Algebraic Inequalities

Arkadii Slinko

1 The Rearrangement Inequality and Its Conse-

quences
Theorem 1 (The Rearrangement Inequality). Let a
1
a
2
a
n
and b
1

b
2
b
n
be real numbers. For any permutation (a

1
, a

2
, . . . , a

n
) of (a
1
, a
2
, . . . , a
n
),
we have
a
1
b
1
+ a
2
b
2
+ + a
n
b
n
a

1
b
1
+ a

2
b
2
+ + a

n
b
n
a
n
b
1
+ a
n1
b
2
+ + a
1
b
n
,
with equality if and only if a
i
= a
i+1
for all i such that b
i
< b
i+1
. In particular, if
b
1
< b
2
< . . . < b
n
, the equality takes place if (a

1
, a

2
, . . . , a

n
) is equal to (a
1
, a
2
, . . . , a
n
)
or (a
n
, a
n1
, . . . , a
1
), respectively.
Proof. We will prove only the rst inequality. Suppose a

i
> a

i+1
, then
a

i
b
i
+ a

i+1
b
i+1
a

i+1
b
i
a

i
b
i+1
= (a

i
a

i+1
)(b
i
b
i+1
) 0.
Moreover, if b
i
< b
i+1
, we have
a

i
b
i
+ a

i+1
b
i+1
a

i+1
b
i
a

i
b
i+1
< 0,
which proves the statement since swapping neighboring pairs like the one above we can
convert (a

1
, a

2
, . . . , a

n
) into (a
1
, a
2
, . . . , a
n
).
Corollary 1. Let a
1
, a
2
, . . . , a
n
be real numbers and (a

1
, a

2
, . . . , a

n
) be a permutation
of (a
1
, a
2
, . . . , a
n
). Then
a
2
1
+ a
2
2
+ + a
2
n
a
1
a

1
+ a
2
a

2
+ + a
n
a

n
.

with borrowings from the article K.Wu and Andy Liu The rearrangement inequality.
1
Corollary 2. Let a
1
, a
2
, . . . , a
n
be positive numbers and (a

1
, a

2
, . . . , a

n
) be a permuta-
tion of (a
1
, a
2
, . . . , a
n
). Then
a

1
a
1
+
a

2
a
2
+ +
a

n
a
n
n.
For the two inequalities above it is hard to formulate a general rule when inequal-
ity becomes an equality. But there are cases, when it is possible. For example, if
(a

1
, a

2
, . . . , a

n
) is cyclic, then the equality takes place i a
1
= a
2
= . . . = a
n
.
Exercise 1. Let a
1
, a
2
, . . . , a
n
be real numbers. Then
a
2
1
+ a
2
2
+ + a
2
n
a
1
a
2
+ a
2
a
3
+ + a
n
a
1
(1)
with the equality i a
1
= a
2
= . . . = a
n
.
Exercise 2. Let a
1
, a
2
, . . . , a
n
be positive numbers. Then
a
n
a
1
+
a
1
a
2
+ +
a
n1
a
n
n (2)
with the equality i a
1
= a
2
= . . . = a
n
.
Exercise 3. Prove that for all non-negative a, b, c the following two inequality hold
a
2
b + b
2
c + c
2
a a
3
+ b
3
+ c
3

a
4
b
+
b
4
c
+
c
4
a
.
Establish when these are equalities.
Simple as it sounds, the Rearrangement Inequality is a result of fundamental im-
portance. We shall derive from it many familiar and useful inequalities.
Theorem 2 (The Arithmetic Mean Geometric Mean Inequality). Let x
1
, x
2
, . . . , x
n
be positive numbers. Then
x
1
+ x
2
+ + x
n
n

n

x
1
x
2
x
n
,
with equality if and only if x
1
= x
2
= = x
n
.
Proof. Let G =
n

x
1
x
2
x
n
, a
1
=
x
1
G
, a
2
=
x
1
x
2
G
2
, . . . , a
n
=
x
1
x
2
x
n
G
n
= 1. By
Corollary 2,
n
a
1
a
n
+
a
2
a
1
+ +
a
n
a
n1
=
x
1
G
+
x
2
G
+ +
x
n
G
,
which is equivalent to
x
1
+ x
2
+ + x
n
n
G. As the permutation was cyclic, equality
holds if and only if a
1
= a
2
= = a
n
, or x
1
= x
2
= = x
n
.
2
Theorem 3 (The Geometric mean Harmonic Mean Inequality). Let x
1
, x
2
, . . . , x
n
be positive numbers. Then
n

x
1
x
2
x
n

n
1
x
1
+
1
x
2
+ +
1
xn
,
with equality if and only if x
1
= x
2
= = x
n
.
Proof. Let G, a
1
, a
2
, . . . , a
n
be as in Example 5. By Corollary 2,
n
a
1
a
2
+
a
2
a
3
+ +
a
n
a
1
=
G
x
1
+
G
x
2
+ +
G
x
n
,
which is equivalent to
G
n
1
x
1
+
1
x
2
+ +
1
xn
.
As the permutation was cyclic, equality holds if and only if x
1
= x
2
= = x
n
.
Theorem 4 (The Root Mean Square Arithmetic Mean Inequality). Let x
1
, x
2
, . . . , x
n
be real numbers. Then
_
x
2
1
+ x
2
2
+ + x
2
n
n

x
1
+ x
2
+ + x
n
n
,
with equality if and only if x
1
= x
2
= = x
n
.
Proof. By Corollary 1, we have
x
2
1
+ x
2
2
+ + x
2
n
x
1
x
2
+ x
2
x
3
+ + x
n
x
1
,
x
2
1
+ x
2
2
+ + x
2
n
x
1
x
3
+ x
2
x
4
+ + x
n
x
2
,

x
2
1
+ x
2
2
+ + x
2
n
x
1
x
n
+ x
2
x
1
+ + x
n
x
n1
.
Adding these and x
2
1
+ x
2
2
+ + x
2
n
= x
2
1
+ x
2
2
+ + x
2
n
, we have
n(x
2
1
+ x
2
2
+ + x
2
n
) (x
1
+ x
2
+ + x
2
n
)
2
,
which is equivalent to the desired result. Equality holds if and only if
x
1
= x
2
= = x
n
.
Exercise 4. Prove that for positive a, b, c
a
a
b
b
c
c
a
b
b
c
c
a
.
When does this become an equality?
3
2 Cauchys Inequality and Its Consequences
The full name of this inequality is Cauchy-Schwarz-Bunyakovski Inequality.
Theorem 5 (Cauchys Inequality). Let a
1
, a
2
, . . . a
n
, b
1
, b
2
, . . . , b
n
be real numbers.
Then
(a
1
b
1
+ a
2
b
2
+ + a
n
b
n
)
2
(a
2
1
+ a
2
2
+ + a
2
n
)(b
2
1
+ b
2
2
+ + b
2
n
),
with equality if and only if for some constant k, a
i
= kb
i
for 1 i n or b
i
= ka
i
for
1 i n.
Proof. If a
1
= a
2
= = a
n
= 0 or b
1
= b
2
= = b
n
= 0, the result is trivial.
Otherwise, dene S =
_
a
2
1
+ a
2
2
+ + a
2
n
and T =
_
b
2
1
+ b
2
2
+ + b
2
n
. Since both
are non-zero, we may let x
i
=
a
i
S
and x
n+i
=
b
i
T
for 1 i n. By Corollary 1,
2 =
a
2
1
+ a
2
2
+ + a
2
n
S
2
+
b
2
1
+ b
2
2
+ + b
2
n
T
2
= x
2
1
+ x
2
2
+ + x
2
2n
x
1
x
n+1
+ x
2
x
n+2
+ + x
n
x
2n
+ x
n+1
x
1
+ x
n+2
x
2
+ + x
2n
x
n
=
2(a
1
b
1
+ a
2
b
2
+ + a
n
b
n
)
ST
,
which is equivalent to the desired result. Equality holds if and only if x
i
= x
n+i
for
1 i n, or a
i
T = b
i
S for 1 i n.
Example 1. For positive real numbers x
1
, . . . , x
n
we have
(x
2
1
+ x
2
2
+ . . . + x
2
n
)(1
2
+ 1
2
+ . . . + 1
2
) (x
1
+ x
2
+ . . . + x
n
)
2
,
from which
x
2
1
+ x
2
2
+ . . . + x
2
n
n

_
x
1
+ x
2
+ . . . + x
n
n
_
2
and we again obtain the Root Mean Square Arithmetic Mean Inequality.
Example 2. Prove the inequality
sin sin + cos + cos 2.
Indeed,
sin sin + cos 1 + 1 cos
_
sin
2
+ cos
2
+ 1
2
_
sin
2
+ 1
2
+ cos
2
= 2.
4
Exercise 5. Use Cauchys inequality to prove that, for all positive x
1
, . . . , x
n
,
(x
1
+ x
2
+ . . . + x
n
)
_
1
x
1
+
1
x
2
+ . . . +
1
x
n
_
n
2
,
and show that this is equivalent to Arithmetic Mean Harmonic Mean Inequality.
Another very useful form of Cauchys inequality is as follows.
Theorem 6. Let x
1
, . . . , x
n
be arbitrary real numbers and y
1
, . . . , y
n
be positive real
numbers. Then
x
2
1
y
1
+
x
2
2
y
2
+ . . . +
x
2
n
y
n

(x
1
+ x
2
+ . . . + x
n
)
2
y
1
+ y
2
+ . . . + y
n
.
Solution. One can reduce this to the Cauchys inequality by substituting a
i
=
x
i

y
i
and
b
i
=

y
i
.
Let us reformulate the Cauchys inequality as follows:
Theorem 7.
n

k=1
x
2
k
= max
_
_
_
_
n

k=1
x
k
y
k
_
2
_
n

k=1
y
2
k
_
1
_
_
_
, (3)
where (y
1
, . . . , y
n
) = (0, . . . , 0) and arbitrary otherwise.
Proof. Obvious from the original Cauchys inequality. Indeed it implies that
n

k=1
x
2
k

_
n

k=1
x
k
y
k
_
2
_
n

k=1
y
2
k
_
1
.
On the other hand both sides are equal, for example, at x
k
= y
k
.
One obvious but very useful corollary.
Corollary 3.
n

k=1
x
2
k
= max
_
n

k=1
x
k
y
k
_
2
, (4)
where

n
k=1
y
k
= 1 and arbitrary otherwise.
Firstly we will exploit the theorem.
5
Example 3 (IMO 95). Let a, b, c be positive real numbers such that abc = 1. Prove
that
1
a
3
(b + c)
+
1
b
3
(c + a)
+
1
c
3
(a + b)

3
2
.
Solution. The idea is to dismantle the complicated expression on the left. We will
represent all three summands on the left as squares:
x
2
1
=
1
a
3
(b + c)
, x
2
2
=
1
b
3
(c + a)
, x
2
3
=
1
c
3
(a + b)
,
and will choose y
1
, y
2
, y
3
as follows:
y
2
1
= a(b + c), y
2
2
= b(c + a), y
3
= c(a + b).
The idea is clear: to make both terms on the right of (4) managable. As abc = 1, we
get
1
a
3
(b + c)
+
1
b
3
(c + a)
+
1
c
3
(a + b)

_
1
a
+
1
b
+
1
c
_
2
(a(b + c) + b(c + a) + c(a + b))
1
=
ab + bc + ca
2

3
2
,
since AM-GM implies
ab + bc + ca 3
3

a
2
b
2
c
2
= 3.
The inequality is proven.
Exercise 6 (Kvant, 1985). Prove that for any positive numbers a, b, c, d
a
b + c
+
b
c + d
+
c
d + a
+
d
a + b
2.
Example 4 (GDR 67). Prove that, if n 2, and a
1
, . . . , a
n
are positive numbers
whose sum is S, then
a
1
S a
1
+
a
2
S a
2
+ . . . +
a
n
S a
n

n
n 1
.
Solution. We will apply the theorem for the numbers
x
2
k
=
a
k
S a
k
, y
2
k
= a
k
(S a
k
).
We will obtain
n

k=1
a
k
S a
k

_
n

k=1
a
k
_
2

_
n

k=1
a
k
(S a
k
)
_
1
= S
2
_
n

k=1
a
k
(S a
k
)
_
1
.
6
Hence it is enough to show that
S
2
_
n

k=1
a
k
(S a
k
)
_
1

n
n 1
.
or (n1)S
2
n

n
k=1
a
k
(S a
k
) = nS
2
n

n
k=1
a
2
k
, which follow from the RootMean
Square - Arithmetic Mean inequality.
Exercise 7 (Mongolia, 1996). Prove that for any positive numbers a, b, c, d satisfying
a + b + c + d = 1
1
1

a
+
1
1

b
+
1
1

c
+
1
1

d
8.
Let us now see what the corollary has to oer.
Example 5 (Beckenbach Inequality). For positive numbers x
1
, . . . , x
n
and y
1
, . . . , y
n
the following inequality holds:

n
k=1
(x
k
+ y
k
)
2

n
k=1
x
k
+

n
k=1
y
k

n
k=1
x
2
k

n
k=1
x
k
+

n
k=1
y
2
k

n
k=1
y
k
.
Solution. Let X = x
1
+ . . . + x
n
and Y = y
1
+ . . . + y
n
. We claim that it is sucient
to show that for any positive numbers z
1
, . . . , z
n
(

n
k=1
(x
k
+ y
k
)z
k
)
2
X + Y

(

n
k=1
x
k
z
k
)
2
X
+
(

n
k=1
y
k
z
k
)
2
Y
(5)
Indeed, if (5) was true, then due to Corollary 4

n
k=1
(x
k
+ y
k
)
2

n
k=1
x
k
+

n
k=1
y
k
=
max (

n
k=1
(x
k
+ y
k
)z
k
)
2
X + Y
=
_
n
k=1
(x
k
+ y
k
)z

k
_
2
X + Y

_
n
k=1
x
k
z

k
_
2
X
+
_
n
k=1
y
k
z

k
_
2
Y

max (

n
k=1
x
k
z
k
)
2
X
+
max (

n
k=1
y
k
z
k
)
2
Y
=

n
k=1
x
2
k

n
k=1
x
k
+

n
k=1
y
2
k

n
k=1
y
k
.
To prove (5) we denote
x =
n

k=1
x
k
z
k

X
, y =
n

k=1
y
k
z
k

Y
.
7
Then (5) becomes
(x

X + y

Y )
2
X + Y
x
2
+ y
2
,
which is easy to prove by expanding of brackets.
Exercise 8 (Minkowskis inequality). Using the same idea prove that for any pos-
itive integers x
1
, . . . , x
n
and y
1
, . . . , y
n
_
x
2
1
+ x
2
2
+ . . . + x
2
n
+
_
y
2
1
+ y
2
2
+ . . . + y
2
n

_
(x
1
+ y
1
)
2
+ (x
2
+ y
2
)
2
+ . . . + (x
n
+ y
n
)
2
.
The trick used in the last two problems and formulated in Corollary is called quasi-
linearisation of Cauchys inequality. You can do it with some other inequalities too.
The quasilinearisation of AM-GM will look as follows.
Theorem 8. For any positive integers x
1
, . . . , x
n
n

x
1
x
2
x
n
= min
n

k=1
x
k
y
k
n
,
where y
1
, . . . , y
n
are positive numbers such that y
1
y
2
y
n
= 1.
Exercise 9. Prove this theorem.
Exercise 10. Use the theorem proved in the previous exercise and prove another
Minkowskis inequality
n

x
1
x
2
x
n
+
n

y
1
y
2
y
n

n
_
(x
1
+ y
1
)(x
2
+ y
2
) (x
n
+ y
n
).
3 Chebyshevs Inequality
This is also known as Tchebychefs inequality.
Denition. Let a
1
, a
2
, . . . , a
n
and b
1
, b
2
, . . . , b
n
be two monotonic (monotone increas-
ing or monotone decreasing) sequences of real numbers. These sequences are called
similarly directed, if
(a
i
a
j
)(b
i
b
j
) 0
for all i j. The sequences are called oppositely directed, if
(a
i
a
j
)(b
i
b
j
) 0
for all i j.
8
Theorem 9 (Chebyshevs Inequality). Let a
1
, a
2
, . . . , a
n
and b
1
, b
2
, . . . , b
n
be two
monotonic sequences of real numbers. Then
a
1
+ a
2
+ . . . + a
n
n

b
1
+ b
2
+ . . . + b
n
n

a
1
b
1
+ a
2
b
2
+ . . . + a
n
b
n
n
, (6)
if the two sequences are similarly directed and
a
1
+ a
2
+ . . . + a
n
n

b
1
+ b
2
+ . . . + b
n
n

a
1
b
1
+ a
2
b
2
+ . . . + a
n
b
n
n
, (7)
if they are oppositely directed.
Proof. Suppose the two sequences are similarly directed. By the Rearrangement In-
equality
a
1
b
1
+ a
2
b
2
+ . . . + a
n
b
n
= a
1
b
1
+ a
2
b
2
+ . . . + a
n
b
n
,
a
1
b
1
+ a
2
b
2
+ . . . + a
n
b
n
a
1
b
2
+ a
2
b
3
+ . . . + a
n
b
1
,

a
1
b
1
+ a
2
b
2
+ . . . + a
n
b
n
a
1
b
n
+ a
2
b
1
+ . . . + a
n
b
n1
.
Adding all of them together we obtain
n(a
1
b
1
+ a
2
b
2
+ . . . + a
n
b
n
) (a
1
+ a
2
+ . . . + a
n
)(b
1
+ b
2
+ . . . + b
n
),
which is equivalent to Chebyshevs Inequality (6). The second Chebyshevs Inequality
(6) can be proved similarly.
Exercise 11. Prove Chebyshevs Inequality (7).
Example 6. Prove that for all real numbers x
1
, . . . , x
n
(x
1
+ x
2
+ . . . + x
n
)
_
x
3
1
+ x
3
2
+ . . . + x
3
n
_
n
_
x
4
1
+ x
4
2
+ . . . + x
4
n
_
.
Indeed, we may assume that x
1
x
2
. . . x
n
. But then x
3
1
x
3
2
. . . x
3
n
and
Chebyshevs Inequality (6) is applicable and gives the result.
Exercise 12. Prove that for all real numbers x
1
, . . . , x
n
_
x
5
1
+ x
5
2
+ . . . + x
5
n
_
_
1
x
2
1
+
1
x
2
2
+ . . . +
1
x
2
n
_
n
_
x
3
1
+ x
3
2
+ . . . + x
3
n
_
.
9
4 Jensens Inequality
This method might require a little bit of Calculus.
Denition 1. Let us recall that a function f(x) on a segment [a, b] is concave up if
for all x
1
, x
2
[a, b]
f
_
x
1
+ x
2
2
_

f(x
1
) + f(x
2
)
2
, (8)
and concave down if for all x
1
, x
2
[a, b]
f
_
x
1
+ x
2
2
_

f(x
1
) + f(x
2
)
2
. (9)
Proposition 1. If the function f(x) is continuous on [a, b] and twice dierentiable on
(a, b), then it is concave up i f

(x) 0 and it is concave down i f

(x) 0.
Theorem 10 (Jensens Inequality). Let n 2 and
1
, . . . ,
n
be nonnegative real
numbers such that
1
+ . . . +
n
= 1. Then
1. For an arbitrary concave up function f(x) on a segment [a, b]
f (
1
x
1
+ +
n
x
n
)
1
f(x
1
) + +
n
f(x
n
)
for all x
1
, . . . , x
n
[a, b].
2. For an arbitrary concave down function f(x) on a segment [a, b]
f (
1
x
1
+ +
n
x
n
)
1
f(x
1
) + +
n
f(x
n
)
for all x
1
, . . . , x
n
[a, b].
Example 7. Prove that for all nonnegative x
1
, . . . , x
n
and y
1
, . . . , y
n
,
_
x
3
1
+ x
3
2
. . . + x
3
n
_ _
y
3
1
+ y
3
2
. . . + y
3
n
_
2

_
x
1
y
2
1
+ x
2
y
2
2
. . . + x
n
y
2
n
_
3
.
To prove this, let us denote S = y
3
1
+. . . +y
3
n
. Then our inequality can be rewritten as
x
3
1
+ x
3
2
. . . + x
3
n
S

(x
1
y
2
1
+ x
2
y
2
2
. . . + x
n
y
2
n
)
3
S
3
,
or
y
3
1
S
_
x
1
y
1
_
3
+
y
3
2
S
_
x
2
y
2
_
3
+. . .+
y
3
n
S
_
x
n
y
n
_
3

_
y
3
1
S
_
x
1
y
1
_
+
y
3
2
S
_
x
2
y
2
_
+ . . . +
y
3
n
S
_
x
n
y
n
__
3
,
which follows from concavity up of the function f(x) = x
3
for nonnegative x. Indeed,
f

(x) = 6x 0 for x 0 and

y
3
1
S
+
y
3
2
S
+ . . . +
y
3
n
S
=
S
S
= 1.
10
Often you need to take logs before applying Jensens inequality.
Example 8. Prove that
_
1 +
1
x
__
1 +
1
y
__
1 +
1
z
_
64,
where x, y, z > 0 and x + y + z = 1.
After taking logs on both sides, we get
log
2
_
1 +
1
x
_
+ log
2
_
1 +
1
y
_
+ log
2
_
1 +
1
z
_
6.
We check that the function f(t) = log
2
(1+1/t) is concave up. Indeed, calculation show
that
f

(t) =
1
t
2

1
(t + 1)
2
> 0.
Hence by Jensens inequality
1
3
log
2
_
1 +
1
x
_
+
1
3
log
2
_
1 +
1
y
_
+
1
3
log
2
_
1 +
1
z
_
log
2
_
1 +
3
x + y + z
_
= 2
and the desired inequality follows.
Exercise 13. Using Jensens inequality prove that
a
a/a+b+c
b
b/a+b+c
c
c/a+b+c

a
2
+ b
2
+ c
2
a + b + c
,
where a, b, c are positive integers.
One of the powerful tools is the so-called weighted AM-GM inequality.
Theorem 11 (Weighted AM-GM inequality). Let w
1
, . . . , w
n
be non-negative reals
such that w
1
+ . . . + w
n
= 1, then for any non-negative reals x
1
, . . . , x
n
w
1
x
1
+ w
2
x
2
+ . . . + w
n
x
n
x
w
1
1
x
w
2
2
. . . x
wn
n
with equality if and only if all the x
i
with w
i
= 0 are equal.
Exercise 14. Deduce Weighted AM-GM inequality from Jensens inequality.
Weighted AM-GM inequality is quite powerful but sometimes dicult to use. We
illustrate this on the following example.
11
Example 9. Prove that for all positive reals a, b, c, d
a
4
b + b
4
c + c
4
d + d
4
a abcd(a + b + c + d).
Solution. By weighted AM-GM
23a
4
b + 7b
4
c + 11c
4
d + 10d
4
a
51

51

a
102
b
51
c
51
d
51
= a
2
bcd.
It remains to symmetrise this inequality.
How can we nd weights w
1
=
23
51
, w
2
=
7
51
, w
3
=
11
51
, w
4
=
10
51
? Aiming to get in
this particular case
(a
4
b)
w
1
(b
4
c)
w
2
(c
4
d)
w
3
(d
4
a)
w
4
= a
2
bcd,
we solving the system of linear equations
w
1
+ w
2
+ w
3
+ w
4
= 1,
4w
1
+ w
4
= 2,
4w
2
+ w
1
= 1,
4w
3
+ w
2
= 1.
and nd the weights.
Weighted AM-GM plus symmetrisation is a powerful tool.
Exercise 15. Prove that if abc > 0, then
1
a
+
1
b
+
1
c

a
8
+ b
8
+ c
8
a
3
b
3
c
3
.
by reducing this inequality to weighted AM-GM inequality.
5 Schurs Inequality
The following simple inequality can be often useful.
Theorem 12. Let t 1 be a real number. Then for all non-negative real numbers
x, y, z
x
r
(x y)(x z) + y
r
(y z)(y x) + z
r
(z x)(z y) 0,
where the equality occurs only when x = y = z = 0 or some two of x, y, z are equal and
the third is zero.
12
Proof. Due to the symmetry of the inequality we may assume that x y z. Then
our inequality can be rewritten as
(x y) [x
r
(x z) y
r
(y z)] + z
r
(x z)(y z) 0,
which is obvious since every term of it is non-negative.
Exercise 16. Prove that for any positive a, b, c
a
3
+ b
3
+ c
3
+ 3abc a
2
(b + c) + b
2
(a + c) + c
2
(a + b). (10)
6 Muirheads Inequality
Let us consider the set S
n
of sequences () = (
1
, . . . ,
n
) with the following two
properties:

1
+
2
+ . . . +
n
= 1, (11)

1

2
. . .
n
0. (12)
For any two sequences () and () from S we say that () majorises () if

1
+
2
+ . . . +
r

1
+
2
+ . . . +
r
for all 1 r < n. We denote this as () (). If () () and () = (), we will
write () ().
Example 10. (1, 0, 0) (
1
2
,
1
2
, 0) (
1
2
,
1
3
,
1
6
) (
1
3
,
1
3
,
1
3
).
Now we introduce one more notation: for () from S
n
we denote
x
()
=
1
n!
(x

1
1
x

2
2
. . . x
n
n
+ . . .),
where the dots denote all n! 1 terms obtained by permutations of s.
Example 11. Let n = 3. Then
x
(1,0,0)
=
1
6
(x
1
1
x
0
2
x
0
3
+ x
1
1
x
0
3
x
0
2
+ x
1
2
x
0
1
x
0
3
+ x
1
2
x
0
3
x
0
1
+ x
1
3
x
0
1
x
0
2
+ x
1
3
x
0
2
x
0
1
) =
1
3
(x
1
+ x
2
+ x
3
);
x
(
1
2
,
1
2
,0)
=
1
6
(x
1
2
1
x
1
2
2
x
0
3
+ x
1
2
2
x
1
2
1
x
0
3
+ x
1
2
1
x
1
2
3
x
0
2
+ x
1
2
3
x
1
2
1
x
0
2
+ x
1
2
2
x
1
2
3
x
0
1
+ x
1
2
3
x
1
2
2
x
0
1
) =
1
3
(

x
1
x
2
+

x
1
x
3
+

x
2
x
3
);
x
(
1
3
,
1
3
,
1
3
)
=
3

x
1
x
2
x
3
.
13
Theorem 13 (Muirhead). If () (), then the inequality
x
()
x
()
holds for all non-negative x
1
, . . . , x
n
. There is equality only when () = () or all the
x
i
are equal.
Proof. We will prove this theorem for the case n = 3. The general case can be proved
similarly. We assume that () = () and not all the x
i
are equal. Let us consider the
following three partial cases from which the general case will follow.
(a) Let () = (
1
,
2
,
3
), and be a positive real number such that <
1

2
and (

) = (
1
,
2
+ ,
3
). Then
3!
_
x
()
x
(

)
_
= (x

1
1
x

2
2
+ x

2
1
x

1
2
x

1
1
x

2
2
x

2
1
x

1
2
)x

3
3
+
(x

1
1
x

2
3
+ x

2
1
x

1
3
x

1
1
x

2
3
x

2
1
x

1
3
)x

3
2
+
(x

1
2
x

2
3
+ x

2
2
x

1
3
x

1
2
x

2
3
x

2
2
x

1
3
)x

3
1
.
The latter expression is positive. Indeed, if x
i
= x
j
, then
x

1
i
x

2
j
+ x

2
i
x

1
j
x

1
i
x

2
j
x

2
i
x

1
j
=
(x
i
x
j
)

2
_
(x

i
x

j
)(x

i
x

j
)

> 0,
since both round brackets are either both positive or both negative. Since not all
the x
i
are equal, there will be x
1
= x
2
, or x
1
= x
3
, or x
2
= x
3
. This secures that
x
()
> x
(

)
.
(b) Similarly, if <
2

3
and (

) = (
1
,
2
,
3
+ ), then x
()
> x
(

)
.
(c) The same argument also shows that if
2
=

2
, then, for any
1

3
, and
(

) = (
1
,
2
,
3
+ ), then x
()
> x
(

)
.
Suppose now that
2
<
2
. Then
2
<
1
, and =
1

1
<
1

2
. Then
by the case (a), x
()
> x
(

)
, where (

) = (
1
,
2
+ ,
3
) = (
1
,
2
+ ,
3
). As

3
<
3
,
2
+ >
2
, hence x
(

)
> x
()
by (b). Thus x
()
> x
()
, as required. The
case
2
>
2
is considered similarly using the cases (b) and (a). If
2
=
2
, then the
statement follows straight from (c).
Example 12. Prove that for all non-negative a, b, c
1
3
(a
2
+ b
2
+ c
2
)
1
3
(ab + ac + bc)
3

a
2
b
2
c
2
.
14
Solution. For all non-negative x
1
, x
2
, x
3
by Muirheads theorem
x
(1,0,0)
x
(
1
2
,
1
2
,0)
x
(
1
3
,
1
3
,
1
3
)
.
Hence
1
3
(x
1
+ x
2
+ x
3
)
1
3
(

x
1
x
2
+

x
1
x
3
+

x
2
x
3
)
3

x
1
x
2
x
3
or, after the substitution a =

x
1
, b =

x
2
, c =

x
3
,
1
3
(a
2
+ b
2
+ c
2
)
1
3
(ab + ac + bc)
3

a
2
b
2
c
2
.
Exercise 17. Prove that for all non-negative a, b, c the following two inequalities hold
2(a
3
+ b
3
+ c
3
) a
2
(b + c) + b
2
(c + a) + c
2
(a + b) 6abc.
Establish when they are equalities.
In fact, we do not need to restrict ourselves with positive n requiring only the rst
of the two conditions (11) but not (12).
Example 13. For example, (
1
2
,
1
2
, 0) (2, 1, 0) and the inequality
2(

xy +

xz +

yz)
x
2
y
+
x
2
z
+
y
2
x
+
y
2
z
+
z
2
x
+
z
2
y
.
holds.
Exercise 18. Prove this extension of the Muirhead Theorem.
Americans call Muirheads Theorem a Bunching Principle. They also use a dif-
ferent notation that can sometimes be useful.
The notion of majorisation can be extended to sequences of the same length. If
s = (s
1
, . . . , s
n
) and t = (t
1
, . . . , t
n
) are two nonincreasing sequences, we say that s
majorizes t if s
1
+ +s
n
= t
1
+ +t
n
and s
1
+ +s
i
t
1
+ +t
i
for i = 1, . . . , n.
Theorem 14 (Bunching Principle). If s and t are two nonincreasing sequences
of nonnegative real numbers such that s majorizes t, then

sym
x
s
1
1
x
sn
n

sym
x
t
1
1
x
tn
n
,
where the sums are taken over all n! permutations of variables.
15
7 Constraints and Homogenisation
Homogenisation is a very important technique, especially in conjunction with Muirhead
Theorem. We illustrate it on the following simple example.
Example 14. Let a, b, c be positive real numbers such that abc = 1. Prove that
a + b + c a
2
+ b
2
+ c
2
.
Proof. The inequality is not homogeneous in the sense that dierent terms have dif-
ferent degrees. However the constraint abc = 1 is not homogeneous either and we may
use this to homogenise the inequality. We make it homogeneous by multiplying the
LHS by
3

abc. We will obtain an equivalent inequality

a
4
3
b
1
3
c
1
3
+ a
1
3
b
4
3
c
1
3
+ a
1
3
b
1
3
c
4
3
a
2
+ b
2
+ c
2
,
which follows from Muirheads inequality
a
2
3
b
1
6
c
1
6
+ a
1
6
b
2
3
c
1
6
+ a
1
6
b
1
6
c
2
3
a + b + c
since (1, 0, 0) (
2
3
,
1
6
,
1
6
). One can also use Weighted AM-GM to show
2
3
a
2
+
1
6
b
2
+
1
6
c
2

a
8
b
2
c
2
= a
4
3
b
1
3
c
1
3
and then symmetrise this expression.
Example 15. Let a, b, c be non-negative real numbers such that a + b + c = 1. Prove
that
a
3
+ b
3
+ c
3
+ 6abc
1
4
.
Proof. Multiplying by 4 and homogenising, we obtain
4(a
3
+ b
3
+ c
3
) + 24abc (a + b + c)
3
.
Simplifying we get
a
3
+ b
3
+ c
3
+ 6abc a
2
(b + c) + b
2
(a + c) + c
2
(a + b),
which follows from (10).
Exercise 19. Let a, b, c be real numbers such that abc = 1. Show that
a
4
+ b
4
+ c
4
+ 3(a + b + c)
a
2
b
+
a
2
c
+
b
2
a
+
b
2
c
+
c
2
a
+
c
2
b
.
16
8 Symmetric Averages
Two more theorems which have been seldom used in Math Olympiad practice so far.
Let x
1
, . . . , x
n
be a sequence of non-negative real numbers. We dene the symmetric
averages of x
1
, . . . , x
n
by
d
i
=
s
i
_
n
i
_,
where s
i
s are the coecients of the polynomial
(x x
1
)(x x
2
) . . . (x x
n
) = s
n
x
n
+ . . . + s
1
x + s
0
.
For example, if n = 4, then
d
1
=
1
4
(x
1
+ x
2
+ x
3
+ x
4
),
d
2
=
1
6
(x
1
x
2
+ x
1
x
3
+ x
1
x
4
+ x
2
x
3
+ x
2
x
4
+ x
3
x
4
),
d
3
=
1
4
(x
1
x
2
x
3
+ x
1
x
2
x
4
+ x
1
x
3
x
4
+ x
2
x
3
x
4
),
d
4
= x
1
x
2
x
3
x
4
.
Theorem 15 (Newton). For all i = 1, . . . , n 1
d
2
i
d
i1
d
i+1
.
Example 16. For example, for n = 4 we have d
2
2
d
1
d
3
, i.e.
(x
1
x
2
+x
1
x
3
+x
1
x
4
+x
2
x
3
+x
2
x
4
+x
3
x
4
)
2

9
4
(x
1
+x
2
+x
3
+x
4
)(x
1
x
2
x
3
+x
1
x
2
x
4
+x
1
x
3
x
4
+x
2
x
3
x
4
).
or
_

sym
x
1
x
2
_
2

9
4

sym
x
1

sym
x
1
x
2
x
3
.
The latter makes sense only if n = 4 is agreed upon and implicitely assumed. Otherwise
it is confusing.
Theorem 16 (Maclaurin). d
1

d
2

3

d
3
. . .
n

d
n
.
Example 17. For example, for n = 4 we have

d
2

3

d
3
, or equivalently
_

sym
x
1
x
2
_
3

27
2
_

sym
x
1
x
2
x
3
_
2
.
Example 18. d
1

n

d
n
is the familiar AM-GM inequality. Maclaurins theorem is a
renement of it.
17
9 Hints and Solutions to Exercises
1. Check that if a
1
< a
n
, then the process of transforming (a
2
, . . . , a
n
, a
1
) into
(a
1
, . . . , a
n
) described in the theorem will give at least one strict inequality.
2. Same as in Exercise 1.
3. Assume a b c. Then a
2
b
2
c
2
. Applying the LHS of Rearrangement
inequality to these two sequences we get
a
2
b + b
2
c + c
2
a a
3
+ b
3
+ c
3
Also we have a
4
b
4
c
4
and 1/c 1/b 1/a. Applying the RHS of Rear-
rangement inequality we obtain
a
3
+ b
3
+ c
3

a
4
b
+
b
4
c
+
c
4
a
.
4. Suppose a b c. As ln x is increasing function of x it is sucient to compare
logarithms of both sides of the inequality, i.e. to establish that ln(a
a
b
b
c
c
)
ln(a
b
b
c
c
a
). This can be written as
a ln a + b ln b + c ln c b ln a + c ln b + a ln c,
which is true by Rearrangement inequality since ln a ln b ln c.
5. Apply Cauchys inequality to a
i
= x
i
and b
i
= 1/x
i
. To obtain AM-HM inequality
divide both sides by n and the sum of reciprocals.
6. Let us make use of Theorem 6 to prove it. We will apply (3) for
x
2
1
=
a
b + c
, x
2
2
=
b
c + d
, x
2
3
=
c
d + a
, x
2
4
=
d
a + b
,
y
2
1
= a(b + c), y
2
2
= b(c + d), y
2
3
= c(d + a), y
2
4
= d(a + b).
We will obtain
a
b + c
+
b
c + d
+
c
d + a
+
d
a + b
(a+b+c+d)
2
(a(b + c) + b(c + d) + c(d + a) + d(a + b))
1
= (a + b + c + d)
2
(2ac + 2bd + ab + bc + cd + da)
1
.
To show that the RHS is greater or equal than 2 we need to prove that
(a + b + c + d)
2
2(2ac + 2bd + ab + bc + cd + da),
which can be checked by simply expanding brackets.
18
7. Let us note that
1
1

a
=

a
1

a
+ 1,
hence we have to prove that

a
1

a
+

b
1

b
+

c
1

c
+

d
1

d
4.
We will prove a more general inequality: for all non-negative a
1
, . . . , a
n
such that

n
k=1
a
k
= 1 it is true that
n

k=1

a
k
1

a
k
4,
which coincides with the Mongolian question for n = 4. Let us use Theorem 6
again and apply (3). We obtain

a
k
1

a
k

_
n

k=1
a
k
_
2
_
n

k=1
a
k

a
k
(1

a
k
_
1
=
_
n

k=1
a
k

a
k
(1

a
k
_
1
.
By AM-GM a
k

a
k
= 2a
k

a
k
2

a
k
4
+ a
2
k
, hence
n

k=1
a
k

a
k

1
4
n

k=1
a
k
+
n

k=1
a
2
k
.
This implies
n

k=1
a
k

a
k
(1

a
k
)
1
4
,
which, in turn, implies the inequality.
8. We will apply (4) of Corollary 3. It implies that for some z
1
, . . . , z
n
such that
z
2
1
+ . . . + z
2
n
= 1
_
(x
1
+ y
1
)
2
+ . . . + (x
n
+ y
n
)
2
= max
_
n

k=1
(x
k
+ y
k
)z
k
_
2
.
Using that max(u + v) max u + max v, we obtain
_
(x
1
+ y
1
)
2
+ . . . + (x
n
+ y
n
)
2
= max
_
n

k=1
(x
k
+ y
k
)z
k
_
2

max
_
n

k=1
x
k
z
k
_
2
+ max
_
n

k=1
y
k
z
k
_
2
=
_
x
2
1
+ . . . + x
2
n
+
_
y
2
1
+ . . . + y
2
n
.
19
9. For any y
1
, . . . , y
n
such that y
1
y
2
y
n
= 1 AM-GM implies
n

x
1
x
2
x
n
=
n
_
(x
1
y
1
)(x
2
y
2
) (x
n
y
n
)
n

k=1
x
i
y
i
n
.
This becomes an equality i x
1
y
1
= . . . = x
n
y
n
. This can be achieved by taking
y
i
=
q
x
i
, where q =
n

x
1
x
2
x
n
.
10. Using the theorem from the previous exercise we get for some z
1
, . . . , z
n
such that
z
1
z
2
z
n
= 1
n
_
(x
1
+ y
1
)(x
2
+ y
2
) (x
n
+ y
n
) = min
_
n

k=1
(x
k
+ y
k
)z
k
_

min
_
n

k=1
x
k
z
k
_
+ min
_
n

k=1
y
k
z
k
_
=
n

x
1
x
2
x
n
+
n

y
1
y
2
y
n
.
11. Use Chebyshevs inequality (6) for sequences a
1
, . . . , a
n
and b
1
, b
2
, . . . , l
n
.
12. We may assume that x
1
x
2
. . . x
n
. But then x
5
1
x
5
2
. . . x
5
n
and
1/x
2
1
1/x
2
2
. . . 1/x
2
n
so these two sequences are oppositely directed. Hence
Chebyshevs Inequality (7) is applicable and gives the result.
13. Since y = ln x is an increasing function, it is sucient to prove the same inequality
for logarithms. Taking logs we will obtain
_
a
a + b + c
_
ln a +
_
b
a + b + c
_
ln b +
_
c
a + b + c
_
ln c
ln
_
a
a + b + c
a +
b
a + b + c
b +
c
a + b + c
c
_
.
This is true. We apply Jensens inequality for the function y = ln x, which is
concave down and

1
=
a
a + b + c
,
2
=
b
a + b + c
,
3
=
c
a + b + c
.
(Note that
1
+
2
+
3
= 1).
14. Take logs on both sides and use concavity of logarithm down.
20
15. The inequality can be rewritten as follows:
(ab + bc + ca)a
2
b
2
c
2
a
8
+ b
8
+ c
8
.
Weighted AM-GM yields
3
8
a
8
+
3
8
b
8
+
2
8
c
8
a
3
b
3
c
2
.
Symmetrising it we obtain the inequality we want.
16. For r = 1 Schurs inequality yields
a(a b)(a c) + b(b c)(b a) + c(c a)(c b) 0,
which can be rewritten as
a
3
+ b
3
+ c
3
+ 3abc a
2
(b + c) + b
2
(a + c) + c
2
(a + b).
17. For all non-negative x
1
, x
2
, x
3
by Muirheads theorem
x
(1,0,0)
x
(
2
3
,
1
3
,0)
x
(
1
3
,
1
3
,
1
3
)
.
Hence
1
3
(x
1
+ x
2
+ x
3
)
1
6
_
3
_
x
2
1
x
2
+
3
_
x
2
1
x
3
+
3
_
x
2
2
x
1
+
3
_
x
2
2
x
3
+
3
_
x
2
3
x
1
+
3
_
x
2
3
x
2
_

x
1
x
2
x
3
or, after the substitution a =
3

x
1
, b =
3

x
2
, c =
3

x
3
,
2(a
3
+ b
3
+ c
3
) a
2
(b + c) + b
2
(a + c) + c
2
(a + b) 6abc.
18. Check that the given proof of Muirheads theorem works.
19. First we homogenise, obtaining
a
4
+ b
4
+ c
4
+ a
3
(b + c) + b
3
(a + c) + c
3
(a + b) 3abc(a + b + c) 0.
We now note that a + b + c is a factor of the RHS and the inequality can be
rewritten as
(a + b + c)(a
3
+ b
3
+ c
3
3abc) 0.
Now we see that for c = (a + b) the second bracket is zero, hence it is divisible
by a + b + c. Finally we nd that the inequality can be written as
(a + b + c)
2
(a
2
+ b
2
+ c
2
ab bc ca) 0
in which form it is obvious.
21
10 Further Exercises
17. Prove that for a, b > 0
n+1

ab
n

a + nb
n + 1
.
18. For positive a, b, c prove that
_
a
3
+ b
3
+ c
3
_
_
1
a
+
1
b
+
1
c
_
(a + b + c)
2
.
19. For positive a, b, c prove that
a + b + c
abc

1
a
2
+
1
b
2
+
1
c
2
.
20. Prove the inequality
a
2
(1 + b
2
) + b
2
(1 + c
2
) + c
2
(1 + a
2
) 6abc.
21. For positive a, b, c prove that
a
3
(b + c) + b
3
(c + a) + c
3
(a + b) 2(a
2
b
2
+ b
2
c
2
+ c
2
a
2
).
22. Let P(x) be a polynomial with positive coecients. Prove that if
P
_
1
x
_

1
P(x)
holds for x = 1, then it holds for all x > 0.
23. (IMO 1978) Let a
1
, . . . , a
n
, . . . be a sequence of pairwise distinct positive integers.
Prove that for all positive integers n
n

k=1
a
k
k
2

n

k=1
1
k
.
24. (Lagranges Identity) Prove that for all positive reals x
1
, . . . , x
n
and y
1
, . . . , y
n
_

x
2
i
__

y
2
i
_

x
i
y
i
_
2
=

i<j
(x
i
y
j
x
j
y
i
)
2
.
22
25. Prove that two triangles with sides a, b, c and a
1
, b
1
, c
1
and semiperimeters p and
p
1
, respectively, are similar if and only if

aa
1
+
_
bb
1
+

cc
1
= 2

pp
1
.
26. (Balkan MO, 84) Prove that, if a
1
, . . . , a
n
are positive numbers whose sum is 1,
then
a
1
2 a
1
+
a
2
2 a
2
+ . . . +
a
n
2 a
n

n
2n 1
.
27. Prove that for all positive a, b, c
a
n
b + c
+
b
n
c + a
+
c
n
a + b

a
n1
+ b
n1
+ c
n1
2
.
28. Prove that for any positive numbers a, b, c
a
b + 2c
+
b
c + 2a
+
c
a + 2b
1.
29. Let a, b, c be positive reals such that a + b + c = 1. Prove that

ab + c +

bc + a +

ca + b 1 +

ab +

bc +

ca.
30. (Russian winter Camp, 98) Prove that for any positive numbers a, b, c, d
a
b + 2c + 3d
+
b
c + 2d + 3a
+
c
d + 2a + 3b
+
d
a + 2b + 3c

2
3
.
31. (APMO, 91) Let a
1
, . . . , a
n
and b
1
, . . . , b
n
be positive numbers such that
a
1
+ a
2
+ + a
n
= b
1
+ b
2
+ + b
n
.
Prove that
a
2
1
a
1
+ b
1
+
a
2
2
a
2
+ b
2
+ . . . +
a
2
n
a
n
+ b
n

a
1
+ a
2
+ . . . + a
n
2
.
23