Inequalities
" 1
3.4.11. cos α ≤ .
8
Proof: Using 3.4.9 and the AM-GM inequality, we get
" 3
cos α 1
cos α ≤ ≤ .
3 8
This simpliﬁes to
2x2 yz cos α + 2xy 2 z cos β − 2xyz 2 cos(α + β) ≤ y2 z2 .
3.4.13. In an obtuse-angled triangle, cos2 α > 1 and sin2 α < 2.
Proof: We have
"
2 2
IH = R 1− cos α = 9R2 − a2 .
Since the triangle is obtuse, cos α < 0. Hence, IH 2 > R2 . This gives
R2 < IH 2 = 9R2 − a2 .
Geometric Inequalities 121
Thus, a2 < 8R2 . This may be put in the form
Equivalently
cos2 α > 1.
√
3.4.14. tan α ≥ 3 3 if the triangle is acute, and tan α < 0 if the
triangle is obtuse.
Proof: Suppose the triangle is acute. In this case tan α, tan β and tan γ are
all positive. We have
" " 1/3
tan α = tan α ≥ 3 tan α .
√
This gives tan α ≥ 3 3. Using tan α = tan α, we get the desired
inequality for acute-angled triangles. Alternatively,
we can also use the fact
that f (x) = tan x is a convex function on 0, π/2 .
If the triangleis obtuse, say α > 90◦ , then
tan α < 0 and tan β, tan γ are
positive. Hence, tan α < 0 and this gives tan α < 0.
√
3.4.15. cot α ≥ 3.
Proof: We have
cos α cos β
cot α + cot β = +
sin α sin β
sin α + β
=
sin α sin β
2 sin γ
=
cos(α − β) + cos γ
2 sin γ
≥ = 2 tan γ/2 .
1 + cos γ
Adding, we obtain
2 cot α + cot β + cot γ ≥ 2 tan α/2 + tan β/2 + tan γ/2 .
122 Inequalities
However using the convexity of f (x) = tan x on 0, π/2 , we obtain
1 α+β+γ
tan α/2 + tan β/2 + tan γ/2 ≥ tan
3 6
1
= tan 30◦ = √ .
3
It follows that
√
tan α/2 + tan β/2 + tan γ/2 ≥ 3.
This gives
√
cot α + cot β + cot γ ≥ tan α/2 + tan β/2 + tan γ/2 ≥ 3.
√
3.4.16. cot α/2 ≥ 3 3.
Proof: Using the convexity of the function f (x) = cot x on (0, π/2), we get
the desired inequality. The convexity part can be proved using the second
derivative of f (x).
2
3.4.17. cot α/2 ≥ cot α/2 cot α .
Proof: Let us put cot α/2 = x, cot β/2 = y, cot γ/2 = z. Then we know
that x + y + z = xyz. We have
(x + y + z)2 = (x + y + z)xyz = x2 − 1 yz + xy.
This gives
x2 + y 2 + z 2 + xy = x2 − 1 yz.
It follows that
2 x2 + y 2 + z 2 ≥ x2 + y 2 + z 2 + xy = x2 − 1 yz.
Thus, we obtain
"
2
cot α/2 ≥ cot α/2 cot α
= cot α/2 cot α .
√
3.4.18. cot α ≤ 1/ 3 3 in an acute-angled triangle and cot α < 0 for
an obtuse-angled triangle.
3.4.19. cot2 α ≥ cot α cot β = 1.
3.4.20. sec α ≥ 6.
Proof: Using cos α ≤ 3/2 (see 3.4.9) and AM-HM inequality, we get
9 2
sec α ≥ ≥ 9 × = 6.
cos α 3
√
3.4.21. cosec α ≥ 2 3.
Proof: This follows from the convexity of f (x) = cosec x on (0, π). In fact
f (x) = cosec x cosec 2 x + 2 cot2 x) ≥ 0,
on (0, π). Hence f (x) = cosec x is convex on (0, π). This gives
1 α+β+γ 2
cosec α ≥ cosec = cosec 60◦ = √ .
3 3 3
Hence, we get √
cosec α ≥ 2 3.
124 Inequalities
3.4.22. cosec 2 α ≥ 4.
We
have used
√ the AM-GM inequality at the end. However we know that
sin α ≤ 3 3/8 (see 3.4.7). It follows that
4
cosec 2 α ≥ 3 × = 4.
3
2
3.4.23. sin α/2 ≤ cos2 α/2 .
This simpliﬁes to
2
sin α/2 ≤ cos2 α/2 .
√
3.4.24. a2 + b2 + c2 ≥ 4 3Δ (Weitzenböck’s inequality).
It follows that √
a2 + b2 + c2 ≥ 4 3Δ.
a2 = b2 + c2 − 2bc cos α
= (b − c)2 + 2bc(1 − cos α)
= (b − c)2 + 4Δ tan(α/2),
since
1 − cos α
2bc(1 − cos α) = 4Δ = 4Δ tan(α/2).
sin α
Thus, we get
a2 + b2 + c2 = 4Δ tan(α/2) + tan(β/2) + tan(γ/2) + (a − b)2 .
126 Inequalities
However, we know that f (x) = tan x is a convex function on (0, π/2). Hence,
it follows that
1 α+β+γ 1
tan(α/2) + tan(β/2) + tan(γ/2) ≥ tan = tan 30◦ = √ .
3 6 3
This shows that √
a2 + b2 + c2 ≥ 4 3Δ + (a − b)2 .
We observe that equality holds if and only if a = b = c.
√
3.4.25. ab + bc + ca ≥ 4 3Δ.
1/3
8 √
ab ≥ 6Δ √ = 4 3Δ.
3 3
3.4.26. a4 ≥ 16Δ2 .
3.4.27. a2 b2 ≥ 16Δ2 .
3
2 4Δ
3.4.28. (abc) ≥ √ .
3
Geometric Inequalities 127
√
Proof: We use a + b + c = 2R sin α + sin β + sin γ ≤ (3 3)R (see 3.4.5) and
abc = 4RΔ. Thus,
4Δ abc 3abc
√ = √ ≥ ≤ (abc)2/3 ,
3 R 3 a+b+c
where we have used the AM-GM inequality in the last step. The result follows
by taking cubes on both the sides.
Proof: We know that a, b, c are the roots of the cubic equation (see section
3.3.1) t3 − 2st2 + s2 + r2 + 4Rr t − 4Rrs = 0. Since all its roots are real, the
derivative polynomial 3t2 − 4st + s2 + r2 + 4Rr = 0 has only real roots. This
imposes a condition on the discriminant, namely, s2 ≥ 3r(r + 4R). This gives
the left inequality.
Again we know that ra , rb , rc are the roots of the equation (see section
3.3.4) t3 − (r + 4R)t2 + s2 t − s2 r = 0. We look at the derivative polynomial:
3t2 − 2(r + 4R)t + s2 = 0. This again has only real roots. Hence, using the
condition on the discriminant, we obtain 3s2 ≤ (r + 4R)2 , which is the other
inequality.
3.4.31. s2 ≥ 27r2 .
Proof: We have
s (s − a) + (s − b) + (s − c) 1/3
= ≥ (s − a)(s − b)(s − c)
3 3
2 1/3
Δ
=
s
2 1/3
= r s .
On cubing the above relation and rearranging the terms, we get the result.
3.4.32. 36r2 ≤ a2 ≤ 9R2 .
128 Inequalities
where we have used the Cauchy-Schwarz inequality. This gives the left side
inequality. On the other hand, using OH 2 = 9R2 − a2 + b2 + c2 , we get the
right side inequality.
a2
3.4.33. ≥ 4.
r b rc
Proof: We know (see 3.3.4) that rb rc = s2 . Using the Cauchy-Schwarz
inequality, we get
a2 2
( a) 4s2
≥ = 2 = 4.
r b rc r b rc s
√
3.4.34. 5R − r ≥ 3s.
Proof: We use ra = 4R + r (3.3.4) and R ≥ 2r (3.4.29). Thus,
5R − r = 4R + r + R − 2r ≥ 4R + r = ra .
3.4.35. a(s − a) ≤ 9rR.
Geometric Inequalities 129
Proof: We know that r = (s − a) tan α/2 , etc. Hence,
ar
a(s − a) = = 4Rr cos2 α/2 .
tan α/2
9
Thus, it is suﬃcient to prove that 2 cos2 α/2 ≤ . This is equivalent to
2
cos α ≤ 3/2. This follows from 3.4.9.
√
3.4.36. ha ≤ 3s.
Proof: We have
3abc
(a + b + c)2 ≥ 3(ab + bc + ca) = ha + hb + hc ) = 6R ha .
2Δ
Thus, we obtain
(a + b + c)2 s(a + b + c)
ha ≤ = .
6R 3R
√ √
But we know that a + b + c ≤ 3 3R. It follows that ha ≤ 3s.
√
1 3
3.4.37. ≤ .
a 2r
Proof: We have
1
ha
= .
a 2Δ
Thus, it is suﬃcient to prove that
√
3Δ
ha ≤ .
r
This follows from Δ = rs and 3.4.36.
√
1 3 3
3.4.38. ≥ .
a 2(r + R)
9
3.4.39. 9r ≤ ra ≤ R.
2
Proof: We know from the section 3.3.4 that ra = 4R + r. Using 2r ≤ R,
we obtain 9r ≤ 4R + r ≤ (9/2)R.
√
3.4.40. 9r ≤ ha ≤ 3 s.
Proof: We have
√
1 3 √
ha = 2Δ ≤ 2Δ = 3 s,
a 2r
and 9
ha ≥ = 9r.
(1/ha )
(See 3.3.2.)
3.4.41. h2a ≤ (3/4) a2 .
Proof: We know that ha ≤ ma , 4m2a = 2b2 + 2c2 − a2 and similar results for
hb , hc , mb , mc . Thus, it follows that
3 2
h2a ≤ m2a = a + b2 + c 2 .
4
Geometric Inequalities 131
a2
3.4.42. ≥ 2.
h2b + h2c
Proof: We write
a2 a 2 b2 c 2
= .
h2b 2
+ hc 4Δ b2 + c2
2
3.4.43. ha ≤ wa ≤ 3(R + r).
Proof: The ﬁrst part is obvious. For the second inequality, we begin with the
expression
2bc
wa = cos α/2 ,
b+c
and similar
expressions for wb and wc . Since 2bc ≤ (b + c)2 , we get wa2 ≤
bc cos α/2 , and similar estimates hold for wb and wc2 . Now the Cauchy-
2 2
We use
" "
1− cos2 α = 2 cos α, cos α = 1 + 4 sin α/2 .
and hence
" r2 16R2 sin2 α/2 "
cos α ≤ = = 8 sin2 α/2 .
2R2 2R2
This completes the proof of the desired inequality.
√
3.4.44. ha hb ≤ 3 3Δ.
3.4.45. ha ≥ 27r3 .
Geometric Inequalities 133
" 1 3
1 1 1
≤ = .
ha 3 ha 27r3
1 3
3.4.46. ≥ .
ha − 2r r
1 1
Proof: Using = , we get
ha r
ha r
ha − 2r
1=3−2= −2 = .
ha ha ha
This gives
ha
≥ 9.
ha − 2r
But the left hand side is equal to
2r
3+ .
ha − 2r
Simpliﬁcation gives
1 3
≥ .
ha − 2r r
√
3.4.47. 3Δ ≤ r(4R + r).
134 Inequalities
Proof: We start with the identity, IH 2 = 2r2 − 4R2 cos α. However
"
4R2 cos α = 2R2 sin2 α − 2
1 2
= a − 4R2
2
1 2
= a − ab − 4R2
2
1 2 2
= a − s + r(4R + r) − 4R2
2
1 2 2
= a − 2R + r .
4
Here we have used ab = s2 + r(4R + r); see 3.3.1. Thus, it follows from
IH 2 ≥ 0 that
s2 ≤ 4 4R2 + 4Rr + 3r2 .
This gives
Δ2 ≤ 4r2 4R2 + 4Rr + 3r2
1
= r2 4R2 + 8/3 Rr + 1/3 r2 + 4Rr + 8r2 .
3
2 2
4
Δ ≤ r 4R + 8/3 Rr + 1/3 r2 + R2
2
3
1 2 2
= r 4R + r .
3
√
This implies that 3Δ ≤ r 4R + r .
3
3.4.48. s< ma < 2s.
2
Using 2ma = AA < AB + BA = b + c and similar inequalities for 2mb ,
2mc , we get
1
m a + mb + mc < b + c + c + a + a + b = 2s.
2
Let C be the point of intersection of BC with a line through A and parallel
to the median BE. Then the sides of AA C are 2ma , 2mb and 2mc . The
lengths of its medians are 3a/2, 3b/2 and 3c/2. Applying the inequality we
have just proved to this triangle, we get
3
a + b + c < 2 m a + mb + m c .
2
This gives the left side inequality.
3.4.49. ma ≤ 4R + r.
√ √ √
3.4.50. s< wa ≤ s s − a ≤ 3s.
136 Inequalities
√
3.4.51. wa2 ≥ 3 3Δ.
min{a, b, c}(OD + OE + OF ) ≤ a · OD + b · OE + c · OF
= 2[BOC] + 2[COA] + 2[AOB]
= 2Δ.
3.4.53. a sin α/2 ≥ s.
Proof: We ﬁrst prove that in a triangle ABC, the following inequality holds:
" "
sin α/2 ≤ 1 − sin α/2 . ()
we obtain
sin(α/2) −1 + sin(α/2)
≤ .
cos(α/2) cos(α/2)
However using r = 4R sin(α/2) and s = 4R cos(α/2), the left side is simply
r/s. Thus, the inequality takes the form
r −1 + sin(α/2)
≤ .
s cos(α/2)
(s − a) sin(α/2) ≤ s − 1 + sin(α/2) .
This reduces to
a sin(α/2) ≥ s.
3.4.54. 3 cos α ≥ 2 sin α sin β.
But then
2
cos2 α ≥ cos α cos β =⇒ cos α ≥3
cos α cos β
3
=⇒ 3 cos α cos β ≤ cos α
2
=⇒ 2 cos α cos β ≤ cos α
Geometric Inequalities 139
=⇒ cos(α − β) + cos(α + β) ≤ cos α
=⇒ cos(α − β) ≤ 2 cos α
=⇒ cos(α − β) − cos(α + β) ≤ 3 cos α
=⇒ 2 sin α sin β ≤ 3 cos α.
We have used the well known result: cos α ≤ 3/2. (3.4.9)
3
3.4.55. max ra , rb , rc ≥ R.
2
Proof: The required inequality is equivalent to
1 1 1 3R
max , , ≥ .
s−a s−b s−c 2Δ
This is further equivalent to
0 1 3Rr 3R
max tan α/2 , tan β/2 , tan γ/2 ≥ = .
2Δ 2s
This may also be written in the form
0 1 3
sin α max tan α/2 , tan β/2 , tan γ/2 ≥ .
2
However, we have
0 1
sin α max tan α/2 , tan β/2 , tan γ/2
≥ sin α tan α/2 = 2 sin2 α/2 .
On the other hand,
3 1
sin2 α/2 = − cos α
2 2
3 1 "
= − 1+4 sin α/2
2 2
3 1 r
= − 1+
2 2 R
r
= 1− .
2R
Since 2r ≤ R, the result follows. In fact we have proved more:
max ra , rb , rc ≥ 2R − r.
3.4.56. R − 2r ≥ wa − ha .
140 Inequalities
Proof: We use
" 4R sin β sin γ
r = 4R sin(α/2), wa = cos(α/2),
sin β + sin γ
and
ha = 2R sin β sin γ.
The inequality to be proved is
" 2 sin β sin γ
1−8 sin(α/2) ≥ 1 − cos (β − γ)/2 .
cos (β − γ)/2
Let us put t = cos (β − γ)/2 and x = sin(α/2). Then t ∈ (0, 1] and x ∈ (0, 1).
We have to prove
f (x; t) = 2 1 + t x2 − 4t2 x + t + 2t2 − t3 ≥ 0.
Consider this as a function of x, say g(x). We observe that
g (x) = 4(1 + t)x − 4t2 , g (x) = 4(1 + t) > 0.
Hence, g has the minimum at x = t2 /(1 + t). But observe that
2
t t(1 − t)(t + 3)
f ;t = ≥ 0.
1+t t+1
Hence, minx f (x; t) ≥ 0. Thus, f (x; t) ≥ 0 for all x ∈ (0, 1) and t ∈ (0, 1].
Equality holds if and only if t = 1. Equivalently x = 1/2. This is equivalent to
α = π/3 and β = γ, which corresponds to the case of an equilateral triangle.
√
3.4.57. s(s − a) + s(s − b) + mc ≤ 3 s.
Here equality holds if and only if D, E, F are the mid-points of the respective
line segments on which they lie.
BD DC CE EA AF FB
= x, = x , = y, = y , = z, = z.
a a b b c c
Let AX and F Y be the altitudes drawn respectively from A and F on to BC.
142 Inequalities
(b) If P QR and P QS are two triangles with the common side P Q and if
∠RP Q = x1 , ∠RQP = y1 , ∠SP Q = x2 , ∠SQP = y2 , then p(P QR) <
p(P QS) or p(P QR) = p(P QS) or p(P QR) > p(P QS) according as
u1 v1 < u2 v2 or u1 v1 = u2 v2 or u1 v1 > u2 v2 respectively. .
proof: It is easy to prove that
Observe that u1 v1 < 1 and u2 v2 < 1. The result follows from comparison.
uu1 + u1 u2 + u2 u = 1,
vv1 + v1 v2 + v2 v = 1,
ww1 + w1 w2 + w2 w = 1,
u2 u1 = uw, v1 w2 = vw, v 2 u1 = kvu.
144 Inequalities
kuv vw uw
u1 = , v1 = , w1 = .
v2 w2 u2
kuv kuv
u· + · u2 + u2 u = 1,
v2 v2
vw vw
v· + · v2 + v2 v = 1,
w2 w2
uw uw
w· + · w2 + w2 w = 1.
u2 u2
w2 − v 2 w uw(w + w2 )
v2 = , u2 = .
v(w2 + w) 1 − ww2
We show that
w2 − v 2 w u2 w2 + u2 ww2 − 1 + ww2 + u2 v 2 1 + w2 w + w2
2
= w 1 + u2 w 2 − v .
Geometric Inequalities 145
We have
w2 − v 2 w u2 w2 + u2 ww2 − 1 + ww2 + u2 v 2 1 + w2 w + w2
= u2 v 2 w + w2 + u2 w2 w2 + u2 ww22 − w2
+ww22 + v 2 w − v 2 w2 w2
= u2 v 2 w + w2 u2 v 2 + v 2 w2 + w2 u2 − 2v 2 w2 w2
+u2 ww22 − w2 + ww22 + v 2 w
0
= w u2 v 2 − 2vw2 uv + vw + wu
1
+ u2 + 1 v 2 + w22 − 2vw2 + 2vw2
2
= w 1 + u2 w 2 − v ;
λ ≤ R(DEF ) ≤ μ,
where
λ = min R(AF E), R(BDF ), R(CED) ,
μ = max R(AF E), R(BDF ), R(CED) .
R(DEF ) sin γ
= .
R(CED) sin ∠EF D
However, α + β + γ = π = ∠F DE + ∠DEF + ∠EF D and hence at least one
of the ratios α : ∠F DE, β : ∠DEF , γ : ∠EF D is ≥ 1, and at least one of
these ratios is also ≤ 1. Since all the angles are acute, at least one of the ratios
R(DEF ) : R(AF E), R(DEF ) : R(BDF ), R(DEF ) : R(CED) is greater than
or equal to 1, and at least one of these ratios is less than or equal to 1. This
proves that
λ ≤ R(DEF ) ≤ μ.
Remark: If r(XY Z) denotes the in-radius of triangle XY Z, it is true that
r(DEF ) ≥ min r(AF E), r(BDF ), r(CED) .
See [5].
Proof: Let BD = x1 , DC = y1 , CE = x2 , EA = y2 , AF = x3 , F B = y3 .
Then
a ≥ a − y3 cos β − x2 cos γ,
b ≥ b − y1 cos γ − x3 cos α,
c ≥ c − y2 cos α − x1 cos β.
Since cos α, cos β, cos γ are positive, the above estimates give
a cos α ≥ a cos α − a cos β cos γ.
However,
a cos α = 2R sin α cos α = R sin 2α = 4R sin α sin β sin γ,
and
a cos β cos γ = 2R sin α cos β cos γ
= 2R cos γ sin(α + β) + 2R cos α cos β sin γ
= 2R sin γ cos γ + 2R cos α cos β sin γ
= 2R sin γ cos γ + cos α cos β
= 2R sin α sin β sin γ.
Geometric Inequalities 147
a cos α ≥ a cos α − a cos β cos γ
= 4R sin α sin β sin γ − 2R sin α sin β sin γ
= 2R sin α sin β sin γ
1
= a cos α.
2
Hence, 2 a cos α ≥ a cos A, as required. It may be seen that equality
holds if and only if D, E, F are mid-points of BC, CA, AB respectively.
3.6.1. Let P be any point in a triangle ABC and s be its semi perimeter.
Prove that
3.6.2. Let ABC be a triangle in which a > b > c. Let P be any interior
Geometric Inequalities 149
point and suppose AP, BP, CP meet the opposite sides in D, E, F respectively.
Prove that
P D + P E + P F < a.
P D + P E + P F < XY + CY + BX = BC = a.
R1 + R2 + R3 ≥ 2(r1 + r2 + r3 ).
Proof: Suppose L and M are arbitrary points chosen on the sides AB and
AC. Complete the parallelograms LAP S and M AP T . Then LM T S is also a
parallelogram and [LAP S] + [M AP T ] = [LM T S].
Observe that [LAP S] = LA · P F , [M AP T ] = M A · P E ,[LM T S] ≤ LM ·
M T = LM · P A. Here, the points L, M can be chosen on the extended lines
AB and AC. Choosing L, M such that AM = AB and AL = AC, we get
AC · P F + AB · P E ≤ LM · P A.
But, the triangle ALM is similar to ACB and hence LM = CB. Thus, we get
bP F + cP E ≤ aP A. It follows that
b c
PA ≥ P F + P E.
a a
Similarly we get
c a a b
P B ≥ P D + P F, PC ≥ P E + P D.
b b c c
150 Inequalities
F E2 = P E 2 + P F 2 − 2P E cot P F · cos ∠F P E
= P E 2 + P F 2 + 2P E cot P F · cos α
= (P E sin γ + P F sin β)2 + (P E cos γ − P F cos β)2
≥ (P E sin γ + P F sin β)2 .
It follows that
FE P E sin γ + P F sin β
PA = ≥ .
sin α sin α
Similarly we get
P D sin γ + P F sin α P E sin α + P D sin β
PB ≥ , PC ≥ .
sin β sin γ
Adding these we get
sin γ sin β sin γ sin α
PA + PB + PC ≥ PD + + PE +
sin β sin γ sin α sin γ
sin α sin β
+P F +
sin β sin α
≥ 2(P D + P E + P F );
Geometric Inequalities 151
2c
KN = [ABC].
a + b2 + c 2
2
Moreover
R1 = 2r1 , R2 = 2r2 , R3 = 2r3 .
Using these we get
1 1
≥2 .
r1 r2 R 1 r1
The same process adopted to the point P and the triangle A B C gives
1 1
≥2 .
R 1 r1 R1 R2
3.6.5. R 1 r1 ≥ 2 r1 r 2 .
Geometric Inequalities 153
BA · P F + CA · P E ≤ BC · P A.
This reduces to cr3 + br2 ≤ aR1 . This may be written in the form
c b
R 1 r1 ≥ r3 r1 + r2 r1 .
a a
Similarly, we obtain
a c
R 2 r2 ≥ r 1 r2 + r3 r2
b b
b a
R 3 r3 ≥ r2 r3 + r 1 r3 .
c c
Adding these three inequalities, we get
b a
R 1 r1 ≥ r1 r2 +
a b
≥ 2 r1 r2 ;
we have used the AM-GM inequality. Equality holds if and only if ABC is
equilateral and P is its centre.
Consequences:
We obtain
cosec α/2 ≥ 6.
and
(iv) If we take P = G, the centroid, then the inequality takes the form
ma h a ≥ ha hb .
AP BP CP x z y x
z y
+ + = + + + + +
PD PE PF z x x y y z
≥ 2 + 2 + 2 = 6.
Geometric Inequalities 155
Similarly,
AP BP CP (x + y)(y + z)(z + x)
· · =
PD PE PF xyz
√ √ √
2 xy · 2 yz · 2 zx
≥ = 8.
xyz
We also observe that
AD AP x+y+z AD x+y+z
=1+ = , = .
PD PD z AP x+y
Thus,
AD BE CF 1 1 1
+ + = (x + y + z) + +
AP BP CP x+y y+z z+x
9
≥ .
2
And
AD BE CF (x + y + z)3
· · = ≥ 27.
PD PE PF xyz
Finally
PD PE PF
+ +
PA PB PC
z x x
= + +
x+y y+z y+z
1 1 1
= (x + y + z) + + −3
x+y y+z z+x
9 3
≥ −3= .
2 2
3.6.7. Prove
ha
(i) ≥ 9;
r1
(ii) ha hb hc ≥ 27r1 r2 r3 ;
(iii) ha − r1 hb − r2 hc − r3 ≥ 8r1 r2 r3 ;
(iv) min ha , hb , hc ≤ r1 + r2 + r3 ≤ max{ha , hb , hc ;
r1 3
(v) ≥ .
ha − r 1 2
156 Inequalities
We easily compute
AP y+z BP z+x CP y+x
= , = , = .
PD x PE y PF z
Thus, we obtain
AD x+y+z BE x+y+z CF x+y+z
= , = , = .
PD x PE y PF z
Using similar triangles P DL and ADT , we have
ha AT AD x+y+z
= = = .
r1 PL PD x
Similarly, we may obtain
hb x+y+z hc x+y+z
= , and = .
r2 y r3 z
Now
ha
1 1 1
= x+y+z + +
r1 x y z
≥ 9.
ha hb h c (x + y + z)3
· · = ≥ 27,
r1 r 2 r3 xyz
Geometric Inequalities 157
We hence obtain
(y + z)2
P A2 = AD2 .
(x + y + z)2
Using Stewart’s theorem, we calculate AD;
(y + z)yb2 + (y + z)zc2 − yza2
AD2 = .
(y + z)2
Thus, we obtain
(y + z)yb2 + (y + z)zc2 − yza2
P A2 = .
(x + y + z)2
Adding, we have
1 1 1 x+y+z a b c 1 1 1
+ + ≥ + + + ha z + y +
rA rB rC 2Δ x y z 2Δ z y
x+y+z a 2
≥ + ha .
2Δ x Δ
However, we know that (see 3.4.38)
√
ha + h b + h c 1 3 3
=2 ≥ .
Δ a R+r
We also observe that
x+y+z a 9
≥ abc)1/3
2Δ x 2Δ
and (3.4.28)
3
2 4Δ
abc ≥ √ .
3
√ 2
Using 3.4.47, we can prove that 3Δ ≤ R + r . Combining all these, we get
√
1 1 1 3 3+2 3
+ + ≥ ,
rA rB rC R+r
Proof: Let us consider the vertices of ABC to be points in the complex plane;
let us denote A, B, C by complex numbers z1 , z2 , z3 respectively. Let us write
z for P . Now consider the function
z − z1 z − z2
g(z) = .
z3 − z 1 z 3 − z 2
This is a quadratic polynomial and g z1 = g z2 = g z3 = 1. Hence, g(z) = 1
for all z. Thus, we have
z − z1 z − z 2 R1 R2
1 = g(z) ≤ =
z3 − z1 z 3 − z 2 .
ab
160 Inequalities
(v) When P = Ω1 , the ﬁrst Brocard point, the sine rule gives
b c c
R1 = sin ω, R2 = sin ω, R3 = sin ω,
sin α sin β sin γ
where ω is the Brocard angle. We then get
a sin α
sin α
≥ .
b sin2 ω
Using the known fact that ω ≤ π/6, we have sin ω ≤ 1/2 and hence
a sin α "
≥4 sin α.
b
r1 a2
= 2 ,
ha a + b2 + c 2
and similar expressions for r2 and r3 . We thus get
r1 r2 12Δ2
= 2 2.
ab ( a )
The inequality now is √
a2 ≥ 4 3Δ.
We obtain 2
sin α
sin4 ω ≤ 2 .
4 sin α sin2 β
Equivalently
1 1
sin4 ω ≤ 2
.
16R 1/a2
1 2
1 1 9
≥ ≥ .
a2 3 a 4(R + r)2
where K is the area- of the. hexagon. Since we are minimising the perimeter of
a hexagon of area 2 ABC , we obtain
2 - .
2 · 2 ABC
2 R1 + R2 + R 3 ≥ 6 √ .
3 3
This simpliﬁes to
2 √ - .
R 1 + R2 + R 3 ≥ 4 3 ABC .
164 Inequalities
Proof:
This simpliﬁes to √
2
3 3
R1 sin α ≥ r1 r2 c.
R
Using a = 2R sin α, etc., this may be written in the form
2 √
R1 sin α ≥ 6 3 r1 r2 sin γ.
However,
R1 sin α)2 = EF 2 = r22 + r32 − 2r1 r2 cos π − α
= r22 + r32 + 2r1 r2 cos α,
or √
3 r1 r2 sin γ ≤ r12 + r2 r3 cos α.
Suppose P = I, the in-centre of ABC. Then r1 = r2 = r3 = r and hence
√ 2
3 r sin α + sin β + sin γ ≤ 3r2 + r2 cos α + cos β + cos γ ,
166 Inequalities
Using
the 2standard identities
4 cos α cos β sin γ = sin 2α + sin 2β − sin 2γ, etc.,
and sin α = 2 + 2 cos α, this reduces to
2
sin 2α + sin 2β + sin 2γ ≤ √ sin2 α + sin2 β + sin2 γ .
3
3.6.13. (AMM, 1978) Let G denote the centroid of a triangle ABC and let
θ = ∠GAB, μ = ∠GBC and ν = ∠GCA. Prove that
3
sin θ + sin μ + sin ν ≤ .
2
Proof: Observe that
Δ Δ Δ
sin θ = , sin μ = , sin ν = .
cma amb bmc
Thus,
1 1 1
sin θ + sin μ + sin ν = Δ + + .
cma amb bmc
Using the Cauchy-Schwarz inequality, we have
2 1 1
sin θ + sin μ + sin ν ≤ Δ2
a2 m2a
2 2 2 2
a b ma mb
= Δ2 2 .
abcma mb mc
Using the standard identities 16Δ2 = 2 a2 b2 − a4 , 4m2a = 2b2 + 2c2 − a2 ,
etc., the above inequality reduces to
4 2 P
sin θ + sin μ + sin ν ≤ ,
9 Q
where
2
P = 2 a 2 b2 − a4 a 2 b2 ,
Q = a2 b2 c2 2a2 + 2b2 − c2 2b2 + 2c2 − a2 2c2 + 2a2 − b2 .
Geometric Inequalities 167
AP · EF = r3 AE + r2 AF
= R1 sin θ2 · R1 cos θ1 + R1 sin θ1 · R1 cos θ2
= R12 sin θ1 + θ2 = R12 sin α,
Consider the triangle DEF and the point P inside it. We show that
DE 2 + EF 2 + F D2 ≤ 3 P D2 + P E 2 + P F 2 , (3.11)
valid for any triangle DEF and any point P inside it. Let K be the mid-point
of EF . Join P K and DK. (See Fig. 3.10.) Appolonius’ theorem applied to
the triangle P EF gives
1
PK = 2P E 2 + 2P F 2 − EF 2 .
2
Similarly,
1
DK = 2DE 2 + 2DF 2 − EF 2 .
2
But DK ≤ P K + KD, with equality if and only if P lies on DK. Thus,
2DE 2 + 2DF 2 − EF 2 ≤ 2P D + 2P E 2 + 2P F 2 − EF 2 .
170 Inequalities
2DE 2 + 2DF 2 ≤ 4P D2 + 2P E 2 + 2P F 2
+P D 2P E 2 + 2P F 2 − EF 2 .
Setting P D2 = x2 and DE 2 = y 2 , the above relation is
1/2
0 ≤ y 2 − 2x2 ≤ x 4x2 − y 2 .
This is equivalent to y 2 y 2 − 3x2 ≤ 0. It follows that y 2 ≤ 3x2 , which is
precisely (3.11). Equality holds if and only if P coincides with the centroid of
DEF . This is equivalent to
[P DE] = [P EF ] = [P F D],
Geometric Inequalities 171
or
r1 r2 sin γ = r2 r3 sin α = r3 r1 sin β.
This shows that
r1 : r2 : r3 = a : b : c.
Thus, equality holds if and only if P is the symmedian point.
Chapter 4
Applications involving inequalities
Some problems, though not direct inequality problems, use inequalities
in their solutions. This is the case when the problems on maximisation and
minimisation are considered. Inequalities are also useful in solving some Dio-
phantine equations by way of estimating the bounds for solutions. They also
help us to determine whether some polynomial equations have real roots. Here
we consider several problems whose solutions use inequalities.
Example 4.1. (IMO, 1987) Let x1 ,x2 ,x3 ,. . .,xn be a sequence of n real num-
n 2
bers such that j=1 xj = 1. Prove that for every k ≥ 2, there are integers
a1 ,a2 ,a3 ,. . .,an , not all zero, such that |aj | ≤ k − 1 and
√
a1 x1 + a2 x2 + · · · + an xn ≤ (k − 1) n .
kn − 1
Solution: For each n-tuple a1 ,a 2n,a3 ,. . .,an of non-negative integersn with
0 ≤ aj ≤ k − 1, associate the sum j=1 aj xj . Observe that there are k such
n-tuples. Using the Cauchy-Schwarz inequality, we obtain
1/2 1/2
n n n
a x ≤ a 2
x 2
j j j j
j=1 j=1 j=1
√
≤ k − 1 n.
- √ .
Now, we split√the*interval 0, (k − 1 n into k n − 1 sub-intervals each of
length k − 1 n kn − 1 . By the pigeonhole principle, we can ﬁnd two n-
n
tuples b1 ,b2 ,b3 ,. . .,bn and c1 ,c2 ,c3 ,. . .,cn such that the sums j=1 bj xj and
n
j=1 cj xj lie in the same sub-interval. Hence, it follows that
√
n n
k−1 n
bj x j −
c j xj ≤ n .
k −1
j=1 j=1
Taking aj = bj − cj , we see that aj ≤ k − 1 for 1 ≤ j ≤ n and
n √
(k − 1) n
aj xj ≤ .
kn − 1
j=1
Example 4.3. (USSR, 1974) Consider a square grid S of 169 points which are
uniformly arranged in 13 rows and 13 columns. Show that no matter what subset
T consisting of 53 points is selected from these 169 points, some four points of T
will always form the vertices of a rectangle R whose sides are parallel to the sides
of S.
13
mj
points of T is . We observe that
j=1
2
13
13
mj 1 2
= m j − mj
j=1
2 2 j=1
13 13
1 1
≥ m2j − mj
2 × 13 j=1
2 j=1
1 2 1
= 53 − 53
26 2
> 2 × 53 − 27
13
= 79 > 78 = .
2
We have used the Cauchy-Schwarz inequality in the second step above. The
total number of pairs of columns in the grid S is 13 2 . Hence, we may apply
the pigeonhole principle to conclude that there is a pair of columns determined
by some two points on some j-th row which coincides with a pair of columns
determined by some two points on some k-th row, j = k. We thus get a
rectangle with sides parallel to the sides of S.
Example 4.4. Consider the set S = a1 , a2 , . . . , an . Let C = {P1 , P2 , . . . , Pn }
be a collection of n 2-element subsets of S such that whenever Pj ∩ Pk
distinct
is not empty then aj , ak is also one of the sets in the collection C . Prove that
each aj appears exactly in two elements of C.
This simpliﬁes to
n
m2j − mj ≤ 2n.
j=1
176 Inequalities
n
Since j=1 mj = 2n, we obtain
n
m2j ≤ 4n.
j=1
m2a + m2b
= 5.
m2c
We obtain
ma + mb √
≤ 10.
mc
√
Equality holds if and only if a = b. Thus, the maximum value is 10.
Example 4.6. Let ABC be a triangle with circum-circle Γ. The medians AD,
BE, CF , when extended, meet Γ in A1 , B1 , C1 respectively. Show that the sum
of the areas of the triangles BA1 C, CB1 A, AC1 B is at least equal to the area of
ABC. When does equality hold?
Applications 177
[BA1 C] a2 a2
= 2
= .
[ABC] 4AD 4m2a
Similarly,
[CB1 A] b2 [AC1 B] c2
= , = .
[ABC] 4m2b [ABC] 4m2c
Thus,
a2 b2 c2
[BA1 C] + [CB1 A] + [AC1 B] = + + [ABC],
4m2a 4m2b 4m2c
a2 b2 c2
+ 2 + ≥ 1.
4m2a 4mb 4m2c
a2 b2 c2
+ + ≥ 1.
2b2 + 2c2 − a2 2c2 + 2a2 − b2 2a2 + 2b2 − c2
Using the Cauchy-Schwarz inequality, observe that
2
2 2
2 2 a
a +b +c = √ 2 2
· a 2b + 2c − a 2
cyclic
2b2 + 2c2 − a2
a2 2 2 2 2
≤ a 2b + 2c − a .
2b2 + 2c2 − a2
cyclic cyclic
Thus, we obtain
2 2
a2 a + b2 + c 2
≥ 2 2 .
2b2 + 2c2 − a2 4 a b + b2 c 2 + c 2 a 2 − a 4 + b4 + c 4
cyclic
xn + a1 xn−1 + · · · + an−1 x + an = 0
= a21 − 2a2 .
Thus,
a21 ≤ na21 − 2na2 ,
and hence (n − 1)a21 − 2na2 ≥ 0. This violates the given condition. Hence not
all α1 , α2 , . . . , αn can be real.
Example 4.8. (Indian Team Selection, 1998) Let M be a positive integer and
consider the set S = {n ∈ N | M 2 ≤ n < (M + 1)2 }. Show that the products of
the form ab where a, b ∈ S are all distinct.
Solution: Suppose there are integers a, b, c, d in the given interval such that
ab = cd. We may assume d is the largest among them so that a < d and b < d.
This implies that c < a and c < b. Put c = a − k and d = a + l for some
positive numbers k and l. Then we have
This shows that a divides kl. Since c, d lie in the set S, we observe that
a − k ≥ M 2 and a + l ≤ M 2 + 2M . Thus, we have
l + k = (a + l) − (a − k) ≤ M 2 + 2M − M 2 = 2M.
Using the AM-GM inequality, we now get
√ k+l
kl ≤ ≤ M.
2
This implies that kl ≤ M 2 . Since a divides kl, we get a ≤ M 2 . If a = M 2 ,
then c < a implies that c is not in the set S. Thus, a < M 2 , a contradiction.
Thus, if ab = cd, then {a, b} and {c, d} must be the same sets.
Example 4.10. ([8]) Determine all real polynomials of degree n with each of its
coeﬃcients in the set {+1, −1} and having only real zeros.
Solution: If n = 1 then p(x) = (x − 1), (x + 1), −(x − 1) and −(x + 1) give all
the polynomials of degree 1. We may assume n > 1. We may also assume that
the leading coeﬃcient of p(x) is 1 by changing the signs of all the coeﬃcients,
if necessary. Thus, p(x) is of the form
n
1
αj2 + = 6.
j=1
αj2
It follows that n ≤ 3. We may verify that for n = 3, ±(x+1)2 (x−1) and ±(x+
1)(x − 1)2 are the required polynomials. In the case n = 2, the polynomials
±(x2 − x − 1) and ±(x2 + x − 1) are the required ones.
Example 4.11. ([8]) Suppose α, β, γ and δ are real numbers such that
α + β + γ + δ = α7 + β 7 + γ 7 + δ 7 = 0.
However we have
0 2 1
4 β 2 + γ 2 + δ 2 + βγ + γδ + βδ + βγδ β + γ + δ
3 2 2 2 42
= β+γ + γ+δ + β+δ − 4αβγδ
3 2 2 2 24
= α+β + α+γ + α+δ − 4αβγδ
3 42
= 3α2 + 2α β + γ + δ + β 2 + γ 2 + δ 2 − 4αβγδ.
This simpliﬁes to
2r4 − 2r2 c2 + c4 − 3c2 + 2) ≤ 0.
For getting a positive value for r2 , we need the discriminant of the above
quadratic expression to be non-negative. This gives
c4 − 6c2 + 4 ≤ 0. (4.3)
Thus, if (4.2) holds for some z, then (4.3) holds. The least number c satisfying
(4.3) is
√
c = 3 − 5.
Thus, for any z, we have
√
max 1 + z , 1 + z 2 ≥ 3− 5.
Taking 2
√
3− 5 2π
z = reiθ , r = ,θ = ,
2 3
√
we see that 1 + z = 1 + z 2 = 3 − 5. Hence, the required minimum is
√
3 − 5.
Applications 183
We show that
n2 (n − 2) + (n − 1)2 √
√ < n + 2.
n(n − 1) n − 2
184 Inequalities
3 2
√ the statement 2n − 6n + 4n − 1 > 0, which is true for
This is equivalent to
n ≥ 4. Thus, xn < n + 1 for n ≥ 4. We obtain
√ √
n < xn < n + 1,
- .
for n ≥ 4. Hence, x2n = n for n ≥ 4.
Example 4.15. ([8]) Show that if all the roots of P (x) = ax4 −bx3 +cx2 −x+1 =
0 are positive, then c ≥ 80a + b.
u + v = 1, αβu = α + β, γδv = γ + δ.
Solution: The second relation shows that xyz(x + y + z)3 ≥ 0 and hence
xyz(x + y + z) ≥ 0. If xyz(x + y + z) = 0, then the second relation implies that
xy = yz = zx = 0. Hence, two of x, y, z are zero. If x = y = 0, then z 2 = 1/3
giving z = ±1/3. We get six solutions:
√ √ √
x, y, z = 0, 0, ±1/ 3 , 0, ±1/ 3, 0 , ± 1/ 3, 0, 0 .
This gives
Example 4.18. Let P (x) = an xn +an−1 xn−1 +· · ·+a1 x+a0 be a real polynomial
of degree n. Suppose n is even and
(i) a0 > 0, an > 0;
4
(ii) a21 + a22 + · · · + a2n−1 ≤ min a20 , a2n .
n−1
Prove that P (x) ≥ 0 for all real values of x.
2
n−1
n−1 x +1 n
≥4 x2j .
j=1
188 Inequalities
Note that after 4(m − 1) terms in the ﬁrst vector, the remaining terms are not
more than m. Hence, it is suﬃcient to prove that
4 2m − 1 + 2m − 2 + · · · + 2m − (m − 1) < 2m(2m − 1) + m(2m − 3).
But the left-side is 6m2 − 12m and the right-side is 6m2 − 5m and hence the
inequality holds. This completes the proof.
Example 4.19. (Nordic Contest-1992) Find all real numbers x > 1, y > 1, z > 1
such that
3 3 3 √ √
x+y+z+ + + =2 x+2+ y+2+ z+2 .
x−1 y−1 z−1
t+2 √
t−1+ ≥ 2 t + 2.
t−1
3 3 3
x+y+z+ + +
x−1 y−1 z−1
x+2 y+2 z+2
= x−1+y−1+z−1+ + +
x−1 y−1 z−1
√ √
≥ 2 x+2+ y+2+ z+2
3 3 3
= x+y+z+ + +
x−1 y−1 z−1
Applications 189
Example 4.20. (Bulgaria) Find the largest real number a such that
x y z
+√ + ≥ a,
y2 + z2 z2 +x 2 x + y2
2
We show that a = 2 is the largest possible. In other words, for each > 0,
there are positive x, y, z such that
x
< 2 + .
cyclic
y + z2
2
√
For this choose x = y = 1 and z = 2. Then
x 2
=√ + < 2 + .
cyclic
y2 + z2 1 + 22
Example 4.21. (Romania, 1993) Prove that if x, y, z are positive integers such
that x2 + y 2 + z 2 = 1993, then x + y + z cannot be the square of an integer.
190 Inequalities
x + y + z ≤ 12 < 49.
8 2
and a + 2b2 is maximum.
3
Applications 191
Thus,
4a2 + 3b2 = 2(a + b)2 + 2(a − b)2 − b2 ≤ 16 − b2 ≤ 16.
Here equality holds only if b = 0. Thus, 4a2 + 3b2 is maximum only if b = 0.
In this case
|a + b| = |a − b| = |a| = 2.
Thus,
|P (1) − P (0)| = |a + c − c| = |a| = 2.
Hence,
2 = |P (1) − P (0)| ≤ |P (1)| + |P (0)| ≤ 2.
We conclude that |P (1)| = |P (0)| = 1 and hence |c| = 1, |a + c| = 1. It follows
that either c = 1, a = −2, b = 0 or c = −1, a = 2, b = 0.
Now the required maximum is
8 2 8 32
a + 2b2 = × 4 = .
3 3 3
Example 4.23. ([7]) What is the maximum value of a2 + b2 , given that the
equation
x4 + ax3 + bx2 + ax + 1 = 0
has real roots.
y 2 + ay + b − 2 = 0.
This gives
(2 − y 2 )2
a 2 + b2 ≥ .
1 + y2
We also note that y = x + 1/x ≤ −2 or y ≥ 2. Hence, y 2 ≥ 4. Consider the
function
(2 − t)2
f (t) = ,
1+t
for t ≥ 4. This is a monotonically increasing function for t ≥ 4, as may be
veriﬁed using its derivatives. Its minimum is attained at t = 4;
4
a2 + b2 ≥ f (4) = .
5
If we take y 2 = 4, we have x + 1/x = 2 and hence x = 1 is a real solution. This
corresponds to 2a + b = −2 and we can take
4 2
a=− , b=− .
5 5
Then a2 + b2 = 4/5. Thus, the minimum value of a2 + b2 is 4/5.
Prove that one of these numbers is the sum of the remaining two.
Solution: We get from the ﬁrst relation, a2 ≤ (b − c)2 , or which is the same
as
(b − c − a)(b − c + a) ≥ 0.
Similarly
Example 4.25. Let ABC be a triangle and P be an interior point of ABC. Let
the lines AP , BP , CP meet BC, CA, AB respectively in D, E, F . Find all
positions of the point P for which the area of triangle DEF is maximal.
Applications 193
If we also introduce α = x/a, β = y/b and γ = z/c, then it takes the form
Now observe that [BDF ] = γ(1 − α)[ABC] and similar expressions for other
corner triangles. Thus,
Hence, we may assume that this never occurs in any subsequent steps.
176
(ii) Suppose b + c − bc > 0. Then again the AM-GM inequality implies
27
that
2
a+d 176
F a, b, c, d ≤ bc(a + d) + b+c− bc
2 27
a+d a+d
= F , b, c, .
2 2
Now we iterate this under the assumption that we are in the second case
in each stage of the iteration and we also exploit the symmetry of F :
a+d a+d
F a, b, c, d ≤ F , b, c,
2 2
a+d a+d
= F b, , ,c
2 2
b+c a+d a+d b+c
≤ F , , ,
2 2 2 2
a+d b+c a+d b+c
= F , , ,
2 2 2 2
1 b+c a+d 1
≤ F , , ,
4 2 2 4
b+c 1 1 a+d
= F , , ,
2 4 4 2
1 1 1 1 1
≤ F , , , =
4 4 4 4 27
n
Example 4.27. Determine the minimum value of j=1 aj where the minimum
is taken over all sequences a1 ,a2 ,a3 ,. . .,an , (n ≥ 2), with non-negative terms
n
such that aj aj+1 = 1, where an+1 = a1 .
j=1
Solution: Let
n
n
Mn = min aj aj ≥ 0, for 1 ≤ j ≤ n, and aj aj+1 = 1 .
j=1 j=1
√ √
that M2 = 2, M3 = 3 and Mn = 2 for n ≥ 4. Put α =
We show aj and
β= aj aj+1 . If n = 2, then α = a1 + a2 , β = 2a1 a2 = 1, so that
α2 − 2 = α2 − 2β = (a1 − a2 )2 ≥ 0,
√ √
with equality if and only if a1 = a2 = 1/ 2. Thus, M2 = 2. If n = 3, then
α2 − 3 = α2 − 3β
= (a1 + a2 + a3 )2 − 3(a1 a2 + a2 a3 + a3 a1 )
= a21 + a22 + a23 − a1 a2 − a2 a3 − a3 a1
10 1
= (a1 − a2 )2 + (a2 − a3 )2 + (a3 − a1 )2
2
≥ 0.
√ √
Here again, equality holds if and only if a1 = a2 = a3 = 1/ 3. Thus, M3 = 3.
Consider the case n = 4. In this case,
α2 − 4 = α2 − 4β
0 12
= (a1 + a3 ) + (a2 + a4 ) − 4(a1 + a3 )(a2 + a4 )
0 12
= (a1 + a3 ) − (a2 + a4 ) ≥ 0.
We show that equality holds in (4.5) if and only if there is a k such that
1 ≤ k ≤ n, ak = 1, ak−1 + ak+1 = 1 and aj = 0 for all other j. (Here a0 = an ,
an+1 = a1 .) Clearly this is suﬃcient. Conversely, suppose equality holds in
(4.5). Then
2 2
a1 + a2 + · · · + an − a1 − a2 + a3 − · · · + (−1)n+1 an
= 4 a1 a2 + a2 + a3 + · · · + an−1 an + an a1 .
a1 + a2 + · · · + an = 2, and a1 − a2 + a3 − · · · ± an = 0.
a1 + a3 + · · · = 1,
a2 + a4 + · · · = 1.
Example 4.28. (Taiwan, 2002) Determine all positive integers n and integers
a1 ,a2 ,a3 ,. . .,an such that aj ≥ 0, for 1 ≤ j ≤ n, and
n
n 2
4
a2j = 1 + aj .
j=1
4n + 1 j=1
This implies that A2 ≤ 4n2 + n. Again the nature of A shows that A ≤ 2n.
Thus, n + 1 ≤ A ≤ 2n. We hence obtain
1 A
1<1+ ≤ ≤ 2.
n n
But we also have
n
2
A A2
aj − =1− < 1,
j=1
n n(4n + 1)
which gives
A
−1 < aj − < 1,
n
for 1 ≤ j ≤ n. Thus,
A A
0 < −1 + < aj < 1 + < 3,
n n
for 1 ≤ j ≤ n. This shows that aj = 1 or 2 for all j. Let b denote number of
those j’s for which aj = 2. Then there are n − b values of j for which aj = 1.
In this case (4.6) takes the form
4
4b + n − b = 1 + (2b + n − b)2 .
4n + 1
198 Inequalities
1
This simpliﬁes to n = b + . Since n and b are integers the only possibility
4b − 3
is b = 1 and n = 2.
We conclude that n = 2 or n = 6. In these cases, aj ’s can be computed
using the bounds we have derived:
n = 2: (a1 , a2 ) = (1, 2), (2, 1);
n = 6: the 6-tuples are (0, 1, 1, 1, 1, 1) and permutations of this.
whee [x] denotes the integer part of a real number x. Determine the range of f .
Applications 199
Solution: We show that the range of f is precisely the set of all those
natural numbers which are not k-th power of some natural number. Suppose
a ∈ N is not a k-th power. Then there is a natural number m such that
mk < a < (m + 1)k . Using the binomial theorem, it is easy to derive that
(x + 1)k ≥ xk + x + 1,
for all x ≥ 0. Taking x = m − 1, it follows that (m − 1)k ≤ mk − m. Taking
n = a − m, we also get mk − m < a − m = n < a. Thus,
mk ≤ a − 1 = (a − m) + (m − 1)
< n + n1/k
< (a − m) + (m + 1)
= a+1
≤ (m + 1)k .
The deﬁnition of f shows that f (n) = m + n = a, i.e., f (a − m) = a. This
shows that if a ∈ N is not a k-th power, then a is in the range of f . Observe
that f is strictly increasing. Since f (1) = 2, it follows that 1 is not in the range
of f . Now consider any k-th power, say, mk . We have proved that if a ∈ N is
not a k-th power, then - .
a = f a − a1/k .
Thus, we obtain
f mk − m = f mk − 1 − (m − 1) = mk − 1,
and
f mk + 1 − m = mk + 1.
Example 4.31. (IMO, 1992) Find all integers a, b, c with 1 < a < b < c such
that (a − 1)(b − 1)(c − 1) is a divisor of abc − 1.
Thus, p = 1 or p = 2.
Case 1. p = 1.
Here we have
1 1 1 1 1 1
+ + + + + = 1.
x y z xy yz zx
Obviously x ≥ 2. If x ≥ 3 then y ≥ 4 and z ≥ 5 and we get an estimate
1 1 1 1 1 1 59
p≤ + + + + + = <1
3 4 5 12 15 20 60
which is not tenable. We conclude that x = 2 and hence
1 1 1 1 1 1
+ + + + + = 1.
2 y z 2y yz 2z
This simpliﬁes to (y − 3)(z − 3) = 11. Keeping in mind that y < z, we see that
the only solution is (y, z) = (4, 14). This leads to the triple (a, b, c) = (3, 5, 15).
Case 2. p = 2.
Here again, we see that x ≥ 2 forces p < 2, and hence x = 1. Using this
information in the expression for p, we obtain (y − 2)(z − 2) = 5. Since y < z,
the only solution is (y, z) = (3, 7) and this gives (a, b, c) = (2, 4, 8).
We obtain two solutions to the problem: (a, b, c) = (3, 5, 15), (2, 4, 8).
Example 4.32. Find all integer sided triangles with the property that the area of
each triangle is numerically equal to its perimeter.
We can rule out isosceles triangles. If, for example, a = b, we get x = y and
the equation reduces to
1 2 1
+ = .
x2 xz 4
This can be written as a quadratic equation in x: zx2 − 8x − 4z = 0. The
discriminant of this equation is D = 16(4 + z 2 ). The equation has an integer
Applications 201
root only if D is the square of an integer. This can happen only if z = 0 which
is impossible.
Thus, we may assume that 1 ≤ x < y < z. If x ≥ 3, we have y ≥ 4 and
z ≥ 5. Going back to the equation (4.7) with this information, we see that
1 1 1 1 1 1 1 1
= + + ≤ + + = .
4 xy yz zx 12 20 15 5
This absurd implication shows that x ≤ 2. Thus, x = 1 or x = 2.
Case 1. x = 1.
Here the equation (4.7) takes the form
1 1 1 1
+ + = .
y yz z 4
This can be written as (y − 4)(z − 4) = 20. In this case, we get (y, z) = (5, 24),
(6, 14), (8, 9). We obtain the triangles (with c < b < a) (6, 25, 29), (7, 15, 20),
and (9, 10, 17).
Case 2. x = 2
By a similar analysis, the only triangles when x = 2 are (5, 12, 13) and (6, 8, 10).
because uv < −1. This leads to the absurd conclusion that 1 ≤ −1.
We conclude that uv = 0. If u = 0, we get v = −1. If v = 0, we get u = 1.
Thus, the only solutions are (u, v) = (1, 0), (0, −1).
Example 4.34. ( Romania, 2003) Two unit squares with parallel sides overlap in
a rectangular region of area 1/8 square units. Find the minimum and maximum
possible distances between the centres of these two squares.
XY 2 = XZ 2 + Y Z 2 = (1 − a)2 + (1 − b)2
= a2 + b2 − 2(a + b) + 2
1
= a2 + 2ab + b2 − 2(a + b) − +2
4
7
= (a + b)2 − 2(a + b) +
4
3
= (a + b − 1)2 +
4
3
≥ .
4
√ √
Thus, XY ≥ 3/2 and hence the minimal distance is 3/2. This is achieved
Applications 203
2. (Czech and Slovak, 1999) If a, b, c are positive real numbers, prove that
a b c
+ + ≥ 1.
b + 2c c + 2a a + 2b
4. (Balkan Olympiads, 2002) For any positive real numbers a, b, c, prove that
2 2 2 27
+ + ≥ .
b(a + b) c(b + c) a(c + a) (a + b + c)2
11. (Austria, 2001) Find all triplets (a, b, c) of positive real numbers which
satisfy the system of equations:
a+b+c = 6,
1 1 1 4
+ + = 2− .
a b c abc
12. (Bosnia and Herzegovina, 2002) Let a, b, c be real numbers such that
a2 + b2 + c2 = 1. Prove that
a2 b2 c2 3
+ + ≥ .
1 + 2bc 1 + 2ca 1 + 2ab 5
13. (Ukraine, 2001) Let a, b, c and α, β, γ be positive real numbers such that
α + β + γ = 1. Prove that
αa + βb + γc + 2 αβ + βγ + γα ab + bc + ca ≤ a + b + c.
14. (Estonia, 1996-97) Prove that for all real numbers a, b, the inequality
a 2 + b2 + 1 > a b2 + 1 + b a 2 + 1
holds.
15. (Poland, 1994-95) For a ﬁxed positive integer n, compute the minimum
value of the sum
x2 x3 xn
x1 + 2 + 3 + · · · + n ,
2 3 n
where x1 , x2 , x3 , . . . , xn are positive real numbers such that
1 1 1 1
+ + + ··· + = n.
x1 x2 x3 xn
17. Let a, b, c be positive real numbers, all less than 1, such that a + b + c = 2.
Prove that
abc ≥ 8(1 − a)(1 − b)(1 − c).
18. (USAMO, 2003) Let a, b, c be three positive real numbers. Prove that
19. Three positive real numbers a, b, c are such that (1 + a)(1 + b)(1 + c) = 8.
Prove that abc ≤ 1.
n
aj n
≥ .
j=1
2 − aj 2n − 1
(Here an+1 = a1 .)
n
a2j 1
≥ .
j=1
1 − 2aj n−2
28. (Czech and Slovak, 2003) Let P (x) = ax2 +bx+c be a quadratic polynomial
with non-negative coeﬃcients and let α be a positive real number. Prove that
32. (Thailand, 2006) Suppose a, b, c are positive real numbers. Prove that
1/3 1/3
a 3 + b3 + c 3
3 a + b + c ≥ 8 abc + .
3
When does equality hold?
33. Let c1 ,c2 ,c3 ,. . .,cn be n real numbers such that either 0 ≤ cj ≤ 1 for all j
or cj ≥ 1 for all j, 1 ≤ j ≤ n. Prove that the inequality
"
n
"
n
1 − p + pcj ≤ 1 − p + p cj
j=1 j=1
34. (Taiwan, 2002) Let x1 , x2 , x3 , x4 be real numbers in the interval (0, 1/2].
Prove that
x1 x2 x3 x4
(1 − x1 )(1 − x2 )(1 − x3 )(1 − x4 )
x41 + x42 + x43 + x44
≤ .
(1 − x1 )4 + (1 − x2 )4 + (1 − x3 )4 + (1 − x4 )4
35. Let x1 ,x2 ,x3 ,. . .,xn be n real numbers such that 0 < xj ≤ 1/2. Prove that
⎛ ⎞ ⎛ ⎞
" / n " /
n n n
n
n
⎝ xj xj ⎠ ≤ ⎝ 1 − xj 1 − xj ⎠.
j=1 j=1 j=1 j=1
36. (CRUX) Consider a sequence an of real numbers satisfying aj+k ≤ aj +ak .
Prove that
a2 a3 an
a1 + + + ··· + ≥ an ,
2 3 n
for all n.
43. Let a, b, c be positive real numbers and let x be a non-negative real number.
Prove that
ax+2 + bx+2 + cx+2 ≥ ax bc + abx c + abcx .
n
n 1/3
n 1/3
n 1/3
a j bj c j ≤ a3j b3j c3j .
j=1 j=1 j=1 j=1
45. ([9]) Prove for any three real numbers a, b, c, the inequality
2
3 a2 − a − 1 b2 − b − 1 c2 − c + 1 ≥ abc − abc + 1.
48. Let a1 <a2 <a3 <. . .<an be n positive integers. Prove that
2
a1 +a2 +a3 +· · ·+an ≤ a31 + a32 + a33 + · · · + a3n .
49. (Poland, 1996) Consider a sequence a1 ,a2 ,a3 ,. . .,an of positive real numbers
which add up to 1, where n ≥ 2 is an integer. Prove that for any positive real
n
numbers x1 ,x2 ,x3 ,. . .,xn with j xj = 1, the inequality
n − 2 aj x2j
n
2 xj xk ≤ + ,
n − 1 j=1 1 − aj
j<k
holds.
Problems 211
50. (Iran, 1998) Let x1 , x2 , x3 , x4 be four positive real numbers such that
x1 x2 x3 x4 = 1. Prove that
1 1 1 1
x31 + x32 + x33 + x34 ≥ min x1 + x2 + x3 + x4 , + + + .
x1 x2 x3 x4
51. (Romania) Let {x} denote the fractional part of x; i.e., {x} = x − [x].
Prove for any positive integer n,
n
n2 − 1
j ≤ .
j=1
2
a2 b2 c2 3
+ + ≥ .
(a + b)(a + c) (b + c)(b + a) (c + a)(c + b) 4
a2 (b + c − a) + b2 (c + a − b) + c2 (a + b − c) ≤ 3abc.
54. Let x, y, z be positive real numbers such that xyz ≥ xy + yz + zx. Prove
that
xyz ≥ 3(x + y + z).
55. Let b1 ,b2 ,b3 ,. . .,bn be n non-negative real numbers and let b denote the
sum of these numbers. Prove that
n−1
b2
bj bj+1 ≤ .
j=1
4
57. Let x, y, z be three real numbers in the interval [0, 1] such that xyz =
(1−x)(1−y)(1−z). Find the least possible value of x(1−z)+y(1−x)+z(1−y).
58. Let x1 ,x2 ,x3 ,. . .,xn be non-negative real numbers such that
n
1
≤ 1.
j=1
1 + xj
59. (Bulgaria, 1998) For positive real numbers a, b, c, prove the inequality
√ √
3
3 a + ab + abc ≤ 4 a + b + c .
60. ([1]) Show that for any two natural numbers m, n, the inequality
1 1 4
− ≤
m + n + 1 (m + 1)(n + 1) 45
holds.
ab + ba > 1.
62. Let a, b be positive real numbers such that a + b = 1 and let p be a positive
real. Prove that p p
1 1 5p
a+ + b+ ≥ p−1 .
a b 2
63. (IMO, 2000) Let a, b, c be positive real numbers such that abc = 1. Prove
that
1 1 1
a−1+ b−1+ c−1+ ≤ 1.
b c a
64. (Proposed for IMO-2001) Let x, y, z be real numbers in the interval [−1, 2]
such that x + y + z = 0. Prove that
2 2 2
(2 − x)(2 − y) (2 − y)(2 − z) (2 − z)(2 − x)
+ + ≥ 3.
(2 + x)(2 + y) (2 + y)(2 + z) (2 + z)(2 + x)
65. (IMO, 1978) Let an be a sequence of distinct positive integers. Prove
that
n
ak
n
1
2
≥ ,
k k
k=1 k=1
for every positive integer n.
66. (IMO, 1984) Let x, y, z be non-negative real numbers such that x+y+z = 1.
Prove that
7
0 ≤ xy + yz + zx − 2xyz ≤ .
27
67. Let x1 , x2 , . . . , xn be n positive real numbers. Prove that
n
x3j 1
n
≥ xj ,
j=1
x2j + xj xj+1 + x2j+1 3 j=1
Problems 213
where xn+1 = x1 .
69. (Indian Team Selection, 2002) Prove that for any positive reals a, b, c, the
inequality,
a b c c+a a+b b+c
+ + ≥ + +
b c a c+b a+c b+a
holds.
76. Let α, β, x1 ,x2 ,x3 ,. . .,xn be positive reals such that α + β = 1, and
x1 +x2 +x3 +· · ·+xn = 1. Prove that
n
x2m+1
j 1
≥ 2m−1 ,
j=1
αxj + βxj+1 n
214 Problems
79. Let ABC be an acute-angled triangle with altitudes AD, BE, CF and
ortho-centre H. Prove that
HD HE HF 3
+ + ≥ .
HA HB HC 2
80. (IMO, 1981) For any point P inside a triangle ABC, let r1 , r2 , r3 denote
the distances of P from the lines BC, CA, AB respectively. Find all points P
for which a/r1 + b/r2 + c/r3 is minimal.
81. (IMO, 1996) Let ABCDEF be a convex hexagon in which AB, BC, CD
are respectively parallel to DE, EF , F A. Let RA , RB , RC be the circum-radii
of the triangles F AB, BCD, DEF respectively, and let p denote the perimeter
of the hexagon. Prove that
p
RA + R B + RC ≥ .
2
82. (AMM) Let ha , mb , wc denote respectively the altitude from A to BC, the
median from B to CA and the internal angle bisector of angle C. Prove that
√
3
h a + mb + w c ≤ (a + b + c).
2
83. (Bulgaria, 1997) Let ABC be a triangle with centroid G. Prove that
2
sin ∠ABG + sin ∠ACG ≤ √ .
3
85. Let Ω be a Brocard point of a triangle ABC. Let AΩ, BΩ,CΩ extended
meet the circum-circle of ABC in K, L, M respectively. Prove that
AΩ BΩ CΩ
+ + ≥ 3.
ΩK ΩL ΩM
88. (Germany, 1995) Let ABC be a triangle, and let D, E be points on BC,
CA such that the in-centre of ABC lies on DE. Prove that [ABC] ≥ 2r2 .
90. (South Korea, 1995) Let the internal bisectors of the angles A, B, C of a
triangle ABC meet the sides BC, CA, AB in D, E, F and the circum-circle
in L, M , N respectively. Prove that
AD BE CF
+ + ≥ 9.
DL EM FN
93. (St. Petersburg Olympiad, 1996) Let M be the point of intersection of two
diagonals of a cyclic quadrilateral. Let N be the point of intersection of two
lines joining the midpoints of opposite pair of sides. If O is the centre of the
circumscribing circle, prove that OM ≥ ON .
94. (APMO, 1997) Let ABC be a triangle with internal angle bisectors AD,
BE, CF . Suppose AD, BE, CF when extended meet the circum-circle again
in K, L, M respectively. If la = AD/AK, lb = BE/BL and lc = CF/CM ,
prove that
la lb lc
2 + 2 + ≥ 3.
sin A sin B sin2 C
102. (Estonia, 1996) Let a, b, c be the sides of a triangle ABC with in-radius
r. Prove that
a sin A + b sin B + c sin C ≥ 9r.
104. (US Team Selection, 2000) Let ABC be a triangle having the circum-
radius R. Let P be an interior point of ABC. Prove that
AP BP CP 1
+ + ≥ .
BC 2 CA2 AB 2 R
105. With every natural number n, associate a real number an by
1 1 1
an = + + ··· + ,
p1 p2 pk
where p1 , p2 , . . . , pk is the set of all prime divisors of n. Show that for any
natural number N ≥ 2,
N
a2 a3 · · · an < 1.
n=2
108. ([9]) Let x1 ,x2 ,x3 ,. . .,x100 be 100 positive integers such that
1 1 1
√ + √ + ··· + √ = 20.
x1 x2 x100
109. Let f (x) be a polynomial with integer coeﬃcients and of degree n > 1.
Suppose f (x) = 0 has n real roots in the interval (0, 1), not all equal. If a is
the leading coeﬃcient of f (x), prove that
|a| ≥ 2n + 1.
4z 2 4x2 4x2
x= , y= , z= ,
1 + 4z 2 1 + 4x2 1 + 4x2
for real numbers x, y, z.
112. Suppose a, b are nonzero real numbers and that all the roots of the real
polynomial
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.
a 4 + b4 + c 4
a cos(β − γ) + b cos(γ − α) + c cos(α − β) = ,
abc
where a, b, c are the sides and α, β, γ are the angles opposite to the sides a, b, c
respectively.
1 1 1 1
x1 + = 4, x2 + = 1, . . . , x1999 + = 4, x2000 + = 1.
x2 x3 x2000 x1
x+y+z = 1,
x3 + y 3 + z 3 + xyz = x4 + y 4 + z 4 + 1.
Problems 219
117. (Bulgaria, 2001) Let a, b be positive integers such that each equation
(a + b − x)2 = a − b, (ab + 1 − x)2 = ab − 1
has two distinct real roots. Suppose the bigger of these roots are the same.
Show that the smaller roots are also the same.
119. (IMO, 1972) Find all the solutions of the following system of inequalities:
2
(i) x1 − x3 x5 x22 − x3 x5 ≤ 0,
2 2
(ii) x2 − x4 x1 x3 − x4 x1 ≤ 0,
2 2
(iii) x3 − x5 x2 x4 − x5 x2 ≤ 0,
2 2
(iv) x4 − x1 x3 x5 − x1 x3 ≤ 0,
2 2
(v) x5 − x2 x4 x1 − x2 x4 ≤ 0.
121. (Indian
n Team Selection,1999)
n Let a1 ,a2 ,a3 ,. . .,an be n positive integers
such that j=1 aj = j=1 aj . Let Vn denote this common value. Show that
Vn ≥ n + s, where s is the least positive integer such that 2s − s ≥ n.
n
122. Let z1 ,z2 ,z3 ,. . .,zn be n complex numbers such that j=1 |zj | = 1. Prove
that there exists a subset S of the set z1 ,z2 ,z3 ,. . .,zn such that
1
z ≥ .
4
z∈S
220 Problems
123. ([1]) Let a1 ,a2 ,a3 ,. . .,an and b1 ,b2 ,b3 ,. . .,bn be two sequences of real
numbers which are not proportional. Let x1 ,x2 ,x3 ,. . .,xn be another sequence
of real numbers such that
n
n
aj xj = 0, bj xj = 1.
j=1 j=1
Prove that
n
n
a2j
x2j
j=1
≥ 2 .
a2j 2 −
n n n
j=1
j=1 b
j=1 j a b
j=1 j j
Prove that
2 2
a1 − a22 − · · · − a2n b21 − b22 − · · · − b2n ≤ a1 b1 − a2 b2 − · · · − an bn ,
and show that equality holds if and only if aj = λbj , 1 ≤ j ≤ n, for some real
constant λ.
125. ([1]) Let x1 ,x2 ,x3 ,. . .,xn be n positive real numbers. Prove that
n
xj
≤ n,
j=1
2xj + xj+1 + · · · + xj+n−2
where xn+k = xk .
126. ([1]) Let x1 ,x2 ,x3 ,. . .,xn be n ≥ 2 positive real numbers and k be a ﬁxed
integer such that 1 ≤ k ≤ n. Show that
x1 + 2x2 + · · · + 2xk−1 + xk 2n(k − 1)
≥ .
xk + xk+1 + · · · + xn n−k+1
cyclic
127. ([1]) Let z and ξ be two complex numbers such that |z| ≤ r, |ξ| ≤ r and
z = ξ. Show that for any natural number n, the inequality
n
z − ξn 1
n−2
z − ξ
z − ξ ≤ 2 n(n − 1)r
holds.
Problems 221
128. For any three vectors, x = x1 ,x2 ,x3 ,. . .,xn , y = y1 ,y2 ,y3 ,. . .,yn , and
z = z1 ,z2 ,z3 ,. . .,zn in Rn , prove that
||x|| + ||y|| + ||z|| − ||x + y|| − ||y + z|| − ||z + x|| + ||x + y + z|| ≥ 0.
131. Suppose a1 ,a2 ,a3 ,. . .,an are n positive real numbers. For each k, deﬁne
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.
for all complex numbers z, where Re(z) denotes the real part of z. Prove that
|b|2 ≤ ac,
and
f (z) ≤ a + c 1 + |z|2 .
Show that |b|2 = ac only if f (z) = 0 for some z ∈ C.
133. (IMO, 2003) Suppose x1 ≤x2 ≤x3 ≤. . .≤xn be n real numbers. Show that
n 2 n n
n
2 2 n − 1 2
x j − x k ≤ xj − xk .
j=1
3 j=1
k=1 k=1
Prove also that equality holds if and only if the sequence xj is in arithmetic
progression.
222 Problems
Prove that c ≥ 1.
137. Let the diagonals of a convex quadrilateral ABCD meet in P . Prove that
[AP B] + [CP D] ≤ [ABCD],
141. Let x1 ,x2 ,x3 ,. . .,xn be n positive reals which add up to 1. Find the
minimum value of
n
x
j .
j=1
1 + k=j xk
a b c d
+ + + ,
a+b+d a+b+c b+c+d a+c+d
when a, b, c, d vary over positive reals.
Prove that
n
Fj
< 2,
j=1
2j
for all n ≥ 1.
prove that x2 y ≤ 1.
147. (IMO, 1991) Let I be the in-centre of a triangle ABC. Suppose the
internal bisectors of angles A, B, C meet the opposite sides at A , B and C .
Prove that
1 AI · BI · CI 8
<
≤ .
4 AA · BB · CC 27
224 Problems
over all n-tuples (x1 ,x2 ,x3 ,. . .,xn ) of reals such that xj ≥ 0 for 1 ≤ j ≤ n.
150. (Indian Team Selection, 1994) Let x1 ,x2 ,x3 ,. . .,xN be positive real num-
bers. Prove that
N
1/j N
x1 x2 · · · xj <3 xj .
j=1 j=1
151. (Indian TeamSelection, 2000) Let a1 ≤a2 ≤a3 ≤· · ·≤an be n real numbers
n
with the property j=1 aj = 0. Prove that
n
na1 an a2j ≤ 0.
j=1
152. (Indian Team Selection, 1997) Suppose a, b, c are positive real numbers.
Prove that
1 1 1 3
+ + ≥ .
a(1 + b) b(1 + c) c(1 + a) 1 + abc
154. (Wolschaum County Problem book) Let 0 ≤ x1 ≤x2 ≤x3 ≤· · ·≤xn be such
n
that j=1 xj = 1, where n ≥ 2 is an integer. If xn ≤ 2/3, prove that there
exists a k such that 1 ≤ k ≤ n and
1
k
2
≤ xj ≤ .
3 j=1 3
x + y + z ≥ xy + yz + zx.
Problems 225
157. Let x, y, z be real numbers and let p, q, r be real numbers in the interval
(0, 1/2) such that p + q + r = 1. Prove that
2
pqr x + y + z ≥ xyr 1 − 2r + yzp 1 − 2p + zxq 1 − 2q .
158. (Romania) Let x1 ,x2 ,x3 ,. . .,xn be n real numbers in the interval [0, 1].
Prove that
n n + ,
n
xj − xj xj+1 ≤ ,
j=1 j=1
2
where xn+1 = x1 .
159. (IMO, 2005) Suppose x, y, z are positive real numbers such that xyz ≥ 1.
Prove that the inequality
x5 − x2 y5 − y2 z5 − z2
+ 5 + 5 ≥0
x5 2
+y +z 2 y +z +x2 2 z + x2 + y 2
holds.
160. (Indian Team Selection, 1992) Consider two sequences of positive real
numbers, a1 ≤a2 ≤a3 ≤· · ·≤an and b1 ≤b2 ≤b3 ≤· · ·≤bn , such that
n
n
aj ≥ bj .
j=1 j=1
"
n "
n
aj ≥ bj .
j=1 j=1
161. (Bulgaria, 1997) Let a, b, c be positive real numbers such that abc = 1.
Prove that
1 1 1 1 1 1
+ + ≤ + + .
1+a+b 1+b+c 1+c+a 2+a 2+b 2+c
226 Problems
162. (PUTNAM) Let n ≥ 4 and let a1 ,a2 ,a3 ,. . .,an be real numbers such that
Prove that
max a1 ,a2 ,a3 ,. . .,an ≥ 2.
n+1 √ 1n2
xj+1 − xj
< .
j=1
xj+1 j=1
j
164. Let a, b, c be three positive real numbers which satisfy abc = 1 and
a3 > 36. Prove that
2 2
a < a2 + b2 + c2 − ab − bc − ca.
3
165. ([1]) Let z1 ,z2 ,z3 ,. . .,zn be n complex numbers and consider
n positive
real numbers λ1 ,λ2 ,λ3 ,. . .,λn which have the property that 1/λj = 1. Prove
that
2
n n
2
z ≤ λ j zj .
j
j=1 j=1
1/2
3 min a, b, c < a− a2 − ab
1/2
< a+ a2 − ab < 3 max a, b, c ,
168. Show that for all complex numbers z with real part of z > 1, the following
inequality holds:
x = a + b − c, y = b + c − a, z = c + a − b.
Prove that
abc(xy + yz + zx) ≥ xyz(ab + bc + ca).
172. (USAMO, 2001) Suppose a, b, c are non-negative real numbers such that
a2 + b2 + c2 + abc = 4. Prove that
0 ≤ ab + bc + ca − abc ≤ 2.
173. (CRUX, 2003) Suppose a, b, c are complex numbers such that |a| = |b| =
|c|. Prove that
ab bc ca √
+ +
a2 − b2 b2 − c2 c2 − a2 ≥ 3.
174. (CRUX, 2001) Suppose x, y, z are non-negative real numbers such that
x2 + y 2 + z 2 = 1. Prove that
√
x 3 3
(a) 1 ≤ ≤ ;
1 − yz 2
cyclic
x √
(b) 1 ≤ ≤ 2.
1 + yz
cyclic
176. (CRUX, 1990) Let a, b, c, d be four positive real numbers such that a +
b + c + d = 2. Prove that
a2 16
2 ≤ .
2 25
cyclic a + 1
179. (CRUX, 2002) Prove that in a triangle with angles α, β, γ, the inequality
15
sin α ≤ + cos(α − β)
4
cyclic
holds.
181. (CRUX, 2003) Let a, b, c be three positive real numbers. Prove that
ab a
≥ ,
c(c + a) c+a
where the sum is taken cyclically over a, b, c.
185. (Romania, 1997) Let a, b, c be positive real numbers such that abc = 1.
Show that
a 9 + b9
≥ 2,
a 6 + a 3 b3 + b 6
where the sum is cyclical.
186. Let a, b, c be the sides of a triangle and set x = 2(s − a), y = 2(s − b),
z = 2(s − c), where s denotes the semi-perimeter. Prove that
187. (AMM) Let a1 ,a2 ,a3 ,. . .,an (n > 2) be positive real numbers and let s
denote their sum. Let 0 < β ≤ 1 be a real number. Prove that
n
β n
β
s − ak 2β ak
≥ (n − 1) .
ak s − ak
k=1 k=1
188. A point D on the segment BC of a triangle ABC is such that the in-radii
of ABD and ACD are equal, say r1 . Similarly deﬁne r2 and r3 . Prove that
s(s − a)
(i) 2r1 + 2 r = ha .
a
(ii) 2 r1 + r2 + r3 + s ≥ ha + hb + hc .
189. (Romania) For n ≥ 4, let a1 ,a2 ,a3 ,. . .,an be n positive real numbers such
n
that a2j = 1. Show that
j=1
a1 a2 an 4 √ √ √ 2
+ + · · · + ≥ a 1 a 1 + a 2 a 2 + · · · + a n an .
a22 + 1 a23 + 1 a21 + 1 5
190. (Belarus) Does there exist an inﬁnite sequence xn of positive real num-
bers such that
√ √
xn+2 = xn+1 − xn ,
for all n ≥ 2.
191. (Belarus) Let a1 ,a2 ,a3 ,. . .,an be n positive real numbers and consider a
permutation b1 ,b2 ,b3 ,. . .,bn of it. Prove that
n
a2j
n
≥ aj .
j=1
bj j=1
230 Problems
n
a2j 1
≥ .
j=1
a j + bj 2
y 2 − x2 z2 − y2 x2 − z 2
+ + ≥ 0.
z+x x+y y+z
194. (Japan, 2003) Find the greatest real value of k such that for every triple
(a, b, c) of positive real numbers, the inequality
2
a2 − bc > k b2 − ca c2 − ab
holds.
195. (Romania, 2003) Let a, b, c, d be positive real numbers such that abcd = 1.
Prove that
1 + ab 1 + bc 1 + cd 1 + da
+ + + ≥ 4.
1+a 1+b 1+c 1+d
198. (Balkan Olympiads, 2005) Let a, b, c be positive real numbers such that
(a + b)(b + c)(c + a) = 1. Prove that
3
ab + bc + ca ≤ .
4
199. (Iran, 2005) Let ABC be a right-angled triangle with A = 90◦ . Let AD
be the bisector of angle A, and Ia be the ex-centre opposite to A. Prove that
AD √
≤ 2 − 1.
DIa
Problems 231
201. Let ABC be a triangle with circum-circle Γ, and G be its centroid. Extend
AG, BG, CG to meet Γ in D, E, F respectively. Prove that
AG + BG + CG ≤ GD + GE + GF.
202. Prove with usual notation that in a triangle ABC, the inequality
a + b + c ha + hb + hc ≥ 18Δ.
203. (Short-list, IMO-2004) Let a, b, c be three positive real numbers such that
ab + bc + ca = 1. Prove that
1/3 1/3 1/3
1 1 1 1
+ 6b + + 6c + + 6a ≤ .
a b c abc
204. (IMO, 1974) Let ABC be a triangle. Show√that there exists a point D
on AB such that CD2 = AD · BD if and only if sin A sin B ≤ sin(C/2).
205. (Short-list, IMO-2004) Let a1 ,a2 ,a3 ,. . .,an be n > 1 positive real numbers.
For each k, 1 ≤ k ≤ n, let Ak = (a1 +a2 +· · ·+ak )/k. Let gn = (a1 a2 · · · an )1/n
and Gn = (A1 A2 · · · An )1/n . Prove that
1/n
Gn gn
n + ≤ n + 1.
An Gn
206. Let x, y, z be real numbers in the interval [0, 1]. Prove that
3 x2 y 2 + y 2 z 2 + z 2 x2 − 2xyz x + y + z ≤ 3.
207. (UK, 1999) Let x, y, z be non-negative real numbers such that x+y+z = 1.
Prove that
7(xy + yz + zx) ≤ 2 + 9xyz.
208. (South Africa, 2003-04) Let x, y, z be real numbers in the interval [0, 1].
Prove that
x y z
+ + ≤ 2.
yz + 1 zx + 1 xy + 1
232 Problems
210. Let x, y, z be positive real numbers such that x + y + z = xyz. Prove that
1 1 1 3
√ + +√ ≤ .
1+x 2 1+y 2 1+z 2 2
211. Let P be an interior point of a triangle ABC whose sides are a, b, c. Let
R1 = P A, R2 = P B, and R3 = P C. Prove that
R1 R2 + R 2 R 3 + R3 R 1 R1 + R 2 + R3 ≥ a 2 R1 + b2 R2 + c 2 R 3 .
When does equality hold?
212. (IMO, 2004) Let t1 ,t2 ,t3 ,. . .,tn be positive real numbers such that
1 1 1
n 2 + 1 > t 1 + t2 + · · · + t n + + ··· + ,
t1 t2 tn
where n ≥ 3 is an integer. Show that for each triple (j, k, l) with 1 ≤ j < k <
l ≤ n, there is a triangle with sides tj , tk , tl .
214. (Romania, 2004) Find all positive real numbers such that
4 ab + bc + ca − 1 ≥ a2 + b2 + c2 ≥ 3 a3 + b3 + c3 .
215. Let a, b, c be positive real numbers such that abc = 1. Prove that
a b c 3
+ + ≥ .
(a + 1)(b + 1) (b + 1)(c + 1) (c + 1)(a + 1) 4
216. (CRUX, 2000) Suppose a, b, c are positive real numbers such that a2 +
b2 + c2 = 1. Prove that
1 1 1 2 a 3 + b3 + c 3
+ 2 + 2 ≥3+ .
a2 b c abc
217. (Taiwan, 1999) Let ABC be a triangle with circum-centre O and circum-
radius R. Suppose the line AO, when extended, meets the circum-circle of
OBC in D; similarly deﬁne E and F . Prove that
OD · OE · OF ≥ 8R3 .
Problems 233
218. (Proposed for IMO-1998) Let x, y, z be positive real numbers such that
xyz = 1. Prove that
x3 y3 z3 3
+ + ≥ .
(1 + y)(1 + z) (1 + z)(1 + x) (1 + x)(1 + y) 4
a3 b3 c3 d3 1
+ + + ≥ .
b+c+d c+d+a d+a+b a+b+c 3
221. (Bulgaria, 1974) Find all real λ for which the inequality
222. Let a, b, c be positive real numbers such that abc = 1. Prove that
a b c 1 1 1
+ + ≥ + + .
b c a a b c
223. (Indian Team Selection, 2007) Let a, b, c be non-negative reals such that
a + b ≤ 1 + c, b + c ≤ 1 + a, c + a ≤ 1 + b.
Prove that
a2 + b2 + c2 ≤ 2abc + 1.
225. (Indian Team Selection, 2006) Let ABC be a triangle with sides a, b, c,
circum-radius R and in-radius r. Prove that
2
R 64a2 b2 c2
≥ 2 .
2r 4a − (b − c)2 4b2 − (c − a)2 4c2 − (a − b)2
234 Problems
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?
229. (Ukraine, 2001) In a triangle ABC, let AA1 , CC1 be the bisectors of the
angles A, C respectively. Let M be point on the segment AC, and I be the
in-centre of ABC. Draw a line through M , parallel to AA1 and let it meet
CC1 in N and BC in Q. Similarly, let the line through M parallel to CC1
meet AA1 in H and AB in P . Let d1 , d1 , d3 respectively denote the distances
of H, I, N from the line P Q. Prove that
d1 d2 d3 2ab 2bc 2ca
+ + ≥ 2 + 2 + 2 .
d2 d3 d1 a + bc b + ca c + ab
230. (CRMO, 1996) Let ABC be a triangle and ha be the altitude through
A. Prove that
(b + c)2 ≥ a2 + 4h2a .
(As usual a, b, c denote the sides BC, CA, AB respectively.)
231. (CRMO, 2003) Let a, b, c be three positive real numbers such that
a + b + c = 1. Prove that among the three numbers a − ab, b − bc, c − ca there
is one which is at most 1/4 and there is one which is at least 2/9.
232. (CRMO, 2004) Let x and y be positive real numbers such that y 3 + y ≤
x − x3 . Prove that
(a) y < x < 1; and
(b) x2 + y 2 < 1.
233. (CRMO, 2005) Let a, b, c be three positive real numbers such that a +
b + c = 1. Let
λ = min a3 + a2 bc, b3 + ab2 c, c3 + abc2 .
Problems 235
234. (CRMO, 2006) If a, b, c are three positive real numbers, prove that
a 2 + 1 b2 + 1 c 2 + 1
+ + ≥ 3.
b+c c+a a+b
237. (USAMO, 1974) Suppose a, b, c are positive real numbers. Prove that
aa bb cc ≥ (abc)(a+b+c)/3 .
238. (Bulgaria, 1988) Find all real p and q for which the equation
8p2 3
x4 − x + 4qx3 − 3px + p2 = 0
q
has four positive roots.
239. (Russia, 2005) Let a1 , a2 , a3 be real numbers, each greater than 1. Let
S = a1 + a2 + a3 and suppose S < a2j /(aj − 1) for j = 1, 2, 3. Prove that
1 1 1
+ + > 1.
a1 + a2 a2 + a3 a3 + a1
241. (Balkan Olympiads, 2010) Suppose a, b, c are positive real numbers. Prove
that
a2 b(b − c) b2 c(c − a) c2 a(a − b)
+ + ≥ 0.
a+b b+c c+a
243. (Thailand, 2014) Determine the largest value of k such that the inequality
a b c a b c b c a
k+ k+ k+ ≥ + + + +
b c ba b c a a b c
246. (Russia, 2014) Let a, b, c be three positive real numbers such that ab +
bc + ca = 1. Prove that
1 1 1 √ √ √
a + + b + + c + ≥ 2( a + b + c).
a b c
247. (Macedonia, 2010) Let a, b, c be positive real numbers such that a+b+c =
3. Prove that
a 3 + 2 b3 + 2 c 3 + 2
+ + ≥ 3.
b+2 c+2 a+2
248. (Russia, 2014) Let a, b, c, d be real numbers such that a2 +b2 +c2 +d2 = 4.
Prove that (2 + a)(2 + b) ≥ cd.
250. (Russia, 2014) Let a, b, c, d be real numbers having absolute value greater
than 1 and such that abc + abd + acd + bcd + a + b + c + d = 0. Prove that
1 1 1 1
+ + + > 0.
a−1 b−1 c−1 d−1
Problems 237
1 1 1
− < ,
x + y + 1 (x + 1)(y + 1) 11
252. (Kazaksthan, 2008) Let a, b, c be three positive real numbers such that
abc = 1. Prove that
1 1 1 3
+ + ≥ .
b(a + b) c(b + c) a(c + a) 2
x3 y 4 z 3 y 3 z 4 x3 z 3 x4 y 3
P = + 4 + 4 .
(x4 4 2
+ y )(xy + z ) 3 4 2
(y + z )(yz + x ) 3 (z + x4 )(zx + y 2 )3
Find the maximum value of P when x, y, z vary over the set of all positive real
numbers.
n
ak bk
n
≤ .
1 − ak 1 − bk
k=1 k=1
260. (Bulgaria, 2014) Suppose a, b, c, d are positive real numbers. Prove that
a4 a+b+c+d
≥ .
a + a b + ab2 + b3
3 2 4
cyclic
263. (Belarus, 2010) For any three positive real numbers a, b, c, prove that
a2 b2 3a + 2b − c
+ ≥ .
a+b b+c 4
264. (APMO, 2003) Let a, b, c be the sides of a triangle with perimeter equal
to 1. Prove that
√
2
2 2 2 2
a +b + b +c + c +a <1+ 2 2 .
2
267. (JBMO, 2014) Let a, b, c be positive real numbers such that abc = 1.
Prove that
2 2 2
1 1 1
a+ + b+ + c+ ≥ 3(a + b + c + 1).
b c a
Problems 239
269. (Estonia, 2014) Let a, b, c be positive real number such that abc = 1.
Prove that
1 1 1
+ + > 1.
1 + a2014 1 + b2014 1 + c2014
270. (CRUX, 2000) For positive real numbers a, b, c, prove the inequality
1 1 1 1 1 1 9
+ + + + ≥ .
a b c 1+a 1+b 1+c 1 + abc
278. (Estonia, 2014) Let a, b, c be positive real numbers such that a+b+c = 1.
Prove that
a2 b2 c2 1
+ + > .
b 3 + c 4 + 1 c 3 + a 4 + 1 a 3 + b4 + 1 5
279. (Balkan Olympiads, 2014) Let x, y, z be three positive real numbers such
that xy + yz + zx = 3xyz. Prove that
x2 y + y 2 z + z 2 x ≥ 2(x + y + z) − 3.
u2 − v 2
u = a + b + c, = ab + bc + ca, w = abc,
3
where v ≥ 0. Then
(u + v)2 (u − 2v) (u − v)2 (u + 2v)
≤r≤ .
27 27
a4 + b4 + c4 ≥ abc(a + b + c).
Problems 241
a+b+c≥λ
√ √ √
for all positive reals a, b, c with a bc + b c + c a ≥ 1.
Show that for any three distinct numbers j, k, l the numbers aj , ak , al form the
sides of a triangle.
293. (Czech and Slovak, 2013) Two circles Γ1 , Γ2 with respective centre S1 , S2
and radii r1 , r2 are externally tangent to each other and lie in a square ABCD
of side a units so that Γ1 touches DC, DA while Γ2 touches CD, CB. Prove
that the area of at least one of the triangles AS1 S2 and BS1 S2 is no more than
3 2
16 a units.
294. (Czech and Slovak, 2013) Find all λ > 0 such that the inequality
√
a2 + λb2 + b2 + λa2 ≥ a + b + (λ − 1) ab
295. (Netherlands, 2013) Let a, b, c be positive real numbers such that abc = 1.
Prove that
1
a+b+c≥ (a + 2)(b + 2)(c + 2).
3
where xn+1 = x1 .
1 1 1
297. (Belarus, 2013) Let a, b, c be positive real number such that ab + bc + ca =
1. Prove that
a2 + b2 + c2 + ab + bc + ca − 3 a b c
≥ + + .
5 b c a
298. (Turkey, 2012) Show that for all positive real numbers x, y, z, the inequal-
ity
x(2x − y) y(2y − z) z(2z − x)
+ + ≥ 1.
y(2z + x) z(2x + y) x(2y + z)
300. (Turkey, 2012) Suppose a, b, c are positive real numbers such that a3 +
b3 + c3 = a4 + b4 + c4 . Prove that
a b c
+ 2 + 2 ≥ 1.
a2 3
+b +c 3 3
b +c +a 3 c + c3 + a3
Prove that
tan(a0 ) tan(a1 ) tan(a2 ) · · · tan(an ) ≥ nn+1 .
304. Suppose a, b, c are positive real numbers such that abc = 1. Prove that
a2 + bc
≥ ab + bc + ca.
a2 (b + c)
cyclic
1 1 1 3
+ + ≤ .
a 4 + b + c b4 + c + a c 4 + a + b a+b+c
1
a4 (b + c) + b4 (c + a) + c4 (a + b) ≤ (a + b + c)5 .
12
1 1 1 1
+ + − ≥ 2.
a+b b+c c+a a+b+c
1 + a 2 b2 1 + b2 c 2 1 + c 2 a2 5
2
+ 2
+ ≥ .
(a + b) (b + c) (c + a)2 2
a4 + b4 + c4 + d4 + 2abcd ≥ a2 b2 + a2 c2 + a2 d2 + b2 c2 + b2 d2 + c2 d2 .
312. (Indian Team Selection, 2017) Let a, b, c be distinct positive real numbers
such that abc = 1. Prove that
a6
> 15.
(a − b)(a − c)
cyclic
313. (Indian Team Selection, 2010) Let a, b, c be real numbers such that a2 +
b2 + c2 = 1. Prove that
√
a + b + c ≤ 2abc + 2.
315. (Ukraine, 2005) Let a, b, c be positive real numbers such that a+b+c = 1.
Prove that
1 1 1 1 1 1
−1 −1+ −1 −1+ −1 − 1 ≥ 6.
a b b c c a
Chapter 6
Solutions to problems
= (n − 1)a2n+1 .
a b c
+ + ≥ 1.
b + 2c c + 2a a + 2b
246 Solutions
It follows that
a2 (a + b + c)2
≥ .
ab + 2ca 3(ab + bc + ca)
cyclic
Equivalently, we need to prove that a2 +b2 +c2 ≥ ab+bc+ca, which is clear from
the Cauchy-Schwarz inequality. (Or one can use (a−b)2 +(b−c)2 +(c−a)2 ≥ 0.)
Thus, we get
2 2
3 a2 + b2 + c2 ≥ a + b + c ≥ abc .
On the other hand, we also have
2/3
a2 + b2 + c2 ≥ 3 abc ,
2 2 2 27
+ + ≥ .
b(a + b) c(b + c) a(c + a) (a + b + c)2
√
3
3
Solution: Taking A = abc and B = (a + b)(b + c)(c + a), we have
2 2 2 6
+ + ≥ ,
b(a + b) c(b + c) a(c + a) AB
a+b+c 2(a + b + c)
A≤ , B≤ .
3 3
Thus
6 27
≥ .
AB (a + b + c)2
1
1 ≤ ab + bc + ca − abc ≤ 1 + .
27
Solution: We have
1
0 ≤ (1 − a)(1 − b)(1 − c) ≤ .
27
But, this is a consequence of the AM-GM inequality: observe a < b + c = 2 − a
so that 1 − a > 0, and similarly 1 − b > 0, 1 − c > 0;
3
1−a+1−b+1−c 1
(1 − a)(1 − b)(1 − c) ≤ = .
3 27
1 + ab 1 + bc 1 + cd 1 + da
+ + + ≥ 4.
1+a 1+b 1+c 1+d
But,
1 1 4
+ ≥
1 + a ab(1 + c) 1 + a + ab + abc
1 1 4
+ ≥ .
1 + b bc(1 + d) 1 + b + bc + bcd
Thus
1 + ab 4(1 + ab) 4(1 + bc)
≥ +
1+a 1 + a + ab + abc 1 + b + bc + bcd
cyclic
4(1 + ab) 4a(1 + bc)
= +
1 + a + ab + abc a + ab + abc + abcd
4(1 + ab) 4a(1 + bc)
= +
1 + a + ab + abc 1 + a + ab + abc
= 4.
This reduces to 1 − xyz < 1 after using x + y + z = 1. The result follows from
xyz > 0.
250 Solutions
11. Find all triplets (a, b, c) of positive real numbers which satisfy the system
of equations:
a+b+c = 6,
1 1 1 4
+ + = 2− .
a b c abc
Solution: By the AM-GM inequality, we have
3
a+b+c
abc ≤ = 8.
3
Thus
4 4 3
2− ≤2− = .
abc 8 2
On the other hand,
1 1 1 9 3
+ + ≥ = .
a b c a+b+c 2
Thus,
3 1 1 1 4 3
≤ + + =2− ≤ .
2 a b c abc 2
This shows that equality holds in the AM-GM inequality. Hence a = b = c,
giving (a, b, c) = (2, 2, 2).
Solutions 251
a2 b2 c2 a2 b2 c2
+ + ≥ + +
1 + 2bc 1 + 2ca 1 + 2ab 2 − a2 2 − b2 2 − c2
1 1 1
= −3 + 2 + + .
2 − a2 2 − b2 2 − c2
Thus, it is suﬃcient to prove that
1 1 1 1 3 9
+ + ≥ +3 = .
2 − a2 2 − b2 2 − c2 2 5 5
This follows from AM-HM inequality:
1 1 1 9 9
+ + ≥ = .
2 − a2 2 − b2 2 − c2 6 − (a2 + b2 + c2 ) 5
holds.
But it is easy to see that x(x + 2) < (x + 1)2 for any positive x. Thus
2
a 2 + b 2 a 2 + b2 + 2 < a 2 + b 2 + 1 ,
15. For a ﬁxed positive integer n, compute the minimum value of the sum
x22 x3 xn
x1 + + 3 + ··· + n,
2 3 n
where x1 , x2 , x3 , . . . , xn are positive real numbers such that
1 1 1 1
+ + + ··· + = n.
x1 x2 x3 xn
17. Let a, b, c be positive real numbers, all less than 1, such that a + b + c = 2.
Prove that
abc ≥ 8(1 − a)(1 − b)(1 − c).
Thus,
4(μ + ν)2 4(μ2 + ν 2 )
2 2
≤ = 8.
2(μ + ν − λ) + 4λ μ 2 + ν 2 + λ2
cyclic cyclic
19. Three positive real numbers a, b, c are such that (1 + a)(1 + b)(1 + c) = 8.
Prove that abc ≤ 1.
√ √ √
Solution: We know that 1 + a ≥ 2 a, 1 + b ≥ 2 b and 1 + c ≥ 2 c. Thus,
we get √
8 = (1 + a)(1 + b)(1 + c) ≥ 8 abc.
It follows that abc ≤ 1.
This also follows from Muirhead’s inequality, since (2, 1, 1) ≺ (3, 0, 1).
(Here an+1 = a1 .)
256 Solutions
Thus, we get
n
a2j n
a2j+1
= .
j=1
aj + aj+1 a + aj+1
j=1 j
Solution: Since aj < 1/2 for each j, we see that 1 − 2aj > 0, for 1 ≤ j ≤ n.
Using the Cauchy-Schwarz inequality, we have
2 2
n n
a
aj = j 1 − 2aj
j=1 j=1
1 − 2aj
n
a2j n
≤ (1 − 2aj )
j=1
1 − 2aj j=1
n 2
aj
= (n − 2) .
j=1
1 − 2aj
n
because of j=1 aj = 1.
x1 x2 xn √
+ + ··· + < n.
1 + x21 1 + x21 + x22 1 + x21 + x22 + · · · + x2n
Solution: We have
2 a b
P (α)P (1/α) = aα + bα + c + +c
α2 α
= a 2 + b2 + c 2
1 21
+(ab + bc) α + + ac α + 2
α α
≥ a2 + b2 + c2 + 2ab + 2bc + 2ac
= (a + b + c)2 = P (1)2 ;
we have used the AM-GM inequality.
Thus, we obtain
a2 b2 c2 d2 e2 (a + b + c + d + e)2 2
+ + + + ≥ ,
ab + ac bc + bd cd + ce de + da ea + eb sym ab
where
ab = ab + ac + ad + ae + bc + bd + be + cd + ce + de.
sym
This is equivalent to
2(a2 + b2 + c2 + d2 + e2 ) ≥ ab + ac + ad + ae + bc + bd + be + cd + ce + de.
260 Solutions
Solution: Let us introduce a2 +b2 −c2 = z, b2 +c2 −a2 = x and c2 +a2 −b2 = y.
Then we have
a 2 + b2 − c 2 a 2 − b2 + c 2
cyclic
√ √ √
= xy + yz + zx
1 0√ √ √ √ √ √ 1
= xy + yz + yz + zx + zx + xy
2
1 0√ √ √ √ √ √ 1
= xy + xz + yz + yx + zx + zy
2
1 0 1
≤ (x + y)(x + z) + (y + z)(y + x) + (z + x)(z + y)
2
1 0√ 2 √ √ 1
= 2c · 2b2 + 2a2 · 2c2 + 2b2 · 2a2
2
= ab + bc + ca.
Solution: We have
a2 b2 c2 1 1 1
+ + =3− + + = 1.
a 2 + 1 b2 + 1 c 2 + 1 a 2 + 1 b2 + 1 c 2 + 1
Thus, we get
a2 + b2 + c2 + 3 ≥ (a + b + c)2 .
This simpliﬁes to
3
ab + bc + ca ≤ .
2
1/3 a3 + b3 + c3 1/3
3 a + b + c ≥ 8 abc + .
3
Solution: Using the concavity of the function f (x) = x1/3 on the interval
(0, ∞), we obtain
1/3 1/3
1/3 a 3 + b3 + c 3 8abc + a3 +b3 +c3
3
8 abc + ≤ 9
3 9
1/3
= 3 24abc + a3 + b3 + c3
24abc + a3 + b3 + c3 ≤ (a + b + c)3 .
This reduces to
33. Let c1 ,c2 ,c3 ,. . .,cn be n real numbers such that either 0 ≤ cj ≤ 1 for all j
or cj ≥ 1 for all j, 1 ≤ j ≤ n. Prove that the inequality
"
n
"
n
1 − p + pcj ≤ 1 − p + p cj
j=1 j=1
"
n
"
n−1
1 − p + pcj = 1 − p + pcj · 1 − p + pcn
j=1 j=1
"
n−1
≤ 1−p+p cj · 1 − p + pcn .
j=1
34. Let x1 , x2 , x3 , x4 be real numbers in the interval (0, 1/2]. Prove that
x1 x2 x3 x4
(1 − x1 )(1 − x2 )(1 − x3 )(1 − x4 )
x41 + x42 + x43 + x44
≤ .
(1 − x1 )4 + (1 − x2 )4 + (1 − x3 )4 + (1 − x4 )4
35. Let x1 ,x2 ,x3 ,. . .,xn be n real numbers such that 0 < xj ≤ 1/2. Prove that
⎛ ⎞ ⎛ ⎞
" / n " /
n n n
n
n
⎝ xj xj ⎠ ≤ ⎝ 1 − xj 1 − xj ⎠.
j=1 j=1 j=1 j=1
Let A = x1 + x2 + · · · + xn /n. Let xl and xs denote respectively the largest
and the smallest among x1 ,x2 ,x3 ,. . .,xn . We show that
n
xj
n j=1
j=1 (1 − xj )
x1 x2 · · · A · · · xl + xs − A · · · xn
≤ ,
1 − x1 1 − x2 · · · 1 − A · · · 1 − xl − xs + A · · · 1 − xn
But
A2 − A xl + xs + xl xs = A − xl A − xs ≤ 0,
since xl ≤ A ≤ xs . Now consider the set
0 1
x1 , x2 . . . , A, . . . , xl + xs − A , . . . , xn
obtained by replacing xl , xs respectively by A, xl +xs −A . This again satisﬁes
the hypothesis of the problem, as 0 < A ≤ 1/2 and 0 < xl + xs − A ≤ 1/2
Solutions 265
(since xs ≥ A and xl ≤ A). Moreover the average of the set does not change
by this process and it remains A. We may continue this process and after at
most n − 1 steps all the numbers in the set are equal to A, the average. Thus
we obtain the inequality
n
xj An
n j=1 ≤ n .
j=1 (1 − xj ) 1−A
However
An (nA)n
n = n
1−A n − nA
n
n
j=1 x j
= n
n
n − j=1 xj
n
n
j=1 xj
= .
n n
j=1 1 − xj
It follows that n
n
n xj
xj j=1
n j=1
≤ n .
j=1 (1 − xj ) n
j=1 1 − xj
Alternate Solution I:
We give here another solution based on induction on n. We use the classical
technique of proving the result whenever n is a power of 2 and then ﬁll up the
remaining gaps by coming back. The case n = 2 is simple. The inequality
required is
x1 x2 1 − x1 1 − x2
2 ≤ 2 .
x1 + x2 2 − x1 − x2
Simpliﬁcation gives the equivalent inequality:
2
x1 − x2 1 − x1 − x2 ≥ 0.
Since x1 + x2 ≤ 1, the result follows for n = 2. We also observe that equality
holds if and only if x1 = x2 . We use the case n = 2 to prove the result
for n = 4.
Consider positive numbers x1 , x2 , x3 , x4 ∈ (0, 1/2]. Taking y1 = x1 + x2 /2
and y2 = x3 + x4 /2, the result for n = 2 gives
y1 y2 1 − y1 1 − y2
2 ≤ 2 .
y1 + y 2 2 − y 1 − y2
266 Solutions
This proves the result for n = 4. Using induction on k, this proves the inequality
for all n of the form n = 2k .
Take any n and ﬁx k such that 2k ≤ n < 2k+1 . Let A be the average of
these n numbers; A = x1 + x2 + · · · + xn /n. We consider 2k+1 numbers
x1 , x2 , . . . , xn , A, A, . . . , A,
where A appears 2k+1 −n times in the above sequence. We apply the inequality
for these 2k+1 numbers. (Note that 0 < A ≤ 1/2.) Thus we obtain
k+1
x 1 x 2 · · · x n A2 −n
2k+1
n
j=1 xj + 2k+1 −n A
2k+1 −n
1 − x1 1 − x2 · · · 1 − xn 1 − A
≤ k+1 .
n 2
j=1 1 − xj + 2 −n 1−A
k+1
This simpliﬁes to
x1 x2 · · · xn 1 − x1 1 − x2 · · · 1 − xn
≤ n ,
An 1−A
Here is another solution using Jensen’s inequality for the convex function
1
f (x) = log −1 ,
x
deﬁned on the interval (0, 1/2). Writing f (x) = log(1 − x) − log x, it is easy to
check that
1 1
f (x) = − − ,
1−x x
1 1
f (x) = − + 2
(1 − x)2 x
1 − 2x
= > 0,
x 1 − x2
2
for x ∈ (0, 1/2). Hence f is convex on (0, 1/2). This implies that
1
n
1 1
log n −1 ≤ log −1 .
j=1 (xj /n) n j=1 xj
an ≤ aj + an−j , 1 ≤ j ≤ (n − 1).
n−1
(n − 1)an ≤ 2 aj .
j=1
But, we have
n−1 a2 a2 a3
bk = a 1 + a 1 + + a1 + +
2 2 3
k=1
a2 an−1
+ · · · + a1 + + ··· + .
2 n−1
This simpliﬁes to
n−1
n−2 n − (n − 1)
bk = (n − 1)a1 + a2 + · · · + an−1 .
2 n−1
k=1
We thus have
n−2 n − (n − 1)
(n − 1)a1 + a2 + · · · + an−1 ≥ a1 + a2 + · · · + an−1 .
2 n−1
Adding a1 + a2 + · · · + an to both sides and simplifying, we get
a2 an−1
n a1 + + ··· + + an ≥ 2 a1 + a2 + · · · + an−1 + an .
2 n−1
But we note that 2 a1 +a2 +· · ·+an−1 +an ≥ (n−1)an +an = nan . Dividing
by n, we get
a2 an−1 an
a1 + + ··· + + ≥ an .
2 n−1 n
This completes the induction step and hence the inequality is valid for all n.
We may assume that not all the aj ’s are equal and not all the bj ’s are equal so
that c1 < cn and dn < d1 . Hence we obtain
cn d1 − c1 dn = cn − c1 d1 + c1 d1 − dn ) > 0.
If k ≥ 2, deﬁne uk and vk by
Then uk ≥ 0, vk ≥ 0 and
1/2 1/2
ck dk = uk c21 + vk c2n uk d21 + vk d2n ≥ uk c 1 d1 + v k c n dn ,
Solutions 271
Since a2j = c2j , b2j = d2j and
n
n
c j dj ≤ a j bj ,
j=1 j=1
n
n
f xj ≤ f xj + n − 1 f 0 .
j=1 j=1
Now we use the induction on n. Suppose the result holds for (n − 1); say
f x1 + f x2 + · · · + f xn−1 ≤ f x1 + x2 + · · · + xn−1 + n − 2 f 0 ,
43. Let a, b, c be positive real numbers and let x be a non-negative real number.
Prove that
ax+2 + bx+2 + cx+2 ≥ ax bc + abx c + abcx .
3
Solution: Consider the function f (a) = 3 a2 −a+1 − a6 +a3 +1 for a ∈ R.
Some computation shows that f (1) = 0, f (1) = 0, f (1) = 0, f (1) = 0, but
f (iv) (1) = 0. Thus (a − 1)4 divides f (a) but not any higher power of (a − 1).
Expanding f (a), we obtain
f (a) = (a − 1)4 2a2 − a + 2 .
But the discriminant of 2a2 − a + 2 is −15 and hence 2a2 − a + 2 ≥ 0 for all
real a. It follows that
3
3 a2 − a + 1 ≥ a6 + a3 + 1,
Hence we get
3 3 3 3
(xyz)2 + (xyz) + 1 ≤ 27 x2 − x + 1 y 2 − y + 1 z 2 − z + 1 .
1 1 8
+ ≥ .
sin A sin B 3 + 2 cos C
Solution: The function f (x) = 1/ sin x is convex on [0, π]. Jensen’s theorem
shows that
1 1 2 2
+ ≥ = .
sin A sin B sin A+B 2
cos(C/2)
Thus, it suﬃces to prove that
1 4
≥ .
cos(C/2) 3 + 2 cos C
Using cos C = 2 cos2 (C/2) − 1, this reduces to
2
2 cos(C/2) − 1 ≥ 0.
Solution: Since the coeﬃcients are all non-negative, we see that P (t) ≥ 1,
for any t ≥ 0. Thus P (x) = 0 has only negative roots; let these be −α1 , −α2 ,
. . . , −αn , where αj ’s are all positive. We have
P (x) = (x + α1 )(x + α2 ) · · · (x + αn ).
Hence
P (2) = (2 + α1 )(2 + α2 ) · · · (2 + αn )
1/3
≥ 3n α 1 α 2 · · · α n
= 3n .
In fact, this may be generalised as follows: for any t ≥ 0, the inequality p(t) ≥
(1 + t)n . We use the weighted AM-GM inequality. As in the above solution,
we have
P (t) = (t + α1 )(t + α2 ) · · · (t + αn ).
276 Solutions
48. Let a1 <a2 <a3 <. . .<an be n positive integers. Prove that
2
a1 +a2 +a3 +· · ·+an ≤ a31 + a32 + a33 + · · · + a3n .
Since aj are integers and 0 < a1 < a2 < · · · < an < an+1 , it follows that
aj ≥ j, for 1 ≤ j ≤ n + 1. Hence an+1 ≥ n + 1, and thus
an+1 − n an+1 − n − 1 ≥ 0.
49. Consider a sequence a1 ,a2 ,a3 ,. . .,an of positive real numbers which add
up to 1, where n ≥ 2is an integer. Prove that for any positive real numbers
n
x1 ,x2 ,x3 ,. . .,xn with j xj = 1, the inequality
n − 2 aj x2j
n
2 xj xk ≤ + ,
n − 1 j=1 1 − aj
j<k
holds.
Solution: We have
n 2
n
1= xj = x2j + 2 xj xk .
j=1 j=1 j<k
This is equivalent to
1
n
x2j
≤ .
n − 1 j=1 1 − aj
Solution: Let us put x31 + x32 + x33 + x34 = A and aj = A − x3j . Then it is easy
to check that
1
A= a1 + a2 + a3 + a4 .
3
Moreover,
1 x3 + x33 + x34 1
a1 = 2 ≥ x2 x3 x4 = .
3 3 x1
Similarly, we get
1 1 1 1 1 1
a2 ≥ , a3 ≥ , a4 ≥ .
3 x2 3 x3 3 x4
Thus it follows that
1 1 1 1 1
A= a1 + a2 + a3 + a4 ≥ + + + .
3 x1 x2 x3 x4
Using Chebyshev’s inequality, we also have
A ≥ x 1 + x2 + x3 + x4 .
51. Let {x} denote the fractional part of x; i.e., {x} = x − [x]. Prove for any
positive integer n,
n
n2 − 1
j ≤ .
j=1
2
Hence
(n+1)2 2 (n+1) 2
n
j = j + j
j=1 j=1 j=n2 +1
2n
n2 − 1 1
≤ + j+0
2 2n j=1
n2 − 1 1 2n(2n + 1)
= + ·
2 2n 2
(n + 1)2 − 1
= .
2
This completes the inductive step and hence the proof as well.
Solution: Introducing x, y, z by
b + c − a = 2x, c + a − b = 2y, a + b − c = 2z,
we get a = y + z, b = z + x, c = x + y and the inequality takes the form
2x(y + z)2 + 2y(z + x)2 + 2z(x + y)2 ≤ 3(x + y)(y + z)(z + x).
Note that x, y, z are positive. The inequality may be reduced to
6xyz ≤ x2 y + y 2 z + z 2 x + xy 2 + yz 2 + zx2 ,
which directly follows from the AM-GM inequality.
54. Let x, y, z be positive real numbers such that xyz ≥ xy + yz + zx. Prove
that
xyz ≥ 3(x + y + z).
280 Solutions
Solution: We have
2
xy + yz + zx − 3xyz(x + y + z)
= (xy)2 + (yz)2 + (zx)2 − (xy)(yz) − (yz)(zx) − (zx)(xy)
10 1
= (xy − yz)2 + (yz − zx)2 + (zx − xy)2 .
2
Hence it follows that
2
3xyz(x + y + z) ≤ xy + yz + zx ≤ (xyz)2 .
55. Let b1 ,b2 ,b3 ,. . .,bn be n non-negative real numbers and let b denote the
sum of these numbers. Prove that
n−1
b2
bj bj+1 ≤ .
j=1
4
n−1
k−1
n−1
bj bj+1 = bj bj+1 + bj bj+1
j=1 j=1 j=k
k−1
n−1
≤ bk bj + bk bj+1
j=1 j=k
= bk b − b k
2
b2 b b2
= − − bk ≤ .
4 2 4
In this case
0 1 0 1 0 1
max 1, |r + s|, |rs| ≥ |rs| > k|r||s| = k max 1, |r| max 1, |s| .
By symmetry, the result follows in the case |r| ≥ 1 and |s| < 1.
57. Let x, y, z be three real numbers in the interval [0, 1] such that xyz =
(1−x)(1−y)(1−z). Find the least possible value of x(1−z)+y(1−x)+z(1−y).
This is equivalent to
⎛ ⎞ ⎛ ⎞
4⎝ a+ ab⎠ ≥ 3 ⎝1 + a+ ab + abc⎠ .
cyclic cyclic cyclic cyclic
This reduces to
a+ ab ≥ 6.
cyclic cyclic
However, we have
1/3
a+b+c ≥ 3 abc = 3,
2/3
ab + bc + ca ≥ 3 abc = 3.
58. Let x1 ,x2 ,x3 ,. . .,xn be non-negative real numbers such that
n
1
≤ 1.
j=1
1 + xj
Solution:
Introduce yj = 1/(1 + xj ), for 1 ≤ j ≤ n.
nThen we get xj =
1 − yj /yj , 1 ≤ j ≤ n. The condition translates to j=1 yj ≤ 1 and the
conclusion we have to derive is
n
"
1 − yj
≥ (n − 1)n .
j=1
yj
n
1 n2
≥ n ≥ n.
y
j=1 j
yj
j=1
Solutions 283
Alternate solution:
We introduce the polynomial
P (z) = z + x1 z + x2 · · · z + xn .
Then
n
1 P (1)
= .
j=1
1 + xj P (1)
The given condition is P (1) ≤ P (1). We now write
where σ’s are elementary symmetric functions in x1 ,x2 ,x3 ,. . .,xn . Then the
derivative is given by
60. Show that for any two natural numbers m, n, the inequality
1 1 4
− ≤
m + n + 1 (m + 1)(n + 1) 45
holds.
Гораздо больше, чем просто документы.
Откройте для себя все, что может предложить Scribd, включая книги и аудиокниги от крупных издательств.
Отменить можно в любой момент.