Académique Documents
Professionnel Documents
Culture Documents
INEQUALITIES
UNIT 1
CLASSICAL INEQUALITIES
(A) A man drove for 2 hours. In the first hour he travelled 16 km, and in the second hour he
travelled 32 km. What is his average speed for the whole journey?
(B) A man drove from City P to City Q at a speed of 16 km/h and returned at 32 km/h. What
is his average speed for the whole journey?
(C) In the traditional Chinese game mahjong, the value (in order not to encourage gambling
we use the term value in place of payoff) of a hand is determined by the number of
points of the hand and usually grows in a geometric sequence. For instance,
Points
Value
16
32
64
128
256
But in this case the value grows too fast. Some people want to modify it so that the value
doubles every 2 points rather than every point. For instance,
Points
Value
16
32
What should be x, y and z? Some people tend to simply put x = 6 , y = 12 and z = 24 , but
then the numbers in the second row will no longer form a geometric sequence. Can we
choose x, y, z so that 4, x, 8, y, 16, z, 32 form a geometric sequence?
These questions should be fairly easy. For (A), the average speed is
16 + 32
= 24 km/h .
2
Page 1 of 15
Mathematical Database
For (B), suppose the distance between P and Q is d km. Then the time taken for the journey
d
d
from P to Q and the return journey are
and
hours, respectively. Hence the average speed is
16
32
d
16
2d
=
+ 32d
1
16
2
64
km/h .
=
1
+ 32
3
For (C), in order that the numbers form a geometric sequence, we must have
x 8
= , i.e.
4 x
The above three situations illustrate the three means that we are going to study. In (A), the
16 + 32
= 24 , the usual mean, is known as the arithmetic mean (A.M.). The mean in (B),
mean
2
2
64
=
, is known as the harmonic mean (H.M.). The mean in (C), 16 32 = 16 2 , is known
1
1
+
3
16
32
as the geometric mean (G.M.).
H.M. =
1
a1
n
+
1
a2
+ a1
From the previous three examples, we see that for the particular case n = 2 , a1 = 16 and
a2 = 32 , A.M. G.M. H.M.. Indeed, for n = 2 this is not hard to prove:
1
a1
2a a
2
= 1 2 =
1
+a
a1 + a2
2
2a1a2
a1 a2
+ 2 a1a2
2a1a2
2 a1a2
= a1a2
( a1 + a2 ) ( a1 a2 )
2
=
=
( a1 + a2 )
a1 + a2
2
Mathematical Database
+ an
n a1a2
an
1
a1
1
a2
n
+
+ a1n
= an .
Example 1.1.
Prove that for positive real numbers a, b, c,
( a + b + c ) 1 + 1 + 1 9 .
a b c
Solution.
By the AM-HM inequality,
a+b+c
3
1 1 1.
3
a + b + c
1 1 1
Hence ( a + b + c ) + + 9 .
a b c
Example 1.2.
Let a and b be real numbers such that a > b > 0 . Determine the least possible value of
a+
1
.
b( a b)
Solution.
By the AM-GM inequality,
a+
1
1
1
= ( a b) + b +
3 3 ( a b) b
=3.
b( a b)
b( a b)
b( a b)
Equality holds if a b = b =
1
, i.e. a = 2 and b = 1 .
b( a b)
Page 3 of 15
Mathematical Database
Example 2.1.
Let a, b, c, d be real numbers. Prove that (a 2 + c 2 )(b 2 + d 2 ) (ab + cd ) 2 .
Determine when equality holds.
Solution.
If any one of a, b, c, d is zero, the inequality is trivial.
In particular, if a = b = 0 or c = d = 0 , then equality holds.
Assume a, b, c, d are non-zero.
Then a 2 d 2 and b 2c 2 are positive. By the AM-GM inequality, we have
a 2 d 2 + b 2 c 2 2 (a 2 d 2 )(b 2 c 2 ) = 2 abcd 2abcd .
a 2b 2 + c 2 d 2 + a 2 d 2 + b 2c 2 a 2b 2 + c 2 d 2 + 2abcd ,
i.e.
(a 2 + c 2 )(b 2 + d 2 ) (ab + cd ) 2 .
Equality holds if and only if a 2 d 2 = b 2c 2 and abcd = abcd , i.e. ad = bc .
Combining all cases, equality holds if and only if ad = bc .
Basically, the above inequality says that the product of the sum of squares is greater than the
square of the sum of products. Indeed, this generalizes to 2n variables. We have
(a12 + a2 2 +
+ an 2 )(b12 + b2 2 +
+ bn 2 ) (a1b1 + a1b2 +
+ anbn ) 2 .
Page 4 of 15
Mathematical Database
a1 a2
=
=
b1 b2
an
or b1 = b2 =
bn
= bn = 0 .
a b ( ab )
2
ai bi aibi .
i =1
i =1
i =1
Remark 2. By a slight modification, the Cauchy-Schwarz inequality holds for complex numbers as
well. In fact, if a1 , a2 , , an , b1 , b2 , , bn are complex numbers, then
(a
+ a2 +
2
+ an
)( b
+ b2 +
+ bn
) ab +ab +
1 1
1 2
+ anbn .
2
Example 2.2.
Prove that for positive real numbers a, b, c,
( a + b + c ) 1 + 1 + 1 9 .
a b c
Solutions.
Note that this is the same inequality as in Example 1.1.
Here we apply the Cauchy-Schwarz inequality. We have
1 1 1
(a + b + c) + + =
a b c
1 2 1 2 1 2
c
+
+
a b c
( a) +( b) +( )
2
1
1
1
a
+ b
+ c
a
b
c
=9
Page 5 of 15
Mathematical Database
a
1
b
1
c
1
, i.e. a = b = c .
Example 2.3.
Let x and y be real numbers such that x 1 y 2 + y 1 x 2 = 1 .
Prove that x 2 + y 2 = 1 .
Solution.
By the Cauchy-Schwarz inequality (with a1 = x , a2 = 1 x 2 , b1 = 1 y 2 , b2 = y ), we have
x 1 y 2 + y 1 x2 x2 +
1 x2
) y + (
2
1 y2
) = 1 .
2
1 x2
,
y
i.e. x 2 y 2 = (1 x 2 )(1 y 2 ) , or x 2 + y 2 = 1 .
Now x 1 y 2 + y 1 x 2 = 1 , so we must have x 2 + y 2 = 1 .
The Cauchy-Schwarz inequality is unique in its special condition for equality case. Most other
inequalities have equality case when the variables are equal. Indeed, observing the equality case is
one important technique in proving inequalities, as we shall see in Unit 2.
A class of 40 students is split into four groups, A, B, C and D, with 7, 8, 12 and 13 students
respectively. Each group is to choose a type of shirts to represent their group. These are four types
of shirts, and their unit prices are $10, $15, $20 and $25 respectively. Of course, each student will
buy a shirt representing his group and each group will choose a different type of shirts. What
matching between groups and shirts will yield a minimum total expenses of the class on shirts?
What matching will yield a maximum?
Page 6 of 15
Mathematical Database
It is quite intuitive that the total expenses will be minimum when Group D, with the largest
number of students, chooses the least expensive type of shirts, Group C chooses the second least
expensive type of shirts, and so on. By contrast, maximality occurs when Group D chooses the most
expensive type of shirts, Group C the second most expensive type of shirts and so on.
The above situation illustrates the following:
an and b1 b2
bn ,
then
a1b1 + a2b2 +
+ anb1 ,
= an or b1 = b2 =
= bn .
Example 3.1.
Let a1 , a2 , , an be positive real numbers and {b1 , b2 , , bn } be a permutation of {a1 , a2 , , an } .
Prove that
a1 a2
+ +
b1 b2
an
n.
bn
Solution.
Without loss of generality, assume a1 a2
an . Then
1
1
an an 1
1
.
a1
Mathematical Database
a1 a2
+ +
b1 b2
1
+ an
bn
1
+ an
an
1
1
an
= a1 + a2 +
bn
b1
b2
1
1
a1 + a2 +
a1
a2
=n
as desired.
4. Other Inequalities
In this section we will look at generalizations and variations to the three inequalities we have
learned.
We first look at two generalizations of the AM-GM inequality. For any real number s and
positive real numbers a1 , a2 , , an , we define
a s + a2 s +
Ms = 1
n
+ an s s
to be the s-power mean of a1 , a2 , , an . It is clear from this definition that M 1 is the familiar AM
and M 1 is the HM. A little knowledge in calculus would enable us to define M 0 as
lim M s = (a1a2
s 0
an )1/ n which is the GM. Hence Theorem 1.1 says M 1 M 0 M 1 . The following
= an .
a12 + a2 2 +
n
+ an 2
Page 8 of 15
Mathematical Database
is usually known as the root mean square (RMS) or quadratic mean (QM) of a1 , a2 , , an .
Also, it is customary to define
M + = lim M s = max {a1 , a2 ,..., an }
s +
This implies that all means lie between the maximum and minimum of the group of numbers.
Example 4.1.
Given a set of data x1 , x2 , , xn with mean (the usual arithmetic mean) x , we define the mean
deviation d and the standard deviation as
d=
1 n
xi x
n i =1
1 n
2
( xi x )
n i =1
Show that d .
Solution.
Consider the n non-negative numbers x1 x , x2 x , , xn x .
If all of these are zeros the result is trivial.
So we assume that exactly k of these numbers are non-zero, and without loss of generality we may
assume that the first k of them are non-zero.
Applying the RMS-AM inequality ( M 2 M 1 ) on these k numbers, we have
1 n
1 n
2
x
x
xi x .
i
k i =1
k i =1
Note that
k
k
k
1 , so
. Multiplying this to the above inequality, we get
n
n
n
1 n
1 n
2
x
( xi x ) ,
i
n i =1
n i =1
i.e.
d .
Page 9 of 15
Mathematical Database
Pn = a1a2
+ an 1an
an
Hence the AM is simply P1 and the GM is Pn1/ n . Theorem 1.1 says P1 Pn1/ n . The following
theorem generalizes this.
Pn1/ n .
= an .
Example 4.2.
Prove that for positive real numbers a, b, c,
1 1 1 a 8 + b8 + c 8
+ +
a b c
a 3b3c 3
Solution.
1
a 8 + b8 + c 8 8 a + b + c
Since M 8 M 1 , we have
.
3
3
i.e.
2
a + b + c (abc) (ab + bc + ca)
3
3
Page 10 of 15
Mathematical Database
a 8 + b8 + c 8
1
(abc) 2 (ab + bc + ca) 1 1 1
a+b+c
= + + .
3
3
a 3b 3 c 3
3
( abc)3
a b c
( abc)
8
Page 11 of 15
Mathematical Database
+ anbn a1 + a2 +
a1 a2
=
=
b1 b2
an
) (b
1
p
+ b2 +
q
bn
1
q q
an
.
bn
Example 4.3.
Prove that for positive real numbers a, b, c,
a 3 + b 3 + c 3 a 2b + b 2 c + c 2 a .
Solution.
Setting p =
3
, q = 3 in Hlders inequality, we have
2
2
3
3
3 3
1
a 2b + b 2 c + c 2 a ( a 2 ) 2 + ( b 2 ) 2 + ( c 2 ) 2 ( a 3 + b 3 + c 3 ) 3
3
3
3
= a +b +c
Alternative Solution.
Without loss of generality, assume a b c . Then a 2 b 2 c 2 .
Since Direct Sum Random Sum, we have
a 2 (a ) + b 2 (b) + c 2 (c) a 2 (b) + b 2 (c) + c 2 ( a) ,
i.e.
a 3 + b 3 + c 3 a 2b + b 2 c + c 2 a .
Page 12 of 15
Mathematical Database
+ anbn
(a1 + a2 +
+ an )(b1 + b2 +
n
= an or b1 = b2 =
an , b1 b2
+ bn )
bn . Then
a1bn + a2bn 1 +
+ anb1 .
= bn .
Example 4.4.
Let a1 , a2 , , an be positive real number less than 1, a1 + a2 +
a1
a
+ 2 +
1 a1 1 a2
+ an = S . Prove that
an
nS
.
1 an n S
Solution.
Without loss of generality assume a1 a2
Then 1 an 1 an 1
an .
1 a1 and so
a1
a
2
1 a1 1 a2
an
.
1 an
By Chebyschevs inequality,
S = a1 + a2 +
+ an
an
(1 an )
1 an
a1
a
(1 a1 ) + 2 (1 a2 ) +
1 a1
1 a2
a
1 a1
+ 2 +
n 1 a1 1 a2
an
[ (1 a1 ) + (1 a2 ) +
1 an
1 a1
a
+ 2 +
n 1 a1 1 a2
an
(n S )
1 an
+ (1 an ) ]
Hence
Page 13 of 15
Mathematical Database
a1
a
+ 2 +
1 a1 1 a2
an
nS
.
1 an n S
Page 14 of 15
Mathematical Database
5. Exercises
1.
Prove that if
(1 + a1 )(1 + a2 )
(1 + an ) = 2n ,
then
a1a2
2.
Prove that
a12 a2 2
+
+
a2
a3
3.
an 1 .
an 2
+
a1 + a2 +
a1
+ an .
4.
Prove that
a
b
c
3
+
+
.
b+c c+a a+b 2
5.
1
+
2
1
a
a1 + 22 +
n
2
an
.
n2
Page 15 of 15