Académique Documents
Professionnel Documents
Culture Documents
C j
L
MATHEMMATICAL
.M
INEQUA
Q ALITIES
D
A
Volume 1
PI
M
LY
SYMM
METRIC
O
POLYNOMIAL INEQUALITIES
POLYNOMIAL
.M
W
W
W
L
Vasile Cîrtoaje
D
.M
Vasile Cîrtoaje is a Professor at the Department of Automatic Control and Computers
A
from Petroleum-Gas University of Ploiesti, Romania, where he teaches university courses
such as Control System Theory and Digital Control Systems.
PI
Since 1970, he published many mathematical problems, solutions and articles in the
M
Romanian journals Gazeta Matematica-B, Gazeta Matematica-A and Mathematical Re-
view of Timisoara. In addition, from 2000 to present, Vasile Cîrtoaje has published
LY
many interesting problems and articles in Art of Problem Solving website, Mathemati-
cal Reflections, Crux with Mayhem, Journal of Inequalities and Applications, Journal
O
He collaborated with Titu Andreescu, Gabriel Dospinescu and Mircea Lascu in writing
W
the book Old and New Inequalities, with Vo Quoc Ba Can and Tran Quoc Anh in writ-
ing the book Inequalities with Beautiful Solution, and he wrote his own books Algebraic
W
Inequalities - Old and New Methods and Mathematical Inequalities (Volume 1 ... 4).
Notice that Vasile Cîrtoaje is the author of some well-known results and strong meth-
ods for proving and creating discrete inequalities, such as:
- Half convex function method (HCF method) for Jensen type discrete inequalities;
- Partial convex function method (PCF method) for Jensen type discrete inequalities;
- Jensen type discrete inequalities with ordered variables;
- Equal variable method (EV method) for real or nonnegative variables;
- Arithmetic compensation method (AC method);
- Best lower and upper bounds for Jensen’s inequality;
- Necessary and sufficient conditions for symmetric homogeneous polynomial inequal-
ities of degree six in real variables;
i
ABOUT THE AUTHOR
L
gree four in real or nonnegative variables;
.M
- Inequalities with power-exponential functions.
D
A
PI
M
LY
O
.M
W
W
W
Foreword
L
well-known for his excellent creations in the mathematical inequality field, ever since
.M
the time when he was student in high school (in Breaza city, Prahova Valley). As a
student (quite some time ago, oh yes!), I was already familiar with the name of Vasile
D
Cîrtoaje. For me, and many others of my age, it is the name of someone who helped
me to grow in mathematics, even though I never met him face to face. It is a name
A
synonymous to hard and beautiful problems involving inequalities. When you say Vasile
PI
Cîrtoaje (Vasc username on the site Art of Problem Solving), you say inequalities. I
remember how happy I was when I could manage to solve one of the problems proposed
M
by professor Cîrtoaje in Gazeta Matematica or Revista Matematica Timisoara.
The first three volumes of this book are a great opportunity to see and know many
LY
old and new elementary methods for solving mathematical inequalities: Volume 1 -
Symmetric polynomial inequalities (in real variables and nonnegative real variables),
O
Volume 2 - Symmetric rational and nonrational inequalities, Volume 3 - Cyclic and non-
.M
cyclic inequalities. As a rule, the inequalities from each section of these volumes are in-
creasingly ordered by the number of variables: two, three, four, five, six and n-variables.
The last volume (Volume 4) contains new beautiful and efficient original methods for
W
creating and solving inequalities: half or partially convex function method - for Jensen’s
type inequalities, equal variable method - for nonnegative or real variables, arithmetic
W
eight in nonnegative or real variables, methods involving either strong or necessary and
sufficient conditions - for cyclic homogeneous inequalities of degree four in real or non-
negative variables and so on.
Many problems, the majority I would say, are made up by the author himself. The
chapters and volumes are relatively independent, and you can open the book somewhere
to solve an inequality or only read its solution. If you carefully make a thorough study
of the book, then you will find that your skills in solving inequalities are considerably
improved.
The book is a rich and meaningful collection of more than 1000 beautiful inequalities,
hints, solutions and methods, some of them being posted in the last ten years by the
author and other inventive mathematicians on Art of Problem Solving website (Vo Quoc
iii
FOREWORD
Ba Can, Pham Kim Hung, Michael Rozenberg, Nguyen Van Quy, Gabriel Dospinescu,
Darij Grinberg, Pham Huu Duc, Tran Quoc Anh, Le Huu Dien Khue, Marius Stanean,
Cezar Lupu, Nguyen Anh Tuan, Pham Van Thuan, Bin Zhao, Ji Chen etc.)
Most inequalities and methods are old and recent own creations of the author. Among
these, I would like to point out the following inequalities:
L
.M
4
X 1
≥ 1, a1 , a2 , a3 , a4 > 0, a1 a2 a3 a4 = 1;
i=1
1 + ai + ai2 + ai3
D
a1 a2 an
+ + ··· + ≥ 1, a1 , a2 , · · · , an ≥ 0;
A
a1 + (n − 1)a2 a2 + (n − 1)a3 an + (n − 1)a1
a ea + b e b ≥ a e b + b ea , a, b > 0,PI e ≈ 2.7182818.
M
a3b + b3a ≤ 2, a, b ≥ 0, a + b = 2.
LY
The book represents a rich source of beautiful, serious and profound mathematics,
dealing with classical and new approaches and techniques which help the reader to de-
velop his inequality-solving skills, intuition and creativity. As a result, it is suitable for
O
a wide variety of audiences; high school students and teachers, college and university
.M
thought and analysis, making the book a rich and rewarding source for anyone inter-
ested in Olympiad-type problems and in the development of the inequality field. Many
W
problems and methods can be used in a classroom for advanced high school students as
group projects.
W
What makes this book so attractive? The answer is simple: the great number of in-
equalities, their quality and freshness, as well as the new approaches and methods for
solving mathematical inequalities. Nevertheless, you will find this book to be delight-
ful, inspired, original and enjoyable. Certainly, any interested reader will remark the
tenacity, enthusiasm and ability of the author in creating and solving nice and difficult
inequalities. And this book is neither more, nor less than a Work of a Master.
The author must be congratulated for publishing such interesting and original in-
equality book. I highly recommend it.
Marian Tetiva
Contents
L
About the author i
.M
Foreword iii
D
1 Some Classic and New Inequalities and Methods 1
A
2 Symmetric Polynomial Inequalities in Real Variables 19
PI
2.1 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2 Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
M
3 Symmetric Polynomial Inequalities in Nonnegative Variables 149
LY
4 Bibliography 309
.M
W
W
W
v
vi Vasile Cîrtoaje
L
D
.M
A
PI
M
LY
O
.M
W
W
W
Chapter 1
L
and Methods
D
.M
A
1. AM-GM (ARITHMETIC MEAN-GEOMETRIC MEAN) INEQUALITY
If a1 , a2 , · · · , an are nonnegative real numbers, then PI
M
a1 + a2 + · · · + an ≥ n n a1 a2 · · · an ,
p
LY
p1 + p2 + · · · + pn = 1.
W
W
p1 a1 + p2 a2 + · · · + pn an ≥ a1 1 a2 2 · · · anpn ,
1 1 1
(a1 + a2 + · · · + an ) + + ··· + ≥ n2 ,
a1 a2 an
1
2 Vasile Cîrtoaje
L
v
.M
t a1 + a22 + · · · + an2
u 2
a1 + a2 + · · · + an p n
≥ ≥ n a1 a2 · · · an ≥ .
n n 1 1 1
+ + ··· +
D
a1 a2 an
A
5. BERNOULLI’S INEQUALITY PI
For any real number x ≥ −1, we have
M
a) (1 + x) r ≥ 1 + r x for r ≥ 1 and r ≤ 0;
b) (1 + x) r ≤ 1 + r x for 0 ≤ r ≤ 1.
LY
(1 + a1 )(1 + a2 ) · · · (1 + an ) ≥ 1 + a1 + a2 + · · · + an .
.M
W
6. SCHUR’S INEQUALITY
W
For any nonnegative real numbers a, b, c and any positive number k, the inequality
holds
W
with equality for a = b = c, and for a = 0 and b = c (or any cyclic permutation).
For k = 1, we get the third degree Schur’s inequality, which can be rewritten as follows
For k = 2, we get the fourth degree Schur’s inequality, which holds for any real numbers
a, b, c, and can be rewritten as follows
L
X
.M
(a − mb)(a − mc)(a − b)(a − c) ≥ 0,
where the equality holds for a = b = c, and for a/m = b = c (or any cyclic permutation).
D
This inequality is equivalent to
A
X X X X
a4 + m(m + 2) a2 b2 + (1 − m2 )a bc a ≥ (m + 1) a b(a2 + b2 )
PI
and
M
X
(b − c)2 (b + c − a − ma)2 ≥ 0.
LY
A more general result is given by the following theorem (Vasile Cirtoaje, 2004).
Theorem. Let
O
X X X X
f4 (a, b, c) = a4 + α a2 b2 + β a bc a−γ a b(a2 + b2 ),
.M
1 + α + β = 2γ.
W
1 + α ≥ γ2 ;
α ≥ (γ − 1) max{2, γ + 1}.
7. CAUCHY-SCHWARZ INEQUALITY
8. HÖLDER’S INEQUALITY
If x i j (i = 1, 2, · · · , m; j = 1, 2, · · · n) are nonnegative real numbers, then
! v !m
m
Y n
X n uY
X m
m
xi j ≥ .
t
xi j
i=1 j=1 j=1 i=1
9. CHEBYSHEV’S INEQUALITY
L
Let a1 ≥ a2 ≥ · · · ≥ an be real numbers.
.M
a) If b1 ≥ b2 ≥ · · · bn , then
D
n n n
A
X X X
n ai bi ≥ ai bi ;
b) If b1 ≤ b2 ≤ · · · ≤ bn , then
i=1 i=1
PI
i=1
M
n n n
LY
X X X
n ai bi ≤ ai bi .
i=1 i=1 i=1
O
For any real number k ≥ 1 and any positive real numbers a1 , a2 , · · · , an and b1 , b2 , · · · , bn ,
the following inequalities hold:
W
1
k k k
n n n
W
X 1 X X
aik + bi k k
≥ ai + bi ;
W
1
k k k k
n n n n
X 1 X X X
aik + bik + ci
k k
≥ ai + bi + ci .
i=1 i=1 i=1 i=1
and
n(a1 b1 + a2 b2 + · · · + an bn ) ≥ (a1 + a2 + · · · + an )(b1 + b2 + · · · + bn ).
(2) If a1 , a2 , · · · , an is decreasing and b1 , b2 , · · · , bn is increasing, then
and
n(a1 b1 + a2 b2 + · · · + an bn ) ≤ (a1 + a2 + · · · + an )(b1 + b2 + · · · + bn ).
(3) Let b1 , b2 , · · · , bn and c1 , c2 , · · · , cn be two real sequences such that
b1 + · · · + bk ≥ c1 + · · · + ck , k = 1, 2, · · · , n.
L
.M
If a1 ≥ a2 ≥ · · · ≥ an ≥ 0, then
a1 b1 + a2 b2 + · · · + an bn ≥ a1 c1 + a2 c2 + · · · + an cn .
D
A
Notice that all these inequalities follow immediately from the identity
X n n
X i
X i
X
!PI
ai (bi − ci ) = (ai − ai+1 ) bj − cj ,
M
i=1 i=1 j=1 j=1
LY
where an+1 = 0.
O
and
W
v X
u
u
u a i1 a i2 · · · a i k
k 1≤i1 <···<ik ≤n
Sk = u
u
.
t n
k
f (αx + β y) ≤ α f (x) + β f ( y)
6 Vasile Cîrtoaje
L
.M
Jensen’s inequality. Let p1 , p2 , . . . , pn be positive real numbers. If f is a convex
function on a real interval I, then for any a1 , a2 , . . . , an ∈ I, the inequality holds
D
p1 f (a1 ) + p2 f (a2 ) + · · · + pn f (an ) p1 a1 + p2 a2 + · · · + pn an
≥f .
A
p1 + p2 + · · · + p n p1 + p2 + · · · + p n
For p1 = p2 = · · · = pn , Jensen’s inequality becomes PI
a + a + ··· + a
M
1 2 n
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f .
n
LY
Based on the following three theorems, we can extend this form of Jensen’s in-
equality to half or partially convex functions.
O
defined on a real interval I and convex on Iu≥s or Iu≤s , where s ∈ I. The inequality
a + a + ··· + a
1 2 n
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f
W
n
holds for all a1 , a2 , . . . , an ∈ I satisfying a1 + a2 + . . . + an = ns if and only if
W
f (x) + (n − 1) f ( y) ≥ n f (s)
W
Half Convex Function Theorem for Ordered Variables (Vasile Cirtoaje, 2007). Let
f (u) be a function defined on a real interval I and convex on Iu≥s / Iu≤s , where s ∈ I. The
inequality a + a + ··· + a
1 2 n
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f
n
holds for all a1 , a2 , . . . , an ∈ I such that a1 + a2 + · · · + an = ns and at least n − m of
a1 , a2 , . . . , an are smaller/greater than or equal to s if and only if
f (x) + m f ( y) ≥ (1 + m) f (s)
Some Classic and New Inequalities and Methods 7
The inequality
a + a + ··· + a
1 2 n
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f
n
L
holds for all a1 , a2 , · · · , an ∈ I satisfying a1 + a2 + · · · + an = ns if and only if
.M
f (x) + (n − 1) f ( y) ≥ n f (s)
D
for all x, y ∈ I such that x ≤ s ≤ y and x + (n − 1) y = ns.
A
PI
Left Partially Convex Function-Theorem (Vasile Cirtoaje, 2012). Let f be a func-
tion defined on a real interval I and convex on [s0 , s], where s0 , s ∈ I, s0 < s. In addition,
M
f is increasing on Iu≥s0 and satisfies
LY
The inequality
.M
a + a + ··· + a
1 2 n
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f
n
W
f (x) + (n − 1) f ( y) ≥ n f (s)
W
Right Partially Convex Function Theorem for Ordered Variables (Vasile Cirtoaje,
2012). Let f be a function defined on a real interval I and convex on [s, s0 ], where s, s0 ∈ I,
s < s0 . In addition, f is decreasing on Iu≤s0 and
The inequality
a + a + ··· + a
1 2 n
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f
n
8 Vasile Cîrtoaje
f (x) + m f ( y) ≥ (1 + m) f (s)
Left Partially Convex Function Theorem for Ordered Variables (Vasile Cirtoaje,
2012). Let f be a function defined on a real interval I and convex on [s0 , s], where s0 , s ∈ I,
s0 < s. In addition, f is increasing on Iu≥s0 and satisfies
L
u∈I
.M
The inequality
a + a + ··· + a
D
1 2 n
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ n f
n
A
holds for all x 1 , x 2 , · · · , x n ∈ I such that a1 + a2 + · · · + an = ns and at least n − m of
PI
a1 , a2 , . . . , an are greater than or equal to s if and only if if and only if
M
f (x) + m f ( y) ≥ (1 + m)n f (s)
LY
f (x) + m f ( y) ≥ (1 + m) f (s),
where
g(x) − g( y) f (u) − f (s)
h(x, y) = g(u) =
W
, .
x−y u−s
The following theorem is also useful to prove some symmetric inequalities.
Left Convex-Right Concave Function Theorem (Vasile Cirtoaje, 2004). Let a < c
be real numbers, let f be a continuous function on I = [a, ∞), strictly convex on [a, c]
and strictly concave on [c, ∞), and let
If a1 , a2 , . . . , an ∈ I such that
a1 + a2 + · · · + an = S = const ant,
Some Classic and New Inequalities and Methods 9
then
(a) E is minimum for a1 = a2 = · · · = an−1 ≤ an ;
(b) E is maximum for either a1 = a or a < a1 ≤ a2 = · · · = an .
On the other hand, it is known the following result concerning the best upper bound
of Jensen’s difference.
Best Upper Bound of Jensen’s Difference-Theorem (Vasile Cirtoaje, 1989). Let
p1 , p2 , . . . , pn be fixed positive real numbers, and let f be a convex function on a closed
interval I = [a, b]. If a1 , a2 , · · · , an ∈ I, then Jensen’s difference
p1 f (a1 ) + p2 f (a2 ) + · · · + pn f (an ) p1 a1 + p2 a2 + · · · + pn an
D= −f
p1 + p2 + · · · + p n p1 + p2 + · · · + pn
L
.M
is maximum when some of ai are equal to a, and the others ai are equal to b; that is, when
all ai ∈ {a, b}.
D
14. KARAMATA’S MAJORIZATION INEQUALITY
A
−
→
We say that a vector A = (a1 , a2 , . . . , an ) with a1 ≥ a2 ≥ · · · ≥ an majorizes a vector
−
→ PI
B = (b1 , b2 , . . . , bn ) with b1 ≥ b2 ≥ · · · ≥ bn , and write it as
M
−
→ − →
A ≥ B,
LY
if
a 1 ≥ b1 ,
O
a1 + a2 ≥ b1 + b2 ,
·····················
.M
a1 + a2 + · · · + an−1 ≥ b1 + b2 + · · · + bn−1 ,
a1 + a2 + · · · + an = b1 + b2 + · · · + bn .
W
A = (a1 , a2 , . . . , an ), ai ∈ I,
W
then
f (a1 ) + f (a2 ) + · · · + f (an ) ≥ f (b1 ) + f (b2 ) + · · · + f (bn ).
p = a + b + c, q = a b + bc + ca, r = a bc,
L
Æ p
s = p2 − 3q = a2 + b2 + c 2 − a b − bc − ca.
.M
From the identity
D
27(a − b)2 (b − c)2 (c − a)2 = 4(p2 − 3q)3 − (2p3 − 9pq + 27r)2 ,
A
it follows that
which is equivalent to
≤r≤ .
27 27
.M
Therefore, for constant p and q, the product r is minimal and maximal when two of
a, b, c are equal.
W
Theorem.
(a) The inequality f4 (a, b, c) ≥ 0 holds for all real numbers a, b, c if and only if
f4 (a, 1, 1) ≥ 0 for all real a;
(b) The inequality f n (a, b, c) ≥ 0 holds for all a, b, c ≥ 0 if and only if f n (a, 1, 1) ≥ 0
and f n (0, b, c) ≥ 0 for all a, b, c ≥ 0.
Any sixth degree symmetric homogeneous polynomial f6 (a, b, c) can be written in the
form
f6 (a, b, c) = Ar 2 + B(p, q)r + C(p, q),
Some Classic and New Inequalities and Methods 11
p = a + b + c, q = a b + bc + ca, r = a bc.
L
2008). This method consists in finding some suitable real numbers B, C and D such
that the following sharper inequality holds
.M
2
q2
3
D
f6 (a, b, c) ≥ A r + Bp + C pq + D .
p
A
Because the function g6 defined by
PI 2
q2
M
3
g6 (a, b, c) = f6 (a, b, c) − A r + Bp + C pq + D
p
LY
has the highest coefficient A1 = 0, we can prove the inequality g6 (a, b, c) ≥ 0 using
O
Theorem above.
Notice that sometimes it is useful to break the problem into two parts, p2 ≤ ξq and
.M
The Equal Variable Theorem (EV-Theorem) for nonnegative real variables has the fol-
lowing statement (Vasile Cirtoaje, 2005).
W
Sn = f (x 1 ) + f (x 2 ) + · · · + f (x n ).
x 1 = x 2 = · · · = x n−1 ≤ x n ,
L
for
x 1 = x 2 = · · · = x n−1 ≤ x n ;
.M
(3) If k > 0 and either f is continuous at x = 0 or f (0+ ) = ∞, then Sn is minimum
D
for
x 1 = · · · = x j−1 = 0, x j+1 = · · · = x n , j ∈ {1, 2, . . . , n}.
A
For f (x) = x m , we get the following corollary.
PI
M
EV-COROLLARY (for nonnegative variables). Let a1 , a2 , · · · , an (n ≥ 3) be fixed
LY
x 1 + x 2 + · · · + x n = a1 + a2 + · · · + an ,
O
and let
Sn = x 1m + x 2m + · · · + x nm .
W
x 1 = x 2 = · · · = x n−1 ≤ x n ,
x 1 = x 2 = · · · = x n−1 ≤ x n ,
x 1 = x 2 = · · · = x n−1 ≤ x n ,
x 1 = x 2 = · · · = x n−1 ≤ x n ;
L
(c) If m ∈ (k, 1), then Sn is maximum for
.M
x 1 = · · · = x j−1 = 0, x j+1 = · · · = x n , j ∈ {1, 2, . . . , n}.
D
Case 3 : k > 1.
A
(a) If m ∈ (0, 1) ∪ (k, ∞), then Sn is maximum for
PI
x 1 = x 2 = · · · = x n−1 ≤ x n ,
M
and is minimum for
LY
0 ≤ x 1 = x 2 = · · · = x n−1 ≤ x n ;
x 1 = · · · = x j−1 = 0, x j+1 = · · · = x n ,
W
j ∈ {1, 2, . . . , n}.
W
The Equal Variable Theorem (EV-Theorem) for real variables has the following
statement (Vasile Cirtoaje, 2012).
EV-Theorem (for real variables). Let a1 , a2 , · · · , an (n ≥ 3) be fixed real numbers,
let x 1 ≤ x 2 ≤ · · · ≤ x n be real variables such that
x 1 + x 2 + · · · + x n = a1 + a2 + · · · + an ,
p
g(x) = f 0
k−1
x
is strictly convex on R. Then, the sum
Sn = f (x 1 ) + f (x 2 ) + · · · + f (x n )
L
Cirtoaje, 2002).
.M
AC-THEOREM. Let s > 0 and let F be a symmetric continuous function on the compact
set in Rn
D
A
S = {(x 1 , x 2 , · · · , x n ) : x 1 + x 2 + · · · + x n = s, x i ≥ 0, i = 1, 2, · · · , n}.
If PI
F (x 1 , x 2 , x 3 , · · · , x n ) ≥
M
n x + x x + x o
1 2 1 2
≥ min F , , x 3 , · · · , x n , F (0, x 1 + x 2 , x 3 , · · · , x n )
LY
2 2
for all (x 1 , x 2 , · · · , x n ) ∈ S, then F (x 1 , x 2 , x 3 , · · · , x n ) is minimal when
O
s
x1 = x2 = · · · = x k = , x k+1 = · · · = x n = 0;
.M
k
that is, s s
F (x 1 , x 2 , x 3 , · · · , x n ) ≥ min F
W
, · · · , , 0, · · · , 0
1≤k≤n k k
for all (x 1 , x 2 , · · · , x n ) ∈ S.
W
W
Notice that if
x + x x + x
1 2 1 2
F (x 1 , x 2 , x 3 , · · · , x n ) < F , , x3, · · · , x n
2 2
involves
F (x 1 , x 2 , x 3 , · · · , x n ) ≥ F (0, x 1 + x 2 , x 3 , · · · , x n ),
then the hypothesis
F (x 1 , x 2 , x 3 , · · · , x n ) ≥
n x + x x + x o
1 2 1 2
≥ min F , , x 3 , · · · , x n , F (0, x 1 + x 2 , x 3 , · · · , x n )
2 2
is satisfied.
Some Classic and New Inequalities and Methods 15
L
.M
Theorem 0 (Vasile Cirtoaje, 2007). Let
X X X X X
f4 (a, b, c) = a4 + A a2 b2 + Ba bc a+C a3 b + D a b3 ,
D
where A, B, C, D are real constants such that 1 + A + B + C + D = 0. The inequality
A
f4 (a, b, c) ≥ 0 holds for all real numbers a, b, c if and only if
3(1 + A) ≥ C 2 + C D + D2 .
PI
M
Notice that
LY
4 C−D 2 4
f4 (a, b, c) = (U + V + C + D)2 + 3 U − V + + (3 + 3A − C 2 − C D − D2 ),
O
S 3 3
where
.M
a3 b − a2 bc a b3 − a2 bc
P P P P
X X
2 2 2
S= a b − a bc, U= , V= .
S S
W
L
X X X X X
f4 (a, b, c) = a4 + A a2 b2 + Ba bc a+C a3 b + D a b3 ,
.M
where A, B, C, D are real constants.
The following Theorem 2 states the necessary and sufficient conditions that f4 (a, b, c) ≥
D
0 for all real numbers a, b, c.
A
Theorem 2 (Vasile Cirtoaje and Yuanzhe Zhou, 2011). The inequality f4 (a, b, c) ≥ 0
PI
holds for all real numbers a, b, c if and only if g4 (t) ≥ 0 for all t ≥ 0, where
M
g4 (t) = 3(2 + A − C − D)t 4 − F t 3 + 3(4 − B + C + D)t 2 + 1 + A + B + C + D,
LY
Æ
F= 27(C − D)2 + E 2 , E = 8 − 4A + 2B − C − D.
Note that in the special case f4 (1, 1, 1) = 0 (when 1 + A + B + C + D = 0), Theorem 1
O
The following Theorem 3 states some strong sufficient conditions that f4 (a, b, c) ≥ 0
for all real numbers a, b, c.
Theorem 3 (Vasile Cirtoaje and Yuanzhe Zhou, 2012). The inequality f4 (a, b, c) ≥ 0
W
holds for all real numbers a, b, c if the following two conditions are satisfied:
W
(a) 1 + A + B + C + D ≥ 0; p p
(b) there exists a real number t ∈ (− 3, 3) such that f (t) ≥ 0, where
W
2E 3
g3 (t) = t + 3t 2 − 1.
F
For F = 0, the inequality f4 (a, b, c) ≥ 0 holds for all a, b, c ≥ 0 if and only if g4 (t) ≥ 0
for all t ∈ [0, 1].
For F 6= 0, the inequality f4 (a, b, c) ≥ 0 holds for all a, b, c ≥ 0 if and only if the
following two conditions are satisfied:
(a) g4 (t) ≥ 0 for all t ∈ [0, t 1 ], where t 1 ∈ [1/2, 1] such that g3 (t 1 ) = 0;
(b) f4 (a, 1, 0) ≥ 0 for all a ≥ 0.
The following Theorem 5 states some strong sufficient conditions that f4 (a, b, c) ≥ 0
for all a, b, c ≥ 0.
L
Theorem 5 (Vasile Cirtoaje and Yuanzhe Zhou, 2013). The inequality f4 (a, b, c) ≥ 0
.M
holds for all a, b, c ≥ 0 if
1 + A+ B + C + D ≥ 0
D
and one of the following two conditions is satisfied:
A
(a) 3(1 + A) ≥ C 2 + C D + D2 ;
(b) 3(1 + A) < C 2 + C D + D2 , and there exists t ≥ 0 such that
PI
Æ
(C + 2D)t 2 + 6t + 2C + D ≥ 2 (t 4 + t 2 + 1)(C 2 + C D + D2 − 3 − 3A).
M
LY
O
.M
W
W
W
18 Vasile Cîrtoaje
L
D
.M
A
PI
M
LY
O
.M
W
W
W
Chapter 2
L
in Real Variables
D
.M
A
2.1 Applications
PI
2.1. Let a, b, c, d be real numbers such that a2 + b2 + c 2 + d 2 = 9. Prove that
M
a3 + b3 + c 3 + d 3 ≤ 27.
LY
O
19
20 Vasile Cîrtoaje
L
.M
2(a2 + b2 )(b2 + c 2 )(c 2 + a2 ) ≥ [a b(a + b) + bc(b + c) + ca(c + a) − 2a bc]2 .
D
2.9. If a, b, c are real numbers, then
A
PI
(a2 + 1)(b2 + 1)(c 2 + 1) ≥ 2(a b + bc + ca).
M
2.10. If a, b, c are real numbers, then
LY
5
(a2 + 1)(b2 + 1)(c 2 + 1) ≥ (a + b + c + 1)2 .
O
16
.M
1 1 1
4 a+ b+ c+ ≥ 9(a + b + c).
a b c
L
2.16. If a, b, c are real numbers, then
.M
(a) (a2 + 2bc)(b2 + 2ca)(c 2 + 2a b) ≤ (a2 + b2 + c 2 )(a b + bc + ca)2 ;
D
(b) (2a2 + bc)(2b2 + ca)(2c 2 + a b) ≤ (a + b + c)2 (a2 b2 + b2 c 2 + c 2 a2 ).
A
PI
2.17. If a, b, c are real numbers such that a b + bc + ca ≥ 0, then
M
27(a2 + 2bc)(b2 + 2ca)(c 2 + 2a b) ≤ (a + b + c)6 .
LY
O
3(a4 + b4 + c 4 ) + a2 + b2 + c 2 + 6 ≥ 6(a3 + b3 + c 3 ).
W
3 1 1 1
2 2 2
a + b +c +6≥ a+b+c+ + + .
2 a b c
22 Vasile Cîrtoaje
L
2.24. If a, b, c are real numbers such that a bc ≥ 0, then
.M
a2 + b2 + c 2 + 2a bc + 4 ≥ 2(a + b + c) + a b + bc + ca.
D
A
2.25. Let a, b, c be real numbers such that a + b + c = 3.
(a) If a, b, c ≥ −3, then
PI
M
1 1 1 1 1 1
2
+ 2+ 2≥ + + .
a b c a b c
LY
1−a 1− b 1−c
+ + ≥ 0.
(1 + a)2 (1 + b)2 (1 + c)2
.M
W
L
.M
1
a4 (b − c)2 + b4 (c − a)2 + c 4 (a − b)2 ≥ (a − b)2 (b − c)2 (c − a)2 .
2
D
A
2.32. If a, b, c are real numbers, then
3
a2 (b2 − c 2 )2 + b2 (c 2 − a2 )2 + c 2 (a2 − b2 )2 ≥ (a − b)2 (b − c)2 (c − a)2 .
8
.M
(a)
3 2
W
2.36. If a, b, c are real numbers, not all of the same sign, then
L
2.39. If a, b, c are real numbers, then
.M
9(1 + a4 )(1 + b4 )(1 + c 4 ) ≥ 8(1 + a bc + a2 b2 c 2 )2 .
D
A
2.40. If a, b, c are real numbers, then
PI
2(1 + a2 )(1 + b2 )(1 + c 2 ) ≥ (1 + a)(1 + b)(1 + c)(1 + a bc).
M
LY
X b2 − bc + c 2 X a2 X X 1
W
+ 2 ≥ a .
a2 bc a
W
a bc − (b + c − a)(c + a − b)(a + b − c) ≤ 1;
a bc − (b + c − a)(c + a − b)(a + b − c) ≤ 4.
Symmetric Polynomial Inequalities in Real Variables 25
L
2.45. Let a, b, c be real numbers such that
.M
a b + bc + ca = a bc + 2.
D
Prove that p
a2 + b2 + c 2 − 3 ≥ (2 + 3)(a + b + c − 3).
A
PI
M
2.46. Let a, b, c be real numbers such that
LY
Prove that
O
(a + b)(b + c)(c + a) = 5.
W
Prove that
W
a + b + c = a3 + b3 + c 3 = 2.
5
/
Prove that a, b, c ∈ ,2 .
4
L
.M
2.51. If a, b, c are real numbers, then
D
A
2.52. If a, b, c are real numbers, then PI
M
X (a − b)2 (b − c)2 (c − a)2
a2 (a − b)(a − c) ≥ .
a2 + b2 + c 2 + a b + bc + ca
LY
O
a + b + c = p, a b + bc + ca = q,
where p and q are fixed real numbers satisfying p2 ≥ 3q. Prove that the product
W
r = a bc
W
a bc + 10 ≥ 2(a + b + c).
a + b + c + a bc = 4,
then
a2 + b2 + c 2 + 3 ≥ 2(a b + bc + ca).
L
.M
2.58. If a, b, c are real numbers such that
D
a b + bc + ca = 3a bc,
A
then PI
4(a2 + b2 + c 2 ) + 9 ≥ 7(a b + bc + ca).
M
LY
2.60. Let f4 (a, b, c) be a symmetric homogeneous polynomial of degree four. Prove that
W
the inequality f4 (a, b, c) ≥ 0 holds for all real numbers a, b, c if and only if f4 (a, 1, 1) ≥ 0
W
3(a4 + b4 + c 4 ) + 33 ≥ 14(a2 + b2 + c 2 ).
28 Vasile Cîrtoaje
1 + α + β = 2γ.
The inequality
X X X X
L
a4 + α a2 b2 + β a bc a≥γ a b(a2 + b2 )
.M
holds for any real numbers a, b, c if and only if
D
1 + α ≥ γ2 .
A
PI
2.65. If a, b, c are real numbers such that a2 + b2 + c 2 = 2, then
M
a b(a2 − a b + b2 − c 2 ) + bc(b2 − bc + c 2 − a2 ) + ca(c 2 − ca + a2 − b2 ) ≤ 1.
LY
O
4 4
(a + b)4 + (b + c)4 + (c + a)4 ≥ (a + b4 + c 4 ).
7
W
p = a + b + c, q = a b + bc + ca, r = a bc.
W
Prove that
p2 + q2 − pq
(3 − p)r + ≥ q.
3
1 1 1 1 1 1 1
a + − 1 b + − 1 c + − 1 + 2 ≥ (a + b + c) + + .
a b c 3 a b c
1 1 1 1 1 1
2 2 2
a + b + c + ≥ a+b− b+c− c+a− .
2 2 2 2 2 2
L
2.71. If a, b, c are real numbers such that a + b + c = 3, then
.M
a(a − 1) b(b − 1) c(c − 1)
+ + 2 ≥ 0.
D
8a2 + 9 8b2 + 9 8c + 9
A
PI
2.72. If a, b, c are real numbers such that a + b + c = 3, then
M
(a − 11)(a − 1) (b − 11)(b − 1) (c − 11)(c − 1)
+ + ≥ 0.
2a2 + 1 2b2 + 1 2c 2 + 1
LY
O
2.75. Any sixth degree symmetric homogeneous polynomial f6 (a, b, c) can be written
in the form
f6 (a, b, c) = Ar 2 + B(p, q)r + C(p, q),
where A is called the highest coefficient of f6 , and
p = a + b + c, q = a b + bc + ca, r = a bc.
In the case A ≤ 0, prove that the inequality f6 (a, b, c) ≥ 0 holds for all real numbers
a, b, c if and only if f6 (a, 1, 1) ≥ 0 for all real a.
30 Vasile Cîrtoaje
L
a2 (a − b)(a − c)(a − 4b)(a − 4c) + 7(a − b)2 (b − c)2 (c − a)2 ≥ 0.
P
(b)
.M
D
2.78. If a, b, c are real numbers, then
A
(a2 + 2bc)(b2 + 2ca)(c 2 + 2a b) + (a − b)2 (b − c)2 (c − a)2 ≥ 0.
PI
M
2.79. If a, b, c are real numbers, then
LY
(2a2 + 5a b + 2b2 )(2b2 + 5bc + 2c 2 )(2c 2 + 5ca + 2a2 ) + (a − b)2 (b − c)2 (c − a)2 ≥ 0.
O
.M
2 2 2 64 2
2 2 2 2 2 2
a + ab + b b + bc + c c + ca + a ≥ (a + bc)(b2 + ca)(c 2 + a b).
W
3 3 3 27
W
W
a2 + 3bc b2 + 3ca c 2 + 3a b
+ 2 + 2 ≥ 0.
b2 + c 2 c + a2 a + b2
a2 + 6bc b2 + 6ca c 2 + 6a b
+ + ≥ 0.
b2 − bc + c 2 c 2 − ca + a2 a2 − a b + b2
L
2.85. If a, b, c are real numbers such that a b + bc + ca ≥ 0, then
.M
4a2 + 23bc 4b2 + 23ca 4c 2 + 23a b
+ + ≥ 0.
D
b2 + c 2 c 2 + a2 a2 + b2
A
PI
2.86. If a, b, c are real numbers such that a b + bc + ca = 3, then
M
20(a6 + b6 + c 6 ) + 43a bc(a3 + b3 + c 3 ) ≥ 189.
LY
X
4 (a2 + bc)(a − b)(a − c)(a − 3b)(a − 3c) ≥ 7(a − b)2 (b − c)2 (c − a)2 .
.M
2.88. Let a, b, c be real numbers such that a b + bc + ca ≥ 0. For any real k, prove that
W
X
4bc(a − b)(a − c)(a − k b)(a − kc) + (a − b)2 (b − c)2 (c − a)2 ≥ 0.
W
W
8
(a) (a2 + 1)(b2 + 1)(c 2 + 1) ≥ p |(a − b)(b − c)(c − a)|;
3 3
(b) (a2 − a + 1)(b2 − b + 1)(c 2 − c + 1) ≥ |(a − b)(b − c)(c − a)|.
L
.M
2.93. If a, b, c are real numbers such that a + b + c = 3, then
D
(1 − a + a2 )(1 − b + b2 )(1 − c + c 2 ) ≥ 1.
A
2.94. If a, b, c are real numbers such that a + b + c = 0, then
PI
M
a(a − 4) b(b − 4) c(c − 4)
+ 2 + 2
LY
≥ 0.
a2 + 2 b +2 c +2
O
2
1 + a bcd
2 2 2 2
(1 − a + a )(1 − b + b )(1 − c + c )(1 − d + d ) ≥ .
2
W
W
2.96. Let a, b, c, d be real numbers such that a bcd > 0. Prove that
W
1 1 1 1 1 1 1 1
a+ b+ c+ d+ ≥ (a + b + c + d) + + + .
a b c d a b c d
a + b + c + d = 4, a2 + b2 + c 2 + d 2 = 7.
Prove that
a3 + b3 + c 3 + d 3 ≤ 16.
Symmetric Polynomial Inequalities in Real Variables 33
12(a4 + b4 + c 4 + d 4 ) ≤ 7(a2 + b2 + c 2 + d 2 )2 .
(a2 + b2 + c 2 + d 2 )3 ≥ 3(a3 + b3 + c 3 + d 3 )2 .
L
.M
(1 + a2 )(1 + b2 )(1 + c 2 )(1 + d 2 ) ≥ (a + b + c + d)2 .
D
2.101. Let a, b, c, d be real numbers such that
A
a2 + b2 + c 2 + d 2 = 4.
PI
M
Prove that
(a bc)3 + (bcd)3 + (cd a)3 + (d a b)3 ≤ 4.
LY
O
−1
2.103. If a, b, c, d ≥ such that a + b + c + d = 4, then
2
W
1−a+a 2 1− b+ b 2 1−c+c 2 1 − d + d2
30(a4 + b4 + c 4 + d 4 + e4 ) ≥ 7(a2 + b2 + c 2 + d 2 + e2 )2 .
34 Vasile Cîrtoaje
p
2.108. Let a1 , a2 , . . . , an ≥ n − 1 − n2 − n + 1 be nonzero real numbers such that a1 +
L
a2 + · · · + an = n. Prove that
.M
1 1 1
+ + ··· + ≥ n.
a12 a22 an2
D
A
2.109. Let a1 , a2 , . . . , an ≤
n
n−2
be real numbers such that PI
M
a1 + a2 + · · · + an = n.
LY
−(3n − 2)
a1 , a2 , . . . , an ≥ , a1 + a2 + · · · + an = n,
n−2
W
then
W
1 − a1 1 − a2 1 − an
+ + ··· + ≥ 0.
(1 + a1 )2 (1 + a2 )2 (1 + an )2
(a1 + a2 + · · · + an )2 (n − 1)n−1
(a) ≤ ;
(a12 + 1)(a22 + 1) · · · (an2 + 1) nn−2
1
a1 + a2 + · · · + an (2n − 1)n− 2
(b) ≤ .
(a12 + 1)(a22 + 1) · · · (an2 + 1) 2n nn−1
Symmetric Polynomial Inequalities in Real Variables 35
2.2 Solutions
a3 + b3 + c 3 + d 3 ≤ 27.
a3 + b3 + c 3 + d 3 ≤ 3(a2 + b2 + c 2 + d 2 ) = 27.
L
The equality holds for a = 3 and b = c = d = 0 (or any cyclic permutation thereof).
D
.M
P 2.2. If a, b, c are real numbers such that a + b + c = 0, then
A
PI
(2a2 + bc)(2b2 + ca)(2c 2 + a b) ≤ 0.
M
First Solution. Among a, b, c there are two with the same sign. Let bc ≥ 0. We need to
LY
show that
(2b2 + ca)(2c 2 + a b) ≤ 0.
O
This is equivalent to
[2b2 − c(b + c)][2c 2 − (b + c)b] ≤ 0,
.M
−a
the conclusion follows. The equality holds for = b = c (or any cyclic permutation).
2
W
L
and a + b ≥ 0. Replacing a by −a, we need to show that 0 ≤ a ≤ b ≤ c involves
.M
−a(c − b)2 + b(c + a)2 + c(a + b)2 ≥ 0.
D
This is true since
A
PI
−a(c − b)2 + b(c + a)2 ≥ −b(c − b)2 + b(c + a)2 = b(a + b)(a − b + 2c) ≥ 0.
x y z
a= p , b= p , c=p ,
3 3 3
we need to show that
(x 2 + 1)( y 2 + 1)(z 2 + 1) ≥ 64
for all real x, y, z such that x y + yz + z x = 9.
First Solution. Applying the Cauchy-Schwarz inequality, we have
y +z
The equality holds for x y + yz + z x = 9 and = yz − 1; that is, for
x
( yz − 1)(9 − yz)
y + z = ( yz − 1)x = ,
y +z
( y + z)2 + ( yz − 1)( yz − 9) = 0,
( y − z)2 + ( yz − 3)2 = 0,
p
y = z = ± 3.
In addition, from x y + yz + z x = 9, we get
L
p
x = y = z = ± 3.
.M
Therefore, the original inequality becomes an equality for
D
a = b = c = ±1.
A
Second Solution. We have PI X X
(x 2 + 1)( y 2 + 1)(z 2 + 1) − 64 = x 2 y 2 z 2 + x2 y2 + x 2 − 63
M
2 X 2
LY
X X X
= x 2 y 2z2 + xy − 2x yz x+ x −2 x y − 63
X X 2 X 2
O
= x 2 y 2 z 2 − 2x yz x+ x = x yz − x ≥ 0.
.M
W
Solution. Since
Clearly, this inequality is true if a ≤ 0 and b ≤ 0. Otherwise, write the inequality in the
form
3(2a + 2b − a b − 1)2 + a2 b2 − 10a b + 1 ≥ 0.
38 Vasile Cîrtoaje
This inequality is obviously true for a b ≤ 0. Therefore, consider further that a > 0
and b > 0. If a2 b2 − 10a b + 1 ≥ 0, then the inequality is true. Assume now that
a2 b2 − 10a b + 1 ≤ 0. Since
p
2a + 2b − a b − 1 ≥ 4 a b − a b − 1
and
p 14a b − a2 b2 − 1 10a b − a2 b2 − 1
4 ab − ab − 1 = p > p ≥ 0,
4 ab + ab + 1 4 ab + ab + 1
it suffices to show that
p
L
3(4 a b − a b − 1)2 + a2 b2 − 10a b + 1 ≥ 0.
.M
This inequality is equivalent to
D
p
(a b − 3 a b + 1)2 ≥ 0,
A
1 p
which is obviously true. The equality holds for a = b = PI (3 ± 5).
2
M
LY
it follows that
W
2(1 − a + a2 )(1 − b + b2 ) ≥ 1 + a2 b2 .
Thus, it is enough to prove that
(3 + a2 b2 )c 2 − (3 + 2a b + 3a2 b2 )c + 1 + 3a2 b2 ≥ 0,
D = −3(1 − a b)4 ≤ 0.
Symmetric Polynomial Inequalities in Real Variables 39
Substituting a, b, c by |a|, |b|, |c|, respectively, the left side of this inequality remains
L
unchanged or decreases, while the right side remains unchanged. Therefore, it suffices
to prove the inequality only for a, b, c ≥ 0. For a = b = c, the inequality is true since
.M
3(1 − a + a2 )3 − (1 + a3 + a6 ) = (1 − a)4 (2 − a + 2a2 ) ≥ 0.
D
A
Multiplying the inequalities
p
3 3
p PI
3(1 − a + a2 ) ≥ 1 + a3 + a6 ,
M
p
3 3
p
3(1 − b + b2 ) ≥ 1 + b3 + b6 ,
LY
p3 3
p
3(1 − c + c 2 ) ≥ 1 + c 3 + c 6 ,
O
we get
.M
Æ
3
3(1 − a + a2 )(1 − b + b2 )(1 − c + c 2 ) ≥ (1 + a3 + a6 )(1 + b3 + b6 )(1 + c 3 + c 6 ).
W
Æ
3
(1 + a3 + a6 )(1 + b3 + b6 )(1 + c 3 + c 6 ) ≥ 1 + a bc + a2 b2 c 2 ,
W
Solution. Substituting a, b, c by |a|, |b|, |c|, respectively, the left side of the inequality
remains unchanged, while the right side either remains unchanged or increases. There-
fore, it suffices to prove the inequality only for a, b, c ≥ 0. Let p = a + b + c and
q = a b + bc + ca. Since
L
Thus, it suffices to show that
.M
(p2 − 2q)3 ≥ q(p4 − 3p2 q + q2 ),
D
which is equivalent to the obvious inequality
A
(p2 − 3q)2 (p2 − q) ≥ 0.
PI
The equality holds for a = b = c.
M
LY
O
Solution. Since
W
and
2(b2 + c 2 ) = (b + c)2 + (b − c)2 ,
the required inequality follows by applying the Cauchy-Schwarz inequality as follows
First Solution. Substituting a, b, c by |a|, |b|, |c|, respectively, the left side of this in-
equality remains unchanged, while the right side remains unchanged or increases. There-
fore, it suffices to prove the inequality only for a, b, c ≥ 0. Without loss of generality,
assume that a ≥ b ≥ c ≥ 0. Since
3(a2 + 1)(b + c)
2(a b + bc + ca) ≤ 3a(b + c) ≤ ,
2
L
it suffices to prove that
.M
2(b2 + 1)(c 2 + 1) ≥ 3(b + c),
which is equivalent to
D
2(b + c)2 − 3(b + c) + 2(bc − 1)2 ≥ 0.
A
PI
Case 1: 4bc ≤ 1. We have
M
2
3 (1 − 4bc)(7 − 4bc)
LY
2 2
2(b + c) − 3(b + c) + 2(bc − 1) = 2 b + c − + ≥ 0.
4 8
O
9(b + c)2
− 3(b + c) + 2 ≥ 0,
8
W
and
W
7(b + c)2
+ 2b2 c 2 − 4bc ≥ 0.
8
W
We have
9(b + c)2 [3(b + c) − 4]2
− 3(b + c) + 2 = ≥0
8 8
and
7(b + c)2 7bc bc(4bc − 1)
+ 2b2 c 2 − 4bc ≥ + 2b2 c 2 − 4bc = .
8 2 2
where
(b + c)2
A = (b2 + 1)(c 2 + 1) − 2bc − .
(b2 + 1)(c 2 + 1)
We need to show that A ≥ 0. By virtue of the Cauchy-Schwarz inequality,
Then,
A ≥ (b2 + 1)(c 2 + 1) − 2bc − 1 = b2 c 2 + (b − c)2 ≥ 0.
L
.M
P 2.10. If a, b, c are real numbers, then
5
D
(a2 + 1)(b2 + 1)(c 2 + 1) ≥ (a + b + c + 1)2 .
16
A
(Vasile Cîrtoaje, 2006)
PI
First Solution. Replacing a, b, c respectively by a/2, b/2, c/2, the inequality becomes
M
(a2 + 4)(b2 + 4)(c 2 + 4) ≥ 5(a + b + c + 2)2 .
LY
Since the equality in this inequality holds for a = b = c = 1, we apply the Cauchy-
Schwarz inequality in the form
O
b+c+2 2
.M
2 2
(a + b + c + 2) ≤ (a + 4) 1 + .
2
2
b+c+2
W
2 2
(b + 4)(c + 4) ≥ 5 1 + .
2
W
Indeed,
1
Equality in the original inequality occurs for a = b = c = .
2
Second Solution. Obviously, among a2 , b2 , c 2 there are two either less than or equal to
1 1
, or greater than or equal to . Let b and c be these numbers; that is,
4 4
(4b2 − 1)(4c 2 − 1) ≥ 0.
Then, we have
2 2
16 2 2 4b − 1 4c − 1
(b + 1)(c + 1) = 5 +1 +1
5 5 5
L
.M
4b2 − 1 4c 2 − 1
≥5 + + 1 = 4b2 + 4c 2 + 3.
5 5
D
Hence, it suffices to prove that
A
(a2 + 1)(4b2 + 4c 2 + 3) ≥ (a + b + c + 1)2 .
PI
Writing this inequality as
M
1 1 1
LY
2
a + + + (1 + 4b2 + 4c 2 + 2) ≥ (a + b + c + 1)2 ,
4 4 2
O
X
( a3 + a bc)2 ≥ 0.
L
X X
( a3 )2 − 4a bc a3 + 4a2 b2 c 2 ≥ 0,
.M
X
( a3 − 2a bc)2 ≥ 0.
D
The equality holds for a3 + b3 + c 3 − 2a bc = 0.
A
PI
M
P 2.12. If a, b, c are real numbers, then
2 6
LY
(a + b6 + c 6 ) + a3 b3 + b3 c 3 + c 3 a3 + a bc(a3 + b3 + c 3 ) ≥ 0.
3
O
4 6
(a + b6 + c 6 ) + 2(a3 b3 + b3 c 3 + c 3 a3 ) + 2a bc(a3 + b3 + c 3 ) ≥ 0,
3
W
1 6
(a + b6 + c 6 ) + (a3 + b3 + c 3 )2 + 2a bc(a3 + b3 + c 3 ) ≥ 0.
W
3
By virtue of the AM-GM inequality, we have
W
1 6
(a + b6 + c 6 ) ≥ a2 b2 c 2 .
3
Therefore, it suffices to show that
a2 b2 c 2 + (a3 + b3 + c 3 )2 + 2a bc(a3 + b3 + c 3 ) ≥ 0,
which is equivalent to
(a bc + a3 + b3 + c 3 )2 ≥ 0.
The equality holds for −a = b = c (or any cyclic permutation).
Symmetric Polynomial Inequalities in Real Variables 45
4x y = (x + y)2 − (x − y)2 ,
we have
L
4(a2 + a b + b2 )(a2 + ac + c 2 ) = [(2a2 + a b + ac + 2bc) + (b − c)2 ]2 − (a + b + c)2 (b − c)2
.M
= (2a2 + a b + ac + 2bc)2 + 3a2 (b − c)2 .
D
From this result and
A
4(b2 + bc + c 2 ) = (b − c)2 + 3(b + c)2 ,
PI
M
we get
LY
Y
(a2 + a b + b2 ) = (2a2 + a b + ac + 2bc)2 + 3a2 (b − c)2 (b − c)2 + 3(b + c)2 .
16
O
Y 2
(a2 + a b + b2 ) ≥ (2a2 + a b + ac + 2bc)(b − c) + 3a(b − c)(b + c)
16
= 4(b − c)2 (a − b)2 (a − c)2 .
W
Y
4 (a2 + a b + b2 ) = 3[a b(a + b) + bc(b + c) + ca(c + a)]2 + (a − b)2 (b − c)2 (c − a)2 .
Solution (by Vo Quoc Ba Can). As we have shown in the proof of the preceding P 2.13,
Y
(a2 + a b + b2 ) = (2a2 + a b + ac + 2bc)2 + 3a2 (b − c)2 3(b + c)2 + (b − c)2 .
16
L
.M
which follows immediately from the AM-GM inequality. The equality holds when two
of a, b, c are equal.
D
Remark. The inequality is a consequence of the identity
A
Y
(a2 + a b + b2 ) = 3(a2 b + b2 c + c 2 a)(a b2 + bc 2 + ca2 ) + (a − b)2 (b − c)2 (c − a)2 .
PI
M
LY
1 1 1
4 a+ b+ c+ ≥ 9(a + b + c).
a b c
.M
p
First Solution. It is easy to checkpthat the equality
p occurs
p for a = b = c = 2. There-
fore, using the substitution a = x 2, b = y 2, c = z 2, the inequality can be written
W
as
(2x 2 + 1)(2 y 2 + 1)2z 2 + 1) ≥ 9x yz(x + y + z).
Since
1X 2
(x y + yz + z x)2 − 3x yz(x + y + z) = x ( y − z)2 ≥ 0,
2
it suffices to prove the stronger inequality
Let
A = ( y 2 − 1)(z 2 − 1), B = (z 2 − 1)(x 2 − 1), C = (x 2 − 1)( y 2 − 1).
Symmetric Polynomial Inequalities in Real Variables 47
From
ABC = (x 2 − 1)2 ( y 2 − 1)2 (z 2 − 1)2 ≥ 0,
A = ( y 2 − 1)(z 2 − 1) ≥ 0.
(x y + yz + z x)2 ≤ (x 2 + 1 + x 2 )( y 2 + y 2 z 2 + z 2 ).
L
Therefore, it suffices to show that
.M
(2 y 2 + 1)(2z 2 + 1) ≥ 3( y 2 + y 2 z 2 + z 2 ),
D
which reduces to the obvious inequality
A
( y 2 − 1)(z 2 − 1) ≥ 0.
PI
M
Second Solution. Since
LY
c 2 + 2a b = q + (c − a)(c − b),
L
X
q2 (a2 + b2 + c 2 ) ≥ q3 + q2 (a − b)(a − c) − (a − b)2 (b − c)2 (c − a)2 .
.M
Since X
(a − b)(a − c) = a2 + b2 + c 2 − q,
D
A
the inequality reduces to the obvious form
PI
(a − b)2 (b − c)2 (c − a)2 ≥ 0.
M
The equality holds for a = b, or b = c, or c = a.
LY
Thus, the required inequality follows immediately from the inequality (a) in P 2.16. The
equality holds for a = b = c.
Symmetric Polynomial Inequalities in Real Variables 49
L
for all a, b, c ≥ 0 satisfying a2 + b2 + c 2 = 2. If b2 −2ac and c 2 −2a b have the same sign,
.M
then the inequality is also trivial. Due to symmetry in b and c, we may assume that
b2 − 2ac ≥ 0 ≥ c 2 − 2a b.
D
A
On the other hand, it is easy to check that the desired inequality becomes an equality
for a = b = 1 and c = 0, when a2 + 2bc = b2 − 2ac = a b − c 2 /2. Then, we rewrite the
PI
desired inequality in the form
M
c2
2 2
(a + 2bc)(b − 2ac) a b − ≤ 1.
LY
2
Using the AM-GM inequality, we have
O
3
c2 c2
.M
2 2 2 2
27(a + 2bc)(b − 2ac) a b − ≤ (a + 2bc) + (b − 2ac) + a b − .
2 2
c2
2 2
(a + 2bc) + (b − 2ac) + a b −
W
≤ 3.
2
W
3(a4 + b4 + c 4 ) + a2 + b2 + c 2 + 6 ≥ 6(a3 + b3 + c 3 ).
L
F (a, b, c) ≥ F (a, x, x) ≥ 0,
.M
where x = (b + c)/2, x ≥ 1. We have
D
F (a, b, c) − F (a, x, x) = 3(b4 + b4 − 2x 4 ) + (b2 + c 2 − 2x 2 ) − 6(b3 + c 3 − 2x 3 )
A
= 3[(b2 + c 2 )2 − 4x 4 ] + 6(x 4 − b2 c 2 ) + (b2 + c 2 − 2x 2 ) − 6(b3 + c 3 − 2x 3 )
PI
= (b2 + c 2 − 2x 2 )[3(b2 + c 2 + 2x 2 ) + 1] + 6(x 2 − bc)(x 2 + bc) − 12x(b2 + c 2 − bc − x 2 ).
M
Since
1
b2 + c 2 − 2x 2 = (b − c)2 ,
LY
2
1
x 2 − bc = (b − c)2 ,
O
4
3
.M
b2 + c 2 − bc − x 2 = (b − c)2 ,
4
we have
W
1
F (a, b, c) − F (a, x, x) =(b − c)2 [3(b2 + c 2 + 2x 2 ) + 1 + 3(x 2 + bc) − 18x]
W
2
1
= (b − c)2 [3(x 2 − bc) + 18x(x − 1) + 1] ≥ 0.
W
2
Also,
F (a, x, x) = F (3 − 2x, x, x) = 6(x − 1)2 (3x − 4)2 ≥ 0.
This completes the proof. The equality holds for a = b = c = 1, and for a = 1/3 and
b = c = 4/3 (or any cyclic permutation).
p3 + 9a bc ≥ 4pq,
we get
L
p3 + 9
.M
q≤ .
4p
Write the required inequality as
D
3(p2 − 2q) + 2p ≥ 5q,
A
3p2 + 2p ≥ 11q. PI
This is true since
M
11(p3 + 9) (p − 3)(p2 + 11p + 33)
LY
Case 2: a > 0 and b, c < 0. Substituting −b for b and −c for c, we need to prove that
.M
Since p
3(b2 + c 2 ) − 5bc b+c bc 1
W
≥ ≥ = p ,
b+c 4 2 2 a
we need to prove that
1
p ≥ 2 − 5a.
2 a
Indeed, by the AM-GM inequality,
v
1 1 1 t 1 1
5a + p = 5a + p + p ≥ 3 3 5a · p · p > 2.
2 a 4 a 4 a 4 a 4 a
This completes the proof. The equality holds for a = b = c = 1.
52 Vasile Cîrtoaje
3 1 1 1
2 2 2
a + b +c +6≥ a+b+c+ + + .
2 a b c
Solution. Substituting a, b, c by |a|, |b|, |c|, respectively, the left side of the inequality
remains unchanged, while the right side either remains unchanged or increases. There-
fore, it suffices to prove the inequality only for a, b, c > 0.
L
.M
3(6x 2 − 3x + 4) ≥ 7(a b + bc + ca),
a+b+c
D
where x = . By virtue of the AM-GM inequality, we get x ≥ 1. The third degree
3
A
Schur’s inequality states that
PI
(a + b + c)3 + 9a bc ≥ 4(a + b + c)(a b + bc + ca),
M
which is equivalent to
LY
3(3x 3 + 1)
a b + bc + ca ≤ .
4x
O
21(3x 3 + 1)
3(6x 2 − 3x + 4) ≥ .
4x
W
3 1
3x 2 − 9x + 7 = 3(x − )2 + > 0.
W
2 4
Second Solution. Due to symmetry, assume that a = min{a, b, c}, and then use the
mixing variable method. Let
3 1 1 1
F (a, b, c) = a2 + b2 + c 2 + 6 − a+b+c+ + + .
2 a b c
p
where x = bc (x ≥ 1). We have
3 1 1 2
p p p
2
F (a, b, c) − F (a, bc, bc) = (b − c) − b + c − 2 bc + + − p
2 b c
bc
1 p p 2 p 2 3
p
= ( b − c) 2( b + c) − 3 −
2 bc
1 p p 2 p 3
≥ ( b − c) 8 bc − 3 −
2 bc
1 p p 2
≥ ( b − c) (8 − 3 − 3) ≥ 0
2
L
and
.M
p p 1
F (a, bc, bc) = F ( , x, x)
x2
x 6 − 6x 5 + 12x 4 − 6x 3 − 3x 2 + 2
D
=
2x 4
A
(x − 1) (x − 4x 3 + 3x 2 + 4x + 2
2 4
= PI
2x 4
(x − 1) [(x − 2x − 1)2 + x 2 + 1]
2 2
M
= ≥ 0.
2x 4
LY
O
1
(1 + a2 )(1 + b2 )(1 + c 2 ) + 8a bc ≥ (1 + a)2 (1 + b)2 (1 + c)2 .
4
W
W
Solution. Consider only the nontrivial case a bc 6= 0, and rewrite the inequality as fol-
lows
2049 4 4
a12 + b12 + (a + b)12 ≥ a b (a + b)4 ,
8
2049 4 4 2
(a6 + b6 )2 − 2a6 b6 + (a2 + b2 + 2a b)6 ≥ a b (a + b2 + 2a b)2 .
8
L
Let us denote
.M
a2 + b2
d= , |d| ≥ 2.
ab
D
Since
a6 + b6 = (a2 + b2 )3 − 3a2 b2 (a2 + b2 ),
A
the inequality can be restated as PI
2049
M
(d 3 − 3d)2 − 2 + (d + 2)6 ≥ (d + 2)2 ,
8
LY
which is equivalent to
O
a2 + b2 + c 2 + 2a bc + 4 ≥ 2(a + b + c) + a b + bc + ca.
Since
x yz = a bc(a − 1)2 (b − 1)2 (c − 1)2 ≥ 0,
at least one of x, y, z is nonnegative; let a(1 − b)(1 − c) ≥ 0. Thus, we have
a bc ≥ a(b + c − 1),
L
a2 + b2 + c 2 + 2a(b + c − 1) + 4 ≥ 2(a + b + c) + a b + bc + ca,
.M
which is equivalent to
D
a2 − (4 − b − c)a + b2 + c 2 − bc − 2(b + c) + 4 ≥ 0,
A
b+c 2 3
a−2+
2 4
PI
+ (b − c)2 ≥ 0.
M
The equality holds for a = b = c = 1, and also for a = 0 and b = c = 2 (or any cyclic
permutation).
LY
O
2
+ 2+ 2≥ + + .
a b c a b c
W
1−a 1− b 1−c
+ + ≥ 0.
(1 + a)2 (1 + b)2 (1 + c)2
(Vasile Cîrtoaje, 2012)
L
(b − c)2 [(b + c)2 − bc(b + c − 2)]
.M
= ≥ 0,
b2 c 2 (b + c)2
since
D
(b + c)2 − bc(b + c − 2) = (b + c)2 − bc(1 − a)
A
(b + c)2 (1 − a) (b + c)2 (3 + a)
≥ (b + c)2 −
4
= PI4
≥ 0.
M
Also,
1 − a 2(1 − t) 3(a − 1)2 (a + 3)
E(a, t, t) = + = ≥ 0.
LY
a2 t2 a2 (3 − a)2
The equality holds for a = b = c = 1, and also for a = −3 and b = c = 3 (or any cyclic
O
permutation).
.M
(b) From
a+b+c 3−a
t≥ = 1, t= ≤ 5,
3 2
W
it follows that
t ∈ [1, 5].
W
E(a, b, c) ≥ E(a, t, t) ≥ 0,
where
1−a 1− b 1−c
E(a, b, c) = + + .
(1 + a)2 (1 + b)2 (1 + c)2
Write the left inequality as follows:
1− b 1− t 1−c 1− t
− + − ≥ 0,
(1 + b)2 (1 + t)2 (1 + c)2 (1 + t)2
(b − 1)t − b − 3 (c − 1)t − c − 3
(b − c) − ≥ 0,
(1 + b)2 (1 + c)2
Symmetric Polynomial Inequalities in Real Variables 57
Also, we have
1−a 2(1 − t) t −1 2(1 − t)
E(a, t, t) = + = +
(1 + a)2 (1 + t)2 2(2 − t)2 (1 + t)2
L
3(1 − t)2 (5 − t)
= ≥ 0.
.M
2(2 − t)2 (1 + t)2
The proof is completed. The equality occurs for a = b = c = 1, and also for a = −7 and
b = c = 5 (or any cyclic permutation).
D
A
P 2.26. If a, b, c are real numbers, then
PI
M
1
a6 + b6 + c 6 − 3a2 b2 c 2 ≥ (a − b)2 (b − c)2 (c − a)2 .
LY
2
(Sungyoon Kim, 2006)
O
1 2
a6 + b6 + c 6 − 3a2 b2 c 2 = (a + b2 + c 2 )[(b2 − c 2 )2 + (c 2 − a2 )2 + (a2 − b2 )2 ]
2
W
1
≥ [a(b2 − c 2 ) + b(c 2 − a2 ) + c(a2 − b2 )]2
2
W
1
= (a − b)2 (b − c)2 (c − a)2 .
2
W
Thus, the proof is completed. The equality holds for a = b = c, and for a = 0 and
b + c = 0 (or any cyclic permutation).
Solution (by Vo Quoc BapCan). Without loss of generality, assume that b and c have the
same sign. Denote x = (b2 + c 2 )/2. Since
3 3
a2 + b2 + c 2 a2 + 2x 2
2 2 2
−a b c = − a2 x 4 + a2 (x 4 − b2 c 2 )
3 3
(a2 − x 2 )2 (a2 + 8x 2 )
= + a2 (x 4 − b2 c 2 )
27
(2a2 − b2 − c 2 )2 (a2 + 4b2 + 4c 2 ) a2 (b2 − c 2 )2
= + ,
108 4
the desired inequality can be rewritten as
L
27
.M
(2a2 − b2 − c 2 )2 (a2 + 4b2 + 4c 2 ) ≥ (b − c)2 [(a − b)2 (a − c)2 − 4a2 (b + c)2 ].
4
D
1
According to the inequalities x 2 − y 2 ≤ 2x(x + y) and 2x y ≤ (x + y)2 , we have
2
A
(a − b)2 (a − c)2 − 4a2 (b + c)2 ≤ 2(a − b)(a − c)[(a − b)(a − c) + 2a(b + c)]
PI
1
= 2(a2 − b2 )(a2 − c 2 ) ≤ (2a2 − b2 − c 2 )2 .
M
2
LY
which is equivalent to the obvious inequality 8a2 + 5b2 + 5c 2 + 54bc ≥ 0. The equality
.M
108 2 2
(a2 + 1 − 2x)3 ≥ a x + 2(1 − 2x)(a2 − a + x)2,
5
2
(4a − 5)2 x 2 + 2(a + 1)(a3 − 9a2 + 5a − 3)x + (a + 1)2 (a4 − 2a3 + 4a2 − 2a + 1) ≥ 0.
5
This inequality holds if
L
(4a − 5)2 (a4 − 2a3 + 4a2 − 2a + 1) ≥ (a3 − 9a2 + 5a − 3)2 .
5
.M
Since
D
10(a4 − 2a3 + 4a2 − 2a + 1) = (a + 1)2 + (3a2 − 4a + 3)2 ≥ (3a2 − 4a + 3)2 ,
A
it suffices to prove that PI
(4a − 5)2 (3a2 − 4a + 3)2 ≥ 25(a3 − 9a2 + 5a − 3)2 .
M
LY
which reduces to
a(a + 1)2 ≤ 0.
Thus, the proof is completed. The equality holds for a = 0 and b + c = 0 (or any cyclic
W
permutation).
W
W
This completes the proof. The equality holds for (a − b)(bc − ac) = (a + b)(a b + c 2 ),
which is equivalent to
(a + b + c)(a b + bc + ca) = 5a bc.
L
Second Solution. Making the substitution
.M
X X
x= a b2 = a b2 + bc 2 + ca2 , y= a2 b = a2 b + b2 c + c 2 a,
D
we have X X X
A
a2 b4 = ( a b2 )2 − 2a bc a2 b = x 2 − 2a bc y,
X
a4 b2 = (
X
a2 b)2 − 2a bc
X PI
a b2 = y 2 − 2a bc x,
M
and hence
LY
Y X X
(a2 + b2 ) = a2 b4 + a4 b2 + 2a2 b2 c 2
O
= x 2 + y 2 − 2a bc(x + y) + 2a2 b2 c 2 .
.M
(x + y − 2a bc)2 ≥ 0.
W
Solution (by Vo Quoc Ba Can). For a, b, c ≥ 0, the inequality is trivial. Otherwise, since
the inequality is symmetric and does not change by substituting −a, −b, −c for a, b, c,
we may assume that a ≤ 0 and b, c ≥ 0. Substituting −a for a, we need to prove that
L
8(b2 − ac)(c 2 − a b) + 9a(b + c)(b − c)2 ≥ 0.
.M
Since
D
(b2 − ac)(c 2 − a b) = bc(bc + a2 ) − a(b3 + c 3 )
p
A
p p
≥ 2a bc bc − a(b3 + c 3 ) = −a(b b − c c)2 ,
p p
Setting b = x and c = y, the inequality can be rewritten as
O
p 2
2
p 2 2 2 2
p
≥ 9[(x − y) + 2 2x y] ≥ 9 (x − y) + 2 2x y
W
3
= 8(x 2 + x y + y 2 )2 ≥ 0.
W
The equality occurs when two of a, b, c are zero, and when −a = b = c (or any cyclic
permutation).
Solution. Since
1 2
b4 (c − a)2 + c 4 (a − b)2 ≥ [b (c − a) + c 2 (a − b)]2
2
1
= (b − c)2 (bc − ca − a b)2 ,
2
which is equivalent to
L
a2 (a2 − 2bc + 2ca + 2a b) ≥ 0.
.M
Therefore, the desired inequality is true if
D
a2 − 2bc + 2ca + 2a b ≥ 0.
A
Indeed, from X X PI
(a2 − 2bc + 2ca + 2a b) = (a + b + c)2 ≥ 0,
M
due to symmetry, we may assume that a2 − 2bc + 2ca + 2a b ≥ 0. Thus, the proof is
LY
completed. The equality occurs when a = b = c, when two of a, b, c are equal to zero,
and when a = 0 and b + c = 0 (or any cyclic permutation).
O
.M
1
a2 (b − c)4 + b2 (c − a)4 + c 2 (a − b)4 ≥ (a − b)2 (b − c)2 (c − a)2 .
W
2
W
Since X X X
a2 b4 = ( a b2 )2 − 2a bc a2 b = x 2 − 2a bc y,
X X X
a4 b2 = ( a2 b)2 − 2a bc a b2 = y 2 − 2a bc x,
X
a2 b2 (a2 + b2 ) = x 2 + y 2 − 2a bc(x + y),
Symmetric Polynomial Inequalities in Real Variables 63
we have
X X
a2 (b − c)4 = a2 (b4 − 4b3 c + 6b2 c 2 − 4bc 3 + c 4 )
X X X
= a2 b2 (a2 + b2 ) − 4a bc( a b2 + a2 b) + 18a2 b2 c 2
= x 2 + y 2 − 6a bc(x + y) + 18a2 b2 c 2 .
1
x 2 + y 2 − 6a bc(x + y) + 18a2 b2 c 2 ≥ (x − y)2 ,
2
L
which is equivalent to the obvious inequality
.M
(x + y − 6a bc)2 ≥ 0.
D
The equality holds for a(b − c)2 + b(c − a)2 + c(a − b)2 = 0.
A
PI
M
P 2.33. If a, b, c are real numbers, then
LY
3
a2 (b2 − c 2 )2 + b2 (c 2 − a2 )2 + c 2 (a2 − b2 )2 ≥ (a − b)2 (b − c)2 (c − a)2 .
8
O
Solution. We see that the inequality remains unchanged and the product
W
(a + b)(b + c)(c + a)
W
changes its sign by replacing a, b, c with −a, −b, −c, respectively. Thus, without loss of
generality, we may assume that (a + b)(b + c)(c + a) ≥ 0. According to this condition,
W
1 2 2 1
b2 (c 2 − a2 )2 + c 2 (a2 − b2 )2 ≥ b(c − a2 ) + c(a2 − b2 ) = (b − c)2 (a2 + bc)2 .
2 2
Thus, it suffices to show that
3
2a2 (b + c)2 + (a2 + bc)2 ≥ (a − b)2 (a − c)2 ,
4
64 Vasile Cîrtoaje
which is equivalent to
L
.M
P 2.34. If a, b, c are real numbers such that a b + bc + ca = 3, then
D
(a) (a2 + a b + b2 )(b2 + bc + c 2 )(c 2 + ca + a2 ) ≥ 3(a + b + c)2 ;
A
3 2
(b) (a2 + a b + b2 )(b2 + bc + c 2 )(c 2 + ca + a2 ) ≥ (a + b2 + c 2 ).
PI 2
(Vasile Cîrtoaje, 1995)
M
LY
Solution. Let
p = a + b + c, q = a b + bc + ca, r = a bc.
O
We have
.M
Y Y
(b2 + bc + c 2 ) = [(b + c)2 − bc]
Y X X
= (b + c)2 − bc(a + b)2 (a + c)2 + a bc a(b + c)2 − a2 b2 c 2 .
W
Since
W
Y
(b + c)2 = (pq − r)2 = r 2 − 2pqr + p2 q2 ,
W
X X X
bc(a + b)2 (a + c)2 = bc(a2 + q)2 = r a3 + 2pqr + q2
L
2(p2 − q)q2 − 2p3 r − (p2 − 2q)q2 ≥ 0,
.M
p2 (q2 − 2pr) ≥ 0,
(a + b + c)2 (a2 b2 + b2 c 2 + c 2 a2 ) ≥ 0.
D
A
The equality holds when a + b + c = 0 and a b + bc + ca = 3.
PI
M
P 2.35. If a, b, c are real numbers, then
LY
Y
(b2 + bc + c 2 ) = (p2 − q)q2 − p3 r,
W
where
W
p = a + b + c, q = a b + bc + ca, r = a bc.
Thus, we can write the desired inequality as
L
.M
q2 (p2 − 3q) ≥ 0.
The inequality is an equality when a = b = c, and when a = 0 and b = c (or any cyclic
D
permutation).
A
PI
M
P 2.36. If a, b, c are real numbers, not all of the same sign, then
LY
Solution. Since the inequality is symmetric and does not change by substituting −a, −b, −c
.M
for a, b, c ≥ 0. Since the left hand side of this inequality is nonnegative, consider further
the nontrivial case
W
bc − a b − ac > 0.
Since
b2 + bc + c 2 − 3(bc − a b − ac) = (b − c)2 + 3a(b + c) ≥ 0,
it suffices to show that
The original inequality is an equality when two of a, b, c are zero, and when a = 0 and
b = c (or any cyclic permutation).
Second Solution. Since
L
by the Cauchy-Schwarz inequality, we get
.M
16(a2 − a b + b2 )(b2 + bc + c 2 ) ≥ [(a + b)(a + c) + 3(a − b)(a − c)]2 .
D
Thus, we only need to show that
A
(a + b)(a + c) + 3(a − b)(a − c) ≥ 4(bc − a b − ac),
PI
which is equivalent to the obvious inequality a(2a + b + c) ≥ 0.
M
LY
O
Solution. If a, b, c have the same sign, then the inequality follows from
W
a2 + a b + b2 ≥ a2 + b2 , b2 + bc + c 2 ≥ b2 + c 2 , c 2 + ca + a2 ≥ c 2 + a2 .
W
Consider now that a, b, c have not the same sign. Since the inequality is symmetric and
does not change by substituting −a, −b, −c for a, b, c, we may assume that a ≤ 0 and
b, c ≥ 0. Substituting −a for a, we need to prove that
3 2
(a2 − a b + b2 )(a2 − ac + c 2 )(b2 + bc + c 2 ) ≥ (a + b2 )(b2 + c 2 )(c 2 + a2 )
8
for a, b, c ≥ 0. Write this inequality in the form
or
(a − b)2 (a − c)2 ) (b − c)2
+
L
≥ .
a2 + b2 a2 + c 2 2(b2 + c 2 )
.M
Consider further two cases.
Case 1: 2a2 ≤ b2 + c 2 . By the Cauchy-Schwarz inequality, we have
D
A
(a − b)2 (a − c)2 ) [(b − a) + (a − c)]2 (b − c)2
+ ≥ = .
a2 + b2 a2 + c 2 PI
(a2 + b2 ) + (a2 + c 2 ) 2a2 + b2 + c 2
≥ ,
2a2 +b +c
2 2 2(b + c 2 )
2
O
≥ .
a2 + b2 a2 + c 2 c 2 (a2 + b2 ) + b2 (a2 + c 2 ) a2 (b2 + c 2 ) + 2b2 c 2
W
a2 1
≥ .
a (b + c ) + 2b c
2 2 2 2 2 2(b + c 2 )
2
Thus, the proof is completed. The equality holds when two of a, b, c are zero, and when
−a = b = c (or any cyclic permutation).
Symmetric Polynomial Inequalities in Real Variables 69
Solution. Since the inequality is symmetric and does not change by substituting −a, −b, −c
for a, b, c, we may assume that a ≤ 0 and b, c ≥ 0. Substituting −a for a, we need to
prove that
L
for a, b, c ≥ 0. Using the notation
.M
A = b2 + c 2 , B = c 2 + a2 , C = a2 + b2 ,
D
we can write the inequality as follows:
A
PI
2ABC ≥ (A − bc)(B + ca)(C + a b),
M
ABC + a2 b2 c 2 ≥ a b(AB − c 2 C) + ac(AC − b2 B) − bc(BC − a2 A),
6a2 b2 c 2 ≥ 0.
Solution. Substituting a, b, c by |a|, |b|, |c|, respectively, the left side of the inequality
remains unchanged, while the right side either remains unchanged or increases. There-
fore, it suffices to prove the inequality only for a, b, c ≥ 0. If a = b = c, then the
inequality reduces to
9(1 + a4 )3 ≥ 8(1 + a3 + a6 )2 ,
1 3 1
9(a2 + 2
) ≥ 8(a3 + 3 + 1)2 .
a a
1
Setting a + = x, this inequality can be written as follows
a
L
9(x 2 − 2)3 ≥ 8(x 3 − 3x + 1)2 ,
.M
x 6 − 6x 4 − 16x 3 + 36x 2 + 48x − 80 ≥ 0,
D
(x − 2)2 [x(x 3 − 8) + 4(x 3 − 5) + 6x 2 ] ≥ 0.
A
PI
Since x ≥ 2, the last inequality is clearly true. Multiplying now the inequalities
9(1 + a4 )3 ≥ 8(1 + a3 + a6 )2 ,
M
LY
9(1 + b4 )3 ≥ 8(1 + b3 + b6 )2 ,
O
9(1 + c 4 )3 ≥ 8(1 + c 3 + c 6 )2 ,
.M
we get
(1 + a3 + a6 )(1 + b3 + b6 )(1 + c 3 + c 6 ) ≥ (1 + a bc + a2 b2 c 2 )3 ,
W
Solution. Substituting a, b, c by |a|, |b|, |c|, respectively, the left side of the inequality
remains unchanged, while the right side either remains unchanged or increases. There-
fore, it suffices to prove the inequality only for a, b, c ≥ 0.
First Solution. For a = b = c, the inequality reduces to
2(1 + a2 )3 ≥ (1 + a)3 (1 + a3 ).
L
Multiplying the inequalities
.M
2(1 + a2 )3 ≥ (1 + a)3 (1 + a3 ),
D
2(1 + b2 )3 ≥ (1 + b)3 (1 + b3 ),
A
2(1 + c 2 )3 ≥ (1 + c)3 (1 + c 3 ),
we get
PI
M
8(1 + a2 )3 (1 + b2 )3 (1 + c 2 )3 ≥ (1 + a)3 (1 + b)3 (1 + c)3 (1 + a3 )(1 + b3 )(1 + c 3 ).
LY
which is just Hölder’s inequality. We can also prove this inequality using the AM-GM
inequality. To do this, we write the inequality as
W
1 + a2 1 + x 2 1 + b2 1 + y 2 1 + c2 1 + z2
= , = , =
1+a 1+ x 1+ b 1+ y 1+c 1+z
and
2(1 + x y + yz + z x)
1 + a bc = ,
(1 + x)(1 + y)(1 + z)
72 Vasile Cîrtoaje
(1 + x 2 )(1 + y 2 )(1 + z 2 ) ≥ 1 + x y + yz + z x,
x 2 y 2 + y 2 z 2 + z 2 x 2 + x 2 + y 2 + z 2 ≥ x y + yz + z x,
1 1 1
x 2 y 2 + y 2 z 2 + z 2 x 2 + (x − y)2 + ( y − z)2 + (z − x)2 ≥ 0.
2 2 2
L
.M
3(a2 − a b + b2 )(b2 − bc + c 2 )(c 2 − ca + a2 ) ≥ a3 b3 + b3 c 3 + c 3 a3 .
D
Solution. Substituting a, b, c by |a|, |b|, |c|, respectively, the left side of the inequality
A
remains unchanged or decreases, while the right side remains unchanged or increases.
PI
Therefore, it suffices to prove the inequality for a, b, c ≥ 0. If a = 0, then the inequality
reduces to b2 c 2 (b − c)2 ≥ 0. Consider further then a, b, c > 0. We first show that
M
3(a2 − a b + b2 )3 ≥ a6 + a3 b3 + b6 .
LY
a b
Indeed, setting x = + , x ≥ 2, we can write this inequality as
O
b a
3(x − 1)3 ≥ x 3 − 3x + 1,
.M
(x − 2)2 (2x − 1) ≥ 0.
W
27(a2 − a b + b2 )3 (b2 − bc + c 2 )3 (c 2 − ca + a2 )3 ≥
≥ (a6 + a3 b3 + b6 )(b6 + b3 c 3 + c 6 )(c 6 + c 3 a3 + a6 ).
W
we see that it is just Hölder’s inequality. The equality holds when a = b = c, when a = 0
and b = c (or any cyclic permutation), and when two of a, b, c are 0.
Symmetric Polynomial Inequalities in Real Variables 73
Solution. We have
X b2 − bc + c 2 X a2 X b2 − bc + c 2 b2 c2
+2 = + +
a2 bc a2 ca a b
X (b2 − bc + c 2 )(a b + bc + ca)
=
L
a2 bc
.M
a b + bc + ca X
= bc(b2 − bc + c 2 ).
a2 b2 c 2
D
Then, we can write the inequality as
A
X X
(a b + bc + ca)[ bc(b2 − bc + c 2 ) − a bc a] ≥ 0.
PI
Since
M
X X X X X X
bc(b2 − bc + c 2 ) − a bc a=( bc)( a2 ) − b2 c 2 − 2a bc a
LY
X X X
=( bc)( a2 ) − ( bc)2
O
X X X
=( bc)( a2 − bc),
.M
a bc − (b + c − a)(c + a − b)(a + b − c) ≤ 1;
a bc − (b + c − a)(c + a − b)(a + b − c) ≤ 4.
L
Indeed, substituting −a, −b, −c for a, b, c, respectively, the left inequality can be restated
.M
as
a bc ≥ (b + c − a)(c + a − b)(a + b − c),
D
where a, b, c ≥ 0. This is just the well-known Schur’s inequality of degree three.
Case 2: a > 0. Since (M − m)2 ≥ (a − b)2 and M ≥ a, we have
A
PI
M (M − m)2 ≥ a(a − c)2 .
Therefore, it suffices to show that
M
a bc − (b + c − a)(c + a − b)(a + b − c) ≤ a(a − c)2 ,
LY
which is equivalent to
O
(b − c)[a2 + (b − 2c)a − b2 + c 2 ] ≥ 0.
.M
Thus, the proof is completed. The equality holds when two of a, b, c are equal to m, and
the other is equal to M .
W
M ≥ a ≥ b ≥ c ≥ m.
L
.M
This is equivalent to
D
A
Case 1: a2 ≥ c 2 . From a2 − c 2 = (a − c)(a + c) ≥ 0, it follows that a + c ≥ 0. Then,
PI
a2 + c 2 + ac − a b − bc − max{a2 , c 2 } = (a + c)(c − b) ≤ 0.
M
Case 2: a2 ≤ c 2 . From a2 − c 2 = (a − c)(a + c) ≤ 0, it follows that a + c ≤ 0. Then,
LY
a2 + c 2 + ac − a b − bc − max{a2 , c 2 } = (a + c)(a − b) ≤ 0.
O
Thus, the proof is completed. For M 2 ≥ m2 , the equality holds when two of a, b, c are
equal to m, and the other is equal to M . For M 2 ≤ m2 , the equality holds when two of
.M
a b + bc + ca = a bc + 2.
Prove that p
a2 + b2 + c 2 − 3 ≥ (2 + 3)(a + b + c − 3).
(Vasile Cîrtoaje, 2011)
a + b + c = a bc
implies p
a2 + b2 + c 2 ≥ 3(a + b + c).
76 Vasile Cîrtoaje
Since
1X 2
(a b + bc + ca)2 − 3a bc(a + b + c) = a (b − c)2 ≥ 0,
2
it suffices to prove that
L
(a2 + b2 + c 2 )2 ≥ (a b + bc + ca)2 ,
.M
which is equivalent to
D
(a2 + b2 + c 2 − a b − bc − ca)(a2 + b2 + c 2 + a b + bc + ca) ≥ 0.
A
This inequality is true, since
PI
2(a2 + b2 + c 2 − a b − bc − ca) = (a − b)2 + (b − c)2 + (c − a)2 ≥ 0,
M
2(a2 + b2 + c 2 + a b + bc + ca) = (a + b)2 + (b + c)2 + (c + a)2 ≥ 0.
p
LY
Prove that
W
Solution. Since
2(b2 + c 2 ) = (b + c)2 + (b − c)2
and
(a2 + b2 )(a2 + c 2 ) = (a2 + bc)2 + a2 (b − c)2 ,
by virtue of the Cauchy-Schwarz inequality, we have
L
P 2.47. Let a, b, c be real numbers such that
.M
(a + b)(b + c)(c + a) = 5.
D
Prove that
A
(a2 + a b + b2 )(b2 + bc + c 2 )(c 2 + ca + a2 ) + 12a2 b2 c 2 ≥ 15.
PI (Vasile Cîrtoaje, 2011)
M
Solution. Since
3 1
b2 + bc + c 2 = (b + c)2 + (b − c)2
LY
4 4
and
O
2
a b + ac 3
2 2 2 2 2
(a + a b + b )(a + ac + c ) = a + + bc + a2 (b − c)2 ,
2 4
.M
p p 2
3 a b + ac 3
2 2
(b + c) a + + bc +
W
≥ a(b − c)
2 2 4
W
3 3
= [(a + b)(b + c)(c + a) − 2a bc]2 = (5 − 2a bc)2 .
4 4
Thus, it suffices to show that
3
(5 − 2a bc)2 + 12a2 b2 c 2 ≥ 15,
4
which is equivalent to (2a bc − 1)2 = 0. The homogeneous inequality
L
.M
b2 − (a + c)b + ac ≤ 0,
which is equivalent to
D
2b2 − b + ac ≤ 0.
A
(a) It suffices to show that |b| ≤ 1. We have
PI
25 − b3 = a3 + c 3 = (a + c)3 − 3ac(a + c) = (1 − b)3 − 3ac(1 − b),
M
which yields
LY
b2 − b − 8
ac = .
1− b
O
(1 + b)(4 − 3b + b2 )
≥ 0,
1− b
W
which involves −1 ≤ b < 1, and hence |b| ≤ 1. The equality |b| = 1 holds for a = b = −1
and c = 3.
W
which yields
b2 − b + 4
ac = .
1− b
Thus, the inequality 2b2 − b + ac ≤ 0 is equivalent to
(b − 2)(b2 + 1)
≥ 0,
1− b
which involves 1 < b ≤ 2. The equality b = 2 holds for a = −3 and b = c = 2.
Symmetric Polynomial Inequalities in Real Variables 79
a + b + c = a3 + b3 + c 3 = 2.
5
/
Prove that a, b, c ∈ ,2 .
4
(Vasile Cîrtoaje, 2011)
Solution. From
L
we get
.M
2(1 − a)2
bc = .
2−a
D
Thus,we can write the false inequality 4bc > (b + c)2 as
A
8(1 − a)2 PI
> (2 − a)2 ,
2−a
M
a(a2 + 2a − 4)
> 0,
LY
2−a
p p
a ∈ (−1 − 5, 0) ∪ (−1 + 5, 2).
O
/ (−1 −
a, b, c ∈ 5, 0) ∪ (−1 + 5, 2].
W
Since
5 p p
, 2 ⊂ (−1 − 5, 0) ∪ (−1 + 5, 2],
W
4
the conclusion follows.
Solution. For a = b = c, the equality holds. Otherwise, using the substitution m = k+2,
u = (1 − k)a, b = a + x, c = a + y, the inequality can be written as
Au2 + Bu + c ≥ 0,
where
A = x 2 − x y + y 2,
B = (x + y)(2A − mx y),
C = (x + y)2 (A − mx y) + m2 x 2 y 2 .
The quadratic Au2 + Bu + C has the discriminant
L
.M
D = B 2 − 4AC = −3m2 x 2 y 2 (x − y)2 .
Since A > 0 and D ≤ 0, the conclusion follows. The equality holds for a = b = c, and
D
for a/k = b = c or any cyclic permutation.
A
Remark 1. The inequality is equivalent to
X
a4 + k(k + 2)
X
a2 b2 + (1 − k2 )a bc
X PI
a ≥ (k + 1)
X
a b(a2 + b2 ).
M
For k = 0, we get Schur’s inequality of degree four
LY
X
a4 + b4 + c 4 + a bc(a + b + c) ≥ a b(a2 + b2 ).
O
X
a4 + b4 + c 4 + 3(a2 b2 + b2 c 2 + c 2 a2 ) ≥ 2 a b(a2 + b2 ),
with equality for a = b = c, and for a/2 = b = c (or any cyclic permutation).
Remark 2. The inequality in P 2.50 is equivalent to
X
(a − b)2 (a + b − c − kc)2 ≥ 0.
Symmetric Polynomial Inequalities in Real Variables 81
(Poland, 1992)
L
(b + c − a)2 (c + a − b)2 ≥ (b2 + c 2 − a2 )(c 2 + a2 − b2 ).
.M
We have
(c + a − b)2 (a + b − c)2 − (c 2 + a2 − b2 )(a2 + b2 − c 2 ) =
D
= [a2 − (b − c)2 ]2 − a4 + (b2 − c 2 )2 = 2(b − c)2 (b2 + c 2 − a2 ) ≥ 0.
A
PI
The equality holds for a = b = c, for a = 0 and b = c (or any cyclic permutation), and
for a = 0 and b + c = 0 (or any cyclic permutation).
M
LY
a2 (a − b)(a − c) ≥ .
a2 + b2 + c 2 + a b + bc + ca
W
X 1X
a2 (a − b)(a − c) = (b − c)2 (b + c − a)2 ,
2
W
which is an identity. The equality holds for a = b = c, for a = 0 and b = c (or any cyclic
permutation), and for a = 0 and b + c = 0 (or any cyclic permutation).
82 Vasile Cîrtoaje
a + b + c = p, a b + bc + ca = q,
where p and q are fixed real numbers satisfying p2 ≥ 3q. Prove that the product
r = a bc
L
p p
.M
p − 2 p2 − 3q p − p2 − 3q
a1 = , a2 = .
3 3
D
From
A
(b − c)2 = (b + c)2 − 4bc = (b + c)2 + 4a(b + c) − 4q
PI
= (p − a)2 + 4a(p − a) − 4q = −3a2 + 2pa + p2 − 4q ≥ 0,
M
we get a ≥ a1 , with equality for b = c. Similarly, from
LY
we get
W
and hence r(a) is minimal for a = a1 , when b = c, and is maximal for a = a2 , when
W
a = b.
Second Solution. From
we get r1 ≤ r ≤ r2 , where
Obviously, r attains its minimal and maximal values when two of a, b, c are equal; that
is, when either a = b or b = c. For a = b, from a + b + c = p and a b + bc + ca = q, we
get
p + 2 p2 − 3q
p p
p − p2 − 3q
a=b= , c= ,
3 3
(p − p2 − 3q)2 (p + 2 p2 − 3q)
p p
r= = r2 .
27
Similar, for b = c, we get
p+
p p
p2 − 3q p − 2 p2 − 3q
L
b=c= , a= ,
3 3
.M
(p + p2 − 3q)2 (p − 2
p p
p2 − 3q)
r= = r1 .
D
27
A
Remark. Using this result, we can prove the following generalization:
• Let a1 , a2 , . . . , an be real numbers such that
PI
M
a1 + a2 + · · · + an = p, a12 + a22 + · · · + an2 = p1 ,
LY
where p and p1 are fixed real numbers satisfying p2 ≤ np1 . Then, the product
O
r = a1 a2 · · · an
.M
of a1 , a2 , . . . , an are distinct, let a1 < a2 < a3 . According to P .53, the product r can be
W
First Solution. Let q = a b + bc + ca. We need to show that (q − 3)2 + 27 ≥ 27a bc.
According to P 2.53, for fixed q, the product a bc is maximal when two of a, b, c are
equal. Therefore, it suffices to prove the desired inequality for b = c; that is, to show
that (2a b + b2 − 3)2 ≥ 27(a b2 − 3) for a + 2b = 3. This inequality is equivalent to
(b − 1)2 (b + 2)2 ≥ 0.
The equality holds for a = b = c = 1, and also for a = 7 and b = c = −2 (or any cyclic
permutation).
Second Solution. Assume that a = max{a, b, c}, a ≥ 1. Since
L
1 1 3
3 − a b − bc − ca ≥ 3 − a(b + c) − (b + c)2 = 3 − a(3 − a) − (3 − a)2 = (a − 1)2
.M
4 4 4
and
1 1 1
D
a bc − 1 ≤ a(b + c)2 − 1 = a(3 − a)2 − 1 = (a − 1)2 (a − 4),
4 4 4
A
it suffices to prove that
9
(a − 1)4 ≥
27
PI
(a − 1)2 (a − 4),
16 4
M
which is equivalent to
LY
(a − 1)2 (a − 7)2 ≥ 0.
O
.M
First Solution. Let q = a b+ bc+ca. We need to show that q2 +9 ≥ 18a bc. According to
W
P 2.53, for fixed q, the product a bc is maximal when two of a, b, c are equal. Therefore,
it suffices to prove the desired inequality for b = c; that is, to show that (2a b+ b2 )2 +9 ≥
18a b2 for a + 2b = 3. This inequality is equivalent to
(b − 1)2 (b + 1)2 ≥ 0.
The equality holds for a = b = c = 1, and also for a = 5 and b = c = −1 (or any cyclic
permutation).
Second Solution. Using the substitution a = x + 1, b = y + 1, c = z + 1, we need to
show that
(x y + yz + z x)2 ≥ 12(x y + yz + z x) + 18x yz,
Symmetric Polynomial Inequalities in Real Variables 85
Since
3
y 2 + yz + z 2 ≥ ( y + z)2 ≥ 3 yz,
4
it suffices to show that
L
which is equivalent to
.M
9( yz + y + z)2 ≥ 0.
D
A
P 2.56. If a, b, c are real numbers such that a2 + b2 + c 2 = 9, then
PI
a bc + 10 ≥ 2(a + b + c).
M
LY
Æ
20 + 5a − a3 ≥ 4 2(9 − a2 )
W
(20 + 5a − a3 )2 ≥ 32(9 − a2 ).
L
Also, if a ≥ b ≥ 0 ≥ c, then
.M
a2 + b2 + c 2 + 3 − 2(a b + bc + ca) ≥ (a − b)2 + c 2 + 3 > 0.
Consider further that a ≥ b ≥ c ≥ 0. Let p = a + b + c, q = a b + bc + ca and r = a bc.
D
We need to show that p2 + 3 ≥ 4q for p + r = 4.
A
First Solution. By Schur’s inequality of degree three, we have p3 +9r ≥ 4pq. Therefore,
we get
PI
p(p2 + 3 − 4q) ≥ p3 + 3p − (p3 + 9r) = 12(p − 3).
M
To complete the proof, we need to show that p ≥ 3. By virtue of the AM-GM inequality,
LY
we get
p3 ≥ 27r,
O
p3 ≥ 27(4 − p),
.M
(p − 3)(p2 + 3p + 36) ≥ 0,
p ≥ 3.
W
suffices to show that p + r > 4. According to P 2.53, for fixed p and q, r is minimal
when two of a, b, c are equal. Therefore, it suffices to consider that b = c; that is, to
W
a b + bc + ca = 3a bc,
then
4(a2 + b2 + c 2 ) + 9 ≥ 7(a b + bc + ca).
(Vasile Cîrtoaje, 2011)
Solution. If one of a, b, c is 0, then the inequality is trivial. Otherwise, write the in-
equality in the homogeneous form
L
81a2 b2 c 2
4(a2 + b2 + c 2 ) + ≥ 7(a b + bc + ca),
.M
(a b + bc + ca)2
or
D
81a2 b2 c 2 ≥ (a b + bc + ca)2 A,
A
where
A = 7(a b + bc + ca) − 4(a2 + b2 + c 2 ).
PI
First Solution (by Vo Quoc Ba Can). Consider the nontrivial case A > 0. Substitut-
M
ing a, b, c by |a|, |b|, |c|, respectively, the left side of the inequality remains unchanged,
LY
while the right side remains unchanged or increases. Therefore, it suffices to prove the
inequality only for a, b, c > 0 and A > 0. Assume that a ≥ b ≥ c > 0. There are two
cases to consider.
O
2
(a b + bc + ca)2
2
4(a b + bc + ca) A ≤ + bA ,
b
W
(a b + bc + ca)2
18a bc ≥ + bA,
b
W
which is equivalent to
The equality holds for a = b = c = 1, and for a = 1/2 and b = c = 2 (or any cyclic
permutation).
Second Solution. Let p = a + b + c, q = a b + bc + ca, r = a bc. Since A = 15q − 4p2 ,we
L
need to show that
.M
81r 2 ≥ q2 (15q − 4p2 )
for all real numbers a, b, c. For fixed p and q, r 2 is minimal when r = 0, or when r is
D
either minimal or maximal. For a = 0, the inequality is true since
A
2(15q − 4p2 ) = −b2 − c 2 − 7(b − c)2 ≤ 0.
PI
According to P 2.53, r is minimal and maximal when two of a, b, c are equal. Therefore,
M
due to symmetry and homogeneity, it suffices to prove the inequality for b = c = 1. In
LY
a2 b2 c 2 + a2 b2 + b2 c 2 + c 2 a2 + a2 + b2 + c 2 ≥ a bc + a b + bc + ca + 3.
a bc + a2 b2 + b2 c 2 + c 2 a2 + a2 + b2 + c 2 ≥ a b + bc + ca + 4,
which is equivalent to
5(1 − a bc) ≥ q(3 − q),
Symmetric Polynomial Inequalities in Real Variables 89
L
P 2.60. Let f4 (a, b, c) be a symmetric homogeneous polynomial of degree four. Prove that
.M
the inequality f4 (a, b, c) ≥ 0 holds for all real numbers a, b, c if and only if f4 (a, 1, 1) ≥ 0
for all real a.
D
Solution. Let p = a + b + c, q = a b + bc + ca and r = a bc. Any symmetric homogeneous
A
polynomial f4 (a, b, c) can be written as
PI
f4 (a, b, c) = Apr + Bp4 + C p2 q + Dq2 ,
M
where A, B, C, D are real constants. For fixed p and q, the linear function g(r) =
Apr + Bp4 + C p2 q + Dq2 is minimal when r is either minimal or maximal. By P 2.53,
LY
r is minimal and maximal when two of a, b, c are equal. Since f4 (a, b, c) is symmet-
ric, homogeneous and satisfies f4 (−a, −b, −c) = f4 (a, b, c), it follows that the inequal-
O
ity f4 (a, b, c) ≥ 0 holds for all real numbers a, b, c if and only if f4 (a, 1, 1) ≥ 0 and
f4 (a, 0, 0) ≥ 0 for all real a. Notice that the condition " f4 (a, 0, 0) ≥ 0 for all real a"
.M
is not necessary because it follows from the condition " f4 (a, 1, 1) ≥ 0 for all real a" as
follows:
W
Remark. Similarly, we can prove the following more general statement, where f4 (a, b, c)
W
(a − 3)2 (5a2 − 3a + 1) ≥ 0.
L
This is true since
.M
3 2 11
5a2 − 3a + 1 = 5(a − ) + > 0.
10 20
The equality holds for a/3 = b = c (or any cyclic permutation).
D
A
PI
M
P 2.62. If a, b, c are real numbers such that a + b + c = 3, then
LY
3(a4 + b4 + c 4 ) + 33 ≥ 14(a2 + b2 + c 2 ).
O
To prove the desired inequality, we use the mixing variable method. We will show that
W
F (a, b, c) ≥ F (a, x, x) ≥ 0.
We have
Since
1
b2 + c 2 − 2x 2 = 2(x 2 − bc) = (b − c)2 ,
2
Symmetric Polynomial Inequalities in Real Variables 91
we get
1
F (a, b, c) − F (a, x, x) =(b − c)2 [3(b2 + c 2 + 2x 2 ) − 14 + 3(x 2 + bc)]
2
1
= (b − c)2 [3(x 2 − bc) + 18x 2 − 14] ≥ 0.
2
Also,
F (a, x, x) = F (3 − 2x, x, x) = 6(x − 1)2 (3x − 5)2 ≥ 0.
This completes the proof. The equality holds for a = b = c = 1, and for a = −1/3 and
b = c = 5/3 (or any cyclic permutation).
L
.M
Second Solution. Write the inequality in the homogeneous form
D
According to P 2.60, it suffices to prove this inequality for b = c = 1, when it becomes
A
PI
25a4 − 40a3 + 6a2 + 8a + 1 ≥ 0,
M
(a − 1)2 (5a + 1)2 ≥ 0.
LY
O
.M
(a − 1)2 (a − 2)2 ≥ 0.
p p
= b = c = ±1, forpa = 2 and b = c = 2/2 (or any cyclic
The equality holds for a p
permutation), and a = − 2 and b = c = − 2/2 (or any cyclic permutation).
92 Vasile Cîrtoaje
1 + α + β = 2γ.
The inequality
X X X X
a4 + α a2 b2 + β a bc a≥γ a b(a2 + b2 )
1 + α ≥ γ2 .
L
(Vasile Cîrtoaje, 2009)
.M
Solution. Let
D
X X X X
f4 (a, b, c) = a4 + α a2 b2 + β a bc a−γ a b(a2 + b2 .
A
According to P 2.60, the inequality f4 (a, b, c) ≥ 0 holds for any real numbers a, b, c if
and only if f4 (a, 1, 1) ≥ 0 for any real a. From PI
M
f4 (a, 1, 1) = (a − 1)2 [(a − γ + 1)2 + 1 + α − γ2 ],
LY
X X X X
a4 + k(k + 2) a2 b2 + (1 − k2 )a bc a ≥ (k + 1) a b(a2 + b2 ), k ∈ R,
.M
X
(a − b)(a − c)(a − k b)(a − kc) ≥ 0,
W
where the equality holds for a = b = c, and also for a/k = b = c (or any cyclic permu-
tation). In addition, for k = 0, we get Schur’s inequality of degree four
W
X
a2 (a − b)(a − c),
with equality for a = b = c, and also for a = 0 and b = c (or any cyclic permutation).
L
.M
P 2.66. If a, b, c are real numbers, then
D
A
4 4
(a + b)4 + (b + c)4 + (c + a)4 ≥ (a + b4 + c 4 ).
7
PI (Vietnam TST, 1996)
M
Solution. Denote the left side of the inequality by f4 (a, b, c). According to P 2.60, it
LY
2
f4 (a, 1, 1) = (5a4 + 28a3 + 42a2 + 28a + 59) > 0
7
.M
23 2 2
5a4 + 28a3 + 42a2 + 28a + 59 = (5a2 − 2a)(a + 3)2 + 9(a + ) + > 0.
9 9
W
p = a + b + c, q = a b + bc + ca, r = a bc.
Prove that
p2 + q2 − pq
(3 − p)r + ≥ q.
3
(Vasile Cîrtoaje, 2011)
94 Vasile Cîrtoaje
1X 1X 2 X
(b − c)2 + a (b − c)2 ≥ a(b − c)2 .
2 2
According to the AM-GM inequality, it suffices to prove that
rX X X
(b − c)2 a2 (b − c)2 ≥ a(b − c)2 .
Clearly, this inequality follows immediately from the Cauchy-Schwarz inequality. The
L
equality holds for a = b = c, and for b = c = 1 (or any cyclic permutation).
.M
Second Solution Write the inequality as f4 (a, b, c) ≥ 0, where
D
A
is a symmetric polynomial of degree four in a, b, c. According to Remark from the proof
PI
of P 2.60, it suffices to prove that f4 (a, b, b) ≥ 0 for all real numbers a and b. Indeed,
we have
M
f4 (a, b, b) = (a − b)2 (b − 1)2 ≥ 0.
LY
O
First Solution. We try to get a stronger homogeneous inequality of third order. Accord-
W
L
3(a bc − 1)2 + f4 (a, b, c) ≥ 0,
.M
where X X 4X
D
f4 (a, b, c) = a2 b2 + a2 + 2a bc − a b(a + b)
3
A
is a symmetric polynomial of degree four. Clearly, it suffices to prove that f4 (a, b, c) ≥ 0.
PI
According to Remark from P 2.60, it suffices to prove that f4 (a, b, b) ≥ 0 for all real
numbers a and b. Indeed, we have
M
3 f4 (a, b, b) = (6b2 − 8b + 3)a2 − 2b2 a + b2 (3b2 − 8b + 6)
LY
2
b2 18b2 (b − 1)4
O
2
= (6b − 8b + 3) a − 2 + ≥ 0.
6b − 8b + 3 6b2 − 8b + 3
.M
1 1 1 1 1 1 1
a + − 1 b + − 1 c + − 1 + 2 ≥ (a + b + c) + + .
a b c 3 a b c
4pq
r 2 + (4 − p − q)r + p2 + q2 − − p − q + 1 ≥ 0.
3
96 Vasile Cîrtoaje
Since the equality holds for a = b = c = 1, that is, for p = q = 3 and r = 1, we write
the inequality as
p − q 2
r −1+ + f (p, q, r) ≥ 0,
2
where
L
f4 (a, b, c) = 3(3 − p)r + p2 + q2 − pq − 3q ≥ 0.
.M
According to Remark from P 2.60, it suffices to prove that f4 (a, b, b) ≥ 0 for all real
D
numbers a and b. Indeed, we have
A
f4 (a, b, b) = (b − 1)2 (a − b)2 ≥ 0.
1 1 1
.M
a + + k b + + k c + + k + (1 − k)(2 + k)2 ≥
a b c
1 1 1 1
W
2
≥ (2 + k) (a + b + c) + + .
3 a b c
W
W
1 1 1 1 1 1
a2 + b2 + c2 + ≥ a+b− b+c− c+a− .
2 2 2 2 2 2
where
Y 1 1 1 a+b+c 2
Y
2
f4 (a, b, c) = a + − a+b− − a bc + −
2 2 2 2
is a symmetric polynomial of degree four. Clearly, it suffices to prove that f4 (a, b, c) ≥ 0.
According to Remark from P 2.60, it suffices to prove that f4 (a, b, b) ≥ 0 for all real
numbers a and b. Indeed, we have
L
.M
(2a bc + 1 − a − b − c)2 + 2(a b + bc + ca − a − b − c)2 ≥ 0.
D
A
P 2.71. If a, b, c are real numbers such that a + b + c = 3, then
PI
a(a − 1) b(b − 1) c(c − 1)
M
+ + 2 ≥ 0.
8a2 + 9 8b2 + 9 8c + 9
LY
Solution (by Michael Rozenberg). We see that the equality holds for a = b = c, and for
a = 3/2 and b = c = 3/4 (or any cyclic permutation). Let k be a positive constant,
.M
≥ 0,
8a2 + 9
W
X (k2 − 8)a(a − 1)
+ 1 ≥ 3.
8a2 + 9
W
Choosing p
k =3+ 17,
the inequality can be written as
X (ka − 3)2
≥ 3.
8a2 + 9
Let m be a real constant. According to the Cauchy-Schwarz inequality, we have
L
X
.M
−3 (ka + a + b + c)2 [8a2 + (a + b + c)2 ].
According to P 2.60, it suffices to prove that f4 (a, 1, 1) ≥ 0 for all real a. Indeed, this
D
inequality is equivalent to (a − 1)2 (a − 2)2 ≥ 0.
A
PI
P 2.72. If a, b, c are real numbers such that a + b + c = 3, then
M
(a − 11)(a − 1) (b − 11)(b − 1) (c − 11)(c − 1)
+ + ≥ 0.
LY
2a2 + 1 2b2 + 1 2c 2 + 1
(Vasile Cîrtoaje, 2013)
O
X (a − 11)(a − 1)
+ 1 ≥ 3,
2a2 + 1
W
X (a − 2)2
≥ 1.
2a2 + 1
W
[ (a − 2)2 ]2
P
X (a − 2)2
≥P ,
2a2 + 1 (a − 2)2 (2a2 + 1)
Therefore, it suffices to show that
X X X X
(a2 + b2 + c 2 )2 ≥ 2 a4 − 8 a3 + 9 a2 − 4 a + 12.
According to P 2.60, it suffices to prove that f4 (a, 1, 1) ≥ 0 for all real a. Indeed,
L
.M
(Vasile Cîrtoaje, 1994)
D
Solution. We will prove the sharper inequality f4 (a, b, c) ≥ 0, where
A
2
a+b+c
2 2 2
f4 (a, b, c) = (a + 2)(b + 2)(c + 2) − 9(a b + bc + ca) − a bc −
PI .
3
M
Since f4 (a, b, c) is a symmetric polynomial of degree four, according to Remark from P
2.60, it suffices to prove that f4 (a, b, b) ≥ 0 for all real numbers a and b. For fixed b,
LY
Indeed, we have
It suffices to prove that there exists a positive number k such that f4 (a, b, c) ≥ 0, where
According to P 2.60, the inequality f4 (a, b, c) ≥ 0 holds for all real a, b, c if and only if
f4 (a, 1, 1) ≥ 0 for all real a. We have
L
.M
f4 (a, 1, 1) = (a − 1)2 (2a + 1)(2a + 3) − k(2a + 1)(a − 1)2
= (a − 1)2 (2a + 1)(2a + 3 − k).
D
Setting k = 2, we get
A
f4 (a, 1, 1) = (a − 1)2 (2a + 1)2 ≥ 0.
The equality holds for a = b = c = ±1.
PI
M
LY
P 2.75. Any sixth degree symmetric homogeneous polynomial f6 (a, b, c) can be written in
O
the form
f6 (a, b, c) = Ar 2 + B(p, q)r + C(p, q),
.M
p = a + b + c, q = a b + bc + ca, r = a bc.
W
In the case A ≤ 0, prove that the inequality f6 (a, b, c) ≥ 0 holds for all real numbers a, b, c
if and only if f6 (a, 1, 1) ≥ 0 for all real a.
W
for all real a" is not necessary because it follows from the condition " f6 (a, 1, 1) ≥ 0 for
all real a" as follows:
L
X
a3 b3 = 3r 2 − 3pqr + q3 ,
.M
X
a2 b2 (a2 + b2 ) = −3r 2 − 2(p3 − 2pq)r + p2 q2 − 2q3 ,
X
a b(a4 + b4 ) = −3r 2 − 2(p3 − 7pq)r + p4 q − 4p2 q2 + 2q3 ,
D
X
A
a6 = 3r 2 + 6(p3 − 2pq)r + p6 − 6p4 q + 9p2 q2 − 2q3 .
PI
According to these relations, the highest coefficient A of the polynomuial f6 (a, b, c) has
the expression
M
(B) A = 3(A1 − A2 − A3 + A4 + A5 − A6 + A7 ).
LY
X
P1 (a, b, c) = (A1 a2 + A2 bc)(B1 a2 + B2 bc)(C1 a2 + C2 bc)
.M
Indeed, since
W
X X X X
P1 (a, b, c) =A1 B1 C1 a6 + A2 B2 C2 b3 c 3 + A1 B1 C2 a bc a3
X
+3 A1 B2 C2 a2 b2 c 2 ,
we have
X X
A = 3A1 B1 C1 + 3A2 B2 C2 + 3 A1 B1 C2 + 3 A1 B2 C2
= 3(A1 + A2 )(B1 + B2 )(C1 + C2 ).
With regard to
L
.M
P4 (a, b, c) = (a − b)2 (b − c)2 (c − a)2 ,
from
D
A
P4 (a, b, c) = (p2 − 2q − c 2 − 2a b)(p2 − 2q − a2 − 2bc)(p2 − 2q − b2 − 2ca),
PI
it follows that P4 has the same highest coefficient as (−c 2 −2a b)(−a2 −2bc)(−b2 −2ca);
that is,
M
A = (−1 − 2)3 = −27.
LY
coefficient A ≤ 0, and let k1 , k2 be two fixed real numbers. The inequality f6 (a, b, c) ≥ 0
.M
k1 (a + b + c)2 + k2 (a b + bc + ca) ≥ 0,
W
Notice that the condition " f6 (a, 0, 0) ≥ 0 for all real a satisfying k1 a2 ≥ 0" is not
necessary because it follows from the condition " f6 (a, 1, 1) ≥ 0 for all real a satisfying
W
k1 (a + 2)2 + k2 (2a + 1) ≥ 0". Indeed, for the non-trivial case k1 ≥ 0, when the condition
" f6 (a, 0, 0) ≥ 0 for all real a satisfying k1 a2 ≥ 0" becomes " f6 (a, 0, 0) ≥ 0 for all real a",
we have
f6 (a, 0, 0) = lim f6 (a, t, t) = lim t 6 f6 (a/t, 1, 1) ≥ 0.
t→0 t→0
Remark 4. The statement in P 2.75 and its extension in Remark 3 are also valid in the
more general case when f6 (a, b, c) is a symmetric homogeneous function of the form
L
.M
From Y Y
(b2 + c 2 ) = (p2 − 2q − a2 ),
D
it follows that f6 (a, b, c) has the highest coefficient A = −5. Then, by P 2.75, it suffices
A
to prove that f6 (a, 1, 1) ≥ 0 for all real a. Indeed, we have
PI
f6 (a, 1, 1) = 2(a + 3)2 (5a2 + 2a + 1) ≥ 0.
M
The homogeneous inequality f6 (a, b, c) ≥ 0 is an equality for −a/3 = b = c (or any
LY
Y Y
(b2 + bc + c 2 ) = (p2 − 2q + bc − a2 ),
W
it follows that f6 (a, b, c) has the same highest coefficient as P3 (a, b, c), where
Y
P3 (a, b, c) = (bc − a2 );
A = P3 (1, 1, 1) = (1 − 1)3 = 0.
Then, by P 2.75, it suffices to prove that f6 (a, 1, 1) ≥ 0 for all real a. Indeed, we have
where
p = a + b + c, q = a b + bc + ca, r = a bc.
Therefore,
1 2X 2
L
f6 (a, b, c) = p2 (q2 − pr) = p a (b + c)2 ≥ 0.
2
D
.M
A
P 2.77. If a, b, c are real numbers, then
(a)
P PI
a2 (a − b)(a − c)(a + 2b)(a + 2c) + (a − b)2 (b − c)2 (c − a)2 ≥ 0;
M
a2 (a − b)(a − c)(a − 4b)(a − 4c) + 7(a − b)2 (b − c)2 (c − a)2 ≥ 0.
P
(b)
LY
X
f (a, b, c) = a2 (a − b)(a − c)(a − k b)(a − kc).
W
Since X
f (a, b, c) = a2 (a2 + 2bc − q)[a2 + (k + k2 )bc − kq],
W
a2 (a − 1)2 (a + 2)2 ≥ 0.
Symmetric Polynomial Inequalities in Real Variables 105
The equality holds for a = b = c, for a + b + c = 0, and for a = 0 and b = c (or any
cyclic permutation).
(b) For k = 4 and m = 7, we get A = 0. Then, by P 2.75, it suffices to prove the
original inequality for b = c = 1; that is,
a2 (a − 1)2 (a − 4)2 ≥ 0.
L
.M
and X X X X 2
a2 − ab a2 − 2 ab ≥ 0,
D
respectively.
A
PI
P 2.78. If a, b, c are real numbers, then
M
(a2 + 2bc)(b2 + 2ca)(c 2 + 2a b) + (a − b)2 (b − c)2 (c − a)2 ≥ 0.
LY
First Solution. Denote the left side of the inequality by f6 (a, b, c). According to Remark
.M
A = (1 + 2)3 − 27 = 0.
W
Then, by P 2.75, it suffices to prove that f6 (a, 1, 1) ≥ 0 for all real a. Indeed,
W
or
[(b + c)a + bc]2 ≥ 0,
which is clearly true. The equality holds for a b + bc + ca = 0.
Remark 1. The inequality is equivalent to
L
Remark 2. The inequality in P 2.78 is a particular case of the following more general
.M
statement.
• If a, b, c are real numbers and
D
9k2 (k2 − k + 1)
A
, 1≤k≤2
4(k + 1)3
αk = PI,
k2
, k≥2
4
M
then
LY
with equality for −ka = b = c, and for b = c = 0 (or any cyclic permutation).
.M
(2a2 + 5a b + 2b2 )(2b2 + 5bc + 2c 2 )(2c 2 + 5ca + 2a2 ) + (a − b)2 (b − c)2 (c − a)2 ≥ 0.
W
Therefore, according to Remark 2 from P 2.75, f6 (a, b, c) has the highest coefficient
A = P3 (1, 1, 1) − 27 = 0.
Then, by P 2.75, it suffices to prove that f6 (a, 1, 1) ≥ 0 for all real a. Indeed,
L
(a + b + c)2 (a b + bc + ca)2 ≥ 0.
.M
Remark 2. The following more general statement holds.
D
• Let a, b, c be real numbers. If k > −2, then
A
Y
4 (b2 + k bc + c 2 ) ≥ (2 − k)(a − b)2 (b − c)2 (c − a)2 .
Notice that this inequality is equivalent to
PI
M
(k + 2)[(a + b + c)(a b + bc + ca) − (5 − 2k)a bc]2 ≥ 0.
LY
O
.M
2 2 2 64 2
W
2 2 2
2 2 2 2 2 2
f (a, b, c) = a + a b + b b + bc + c c + ca + a .
3 3 3
64 2
f6 (a, b, c) = f (a, b, c) − (a + bc)(b2 + ca)(c 2 + a b).
27
Since
2 2 2
f (a, b, c) = p2 − 2q + a b − c 2 p2 − 2q + bc − a2 p2 − 2q + ca − b2 ,
3 3 3
108 Vasile Cîrtoaje
L
f6 (a, 1, 1) = a + a+1 − (a + 1)(a + 1)2 = (a − 1)4 ≥ 0.
3 3 27 27
.M
The equality holds for a = b = c.
D
A
P 2.81. If a, b, c are real numbers, then PI
2(a − b)2 (b − c)2 (c − a)2
M
X
a2 (a − b)(a − c) ≥ .
a2 + b2 + c 2
LY
X
f6 (a, b, c) = (a2 + b2 + c 2 ) a2 (a − b)(a − c) − 2(a − b)2 (b − c)2 (c − a)2 .
.M
A = −2(−27) = 54.
Since A > 0, we will use the highest coefficient cancellation method. It is easy to check
W
that
W
f (1, 1, 1) = 0, f (0, 1, 1) = 0.
Therefore, we define the symmetric homogeneous polynomial of degree three
P(a, b, c) = r + Bp3 + C pq
Since g6 (a, b, c) has the highest coefficient A1 = 0, it suffices to show that g6 (a, 1, 1) ≥ 0
for all real a (see P 2.75). Indeed, we have
1
f6 (a, 1, 1) = a2 (a2 + 2)(a − 1)2 , P(a, 1, 1) = a(a − 1)2 ,
81
hence
1 2
g6 (a, 1, 1) = f6 (a, 1, 1) − 54P(a, 1, 1) = a (a − 1)2 (a + 2)2 ≥ 0.
3
The equality holds for a = b = c, for a = 0 and b = c (or any cyclic permutation), and
also for a = 0 and b + c = 0 (or any cyclic permutation).
L
Remark. In the same manner, we can prove the following generalization (Vasile Cîrtoaje,
.M
2014).
• Let x, y, z be real numbers. If k ∈ [−1, 2), then
D
A
X (2 − k)(x − y)2 ( y − z)2 (z − x)2
x 2 (x − y)(x − z) ≥ ,
x 2 + y 2 + z 2 + k(x y + yz + z x)
PI
with equality for x = y = z, and for x = 0 and y 2 = z 2 (or any cyclic permutation).
M
LY
O
Solution. Let
W
X
f6 (a, b, c) = (a2 + b2 + c 2 ) (a − b)(a − c)(a − 2b)(a − 2c) − 8(a − b)2 (b − c)2 (c − a)2 .
W
A = (−8)(−27) = 216.
Since A > 0, we will use the highest coefficient cancellation method. Since
f (1, 1, 1) = 0, f (2, 1, 1) = 0,
such that P(1, 1, 1) = 0 and P(2, 1, 1) = 0. We get B = 1/18 and C = −5/18, hence
1 5
P(a, b, c) = a bc + (a + b + c)3 − (a + b + c)(a b + bc + ca).
18 18
Consider now the sharper inequality g6 (a, b, c) ≥ 0, where
Clearly, g6 (a, b, c) has the highest coefficient A1 = 0. By P 2.75, it suffices to prove that
g6 (a, 1, 1) ≥ 0 for all real a. We have
L
f6 (a, 1, 1) = (a2 + 2)(a − 1)2 (a − 2)2 , P(a, 1, 1) = (a − 1)2 (a − 2),
18
.M
hence
1
g6 (a, 1, 1) = f6 (a, 1, 1) − 216P 2 (a, 1, 1) =
(a − 1)2 (a2 − 4)2 ≥ 0.
D
3
A
The equality holds for a = b = c, for a = 0 and b + c = 0 (or any cyclic permutation),
and also for a/2 = b = c (or any cyclic permutation). PI
M
Remark. In the same manner, we can prove the following generalization (Vasile Cîrtoaje,
2014).
LY
with equality for x = y = z, for x/k = y = z (or any cyclic permutation) if k 6= 0, and for
x = 0 and y + z = 0 (or any cyclic permutation).
W
W
W
a2 + 3bc b2 + 3ca c 2 + 3a b
+ 2 + 2 ≥ 0.
b2 + c 2 c + a2 a + b2
(Vasile Cîrtoaje, 2014)
Let
p = a + b + c, q = a b + bc + ca, r = a bc.
Symmetric Polynomial Inequalities in Real Variables 111
From X
f6 (a, b, c) = (a2 + 3bc)(p2 − 2q − c 2 )(p2 − 2q − b2 ),
it follows that f6 (a, b, c) has the same highest coefficient A as f (a, b, c), where
X X
f (a, b, c) = (a2 + 3bc)b2 c 2 = 3r 2 + 3 b3 c 3 = 12r 2 − 9pqr + 3q3 ;
that is, A = 12. Since A > 0, we will use the highest coefficient cancellation method. It is
easy to check that
f6 (−1, 1, 1) = 0.
So, we define the homogeneous polynomial
L
.M
P(a, b, c) = r + Bp3 + (B − 1)pq,
which satisfies the property P(−1, 1, 1) = 0. We will show that there is at least a real
D
value of B such that the following sharper inequality holds
A
f6 (a, b, c) ≥ 12P 2 (a, b, c).
PI
M
Let us denote
g6 (a, b, c) = f6 (a, b, c) − 12P 2 (a, b, c).
LY
Clearly, g6 (a, b, c) has the highest coefficient A1 = 0. By P 2.75, it suffices to prove that
g6 (a, 1, 1) ≥ 0 for all real a. We have
O
and
W
where
g(a) = (a2 + 1)(a2 − 2a + 7) − 12[B(a + 2)(a + 5) − 2(a + 1)]2 .
Choosing B = 1/4, we get
hence g6 (a, 1, 1) ≥ 0 for all real a. The proof is completed. The equality holds for
−a = b = c (or any cyclic permutation).
112 Vasile Cîrtoaje
a2 + 6bc b2 + 6ca c 2 + 6a b
+ + ≥ 0.
b2 − bc + c 2 c 2 − ca + a2 a2 − a b + b2
(Vasile Cîrtoaje, 2014)
Let
p = a + b + c, q = a b + bc + ca, r = a bc.
L
.M
From X
f6 (a, b, c) = (a2 + 6bc)(p2 − 2q − c 2 − a b)(p2 − 2q − b2 − ac),
D
it follows that f6 (a, b, c) has the same highest coefficient A as f (a, b, c), where
A
X
f (a, b, c) = (a2 + 6bc)(b2 + ca)(c 2 + a b);
Since A > 0, we use the highest coefficient cancellation method. We will show that there
O
are two real numbers B and C such that the following sharper inequality holds
where
P(a, b, c) = r + Bp3 + C pq.
W
Let us denote
W
Clearly, g6 (a, b, c) has the highest coefficient equal to zero. Then, it suffices to prove
that g6 (a, 1, 1) ≥ 0 for all real a.
We have
g6 (a, 1, 1) = f6 (a, 1, 1) − 84P 2 (a, 1, 1),
where
f6 (a, 1, 1) = (a2 − a + 1)(a2 + a + 1)(a2 − 2a + 8)
and
P(a, 1, 1) = a + B(a + 2)3 + C(a + 2)(2a + 1).
Let us denote g(a) = g6 (a, 1, 1). Since g(−2) = 0, we can have g(a) ≥ 0 in the vicinity
of a = −2 only if g 0 (−2) = 0, which involves C = −61/168. On the other hand,
Symmetric Polynomial Inequalities in Real Variables 113
from g(1) = 0, we get B = 155/1512. Using these values of B and C, the inequality
g6 (a, 1, 1) ≥ 0 is equivalent to
that is,
(a + 2)2 (a − 1)2 (3191a2 − 8734a + 49391) ≥ 0,
which is true for all real a.
The proof is completed. The equality holds for a = b+c = 0 (or any cyclic permutation).
L
.M
P 2.85. If a, b, c are real numbers such that a b + bc + ca ≥ 0, then
D
4a2 + 23bc 4b2 + 23ca 4c 2 + 23a b
A
+ + ≥ 0.
b2 + c 2 c 2 + a2 a2 + b2
PI (Vasile Cîrtoaje, 2014)
M
Solution. Write the inequality as f6 (a, b, c) ≥ 0, where
LY
X
f6 (a, b, c) = (4a2 + 23bc)(a2 + b2 )(a2 + c 2 ).
O
Let
p = a + b + c, q = a b + bc + ca, r = a bc.
.M
From X
f6 (a, b, c) = (4a2 + 23bc)(p2 − 2q − c 2 )(p2 − 2q − b2 ),
W
it follows that f6 (a, b, c) has the same highest coefficient A as f (a, b, c), where
W
X X
f (a, b, c) = (4a2 + 23bc)b2 c 2 = 12r 2 + 23 b3 c 3 = 81r 2 − 69pqr + 23q3 ;
W
that is, A = 81. Since A > 0, we will use the highest coefficient cancellation method. It is
easy to check that
f (−1, 2, 2) = 0.
Therefore, define the homogeneous polynomial
4 3
P(a, b, c) = r + p + C pq,
27
which satisfies the property P(−1, 2, 2) = 0. We will show that there is at least a real C
such that the following sharper inequality holds for a b + bc + ca ≥ 0:
Let us denote
g6 (a, b, c) = f6 (a, b, c) − 81P 2 (a, b, c).
Clearly, g6 (a, b, c) has the highest coefficient A1 = 0. Then, by Remark 3 from P 2.75, it
suffices to prove that g6 (a, 1, 1) ≥ 0 for all real a such that 2a + 1 ≥ 0.
We have
f6 (a, 1, 1) = (2a + 1)(a2 + 1)(2a3 − a2 + 14a + 39),
1
P(a, 1, 1) = (2a + 1)[2a2 + (27C + 11)a + 54C + 32],
27
g6 (a, 1, 1) = f6 (a, 1, 1) − 81P 2 (a, 1, 1).
L
From the condition g6 (1, 1, 1) = 0, we get C = −1/3. For this value of C, we find
.M
2
P(a, 1, 1) = (2a + 1)(a2 + a + 7),
27
D
then
A
1
g6 (a, 1, 1) =
9
PI
(2a + 1)(10a5 − 29a4 + 16a3 + 170a2 − 322a + 155)
M
1
= (2a + 1)(a − 1)2 (10a3 − 9a2 − 12a + 155).
9
LY
We need to show that 10a3 − 9a2 − 12a + 155 ≥ 0 for a ≥ −1/2. This is clearly true for
−1/2 ≤ a ≤ 0. Also, for a > 0, we have
O
The proof is completed. The equality holds for −2a = b = c (or any cyclic permutation).
W
W
A = 20 · 3 + 43 · 3 = 189,
Symmetric Polynomial Inequalities in Real Variables 115
it follows that
f6 (1, 1, 1) = 0, f6 (−1/2, 1, 1) = 0.
Define the homogeneous function
L
4 3 5
P(a, b, c) = r + p − pq,
.M
27 9
hence
D
27a + 4(a + 2)3 − 15(a + 2)(2a + 1) 2(a − 1)2 (2a + 1)
P(a, 1, 1) = = .
A
27 27
PI
We will show that the following sharper inequality holds for a b + bc + ca ≥ 0:
Since the highest coefficient of g6 (a, b, c) is zero, it suffices to prove that g6 (a, 1, 1) ≥ 0
for all real a such that 2a + 1 ≥ 0 (see Remark 3 from P 2.75). We have
.M
where
W
28
g(a) = 10a3 + 15a2 + 44a + 33 − (a − 1)2 (2a + 1).
27
W
Since
W
g(a) ≥ 10a3 + 15a2 + 44a + 33 − 5(a − 1)2 (2a + 1) = 22(a + 1)2 + 8a2 + 6 > 0,
we have g6 (a, 1, 1) ≥ 0 for all a ≥ −1/2. Thus, the proof is completed. The equality
holds for a = b = c = 1.
L
the same highest coefficient A0 as g(a, b, c), where
.M
X
g(a, b, c) = (a2 + bc)(a2 + 2bc)[a2 + 12bc];
D
that is, according to Remark 2 from P 2.75,
A
PI
A0 = g(1, 1, 1) = 3 · 2 · 3 · 13 = 234.
M
Therefore, f6 (a, b, c) has the highest coefficient
LY
Since the highest coefficient A is positive, we will use the highest coefficient cancellation
method. There are two cases to consider: q ≥ 0 and q < 0.
.M
Case 1: q ≥ 0. Since
f6 (1, 1, 1) = f6 (3, 1, 1) = 0,
W
P(a, b, c) = r + Bp3 + C pq
W
2 3 11
P(a, b, c) = r + p − pq,
45 45
hence
Let us denote
g6 (a, b, c) = f6 (a, b, c) − 1125P 2 (a, b, c).
Since the highest coefficient of g6 (a, b, c) is zero, it suffices to prove that g6 (a, 1, 1) ≥ 0
for all real a such that 2a + 1 ≥ 0 (see Remark 3 from P 2.75). We have
1
3
P(a, b, c) = r + Bp − 3B + pq,
L
9
.M
which satisfies P(1, 1, 1) = 0. We will show that there is a real number B such that the
following sharper inequality holds
D
f6 (a, b, c) ≥ 1125P 2 (a, b, c).
A
Let us denote PI
g6 (a, b, c) = f6 (a, b, c) − 1125P 2 (a, b, c).
M
Clearly, g6 (a, b, c) has the highest coefficient equal to zero. Then, by Remark 3 from P
2.75, it suffices to prove that g6 (a, 1, 1) ≥ 0 for 2a + 1 < 0. We have
LY
where
.M
2
1
2 3
P (a, 1, 1) = a + B(a + 2) − 3B + (a + 2)(2a + 1) .
9
Let us denote g(a) = g6 (a, 1, 1). Since g(−2) = 0, we can have g(a) ≥ 0 in the vicinity
W
25 · 729
5
2 2 2
g6 (a, 1, 1) = 4(a − 1) (a + 1)(a − 3) − (a − 1)2 (4a − 7)2
81
4
= (a − 1)2 (a + 2)2 (a2 − 50a + 121) ≥ 0.
81
The proof is completed. The equality holds for a = b = c, for a/3 = b = c (or any cyclic
permutation), and for a = 0 and b + c = 0 (or any cyclic permutation).
118 Vasile Cîrtoaje
P 2.88. Let a, b, c be real numbers such that a b + bc + ca ≥ 0. For any real k, prove that
X
4bc(a − b)(a − c)(a − k b)(a − kc) + (a − b)2 (b − c)2 (c − a)2 ≥ 0.
L
Let
.M
p = a + b + c, q = a b + bc + ca, r = a bc.
Since (a − b)(a − c) = a2 + 2bc − q and (a − k b)(a − kc) = a2 + (k + k2 )bc − kq, f (a, b, c)
D
has the same highest coefficient A0 as P1 (a, b, c), where
A
X
P1 (a, b, c) = bc(a2 + 2bc)[a2 + (k + k2 )bc];
PI
that is, according to Remark 2 from P 2.75,
M
A0 = P1 (1, 1, 1) = 3(1 + 2)(1 + k + k2 ) = 9(1 + k + k2 ).
LY
We have
f6 (a, 1, 1) = 4(a − 1)2 (a − k)2 .
W
Consider first that k = −1/2, when A = 0. By P 2.75, it suffices to prove that f6 (a, 1, 1) ≥
W
0 for all real a. Clearly, these conditions are fulfilled. Consider further that k 6= −1/2,
when the highest coefficient A is positive. We will use the highest coefficient cancellation
W
method. Since
f6 (1, 1, 1) = f6 (k, 1, 1) = 0,
define the homogeneous function
q2
P(a, b, c) = r + C pq + D
p
pq 2(k + 2)q2
P(a, b, c) = r + − .
3(2k + 1) 3(2k + 1)p
Symmetric Polynomial Inequalities in Real Variables 119
Let us denote
g6 (a, b, c) = f6 (a, b, c) − 9(2k + 1)2 P 2 (a, b, c).
Clearly, g6 (a, b, c) has the highest coefficient A1 = 0. Then, by Remark 4 from P 2.75, it
suffices to prove that g6 (a, 1, 1) ≥ 0 for all real a such that 2a + 1 ≥ 0. We have
L
2(a − 1) (a − k)
2
.M
= ,
3(2k + 1)(a + 2)
D
then
A
g6 (a, 1, 1) = f6 (a, 1, 1) − 9(2k + 1)2 P 2 (a, 1, 1)
=
PI
12(a − 1)2 (a − k)2 (2a + 1)
≥ 0.
(a + 2)2
M
The proof is completed. The equality holds for a = b = c, for a/k = b = c (or any cyclic
LY
2
(a2 b + b2 c + c 2 a) + (a b2 + bc 2 + ca2 ) ≥ 4(a b + bc + ca)(a2 b2 + b2 c 2 + c 2 a2 ).
W
First Solution. Consider the nontrivial case a b + bc + ca ≥ 0, and write the inequality
W
as f6 (a, b, c) ≥ 0, where
Since
f6 has the highest coefficient A = (−3)2 = 9. Since A > 0, we will use the highest
coefficient cancellation method. Because
f (1, 1, 1) = f (0, 1, 1) = 0,
120 Vasile Cîrtoaje
(a b + bc + ca)2
P(a, b, c) = a bc + C(a + b + c)(a b + bc + ca) + D
a+b+c
such that P(1, 1, 1) = P(0, 1, 1) = 0; that is,
L
f6 (a, b, c) ≥ 9P 2 (a, b, c).
.M
Let us denote
g6 (a, b, c) = f6 (a, b, c) − 9P 2 (a, b, c).
D
Clearly, g6 (a, b, c) has the highest coefficient A1 = 0. Then, by Remark 4 from P 2.75, it
A
suffices to prove that g6 (a, 1, 1) ≥ 0 for all real a such that 2a + 1 ≥ 0. We have
PI
f6 (a, 1, 1) = 4a2 (a − 1)2 ,
M
(a + 2)(2a + 1) 4(2a + 1)2 2a(a − 1)2
P(a, 1, 1) = a + − = ,
LY
3 3(a + 2) 3(a + 2)
hence
O
The proof is completed. The equality holds for a = b = c, for a = 0 and b = c (or any
cyclic permutation), and for b = c = 0 (or any cyclic permutation).
W
Second Solution (by Nguyen Van Quy). Since the inequality remains unchanged by
replacing a, b, c with −a, −b, −c, we may assume that a+ b+c ≥ 0. In addition, consider
W
Solution. Let p = a+b+c. Write the inequality in the homogeneous form f6 (a, b, c) ≥ 0,
where X
f6 (a, b, c) = (3a − p)(3a − 25p)(9b2 + 23p2 )(9c 2 + 23p2 ).
L
A = 3 · 93 ,
.M
we use the highest coefficient cancellation method. Thus, we will prove that there exist
D
two real numbers B and C such that g6 (a, b, c) ≥ 0, where
A
g6 (a, b, c) = f6 (a, b, c) − A[a bc + B(a + b + c)3 + C(a + b + c)(a b + bc + ca)]2 .
PI
Since g6 has the highest coefficient equal to zero, it suffices to show that g6 (a, 1, 1) ≥ 0
M
for all real a. Notice that
LY
and
g6 (a, 1, 1) = f6 (a, 1, 1) − 3 · 93 [a + B(a + 2)3 + C(a + 2)(2a + 1)]2 .
.M
Let us denote g(a) = g6 (a, 1, 1). Since g(−2) = 0, we can have g(a) ≥ 0 in the vicinity
of a = −2 only if g 0 (−2) = 0; this involves C = −13/9, hence
W
Case 1: 5p2 + q ≤ 0. By Remark 3 from the proof of P 2.75, we only need to show that
there exist a real number B such that g6 (a, 1, 1) ≥ 0 for all real a satisfying 5(a + 2)2 +
2a + 1 ≤ 0; that is, for a ∈ [−3, −7/5]. From g6 (−11/7, 1, 1) = 0, we get B = 28/9,
then
g6 (a, 1, 1) = 12(a − 1)2 (7a + 11)2 [23(a + 2)2 + 9] − 108(7a + 11)2 (2a2 + 7a + 9)2
= −12(a + 2)2 (7a + 11)2 (13a2 + 154a + 157) ≥ 0.
Case 2: 5p2 + q ≥ 0. By Remark 3 from the proof of P 2.75, we only need to show
that there exist a real number B such that g6 (a, 1, 1) ≥ 0 for all real a satisfying 5(a +
122 Vasile Cîrtoaje
2)2 + 2a + 1 ≥ 0; that is, for a ∈ (−∞, −3] ∪ [−7/5, ∞). From g6 (1, 1, 1) = 0, we get
B = 4/9, then
g6 (a, 1, 1) = 12(a − 1)2 (7a + 11)2 [23(a + 2)2 + 9] − 108(a − 1)4 (2a + 3)2
= 12(a + 2)2 (a − 1)2 (1091a2 + 3650a + 3035) ≥ 0.
The proof is completed. The equality holds for a = b = c = 1, and for a = −11 and
b = c = 7 (or any cyclic permutation).
L
.M
b + c 2 c + a 2 a+b 2
+ + > 2.
a b c
D
(Michael Rozenberg, 2014)
A
Solution. Assume that a2 = min{a2 , b2 , c 2 }. By the Cauchy-Schwarz inequality, we
have
c + a 2 a + b 2 [(c + a) + (−a − b)]2
PI (b − c)2
+ ≥ = .
M
b c b2 + c 2 b2 + c 2
On the other hand,
LY
b + c 2 (b + c)2
≥ 2 .
a b + c2
O
Therefore,
.M
b + c 2 c + a 2 a + b 2 (b + c)2 (b − c)2
+ + ≥ 2 + 2 = 2.
a b c b + c2 b + c2
W
+ =
b c b2 + c 2
W
Solution. (a) First Solution. Without loss of generality, assume that a ≤ b ≤ c, when
Denote
4
k= p
3 3
and write the inequality as
Aa2 + 2Ba + C ≥ 0,
where
A = (b2 + 1)(c 2 + 1) + 2k(b − c),
L
.M
B = −k(b2 − c 2 ),
C = (b2 + 1)(c 2 + 1) + 2k bc(b − c).
D
−x y
Substituting b = p and c = p , by the Cauchy-Schwarz inequality, we get
A
3 3
PI
9A = (x 2 + 1 + 2)(1 + y 2 + 2) − 8(x + y) ≥ (x + y + 2)2 − 8(x + y) = (x + y − 2)2 ≥ 0.
M
p p
We have A = 0 only for b = −1/ 3, c = 1/ 3, when Aa2 + Ba + C = 64/27. Otherwise,
for A > 0, it suffices to prove that AC − B 2 ≥ 0. Let us denote
LY
E = b − c, F = bc + 1.
O
Since
.M
we have
W
and hence
1 p p
= (E 2 + F 2 )(E 2 + F 2 + 2kE F ) − k2 E 4 = (E + 3F )2 (11E 2 − 2 3E F + 9F 2 ) ≥ 0.
27
The equality holds for
p b+c
b−c+ 3(bc + 1) = 0, a+ =0
1 + 3bc
124 Vasile Cîrtoaje
where
L
.M
X X
E = (3x 2 + 1)(3 y 2 + 1)(3z 2 + 1) = 27x 2 y 2 z 2 + 9 x2 y2 + 3 x 2 + 1.
It it easy to check that the equality holds for x = −1, y = 0 and z = 1 (or any cyclic
D
permutation), when
A
x + y + z = 0, x y + yz + z x = −1, x yz = 0.
PI
From
M
X 2
xy +1 ≥ 0,
LY
we get X X X
1≥− x 2 y 2 − 2x y x x −2 x y,
O
and from
.M
X
(9x yz + x)2 ≥ 0,
we get
W
X X X
81x 2 y 2 z 2 ≥ −18x yz x− x2 − 2 x y.
W
Therefore,
W
X X X X X
3E ≥ −18x yz x− x2 − 2 x y + 27 x2 y2 + 9 x2
X X X
+3 − x 2 y 2 − 2x y x x −2 xy
X X X X
= 24 x 2 y 2 − x yz x +8 x2 − xy
X 4X
= 12 x 2 ( y − z)2 + (2x − y − z)2 .
3
By the AM-GM inequality, we have
rX X
3E ≥ 8 x 2 ( y − z)2 (2x − y − z)2 .
Symmetric Polynomial Inequalities in Real Variables 125
1 2 3 1 2 3 1 2 3
a− + b− + c− + ≥ (a − b)(b − c)(c − a).
L
2 4 2 4 2 4
.M
Using the substitution
p p p
1 3 1 3 1 3
D
a− = x, b− = y, c− = z,
2 2 2 2 2 2
A
the inequality turns out into the inequality in (a). From the equality conditions in (a),
namely
p
PI
y +z
y − z + 3( yz + 1) = 0, x + = 0,
M
1 + 3 yz
LY
c 1−c
(or any cyclic permutation).
.M
W
(1 − a + a2 )(1 − b + b2 )(1 − c + c 2 ) ≥ 1.
W
Solution (by Marian Tetiva). Assume first that a, b, c ≥ 0. Among the numbers a, b, c
always there exist two (let b and c) which are either less than or equal to 1, or greater
than or equal to 1. Then,
bc(b − 1)(c − 1) ≥ 0,
hence
(1 − a + a2 )(5 − 4a + a2 ) ≥ 2.
Indeed,
(1 − a + a2 )(5 − 4a + a2 ) − 2 = (a − 1)2 (a2 − 3a + 3) ≥ 0.
Assume now that a ≤ b ≤ c and a<0. We have
b+c a+b+c 3
c≥ > = .
2 2 2
L
The desired inequality is true since
.M
1 − a + a2 > 1,
D
2
1 3 3
2
1− b+ b = −b + ≥ ,
A
2 4 4
1 − c + c2 > 1 − c +
3c
2
c 3 7
=1+ >1+ = .
2 4 4
PI
M
The proof is completed. The equality holds for a = b = c = 1.
LY
O
X a(a − 4)
+ 1 ≥ 3,
W
a2 + 2
X (a − 1)2 3
≥ .
a2 +2 2
From
a2 = (b + c)2 ≤ 2(b2 + c 2 ),
we get
3a2 ≤ 2(a2 + b2 + c 2 ).
Similarly,
3b2 ≤ 2(a2 + b2 + c 2 ), 3c 2 ≤ 2(a2 + b2 + c 2 ).
Symmetric Polynomial Inequalities in Real Variables 127
Therefore, we have
X (a − 1)2 X 3(a − 1)2 3(a − 1)2
X
= ≥
a2 + 2 3a2 + 6 2(a2 + b2 + c 2 ) + 6
3 X 3
= (a − 1)2 = .
2(a + b + c + 3)
2 2 2 2
Thus, the proof is completed. The equality holds for a = b = c = 0, and also for a = −2
and b = c = 1 (or any cyclic permutation).
L
P 2.95. If a, b, c, d are real numbers, then
.M
2
1 + a bcd
2 2 2 2
(1 − a + a )(1 − b + b )(1 − c + c )(1 − d + d ) ≥ .
D
2
A
(Vasile Cîrtoaje, 1992)
PI
Solution. For a = b = c = d, the inequality can be written as
M
2(1 − a + a2 )2 ≥ 1 + a4 .
LY
It is true, since
2(1 − a + a2 )2 − 1 − a4 = (1 − a)4 ≥ 0.
O
4(1 − a + a2 )2 (1 − b + b2 )2 ≥ (1 + a4 )(1 + b4 ) ≥ (1 + a2 b2 )2 .
2(1 − a + a2 )(1 − b + b2 ) ≥ 1 + a2 b2 ,
W
2(1 − c + c 2 )(1 − d + d 2 ) ≥ 1 + c 2 d 2 ,
W
(1 + a2 b2 )(1 + c 2 d 2 ) ≥ (1 + a bcd)2 .
The equality holds for a = b = c = d = 1.
P 2.96. Let a, b, c, d be real numbers such that a bcd > 0. Prove that
1 1 1 1 1 1 1 1
a+ b+ c+ d+ ≥ (a + b + c + d) + + + .
a b c d a b c d
(Vasile Cîrtoaje, 2011)
128 Vasile Cîrtoaje
Then,
1X 2 1X 2
A − B =(1 − a bcd)2 + (ac − bd)2 + a (1 − bc)2 + a (1 − cd)2
2 2
L
X X
+ a2 b2 − a2 bd,
.M
and hence
X X 1X 2
a2 b2 − a2 bd = a (b − d)2 ≥ 0.
D
A− B ≥
2
A
The equality holds for a = b = c = d = 1.
(c 2 + 1)(1 + d 2 ) ≥ (c + d)2 ,
(d 2 + 1)(1 + a2 ) ≥ (d + a)2 .
a + b + c + d = 4, a2 + b2 + c 2 + d 2 = 7.
Prove that
a3 + b3 + c 3 + d 3 ≤ 16.
(Vasile Cîrtoaje, 2010)
L
.M
c + d = 4 − s, c 2 + d 2 = 7 − (a2 + b2 ) = 7 − s2 + 2p,
D
we have
A
2(a3 + b3 + c 3 + d 3 − 16) = 12p(2 − s) + 6s3 − 24s2 + 27s − 12
PI
≤ 3s2 (2 − s) + 6s3 − 24s2 + 27s − 12
M
= 3(s − 1)2 (s − 4) ≤ 0.
LY
This completes the proof. The equality holds for a = b = c = 1/2 and d = 5/2 (or any
cyclic permutation).
O
1 1
7 = a2 + b2 + c 2 + d 2 ≥ a2 + (b + c + d)2 = a2 + (4 − a)2 ,
3 3
W
−1 5 −1 5
it follows that a ∈ [ , ]. Similarly, we have b, c, d ∈ [ , ]. On the other hand,
W
2 2 2 2
W
5X 2 X 3 5 2
a3 + b3 + c 3 + d 3 = a + (a − a )
2 2
35 1 X
2
= − a (5 − 2a)
2 2
and, by virtue of the Cauchy-Schwarz inequality,
[ a(5 − 2a)]2 (5 a − 2 a2 )2
P P P
X
2
a (5 − 2a) ≥ P = = 3.
(5 − 2a)
P
20 − 2 a
Therefore,
35 3
a3 + b3 + c 3 + d 3 ≤ − = 16.
2 2
130 Vasile Cîrtoaje
Remark. In the same manner as in the second solution, we can prove the following
generalization.
• If a1 , a2 , . . . , an are real numbers such that
then
a13 + a23 + · · · + an3 ≤ n(n2 + 3n + 4),
with equality for a1 = ... = an−1 = 1 and an = n + 1 (or any cyclic permutation).
L
.M
P 2.98. Let a, b, c, d be real numbers such that a + b + c + d = 0. Prove that
12(a4 + b4 + c 4 + d 4 ) ≤ 7(a2 + b2 + c 2 + d 2 )2 .
D
A
(Vasile Cîrtoaje, 2010)
3
From
O
2
b2 + c 2 + d 2 b+c+d a2
x2 = ≥ = ,
3 3 9
.M
b4 + c 4 + d 4 = (b2 + c 2 + d 2 )2 − 2(b2 c 2 + c 2 d 2 + d 2 b2 )
= 9x 4 − 2(b2 c 2 + c 2 d 2 + d 2 b2 )
W
2
≤ 9x 4 − (bc + cd + d b)2
W
3
1
= 9x − [(b + c + d)2 − b2 − c 2 − d 2 ]2
4
6
1 45x 4 + 6a2 x 2 − a4
= 9x 4 − (a2 − 3x 2 )2 = .
6 6
and hence
45x 4 + 6a2 x 2 + 5a4
a4 + b4 + c 4 + d 4 ≤ .
6
Therefore, it suffices to prove that
(x 2 − a2 )(9x 2 − a2 ) ≤ 0.
a1 + a2 + · · · + an = 0,
L
then
(a12 + a22 + · · · + an2 )2 n(n − 1)
.M
≥ ,
a14 + a24 + · · · + an4 n2 − 3n + 3
D
with equality for −a1 /(n − 1) = a2 = · · · = an (or any cyclic permutation).
A
PI
M
P 2.99. Let a, b, c, d be real numbers such that a + b + c + d = 0. Prove that
LY
(a2 + b2 + c 2 + d 2 )3 ≥ 3(a3 + b3 + c 3 + d 3 )2 .
we have
W
a1 + a2 + · · · + an = 0,
132 Vasile Cîrtoaje
then
n(n − 1) 3
(a12 + a22 + · · · + an2 )3 ≥ (a + a23 + · · · + an3 )2 .
(n − 2)2 1
Moreover,
• If k ≥ 3 is an odd number, and a1 , a2 , . . . , an are real numbers such that
a1 + a2 + · · · + an = 0,
then
nk (n − 1)k−2
(a12 + a22 + · · · + an2 )k ≥ (a k + a2k + · · · + ank )2 ,
L
[(n − 1)k−1 − 1]2 1
.M
with equality for a1 = ... = an−1 = −an /(n − 1) (or any cyclic permutation).
D
A
PI
P 2.100. If a, b, c, d are real numbers such that a bcd = 1. Prove that
M
(1 + a2 )(1 + b2 )(1 + c 2 )(1 + d 2 ) ≥ (a + b + c + d)2 .
LY
Solution. Substituting a, b, c, d by |a|, |b|, |c|, |d|, respectively, the left side of the in-
equality remains unchanged, while the right side either remains unchanged or increases.
.M
(1 − b)(1 − d) ≥ 0.
W
c + a + bd + 1 ≥ a + b + c + d,
a2 + b2 + c 2 + d 2 = 4.
Prove that
(a bc)3 + (bcd)3 + (cd a)3 + (d a b)3 ≤ 4.
(Vasile Cîrtoaje, 2004)
Solution. Substituting a, b, c, d by |a|, |b|, |c|, |d|, respectively, the hypothesis and the
right side of the inequality remains unchanged, while the left side either remains un-
changed or decreases. Therefore, it suffices to prove the inequality only for a, b, c, d ≥ 0.
L
Setting x = a2 , y = b2 , z = c 2 and t = d 2 , we need to prove that
.M
(x yz)3/2 + ( yz t)3/2 + (z t x)3/2 + (t x y)3/2 ≤ 4
D
for x + y + z + t = 4. By the AM-GM inequality, we have
A
p
PI
4 4 x yz ≤ 1 + x + y + z = 5 − t,
M
p x yz(5 − t)2
(x yz)3/2 = x yz x yz ≤ .
16
LY
Analogously,
O
To prove this inequality, we use the mixing variable method.Without loss of generality,
x + y +z 4
assume that x ≥ y ≥ z ≥ t. Setting u = , we have 3u + t = 4, t ≤ u ≤ and
3 3
u3 ≥ x yz. We will show that
E(x, y, z, t) ≥ E(u, u, u, t) ≥ 0.
By Schur’s inequality
we get
9u3 + 3x yz ≥ 4u(x y + yz + z x),
and hence
3(u3 − x yz)
3u2 − x y − yz − z x ≥ ≥ 0.
4u
Therefore, it suffices to prove that
L
.M
3t
25(1 + ) ≥ 36t.
4u
D
Write this inequality in the homogeneous form
A
25(3u + t)(4u + 3t) ≥ 576ut,
PI
or, equivalently,
M
75(4u2 + t 2 ) ≥ 251ut.
LY
(1 − a)4 + (1 − b)4 ≥ c 4 + d 4 ,
(1 − c)4 + (1 − d)4 ≥ a4 + b4 .
Since
(1 − a)4 + (1 − b)4 ≥ 2(1 − a)2 (1 − b)2
and
1 2
c4 + d 4 ≥ (c + d 2 )2 ,
2
the former inequality holds if
L
.M
2(1 − a)(1 − b) ≥ c 2 + d 2 .
Indeed,
D
2(1 − a)(1 − b) − c 2 − d 2 = 2(1 − a)(1 − b) + a2 + b2 − 1 = (a + b − 1)2 ≥ 0.
A
The equality holds for a = b = c = d =
1
2
. PI
M
LY
−1
P 2.103. If a, b, c, d ≥ such that a + b + c + d = 4, then
2
O
1−a+a 2 1− b+ b 2 1−c+c 2 1 − d + d2
(Vasile Cîrtoaje, 2014)
W
Solution (by Nguyen Van Quy). Assume that a ≤ b ≤ c ≤ d and consider two cases:
a > 0 and a ≤ 0.
W
a2 b2 c2 d2
+ + + ≤ 4.
1 − a + a2 1 − b + b2 1 − c + c 2 1 − d + d 2
We have
a2 b2 c2 d2 a2 b2 c 2 d 2
+ + + ≤ + + + = 4.
1 − a + a2 1 − b + b2 1 − c + c 2 1 − d + d 2 a b c d
Case 2: −1/2 ≤ a ≤ 0. We can check that the equality holds for a = −1/2 and b = c =
d = 3/2 (or any cyclic permutation). Define the function
1− x −1
f (x) = + k1 x + k2 , x≥ ,
1 − x + x2 2
136 Vasile Cîrtoaje
such that
f (3/2) = f 0 (3/2) = 0.
We get
12 −4
k1 = , k2 = ,
49 49
when
1− x 12x − 4 (2x − 3)2 (3x + 5)
f (x) = + = .
1 − x + x2 49 49(1 − x + x 2 )
Since f (x) ≥ 0 for x ≥ −1/2, we have
L
1− x 4 − 12x
≥ .
.M
1 − x + x2 49
Therefore,
D
1− b 1−c 1−d 12 − 12(b + c + d) 12(a − 3)
A
+ + ≥ = .
1− b+ b 2 1−c+c 2 1−d +d 2 49 49
1−a+a 2 49
Indeed,
O
49
The proof is completed. The equality holds for a = b = c = d = 1, and also for a = −1/2
and b = c = d = 3/2 (or any cyclic permutation).
W
W
W
1− x 1− x (x − 1)2 (x + 2)
− = ≥ 0.
1 + x + x2 3 3(1 + x + x 2 )
Symmetric Polynomial Inequalities in Real Variables 137
1− x
f (x) = + k1 x + k2 , x ≥ −3,
1 + x + x2
L
such that
.M
f (2) = f 0 (2) = 0.
We get
D
2 3
k1 = , k2 = ,
A
49 49
when PI
1− x 2x + 3 (x − 2)2 (2x + 13)
f (x) = + = .
1 + x + x2 49(1 + x + x 2 )
M
49
Since f (x) ≥ 0 for x ≥ −3, we have
LY
1− x −2x − 3
≥ .
O
1 + x + x2 49
.M
1−a 2b + 3 2c + 3 2d + 3 2e + 3
− − − − ≥ 0,
1+a+a 2
W
49 49 49 49
which is equivalent to
W
1−a 2(b + c + d + e) + 12
W
− ≥ 0,
1+a+a 2 49
1−a 2(5 − a) + 12
− ≥ 0,
1 + a + a2 49
(a + 3)(2a2 − 26a + 9)
≥ 0.
49(1 + a + a2 )
Clearly, the last inequality is true for −3 ≤ a ≤ 0. The equality holds for a = b = c =
d = e = 1, and also for a = −3 and b = c = d = e = 2 (or any cyclic permutation).
138 Vasile Cîrtoaje
30(a4 + b4 + c 4 + d 4 + e4 ) ≥ 7(a2 + b2 + c 2 + d 2 + e2 )2 .
Among the numbers a, b, c, d, e there exist three with the same sign. Let a, b, c be these
numbers. We will show that
L
.M
E(a, b, c, d, e) ≥ E(a, b, c, x, x) ≥ 0,
where
D
d+e −(a + b + c)
x= = .
A
2 2
PI
The inequality E(a, b, c, d, e) ≥ E(a, b, c, x, x) is equivalent to
M
30(d 4 + e4 − 2x 4 ) ≥ 7(d 2 + e2 − 2x 2 )(2a2 + 2b2 + 2c 2 + d 2 + e2 + 2x 2 ).
LY
Since
(d − e)2 (7d 2 + 10d e + 7e2 )
d 4 + e4 − 2x 4 =
O
8
and
.M
(d − e)2
d 2 + e2 − 2x 2 = ,
2
W
which reduces to
21(d 2 + e2 ) + 34d e ≥ 7(a2 + b2 + c 2 ).
Since
a2 + b2 + c 2 ≤ (a + b + c)2 = (d + e)2 ,
it suffices to prove that
−(a + b + c)
The inequality E(a, b, c, x, x) ≥ 0, where x = , can be written as
2
X X X X X
23 a4 + 2( a)4 ≥ 7( a2 )( a)2 + 14 a2 b2 .
X X X X X X
23 a4 + 6( a b)( a)2 ≥ 7( a2 )( a)2 + 14 a2 b2 .
This is equivalent to
X X 1X X
a4 + a bc a b(a2 + b2 ) + a2 b2 ,
L
a≥
2
.M
which follows by summing Schur’s inequality of degree four
D
X X X
a4 + a bc a≥ a b(a2 + b2 )
A
and the obvious inequality
1X X
PI
a b(a2 + b2 ) ≥ a2 b2 .
M
2
LY
This completes the proof. The equality holds for a = b = c = 2 and d = e = −3 (or any
permutation thereof).
O
Remark. Notice that the following generalization holds (Vasile Cîrtoaje, 2010).
.M
a1 + a2 + · · · + an = 0,
W
then
(a12 + a22 + · · · + an2 )2
W
n(n2 − 1)
≤ ,
a14 + a24 + · · · + an4 n2 + 3
W
with equality when (n+1)/2 of a1 , a2 , . . . , an are equal to (n−1)/2 and the other numbers
are equal to −(n + 1)/2.
a1 ≥ a2 ≥ · · · ≥ an .
where
f (x) = (n − 2)x − x 2 .
Since
L
f (ai ) − f (ai+1 ) = (ai − ai+1 )[n − 2 − (ai + ai+1 )]
.M
≥ (ai − ai+1 )[n − 2 − (a1 + a2 )] = (ai − ai+1 )(n − 2 + a3 + · · · + an )]
D
= (ai − ai+1 )[(1 + a3 ) + · · · + (1 + an )] ≥ 0,
A
we have a1 ≥ a2 ≥ · · · ≥ an and f (a1 ) ≥ f (a2 ) ≥ · · · ≥ f (an ). Therefore, by Chebyshev’s
inequality, we get
n
PI
M
X
n ai f (ai ) ≥ (a1 + a2 + · · · + an )[ f (a1 ) + f (a2 ) + · · · + f (an )] = 0.
LY
i=1
Solution. For the nontrivial case a12 + a22 + · · · + an2 6= 0, write the inequality as
S3
(n − 1) + n − 2 ≥ 0,
S2
where
S2 = a12 + a22 + · · · + an2 , S3 = a13 + a23 + · · · + an3 .
Without loss of generality, assume that
a1 ≤ a2 ≤ · · · ≤ an .
Symmetric Polynomial Inequalities in Real Variables 141
L
n(n − 2)
S2 + p ≥ np2 .
.M
n−1
n
Case 1: S2 ≥ . Choosing p = 1, we have a1 + p ≥ 0, and
D
n−1
A
n(n − 2) n
S2 + p − np2 = S2 − ≥ 0.
n
n−1 n−1 PI
Case 2: 0 < S2 ≤ . We set
M
n−1 v
tn − 1
LY
p= S2 .
n
Since
O
n−1 n−1 2 1
p2 − a12 = S2 − a12 = (a2 + · · · + an2 ) − a12
n n n
.M
2
(a2 + · · · + an )2 a1
≥ − = 0,
n n
W
we have a1 + p ≥ 0. In addition,
W
s
n(n − 2) 2
p n p
S2 + p − np = (n − 2) S2 − S2 ≥ 0.
n−1 n−1
W
p
P 2.108. Let a1 , a2 , . . . , an ≥ n − 1 − n2 − n + 1 be nonzero real numbers such that
a1 + a2 + · · · + an = n. Prove that
1 1 1
+ + ··· + ≥ n.
a12 a22 an2
1 1 1 1 1 1 1 2
+ + ··· + 2 ≥ + + ··· +
a12 a22 an n a1 a2 an
2
n2
1
≥ = n.
n a1 + a2 + · · · an
L
a1 + · · · + ak ak+1 + · · · + an
.M
Let us denote x = and y = . We have
k n−k
p
D
−1 < n − 1 − n2 − n + 1 ≤ x < 0, y > 1, k x + (n − k) y = n.
A
From k ≥ 1 and k( y − x) = n( y − 1) > 0, we get y − x ≤ n( y − 1), and hence
y≥
n− x
.
PI
n−1
M
p
In addition, from n − 1 − n2 − n + 1 ≤ x, we get
LY
n + 2(n − 1)x − x 2 ≥ 0.
O
2
1 2 1 k2
1 1 1 1
k
2
+ ··· + 2 ≥ + ··· + ≥ = 2
a1 ak k −a1 −ak k −a1 − · · · − ak x
W
and
W
2 2
(n − k)2
1 1 1 1 1 1 n−k
+ ··· + 2 ≥ + ··· + ≥ = .
W
2
ak+1 an n−k ak+1 an n−k ak+1 + · · · + an y2
n
P 2.109. Let a1 , a2 , . . . , an ≤ be real numbers such that
L
n−2
.M
a1 + a2 + · · · + an = n.
D
a1k + a2k + · · · + ank ≥ n.
A
PI (Vasile Cîrtoaje, 2012)
Solution. First we show that at most one of ai is negative. Assume, for the sake of
M
contradiction, that an−1 < 0 and an < 0. Then,
LY
n
an−1 + an = n − (a1 + · · · + an−2 ) ≥ n − (n − 2) · = 0,
n−2
O
a1 + a2 + · · · + an−1
x= , y = −an ,
W
n−1
where
W
(n − 1)x − y = n, x ≥ y > 0.
The condition x ≥ y follows from
x − y = n − (n − 2)x ≥ 0.
(n − 1)x k − y k ≥ n.
144 Vasile Cîrtoaje
(n − 2)x k + (x − y)k ≥ n.
L
.M
By Jensen’s inequality, we have
D
k
(n − 2)x + (x − y) n k−1
k k
(n − 2)x + (x − y) ≥ [(n − 2) + 1] =n > n.
(n − 2) + 1
A
n−1
PI
This completes the proof. The equality holds for a1 = a2 = · · · an = 1.
M
LY
−(3n − 2)
a1 , a2 , . . . , an ≥ , a1 + a2 + · · · + an = n,
n−2
.M
then
1 − a1 1 − a2 1 − an
+ + ··· + ≥ 0.
W
(1 + a1 )2 (1 + a2 )2 (1 + an )2
(Vasile Cîrtoaje, 2014)
W
Solution. Since the inequality holds for n = 3 (see P 2.25), consider further that n ≥ 4.
W
1− x x − 1 (x − 1)2 (x + 3)
+ = ≥ 0.
(1 + x)2 4 4(1 + x)2
Case 2: −(3n − 2)/(n − 2) ≤ a1 ≤ 0. We can check that the equality holds for
−(3n − 2) n+2
a1 = , a2 = a3 = · · · = an = .
n−2 n−2
Based on this, define the function
1− x −(3n − 2)
f (x) = + k1 x + k2 , x≥ ,
(1 + x)2 n−2
such that
n+2 0 n+2
f =f = 0.
n−2 n−2
We get
L
(n − 4)(n − 2)2 (n − 2)(−n2 + 6n + 8)
k1 = =
.M
, k 2 ,
4n3 4n3
[(n − 2)x − n − 2]2 [(n − 4)x + 3n − 4]
f (x) = .
D
4n3 (1 + x)2
A
Since f (x) ≥ 0 for n ≥ 4 and x ≥ −(3n − 2)/(n − 2), we have
1− x
(1 + x)2
PI
≥ −k1 x − k2 .
M
Thus, it suffices to show that
LY
1 − a1
− k1 (a2 + a3 + · · · + an ) − (n − 1)k2 ≥ 0,
(1 + a1 )2
O
which is equivalent to
1 − a1
.M
− k1 (n − a1 ) − (n − 1)k2 ≥ 0,
(1 + a1 )2
[(n − 2)a1 + 3n − 2][(n − 4)(n − 2)a12 − 2(n2 + 4n − 8)a1 + n2 − 2n + 8] ≥ 0.
W
Clearly, the last inequality is true for n ≥ 4 and −(3n − 2)/(n − 2) ≤ a1 ≤ 0. This
W
completes the proof. The equality holds for a1 = a2 = · · · an = 1, and also for x 1 =
−(3n − 2) n+2
and x 2 = · · · = x n = (or any cyclic permutation).
W
n−2 n−2
(a1 + a2 + · · · + an )2 (n − 1)n−1
(a) ≤ ;
(a12 + 1)(a22 + 1) · · · (an2 + 1) nn−2
1
a1 + a2 + · · · + an (2n − 1)n− 2
(b) ≤ .
(a12 + 1)(a22 + 1) · · · (an2 + 1) 2n nn−1
xi
Solution. Let m be a positive integer (m ≥ n), and let ai = p for all i. Assume
m−1
that
x 12 ≤ · · · ≤ x k2 ≤ 1 ≤ x k+1
2
≤ · · · ≤ x n2 ,
where 0 ≤ k ≤ n. By Bernoulli’s inequality and the Cauchy-Schwarz inequality, we have
n n 2
m−1 nY 2 m−1 nY
xi
(ai + 1) = +1
m i=1
m i=1
m−1
n k n
x i2 − 1 x i2 − 1 Y x i2 − 1
Y Y
= 1+ = 1+ 1+
m m m
L
i=1 i=1 i=k+1
.M
k n
x i2 − 1 x i2 − 1
X X
≥ 1+ 1+
i=1
m i=k+1
m
D
1
= [x 2 + · · · + x k2 + (m − n) + (n − k)][k + (m − n) + x k+1
2
+ · · · + x n2 ]
A
m2 1
1 PI
≥ 2 (x 1 + · · · + x k + m − n + x k+1 + · · · + x n )2
m
M
1
= 2 (m − n + x 1 + x 2 + · · · + x n )2 .
LY
m
Therefore,
O
n Yn
m−1 1
(ai2 + 1) ≥ 2 (m − n + x 1 + x 2 + · · · + x n )2 ,
.M
m i=1
m
and hence
W
n 2
mn−2 m−n
Y
(ai2 + 1) ≥ p + a1 + a2 + · · · + an .
W
i=1
(m − 1)n−1 m−1
W
(a) Choosing m = n, we get the desired inequality. The equality holds for
1
a1 = a2 = · · · = an = p
n−1
or
−1
a1 = a2 = · · · = an = p .
n−1
Symmetric Polynomial Inequalities in Real Variables 147
(b) Since
2
m−n 4(m − n)
p + a1 + a2 + · · · + an ≥ p (a1 + a2 + · · · + an ),
m−1 m−1
we get
n
Y 4mn−2 (m − n)
(ai2 + 1) ≥ 1
(a1 + a2 + · · · + an ).
i=1 (m − 1)n− 2
Choosing m = 2n, we get the desired inequality. The equality holds for
1
a1 = a2 = · · · = an = p .
L
2n − 1
D
.M
A
PI
M
LY
O
.M
W
W
W
148 Vasile Cîrtoaje
L
D
.M
A
PI
M
LY
O
.M
W
W
W
Chapter 3
L
in Nonnegative Variables
D
.M
A
3.1 Applications
PI
M
3.1. If a, b, c are positive real numbers, then
a2 + b2 + c 2 + 2a bc + 1 ≥ 2(a b + bc + ca).
LY
O
p
3.2. Let a, b, c be nonnegative real numbers. If 0 ≤ k ≤ 2, then
.M
a2 + b2 + c 2 + ka bc + 2k + 3 ≥ (k + 2)(a + b + c).
W
149
150 Vasile Cîrtoaje
L
.M
3.8. If a, b, c are positive real numbers, then
D
4(a3 + b3 + c 3 ) + 15a bc + 54 ≥ 27(a + b + c).
A
(b)
PI
M
3.9. Let a, b, c be nonnegative real numbers such that
a + b + c = a2 + b2 + c 2 .
LY
Prove that
O
a b + bc + ca ≥ a2 b2 + b2 c 2 + c 2 a2 .
.M
L
.M
(3a2 − 2a b + 3b2 )(3b2 − 2bc + 3c 2 )(3c 2 − 2ca + 3a2 ) ≤ 36.
D
A
3.16. Let a, b, c be nonnegative real numbers such that a + b + c = 3. Prove that
PI
(a2 − 4a b + b2 )(b2 − 4bc + c 2 )(c 2 − 4ca + a2 ) ≤ 3.
M
LY
12
a bc + ≥ 5.
a b + bc + ca
.M
3
5(a + b + c) +
W
≥ 18.
a bc
W
12 + 9a bc ≥ 7(a b + bc + ca).
L
3.23. If a, b, c are positive real numbers such that a bc = 1, then
.M
a3 + b3 + c 3 + a−3 + b−3 + c −3 + 21 ≥ 3(a + b + c)(a−1 + b−1 + c −1 ).
D
A
3.24. If a, b, c are positive real numbers such that a bc = 1, then
a2 + b2 + c 2 − a b − bc − ca ≥
9
PI
(a + b + c − 3).
M
4
LY
a2 + b2 + c 2 + a + b + c ≥ 2(a b + bc + ca).
.M
6
a b + bc + ca + ≥ 5.
a+b+c
Symmetric Polynomial Inequalities in Nonnegative Variables 153
L
3.31. If a, b, c are positive real numbers such that a + b + c = 3, then
.M
1 1 1
2
+ 2 + 2 ≥ a2 + b2 + c 2 .
a b c
D
A
3.32. If a, b, c are positive real numbers such that a b + bc + ca = 3, then
PI
a3 + b3 + c 3 + 7a bc ≥ 10.
M
LY
a4 b4 + b4 c 4 + c 4 a4 ≤ 3.
.M
W
1 + 8a bc ≥ 9 min{a, b, c}.
1 + 4a bc ≥ 5 min{a, b, c}.
154 Vasile Cîrtoaje
L
.M
(8a2 + bc)(8b2 + ca)(8c 2 + a b) ≤ (a + b + c)6 .
D
3.40. If a, b, c are positive real numbers such that a2 b2 + b2 c 2 + c 2 a2 = 3, then
A
PI
a + b + c ≥ a bc + 2.
M
LY
a2 + b2 + c 2 + a bc + 2 ≥ a + b + c + a b + bc + ca.
W
W
a + b + c + 4a bc ≥ k(a b + bc + ca)
2
3.45. If a, b, c ≥ such that a + b + c = 3, then
3
a2 b2 + b2 c 2 + c 2 a2 ≥ a b + bc + ca.
a + b + c = 3,
then
1 1 1
L
+ + ≥ a2 + b2 + c 2 .
a b c
D
.M
3.47. If a, b, c are positive real numbers such that a ≤ 1 ≤ b ≤ c and
A
1 1 1
a+b+c = + + ,
a b c PI
then
M
1 1 1
a2 + b2 + c 2 ≤ 2
+ 2 + 2.
a b c
LY
O
1 1 1
a+b+c = + + ,
a b c
W
then
1 1 1
(1 − a bc) a n + b n + c n − n − n − n ≥ 0
W
a b c
for any integer n ≥ 2.
W
Prove that
1 1 1
(b) 2 + + E(a, b, c) ≥ (a − b)2 + (b − c)2 + (c − a)2 .
a b c
156 Vasile Cîrtoaje
L
and (b) is sharper than (a) if p p p
b+
.M
c≥ a.
D
3.51. If a, b, c are nonnegative real numbers such that
A
(a + b)(b + c)(c + a) = 8, PI
then
M
p p p
a+ b+ c ≥ a b + bc + ca.
LY
p
O
a b + bc + ca ≥ 1 + 2a bc.
a2 b2 + b2 c 2 + c 2 a2 ≥ a b + bc + ca.
Symmetric Polynomial Inequalities in Nonnegative Variables 157
1 1 1 41
+ + + ≥ 3(a2 + b2 + c 2 ).
a b c 6
L
c
then the product r = a bc is maximal when a = b ≥ or a + b = c, and is minimal
.M
2
when b = c ≥ a.
D
A
3.58. If a, b, c are positive real numbers such that a + b + c = 3, then
9
+ 16 ≥
75
PI.
a b + bc + ca
M
a bc
LY
1 1 1
8 + + + 9 ≥ 10(a2 + b2 + c 2 ).
.M
a b c
W
L
3.65. If a, b, c ∈ [0, 1], then
.M
p p p 3
a a + b b + c c ≥ (a b + bc + ca − a bc).
2
D
A
3.66. If a, b, c ∈ [0, 1], then PI
p p p 500
M
3(a a + b b + c c) + a bc ≥ 5(a b + bc + ca).
81
LY
O
a + b + c ≥ 2 + a bc.
W
that
(a) the inequality f n (a, b, c) ≥ 0 holds for all nonnegative real numbers a, b, c if and
W
4(a4 + b4 + c 4 ) + 45 ≥ 19(a2 + b2 + c 2 ).
Symmetric Polynomial Inequalities in Nonnegative Variables 159
L
X
.M
a(a − 2b)(a − 2c)(a − 5b)(a − 5c) ≥ 0.
D
3.73. If a, b, c are the lengths of the sides of a triangle, then
A
a4 + b4 + c 4 + 9a bc(a + b + c) ≤ 10(a2 b2 + b2 c 2 + c 2 a2 ).
PI
M
3.74. If a, b, c are the lengths of the sides of a triangle, then
LY
X
3(a4 + b4 + c 4 ) + 7a bc(a + b + c) ≤ 5 a b(a2 + b2 ).
O
.M
b2 + c 2 − 6bc c 2 + a2 − 6ca a2 + b2 − 6a b
+ + + 4(a + b + c) ≤ 0.
W
a b c
W
the form
f6 (a, b, c) = Ar 2 + B(p, q)r + C(p, q), A ≤ 0,
where
p = a + b + c, q = a b + bc + ca, r = a bc.
Prove that
(a) the inequality f6 (a, b, c) ≥ 0 holds for all nonnegative real numbers a, b, c if and
only if f6 (a, 1, 1) ≥ 0 and f6 (0, b, c) ≥ 0 for all nonnegative real numbers a, b, c;
(b) the inequality f6 (a, b, c) ≥ 0 holds for all lengths a, b, c of the sides of a non-
degenerate or degenerate triangle if and only if f6 (x, 1, 1) ≥ 0 for 0 ≤ x ≤ 2, and
f6 ( y + z, y, z) ≥ 0 for all y, z ≥ 0.
160 Vasile Cîrtoaje
(a) If 2 ≤ k ≤ 6, then
X 4(k − 2)(a − b)2 (b − c)2 (c − a)2
a(a − b)(a − c)(a − k b)(a − kc) + ≥ 0;
a+b+c
L
.M
(b) If k ≥ 6, then
X (k + 2)2 (a − b)2 (b − c)2 (c − a)2
a(a − b)(a − c)(a − k b)(a − kc) + ≥ 0.
D
4(a + b + c)
A
3.79. If a, b, c are nonnegative real numbers, then
PI
M
(3a2 + 2a b + 3b2 )(3b2 + 2bc + 3c 2 )(3c 2 + 2ca + 3a2 ) ≥ 8(a2 + 3bc)(b2 + 3ca)(c 2 + 3a b).
LY
O
−2 11
≤k≤ ,
3 8
W
then
(a2 + ka b + b2 )(b2 + k bc + c 2 )(c 2 + kca + a2 ) ≤ k + 2.
W
W
3.82. Let a, b, c be nonnegative real numbers, no two of which are zero. Then,
X 5(a − b)2 (b − c)2 (c − a)2
(a − b)(a − c)(a − 2b)(a − 2c) ≥ .
a b + bc + ca
Symmetric Polynomial Inequalities in Nonnegative Variables 161
a + b + c = p, a bc = r,
where p and r are fixed positive numbers satisfying p3 ≥ 27r. Prove that
q = a b + bc + ca
L
3.84. If a, b, c are positive real numbers, then
.M
1 1 1 1
(a + b + c − 3) + + − 3 + a bc + ≥ 2.
a b c a bc
D
A
3.85. If a, b, c are positive real numbers such that a bc = 1, then
PI
s
3 2 2
M
(a) a b + bc + ca − ≥ (a + b + c) − 1;
7 3 3
LY
46 p
(b) a b + bc + ca − 3 ≥ ( a + b + c − 2 − 1).
27
O
.M
50 37
W
a b + bc + ca + ≥ .
a+ b+c+5 4
W
W
a2 + b2 + c 2
(a + b + c − 3)2 + 1 ≥ ;
3
1
(b) If a bc = , then
2
a2 + b2 + c 2 + 3(3 − a − b − c)2 ≥ 3.
162 Vasile Cîrtoaje
a b + bc + ca = a bc + 2,
then
a2 + b2 + c 2 + a bc ≥ 4.
L
.M
3.90. If a, b, c are positive real numbers, then
D
2
b+c p c + a p 2 a+b p 2
+ 2 +
A
−2− 2 −2− − 2 − 2 ≥ 6.
a b c
PI
M
3.91. If a, b, c are positive real numbers, then
LY
a4 + b4 + c 4 − a2 b2 − b2 c 2 − c 2 a2 ≥ 2|a3 b + b3 c + c 3 a − a b3 − bc 3 − ca3 |.
W
1 4
3.96. Let a, b, c 6= be nonnegative real numbers such that a + b + c = 3. If k ≥ ,
k 3
then
1−a 1− b 1−c
+ + ≥ 0.
(1 − ka)2 (1 − k b)2 (1 − kc)2
a2 + b2 + c 2 + d 2 = 1.
Prove that
(1 − a)(1 − b)(1 − c)(1 − d) ≥ a bcd.
L
.M
3.98. Let a, b, c, d and x be positive real numbers such that
D
1 1 1 1 4
+ 2 + 2 + 2 = 2.
A
a 2 b c d x
If x ≥ 2, then PI
(a − 1)(b − 1)(c − 1)(d − 1) ≥ (x − 1)4 .
M
LY
1 1 1 1 1 1 1 1
a+ −1 b+ −1 c+ −1 d + −1 +3≥ + + + .
W
a b c d a b c d
W
a + b + c + d = 4.
Prove that
1 + 2(a bc + bcd + cd a + d a b) ≥ 9 min{a, b, c, d}.
164 Vasile Cîrtoaje
a + b + c + d = 4.
Prove that
5(a2 + b2 + c 2 + d 2 ) ≥ a3 + b3 + c 3 + d 3 + 16.
a + b + c + d = 4.
L
.M
Prove that
3(a2 + b2 + c 2 + d 2 ) + 4a bcd ≥ 16.
D
A
3.105. Let a, b, c, d be nonnegative real numbers such that
a + b + c + d = 4.
PI
M
Prove that
LY
1 1 1 1
a+b+c+d = + + + .
W
a b c d
Prove that
W
1 1 1 1
(1 − a bcd) a2 + b2 + c 2 + d 2 − 2 − 2 − 2 − 2 ≥ 0.
W
a b c d
a + b + c + d = 1.
Prove that
1 1 1 1 81
(1 − a)(1 − b)(1 − c)(1 − d) + + + ≥ .
a b c d 16
Symmetric Polynomial Inequalities in Nonnegative Variables 165
a + b + c + d = a3 + b3 + c 3 + d 3 = 2.
Prove that
7
a2 + b2 + c 2 + d 2 ≥ .
4
L
.M
3.110. Let a, b, c, d and k be positive real numbers such that
1 1 1 1
D
(a + b + c + d) + + + = k.
a b c d
A
p
If 16 ≤ k ≤ (1 + 10)2 , then any three of a, b, c, d are the lengths of the sides of a
triangle (non-degenerate or degenerate).
PI
M
LY
a b c d
.M
119
If 16 ≤ k ≤ , then there exist three numbers of a, b, c, d which are the lengths of
6
the sides of a triangle (non-degenerate or degenerate).
W
W
(a + b + c + d)2 = k(a2 + b2 + c 2 + d 2 ).
W
11
If ≤ k ≤ 4, then any three of a, b, c, d are the lengths of the sides of a triangle
3
(non-degenerate or degenerate).
(a + b + c + d)2 = k(a2 + b2 + c 2 + d 2 ).
49
If ≤ k ≤ 4, then there exist three numbers of a, b, c, d which are the lengths of the
15
sides of a triangle (non-degenerate or degenerate).
166 Vasile Cîrtoaje
4(a4 + b4 + c 4 + d 4 + e4 ) ≥ (a2 + b2 + c 2 + d 2 + e2 )2 ;
(b) If a + b + c = d + e, then
12(a4 + b4 + c 4 + d 4 + e4 ) ≤ 7(a2 + b2 + c 2 + d 2 + e2 )2 .
L
.M
a + b + c + d + e = 5.
Prove that
D
a4 + b4 + c 4 + d 4 + e4 + 150 ≤ 31(a2 + b2 + c 2 + d 2 + e2 ).
A
PI
M
3.116. Let a, b, c, d, e be positive real numbers such that
LY
a2 + b2 + c 2 + d 2 + e2 = 5.
Prove that
O
a bcd e(a4 + b4 + c 4 + d 4 + e4 ) ≤ 5.
.M
a + b + c + d + e = 5.
W
Prove that
W
1 1 1 1 1 20
+ + + + + 2 ≥ 9.
a b c d e a + b + c 2 + d 2 + e2
2
3.118. If a, b, c, d, e ≥ 1, then
1 1 1 1 1
a+ b+ c+ d+ e+ + 68 ≥
a b c d e
1 1 1 1 1
≥ 4(a + b + c + d + e) + + + + .
a b c d e
Symmetric Polynomial Inequalities in Nonnegative Variables 167
Prove that
1
a bc x yz < .
36
L
.M
Prove that
1 1 1 n2 (2n − 3)
+ + ··· + ≥ .
a1 a2 an 2(n − 1)(n − 2)
D
A
3.121. Let a1 , a2 , · · · , an be positive real numbers such that a1 + a2 +· · ·+ an = n. Prove
that
1 1 1
PI
M
n2 + + ··· + − n ≥ 4(n − 1)(a12 + a22 + · · · + an2 − n).
a1 a2 an
LY
Prove that
(n + 1)(a12 + a22 + · · · + an2 ) ≥ n2 + a13 + a23 + · · · + an3 .
.M
W
a1 + a2 + · · · + an = 1.
Prove that
n
1 1 1 1
a1 + −2 a2 + −2 an + −2 ≥ n+ −2 .
a1 a2 an n
168 Vasile Cîrtoaje
a1 + a2 + · · · + an = n.
Prove that
n
a12 + a22 + · · · + an2 − n ≥ (1 − a1 a2 · · · an ).
n−1
a1 ≤ a2 ≤ · · · ≤ an ,
L
.M
1 1 1
(a1 + a2 + · · · + an ) + + ··· + = k.
a1 a2 an
D
i(n − i)
(a) If n2 ≤ k ≤ n2 + , i ∈ {2, 3, · · · , n − 1}, then ai−1 , ai and ai+1 are the
2
A
lengths of the sides of a non-degenerate or degenerate triangle;
(b) If n2 ≤ k ≤ αn , where αn =
9n2
for even n, and αn =
PI9n2 − 1
for odd n, then
8 8
M
there exist three numbers ai which are the lengths of the sides of a non-degenerate or
degenerate triangle.
LY
O
a1 ≤ a2 ≤ · · · ≤ an ,
(2n − i)2
W
(a) If ≤ k ≤ n, i ∈ {2, 3, · · · , n − 1}, then ai−1 , ai and ai+1 are the lengths
4n − 3i
of the sides of a non-degenerate or degenerate triangle;
W
8n + 1
(b) If ≤ k ≤ n, then there exist three numbers ai which are the lengths of the
9
sides of a non-degenerate or degenerate triangle.
Symmetric Polynomial Inequalities in Nonnegative Variables 169
3.2 Solutions
a2 + b2 + c 2 + 2a bc + 1 ≥ 2(a b + bc + ca).
L
Using Schur’s inequality and the AM-GM inequality, we have
.M
X X
x 6 + y 6 + z 6 + 3x 2 y 2 z 2 ≥ x 2 y 2 (x 2 + y 2 ) ≥ 2 x 3 y 3.
D
Thus, it suffices to show that
A
2x 3 y 3 z 3 − 3x 2 y 2 z 2 + 1 ≥ 0,
PI
which is equivalent to
M
(x yz − 1)2 (2x yz + 1) ≥ 0.
LY
Second Solution. Among the numbers 1 − a, 1 − b and 1 − c there are always two with
the same sign; let us say (1 − b)(1 − c) ≥ 0. Then, we have
.M
a2 + b2 + c 2 + 2a bc + 1 − 2(a b + bc + ca) =
= (a − 1)2 + (b − c)2 + 2a + 2a bc − 2a(b + c)
W
Since the both sides of the inequality are linear of k, it suffices to prove it for only k = 0
and k = 1. For k = 0, the inequality reduces to the the known a2 + b2 +c 2 ≥ a b+ bc +ca.
p
P 3.2. Let a, b, c be nonnegative real numbers. If 0 ≤ k ≤ 2, then
a2 + b2 + c 2 + ka bc + 2k + 3 ≥ (k + 2)(a + b + c).
170 Vasile Cîrtoaje
L
p
x 2 + y 2 + z 2 + 2 (x yz + x y + yz + z x) ≥ 0
.M
for x, y, z ≥ −1. Among the numbers x, y and z there are always two of them with the
same sign; let us say yz ≥ 0. Since
D
A
1
y 2 + z2 ≥ ( y + z)2
2 PI
and
M
x yz + x y + yz + z x = (x + 1) yz + x( y + z) ≥ x( y + z),
LY
2
.M
which is equivalent to
1
[x + p ( y + z)]2 ≥ 0.
2
W
permutation).
W
Second Solution. Let a + b + c = p. For the nontrivial case p > 0, we write the original
inequality in the homogeneous form
L
.M
18[p2 − 2(k + 2)p + 2(2k + 3)] + kp3 − 9p2 + 9(k + 2)p − 9(2k + 3) ≥ 0;
that is,
D
kp3 + 9p2 − 27(k + 2)p + 27(2k + 3) ≥ 0,
A
(kp + 6k + 9)(p − 3)2 ≥ 0.
PI
M
LY
Æ 18a bc
a bc(a + b + c) + 3 ≥ 2 3a bc(a + b + c) ≥ .
a+b+c
W
18a bc
a2 + b2 + c 2 + ≥ 2(a b + bc + ca),
a+b+c
which is just Schur’s inequality of third degree. The equality holds for a = b = c = 1.
Second Solution. Applying the AM-GM, we get
p p p
a bc(a + b + c) + 3 = (a2 bc + 1) + (a b2 c + 1) + (a bc 2 + 1) ≥ 2a bc + 2b ca + 2c a b.
p p p
Substituting x = a, y = b, z = c, where x, y, z ≥ 0, we need to show that
x 4 + y 4 + z 4 + x yz(x + y + z) ≥ 2(x 2 y 2 + y 2 z 2 + z 2 x 2 ).
This inequality can be obtained by summing Schur’s inequality of degree four
x 4 + y 4 + z 4 + x yz(x + y + z) ≥ x y(x 2 + y 2 ) + yz( y 2 + z 2 ) + z x(z 2 + x 2 )
to
x y(x 2 + y 2 ) + yz( y 2 + z 2 ) + z x(z 2 + x 2 ) ≥ 2(x 2 y 2 + y 2 z 2 + z 2 x 2 ),
which is equivalent to
x y(x − y)2 + yz( y − z)2 + z x(z − x)2 ≥ 0.
L
D
.M
P 3.4. If a, b, c are positive real numbers, then
A
a(b2 + c 2 ) + b(c 2 + a2 ) + c(a2 + b2 ) + 3 ≥ 3(a b + bc + ca).
which is equivalent to
W
For the nontrivial case 2(a + b + c) − 9 < 0, using the known inequality
(a + b + c)2 ≥ 3(a b + bc + ca),
it is enough to show that
[2(a + b + c) − 9](a + b + c)2 + 27 ≥ 0.
This inequality is equivalent to the obvious inequality
(a + b + c − 3)2 [2(a + b + c) + 3] ≥ 0.
The equality holds for a = b = c = 1.
Symmetric Polynomial Inequalities in Nonnegative Variables 173
Solution (by Vo Quoc Ba Can). Assume that a = min{a, b, c}. By virtue of the AM-GM
inequality, we have
3 3
a2 + b2 + c 2 (a2 + b2 + c 2 − 2bc) + bc + bc
= ≥ (a2 + b2 + c 2 − 2bc)b2 c 2
3 3
L
= a2 b2 c 2 + (b − c)2 b2 c 2 .
.M
Thus, it suffices to prove that
D
(b − c)2 b2 c 2 ≥ (b − c)2 (b − a)2 (c − a)2 .
A
PI
This is obvious, because b2 > (b − a)2 and c 2 > (c − a)2 . The equality occurs for
a = b = c.
M
LY
First Solution. Assume that a ≥ b ≥ c. For the nontrivial case b > 0, by the AM-GM
W
inequality, we have
W
2
a2 b2 + b2 c 2 + c 2 a2
2 2 2 2 2 2
4(a b + bc + ca)(a b + b c + c a ) ≤ b(a b + bc + ca) + .
W
a2 b2 + b2 c 2 + c 2 a2
a b(a + b) + bc(b + c) + ca(c + a) ≥ b(a b + bc + ca) + .
b
This inequality reduces to the obvious form
ac(a − b)(b − c) ≥ 0.
The equality holds for a = b = c, for b = c = 0 (or any cyclic permutation), and for
a = 0 and b = c (or any cyclic permutation).
174 Vasile Cîrtoaje
where
A = (a − b)2 (b − c)2 (c − a)2 .
Let p = a + b + c, q = a b + bc + ca, r = a bc. Since
L
.M
(pq − 3r)2 ≥ 4q(q2 − 2pr) − 27r 2 + 2(9pq − 2p3 )r + p2 q2 − 4q3 ,
which reduces to
D
r(p3 + 9r − 4pq) ≥ 0.
A
This is true since p3 + 9r − 4pq ≥ 0 (by the third degree Schur’s inequality).
PI
M
LY
we get
a3 + b3 + c 3 ≥ 2(a2 + b2 + c 2 ) − a − b − c = 2p2 − p − 4q.
(a) Using the result above and the known inequality p2 ≥ 3q, we have
a3 + b3 + c 3 + a b + bc + ca + 9 − 5(a + b + c) ≥
≥ (2p2 − p − 4q) + q + 9 − 5p = 2p2 − 6p + 9 − 3q
≥ 2p2 − 6p + 9 − p2 = (p − 3)2 ≥ 0.
L
P 3.8. If a, b, c are positive real numbers, then
.M
(a) a3 + b3 + c 3 + a bc + 8 ≥ 4(a + b + c);
D
(b) 4(a3 + b3 + c 3 ) + 15a bc + 54 ≥ 27(a + b + c).
A
PI
Solution. Let p = a + b + c and q = a b + bc + ca. By Schur’s inequality of third degree,
we have
M
p3 + 9a bc ≥ 4pq,
LY
p(4q − p2 )
a bc ≥ .
9
O
a) We have
.M
a3 + b3 + c 3 + a bc = 4a bc + p(p2 − 3q)
4p(4q − p2 )
≥ + p(p2 − 3q)
W
9
p(5p2 − 11q)
= .
W
9
W
p(5p2 − 11q)
+ 8 ≥ 4p,
9
which is equivalent to
5p3 − 36p + 72 ≥ 11pq.
Since p2 ≥ 3q, we have
(b) We have
(p − 3)2 (p + 6) ≥ 0.
L
.M
The equality holds for a = b = c = 1, and also for a = 0 and b = c = 3/2 (or any cyclic
permutation).
D
Remark. Similarly, we can prove the following generalization (Vasile Cîrtoaje, 2010):
A
• Let a, b, c be nonnegative real numbers. If 0 ≤ k ≤ 27/4, then
PI
a3 + b3 + c 3 + (k − 3)a bc + 2k ≥ k(a + b + c).
M
LY
a + b + c = a2 + b2 + c 2 .
.M
Prove that
a b + bc + ca ≥ a2 b2 + b2 c 2 + c 2 a2 .
W
Solution (by Michael Rozenberg). From the hypothesis condition, by squaring, we get
a4 + b4 + c 4 − a2 − b2 − c 2 = 2(a b + bc + ca − a2 b2 − b2 c 2 − c 2 a2 ).
W
a4 + b4 + c 4 ≥ a2 + b2 + c 2 .
which follows immediately from Hölder’s inequality. The equality holds for a = b =
c = 1, for a = b = c = 0, for (a, b, c) = (0, 1, 1) (or any cyclic permutation), and for
(a, b, c) = (1, 0, 0) (or any cyclic permutation).
Symmetric Polynomial Inequalities in Nonnegative Variables 177
Solution. We have
X X
(a2 + 2bc)(b2 + 2ca)(c 2 + 2a b) = 9a2 b2 c 2 + 2 a3 b3 + 4a bc a3
and X X
(a b + bc + ca)3 = 6a2 b2 c 2 + a3 b3 + 3a bc a b(a + b).
L
.M
So, we can rewrite the inequality as
X X X
3a2 b2 c 2 + a3 b3 + 4a bc a3 ≥ 3a bc a b(a + b).
D
A
a3 b3 ≥ 3a2 b2 c 2 (by the AM-GM inequality), it suffices to prove that
P
Since
6a bc + 4
X
a3 ≥ 3
X PI
a b(a + b).
M
We can get this inequality by summing the inequalities
LY
1X 3
a ≥ a bc
O
3
and
.M
X X
3a bc + a3 ≥ a b(a + b).
The first inequality follows from the AM-GM inequality, while the second is just the third
W
degree Schur’s inequality. The equality holds when a = b = c, and also when two of
W
a, b, c are zero.
Remark. Similarly, we can also prove the following inequality
W
and X X
(a b + bc + ca)3 = 6a2 b2 c 2 + a3 b3 + 3a bc a b(a + b),
L
We can get this inequality by summing
.M
2 X
a bc a3 ≥ 2a2 b2 c 2
3
D
A
and X X
a3 b3 + 3a2 b2 c 2 ≥ a bc a b(a + b).
PI
The first inequality follows from the AM-GM inequality, while the second is just the
M
third degree Schur’s inequality applied to the numbers a b, bc and ca. The equality
holds when a = b = c, and also when two of a, b, c are zero.
LY
O
Remark. Using the first method, we can also prove the following inequality
W
Solution. Assume that a = min{a, b, c}. It is easy to check that the equality holds in
both of the inequalities for a = 0 and b = c = 1.
(a) Since
a2 + b2 ≤ b(a + b)
and
c 2 + a2 ≤ c(c + a),
it suffices to show that
bc(b2 + c 2 ) ≤ b + c.
By virtue of the AM-GM inequality and the hypothesis b + c ≤ 2, we have
L
.M
2
2bc + (b2 + c 2 ) (b + c)4
2 2
2bc(b + c ) ≤ = ≤ 2(b + c).
2 4
D
The equality holds for a = 0 and b = c = 1 (or any cyclic permutation).
A
(b) First Solution. Since PI
a2 + b2 ≤ b(a + b)
M
and
c 2 + a2 ≤ c(a + c),
LY
3
W
b2 + c 2 + 4bc + 2a b + 2ac ≤ 6.
is true since
a2 + b2 ≤ (b + a/2)2 ,
b2 + c 2 ≤ (b + a/2)2 + (c + a/2)2 ,
c 2 + a2 ≤ (c + a/2)2 .
L
The right inequality holds if the original inequality holds for a = 0; that is,
.M
b2 c 2 (b2 + c 2 ) ≤ 2
D
for b + c = 2. Indeed, by virtue of the AM-GM inequality, we have
A
2
b+c PI
bc ≤ = 1,
2
M
2
2bc + (b2 + c 2 )
LY
2 2 2 2 2 2
2b c (b + c ) ≤ 2bc(b + c ) ≤
2
O
(b + c)4
= = 4.
4
.M
W
Solution. Due to symmetry, we may assume that a = min{a, b, c}. It is easy to check
that the equality holds for a = 0 and b = c = 1. Write the inequality as
Y Y
(a + b) (a2 − a b + b2 ) ≤ 2.
Since Y
(a + b) ≤ (a + b + c)(a b + bc + ca) = 2(a b + bc + ca),
it suffices to show that
Y
(a b + bc + ca) (a2 − a b + b2 ) ≤ 1.
Symmetric Polynomial Inequalities in Nonnegative Variables 181
Since
a2 − a b + b2 ≤ b2
and
c 2 + ca + a2 ≤ c 2 ,
it suffices to show that
b2 c 2 (a b + bc + ca)(b2 − bc + c 2 ) ≤ 1.
L
2 2 2 2
b c (a b + bc + ca)(b − bc + c ) ≤ .
.M
4
D
b2 + c 2 + 2bc + a b + ca ≤ 4.
A
This is true since PI
4 − (b2 + c 2 + 2bc + a b + ca) = (a + b + c)2 − (b2 + c 2 + 2bc + a b + ca)
M
= a(a + b + c) ≥ 0.
LY
Due to symmetry, we may assume that x = min{x, y, z}. It is easy to check that the
equality holds for x = 0 and y = z = 1. Since
and
x 2 + y 2 ≤ y(x + y), z 2 + x 2 ≤ z(x + z),
it suffices to show that
L
4
2 yz + 2(x y + yz + z x) + 2(x + y)(x + z) + ( y 2 + z 2 )
.M
≤ 16.
4
D
A
2x 2 + y 2 + z 2 + 6 yz + 4x y + 4z x ≤ 8,
PI
2x 2 + y 2 + z 2 + 6 yz + 4x y + 4z x ≤ 2(x + y + z)2 ,
M
= ( y − z)2 ≥ 0.
The equality holds for a = 0 and b = c = 1 (or any cyclic permutation).
LY
O
.M
Solution. Due to symmetry, assume that a = min{a, b, c}. Under this assumption, we
can check that the equality holds for (a, b, c) = (0, 1, 1). Since
W
and
0 < 3c 2 − 2ca + 3a2 ≤ c(a + 3c),
it suffices to show that
L
The equality holds for a = 0 and b = c = 1 (or any cyclic permutation).
.M
Remark. Similarly, we can prove the following more general statement.
D
2
• Let a, b, c be nonnegative real numbers. If ≤ k ≤ 2, then
3
A
4
PI
(a2 − ka b + b2 )(b2 − k bc + c 2 )(c 2 − kca + a2 ) ≤
27(2 + k)2
(a + b + c)6 ,
M
b c 3k
with equality for a = 0 and + =1+ (or any cyclic permutation).
c b 2
LY
2
Actually, this inequality holds for ≤ k ≤ 5.
3
O
.M
Solution. Assume that a ≤ b ≤ c. It is easy to show that the equality holds for a = 0
and b2 + c 2 = 7bc.pIf c 2 − 4ca + a2 < 2
p0, then a −p4a b + b < 0 and b − 4bc + c < 0,
2 2 2
W
since b ≤ c < (2+ 3)a and c < (2+ 3)a ≤ (2+ 3)b. Therefore, there two non-trivial
cases when the left hand side of the inequality is nonnegative: when all three factors
are nonnegative and when a2 − 4a b + b2 ≤ 0, b2 − 4bc + c 2 ≤ 0, c 2 − 4ca + a2 ≥ 0.
Case 1: a2 − 4a b + b2 ≥ 0, b2 − 4bc + c 2 ≥ 0, c 2 − 4ca + a2 ≥ 0. Since a2 − 4a b + b2 ≤ b2
and c 2 − 4ca + a2 ≤ c 2 , it suffices to prove that
b2 c 2 (b2 − 4bc + c 2 ) ≤ 3.
3
1 3bc + 3bc + (b2 − 4bc + c 2 )
≤
9 3
6
b + c) a + b + c) 6
=3 ≤3 = 3.
3 3
Since
L
(4a b − a2 − b2 )(4bc − b2 − c 2 )(c 2 − 4ca + a2 ) ≤
.M
3
(4a b − a2 − b2 ) + (4bc − b2 − c 2 ) + (c 2 − 4ca + a2 )
≤
3
D
8
A
= (2a b + 2bc − 2ca − b2 )3 ,
27
it suffices to prove that
PI
M
81 1
(2a b + 2bc − 2ca − b2 )3 ≤ = (a + b + c)6 ,
8 72
LY
which is equivalent to
O
p
3
2 9(2a b + 2bc − 2ca − b2 ) ≤ (a + b + c)2 .
.M
p
3 21
Since 2 9 < , it suffices to show that
5
W
p b2
From a ≥ (2 − 3)b, we get 4a > b, 5a2 > , and hence
20
b2 1
f (a) > + (13c − 8b)b + 26b2 + 5c 2 − 32bc = (19b − 10c)2 ≥ 0.
20 20
b c
This completes the proof. The equality holds for a = 0 and + = 7 (or any cyclic
c b
permutation).
Symmetric Polynomial Inequalities in Nonnegative Variables 185
L
4(a b + bc + ca) − 9 + ≥ 15.
a b + bc + ca
.M
This inequality is equivalent to
D
(a b + bc + ca − 3)2 ≥ 0.
A
The equality holds for a = b = c = 1.
PI
M
P 3.18. If a, b, c are positive real numbers such that a2 + b2 + c 2 = 3, then
LY
3
5(a + b + c) + ≥ 18.
O
a bc
.M
p
we get x > 1/ 3. By the known inequality
W
we get
1 4x
≥ .
a bc (3x − 1)2
2
1 2 p
15x 3 + 12x 2 − x − 2 > x 2 (12 − − 2 ) > x 2 (12 − 3 − 6) > 0.
x x
The equality holds for a = b = c = 1.
L
.M
12 + 9a bc ≥ 7(a b + bc + ca).
D
(Vasile Cîrtoaje, 2005)
A
Solution. Let x = (a + b + c)/3. Since
PI
2(a b + bc + ca) = (a + b + c)2 − (a2 + b2 + c 2 ) = 3(3x 2 − 1),
M
we can write the inequality as
LY
5 + 2a bc ≥ 7x 2 .
By Schur’s inequality of degree three
O
we get
W
3x 3 + a bc ≥ 2x(3x 2 − 1),
a bc ≥ 3x 3 − 2x.
W
Then,
W
(p2 − q)(4q − p2 )
a bc ≥ ,
6p
we get
(p2 + 3)(p2 − 6)
a bc ≥ .
12p
L
Therefore,
.M
3(p2 + 3)(p2 − 6) 13(p2 − 3)
21 + 18a bc − 13(a b + bc + ca) ≥ 21 + −
2p 2
D
(p − 3) (3p + 5p − 6)
2 2
A
= ≥ 0.
2p
The equality holds for a = b = c = 1.
PI
M
LY
Since
W
= 8 − 2(p2 − 3) + 2a bc p − a2 b2 c 2
= 14 − p2 − (p − a bc)2 ,
13 − p2 − (p − a bc)2 ≥ 0.
Clearly,
3(p − a bc) = (a2 + b2 + c 2 )(a + b + c) − 3a bc > 0.
By Schur’s inequality
we get
p3 + 9a bc ≥ 2p(p2 − 3),
p(p2 − 6)
a bc ≥ .
9
Since
p(p2 − 6) p(15 − p2 )
0 < p − a bc ≤ p − = ,
9 9
it suffices to prove that
p2 (15 − p2 )2
13 − p2 − ≥ 0.
L
81
p
.M
Setting p = 3 x, 0 < x ≤ 1, this inequality becomes
13 − 34x + 30x 2 − 9x 3 ≥ 0.
D
A
It is true because
PI
13 − 34x + 30x 2 − 9x 3 = (1 − x)(13 − 21x + 9x 2 )
M
= (1 − x)[1 + 3(1 − x)(4 − 3x)] ≥ 0.
LY
Second Solution. We use the mixing variable method. Assume that a ≤ 1 and show
that
.M
inequalities
2 − bc ≥ 2 − x 2
W
and
(2 − ca)(2 − a b) ≥ (2 − a x)2 .
and
p
4 − a(2x + b + c) ≥ 4(1 − a x) = 2(2 − a 6 − 2a2 )
4(1 − a2 )(2 − a2 )
= p ≥ 0.
2 + a 6 − 2a2
The right inequality (2 − x 2 )(2 − a x)2 ≥ 1 is equivalent to
(1 + a2 )(2 − a x)2 ≥ 2.
L
(1 + a)(2 − a x) ≥ 2.
.M
Indeed,
D
a(a4 + 2a3 − 2a2 − 6a + 5)
(1 + a)(2 − a x) − 2 = a(2 − x − a x) =
A
2(2 + x + a x)
a(a − 1)2 (a2 + 4a + 5)
PI
= ≥ 0.
2(2 + x + a x)
M
LY
a + b + c 5 a2 + b2 + c 2
.M
≥ .
3 3
W
b+c b+c
E(a, b, c) ≥ E(a, , ) ≥ 0,
2 2
where
E(a, b, c) = (a + b + c)5 − 81a bc(a2 + b2 + c 2 ).
Indeed, we have
1 b+c b+c
L
[E(a, b, c) − E(a, , )] =
81 2 2
.M
b+c 2 b+c 2
= a3 [( ) − bc] + a[2( ) − bc(b2 + c 2 )]
2 2
D
1 1
= a3 (b − c)2 + a(b − c)4 ≥ 0
A
4 8
and PI
b+c b+c
M
E(a, , )=
2 2
LY
81
= (a + b + c)5 − a(b + c)2 [2a2 + (b + c)2 ]
8
1
O
since
Solution. Since
a b c b c a
a3 + b3 + c 3 + a−3 + b−3 + c −3 + 3 = + + + +
b c a a b c
Symmetric Polynomial Inequalities in Nonnegative Variables 191
and
−1 −1 −1 a b c b c a
(a + b + c)(a +b +c )= + + + + + + 3,
b c a a b c
we can write the desired inequality in the homogeneous form
a b c b c a a b c b c a
+ + + + +9≥3 + + +3 + + ,
b c a a b c b c a a b c
or
a b c b c a
+ + −3 + + − 3 ≥ 0.
b c a a b c
This is true since, by the AM-GM inequality,
L
.M
a b c b c a
+ + ≥ 3, + + ≥ 3.
b c a a b c
D
The equality holds for a = b = c = 1.
A
PI
P 3.24. If a, b, c are positive real numbers such that a bc = 1, then
M
9
LY
a2 + b2 + c 2 − a b − bc − ca ≥ (a + b + c − 3).
4
O
3
where x = (a + b + c)/3. By the AM-GM inequality, we have x ≥ a bc = 1. The third
degree Schur’s inequality states that
W
which is equivalent to
3(3x 3 + 1)
4(a b + bc + ca) ≤ .
x
Therefore, it suffices to show that
3(3x 3 + 1)
3(4x 2 − 3x + 3) ≥ .
x
This inequality reduces to (x − 1)3 ≥ 0, which is obviously true for x ≥ 1. The equality
holds for a = b = c = 1.
192 Vasile Cîrtoaje
a2 + b2 + c 2 + a + b + c ≥ 2(a b + bc + ca).
p3 + 9a bc ≥ 4pq,
we get
p3 + 9
L
4q ≤ .
p
.M
Therefore,
D
p3 + 9
a2 + b2 + c 2 + a + b + c − 2(a b + bc + ca) = p2 + p − 4q ≥ p2 + p −
A
p
(p − 3)(p + 3)
PI
= ≥ 0.
p
M
The equality holds for a = b = c = 1.
LY
O
1 1 1
F (a, b, c) = a2 + b2 + c 2 + 15 + + − 16(a + b + c).
W
a b c
1 1 2 p
p
= (b − c)2 + 15 p − p − 16( b − c)2
b c
p p 15 p 15
p p p p
= ( b − c)2 ( b + c)2 + − 16 ≥ ( b − c)2 4 bc + − 16 ,
bc bc
it suffices to show that
15
4t + − 16 ≥ 0.
t2
Indeed,
15 15 (1 − t)(15 − t)
4t + − 16 > t + − 16 = ≥ 0.
t2 t t
L
The inequality F (a, t, t) ≥ 0 is equivalent to
.M
(t − 1)2 (17t 4 + 2t 3 − 13t 2 + 2t + 1) ≥ 0.
D
We have
A
17t 4 + 2t 3 − 13t 2 + 2t + 1 = (2t − 1)4 + t(t 3 + 34t 2 − 37t + 10)
t
PI
= (2t − 1)4 + [t(2t − 1)2 + 140t 2 − 149t + 40] > 0
M
4
since D = 1492 − 4 · 140 · 40 = −199. The equality holds for a = b = c = 1.
LY
O
2 1 3
+ ≥ .
a+b+c 3 a b + bc + ca
W
W
p
3
x ≥ a b · bc · ca = 1.
The third degree Schur’s inequality applied to a b, bc, ca, states that
which is equivalent to
3 4x
≥ .
a+b+c 3x 3 + 1
Therefore,
2 1 3 8x 3
3 + − ≥ + 1 −
a + b + c 3 a b + bc + ca 3x 3 + 1 x
194 Vasile Cîrtoaje
3x 4 − 9x 3 + 8x 2 + x − 3 (x − 1)(3x 3 − 6x 2 + 2x + 3)
= = .
x(3x 3 + 1) x(3x 3 + 1)
Since x ≥ 1, we need to show that 3x 3 − 6x 2 + 2x + 3 ≥ 0. For x ≥ 2, we have
3x 3 − 6x 2 + 2x + 3 > 3x 3 − 6x 2 = 3x 2 (x − 2) ≥ 0,
L
D
.M
P 3.28. If a, b, c are positive real numbers such that a bc = 1, then
A
6
a b + bc + ca + ≥ 5.
PI
a+b+c
M
(Vasile Cîrtoaje, 2005)
LY
(a + b + c)(3x − 5) + 6 ≥ 0.
O
In virtue of the AM-GM inequality, we get x ≥ 1. Since the inequality holds for x ≥ 5/3,
.M
consider next that 1 ≤ x < 5/3. Applying the third degree Schur’s inequality to the
numbers a b, bc and ca, we have
W
which is equivalent to
W
3(3x 3 + 1)
a+b+c ≤ .
4x
Having in view that 3x − 5 < 0, it suffices to prove that
3(3x 3 + 1)(3x − 5)
+ 6 ≥ 0.
4x
This inequality is equivalent to
9x 4 − 15x 3 + 11x − 5 ≥ 0,
(x − 1)(9x 3 − 6x 2 − 6x + 5) ≥ 0.
Symmetric Polynomial Inequalities in Nonnegative Variables 195
Since
Second Solution (by Vo Quoc Ba Can). Among a − 1, b − 1, c − 1 there are two with
the same sign. Due to symmetry, assume that (b − 1)(c − 1) ≥ 0; that is, b + c ≤ 1 + bc.
Then,
6 6 6a
≥ = 2 .
a+b+c a + 1 + bc a +a+1
L
On the other hand, using the AM-GM inequality yields
.M
p p 1
a b + bc + ca = a(b + c) + bc ≥ 2a bc + bc = 2 a + .
a
D
Therefore, it suffices to prove that
A
p 1
2 a+ + 2
6a
a a +a+1
≥ 5.
PI
M
p
Setting a = x, this inequality becomes as follows
LY
1 6x 2
2x + + ≥ 5,
O
x2 x4 + x2 + 1
.M
1 6x 2
2x + − 3 ≥ 2 − ,
x2 x4 + x2 + 1
W
(x − 1)2 (2x 5 − x 4 − 2x 3 − x 2 + 2x + 1) ≥ 0,
W
Solution. Since
which is equivalent to
(1 + q)2 ≥ 4(1 + p),
L
where p = a + b + c, q = a b + bc + ca. Setting x = bc, y = ca, z = a b in Schur’s
.M
inequality
(x + y + z)3 + 9x yz ≥ 4(x + y + z)(x y + yz + z x),
D
we get
A
q3 + 9 ≥ 4pq.
Since
PI
M
q3 + 9
(1 + q)2 − 4(1 + p) ≥ (1 + q)2 − 4 −
LY
q
(q − 3)(2q + 3)
= ,
O
q
.M
Solution. Due to homogeneity, we may assume that a bc = 1, when the inequality can
be written as
a6 + b6 + c 6 − 3 ≥ 18(a3 + b3 + c 3 − a3 b3 − b3 c 3 − c 3 a3 ).
Symmetric Polynomial Inequalities in Nonnegative Variables 197
To do this, we use the mixing variable method. Without loss of generality, assume that
a ≥ 1. We claim that p p
F (a, b, c) ≥ F (a, bc, bc) ≥ 0.
We have
p p p p p p
F (a, b, c) − F (a, bc, bc) = (b − c)2 − 18( b − c)2 + 18a( b − c)2
L
p p
.M
= (b − c)2 + 18(a − 1)( b − c)2 ≥ 0.
p
Also, putting bc = t, we have
D
A
p p 1 1 18 36
F (a, bc, bc) = F ( 2
, t, t) = 4 + 20t 2 − 3 − 2 − 36t +
t t PI t t
(t − 1)2 (2t − 1)2 (t + 1)(5t + 1)
= ≥ 0.
M
t4
The equality holds for a = b = c, and for a/2 = b = c (or any cyclic permutation).
LY
O
.M
1 1 1
+ 2 + 2 ≥ a2 + b2 + c 2 .
W
a 2 b c
W
Then,
a bc(a2 + b2 + c 2 ) − 3 ≤ x 2 (9 − 6x) − 3 = −3(x − 1)2 (2x + 1) ≤ 0.
L
X 1
( 2 − a2 + 4a − 4) ≥ 0,
.M
a
which is equivalent to
D
X (1 − a)2 (1 + 2a − a2 )
A
≥ 0.
a2
p p
PI
Without loss of generality, assume that a = max{a, b, c}. We have two cases to consider.
Case 1: a ≤ 1 + 2. Since a, b, c ≤ 1 + 2, we have
M
1 + 2a − a2 ≥ 0, 1 + 2b − b2 ≥ 0, 1 + 2c − c 2 ≥ 0.
LY
O
1 1
bc ≤ (b + c)2 < ,
W
4 9
W
and hence
1 1 1 1 1 2
W
2
+ 2+ 2> 2+ 2≥ > 18 > (a + b + c)2 > a2 + b2 + c 2 .
a b c b c bc
a3 + b3 + c 3 + 7a bc ≥ 10.
we get x ≥ 1. Since
L
.M
4
For x ≥ , this inequality is true, since
3
D
16
27x 3 − 27x − 10 = 27x(x 2 − 1) − 10 ≥ 36( − 1) − 10 = 18.
A
9
For 1 ≤ x ≤
4
, we use Schur’s inequality
PI
3
M
(a + b + c)3 + 9a bc ≥ 4(a + b + c)(a b + bc + ca),
LY
which is equivalent to
O
a bc + 3x 3 − 4x ≥ 0.
.M
Therefore,
= (x − 1)[4 − 3x + 3(2 − x 2 )] ≥ 0.
W
a4 b4 + b4 c 4 + c 4 a4 ≤ 3.
b3 + c 3 + 1 4 − a3
bc ≤ = .
3 3
200 Vasile Cîrtoaje
Then, we have
4b3 c 3 − a3 b3 c 3
b4 c 4 ≤ .
3
Similarly,
4c 3 a3 − a3 b3 c 3 4a3 b3 − a3 b3 c 3
c 4 a4 ≤ , a4 b4 ≤ .
3 3
Summing these inequalities, we obtain
4(a3 b3 + b3 c 3 + c 3 a3 )
a4 b4 + b4 c 4 + c 4 a4 ≤ − a3 b3 c 3 .
3
L
Thus, using the substitutions x = a3 , y = b3 , z = c 3 , it suffices to prove that
.M
4(x y + yz + z x) ≤ 3x yz + 9,
D
where x, y, z are nonnegative real numbers satisfying x + y + z = 3. This follows imme-
A
diately from Schur’s inequality
PI
4(x + y + z)(x y + yz + z x) ≤ 9x yz + (x + y + z)3 .
M
The equality holds for a = b = c = 1.
LY
Remark 1. Using the contradiction method, it is easy to prove the reverse statement.
• If a, b, c are nonnegative real numbers such that a4 b4 + b4 c 4 + c 4 a4 = 3, then
O
a3 + b3 + c 3 ≥ 3.
.M
Remark 2. The inequality in P 3.33 is a particular case of the following more general
statement (Vasile Cîrtoaje, 2003).
W
ln 3
k0 = ≈ 1.355,
W
ln 9 − ln 4
then
a k b k + b k c k + c k a k ≤ 3.
L
Æ
a bc(a + b + c)(a b + bc + ca) ≥ 3a bc.
.M
This is true since
D
(a + b + c)(a b + bc + ca) − 9a bc = a(b − c)2 + b(c − a)2 + c(a − b)2 ≥ 0.
A
The equality holds for a = b = c = 1.
PI
M
LY
1 + 8a bc ≥ 9 min{a, b, c}.
.M
Solution. Without loss of generality, assume that a = min{a, b, c}, a ≤ 1. The inequality
W
becomes
1 + 8a bc ≥ 9a.
W
Therefore,
1 1 5
The equality holds for a = b = c = 1, and also for (a, b, c) = , , or any cyclic
4 4 2
permutation.
202 Vasile Cîrtoaje
1 + 4a bc ≥ 5 min{a, b, c}.
Solution. Without loss of generality, assume that a = min{a, b, c}, a ≤ 1. The inequality
can be written as
1 + 4a bc ≥ 5a.
From (a2 − b2 )(a2 − c 2 ) ≥ 0, we get
L
p p
bc ≥ a b2 + c 2 − a2 = a 3 − 2a2 .
.M
Therefore, it suffices to prove that
D
p
4a2 3 − 2a2 ≥ 5a − 1.
A
We consider two cases. p p PI
Case 1: 0 < a ≤ 1/2. Since 3 − 2a2 ≥ 5/2 > 25/16, it is enough to show that
M
25 2
a ≥ 5a − 1.
4
LY
or
W
It is true, since
32a5 + 32a4 − 16a3 − 16a2 + 9a − 1 =
W
L
t > 0. By the AM-GM inequality, we have
.M
Æ3
2 = x y + yz + z x + x yz ≥ 3 x 2 y 2 z 2 + x yz = 3t 2 + t 3 ,
D
and hence
A
t 3 + 3t 2 − 2 ≤ 0,
PI
(t + 1)(t 2 + 2t − 2) ≤ 0,
M
t 2 + 2t − 2 ≤ 0,
p
LY
t ≤ 3 − 1,
p
x yz ≤ ( 3 − 1)3 .
O
p
The equality holds for a = b = c = 3.
.M
W
Let
E(a, b, c) = (2a + c)(b2 + c 2 + a b + ac).
We will show that
E(a, b, c) ≤ E(a, b + c, 0) ≤ 4.
Indeed,
E(a, b, c) − E(a, b + c, 0) = c(b2 + c 2 + ac − 3a b) ≤ 0
and
L
= 4a(2 − a) − 4 = −4(a − 1)2 ≤ 0.
.M
The equality occurs for (a, b, c) = (1, 1, 0) or any cyclic permutation.
D
A
P 3.39. If a, b, c are nonnegative real numbers, then
PI
M
(8a2 + bc)(8b2 + ca)(8c 2 + a b) ≤ (a + b + c)6 .
LY
Solution. We use the mixing variable method. Without loss of generality, assume that
O
E(a, b, c) ≤ E(a, x, x) ≤ 0.
W
8a2 + x 2 ≥ 8a2 + bc
and
(8x 2 + a x)2 ≥ (8b2 + ca)(8c 2 + a b).
Write the last inequality as
Since
3(b + c)(b − c)2
b3 + c 3 − 2x 3 = = 6x(x 2 − bc) ≥ 0,
4
we need to show that
64(x 2 + bc) + a2 − 48a x ≥ 0.
This is true, since
L
.M
176x 5 − 273a x 4 + 32a2 x 3 + 52a3 x 2 + 12a4 x + a5 ≥ 0,
(x − a)2 (176x 3 + 79a x 2 + 14a2 x + a3 ) ≥ 0,
D
the last being clearly true. The equality holds for a = b = c, and for a = 0 and b = c
A
(or any cyclic permutation).
PI
M
P 3.40. If a, b, c are positive real numbers such that a2 b2 + b2 c 2 + c 2 a2 = 3, then
LY
a + b + c ≥ a bc + 2.
O
v v
t 3 − a2 b2 t 3 − a2 b2
c= ≤ .
a2 + b2
W
2a b
We have
W
a + b + c − a bc − 2 = a + b − 2 − (a b − 1)c
v
p t 3 − a2 b2
≥ 2 a b − 2 − (a b − 1)
2a b
v
p p t 3 − a2 b2
= ( a b − 1) 2 − ( a b + 1) .
2a b
L
.M
P 3.41. Let a, b, c be nonnegative real numbers such that a + b + c = 5. Prove that
D
(a2 + 3)(b2 + 3)(c 2 + 3) ≥ 192.
A
PI
First Solution. Without loss of generality, assume that a = min{a, b, c}, a ≤
5
. By
3
M
virtue of the Cauchy-Schwarz inequality, we have
LY
5
for 0 ≤ a ≤ . Indeed,
W
3
W
since
5 19 4
2
a − 8a + 11 = −a − a + > 0.
3 3 9
The equality holds for a = 3 and b = c = 1 (or any cyclic permutation).
Second Solution. Without loss of generality, assume that a = ma x{a, b, c}. First, we
show that
(b2 + 3)(c 2 + 3) ≥ (x 2 + 3)2 ,
b+c 5
where x = , 0 ≤ x ≤ . This inequality is equivalent to
2 3
(b − c)2 (6 − bc − x 2 ) ≥ 0,
Symmetric Polynomial Inequalities in Nonnegative Variables 207
which is equivalent to
[(5 − 2x)2 + 3](x 2 + 3)2 ≥ 192,
(x 2 − 5x + 7)(x 2 + 3)2 ≥ 48,
(x − 1)2 (x 4 − 3x 3 + 6x 2 − 15x + 15) ≥ 0.
L
This inequality is true since
.M
2 2
3
x
4 3 2 2
x − 3x + 6x − 15x + 15 = x x− + 15 −1 > 0.
2 2
D
A
P 3.42. If a, b, c are nonnegative real numbers, then
PI
M
a2 + b2 + c 2 + a bc + 2 ≥ a + b + c + a b + bc + ca.
LY
Solution. Among the numbers 1− a, 1− b and 1− c there are always two with the same
sign; let us say (1 − b)(1 − c) ≥ 0. Thus, it suffices to show that
.M
a2 + b2 + c 2 + a(b + c − 1) + 2 ≥ a + b + c + a b + bc + ca,
which is equivalent to
W
a2 − 2a + b2 + c 2 − bc − (b + c) + 2 ≥ 0.
W
Since
W
1
b2 + c 2 − bc ≥ (b + c)2 ,
4
it suffices to show that
1
a2 − 2a + (b + c)2 − (b + c) + 2 ≥ 0,
4
which can be written in the obvious form
2
b+c
(a − 1)2 + − 1 ≥ 0.
2
The equality holds for a = b = c = 1.
208 Vasile Cîrtoaje
a3 (b + c)(a − b)(a − c) ≥ 0
and
b3 (c + a)(b − c)(b − a) + c 3 (a + b)(c − a)(c − b) =
L
.M
(b − c)[bc(b3 − c 3 ) + (b − c)(b3 + c 3 )a − (b3 − c 3 )a2 ]
D
A
≥ (b − c)2 (b2 + bc + c 2 )(bc − a2 ),
Since
bc − a2 = (a − b)(a − c) + a(b + c − 2a) ≥ (a − b)(a − c),
O
(b − c)2 + 3a(b + c − a) ≥ 0.
W
The equality holds for a = b = c, for a = 0 and b = c (or any cyclic permutation), and
W
a + b + c + 4a bc ≥ k(a b + bc + ca)
Since b + c − 2bc = b(1 − c) + c(1 − b) ≥ 0, the inequality is clearly true for 1 + 4bc −
2b − 2c ≥ 0. Consider further that 1 + 4bc − 2b − 2c < 0. It suffices to show that
L
(1 + 4bc − 2b − 2c) + b + c − 2bc ≥ 0,
.M
which is equivalent to the obvious inequality
D
bc + (1 − b)(1 − c) ≥ 0.
A
PI
Thus, the proof is completed. For k = 2, the equality holds when a = b = c = 0, and
when one of a, b, c is zero and the others are 1.
M
Remark. From the proof above it follows that the following stronger inequality holds
LY
a + b + c + 3a bc ≥ 2(a b + bc + ca),
.M
2
P 3.45. If a, b, c ≥ such that a + b + c = 3, then
W
a2 b2 + b2 c 2 + c 2 a2 ≥ a b + bc + ca.
Solution. We use the mixing variable method. Assume that a = max{a, b, c} and denote
x = (b + c)/2. From a, b, c ≥ 2/3 and a + b + c = 3, it follows that 2/3 ≤ x ≤ 1. We
will show that
E(a, b, c) ≥ E(a, x, x) ≥ 0,
where
E(a, b, c) = a2 b2 + b2 c 2 + c 2 a2 − a b − bc − ca.
210 Vasile Cîrtoaje
We have
L
9E(a, x, x) = 18a2 x 2 + 9x 4 − (2a x + x 2 )(a + 2x)2
.M
= x(5x 3 − 12a x 2 + 9a2 x − 2a3 ) = x(x − a)2 (5x − 2a)
= 3x(x − a)2 (3x − 2) ≥ 0.
D
The equality holds for a = b = c = 1, and also for a = 5/3, b = 2/3, c = 2/3 (or any
A
cyclic permutation).
PI
M
LY
a + b + c = 3,
O
then
.M
1 1 1
+ + ≥ a2 + b2 + c 2 .
a b c
W
Solution. Let
1 1 1
W
F (a, b, c) = + + − a2 − b2 − c 2 .
a b c
W
F (a, b, c) − F (a, 1, b + c − 1) =
1 1 1
= + −1− + 1 + (b + c − 1)2 − b2 − c 2
b c b+c−1
1 1
= (b + c) − + 2(b − 1)(c − 1)
bc b + c − 1
b+c
= (b − 1)(c − 1) 2 −
bc(b + c − 1)
Symmetric Polynomial Inequalities in Nonnegative Variables 211
and
1 1 (x + a)4 (x − a)4
F (a, 1, x) = + − a2 − x 2 = − a2 − x 2 = ≥ 0.
a x 8a x 8a x
L
.M
The equality holds for a = b = c = 1.
D
A
P 3.47. If a, b, c are positive real numbers such that a ≤ 1 ≤ b ≤ c and
a+b+c =
1 1 1
+ + ,
PI
M
a b c
then
LY
1 1 1
a2 + b2 + c 2 ≤ 2
+ 2 + 2.
a b c
O
1 1
b2 − ≤ (a 2
+ c 2
) − 1 .
W
b2 a2 c 2
W
1 1 1
From a + b + c = + + , we have
a b c
W
1 1
b − = (a + c) − 1 ≥ 0.
b ac
Thus, the desired inequality holds if
1 1
(a + c)(b + ) ≤ (a2 + c 2 )( + 1).
b ac
1
On the other hand, from (b − c)(1 − ) ≤ 0, one gets
bc
1 1
b+ ≥c+ .
b c
212 Vasile Cîrtoaje
L
1 1 1
a+b+c = + + ,
.M
a b c
then
1 1 1
D
(1 − a bc) a n + b n + c n − n − n − n ≥ 0
a b c
A
for any integer n ≥ 2.
PI (Vasile Cîrtoaje, 2007)
Solution. Since the statement remains unchanged by substituting a, b, c with 1/a, 1/b,
M
1/c, respectively, it suffices to prove that
LY
1 1 1
an + bn + c n −
n
− n − n ≤0
a b c
O
(a b − 1)(bc − 1)(ca − 1) = a2 b2 c 2 − 1,
W
(x − 1)( y − 1)(z − 1) = x yz − 1 ≥ 0
involves
(x n − 1)( y n − 1)(z n − 1) ≥ x n y n z n − 1.
This inequality holds if
(x n−1 + x n−2 + · · · + 1)( y n−1 + y n−2 + · · · + 1)(z n−1 + z n−2 + · · · + 1) ≥
≥ x n−1 y n−1 z n−1 + x n−2 y n−2 z n−2 + · · · + 1.
Since the last inequality is clearly true, the proof is completed. The equality occurs for
a = bc = 1, or b = ca = 1, or c = a b = 1.
Symmetric Polynomial Inequalities in Nonnegative Variables 213
L
X
a2 (a − b)(a − c) ≥ 0,
.M
we have
D
X X
(a + b + c)E(a, b, c) = a2 (a − b)(a − c) + a(b + c)(a − b)(a − c)
A
X
≥ a(b + c)(a − b)(a − c)
=
X
a b(a − b)(a − c) +
X PI
ac(a − b)(a − c)
M
X X
= a b(a − b)(a − c) + ba(b − c)(b − a)
LY
X
= a b(a − b)2 ≥ 0.
O
The equality holds for a = b = c. If a, b, c are nonnegative real numbers, then the
equality also holds for a = 0 and b = c (or any cyclic permutation).
.M
(b) Since
(a b + bc + ca)E(a, b, c) =
W
X X
= a bc (a − b)(a − c) + (a2 b + a2 c)(a − b)(a − c)
W
1 X X
= a bc (a − b)2 + [a2 b(a − b)(a − c) + b2 a(b − c)(b − a)]
2
W
1 X X
= a bc (a − b)2 + a b(a − b)2 (a + b − c),
2
the required inequality is equivalent to
X
a b(a − b)2 (a + b − c) ≥ 0.
Without loss of generality, assume that a ≥ b ≥ c. Then,
X
a b(a − b)2 (a + b − c) ≥ bc(b − c)2 (b + c − a) + ac(a − c)2 (a + c − b)
≥ bc(b − c)2 (b + c − a) + ac(b − c)2 (a + c − b)
= c(b − c)2 [(a − b)2 + c(a + b)] ≥ 0.
214 Vasile Cîrtoaje
L
Prove that (a) is sharper than (b) if
.M
p p p
b+ c≤ a,
D
and (b) is sharper than (a) if
A
p p p
b+ c≥ a.
PI (Vasile Cîrtoaje, 2005)
M
Solution. Let p = a + b + c and q = a b + bc + ca. If we rewrite Schur’s inequalities
LY
above as
a bc ≥ f (p, q)
O
and
.M
a bc ≥ g(p, q),
respectively, then (a) is sharper than (b) if f (p, q) ≥ g(p, q), while (b) is sharper than
(a) if g(p, q) ≥ f (p, q). Therefore, we need to show that
W
p p p
( b + c − a)[g(p, q) − f (p, q)] ≥ 0.
W
W
from which
p(4q − p2 )
f (p, q) = .
9
Analogously, write the inequality in (b) as
a4 + b4 + c 4 + 2a bc(a + b + c) ≥ (a b + bc + ca)(a2 + b2 + c 2 ).
Since
a2 + b2 + c 2 = p2 − 2q
L
.M
and
a4 + b4 + c 4 = (a2 + b2 + c 2 )2 − 2(a2 b2 + b2 c 2 + c 2 a2 )
D
= (p2 − 2q)2 − 2q2 + 4a bc p,
A
we get PI
(p2 − q)(4q − p2 )
M
g(p, q) = .
6p
LY
Therefore, we have
(p2 − 3q)(4q − p2 )
g(p, q) − f (p, q) =
O
,
18p
.M
and hence
(p2 − 3q)(4q − p2 )2
(4q − p2 )[g(p, q) − f (p, q)] = ≥ 0.
18p
W
Remark. If a, b, c are the lengths of the sides of a triangle, then Schur’s inequality of
W
(a + b)(b + c)(c + a) = 8,
then
p p p
a+ b+ c ≥ a b + bc + ca.
First Solution. Assume that a ≥ b ≥ c, and write the inequality in the equivalent ho-
mogeneous forms
p p p Æ p
( a + b + c) (a + b)(b + c)(c + a) ≥ 2 2(a b + bc + ca),
XÆ Æ Æ
a(b + c)[ (a + b)(a + c) − 2a(b + c)] ≥ 0,
p
X (a − b)(a − c) a(b + c)
p p ≥ 0.
(a + b)(a + c) + 2a(b + c)
Since (c − a)(c − b) ≥ 0, it suffices to prove that
L
p p
(a − b)(a − c) a(b + c) (b − c)(b − a) b(c + a)
.M
p p +p p ≥ 0,
(a + b)(a + c) + 2a(b + c) (b + c)(b + a) + 2b(c + a)
D
which is true if
A
p p
(a − c) a(b + c) (b − c) b(c + a)
p p
(a + b)(a + c) + 2a(b + c)
≥p p PI
(b + c)(b + a) + 2b(c + a)
.
M
p p
Since a≥ b,
LY
Æ Æ
(a + b)(a + c) ≥ 2a(b + c)
and
O
Æ Æ
(b + c)(b + a) ≤ 2b(a + c),
.M
p ≥p .
(a + b)(a + c) (b + c)(b + a)
W
q2 + a bc p
p+ ≥ q2 ,
2
or
p(2 + a bc) ≥ q2 .
According to the hypothesis pq − a bc = 8, we can write this inequality in the homoge-
neous forms
pq − a bc
p + a bc ≥ q2 ,
4
L
p2 q + 3a bc p ≥ 4q2 .
.M
Since p2 ≥ 3q and p3 + 9a bc ≥ 4pq (Schur’s inequality), we have
D
p(p2 q + 3a bc p − 4q2 ) ≥ q(p3 + 9a bc − 4pq) ≥ 0.
A
PI
p
M
P 3.52. If a, b, c ∈ [1, 4 + 3 2], then
LY
= (A − B)(4B − A)
W
and
2(A − B) = (a − b)2 + (b − c)2 + (c − a)2 ≥ 0,
W
From
E(a, b, c) − E(1, b, c) = (a − 1)(a + 1 − 4b − 4c) ≥ 0,
218 Vasile Cîrtoaje
we get
4(b + c) − a ≤ 1,
and from
E(a, b, c) − E(w, b, c) = (a − w)(a + w − 4b − 4c) ≥ 0,
we get
4(b + c) − a ≥ w.
These results involve w ≤ 1, which is false. Therefore, since E(a, b, c) is symmetric, we
have
E(a, b, c) ≤ max{E(1, 1, 1), E(1, 1, w), E(1, w, w), E(w, w, w)}
L
= max{−9, w2 − 8w − 2, 1 − 2w2 − 8w, −9w2 } = w2 − 8w − 2 = 0.
.M
This completes
p the proof. The equality holds for a = b = c, and also for a = b = 1 and
c = 4 + 3 2 (or any cyclic permutation).
D
A
PI
P 3.53. If a, b, c are nonnegative real numbers such that a + b + c + a bc = 4, then
M
(a) a2 + b2 + c 2 + 12 ≥ 5(a b + bc + ca);
LY
p3 ≥ 27(4 − p),
(p − 3)(p2 + 3p + 36) ≥ 0,
p ≥ 3.
The equality holds for a = b = c = 1, and for a = 0 and b = c = 2 (or any cyclic
permutation).
Symmetric Polynomial Inequalities in Nonnegative Variables 219
(b) We need to show that 3p2 + 7q ≥ 48 for p + r = 4. Using the known inequality
pq ≥ 9r, we get
L
.M
a b + bc + ca ≥ 1 + 2a bc.
D
A
Solution. Write the inequality such that the left-hand side and right-hand side are ho-
mogeneous expressions
PI
3(a b + bc + ca) − a2 − b2 − c 2 ≥ 6a bc.
M
From
LY
3(a2 + b2 + c 2 ) ≥ (a + b + c)2 ,
we get a + b + c ≤ 3. Therefore, it suffices to prove the homogeneous inequality
O
This is equivalent to
W
can be written as
which is just the third degree Schur’s inequality. The equality holds for an equilateral
triangle.
a2 b2 + b2 c 2 + c 2 a2 ≥ a b + bc + ca.
220 Vasile Cîrtoaje
3[3(a2 b2 + b2 c 2 + c 2 a2 ) − (a b + bc + ca)2 ] ≥
L
X
Sa (b − c)2 ≥ 0,
.M
where
Sa = 6a2 − a b − bc − ca.
D
A
Without loss of generality, assume that a ≥ b ≥ c. It suffices to show that
PI
S b (a − c)2 + Sc (a − b)2 ≥ 0.
M
Since
LY
(a − c)2 ≥ (a − b)2 ,
and
.M
we get
S b (a − c)2 + Sc (a − b)2 ≥ (S b + Sc )(a − b)2 ≥ 0.
W
1 1 1 41
+ + + ≥ 3(a2 + b2 + c 2 ).
a b c 6
1 1 1 41 3
+ + + ≥ [( y + z)2 + (z + x)2 + (x + y)2 ],
y + z z + x x + y 12 8
X x + y + z 41 9 X X
+ ≥ x2 + xy ,
y +z 4 4
L
X x 41 9 X
+3+ ≥ 9− xy ,
.M
y +z 4 4
X x 9X
≥7− x y.
D
y +z 4
A
Let us denote t = x y + yz + z x. Since
X x
= ≥
PI
1 X x(x y + yz + z x) 1 X x(x y + z x)
y +z t y +z t y +z
M
1 X 2 9 − 2t
= x = ,
LY
t t
it suffices to show that
O
9 − 2t 9
≥ 7 − t,
t 4
.M
which is equivalent to
(t − 2)2 ≥ 0.
W
The equality holds for a degenerate triangle having a = 3/2, b = 1, c = 1/2 (or any
permutation thereof).
W
W
Solution. (a) The proof is similar to the first proof of problem P 2.53. We have here
p
p − 2 p2 − 3q
, 3q ≤ p2 ≤ 4q
a1 = 3 .
2
0, p ≥ 4q
Therefore, a attains its minimum value a1 when b = c (if p2 ≤ 4q) or a = 0 (if p2 ≥ 4q).
This means that r is minimal when b = c or a = 0.
(b) Using the known substitutions a = y+z, b = z+x, c = x + y, where x ≥ y ≥ z ≥ 0,
L
from
.M
a + b + c = 2(x + y + z),
a b + bc + ca = (x + y + z)2 + x y + yz + z x,
D
a bc = (x + y + z)(x y + yz + z x) − x yz,
A
it follows that
x + y +z =
p
,
PI
x y + yz + z x = q −
p2
,
M
2 4
and hence
LY
p2
p
a bc = q− − x yz.
2 4
O
Therefore, the product a bc is maximal when x yz is minimal; that is, according to (a),
.M
Remark. Using the result in (a), we can prove by the contradiction method (as in
W
where p and p1 are fixed real numbers satisfying p2 ≤ np1 . Then, the product
r = a1 a2 · · · an
L
9 25
= 2 + 16 −
.M
a (3 − 2a) a(2 − a)
2(16a − 56a + 73a2 − 42a + 9)
4 3
=
D
a2 (3 − 2a)(2 − a)
2(a − 1)2 (4a − 3)2
A
= 2 ≥ 0,
a (3 − 2a)(2 − a)
PI 3 3 3
as desired. The equality holds for (a, b, c) = (1, 1, 1), and also for (a, b, c) =
M
, ,
4 4 2
or any cyclic permutation.
LY
O
1 1 1
8 + + + 9 ≥ 10(a2 + b2 + c 2 ).
a b c
W
8q
+ 20q ≥ 81.
a bc
By P 3.57-(a), the product a bc is maximal for fixed q when two of a, b, c are equal.
Therefore, it suffices to prove the inequality for b = a and c = 3 − 2a, a < 3/2. We have
8q 24a(2 − a)
+ 20q − 81 = 2 + 60a(2 − a) − 81
a bc a (3 − 2a)
3a(40a4 − 140a3 + 174a2 − 89a + 16)
=
a2 (3 − 2a)
3a(2a − 1)2 (10a2 − 25a + 16)
= .
a2 (3 − 2a)
224 Vasile Cîrtoaje
Since
5 2 3
10a2 − 25a + 16 = 10 a − + > 0,
4 8
1 1
the proof is completed. The equality holds for (a, b, c) = , , 2 or any cyclic permu-
2 2
tation.
Second Solution (by Vo Quoc Ba Can). It is easy to check that the equality holds when
two of a, b, c are 1/2. Then, let us define
8
f (x) = − 10x 2 − αx − β,
x
L
such that (2x − 1)2 divides f (x). From f (1/2) = 0, we get α + 2β = 27. Therefore,
.M
8 (1 − 2x)h(x)
f (x) = − 10x 2 − (27 − 2β)x − β = ,
D
x x
A
where h(x) = 5x 2 − (β − 16)x + 8. From h(1/2) = 0, we get β = 69/2, and hence
h(x) =
PI
(1 − 2x)(16 − 5x)
,
2
M
(1 − 2x)2 (16 − 5x)
LY
f (x) = .
2x
Thus, we can write the inequality in the form
O
which is equivalent to
a b c
≥ + .
5a − 1 16 − 5b 16 − 5c
Indeed,
a b c a b+c
− − ≥ −
5a − 1 16 − 5b 16 − 5c 5a − 1 16 − 5a
3
= > 0.
(5a − 1)(16 − 5a)
L
Remark. Using the second method, we can prove the following more general inequality.
.M
• If x 1 , x 2 , · · · , x n are positive real numbers such that
D
x 1 + x 2 + · · · + x n = n,
A
then PI
1 1
M
(n + 1)2 + ··· + ≥ n(n2 − 3n − 6) + 4(n + 2)(x 12 + x 22 + · · · + x n2 ),
x1 xn
LY
with equality for x 1 = (n + 1)/2 and x 2 = · · · = x n = 1/2 (or any cyclic permutation).
O
.M
a2 + b2 + c 2 = 9 − 2q,
a2 b2 + b2 c 2 + c 2 a2 = q2 − 6a bc,
we can rewrite the inequality as
Since 3
a+b+c
a bc ≤ = 1,
3
226 Vasile Cîrtoaje
1 1
The equality holds for a = b = c = 1, and for (a, b, c) = , , 2 or any cyclic permu-
2 2
tation.
L
.M
P 3.61. If a, b, c are nonnegative real numbers, then
D
(a3 + b3 + c 3 + a bc)2 ≥ 2(a2 + b2 )(b2 + c 2 )(c 2 + a2 ).
A
PI (Aleksandar Bulj, 2011)
and
O
Since
3p2 − 8q = 3(p2 − 3q) + q ≥ 0,
for fixed p and q, f6 is an increasing function of r. Therefore, it suffices to prove the
inequality f6 (a, b, c) ≥ 0 for the case when r is minimal; that is, when a = 0 or b = c
(see P 3.57). In this cases, the original inequality becomes respectively
and
a(a − b)2 (a3 + 2a2 b + a b2 + 4b3 ) ≥ 0.
The equality holds for a = b = c, and for a = 0 and b = c (or any cyclic permutation).
Symmetric Polynomial Inequalities in Nonnegative Variables 227
Second Solution. Without loss of generality, assume that a = min{a, b, c}. From
a(a − b)(a − c) ≥ 0,
we get
a3 + a bc ≥ a2 (b + c),
and hence
a3 + b3 + c 3 + a bc ≥ a2 (b + c) + b3 + c 3 = (b + c)(a2 + b2 + c 2 − bc).
L
1
2(a2 + b2 )(c 2 + a2 ) ≤ (2a2 + b2 + c 2 )2 .
.M
2
Therefore, it suffices to prove that
D
2(b + c)2 (a2 + b2 + c 2 − bc)2 ≥ (b2 + c 2 )(2a2 + b2 + c 2 )2 .
A
PI
We can obtain this inequality by multiplying the obvious inequality
and
(b + c)2 (a2 + b2 + c 2 − bc) ≥ (b2 + c 2 )(2a2 + b2 + c 2 ).
O
(b − c)2 (bc − a2 ) ≥ 0,
Remark. Using the first method, we can prove the following stronger inequality (Vasile
W
Cîrtoaje, 2011).
• If a, b, c are nonnegative real numbers then
W
Since
(a − b)2 (b − c)2 (c − a)2 = −27r 2 + 2(9pq − 2p3 )r + p2 q2 − 4q3 ,
we can write this inequality as f6 (a, b, c) ≥ 0, where
where
g(r) = 9r 2 + 2(p3 − 4pq)r,
it suffices to show that g(r) is increasing. Indeed, for r1 ≥ r2 ≥ 0, by the third degree
Schur’s inequality, we have
Therefore, it suffices to prove the inequality f6 (a, b, c) ≥ 0 for the case when r is min-
imal; that is, when a = 0 or b = c (see P 3.57). In this cases, the original inequality
L
becomes respectively
.M
(b − c)4 (b2 + 4bc + c 2 ) ≥ 0
and
D
a(a − b)2 (a3 + 2a2 b + a b2 + 4b3 ) ≥ 0.
A
PI
M
P 3.62. If a, b, c are positive real numbers, then
LY
Solution. Setting
p = a + b + c, q = a b + bc + ca, r = a bc,
W
or
3r(q − p) + pq − 6q + 9 ≥ 0.
First Solution. For fixed p and q, the linear function f (r) = 3r(q − p) + pq − 6q + 9 is
minimal when r is either minimal or maximal. Thus, according to P 3.57-(a), we need
only to prove that f (r) ≥ 0 for a = 0 and for b = c.
For a = 0, we need to show that
(b + c)bc − 6bc + 9 ≥ 0.
Symmetric Polynomial Inequalities in Nonnegative Variables 229
p
Indeed, putting x = bc, we have
that is,
Aa2 + Ba + C ≥ 0,
where
L
A = b(6b2 − 3b + 2), B = b(3b3 − 6b2 + 5b − 12), C = 2b3 − 6b2 + 9.
.M
Consider two cases.
D
Case 1: b ≥ 12/5. Since A > 0, B = 3b2 (b − 2) + b(5b − 12) > 0, C > 0, we have
A
Aa2 + Ba + C > 0.
Case 2: 0 < b < 12/5. Since PI
p
M
Aa2 + Ba + C ≥ (Aa2 + C) + Ba ≥ a(2 AC + B),
LY
b(b − 1)4 (8 + 4b − b3 ) ≥ 0.
This inequality is true since
W
since
q2
q − 3r ≥ q − ≥ 0.
p
Consider further that p < 6. From p2 ≥ 3q ≥ 3p, we get p ≥ 3, and from Schur’s
inequality
p3 + 9r ≥ 4pq,
we get
p3 + p2 r ≥ 4pq,
p2 + pr ≥ 4q.
Using this result, we have
L
= 12q2 − 2p(p + 9)q + 3p(p2 + 3)
.M
2
p2 + 9p p(12 − p)(p − 3)2
= 12 q − + ≥ 0.
12 12
D
A
PI
P 3.63. Let a, b, c be nonnegative real numbers such that a b + bc + ca = 3. Prove that
M
LY
Solution. Since
.M
Since
b3 + c 3 = (b + c)3 − 3bc(b + c) = (b + c)3 − 9(b + c)
and p p
b+c ≥2 bc = 2 3,
Symmetric Polynomial Inequalities in Nonnegative Variables 231
we have
L
(b − 1)2 (2b − 3)2 (2b2 + 10b + 3) ≥ 0.
.M
Since the last inequality is true, the proof is completed.
The equality holds for a = b = c, and also for a = 1/4 and b = c = 3/2 (or any cyclic
D
permutation).
A
PI
M
P 3.64. If a, b, c ∈ [0, 1], then
LY
p p p
a a + b b + c c + 4a bc ≥ 2(a b + bc + ca).
O
x 3 + y 3 + z 3 + 4x 2 y 2 z 2 ≥ 2(x 2 y 2 + y 2 z 2 + z 2 x 2 ),
W
y 3 + z 3 − 2 y 2 z 2 ≥ y 4 + z 4 − 2 y 2 z 2 = ( y 2 − z 2 )2 ≥ 0.
For y = 0 and y = 1, we have f (x) = x 3 ≥ 0. Consider further that y ∈ (0, 1). From
8 y 2 (1 − y 2 ) 2
x1 = , x 1 ∈ 0, .
3 3
Therefore, it remains to show that f (x 1 ) ≥ 0, which is equivalent to
L
Since
.M
1
y 2 (1 − y 2 ) ≤ ,
4
it suffices to show that
D
32 y(1 − y)(1 + y)2 ≤ 27.
A
Using the AM-GM inequality, we get
y
1+ y 2
PI
2
32 y(1 − y)(1 + y) = 1024 · (1 − y)
M
2 4
1+ y 4
LY
y
+ (1 − y) + 2 ·
2 4 81
≤ 1024 = < 27.
O
4 4
.M
The equality holds for a = 0 and b = c = 1 (or any cyclic permutation), and also for
a = b = c = 0.
W
W
p p p 3
a a + b b + c c ≥ (a b + bc + ca − a bc).
2
(Vasile Cîrtoaje, 2012)
2(x 3 + y 3 + z 3 ≥ 3(x 2 y 2 + y 2 z 2 + z 2 x 2 − x 2 y 2 z 2 ),
2( y 3 + z 3 ) − 3 y 2 z 2 ≥ 2( y 4 + z 4 ) − 4 y 2 z 2 = 2( y 2 − z 2 )2 ≥ 0.
L
f 0 (x) = 6x[x − y 2 (2 − y 2 )],
.M
it follows that f (x)is decreasing on [0, x 1 ] and increasing on [x 1 , 1], where
D
x 1 = y 2 (2 − y 2 ), x 1 ∈ (0, 1].
A
PI
Therefore, we only need to show that f (x 1 ) ≥ 0, which is equivalent to
y 3 (2 − y 2 )3 ≤ 4 − 3 y.
M
LY
Indeed,
y 3 (2 − y 2 )3 − (4 − 3 y) ≤ y(2 − y 2 )2 − (4 − 3 y) = ( y − 1)2 ( y 3 + 2 y 2 − y − 4) ≤ 0.
O
p p p 500
3(a a + b b + c c) + a bc ≥ 5(a b + bc + ca).
W
81
(Vasile Cîrtoaje, 2012)
500 2 2 2
3(x 3 + y 3 + z 3 ) + x y z ≥ 5(x 2 y 2 + y 2 z 2 + z 2 x 2 ),
81
where x, y, z ∈ [0, 1]. In addition, using the substitutions p = x + y + z and q =
x y + yz + z x, we can rewrite this inequality as
500 2 2 2
x y z + 3(3 + 5p)x yz + 3p3 − 9pq − 5q2 ≥ 0.
81
234 Vasile Cîrtoaje
3( y 3 + z 3 ) − 5 y 2 z 2 ≥ 3( y 4 + z 4 ) − 6 y 2 z 2 = 3( y 2 − z 2 )2 ≥ 0.
50 2
3 2 2
f (x) = 3x − 10x y 1 − y + y 3 (6 − 5 y), y ∈ [0, 1].
81
L
.M
50 2
0 2
f (x) = x 9x − 20 y 1 − y ,
81
D
it follows that f (x)is decreasing on [0, x 1 ] and increasing on [x 1 , 1], where
A
20 2 50 2
x1 =
9
y 1−
81
PI
y , x 1 ∈ (0, 1).
M
Therefore, it remains to show that f (x 1 ) ≥ 0, which is equivalent to
LY
4000 3 50 2 3
y 1− y ≤ 6 − 5 y.
243 81
O
Substituting
9t 10
.M
y= , 0<t ≤ ,
10 9
this inequality can be written as
W
t 3 (2 − t 2 )3 ≤ 4 − 3t,
W
Indeed,
W
a + b + c ≥ 2 + a bc.
3 = 2b2 + c 2 ≥ 3b2 ,
L
it follows that 2b2 ≥ 1. Therefore, we need to prove that 2b2 + c 2 = 3, b ≤ 1, 2b2 ≥ 1
.M
involve 2b + c ≥ 2 + b2 c. Since
2b + c − 2 − b2 c = (1 − b)(c + bc − 2),
D
A
it suffices to show that c(1 + b) ≥ 2. This is true, since
PI
c 2 (1 + b)2 − 4 = (3 − 2b2 )(1 + b)2 − 4
M
= −1 + 6b + b2 − 4b3 − 2b4 = (1 − b)(−1 + 5b + 6b2 + 2b3 ) ≥ 0.
LY
3 −2c 3 + 7c − 4
a + b + c − 2 − a bc = 2c − 2 − c(c 2 − ) = ,
.M
2 2
we need to show that
2c 3 − 7c + 4 ≤ 0.
W
From
W
(c 2 − 2)(2c 2 − 3) ≤ 0,
W
c(2c 3 − 7c + 4) ≤ −6 + 4c < 0.
E(a, b, c) = a + b + c − 2 − a bc
236 Vasile Cîrtoaje
and v
t b2 + c 2
t= , t ≤ 1 ≤ a.
2
We will show that
E(a, b, c) ≥ E(a, t, t) ≥ 0.
We have
L
= −
2 a2 + 2t 2 2t + b + c
.M
(b − c)2 [2t(3a − 2t) + a(3b + 3c − 2a)]
= ≥ 0,
2(a2 + 2t 2 )(2t + b + c)
D
because 3a − 2t > 2(a − t) ≥ 0 and 3(b + c) − 2a > 2(b + c − a) ≥ 0. On the other hand,
A
from
PI
E(a, t, t) = a + 2t − 2 − at 2 = (1 − t)(a + at − 2),
M
it follows that E(a, t, t) ≥ 0 if at ≥ 2 − a; that is,
LY
v
t 3 − a2
a ≥ 2 − a.
2
O
(a − 1)(8 − a2 − a3 ) ≥ 0.
W
p p
This is true, since 1 ≤ a ≤ 2 implies a − 1 ≥ 0 and 8 − a2 − a3 ≥ 8 − 2 − 2 2 > 0.
W
W
where An (p, q) and Bn (p, q) are polynomial functions. For fixed p and q, the linear
function g n (r) = An (p, q)r + Bn (p, q) is minimal when r is either minimal or maximal.
(a) By P 3.57-(a), for fixed p and q, the product r is minimal and maximal when
two of a, b, c are equal or one of a, b, c is 0. Due to symmetry and homogeneity, the
conclusion follows.
L
(b) By P 3.57-(b), for fixed p and q, the product r is minimal and maximal when
.M
two of a, b, c are equal or one of a,b,c is the sum of the others. Due to symmetry and
homogeneity, the conclusion follows.
D
Remark. Similarly, we can prove the following statement, which does not involve the
A
homogeneity property.
PI
• Let f n (a, b, c) be a symmetric polynomial function of degree n ≤ 5. The inequality
f n (a, b, c) ≥ 0 holds for all nonnegative real numbers a, b, c if and only if it holds for a = 0
M
and for b = c.
LY
O
4(a4 + b4 + c 4 ) + 45 ≥ 19(a2 + b2 + c 2 ).
W
First Solution. We use the mixing variable method. Write the inequality as F (a, b, c) ≥
W
0, where
W
Since
1
b2 + c 2 − 2x 2 = 2(x 2 − bc) = (b − c)2 ,
2
we get
1
F (a, b, c) − F (a, x, x) =(b − c)2 [4(b2 + c 2 + 2x 2 ) − 19 + 4(x 2 + bc)]
2
1
= (b − c)2 [4(x 2 − bc) + 24x 2 − 19] ≥ 0.
2
Also,
F (a, x, x) = F (3 − 2x, x, x) = 6(x − 1)2 (3 − 2x)(11 − 6x) ≥ 0.
L
This completes the proof. The equality holds for a = b = c = 1, and for a = 0 and
.M
b = c = 3/2 (or any cyclic permutation).
Second Solution. Write the inequality in the homogeneous form f4 (a, b, c) ≥ 0, where
D
f4 (a, b, c) = 36(a4 + b4 + c 4 ) + 5(a + b + c)4 − 19(a2 + b2 + c 2 )(a + b + c)2 .
A
According to P 3.68-(a), it suffices to prove that f4 (a, 1, 1) ≥ 0 and f4 (0, b, c) ≥ 0 for all
PI
a, b, c ≥ 0. We have
M
f4 (a, 1, 1) = 2a(11a3 − 18a2 + 3a + 4) = 2a(a − 1)2 (11a + 4) ≥ 0,
LY
Remark. Similarly, we can prove the following more general statement (Vasile Cîrtoaje
and Le Huu Dien Khue, 2008).
.M
1 + α + β = 2γ.
W
The inequality
W
X X X X
a4 + α a2 b2 + β a bc a≥γ a b(a2 + b2 )
W
α ≥ (γ − 1) max{2, γ + 1}.
Solution. Let us denote by f5 (a, b, c) the left hand side. By P 3.68-(a), it suffices to
show that f5 (a, 1, 1) ≥ 0 and f5 (0, b, c) ≥ 0 for all a, b, c ≥ 0. Indeed, we have
and
f5 (0, b, c) = (b + c)(b − c)2 (b2 − k bc + c 2 ) ≥ 0.
The equality holds for a = b = c, for a = 0 and b = c (or any cyclic permutation), and
for a/k = b = c, k > 0 (or any cyclic permutation).
L
.M
P 3.71. Let a, b, c be nonnegative real numbers. If k ∈ R, then
D
X
(b + c)(a − b)(a − c)(a − k b)(a − kc) ≥ 0.
A
(Vasile Cîrtoaje, 2008)
PI
Solution. Let us denote by f5 (a, b, c) the left hand side. By P 3.68-(a), it suffices to
M
show that f5 (a, 1, 1) ≥ 0 and f5 (0, b, c) ≥ 0 for all a, b, c ≥ 0. Indeed, we have
LY
and
f5 (0, b, c) = k2 (b + c)b2 c 2 + bc(b + c)(b − c)2 ≥ 0.
.M
The equality holds for a = b = c, for b = c = 0 (or any cyclic permutation), and for
a/k = b = c, k > 0 (or any cyclic permutation).
W
X
(ma + b + c)(a − b)(a − c)(a − k b)(a − kc) ≥ 0.
Solution. Let f5 (a, b, c) = a(a −2b)(a −2c)(a −5b)(a −5c). By P 3.68-(a), it suffices
P
since
20a3 − 60a2 + 44a + 8 > 20a2 (a − 3) ≥ 0
for a ≥ 3, and
L
.M
for a ≤ 8. Also,
f5 (0, b, c) = (b + c)(b2 − 4bc + c 2 )2 ≥ 0.
D
The equality holds for a = 0 and b2 − 4bc + c 2 = 0 (or any cyclic permutation).
A
PI
M
P 3.73. If a, b, c are the lengths of the side of a triangle, then
LY
a4 + b4 + c 4 + 9a bc(a + b + c) ≤ 10(a2 b2 + b2 c 2 + c 2 a2 ).
O
and
f4 ( y + z, y, z) = 8( y 2 + z 2 )2 − 2 yz( y 2 + z 2 ) − 28 y 2 z 2
= 2( y − z)2 (4 y 2 + 4z 2 + 7 yz) ≥ 0,
the proof is completed. The equality holds for an equilateral triangle and for a degen-
erate triangle with a/2 = b = c (or any cyclic permutation).
Second Solution. We use the sum-of-squares method (SOS method). Write the inequal-
ity as follows X X X X
9( b2 c 2 − a bc a) − ( a4 − b2 c 2 ) ≥ 0,
Symmetric Polynomial Inequalities in Nonnegative Variables 241
X X
9 a2 (b − c)2 − (b2 − c 2 )2 ≥ 0,
X
(b − c)2 (3a − b − c)(3a + b + c) ≥ 0.
Without loss of generality, assume that a ≥ b ≥ c. Since
L
Since
.M
3b − c − a ≥ 2b − a ≥ b + c − a ≥ 0
and (c − a)2 ≥ (a − b)2 ), it is enough to prove that
D
A
(3b − c − a)(3b + c + a) + (3c − a − b)(3c + a + b) ≥ 0.
We have
PI
(3b + c + a) − (3c + a + b) = 2(b − c) ≥ 0,
M
and hence
LY
= 2(b + c − a)(3c + a + b) ≥ 0.
W
W
X
3(a4 + b4 + c 4 ) + 7a bc(a + b + c) ≤ 5 a b(a2 + b2 ).
Solution. Let
X
f4 (a, b, c) = 5 a b(a2 + b2 ) − 3(a4 + b4 + c 4 ) − 7a bc(a + b + c).
and
f4 ( y + z, y, z) = 4( y 2 + z 2 )2 + 4 yz( y 2 + z 2 ) − 24 y 2 z 2
= 4( y − z)2 ( y 2 + z 2 + 3 yz) ≥ 0,
the proof is completed. The equality holds for an equilateral triangle and for a degen-
erate triangle with a/2 = b = c (or any cyclic permutation).
L
P 3.75. If a, b, c are the lengths of the sides of a triangle, then
.M
b2 + c 2 − 6bc c 2 + a2 − 6ca a2 + b2 − 6a b
+ + + 4(a + b + c) ≤ 0.
a b c
D
(Vasile Cîrtoaje, 2005)
A
PI
First Solution. Write the inequality as f4 (a, b, c) ≥ 0, where
M
X
f4 (a, b, c) = bc(6bc − b2 − c 2 ) − 4a bc(a + b + c).
LY
and
W
f4 ( y + z, y, z) = 4( y 2 + z 2 )2 − 2 yz( y 2 + z 2 ) − 12 y 2 z 2
= 2( y − z)2 (2 y 2 + 3 yz + 2z 2 ) ≥ 0,
W
W
the proof is completed. The equality holds for an equilateral triangle and for a degen-
erate triangle with a/2 = b = c (or any cyclic permutation).
Second Solution. We use the SOS method. Write the inequality as follows:
X X
bc(b2 + c 2 − 6bc) + 4a bc a ≤ 0,
X X X
bc(b2 + c 2 − 2bc) − 4( b2 c 2 − a bc a) ≤ 0,
X X
bc(b − c)2 − 2 a2 (b − c)2 ≤ 0,
X
(b − c)2 (2a2 − bc) ≥ 0.
Symmetric Polynomial Inequalities in Nonnegative Variables 243
Since
2b2 − ca ≥ 2b2 − c(b + c) = (b − c)(2b + c) ≥ 0
and (c − a)2 ≥ (a − b)2 , it is enough to show that
L
Indeed,
.M
(2b2 − ca) + (2c 2 − a b) = (b − c)2 + (b + c)(b + c − a) ≥ 0.
D
A
P 3.76. Let f6 (a, b, c) be a sixth degree symmetric homogeneous polynomial written in the
PI
form
f6 (a, b, c) = Ar 2 + B(p, q)r + C(p, q), A ≤ 0,
M
where
LY
p = a + b + c, q = a b + bc + ca, r = a bc.
O
Prove that
(a) the inequality f6 (a, b, c) ≥ 0 holds for all nonnegative real numbers a, b, c if and
.M
(a2) For A ≤ 0, the inequality f6 (a, b, c) ≥ 0 holds for all nonnegative real numbers
a, b, c satisfying p2 > 4q if and only if f6 (a, 1, 1) ≥ 0 for all a > 4 and f6 (0, b, c) ≥ 0 for
all b, c ≥ 0.
Notice that the restriction 0 ≤ a ≤ 4 in (a1) follows by setting b = c = 1 in p2 ≤ 4q.
In addition, since a = 0 and p2 ≤ 4q involve b = c, the condition f6 (0, b, c) ≥ 0 is
not necessary in (a1) because it is equivalent to f6 (0, 1, 1) ≥ 0, which follows from
f6 (a, 1, 1) ≥ 0 for all 0 ≤ a ≤ 4. Also, the restriction a > 4 in (a2) follows by setting
b = c = 1 in p2 > 4q.
Remark 2. The statement in P 3.76 and its extension in Remark 1 are also valid in the
more general case when f6 (a, b, c) is a symmetric homogeneous function of the form
L
.M
f6 (a, b, c) = Ar 2 + B(p, q)r + C(p, q),
D
A
PI
P 3.77. If a, b, c are nonnegative real numbers, then
M
X
a(b + c)(a − b)(a − c)(a − 2b)(a − 2c) ≥ (a − b)2 (b − c)2 (c − a)2 .
LY
where X
f (a, b, c) = a(b + c)(a − b)(a − c)(a − 2b)(a − 2c).
W
Since
W
X
a(b + c)(a − b)(a − c)(a − 2b)(a − 2c) =
X
= a(p − a)(a2 + 2bc − q)(a2 + 6bc − 2q),
f (a, b, c) has the same highest coefficient A0 as
X
P1 (a, b, c) = − a2 (a2 + 2bc)(a2 + 6bc);
that is, according to Remark 2 from P 2.75, A0 = P1 (1, 1, 1) = −3(1 + 2)(1 + 6) = −63.
Then, f6 (a, b, c) has the highest coefficient
A = A0 + 27 = −36.
Symmetric Polynomial Inequalities in Nonnegative Variables 245
and
f6 (0, b, c) = bc(b − c)4 ≥ 0.
The equality holds for a = b = c, for a = 0 and b = c (or any cyclic permutation), and
for a/2 = b = c (or any cyclic permutation).
L
.M
P 3.78. Let a, b, c be nonnegative real numbers.
D
(a) If 2 ≤ k ≤ 6, then
A
X 4(k − 2)(a − b)2 (b − c)2 (c − a)2
PI
a(a − b)(a − c)(a − k b)(a − kc) + ≥ 0;
a+b+c
M
(b) If k ≥ 6, then
LY
X
f6 (a, b, c) = (a + b + c) a(a − b)(a − c)(a − k b)(a − kc)
W
and (a − b)2 (b − c)2 (c − a)2 has the highest coefficient −27, it follows that f6 (a, b, c)
has the highest coefficient
A = −108(k − 2).
Since A ≤ 0, according to P 3.76-(a), it suffices to prove that f6 (a, 1, 1) ≥ 0 and
f6 (0, b, c) ≥ 0 for all a, b, c ≥ 0. Indeed, we have
and
f6 (0, b, c) = (b − c)6 + (6 − k)bc(b − c)4 ≥ 0.
The equality holds for a = b = c, for a = 0 and b = c (or any cyclic permutation, and
for and for a/k = b = c (or any cyclic permutation).
(b) We need to prove that f6 (a, b, c) ≥ 0, where
X
f6 (a, b, c) = 4(a + b + c) a(a − b)(a − c)(a − k b)(a − kc)
L
.M
(k + 2)2 (a − b)2 (b − c)2 (c − a)2
and (a − b)2 (b − c)2 (c − a)2 has the highest coefficient −27, it follows that f6 (a, b, c)
D
has the highest coefficient
A
A = −27(k + 2)2 < 0.
PI
According to P 3.76-(a), it suffices to prove that f6 (a, 1, 1) ≥ 0 and f6 (0, b, c) ≥ 0 for all
a, b, c ≥ 0. Indeed, we have
M
and
O
The equality holds for a = b = c, for a = 0 and b = c (or any cyclic permutation, and
b c k−2
for and for a/k = b = c (or any cyclic permutation), and for a = 0 and + =
c b 2
W
(3a2 + 2a b + 3b2 )(3b2 + 2bc + 3c 2 )(3c 2 + 2ca + 3a2 ) ≥ 8(a2 + 3bc)(b2 + 3ca)(c 2 + 3a b).
Since
P2 (a, b, c) = (2a b − 3c 2 )(2bc − 3a2 )(2ca − 3b2 ) − 8(a2 + 3bc)(b2 + 3ca)(c 2 + 3a b);
Then, by P 3.76-(a), it suffices to prove that f6 (a, 1, 1) ≥ 0 and f6 (0, b, c) ≥ 0 for all
L
nonnegative real a, b, c. Indeed,
.M
f6 (a, 1, 1) = 8(3a2 + 2a + 3)2 − 8(a2 + 3)(3a + 1)2 = 48(a + 1)(a − 1)2 ≥ 0,
D
A
f6 (0, b, c) = 3b2 c 2 (9b2 − 2bc + 9c 2 ) ≥ 0.
The equality holds for a = b = c. PI
M
LY
−2 11
≤k≤ ,
3 8
.M
then
(a2 + ka b + b2 )(b2 + k bc + c 2 )(c 2 + kca + a2 ) ≤ k + 2.
W
Since f6 (a, b, c) has the same highest coefficient A as P2 (a, b, c), where
Also,
and
11
Case 1: 1 ≤ k ≤ . Since A ≤ 0, according to P 3.76-(a), it suffices to prove that
8
f6 (a, 1, 1) ≥ 0 and f6 (0, b, c) ≥ 0 for all a, b, c ≥ 0. Clearly, these conditions are satisfied.
The equality holds for a = 0 and b = c = 1 (or any cyclic permutation). If k = 11/8,
then the equality holds also for a = b = c = 2/3.
−2
Case 2: ≤ k < 1. Since A > 0, we will use the highest coefficient cancellation method.
L
3
We will prove the sharper inequality g6 (a, b, c) ≥ 0, where
.M
g6 (a, b, c) = f6 (a, b, c) − 64(1 − k)3 a2 b2 c 2 .
D
Since g6 (a, b, c) has the highest coefficient A1 = 0, it suffices to show that g6 (a, 1, 1) ≥ 0
A
and g6 (0, b, c) ≥ 0 for all a, b, c ≥ 0 (see P 3.76). The inequality g6 (a, 1, 1) ≥ 0 is true
if PI
(k + 2)[(a − 1)2 + 11 − 8k][a3 + 14a2 + (8k + 12)a + 16] ≥ 64(1 − k)3 a.
M
It suffices to show that
LY
Since
W
3a2 + (4k + 6)a + 8 ≥ 3(2a − 1) + (4k + 6)a + 8 = (4k + 12)a + 5 > 4(k + 3)a,
W
Indeed,
2 1
2
(k + 2)(k + 3) − (1 − k) = 7k + 5 = 7 k + + > 0.
3 3
The inequality g6 (0, b, c) ≥ 0 is also true since
Thus, the proof is completed. The equality holds for a = 0 and b = c = 1 (or any cyclic
permutation). If k = 11/8, then the equality holds also for a = b = c = 2/3.
Symmetric Polynomial Inequalities in Nonnegative Variables 249
Since f6 (a, b, c) has the same highest coefficient A as P2 (a, b, c), where
L
.M
according to Remark 2 from P 2.75, we have
D
A = P2 (1, 1, 1) = −432.
A
Since A < 0, according to P 3.76-(a), it suffices to prove that f6 (a, 1, 1) ≥ 0 and
f6 (0, b, c) ≥ 0 for all a, b, c ≥ 0. We have PI
M
f6 (a, 1, 1) = a(a + 2)2 (a3 + 8a2 − 8a + 32) = a(a + 2)2 [a3 + 4a2 + 28 + 4(a − 1)2 ] ≥ 0,
LY
P 3.82. Let a, b, c be nonnegative real numbers, no two of which are zero. Then,
W
X
(a − b)(a − c)(a − 2b)(a − 2c) ≥ .
a b + bc + ca
W
Solution. Denote
p = a + b + c, q = a b + bc + ca,
and write the inequality as f6 (a, b, c) ≥ 0, where
X
f6 (a, b, c) = q (a − b)(a − c)(a − 2b)(a − 2c) − 5(a − b)2 (b − c)2 (c − a)2 .
A = (−5)(−27) = 135.
250 Vasile Cîrtoaje
Since A > 0, we will use the highest coefficient cancellation method. We have
L
such that P(1, 1, 1) = 0 and P(2, 1, 1) = 0. We get B = 1/18 and C = −5/18, hence
.M
1 5
P(a, b, c) = a bc + (a + b + c)3 − (a + b + c)(a b + bc + ca).
18 18
D
Consider now the sharper inequality g6 (a, b, c) ≥ 0, where
A
PI
g6 (a, b, c) = f6 (a, b, c) − 135P 2 (a, b, c).
M
Clearly, g6 (a, b, c) has the highest coefficient A = 0. Then, according to Remark 1 from
the proof of P 3.76, it suffices to prove that g6 (a, 1, 1) ≥ 0 for 0 ≤ a ≤ 4. We have
LY
1
P(a, 1, 1) = (a − 1)2 (a − 2),
O
18
hence
.M
1
g6 (a, 1, 1) = f6 (a, 1, 1) − 135P 2 (a, 1, 1) = (a − 1)2 (a − 2)2 (7 + 34a − 5a2 ) ≥ 0.
12
W
(a b + bc + ca)2
W
which satisfies R(1, 1, 1) = 0, and consider the sharper inequality g6 (a, b, c) ≥ 0, where
Clearly, g6 (a, b, c) has the highest coefficient A = 0. Then, according to Remark 2 from
the proof of P 3.76, it suffices to prove that g6 (a, 1, 1) ≥ 0 for a > 4, and g6 (0, b, c) ≥ 0
for all b, c ≥ 0. We have
9C + 1
= 6;
3C
that is, C = 1/9. For this value of C, we have
hence
(a − 1)2
g6 (a, 1, 1) = f6 (a, 1, 1) − 135R2 (a, 1, 1) = g(a),
L
3(a + 2)2
.M
where
g(a) = 3(2a + 1)(a2 − 4)2 − 20(a − 1)4 ,
D
and
A
bc(3b2 + bc + 3c 2 )[(b + c)2 − 6bc]2
g6 (0, b, c) = f6 (0, b, c) − 135R2 (0, b, c) = PI 3(b + c)2
≥ 0.
M
To complete the proof, we need to show that g(a) ≥ 0 for a > 4. This is true since
a2 − 4 > (a − 1)2 and 3(2a + 1) > 20.
LY
The equality holds for a = b = c, for a/2 = b = c (or any cyclic permutation), and
for a = 0 and b/c + c/b = 4 (or any cyclic permutation).
O
.M
a + b + c = p, a bc = r,
W
where p and r are fixed positive numbers satisfying p3 ≥ 27r. Prove that
W
q = a b + bc + ca
Solution. First, we show that b ∈ [b1 , b2 ], where b1 and b2 are the positive roots of
the equation
2x 3 − p x 2 + r = 0.
Let f (x) = 2x 3 − p x 2 + r, x ≥ 0. From f 0 (x) = 2x(3x − p), it follows that f (x)
is decreasing for x ≤ p/3 and increasing for x ≥ p/3. Since f (0) > 0, f (p/3) =
252 Vasile Cîrtoaje
(27r − p3 )/27 ≤ 0 and f (p) = p3 + r > 0, there exists two positive numbers b1 ≤ b2
such that f (b1 ) = f (b2 ) = 0 and f (x) ≤ 0 for x ∈ [b1 , b2 ]. From
f (b) r
= b2 − b(p − b) + = b2 − b(a + c) + bc = (b − a)(b − c) ≤ 0,
b b
L
we get
r
.M
q(b) = p b − b2 + .
b
Since
D
r −(b − a)(b − c)
q0 (b) = p − 2b − = ≥ 0,
A
b2 b
q(b) is increasing on [b1 , b2 ], and hence q(b) is maximal for b = b2 , when b = c, and
PI
is minimal for b = b1 , when b = a.
M
Remark 1. Substituting 1/a, 1/b, 1/c for a, b, c, respectively, we get the following
LY
similar statement.
• Let a ≤ b ≤ c be positive real numbers such that
O
a b + bc + ca = q, a bc = r,
.M
where q and r are fixed positive numbers satisfying q3 ≥ 27r 2 . The sum
W
p=a+b+c
W
following generalizations:
• Let a1 , a2 , . . . , an be positive real numbers such that
a1 + a2 + · · · + an = np, a1 a2 · · · an = p1 ,
where p and p1 are fixed positive numbers satisfying p n ≥ p1 . Then, the sum
1 1 1
+ + ··· +
a1 a2 an
1 1 1 n
+ + ··· + = , a1 a2 · · · an = p1 ,
a1 a2 an p
where p and p1 are fixed positive numbers satisfying p n ≤ p1 . Then, the sum
a1 + a2 + · · · + an
L
.M
P 3.84. If a, b, c are positive real numbers, then
D
A
1 1 1 1
(a + b + c − 3) + + − 3 + a bc + ≥ 2.
a b c a bc
PI
(Vasile Cîrtoaje, 2004)
M
1 1 1
LY
a b + bc + ca 1
(p − 3) − 3 + r + ≥ 2.
r r
W
2 1 1 1
a b2 + − 3 + + 2b − 3 ≥ 6 b + − 2 ,
b a b2 b
1 1
(b − 1)2 ab + − 2 + 2 a + − 2 ≥ 0.
ab a
1 1
Since a b + ≥ 2 and a + ≥ 2, the conclusion follows. The equality holds for
ab a
a = b = 1, or b = c = 1, or c = a = 1.
254 Vasile Cîrtoaje
46 p
(b) a b + bc + ca − 3 ≥ ( a + b + c − 2 − 1).
27
(Vasile Cîrtoaje, 2009)
L
.M
vt 2p
3 2
a b + bc + ca − ≥ − 1.
7 3 3
D
For fixed p, the expression q = a b + bc + ca is minimal when two of a, b, c are equal
A
(see P 3.83). Thus, it suffices to consider the case a = b, when the inequality becomes
v PI
t 4a3 − 3a2 + 2
3a3 − 2a + 6 ≥ 7 .
M
3
LY
By squaring, we get
(a − 1)2 (3a − 1)2 (3a2 + 8a + 10) ≥ 0,
O
1 1
which is true. The equality holds for a = b = c = 1, and also for (a, b, c) = , , 9 or
3 3
.M
46 p
a b + bc + ca − 3 ≥ ( p − 2 − 1).
W
27
For fixed p, the expression q = a b + bc + ca is minimal when two of a, b, c are equal
W
(see P 3.83). Thus, it suffices to consider the case a = b, when the inequality becomes
p
27a3 − 35a + 54 ≥ 46 2a3 − 2a2 + 1.
By squaring, we get
50 37
a b + bc + ca + ≥ .
a+ b+c+5 4
50 37
a b + bc + ca + ≥ .
p+5 4
L
For fixed p, the expression q = a b + bc + ca is minimal when two of a, b, c are equal
.M
(see P 3.83). Thus,it suffices to prove the desired inequality for a = b; that is, to show
that a2 c = 1 involves
D
50 37
a2 + 2ac + ≥ .
2a + c + 5 4
A
This is equivalent to
a2 +
2
+ 3
50a2
≥
37
PI
,
a 2a + 5a + 1
M
2 4
which can be written in the obvious form
LY
The equality holds for a = b = c = 1, and for a = b = 1/2 and c = 4 (or any cyclic
.M
permutation).
W
W
(a) If a bc = 2, then
a2 + b2 + c 2
(a + b + c − 3)2 + 1 ≥ ;
3
1
(b) If a bc = , then
2
a2 + b2 + c 2 + 3(3 − a − b − c)2 ≥ 3.
p2 − 2(a b + bc + ca)
(p − 3)2 + 1 ≥ .
3
For fixed p, the expression q = a b + bc + ca is minimal when two of a, b, c are equal
(see P 3.83). Thus, it suffices to consider the case a = b, when the inequality becomes
in succession 2
2 2a2 4
2a + 2 − 3 + 1 ≥ + 4,
a 3 3a
L
5a6 − 18a5 + 15a4 + 12a3 − 18a2 + 4 ≥ 0,
.M
(a − 1)2 (5a4 − 8a3 − 6a2 + 8a + 4) ≥ 0.
D
Since
5a4 − 8a3 − 6a2 + 8a + 4 = 4(a − 1)4 + a(a3 + 8a3 − 30a + 24),
A
PI
it suffices to prove that a3 + 8a3 − 30a + 24 ≥ 0. Indeed, for a ≥ 1, we have
M
a3 + 8a3 − 30a + 24 = (a − 1)3 + 11a2 − 33a + 25
3 1
LY
1
The equality holds for (a, b, c) = 1, 1, or any cyclic permutation.
2
L
bc ca a b
.M
4 + + + 9a bc ≥ 21.
a b c
D
Solution. Let p = a + b + c, q = a b + bc + ca, r = a bc. We write the required inequality
A
in the homogeneous form
4p2 q2
+ 9r ≥
5p3
PI
.
M
9r 3
For fixed p and r, it suffices to prove this inequality for the case when q is minimal; that
LY
is, when two of a, b, c are equal (see P 3.83). Due to symmetry and homogeneity, we
can set b = c = 1. Since p = a + 2, q = 2a + 1, r = a, the inequality becomes
O
which is equivalent to
(a − 1)2 (a − 4)2 ≥ 0.
W
The equality holds for a = b = c = 1, and for a = 2 and b = c = 1/2 (or any cyclic
W
permutation).
W
a b + bc + ca = a bc + 2,
then
a2 + b2 + c 2 + a bc ≥ 4.
First Solution. Among the numbers 1 − a, 1 − b and 1 − c there are always two with
the same sign; let us say (1 − b)(1 − c) ≥ 0. Thus, we have
a(1 − b)(1 − c) ≥ 0,
a + a bc ≥ a b + ac,
a + (a b + bc + ca − 2) ≥ a b + ac,
a + bc ≥ 2,
and hence
L
a2 + b2 + c 2 + a bc − 4 ≥ a2 + 2bc + a bc − 4
.M
= (a + 2)(a + bc − 2) ≥ 0.
p
The equality holds for a = b = c = 1, and for a = 0 and b = c = 2 (or any cyclic
D
permutation).
A
Second Solution For a = 0, we need to show that bc = 2 involves b2 + c 2 ≥ 0. This is
true since PI
b2 + c 2 ≥ 2bc = 4.
M
Consider further that a, b, c are positive, and write the required inequality as
LY
a2 + b2 + c 2 + a bc ≥ 2(a b + bc + ca − a bc),
O
or
3a bc ≥ 2(a b + bc + ca) − a2 − b2 − c 2 .
.M
are equal (see Remark 1 from P 3.83). Thus, it suffices to consider the case b = c, when
p = a + 2b, q = 2a b + b2 , r = a b2 . We need to prove that
W
2a b + b2 = a b2 + 2
W
implies
3a b2 ≥ 4(2a b + b2 ) − (a + 2b)2 ,
which reduces to
a(a + 3b2 − 4b) ≥ 0.
This is true since, for the nontrivial case b < 4/3, we have
2 − b2 (1 − b)2 (2 + 4b − 3b2 )
a + 3b2 − 4b = + 3b2 − 4b = ≥ 0.
b(2 − b) b(2 − b)
Symmetric Polynomial Inequalities in Nonnegative Variables 259
L
.M
According to the identity 2p2 + 2q2 = (p − q)2 + (p + q)2 , we have
2
c+a a+b 2
c + a 2
2 a+b c+a a+b
D
2 −m +2 −m = − + + − 2m .
b c b c b c
A
Thus, we can rewrite the inequality (*) as PI
2 2 2
c+a a+b 2a c+a a+b
M
− ≥4 −m+1 − + − 2m ,
b c b+c b c
LY
b c
This is true if f (a) ≥ 0, where
.M
b+c
Since f (a) is increasing and a = max{a, b, c}, it suffices to show that f ≥ 0.
W
2
Indeed,
W
b+c
Setting = t, this inequality becomes
a
2
2
(t − m)2 + 2 − m + 1 ≥ 6,
t
260 Vasile Cîrtoaje
p 2
(t − 2)2 (t − 2)
≥ 0.
t2
a
The proof is completed. The equality holds for a = b = c, and for p = b = c (or any
2
cyclic permutation).
L
.M
(Vasile Cîrtoaje, 2010)
D
we can write the inequality as
A
6a bc + 2p3 + 3(3 − 2p)q + 39 ≥ 24p.
PI
By Schur’s inequality of degree three, we have
M
9a bc ≥ 4pq − p3 .
LY
2
(4pq − p3 ) + 2p3 + 3(3 − 2p)q + 39 ≥ 24p,
.M
3
which is equivalent to
W
(10p − 27)p2
4p3 + 117 − 72p − (10p − 27)q ≥ 4p3 + 117 − 72p −
3
1
= (p − 3)2 (2p + 39) ≥ 0.
3
9p2 − 48p + 64 ≥ 0,
X
18q ≥ −9 a2 + 48p − 64.
Symmetric Polynomial Inequalities in Nonnegative Variables 261
we get
X X X
2 2 a3 + 9q + 39 − 24p ≥ 4 a3 + −9 a2 + 48p − 64 + 78 − 48p
X 14 14 14 X
X
= 4a3 − 9a2 + < 4a3 − 9a2 + + (10a − 9)
3 3 27
X 140 81 9 2
X X
= a 4a2 − 9a + > a 4a2 − 9a + = a 2a − ≥ 0.
27 16 4
L
The equality holds for a = b = c = 1.
D
.M
P 3.92. If a, b, c are positive real numbers such that a2 + b2 + c 2 = 3, then
A
PI
a3 + b3 + c 3 − 3 ≥ |(a − b)(b − c)(c − a)|.
M
Solution. Assume that a ≤ b ≤ c and write the inequality in the homogeneous form
LY
3/2
a + b2 + c 2
2
3 3 3
a + b +c −3 ≥ |(a − b)(b − c)(c − a)|.
O
3
The left hand side is nonnegative because, by the Cauchy-Schwarz inequality, we have
.M
Thus, it suffices to consider that a < b < c. Using the substitution b = a + p and
c = a + q, where 0 < p < q, we need to show that
W
where
3/2
a2 + (a + p)2 + (a + q)2
3 3 3
f (a) = a + (a + p) + (a + q) − 3 .
3
From
v
t a2 + b2 + c 2
f 0 (a) = 3(a2 + b2 + c 2 ) − 3(a + b + c)
3
v v
t a2 + b2 + c 2 t a2 + b2 + c 2 a + b + c
=9 − ≥ 0,
3 3 3
262 Vasile Cîrtoaje
it follows that f is increasing, hence f (a) ≥ f (0). Therefore, it suffices to show that
f (0) ≥ pq(q − p); that is,
3/2
p2 + q2
3 3
p +q −3 ≥ pq(q − p).
3
Due to homogeneity, we may assume that p = 1 and q > 1, when the inequality becomes
as follows: 3/2
1 + q2
3 2
q −q +q+1≥3 ,
3
L
3(q3 − q2 + q + 1)2 ≥ (q2 + 1)3 ,
.M
q6 − 3q5 + 3q4 − 3q2 + 3q + 1 ≥ 0,
q3 (q − 1)3 + q3 − 3q2 + 3q + 1 ≥ 0,
D
(q3 + 1)(q − 1)3 + 2 ≥ 0.
A
The last inequality is clearly true. The equality holds for a = b = c = 1.
PI
M
LY
a4 + b4 + c 4 − a2 b2 − b2 c 2 − c 2 a2 ≥ 2|a3 b + b3 c + c 3 a − a b3 − bc 3 − ca3 |.
.M
4Aa2 + 4Ba + C ≥ 0,
where
A = p2 − pq + q2 , B = p3 + q(p − q)2 ,
C = p4 + 2p3 q − p2 q2 − 2pq3 + q4 = (p2 + pq − q2 )2 .
Since A ≥ 0, B ≥ 0 and C ≥ 0, the inequality
p is obviously true. The equality occurs for
c 1+ 5
a = b = c, and also for a = 0 and = (or any cyclic permutation).
b 2
Symmetric Polynomial Inequalities in Nonnegative Variables 263
Aa2 + Ba + C ≥ 0,
L
.M
where
p p
A = 5(p2 − pq + q2 ), B = 4p3 + (6 2 − 1)p2 q − (6 2 + 1)pq2 + 4q3 ,
D
p
C = p4 + q4 + 2 2pq(p2 − q2 ).
A
Since
A ≥ 0,
PI
M
2
25 2 5p
2 3
B≥ p q − 10pq + 4q = q − 2q ≥ 0
LY
4 2
and p
O
C = (p2 + 2pq − q2 )2 ≥ 0,
p
c 2+ 6
= (or any cyclic permutation).
b 2
W
W
1−a 1− b 1−c
+ + ≥ 0.
1 + a + a2 1 + b + b2 1 + c + c 2
(Vasile Cîrtoaje, 2014)
a = x − 5, b = y − 5, c = z − 5,
Denoting
x + y +z
p= ,
18
we can write this inequality as f5 (x, y, z) ≥ 0, where
X
f5 (x, y, z) = (6p − x)( y 2 − 9 y p + 21p2 )(z 2 − 9z p + 21p2 )
L
. Since
2
.M
1− b 1−c 2(a − 1)
= = 2 ,
1+ b+ b 2 1+c+c 2 a − 8a + 19
D
we need to show that
A
1−a 4(a − 1)
+ 2 ≥ 0,
1+a+a 2 a − 8a + 19 PI
which is equivalent to
M
(a − 1)2 (a + 5) ≥ 0.
LY
1 1− b 1 1−c
+ + + ≥ 0,
7 1 + b + b2 7 1 + c + c2
.M
(b − 4)(b − 2) (c − 4)(c − 2)
+ ≥ 0,
1 + b + b2 1 + c + c2
W
2(1 + b + b2 )(1 + c + c 2 )
The last inequality is true since
2 2
b+c b−c
2(b + c) − bc = − bc = ≥ 0.
2 2
The proof is completed. The equality occurs for a = b = c = 1, and also for a = −5 and
b = c = 4 (or any cyclic permutation).
Second Solution. Assume that a ≤ b ≤ c and denote
b+c 1−a 1− b 1−c
t= , E(a, b, c) = + + .
2 1+a+a 2 1+ b+ b 2 1 + c + c2
Symmetric Polynomial Inequalities in Nonnegative Variables 265
From
a+b+c 3−a
t≥ = 1, t= ≤ 4,
3 2
it follows that
t ∈ [1, 4].
L
1− b 1− t 1−c 1− t
− + − ≥ 0,
1 + b + b2 1 + t + t 2 1 + c + c2 1 + t + t 2
.M
(b − 1)t − b − 2 (c − 1)t − c − 2
(b − c)
D
− ≥ 0,
1 + b + b2 1 + c + c2
A
(b − c)2 [(2 + b + c − bc)t + 1 + 2(b + c) + bc] ≥ 0,
PI
(b − c)2 [2t 2 + 6t + 1 − bc(t − 1)] ≥ 0.
M
The last inequality is true since
LY
2t 2 + 6t + 1 − bc(t − 1) ≥ 2t 2 + 6t + 1 − t 2 (t − 1)
O
Also, we have
W
6(1 − t)2 (4 − t)
=
W
≥ 0.
(4t 2 − 14t + 13)(1 + t + t 2 )
1 4
P 3.96. Let a, b, c 6= be nonnegative real numbers such that a + b + c = 3. If k ≥ ,
k 3
then
1−a 1− b 1−c
+ + ≥ 0.
(1 − ka)2 (1 − k b)2 (1 − kc)2
(Vasile Cîrtoaje, 2012)
266 Vasile Cîrtoaje
L
≥ 0,
(1 − 3k + 2k b)2 (1 − k b)2
.M
k(b − 1)2 [k(3 − b) − 2] ≥ 0.
The last inequality holds since
D
4 2(3 − 2b) 2a
A
k(3 − b) − 2 ≥ (3 − b) − 2 = = ≥ 0.
3 3 3
PI
Case 2: a = 0. Since b + c = 3, the original inequality becomes as follows:
M
b−1 c−1
1≥ + ,
(1 − k b)2 (1 − kc)2
LY
which is equivalent to
The proof is completed. The equality occurs for a = b = c = 1. If k = 4/3, then the
equality holds also for a = 0 and b = c = 3/2 (or any cyclic permutation).
Symmetric Polynomial Inequalities in Nonnegative Variables 267
a2 + b2 + c 2 + d 2 = 1.
Prove that
(1 − a)(1 − b)(1 − c)(1 − d) ≥ a bcd.
(Vasile Cîrtoaje, 2001)
(1 − a)(1 − b) ≥ cd,
L
.M
(1 − c)(1 − d) ≥ a b.
With regard to the first inequality, we have
D
2cd ≤ c 2 + d 2 = 1 − a2 − b2 ,
A
and hence PI
M
2(1 − a)(1 − b) − 2cd ≥ 2(1 − a)(1 − b) − 1 + a2 + b2
= (1 − a − b)2 ≥ 0.
LY
The second inequality can be proved similarly. The equality holds for a = b = c = d =
O
1 1 1 1 4
2
+ 2 + 2 + 2 = 2.
a b c d x
W
If x ≥ 2, then
(a − 1)(b − 1)(c − 1)(d − 1) ≥ (x − 1)4 .
(Vasile Cîrtoaje, 2001)
ab cd
x + x −2 x + x − 2 ≥ 4(x − 1)2 .
cd ab
With regard to the first inequality, we write it as
ab
2a b − 2(a + b) + x(3 − x) ≥ x(x − 1) .
cd
Since
2 1 1 4 1 1
≤ 2 + 2 = 2 − 2 − 2,
cd c d x a b
it suffices to show that
L
4 1 1
4a b − 4(a + b) + 2x(3 − x) ≥ x(x − 1)a b
.M
− − .
x 2 a2 b2
This is equivalent to
D
A
4a2 b2 − 4a b(a + b)x + 2x 2 (3 − x)a b + x 2 (x − 1)(a2 + b2 ) ≥ 0,
The second inequality can be proved similarly. With regard to the third inequality, we
have
O
ab cd
x + x −2 x + x −2 =
.M
cd ab
2 a b cd
= 2x − 4x + 4 + + x(x − 2)
cd a b
W
(1 + a3 )(1 + a) ≥ (1 + a2 )2 ,
L
.M
which reduces to a(1 − a)2 ≥ 0, while the right side inequality is equivalent to
D
2(1 − a + a2 )2 ≥ 1 + a4 ,
A
which reduces to (1 − a)4 ≥ 0. Multiplying the inequalities
4
PI 4
1 + a3 1 + a4 1 + b3 1 + b4
M
≥ , ≥ ,
1 + a2 2 1 + b2 2
LY
4 4
1 + c3 1 + c4 1 + d3 1 + d4
≥ , ≥ ,
O
1 + c2 2 1 + d2 2
.M
yields
from which the desired inequality follows. The equality holds for a = b = c = d = 1.
1 1 1 1 1 1 1 1
a+ −1 b+ −1 c+ −1 d + −1 +3≥ + + + .
a b c d a b c d
270 Vasile Cîrtoaje
L
1+ a+ −2 ≥ − 3.
a a
.M
This is an identity, and then the proof is completed. The equality holds for a = b = c =
d = 1.
D
A
P 3.101. If a, b, c, d are nonnegative real numbers, then
PI
M
4(a3 + b3 + c 3 + d 3 ) + 15(a bc + bcd + cd a + d a b) ≥ (a + b + c + d)3 .
LY
Solution. Let
O
E(a, b, c, d) ≥ E(0, a + b, c, d) ≥ 0.
W
We have
W
The equality holds for a = 0 and b = c = d (or any cyclic permutation), and also for
a = b = 0 and c = d (or any permutation thereof).
Symmetric Polynomial Inequalities in Nonnegative Variables 271
a + b + c + d = 4.
Prove that
1 + 2(a bc + bcd + cd a + d a b) ≥ 9 min{a, b, c, d}.
(Vasile Cîrtoaje, 2008)
L
(a + b + c + d)3 + 128bcd + 128a(bc + cd + d b) ≥ 36a(a + b + c + d)2 ,
.M
(4a + t)3 + 128bcd + 128a(bc + cd + d b) ≥ 36a(4a + t)2 .
D
Since
A
bcd = (a + x)(a + y)(a + z) ≥ a3 + a2 t
and PI
M
bc + cd + d b = (a + x)(a + y) + (a + y)(a + z) + (a + z)(a + x)
= 3a2 + 2at + x y + yz + z x ≥ 3a2 + 2at,
LY
t(t − 12a)2 ≥ 0.
W
1 1 1 13
The equality holds for a = b = c = d = 1, and also for (a, b, c, d) = , , , or
4 4 4 4
W
a + b + c + d = 4.
Prove that
5(a2 + b2 + c 2 + d 2 ) ≥ a3 + b3 + c 3 + d 3 + 16.
(Vasile Cîrtoaje, 2005)
272 Vasile Cîrtoaje
E(a, b, c, d) ≥ E(a, x, x, x) ≥ 0,
where
E(a, b, c, d) = 5(a2 + b2 + c 2 + d 2 ) − a3 − b3 − c 3 − d 3 − 16.
The left side inequality is equivalent to
5(b2 + c 2 + d 2 − 3x 2 ) − (b3 + c 3 + d 3 − 3x 3 ) ≥ 0.
L
.M
Since b2 + c 2 + d 2 − 3x 2 ≥ 0 and x ≤ 1, it suffices to prove the homogeneous inequality
5x(b2 + c 2 + d 2 − 3x 2 ) − (b3 + c 3 + d 3 − 3x 3 ) ≥ 0,
D
which is equivalent to
A
PI
2(b3 + c 3 + d 3 ) + 3[b(c 2 + d 2 ) + c(d 2 + b2 ) + d(b2 + c 2 )] ≥ 24bcd.
E(a, x, x, x) = 5(a2 + 3x 2 ) − a3 − 3x 3 − 16
.M
This completes the proof. The equality holds for a = b = c = d = 1, and also for
W
5a2 ≥ a3 + 16
and
5(b2 + c 2 + d 2 ) ≥ b3 + c 3 + d 3 .
Symmetric Polynomial Inequalities in Nonnegative Variables 273
We have
and
5(b2 + c 2 + d 2 ) ≥ b2 + c 2 + d 2 ≥ b3 + c 3 + d 3 .
L
or
.M
X
(1 − a) f (a) ≤ 0,
where
D
f (a) = a2 − 4a.
A
Since a + b ≤ 4, we have
PI
f (a) − f (b) = (a − b)(a + b − 4) ≤ 0,
M
and, similarly,
LY
and
f (a) ≤ f (b) ≤ f (c) ≤ f (d),
W
X X X
4 (a − 1) f (a) ≤ (a − 1) f (a) = 0.
W
a1 + a2 + · · · + an = n,
then
(n + 1)(a12 + a22 + · · · + an2 ) ≥ a13 + a23 + · · · + an3 + n2 .
274 Vasile Cîrtoaje
a + b + c + d = 4.
Prove that
3(a2 + b2 + c 2 + d 2 ) + 4a bcd ≥ 16.
(Vasile Cîrtoaje, 2004)
Solution. We use the mixing variable method. Assume that a = min{a, b, c, d}, a ≤ 1.
Setting x = (b + c + d)/3, we have a + 3x = 4 and x ≤ 4/3. We will show that
E(a, b, c, d) ≥ E(a, x, x, x) ≥ 0,
L
.M
where
E(a, b, c, d) = 3(a2 + b2 + c 2 + d 2 ) + 4a bcd − 16.
D
The left side inequality is equivalent to
A
3(3x 2 − bc − cd − d b) ≥ 2a(x 3 − bcd).
By Schur’s inequality PI
M
(b + c + d)3 + 9bcd ≥ 4(b + c + d)(bc + cd + d b),
LY
we get
4x
x 3 − bcd ≤ (3x 2 − bc − cd − d b).
O
3
Therefore, it suffices to prove that
.M
8a x
3(3x 2 − bc − cd − d b) ≥ (3x 2 − bc − cd − d b);
3
W
that is,
(3x 2 − bc − cd − d b)(9 − 8a x) ≥ 0.
W
and
3(9 − 8a x) = 27 − 8a(4 − a) = 8(1 − a)2 + 16(1 − a) + 3 > 0.
The right side inequality is also true, since
E(a, x, x, x) = 3a2 + 9x 2 + 4a x 3 − 16
= 3(4 − 3x)2 + 9x 2 + 4(4 − 3x)x 3 − 16
= 4(8 − 18x + 9x 2 + 4x 3 − 3x 4 )
= 4(1 − x)2 (2 + x)(4 − 3x) ≥ 0.
Symmetric Polynomial Inequalities in Nonnegative Variables 275
a1 + a2 + · · · + an = n.
L
n−1
.M
then
a1k + a2k + · · · + ank − n ≥ nr(1 − a1 a2 · · · an ).
D
A
PI
P 3.105. Let a, b, c, d be nonnegative real numbers such that
M
a + b + c + d = 4.
LY
Prove that
O
Solution. Use the mixing variable method. Without loss of generality, assume that a ≥
b ≥ c ≥ d. Setting x = (a + b + c)/3, we have 3x + d = 4, d ≤ x ≤ 4/3 and x 3 ≥ a bc.
W
E(a, b, c, d) ≥ E(x, x, x, d) ≥ 0.
The left inequality is equivalent to
W
By Schur’s inequality
we get
9x 3 + 3a bc ≥ 4x(a b + bc + ca),
and hence
3(x 3 − a bc)
3x 2 − a b − bc − ca ≥ ≥ 0.
4x
276 Vasile Cîrtoaje
3d
27(1 + ) ≥ 44d.
4x
Write this inequality in the homogeneous form
or, equivalently,
81(4x 2 + d 2 ) ≥ 353x d.
This inequality is true, since
L
.M
81(4x 2 + d 2 ) − 353x d ≥ 81(4x 2 + d 2 − 5ut) = 81(x − d)(4x − d) ≥ 0.
D
The right inequality E(x, x, x, t) ≥ 0 is also true, since
A
E(x, x, x, d) = (44x 3 − 81x 2 )d − 27x 3 + 64
PI
= 4(16 − 81x 2 + 98x 3 − 33x 4 )
= 4(1 − x)2 (16 + 32x − 33x 2 )
M
= 4(1 − x)2 (4 − 3x)(4 + 11x) ≥ 0.
LY
4 4 4
(a, b, c, d) = 0, , , or any cyclic permutation.
3 3 3
.M
W
1 1 1 1
a+b+c+d = + + + .
a b c d
W
Prove that
1 1 1 1
2 2 2 2
(1 − a bcd) a + b + c + d − 2 − 2 − 2 − 2 ≥ 0.
a b c d
(Vasile Cîrtoaje, 2007)
Solution. From 2
1 1 1 1
2
(a + b + c + d) = + + + ,
a b c d
we get
X X 1 X 1 X
a2 − = 2 − 2 ab
a2 s ym
ab s ym
Symmetric Polynomial Inequalities in Nonnegative Variables 277
X 1 X 1
=2 − cd = 2(1 − a bcd) .
s ym
ab s ym
ab
Thus, the inequality can be restated as
X 1
2(1 − a bcd)2 ≥ 0,
s ym
ab
L
1 1 1 1
a+b+c+d = + + + ,
.M
a b c d
then
D
1 1 1 1
n n n n
(1 − a bcd) a + b + c + d − n − n − n − n ≥0
a b c d
A
for any integer n ≥ 2.
PI
M
LY
a + b + c + d = 1.
O
Prove that
.M
1 1 1 1 81
(1 − a)(1 − b)(1 − c)(1 − d) + + + ≥ .
a b c d 16
W
(Keira, 2007)
81
W
E(a, b, c, d) ≥ ,
16
where
1 1 1 1
E(a, b, c, d) = (1 − a)(1 − b)(1 − c)(1 − d) + + + .
a b c d
Without loss of generality, assume that a ≤ b ≤ c ≤ d. First, we show that for a ≤ b ≤
c ≤ d and a + b + c + d = 1, F (a, b, c, d) is minimal when a = c. This is true if
a+c a+c
E(a, b, c, d) ≥ E( , b, , d).
2 2
Since
(1 − a)(1 − c) = 1 − a − c + ac = b + d + ac
278 Vasile Cîrtoaje
and
(1 − b)(1 − d) = 1 − b − d + bd = a + c + bd,
we have
a+c b+d
E(a, b, c, d) = (b + d + ac)(a + c + bd) + ,
ac bd
and the inequality is equivalent to
L
or
4bd
.M
(a − c)2 − a − c ≥ 0.
ac
Since
D
4bd
− a − c ≥ 4 − a − c = 3 + b + d > 0,
A
ac
PI
the last inequality is clearly true. Since E(a, b, c, d) is minimal when a = c, from a ≤
b ≤ c ≤ d it follows that E(a, b, c, d) is minimal when a = b = c. Therefore, it suffices
M
to prove that 3a + d = 4 involves
LY
81
E(a, a, a, d) ≥ .
16
O
This is equivalent to
.M
a + b + c + d = a3 + b3 + c 3 + d 3 = 2.
Prove that
7
a2 + b2 + c 2 + d 2 ≥ .
4
(Vasile Cîrtoaje, 2010)
Symmetric Polynomial Inequalities in Nonnegative Variables 279
L
[ a(5 − 4a)]2 (5 a − 4 a2 )2
P P P
X (5 − 2x)2
.M
2
a (5 − 4a) ≥ P = = .
(5 − 4a)
P
20 − 4 a 3
D
Therefore, we have
(5 − 2x)2
5x ≥ 8 +
A
.
3
PI
7
This is equivalent to (4x − 7)(x − 7) ≤ 0, and involves x ≥ . The equality holds for
4
M
1 1 1 7
(a, b, c, d) = , , , or any cyclic permutation.
12 12 12 4
LY
O
Solution. Assume that a ≥ b ≥ c ≥ d. For the nontrivial case where the right side of
the inequality is positive, there are two cases to consider.
W
4
(5 − 2a) + (5 − 2b) + (5 − 2c) + (5 − 2d)
(5 − 2a)(5 − 2b)(5 − 2c)(5 − 2d) ≤
4
4 p
4 4
10 − (a + b + c + d) 10 − 4 a bcd
= ≤ = 81,
2 2
L
.M
This is equivalent to
p p
2 a b + 1 5 − 2 cd
p ≥ p ,
D
2 a b − 5 1 + 2 cd
p p
A
2 ab + 1 5 ab − 2
p ≥ p ,
2 ab − 5
p p
ab + 2PI
M
(4 a b − 1)(4 − a b) ≥ 0.
p
The last inequality is true, since a, b ∈ (0, 4] involves 4 − a b ≥ 0.
LY
The equality holds for a = b = c = d = 1, and for (a, b, c, d) = (4, 4, 1/4, 1/4) or any
O
permutation thereof.
.M
W
1 1 1 1
(a + b + c + d) + + + = k.
a b c d
W
p
If 16 ≤ k ≤ (1 + 10)2 , then any three of a, b, c, d are the lengths of the sides of a triangle
(non-degenerate or degenerate).
1 1 1 1
(a + b + c + d) + + + ≥ 16.
a b c d
and hence
L
1 4 p
Æ
(a + c + d) + ≤ ( 11 + 2 10 − 1)2 = 10.
.M
a c+d
Writing this inequality as
(a − c − d)(4a − c − d) ≤ 0,
D
A
we get a ≤ c + d. Thus, the proof is completed.
p
PI
Remark 1. The interval [16, 11 + 2 10] is the largest possible range of k such that any
three of a, b, c, d are the lengths of the sides ofpa triangle. In order to prove this, for the
M
sake of contradiction, assume that k > 11 + 2 10. The given relation
LY
1 1 1 1
(a + b + c + d) + + + =k
a b c d
O
v
t a(a + 2)
is satisfied for a = p + p2 − 1, b = and c = d = 1, where
p
.M
2a + 1
p
( k − 1)2 − 5
W
p= .
4
p
W
If k > 11 + 2 10, then we get p > 5/4 and a > 2. Clearly, the numbers a, c and d are
not the lengths of the sides of a triangle.
W
1 1 1
(a1 + a2 + · · · + an ) + + ··· + = k.
a1 a2 an
p
If n ≥ 3 and n2 ≤ k ≤ (n + 10 − 3)2 , then any three of a1 , a2 , . . . , an are the lengths of
the sides of a triangle.
282 Vasile Cîrtoaje
1 1 1 1
(a + b + c + d) + + + = k.
a b c d
119
If 16 ≤ k ≤ , then there exist three numbers of a, b, c, d which are the lengths of the
6
sides of a triangle (non-degenerate or degenerate).
L
1 1 1 1
.M
(a + b + c + d) + + + ≥ 16.
a b c d
Without loss of generality, assume that a ≥ b ≥ c ≥ d. We need to show that either
D
a ≤ b + c or b ≤ c + d. For the sake of contradiction, consider that a > b + c and
A
b > c + d. To complete the proof, it suffices to show that k > 119/6; that is,
(a + b + c + d)
1 1 1 1
+ + +
PI
>
119
.
M
a b c d 6
Notice that for a = 3, b = 2 and c = d = 1, we have a = b + c, b = c + d and
LY
1 1 1 1 119
(a + b + c + d) + + + = .
O
a b c d 6
.M
9 16
[a + (b + c + d)] + ≥ (3 + 4)2 = 49.
a b+c+d
W
1 1 1 1 17 9 16
+ + + > +
W
.
a b c d 42 a b+c+d
Since
1 1 1 1 4
(b + c + d) + + − 10 ≥ (b + c + d) + − 10
b c d b c+d
(b − c − d)(4b − c − d)
= > 0,
b(c + d)
we need to prove that
1 10 17 9 16
+ ≥ + .
a b+c+d 42 a b+c+d
Symmetric Polynomial Inequalities in Nonnegative Variables 283
This is equivalent to
4a ≥ 3(b + c + d),
which is true, since
Remark. The interval [16, 119/6] is the largest domain of k such that among a, b, c, d
there exist three numbers as lengths of the sides of a non-degenerate or degenerate tri-
angle. In the case of the non-degenerate triangles, the largest domain of k is [16, 119/6).
To prove this, for the sake of contradiction, assume that k ≥ 119/6. The hypothesis
L
1 1 1 1
(a + b + c + d) + + + =k
.M
a b c d
p
is satisfied for a = (k − 11 + k2 − 22k + 81 )/5, b = 2 and c = d = 1. Since a ≥ 3 for
D
k ≥ 119/6, there exist not three numbers of a, b, c, d as lengths of the sides of a triangle.
A
PI
M
P 3.112. Let a, b, c, d and k be positive real numbers such that
LY
(a + b + c + d)2 = k(a2 + b2 + c 2 + d 2 ).
O
11
If ≤ k ≤ 4, then any three of a, b, c, d are the lengths of the sides of a triangle (non-
3
.M
degenerate or degenerate).
(a + b + c + d)2 ≤ 4(a2 + b2 + c 2 + d 2 ).
W
(a + b + 2x)2 ≥ k(a2 + b2 + 2x 2 ),
a + 2x 2
k
(k − 1) b − + k(a2 + 2x 2 ) − (a + 2x)2 ≤ 0.
k−1 k−1
284 Vasile Cîrtoaje
This involves
1
a2 + 2x 2 − (a + 2x)2 ≤ 0.
k−1
11
Since k ≥ , we get
3
3
(a2 + 2x 2 ) − (a + 2x)2 ≤ 0,
8
(a − 2x)(5a − 2x) ≤ 0,
a − b − c ≤ 0.
Thus, the proof is completed.
11
L
Remark 1. The interval [ , 4] is the largest possible. The obvious inequality
.M
3
(a + b + c + d)2 > a2 + b2 + c 2 + d 2
D
involves k > 1. For the sake of contradiction, assume that 1 < k < 11/3. It is easy to
A
check that the hypothesis
PI
(a + b + c + d)2 = k(a2 + b2 + c 2 + d 2 )
p
7 + k(66 − 17k)
M
is satisfied for a = > 4, b = 3 and c = d = 2. Since a > 4 for
k−1
1 < k < 11/3, a, c and d are not the lengths of the sides of a triangle.
LY
3
a triangle.
Notice that the interval [n−1/3, n] is the largest possible of k. From the known inequal-
W
ities
W
(a + b + c + d)2 = k(a2 + b2 + c 2 + d 2 ).
If 49/15 ≤ k ≤ 4, then there exist three numbers of a, b, c, d which are the lengths of the
sides of a triangle (non-degenerate or degenerate).
(a + b + c + d)2 ≤ 4(a2 + b2 + c 2 + d 2 ).
L
.M
Without loss of generality, assume that a ≥ b ≥ c ≥ d. We need to show that either
a ≤ b + c or b ≤ c + d. For the sake of contradiction, consider that a > b + c and
D
b > c + d. To complete the proof, it suffices to show that k < 49/15; that is,
A
49 2
(a + b2 + c 2 + d 2 ) > (a + b + c + d)2 .
PI
15
M
Notice that for a = 3, b = 2 and c = d = 1, we have a = b + c, b = c + d and
LY
49 2
(a + b2 + c 2 + d 2 ) = (a + b + c + d)2 .
15
O
a2 (b + c + d)2
(3 + 4) + ≥ (a + b + c + d)2 .
3 4
W
7 2 a2 (b + c + d)2
(a + b2 + c 2 + d 2 ) > + .
W
15 3 4
This is equivalent to
15
2a2 + 7(b2 + c 2 + d 2 ) > (b + c + d)2 .
4
Since
21 15
2a2 + (b + c + d)2 ≥ (b + c + d)2 .
8 4
This reduces to
4a ≥ 3(b + c + d),
L
Remark. The interval [49/15, 4] is the largest possible domain of k. In the case of the
.M
non-degenerate triangles, the interval (49/15, 4] of k is the largest possible domain. To
prove this, we see that the known inequalities
D
(a + b + c + d)2 > a2 + b2 + c 2 + d 2
A
and PI
(a + b + c + d)2 ≤ 4(a2 + b2 + c 2 + d 2 )
M
involves 1 < k ≤ 4. For the sake of contradiction, assume that 1 < k ≤ 49/15. The
LY
hypothesis
(a + b + c + d)2 = k(a2 + b2 + c 2 + d 2 )
O
p
4 + 2k(11 − 3k)
.M
4(a4 + b4 + c 4 + d 4 + e4 ) ≥ (a2 + b2 + c 2 + d 2 + e2 )2 ;
(b) If a + b + c = d + e, then
12(a4 + b4 + c 4 + d 4 + e4 ) ≤ 7(a2 + b2 + c 2 + d 2 + e2 )2 .
L
3
.M
= 3(d + e)2 − 3d 2 − 3e2 − 5d e = d e ≥ 0.
Also, in virtue of the Cauchy-Schwarz inequality, we have
D
E(a, b, c, d + e, 0) = 4[a4 + b4 + c 4 + (d + e)4 ] − [a2 + b2 + c 2 + (d + e)2 ]2 ≥ 0.
A
The equality holds for a = b = c = d and e = 0, or for a = b = c = e and d = 0.
PI
(b) Let
M
E(a, b, c, d, e) = 7(a2 + b2 + c 2 + d 2 + e2 )2 − 12(a4 + b4 + c 4 + d 4 + e4 ).
LY
Also, we have
1 1
E(a, b, c, d + e, 0) = E(a, b, c, a + b + c, 0)
4 4
X X X X
= a4 + 2 a b(a2 + b2 ) + 3 a2 b2 − 8a bc a
X X X X
≥ a2 b2 + 4 a2 b2 + 3 a2 b2 − 8a bc a
X X X
= 8( a2 b2 − a bc a) = 4 a2 (b − c)2 ≥ 0.
The equality holds for a = b = c = d/3 and e = 0, or for a = b = c = e/3 and d = 0.
288 Vasile Cîrtoaje
a + b + c + d + e = 5.
Prove that
a4 + b4 + c 4 + d 4 + e4 + 150 ≤ 31(a2 + b2 + c 2 + d 2 + e2 ).
(Vasile Cîrtoaje, 2007)
Solution. Write the inequality as
X
(a4 − 31a2 + 30a) ≤ 0,
L
or X
(1 − a) f (a) ≤ 0,
.M
where
f (a) = a3 + a2 − 30a.
D
Without loss of generality, assume that a ≥ b ≥ c ≥ d ≥ e. Since a + b ≤ 5, we have
A
PI
f (a) − f (b) = (a − b)(a2 + a b + b2 + a + b − 30)
≤ (a − b)[(a + b)2 + a + b − 30]
M
= (a − b)(a + b − 5)(a + b + 6) ≤ 0.
LY
Similarly,
f (b) − f (c) ≤ 0, f (c) − f (d) ≤ 0, f (d) − f (e) ≤ 0.
O
Since
.M
X X X
5 (a − 1) f (a) ≤ (a − 1) f (a) = 0.
W
a1 + a2 + · · · + an = n,
then
a14 + a24 + · · · + an4 + n2 (n + 1) ≤ (n2 + n + 1)(a12 + a22 + · · · + an2 ).
Symmetric Polynomial Inequalities in Nonnegative Variables 289
a2 + b2 + c 2 + d 2 + e2 = 5.
Prove that
a bcd e(a4 + b4 + c 4 + d 4 + e4 ) ≤ 5.
(Vasile Cîrtoaje, 2006)
L
E(a, b, c, d, e) = a bcd e(a4 + b4 + c 4 + d 4 + e4 )
.M
is maximal for a = d. We need to show that
D
v v
t a2 + d 2 t a2 + d 2
E(a, b, c, d, e) ≤ E , b, c, ,e .
A
2 2
which is equivalent to
O
b4 + c 4 + e4 ≥ 2a2 d 2 .
W
Indeed, we have
W
From
9 1 4x 4(x − 1)(x − 2)2
f 0 (x) = − − = ,
4 + x x 4 + x2 x(4 + x)(4 + x 2 )
it follows that f (x) is decreasing for 0 < x ≤ 1 and increasing for x ≥ 1. Therefore,
f (x) ≥ f (1) = 0. This completes the proof. The equality holds if and only if a = b =
c = d = e = 1.
p p p p p
Second Solution. Replacing a, b, c, d, e by a, b, c, d, e, we need to show the
homogeneous inequality
a+b+c+d+e 4
p
2 2 2 2 2
a bcd e(a + b + c + d + e ) ≤ 5 ,
5
L
.M
where a, b, c, d, e are positive real numbers. According to Remark from P 3.57, it suffices
to prove this inequality for b = c = d = e = 1; that is, to show that
D
p a + 4 9/2
a(a2 + 4) ≤ 5 .
A
5
Taking logarithms of both sides, we need to prove that f (a) ≥ 0, where
PI
f (a) = 9 ln (a + 4) − 7 ln 5 − ln a − 2 ln (a2 + 4).
M
From
LY
it follows that f (a) is decreasing for 0 < a ≤ 1 and increasing for a ≥ 1; therefore,
f (a) ≥ f (1) = 0.
.M
Remark. The following more general statement holds (Vasile Cirtoaje, 2006):
• If a1 , a2 , . . . , an are positive real numbers such that
W
a1 + a2 + · · · + an = n,
W
then
W
p1
(a1 a2 · · · an ) n−1 (a12 + a22 + · · · + an2 ) ≤ n.
a + b + c + d + e = 5.
Prove that
1 1 1 1 1 20
+ + + + + 2 ≥ 9.
a b c d e a + b + c 2 + d 2 + e2
2
1 1 1 1 20
E(a, b, c, d, e) = + + + + 2
a b c e a + b2 + c 2 + d 2 + e2
is minimal when a = d. If this is true, then E(a, b, c, d, e) is minimal when a = b = c = d,
and it suffices to prove the desired inequality for a = b = c = d, when it is equivalent
to the obvious inequality
(a − 1)2 (6a − 5)2 ≥ 0.
Therefore, it remains to show that
L
a+d a+d
.M
E(a, b, c, d, d) ≥ E , b, c, ,e .
2 2
D
This inequality is equivalent to
A
(a − d)2 20(a − d)2
≥ 2 .
PI
ad(a + d) (a + b2 + c 2 + d 2 + e2 )[(a + d)2 + 2b2 + 2c 2 + 2e2 ]
M
Since (a − d)2 ≥ 0 and
LY
1
a2 + b2 + c 2 + d 2 + e2 ≥ (a + b + c + d + e)2 = a + b + c + d + e,
5
O
Since
W
a + b + c + d + e ≥ a + a + a + d + d = 3a + 2d
W
and
W
Remark. The following more general statement holds (Vasile Cirtoaje, 2006):
• If a1 , a2 , . . . , an are positive real numbers such that a1 + a2 + · · · + an = n, then
p
1 1 1 2n n − 1 p
+ + ··· + + 2 ≥ n + 2 n − 1.
a1 a2 an a1 + a2 + · · · + an2
2
L
.M
P 3.118. If a, b, c, d, e ≥ 1, then
D
1 1 1 1 1
a+ b+ c+ d+ e+ + 68 ≥
A
a b c d e
≥ 4(a + b + c + d + e)
1 1 1 1 1
+ + + +
a b c d e
PI
.
M
(Vo Quoc Ba Can and Vasile Cîrtoaje, 2011)
LY
1 1 1 1
A= a+ b+ c+ d+ .
a b c d
.M
We claim that
E(a, b, c, d, e) ≥ E(a, b, c, d, 1).
W
From A ≥ 16 and
1 1 1 1
+ + + ≤ 4,
a b c d
it follows that B ≥ 0. If C ≤ 0, then B − C/e ≥ 0. If C ≥ 0, then
1 1 1 1
C
B− ≥B−C =4 a− +4 b− +4 c− +4 d − ≥ 0.
e a b c d
L
.M
P 3.119. Let a, b, c and x, y, z be positive real numbers such that
D
A
Prove that
1
a bc x yz <
36
. PI
M
(Vasile Cîrtoaje and Mircea Lascu, 1997)
LY
Solution. Using the given relations and the AM-GM inequality, we have
4(a b + bc + ca)(x y + yz + z x) =
O
Æ
≤ 20 − 2(a + b + c)(x + y + z) (a2 + b2 + c 2 )(x 2 + y 2 + z 2 ) = 4,
W
and hence
(a b + bc + ca)(x y + yz + z x) ≤ 1.
W
(a b + bc + ca)2 ≥ 3a bc(a + b + c)
and
(x y + yz + z x)2 ≥ 3x yz(x + y + z),
we get
(a b + bc + ca)2 (x y + yz + z x)2 ≥ 36a bc x yz,
and hence
1 ≥ (a b + bc + ca)2 (x y + yz + z x)2 ≥ 36a bc x yz.
294 Vasile Cîrtoaje
L
a1 + a2 + · · · + an = a12 + a22 + · · · + an2 = n − 1.
.M
Prove that
1 1 1 n2 (2n − 3)
+ + ··· + ≥ .
D
a1 a2 an 2(n − 1)(n − 2)
A
(Vasile Cîrtoaje, 2010)
Solution. By the Cauchy-Schwarz inequality, we have PI
(a1 + a2 + · · · + an−1 )2
M
n − 1 = a12 + a22 + · · · + an2 ≥ + an2
n−1
LY
(n − 1 − an )2
= + an2 ,
n−1
O
which provides an ≤ 2(n − 1)/n. Similarly, ai ≤ 2(n − 1)/n for all i. The hint for
proving the given inequality is to apply the Cauchy-Schwarz inequality after we made
.M
X 1 X 1 n2
n
= − +
a1 a1 2n − 2 2n − 2
W
1 X 2n − 2 − na1 n2
= +
W
2(n − 1) a1 2n − 2
[ (2n − 2 − na1 )]2
P
1 n2
≥ ·P +
2(n − 1) a1 (2n − 2 − na1 ) 2n − 2
[n(2n − 2) − n a1 ]2
P
1 n2
= · P 2+
2(n − 1) (2n − 2) a1 − n a1 2n − 2
P
1 n2 (n − 1)2 n2 n2 (2n − 3)
= · + = ,
2(n − 1) (n − 1)(n − 2) 2n − 2 2(n − 1)(n − 2)
from where the conclusion follows. The equality holds for a1 = a2 = · · · = an−1 =
1 − 2/n and an = 2 − 2/n (or any cyclic permutation).
Symmetric Polynomial Inequalities in Nonnegative Variables 295
L
.M
a12 + a22 + · · · + an2 = n + n(n − 1)t 2 ,
D
(a2 + · · · + an )2 (n − a1 )2
A
a12 + a22 + · · · + an2 ≥ a12 + = a12 + ,
n−1 n−1
we get
PI
(n − a1 )2
M
n + n(n − 1)t 2 ≥ a12 + ,
n−1
LY
ai ≤ 1 + (n − 1)t
.M
for any i. The hint for proving the given inequality is to apply the Cauchy-Schwarz
inequality after we made the numerators nonnegative and as small as possible. Since
X 1 1 X 1 1
W
X
= + −
a1 1 + (n − 1)t a1 1 + (n − 1)t
W
n 1 X 1 + (n − 1)t − a
1
= + ,
1 + (n − 1)t 1 + (n − 1)t a1
W
[ (1 + (n − 1)t − a1 )]2
P
X 1 + (n − 1)t − a
1
≥ P
a1 a1 (1 + (n − 1)t − a1 )
[n + n(n − 1)t − a1 ]2
P
n(n − 1)t
= P 2= ,
(1 + (n − 1)t) a1 − a1
P
1− t
296 Vasile Cîrtoaje
Indeed,
L
The equality holds when a1 = a2 = · · · = an = 1, as well as when one of a1 , a2 , . . . , an is
.M
n/2 and the others are n/(2n − 2).
D
A
P 3.122. Let a1 , a2 , · · · , an be nonnegative real numbers such that a1 + a2 + · · · + an = n.
PI
Prove that
M
(n + 1)(a12 + a22 + · · · + an2 ) ≥ n2 + a13 + a23 + · · · + an3 .
(Vasile Cîrtoaje, 2002)
LY
ai ≤ 1 + (n − 1)t.
W
[ (1 + (n − 1)t − a1 )a1 ]2
P
X
2
[1 + (n − 1)t − a1 ]a1 ≥ P = n(n − 1)t(1 − t)2 ,
[1 + (n − 1)t − a1 ]
hence
X X
a13 ≤ [1 + (n − 1)t] a12 − n(n − 1)t(1 − t)2
= n[(n − 1)(n − 2)t 3 + 3(n − 1)t 2 + 1].
For n = 2, the original inequality is an identity. For n ≥ 3, the equality holds when
a1 = a2 = · · · = an = 1, as well as when n − 1 of a1 , a2 , . . . , an are zero.
Second Solution. Assume that a1 ≥ a2 ≥ · · · ≥ an . Replacing n2 by n(a1 + a2 + · · · + an ),
the desired inequality becomes as follows
X
[(n + 1)a12 − na1 − a13 ] ≥ 0,
L
X
(a1 − 1)(na1 − a12 ) ≥ 0.
.M
Since
a1 − 1 ≥ a2 − 1 ≥ · · · ≥ an − 1
D
and
A
na1 − a12 ≥ na2 − a22 ≥ · · · ≥ nan − an2 ,
we apply Chebyshev’s inequality to get
PI
M
X X X
n (a1 − 1)(na1 − a12 ) ≥ (a1 − 1) (na1 − a12 ) = 0.
LY
O
.M
ai ≥ 1 − (n − 1)t.
[ (a1 − 1 + (n − 1)t)a1 ]2
P
X
2
[a1 − 1 + (n − 1)t]a1 ≥ P = n(n − 1)t(t + 1)2 ,
[a1 − 1 + (n − 1)t]
298 Vasile Cîrtoaje
which yields
X X
a13 ≥ n(n − 1)t(t + 1)2 + [1 − (n − 1)t] a12
= n[1 + 3(n − 1)t 2 − (n − 1)(n − 2)t 3 ].
L
n(n − 1)(n − 2)t 2 [1 − (n − 1)t] ≥ 0.
.M
The equality holds for a1 = a2 = · · · = an = 1, and also for a1 = 0 and a2 = a3 = · · · =
n
an = (or any cyclic permutation).
D
n−1
A
P 3.124. Let a1 , a2 , . . . , an (n ≥ 3) be positive real numbers such that
PI
M
a1 + a2 + · · · + an = 1.
LY
Prove that
O
n
1 1 1 1
a1 + −2 a2 + −2 an + −2 ≥ n+ −2 .
.M
a1 a2 an n
W
Solution. Applying Popoviciu’s inequality for the convex function f (x) = − ln x, x > 0,
gives
W
a + a + · · · + a n(n−2)
1 2 n
(b1 b2 · · · bn )n−1 ≥ (a1 a2 · · · an ) ,
n
W
where
1 X
bi = a j , i = 1, 2, · · · , n.
n − 1 j6=i
1 n
1− ≥ (1 − a1 )(1 − a2 ) · · · (1 − an ).
n
1 n(n−3)
1− ≥ (1 − a1 )n−3 (1 − a2 )n−3 · · · (1 − an )n−3 ,
n
1 n(n−3)
1− (1 − a1 )2 (1 − a2 )2 · · · (1 − an )2 ≥ (1 − a1 )n−1 (1 − a2 )n−1 · · · (1 − an )n−1 .
n
Multiplying this inequality and (*) yields the desired inequality. The equality holds for
a1 = a2 = · · · = an = 1/n.
L
.M
P 3.125. Let a1 , a2 , . . . , an (n ≥ 3) be positive real numbers such that
a1 + a2 + · · · + an = n.
D
A
Prove that
n
a12 + a22 + · · · + an2 − n ≥ (1 − a1 a2 · · · an ).
PI
n−1
M
Solution. For fixed a12 + a22 + · · · + an2 , according to Remark from P 3.57, the product
LY
n2
a22 + a32 + · · · + an2 ≥ .
n−1
W
1 n2
a22 + a32 + · · · + an2 ≥ (a2 + a3 + · · · + an )2 = .
n−1 n−1
W
x + (n − 1) y = n
involves
n
x 2 + (n − 1) y 2 − n +
x y n−1 − 1 ≥ 0.
n−1
By Bernoulli’s inequality, we have
1 − x n−1
y n−1 = 1 + ≥ 1 + (1 − x) = 2 − x.
n−1
300 Vasile Cîrtoaje
L
P 3.126. Let a1 , a2 , . . . , an (n ≥ 3) be positive real numbers such that
.M
a1 ≤ a2 ≤ · · · ≤ an ,
D
1 1 1
(a1 + a2 + · · · + an ) + + ··· + = k.
A
a1 a2 an
(a) If n2 ≤ k ≤ n2 +
i(n − i)
2
PI
, i ∈ {2, 3, · · · , n−1}, then ai−1 , ai and ai+1 are the lengths
of the sides of a non-degenerate or degenerate triangle;
M
9n2 9n2 − 1
LY
triangle.
.M
1 1 1
(a1 + a2 + · · · + an ) + + ··· + ≥ n2 .
W
a1 a2 an
W
(a) For the sake of contradiction, assume that ai−1 , ai and ai+1 are not the lengths of
the sides of a triangle; that is, ai+1 > ai−1 + ai . Let x and y be positive numbers such
that
(i − 1)x = a1 + · · · + ai−1 ,
(n − i) y = ai+1 + · · · + an ,
x ≤ ai−1 ≤ ai < ai+1 ≤ y, x < y.
Let us denote
A(x, y) = (i − 1)x + ai + (n − i) y,
i−1 1 n−i
B(x, y) = + + ,
x ai y
Symmetric Polynomial Inequalities in Nonnegative Variables 301
L
On the other hand, we claim that
.M
f (x, y) > f (ai−1 , ai−1 + ai ).
D
This inequality is equivalent to
A
[ai−1 y − (ai−1 + ai )x][(ai−1 + ai ) y − ai−1 x] ≥ 0,
and is true since y ≥ ai+1 > ai−1 + ai and x ≤ ai−1 imply PI
ai−1 y − (ai−1 + ai )x > ai−1 (ai−1 + ai ) − (ai−1 + ai )ai−1 = 0.
M
Then, we have
LY
i−1 1 n−i
= [(n − 1)ai−1 + (n − i + 1)ai ] + + .
ai−1 ai ai−1 + ai
.M
k> ≥ n2 + ,
t(1 + t) 2
W
which contradicts the hypothesis k ≤ n2 + i(n − i)/2. The last inequality is equivalent
to (t − 1)(C t 2 + Dt + E) ≥ 0, where
W
a1 ≤ a2 ≤ · · · ≤ an ,
(2n − i)2
(a) If ≤ k ≤ n, i ∈ {2, 3, · · · , n − 1}, then ai−1 , ai and ai+1 are the lengths of
4n − 3i
the sides of a non-degenerate or degenerate triangle;
8n + 1
(b) If ≤ k ≤ n, then there exist three numbers ai which are the lengths of the
9
sides of a non-degenerate or degenerate triangle.
L
.M
(Vasile Cîrtoaje, 2010)
D
Solution. The condition k ≤ n follows from the Cauchy-Schwarz inequality
A
(a1 + a2 + · · · + an )2 ≤ n(a12 + a22 + · · · + an2 ).
PI
(a) For the sake of contradiction, assume that ai−1 , ai and ai+1 are not the lengths of
M
the sides of a triangle; that is, ai+1 > ai−1 + ai . Let x and y be positive numbers such
that
LY
(i − 1)x = a1 + · · · + ai−1 ,
(n − i) y = ai+1 + · · · + an ,
O
Let us denote
A(x, y) = (i − 1)x + ai + (n − i) y,
W
A2 (x, y)
f (x, y) = .
W
B(x, y)
By the Cauchy-Schwarz inequality, we have
(n − i) y 2 ≤ ai+1
2
+ · · · + an2 ,
and hence
B(x, y) ≤ a12 + a22 + · · · + an2 .
Since
A(x, y) = a1 + a2 + · · · + an ,
Symmetric Polynomial Inequalities in Nonnegative Variables 303
we obtain
(a1 + a2 + · · · + an )2
f (x, y) ≥ ,
a12 + a22 + · · · + an2
and hence
f (x, y) ≥ k.
On the other hand, from
∂ f (x, y) 2(i − 1)AC
= >0
∂x B2
and
∂ f (x, y) 2(n − i)AD
= < 0,
L
∂y B2
.M
where
C = ai (ai − x) + (n − i) y( y − x) > 0,
D
D = ai (ai − y) + (i − 1)x(x − y) < 0,
A
it follows that f (x, y) is strictly increasing with respect to x > 0 and strictly decreasing
PI
with respect to y > 0. Then, since x ≤ ai−1 and y ≥ ai+1 > ai−1 + ai , we have
M
f (x, y) < f (ai−1 , ai−1 + ai ).
LY
This involves
k < f (ai−1 , ai−1 + ai ),
O
and hence
.M
=
(n − 1)ai−1
2
+ 2(n − i)ai−1 ai + (n − i + 1)ai2
W
(2n − i)2
≤ ,
W
4n − 3i
which is false. Setting ai−1 = 1 and ai = t, t ≥ 1, the last inequality
Since
E = (n − i + 1)[(3i − 4)(n − i + 1) + i 2 + 2i − 4] ≥ 0,
we get
E t − F ≥ E − F = 2(i − 2)(n − i)(2n − i) ≥ 0.
(b) According to (a), it suffices to show that there exists i ∈ {2, 3, · · · , n − 1} such that
(2n − i)2 8n + 1
≤ ;
4n − 3i 9
that is,
4n − 3i ≥ (2n − 3i)2 .
L
.M
2n − 1 2n 2n + 1
Since one of the numbers , and is integer, it suffices to prove this in-
3 3 3
2n − 1 2n 2n + 1
D
equality for all i ∈ { , , }. Indeed, for these cases, the inequality reduces
3 3 3
to 2n ≥ 0, 2n ≥ 0 and 2n − 2 ≥ 0, respectively.
A
PI
M
LY
O
.M
W
W
W
Bibliography
[1] Andreescu T., Cîrtoaje V., Dospinescu G., Lascu M., Old and New Inequalities, GIL
L
Publishing House, 2004.
.M
[2] Bin X., Boreico I., Can V.Q.B., Bulj A., Lascu M., Opympiad Inequalities, GIL Pub-
lishing House, 2015.
D
A
[3] Bin X., Boreico I., Can V.Q.B., Cîrtoaje V., Lascu M., An Introduction to Inequalities,
GIL Publishing House, 2015. PI
[4] Can V.Q.B., Pohoaţă C., Old and New Inequalities, GIL Publishing House, 2008.
M
[5] Can V.Q.B., Anh T. Q., Su Dung Phuong Phap Cauchy-Schwarz De Chung Minh Bat
LY
[7] Cîrtoaje V., The Equal Variable Method, Journal of Inequalities In Pure and Applied
Mathematics, Volume 7, Issue 3, 2006.
W
[9] Cîrtoaje V., Algebraic Inequalities-Old and New Methods, GIL Publishing House,
W
2006.
[10] Cîrtoaje V., Four Applications of RCF and LCF Theorems, Math. Reflections, 1, 2007.
[11] Cîrtoaje V., On Jensen Type Inequalities with Ordered Variables, Journal of Inequal-
ities In Pure and Applied Mathematics, Volume 9, Issue 1, 2008.
[12] Cîrtoaje V., The Proof of Three Open Inequalities, Crux Mathematicorum, Volume
34, Issue 4, 2008, 231-238.
[13] Cîrtoaje V., Can V. Q. B., Anh T. Q., Inequalities with Beautiful Solutions, GIL Pub-
lishing House, 2009.
305
306 Vasile Cîrtoaje
[14] Cîrtoaje V., On Some Inequalities with Power-Exponential Functions, Journal of In-
equalities In Pure and Applied Mathematics, Volume 10, Issue 1, 2009.
[15] Cîrtoaje V., On the Cyclic Homogeneous Polynomial Inequalities of Degree Four, Jour-
nal of Inequalities in Pure and Applied Mathematics, Volume 10, Issue 3, 2009.
[16] Cîrtoaje V., The Best Lower Bound Depended on Two Fixed Variables for Jensen’s In-
equality with Ordered Variables, Journal of Inequalities and Applications, Volume
2010.
[17] Cîrtoaje V., Baiesu A., An Extension of Jensen’s discrete inequality to half convex
functions, Journal of Inequalities and Applications, Volume 2011.
L
.M
[18] Cîrtoaje V., Proofs of Three Open Inequalities with Power-Exponential Functions,
Journal of Nonlinear Sciences and Applications, Volume 4, no. 2, 2011.
D
[19] Cîrtoaje V., The Best Upper Bound for Jensen’s Inequality, Australian Journal of
A
Mathematical Analysis and Aplications, Volume 7, Issue 2, Art. 22, 2011.
PI
[20] Cîrtoaje V., Zhou Y., Necessary and Sufficient Conditions for Cyclic Homogeneous
M
Polynomial Inequalities of Degree Four in Real Variables, The Australian Journal of
Mathematical Analysis and Applications, Volume 9, Issue 1, Article 15, 2012.
LY
[21] Cîrtoaje V., Zhou Y., Some Strong Sufficient Conditions for Cyclic Homogeneous Poly-
O
[22] Cîrtoaje V., Can V.Q.B., On Some Cyclic Homogeneous Polynomial Inequalities of De-
gree Four in Real Variables, International Journal of Pure and Applied Mathematics,
W
[23] Cîrtoaje V., On the Arithmetic Compensation Methods, International Journal of Pure
and Applied Mathematics, Volume 80, No. 3, 2012.
W
[24] Cîrtoaje V., Necessary and Sufficient Conditions for Symmetric Homogeneous Poly-
nomial Inequalities of Degree Four and Six in Real Variables, Journal of Nonlinear
Science and Applications, Volume 5, Issue 5, 2012.
[25] Cîrtoaje V., Necessary and Sufficient Conditions for Symmetric Homogeneous Poly-
nomial Inequalities in Nonnegative Real Variables, Mathematical Inequalities and
Applications, Volume 16, No. 2, 2013.
[26] Cîrtoaje V., The Best Lower Bound for Jensen’s Inequality with three fixed ordered
variables Journal, Banach Journal of Mathematical Analysis, Volume 7, Issue 1,
2013.
Symmetric Polynomial Inequalities in Nonnegative Variables 307
[27] Cîrtoaje V., An Extension of Jensen’s discrete inequality to partially convex functions,
Journal of Inequalities and Applications, Volume 2013:54.
[28] Cîrtoaje V., A Strong Method for Symmetric Homogeneous Polynomial Inequalities
of Degree Six in Nonnegative Real Variables, British Journal of Mathematical and
Computers Science, 4(5), 2014.
[30] Drimbe M.O., Inegalitati - Idei si Metode, GIL Publishing House, 2003.
L
[31] Engel A., Problem-Solving Strategies, Springer, 1998.
.M
[32] Hung P. K., Secrets in Inequalities, Volume 1: Basic Inequalities, GIL Publishing
House, 2007.
D
[33] Hung P. K., Secrets in Inequalities, Volume 2: Advanced Inequalities, GIL Publishing
A
House, 2008.
PI
[34] Lee Hojoo, Topics in Inequalities-Theorems and Techniques,
M
http://www.artofproblemsolving.com/Forum/viewtopic.php?t=602698
LY
[35] Littlewood G. H., Polya, J. E., Inequalities, Cambridge University Press, 1967.
O
[36] Mitrinovic̆ D. S., Pecaric̆ J. E., Fink A. M., Classical and New Inequalities in Analysis,
Kluwer, 1993.
.M
[37] Panaitopol L., Băndilă V., Lascu M., Inegalităţi, GIL Publishing House, 1995.
[38] Zhou Y., Cîrtoaje V., Some strong sufficient conditions for cyclic homogeneous poly-
W
L
D
.M
A
PI
M
LY
O
.M
W
W
W
Chapter 4
Bibliography
L
D
.M
A
PI
M
LY
O
.M
W
W
W
309