Inspired Inequalities

Thomas Mildorf
Aqua Lecture, June 15, 2011
Let x1 , . . . , xn be nonnegative real numbers. Define the elementary symmetric polynomials s1 , . . . , sn
by
(x + x1 ) (x + xn ) = xn + + sn1 x + sn . The symmetric averages di are then defined by di = si / ni .
Trick 1 (Symmetric Polynomials) If x1 , . . . , xm are real numbers and m > n, then there exist reals
x01 , . . . , x0m1 such that d0i = di for i = 1, . . . , n. In particular, an inequality in symmetric averages d1 , . . . , dn
need only be proved for the n-variable case.
Theorem 1 (Newton) d2i di+1 di1 .

Theorem 2 (Maclaurin) d1 d2 3 d3 n dn . b
Theorem 3 (Bernoulli) For all r 1 and x 1,
(1 + x)r 1 + xr
A variety of substitutions can be useful when solving inequalities. There is no special prize for using the
given variables. Substitutions can be a particularly handy way to deal with constraints; if a constraint is
unwieldy, change it!
Positive reals x1 , . . . , xn such that the product x1 xn = 1 can be written as xi = yi /yi1 or xi =
yi1 /yi for unconstrained positive reals y1 , . . . , yn .
Positive reals x1 , . . . , xn such that the product x1 xn = 1 can be written as xi = eyi for reals
y1 , . . . , yn with sum 0.
Three positive reals a, b, c which are the sides of a triangle can be written a = x + y, b = y + z, c = z + x
for unconstrained positive reals x, y, z.
Three positive reals a, b, c such that a + b + c = abc are expressible as a = tan A, b = tan B, c = tan C
where ABC is a triangle. (What is the related cotangent substitution?)
Three positive reals a, b, c such that a2 + b2 + c2 + 2abc = 1 are expressible as a = cos(A), b = cos B, c =
cos C where ABC is an acute triangle.

Isolated Fudging

Some contest inequalities are solved by blending all of the terms under consideration together with one of
the standard inequalities. That strategy sometimes fails, in which case the alternative is manipulating each
term by itself (hence the name) so that the resultant sum is simple.
One idea is to give a linear approximation, which can be accomplished in several ways. We can compare
given terms to
k ar
,
ar + br + cr
1

which gives a sum independent of r. (The free variable r is calculated by taking derivatives at an equality
case, and actually can be at most one value, while k is the desired sum.) Equivalently, we can fix a + b + c
and attempt to give a linear approximation. In the following solution, r = 1 can be found by computing


2/3 
2/3



2a
2

1/3 
(2/3)a
=
= 2/3




a b + c
b+c
(a,b,c)=(1,1,1)

(a,b,c)=(1,1,1)

and equating it with

3ar

2r
=
r
a a + br + cr
3
It is then an easy check to see that the choice r = 1 is suitable. Importantly, a complete solution need not,
and should not, show the work behind finding r.
(MOP 2002) a, b, c are positive reals. Show that


2a
b+c

2/3


+

2b
c+a

2/3


+

2c
a+b

2/3
3

Solution. Consider the inequality



2a
b+c

 32

3a
a+b+c

This is equivalent to
s 
2
b+c
3
a+b+c3 a
2
and is a direct consequence of AM-GM on a, b+c
2 , and

b+c
2 .

It follows that

X  2a  3
X
3a
=3

b
+
c
a
+
b+c
cyc
cyc
as desired.

Mixing Variables

The idea behind mixing variables is to blend the given variables into a more convenient form. Examples

x+y
include (x, y) 7 ( x+y
2 , 2 ), (x, y) 7 ( xy, xy), (x, y) 7 (x + y, 0), and (x, y) 7 (xy, 1). In this sense,
mixing variables can be thought of as a variant of smoothing. The requisite computations are often not as
clear as they are for Jensen or Karamata, but fortunately they typically do not include Calculus. Moreover,
that the computations are less deterministic is a consequence of the number of possibilities.
Show that for reals x, y, z, not all of which are positive,
16 2
(x x + 1)(y 2 y + 1)(z 2 z + 1) (xyz)2 xyz + 1
9
Solution Suppose without loss of generality that x y z. Note that x2 x + 1 |x|2 |x| + 1. Thus, if
0 > x, exchanging x and y for x0 = x and y 0 = y decreases the left hand side while fixing the right hand
side. If instead x 0 > y, simply exhanging y for y 0 = y decreases the left hand side while increasing the
right hand side. Thus, we may assume x y 0 z. We now prove a lemma:
Lemma 1 For any reals a 0 b, the following inequality holds:
4 2
(a a + 1)(b2 b + 1) (ab)2 ab + 1
3
2

Proof. Let c = b. The desired inequality is equivalent to

4 2
(a a + 1)(c2 + c + 1) (ac)2 + ac + 1,
3
where a, c 0, but this equivalent inequality is easily realized as the sum of three inequalities:
4 2
(a a + 1)c2 a2 c2
3
4 2
(a a + 1)c ac
3
4 2
(a a + 1) 1
3

(a 2)2 c2 0

(4a2 7a + 4)c 0

(2a 1)2 0.

Now we are done, for by the lemma, we have

4
16 2
(x x + 1)(y 2 y + 1)(z 2 z + 1)
(xz)2 xz + 1 (y 2 y + 1) (xyz)2 xyz + 1,
9
3
where y 0 xz.

Sum of Squares

Sometimes an inequality is so sharp that none of the usual theorems can be used to solve it. Sharpness
clearly greater than Schur, or unusual equality cases are good indicators of this situation. For these problems,
carefully chosen squares, or sharp bounds following from pure algebra, or both, may be useful. The reasoning
is that weighted sums of squares such that both the squares and the weights become small near equality
cases give rise to particularly sharp inequalities.
P
A good goal for three variable inequalities is to manipulate to the form
f (a, b, c)(a b)2 0 for some
reasonable function f. A variety of conditions then demonstrate the inequality.
(Anh-Cuong) a, b, c are positive reals. Show that
p
p
p
a3 + b3 + c3 + 3abc ab 2a2 + 2b2 + bc 2b2 + 2c2 + ca 2c2 + 2a2
q
2
2
2
+2ab+3b2
2
2 2
2
2
2
Solution. Note that 3a 4(a+b)
a +b
2 , which is a consequence of (3a +2ab+3b ) 8(a +b )(a+b) =
a4 4a3 b + 6a2 b2 4ab3 + b4 = (a b)4 0. It therefore suffices to check that
a3 + b3 + c3 + 3abc

ab

cyc

or that
X

a(a b)(a c)

cyc

X  3a2 + 2ab + 3b2

cyc

2(a + b)

3a2 + 2ab + 3b2

2(a + b)
 X
ab(a b)2
ab(a + b) =
2(a + b)
cyc

Consider the identity

2

ar (a b)(a c) =

cyc

(ar + br cr )(a b)2 .

cyc

Taking r = 1, we need only prove

X
cyc

ab
a+bc
a+b

(a b)2 0

This is evident, for we may assume without loss of generality that a b c and note that the desired is the
sum of the three inequalities


ab
(a b)2 0
a+bc
a+b




ac
0
(a c)2 (b c)2
c+ab
a+c




bc
ac
b+ca
(b c)2 + c + a b
(b c)2 0. 
b+c
a+c

Problems
1. Prove the weighted form of Jensens inequality from the definition of a convex function: if f be a
convex function on an interval I and x1 , . . . , xn I, then for any nonnegative reals 1 , . . . , n with
sum 1,
1 f (x1 ) + + n f (xn ) f (1 x1 + + n xn ).
2. Show that the AM-GM inequality follows from Jensens inequality: if a1 , . . . , an are positive reals and
1 , . . . , n are nonnegative reals with sum 1, then
1 a1 + + n an a1 1 ann .
3. Prove H
olders inequalty using only AM-GM: if a1 , . . . , an ; b1 , . . . , bn ; . . . ; z1 , . . . , zn are sequences of
nonnegative reals and a , . . . , z are nonnegative reals with sum 1, then
(a1 + + an )a (b1 + + bn )b (z1 + + zn )z a1 a b1 b z1z + + ana bnb znz .
4. Use H
olders inequality to prove the power-mean inequality: if a1 , . . . , an are positive reals and
1 , . . . , n are nonnegative reals with sum 1, then for nonzero reals r and s with r > s,
1/r

(1 ar1 + + n arn )

1/s

(1 as1 + + n asn )

5. (Titu) Prove that

b
c
d
a
+
+
+
1
3b + c 3c + d 3d + a 3a + b
for all positive real numbers a, b, c, d.
6. Let a1 , a2 , . . . , an be nonnegative reals with a sum of 1. Prove that
a1 a2 + a2 a3 + + an1 an

1
4

7. (USAMO 97/5) Prove that for all positive reals a, b, c,

a3

1
1
1
1
+ 3
+ 3

3
3
3
+ b + abc b + c + abc c + a + abc
abc

8. (Japan 97) Let a, b, c be positive real numbers. Prove that

(b + c a)2
(c + a b)2
(a + b c)2
3
+
+
.
2
2
2
2
(b + c) + a
(c + a) + b
(a + b)2 + c2
5

9. (IMO 00) Let a, b, c be positive real numbers such that abc = 1. Prove that




1
1
1
a1+
b1+
c1+
1.
b
c
a
10. Let a0 , a1 , . . . , an be numbers from the interval (0, /2) such that






+ tan a1
+ + tan an
n 1.
tan a0
4
4
4
Prove that
tan a0 tan a1 tan an nn+1 .
11. (USAMO 01/3) Let a, b, c be nonnegative reals such that
a2 + b2 + c2 + abc = 4
Prove that
0 ab + bc + ca abc 2
12. (Poland 96?) Let a, b, c be real numbers with a + b + c = 1 and a, b, c 3/4. Prove that
a2

a
b
c
9
+ 2
+ 2

.
+1 b +1 c +1
10

13. Show that for all nonnegative reals a, b, c,

p
(a + b + c)3 3(ab + bc + ca) 3(a2 + b2 + c2 )
14. (Michael Rozenberg) Show that for all positive reals a, b, c,
a2
b2
c2
3 a3 + b3 + c3
.
+
+
2
b+c c+a a+b
2 a + b2 + c2
15. (MOP 2006) Show that for nonnegative reals a, b, c,
p
p
(a2 b + b2 c + c2 a)(a2 c + c2 b + b2 a) abc + 3 (a3 + abc)(b3 + abc)(c3 + abc)
16. Show that for nonnegative reals a, b, c,
2a6 + 2b6 + 2c6 + 16a3 b3 + 16b3 c3 + 16c3 a3 9a4 (b2 + c2 ) + 9b4 (c2 + a2 ) + 9c4 (a2 + b2 )
17. Show that for all positive reals a, b, c,
p
3

4a3 + 4b3 +

p
3

4b3 + 4c3 +

p
4a2
4b2
4c2
3
4c3 + 4a3
+
+
a+b b+c c+a

18. Show that for all positive numbers x1 , . . . , xn ,

x21

x32
x3n
x31
x1 + + xn
+ 2
+ + 2

2
2
+ x1 x2 + x2
x2 + x2 x3 + x3
xn + xn x1 + x21
3

19. Prove that for all positive reals a, b, c, d,

a4 b + b4 c + c4 d + d4 a abcd(a + b + c + d)
5

20. (USAMO 2003) Let a, b, c be positive real numbers. Prove that

(2a + b + c)2
(2b + c + a)2
(2c + a + b)2
+
+
8.
2a2 + (b + c)2
2b2 + (c + a)2
2c2 + (a + b)2
21. Prove that for all positive reals a, b, c,
8abc
a2 + b2 + c2
+
2
ab + bc + ca (a + b)(b + c)(c + a)
22. (Bulgaria 97) 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
23. (Darij Grinberg) Show that for any positive reals a, b, c,
(b + c)2
(c + a)2
(a + b)2
+
+
6
a2 + bc
b2 + ca
c2 + ab
24. (IMO 99) Let n 2 be a fixed integer. Find the smallest constant C such that for all nonnegative
reals x1 , . . . , xn ,

4
X
X
xi xj (x2i + x2j ) C
xi .
1i<jn

1in

Determine when equality occurs.

25. (Vietnam 98) Let x1 , . . . , xn be positive reals such that
1
1
1
1
+
+ +
=
x1 + 1998 x2 + 1998
xn + 1998
1998
Prove that

x1 x2 xn
1998
n1

26. (Vietnam 96) Let a, b, c, d be four nonnegative real numbers satisfying the condition
2(ab + ac + ad + bc + bd + cd) + abc + bcd + acd + bcd = 16.
Prove that
a+b+c+d

2
(ab + ac + ad + bc + bd + cd) ,
3

and determine when equality holds.

27. (USAMO 00/6) Let n 2 be an integer and S = {1, 2, . . . , n}. Show that for all nonnegative reals
a1 , a2 , . . . , an , b1 , b2 , . . . , bn ,
X
X
min{ai aj , bi bj }
min{ai bj , aj bi }
i,jS

i,jS

28. (George Tsintifas) Prove that for any a, b, c, d > 0 we have the inequality
(a + b)3 (b + c)3 (c + d)3 (d + a)3 16a2 b2 c2 d2 (a + b + c + d)4 .
6

29. (MOP 2003) Show that for positive reals x1 , . . . , xn satisfying

x1 + xn =

1
1
+ +
x1
xn

the following inequality holds:

1
1
+ +
1.
n 1 + x1
n 1 + xn
30. (???) Let n be an integer greater than 2, and let x1 , x2 , . . . , xn be positive real numbers. Prove that
x1
x2
xn1
xn
5n
+
+ ... +
+
>
x2 + x3
x3 + x4
xn + x1
x1 + x2
12

Homework
1. Prove the symmetric polynomials trick by considering the polynomial P (x) = (x x1 ) (x xm ) and
its derivative.
2. Using the symmetric polynomials trick, prove the Newton inequalities.
3. Prove the Maclaurin inequalities using the symmetric polynomials trick.
4. Suppose that a, b, c, Sa , Sb , Sc are arbitrary reals such that |Sa |, |Sb |, |Sc | satisfy the triangle inequality.
Prove that
Sa2 (a b)(a c) + Sb2 (b c)(b a) + Sc2 (c a)(c b) 0.
5. Let a, b, c be arbitrary reals. Show that the inequality
Sa (b c)2 + Sb (c a)2 + Sc (a b)2 0
holds for each of the following situations:
Sa , Sb , Sc 0,
a b c and Sb , Sb + Sc , Sb + Sa 0,
a b c and Sa , Sc , Sa + 2Sb , Sc + 2Sb 0,
a b c and Sb , Sc 0, and a2 Sb + b2 Sa 0.